REAL AND COMPLEX MANIFOLDS
We shall begin by deﬁning holomorphic functions and the CauchyRiemann equa
tions in C
n
. In Sections 1.21.4 of this chapter we will review the deﬁnitions and
various properties of a smooth real or complex manifold. In Section 1.5, the Cauchy
Riemann complex is introduced on complex manifolds. Section 1.6 is devoted to
the Frobenius theorem. In the last section, in contrast to the Riemann mapping
theorem in one complex variable, we prove the inequivalence between the ball and
the polydisc in several variables.
1.1 Holomorphic Functions in Complex Euclidean Spaces
Let C
n
= C C denote the ndimensional complex Euclidean space with
product topology. The coordinates of C
n
will be denoted by . = (.
1
, , .
n
) with
.
j
= r
j
+ij
j
, 1 ≤ , ≤ n. Thus, C
n
can be identiﬁed with 1
2n
in a natural manner,
. →(r
1
, j
1
, , r
n
, j
n
).
Deﬁnition 1.1.1. A complexvalued C
1
function )(.) deﬁned on an open subset 1
of C
n
is called holomorphic, denoted by ) ∈ O(1), if )(.) is holomorphic in each
variable .
j
when the other variables are ﬁxed. In other words, )(.) satisﬁes
(1.1.1)
∂)
∂.
j
= 0,
for each , = 1, , n, where
(1.1.2)
∂
∂.
j
=
1
2
_
∂
∂r
j
+i
∂
∂j
j
_
is the socalled CauchyRiemann operator.
The objective of this book is to study the behavior of holomorphic functions. It
is closely related to the solvability and regularity of the inhomogeneous Cauchy
Riemann equations
(1.1.3)
∂n
∂.
j
= )
j
, for , = 1, , n,
where )
j
’s are given functions.
1
2 Real and Complex Manifolds
Some of the properties of holomorphic functions, like power series expansion, do
extend from one variable to several variables. They diﬀer, however, in many impor
tant aspects. It is therefore, not correct to consider the theory of several complex
variables as a straightforward generalization of that of one complex variable. For
example, in one variable the zero set of a holomorphic function is a discrete set. The
zero set of a holomorphic function in C
n
, n ≥ 2, has a real 2n −2 dimension. In C,
it is trivial to construct a holomorphic function in a domain 1 which is singular at
one boundary point j ∈ /1. In contrast, in C
n
when n ≥ 2, it is not always possible
to construct a holomorphic function in a given domain 1 ⊂ C
n
which is singular
at one boundary point. This leads to the existence of a domain in several variables
such that any holomorphic function deﬁned on this domain can be extended holo
morphically to a ﬁxed larger set, a feature that does not exist in one variable. In
Chapter 3 we will discuss this phenomenon in detail. Another main diﬀerence is
that there is no analog to the Riemann mapping theorem of one complex variable
in higher dimensional spaces. This phenomenon is analyzed in Section 1.7. Many
of these important diﬀerences will be further investigated in Chapters 46 using
solutions of the inhomogeneous CauchyRiemann equations (1.1.3).
There is yet another major diﬀerence in solving (1.1.3) in one and several vari
ables. When n ≥ 2, a compatibility condition must be satisﬁed in order for Equa
tions (1.1.3) to be solvable:
(1.1.4)
∂)
i
∂.
j
=
∂)
j
∂.
i
, for 1 ≤ i < , ≤ n.
This will be discussed in the next few chapters on bounded domains in C
n
.
We recall here the deﬁnition concerning the diﬀerentiability of the boundary of
a domain.
Deﬁnition 1.1.2. A domain 1 in 1
n
, n ≥ 2, is said to have C
k
(1 ≤ / ≤ ∞)
boundary at the boundary point j if there exists a realvalued C
k
function : deﬁned
in some open neighborhood l of j such that 1∩l = ¦r ∈ l[ :(r) < 0¦, /1∩l =
¦r ∈ l[ :(r) = 0¦ and d:(r) ,= 0 on /1 ∩ l. The function : is called a C
k
local
deﬁning function for 1 near j. If l is an open neighborhood of 1, then : is called
a global deﬁning function for 1, or simply a deﬁning function for 1.
The relationship between two deﬁning functions is clariﬁed in the next lemma.
Lemma 1.1.3. Let :
1
and :
2
be two local deﬁning functions for 1 of class C
k
(1 ≤ / ≤ ∞) in a neighborhood l of j ∈ /1. Then there exists a positive C
k−1
function / on l such that
(1) :
1
= /:
2
on l,
(2) d:
1
(r) = /(r)d:
2
(r) for r ∈ l ∩ /1.
Proof. Since d:
2
,= 0 on the boundary near j, after a C
k
change of coordinates,
we may assume that j = 0, r
n
= :
2
(r) and /1 ∩ l = ¦r ∈ l[ r
n
= 0¦. Let
r
t
= (r
1
, , r
n−1
). Then :
1
(r
t
, 0) = 0. By the fundamental theorem of calculus,
:
1
(r
t
, r
n
) = :
1
(r
t
, r
n
) −:
1
(r
t
, 0) = r
n
_
1
0
∂:
1
∂r
n
(r
t
, tr
n
)dt.
1.2 Real and Complex Manifolds 3
This shows :
1
= /:
2
for some C
k−1
function / on l. For / ≥ 2, we clearly have
(2) and / 0 on l. When / = 1, (2) also follows directly from the deﬁnition of
diﬀerentiation at 0. This proves the lemma.
1.2 Real and Complex Manifolds
Let ` be a Hausdorﬀ space. ` is called a topological manifold of dimension
n if each point j of ` has a neighborhood l
p
homeomorphic to an open subset
\
p
in 1
n
. Let the homeomorphism be given by ϕ
p
: l
p
→ \
p
. We call the pair
(l
p
, ϕ
p
) a coordinate neighborhood of ` near j. Since, for any ¡ ∈ l
p
, ϕ
p
(¡) is
a point in 1
n
, we have the usual Euclidean coordinates (r
1
(ϕ
p
(¡)), , r
n
(ϕ
p
(¡)))
for ϕ
p
(¡). We shall call the set (r
1
(ϕ
p
(¡)), , r
n
(ϕ
p
(¡))) the local coordinates for
the points ¡ in l
p
with respect to the coordinate neighborhood (l
p
, ϕ
p
), and it will
be abbreviated by (r
1
(¡), , r
n
(¡)), and the ntuple (r
1
, , r
n
) of functions on
l
p
will be called the local coordinate system on (l
p
, ϕ
p
).
Let ` be a topological manifold, then ` is covered by a family of such coordinate
neighborhoods ¦(l
α
, ϕ
α
)¦
α∈Λ
, where Λ is an index set. If for some α, β in Λ we
have l
αβ
= l
α
∩ l
β
,= ∅, then there is a welldeﬁned homeomorphism
)
βα
= ϕ
β
◦ ϕ
−1
α
: ϕ
α
(l
αβ
) →ϕ
β
(l
αβ
).
These will be called the transition functions with respect to the coordinate neigh
borhood system ¦(l
α
, ϕ
α
)¦
α∈Λ
. Obviously, we have )
βα
= )
−1
αβ
. Now we give the
deﬁnition of a diﬀerentiable manifold.
Deﬁnition 1.2.1. Let ` be a topological manifold together with a coordinate neigh
borhood system ¦(l
α
, ϕ
α
)¦
α∈Λ
. We call ` an ndimensional diﬀerentiable manifold
of class C
r
, or a C
r
manifold, 1 ≤ : ≤ ∞, if all of the corresponding transition
functions are of class C
r
. If : = ∞, we call ` a smooth manifold. If all of the
corresponding transition functions are real analytic, ` will be called a real analytic
manifold, or a C
ω
manifold.
Next we deﬁne complex manifolds.
Deﬁnition 1.2.2. Let ` be a topological manifold together with a coordinate neigh
borhood system ¦(l
α
, ϕ
α
)¦
α∈Λ
, where ϕ
α
(l
α
) = \
α
are open sets in C
n
. ` is called
a complex manifold of complex dimension n if the transition function )
βα
= ϕ
β
◦ϕ
−1
α
is holomorphic on ϕ
α
(l
αβ
) ⊂ C
n
, whenever l
αβ
= l
α
∩ l
β
,= ∅ for all α, β.
It follows that a complex manifold is automatically a real analytic manifold. Here
are some important examples of real and complex manifolds.
Example 1.2.3. Any connected open subset ` of 1
n
is a real analytic manifold.
The local chart (`, ι) is simply the induced one given by the identity mapping
ι from ` into 1
n
. Similarly, any connected open subset ` of C
n
is a complex
manifold of complex dimension n.
Example 1.2.4 (Real projective space, 1P
n
). Deﬁne an equivalence relation
on the set 1
n+1
¸ ¦0¦. Two points r and j in 1
n+1
¸ ¦0¦ are said to be equivalent
4 Real and Complex Manifolds
if there is a nonzero real number λ ∈ 1
∗
= 1 ¸ ¦0¦ such that r = λj. The set of
equivalence classes given by this equivalence relation is called the real projective
space 1P
n
of dimension n. In other words, 1P
n
can be identiﬁed with the space
of all lines passing through the origin in 1
n+1
. The mapping π from 1
n+1
¸ ¦0¦
onto 1P
n
so that π(r) is the equivalence class containing the point r is continuous,
provided that 1P
n
is equipped with the quotient topology, namely, a subset l of
1P
n
is open if and only if π
−1
(l) is open in 1
n+1
¸ ¦0¦. Since π also maps the
compact set o
n
onto 1P
n
, we see that 1P
n
is compact.
The coordinate neighborhood system ¦(l
j
, ϕ
j
)¦
n+1
j=1
is constructed as follows: for
each j ∈ 1P
n
pick an element r = (r
1
, , r
n+1
) ∈ 1
n+1
¸ ¦0¦ such that π(r) = j.
The point j can be represented by the corresponding homogeneous coordinates
[r
1
: r
2
: : r
n+1
]. This representation is clearly independent of the choice of r.
Let l
j
= ¦[r
1
: : r
j
: : r
n+1
][r
j
,= 0¦ be an open subset of 1P
n
, and the
homeomorphism ϕ
j
from l
j
onto 1
n
is given by
ϕ
j
([r
1
: : r
j
: : r
n+1
])
= (r
1
,r
j
, , r
j−1
,r
j
, r
j+1
,r
j
, , r
n+1
,r
j
).
Hence, if l
i
∩ l
j
= l
ij
,= ∅, say, i < ,, then the transition function )
ij
is
)
ij
(j) = ϕ
i
◦ ϕ
−1
j
(j)
= ϕ
i
([j
1
: : j
j−1
: 1 : j
j
: : j
n
])
=
_
j
1
j
i
, ,
j
i−1
j
i
,
j
i+1
j
i
, ,
j
j−1
j
i
,
1
j
i
,
j
j
j
i
, ,
j
n
j
i
_
.
It follows that the real projective space 1P
n
is a real analytic compact manifold.
Example 1.2.5 (Complex projective space, CP
n
). If C is substituted for 1
in the deﬁnition of the real projective space 1P
n
, we will end up with a compact
complex manifold of complex dimension n which we call the complex projective
space and denote by CP
n
.
Example 1.2.6 (Riemann surface). A Riemann surface ` is by deﬁnition a
complex manifold of complex dimension one. Hence, any open subset l of C is a
Riemann surface. Complex projective space CP
1
is a compact Riemann surface,
also known as the Riemann sphere.
From now on we shall concentrate on complex manifolds, unless the contrary is
stated explicitly in the text. Let ) be a continuous complexvalued function deﬁned
on an open subset l of a complex manifold `, and let j be a point in l. We
say that ) is holomorphic at j if there exists a small open neighborhood \ of j,
contained in l ∩l
α
for some local coordinate neighborhood l
α
, such that ) ◦ϕ
α
−1
is holomorphic on the open subset ϕ
α
(\ ) in C
n
. Clearly, the above deﬁnition of
holomorphic functions at a point j ∈ l is independent of the choice of the local
coordinate system (l
α
, ϕ
α
). The function ) is said to be holomorphic on l if )
is holomorphic at every point j ∈ l. In particular, the local coordinate functions
.
i
, 1 ≤ i ≤ n, on l
α
of a complex manifold are holomorphic.
Let ` and · be two complex manifolds of complex dimensions : and n with
local coordinate systems ¦(l
α
, ϕ
α
)¦
α∈Λ
and ¦(\
β
, ψ
β
)¦
β∈Γ
respectively, and let ) be
1.3 Tangent Spaces and the Hermitian Metric 5
a continuous mapping from ` into ·. We shall say that ) deﬁnes a holomorphic
mapping at j ∈ `, if there exists an open neighborhood l
p
of j, contained in
a local coordinate neighborhood l
α
, with )(l
p
) contained in a local coordinate
neighborhood \
β
such that ψ
β
◦) ◦ϕ
−1
α
deﬁnes a holomorphic mapping from ϕ
α
(l
p
)
into ψ
β
(\
β
). The deﬁnition is easily seen to be independent of the choice of the
local coordinate systems.
If ) is a holomorphic mapping between two complex manifolds ` and · of
equal dimensions such that ) is onetoone, onto and the inverse mapping )
−1
is
also holomorphic, then ) will be called a biholomorphic map or a biholomorphism
from ` onto ·.
1.3 Tangent Spaces and the Hermitian Metric
Let C
n
be identiﬁed with 1
2n
via the map (.
1
, , .
n
) → (r
1
, j
1
, , r
n
, j
n
).
For any point j ∈ C
n
the tangent space T
p
(C
n
) is spanned by
_
∂
∂r
1
_
p
,
_
∂
∂j
1
_
p
, ,
_
∂
∂r
n
_
p
,
_
∂
∂j
n
_
p
.
Deﬁne an 1linear map J from T
p
(C
n
) onto itself by
J
_
∂
∂r
j
_
p
=
_
∂
∂j
j
_
p
, J
_
∂
∂j
j
_
p
= −
_
∂
∂r
j
_
p
,
for all , = 1, , n. Obviously, we have J
2
= −1, and J is called the complex
structure on C
n
.
The complex structure J induces a natural splitting of the complexiﬁed tangent
space CT
p
(C
n
) = T
p
(C
n
)⊗
R
C. First we extend J to the whole complexiﬁed tangent
space by J(r⊗α) = (Jr)⊗α. It follows that J is a Clinear map from CT
p
(C
n
) onto
itself with J
2
= −1, and the eigenvalues of J are i and −i. Denote by T
1,0
p
(C
n
)
and T
0,1
p
(C
n
) the eigenspaces of J corresponding to i and −i respectively. It is
easily veriﬁed that T
0,1
p
(C
n
) = T
1,0
p
(C
n
) and T
1,0
p
(C
n
) ∩ T
0,1
p
(C
n
) = ¦0¦, and that
T
1,0
p
(C
n
) is spanned by
_
∂
∂.
1
_
p
, ,
_
∂
∂.
n
_
p
,
where (∂,∂.
j
)
p
=
1
2
(∂,∂r
j
− i∂,∂j
j
)
p
for 1 ≤ , ≤ n. Any vector · ∈ T
1,0
p
(C
n
) is
called a vector of type (1, 0), and we call · ∈ T
0,1
p
(C
n
) a vector of type (0, 1). The
space T
1,0
p
(C
n
) is called the holomorphic tangent space at j.
Let CT
∗
p
(C
n
) be the dual space of CT
p
(C
n
). By duality, J also induces a splitting
on
CT
∗
p
(C
n
) = Λ
1,0
p
(C
n
) ⊕Λ
0,1
p
(C
n
),
where Λ
1,0
p
(C
n
) and Λ
0,1
p
(C
n
) are eigenspaces corresponding to the eigenvalues i and
−i respectively. It is easy to see that the vectors (d.
1
)
p
, , (d.
n
)
p
span Λ
1,0
p
(C
n
)
and the space Λ
0,1
p
(C
n
) is spanned by (d.
1
)
p
, ,(d.
n
)
p
.
6 Real and Complex Manifolds
Let ` be a complex manifold of complex dimension n and j be a point of `.
Let (.
1
, , .
n
) be a local coordinate system near j with .
j
= r
j
+ij
j
, , = 1, , n.
Then the real tangent space T
p
(`) is spanned by
(∂,∂r
1
)
p
, (∂,∂j
1
)
p
, , (∂,∂r
n
)
p
, (∂,∂j
n
)
p
.
Deﬁne as before an 1linear map J from T
p
(`) onto itself by
J
_
∂
∂r
j
_
p
=
_
∂
∂j
j
_
p
, J
_
∂
∂j
j
_
p
= −
_
∂
∂r
j
_
p
,
for 1 ≤ , ≤ n. We observe that the deﬁnition of J is independent of the choice
of the local coordinates (.
1
, , .
n
) and that J
2
= −1. Therefore, an argument
similar to the one given above shows that
CT
p
(`) = T
p
(`) ⊗
R
C = T
1,0
p
(`) ⊕T
0,1
p
(`),
and
CT
∗
p
(`) = Λ
1,0
p
(`) ⊕Λ
0,1
p
(`).
Next we introduce a Hermitian metric on `. By that, we mean at each point j ∈
`, a Hermitian inner product /
p
(n, ·) is deﬁned for n, · ∈ T
1,0
p
(`). If (.
1
, , .
n
)
is a local coordinate system on a neighborhood l of j, then
/
ij
(j) = /
p
_
∂
∂.
i
,
∂
∂.
j
_
is a complexvalued function on l, and (/
ij
(j))
n
i,j=1
is a positive deﬁnite Hermitian
matrix deﬁned for each point j of l. We shall assume the metric is smooth; namely,
that all the /
ij
’s vary smoothly on `. Then, we extend the metric to the whole
complexiﬁed tangent space in a natural way by requiring T
1,0
(`) to be orthogonal
to T
0,1
(`). If a complex manifold ` is equipped with a Hermitian metric /, we
shall call (`, /) a Hermitian manifold.
1.4 Vector Bundles
Let ` be a smooth manifold of real dimension n. The union of all the tan
gent spaces T
p
(`), j ∈ `, inherits a natural geometric structure called the vector
bundle.
Deﬁnition 1.4.1. Let 1 and ` be two smooth manifolds. 1 is called a vector
bundle over ` of rank / if there exists a smooth mapping π, called the projection
map, from 1 onto ` such that the following conditions are satisﬁed:
(1) For each j ∈ `, 1
p
= π
−1
(j) is a vector space over 1 of dimension /. 1
p
is called the ﬁbre space over j.
(2) For each j ∈ `, there exists an open neighborhood l containing j and a
diﬀeomorphism
/ : π
−1
(l) →l 1
k
,
such that /(π
−1
(¡)) = ¦¡¦1
k
and the restriction /
q
: π
−1
(¡) →¦¡¦1
k
≡
1
k
is a linear isomorphism, for every ¡ ∈ l.
1.4 Vector Bundles 7
The pair (l, /) is called a local trivialization.
When / = 1, 1 is also called a line bundle over `.
For a vector bundle π : 1 → `, the manifold 1 is called the total space and
` is called the base space, and 1 is called a vector bundle over `. Notice that
if two local trivializations (l
α
, /
α
) and (l
β
, /
β
) have nonempty intersection, i.e.,
l
α
∩l
β
,=∅, then a map p
αβ
is induced on l
α
∩l
β
:
p
αβ
: l
α
∩l
β
→ G1(/, 1),
such that
p
αβ
(j) = (/
α
)
p
◦ (/
β
)
−1
p
: 1
k
→1
k
.
The matrices p
αβ
’s are called transition matrices. Clearly, they are smooth and
satisfy the following conditions:
(1) p
αβ
= p
−1
βα
,
(2) p
αβ
◦ p
βγ
◦ p
γα
= 1,
where 1 is the identity matrix of rank /.
Let 1 be a vector bundle over `, and let l be an open subset of `. Any smooth
mapping : from l to 1 such that π ◦ : = id
U
, where id
U
is the identity mapping
on l, will be called a section over l. The space of all the sections over l will be
denoted by Γ(l, 1).
Notice that the concept of vector bundle can obviously be deﬁned for other cate
gories. For instance, if 1 and ` are complex manifolds and the ﬁbers are complex
vector spaces, then one can deﬁne a holomorphic vector bundle 1 over ` by re
quiring the morphisms and the transition matrices to be holomorphic mappings.
Here are some typical examples of vector bundles.
Example 1.4.2 (Tangent Bundles). Let ` be a manifold of real dimension n.
The set formed by the disjoint union of all tangent spaces T
p
(`) of j ∈ `, namely,
T(`) = ∪
p∈M
T
p
(`),
has a natural vector bundle structure of rank n over `. The local coordinate
neighborhoods of T(`) and the local trivializations of the bundle are given by the
local coordinate systems of ` as follows: let (r
1
, , r
n
) be a local coordinate
system on l of `, and let j ∈ l. Then any tangent vector · at j can be written
as
· =
n
i=1
·
i
(r)
_
∂
∂r
i
_
p
.
Thus, we obtain a map φ from π
−1
(l) onto l 1
n
by
φ : π
−1
(l) →l 1
n
,
(j, ·) →(j, ·
1
(r), , ·
n
(r)).
If two local coordinate systems have nontrivial intersection, then the transition
matrix is clearly deﬁned by the Jacobian matrix, with respect to these two local
8 Real and Complex Manifolds
coordinate systems, which by deﬁnition is smooth. It is also clear that any global
section : in Γ(`, T(`)) is a smooth vector ﬁeld A deﬁned on `.
Next, we can also form a new vector bundle from a given one. The most important
examples of such algebraically derived vector bundles are those originating from the
tangent bundle T(`). For instance, by considering the dual space and the exterior
algebra of the tangent space T
p
(`), we obtain the following new vector bundles:
Example 1.4.3 (Cotangent Bundle). Let ` be a smooth manifold of real
dimension n. The ﬁbre of the cotangent bundle, T
∗
(`), at each point j ∈ ` is
the 1linear dual space of T
p
(`), denoted by T
∗
p
(`). Clearly, T
∗
(`) is a vector
bundle of rank n over `. A section : of this bundle over an open set l of ` is
called a smooth 1form over l. We also have the complexiﬁed cotangent bundle,
denoted by Λ
1
(`) = T
∗
(`) ⊗
R
C, over `.
Example 1.4.4 (Exterior Algebra Bundles). Let ` be a complex manifold
of complex dimension n. Then the exterior algebra bundles over ` are the vector
bundles Λ
r
(`) whose ﬁbers at each point .
0
∈ ` are the wedge product of degree
: of Λ
1
p
(`), and
Λ(`) =
2n
⊕
r=0
Λ
r
(`).
Any smooth section : of Λ
r
(`) over an open subset l of ` is a smooth :form
on l. If, at each point .
0
of `, we take the wedge product of j copies of Λ
1,0
(`)
and ¡ copies of Λ
0,1
(`), where j ≤ n and ¡ ≤ n, we obtain the vector bundle of
bidegree (j, ¡), denoted by Λ
p,q
(`), and we have
Λ
r
(`) = ⊕
p+q=r
Λ
p,q
(`).
Smooth sections of Λ
p,q
(`), denoted by C
∞
(p,q)
(`), are called (j, ¡)forms on `.
1.5 Exterior Derivatives and the CauchyRiemann Complex
Let ` be a complex manifold of complex dimension n, and let (.
1
, , .
n
) be
a local coordinate system on an open neighborhood l of a point j of `, with
.
j
= r
j
+ij
j
for 1 ≤ , ≤ n. Let ) be a C
1
complexvalued function deﬁned on `.
Then, locally on l one can express d) as
(1.5.1)
d) =
n
j=1
∂)
∂r
j
dr
j
+
n
j=1
∂)
∂j
j
dj
j
=
n
j=1
∂)
∂.
j
d.
j
+
n
j=1
∂)
∂.
j
d.
j
,
where we have used the notation
∂)
∂.
j
=
1
2
_
∂)
∂r
j
−i
∂)
∂j
j
_
,
∂)
∂.
j
=
1
2
_
∂)
∂r
j
+i
∂)
∂j
j
_
,
1.5 Exterior Derivatives and the CauchyRiemann Complex 9
and
d.
j
= dr
j
+idj
j
, d.
j
= dr
j
−idj
j
,
for 1 ≤ , ≤ n. Deﬁne the operators ∂ and ∂ on functions by
∂) =
n
j=1
∂)
∂.
j
d.
j
, and ∂) =
n
j=1
∂)
∂.
j
d.
j
.
Then, (1.5.1) can be written as
d) = ∂) +∂).
This means that the diﬀerential d) of a C
1
function ) on l can be decomposed into
the sum of a (1, 0)form ∂) and (0, 1)form ∂).
It is easily veriﬁed that the deﬁnitions of ∂ and ∂ are invariant under holomorphic
change of coordinates. Hence, the operators ∂ and ∂ are well deﬁned for functions
on a complex manifold. A C
1
complexvalued function on a complex manifold is
holomorphic if and only if
∂) = 0.
Next we extend the deﬁnition of the operators ∂ and ∂ to diﬀerential forms of
arbitrary degree. Let ) be a (j, ¡)form on l. Write ) as
) =
]I]=p,]J]=q
)
IJ
d.
I
∧ d.
J
,
where 1 = (i
1
, , i
p
) and J = (,
1
, , ,
q
) are multiindices of length j and ¡
respectively, d.
I
= d.
i1
∧ ∧ d.
ip
, d.
J
= d.
j1
∧ ∧ d.
jq
.
The exterior derivative d) of ) is then deﬁned by
d) =
I,J
d)
I,J
∧ d.
I
∧ d.
J
= ∂) +∂),
where ∂) and ∂) are deﬁned by
∂) =
I,J
∂)
I,J
∧ d.
I
∧ d.
J
, ∂) =
I,J
∂)
I,J
∧ d.
I
∧ d.
J
,
which are of type (j + 1, ¡) and (j, ¡ + 1) respectively.
Since the transition matrices of a complex manifold ` are holomorphic, the
operators ∂ and ∂ are well deﬁned for (j, ¡)forms on `, and we have
d = ∂ +∂.
Since
0 = d
2
) = ∂
2
) + (∂∂ +∂∂)) +∂
2
)
10 Real and Complex Manifolds
and all terms are of diﬀerent types, we obtain
(1.5.2) ∂
2
= 0, ∂∂ +∂∂ = 0, ∂
2
= 0.
Since
(1.5.3) ∂
2
= 0,
It follows that the sequence
0 →Λ
p,0
(`)
∂
−→Λ
p,1
(`)
∂
−→
∂
−→Λ
p,n−1
(`)
∂
−→Λ
p,n
(`) →0,
for 0 ≤ j ≤ n, is a complex. This is called the CauchyRiemann complex. Denote
∂
p,q
= ∂ : Λ
p,q
(`) →Λ
p,q+1
(`). It follows that the image of ∂
p,q
lies in the kernel
of ∂
p,q+1
. To measure the exactness of the sequence, we have to solve the following
inhomogeneous equation
(1.5.4) ∂n = ),
under the compatibility condition
(1.5.5) ∂) = 0.
The solvability of the ∂equation as well as the smoothness of the solution is one of
the main issues throughout this book.
1.6 The Frobenius Theorem
Let l be an open neighborhood of the origin in 1
n
, and let / be an integer with
1 ≤ / < n. Then, the set ·
c
= ¦r = (r
1
, , r
n
) ∈ l[ r
k+1
= c
k+1
, , r
n
=
c
n
¦, where c = (c
k+1
, , c
n
) ∈ 1
n−k
is a constant vector, forms a / dimensional
submanifold of l. By a submanifold we mean that ·
c
is a closed subset of l and ·
c
forms a manifold itself. Notice that ·
c1
∩·
c2
= ∅ if c
1
,= c
2
. Also ∪
c∈R
n−k·
c
= l.
With such a submanifold structure, we shall say that l is foliated by / dimensional
submanifolds ·
c
, and call ·
c
a leaf of the foliation.
Let A
1
, , A
k
be / linearly independent vector ﬁelds on l such that they are
tangent to some ·
c
, c ∈ 1
n−k
, everywhere. Since the restriction of the vector ﬁeld
A
i
, 1 ≤ i ≤ / to each ·
c
deﬁnes a vector ﬁeld on ·
c
, it is easily seen that the
commutator [A
i
, A
j
] = A
i
A
j
− A
j
A
i
, 1 ≤ i, , ≤ /, is still a smooth vector ﬁeld
tangent to ·
c
everywhere on l. It follows that on l we have
(1.6.1) [A
i
, A
j
] =
k
l=1
o
ijl
(r)A
l
,
where o
ijl
(r) ∈ C
∞
(l).
In this section we shall show that condition (1.6.1) is also suﬃcient for a man
ifold to be foliated locally by submanifolds whose tangent vectors are spanned by
A
t
i
:. Since the result is purely local, we shall formulate the theorem in an open
neighborhood l of the origin in 1
n
.
1.6 The Frobenius Theorem 11
Theorem 1.6.1 (Frobenius). Let A
1
, , A
k
, 1 ≤ / < n, be smooth vector ﬁelds
deﬁned in an open neighborhood l of the origin in 1
n
. If
(1) A
1
(0), , A
k
(0) are linearly independent, and
(2) [A
i
, A
j
] =
k
l=1
o
ijl
(r)A
l
, 1 ≤ i, , ≤ /, for some o
ijl
(r) ∈ C
∞
(l),
then there exist new local coordinates (j
1
, , j
n
) in some open neighborhood \ of
the origin such that
A
i
=
k
j=1
/
ij
(j)
∂
∂j
j
, i = 1, , /,
where (/
ij
(j)) is an invertible matrix. In other words, \ is foliated by the /
dimensional submanifolds ¦j ∈ \ [ j
i
= c
i
, i = / + 1, , n¦.
Proof. The theorem will be proved by induction on the dimension n of the ambient
space. When n = 1, the assertion is obviously true. Let us assume that the assertion
is valid up to dimension n −1.
First, we may simplify the vector ﬁeld A
1
(r) = (o
1
(r), , o
n
(r)). From the
basic existence theorem for a system of ﬁrst order ordinary diﬀerential equation,
through every point j in a small open neighborhood of the origin, there exists
exactly one integral curve γ(t) = (γ
1
(t), , γ
n
(t)), where t ∈ (−δ, δ) for some real
number δ 0, such that
dγ
i
dt
(t) = o
i
(γ(t)), i = 1, , n.
It follows that for any smooth function ) in a small neighborhood \ of 0,
A
1
)(γ(t)) =
k
i=1
o
i
(γ(t))
∂)(γ(t))
∂r
i
=
∂)(γ(t))
∂t
.
Locally one can introduce new independent variables, also denoted by (r
1
, , r
n
),
which straighten out the integral curves so that A
1
= ∂,∂r
1
. Next, by subtracting
a multiple of A
1
from A
i
for 2 ≤ i ≤ /, we may also assume that
A
i
(r) =
n
l=2
c
il
(r)
∂
∂r
l
, i = 2, , /.
Denote r
t
= (r
2
, , r
n
). On the submanifold \ ∩ ¦r = (r
1
, r
t
) [ r
1
= 0¦, these
vector ﬁelds A
2
(0, r
t
), , A
k
(0, r
t
), satisfy both conditions (1) and (2) in an open
neighborhood of the origin in 1
n−1
. Hence, by the induction hypotheses, there exist
new local coordinates near the origin in 1
n−1
, denoted also by r
t
= (r
2
, , r
n
),
such that c
il
(0, r
t
) = 0 for 2 ≤ i ≤ / and  /. For 2 ≤ i ≤ / and 2 ≤  ≤ n, we
have
∂c
il
∂r
1
(r) = A
1
A
i
(r
l
) = [A
1
, A
i
](r
l
)
=
k
α=2
o
1iα
(r)A
α
(r
l
)
=
k
α=2
o
1iα
(r)c
αl
(r).
12 Real and Complex Manifolds
Hence, the uniqueness part of the Cauchy problem for a system of ﬁrst order ordi
nary diﬀerential equations implies that c
il
(r) ≡ 0 for  / in an open neighborhood
\ of the origin. This completes the proof of the theorem.
It should be pointed out that the existence of the local coordinates (j
1
, , j
n
)
guaranteed by the Frobenius theorem is not unique. Suppose that there are /
smooth vector ﬁelds A
1
, , A
k
deﬁned in some open neighborhood of the origin in
1
n
such that conditions (1) and (2) of Theorem 1.6.1 are satisﬁed. Let us consider
a system of overdetermined partial diﬀerential equations
(1.6.2) A
j
n = )
j
, , = 1, , /,
where the data )
j
’s are smooth functions given in an open neighborhood of the
origin. It is clear from condition (2) that the system (1.6.2) is solvable only if the
given data satisfy the following compatibility condition
(1.6.3) A
i
)
j
−A
j
)
i
=
k
l=1
o
ijl
(r))
l
(r), 1 ≤ i, , ≤ /.
With the aid of Theorem 1.6.1, the next theorem shows that condition (1.6.3) is, in
fact, also suﬃcient for the solvability of (1.6.2).
Theorem 1.6.2. Under the same hypotheses as in Theorem 1.6.1, let )
1
, , )
k
be
smooth functions deﬁned on l. Then, the system (1.6.2) has a smooth solution n in
an open neighborhood of the origin if and only if the compatibility conditions (1.6.3)
are satisﬁed. Furthermore, if H is a closed submanifold of l through the origin of
dimension n−/ such that the tangent plane of H at the origin is complementary to
the space spanned by A
1
(0), , A
k
(0) , then given any smooth function n
h
on H,
there exists an unique solution n to the equations (1.6.2) in an open neighborhood
of the origin with n[
H
= n
h
.
Notice that in the language of partial diﬀerential equations, the hypothesis on
H is equivalent to stating that the manifold H is noncharacteristic with respect to
A
1
, , A
k
, or that, geometrically, H is transversal to the leaves of the foliation
deﬁned by the vector ﬁelds A
1
, , A
k
in some open neighborhood of the origin.
Proof. Notice ﬁrst that conditions (1) and (2) of Theorem 1.6.1 and equations
(1.6.2),(1.6.3) are invariant if we change variables or make linear combinations of the
equations. Hence, by Theorem 1.6.1, we may thus assume that A
j
(r) = ∂,∂r
j
, , =
1, , /. It follows that o
ijl
(r) ≡ 0 for all 1 ≤ i, ,,  ≤ /, and equation (1.6.3) is
reduced to
(1.6.4)
∂)
j
∂r
i
−
∂)
i
∂r
j
= 0, 1 ≤ i, , ≤ /.
Let us write the local coordinates r = (r
t
, r
tt
) with r
t
= (r
1
, , r
k
) and r
tt
=
(r
k+1
, , r
n
). Then, locally in some open neighborhood of the origin one may
express H as a graph over an open subset \ containing the origin in 1
n−k
, namely,
1.7 Inequivalence between the Ball and the Polydisc 13
H is deﬁned by r
t
= /(r
tt
) where /(r
tt
) is a smooth function on \ . Now, for each
ﬁxed r
tt
∈ \ ,
k
j=1
)
j
(r
t
, r
tt
)dr
j
is a diﬀerential of r
t
which in turn by (1.6.4) is closed. Hence, the line integral
(1.6.5) n(r) =
_
x
h(x
)
k
j=1
)
j
(r
t
, r
tt
)dr
j
+n
h
(/(r
tt
), r
tt
),
is welldeﬁned, i.e., independent of the paths in r
t
space from/(r
tt
) to r
t
. Obviously,
(1.6.5) deﬁnes the unique solution n which is equal to the initial datum n
h
on H to
the equations (1.6.2). This completes the proof of the theorem.
Now we turn to the complex analog of the Frobenius theorem. Let 1
1
, , 1
k
,
1 ≤ / < n, be type (1, 0) vector ﬁelds deﬁned in some open neighborhood l of
the origin in C
n
such that 1
1
, , 1
k
are linearly independent over C on l. If
there exist local holomorphic coordinates (.
1
, , .
n
) on l such that 1
1
, , 1
k
are tangent to the / dimensional complex submanifolds ·
c
= ¦(.
1
, , .
n
) ∈
C
n
[.
k+1
= c
1
, , .
n
= c
n−k
¦ for c = (c
1
, , c
n−k
) ∈ C
n−k
, then we see imme
diately that the subbundle E spanned by 1
1
, , 1
k
is closed under the Lie bracket
operation, and so is the subbundle E ⊕E.
Conversely, if both the subbundles E and E ⊕ E are closed respectively under
the Lie bracket operation, then locally on l one may introduce new holomorphic
coordinates (n
1
, , n
n
) so that l is foliated by the complex submanifolds ·
c
=
¦(n
1
, , n
n
) ∈ C
n
[ n
k+1
= c
1
, , n
n
= c
n−k
¦ for c = (c
1
, , c
n−k
) ∈ C
n−k
,
and E = T
1,0
(·
c
). This is the socalled complex Frobenius theorem which can be
deduced from the NewlanderNirenberg theorem proved in Chapter 5. When / = 1,
this will be proved in Chapter 2.
1.7 Inequivalence between the Ball and the Polydisc in C
n
In one complex variable the Riemann mapping theorem states that any simply
connected region not equal to the whole complex plane is biholomorphically equiv
alent to the unit disc.
However, the situation is completely diﬀerent in higher dimensional spaces. Let
´ = ¦. ∈ C[ [.[ < 1¦ and 1
n
= ¦(.
1
, , .
n
) ∈ C
n
[ [.
1
[
2
+ + [.
n
[
2
< 1¦. The
following theorem shows that an analog of the Riemann mapping theorem in several
variables is impossible.
Theorem 1.7.1 (Poincar´e). There exists no biholomorphic map
) : ´
n
→1
n
, for n ≥ 2,
where ´
n
is the Cartesian product of n copies of ´ in C
n
.
Proof. We shall assume that n = 2. The proof is the same for n 2. Suppose
that ) = ()
1
, )
2
) : ´
2
→ 1
2
is a biholomorphism. Let (., n) be the coordinates
14 Real and Complex Manifolds
in C
2
. For any point sequence ¦.
j
¦ in ´ with [.
j
[ → 1 as , → ∞, the sequence
p
j
(n) = )(.
j
, n) : ´ → 1
2
is uniformly bounded. Hence, by Montel’s theorem,
there is a subsequence, still denoted by p
j
(n), that converges uniformly on compact
subsets of ´ to a holomorphic map p(n) = (p
1
(n), p
2
(n)) : ´ → 1
2
. Since ) is a
biholomorphism, we must have [p(n)[
2
= 1 for all n ∈ ´. Hence, [p
t
(n)[ = 0 for all
n ∈ ´ which implies p
t
(n) ≡ 0 on ´. It follows that
(1.7.1) lim
j→∞
)
w
(.
j
, n) = p
t
(n) ≡ 0.
Equation (1.7.1) implies that for each ﬁxed n ∈ ´, )
w
(., n), when viewed as a
function of . alone, is continuous up to the boundary with boundary value identically
equal to zero. Therefore, by the maximum modulus principle we get
)
w
(., n) ≡ 0, for all (., n) ∈ ´
2
.
This implies ) is independent of n, a contradiction to the fact that ) is a biholo
morphic map. This completes the proof of the theorem.
Thus, according to Theorem 1.7.1, the classiﬁcation problem in several variables
is considerably more complicated than in one variable. An approach towards the
classiﬁcation of certain domains in C
n
, n ≥ 2, will be discussed in Section 6.3.
NOTES
For a general background on complex manifolds, the reader may consult books
by S. S. Chern [Cher 2], J. Morrow and K. Kodaira [MoKo 1] and R. O. Wells [Wel
1]. For a proof of the complex Frobenius theorem, the reader is referred to [Nir 1].
See also [H¨or 5]. The inequivalence between the polydisc and the unit ball was ﬁrst
discovered by H. Poincar´e by counting the dimensions of the automorphism groups
of both domains. The proof of Theorem 1.7.1 that we present here is based on the
ideas of R. Remmert and K. Stein [ReSt 1]. See also [Nar 1] and [Ran 6].
15
CHAPTER 2
THE CAUCHY INTEGRAL FORMULA
AND ITS APPLICATIONS
The main task of this chapter is to study the solvability and regularity of the
CauchyRiemann operator on the complex plane. We will ﬁrst show that the so
lution to the CauchyRiemann operator can be obtained via the Cauchy integral
formula. Then we shall prove the Plemelj jump formula associated with the Cauchy
transform. As an application of the Cauchy integral formula, given a (j, ¡)form )
on a polydisc satisfying the compatibility condition ∂) = 0, we will solve the inho
mogeneous ∂equation, ∂n = ), on a relatively smaller polydisc in several complex
variables.
Next we shall present the BochnerMartinelli formula which can be viewed as a
generalization of the Cauchy integral formula in several variables. Then, in a similar
manner, we will prove the jump formula associated with the BochnerMartinelli
transform.
In Section 2.3, we will determine when a ﬁrstorder partial diﬀerential equation
in two real variables is locally equivalent to the CauchyRiemann equation.
2.1 The Cauchy Integral Formula
All functions in this chapter are complexvalued unless otherwise stated. Then
the following formula, known as Cauchy’s integral formula, holds:
Theorem 2.1.1. Let 1 be a bounded open set in C with C
1
boundary /1. If
n ∈ C
1
(1), we have
(2.1.1) n(.) =
1
2πi
__
bD
n(ζ)
ζ −.
dζ +
__
D
∂u
∂ζ
ζ −.
dζ ∧ dζ
_
for any . ∈ 1.
Proof. The proof is an easy consequence of Stokes’ theorem. Let c be any small
positive number less than the distance from . to the boundary of 1. Denote by
1
(.) the open disc centered at . with radius c. Applying Stokes’ theorem to the
form n(ζ)dζ,(ζ −.) on the punctured domain 1
= 1 ¸ 1
(.), we obtain
_
bD
n(ζ)
ζ −.
dζ −i
_
2π
0
n(. +cc
iθ
) dθ =
__
D
∂u
∂ζ
ζ −.
dζ ∧ dζ.
Letting c →0, we have (2.1.1).
Next we show how to apply the Cauchy integral formula to solve the Cauchy
Riemann equation.
16 The Cauchy Integral Formula and its Applications
Theorem 2.1.2. Let 1 be a bounded domain in C, and let ) ∈ C
k
(1) for / ≥ 1.
Deﬁne
(2.1.2) n(.) =
1
2πi
__
D
)(ζ)
ζ −.
dζ ∧ dζ.
Then n(.) is in C
k
(1) and satisﬁes
(2.1.3)
∂n
∂.
= )(.)
on 1. When / = 0, n deﬁned by (2.1.2) is in C(1) and satisﬁes (2.1.3) in the
distribution sense.
Proof. For the case / ≥ 1, we ﬁrst assume ) ∈ C
k
0
(C). Setting −η = ζ −., we have
n(.) =
−1
2πi
__
C
)(. −η)
η
dη ∧ dη.
Diﬀerentiation under the integral sign gives that n ∈ C
k
(C). Using Theorem 2.1.1
we obtain
∂n
∂.
(.) =
1
2πi
__ ∂f
∂ζ
ζ −.
dζ ∧ dζ = )(.).
For the general situation, let .
0
∈ 1, and let χ be a cutoﬀ function, 0 ≤ χ ≤ 1,
χ ≡ 1 in some neighborhood \ of .
0
and suppχ ⊂ 1. Thus,
n(.) =
1
2πi
__
D
)(ζ)
ζ −.
dζ ∧ dζ
=
1
2πi
__
D
χ(ζ))(ζ)
ζ −.
dζ ∧ dζ +
1
2πi
__
D
(1 −χ(ζ)))(ζ)
ζ −.
dζ ∧ dζ
≡ n
1
(.) +n
2
(.).
It is easy to see that n
2
(.) is holomorphic in \ . Hence, from the previous argument
for 1 = C, we obtain
∂n
∂.
=
∂n
1
∂.
+
∂n
2
∂.
= χ(.))(.) = )(.),
for . ∈ \ .
To prove the case for / = 0, we observe that 1,(ζ − .) is an integrable kernel
after changing to polar coordinates. The following estimate holds for n deﬁned by
(2.1.2):
 n 
∞
≤ C  ) 
∞
.
Approximate ) by )
n
∈ C
1
(1) in the sup norm on 1. Deﬁne n
n
by (2.1.2) with
respect to )
n
. Then n
n
converges to n uniformly on 1. This shows that n ∈ C(1).
Also, in the distribution sense, we have ∂n,∂. = ), by letting n pass to inﬁnity.
This proves the theorem.
2.1 The Cauchy Integral Formula 17
We recall that a function ) deﬁned in some domain 1 contained in 1
n
is said to
be H¨older continuous of order λ, 0 < λ < 1, denoted by ) ∈ C
λ
(1), if for any two
distinct points r
1
and r
2
in 1, we have
[)(r
1
) −)(r
2
)[ ≤ 1[r
1
−r
2
[
λ
,
where the constant 1 is independent of r
1
and r
2
. When λ is equal to 1, ) is called
Lipschitz. The space of Lipschitz continuous function on 1 is denoted by Λ
1
(1).
Notice that C
1
(1) ⊂ Λ
1
(1) ⊂ C
λ
(1) ⊂ C(1). A function ) is said to be H¨older
continuous of order / +λ with / ∈ N and 0 < λ < 1, if all the partial derivatives of
) of order / are H¨older continuous of order λ.
For any continuous function ) on the boundary, the Cauchy transform of ), i.e.,
(2.1.4) 1(.) =
1
2πi
_
bD
)(ζ)
ζ −.
dζ for . ∈ C ¸ /1,
deﬁnes a holomorphic function 1(.) oﬀ the boundary. The Cauchy transform 1(.)
and the given data ) on the boundary are related by the socalled Plemelj jump
formula as shown in the following theorem.
Theorem 2.1.3 (Jump formula). Let 1 be a bounded domain in C with C
k+1
boundary, / ∈ N, such that C ¸ 1 is connected, and let ) be a C
k
function deﬁned
on the boundary. Deﬁne 1(.) as in (2.1.4), and set 1
−
(.) = 1(.) for . ∈ 1 and
1
+
(.) = 1(.) for . , ∈ 1. Then, for any given 0 < c < 1, 1
−
(.) ∈ C
k−
(1) ∩O(1)
and 1
+
(.) ∈ C
k−
(C ¸ 1) ∩ O(C ¸ 1) and
(2.1.5) )(.) = 1
−
(.) −1
+
(.) for . ∈ /1.
In particular, if 1 has C
∞
boundary and ) is smooth on /1, then both 1
−
(.) and
1
+
(.) are smooth up to the boundary.
Proof. First we prove the identity (2.1.5). Let )
e
(.) be any C
k
extension of ) to
the whole complex plane. Then, we have, for . ∈ 1,
(2.1.6) 1
−
(.) −)
e
(.) =
1
2πi
_
bD
)(ζ) −)
e
(.)
ζ −.
dζ,
and, for . ∈ C ¸ 1,
(2.1.7) 1
+
(.) =
1
2πi
_
bD
)(ζ) −)
e
(.)
ζ −.
dζ.
Since ) is, at least, of class C
1
on the boundary, the integral on the righthand sides
of (2.1.6) and (2.1.7) deﬁnes a continuous function on the whole complex plane.
Therefore, letting . approach the same point on the boundary from either sides, we
obtain
)(.) = 1
−
(.) −1
+
(.) for . ∈ /1.
This proves (2.1.5).
18 The Cauchy Integral Formula and its Applications
For the regularity of 1
−
(.) and 1
+
(.) near the boundary we shall invoke the
HardyLittlewood lemma (see Theorem C.1 in the Appendix). It is clear that,
without loss of generality, we may assume that ) is compactly supported in a
boundary coordinate chart, with ∂:,∂. ,= 0 on this coordinate chart, where : is a
C
k+1
deﬁning function for 1. When / = 1, using (2.1.6), we have
(2.1.8)
[d(1
−
(.) −)
e
(.))[
_
bD
[d)
e
(.)[
[. −ζ[
d:(ζ) +
_
bD
[)(ζ) −)
e
(.)[
[. −ζ[
2
d:(ζ)
_
bD
1
[. −ζ[
d:(ζ).
Here, ¹ 1 means there is an universal constant C, independent of ¹ and 1,
such that ¹ ≤ C1. For any given c 0, to show 1
−
(.) ∈ C
1−
(1), it suﬃces
to estimate (2.1.8) over a small neighborhood l of π(.) on the boundary, where
π(.) is the projection of . on the boundary. Let d(.) be the distance from . to the
boundary. If . is suﬃciently close to the boundary, it is easily seen that, for ζ ∈ l,
[. −ζ[ is equivalent to d(.) +:(ζ), where :(ζ) is the distance from ζ to π(.) along
the boundary. It follows that
_
U
d:(ζ)
[. −ζ[
≤ d(.)
−
_
U
d:(ζ)
[. −ζ[
1−
d(.)
−
_
1
0
d:(ζ)
(d(.) +:(ζ))
1−
d(.)
−
.
This proves that 1
−
(.) ∈ C
1−
(1). Similarly, we have 1
+
(.) ∈ C
1−
(C ¸ 1).
For / 1, observe that
(2.1.9) T
z
=
∂
∂.
−
∂:
∂.
(
∂:
∂.
)
−1
∂
∂.
satisﬁes T
z
(:) = 0. Hence, T
z
is a tangential vector ﬁeld with C
k
coeﬃcients along
the level sets of :. Then, integration by parts shows, for . ∈ 1,
∂
∂.
1
−
(.) =
1
2πi
_
bD
)(ζ)
∂
∂.
_
1
ζ −.
_
dζ
= −
1
2πi
_
bD
)(ζ)
∂
∂ζ
_
1
ζ −.
_
dζ
= −
1
2πi
_
bD
)(ζ)T
ζ
_
1
ζ −.
_
dζ
=
1
2πi
_
bD
T
∗
ζ
)(ζ)
_
1
ζ −.
_
dζ,
where T
∗
ζ
is a ﬁrst order diﬀerential operator with C
k−1
coeﬃcients on the boundary.
It follows that
(2.1.10)
∂
k
∂.
k
1
−
(.) =
1
2πi
_
bD
(T
∗
ζ
)
k
)(ζ)
_
1
ζ −.
_
dζ.
2.1 The Cauchy Integral Formula 19
Since ) is of class C
k
on the boundary, a similar argument shows that, for any small
c 0, we have
¸
¸
¸
¸
∂
k
∂.
k
1
−
(.)
¸
¸
¸
¸
d(.)
−
.
This proves 1
−
(.) ∈ C
k−
(1) from the HardyLittlewood lemma. Similarly, we
have 1
+
(.) ∈ C
k−
(C ¸ 1). The proof of the theorem is now complete.
Corollary 2.1.4. Under the same hypotheses as in Theorem 2.1.3, ) is the restric
tion of a holomorphic function 1 ∈ C
k−
(1) ∩ O(1) if and only if ) is orthogonal
to ¦.
m
¦
∞
m=0
on the boundary, namely,
(2.1.11)
_
bD
)(.).
m
d. = 0 for : ∈ ¦0¦ ∪ N.
Proof. Assume that ) is orthogonal to ¦.
m
¦
∞
m=0
on the boundary. If . satisﬁes
[.[ [ζ[ for all ζ ∈ /1, we have
1
+
(.) =
1
2πi
_
bD
)(ζ)
ζ −.
dζ
=
−1
2πi
∞
m=0
__
bD
)(ζ)ζ
m
dζ
_
.
−m−1
= 0.
Since 1
+
(.) is holomorphic on C ¸ 1 from Theorem 2.1.3, the identity theorem
shows that 1
+
(.) ≡ 0 for all . ∈ C¸ 1. It is now clear from the jump formula that
1
−
(.) is a C
k−
holomorphic extension of ) to 1.
Conversely, if ) is the restriction of a function in C
k−
(1) ∩O(1), we must have
1
+
(.) ≡ 0 on C ¸ 1. Thus, by reversing the above arguments for . outside a large
disc centered at the origin, we have
∞
m=0
__
bD
)(ζ)ζ
m
dζ
_
.
−m−1
= 0.
This implies
_
bD
)(ζ)ζ
m
dζ = 0 for : ∈ ¦0¦ ∪ N,
and hence proves the corollary.
Combining with Theorem 2.1.2, the arguments for proving Corollary 2.1.4 can be
applied, almost verbatim, to obtain necessary and suﬃcient conditions for solving
the ∂equation with compactly supported solution, via the Cauchy integral formula
in C.
Corollary 2.1.5. Let the domain 1 be as in Theorem 2.1.3, and let ) ∈ C
k
(1),
/ ≥ 1. Deﬁne n(.) by (2.1.2). Then n(.) satisﬁes ∂n,∂. = ) in C and is supported
in 1 if and only if
__
D
)(ζ)ζ
m
dζ ∧ dζ = 0 for : ∈ ¦0¦ ∪ N.
20 The Cauchy Integral Formula and its Applications
As an application, we will apply the Cauchy integral formula to solve the ∂
equation on a polydisc in C
n
, n ≥ 2. By a polydisc 1(ζ; :) centered at ζ =
(ζ
1
, , ζ
n
) with multiradii : = (:
1
, , :
n
) in C
n
, we mean 1(ζ; :) =
n
j=1
1
rj
(ζ
j
)
where 1
rj
(ζ
j
) = ¦. ∈ C[ [. −ζ
j
[ < :
j
¦. Let 1
t
(ζ; :
t
) =
n
j=1
1
r
j
(ζ
j
) be another
polydisc with :
t
j
< :
j
for 1 ≤ , ≤ n. Then, we have the following result:
Theorem 2.1.6. Let 1 and 1
t
be deﬁned as above with ζ = 0, and let ) be a
smooth (j, ¡ +1)form, j ≥ 0, ¡ ≥ 0, deﬁned on 1, which satisﬁes the compatibility
condition ∂) = 0. Then there exists a smooth (j, ¡)form n on 1
t
such that ∂n = ).
Note that we have solved the ∂equation on any slightly smaller subdomain 1
t
.
In fact, it will be clear later that the ∂equation can be solved on the whole polydisc.
Proof. Write ) as
) =
t
]I]=p,]J]=q+1
)
IJ
d.
I
∧ d.
J
,
where the prime means that we sum over only increasing multiindices. We shall
inductively prove the following statement:
o
k
: The assertion holds if ) involves only (0, 1)forms from the set
¦d.
1
, , d.
k−1
and d.
k
¦.
When / = n, o
n
gives the desired result.
o
k
obviously holds when 0 ≤ / ≤ ¡, since ) is of type (j, ¡ + 1). Hence, we
assume the statement is valid up to o
k−1
for some / with / −1 ≥ ¡, and we proceed
to prove the statement o
k
. Write
) = d.
k
∧ β +α,
where β is a (j, ¡)form and α is a (j, ¡ + 1)form, and both α and β involve only
(0, 1)forms from d.
1
, , d.
k−1
. Express
β =
t
]I]=p,]J]=q
β
IJ
d.
I
∧ d.
J
.
It is easy to see by type consideration that ∂β
IJ
,∂.
j
= 0 for , / and all 1, J.
Now choose a cutoﬀ function χ(.
k
) ∈ C
∞
0
(1
r
k
) such that 0 ≤ χ ≤ 1 and χ ≡ 1
in some open neighborhood of 1
r
k
. Then Theorem 2.1.2 shows that, for each 1, J,
the function
1
IJ
(.) =
1
2πi
__
C
χ(ζ
k
)β
IJ
(.
1
, , .
k−1
, ζ
k
, .
k+1
, , .
n
)
ζ
k
−.
k
dζ
k
∧ dζ
k
is smooth and solves the ∂equation
∂1
IJ
∂.
k
= χ(.
k
)β
IJ
(.) = β
IJ
(.)
2.1 The Cauchy Integral Formula 21
on some neighborhood of 1
t
. We also have, for , / and all 1, J,
∂1
IJ
∂.
j
(.) = 0.
Put
1 =
t
]I]=p,]J]=q
1
IJ
d.
I
∧ d.
J
,
then
∂1 =
t
]I]=p,]J]=q
_
n
j=1
∂1
IJ
∂.
j
d.
j
_
∧ d.
I
∧ d.
J
= d.
k
∧ β +α
0
,
where α
0
is a (j, ¡ + 1)form that involves only (0, 1)forms from d.
1
, , d.
k−1
.
Hence,
) −∂1 = α −α
0
is a smooth (j, ¡ + 1)form which is ∂closed and involves only (0, 1)forms from
d.
1
, , d.
k−1
. It follows now from the induction hypotheses that there exists a
smooth (j, ¡)form n
0
that satisﬁes
∂n
0
= α −α
0
= ) −∂1.
Clearly, n = n
0
+1 is a solution of ∂n = ), and the proof is complete.
Theorem 2.1.7 (Cauchy integral formula for polydiscs). Let 1(η; :) be a
polydisc in C
n
, n ≥ 2. Suppose that ) is continuous on 1(η; :) and holomorphic in
1(η; :). Then for any . ∈ 1(η; :),
)(.) =
1
(2πi)
n
_
]ζn−ηn]=rn
_
]ζ1−η1]=r1
)(ζ
1
, , ζ
n
)
(ζ
1
−.
1
) (ζ
n
−.
n
)
dζ
1
dζ
n
.
Proof. It is easily seen that the integral representation of )(.) is obtained by re
peated application of the Cauchy integral formula in one variable. This proves the
theorem.
Here are some easy consequences of Theorem 2.1.7:
Theorem 2.1.8 (Cauchy estimates). Under the same hypotheses as in Theorem
2.1.7. Suppose that [)[ ≤ ` for all . ∈ 1(η; :). Then
¸
¸
¸
¸
_
∂
∂.
_
α
)(η)
¸
¸
¸
¸
≤
`α!
:
α1
1
:
n
αn
,
where α = (α
1
, , α
n
) is a multiindex with α
j
∈ ¦0¦ ∪ N, α! = α
1
! α
n
! and
_
∂
∂z
_
α
=
_
∂
∂z1
_
α1
_
∂
∂zn
_
αn
.
22 The Cauchy Integral Formula and its Applications
Theorem 2.1.9. If ) is holomorphic in 1 ⊂ C
n
, then locally near any point n in
1, ) has a power series representation. In particular, ) is real analytic.
By a power series representation for ) near n, we mean
)(.) =
α
o
α
(. −n)
α
such that the series converges absolutely in some open neighborhood of n. Here
the summation is over multiindices α and (. − n)
α
= (.
1
− n
1
)
α1
(.
n
− n
n
)
αn
.
It follows now from the power series expansion of holomorphic functions, we have
Theorem 2.1.10 (Identity Theorem). Let ) and p be two holomorphic functions
deﬁned on a connected open set 1 ⊂ C
n
. If ) and p coincide on an open subset of
1, then ) = p on 1.
2.2 The BochnerMartinelli Formula
In this section, we shall extend the Cauchy kernel from the complex plane to
higher dimensional space. Deﬁne the BochnerMartinelli kernel by
(2.2.1) 1(ζ, .) =
(n −1)!
(2πi)
n
1
[ζ −.[
2n
n
j=1
(ζ
j
−.
j
)dζ
j
∧ ( ∧
k,=j
dζ
k
∧ dζ
k
),
for ζ = (ζ
1
, , ζ
n
), . = (.
1
, , .
n
) ∈ C
n
and ζ ,= .. 1(ζ, .) is a form of type
(n, n −1) in ζ. It is clear that when n = 1,
1(ζ, .) =
1
2πi
1
ζ −.
dζ
which is the Cauchy kernel in C.
The following theorem is a generalized version of the Cauchy integral formula in
several variables.
Theorem 2.2.1. Let 1 be a bounded domain with C
1
boundary in C
n
, n ≥ 2, and
let ) ∈ C
1
(1). Then
(2.2.2) )(.) =
_
bD
)(ζ)1(ζ, .) −
_
D
∂) ∧ 1(ζ, .) for . ∈ 1,
and
(2.2.3) 0 =
_
bD
)(ζ)1(ζ, .) −
_
D
∂) ∧ 1(ζ, .) for . , ∈ 1.
Proof. A direct calculation shows that ∂
ζ
1(ζ, .) = 0 for ζ ,= .. Since 1(ζ, .) is of
type (n, n −1) in ζ, by Stokes’ theorem we have, for . ∈ 1,
_
bD
)(ζ)1(ζ, .) =
_
D(z)
d()(ζ)1(ζ, .)) +
_
bB(z)
)(ζ)1(ζ, .)
=
_
D(z)
∂)(ζ) ∧ 1(ζ, .) +
_
bB(z)
)(ζ)1(ζ, .),
2.2 The BochnerMartinelli Formula 23
where 1
(.) = ¦ζ ∈ C
n
[ [ζ −.[ < c¦ for small c 0 and 1
(.) = 1 ¸ 1
(.). Using
homogeneity of the kernel and Stokes’ theorem, we easily get
_
bB(z)
1(ζ, .) =
(n −1)!
(2πi)
n
n
j=1
_
bB(0)
ζ
j
[ζ[
2n
dζ
j
∧ ( ∧
k,=j
dζ
k
∧ dζ
k
)
=
(n −1)!
(2πi)
n
1
c
2n
n
j=1
_
B(0)
dζ
j
∧ dζ
j
∧ ( ∧
k,=j
dζ
k
∧ dζ
k
)
= 1,
for all c 0. Now, letting c →0, we obtain
)(.) =
_
bD
)(ζ)1(ζ, .) −
_
D
∂)(ζ) ∧ 1(ζ, .).
This proves (2.2.2).
Now for the proof of (2.2.3), since . , ∈ 1, the kernel is regular on 1. Hence, an
application of Stokes’ theorem gives
_
bD
)(ζ)1(ζ, .) =
_
D
∂)(ζ) ∧ 1(ζ, .).
This proves (2.2.3) and hence the theorem.
An immediate consequence of Theorem 2.2.1 is the following reproducing prop
erty of the BochnerMartinelli kernel for holomorphic functions:
Corollary 2.2.2. Let 1 be a bounded domain with C
1
boundary in C
n
, n ≥ 2. For
any ) ∈ O(1) ∩ C(1), we have
(2.2.4) )(.) =
_
bD
)(ζ)1(ζ, .) for . ∈ 1.
The integral (2.2.4) is zero if . , ∈ 1.
Proof. First we assume that ) ∈ C
1
(1). Then the assertion follows immediately
from Theorem 2.2.1. The general case now follows from approximation. This proves
the corollary.
Thus, the BochnerMartinelli kernel also enjoys the reproducing property for
holomorphic functions, although 1(ζ, .) is no longer holomorphic in ..
A more systematic treatment of kernels in several variables will be given in Chap
ter 11 where a reproducing kernel holomorphic in . variables will be constructed
for convex domains.
Theorem 2.2.3 (Jump formula). Let 1 be a bounded domain in C
n
, n ≥ 2, with
connected C
1
boundary, and let ) be a C
1
function deﬁned on the boundary. Deﬁne
1(.) =
_
bD
)(ζ)1(ζ, .) for . ∈ C
n
¸ /1,
24 The Cauchy Integral Formula and its Applications
and let 1
−
(.) = 1(.) for . ∈ 1 and 1
+
(.) = 1(.) for . , ∈ 1. Then, for any small
c 0, we have 1
−
(.) ∈ C
1−
(1), 1
+
(.) ∈ C
1−
(C
n
¸ 1) and
(2.2.5) )(.) = 1
−
(.) −1
+
(.) for . ∈ /1.
Equation (2.2.5) is the socalled jump formula associated with the Bochner
Martinelli transform. When n = 1, this is the Plemelj jump formula proved in
Theorem 2.1.3 where 1
−
and 1
+
are also holomorphic.
Proof. Let )
e
(.) be any C
1
extension of ) to the whole space. Then, for any . ∈ 1,
we have
(2.2.6) 1
−
(.) −)
e
(.) =
_
bD
()(ζ) −)
e
(.))1(ζ, .).
Since
[1(ζ, .)[ [. −ζ[
1−2n
from the deﬁnition of BochnerMartinelli kernel, the righthand side of (2.2.6) de
ﬁnes a continuous function on the whole space. Thus, we have 1
−
(.) ∈ C(1). For
. , ∈ 1, we get
(2.2.7) 1
+
(.) =
_
bD
()(ζ) −)
e
(.))1(ζ, .).
Letting . tend to the same point on the boundary from either side, we obtain
)(.) = 1
−
(.) −1
+
(.) for . ∈ /1.
This proves (2.2.5).
For the regularity of 1
−
(.) and 1
+
(.) we again use the HardyLittlewood lemma
(see Theorem C.1 in the Appendix). Thus, we need to estimate the diﬀerential of
1
−
(.). An easy exercise shows that
[d(1
−
(.) −)
e
(.))[
≤
_
bD
[d)
e
(.)[[1(ζ, .)[ +
_
bD
[)(ζ) −)
e
(.)[[d1(ζ, .)[
_
bD
1
[. −ζ[
2n−1
dσ(ζ)
d(.)
−
,
for any small c 0, where d(.) is the distance from . to the boundary of 1. This
proves that 1
−
(.) ∈ C
1−
(1). Using (2.2.7), we obtain through a similar argument
the same assertion for 1
+
(.). This proves the theorem.
2.3 The CauchyRiemann Operator in C 25
2.3 The CauchyRiemann Operator in C
Let
(2.3.1) A = A
1
+iA
2
be a ﬁrst order partial diﬀerential operator deﬁned in some open neighborhood l
of the origin in 1
2
, where
(2.3.2) A
j
= o
j
(r, j)
∂
∂r
+/
j
(r, j)
∂
∂j
, , = 1, 2,
and o
j
(r, j), /
j
(r, j) are realvalued functions on l. We wish to study the solvability
of such operator. If A
1
and A
2
are linearly dependent everywhere on l, then A
1
and A
2
will be multiples of the same ﬁrst order operator A
0
with real coeﬃcients
in some neighborhood of the origin. It follows that A is reduced to
(2.3.3) A = λ(.)A
0
,
and the solvability of (2.3.3) will then follow from the basic theory of the ordinary
diﬀerential equations.
Thus, let us assume that A
1
and A
2
are linearly independent everywhere on l.
The most famous operator of this type is the CauchyRiemann operator,
(2.3.4)
∂
∂.
=
1
2
_
∂
∂r
+i
∂
∂j
_
.
In Theorem 2.1.2, we have shown how to obtain a solution for the inhomogeneous
CauchyRiemann equation via the Cauchy integral formula.
As another application of the Cauchy integral formula, we shall show that, under
certain regularity hypotheses on the coeﬃcients o
j
(r, j) and /
j
(r, j), locally one
may introduce a new holomorphic coordinate n so that A can be converted to the
CauchyRiemann operator in n. Hence, one can deduce the solvability of A from
the knowledge of the CauchyRiemann operator. More precisely, we will prove the
following theorem.
Theorem 2.3.1. Let A be given as in (2.3.1) and (2.3.2) in some open neigh
borhood l of the origin in 1
2
. Suppose that the coeﬃcients o
j
(r, j) and /
j
(r, j),
, = 1, 2, are H¨older continuous of order λ, 0 < λ < 1. Then there exists a new local
holomorphic coordinate n in some neighborhood of the origin so that
A(n) =
∂
∂n
.
By a linear transformation on r, j with constant coeﬃcients we may assume that
the operator A takes the following form
(2.3.5) A(.) =
∂
∂.
−o(.)
∂
∂.
,
26 The Cauchy Integral Formula and its Applications
with o(0) = 0, and that o(.) is H¨older continuous of order λ. Hence, the assertion
of Theorem 2.3.1 is equivalent to the existence of a solution n(.) to the equation
n
z
(.) −o(.)n
z
(.) = 0,
or, in terms of the partial diﬀerential operator 7,
(2.3.6) n
z
(.) = 7n(.),
where 7 = o(.)(∂,∂.), with n
z
(0) ,= 0.
We shall use an iteration process to construct a solution to equation (2.3.6) for
the remaining parts of this section. This is where one needs H¨older regularity for
the coeﬃcients of A. We shall ﬁrst prove some lemmas and estimates that are
needed in the sequel. We denote by 1
R
= 1
R
(0) the disc centered at the origin
with radius 1 in 1
2
.
Lemma 2.3.2. Let ζ = (o, /) ∈ 1
R
. Put : = [. −ζ[. Then, for any λ 0, we
have
(2.3.7)
__
D
R
:
λ
:
2
drdj ≤
2π
λ
(21)
λ
.
Proof. This is obvious if we apply polar coordinates to the disc centered at ζ with
radius 21.
Lemma 2.3.3. Let 0 < c
1
, c
2
≤ 1 with c
1
+ c
2
,= 2. Then, for any two distinct
points ζ
1
and ζ
2
contained in 1
R
, we have
i
2
__
D
R
d. ∧ d.
[. −ζ
1
[
2−1
[. −ζ
2
[
2−2
≤ c(c
1
, c
2
)
1
[ζ
1
−ζ
2
[
2−1−2
.
What is essential in this lemma is that the constant c(c
1
, c
2
) depends only on c
1
and c
2
, but not on ζ
1
and ζ
2
.
Proof. By changing to polar coordinates it is easy to see that the integral exists.
For the estimate of the integral, let 2δ = [ζ
1
−ζ
2
[ and
´
1
= ¦. ∈ 1
R
[ [. −ζ
1
[ < δ¦,
´
2
= ¦. ∈ 1
R
[ [. −ζ
2
[ < δ¦,
´
3
= 1
R
¸ ¦´
1
∪ ´
2
¦.
Then, by changing to polar coordinates, the integral over ´
1
can be estimated as
follows:
i
2
__
Z1
d. ∧ d.
[. −ζ
1
[
2−1
[. −ζ
2
[
2−2
≤ 2πδ
2−2
_
δ
0
:
1−1
d: =
2π
c
1
δ
1+2−2
.
Similarly, we have the estimate over ´
2
,
i
2
__
Z2
d. ∧ d.
[. −ζ
1
[
2−1
[. −ζ
2
[
2−2
≤
2π
c
2
δ
1+2−2
.
2.3 The CauchyRiemann Operator in C 27
Both estimates are of the desired form. For the estimate over ´
3
, note that the
function (. −ζ
1
),(. −ζ
2
) is smooth on ´
3
. Hence, we obtain
1
3
≤
¸
¸
¸
¸
. −ζ
1
. −ζ
2
¸
¸
¸
¸
≤ 3, for . ∈ ´
3
.
It follows we have
i
2
__
Z3
d. ∧ d.
[. −ζ
1
[
2−1
[. −ζ
2
[
2−2
≤ 3
2−2
__
Z3
drdj
[. −ζ
1
[
4−1−2
≤ 3
2−2
__
C\Z1
drdj
[. −ζ
1
[
4−1−2
≤ 2π3
2−2
_
∞
δ
:
1+2−3
d:
=
2π3
2−2
2 −c
1
−c
2
δ
1+2−2
.
This completes the proof of Lemma 2.3.3.
The following lemma is the key for the regularity of the ∂equation.
Lemma 2.3.4. Let )(.) be a complexvalued continuous function deﬁned on 1
R
which satisﬁes
(2.3.8) [)(.
1
) −)(.
2
)[ ≤ 1[.
1
−.
2
[
λ
,
for any two points .
1
, .
2
∈ 1
R
, where λ, 1 are positive constants with 0 < λ < 1.
Deﬁne the function 1(ζ) for ζ ∈ 1
R
by
1(ζ) =
1
2πi
__
D
R
)(.)
. −ζ
d. ∧ d..
Then 1 ∈ C
1+λ
(1
R
). If [)(.)[ ≤ ¹ for all . ∈ 1
R
, then we have
(1) 1
ζ
and 1
ζ
exist, and 1
ζ
(ζ) = )(ζ), ζ ∈ 1
R
.
(2) [1(ζ)[ ≤ 41¹, ζ ∈ 1
R
.
(3) [1
ζ
(ζ)[ ≤ (
2
λ+1
λ
)1
λ
1, ζ ∈ 1
R
.
(4) [1(ζ
1
) −1(ζ
2
)[ ≤ 2(¹+ (
2
λ+1
λ
)1
λ
1)[ζ
1
−ζ
2
[, ζ
1
, ζ
2
∈ 1
R
.
(5) [1
ζ
(ζ
1
) −1
ζ
(ζ
2
)[ ≤ j(λ)1[ζ
1
−ζ
2
[
λ
, ζ
1
, ζ
2
∈ 1
R
.
where j(λ) 0 is independent of ζ
1
and ζ
2
.
Proof. The existence of 1
ζ
and the equality 1
ζ
(ζ) = )(ζ) are guaranteed by Theo
rem 2.1.2. To prove the existence of 1
ζ
(ζ), we write
1(ζ) =
1
2πi
__
D
R
)(.) −)(ζ)
. −ζ
d. ∧ d. +
)(ζ)
2πi
__
D
R
1
. −ζ
d. ∧ d..
Note that by Cauchy’s integral formula (2.1.1), we get
ζ =
1
2πi
__
D
R
1
. −ζ
d. ∧ d..
28 The Cauchy Integral Formula and its Applications
Hence, if ) ∈ C
1
(1), we clearly have
(2.3.9) 1
ζ
(ζ) =
1
2πi
__
D
R
)(.) −)(ζ)
(. −ζ)
2
d. ∧ d..
In general, if ) is only H¨older continuous of order λ, 0 < λ < 1, we approximate )
by functions in C
1
(1) to get (2.3.9). This proves (1).
Estimate (2) then follows from (2.1.2) and Lemma 2.3.2 with λ = 1 since
[1(ζ)[ ≤
¹
π
__
D
R
1
[. −ζ[
drdj ≤ 41¹.
Similarly, from (2.3.8) and (2.3.9) we have
[1
ζ
(ζ)[ ≤
1
π
__
D
R
[. −ζ[
λ
[. −ζ[
2
drdj ≤
_
2
λ+1
λ
_
1
λ
1.
This gives (3). Now (4) follows immediately from the Mean Value Theorem and
estimate (3).
Finally, we estimate (5). Let ζ
1
and ζ
2
be two ﬁxed distinct points in 1
R
, and
set
β =
)(ζ
2
) −)(ζ
1
)
ζ
1
−ζ
2
.
By the assumption (2.3.8) on )(ζ), we have
[β[ ≤ 1[ζ
1
−ζ
2
[
λ−1
.
Set
ˆ
)(.) = )(.) +β., then we have
´
1(ζ) ≡
1
2πi
__
D
R
ˆ
)(.)
. −ζ
d. ∧ d. = 1(ζ) −β1
2
+β[ζ[
2
.
It follows that
´
1
ζ
(ζ) = 1
ζ
(ζ) +βζ.
Note also that the deﬁnition of β gives the following:
(1)
ˆ
)(ζ
1
) =
ˆ
)(ζ
2
),
(2)
ˆ
)(.) −
ˆ
)(ζ
1
) = )(.) −)(ζ
1
) +β(. −ζ
1
),
(3)
ˆ
)(.) −
ˆ
)(ζ
2
) = )(.) −)(ζ
2
) +β(. −ζ
2
),
Based on these observations, we obtain
2πi(
´
1
ζ
(ζ
1
) −
´
1
ζ
(ζ
2
))
=
__
D
R
_
ˆ
)(.) −
ˆ
)(ζ
1
)
(. −ζ
1
)
2
−
ˆ
)(.) −
ˆ
)(ζ
2
)
(. −ζ
2
)
2
_
d. ∧ d.
=
__
D
R
(
ˆ
)(.) −
ˆ
)(ζ
1
))(ζ
1
−ζ
2
)((. −ζ
1
) + (. −ζ
2
))
(. −ζ
1
)
2
(. −ζ
2
)
2
d. ∧ d.
2.3 The CauchyRiemann Operator in C 29
= (ζ
1
−ζ
2
)
__
D
R
)(.) −)(ζ
2
)
(. −ζ
1
)(. −ζ
2
)
2
d. ∧ d.
+ (ζ
1
−ζ
2
)
__
D
R
)(.) −)(ζ
1
)
(. −ζ
1
)
2
(. −ζ
2
)
d. ∧ d.
+ 2β(ζ
1
−ζ
2
)
__
D
R
1
(. −ζ
1
)(. −ζ
2
)
d. ∧ d.
= 1
1
+1
2
+1
3
.
By Lemma 2.3.3 the term 1
1
can be estimated by
[1
1
[ ≤ 21[ζ
1
−ζ
2
[
__
D
R
drdj
[. −ζ
2
[
2−λ
[. −ζ
1
[
≤ c(λ)1[ζ
1
−ζ
2
[
λ
,
where the constant c(λ) depends only on λ. A similar estimate holds for 1
2
. The
term 1
3
can be written as
1
3
= 2β(ζ
1
−ζ
2
)
__
D
R
1
(. −ζ
1
)(. −ζ
2
)
d. ∧ d.
= 2β
__
D
R
_
1
. −ζ
1
−
1
. −ζ
2
_
d. ∧ d.
= (4πi)β(ζ
1
−ζ
2
).
Hence, we have
[1
3
[ ≤ 4π1[ζ
1
−ζ
2
[
λ
.
These estimates together show that
[1
ζ
(ζ
1
) −1
ζ
(ζ
2
)[ ≤ [
´
1
ζ
(ζ
1
) −
´
1
ζ
(ζ
2
)[ +[β[[ζ
1
−ζ
2
[
≤ j(λ)1[ζ
1
−ζ
2
[
λ
.
The constant j(λ) is obviously independent of ζ
1
and ζ
2
, and the proof of Lemma
2.3.4 is now complete.
With the aid of Lemma 2.3.4 we now prove the following existence and uniqueness
theorem of an integrodiﬀerential equation from which Theorem 2.3.1 will follow.
Proposition 2.3.5. Let
(2.3.10) 7n = o(.)n
z
,
be a partial diﬀerential operator whose coeﬃcient o(.) is H¨older continuous of order
λ, 0 < λ < 1, and vanishes at . = 0. Then the equation
(2.3.11) 2πin(ζ) −
__
D
R
7n(.)
. −ζ
d. ∧ d. = σ(ζ), ζ ∈ 1
R
,
where σ(ζ) is a holomorphic function with σ(0) = 0, has exactly one solution n(.) ∈
C
1+λ
(1
R
), provided that 1 is suﬃciently small.
30 The Cauchy Integral Formula and its Applications
Before proceeding to the proof of this proposition, based on Lemma 2.3.4, we shall
ﬁrst make some further estimates of integrals. The hypotheses of the proposition
imply the existence of a number ` 0 so that the following estimates hold,
(2.3.12) 1 < `, [σ(ζ)[ ≤ `,
(2.3.13) [/(ζ
1
) −/(ζ
2
)[ ≤ `[ζ
1
−ζ
2
[
λ
,
(2.3.14) [7σ(ζ
1
) −7σ(ζ
2
)[ ≤
`
2
2
λ
[ζ
1
−ζ
2
[
λ
,
where ζ, ζ
1
, ζ
2
denote any three points of 1
R
and /(.) stands for the functions o(.)
and σ(.). Since o(.) and 7σ(.) both vanish at . = 0, we have
(2.3.15) [o(ζ)[ ≤ `[ζ[
λ
≤ `1
λ
,
and
(2.3.16) [7σ(ζ)[ ≤ `
2
1
λ
.
Now we consider the function 1(ζ) deﬁned in Lemma 2.3.4 and using the notation
of the lemma, we obtain
(2.3.17) [71(ζ)[ = [o(ζ)1
ζ
[ ≤
_
2
λ+1
λ
_
`1
2λ
1,
and
(2.3.18)
[71(ζ
1
) −71(ζ
2
)[
= [o(ζ
1
)1
ζ
(ζ
1
) −o(ζ
2
)1
ζ
(ζ
2
)[
≤ `[ζ
1
−ζ
2
[
λ
_
2
λ+1
λ
_
1
λ
1 +`1
λ
j(λ)1[ζ
1
−ζ
2
[
λ
≤ `[ζ
1
−ζ
2
[
λ
p(1)1,
where
p(1) =
__
2
λ+1
λ
_
+j(λ)
_
1
λ
,
is a function of 1 which tends to zero as 1 approaches zero.
Proof of Proposition 2.3.5. Based on the estimates obtained above, we shall ﬁrst
prove the existence of a solution to (2.3.11) by successive approximations. In order
to make the iteration converge, we shall choose the radius 1 to be suﬃciently small
so that it satisﬁes
(2.3.19) 41
λ
≤ 1 and 2
2−λ
_
1 +
2
λ
_
1
1−λ
≤ 1.
2.3 The CauchyRiemann Operator in C 31
Denote by c 0 another universal constant such that
(2.3.20) j(λ) +
2
λ+1
λ
≤ c.
Now we construct a sequence of functions ¦n
j
(ζ)¦
∞
j=0
to generate a solution of
(2.3.11). We ﬁrst set
(2.3.21) 2πin
0
(ζ) = σ(ζ),
and inductively deﬁne for ζ ∈ 1
R
,
(2.3.22) 2πin
n+1
(ζ) =
__
D
R
7n
n
(.)
. −ζ
d. ∧ d., n = 0, 1, 2, .
Claim 2.3.6. The functions ¦n
j
(ζ)¦
∞
j=0
satisfy the following estimates:
(1) [n
n
(ζ)[ ≤ `(c`1
λ
)
n
, ζ ∈ 1
R
.
(2) [7n
n
(ζ)[ ≤ `(c`1
λ
)
n+1
, ζ ∈ 1
R
.
(3) [n
n
(ζ
1
) −n
n
(ζ
2
)[ ≤ `(c`1
λ
)
n
[ζ
1
−ζ
2
[
λ
, ζ
1
, ζ
2
∈ 1
R
.
(4) [7n
n
(ζ
1
) −7n
n
(ζ
2
)[ ≤ (
cM
2
2
λ
)(c`1
λ
)
n
[ζ
1
−ζ
2
[
λ
. ζ
1
, ζ
2
∈ 1
R
.
In particular, estimate (4) of the claim implies that the function under the integral
sign in (2.3.22) is H¨older continuous of order λ, thus allowing the deﬁnition of the
next integral, and the iteration continues.
Proof of the claim. The claim will be proved by an induction on n. The initial
step n = 0 follows easily from (2.3.12),(2.3.13),(2.3.14) and (2.3.16). Hence, let us
assume that the claim is valid up to step n and proceed to prove the statement for
n + 1.
By estimates (2) and (4) of Lemma 2.3.4, the choice of 1 in (2.3.19) and the
induction hypotheses, we obtain
[n
n+1
(ζ)[ ≤ 41`(c`1
λ
)
n+1
≤ `(c`1
λ
)
n+1
,
and
[n
n+1
(ζ
1
) −n
n+1
(ζ
2
)[
≤ 2[ζ
1
−ζ
2
[
_
`(c`1
λ
)
n+1
+
_
2
λ+1
λ
_
1
λ
_
c`
2
2
λ
_
(c`1
λ
)
n
_
= `(c`1
λ
)
n+1
[ζ
1
−ζ
2
[
λ
_
1 +
2
λ
_
2[ζ
1
−ζ
2
[
1−λ
≤ `(c`1
λ
)
n+1
[ζ
1
−ζ
2
[
λ
_
2
2−λ
_
1 +
2
λ
_
1
1−λ
_
≤ `(c`1
λ
)
n+1
[ζ
1
−ζ
2
[
λ
.
This proves (1) and (3) of the claim for n + 1.
32 The Cauchy Integral Formula and its Applications
Next, we apply (2.3.17), (2.3.19) and (2.3.20) and the induction hypotheses to
get (2) for n + 1,
[7n
n+1
(ζ)[ ≤
_
2
λ+1
λ
_
`1
2λ
(
c`
2
2
λ
)
_
c`1
λ
_
n
= `(c`1
λ
)
n+2
_
2
λc
_
≤ `(c`1
λ
)
n+2
.
Estimate (4) of the claim for n + 1 can be obtained from (2.3.18), (2.3.19) and
(2.3.20) as follows:
[7n
n+1
(ζ
1
) −7n
n+1
(ζ
2
)[ ≤
_
c`
2
2
λ
_
(c`1
λ
)
n+1
[ζ
1
−ζ
2
[
λ
_
1
c1
λ
_
p(1)
≤
_
c`
2
2
λ
_
(c`1
λ
)
n+1
[ζ
1
−ζ
2
[
λ
.
This completes the induction procedure, and hence the proof of the claim.
We return to the proof of Proposition 2.3.5. In addition to (2.3.19), let the radius
1 of the domain be chosen so small that it also satisﬁes c`1
λ
< 1. It follows that
the series
(2.3.23)
∞
j=0
n
j
(.)
converges absolutely and uniformly in 1
R
, and deﬁnes a solution n(.) ∈ C
1+λ
(1
R
)
which satisﬁes (2.3.11).
For the uniqueness of the solution when 1 is suﬃciently small, let η(.) be another
solution of (2.3.11) such that 7η(.) satisﬁes a H¨older condition of order λ. Then
the function
´ n(.) = n(.) −η(.)
satisﬁes the equation
(2.3.24) 2πi ´ n(ζ) =
__
D
R
7 ´ n(.)
. −ζ
d. ∧ d..
Put
¹ = ¹(1) = sup
ζ∈D
R
[7 ´ n(ζ)[,
and
1 = 1(1) = sup
ζ1,ζ2∈D
R
ζ1,=ζ2
[7 ´ n(ζ
1
) −7 ´ n(ζ
2
)[
[ζ
1
−ζ
2
[
λ
.
Obviously, ¹ and 1 will in general depend on 1, and both decrease as 1 tends to
zero. Then, from (2.3.17) and (2.3.18) we obtain
¹ ≤
_
2
λ+1
λ
_
`1
2λ
1,
33
and
1 ≤ `p(1)1.
Since p(1) approaches zero when 1 tends to zero, we must have 1 = 0 for suﬃ
ciently small 1. This implies ¹ = 0 when 1 is suﬃciently small. Hence n(.) = η(.)
on 1
R
for some suﬃciently small 1. This proves the uniqueness part of the propo
sition, and the proof of Proposition 2.3.5 is now complete.
Proof of Theorem 2.3.1. We are now in a position to prove Theorem 2.3.1. To solve
equation (2.3.6) we set σ(.) = . in the statement of Proposition 2.3.5. Then there
exists a unique solution n(.) ∈ C
1+λ
(1
R
) to the equation
2πin(ζ) −
__
D
R
7n(.)
. −ζ
d. ∧ d. = ζ, for ζ ∈ 1
R
,
with n
ζ
(0) ,= 0 if 1 is suﬃciently small. Since 7n(.) satisﬁes a H¨older condition
of order λ, we see by Lemma 2.3.4 that n(.) satisﬁes equation (2.3.6). This proves
Theorem 2.3.1.
NOTES
The Plemelj jump formula associated with the Cauchy transform was proved in
[Ple 1]. Theorem 2.1.6 which is the analog of the Poincar´e lemma for the ∂ operator
is often known as the DolbeaultGrothendieck lemma (see [Dol 1,2]).
Theorem 2.2.1 is a special case of the socalled BochnerMartinelliKoppelman
formula due to W. Koppelman [Kop 1]. Corollary 2.2.2 concerning the reproducing
property of the BochnerMartinelli kernel for holomorphic functions was discovered
independently by S. Bochner [Boc 1] and E. Martinelli [Mar 1]. The jump for
mula stated in Theorem 2.2.3, which extends the jump formula associated with the
Cauchy transform on the complex plane, can be found in [HaLa 1]. See also the
book by R. M. Range [Ran 6] for more discussions.
Theorem 2.3.1 is known to geometers as the theorem of Korn and Lichtenstein
which states that given a Riemannian metric
d:
2
= p
11
(r, j)dr
2
+ 2p
12
(r, j)drdj +p
22
(r, j)dj
2
,
in some open neighborhood l of the origin in 1
2
, where the coeﬃcient functions
p
ij
(r, j), 1 ≤ i, , ≤ 2, are H¨older continuous of order λ, 0 < λ < 1, we have
near every point there is a neighborhood whose local coordinates are isothermal
parameters. By isothermal parameters we mean that, under new local coordinates,
the metric d:
2
takes the following normal form
d:
2
= λ(n, ·)(dn
2
+d·
2
),
for some λ(n, ·) 0. If the coeﬃcient functions p
ij
(r, j), 1 ≤ i, , ≤ 2, are as
sumed to be continuous only, then the Riemannian metric d:
2
cannot always be
transformed to the normal form. A counterexample was found by P. Hartman and
A. Wintner [HaWi 1]. The proof we present here for Theorem 2.3.1 is essentially
taken from [Cher 1] and [Ber 1]. See also Chapter IV of the book, Volume II, by R.
Courant and D. Hilbert [CoHi 1].
34 The Cauchy Integral Formula and its Applications
CHAPTER 3
HOLOMORPHIC EXTENSION
AND PSEUDOCONVEXITY
Let ` be a C
k
hypersurface in C
n
, and let j be a point on `, where / ∈ N.
By this we mean that there exists a C
k
realvalued deﬁning function ρ and an open
neighborhood l of j such that ` ∩ l = ¦. ∈ l[ρ(.) = 0¦ and dρ(.) ,= 0 on
`∩l. ` divides l into two sides, l
+
and l
−
, where l
+
= ¦. ∈ l[ρ(.) 0¦ and
l
−
= ¦. ∈ l[ρ(.) < 0¦. Deﬁne by 1, a type (0, 1) vector ﬁeld on `, such that
1 =
n
j=1
o
j
(.)
∂
∂.
j
on ` ∩ l,
where the o
t
j
: satisfy
n
j=1
o
j
(.)
∂ρ
∂.
j
(.) = 0 for all . ∈ ` ∩ l.
Any such vector ﬁeld 1 on ` is called a tangential CauchyRiemann equation.
Suppose that ) ∈ C
1
(l
−
) ∩ O(l
−
). By continuity, we see that 1) = 0 on ` ∩ l.
This shows that the restriction of a holomorphic function ) to a hypersurface will
automatically satisfy the homogeneous tangential CauchyRiemann equations.
Deﬁnition 3.0.1. Let ` be a C
1
hypersurface in C
n
, n ≥ 2. A C
1
function )
on ` is called a C1 function if ) satisﬁes the homogeneous tangential Cauchy
Riemann equations
n
j=1
o
j
∂)
∂.
j
(.) = 0, . ∈ `,
for all o = (o
1
, , o
n
) ∈ C
n
with
n
j=1
o
j
(∂ρ,∂.
j
)(.) = 0, . ∈ `, where ρ(.) is
a C
1
deﬁning function for `.
The restriction of a holomorphic function ) to a hypersurface is a C1 function.
Conversely, given any C1 function ) on `, can one extend ) holomorphically into
one side of `? This is the socalled holomorphic extension of C1 functions. In
general, the converse part is not true.
For instance, let ` be deﬁned by j
1
= 0 in C
n
, where .
j
= r
j
+ ij
j
, 1 ≤ , ≤
n. Consider a realvalued smooth function )(r
1
, .
2
, , .
n
) = )(r
1
), which is
independent of .
2
, , .
n
, in some open neighborhood of the origin. Suppose that
3.1 The Hartogs Extension Theorem 35
)(r
1
) is not real analytic at the origin. Note that )(r
1
, .
2
, , .
n
) is annihilated by
the tangential type (0, 1) vector ﬁelds ∂,∂.
2
, , ∂,∂.
n
, hence ) is a C1 function
on `. Still, ) can not be holomorphically extended to some open neighborhood of
the origin, or to just one side of the hypersurface `.
In this chapter, we ﬁrst consider the problem of global holomorphic extension
of a C1 function on a compact hypersurface. We then study the local onesided
holomorphic extension of a C1 function. In Sections 4 and 5, we deﬁne plurisub
harmonic functions, pseudoconvex domains and domains of holomorphy. We study
their relations with each other, and give several equivalent deﬁnitions of pseudocon
vexity. Finally, we discuss the Levi problem and its relations with the
¯
∂equation.
3.1 The Hartogs Extension Theorem
One of the major diﬀerences between one and several complex variables is the
socalled Hartogs extension theorem, which states that if a bounded domain 1 in
C
n
, n ≥ 2, has connected boundary, then any holomorphic function )(.) deﬁned
in some open neighborhood of the boundary /1 can be holomorphically extended
to the entire domain 1. This sort of extension phenomenon fails in one complex
variable. For instance, )(.) = 1,. is holomorphic on the entire complex plane
except the origin, but there is no way to extend it as an entire function.
We consider the inhomogeneous CauchyRiemann equations in C
n
(3.1.1) ∂n = ),
where ) is a (0,1)form of class C
k
with / ≥ 1. Write ) as ) =
n
j=1
)
j
d.
j
. Since
∂ is a complex, a necessary condition for solving the ∂equation is ∂) = 0. More
explicitly, the equation (3.1.1) is overdetermined. In order to solve (3.1.1) for some
function n, it is necessary that the )
i
’s satisfy the following compatibility conditions:
(3.1.2)
∂)
j
∂.
k
=
∂)
k
∂.
j
,
for all 1 ≤ , < / ≤ n.
First we prove the following theorem:
Theorem 3.1.1. Let )
j
∈ C
k
0
(C
n
), n ≥ 2, , = 1, , n, and let / ≥ 1 be a positive
integer such that (3.1.2) is satisﬁed. Then there is a function n ∈ C
k
0
(C
n
) satisfying
(3.1.1). When / = 0, if (3.1.2) is satisﬁed in the distribution sense, then there exists
a function n ∈ C
0
(C
n
) such that n satisﬁes (3.1.1) in the distribution sense.
Proof. For / ≥ 1, set
n(.) =
1
2πi
_
C
)
1
(ζ, .
2
, , .
n
)
ζ −.
1
dζ ∧ dζ
=
1
2πi
_
C
)
1
(ζ +.
1
, .
2
, , .
n
)
ζ
dζ ∧ dζ.
It is easily seen that n ∈ C
k
(C
n
) from diﬀerentiation under the integral sign. We
also have n(.) = 0 when [.
2
[ + +[.
n
[ is suﬃciently large, since ) vanishes on the
set.
36 Holomorphic Extension and Pseudoconvexity
By Theorem 2.1.2, we have
∂n
∂.
1
= )
1
(.).
For , 1, using the compatibility condition (3.1.2), we obtain
∂n
∂.
j
=
1
2πi
_
C
∂fj
∂ζ
(ζ, .
2
, , .
n
)
ζ −.
1
dζ ∧ dζ = )
j
(.).
Hence, n(.) is a solution to the ∂equation (3.1.1). In particular, n is holomorphic on
the unbounded component of the complement of the support of ). Since n(.) = 0
when [.
2
[ + + [.
n
[ is suﬃciently large, we see from the Identity Theorem for
holomorphic functions that n must be zero on the unbounded component of the
complement of the support of ). This completes the proof of the theorem for / ≥ 1.
When / = 0, deﬁne n(.) by the same equation. We see that n ∈ C
0
(C
n
), and
that n(.) = 0 when [.
2
[ + +[.
n
[ is suﬃciently large.
By Theorem 2.1.2,
∂n
∂.
1
= )
1
(.).
For , 1, let φ ∈ C
∞
0
(C
n
). Then, using ( , )
C
n to denote the pairing between
distributions and test functions, we have
_
∂n
∂.
j
, φ
_
C
n
=
_
n, −
∂φ
∂.
j
_
C
n
= −
_
C
n
_
1
2πi
_
C
)
1
(ζ +.
1
, .
2
, , .
n
)
ζ
dζ ∧ dζ
_
∂φ
∂.
j
dλ(.)
=
−1
2πi
_
C
1
ζ
__
C
n
)
1
(ζ +.
1
, .
2
, , .
n
)
∂φ
∂.
j
dλ(.)
_
dζ ∧ dζ
=
1
2πi
_
C
1
ζ
_
∂)
1
∂.
j
(ζ +.
1
, .
2
, , .
n
), φ
_
C
n
dζ ∧ dζ
=
1
2πi
_
C
1
ζ
_
∂)
j
∂.
1
(ζ +.
1
, .
2
, , .
n
), φ
_
C
n
dζ ∧ dζ
=
−1
2πi
_
C
1
ζ
__
C
n
)
j
(ζ +.
1
, .
2
, , .
n
)
∂φ
∂.
1
dλ(.)
_
dζ ∧ dζ
= −
_
C
n
_
1
2πi
_
C
)
j
(ζ +.
1
, .
2
, , .
n
)
ζ
dζ ∧ dζ
_
∂φ
∂.
1
dλ(.)
=
_
∂
∂.
1
_
1
2πi
_
C
)
j
(ζ, .
2
, , .
n
)
ζ −.
1
dζ ∧ dζ
_
, φ
_
C
n
=
_
)
j
, φ
_
C
n
,
where the last equality is again guaranteed by Theorem 2.1.2. Hence,
∂n
∂.
j
= )
j
(.),
for , 1 in the distribution sense. One shows, similarly, that n must vanish on
the unbounded component of the complement of the support of ). This proves the
theorem.
3.1 The Hartogs Extension Theorem 37
Theorem 3.1.2 (Hartogs). Let 1 be a bounded domain in C
n
with n ≥ 2, and
let 1 be a compact subset of 1 so that 1¸ 1 is connected. Then any holomorphic
function ) deﬁned on 1 ¸ 1 can be extended holomorphically to 1.
Proof. Choose a cutoﬀ function χ ∈ C
∞
0
(1) such that χ = 1 in some open neigh
borhood of 1. Then −)(∂χ) ∈ C
∞
(0,1)
(C
n
) satisﬁes the compatibility conditions,
and it has compact support. By Theorem 3.1.1 there is a n ∈ C
∞
0
(C
n
) such that
∂n = −)∂χ,
and that n = 0 in some open neighborhood of C
n
¸ 1. Then, it is easily seen that
1 = (1 −χ)) −n
is the desired holomorphic extension of ).
Theorem 3.1.1 is the key for proving the Hartogs extension theorem. The hy
pothesis n ≥ 2 made in Theorem 3.1.1 is crucial. Using Corollary 2.1.5 it is clear
that in general, we cannot solve the equation ∂n,∂. = ) for a solution with compact
support in C when the given function ) has compact support.
Next, we prove another version of the holomorphic extension theorem which is
an easy application of the Cauchy integral formula.
Theorem 3.1.3. Let ) be a continuous function on a domain 1 in C
n
, and let o
be a smooth real hypersurface in C
n
. Suppose that ) is holomorphic in 1¸ o. Then
) is holomorphic on 1.
Proof. It suﬃces to show ) is holomorphic near each j ∈ 1 ∩ o. Let us ﬁx such a
point j. We may assume that j is the origin, and that locally near j, the hypersur
face o is realized as a graph which can be represented as
o = ¦(.
1
= r +ij, .
t
) ∈ C C
n−1
[ j = φ(r, .
t
)¦,
for some smooth function φ such that φ(0) = 0 and dφ(0) = 0.
Hence, for any small β 0, there exists a δ
β
0 and a polydisc l
β
in C
n−1
centered at the origin, such that
[φ(r, .
t
)[ < β
for all [r[ < δ
β
and .
t
∈ l
β
. Let β
1
0 be suﬃciently small, and let β
2
β
1
be a
positive number suﬃciently close to β
1
, then we may assume that \
0
l
β1
, where
\
0
= ¦.
1
∈ C[ [r[ < δ
β1
and β
1
< j < β
2
¦
is contained in 1 ¸ o. Thus, ) ∈ O(\
0
l
β1
).
Next, for each .
t
∈ l
β1
, )(.
1
, .
t
) is continuous on
\ = ¦.
1
∈ C[ [r[ < δ
β1
and [j[ < β
2
¦
and holomorphic on \ except for the smooth curve j = φ(r, .
t
). By Morera’s
theorem in one complex variable, )(.
1
, .
t
) is holomorphic in .
1
on \ . Now choose
38 Holomorphic Extension and Pseudoconvexity
a contour of integration Γ in l
β1
. Namely, let Γ = Γ
2
Γ
n
, where Γ
j
= ¦.
j
∈
C[ [.
j
[ = :
j
¦, for 2 ≤ , ≤ n, so that Γ ⊂ l
β1
.
Deﬁne
1(.
1
, .
t
) =
1
(2πi)
n−1
_
Γ
)(.
1
, ζ
t
)
(ζ
2
−.
2
) (ζ
n
−.
n
)
dζ
2
dζ
n
.
It is easily seen that 1(.
1
, .
t
) is holomorphic on \ l, where l = 1
2
1
n
and 1
j
= ¦.
j
∈ C[ [.
j
[ < :
j
¦. Since for (.
1
, .
t
) ∈ \
0
l, we have
1(.
1
, .
t
) = )(.
1
, .
t
).
Since ) is continuous, it follows from the Identity Theorem that ) is holomorphic
on \ l. The proof of the theorem is now complete.
3.2 The Holomorphic Extension Theorem from a Compact Hypersurface
In this section we shall prove a generalized version of the Hartogs extension
theorem. The following lemma is useful.
Lemma 3.2.1. Let ` be a C
k
hypersurface with a C
k
deﬁning function :, / ≥ 1.
Then any C1 function of class C
k
on ` can be extended to a C
k−1
function
˜
) in
some open neighborhood of ` such that
¯
∂
˜
) = 0 on `.
Proof. We ﬁrst extend ) to a C
k
function in some open neighborhood of `, still
denoted by ). Since ) is C1 on `, we have
1
i
) = 0, on `, i = 1, , n −1,
where 1
1
, , 1
n−1
forms a basis of the tangential Cauchy Riemann equations. Let
: be a deﬁning function for ` such that [d:[ = 1 on `. Then we simply modify )
to be
˜
) = ) −4:(1
n
)), where 1
n
=
n
j=1
(∂:,∂.
j
)(∂,∂.
j
) is the type (1, 0) vector
ﬁeld transversal to the boundary everywhere. When / ≥ 2, we have that :(1
n
)) is
a C
k−1
function and
(3.2.1) 1
n
˜
) = (1
n
)) −4(1
n
:)(1
n
)) = 0 on `.
This proves
¯
∂
˜
) = 0 on `.
When / = 1, (1
n
)) is only a C
0
function but it is easy to check, from the
deﬁnition, that :(1
n
)) is C
1
on ` and (3.2.1) still holds. This proves the lemma.
Theorem 3.2.2. Let 1 be a bounded domain in C
n
, n ≥ 2, with connected C
1
boundary. Let ) be a C1 function of class C
1
deﬁned on /1. Then, for any small
c 0, ) extends holomorphically to a function 1 ∈ C
1−
(1) ∩ O(1) such that
1[
bD
= ).
Proof. We deﬁne 1
−
(.) and 1
+
(.) as the BochnerMartinelli transform of ) on 1
and C
n
¸ 1. From Theorem 2.2.3, we have 1
−
∈ C
1−
(1), 1
+
∈ C
1−
(C
n
¸ 1) and
)(.) = 1
−
(.) −1
+
(.) for . ∈ /1.
3.2 The Holomorphic Extension Theorem from a Compact Hypersurface 39
We will ﬁrst show that 1
−
(.) ∈ O(1) and 1
+
(.) ∈ O(C
n
¸ 1). Deﬁne 1(ζ, .) by
(2.2.1) and
1
1
(ζ, .)
= −
n −1
(2πi)
n
_
n
j=1
ζ
j
−.
j
[ζ −.[
2n
dζ
j
_
∧
_
n
j=1
dζ
j
∧ dζ
j
_
n−2
∧
_
n
j=1
d.
j
∧ dζ
j
_
.
Then we have
(3.2.2) ∂
z
1(ζ, .) +∂
ζ
1
1
(ζ, .) = 0, for ζ ,= ..
Identity (3.2.2) is proved by a straightforward calculation as follows: For ζ ,= .,
−∂
z
1(ζ, .)
=
(n −1)!
(2πi)
n
n
j=1
d.
j
∧ dζ
j
[ζ −.[
2n
∧ ( ∧
k,=j
dζ
k
∧ dζ
k
)
−
n!
(2πi)
n
n
j=1
_
n
l=1
(ζ
j
−.
j
)(ζ
l
−.
l
)
[ζ −.[
2n+2
d.
l
∧ dζ
j
_
∧ ( ∧
k,=j
dζ
k
∧ dζ
k
)
= −
n −1
(2πi)
n
1
[ζ −.[
2n
_
n
j=1
d.
j
∧ dζ
j
_
∧
_
n
j=1
dζ
j
∧ dζ
j
_
n−1
−
n(n −1)
(2πi)
n
_
_
j,=l
(ζ
j
−.
j
)(ζ
l
−.
l
)
[ζ −.[
2n+2
d.
l
∧ dζ
j
∧ dζ
l
∧ dζ
l
_
_
∧
_
n
j=1
dζ
j
∧ dζ
j
_
n−2
+
n!
(2πi)
n
n
j=1
_
1
[ζ −.[
2n
−
[ζ
j
−.
j
[
2
[ζ −.[
2n+2
_
d.
j
∧ dζ
j
∧ ( ∧
k,=j
dζ
k
∧ dζ
k
)
40 Holomorphic Extension and Pseudoconvexity
= −
n −1
(2πi)
n
1
[ζ −.[
2n
_
n
j=1
d.
j
∧ dζ
j
_
∧ (
n
j=1
dζ
j
∧ dζ
j
)
n−1
+
n(n −1)
(2πi)
n
_
_
j,=l
(ζ
j
−.
j
)(ζ
l
−.
l
)
[ζ −.[
2n+2
dζ
l
∧ dζ
j
∧ d.
l
∧ dζ
l
_
_
∧
_
n
j=1
dζ
j
∧ dζ
j
_
n−2
+
n(n −1)
(2πi)
n
_
_
n
j=1
[ζ
j
−.
j
[
2
[ζ −.[
2n+2
dζ
j
∧ dζ
j
_
_
∧
_
n
j=1
d.
j
∧ dζ
j
_
∧
_
n
j=1
dζ
j
∧ dζ
j
_
n−2
= −
n −1
(2πi)
n
1
[ζ −.[
2n
_
n
j=1
dζ
j
∧ dζ
j
_
n−1
∧
_
n
j=1
d.
j
∧ dζ
j
_
+
n(n −1)
(2πi)
n
_
_
n
j=1
n
l=1
(ζ
j
−.
j
)(ζ
l
−.
l
)
[ζ −.[
2n+2
dζ
l
∧ dζ
j
_
_
∧
_
n
j=1
dζ
j
∧ dζ
j
_
n−2
∧
_
n
j=1
d.
j
∧ dζ
j
_
= ∂
ζ
1
1
(ζ, .).
This proves (3.2.2).
Since ) is a C1 function of class C
1
on /1, using Lemma 3.2.1, one can extend
) to be a continuous function in some open neighborhood of the boundary so that
) is diﬀerentiable at /1 and ∂) = 0 on /1. It follows now from (3.2.2) that for
. ∈ 1, we have
∂
z
1
−
(.) =
_
bD
)(ζ)∂
z
1(ζ, .)
= −
_
bD
)(ζ)∂
ζ
1
1
(ζ, .)
= −
_
bD
∂
ζ
()(ζ)1
1
(ζ, .))
= −
_
bD
d
ζ
()(ζ)1
1
(ζ, .))
= 0.
Thus, 1
−
(.) ∈ O(1) and, similarly, 1
+
(.) ∈ O(C
n
¸ 1).
Finally, we claim 1
+
(.) ≡ 0. We let [.
2
[ + + [.
n
[ be suﬃciently large, then
1
+
(, .
2
, , .
n
) is an entire function in .
1
which tends to zero as [.
1
[ tends to
inﬁnity. Thus, by Liouville’s theorem 1
+
(.) ≡ 0 if [.
2
[ + +[.
n
[ is large enough.
3.3 A Local Extension Theorem 41
It follows now from the identity theorem that 1
+
(.) ≡ 0. Setting 1 = 1
−
, we have
1
−
[
bD
= ) and this proves the theorem.
3.3 A Local Extension Theorem
The result in this section deals with the local onesided holomorphic extension of
a smooth C1 function deﬁned on some neighborhood of j on `. It turns out that
this question is related to the geometry of the domain. In particular, it is related
to the socalled Levi form of the domain.
Deﬁnition 3.3.1. Let 1 be a bounded domain in C
n
with n ≥ 2, and let : be a C
2
deﬁning function for 1. The Hermitian form
(3.3.1) L
p
(:; t) =
n
j,k=1
∂
2
:
∂.
j
∂.
k
(j)t
j
t
k
, j ∈ /1,
deﬁned for all t = (t
1
, , t
n
) ∈ C
n
with
n
j=1
t
j
(∂:,∂.
j
)(j) = 0 is called the Levi
form of the function : at the point j, denoted by L
p
(:; t).
If ρ is another C
2
deﬁning function for 1, then ρ = /: for some C
1
function
/ with / 0 on some open neighborhood of /1. Hence, for j ∈ /1 and t =
(t
1
, , t
n
) ∈ C
n
with
n
j=1
t
j
(∂:,∂.
j
)(j) = 0, we have
n
j,k=1
∂
2
ρ
∂.
j
∂.
k
(j)t
j
t
k
=
n
j,k=1
∂:
∂.
j
(j)
∂/
∂.
k
(j)t
j
t
k
+
n
j,k=1
∂/
∂.
j
(j)
∂:
∂.
k
(j)t
j
t
k
+/(j)
n
j,k=1
∂
2
:
∂.
j
∂.
k
(j)t
j
t
k
= /(j)
n
j,k=1
∂
2
:
∂.
j
∂.
k
(j)t
j
t
k
.
This shows that the Levi form associated with 1 is independent of the deﬁning
function up to a positive factor. In particular, the number of positive and negative
eigenvalues of the Levi form are independent of the choice of the deﬁning function.
For j ∈ /1, let
T
1,0
p
(/1) = ¦t = (t
1
, , t
n
) ∈ C
n
[
n
j=1
t
j
(∂:,∂.
j
)(j) = 0¦.
Then T
1,0
p
(/1) is the space of type (1, 0) vector ﬁelds which are tangent to the
boundary at the point j. Smooth sections in T
0,1
(/1) are the tangential Cauchy
Riemann operators deﬁned in the introduction. By deﬁnition, the Levi form is
applied only to the tangential type (1, 0) vector ﬁelds. We now state and prove the
local extension theorem for C1 functions.
42 Holomorphic Extension and Pseudoconvexity
Theorem 3.3.2. Let : be a C
2
deﬁning function for a hypersurface ` in a neigh
borhood l of j where j ∈ `. Assume that the Levi form L
p
(:; t) < 0 for some
t ∈ T
1,0
p
(`). Then there exists a neighborhood l
t
⊂ l of j such that for any
C1 function )(.) of class C
2
on ` ∩ l
t
, one can ﬁnd an 1(.) ∈ C
0
(l
t
+
),
where l
t
+
= ¦. ∈ l
t
[:(.) ≥ 0¦, so that 1 = ) on ` ∩ l
t
and ∂1 = 0 on
l
t
+
= ¦. ∈ l
t
[:(.) 0¦.
Proof. First we introduce new local coordinates near j. By a linear coordinate
change we may assume that j = 0 and that the Taylor expansion at 0 gives
:(.) = j
n
+¹(.) +O([.[
3
),
where .
n
= r
n
+ij
n
and
¹(.) =
n
j,k=1
∂
2
:
∂.
j
∂.
k
(0).
j
.
k
+ Re
_
n
j,k=1
∂
2
:
∂.
j
∂.
k
(0).
j
.
k
_
.
Consider the following holomorphic coordinate change. Let
n
j
= .
j
for 1 ≤ , ≤ n −1,
n
n
= .
n
+i
n
j,k=1
∂
2
:
∂.
j
∂.
k
(0).
j
.
k
.
Then the Taylor expansion becomes
:(n) = Imn
n
+
n
j,k=1
∂
2
:
∂n
j
∂n
k
(0)n
j
n
k
+O([n[
3
).
Therefore, we may assume that we are working in a local coordinate system . =
(.
1
, , .
n
) so that
:(.) = Im.
n
+
n
j,k=1
`
jk
.
j
.
k
+O([.[
3
),
where (`
jk
) is a Hermitian symmetric matrix. The hypothesis on the Levi form im
plies that the submatrix (`
jk
)
n−1
j,k=1
is not positive semideﬁnite. Hence, by another
linear change of coordinates, we may assume that `
11
< 0. Notice that
:(.
1
, 0, , 0) = `
11
[.
1
[
2
+O([.
1
[
3
).
Thus, we can ﬁrst choose δ 0, and then c 0 so that
∂
2
:
∂.
1
∂.
1
(.) < 0
on
l
t
= ¦. ∈ C
n
[ [.
1
[ < δ and [.
2
[ + +[.
n
[ < c¦ ⊂ l,
3.4 Pseudoconvexity 43
and :(.) < 0 on the part of the boundary where [.
1
[ = δ. For any ﬁxed .
t
=
(.
2
, , .
n
) with [.
2
[ + + [.
n
[ < c, the set of all .
1
with [.
1
[ < δ such that
:(.
1
, .
t
) < 0 must be connected. Otherwise, :(.
1
, .
t
) will attain a local minimum
at some point [.
1
[ < δ and we will have ´
z1
:(.
1
, .
t
) ≥ 0. This is a contradiction.
Consider now, a C1 function ) of class C
2
on l
t
∩ `. Using Lemma 3.2.1,
extend ) to l
t
+
, also denoted by ), so that ) ∈ C
1
(l
t
+
) and satisﬁes ∂) = 0 on
l
t
∩ `. If we write
∂) = p =
n
j=1
p
j
d.
j
,
then p
j
∈ C(l
t
+
) and p
j
= 0 on the boundary l
t
∩ `.
The p
j
t
:, extending by zero outside l
t
+
, will be viewed as functions deﬁned on
\ = C \ , where \ = ¦.
t
= (.
2
, , .
n
) ∈ C
n−1
[[.
2
[ + + [.
n
[ < c¦. For any
.
t
∈ \ , deﬁne
(3.3.2) G(.
1
, .
t
) =
1
2πi
_
C
p
1
(ζ, .
t
)
ζ −.
1
dζ ∧ dζ.
We have immediately that G(.) ∈ C
0
(\). Since ∂p = 0 in the sense of distribution,
Theorem 3.1.1 implies ∂G = p in the distribution sense.
Similarly, (∂G,∂.
j
)(.) = 0 on \ ¸ l
t
+
for 1 ≤ , ≤ n. It follows that G is
holomorphic on \¸l
t
+
. Also, notice that for any suﬃciently small positive number
0 < η << 1, there is a small open neighborhood \
0
of (0, , 0, −iη) in C
n−1
=
¦.
t
= (.
2
, , .
n
)¦ such that C\
0
is contained in \ and :(.
1
, .
t
) < 0 on C\
0
.
It implies that G(.) ≡ 0 on C \
0
. Hence, by the identity theorem, G(.) ≡ 0 on
\¸l
t
+
. In particular, G(.) = 0 on ¦. ∈ l
t
[ :(.) = 0¦. Now the function 1 = ) −G
is in C
0
(l
t
+
) with 1 = ) on ¦. ∈ l
t
[ :(.) = 0¦ and satisﬁes ∂1 = 0 on l
t
+
. This
proves the theorem.
Theorem 3.3.2 states that if the Levi form associated with the hypersurface `
has one nonzero eigenvalue, then we have onesided holomorphic extension of the
C1 functions. In particular, if the Levi form has eigenvalues of opposite signs,
then the given C1 function )(.) on ` can be extended holomorphically to both
sides, say, 1
+
(.) and 1
−
(.) respectively, such that 1
+
(.)[
M
= 1
−
(.)[
M
= )(.) on
`. Hence, by Theorem 3.1.3, 1
+
(.) and 1
−
(.) can be patched together to form a
holomorphic function deﬁned in some open neighborhood of the reference point j
in C
n
.
3.4 Pseudoconvexity
Let 1 be a bounded domain in C
n
with n ≥ 2. In this section we deﬁne the
concept of pseudoconvexity. We also discuss the relations between pseudoconvexity
and plurisubharmonic functions.
Deﬁnition 3.4.1. Let 1 be a bounded domain in C
n
with n ≥ 2, and let : be a C
2
deﬁning function for 1. 1 is called pseudoconvex, or Levi pseudoconvex, at j ∈ /1,
44 Holomorphic Extension and Pseudoconvexity
if the Levi form
(3.4.1) L
p
(:; t) =
n
j,k=1
∂
2
:
∂.
j
∂.
k
(j)t
j
t
k
≥ 0
for all t ∈ T
1,0
p
(/1). The domain 1 is said to be strictly (or strongly) pseudoconvex
at j, if the Levi form (3.4.1) is strictly positive for all such t ,= 0. 1 is called a
(Levi) pseudoconvex domain if 1 is (Levi) pseudoconvex at every boundary point
of 1. 1 is called a strictly (or strongly) pseudoconvex domain if 1 is strictly (or
strongly) pseudoconvex at every boundary point of 1.
Note that Deﬁnition 3.4.1 is clearly independent of the choice of the deﬁning
function :.
Deﬁnition 3.4.2. A function φ deﬁned on an open set 1 ⊂ C
n
, n ≥ 2, with values
in [−∞, +∞) is called plurisubharmonic if
(1) φ is upper semicontinuous,
(2) for any . ∈ 1 and n ∈ C
n
, φ(. + τn) is subharmonic in τ ∈ C whenever
¦. +τn[ τ ∈ C¦ ⊂ 1.
Theorem 3.4.3. A C
2
realvalued function φ on 1 is plurisubharmonic if and only
if
(3.4.2)
n
j,k=1
∂
2
φ
∂.
j
∂.
k
(.)t
j
t
k
≥ 0,
for all t = (t
1
, , t
n
) ∈ C
n
and all . ∈ 1.
Proof. The assertion follows immediately from the nonnegativeness of the Laplacian
of the subharmonic function φ(. +τn) in τ ∈ C whenever it is deﬁned.
If (3.4.2) is strictly positive, we shall call φ a strictly plurisubharmonic function.
It is obvious from Deﬁnition 3.4.2 that any plurisubharmonic function satisﬁes the
submean value property on each complex line where it is deﬁned. The following
theorem shows that there always exists a strictly plurisubharmonic deﬁning function
for any strongly pseudoconvex domain.
Theorem 3.4.4. Let 1 be a bounded strongly pseudoconvex domain in C
n
, n ≥ 2,
with a C
k
(2 ≤ / ≤ ∞) deﬁning function :(.). Then there exists a C
k
(2 ≤ / ≤ ∞)
strictly plurisubharmonic deﬁning function for 1.
Proof. For any λ 0, set
ρ(.) = c
λr
−1 for . ∈ 1.
We will show that ρ(.) is the desired strictly plurisubharmonic deﬁning function
for 1 if λ is chosen to be suﬃciently large.
First, ρ is a C
k
deﬁning function for 1 since
∇ρ(.) = λ∇:(.) ,= 0 for . ∈ /1.
3.4 Pseudoconvexity 45
Next, we calculate the Levi form of ρ(.) for . ∈ /1 to get that
n
j,k=1
∂
2
ρ
∂.
j
∂.
k
(.)t
j
t
k
= λ
n
j,k=1
∂
2
:
∂.
j
∂.
k
(.)t
j
t
k
+λ
2
¸
¸
n
j=1
∂:
∂.
j
(.)t
j
¸
¸
2
,
for t ∈ C
n
. By homogeneity, we may assume that [t[ = 1. Since 1 is of strong
pseudoconvexity, by continuity there exists an c 0 such that
n
j,k=1
(
∂
2
ρ
∂.
j
∂.
k
)(.)t
j
t
k
0
on the set ¦(., t) [ . ∈ /1, t ∈ C
n
, [t[ = 1, [
n
j=1
(∂:,∂.
j
)(.)t
j
[ < c¦. On the other
hand, if t ∈ C
n
is of unit length and satisﬁes [
n
j=1
(∂:,∂.
j
)(.)t
j
[ ≥
2
for . ∈ /1,
we may also achieve
n
j,k=1
(∂
2
ρ,∂.
j
∂.
k
)(.)t
j
t
k
0 simply by choosing λ to be
suﬃciently large. This shows that ρ is strictly plurisubharmonic near the boundary
by continuity if λ is large enough. This proves the theorem.
We recall that a bounded domain 1 ⊂ 1
N
with C
2
boundary is called strictly
convex if there is a C
2
deﬁning function ρ for 1 such that
N
j,k=1
∂
2
ρ
∂r
j
∂r
k
(j)t
j
t
k
0, j ∈ /1,
for all t = (t
1
, , t
N
) ∈ 1
N
with
N
j=1
∂ρ,∂r
j
(j)t
j
= 0.
Corollary 3.4.5. Let 1 be a bounded pseudoconvex domain with C
2
boundary in
C
n
, n ≥ 2. Then 1 is strongly pseudoconvex if and only if 1 is locally biholomor
phically equivalent to a strictly convex domain near every boundary point.
Proof. Suppose ﬁrst that 1 is strongly pseudoconvex. By Theorem 3.4.4 there is
a C
2
strictly plurisubharmonic deﬁning function :(.) for 1. Let j be a boundary
point. After a holomorphic coordinate change as we did in Theorem 3.3.2, we may
assume that j is the origin and the deﬁning function takes the following form
:(.) = Re.
n
+
n
j,k=1
∂
2
:
∂.
j
∂.
k
(0).
j
.
k
+O([.[
3
).
Since the quadratic term is positive by hypothesis for any . ,= 0, it is now easy
to see that 1 is strictly convex near j. The other direction is trivially true. This
proves the corollary.
Deﬁnition 3.4.6. A function ϕ : 1 → 1 on an open subset 1 in 1
n
is called an
exhaustion function for 1 if for every c ∈ 1 the set ¦r ∈ 1[ ϕ(r) < c¦ is relatively
compact in 1.
Clearly, if ϕ is an exhaustion function for 1, then ϕ(r) → ∞ as r → /1. This
condition is also suﬃcient if the domain 1 is bounded. Next, we show the exis
tence of a smooth strictly plurisubharmonic exhaustion function on a pseudoconvex
domain. Let d
D
(.) denote the Euclidean distance from . ∈ 1 to /1.
46 Holomorphic Extension and Pseudoconvexity
Theorem 3.4.7. Let 1 be a bounded pseudoconvex domain in C
n
, n ≥ 2, with a
C
2
boundary. Then −log(d
D
(.)) is plurisubharmonic near the boundary.
Proof. First set
(3.4.3) :(.) =
_
−d
D
(.) = −dist(., /1), for . ∈ 1,
dist(., /1), for . , ∈ 1.
Then it follows from the implicit function theorem that :(.) is a C
2
deﬁning function
for 1 in some small open neighborhood of the boundary. Hence the Levi form
deﬁned by : is positive semideﬁnite.
If −log(d
D
(.)) is not plurisubharmonic near the boundary, then
∂
2
∂τ∂τ
logd
D
(. +τn)[
τ=0
0
for some n ∈ C
n
and . close to the boundary where d
D
(.) is C
2
. Expand logd
D
(. +
τn) at τ = 0 to get
logd
D
(. +τn) = log(d
D
(.)) + Re(ατ +βτ
2
) +γ[τ[
2
+O([τ[
3
),
for small τ. Here α, β ∈ C, γ 0 are constants. Choose η ∈ C
n
such that .+η ∈ /1
and [η[ = d
D
(.). Then consider the analytic disc
´
δ
= ¦.(τ) = . +τn +ηc
ατ+βτ
2
[ [τ[ ≤ δ¦
for some suﬃciently small δ 0. Using Taylor’s expansion, for [τ[ ≤ δ, τ ,= 0, we
get
d
D
(.(τ)) ≥ d
D
(. +τn) −[η[[c
ατ+βτ
2
[
≥ [η[(c
γ
2
]τ]
2
−1)[c
ατ+βτ
2
[
0,
if δ is small enough. Since .(0) = . + η ∈ /1, this implies that ´
δ
is tangent to
the boundary at .(0). Hence,
∂
∂τ
d
D
(.(τ))[
τ=0
= 0 and
∂
2
∂τ∂τ
d
D
(.(τ))[
τ=0
0.
From the deﬁnition of :, this means
n
k=1
∂:
∂.
k
(. +η).
t
k
(0) = 0 and
n
j,k=1
∂
2
:
∂.
j
∂.
k
(. +η).
t
j
(0).
t
k
(0) < 0.
This contradicts the nonnegativeness of the Levi form at .(0) = . + η. Hence
−log(d
D
(.)) is plurisubharmonic near the boundary. This proves the theorem.
3.4 Pseudoconvexity 47
Corollary 3.4.8. Let 1 be a bounded pseudoconvex domain in C
n
, n ≥ 2, with
a C
2
boundary. Then there exists a smooth strictly plurisubharmonic exhaustion
function on 1.
Proof. By Theorem 3.4.7, −log(d
D
(.)) is a C
2
plurisubharmonic function for . ∈ 1
near the boundary. Let η(.) be a C
2
function on 1 such that η(.) = −log(d
D
(.)) on
l ∩1, where l is an open neighborhood of /1. We may assume that −log(d
D
(.))
is plurisubharmonic on l ∩ 1. Then it is easily seen that
λ(.) = η(.) +`[.[
2
is a C
2
strictly plurisubharmonic exhaustion function on 1 if ` is chosen large
enough.
The next step is to regularize λ(.). For each , ∈ N, we set 1
j
= ¦. ∈ 1[λ(.) <
,¦, then 1
j
⊂⊂ 1. Choose a function χ(.) = χ([.[) ∈ C
∞
0
(1(0; 1)) such that
χ(.) ≥ 0 and
_
χ(.)d\ = 1. Set χ
(.) = c
−2n
χ(.,c). For . ∈ 1
j
, the function
λ
(.) =
_
λ(. −ζ)χ
(ζ)d\ (ζ) =
_
λ(. −cζ)χ(ζ)d\ (ζ)
is deﬁned and smooth on 1
j
if c is suﬃciently small. Since λ is strictly plurisub
harmonic of class C
2
, it is clear that λ
(.) is strictly plurisubharmonic and, by the
submean value property, λ
1
≤ λ
2
if c
1
< c
2
, and λ
(.) converges uniformly to λ(.)
on any compact subset of 1.
Therefore, by extending λ
(.) in a smooth manner to 1, we see that there are
functions λ
j
(.) ∈ C
∞
(1) for , ∈ N such that λ
j
(.) is strictly plurisubharmonic
on 1
j+2
, λ(.) < λ
1
(.) < λ(.) + 1 on 1
2
and λ(.) < λ
j
(.) < λ(.) + 1 on 1
j
for
, ≥ 2. It follows that
λ
j
(.) −, + 1 < 0 on 1
j−2
for , ≥ 3,
and
λ
j
(.) −, + 1 0 on 1
j
−1
j−1
for , ≥ 3.
Now choose a β(r) ∈ C
∞
(1) with β(r) = 0 for r ≤ 0 and β(r), β
t
(r), β
tt
(r)
positive for r 0. Then, β(λ
j
(.) −, +1) ≥ 0 and β(λ
j
(.) −, +1) ≡ 0 on 1
j−2
.
A direct computation shows that β(λ
j
(.)−, +1) is plurisubharmonic on 1
j+2
and
strictly plurisubharmonic on 1
j
−1
j−1
. Thus, one may choose inductively :
j
∈ N
so that, for / ≥ 3,
ϕ
k
(.) = λ
1
(.) +
k
j=3
:
j
β(λ
j
(.) −, + 1)
is strictly plurisubharmonic and ϕ
k
(.) ≥ λ(.) on 1
k
. Clearly, ϕ
k
(.) = ϕ
k−1
(.) on
1
k−2
. Thus, ϕ(.) = lim
k→∞
ϕ
k
(.) is the desired smooth strictly plurisubharmonic
exhaustion function on 1. This completes the proof of Corollary 3.4.8.
Now if 1 is a bounded pseudoconvex domain in C
n
with a C
2
boundary, accord
ing to Corollary 3.4.8, there exists a smooth strictly plurisubharmonic exhaustion
48 Holomorphic Extension and Pseudoconvexity
function ϕ(.) on 1. Deﬁne 1
c
= ¦. ∈ 1[ ϕ(.) < c¦ for every c ∈ 1. It follows
from Sard’s Theorem that, for almost every c ∈ 1, 1
c
is a strictly pseudoconvex
domain with smooth boundary. In other words, any bounded pseudoconvex domain
1 in C
n
with a C
2
boundary can be exhausted by a sequence of smooth bounded
strictly pseudoconvex domains 1
c
.
When the domain 1 does not have smooth boundary or 1 is not bounded, we
deﬁne pseudoconvexity by the following
Deﬁnition 3.4.9. An open domain 1 in C
n
is called pseudoconvex if there exists
a smooth strictly plurisubharmonic exhaustion function ϕ(.) on 1.
Theorem 3.4.10. Let 1 be a pseudoconvex domain in C
n
, n ≥ 2, in the sense of
Deﬁnition 3.4.9. Then −log(d
D
(.)) is plurisubharmonic and continuous on 1.
Proof. Let ϕ be a smooth plurisubharmonic exhaustion function on 1. We shall
show that if .
0
is a point in 1 and n ∈ C
n
is a nonzero vector, then −logd
D
(.
0
+τn)
is subharmonic in τ ∈ C whenever .
0
+τn ∈ 1. Choose δ 0 so that
´
0
= ¦.
0
+τn[ [τ[ ≤ δ¦ ⊆ 1,
and let )(τ) be a holomorphic polynomial such that
(3.4.4) −logd
D
(.
0
+τn) ≤ Re)(τ) for [τ[ = δ.
We want to show that
−logd
D
(.
0
+τn) ≤ Re)(τ) for [τ[ ≤ δ.
Equation (3.4.4) is equivalent to
(3.4.5) d
D
(.
0
+τn) ≥ [c
−f(τ)
[ for [τ[ = δ.
Now, for any η ∈ C
n
with [η[ < 1, we consider the mapping with 0 ≤ t ≤ 1,
(3.4.6) τ →.
0
+τn +tηc
−f(τ)
for [τ[ ≤ δ.
The image of (3.4.6) is an analytic disc. Let ´
t
= ¦.
0
+τn +tηc
−f(τ)
[ [τ[ ≤ δ¦.
Set 1 = ¦t ∈ [0, 1][ ´
t
⊆ 1¦. Clearly, 0 ∈ 1 and 1 is open. To show that 1 is
closed, set 1 = ∪
0≤t≤1
/´
t
. Estimate (3.4.5) implies that 1 is a compact subset of
1. Now, if ´
t
⊆ 1 for some t, ϕ(.
0
+τn +tηc
−f(τ)
) would deﬁne a subharmonic
function in some open neighborhood of the closure of the unit disc in C. Therefore,
by the maximum principle for subharmonic functions and the exhaustion property
of ϕ, we see that ´
t
must be contained in ¦. ∈ 1[ ϕ(.) ≤ sup
K
ϕ¦, a compact
subset of 1. It follows that 1 is closed, and hence 1 = [0, 1]. This implies, for any
η ∈ C
n
with [η[ < 1 and [τ[ ≤ δ, that
.
0
+τn +ηc
−f(τ)
∈ 1.
Thus, we have
d
D
(.
0
+τn) ≥ [c
−f(τ)
[ for [τ[ ≤ δ,
or equivalently,
−logd
D
(.
0
+τn) ≤ Re)(τ) for [τ[ ≤ δ.
Hence, −logd
D
(.
0
+ τn) is subharmonic in τ ∈ C whenever .
0
+ τn ∈ 1. This
proves the theorem.
The equivalence between Deﬁnitions 3.4.1 and 3.4.9 on domains with smooth
boundaries is proved in the following theorem.
3.4 Pseudoconvexity 49
Theorem 3.4.11. Let 1 be a bounded domain in C
n
, n ≥ 2, with C
2
bound
ary. Then 1 is Levi pseudoconvex if and only if 1 is pseudoconvex according to
Deﬁnition 3.4.9.
Proof. If 1 is Levi pseudoconvex, then by Corollary 3.4.8 1 is pseudoconvex in the
sense of Deﬁnition 3.4.9.
On the other hand, assume 1 is pseudoconvex according to Deﬁnition 3.4.9.
Deﬁne : by(3.4.3). Then :(.) is a C
2
deﬁning function for 1 in some small open
neighborhood of the boundary.
Now Theorem 3.4.10 asserts that −log(d
D
(.)) is a C
2
plurisubharmonic function
if . ∈ 1 is suﬃciently close to the boundary. Thus, following from the plurisubhar
monicity of −log(d
D
(.)), we obtain that
n
j,k=1
_
−
1
d
D
∂
2
d
D
∂.
j
.
k
o
j
o
k
_
+
1
d
2
D
¸
¸
¸
¸
n
j=1
∂d
D
∂.
j
o
j
¸
¸
¸
¸
2
≥ 0,
for any o ∈ C
n
and . ∈ 1 suﬃciently close to /1. Therefore,
n
j,k=1
∂
2
:
∂.
j
.
k
(.)o
j
o
k
≥ 0 if
n
j=1
∂:
∂.
j
(.)o
j
= 0.
Passing to the limit, we obtain the desired assertion. This proves the theorem.
We note that by Deﬁnition 3.4.9 every pseudoconvex domain 1 can be exhausted
by strictly pseudoconvex domains, i.e.,
1 = ∪1
ν
,
where 1
ν
⊂⊂ 1
ν+1
⊂⊂ 1 and each 1
ν
is a strictly pseudoconvex domain.
To end this section we show that there always exists a bounded strictly plurisub
harmonic exhaustion function on any smooth bounded pseudoconvex domain.
Theorem 3.4.12. Let 1 ⊂ C
n
, n ≥ 2, be a smooth bounded pseudoconvex domain.
Let : be a smooth deﬁning function for 1. Then there exist constants 1 0 and
0 < η
0
< 1, such that for any η with 0 < η ≤ η
0
, ρ = −(−:c
−K]z]
2
)
η
is a smooth
bounded strictly plurisubharmonic exhaustion function on 1.
Note that ρ is continuous on 1 and vanishes on /1.
Proof. We ﬁrst assume that . ∈ 1 with [:(.)[ ≤ c for some small c 0. With
ρ = −(−:c
−K]z]
2
)
η
, a direct calculation shows, for t ∈ C
n
,
L
z
(ρ; t) = η(−:)
η−2
c
−ηK]z]
2
_
1:
2
_
[t[
2
−η1
¸
¸
¸
¸
n
j=1
.
j
t
j
¸
¸
¸
¸
2
_
+ (−:)
_
L
z
(:; t) −2η1Re
_
n
i=1
∂:
∂.
i
t
i
__
n
j=1
.
j
t
j
__
+ (1 −η)
¸
¸
¸
¸
n
i=1
∂:
∂.
i
t
i
¸
¸
¸
¸
2
_
.
50 Holomorphic Extension and Pseudoconvexity
For each . with [:(.)[ ≤ c and t = (t
1
, , t
n
) ∈ C
n
, write t = t
τ
+ t
ν
, where
t
ν
= (t
ν
1
, , t
ν
n
) with
t
ν
k
=
_
n
j=1
t
j
∂r
∂zj
(.)
n
j=1
[
∂r
∂zj
(.)[
2
_
∂:
∂.
k
(.),
and t
τ
= (t
τ
1
, , t
τ
n
) ∈ T
τ
z
= ¦o ∈ C
n
[
n
j=1
(∂:,∂.
j
)(.)o
j
= 0¦. Such a decompo
sition is clearly smooth when c is suﬃciently small. Also, let π(.) be the projection
of . along the normal on the boundary. Obviously, π is smooth for small c. Then
the Levi form of : at . is
L
z
(:; t
τ
) =
n
i,j=1
∂
2
:
∂.
i
∂.
j
(.)t
τ
i
(.)t
τ
j
(.)
=
n
i,j=1
∂
2
:
∂.
i
∂.
j
(.)(t
i
−t
ν
i
(.))(t
j
−t
ν
j
(.))
=
n
i,j=1
/
ij
(.)t
i
t
j
,
where /
ij
(.) is deﬁned by the last equality. Hence, by pseudoconvexity of the
domain, we have
(3.4.7)
L
z
(:; t
τ
(.)) ≥ L
z
(:; t
τ
(.)) −L
π(z)
(:; t
τ
(π(.)))
=
n
i,j=1
(/
ij
(.) −/
ij
(π(.)))t
i
t
j
≥ −C[:(.)[[t[
2
,
for some constant C 0. Since
(3.4.8) [t
ν
[ = O
_¸
¸
¸
¸
n
i=1
∂:
∂.
i
t
i
¸
¸
¸
¸
_
,
(3.4.7) and (3.4.8) together imply
L
z
(:; t) ≥ −C[:(.)[[t[
2
−C[t[
¸
¸
¸
¸
n
j=1
∂:
∂.
j
t
j
¸
¸
¸
¸
.
Hence
L
z
(ρ; t) ≥ η(−:)
η−2
c
−ηK]z]
2
_
1:
2
(1 −Cη1)[t[
2
−C:
2
[t[
2
+C:[t[
¸
¸
¸
¸
n
j=1
∂:
∂.
j
t
j
¸
¸
¸
¸
+ (1 −η)
¸
¸
¸
¸
n
j=1
∂:
∂.
j
t
j
¸
¸
¸
¸
2
_
,
3.5 Domains of Holomorphy 51
for some constant C 0. Since
C[:[[t[
¸
¸
¸
¸
n
j=1
∂:
∂.
j
t
j
¸
¸
¸
¸
≤
1
4
¸
¸
¸
¸
n
j=1
∂:
∂.
j
t
j
¸
¸
¸
¸
2
+C
1
:
2
[t[
2
,
we have
L
z
(ρ; t) ≥ η(−:)
η−2
c
−ηK]z]
2
_
1:
2
(1 −Cη1)[t[
2
−(C +C
1
):
2
[t[
2
+
_
3
4
−η
_¸
¸
¸
¸
n
j=1
∂:
∂.
j
t
j
¸
¸
¸
¸
2
_
.
Now, if we ﬁrst choose 1 2(C + C
1
) + 10 and then η to be suﬃciently small so
that η < 1,4 and Cη1 < 1,2, we have
L
z
(ρ; t) 0, for t ∈ C
n
¸ ¦0¦.
For this case we may take η
0
= min(1,4, 1,(2C1)).
If [:(.)[ ≥ c, the situation is even simpler. This proves the theorem.
3.5 Domains of Holomorphy
Throughout this section, 1 will denote a domain in C
n
, n ≥ 1. Here we give the
deﬁnition of a domain of holomorphy.
Deﬁnition 3.5.1. A domain 1 in C
n
is called a domain of holomorphy if we
cannot ﬁnd two nonempty open sets 1
1
and 1
2
in C
n
with the following properties:
(1) 1
1
is connected, 1
1
_ 1 and 1
2
⊂ 1
1
∩ 1.
(2) For every ) ∈ O(1) there is a
˜
) ∈ O(1
1
) satisfying ) =
˜
) on 1
2
.
According to Hartogs’ theorem (Theorem 3.1.2), if we remove a compact subset
1 from the unit ball 1(0; 1) in C
n
, n ≥ 2, such that 1(0; 1) ¸ 1 is connected, then
the remaining set 1(0; 1) ¸ 1 is not a domain of holomorphy. Also, from Theorem
3.3.2, if the Levi form of a smooth bounded domain 1 in C
n
, n ≥ 2, has one
negative eigenvalue, then 1 is not a domain of holomorphy.
In this section, we shall characterize the domain of holomorphy in C
n
for n ≥ 2.
Let 1 be a compact subset of 1. Deﬁne the holomorphically convex hull
´
1
D
of 1
in 1 by
(3.5.1)
´
1
D
= ¦. ∈ 1[ [)(.)[ ≤ sup
K
[)[, for all ) ∈ O(1)¦.
A compact subset 1 of 1 is called holomorphically convex if
´
1
D
= 1. We obviously
have
´
1
D
=
´
´
1
D
. Using )(.) = exp(o
1
.
1
+ +o
n
.
n
) with o
i
∈ C for i = 1, , n,
it is clear that
´
1
D
must be contained in the geometrically convex hull of 1, and
is a closed subset of 1. However,
´
1
D
in general is not a closed subset of C
n
,
i.e.,
´
1
D
in general is not a compact subset of 1. In one complex variable
´
1
D
is obtained from 1 by ﬁlling up all the bounded components of the complement
1
c
. For higher dimensional spaces, the situation is more subtle. In addition to the
concept of holomorphically convex hull, we deﬁne:
52 Holomorphic Extension and Pseudoconvexity
Deﬁnition 3.5.2. A domain 1 in C
n
is called holomorphically convex if
´
1
D
is
relatively compact in 1 for every compact subset 1 of 1.
The main task of this section is to prove the following characterization of domains
of holomorphy.
Theorem 3.5.3. Let 1 be a domain in C
n
, n ≥ 2. The following statements are
equivalent:
(1) 1 is a domain of holomorphy.
(2) dist(1, 1
c
) = dist(
´
1
D
, 1
c
) for every compact subset 1 in 1, where dist(1,
1
c
) denotes the distance between 1 and 1
c
= C
n
¸ 1.
(3) 1 is holomorphically convex.
(4) There exists a holomorphic function ) on 1 which is singular at every bound
ary point of 1.
Proof. (2) ⇒ (3) and (4) ⇒ (1) are obvious. We need to show (1) ⇒ (2) and (3)
⇒ (4).
If 1(0; :) is a polydisc centered at zero with multiradii : = (:
1
, , :
n
), for each
. ∈ 1, we set
d
r
(.) = sup¦λ 0[ ¦.¦ +λ1(0; :) ⊂ 1¦.
To prove (1) ⇒ (2), we ﬁrst show:
Lemma 3.5.4. Let 1 be a compact subset of a domain 1 in C
n
, and let ) ∈ O(1).
Suppose that
[)(.)[ ≤ d
r
(.) for . ∈ 1.
Let ζ be a ﬁxed point in
´
1
D
. Then any / ∈ O(1) extends holomorphically to
1 ∪ ¦¦ζ¦ +[)(ζ)[1(0; :)¦.
Proof. For each 0 < t < 1, the union of the polydiscs with centers at . ∈ 1
(3.5.2) 1
t
= ∪
z∈K
¦¦.¦ +t[)(.)[1(0; :)¦
is a compact subset of 1. Hence, for any / ∈ O(1), there exists `
t
0 such that
[/(.)[ ≤ `
t
on 1
t
. Using Cauchy’s estimates of /, we obtain
(3.5.3)
[
∂
α
h
∂z
α
(.)[t
]α]
[)(.)[
]α]
:
α
α!
≤ `
t
for . ∈ 1 and all multiindices α = (α
1
, , α
n
) with [α[ = α
1
+ + α
n
. Since
(∂
α
/,∂.
α
)(.))(.)
]α]
is holomorphic on 1, by deﬁnition, (3.5.3) also holds for . ∈
´
1
D
. Letting t tend to one, we see that /(.) extends holomorphically to 1∪¦¦ζ¦ +
[)(ζ)[1(0; :)¦. This proves the lemma.
We write
dist(., 1
c
) = sup¦: 0 [ . +on ∈ 1, for all n ∈ C
n
, [n[ ≤ 1 and
o ∈ C, [o[ < :¦
= inf
]w]≤1
d
w
(.),
3.5 Domains of Holomorphy 53
where
d
w
(.) = sup¦: 0 [ . +on ∈ 1, for all o ∈ C, [o[ < :¦.
Fix a n, we may assume that n = (1, 0, , 0). Denote by 1
j
= 1(0; :(,)) the
polydisc with multiradii :(,) = (1, 1,,, , 1,,) for , ∈ N. Then it is easily seen
that
lim
j→∞
d
r(j)
(.) = d
w
(.).
Thus, given c 0, if , is suﬃciently large, we have
(3.5.4) dist(1, 1
c
) ≤ (1 +c)d
r(j)
(.), . ∈ 1.
We let )(.) = dist(1, 1
c
),(1 +c) be the constant function. Since 1 is a domain of
holomorphy, using estimate (3.5.4), Lemma 3.5.4 shows that
dist(1, 1
c
) ≤ (1 +c)d
r(j)
(ζ) ≤ (1 +c)d
w
(ζ), for all ζ ∈
´
1
D
.
Letting c tend to zero, we get
dist(1, 1
c
) ≤ inf
ζ∈
b
K
D
( inf
]w]≤1
d
w
(ζ))
= inf
ζ∈
b
K
D
dist(ζ, 1
c
)
= dist(
´
1
D
, 1
c
).
This proves that (1) ⇒ (2).
Finally, we show (3) ⇒(4). Assume that 1 is holomorphically convex. Let T be
the set containing all points in 1 with rational coordinates. Clearly, T is countable
and dense in 1. Let ¦ζ
i
¦
∞
i=1
be a sequence of points in 1 such that every point
belonging to T appears inﬁnitely many times in the sequence. Now, we exhaust
1 by a sequence of increasingly holomorphically convex compact subsets ¦1
j
¦
∞
j=1
of 1 with 1
j
⊂
o
1
j+1
, where
o
1
j+1
is the interior of 1
j+1
. For each i, denote
by 1
ζi
the largest polydisc of the form 1
ζi
= ¦ζ
i
¦ + η1(0; 1) that is contained in
1, where η 0 and 1(0; 1) is the polydisc centered at the origin with multiradii
: = (1, , 1). Then, inductively for each ,, pick a .
j
∈ (1
ζj
¸ 1
nj
) ∩
o
1
nj+1
, where
¦1
nj
¦ is a suitable subsequence of ¦1
j
¦, and a )
j
(.) ∈ O(1) satisfying
[)
j
(.)[ <
1
2
j
, . ∈ 1
nj
,
and
[)
j
(.
j
)[ ≥
j−1
i=1
[)
i
(.
j
)[ +, + 1.
It follows that
/(.) =
∞
j=1
)
j
(.)
54 Holomorphic Extension and Pseudoconvexity
deﬁnes a holomorphic function on 1 and that
[/(.
j
)[ ≥ [)
j
(.
j
)[ −
j−1
i=1
[)
i
(.
j
)[ −
∞
i=j+1
[)
i
(.
j
)[ ≥ ,,
which implies /(.) is singular at every boundary point of 1. Otherwise, if /(.)
extends holomorphically across some boundary point, then /(.) would be bounded
on 1
ζi
for some ζ
i
. Obviously, it contradicts the construction of /. This proves (3)
⇒ (4), and, hence the theorem.
We see from Theorem 3.5.3 that the concept of domains of holomorphy is equiva
lent to that of holomorphic convexity. With this characterization, the next theorem
shows that a domain of holomorphy is pseudoconvex.
Theorem 3.5.5. If 1 is a domain of holomorphy, then 1 is pseudoconvex in the
sense of Deﬁnition 3.4.9.
Proof. Let 1 be a domain of holomorphy and ¦1
j
¦
∞
j=1
be a sequence of increas
ingly holomorphically convex compact subsets of 1 which exhausts 1. We may
assume that 1
j
⊂
o
1
j+1
for all ,. Then, by hypothesis, for each , ∈ N there exist
)
j1
, , )
jmj
∈ O(1) such that the function φ
j
(.) =
mj
k=1
[)
jk
(.)[
2
satisﬁes
(3.5.5) φ
j
(.) <
1
2
j
for . ∈ 1
j
,
and
φ
j
(.) , for . ∈ 1
j+2
¸
o
1
j+1
.
Hence,
ϕ(.) =
∞
j=1
φ
j
(.)
is a continuous exhaustion function deﬁned on 1. In fact, ϕ(.) is real analytic. It
can be seen easily from (3.5.5) that the series
∞
j=1
_mj
k=1
)
jk
(.))
jk
(n)
_
converges uniformly on compact subsets of 1 1
∗
, where 1
∗
= ¦. [ . ∈ 1¦ de
notes the conjugate domain of 1. Thus the series deﬁnes a holomorphic function on
11
∗
. By substituting . for n in the above series, we obtain the real analyticity
of ϕ(.) on 1, and that one can diﬀerentiate ϕ(.) term by term. Obviously, ϕ(.)
is plurisubharmonic on 1. It follows that [.[
2
+ϕ(.) is a smooth strictly plurisub
harmonic exhaustion function on 1, and by deﬁnition, 1 is pseudoconvex. This
proves the theorem.
3.6 The Levi Problem and the
¯
∂ Equation 55
3.6 The Levi Problem and the
¯
∂ Equation
Let 1 be a pseudoconvex domain in C
n
with n ≥ 2. One of the major problems
in complex analysis is to show that a pseudoconvex domain 1 is a domain of
holomorphy. Near each boundary point j ∈ /1, one must ﬁnd a holomorphic
function )(.) on 1 which cannot be continued holomorphically near j. This problem
is called the Levi problem for 1 at j. It involves the construction of a holomorphic
function with certain speciﬁc local properties.
If the domain 1 is strongly pseudoconvex with C
∞
boundary /1 and j ∈ /1,
one can construct a local holomorphic function ) in an open neighborhood l of
j, such that ) is holomorphic in l ∩ 1, ) ∈ C(1 ∩ l ¸ ¦j¦) and )(.) → ∞ as
. ∈ 1 approaches j. In fact ) can be easily obtained as follows: let : be a strictly
plurisubharmonic deﬁning function for 1 and we assume that j = 0. Let
1(.) = −2
n
i=1
∂:
∂.
i
(0).
i
−
n
i,j=1
∂
2
:
∂.
i
∂.
j
(0).
i
.
j
.
1(.) is holomorphic, and it is called the Levi polynomial of : at 0. Using Taylor’s
expansion at 0, there exists a suﬃciently small neighborhood l of 0 and C 0
such that for any . ∈ 1 ∩ l,
Re1(.) = −:(.) +
n
i,j=1
∂
2
:
∂.
i
∂¯ .
j
(0).
i
¯ .
j
+O([.[
3
) ≥ C[.[
2
.
Thus, 1(.) ,= 0 when . ∈ 1 ∩ l ¸ ¦0¦. Setting
) =
1
1
,
it is easily seen that ) is locally a holomorphic function which cannot be extended
holomorphically across 0.
Global holomorphic functions cannot be obtained simply by employing smooth
cutoﬀ functions to patch together the local holomorphic data, since the cutoﬀ
functions are no longer holomorphic. Let χ be a cutoﬀ function such that χ ∈
C
∞
0
(l) and χ = 1 in a neighborhood of 0. We note that χ) is not holomorphic in
1. However, if χ) can be corrected by solving a
¯
∂equation, then the Levi problem
will be solved.
Let us consider the (0,1)form p deﬁned by
p =
¯
∂(χ)) = (
¯
∂χ)).
This form p can obviously be extended smoothly up to the boundary. It is easy
to see that p is a
¯
∂closed form in 1 and p ∈ C
∞
(0,1)
(1). If we can ﬁnd a solution
n ∈ C
∞
(1) such that
(3.6.1)
¯
∂n = p in 1,
56 Holomorphic Extension and Pseudoconvexity
then we deﬁne for . ∈ 1,
/(.) = χ(.))(.) −n(.).
It follows that / is holomorphic in 1, / ∈ C
∞
(1¸¦0¦) and / is singular at 0. Thus
one can solve the Levi problem for strongly pseudoconvex domains provided one
can solve equation (3.6.1) with solutions smooth up to the boundary.
Problems of this sort are among the most diﬃcult in complex analysis and they
are the main topics of the next three chapters. In Chapter 4, we will solve the Levi
problem using the 1
2
estimate method for
¯
∂ (H¨ormander’s solution) on pseudo
convex domains (Theorem 4.5.2). In Chapter 5, we study the boundary regularity
for
¯
∂ on strongly pseudoconvex domains. This gives another solution (Kohn’s so
lution) of the Levi problem on complex manifolds (Theorem 5.3.11). In Chapter 6,
we further investigate the boundary regularity of
¯
∂ on pseudoconvex domains with
smooth boundaries for other applications.
NOTES
Theorem 3.1.2 is a theorem due to F. Hartogs [Har 1]. The present proof of The
orem 3.1.2 as pointed out by L. Ehrenpreis [Ehr 2] is essentially based on Theorem
3.1.1, i.e., the existence of compactly supported solutions to the CauchyRiemann
equation. The proof of Theorem 3.1.3 is based on an idea of F. Hartogs [Har 1].
Using a more delicate proof found in Harvey and Lawson [HaLa 1], one can prove
Theorem 3.2.2 in an optimal way. Namely, if the domain 1 has C
k
(1 ≤ / ≤ ∞)
boundary and ) is a C1 function of class C
k
on the boundary, then the holomorphic
extension 1 is also in C
k
(1). See also the book by R. M. Range [Ran 6].
Theorem 3.3.2 is concerned with the local onesided holomorphic extension of C1
functions which is essentially due to H. Lewy [Lew 1]. Another way to prove the
local extension theorem for the C1 functions is to invoke the result discovered by
Baouendi and Treves [BaTr 1]. This is the socalled analytic disc method. See the
books by A. Boggess [Bog 2] and M. S. Baouendi, P. Ebenfelt and L. P. Rothschild
[BER 1] for details and the references therein.
Corollary 3.4.5 in general is false for weakly pseudoconvex domains. A coun
terexample was discovered by J. J. Kohn and L. Nirenberg [KoNi 3]. The concept
of plurisubharmonicity (Deﬁnition 3.4.2) was ﬁrst introduced in two variables by K.
Oka [Oka 1], and by K. Oka [Oka 2] and P. Lelong [Lel 1] in arbitrary dimension. It
was K. Oka [Oka 1] who ﬁrst proved the plurisubharmonicity of −log(d
D
(.)) for a
pseudoconvex domain (Theorem 3.4.10) in C
2
. Later, similar results were obtained
independently by K. Oka [Oka 2], P. Lelong [Lel 2] and H. Bremermann [Bre 2]
in C
n
. The existence of a H¨older bounded strictly plurisubharmonic exhaustion
function on a C
2
bounded pseudoconvex domain was ﬁrst proved by K. Diederich
and J. E. Fornaess [DiFo 2]. The proof we present here for Theorem 3.4.12, based
on an idea of J. J. Kohn [Koh 6], is due to R. M. Range [Ran 4].
The characterization of domains of holomorphy in Theorem 3.5.3 is due to H.
Cartan and P. Thullen [CaTh 1]. For more discussion on pseudoconvexity and
domains of holomorphy, we refer the reader to the books by L. H¨ormander [H¨or 9]
and R. M. Range [Ran 6].
57
58
CHAPTER 4
1
2
THEORY FOR
¯
∂
ON PSEUDOCONVEX DOMAINS
Let 1 be a domain in C
n
. We study the existence of solutions of the Cauchy
Riemann equations
(4.0.1)
¯
∂n = ) in 1,
where ) is a (j, ¡)form and n is a (j, ¡ −1)form on 1, 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n. Since
¯
∂
2
= 0, it is necessary that
(4.0.2)
¯
∂) = 0 in 1
in order for equation (4.0.1) to be solvable.
In this chapter, we prove H¨ormander’s 1
2
existence theorems for the
¯
∂ oper
ator on pseudoconvex domains in C
n
. To study Equation (4.0.1), Hilbert space
techniques are used in the context of the
¯
∂Neumann problem. First, we set up
the
¯
∂Neumann problem with weights and derive the basic a priori estimates of
MorreyKohnH¨ormander. We then choose suitable weight functions in order to
obtain existence theorems with 1
2
estimates.
The 1
2
existence theorems for
¯
∂ also give existence theorems for the
¯
∂Neumann
operator. We will conclude the chapter with a discussion of existence theorems in
other function spaces. The solution of the Levi problem will be given at the end.
4.1 Unbounded Operators in Hilbert Spaces
We shall use Hilbert space techniques to study the
¯
∂ operator. To do this we
need to formulate the
¯
∂ operator as a linear, closed, densely deﬁned operator from
one Hilbert space to another. This will be done in the next section. We ﬁrst discuss
some basic facts for unbounded operators in Hilbert spaces.
Let H
1
and H
2
be two Hilbert spaces and let T : H
1
→ H
2
be a linear, closed,
densely deﬁned operator. We recall that T is closed if and only if the graph of
T is closed. The domain of deﬁnition for T is denoted by Dom(T). If T is an
unbounded operator, Dom(T) is a proper subset of H
1
by the closed graph theorem.
The norms in H
1
, H
2
are denoted by  
1
,  
2
, respectively. Then the adjoint of
T, T
∗
: H
2
→ H
1
is also a linear, closed, densely deﬁned operator and T
∗∗
= T.
(See [RiNa 1].)
59
We use Ker(T) and ¹(T) to denote the kernel and the range of T respectively.
Since T is a closed operator, Ker(T) is closed. Let ¹(T) denote the closure of the
range of T. By the deﬁnition of the adjoint operator, it is easy to see that
(4.1.1) H
1
= Ker(T) ⊕¹(T
∗
)
and
(4.1.2) H
2
= Ker(T
∗
) ⊕¹(T).
In later applications, the operator T will be a system of diﬀerential operators
associated with the CauchyRiemann equations and H
1
, H
2
will be spaces of forms
with 1
2
coeﬃcients. To solve Equation (4.0.1) in the Hilbert space sense is to
show that the range of T is closed. Using (4.1.2), the range of T is then equal to
Ker(T
∗
)
⊥
.
In order to show that the range of T is closed, we use the following lemma for
unbounded operators in Hilbert spaces to reduce the proof to verifying an estimate.
Lemma 4.1.1. Let T : H
1
→ H
2
be a linear, closed, densely deﬁned operator.
The following conditions on T are equivalent:
(1) ¹(T) is closed.
(2) There is a constant C such that
(4.1.3)  ) 
1
≤ C  T) 
2
for all ) ∈ Dom(T) ∩ ¹(T
∗
).
(3) ¹(T
∗
) is closed.
(4) There is a constant C such that
(4.1.4)  ) 
2
≤ C  T
∗
) 
1
for all ) ∈ Dom(T
∗
) ∩ ¹(T).
The best constants in (4.1.3) and (4.1.4) are the same.
Proof. We assume that (1) holds. From (4.1.1),
T : Dom(T) ∩ ¹(T
∗
) →¹(T)
is onetoone, and its inverse
T
−1
: ¹(T) →Dom(T) ∩ ¹(T
∗
)
is welldeﬁned and is also a closed operator. Thus from the closed graph theorem,
T
−1
is continuous and this proves (2). It is obvious that (2) implies (1). Similarly,
(3) and (4) are equivalent.
To prove that (2) implies (4), notice that
[(p, T))
2
[ = [(T
∗
p, ))
1
[ ≤ C  T
∗
p 
1
 T) 
2
,
for p ∈ Dom(T
∗
) and ) ∈ Dom(T) ∩ ¹(T
∗
). Thus
[(p, /)
2
[ ≤ C  T
∗
p 
1
 / 
2
, for p ∈ Dom(T
∗
) and / ∈ ¹(T),
60 L
2
Theory for ∂ on Pseudoconvex Domains
which implies (4). Similarly, (4) implies (2).
4.2 The ∂Neumann Problem
Let 1 be a bounded domain in C
n
, n ≥ 2, not necessarily with a smooth bound
ary. Let C
∞
(p,q)
(1) denote the smooth (j, ¡)forms on 1, where 0 ≤ j ≤ n, 0 ≤ ¡ ≤
n. We use C
∞
(p,q)
(1) to denote the smooth (j, ¡)forms on 1, i.e., the restriction
of smooth (j, ¡)forms in C
n
to 1. Let (.
1
, , .
n
) be the complex coordinates for
C
n
. Then any (j, ¡)form ) ∈ C
∞
(p,q)
(1) can be expressed as
(4.2.0) ) =
I,J
t
)
I,J
d.
I
∧ d¯ .
J
,
where 1 = (i
1
, , i
p
) and J = (,
1
, , ,
q
) are multiindices and d.
I
= d.
i1
∧ ∧
d.
ip
, d¯ .
J
= d¯ .
j1
∧ ∧ d¯ .
jq
. The notation
t
means the summation over strictly
increasing multiindices and the )
I,J
’s are deﬁned for arbitrary 1 and J so that they
are antisymmetric. The operator
¯
∂ =
¯
∂
(p,q)
: C
∞
(p,q)
(1) →C
∞
(p,q+1)
(1)
is deﬁned by
(4.2.1)
¯
∂) =
I,J
t
n
k=1
∂)
I,J
∂¯ .
k
d¯ .
k
∧ d.
I
∧ d¯ .
J
.
Let 1
2
(1) denote the space of square integrable functions on 1 with respect to
the Lebesgue measure in C
n
such that the volume element is d\ = i
n
d.
1
∧ d¯ .
1
∧
∧d.
n
∧d¯ .
n
. This volume element diﬀers from the usual Euclidean measure by a
factor of 2
n
and it is more suitable for our purpose. We use 1
2
(p,q)
(1) to denote the
space of (j, ¡)forms whose coeﬃcients are in 1
2
(1). If ) =
t
I,J
)
I,J
d.
I
∧ d¯ .
J
,
p =
t
I,J
p
I,J
d.
I
∧ d¯ .
J
are two (j, ¡)forms in 1
2
(p,q)
(1), we deﬁne
¸), p) =
I,J
t
¸)
I,J
, p
I,J
), [)[
2
= ¸), )) =
I,J
t
[)
I,J
[
2
,
 ) 
2
=
_
D
¸), ))d\ =
I,J
t
_
D
[)
I,J
[
2
d\.
We use ( , )
D
to denote the inner product in 1
2
(p,q)
(1) and when there is no
danger of confusion, we drop the subscript 1 in the notation. If φ is a continuous
function in 1, then 1
2
(1, φ) is the space of functions in 1 which are square inte
grable with respect to the weight function c
−φ
. The norm in 1
2
(p,q)
(1, φ) is deﬁned
by
)
2
φ
=
_
D
[)[
2
c
−φ
d\, ) ∈ 1
2
(p,q)
(1, φ).
4.2 The ∂Neumann Problem 61
The inner product in 1
2
(p,q)
(1, φ) is denoted by ( , )
φ
. Notice that the space
1
2
(p,q)
(1, φ) is equal to 1
2
(p,q)
(1) if φ is continuous on 1. Let 1
2
(1, loc) denote the
space of locally square integrable functions. A function ) is in 1
2
(1, loc) if and only
if ) is in 1
2
(/) for every compact subset / of 1. 1
2
(p,q)
(1, loc) is deﬁned similarly.
When there is no danger of confusion, we also use 1
2
(1) to denote 1
2
(p,q)
(1).
The formal adjoint of
¯
∂ : C
∞
(p,q−1)
(1) → C
∞
(p,q)
(1), 1 ≤ ¡ ≤ n, under the usual
1
2
norm is denoted by ϑ, where
ϑ = ϑ
(p,q)
: C
∞
(p,q)
(1) →C
∞
(p,q−1)
(1).
The operator ϑ is deﬁned by the requirement that
(4.2.2) (ϑ), p) = (),
¯
∂p)
for all smooth p ∈ C
∞
(p,q−1)
(1) with compact support in 1. If ) is expressed by
(4.2.0) and p =
t
]I]=p,]K]=q−1
p
I,K
d.
I
∧ d¯ .
K
, we have
(),
¯
∂p) = (−1)
p
I,K
t
n
k=1
_
)
I,kK
,
∂p
I,K
∂¯ .
k
_
= (−1)
p+1
I,K
t
n
k=1
_
∂)
I,kK
∂.
k
, p
I,K
_
= (ϑ), p).
Therefore, ϑ can be expressed explicitly by
(4.2.3) ϑ) = (−1)
p−1
I,K
t
n
j=1
∂)
I,jK
∂.
j
d.
I
∧ d¯ .
K
,
where 1 and 1 are multiindices with [1[ = j and [1[ = ¡ − 1. It is easy to check
that
¯
∂
2
= ϑ
2
= 0. Let ϑ
φ
be the formal adjoint of
¯
∂ under the 1
2
(1, φ) norm, i.e.,
(4.2.4) (ϑ
φ
), p)
φ
= (),
¯
∂p)
φ
for every compactly supported p ∈ C
∞
(p,q−1)
(1). We have the following relation
between ϑ and ϑ
φ
: for any ) ∈ C
∞
(p,q)
(1),
(4.2.5) ϑ
φ
) = c
φ
ϑ(c
−φ
)).
Thus
¯
∂, ϑ and ϑ
φ
are systems of ﬁrst order diﬀerential operators.
We take the (weak) 1
2
closure of the unbounded diﬀerential operator
¯
∂, still
denoted by
¯
∂. Let
¯
∂ =
¯
∂
p,q−1
: 1
2
(p,q−1)
(1, φ) →1
2
(p,q)
(1, φ)
be the maximal closure of the CauchyRiemannn operator deﬁned as follows: an
element n ∈ 1
2
(p,q−1)
(1, φ) is in the domain of
¯
∂ if
¯
∂n, deﬁned in the distribution
62 L
2
Theory for ∂ on Pseudoconvex Domains
sense, belongs to 1
2
(p,q)
(1, φ). Then
¯
∂ deﬁnes a linear, closed, densely deﬁned
operator.
¯
∂ is closed since diﬀerentiation is a continuous operation in distribution
theory. It is densely deﬁned since Dom(
¯
∂) contains all the compactly supported
smooth (j, ¡ −1)forms. If 1 is bounded, any ) ∈ C
∞
(p,q−1)
(1) is in Dom(
¯
∂).
The Hilbert space adjoint of
¯
∂, denoted by
¯
∂
∗
φ
, is a linear, closed, densely deﬁned
operator and
¯
∂
∗
φ
: 1
2
(p,q)
(1, φ) →1
2
(p,q−1)
(1, φ).
When φ = 0, we denote the adjoint by
¯
∂
∗
. Let Dom(
¯
∂
∗
) and Dom(
¯
∂
∗
φ
) denote the
domains for
¯
∂
∗
and
¯
∂
∗
φ
, respectively. An element ) belongs to Dom(
¯
∂
∗
φ
) if there
exists a p ∈ 1
2
(p,q−1)
(1, φ) such that for every ψ ∈ Dom(
¯
∂) ∩ 1
2
(p,q−1)
(1, φ), we
have
(),
¯
∂ψ)
φ
= (p, ψ)
φ
.
We then deﬁne
¯
∂
∗
φ
) = p. Note that if ) ∈ Dom(
¯
∂
∗
φ
), then it follows from (4.2.4)
that
¯
∂
∗
φ
) = ϑ
φ
) where ϑ
φ
is deﬁned in the distribution sense in 1.
If 1 is bounded, we have C
∞
(p,q−1)
(1) ⊂ Dom(
¯
∂). However, not every element
in C
∞
(p,q)
(1) is in Dom(
¯
∂
∗
φ
). Any element in Dom(
¯
∂
∗
) (or Dom(
¯
∂
∗
φ
)) must satisfy
certain boundary conditions in the weak sense. If 1 has C
1
boundary /1, then any
) ∈ Dom(
¯
∂
∗
φ
) ∩ C
1
(p,q)
(1) must satisfy the following:
Lemma 4.2.1. Let 1 be a bounded domain with C
1
boundary /1 and ρ be a C
1
deﬁning function for 1. For any ) ∈ Dom(
¯
∂
∗
φ
) ∩ C
1
(p,q)
(1), where φ ∈ C
1
(1), )
must satisfy the boundary condition
(4.2.6) σ(ϑ, dρ))(.) = 0, . ∈ /1,
where σ(ϑ, dρ))(.) = ϑ(ρ))(.) denotes the symbol of ϑ in the dρ direction evaluated
at .. More explicitly, if ) is expressed as in (4.2.0), then ) must satisfy
(4.2.6
t
)
k
)
I,kK
∂ρ
∂.
k
= 0 on /1 for all 1, 1,
where [1[ = j and [1[ = ¡ −1.
Proof. We ﬁrst assume that φ = 0. Note that (4.2.6) and (4.2.6
t
) are independent
of the deﬁning function ρ. We normalize ρ such that [dρ[ = 1 on /1.
Let ) be a (0,1)form and ) =
n
i=1
)
i
d¯ .
i
. Using integration by parts and (4.2.3),
we have for any ψ ∈ C
∞
(1) ⊂ Dom(
¯
∂),
(ϑ), ψ) =
n
i=1
_
−
∂)
i
∂.
i
, ψ
_
=
n
i=1
_
)
i
,
∂ψ
∂¯ .
i
_
−
n
i=1
_
bD
)
i
∂ρ
∂.
i
¯
ψdo
= (),
¯
∂ψ) +
_
bD
¸σ(ϑ, dρ)), ψ)do,
4.2 The ∂Neumann Problem 63
where do is the surface measure of /1. Similarly, for a (j, ¡)form ) and ψ ∈
C
∞
(p,q−1)
(1) ⊂ Dom(
¯
∂), using integration by parts, we obtain
(4.2.7) (ϑ), ψ) = (),
¯
∂ψ) +
_
bD
¸σ(ϑ, dρ)), ψ)do.
If, in addition, ψ has compact support in 1, we have
(
¯
∂
∗
), ψ) = (ϑ), ψ) = (),
¯
∂ψ),
where the ﬁrst equality follows from ) ∈ Dom(
¯
∂
∗
) ∩ C
1
(p,q)
(1). Since compactly
supported smooth (j, ¡ −1)forms are dense in 1
2
(p,q−1)
(1), we must have
_
bD
¸σ(ϑ, dρ)), ψ)do = 0, for any ψ ∈ C
∞
(p,q−1)
(1).
This implies that σ(ϑ, dρ))(.) = 0 for . ∈ /1.
If ) is expressed by (4.2.0), one can easily show that (4.2.6) implies that (4.2.6
t
)
holds on /1 for each 1, 1. The case for φ ,= 0 can be proved similarly and is left
to the reader. This proves the lemma.
Another way to express condition (4.2.6) or (4.2.6
t
) is as follows. Let ∨ be the
interior product deﬁned as the dual of the wedge product. For any (j, ¡)form ),
¯
∂ρ ∨ ) is deﬁned as the (j, ¡ −1)form satisfying
¸p ∧
¯
∂ρ, )) = ¸p,
¯
∂ρ ∨ )), p ∈ C
∞
(p,q−1)
(C
n
).
Using this notation, condition (4.2.6) or (4.2.6
t
) can be expressed as
(4.2.6
tt
)
¯
∂ρ ∨ ) = 0 on /1.
It is also easy to see that ) ∈ C
1
(p,q)
(1) ∩ Dom(
¯
∂
∗
φ
) if and only if ) satisﬁes one of
the three equivalent conditions (4.2.6), (4.2.6
t
) or (4.2.6
tt
).
For a ﬁxed 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n, we deﬁne the Laplacian of the
¯
∂ complex
1
2
(p,q−1)
(1)
¯
∂(p,q−1)
−−−−−→
←−−−−−−
¯
∂
∗
(p,q)
1
2
(p,q)
(1)
¯
∂
(p,q)
−−−−−−→
←−−−−−−
¯
∂
∗
(p,q+1)
1
2
(p,q+1)
(1).
Deﬁnition 4.2.2. Let
(p,q)
=
¯
∂
(p,q−1)
¯
∂
∗
(p,q)
+
¯
∂
∗
(p,q+1)
¯
∂
(p,q)
be the operator from
1
2
(p,q)
(1) to 1
2
(p,q)
(1) such that Dom(
(p,q)
) = ¦) ∈ 1
2
(p,q)
(1) [ ) ∈ Dom(
¯
∂
(p,q)
) ∩
Dom(
¯
∂
∗
(p,q)
);
¯
∂
(p,q)
) ∈ Dom(
¯
∂
∗
(p,q+1)
) and
¯
∂
∗
(p,q)
) ∈ Dom(
¯
∂
(p,q−1)
)¦.
Proposition 4.2.3.
(p,q)
is a linear, closed, densely deﬁned selfadjoint operator.
Proof.
(p,q)
is densely deﬁned since Dom(
(p,q)
) contains all smooth forms with
compact support. To show that
(p,q)
is closed, one needs to prove that for every
64 L
2
Theory for ∂ on Pseudoconvex Domains
sequence )
n
∈ Dom(
(p,q)
) such that )
n
→ ) in 1
2
(p,q)
(1) and
(p,q)
)
n
converges,
we have ) ∈ Dom(
(p,q)
) and
(p,q)
)
n
→
(p,q)
). Since )
n
∈ Dom(
(p,q)
),
(
(p,q)
)
n
, )
n
) = (
¯
∂
¯
∂
∗
)
n
, )
n
) + (
¯
∂
∗
¯
∂)
n
, )
n
)
= 
¯
∂
∗
)
n

2
+
¯
∂)
n

2
,
thus
¯
∂
∗
)
n
and
¯
∂)
n
converge in 1
2
(p,q−1)
(1) and 1
2
(p,q+1)
(1), respectively. Since
¯
∂
and
¯
∂
∗
are closed operators, we have ) ∈ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
) and
¯
∂)
n
→
¯
∂) and
¯
∂
∗
)
n
→
¯
∂
∗
) in 1
2
.
To show that
¯
∂) ∈ Dom(
¯
∂
∗
) and
¯
∂
∗
) ∈ Dom(
¯
∂), we note that since
(p,q)
)
n
=
¯
∂
¯
∂
∗
)
n
+
¯
∂
∗ ¯
∂)
n
converges, both
¯
∂
¯
∂
∗
)
n
and
¯
∂
∗ ¯
∂)
n
converge. This follows from the
fact that
¯
∂
¯
∂
∗
)
n
and
¯
∂
∗ ¯
∂)
n
are orthogonal to each other since
(
¯
∂
¯
∂
∗
)
n
,
¯
∂
∗
¯
∂)
n
) = (
¯
∂
2
¯
∂
∗
)
n
,
¯
∂)
n
) = 0.
It follows again from the fact that
¯
∂ and
¯
∂
∗
are closed operators that
¯
∂
¯
∂
∗
)
n
→
¯
∂
¯
∂
∗
) and
¯
∂
∗
¯
∂)
n
→
¯
∂
∗
¯
∂).
Therefore, we have proved that
(p,q)
)
n
→
(p,q)
) and
(p,q)
is a closed operator.
Let
∗
(p,q)
be the Hilbert space adjoint of
(p,q)
. It is easy to see that
(p,q)
=
∗
(p,q)
on Dom(
(p,q)
)∩Dom(
∗
(p,q)
). To show that Dom(
(p,q)
)=Dom(
∗
(p,q)
), de
ﬁne
1
1
=
¯
∂
¯
∂
∗
+
¯
∂
∗
¯
∂ +1 =
(p,q)
+1 on Dom(
(p,q)
).
We shall prove that 1
−1
1
is selfadjoint. By a theorem of Von Neumann [RiNa 1],
(1 +
¯
∂
¯
∂
∗
)
−1
and (1 +
¯
∂
∗
¯
∂)
−1
are bounded selfadjoint operators. We deﬁne
Q
1
= (1 +
¯
∂
¯
∂
∗
)
−1
+ (1 +
¯
∂
∗
¯
∂)
−1
−1.
Then Q
1
is bounded and selfadjoint. We claim that Q
1
= 1
−1
1
. Since
(1 +
¯
∂
¯
∂
∗
)
−1
−1 = (1 −(1 +
¯
∂
¯
∂
∗
))(1 +
¯
∂
¯
∂
∗
)
−1
= −
¯
∂
¯
∂
∗
(1 +
¯
∂
¯
∂
∗
)
−1
,
we have that ¹(1 +
¯
∂
¯
∂
∗
)
−1
⊂ Dom(
¯
∂
¯
∂
∗
). Similarly, we have ¹(1 +
¯
∂
∗ ¯
∂)
−1
⊂
Dom(
¯
∂
∗ ¯
∂) and
Q
1
= (1 +
¯
∂
∗
¯
∂)
−1
−
¯
∂
¯
∂
∗
(1 +
¯
∂
¯
∂
∗
)
−1
.
Since
¯
∂
2
= 0, we have ¹(Q
1
) ⊂ Dom(
¯
∂
∗ ¯
∂) and
¯
∂
∗
¯
∂Q
1
=
¯
∂
∗
¯
∂(1 +
¯
∂
∗
¯
∂)
−1
.
Similarly, we have ¹(Q
1
) ⊂ Dom(
¯
∂
¯
∂
∗
) and
¯
∂
¯
∂
∗
Q
1
=
¯
∂
¯
∂
∗
(1 +
¯
∂
¯
∂
∗
)
−1
.
Thus, ¹(Q
1
) ⊂ Dom(1
1
) and
1
1
Q
1
=
¯
∂
¯
∂
∗
(1 +
¯
∂
¯
∂
∗
)
−1
+
¯
∂
∗
¯
∂(1 +
¯
∂
∗
¯
∂)
−1
+Q
1
= 1.
Since 1
1
is injective, we have that Q
1
= 1
−1
1
. This proves that 1
1
is selfadjoint
which implies
(p,q)
= 1
1
−1 is selfadjoint. The proposition is proved.
The following proposition shows that smooth forms in Dom(
(p,q)
) must satisfy
two sets of boundary conditions, namely, the ∂Neumann boundary conditions.
4.2 The ∂Neumann Problem 65
Proposition 4.2.4. Let 1 be a bounded domain with C
1
boundary and ρ be a C
1
deﬁning function. If ) ∈ C
2
(p,q)
(1), then
) ∈ Dom(
(p,q)
)
if and only if
σ(ϑ, dρ)) = 0 and σ(ϑ, dρ)∂) = 0 on /1.
If ) =
t
I,J
)
I,J
d.
I
∧ d¯ .
J
∈ C
2
(p,q)
(1) ∩ Dom(
(p,q)
), we have
(4.2.8)
(p,q)
) = −
1
4
I,J
t
´)
I,J
d.
I
∧ d¯ .
J
,
where ´ = 4
n
k=1
∂
2
,∂.
k
∂¯ .
k
=
n
k=1
(∂
2
,∂r
2
k
+∂
2
,∂j
2
k
) is the usual Laplacian on
functions.
Proof. If ) ∈ C
2
(p,q)
(1) ∩ Dom(
(p,q)
), then ) ∈ Dom(∂
∗
) and ∂) ∈ Dom(∂
∗
).
Thus from the same arguments as in Lemma 4.2.1, ) must satisfy σ(ϑ, dρ)) =
σ(ϑ, dρ)∂) = 0 on /1. Conversely, if σ(ϑ, dρ)) = σ(ϑ, dρ)∂) = 0, then ) ∈
Dom(∂
∗
) and ∂) ∈ Dom(∂
∗
) from integration by parts. Also, it is easy to see that
) and
¯
∂
∗
) = ϑ) are in Dom(
¯
∂). Thus ) ∈ Dom(
(p,q)
).
If ) ∈ C
2
(p,q)
(1) ∩ Dom(
(p,q)
), we have
(p,q)
) = (
¯
∂ϑ +ϑ
¯
∂)).
A direct calculation, using (4.2.1) and (4.2.3), gives us that
(4.2.9)
ϑ
¯
∂) = −
I,J
t
k
∂
2
)
I,J
∂.
k
∂¯ .
k
d.
I
∧ d¯ .
J
+ (−1)
p
I,K
t
k
j
∂
2
)
I,jK
∂.
j
∂¯ .
k
d¯ .
k
∧ d.
I
∧ d¯ .
K
,
and
(4.2.10)
¯
∂ϑ) = (−1)
p−1
I,K
t
j
k
∂)
I,jK
∂¯ .
k
∂.
j
d¯ .
k
∧ d.
I
∧ d¯ .
K
.
Adding (4.2.9) and (4.2.10), we get that
(p,q)
) = −
I,J
t
k
∂
2
)
I,J
∂.
k
∂¯ .
k
d.
I
∧ d¯ .
J
= −
1
4
I,J
t
´)
I,J
d.
I
∧ d¯ .
J
.
66 L
2
Theory for ∂ on Pseudoconvex Domains
Example. Let 1 be a smooth bounded domain in C
n
with the origin 0 ∈ /1. We
assume that for some neighborhood l of 0
1 ∩ l = ¦Im .
n
= j
n
< 0¦ ∩ l.
Let ) =
k
)
k
d¯ .
k
∈ C
2
(0,1)
(1) and the support of ) lies in l ∩ 1. Then ) is in
Dom(
(0,1)
) if and only if ) satisﬁes
)
n
= 0 on /1 ∩ l, (a)
∂)
i
∂¯ .
n
= 0 on /1 ∩ l, i = 1, , n −1. (b)
Proof. (a) follows from the condition that ) ∈ Dom(
¯
∂
∗
). To see that (b) holds, we
note that
¯
∂) ∈ Dom(
¯
∂
∗
), implying
∂)
i
∂¯ .
n
−
∂)
n
∂¯ .
i
= 0 on /1 ∩ l.
From (a), we have ∂)
n
,∂¯ .
i
= 0 on /1 ∩ l for i = 1, , n − 1 since each ∂,∂¯ .
i
is
tangential. This proves (b).
We note that the ﬁrst boundary condition (a) is just the Dirichlet boundary value
problem. The second condition (b) is the complex normal derivative ∂,∂¯ .
n
on each
)
i
instead of the usual normal derivative ∂,∂j
n
. It is the second boundary condi
tion which makes the system noncoercive, i.e., it is not an elliptic boundary value
problem (One can check easily that it does not satisfy the Lopatinski’s conditions,
see e.g., Treves [Tre 1]).
There are two objectives to the study of the
¯
∂Neumann problem: one is to show
that the range of
(p,q)
is closed in 1
2
and that there exists a bounded inverse of
the operator
(p,q)
on any bounded pseudoconvex domain; the other is to study the
regularity of the solution of
(p,q)
up to the boundary. In the next sections we shall
prove 1
2
existence theorems for
¯
∂ and
(p,q)
. We discuss the boundary regularity
for the solution of
(p,q)
in Chapters 5 and 6.
4.3 1
2
Existence Theorems for ∂ in Pseudoconvex Domains
In this section we prove the 1
2
estimates and existence theorems for the
¯
∂ oper
ator with precise bounds on any bounded pseudoconvex domains.
Let 1 be a domain with C
2
boundary /1. Let ρ be a C
2
deﬁning function in
a neighborhood of 1 such that 1 = ¦. [ ρ(.) < 0¦ and [dρ[ = 1 on /1. For each
/ ∈ N, we set
T
(p,q)
= Dom(
¯
∂
∗
) ∩ C
(p,q)
(1)
and
T
(p,q)
= Dom(
¯
∂
∗
) ∩ C
∞
(p,q)
(1).
Let φ ∈ C
2
(1) be a ﬁxed function. Let
T
φ
(p,q)
= Dom(
¯
∂
∗
φ
) ∩ C
∞
(p,q)
(1).
4.3 L
2
Existence Theorems for ∂ in Pseudoconvex Domains 67
It is easy to see from the arguments in the proof of Lemma 4.2.1 that ) ∈ T
φ
(p,q)
if
and only if σ(ϑ, dρ))(.) = 0 for any . ∈ /1, a condition independent of φ. Thus
we have
T
φ
(p,q)
= T
(p,q)
,
which is also independent of φ. Similarly, we also have
Dom(
¯
∂
∗
φ
) ∩ C
(p,q)
(1) = T
(p,q)
.
Let Q
φ
be the form on T
(p,q)
deﬁned by
Q
φ
(), )) = 
¯
∂)
2
φ
+
¯
∂
∗
φ
)
2
φ
.
We shall ﬁrst prove the following basic a priori identity:
Proposition 4.3.1 (MorreyKohnH¨ormander). Let 1 ⊂⊂ C
n
be a domain
with C
2
boundary /1 and ρ be a C
2
deﬁning function for 1 such that [dρ[ = 1 on
/1. Let φ ∈ C
2
(1). For any ) =
t
]I]=p,]J]=q
)
I,J
d.
I
∧d¯ .
J
∈ T
1
(p,q)
,
(4.3.1)
Q
φ
(), )) = 
¯
∂) 
2
φ
+  ϑ
φ
) 
2
φ
=
t
]I]=p,]K]=q−1
i,j
_
D
∂
2
φ
∂.
i
∂¯ .
j
)
I,iK
¯
)
I,jK
c
−φ
d\
+
t
]I]=p,]J]=q
k
_
D
¸
¸
¸
¸
∂)
I,J
∂¯ .
k
¸
¸
¸
¸
2
c
−φ
d\
+
t
]I]=p,]K]=q−1
i,j
_
bD
∂
2
ρ
∂.
i
∂¯ .
j
)
I,iK
¯
)
I,jK
c
−φ
do.
Proof. From (4.2.1), (4.2.3) and (4.2.5), we have
(4.3.2) ∂) =
I,J
t
j
∂)
I,J
∂.
j
d¯ .
j
∧ d.
I
∧ d.
J
and
(4.3.3) ϑ
φ
) = (−1)
p−1
I,K
t
j
δ
φ
j
)
I,jK
d.
I
∧ d.
K
,
where δ
φ
j
n = c
φ
∂
∂.
j
(c
−φ
n). Thus, setting 1
j
= ∂,∂¯ .
j
, we get
(4.3.4)
∂)
2
φ
+ ϑ
φ
)
2
φ
=
I,J,L
t
j,
c
jJ
L
(1
j
()
I,J
), 1
()
I,L
))
φ
+
I,K
t
j,k
(δ
φ
j
)
I,jK
, δ
φ
k
)
I,kK
)
φ
,
68 L
2
Theory for ∂ on Pseudoconvex Domains
where c
jJ
L
= 0, unless , , ∈ J, / , ∈ 1 and ¦,¦ ∪ J = ¦/¦ ∪ 1, in which case c
jJ
L
is the
sign of permutation
_
jJ
L
_
. Rearranging the terms in (4.3.4) gives
(4.3.5)
∂)
2
φ
+ ϑ
φ
)
2
φ
=
I,J
t
j
1
j
)
I,J

2
φ
−
I,K
t
j,k
(1
k
)
I,jK
, 1
j
)
I,kK
)
φ
+
I,K
t
j,k
(δ
φ
j
)
I,jK
, δ
φ
k
)
I,kK
)
φ
.
We now apply integration by parts to the last term in (4.3.4). Note that for each
n, · ∈ C
2
(1),
(n, δ
φ
j
·)
φ
= −(1
j
n, ·)
φ
+
_
bD
∂ρ
∂¯ .
j
n·c
−φ
do
and
[δ
φ
j
, 1
k
]n = δ
φ
j
1
k
n −1
k
δ
φ
j
n = n
∂
2
φ
∂.
j
∂.
k
.
Thus, we ﬁnd that
(4.3.6)
(δ
φ
j
n, δ
φ
k
·)
φ
= (−1
k
δ
φ
j
n, ·)
φ
+
_
bD
∂ρ
∂.
k
(δ
φ
j
n)·c
−φ
do
= (−δ
φ
j
1
k
n, ·)
φ
+ ([δ
φ
j
, 1
k
]n, ·)
φ
+
_
bD
∂ρ
∂.
k
(δ
φ
j
n)·c
−φ
do
= (1
k
n, 1
j
·)
φ
+ (
∂
2
φ
∂.
j
∂¯ .
k
n, ·)
φ
+
_
bD
∂ρ
∂¯ .
k
(δ
φ
j
n)·c
−φ
do −
_
bD
∂ρ
∂.
j
(1
k
n)·c
−φ
do.
When n, · are in C
1
(1), (4.3.6) also holds by approximation since C
2
(1) is a dense
subset in C
1
(1). Using (4.3.6) for each ﬁxed 1, 1, it follows that
(4.3.7)
j,k
(δ
φ
j
)
I,jK
, δ
φ
k
)
I,kK
)
φ
=
j,k
(1
k
)
I,jK
, 1
j
)
I,kK
)
φ
+
j,k
(
∂
2
φ
∂.
j
∂¯ .
k
)
I,jK
, )
I,kK
)
φ
+
j,k
_
bD
∂ρ
∂¯ .
k
(δ
φ
j
)
I,jK
)
¯
)
I,kK
c
−φ
do
−
j,k
_
bD
∂ρ
∂.
j
_
∂
∂¯ .
k
)
I,jK
_
¯
)
I,kK
c
−φ
do.
If ) ∈ T
1
(p,q)
, Lemma 4.2.1 and (4.2.6
t
) show that
(4.3.8)
k
∂ρ
∂¯ .
k
¯
)
I,kK
= 0 on /1
4.3 L
2
Existence Theorems for ∂ in Pseudoconvex Domains 69
for each 1, 1. Since
k
¯
)
I,kK
∂
∂z
k
is tangential to /1, we conclude from (4.3.8)
that
k
¯
)
I,kK
∂
∂¯ .
k
_
_
j
∂ρ
∂.
j
)
I,jK
_
_
= 0 on /1 for each 1, 1.
This implies
(4.3.9)
k
j
¯
)
I,kK
∂ρ
∂.
j
∂)
I,jK
∂¯ .
k
+
k
j
)
I,jK
¯
)
I,kK
∂
2
ρ
∂.
j
∂¯ .
k
= 0
on /1. Combining (4.3.5)(4.3.9), we have proved (4.3.1) and the proposition.
In order to pass from a priori estimates (4.3.1) to the real estimates, the following
density lemma is crucial:
Lemma 4.3.2 (A density lemma). Let 1 be a bounded domain with C
+1
bound
ary /1, / ≥ 1 and φ ∈ C
2
(1). Then T
(p,q)
is dense in Dom(
¯
∂) ∩ Dom(
¯
∂
∗
φ
) in the
graph norm
) →)
φ
+
¯
∂)
φ
+
¯
∂
∗
φ
)
φ
.
Proof. The proof is essentially a variation of Friedrichs’ lemma (see Appendix D).
We divide the proof of the lemma into three steps.
(i). C
∞
(p,q)
(1) is dense in Dom(
¯
∂) ∩ Dom(
¯
∂
∗
φ
) in the graph norm.
By this we mean that if ) ∈ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
φ
), one can construct a sequence
)
n
∈ C
∞
(p,q)
(1) such that )
n
→),
¯
∂)
n
→
¯
∂) and ϑ
φ
)
n
→ϑ
φ
) in 1
2
(1, φ). We ﬁrst
show that this can be done on a compact subset in 1 from the usual regularization
by convolution.
Let χ ∈ C
∞
0
(C
n
) be a function such that χ ≥ 0,
_
χd\ = 1, χ(.) depends only on
[.[ and vanishes when [.[ ≥ 1. We deﬁne χ
ε
(.) = ε
−2n
χ(.,ε) for ε 0. Extending
) to be 0 outside 1, we deﬁne for ε 0 and . ∈ C
n
,
)
ε
(.) = ) ∗ χ
ε
(.)
=
_
)(.
t
)χ
ε
(. −.
t
)d\ (.
t
) =
_
)(. −ε.
t
)χ(.
t
)d\ (.
t
),
where the convolution is performed on each component of ). In the ﬁrst integral
deﬁning )
ε
, we can diﬀerentiate under the integral sign to show that )
ε
is C
∞
(C
n
).
From Young’s inequality for convolution, we have
 )
ε
 ≤  )  .
Since )
ε
→) uniformly when ) ∈ C
∞
0
(C
n
), a dense subset of 1
2
(C
n
), we have that
)
ε
→) in 1
2
(C
n
) for every ) ∈ 1
2
(C
n
).
Obviously, this implies that )
ε
→) in 1
2
(1, φ).
70 L
2
Theory for ∂ on Pseudoconvex Domains
Let δ
ν
be a sequence of small numbers with δ
ν
¸ 0. For each δ
ν
, we deﬁne
1
δν
= ¦. ∈ 1[ ρ(.) < −δ
ν
¦. Then 1
δν
is a sequence of relatively compact open
subsets of 1 with union equal to 1. Using similar arguments as before, for any
ﬁrst order diﬀerential operator 1
i
with constant coeﬃcients, if 1
i
) ∈ 1
2
(1, φ), we
have
1
i
)
ε
= 1
i
() ∗ χ
ε
) = 1
i
) ∗ χ
ε
→1
i
) in 1
2
(1
δν
, φ)
as ε →0. Since ϑ
φ
= ϑ+¹
0
where ¹
0
is an operator of degree 0, we have
¯
∂)
ε
→
¯
∂)
and ϑ
φ
)
ε
→ϑ
φ
) in 1
2
(1
δν
, φ) on 1
δν
, where )
ε
∈ C
∞
(p,q)
(1
δν
).
To see that this can be done up to the boundary, we ﬁrst assume that the
domain 1 is starshaped and 0 ∈ 1 is a center. We approximate ) ﬁrst by dilation
componentwise. Let 1
= ¦(1 +c). [ . ∈ 1¦ and
)
= )
_
.
1 +c
_
,
where the dilation is performed for each component of ). Then 1 ⊂⊂ 1
and
)
∈ 1
2
(1
). Also
¯
∂)
→
¯
∂) ∈ 1
2
(1) and ϑ
φ
)
→ ϑ
φ
) ∈ 1
2
(1). By regularizing
)
componentwise as before, we can ﬁnd a family of )
()
∈ C
∞
(p,q)
(1) deﬁned by
(4.3.10) )
()
= )
_
.
1 +c
_
∗ χ
δ
,
where δ
¸ 0 as c ¸ 0 and δ
is chosen suﬃciently small. We have )
()
→ ) in
1
2
(1, φ),
¯
∂)
()
→
¯
∂) and ϑ
φ
)
()
→ϑ
φ
) in 1
2
(1, φ). Thus, C
∞
(p,q)
(1) is dense in the
graph norm when 1 is starshaped. The general case follows by using a partition of
unity since we assume our domain has at least C
2
boundary. (In fact, C
1
boundary
will suﬃce in this step).
(ii). Compactly supported smooth forms (i.e., forms with coeﬃcients in C
∞
0
(1))
are dense in Dom(
¯
∂
∗
φ
) in the graph norm
) →)
φ
+
¯
∂
∗
φ
)
φ
.
We ﬁrst assume that φ = 0. Since
¯
∂ is the maximal closure (i.e., the domain of
¯
∂
contains all elements in C
∞
(p,q)
(1)) of the CauchyRiemann operator, its 1
2
adjoint,
¯
∂
∗
, is minimal. This means that if ) ∈ Dom(
¯
∂
∗
) and we extend ) to
˜
) on the whole
space C
n
by setting
˜
) to be zero in 1
c
, then ϑ
˜
) ∈ 1
2
(C
n
) in the distribution sense.
In fact, for ) ∈ Dom(
¯
∂
∗
), we have
ϑ
˜
) =
ϑ)
where
ϑ) = ϑ) in 1 and
ϑ) = 0 in 1
c
. This can be checked from the deﬁnition of
¯
∂
∗
, since for any · ∈ C
∞
(p,q−1)
(C
n
),
(
˜
),
¯
∂·)
C
n = (),
¯
∂·)
D
= (
¯
∂
∗
), ·)
D
= (ϑ), ·)
D
= (
ϑ), ·)
C
n.
4.3 L
2
Existence Theorems for ∂ in Pseudoconvex Domains 71
Again, we can assume that 1 is starshaped and 0 is a center. The general case can
be proved using a partition of unity. We ﬁrst approximate
˜
) by
˜
)
−
=
˜
)
_
.
1 −c
_
.
Then
˜
)
−
has compact support in 1 and ϑ
˜
)
−
→ϑ
˜
) in 1
2
(C
n
). Regularizing
˜
)
−
by convolution as before, we deﬁne
(4.3.11) )
(−)
=
˜
)
_
.
1 −c
_
∗ χ
δ
.
Then the )
(−)
are (j, ¡)forms with coeﬃcients in C
∞
0
(1) such that )
(−)
→ ) in
1
2
(1) and ϑ)
(−)
→ϑ) in 1
2
(1). This proves (ii) when φ = 0. Again, in this step
we only require that the boundary be C
1
. The case for
¯
∂
∗
φ
can be proved similarly.
However, compactly supported smooth forms are not dense in Dom (
¯
∂) in the
graph norm ) →)
φ
+
¯
∂)
φ
. Nevertheless, we have:
(iii). T
(p,q)
is dense in Dom(
¯
∂) in the graph norm
) →)
φ
+
¯
∂)
φ
.
To prove (iii), we must use Friedrichs’ lemma in a more subtle way. From (i), it
suﬃces to show that for any ) ∈ C
∞
(p,q)
(1) that one can ﬁnd a sequence )
n
∈ T
(p,q)
such that )
n
→ ) in 1
2
(1, φ) and
¯
∂)
n
→
¯
∂) in 1
2
(1, φ). We may assume φ = 0
and the general case is similar.
We regularize near a boundary point .
0
∈ /1. Let l be a small neighborhood of
.
0
. By a partition of unity, we may assume that 1∩l is starshaped and ) is sup
ported in l∩1. Let ρ be a C
+1
deﬁning function such that [dρ[ = 1 on /1. Shrink
ing l if necessary, we can choose a special boundary chart (t
1
, t
2
, , t
2n−1
, ρ)
where (t
1
, t
2
, , t
2n−1
), when restricted to /1, forms a coordinate system on /1.
Let ¯ n
1
, , ¯ n
n
be an orthonormal basis for (0,1)forms on l such that
¯
∂ρ = ¯ n
n
.
Written in this basis,
) =
t
]I]=p,]J]=q
)
I,J
n
I
∧ ¯ n
J
,
where 1 = (i
1
, , i
p
) and J = (,
1
, , ,
q
) are increasing multiindices, and n
I
=
n
i1
∧ ∧n
ip
, ¯ n
J
= ¯ n
j1
∧ ∧ ¯ n
jq
. Each )
I,J
is a function in C
(1). We note that
both
¯
∂ and ϑ are ﬁrst order diﬀerential operators with variable coeﬃcients which
are in C
(1) when computed in the special frame n
1
, , n
n
. We write
) = )
τ
+)
ν
,
where
)
τ
=
t
]I]=p,]J]=q, n/ ∈J
)
I,J
n
I
∧ ¯ n
J
,
)
ν
=
t
]I]=p,]J]=q, n∈J
)
I,J
n
I
∧ ¯ n
J
.
72 L
2
Theory for ∂ on Pseudoconvex Domains
)
τ
is the complex tangential part of ), and )
ν
is the complex normal part of ).
From Lemma 4.2.1 and (4.2.6
tt
), it follows that
) ∈ T
(p,q)
if and only if )
ν
= 0 on /1.
We also observe that from integration by parts, for ) ∈ C
∞
(p,q)
(1), p ∈ C
∞
(p,q+1)
(1),
(4.3.12) (
¯
∂), p) = (), ϑp) +
_
bD
¸σ(
¯
∂, dρ)), p)do,
where do is the surface measure of /1 and
σ(
¯
∂, dρ)) =
¯
∂(ρ)) =
¯
∂ρ ∧ ) =
¯
∂ρ ∧ )
τ
on /1.
Thus, when we do integration by parts for
¯
∂), only the tangential part )
τ
will
appear in the boundary term. This is called the Cauchy data of ) with respect to
the operator
¯
∂. The Cauchy data of ) with respect to
¯
∂ contains the tangential
part of ), and it does not contain the complex normal part )
ν
(From (4.2.7), it is
easy to see that the Cauchy data of ) for ϑ is the complex normal part )
ν
).
We regularize only the complex normal part of ) and leave the complex tangential
part )
τ
unchanged. Let
˜
)
ν
be the extension of )
ν
to C
n
by setting
˜
)
ν
equal to zero
outside 1. We approximate
˜
)
ν
by the dilation and regularization by convolution
as in (4.3.11),
)
ν
(−)
=
˜
)
ν
_
.
1 −c
_
∗ χ
δ
.
Thus, )
ν
(−)
is smooth and supported in a compact subset in 1 ∩ l. By this, we
approximate )
ν
by )
ν
(−)
∈ C
∞
0
(1∩l) in the 1
2
norm. Furthermore, by extending
¯
∂)
ν
to be zero outside 1 ∩ l and denoting the extension by
¯
¯
∂)
ν
, we have
¯
∂
˜
)
ν
=
¯
¯
∂)
ν
in 1
2
(C
n
)
in the distribution sense. This follows from (4.3.12) since )
ν
∈ C
(1) and for any
p ∈ C
∞
(p,q+1)
(C
n
),
(
˜
)
ν
, ϑp)
C
n = (
¯
∂)
ν
, p)
D
−
_
bD
¸
¯
∂ρ ∧ )
ν
, p)do = (
¯
¯
∂)
ν
, p)
C
n.
Since
¯
∂ is a ﬁrst order diﬀerential operator with variable coeﬃcients, using the
arguments for (ii), but now applying Friedrichs’ lemma (see Appendix D), we have
(4.3.13)
¯
∂)
ν
(−)
→
¯
∂
˜
)
ν
in 1
2
(C
n
).
We set
)
(−)
= )
τ
+)
ν
(−)
.
It follows )
(−)
∈ T
(p,q)
since each coeﬃcient and n
i
is in C
(1 ∩ l). Also we see
that
)
(−)
∈ T
(p,q)
and )
(−)
→) in 1
2
(1).
4.3 L
2
Existence Theorems for ∂ in Pseudoconvex Domains 73
To see that
¯
∂)
(−)
→
¯
∂) in the 1
2
(1) norm, using (4.3.13), we ﬁnd that
¯
∂)
(−)
=
¯
∂)
τ
+
¯
∂)
ν
(−)
→
¯
∂) in 1
2
(1) as c →0.
Thus, T
(p,q)
is dense in Dom(
¯
∂) in the graph norm ) →)
φ
+
¯
∂)
φ
. This proves
(iii).
To ﬁnish the proof of the lemma, we assume that φ = 0. For any ) ∈ Dom(
¯
∂) ∩
Dom(
¯
∂
∗
), we use a partition of unity and the same notation as before to regularize
) in each small starshaped neighborhood near the boundary. We regularize the
complex tangential and normal part separately by setting
˜
)
()
= )
τ
()
+)
ν
(−)
,
where )
τ
()
is the regularization deﬁned by (4.3.10) for each coeﬃcient in the com
plex tangent component and )
ν
(−)
is the regularization deﬁned by (4.3.11) for each
coeﬃcient in the complex normal component. It follows that for suﬃciently small
c 0, )
ν
(−)
has coeﬃcients in C
∞
0
(1) and )
τ
()
has coeﬃcients in C
∞
(1). Thus we
see that
˜
)
()
∈ T
(p,q)
,
˜
)
()
→) in 1
2
(1).
It follows from steps (i), (iii) and Friedrichs’s lemma that
¯
∂
˜
)
()
→
¯
∂) in 1
2
(1).
Also, from steps (i) and (ii), it follows that
ϑ
˜
)
()
→ϑ
˜
) in 1
2
(C
n
),
where
˜
) is the extension of ) to be zero outside 1. This shows that T
(p,q)
is dense
in Dom(
¯
∂)∩Dom(
¯
∂
∗
) in the graph norm ) →)+
¯
∂)+
¯
∂
∗
). Thus, the lemma
is proved for φ = 0. For φ ,= 0 the proof is similar and the density lemma is proved.
Proposition 4.3.3. Let 1 be a bounded pseudoconvex domain in C
n
with C
2
boundary and φ ∈ C
2
(1). We have for every ) ∈ 1o:(∂) ∩ 1o:(∂
∗
φ
),
(4.3.14)
I,K
t
j,k
_
D
∂
2
φ
∂.
j
∂.
k
)
I,jK
¯
)
I,kK
c
−φ
d\ ≤  ∂) 
2
φ
+  ∂
∗
φ
) 
2
φ
.
Proof. From the assumption that 1 is pseudoconvex and has C
2
boundary, we have
for any ) ∈ T
1
(p,q)
,
i,j
_
bD
∂
2
ρ
∂.
i
∂¯ .
j
)
I,iK
¯
)
I,jK
c
−φ
do ≥ 0,
since ) satisﬁes (4.2.6
t
). The proposition follows directly from Proposition 4.3.1
and Lemma 4.3.2.
74 L
2
Theory for ∂ on Pseudoconvex Domains
Theorem 4.3.4 (1
2
existence theorems for
¯
∂). Let 1 be a bounded pseudo
convex domain in C
n
. For every ) ∈ 1
2
(p,q)
(1), where 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n with
∂) = 0, one can ﬁnd n ∈ 1
2
(p,q−1)
(1) such that ∂n = ) and
(4.3.15) ¡
_
D
[ n [
2
d\ ≤ cδ
2
_
D
[ ) [
2
d\,
where δ = sup
z,z
∈D
[. −.
t
[ is the diameter of 1.
Proof. We ﬁrst prove the theorem for 1 with C
2
boundary. Without loss of gen
erality, we may assume that 0 ∈ 1. We shall choose φ = t[.[
2
for some positive
number t. From Proposition 4.3.3, we have for any p ∈ Dom(∂) ∩ Dom(∂
∗
φ
),
(4.3.16) t¡
_
D
[ p [
2
c
−t]z]
2
d\ ≤  ∂p 
2
φ
+  ∂
∗
φ
p 
2
φ
.
Since
¯
∂
2
= 0, we have
(4.3.17) ¹(
¯
∂) ⊂ Ker(
¯
∂) and ¹(∂
∗
φ
) ⊂ Ker(∂
∗
φ
).
It follows from (4.3.16) that for any p ∈ Dom(∂
∗
φ
) ∩ Ker(∂),
(4.3.18) t¡
_
D
[ p [
2
c
−t]z]
2
d\ ≤  ∂
∗
φ
p 
2
φ
.
Using Lemma 4.1.1, we have that ¹(
¯
∂) is closed in 1
2
(p,q)
(1, φ). To show that
(4.3.19) ¹(
¯
∂) = Ker(
¯
∂),
we claim that for any ) ∈ 1
2
(p,q)
(1) with ∂) = 0, there exists a constant C 0
such that
(4.3.20) [ (), p)
φ
[ ≤ C  ∂
∗
φ
p 
φ
, for all p ∈ Dom(∂
∗
φ
).
Using Lemma 4.1.1, ¹(∂
∗
φ
) is also closed. From (4.1.1), we have
1
2
(p,q)
(1, φ) = Ker(
¯
∂) ⊕Ker(
¯
∂)
⊥
= Ker(
¯
∂) ⊕¹(∂
∗
φ
).
For any p
1
∈ Dom(∂
∗
φ
) ∩ Ker(∂), using (4.3.18),
[ (), p
1
)
φ
[ ≤  ) 
φ
 p
1

φ
≤
1
√
t¡
 ) 
φ
 ∂
∗
φ
p
1

φ
.
If p
2
∈ Dom(∂
∗
φ
) ∩ Ker(∂)
⊥
, we have
(), p
2
)
φ
= 0,
4.3 L
2
Existence Theorems for ∂ in Pseudoconvex Domains 75
since ) ∈ Ker(∂). For any p ∈ Dom(∂
∗
φ
), we write p = p
1
+ p
2
where p
1
∈ Ker(∂)
and p
2
∈ Ker(∂)
⊥
= ¹(∂
∗
φ
) ⊂ Ker(∂
∗
φ
). Thus, p
2
∈ Dom(∂
∗
φ
) and ∂
∗
φ
p
2
= 0. This
implies that p
1
∈ Dom(∂
∗
φ
) and ∂
∗
φ
p = ∂
∗
φ
p
1
. Hence, we have for any p ∈ Dom(∂
∗
φ
),
[ (), p)
φ
[ = [ (), p
1
)
φ
[
≤
1
√
t¡
 ) 
φ
 ∂
∗
φ
p
1

φ
=
1
√
t¡
 ) 
φ
 ∂
∗
φ
p 
φ
.
This proves the claim (4.3.20). Using the HahnBanach theorem and the Riesz
representation theorem applied to the antilinear functional ∂
∗
φ
p → (), p)
φ
, there
exists n ∈ 1
2
(p,q−1)
(1, φ) such that for every p ∈ Dom(∂
∗
φ
),
(), p)
φ
= (n, ∂
∗
φ
p)
φ
,
and
 n 
φ
≤
1
√
t¡
 ) 
φ
.
This implies that ∂n = ) in the distribution sense and n satisﬁes
¡
_
D
[ n [
2
d\ ≤ ¡c
tδ
2
_
D
[ n [
2
c
−t]z]
2
d\
≤
1
t
c
tδ
2
_
D
[ ) [
2
c
−t]z]
2
d\
≤
1
t
c
tδ
2
_
D
[ ) [
2
d\.
Since the function
1
t
c
tδ
2
achieves its minimum when t = δ
−2
, we have
¡
_
D
[ n [
2
d\ ≤ cδ
2
_
D
[ ) [
2
d\.
This proves the theorem when the boundary /1 is C
2
.
For a general pseudoconvex domain, from Deﬁnition 3.4.9, one can exhaust 1
by a sequence of pseudoconvex domains with C
∞
boundary 1
ν
. We write
1 =
∞
∪
ν=1
1
ν
,
where each 1
ν
is a bounded pseudoconvex domain with C
∞
boundary and 1
ν
⊂
1
ν+1
⊂ 1 for each ν. Let δ
ν
denote the diameter for 1
ν
. On each 1
ν
, there exists
a n
ν
∈ 1
2
(p,q−1)
(1
ν
) such that ∂n
ν
= ) in 1
ν
and
¡
_
Dν
[ n
ν
[
2
d\ ≤ cδ
2
ν
_
Dν
[ ) [
2
d\ ≤ cδ
2
_
D
[ ) [
2
d\.
76 L
2
Theory for ∂ on Pseudoconvex Domains
We can choose a subsequence of n
ν
, still denoted by n
ν
, such that n
ν
÷ n weakly
in 1
2
(p,q−1)
(1). Furthermore, n satisﬁes the estimate
¡
_
D
[ n [
2
d\ ≤ liminf cδ
2
ν
_
Dν
[ ) [
2
d\ ≤ cδ
2
_
D
[ ) [
2
d\,
and ∂n = ) in 1 in the distribution sense. Theorem 4.3.4 is proved.
Theorem 4.3.5. Let 1 be a pseudoconvex domain in C
n
. For every ) ∈ 1
2
(p,q)
(1,
loc), where 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n with ∂) = 0, one can ﬁnd n ∈ 1
2
(p,q−1)
(1, loc)
such that ∂n = ).
Proof. Since 1 is pseudoconvex, from Deﬁnition 3.4.9, there exists a C
∞
strictly
plurisubharmonic exhaustion function σ for 1. For any ) ∈ 1
2
(p,q)
(1, loc), we can
choose a rapidly increasing convex function η(t), t ∈ 1 such that η(t) = 0 when
t ≤ 0 and ) ∈ 1
2
(p,q)
(1, η(σ)). Let 1
ν
= ¦. ∈ 1[ σ(.) < ν¦, then
1 =
∞
∪
ν=1
1
ν
,
where each 1
ν
is a bounded pseudoconvex domain with C
∞
boundary and 1
ν
⊂
1
ν+1
⊂ 1 for each ν. Since η(σ) is plurisubharmonic, the function φ = η(σ) +[.[
2
is strictly plurisubharmonic with
n
j,k=1
∂
2
φ
∂.
j
∂¯ .
k
(.)o
j
¯ o
k
≥ [o[
2
for all (o
1
, , o
n
) ∈ C
n
and all . ∈ 1. Applying Proposition 4.3.3 to each 1
ν
we
have for any p ∈ Dom(∂) ∩ Dom(∂
∗
φ
),
¡  p 
2
φ(Dν)
≤
_
Dν
I,K
t
j,k
∂
2
φ
∂.
j
∂¯ .
k
p
I,jK
p
I,kK
c
−φ
d\
≤  ∂p 
2
φ(Dν)
+  ∂
∗
φ
p 
2
φ(Dν)
.
Repeating the same argument as in Theorem 4.3.4, there exists a n
ν
∈ 1
2
(p,q−1)
(1
ν
,
φ) such that
¯
∂n
ν
= ) in 1
ν
and
¡
_
Dν
[n
ν
[
2
c
−φ
d\ ≤
_
Dν
[)[
2
c
−φ
d\ ≤
_
D
[)[
2
c
−φ
d\ < ∞.
Taking a weak limit n of n
ν
as ν →∞, we have shown that there exists n such that
¯
∂n = ) in 1 and
¡
_
D
[n[
2
c
−φ
d\ ≤
_
D
[)[
2
c
−φ
d\.
This proves the theorem.
4.4 L
2
Existence Theorems for the
¯
∂Neumann Operator 77
4.4 1
2
Existence Theorems for the
¯
∂Neumann Operator
We shall use the 1
2
existence theorems for
¯
∂ in Section 4.3 to establish the exis
tence theorem for the
¯
∂Neumann operator on any bounded pseudoconvex domain
1 in C
n
. Using Proposition 4.2.3, the operator
(p,q)
is closed and selfadjoint.
Thus, the kernel of
(p,q)
, denoted by Ker(
(p,q)
), is closed. From the Hilbert
space theory, we have the following weak Hodge decomposition
(4.4.1) 1
2
(p,q)
(1) = ¹(
(p,q)
) ⊕Ker(
(p,q)
),
where ¹(
(p,q)
) denotes the range of
(p,q)
. We shall show that ¹(
(p,q)
) is closed
and Ker(
(p,q)
) = ¦0¦. We claim that
(4.4.2) Ker(
(p,q)
) = Ker(
¯
∂) ∩ Ker(
¯
∂
∗
) = ¦0¦ for ¡ ≥ 1.
For any α ∈ Ker(
(p,q)
), we have α ∈ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
) and
(α,
(p,q)
α) = 
¯
∂α
2
+
¯
∂
∗
α
2
= 0.
Thus, Ker(
(p,q)
) ⊂ Ker(
¯
∂) ∩Ker(
¯
∂
∗
). On the other hand, if α ∈ Ker(
¯
∂) ∩Ker(
¯
∂
∗
),
then α ∈ Dom(
(p,q)
) and
(p,q)
α = 0. Thus, Ker(
(p,q)
) ⊃ Ker(
¯
∂) ∩ Ker(
¯
∂
∗
) and
the ﬁrst equality in (4.4.2) is proved. To see that Ker(
¯
∂) ∩ Ker(
¯
∂
∗
) = ¦0¦, we note
that if α ∈ Ker(
¯
∂), from Theorem 4.3.4, there exists n ∈ 1
2
(p,q−1)
(1) such that
α =
¯
∂n. If α is also in Ker(
¯
∂
∗
), we have
0 = (
¯
∂
∗
¯
∂n, n) = 
¯
∂n
2
and α = 0. This proves (4.4.2).
We shall show that ¹(
(p,q)
) is closed and the following 1
2
existence theorem
holds for the
¯
∂Neumann operator.
Theorem 4.4.1. Let 1 be a bounded pseudoconvex domain in C
n
, n ≥ 2. For each
0 ≤ j ≤ n, 1 ≤ ¡ ≤ n, there exists a bounded operator ·
(p,q)
: 1
2
(p,q)
(1) →1
2
(p,q)
(1)
such that
(1) ¹(·
(p,q)
) ⊂ Dom(
(p,q)
),
·
(p,q)
(p,q)
=
(p,q)
·
(p,q)
= 1 on Dom(
(p,q)
).
(2) For any ) ∈ 1
2
(p,q)
(1), ) =
¯
∂
¯
∂
∗
·
(p,q)
) ⊕
¯
∂
∗ ¯
∂·
(p,q)
).
(3)
¯
∂·
(p,q)
= ·
(p,q+1)
¯
∂ on Dom(
¯
∂), 1 ≤ ¡ ≤ n −1.
(4)
¯
∂
∗
·
(p,q)
= ·
(p,q−1)
¯
∂
∗
on Dom(
¯
∂
∗
), 2 ≤ ¡ ≤ n.
(5) Let δ be the diameter of 1. The following estimates hold for any ) ∈
1
2
(p,q)
(1):
 ·
(p,q)
)  ≤
cδ
2
¡
 ) ,

¯
∂·
(p,q)
)  ≤
¸
cδ
2
¡
 ) ,

¯
∂
∗
·
(p,q)
)  ≤
¸
cδ
2
¡
 )  .
78 L
2
Theory for ∂ on Pseudoconvex Domains
Proof. Using Theorem 4.3.4, for any ) ∈ 1
2
(p,q)
(1), ¡ 0 with
¯
∂) = 0, there exists
n ∈ 1
2
(p,q−1)
(1) such that
¯
∂n = ) and n satisﬁes the estimate (4.3.15). Thus, ¹(
¯
∂)
is closed in every degree and is equal to Ker(
¯
∂). It follows from Lemma 4.1.1 that
¹(
¯
∂
∗
) is closed also for every ¡, and we have the following orthogonal decomposition:
(4.4.3) 1
2
(p,q)
(1) = Ker(
¯
∂) ⊕¹(
¯
∂
∗
) = ¹(
¯
∂) ⊕¹(
¯
∂
∗
).
For every ) ∈ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
), we have
) = )
1
⊕)
2
where )
1
∈ ¹(
¯
∂) and )
2
∈ ¹(
¯
∂
∗
). Also, we have )
1
, )
2
∈ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
),
¯
∂) =
¯
∂)
2
,
¯
∂
∗
) =
¯
∂
∗
)
1
.
Using Theorem 4.3.4 and (4.1.3) and (4.1.4) in Lemma 4.1.1, we have the follow
ing estimates:
(4.4.4) )
1

2
≤ c
q

¯
∂
∗
)
1

2
and
(4.4.5) )
2

2
≤ c
q+1

¯
∂)
2

2
,
where the constant c
q
= cδ
2
,¡. Combining (4.4.4), (4.4.5) we have
(4.4.6) )
2
= )
1

2
+)
2

2
≤ c
q
(
¯
∂)
2
+
¯
∂
∗
)
2
)
for every ) ∈ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
). Thus for any ) ∈ Dom(
(p,q)
), we have
(4.4.7)
)
2
≤ c
q
[(
¯
∂),
¯
∂)) + (
¯
∂
∗
),
¯
∂
∗
))]
= c
q
[(
¯
∂
∗
¯
∂), )) + (
¯
∂
¯
∂
∗
), ))]
= c
q
(
(p,q)
), ))
≤ c
q

(p,q)
)).
Hence,
(4.4.8) ) ≤ c
q

(p,q)
).
It follows from Lemma 4.1.1 (since
(p,q)
is a closed operator from Proposition
4.2.3) that the range of
(p,q)
is closed. We have the strong Hodge decomposition
1
2
(p,q)
(1) = ¹(
(p,q)
) =
¯
∂
¯
∂
∗
(Dom(
(p,q)
)) ⊕
¯
∂
∗
¯
∂(Dom(
(p,q)
)).
Also, from (4.4.8),
(p,q)
is one to one and the range of
(p,q)
is the whole space
1
2
(p,q)
(1). There exists a unique inverse ·
(p,q)
: 1
2
(p,q)
(1) →Dom(
(p,q)
) such that
·
(p,q)
= ·
(p,q)
= 1. Using (4.4.8), ·
(p,q)
is bounded. The assertions (1) and
(2) in Theorem 4.4.1 have been established.
4.4 L
2
Existence Theorems for the
¯
∂Neumann Operator 79
To show that
¯
∂·
(p,q)
= ·
(p,q+1)
¯
∂ on Dom(
¯
∂), we note that from (2),
¯
∂) =
¯
∂
¯
∂
∗ ¯
∂·
(p,q)
) for ) ∈ Dom(
¯
∂). It follows that
·
(p,q+1)
¯
∂) = ·
(p,q+1)
¯
∂
¯
∂
∗
¯
∂·
(p,q)
)
= ·
(p,q+1)
(
¯
∂
¯
∂
∗
+
¯
∂
∗
¯
∂)
¯
∂·
(p,q)
)
=
¯
∂·
(p,q)
).
If 2 ≤ ¡ ≤ n, one can prove ·
(p,q−1)
¯
∂
∗
=
¯
∂
∗
·
(p,q)
on Dom(
¯
∂
∗
) similarly.
To prove (5), we see from (4.4.8) that
 ·
(p,q)
)  ≤
cδ
2
¡
 )  for ) ∈ 1
2
(p,q)
(1).
Using (2), we have
(
¯
∂·
(p,q)
),
¯
∂·
(p,q)
)) + (
¯
∂
∗
·
(p,q)
),
¯
∂
∗
·
(p,q)
))
= ((
¯
∂
¯
∂
∗
+
¯
∂
∗
¯
∂)·
(p,q)
), ·
(p,q)
))
= (), ·
(p,q)
))
≤ )·
(p,q)
)
≤
cδ
2
¡
)
2
.
This proves (5). The proof of Theorem 4.4.1 is complete.
Corollary 4.4.2. Let 1 and ·
(p,q)
be the same as in Theorem 4.4.1, where 0 ≤
j ≤ n, 1 ≤ ¡ ≤ n. For any α ∈ 1
2
(p,q)
(1) such that
¯
∂α = 0, the (j, ¡ −1)form
(4.4.9) n =
¯
∂
∗
·
(p,q)
α
satisﬁes the equation
¯
∂n = α and the estimate
n
2
≤
cδ
2
¡
α
2
.
The solution n is called the canonical solution to the equation (4.0.1) and it is
the unique solution which is orthogonal to Ker(
¯
∂).
Proof. We have from (2) of Theorem 4.4.1,
α =
¯
∂
¯
∂
∗
·
(p,q)
α +
¯
∂
∗
¯
∂·
(p,q)
α.
Using (3) in Theorem 4.4.1, we have
¯
∂·
(p,q)
α = ·
(p,q+1)
¯
∂α = 0,
since
¯
∂α = 0. Thus we have α =
¯
∂
¯
∂
∗
·
(p,q)
α. The estimate of n follows from (5) in
Theorem 4.4.1. If · is another solution orthogonal to Ker(
¯
∂), then n−· ∈ Ker(
¯
∂) ∩
Ker(
¯
∂
∗
) = 0. This proves the uniqueness of the canonical solution. Corollary 4.4.2
is proved.
The existence of the
¯
∂Neumann operator for ¡ = 0, ·
(p,0)
, is also important.
Let
(p,0)
=
¯
∂
∗ ¯
∂ on 1
2
(p,0)
(1). We deﬁne
H
(p,0)
(1) = ¦) ∈ 1
2
(p,0)
(1)[
¯
∂) = 0¦.
H
(p,0)
(1) is a closed subspace of 1
2
(p,0)
since
¯
∂ is a closed operator. Let H
(p,0)
denote the projection from 1
2
(p,0)
(1) onto the set H
(p,0)
(1). We have the following
theorem.
80 L
2
Theory for ∂ on Pseudoconvex Domains
Theorem 4.4.3. Let 1 be a bounded pseudoconvex domain in C
n
, n ≥ 2. There
exists an operator ·
(p,0)
: 1
2
(p,0)
(1) →1
2
(p,0)
(1) such that
(1) ¹(·
(p,0)
) ⊂ Dom(
(p,0)
),
·
(p,0)
(p,0)
=
(p,0)
·
(p,0)
= 1 −H
(p,0)
.
(2) For every ) ∈ 1
2
(p,0)
(1), ) =
¯
∂
∗ ¯
∂·
(p,0)
) ⊕H
(p,0)
).
(3)
¯
∂·
(p,0)
= ·
(p,1)
¯
∂ on Dom(
¯
∂),
¯
∂
∗
·
(p,1)
= ·
(p,0)
¯
∂
∗
on Dom (
¯
∂
∗
).
(4) ·
(p,0)
=
¯
∂
∗
·
2
(p,1)
¯
∂.
(5) Let δ be the diameter of 1. For any ) ∈ 1
2
(p,0)
(1),
 ·
(p,0)
)  ≤ cδ
2
 ) ,

¯
∂·
(p,0)
)  ≤
√
cδ  )  .
Proof. Note that H
(p,0)
(1) = H
(p,0)
= Ker(
(p,0)
). We ﬁrst show that
(p,0)
is
bounded away from zero on (H
(p,0)
)
⊥
. Since
¯
∂ has closed range in every degree,
¯
∂
∗
also has closed range by Lemma 4.1.1. If ) ∈ Dom(
(p,0)
) ∩ (H
(p,0)
)
⊥
, we have
) ⊥ Ker(
¯
∂) and ) ∈ ¹(
¯
∂
∗
).
Let α =
¯
∂), then α ∈ 1
2
(p,1)
(1) since ) ∈ Dom(
(p,0)
). Using (4) in Theorem
4.4.1, we have that φ ≡
¯
∂
∗
·
(p,1)
α is the unique solution satisfying
¯
∂φ = α and
φ ⊥ Ker(
¯
∂). Thus, φ = ). Applying Corollary 4.4.2, we have
)
2
≤ cα
2
= c
¯
∂)
2
= c(
(p,0)
), )) ≤ c
(p,0)
)),
where c = cδ
2
. This implies that
) ≤ cδ
2

(p,0)
) for ) ∈ Dom(
(p,0)
) ∩ (H
(p,0)
)
⊥
.
Using Lemma 4.1.1, we see that
(p,0)
has closed range. From (4.1.1), the following
strong Hodge decomposition holds:
1
2
(p,0)
(1) = ¹(
(p,0)
) ⊕H
(p,0)
=
¯
∂
∗
¯
∂(Dom(
(p,0)
)) ⊕H
(p,0)
.
For any α ∈ ¹(
(p,0)
), there is a unique ·
(p,0)
α ⊥ H
(p,0)
such that
(p,0)
·
(p,0)
α =
α. Extending ·
(p,0)
to 1
2
(p,0)
(1) by requiring ·
(p,0)
H
(p,0)
= 0, ·
(p,0)
satisﬁes (1)
and (2) in Theorem 4.4.3. That ·
(p,0)
commutes with
¯
∂ and
¯
∂
∗
is proved exactly
as before, and we omit the details. If ) ∈ Dom(
¯
∂), it follows that
(4.4.10) ·
(p,0)
) = (1 −H
(p,0)
)·
(p,0)
) = ·
(p,0)
(
¯
∂
∗
¯
∂)·
(p,0)
) =
¯
∂
∗
·
2
(p,1)
¯
∂).
Thus, (4) holds on Dom(
¯
∂). In fact, we can show that (4.4.10) holds on all of
1
2
(p,0)
(1). From (5) in Theorem 4.4.1, we have
 ·
(p,1)
α  ≤ cδ
2
α.
4.4 L
2
Existence Theorems for the
¯
∂Neumann Operator 81
We obtain, for any ) ∈ C
∞
(p,0)
(1),
(4.4.11)
·
(p,0)
)
2
= (
¯
∂
¯
∂
∗
·
2
(p,1)
¯
∂), ·
2
(p,1)
¯
∂))
= (·
(p,1)
¯
∂), ·
2
(p,1)
¯
∂))
≤ ·
(p,1)
¯
∂)·
2
(p,1)
¯
∂)
≤ cδ
2
·
(p,1)
¯
∂)
2
.
On the other hand, we have
(4.4.12)
(·
(p,1)
¯
∂), ·
(p,1)
¯
∂)) = (·
2
(p,1)
¯
∂),
¯
∂))
= (
¯
∂
∗
·
2
(p,1)
¯
∂), ))
≤ ·
(p,0)
)).
Combining (4.4.11) and (4.4.12), we have proved that
·
(p,0)
) ≤ cδ
2
).
Thus, ·
(p,0)
deﬁned by (4.4.10) is bounded on all smooth (j, 0)forms and it can be
extended to 1
2
(p,0)
(1) as a bounded operator. This proves (4). It follows from (1)
that

¯
∂·
(p,0)
)
2
= (
¯
∂
∗
¯
∂·
(p,0)
), ·
(p,0)
))
= ((1 −H
(p,0)
)), ·
(p,0)
))
≤ )·
(p,0)
) ≤ cδ
2
)
2
.
Theorem 4.4.3 is proved.
Corollary 4.4.4. Let 1 be a bounded pseudoconvex domain in C
n
. For any ) ∈
1
2
(p,0)
(1), we have
(4.4.13) H
(p,0)
) = ) −
¯
∂
∗
·
(p,1)
¯
∂).
Proof. For any ) ∈ Dom(
¯
∂), this follows from (2) in Theorem 4.4.3, since
H
(p,0)
= 1 −
¯
∂
∗
¯
∂·
(p,0)
= 1 −
¯
∂
∗
·
(p,1)
¯
∂.
From (3) in Theorem 4.4.3,
¯
∂
∗
·
(p,1)
¯
∂ =
¯
∂
∗ ¯
∂·
(p,0)
= 1 − H
(p,0)
is a bounded
operator on Dom(
¯
∂), it can be extended to any ) ∈ 1
2
(p,0)
(1) by continuity. This
proves the corollary.
Corollary 4.4.4 is especially important when j = 0, so we state it below as a
theorem. Let 1 denote the projection onto the closed subspace of holomorphic
square integrable functions
H(1) = ¦) ∈ 1
2
(1) [
¯
∂) = 0¦.
1 is called the Bergman projection and H(1) is called the Bergman space.
82 L
2
Theory for ∂ on Pseudoconvex Domains
Theorem 4.4.5 (Bergman projection). Let 1 be a bounded pseudoconvex do
main in C
n
. For any ) ∈ 1
2
(1), the Bergman projection 1) is given by
(4.4.14) 1) = ) −
¯
∂
∗
·
(0,1)
¯
∂).
4.5 Pseudoconvexity and the Levi Problem
In this section we show that pseudoconvex domains are domains of holomorphy.
We ﬁrst examine the solvability of
¯
∂ in the C
∞
(1) category.
Theorem 4.5.1. Let 1 be a pseudoconvex domain in C
n
. For every ) ∈ C
∞
(p,q)
(1),
where 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n with ∂) = 0, one can ﬁnd n ∈ C
∞
(p,q−1)
(1) such that
∂n = ).
Proof. Let ) ∈ C
∞
(p,q)
(1). From Theorem 4.3.5, there exists a strictly plurisub
harmonic function φ ∈ C
∞
(1) such that ) is in 1
2
(p,q)
(1, φ) and there exists
· ∈ 1
2
(p,q−1)
(1, φ) with
¯
∂· = ) and
 · 
φ
≤  ) 
φ
.
Repeating the same arguments as in Section 4.4, there exists a weighted
¯
∂
Neumann operator ·
φ
such that for any ) ∈ 1
2
(p,q)
(1, φ), we have
) =
¯
∂
¯
∂
∗
φ
·
φ
) +
¯
∂
∗
φ
¯
∂·
φ
).
Since
¯
∂) = 0, we have that ) =
¯
∂
¯
∂
∗
φ
·
φ
). Setting n =
¯
∂
∗
φ
·
φ
), we shall show that
n ∈ C
∞
(p,q−1)
(1). Since
¯
∂
∗
φ
n = ϑn +¹
0
n = 0 for some zeroth order operator ¹
0
, we
have
_
¯
∂n = ),
ϑn = −¹
0
n ∈ 1
2
(1, loc).
However,
¯
∂ ⊕ ϑ is an elliptic system. By this we mean that for any α ∈ C
∞
(p,q)
(1)
such that α has compact support in 1, the following inequality holds:
(4.5.1)  α 
1
≤ C(
¯
∂α  +  ϑα  +  α ).
Inequality (4.5.1) is called G˚arding’s inequality. To prove (4.5.1), we use Proposition
4.2.4 and (4.2.8) to get
(4.5.2) 4(
¯
∂α 
2
+  ϑα 
2
) = 4(α, α) = (−´α, α) =  ∇α 
2
,
where ´ is the real Laplacian and ∇ is the gradient, both act on α componentwise.
When ¡ = 0, (4.5.2) also holds since = ϑ
¯
∂ is also equal to −
1
4
´. Thus (4.5.1)
holds for any compactly supported smooth form α. Let ˜ n = ζn where ζ ∈ C
∞
0
(1)
and deﬁne n
= ˜ n ∗ χ
where χ and χ
are the same as in Lemma 4.3.2. It follows
4.5 Pseudoconvexity and the Levi Problem 83
that  n
 ≤  ˜ n ,
¯
∂n
=
¯
∂˜ n ∗ χ
and ϑn
= ϑ˜ n ∗ χ
. Substituting n
into (4.5.1),
we have
(4.5.3)
 n

1
≤ C(
¯
∂n
 +  ϑn
 +  n
)
≤ C(
¯
∂˜ n  +  ϑ˜ n  +  ˜ n ).
Thus, n
converges in \
1
(1) to ˜ n, and we have n ∈ \
1
(1, loc). Continuing
this process to 1
k
˜ n where 1
k
is any /th order diﬀerential operator with constant
coeﬃcients, we conclude by induction that n ∈ \
k+1
(1, loc) for any / ∈ N. The
theorem follows from the Sobolev embedding theorem (see Theorem A.7 in the
Appendix).
The following theorem uniﬁes domains of holomorphy, pseudoconvexity and ex
istence theorems for the CauchyRiemann equations:
Theorem 4.5.2. Let 1 be a domain in C
n
, n ≥ 1. Then the following conditions
are equivalent:
(1) 1 is pseudoconvex.
(2) 1 is a domain of holomorphy.
(3) For every ) ∈ C
∞
(p,q)
(1), where 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n with ∂) = 0, one can
ﬁnd n ∈ C
∞
(p,q−1)
(1) such that ∂n = ).
Proof. We have already proved in Theorem 3.5.5 that (2) implies (1). That (1)
implies (3) follows from Theorem 4.5.1.
To prove that (3) implies (2), we use an induction argument. This is obviously
true for n = 1, since any open set in C is a domain of holomorphy. We shall show
that if (3) implies (2) for n −1, then it is true for n.
To prove this, for any .
0
∈ /1, we need to construct a holomorphic function in
1 which cannot be extended holomorphically across any neighborhood containing
.
0
. It suﬃces to prove this in a dense subset of /1.
Let .
0
be a boundary point such that there exists a complex (n−1)dimensional
hyperplane Σ and .
0
∈ /(Σ ∩ 1). Such boundary points are dense in /1. To see
this, we note that for almost every boundary point .
0
, one can ﬁnd a ball 1 ⊂ 1
such that .
0
∈ /(1∩1). At such a .
0
, we obviously can ﬁnd a complex hyperplane
Σ passing through the center of the ball and .
0
. It is easy to see that .
0
∈ /(Σ∩1).
By a linear transformation, we may assume that .
0
= 0 and Σ
0
= 1 ∩ ¦.
n
= 0¦ is
nonempty.
We shall show that on Σ
0
, (3) is fulﬁlled. Let ) be a smooth
¯
∂closed (j, ¡)form
on Σ
0
, where 0 ≤ j ≤ n−1, 1 ≤ ¡ ≤ n−1. We claim that ) can be extended to be a
smooth
¯
∂closed form in 1. We ﬁrst extend ) to
˜
) in 1 such that
˜
) ∈ C
∞
(1) and
˜
)(.) =
˜
)(.
1
, , .
n
) = )(.
t
, 0) in an open neighborhood of Σ
0
. This can be done
as follows: Let π : 1 →C
n−1
be the projection such that π(.) = (.
1
, , .
n−1
, 0).
Then the set 1
0
= 1¸ π
−1
(Σ
0
) is a closed subset of 1. Since Σ
0
and 1
0
are closed
(with respect to 1) disjoint subsets of 1, using Urysohn’s lemma, we see that there
exists a function η ∈ C
∞
(1) such that η = 1 in a neighborhood of Σ
0
and η = 0
in a neighborhood of 1
0
. Then we can choose our
˜
) = ηπ
∗
)(.
t
), where π
∗
is the
pullback of the form ). Let
1(.) =
˜
)(.) −.
n
n(.)
84 L
2
Theory for ∂ on Pseudoconvex Domains
where n(.) is chosen such that
(4.5.4)
¯
∂n(.) =
¯
∂
˜
)
.
n
.
We note that the righthand side of (4.5.4) is
¯
∂closed and is in C
∞
(p,q+1)
(1), since
˜
) is
¯
∂closed in a neighborhood of Σ
0
. Thus, from (3), there exists n ∈ C
∞
(p,q)
(1)
satisfying the Equation (4.5.4). This implies that 1 is
¯
∂closed on 1 and 1 = ) on
Σ
0
. Thus, any
¯
∂closed form ) on Σ
0
can be extended to a
¯
∂ closed form 1 on 1.
This is also true for ¡ = 0.
From (3), we can ﬁnd l(.) ∈ C
∞
(p,q−1)
(1) such that
¯
∂l = 1 in 1. Restricting
l to Σ
0
, we have shown that (3) is fulﬁlled on Σ
0
.
By the induction hypothesis, Σ
0
is a domain of holomorphy. Hence, there exists
a holomorphic function )(.
t
) = )(.
1
, , .
n−1
) such that ) is singular at 0. Since
¯
∂) = 0 in Σ
0
, repeating the same argument above for ¡ = 0, there exists 1 in 1
such that
¯
∂1 = 0 in 1 and 1 = ) on Σ
0
. 1(.) is holomorphic in 1 and is equal to
)(.
t
, 0) on Σ
0
. Thus, it is a holomorphic function in 1 which cannot be extended
across 0. This shows that 1 is a domain of holomorphy. Thus, (3) implies (2) and
the theorem is proved.
Theorem 4.5.2 solves the Levi problem on pseudoconvex domains in C
n
.
NOTES
The
¯
∂Neumann problem was suggested by P. R. Garabedian and D. C. Spencer
[GaSp 1] to study the CauchyRiemann equations. This approach generalizes the
Hodgede Rham theorem from compact manifolds to complex manifolds with bound
aries. The basic a priori estimates were ﬁrst proved for (0, 1)forms by C. B. Morrey
[Mor 1]. J. J. Kohn [Koh 1] has derived the general estimates and has proved the
boundary regularity for the
¯
∂Neumann operator on strongly pseudoconvex mani
folds. This latter result is actually required in Kohn’s approach to the
¯
∂Neumann
problem which will be discussed in Chapter 5. The use of weighted 1
2
estimates
which depend on a parameter, combined with the basic estimates of Morrey and
Kohn, to study the overdetermined system was introduced by L. H¨ormander [H¨or
3] in order to bypass the boundary regularity problem. Related arguments are used
in A. Andreotti and E. Vesentini [AnVe 1].
Lemma 4.1.1 and much of the material on 1
2
existence theorems presented in
Section 4.3 are taken from the paper of L. H¨ormander [H¨or 3]. Theorem 4.3.4 is a
special case of Theorem 2.2.3 in [H¨or 3] where the precise bounds are given. The
density lemma 4.3.2 was also proved in [H¨or 3] in a much more general setting. Using
three diﬀerent weight functions which are singular near the boundary, L. H¨ormander
[H¨or 9] gives another approach to 1
2
existence theorems. The canonical solution
formula given by (4.4.9) and the Bergman projection formula (4.4.14) are due to
J. J. Kohn [Koh 1]. The proof of Theorem 4.5.2 is due to K. Oka [Oka 2] and H.
Bremermann [Bre 1], and F. Norguet [Nor 1] and our presentation follows that of
Section 4.2 in [H¨or 9].
85
86
CHAPTER 5
THE
¯
∂NEUMANN PROBLEM
ON STRONGLY PSEUDOCONVEX MANIFOLDS
In this chapter we study boundary regularity for the
¯
∂Neumann problem on
a strongly pseudoconvex domain Ω. Let ρ be a C
2
deﬁning function for Ω. The
¯
∂Neumann problem for (j, ¡)forms, 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n, in Ω is the boundary
value problem:
(5.0.1)
_
¸
_
¸
_
n = ) in Ω,
¯
∂ρ ∨ n = 0 on /Ω,
¯
∂ρ ∨
¯
∂n = 0 on /Ω,
where n, ) are (j, ¡)forms, =
¯
∂ϑ +ϑ
¯
∂, ∨ denotes the interior product of forms.
On any bounded pseudoconvex domain Ω in C
n
, we have derived the following
estimates: for any ) ∈ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
),
(5.0.2)  ) 
2
Ω
≤
cδ
2
¡
(
¯
∂) 
2
Ω
+ 
¯
∂
∗
) 
2
Ω
),
where δ is the diameter of Ω (see (4.4.6)). It follows from Theorem 4.4.1 that the
¯
∂Neumann operator ·
(p,q)
exists in Ω, which solves (5.0.1) in the Hilbert space
sense.
The operator is elliptic in the interior, but the boundary conditions are not
coercive except when ¡ = n. It only satisﬁes G˚arding’s inequality in the interior, but
not near the boundary. However, under the assumption of strong pseudoconvexity,
we will show that it satisﬁes subelliptic 1,2estimates near the boundary.
In Section 5.1, we prove that the following subelliptic 1,2estimate holds on a
strongly pseudoconvex domain Ω in C
n
: for any ) ∈ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
),
(5.0.3)  ) 
2
1
2
(Ω)
≤ C(
¯
∂) 
2
Ω
+ 
¯
∂
∗
) 
2
Ω
),
where  1
2
(Ω)
is the Sobolev norm in the Sobolev space \
1/2
(p,q)
(Ω).
The regularity of the
¯
∂Neumann operator in other Sobolev spaces when the
boundary /Ω is C
∞
is discussed in Section 5.2. In Section 5.3, we discus the
existence and regularity of the
¯
∂Neumann operator in an open subset of a complex
manifold with a Hermitian metric. In particular, a solution to the Levi problem
on strongly pseudoconvex manifolds is obtained using the
¯
∂Neumann operator.
Finally, the NewlanderNirenberg theorem is proved by using the solution for
¯
∂ on
an almost complex manifold in the last section.
5.1 Subelliptic Estimates for the
¯
∂Neumann Operator 87
5.1 Subelliptic Estimates for the
¯
∂Neumann Operator
In this section we shall derive the subelliptic 1,2estimate for the
¯
∂Neumann
operator when Ω is strongly pseudoconvex with C
2
boundary. We shall use ·
instead of ·
(p,q)
to simplify the notation. We also use 1
2
(Ω), \
s
(Ω) and \
s
(Ω, loc)
to denote the spaces 1
2
(p,q)
(Ω), \
s
(p,q)
(Ω) and \
s
(p,q)
(Ω, loc) respectively, where
\
s
(Ω) is the Sobolev space, : ∈ 1. (See Appendix A for its deﬁnition and basic
properties.) The norm in \
s
is denoted by  
s(Ω)
.
We ﬁrst observe that the ﬁrst order system
¯
∂ ⊕ϑ : C
∞
(p,q)
(Ω) → C
∞
(p,q+1)
(Ω) ⊕ C
∞
(p,q−1)
(Ω)
is elliptic in the interior. This means that we have G˚arding’s inequality in the
interior.
Proposition 5.1.1. Let Ω be a bounded domain in C
n
and η ∈ C
∞
0
(Ω). For any
(j, ¡)form ) ∈ 1
2
(p,q)
(Ω) such that
¯
∂) ∈ 1
2
(p,q+1)
(Ω) and ϑ) ∈ 1
2
(p,q−1)
(Ω), where
0 ≤ j ≤ n and 0 ≤ ¡ ≤ n, we have the following estimates:
(5.1.1)  η) 
2
1(Ω)
≤ C(
¯
∂) 
2
Ω
+  ϑ) 
2
Ω
+  ) 
2
Ω
),
where C is a constant depending only on η but not on ).
Proof. Using the basic estimates proved in Proposition 4.3.1, when φ = 0, we have
for any η ∈ C
∞
0
(Ω) and ) ∈ C
∞
(p,q)
(Ω), 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n,
(5.1.2)
I,J
t
k

∂(η)
I,J
)
∂¯ .
k

2
Ω
= (
¯
∂(η)) 
2
Ω
+  ϑ(η)) 
2
Ω
)
≤ C( η
¯
∂) 
2
Ω
+  ηϑ) 
2
Ω
+  ) 
2
Ω
).
(5.1.2) also holds trivially for ¡ = 0 (in this case, ϑ) = 0). But
(5.1.3) 
∂(η)
I,J
)
∂¯ .
k

2
Ω
= 
∂(η)
I,J
)
∂.
k

2
Ω
from integration by parts. We have for any smooth ),
 η) 
2
1(Ω)
≤ C
_
_
I,J
t
k

∂(η)
I,J
)
∂.
k

2
Ω
+
I,J
t
k

∂(η)
I,J
)
∂¯ .
k

2
Ω
+  ) 
2
Ω
_
_
≤ C( η
¯
∂) 
2
Ω
+  ηϑ) 
2
Ω
+  ) 
2
Ω
).
We note that one can also use (4.5.2) to prove the above a priori estimates. Estimate
(5.1.1) follows from regularization similar to (4.5.3) and we omit the details.
If ) ∈ Dom(
¯
∂)∩Dom(
¯
∂
∗
), then ) ∈ \
1
(Ω, loc). The diﬃculty for the
¯
∂Neumann
problem is only on the boundary. The following subelliptic 1,2estimates for a
strongly pseudoconvex domain are of fundamental importance:
88 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
Theorem 5.1.2. Let Ω be a bounded strongly pseudoconvex domain in C
n
with C
2
boundary. The following estimate holds: for any 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n − 1,
there exists C 0 such that for any ) ∈ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
),
(5.1.4)  ) 
2
1
2
(Ω)
≤ C(
¯
∂) 
2
Ω
+ 
¯
∂
∗
) 
2
Ω
),
where C is independent of ).
Theorem 5.1.3. Let Ω be a bounded strongly pseudoconvex domain in C
n
with C
2
boundary. For any 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n − 1, the
¯
∂Neumann operator ·
satisﬁes the estimates:
(5.1.5)  ·) 
2
1
2
(Ω)
≤ C  ) 
2
−
1
2
(Ω)
, ) ∈ 1
2
(p,q)
(Ω),
where C is independent of ). · can be extended as a bounded operator from
\
−1/2
(p,q)
(Ω) into \
1/2
(p,q)
(Ω). In particular, · is a compact operator on 1
2
(p,q)
(Ω).
We divide the proof of Theorem 5.1.2 into several lemmas. From Lemma 4.3.2,
we have that C
1
(p,q)
(Ω) ∩ Dom(
¯
∂
∗
) = T
1
(p,q)
is dense in Dom(
¯
∂) ∩ Dom(
¯
∂
∗
) in the
graph norm. We only need to prove (5.1.4) for C
1
smooth forms ). The starting
point is the following basic a priori estimate of MorreyKohn:
Lemma 5.1.4. Let Ω be a bounded strongly pseudoconvex domain in C
n
with C
2
boundary /Ω. There exists a constant C 0, such that for any ) ∈ C
1
(p,q)
(Ω) ∩
Dom(
¯
∂
∗
) = T
1
(p,q)
,
(5.1.6)
_
bΩ
[)[
2
do ≤ C(
¯
∂) 
2
Ω
+  ϑ) 
2
Ω
),
where do is the surface element on /Ω and C is independent of ).
Proof. Let ) =
t
]I]=p,]J]=q
)
I,J
d.
I
∧ d¯ .
J
, where 1 = (i
1
, , i
p
) and J = (,
1
, ,
,
q
) are increasing multiindices. Let ρ be a C
2
deﬁning function for Ω normalized
such that [dρ[ = 1 on /Ω. Following Proposition 4.3.1 with φ = 0, we have for each
) ∈ C
1
(p,q)
(Ω) ∩ Dom(
¯
∂
∗
),
(5.1.7)

¯
∂) 
2
Ω
+  ϑ) 
2
Ω
=
I,J
t
k

∂)
I,J
∂¯ .
k

2
Ω
+
I,K
t
n
i,j=1
_
bΩ
∂
2
ρ
∂.
i
∂¯ .
j
)
I,iK
¯
)
I,jK
do,
where 1 is an increasing multiindex and [1[ = ¡ −1.
Since /Ω is strongly pseudoconvex with C
2
boundary, there exists C
0
0 such
that for any . ∈ /Ω,
(5.1.8)
n
i,j=1
∂
2
ρ
∂.
i
∂¯ .
j
o
i
¯ o
j
≥ C
0
[o[
2
, if
n
i=1
o
i
∂ρ
∂.
i
= 0.
5.1 Subelliptic Estimates for the
¯
∂Neumann Operator 89
Since ) ∈ Dom(
¯
∂
∗
) ∩ C
1
(p,q)
(Ω), we have from Lemma 4.2.1,
n
j=1
)
I,jK
∂ρ
∂.
j
= 0 on /Ω for each 1, 1.
Substituting (5.1.8) into (5.1.7), we have

¯
∂) 
2
Ω
+  ϑ) 
2
Ω
≥
I,J
t
k

∂)
I,J
∂¯ .
k

2
Ω
+ ¡ C
0
_
bΩ
[)[
2
do.
This proves the lemma.
Lemma 5.1.5. Let Ω be a bounded domain in C
n
with C
2
boundary /Ω and let ρ
be a C
2
deﬁning function for Ω. There exists a constant C 0 such that for any
) ∈ C
2
(Ω)
(5.1.9)
_
Ω
[ρ[[∇)[
2
d\ ≤ C
__
bΩ
[)[
2
do +
_
Ω
[)[
2
d\
_
+ Re
_
Ω
ρ(´))
¯
)d\,
where do is the surface element on /Ω and C 0 is independent of ).
Proof. By Green’s theorem, for any n, · ∈ C
2
(Ω),
(5.1.10)
_
Ω
(n´· −·´n)d\ =
_
bΩ
_
n
∂·
∂n
−·
∂n
∂n
_
do,
where ∂,∂n is the directional derivative along the outward normal. Let n = −ρ
and · = [)[
2
,2, then ´· = Re(
¯
)´)) +[∇)[
2
. We have from (5.1.10)
(5.1.11)
_
Ω
−ρ[∇)[
2
d\ + Re
_
Ω
−ρ(´))
¯
)d\ +
_
Ω
[)[
2
2
´ρd\ =
_
bΩ
[)[
2
2
∂ρ
∂n
do.
Since ρ is in C
2
(Ω), there exists C 0 such that [´ρ[ ≤ C in Ω and [∂ρ,∂n[ ≤ C
on /Ω. This implies that
(5.1.12)
_
Ω
[ρ[[∇)[
2
d\ ≤ C
__
bΩ
[)[
2
do +
_
Ω
[)[
2
d\
_
+ Re
_
Ω
ρ(´))
¯
)d\,
where C is independent of ). This proves the Lemma.
Lemma 5.1.6. Let Ω be a bounded domain in C
n
with C
2
boundary /Ω. There
exists a constant C 0 such that for any ) ∈ C
1
(p,q)
(Ω),
(5.1.13)  ) 
2
1
2
(Ω)
≤ C
__
bΩ
[)[
2
do+ 
¯
∂) 
2
Ω
+  ϑ) 
2
Ω
+  ) 
2
Ω
_
,
where C is independent of ).
90 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
Proof. When ) ∈ C
2
(p,q)
(Ω), one sees from Proposition 4.2.4 that ) = −
1
4
´),
where =
¯
∂ϑ + ϑ
¯
∂ and ´ is deﬁned componentwise. Applying Lemma 5.1.5 to
each component of ), we get
(5.1.14)
_
Ω
[ρ[[∇)[
2
d\ ≤ C
__
bΩ
[)[
2
do +
_
Ω
[)[
2
d\
_
+ 4[(ρ), ))
Ω
[.
Since ) ∈ C
1
(p,q)
(Ω), we have
(5.1.15)  ) 
2
1
2
(Ω)
≤ C
__
Ω
[ρ[[∇)[
2
d\ +
_
Ω
[)[
2
d\
_
.
(5.1.15) follows from Theorem C.2 in the Appendix applied to each component of
).
Now integration by parts gives that
(5.1.16)
[ (ρ), ))
Ω
[ ≤ [ (
¯
∂(ρϑ)), ))
Ω
[ + [ ([
¯
∂, ρ]ϑ), ))
Ω
[
+ [ (ϑ(ρ
¯
∂)), ))
Ω
[ + [ ([ϑ, ρ]
¯
∂), ))
Ω
[
≤ [ (ρϑ), ϑ))
Ω
[ + [ (ρ
¯
∂),
¯
∂))
Ω
[
+ sup
Ω
[∇ρ[(
¯
∂) 
Ω
+  ϑ) 
Ω
)  ) 
Ω
≤ C(
¯
∂) 
2
Ω
+  ϑ) 
2
Ω
+  ) 
2
Ω
).
Combining (5.1.14)(5.1.16), we see that (5.1.13) holds for any ) ∈ C
2
(p,q)
(Ω). An
approximation argument shows that (5.1.13) holds for any ) ∈ C
1
(p,q)
(Ω) since C
2
(Ω)
is dense in C
1
(Ω).
Proof of Theorem 5.1.2. From the assumption of strong pseudoconvexity and Lem
ma 5.1.4, we have for any ) ∈ C
1
(p,q)
(Ω) ∩ Dom(
¯
∂
∗
),
(5.1.17)
_
bΩ
[)[
2
do ≤ C(
¯
∂) 
2
Ω
+  ϑ) 
2
Ω
).
Using (4.4.6), we also get
(5.1.18)  ) 
2
Ω
≤
cδ
2
¡
(
¯
∂) 
2
Ω
+  ϑ) 
2
Ω
).
Thus, Lemmas 5.1.4 and 5.1.6 show that (5.1.4) holds for all forms ) ∈ C
1
(p,q)
(Ω) ∩
Dom(
¯
∂
∗
) = T
1
(p,q)
. Since T
1
(p,q)
is dense in Dom(
¯
∂) ∩ Dom(
¯
∂
∗
) in the graph norm

¯
∂) 
Ω
+ 
¯
∂
∗
) 
Ω
from Lemma 4.3.2, Theorem 5.1.2 is proved.
Proof of Theorem 5.1.3. By the deﬁnition of the space \
−1/2
(Ω) (see Appendix
A), we have
(5.1.19) [(/, p)
Ω
[ ≤  / 1
2
(Ω)
 p 
−
1
2
(Ω)
5.1 Subelliptic Estimates for the
¯
∂Neumann Operator 91
for any / ∈ \
1/2
(Ω) and p ∈ \
−1/2
(Ω). There exists a constant C 0 such that
for any ) ∈ 1
2
(p,q)
(Ω) ∩ Dom(), 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n −1,
(5.1.20)
 ) 
2
1
2
(Ω)
≤ C(
¯
∂) 
2
Ω
+ 
¯
∂
∗
) 
2
Ω
)
= C(), ))
Ω
≤ C  ) 
−
1
2
(Ω)
 ) 1
2
(Ω)
,
where C is independent of ). Substituting ·) into (5.1.20), we have
 ·) 1
2
(Ω)
≤ C  ·) 
−
1
2
(Ω)
= C  ) 
−
1
2
(Ω)
.
Thus, · can be extended as a bounded operator from \
−1/2
(Ω) to \
1/2
(Ω). Using
the Rellich lemma (see Theorem A.8 in the Appendix), · is a compact operator on
\
−1/2
(Ω) and 1
2
(p,q)
(Ω). This proves Theorem 5.1.3.
Corollary 5.1.7. Let Ω and ) be as in Theorem 5.1.2. Then
¯
∂
∗
· and
¯
∂· are
compact operators on 1
2
(p,q)
(Ω). Moreover, the following estimates hold: there exists
a C 0 such that for any ) ∈ 1
2
(p,q)
(Ω),
(5.1.21) 
¯
∂
∗
·) 1
2
(Ω)
≤ C  ) 
Ω
, 1 ≤ ¡ ≤ n,
(5.1.22) 
¯
∂·) 1
2
(Ω)
≤ C  ) 
Ω
, 0 ≤ ¡ ≤ n −1,
where C is independent of ).
Proof. Estimate (5.1.22) follows easily from (5.1.4). Since
¯
∂·) is in Dom(
¯
∂) ∩
Dom(
¯
∂
∗
) when 0 ≤ ¡ ≤ n −1, substituting
¯
∂·) into (5.1.4), we have

¯
∂·) 
2
1
2
(Ω)
≤ C(
¯
∂
∗
¯
∂·) 
2
Ω
+ 
¯
∂
¯
∂·) 
2
Ω
) ≤ C  ) 
2
Ω
.
When ¡ 1, (5.1.21) can be proved similarly since in this case, we also have
¯
∂
∗
·)
is in Dom(
¯
∂) ∩ Dom(
¯
∂
∗
). When ¡ = 1,
¯
∂
∗
·) is a (j, 0)form and (5.1.4) does not
hold for ¡ = 0. The proof of (5.1.21) for ¡ = 1 is much more involved, and will be
postponed. A more general result will be proved in the next section (see Theorem
5.2.6).
Remark. When ¡ = n, the
¯
∂Neumann problem is the classical Dirichlet problem
_
n = ) in Ω,
n = 0 on /Ω,
where n and ) are (j, n)forms. In this case, we have G˚arding’s inequality in place
of (5.1.4):
 ) 
2
1(Ω)
≤ C 
¯
∂
∗
) 
2
Ω
, ) ∈ Dom(
¯
∂
∗
),
where C is independent of ). Since for any ) ∈ 1
2
(p,n)
(Ω), ·) ∈ Dom(
¯
∂
∗
) and we
obtain
 ·) 
2
1(Ω)
≤ C 
¯
∂
∗
·) 
2
Ω
= C(
¯
∂
¯
∂
∗
·), ·))
Ω
= C(), ·))
Ω
≤ C  ) 
−1(Ω)
 ·) 
1(Ω)
.
92 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
·
(p,n)
can be extended as a bounded operator from \
−1
(Ω) into \
1
0
(Ω) (see Ap
pendix A for its deﬁnition) for any bounded domain with C
2
boundary.
5.2 Boundary Regularity for · and
¯
∂
∗
·
In this section, we assume that Ω is strongly pseudoconvex and its boundary
/Ω is C
∞
; i.e., there exists a C
∞
deﬁning function ρ. We normalize ρ such that
[dρ[ = 1 on /Ω. First we derive estimates for the
¯
∂Neumann operator · in Sobolev
spaces for : ≥ 0. As observed in Proposition 5.1.1, the system
¯
∂ ⊕ ϑ is elliptic in
the interior. The interior regularity follows from the usual elliptic theory (see e.g.
LionsMagenes [LiMa 1], BersJohnSchechter [BJS 1] or Treves [Tre 1]). Based on
G˚arding’s inequality (5.1.1), we have for every η ∈ C
∞
0
(Ω) and : ≥ 0,
η·) ∈ \
s+2
(p,q)
(Ω), for any ) ∈ \
s
(p,q)
(Ω).
The main result in this section is to prove the following estimates on boundary
regularity for ·.
Theorem 5.2.1. Let Ω ⊂ C
n
be a bounded strongly pseudoconvex domain with
C
∞
boundary. The
¯
∂Neumann operator · is a bounded operator from \
s
(p,q)
(Ω)
to \
s+1
(p,q)
(Ω) where : ≥ −
1
2
, 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n − 1, and · satisﬁes the
estimate: there exists a constant C
s
such that for any ) ∈ \
s
(p,q)
(Ω),
(5.2.1)  ·) 
2
s+1(Ω)
≤ C
s
 ) 
2
s(Ω)
,
where C
s
is independent of ).
In order to obtain the boundary regularity, we shall distinguish the tangential
derivatives from the normal derivatives. Restricting to a small neighborhood l
near a boundary point, we shall choose special boundary coordinates t
1
, , t
2n−1
, ρ
such that t
1
, , t
2n−1
restricted to /Ω are coordinates for /Ω. Let 1
tj
= ∂,∂t
j
,
, = 1, , 2n − 1, and 1
ρ
= ∂,∂ρ. Thus 1
tj
’s are the tangential derivatives
on /Ω, and 1
ρ
is the normal derivative. For a multiindex α = (α
1
, , α
2n−1
),
where each α
i
is a nonnegative integer, 1
α
t
denotes the product of 1
tj
’s with order
[α[ = α
1
+ +α
2n−1
, i.e.,
1
α
t
= 1
α1
t1
1
α2n−1
t2n−1
.
For any n ∈ C
∞
0
(l ∩ Ω), we deﬁne the tangential Fourier transform for n in a
special boundary chart by
˜ n(τ, ρ) =
_
R
2n−1
c
−i¸t,τ)
n(t, ρ)dt,
where τ = (τ
1
, , τ
2n−1
) and ¸t, τ) = t
1
τ
1
+ + t
2n−1
τ
2n−1
. For each ﬁxed
−c < 0, we deﬁne Γ
= ¦. ∈ Ω[ ρ(.) = −c¦ and set
 n(, −c) 
2
s(Γ)
=
_
R
2n−1
(1 +[τ[
2
)
s
[˜ n(τ, −c)[
2
dτ.
5.2 Boundary Regularity for N and
¯
∂
∗
N 93
We deﬁne the tangential Sobolev norms [[[ [[[
s
by
[[[n[[[
2
s
=
_
R
2n−1
_
0
−∞
(1 +[τ[
2
)
s
[˜ n(τ, ρ)[
2
dρdτ
=
_
0
−∞
 n(, ρ) 
2
s(Γρ)
dρ.
As usual, tangential norms for forms are deﬁned as the sum of the norms of the
components. The operator Λ
s
t
for any : ∈ 1 is given by
Λ
s
t
n(t, ρ) = (2π)
1−2n
_
R
2n−1
c
i¸t,τ)
(1 +[τ[
2
)
s
2
˜ n(τ, ρ)dτ.
Using this notation and Plancherel’s theorem (see Theorem A.5 in the Appendix),
we have
[[[n[[[
s
=  Λ
s
t
n  .
The tangential norms have the following properties:
Properties for the tangential norms. Let \
s
t
denote the completion of n ∈
C
∞
0
(l ∩ Ω) under the [[[ [[[
s
norm. Then the following hold:
(1) When : is a nonnegative integer, we have
(5.2.2 i) [[[n[[[
s
≈
0≤]α]≤s
 1
α
t
n ,
where ≈ means that the two norms are equivalent.
(2) For any : ∈ 1, there exists a constant C such that
(5.2.2 ii) [[[1
α
t
)[[[
s
≤ C[[[)[[[
k+s
for [α[ = /,
where C is independent of ).
(3) The Schwarz inequality holds; for any : ≥ 0, ) ∈ \
s
t
and p ∈ \
−s
t
, we have
(5.2.2 iii) [(), p)[ ≤ C[[[)[[[
s
[[[p[[[
−s
,
where C = (2π)
1−2n
.
(4) Given two spaces \
s1
t
and \
s2
t
, where :
1
and :
2
are real numbers, :
1
:
2
, the
interpolation space [\
s1
t
, \
s2
t
]
θ
= \
s1(1−θ)+s2θ
t
for any 0 < θ < 1.
Properties (1) and (2) are easily checked from Plancherel’s Theorem (see Theorem
A.5 in the Appendix). (3) follows immediately from the deﬁnition. The proof of (4)
is the same as that for the usual Sobolev spaces (see Appendix B). The following
simple fact will also play a crucial rule in proving the boundary regularity.
94 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
Commutator of two operators. For any smooth diﬀerential operators ¹ and 1
of order /
1
and /
2
, /
1
and /
2
are nonnegative integers, the commutator [¹, 1] =
¹1 −1¹ is an operator of order /
1
+/
2
−1.
The proof follows directly from the deﬁnition. We shall denote Λ
1
t
n = Λ
t
n and
deﬁne for any real :,
[[[1n[[[
2
s
= [[[1
ρ
n[[[
2
s
+[[[Λ
t
n[[[
2
s
= [[[1
ρ
n[[[
2
s
+[[[n[[[
2
s+1
.
The norm [[[1n[[[
s
is stronger than the norm [[[n[[[
s+1
in general, since the normal
derivatives are not controlled in the tangential norms.
By ﬁxing a partition of unity, we can also deﬁne [[[ [[[
s(Ω
δ
)
for some tubular
neighborhood Ω
δ
= ¦. ∈ Ω [ −δ < ρ(.) < 0¦. Let l
i
, i = 1, 2, , 1, be boundary
coordinate patches such that each l
i
∩/Ω ,= ∅, ∪
K
i=1
l
i
covers Ω
δ
and there exists a
special boundary chart on each l
i
∩Ω. We choose η
i
∈ C
∞
0
(l
i
), i = 1, , 1, such
that
K
i=1
η
2
i
= 1 in Ω
δ
.
For each ﬁxed c, the norm  
s(Γ)
is deﬁned by a partition of unity.
We set
[[[n[[[
2
s(Ω
δ
)
=
_
0
−δ
 n(, ρ) 
2
s(Γρ)
dρ.
We choose a special boundary frame such that n
1
, , n
n
is an orthonormal basis
for (1,0)forms with n
n
= ∂ρ in a boundary patch l as before. Written in this
basis, a smooth (j, ¡)form supported in l ∩ Ω can be expressed as
) =
t
]I]=p,]J]=q
)
I,J
n
I
∧ ¯ n
J
,
where 1 = (i
1
, , i
p
) and J = (,
1
, , ,
q
) are increasing multiindices and n
I
=
n
i1
∧ ∧ n
ip
, ¯ n
J
= ¯ n
j1
∧ ∧ ¯ n
jq
. From Lemma 4.2.1 and (4.2.6
tt
), it follows
that
) ∈ T
(p,q)
if and only if )
I,J
= 0 on /Ω, whenever n ∈ J,
where T
(p,q)
= C
∞
(p,q)
(Ω)∩Dom(
¯
∂
∗
). Thus, the tangential derivatives preserve T
(p,q)
in the following sense:
Lemma 5.2.2. Let ) ∈ T
(p,q)
. Assume that ) is supported in l ∩ Ω and ) is
expressed in the special boundary frame. Let T be a ﬁrst order tangential diﬀerential
operator with smooth coeﬃcients acting componentwise. Then T) ∈ T
(p,q)
.
In order to obtain estimates for the
¯
∂Neumann operator on the boundary, our
ﬁrst observation is that when ) satisﬁes an elliptic system, then there is no distinc
tion between the tangential Sobolev norms and the Sobolev norms. We have the
following lemma:
5.2 Boundary Regularity for N and
¯
∂
∗
N 95
Lemma 5.2.3. Let Ω be a bounded domain with C
∞
boundary and let l be a
special boundary patch. Let ) ∈ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
) where the support of ) lies in
l ∩ Ω. The following conditions are equivalent:
(a)  ) 1
2
(Ω)
< ∞.
(b) [[[)[[[ 1
2
(Ω)
< ∞.
(c) [[[1)[[[
−
1
2
(Ω)
< ∞.
(d)  ) 
L
2
(bΩ)
< ∞.
Proof. It is obvious from the deﬁnition that (a) implies (b).
From the density lemma 4.3.2, we can assume ) ∈ C
1
(l ∩Ω) and the other cases
follow by approximation. To show that (b) implies (c), we note that
¯
∂⊕ϑ is elliptic.
We can express 1
ρ
) by the sum of the components of
¯
∂), ϑ) and the tangential
derivatives of ). Thus
[[[1
ρ
)[[[
−
1
2
≤ C([[[
¯
∂)[[[
−
1
2
+[[[ϑ)[[[
−
1
2
+[[[Λ
t
)[[[
−
1
2
)
≤ C(
¯
∂)  +  ϑ)  + [[[)[[[ 1
2
).
To see that (c) implies (d), we use
[
˜
)(τ, 0) [
2
=
_
0
−∞
1
ρ
[
˜
)(τ, ρ) [
2
dρ = Re
_
0
−∞
21
ρ
˜
)(τ, ρ)
˜
)(τ, ρ)dρ
≤ ¹
_
0
−∞
[
˜
)(τ, ρ) [
2
dρ +
1
¹
_
0
−∞
[ 1
ρ
˜
)(τ, ρ) [
2
dρ,
for any ¹ 0. Choosing ¹ = (1 + [τ[
2
)
1/2
and integrating over 1
2n−1
, we have
from Plancherel’s theorem,
 ) 
2
L
2
(bΩ)
=
_
1
2π
_
2n−1
_
R
2n−1
[
˜
)(τ, 0) [
2
dτ
≤ C
__
R
2n−1
_
0
−∞
(1 +[τ[
2
)
1
2
[
˜
)(τ, ρ) [
2
dρdτ
_
+C
__
R
2n−1
_
0
−∞
(1 +[τ[
2
)
−
1
2
[ 1
ρ
˜
)(τ, ρ) [
2
dρdτ
_
≤ C [[[1)[[[
2
−
1
2
(Ω)
.
Finally, (d) implies (a) follows from Lemma 5.1.6 and Lemma 4.3.2, since we
have for any ) ∈ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
),
 ) 
2
1
2
(Ω)
≤ C
__
bΩ
[)[
2
do+ 
¯
∂) 
2
Ω
+  ϑ) 
2
Ω
+  ) 
2
Ω
_
.
We remark that (c) implies (d) can be viewed as a version of the trace theorem
for Sobolev spaces. In general, a function satisfying (a) does not necessarily satisfy
(d) (see Theorem A.9 in the Appendix).
To prove Theorem 5.2.1, we ﬁrst prove a priori estimates; i.e., we assume that
·) is smooth up to the boundary.
96 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
Lemma 5.2.4. Let Ω be a bounded pseudoconvex domain with C
∞
boundary and
let ρ be a C
∞
deﬁning function. Choose δ 0, such that the tangential norms are
deﬁned on the ﬁxed tubular neighborhood Ω
δ
= ¦. ∈ Ω [ ρ(.) −δ¦. Then there
exists a constant C
k
, / = 1, 2, such that for any ) ∈ Dom() ∩ C
∞
(p,q)
(Ω),
(5.2.3)  ) 
k(Ω)
≤ C
k
( ) 
k−1(Ω)
+ [[[)[[[
k(Ω
δ
)
),
where C
k
is independent of ).
Proof. We use induction on / for / = 1, 2, . Since
k
m=0
[[[1
m
ρ
)[[[
k−m
≈  ) 
k
,
(5.2.3) is proved if we can show
(5.2.4)
k
m=0
[[[1
m
ρ
)[[[
k−m
≤ C
k
( ) 
k−1
+ [[[)[[[
k
).
We ﬁrst prove (5.2.4) under the assumption that ) is supported in Ω ∩ l where l
is a special coordinate chart near the boundary.
When / = 1, we use the same argument as in Lemma 5.2.2 to express 1
ρ
) by
the components of
¯
∂), ϑ) and the tangential derivatives of ). We have
 1
ρ
) 
2
≤ C
_

¯
∂) 
2
+  ϑ) 
2
+
2n−1
i=1
 1
ti
) 
2
_
≤ C
_
(), )) + [[[)[[[
2
1
_
≤ C( ) 
2
+ [[[)[[[
2
1
).
This proves (5.2.4) for / = 1. Assuming the lemma holds for / − 1, we shall show
that (5.2.4) holds for /. If : = 1, again we get
(5.2.5)
[[[1
ρ
)[[[
2
k−1
≤ C( [[[
¯
∂)[[[
2
k−1
+[[[ϑ)[[[
2
k−1
+
2n−1
i=1
[[[1
ti
)[[[
2
k−1
)
≤ C( [[[
¯
∂)[[[
2
k−1
+[[[ϑ)[[[
2
k−1
+[[[)[[[
2
k
).
Notice that
[[[
¯
∂)[[[
2
k−1
+[[[ϑ)[[[
2
k−1
=  Λ
k−1
t
¯
∂) 
2
+  Λ
k−1
t
ϑ) 
2
≤ C
_
_
0≤]α]≤k−1
 1
α
t
¯
∂) 
2
+
0≤]α]≤k−1
 1
α
t
ϑ) 
2
_
_
.
For any nonnegative integer /, let T
k
denote any tangential diﬀerential operator of
the form 1
α
t
, where [α[ = /. Using Lemma 5.2.2, we ﬁnd that T
k
) ∈ T
(p,q)
and
5.2 Boundary Regularity for N and
¯
∂
∗
N 97
T
k ¯
∂) ∈ T
(p,q+1)
since ) ∈ Dom(). We see that
(5.2.6)
(T
k−1
¯
∂), T
k−1
¯
∂)) + (T
k−1
ϑ), T
k−1
ϑ))
= (
¯
∂T
k−1
), T
k−1
¯
∂)) + (
_
T
k−1
,
¯
∂
¸
), T
k−1
¯
∂))
+ (ϑT
k−1
), T
k−1
ϑ)) + (
_
T
k−1
, ϑ
¸
), T
k−1
ϑ))
= (T
k−1
), ϑT
k−1
¯
∂)) + (
_
T
k−1
,
¯
∂
¸
), T
k−1
¯
∂))
+ (T
k−1
),
¯
∂T
k−1
ϑ)) + (
_
T
k−1
, ϑ
¸
), T
k−1
ϑ))
= (T
k−1
), T
k−1
ϑ
¯
∂)) + (
_
T
k−1
,
¯
∂
¸
), T
k−1
¯
∂))
+ (T
k−1
), T
k−1
¯
∂ϑ)) + (
_
T
k−1
, ϑ
¸
), T
k−1
ϑ))
+ (T
k−1
),
_
ϑ, T
k−1
¸
¯
∂)) + (T
k−1
),
_
¯
∂, T
k−1
¸
ϑ))
= (T
k−1
), T
k−1
)) +¹+O( ) 
k−1
([[[
¯
∂)[[[
k−1
+[[[ϑ)[[[
k−1
)),
where
¹ = (T
k−1
),
_
ϑ, T
k−1
¸
¯
∂)) + (T
k−1
),
_
¯
∂, T
k−1
¸
ϑ)).
The term (T
k−1
),
_
ϑ, T
k−1
¸
¯
∂)) in ¹ can be estimated by
(5.2.7)
(T
k−1
),
_
ϑ, T
k−1
¸
¯
∂))
= (T
k−1
),
¯
∂
_
ϑ, T
k−1
¸
)) + (T
k−1
),
__
ϑ, T
k−1
¸
,
¯
∂
¸
))
= (ϑT
k−1
),
_
ϑ, T
k−1
¸
)) + (T
k−1
),
__
ϑ, T
k−1
¸
,
¯
∂
¸
))
= (T
k−1
ϑ),
_
ϑ, T
k−1
¸
)) + (
_
ϑ, T
k−1
¸
),
_
ϑ, T
k−1
¸
))
+ (T
k−1
),
__
ϑ, T
k−1
¸
,
¯
∂
¸
)).
Similarly, one can estimate the term (T
k−1
),
_
¯
∂, T
k−1
¸
ϑ)) in ¹. Thus [¹[ can be
estimated by
(5.2.8) C( ) 
k−1
[[[
¯
∂)[[[
k−1
+  ) 
k−1
[[[ϑ)[[[
k−1
+  ) 
2
k−1
).
If we apply (5.2.6), (5.2.7) to each term of the form T
]α]
= 1
α
t
, where 0 ≤ [α[ ≤ /−1,
we can conclude that
0≤]α]≤k−1
( 1
α
t
¯
∂) 
2
+  1
α
t
ϑ) 
2
)
≤ C( Λ
k−1
t
)  Λ
k−1
t
)  +  ) 
k−1
([[[
¯
∂)[[[
k−1
+[[[ϑ)[[[
k−1
) +  ) 
2
k−1
).
Using the inequality that o/ ≤ co
2
+(1,c)/
2
for any c 0, we see from (5.2.5) that
[[[
¯
∂)[[[
2
k−1
+[[[ϑ)[[[
2
k−1
≤ C( Λ
k−1
t
)  Λ
k−1
t
) 
+  ) 
k−1
([[[
¯
∂)[[[
k−1
+[[[ϑ)[[[
k−1
) +  ) 
2
k−1
)
≤ C[[[)[[[
k−1
[[[)[[[
k−1
+c([[[
¯
∂)[[[
2
k−1
+[[[ϑ)[[[
2
k−1
) +C
 ) 
2
k−1
.
98 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
Choosing c suﬃciently small, the induction hypothesis yields
(5.2.9) [[[
¯
∂)[[[
2
k−1
+[[[ϑ)[[[
2
k−1
≤ C( ) 
2
k−1
+ [[[)[[[
2
k−1
).
Substituting (5.2.9) back into (5.2.5), we have proved (5.2.4) when : = 1. For
: ≥ 2, we shall repeat the procedure and use induction on 1 ≤ : ≤ /.
Since is a constant multiple of the Laplacian operator on each component, we
can express 1
2
ρ
) by the sum of the components of terms of the form
), 1
ρ
1
tj
), 1
ti
1
tj
), i, , = 1, , 2n −1,
and lower order terms. If : 2, diﬀerentiation shows that one can express 1
m
ρ
)
by the sum of terms of the form
1
m−2
ρ
), 1
m−1
ρ
1
tj
), 1
m−2
ρ
1
ti
1
tj
), i, , = 1, , 2n −1,
and lower order terms. Assuming that (5.2.4) holds for 1, , : − 1, we use the
induction hypothesis to show that
(5.2.10)
[[[1
m
ρ
)[[[
k−m
≤ C( [[[1
m−2
ρ
)[[[
k−m
+
2n−1
i=1
[[[1
m−1
ρ
1
ti
)[[[
k−m
+
2n−1
i,j=1
[[[1
m−2
ρ
1
ti
1
tj
)[[[
k−m
+  ) 
k−1
)
≤ C( ) 
k−2
+ [[[1
m−1
ρ
)[[[
k−m+1
+ [[[1
m−2
ρ
)[[[
k−m+2
+  ) 
k−1
)
≤ C( ) 
k−2
+  ) 
k−1
+ [[[)[[[
k
+  ) 
k−1
)
≤ C( ) 
k−1
+ [[[)[[[
k
).
Thus, (5.2.4) holds for all : ≤ /. This ﬁnishes the proof of Lemma 5.2.4 when ) has
compact support in a coordinate patch. The general case follows from a partition of
unity. We shall do this in detail for the case when / = 1. Letting η ∈ C
∞
(Ω ∩ l),
we have that
 η) 
2
1
≤ C( 
¯
∂(η)) 
2
+  ϑ(η)) 
2
+ [[[)[[[
2
1(Ω
δ
)
)
≤ C(  η
¯
∂) 
2
+  ηϑ) 
2
+  ) 
2
+ [[[)[[[
2
1(Ω
δ
)
)
≤ C((η
2
), ))
Ω
+c(
¯
∂(η)) 
2
+  ϑ(η)) 
2
)
+C
 ) 
2
+ [[[)[[[
2
1(Ω
δ
)
).
If we choose c suﬃciently small, then the term
c(
¯
∂(η)) 
2
+  ϑ(η)) 
2
) ≤ c  η) 
2
1
can be absorbed by the lefthand side. It follows that
 η) 
2
1
≤ C( η)  η)  +  ) 
2
+ [[[)[[[
2
1(Ω
δ
)
)
≤ C( η) 
2
+  ) 
2
+ [[[)[[[
2
1(Ω
δ
)
)
≤ C( ) 
2
+ [[[)[[[
2
1(Ω
δ
)
).
5.2 Boundary Regularity for N and
¯
∂
∗
N 99
In the last inequality above, we have used  )  ≤ C  )  which was proved in
Theorem 4.4.1 under the assumption of pseudoconvexity.
If η
0
∈ C
∞
0
(Ω), using Proposition 5.1.1, one sees that
 η
0
) 
2
1
≤ C  ) 
2
.
Summing over a partition of unity η
i
, i = 0, , 1 such that η
0
∈ C
∞
0
(Ω), each η
i
is supported in a boundary coordinate patch and
K
i=0
η
2
i
= 1, we have proved
 ) 
2
1
≤ C( ) 
2
+ [[[)[[[
2
1(Ω
δ
)
).
The lemma is proved when / = 1. The other cases are proved similarly by using a
partition of unity and induction.
Proposition 5.2.5. Let Ω ⊂⊂ C
n
be a bounded strongly pseudoconvex domain with
C
∞
deﬁning function. Choose δ 0, such that the tangential norms are deﬁned on
the ﬁxed tubular neighborhood Ω
δ
= ¦. ∈ Ω [ ρ(.) −δ¦. For each / = 0, 1, 2, ,
there exists a constant C
k
such that for any ) ∈ Dom() ∩ C
∞
(p,q)
(Ω),
(5.2.11) [[[)[[[
2
k+
1
2
(Ω
δ
)
≤ C
k
 ) 
2
k−
1
2
(Ω)
,
where C
k
is independent of ).
Proof. We shall prove the proposition by induction on / = 0, 1, . When / = 0,
this is already proved in Theorem 5.1.3. Thus we have
(5.2.12)  ) 1
2
≤ C  ) 
−
1
2
.
(We even have the actual estimates instead of just a priori estimates.) Assume that
(5.2.11) holds for / −1. Let l be a special coordinate patch. We ﬁrst assume that
) is supported in Ω∩l and written in the special frame as before. Let T
k
be a /th
order tangential diﬀerential operator of the form 1
α
t
where [α[ = /. From Lemma
5.2.2, we know that T
k
) ∈ T
(p,q)
. Substituting T
k
) into the estimate (5.1.4), we
see that
(5.2.13)  T
k
) 
2
1
2
≤ C(
¯
∂T
k
) 
2
+  ϑT
k
) 
2
).
Using arguments similar to those in (5.2.6) and (5.2.7), it follows that
(5.2.14)

¯
∂T
k
) 
2
+  ϑT
k
) 
2
=  T
k
¯
∂) 
2
+  T
k
ϑ) 
2
+ O( ) 
2
k
)
= (T
k
), T
k
)) +O( ) 
k
[[[
¯
∂)[[[
k
+  ) 
k
[[[ϑ)[[[
k
+  ) 
2
k
).
Applying the inequality o/ ≤ co
2
+
1
/
2
again and (5.2.2 ii), (5.2.2 iii), we obtain
(5.2.15)
[ (T
k
), T
k
)) [ ≤ C[[[T
k
)[[[ 1
2
[[[T
k
)[[[
−
1
2
≤ C
[[[)[[[
2
k−
1
2
+c[[[T
k
)[[[
2
1
2
≤ C
[[[)[[[
2
k−
1
2
+c[[[)[[[
2
k+
1
2
.
100 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
Now since
[[[
¯
∂)[[[
k
+[[[ϑ)[[[
k
≤ C
]α]≤k
( 1
α
t
¯
∂)  +  1
α
t
ϑ) )
≤ C
]α]≤k
(
¯
∂1
α
t
)  +  ϑ1
α
t
) ) +O( ) 
k
),
it follows from from (5.2.14) that
(5.2.16)
]α]≤k
(
¯
∂1
α
t
) 
2
+  ϑ1
α
t
) 
2
) ≤ C
_
]α]≤k
(1
α
t
), 1
α
t
)) +  ) 
2
k
_
.
Combining (5.2.13)(5.2.16) and summing up all the tangential derivatives of the
form 1
α
t
, where [α[ ≤ /, we deduce that
(5.2.17)
[[[)[[[
2
k+
1
2
≤ C
_
]α]≤k
[[[1
α
t
)[[[ 1
2
[[[1
α
t
)[[[
−
1
2
+  ) 
2
k
_
≤ c[[[)[[[
2
k+
1
2
+C
[[[)[[[
2
k−
1
2
+C  ) 
2
k
.
Using the interpolation inequality for Sobolev spaces (see Theorem B.2 in the Ap
pendix), for any c
t
0 there exists a C
such that
(5.2.18) [[[)[[[
k
≤ c
t
[[[)[[[
k+
1
2
+C
[[[)[[[ 1
2
.
Applying Lemma 5.2.4, we observe that
 ) 
k
≤ C( [[[)[[[
k
+  ) 
k−1
)
≤ C(c
t
[[[)[[[
k+
1
2
+C
[[[)[[[ 1
2
+  ) 
k−1
).
Choosing ﬁrst c and then c
t
suﬃciently small, using (5.2.17) and (5.2.18), one obtains
[[[)[[[
2
k+
1
2
≤ C( ) 
2
k−
1
2
+  ) 
2
1
2
).
From (5.2.12), we have established
(5.2.19) [[[)[[[
2
k+
1
2
≤ C  ) 
2
k−
1
2
,
when ) is supported in a special coordinate patch.
The general case will be derived from a partition of unity. Let η, η
t
∈ C
∞
0
(l)
and η
t
= 1 on the support of η. We have
[[[T
k
η)[[[
2
1
2
≤ C( [[[ηT
k
)[[[
2
1
2
+[[[η
t
)[[[
2
k−
1
2
)
≤ C(
¯
∂ηT
k
) 
2
+  ϑηT
k
) 
2
+ [[[η
t
)[[[
2
k−
1
2
).
Repeating the previous argument with ηT
k
substituted for T
k
, we see that
[[[η)[[[
2
k+
1
2
≤ c[[[η
t
)[[[
2
k+
1
2
+C
[[[)[[[
2
k−
1
2
+C  η
t
) 
2
k
.
5.2 Boundary Regularity for N and
¯
∂
∗
N 101
Now for any η
0
∈ C
∞
0
(Ω), we already know that
 η
0
) 
k+
3
2
≤ C  ) 
k−
1
2
.
Summing over a partition of unity η
i
, i = 1, , 1 for the tubular neighborhood
Ω
δ
, yields that for some η
0
∈ C
∞
0
(Ω),
[[[)[[[
2
k+
1
2
(Ω
δ
)
≤ C( ) 
2
k−
1
2
+  η
0
) 
2
k
) ≤ C  ) 
2
k−
1
2
.
This proves the proposition.
Proof of Theorem 5.2.1. Using Theorem 5.1.3, we already know that Theorem 5.2.1
holds when : = −1,2 and
(5.2.20)  ·) 1
2
(Ω)
≤ C  ) 
−
1
2
(Ω)
, for any ) ∈ \
−
1
2
(p,q)
(Ω).
We shall prove the theorem for : = / when / ∈ N. Since C
∞
(p,q)
(Ω) is dense in
\
s
(p,q)
(Ω), it suﬃces to prove the following estimates:
(5.2.21)  ·) 
s+
1
2
(Ω)
≤ C  ) 
s−
1
2
(Ω)
, for any ) ∈ C
∞
(p,q)
(Ω).
When : is a nonnegative integer, (5.2.21) has already been established in Proposition
5.2.5 and Lemma 5.2.4 assuming that ·) is smooth up to the boundary. To pass
from a priori estimates to the real estimates, we can use the following elliptic
regularization method:
Let Q be deﬁned by
Q(p, p) = 
¯
∂p
2
+
¯
∂
∗
p
2
, p ∈ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
).
We deﬁne
Q
(p, p) = Q(p, p) +c  ∇p 
2
, p ∈ T
,
where T
is the completion of T
(p,q)
under the Q
norm. From inequality (4.4.6),
we see that
(5.2.22) Q
(p, p) ≥ C  p 
2
for every p ∈ T
,
where C 0 is independent of c (C can be chosen as cδ
2
,¡ where δ is the diameter
of Ω). Thus, for any ) ∈ 1
2
(p,q)
(Ω), p ∈ T
, we can deduce that
(5.2.23) [(), p)[ ≤ C
−
1
2
 )  Q
(p, p)
1
2
.
This implies that the map from p →(), p) is a bounded conjugate linear functional
on T
. By the Riesz representation theorem, there exists an element ·
) ∈ T
such that
(), p) = Q
(·
), p) for every p ∈ T
.
102 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
Moreover, we have
 ·
)  ≤ C  ) ,
where C is the same constant as in (5.2.22). Note that Q
satisﬁes G˚arding’s
inequality
Q
(), )) ≥ c  ) 
2
1
for every ) ∈ T
.
Thus, the bilinear form Q
is elliptic on T
and we can use the theory for elliptic
boundary value problems on a smooth domain to conclude that ·
) ∈ C
∞
(p,q)
(Ω) if
) ∈ C
∞
(p,q)
(Ω). Applying the a priori estimates (5.2.11) to the form ·
), we get
(5.2.24) [[[·
)[[[
k+
1
2
(Ω
δ
)
≤ C
k
 ) 
k−
1
2
(Ω)
,
where C
k
is independent of c. The interpolation theorem for the operator ·
on
Sobolev spaces \
s
(Ω) and tangential Sobolev spaces \
s
t
(Ω
δ
) (see Theorem B.3 in
the Appendix) gives
[[[·
)[[[
k(Ω
δ
)
≤ C  ) 
k−1(Ω)
for / = 1, 2, .
Repeating the argument of Lemma 5.2.4, we obtain
 ·
) 
k(Ω)
≤ C  ) 
k−1(Ω)
for / = 1, 2, .
Thus, a subsequence of ·
) converges weakly in \
k
(p,q)
(Ω) to some element β ∈
\
k
(p,q)
(Ω). We claim that β = ·).
For any p ∈ T
(p,q)
,
(), p) = Q(·), p) = Q
(·
), p).
It follows from the deﬁnition of Q
that for any p ∈ T
(p,q)
,
[Q(·
) −·), p)[ ≤ c  ·
) 
1
 p 
1
≤ cC  )  p 
1
→0
as c → 0. Since T
(p,q)
is dense in Dom(
¯
∂) ∩ Dom(
¯
∂
∗
) = T from Lemma 4.3.2, we
see that
Q(·
) −·), p) →0 for every p ∈ T.
Thus, ·
) converges to ·) weakly in the Qnorm. But (a subsequence of) ·
) →β
weakly in the \
k
norm. Therefore, we must have
·) = β
and
 ·) 
k
≤ liminf  ·
) 
k
≤ C  ) 
k−1
, / = 1, 2, .
Thus (5.2.21) is proved for : = 0 and : =
1
2
+/, / = 0, 1, 2, . Using the interpo
lation theorem for operators on Sobolev spaces again, we have for any : ≥ −
1
2
,
 ·) 
s+1
≤ C
s
 ) 
s
, ) ∈ \
s
(p,q)
(Ω).
This proves Theorem 5.2.1.
5.2 Boundary Regularity for N and
¯
∂
∗
N 103
Theorem 5.2.6. Let Ω ⊂⊂ C
n
be a bounded strongly pseudoconvex domain with
C
∞
boundary. Then
¯
∂
∗
· and
¯
∂· are bounded operators from \
s
(p,q)
(Ω) into
\
s+1/2
(p,q−1)
(Ω) and \
s+1/2
(p,q+1)
(Ω) respectively, where : ≥ −1,2, 0 ≤ j ≤ n and 1 ≤
¡ ≤ n −1. There exists a constant C
s
such that for any ) ∈ \
s
(p,q)
(Ω),
(5.2.25) 
¯
∂
∗
·) 
2
s+
1
2
(Ω)
+ 
¯
∂·) 
2
s+
1
2
(Ω)
≤ C
s
 ) 
2
s(Ω)
,
where C
s
is independent of ).
Proof. When : = −1,2, we have shown in Theorem 5.1.3 that

¯
∂
∗
·) 
2
+ 
¯
∂·) 
2
= (
¯
∂
¯
∂
∗
·), ·)) + (
¯
∂
∗
¯
∂·), ·)) = (), ·))
≤  ) 
−
1
2
 ·) 1
2
≤ C  ) 
2
−
1
2
.
We shall prove by induction that
(5.2.26) 
¯
∂
∗
·) 
2
k
+ 
¯
∂·) 
2
k
≤ C
k
 ) 
2
k−
1
2
,
for / = 1, 2, . Let η be a cutoﬀ function. If η is supported in a boundary
coordinate patch, let T
k
denote the same /th order tangential diﬀerential operator
of the form η1
α
t
, [α[ = /. Observe that
(5.2.27)
(T
k
¯
∂
∗
·), T
k
¯
∂
∗
·)) + (T
k
¯
∂·), T
k
¯
∂·))
= (T
k
¯
∂
∗
·),
¯
∂
∗
T
k
·)) + (T
k
¯
∂·),
¯
∂T
k
·))
+ (T
k
¯
∂
∗
·), [T
k
,
¯
∂
∗
]·)) + (T
k
¯
∂·), [T
k
,
¯
∂]·))
≤ C
_
(T
k
¯
∂
¯
∂
∗
·), T
k
·)) + (T
k
¯
∂
∗
¯
∂·), T
k
·)) +1
_
≤ C(T
k
·), T
k
·)) +1)
≤ C([[[T
k
)[[[
−
1
2
[[[T
k
·)[[[ 1
2
+1)
≤ C([[[)[[[
k−
1
2
[[[·)[[[
k+
1
2
+1)
≤ C([[[)[[[
2
k−
1
2
+1),
where 1 denotes terms which can be bounded by
(5.2.28)
C
_
 ·) 
k
([[[
¯
∂
∗
·)[[[
k
+[[[
¯
∂·)[[[
k
) +  ·) 
2
k
_
≤ c([[[
¯
∂
∗
·)[[[
2
k
+[[[
¯
∂·)[[[
2
k
) +C
 ·) 
2
k
.
Using a partition of unity and summing up all the tangential derivatives up to order
/ in (5.2.27), we obtain, using (5.2.28), that
[[[
¯
∂
∗
·)[[[
2
k
+[[[
¯
∂·)[[[
2
k
≤ C( ) 
2
k−1
+ [[[)[[[
2
k−
1
2
),
if c in (5.2.28) is chosen suﬃciently small.
Again, using the fact
¯
∂ ⊕ ϑ is an elliptic system, the normal derivative can be
expressed as the linear combination of terms which have been estimated before.
The interior regularity is easier. This proves the inequality (5.2.26). The theorem
follows from interpolation.
104 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
Corollary 5.2.7. Let Ω ⊂⊂ C
n
be a bounded strongly pseudoconvex domain with
C
∞
boundary and 0 ≤ j ≤ n and 0 ≤ ¡ ≤ n − 1. Then · maps C
∞
(p,q)
(Ω) into
C
∞
(p,q)
(Ω). In particular, the Bergman projection 1 maps C
∞
(Ω) into C
∞
(Ω). Also
if ) ∈ C
∞
(p,q)
(Ω) and
¯
∂)=0, the canonical solution n =
¯
∂
∗
·) ∈ C
∞
(p,q−1)
(Ω).
Proof. The corollary is an easy consequence from the Sobolev embedding theorem
for ¡ ≥ 1. For ¡ = 0, we use (4) in Theorem 4.4.3. The regularity of the Bergman
projection follows from the formula 1) = ) −
¯
∂
∗
·
¯
∂) = ) −ϑ·
¯
∂).
In Chapter 6, we prove a more precise result for the Bergman projection. In fact,
1 preserve \
s
(Ω) for all : ≥ 0 (see Theorem 6.2.2).
5.3 Function Theory on Manifolds
Let ` be a complex manifold of dimension n (For the deﬁnition of complex
manifolds, see Chapter 1). The decomposition of diﬀerential forms into forms
of type (j, ¡), the deﬁnition of the
¯
∂ operator and the deﬁnition of plurisubhar
monic functions for domains in C
n
can immediately be extended to forms and
functions on the complex manifold `. In order to study the operator
¯
∂ with
Hilbert space techniques, we must equip ` with a Hermitian metric such that
CT(`) = T
1,0
(`) ⊕ T
0,1
(`) and T
1,0
(`) ⊥ T
0,1
(`). A Hermitian metric in
local coordinates .
1
, , .
n
is of the form
n
i,j=1
/
ij
d.
i
⊗d¯ .
j
,
where /
ij
is a positive deﬁnite Hermitian matrix with C
∞
coeﬃcients. The existence
of a Hermitian metric is trivial locally and is proved globally by a partition of unity.
We ﬁx a Hermitian metric in all that follows. This induces an inner product in
C
∞
(p,q)
(`) for each j ∈ `. If φ, ψ ∈ C
∞
(p,q)
(`), this inner product is denoted by
¸φ, ψ). We have the following deﬁnition:
Deﬁnition 5.3.1. Let j ∈ ` and φ ∈ C
2
(`). If 1 ∈ T
1,0
p
(`), the complex
Hessian of φ at j is deﬁned to be the Hermitian form
1 →(∂
¯
∂φ)
p
(1 ∧ 1).
The function φ is called plurisubharmonic at j if the complex Hessian is positive
semideﬁnite. φ is called strictly plurisubharmonic at j if the complex Hessian is
positive deﬁnite.
Let Ω be an open subset in ` whose closure is compact in `, i.e., Ω is relatively
compact in ` and denoted by Ω ⊂⊂ `. Ω is called a complex manifold with
C
k
boundary /Ω if there exists a neighborhood \ of Ω and a realvalued function
ρ ∈ C
k
(\ ) such that Ω = ¦. ∈ \ [ ρ < 0¦, ρ 0 in \ ¸ Ω and [dρ[ ,= 0 on /Ω. Let
CT
p
(/Ω) be the complexiﬁed tangent bundle of /Ω at j.
5.3 Function Theory on Manifolds 105
Deﬁnition 5.3.2. Let Ω be a complex manifold with C
2
boundary and ρ be a C
2
deﬁning function. Ω is called pseudoconvex (strictly pseudoconvex) if for each j ∈
/Ω, the restriction of the complex Hessian of ρ to T
1,0
p
(`) ∩ CT
p
(/Ω) is positive
semideﬁnite (positive deﬁnite).
In local coordinates, by the usual GramSchmidt orthogonalization process we
can choose an orthonormal basis n
1
, , n
n
for (1,0)forms locally on a suﬃciently
small neighborhood l such that ¸n
i
, n
k
) = δ
ik
, i, / = 1, , n. Then written in
this basis, for any n ∈ C
1
(l), we can write
dn =
n
i=1
∂n
∂n
i
n
i
+
n
i=1
∂n
∂ ¯ n
i
¯ n
i
,
where the ﬁrst order linear diﬀerential operators ∂,∂n
i
and ∂,∂ ¯ n
i
are duals of n
i
and ¯ n
i
respectively. Then we have
¯
∂n =
n
i=1
∂n
∂ ¯ n
i
¯ n
i
.
If ) is a (j, ¡)form on l, then we can write ) as
(5.3.1) ) =
t
]I]=p,]J]=q
)
I,J
n
I
∧ ¯ n
J
,
where 1 = (i
1
, , i
p
) and J = (,
1
, , ,
q
) are multiindices and n
I
= n
i1
∧ ∧n
ip
,
¯ n
J
= ¯ n
j1
∧ ∧ ¯ n
jq
. If n ∈ C
2
(l), we set n
ij
to be the coeﬃcients of ∂
¯
∂n, i.e.,
(5.3.2) ∂
¯
∂n =
i,j
n
ij
n
i
∧ ¯ n
j
.
Let c
i
jk
be the smooth functions such that
¯
∂n
i
=
j,k
c
i
jk
¯ n
j
∧ n
k
.
Then n
ij
can be calculated as follows:
∂
¯
∂n = ∂
_
k
∂n
∂ ¯ n
k
¯ n
k
_
=
j,k
_
∂
2
n
∂n
j
∂ ¯ n
k
+
i
∂n
∂ ¯ n
i
¯ c
i
jk
_
n
j
∧ ¯ n
k
.
From the fact that ∂
¯
∂ +
¯
∂∂ = 0, we have
(5.3.3) n
jk
=
∂
2
n
∂n
j
∂ ¯ n
k
+
i
∂n
∂ ¯ n
i
¯ c
i
jk
=
∂
2
n
∂ ¯ n
k
∂n
j
+
i
∂n
∂n
i
c
i
kj
.
A function φ ∈ C
2
is plurisubharmonic (strictly plurisubharmonic) if the form
n
j,k=1
φ
jk
o
j
¯ o
k
, o = (o
1
, , o
n
) ∈ C
n
,
106 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
is positive semideﬁnite (positive deﬁnite).
We shall normalize ρ such that [dρ[ = 1 on /Ω. Ω is pseudoconvex at a point
. on /Ω if there exists a neighborhood l of . and a local (1,0) orthonormal frame
n
1
, , n
n
such that
(5.3.4)
n
j,k=1
ρ
jk
o
j
¯ o
k
≥ 0 if
n
j=1
o
j
∂ρ(.)
∂n
j
= 0.
Here o = (o
1
, , o
n
) is a vector in C
n
. If the Hermitian form is strictly positive
for all such o ,= 0, the boundary is strongly pseudoconvex at ..
Note that these deﬁnitions are independent of the choice of the deﬁning function
ρ and are independent of the choice of n
1
, , n
n
. If we choose a special boundary
chart such that n
n
= ∂ρ, then ∂,∂n
i
, i = 1, , n − 1 are tangential operators.
We have, substituting ρ for n in (5.3.3),
(5.3.5) ∂
¯
∂ρ =
¯
∂n
n
=
j,k
ρ
jk
n
j
∧ ¯ n
k
,
where (ρ
jk
) = (¯ c
n
jk
) = (c
n
kj
) is the Levi matrix. In this case, /Ω is pseudoconvex if
and only if (ρ
jk
)
n−1
j,k=1
is positive semideﬁnite.
We shall use the same Hilbert space theory as that in Chapter 4 to study the
function theory on pseudoconvex manifolds. We ﬁx a function φ ∈ C
2
(Ω). Let
¯
∂ : 1
2
(p,q)
(Ω, φ) → 1
2
(p,q+1)
(Ω, φ) be the closure of the CauchyRiemann operator
and we deﬁne the Hilbert space adjoint for
¯
∂
∗
φ
as before. Let .
0
∈ /Ω be a boundary
point and l be an open neighborhood of .
0
. We shall ﬁx a special orthonormal
boundary frame n
1
, , n
n
= ∂ρ. Writing 1
i
= ∂,∂n
i
, then 1
1
, , 1
n
are dual
to the (1,0)forms n
1
, , n
n
and we have
(5.3.6)
_
1
i
(ρ) = 0, when . ∈ /Ω ∩ l, i = 1, , n −1,
1
n
(ρ) = 1, when . ∈ /Ω ∩ l.
We compute
¯
∂) and ϑ) in this special coordinate chart. We can write any ) ∈
C
∞
(p,q)
(Ω ∩ l) as ) =
t
]I]=p,]J]=q
)
I,J
n
I
∧ ¯ n
J
. Then
(5.3.7)
) ∈ C
∞
0
(Ω ∩ l) ∩ Dom(
¯
∂
∗
φ
)
if and only if )
I,J
= 0 whenever n ∈ J,
where C
∞
0
(l ∩Ω) denotes the space of functions in C
∞
(Ω) which are supported in
l ∩ Ω.
We denote the space C
∞
(p,q)
(Ω) ∩Dom(
¯
∂
∗
φ
) by T
(p,q)
and C
(p,q)
(Ω) ∩Dom(
¯
∂
∗
φ
) by
T
(p,q)
, / ∈ N, as before. It follows that
(5.3.8) ∂) =
I,J
t
j
∂)
I,J
∂n
j
n
j
∧ n
I
∧ n
J
+ = ¹) + ,
5.3 Function Theory on Manifolds 107
and
(5.3.9) ϑ
φ
) = (−1)
p−1
I,K
t
j
δ
φ
j
)
I,jK
n
I
∧ n
K
+ = 1) + ,
where δ
φ
j
n = c
φ
1
j
(c
−φ
n) and dots indicate terms where no derivatives of )
I,J
occur and which do not involve φ. The second equalities in (5.3.8) and (5.3.9) are
deﬁnitions of ¹ and 1 which are ﬁrst order diﬀerential operators.
Thus, we have
(5.3.10)
∂)
2
φ
+ ϑ
φ
)
2
φ
=
I,J,L
t
j,
c
jJ
L
(1
j
()
I,J
), 1
()
I,L
))
φ
+
I,K
t
j,k
(δ
φ
j
)
I,jK
, δ
φ
k
)
I,kK
)
φ
+1()),
where 1()) involves terms that can be controlled by O(( ¹) 
φ
+  1) 
φ
)  ) 
φ
)
and c
jJ
L
is deﬁned as before. Rearranging the terms in (5.3.10), we have
(5.3.11)
∂)
2
φ
+ ϑ
φ
)
2
φ
=
I,J
t
j
1
j
)
I,J

2
φ
−
I,K
t
j,k
(1
k
)
I,jK
, 1
j
)
I,kK
)
φ
+
I,K
t
j,k
(δ
φ
j
)
I,jK
, δ
φ
k
)
I,kK
)
φ
+1()).
We apply integration by parts to the terms (δ
φ
j
)
I,jK
, δ
φ
k
)
I,kK
)
φ
. For each n, · ∈
C
1
0
(Ω ∩ l), Green’s formula gives
(5.3.12) (n, δ
φ
j
υ)
φ
= −(1
j
n, υ)
φ
+ (σ
j
n, ·)
φ
+
_
bΩ
(1
j
ρ)n·c
−φ
do,
where do is the surface element on /Ω and σ
j
is in C
1
(Ω∩l). The boundary term
in (5.3.12) will vanish if , < n from (5.3.6). If ) ∈ T
(p,q)
, when we apply (5.3.12)
to the terms (δ
φ
j
)
I,jK
, δ
φ
k
)
I,kK
)
φ
, no boundary terms arise since
(5.3.13) )
I,J
= 0 on /Ω, if n ∈ J,
and if , < n and n ∈ J,
(5.3.14) 1
j
(ρ) = 1
j
(ρ) = 1
j
()
I,J
) = 1
j
()
I,J
) = 0 on /Ω.
In order to calculate the commutator [δ
φ
j
, 1
k
], we use (5.3.2) and (5.3.3),
(5.3.15)
[δ
φ
j
, 1
k
] n = [1
j
−1
j
(φ), 1
k
]n = [1
j
, 1
k
]n +1
k
1
j
(φ)n
=
i
c
i
kj
1
i
(n) −
i
¯ c
i
jk
1
i
(n) +1
k
1
j
(φ)n
= φ
jk
n +
i
c
i
kj
δ
φ
i
(n) −
i
¯ c
i
jk
1
i
(n).
108 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
Using (5.3.12)(5.3.15), for each ﬁxed 1, 1, ,, /, we have
(5.3.16)
(δ
φ
j
)
I,jK
, δ
φ
k
)
I,kK
)
φ
= (−1
k
δ
φ
j
)
I,jK
, )
I,kK
)
φ
+ (δ
φ
j
)
I,jK
, ¯ σ
k
)
I,kK
)
φ
= (1
k
)
I,jK
, 1
j
)
I,kK
)
φ
+ ([δ
φ
j
, 1
k
])
I,jK
, )
I,kK
)
φ
−(1
k
)
I,jK
, σ
j
)
I,kK
)
φ
+ (δ
φ
j
)
I,jK
, ¯ σ
k
)
I,kK
)
φ
.
In the above calculation, no boundary terms arise since ) ∈ T
(p,q)
and by (5.3.13)
and (5.3.14). Introducing the notation
 1) 
2
φ
=
I,J
t
j
 1
j
)
I,J

2
φ
+  ) 
2
φ
and applying integration by parts to the last terms of (5.3.16), we see from (5.3.13),
(5.3.14) that
(5.3.17) [(δ
φ
j
)
I,jK
, ¯ σ
k
)
I,kK
)
φ
[ ≤ C  1) 
φ
 ) 
φ
,
where C is a constant independent of φ. We shall use O( 1) 
φ
 ) 
φ
) to denote
terms which are bounded by C  1) 
φ
 ) 
φ
where C is a constant independent
of φ. Thus, (5.3.16) reads
(5.3.18)
(δ
φ
j
)
I,jK
, δ
φ
k
)
I,kK
)
= (1
k
)
I,jK
, 1
j
)
I,kK
)
φ
+ ([δ
φ
j
, 1
k
])
I,jK
, )
I,kK
)
φ
+O( 1) 
φ
 ) 
φ
)
= (1
k
)
I,jK
, 1
j
)
I,kK
)
φ
+ (φ
jk
)
I,jK
, )
I,kK
)
φ
+
_
i
c
i
kj
δ
φ
i
)
I,jK
, )
I,kK
_
φ
+O( 1) 
φ
 ) 
φ
).
If i < n, integration by parts gives
[(c
i
kj
δ
φ
i
)
I,jK
, )
I,kK
)
φ
[ ≤ C  1) 
φ
 ) 
φ
.
If i = n, we get, using (5.3.5),
(5.3.19)
(c
n
kj
δ
φ
n
)
I,jK
, )
I,kK
)
φ
=
_
bΩ∩U
c
n
kj
)
I,jK
¯
)
I,kK
c
−φ
do +O( 1) 
φ
 ) 
φ
)
=
_
bΩ∩U
ρ
jk
)
I,jK
¯
)
I,kK
c
−φ
do +O( 1) 
φ
 ) 
φ
).
Combining (5.3.11), (5.3.18) and (5.3.19), we obtain
(5.3.20)
∂)
2
φ
+ϑ
φ
)
2
φ
=
I,J
t
j
1
j
)
I,J

2
φ
+
I,K
t
j,k
(φ
jk
)
I,jK
, )
I,kK
)
φ
+
I,K
t
j,k
_
bΩ∩U
ρ
jk
)
I,jK
¯
)
I,kK
c
−φ
do +1()) +1()),
5.3 Function Theory on Manifolds 109
where [1())[ ≤ C( 1) 
φ
 ) 
φ
). Also for any c 0, there exists a C
0 such
that
(5.3.21) [1())[ ≤ c(∂)
2
φ
+ ϑ
φ
)
2
φ
) +C
)
2
φ
,
where C
is independent of φ. Thus combining (5.3.20) and (5.3.21), we have proved
the following proposition (notice only three derivatives of ρ are required).
Proposition 5.3.3. Let Ω ⊂⊂ ` be a complex manifold with C
3
boundary and ρ
be a C
3
deﬁning function for Ω. For any ) ∈ T
2
(p,q)
such that ) vanishes outside a
coordinate patch l near a boundary point in /Ω and φ ∈ C
2
(Ω), we have
(5.3.22)
∂)
2
φ
+ ϑ
φ
)
2
φ
≥ (1 −c)
I,K
t
n
j,k=1
(φ
jk
)
I,jK
, )
I,kK
)
φ
+ (1 −c)
I,K
t
n
j,k=1
_
bΩ∩U
ρ
jk
)
I,jK
¯
)
I,kK
c
−φ
do
+ (1 −c)
I,J
t
n
j=1
1
j
)
I,J

2
φ
+O(c)( 1) 
φ
 ) 
φ
),
where c 0 can be chosen arbitrarily small and O(c)( 1) 
φ
 ) 
φ
) denotes terms
which can be bounded by C
 1) 
φ
 ) 
φ
for some constant C
independent of φ.
Let λ be the smallest eigenvalue of the Hermitian symmetric form
(5.3.23)
n
j,k=1
φ
jk
o
j
¯ o
k
, o = (o
1
, , o
n
) ∈ C
n
.
Let j be the smallest eigenvalue of the Levi form
(5.3.24)
n
j,k=1
ρ
jk
o
j
¯ o
k
, where
n
i=1
o
i
∂ρ
∂n
i
= 0.
Note that λ and j are independent of the choice of the basis n
1
, , n
n
. We have
the following global a priori estimates:
Proposition 5.3.4. Let Ω ⊂⊂ ` be a complex manifold with C
3
boundary /Ω and
φ ∈ C
2
(Ω). We have the following estimates: for every ) ∈ T
(p,q)
,
(5.3.25) ∂)
2
φ
+ ϑ
φ
)
2
φ
≥
1
2
__
Ω
(λ −C)[)[
2
c
−φ
d\ +
_
bΩ
j [)[
2
c
−φ
do
_
,
where λ is the smallest eigenvalue of the form (5.3.23), j is the smallest eigenvalue
of the Levi form (5.3.24) and C is a constant independent of φ.
110 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
Proof. Let ¦η
i
¦
N
i=0
be a partition of unity such that η
0
∈ C
∞
0
(Ω) and each η
i
,
1 ≤ i ≤ ·, is supported in a coordinate patch l
i
, η
i
∈ C
∞
0
(l
i
), Ω ⊂ Ω ∪ (∪
i
l
i
),
and
N
i=0
η
2
i
= 1 on Ω.
Since )
I,nK
= 0 on /Ω, we have, for 1 ≤ i ≤ ·,
I,K
t
j,k
_
bΩ∩Ui
ρ
jk
)
I,jK
¯
)
I,kK
c
−φ
do ≥
_
bΩ∩Ui
jη
2
i
[)[
2
c
−φ
do.
Applying Proposition 5.3.3 to each η
i
), choosing c suﬃciently small, we have
(5.3.26)
_
bΩ∩Ui
jη
2
i
[)[
2
c
−φ
do +
_
Ui
λη
2
i
[)[
2
c
−φ
d\
≤ 2( η
i
¯
∂) 
2
φ
+  η
i
ϑ
φ
) 
2
φ
) +C
i
_
Ui∩Ω
[)[
2
c
−φ
d\.
The constant C
i
depends only on η
i
but not on φ. Summing up over i, the propo
sition is proved.
From (5.3.25), we can repeat the same argument as in Chapter 4 to prove the
following 1
2
existence theorem for
¯
∂ if there exists a strictly plurisubharmonic
function on Ω.
Theorem 5.3.5. Let Ω ⊂⊂ ` be a pseudoconvex manifold with C
3
boundary /Ω
such that there exists a strictly plurisubharmonic function φ on Ω. Then for any
) ∈ 1
2
(p,q)
(Ω) with
¯
∂) = 0, there exists n ∈ 1
2
(p,q−1)
(Ω) such that
¯
∂n = ).
Proof. From pseudoconvexity, j ≥ 0 on /Ω where j is the smallest eigenvalue of
the Levi form. We have that the last term in (5.3.25) is nonnegative. Since φ is
strictly plurisubharmonic in Ω which is relatively compact in `, it follows that
λ 0 where λ is the smallest eigenvalue of the form (5.3.23). If we choose t 0
such that tλ ≥ C + 2, where C is as in (5.3.25), we see that for any p ∈ T
2
(p,q)
,
(5.3.27) p
2
tφ
≤ ∂p
2
tφ
+
¯
∂
∗
tφ
p
2
tφ
.
Using the same arguments as in the proof of the density lemma, Lemma 4.3.2, we
can show that T
2
(p,q)
is dense in Dom(
¯
∂) ∩ Dom(
¯
∂
∗
tφ
) in the graph norm p
tφ
+
∂p
tφ
+
¯
∂
∗
tφ
p
tφ
and (5.3.27) holds for any p ∈ Dom(
¯
∂)∩Dom(
¯
∂
∗
tφ
). Using Lemma
4.1.1, this implies that ¹(
¯
∂) and ¹(
¯
∂
∗
tφ
) are closed. To show that ¹(
¯
∂) = ker(
¯
∂),
we repeat the arguments of the proof of Theorem 4.3.4. For any p ∈ 1
2
(p,q)
(Ω) ∩
Dom(
¯
∂
∗
tφ
), one has
(5.3.28) [ (), p)
tφ
[ ≤  ) 
tφ
 p 
tφ
≤  ) 
tφ

¯
∂
∗
tφ
p 
tφ
.
Thus, there exists n ∈ 1
2
(p,q−1)
(Ω) such that
¯
∂n = ) in Ω and
 n 
tφ
≤  ) 
tφ
.
5.3 Function Theory on Manifolds 111
This proves the theorem.
We note that if ` = C
n
, one can take φ = [.[
2
. However, on a general complex
manifold, there does not always exist a plurisubharmonic function on `.
Let
(p,q)
=
¯
∂
¯
∂
∗
+
¯
∂
∗ ¯
∂ on Dom(
(p,q)
) and Dom(
(p,q)
) be deﬁned as in Deﬁni
tion 4.2.2. The arguments of the proof of Proposition 4.2.3 can be applied to show
that
(p,q)
is a linear, closed, densely deﬁned selfadjoint operator on 1
2
(p,q)
(Ω). Us
ing the 1
2
existence theorem 5.3.5, we can obtain the following existence theorem
for the
¯
∂Neumann operator on pseudoconvex manifolds.
Theorem 5.3.6. Let Ω ⊂⊂ ` be a pseudoconvex Hermitian manifold with C
3
boundary /Ω such that there exists a strictly plurisubharmonic function φ on Ω.
For each j, ¡ such that 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n, there exists a bounded operator
·
(p,q)
: 1
2
(p,q)
(Ω) →1
2
(p,q)
(Ω) such that
(1) ¹(·
(p,q)
) ⊂ Dom(
(p,q)
),
·
(p,q)
(p,q)
=
(p,q)
·
(p,q)
= 1 on Dom(
(p,q)
).
(2) For any ) ∈ 1
2
(p,q)
(Ω), ) =
¯
∂
¯
∂
∗
·
(p,q)
) ⊕
¯
∂
∗ ¯
∂·
(p,q)
).
(3)
¯
∂·
(p,q)
= ·
(p,q+1)
¯
∂ on Dom(
¯
∂), 1 ≤ ¡ ≤ n −1.
(4)
¯
∂
∗
·
(p,q)
= ·
(p,q−1)
¯
∂
∗
on Dom(
¯
∂
∗
), 2 ≤ ¡ ≤ n.
(5) For any ) ∈ 1
2
(p,q)
(Ω) such that
¯
∂) = 0, ) =
¯
∂
¯
∂
∗
·
(p,q)
).
The proof is exactly the same as the proof for Theorem 4.4.1. One can also show
the existence of ·
(p,0)
for ¡ = 0 following the same arguments as in Theorem 4.4.3
and we omit the details.
If /Ω is strongly pseudoconvex, there exists a c 0 such that the smallest
eigenvalue of the Levi form j c 0 on /Ω, and we have from (5.3.25) (setting
φ = 0),
(5.3.29) ∂)
2
+ ϑ)
2
≥
c
2
_
bΩ
[)[
2
do −C)
2
.
When /Ω is a strongly pseudoconvex manifold with C
∞
boundary, we can also
use the boundary term in the estimates (5.3.29) to obtain the existence and the
regularity for the
¯
∂Neumann operator. By using a partition of unity, the Sobolev
spaces \
s
(Ω) can be deﬁned on a manifold Ω for any : ∈ 1 (see Appendix A). Using
the same arguments as in Section 5.2, we have the following subelliptic estimates.
Theorem 5.3.7. Let Ω ⊂⊂ ` be a strongly pseudoconvex Hermitian manifold with
C
3
boundary /Ω. There exists a constant C 0 such that for any ) ∈ Dom(
¯
∂) ∩
Dom(
¯
∂
∗
), 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n −1,
(5.3.30)  ) 
2
1
2
(Ω)
≤ C(
¯
∂) 
2
Ω
+ 
¯
∂
∗
) 
2
Ω
+  ) 
2
Ω
),
where C is independent of ).
The proof is similar to Theorem 5.1.2. We note that in each coordinate patch
with a special frame, the operators
¯
∂ and ϑ given by (5.3.8) and (5.3.9) diﬀer from
(4.2.1) and (4.2.3) only by lower order terms. Thus, the arguments used in proving
Theorem 5.1.2 can be easily modiﬁed to prove (5.3.30). We omit the details.
112 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
We can use (5.3.30) to prove that there exists a
¯
∂Neumann operator which
inverts
(p,q)
. Let
H
(p,q)
(Ω) = ¦) ∈ 1
2
(p,q)
(Ω) ∩ Dom(
¯
∂) ∩ Dom(
¯
∂
∗
)[
¯
∂) =
¯
∂
∗
) = 0¦
= Ker(
¯
∂) ∩ Ker(
¯
∂
∗
)
= Ker(
(p,q)
).
The last equality can be veriﬁed in the same way as in the proof of (4.4.2). However,
on a strongly pseudoconvex complex manifold, H
(p,q)
(Ω) = H
(p,q)
is not always triv
ial for ¡ ≥ 1. The following theorem shows that H
(p,q)
is always ﬁnite dimensional
when ¡ ≥ 1.
Theorem 5.3.8. Let Ω ⊂⊂ ` be a strongly pseudoconvex Hermitian manifold with
a C
3
boundary /Ω. For any 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n, the space H
(p,q)
is ﬁnite
dimensional. Furthermore, the following estimate holds: for any ) ∈ Dom(
¯
∂) ∩
Dom(
¯
∂
∗
) ∩ H
⊥
(p,q)
,
(5.3.31)  ) 
2
Ω
≤ C(
¯
∂) 
2
Ω
+ 
¯
∂
∗
) 
2
Ω
).
Proof. We have from (5.3.30),
(5.3.32)  ) 
2
1
2
(Ω)
≤ C  ) 
2
Ω
, ) ∈ H
(p,q)
.
Since \
1/2
is compact in 1
2
(Ω) by the Rellich lemma (see Theorem A.8 in the
Appendix), we have that the unit sphere in H
(p,q)
is compact. Thus, H
(p,q)
is ﬁnite
dimensional.
If (5.3.31) does not hold, there exists a sequence )
n
such that )
n
∈ Dom(
¯
∂) ∩
Dom(
¯
∂
∗
) ∩ H
⊥
(p,q)
,
 )
n

2
Ω
≥ n(
¯
∂)
n

2
Ω
+ 
¯
∂
∗
)
n

2
Ω
).
Let θ
n
= )
n
, )
n

Ω
. Then 
¯
∂θ
n

Ω
+ 
¯
∂
∗
θ
n

Ω
→ 0 and  θ
n

Ω
= 1. From
(5.3.30) we have  θ
n
1
2
(Ω)
≤ C. Using the Rellich lemma, there exists a subsequence
of θ
n
which converges to some element θ ∈ 1
2
(p,q)
∩ H
⊥
(p,q)
. However,
¯
∂θ =
¯
∂
∗
θ = 0
and  θ 
Ω
= 1, giving a contradiction. This proves (5.3.31).
Let H
(p,q)
denote the projection onto the subspace H
(p,q)
where 0 ≤ j ≤ n and
0 ≤ ¡ ≤ n. We note that when ¡ = 0, H
(p,0)
is the projection onto 1
2
holomorphic
forms, i.e., forms with 1
2
holomorphic coeﬃcients. The following theorem gives the
existence and regularity of the
¯
∂Neumann operator on any strongly pseudoconvex
Hermitian manifold.
Theorem 5.3.9. Let Ω ⊂⊂ ` be a strongly pseudoconvex Hermitian manifold with
C
∞
boundary /Ω. For each j, ¡ such that 0 ≤ j ≤ n, 0 ≤ ¡ ≤ n, there exists a
compact operator ·
(p,q)
: 1
2
(p,q)
(Ω) →1
2
(p,q)
(Ω) such that
(1) ¹(·
(p,q)
) ⊂ Dom(
(p,q)
),
·
(p,q)
(p,q)
=
(p,q)
·
(p,q)
= 1− H
(p,q)
on Dom(
(p,q)
).
(2) For any ) ∈ 1
2
(p,q)
(Ω), ) =
¯
∂
¯
∂
∗
·
(p,q)
) ⊕
¯
∂
∗ ¯
∂·
(p,q)
) ⊕H
(p,q)
).
5.3 Function Theory on Manifolds 113
(3)
¯
∂·
(p,q)
= ·
(p,q+1)
¯
∂ on Dom(
¯
∂).
(4)
¯
∂
∗
·
(p,q)
= ·
(p,q−1)
¯
∂
∗
on Dom(
¯
∂
∗
).
(5) ·
(p,q)
(C
∞
(p,q)
(Ω)) ⊂ C
∞
(p,q)
(Ω), ¡ ≥ 0.
H
(p,q)
(C
∞
(p,q)
(Ω)) ⊂ C
∞
(p,q)
(Ω), ¡ ≥ 0.
Proof. We ﬁrst prove that ¹(
(p,q)
) is closed for any 1 ≤ ¡ ≤ n. From (5.3.31), we
know that for any ) ∈ Dom(
(p,q)
) ∩ H
⊥
(p,q)
,
 ) 
2
Ω
≤ C(
¯
∂) 
2
Ω
+ 
¯
∂
∗
) 
2
Ω
)
= C(
(p,q)
), ))
Ω
≤ C 
(p,q)
) 
Ω
 ) 
Ω
.
Now applying Lemma 4.1.1, we see that ¹(
(p,q)
) is closed and
1
2
(p,q)
(Ω) = ¹(
(p,q)
) ⊕Ker(
(p,q)
) = ¹(
(p,q)
) ⊕H
(p,q)
.
Let ·
(p,q)
to be the bounded inverse operator of
(p,q)
on ¹(
(p,q)
). We extend
·
(p,q)
to 1
2
(p,q)
(Ω) by setting ·
(p,q)
H
(p,q)
= 0. One can easily show that (1) and
(2) hold. Using (5.3.30), we observe that for any ) ∈ 1
2
(p,q)
(Ω),
(5.3.33)
 ·
(p,q)
) 
2
1
2
(Ω)
≤ C(
(p,q)
·
(p,q)
) 
2
Ω
+  ·
(p,q)
) 
2
Ω
)
≤ C  ) 
2
Ω
.
The Rellich lemma implies that · is a compact operator. (3) and (4) can be veriﬁed
by repeating the proofs of (3) and (4) in Theorem 4.4.1. We can establish (5) using
the same arguments in the proof of Theorem 5.2.1. From the proof of Theorem
4.4.3, one can show that ·
(p,0)
exists and can be expressed as
(5.3.34) ·
(p,0)
= ϑ·
2
(p,1)
¯
∂.
Also, ·
(p,0)
is bounded. The compactness of ·
(p,0)
follows since ϑ·
(p,1)
is compact
and ·
(p,1)
¯
∂ is bounded (see (4.4.12)).
Using the same arguments as the proof of Theorems 5.2.1 and 5.2.6, we have the
following more precise estimates.
Theorem 5.3.10. Let Ω ⊂⊂ ` be a strongly pseudoconvex Hermitian manifold
with C
∞
boundary /Ω. For ¡ ≥ 1 and each / = 0, 1, 2, , there exists a constant
C
k
0 such that for any ) ∈ \
k
(p,q)
(Ω),
(5.3.35) ·
(p,q)
)
k+1
≤ C
k
)
k
,
(5.3.36) 
¯
∂
∗
·
(p,q)
)
k+
1
2
+
¯
∂·
(p,q)
)
k+
1
2
≤ C
k
)
k
.
114 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
Corollary 5.3.11. Let Ω ⊂⊂ ` be a strongly pseudoconvex Hermitian manifold
with C
∞
boundary /Ω. For any ) ∈ \
k
(p,q)
(Ω), ¡ ≥ 1 and / ≥ 0, such that
¯
∂) = 0
in Ω and H
(p,q)
) = 0, we have n =
¯
∂
∗
·) is a solution of
¯
∂n = ) in Ω and
n
k+
1
2
≤ C)
k
,
where C is a constant independent of ). In particular, if ) ∈ C
∞
(p,q)
(Ω),
¯
∂) = 0 and
H
(p,q)
) = 0, there exists a solution n ∈ C
∞
(p,q−1)
(Ω) such that
¯
∂n = ) in Ω.
The solution n is called the canonical solution (or Kohn’s solution) to the equation
¯
∂n = ) and it is the unique solution which is orthogonal to Ker(
¯
∂).
An important consequence of Corollary 5.3.11 is the solution to the Levi problem
on a strongly pseudoconvex manifold with smooth boundary. A complex manifold
Ω with smooth boundary /Ω is called a domain of holomorphy if for every j ∈
/Ω there is a holomorphic function on Ω which is singular at j (c.f. Deﬁnition
3.5.1). In Theorem 4.5.2, we have already proved that pseudoconvex domains in C
n
are domains of holomorphy. The next theorem shows that strongly pseudoconvex
domains in complex manifolds are domains of holomorphy.
Theorem 5.3.12. Let Ω ⊂⊂ ` be a strongly pseudoconvex manifold with C
∞
boundary /Ω. Then Ω is a domain of holomorphy.
Proof. For each boundary point j ∈ /Ω, we will construct a function /(.) such that
/ ∈ C
∞
(Ω ¸ ¦j¦), / is holomorphic in Ω and
lim
z→p
/(.) = ∞.
Since Ω is strongly pseudoconvex with C
∞
boundary /Ω, one can construct a
local holomorphic function ) on Ω ∩ l where l is an open neighborhood of j such
that ) is singular at j. To do this, let .
1
, , .
n
be a coordinate system in the
neighborhood l of j with origin at j. Let : be a smooth deﬁning function for Ω
such that : is strictly plurisubharmonic near j. That such a deﬁning function exists
follows from the same arguments as before (see Theorem 3.4.4). Let
1(.) = −2
n
i=1
∂:
∂.
i
(0).
i
−
n
i,j=1
∂
2
:
∂.
i
∂.
j
(0).
i
.
j
.
1(.) is holomorphic in l. Using Taylor’s expansion at 0, there exists a suﬃciently
small neighborhood \ ⊂ l of 0 and C 0 such that for any . ∈ Ω ∩ \ ,
Re1(.) = −:(.) +
n
i,j=1
∂
2
:
∂.
i
∂¯ .
j
(0).
i
¯ .
j
+O([.[
3
) ≥ C[.[
2
.
Thus, 1(.) ,= 0 when . ∈ Ω ∩ \ ¸ ¦0¦. Setting
) =
1
1
,
5.4 Almost Complex Structures 115
it is easily seen that ) is locally a holomorphic function which cannot be extended
holomorphically across 0.
Let χ be a cutoﬀ function such that χ ∈ C
∞
0
(\ ) and χ = 1 in a neighborhood
of 0. We extend χ to be 0 on Ω ¸ \ . Let p be the (0, 1)form deﬁned by
p =
¯
∂(χ)) = (
¯
∂χ)) in Ω.
Obviously p can be extended smoothly up to the boundary. Thus, p ∈ C
∞
(0,1)
(Ω) and
¯
∂p = 0 in Ω. To show that H
(0,1)
p = 0, notice that when n ≥ 3, χ) ∈ 1
2
(Ω). Thus,
p ∈ ¹(
¯
∂) = Ker(
¯
∂
∗
)
⊥
and p ⊥ H
(0,1)
. When n = 2, we approximate ) by )
=
1
P+
for c ¸ 0. Then χ)
∈ C
∞
(Ω) and
¯
∂(χ)
) → p in 1
2
. However, each
¯
∂(χ)
) is
in the ¹(
¯
∂) which is closed from Theorem 5.3.9. This implies that p ∈ ¹(
¯
∂) and
H
(0,1)
p = 0 for n = 2 also. We deﬁne
n =
¯
∂
∗
·
(0,1)
p.
It follows from Corollary 5.3.11 that n ∈ C
∞
(Ω) and
¯
∂n = p in Ω. Let / be deﬁned
by
/ = χ) −n.
Then, / is holomorphic in Ω, / ∈ C
∞
(Ω ¸ ¦j¦) and / is singular at j. This proves
the theorem.
Thus, the Levi problem for strongly pseudoconvex manifolds with smooth bound
aries is solved.
5.4 Almost Complex Structures
In Chapter 2 we study when a complex vector ﬁeld in 1
2
is actually a Cauchy
Riemann equation in other coordinates. In this section we study the ndimensional
analog of this problem.
Deﬁnition 5.4.1. Let ` be a real C
∞
manifold of dimension 2n. An almost
complex structure T
1,0
(`) is a subbundle of the complexiﬁed tangent bundle CT(`)
such that
(1) CT(`) = T
1,0
(`) +T
0,1
(`),
(2) T
1,0
(`) ∩ T
0,1
(`) = ¦0¦,
where T
0,1
(`) = T
1,0
(`). ` is called an almost complex manifold with an almost
complex structure T
1,0
(`).
When ` is a complex manifold, there is a canonical T
1,0
deﬁned on `, namely,
the holomorphic vector bundle. In local holomorphic coordinates .
1
, , .
n
, in a
neighborhood l, we have that
(5.4.1) T
1,0
(` ∩ l) =
_
∂
∂.
1
, ,
∂
∂.
n
_
,
where the righthand side denotes the linear span by vector ﬁelds ∂,∂.
1
, , ∂,∂.
n
.
Let C
∞
(T
1,0
(`)) denote the smooth sections of T
1,0
(`). If ` is a complex man
ifold, we have
(5.4.2) [1, 1
t
] ∈ C
∞
(T
1,0
(`)), for any 1, 1
t
∈ C
∞
(T
1,0
(`)).
116 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
Deﬁnition 5.4.2. An almost complex structure T
1,0
(`) is called integrable if
(5.4.2) is satisﬁed.
A complex manifold is an integrable almost complex manifold. The Newlander
Nirenberg theorem states that the converse is also true. Before we state and prove
the theorem, we ﬁrst note that on an almost complex manifold, there is a notion of
the CauchyRiemann equations and the
¯
∂ operator.
Let Π
1,0
, Π
0,1
denote the projection from CT(`) onto T
1,0
(`) and T
0,1
(`)
respectively. Then, one has
Π
1,0
+ Π
0,1
= 1, Π
1,0
Π
0,1
= 0, and Π
0,1
= Π
1,0
.
The last equation means that Π
0,1
ζ = Π
1,0
¯
ζ, for ζ ∈ CT(`). Thus there is a
natural splitting of the diﬀerential 1forms Λ
1
(`) into (1,0)forms, Λ
1,0
(`), and
(0,1)forms, Λ
0,1
(`), which are deﬁned to be the dual of T
1,0
(`) and T
0,1
(`)
respectively. We shall still use Π
1,0
, Π
0,1
to denote the projection from Λ
1
(`)
onto Λ
1,0
(`) and Λ
0,1
(`) respectively. For any smooth function n, we have dn =
Π
1,0
dn + Π
0,1
dn. On an almost complex manifold, one can deﬁne the Cauchy
Riemann equation by
¯
∂n = Π
0,1
dn and ∂n = Π
1,0
dn,
where n is any smooth function on `. We can also extend this deﬁnition to (j, ¡)
forms and deﬁne ∂ and
¯
∂ on (j, ¡)forms ) to be the projection of the exterior
derivative d) into the space of (j+1, ¡)forms and (j, ¡ +1)forms respectively. The
integrability condition guarantees that
¯
∂ is a complex.
Lemma 5.4.3. If an almost complex structure is integrable, then d = ∂ +
¯
∂ and
∂
2
= ∂
¯
∂ +
¯
∂∂ =
¯
∂
2
= 0.
Proof. If one can show that d = ∂ +
¯
∂, then ∂
2
= ∂
¯
∂ +
¯
∂∂ =
¯
∂
2
= 0 follows easily
from degree consideration. It is obvious that d = ∂+
¯
∂ on functions. For 1forms, we
have Λ
1
(`) = Λ
1,0
(`)+Λ
0,1
(`). To verify for 1forms, it suﬃces to prove for each
(0,1)form and each (1,0)form. If ) is a (0,1)form, for any 1, 1
t
∈ C
∞
(T
1,0
(`)),
we have
d)(1, 1
t
) =
1
2
(1(), 1
t
) −1
t
(), 1) −(), [1, 1
t
])) = 0
since T
1,0
(`) is integrable. This shows that d) has no component of (2,0)forms.
Similarly, if ) is a (1,0)form, d) has no component of (0,2)forms. In each case,
d) = ∂) +
¯
∂). The general case follows from the fact that each (j, ¡)form can be
written as linear combination of forms of the type
/ = )
1
∧ ∧ )
p
∧ p
1
∧ ∧ p
q
,
where )
i
’s are (1,0)forms and p
j
’s are (0,1)forms. Since d/ is a sum of a type
(j + 1, ¡)form and a type (j, ¡ + 1)form, we have d/ = ∂/ +
¯
∂/. This proves the
lemma.
5.4 Almost Complex Structures 117
Theorem 5.4.4 (NewlanderNirenberg). An integrable almost complex mani
fold is a complex manifold.
Proof. This problem is purely local and we shall assume that ` is a small neighbor
hood of 0 in 1
2n
. Let 1
1
, , 1
n
be a local basis for smooth sections of T
1,0
(`).
If we can ﬁnd complexvalued functions ζ
1
, , ζ
n
such that
(5.4.3) 1
i
ζ
j
= 0, i, , = 1, , n,
where dζ
1
, , dζ
n
are linearly independent, then the theorem will be proved since
(5.4.4) ¸1
1
, , 1
n
) =
_
∂
∂ζ
1
, ,
∂
∂ζ
n
_
,
where ¸1
1
, , 1
n
) denotes the linear span of 1
1
, , 1
n
over C.
Let r
1
, , r
2n
be the real coordinates for ` and we write .
j
= r
j
+ir
n+j
. We
can, after a quadratic change of coordinates, assume that
(5.4.5) 1
i
=
∂
∂.
i
+
n
j=1
o
ij
∂
∂¯ .
j
, i = 1, , n,
where the o
ij
’s are smooth functions and o
ij
(0) = 0 for all i, , = 1, , n. At the
origin, 1
i
is the constant coeﬃcient operator ∂,∂.
i
. We shall show that (5.4.4) can
be solved in a small neighborhood of 0. Let
(5.4.6) 1
i
=
∂
∂.
i
+
n
j=1
o
ij
(cr)
∂
∂¯ .
j
, i = 1, , n,
where c 0 is small. Then
T
0,1
= ¸1
1
, , 1
n
)
deﬁnes an almost complex structure that is integrable for each c < c
0
for some
suﬃciently small c
0
0.
From Lemma 5.4.3, there is a CauchyRiemann complex, denoted by
¯
∂
, associ
ated with each almost complex structure T
0,1
. We shall equip the almost complex
structure with a Hermitian metric. Then the existence and regularity theory devel
oped for
¯
∂ in the previous section on any complex manifold can be applied to `
with
¯
∂ substituted by
¯
∂
. Let φ =
n
i=1
[.
i
[
2
= [r[
2
, then at 0 we see that
n
j,k=1
φ
jk
o
j
¯ o
k
=
n
i=1
[o
i
[
2
.
Thus, φ is a strictly plurisubharmonic function near 0. If we set
Ω = ¦r ∈ ` [ [r[
2
< δ¦
118 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
for some small δ 0, then Ω is strongly pseudoconvex with respect to the almost
complex structure T
0,1
(`). Using Corollary 5.3.11, there exists a solution n
i
on Ω
such that
(5.4.7)
¯
∂
n
i
=
¯
∂
.
i
and
(5.4.8)  n
i

s
≤ C
s

¯
∂
.
i

s
,
where C
s
can be chosen uniformly for c < c
0
. Since
1
i
.
j
= ¯ o
ij
(cr),
we have
1
α
¯
∂
.
i
= O(c)
for any 1
α
= (∂,∂r
1
)
α1
(∂,∂r
2n
)
α2n
, where the α
i
’s are nonnegative integers.
This implies that
(5.4.9) 
¯
∂
.
i

s
→0 if c →0.
Let
(5.4.10) ζ
i
= .
i
−n
i
.
The Sobolev embedding theorem (see Theorem A.7 in the Appendix) shows that if
we choose : n + 1, then
[dn
i
(0)[ ≤  n
i

s
≤ C
s

¯
∂
.
i

s
→0 if c →0.
We have from (5.4.7) that
¯
∂
ζ
i
= 0 in Ω and also dζ
i
(0) = d.
i
−dn
i
(0) are linearly
independent if c is suﬃciently small. If we pull back ζ
i
to cΩ by setting ζ
i
= ζ
i
(r,c),
then we have that
¯
∂ζ
i
= 0 and dζ
i
are linearly independent in cΩ provided we choose
c suﬃciently small. This proves the theorem.
NOTES
The subelliptic 1,2estimates and boundary regularity for the
¯
∂Neumann oper
ator on strongly pseudoconvex manifolds were proved in J. J. Kohn [Koh 1]. Much
of the material concerning strong pseudoconvex domains in this chapter was ob
tained there. The use of a special boundary frame was due to M. E. Ash [Ash 1]. A
simpliﬁcation of the proof of the boundary regularity for subelliptic operators using
pseudodiﬀerential operators was given in J. J. Kohn and L. Nirenberg [KoNi 1]
where the method of elliptic regularization was used in order to pass from a priori
estimates to actual estimates. In [KoNi 1], a systematic treatment of the subel
liptic boundary value problem with any subellipticity 0 < c ≤ 1,2 was discussed.
119
Pseudodiﬀerential operators and subelliptic estimates will be discussed in Chapter
8.
The proof of subelliptic 1,2estimates in Theorem 5.1.2 follows the approach of J.
Michel and M.C. Shaw [MiSh 1]. The proof of the boundary regularity discussed in
5.2 is a variation of the proof used in [KoNi 1] since only commutators of diﬀerential
operators are used but not pseudodiﬀerential operators. The discussion of function
theory on manifolds mainly follows that of L. H¨ormander [H¨or 3]. Global strictly
plurisubharmonic functions do not always exists on general complex manifolds. If
` is a Stein manifold, then there exists a strictly plurisubharmonic exhaustion
function for `. Thus Theorem 5.3.5 can be applied to any relatively compact
pseudoconvex manifold Ω which lies in a Stein manifold. For a detailed discussion
of function theory on Stein manifolds, see Chapter 5 in L. H¨ormander [H¨or 9].
The Levi problem on a strongly pseudoconvex manifold (Theorem 5.3.12) was
ﬁrst solved by H. Grauert [Gra 1] using sheaf theory. The proof of Theorem 5.3.12
using the existence and the regularity of the
¯
∂Neumann operator was due to J. J.
Kohn [Koh 1]. Since a pseudoconvex domain in C
n
by deﬁnition can be exhausted by
strongly pseudoconvex domains, using a result of H. Behnke and K. Stein (see [BeSt
1] or [GuRo 1], one can deduce that any pseudoconvex domain in C
n
is a domain
of holomorphy (c.f. Theorem 4.5.2). This needs not be true for pseudoconvex
domains in complex manifolds (see J. E. Fornaess [For 1]). More discussions on the
Levi problem on pseudoconvex manifolds can be found in [For 3], [Siu 2].
Theorem 5.4.4 was ﬁrst proved by A. Newlander and L. Nirenberg [NeNi 1].
B. Malgrange has given a totally diﬀerent proof (see B. Malgrange [Mal 2] or L.
Nirenberg [Nir 3]). There is yet another proof, due to S. Webster [Web 1], using
integral kernel methods. Our proof was essentially given in J. J. Kohn [Koh 1] as
an application of the
¯
∂Neumann problem.
There is a considerable amount of literature on the
¯
∂Neumann operator, the
canonical solution and the Bergman projection on strongly pseudoconvex domains
in other function spaces, including H¨older and 1
p
spaces (See R. Beals, P. C. Greiner
and N. Stanton [BGS 1], R. Harvey and J. Polking [HaPo 3,4], I. Lieb and R. M.
Range [LiRa 2,3,4], A. Nagel and E. M. Stein [NaSt 1], D. H. Phong and E. M.
Stein [PhSt 1], R. M. Range [Ran 5] and the references therein). We also refer the
reader to the article by M. Beals, C. Feﬀerman and R. Grossman [BFG 1] for more
discussions on strongly pseudoconvex domains.
120 The ∂Neumann Problem on Strongly Pseudoconvex Manifolds
CHAPTER 6
BOUNDARY REGULARITY FOR ∂
ON PSEUDOCONVEX DOMAINS
Let 1 be a bounded pseudoconvex domain in C
n
with smooth boundary /1. In
this chapter, we study the global regularity of the equation
(6.0.1) ∂n = ) on 1,
where ) ∈ C
∞
(p,q)
(1) with 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n and ) is ∂closed.
The existence theorems for ∂ and the ∂Neumann operator · on any bounded
pseudoconvex domain have been proved in Chapter 4 in 1
2
spaces. In this chapter
we are interested in the following questions:
(1) Can one solve equation (6.0.1) with a smooth solution n ∈ C
∞
(p,q−1)
(1) if )
is in C
∞
(p,q)
(1)?
(2) Does the canonical solution ∂
∗
·) belong to \
s
(p,q−1)
(1) if ) is in \
s
(p,q)
(1)?
(3) Does the Bergman projection 1 preserve C
∞
(1) or \
s
(1)?
(4) Under what conditions can a biholomorphic mapping between two smooth
bounded domains be extended smoothly up to the boundaries?
From the results in Chapter 5, both · and ∂
∗
· are regular in Sobolev spaces
or the C
∞
category if we assume that /1 is smooth and strongly pseudoconvex.
In fact, the canonical solution has a “gain” of 1,2 derivative in the Sobolev spaces.
Here, we study the global boundary regularity for ∂ and the ∂Neumann operator
· on a smooth bounded weakly pseudoconvex domain in C
n
.
In Section 6.1, we prove that the ﬁrst question can be answered aﬃrmatively.
This result is proved using the weighted ∂Neumann problem. However, the smooth
solution might not be the canonical solution. In Section 6.2, we study the global
regularity for · when the domain has either a smooth plurisubharmonic deﬁning
function or certain transverse symmetry. We establish the Sobolev estimates for
the ∂Neumann operator and the regularity of the canonical solution on such do
mains. This result implies the regularity of the Bergman projection and is used
to prove the boundary regularity of biholomorphic mappings between pseudocon
vex domains. In general, a smooth pseudoconvex domain does not necessarily have
a plurisubharmonic deﬁning function or transverse symmetry. A counterexample,
known as the worm domain, is constructed in Section 6.4. Finally, we prove in
Section 6.5 that, for any : 0, there is a smooth bounded pseudoconvex domain
on which the ∂Neumann operator fails to be regular in any Sobolev spaces \
k
for
/ ≥ :.
6.1 Global Regularity for ∂ on Pseudoconvex Domains 121
6.1 Global Regularity for ∂ on Pseudoconvex Domains with Smooth
Boundaries
The main result in the section is the following theorem:
Theorem 6.1.1. Let 1 be a smooth bounded pseudoconvex domain in C
n
with
n ≥ 2. For every ) ∈ C
∞
(p,q)
(1), where 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n with ∂) = 0, one can
ﬁnd n ∈ C
∞
(p,q−1)
(1) such that ∂n = ).
We will prove the theorem using the weighted ∂Neumann operator with respect
to the weighted 1
2
norm 1
2
(1, φ
t
) introduced in Section 4.2, where φ
t
= t[.[
2
for
some t 0. Theorem 6.1.1 will be proved at the end of this section. We note that
1
2
(1, φ
t
) = 1
2
(1). The existence for the weighted
¯
∂Neumann operator on any
pseudoconvex domain with smooth boundary follows from the discussion in Chapter
4. We brieﬂy describe below.
From Proposition 4.3.3, we have for any (j, ¡)form ) ∈ Dom(∂) ∩ Dom(∂
∗
φt
),
_
D
I,K
t
j,k
∂
2
φ
t
∂.
j
∂.
k
)
I,jK
)
I,kK
c
−φt
d\ ≤  ∂) 
2
φt
+  ∂
∗
φt
) 
2
φt
.
Using the notation  
(t)
=  
φt
and ∂
∗
t
= ∂
∗
φt
, we see that for any ) ∈ Dom(∂) ∩
Dom(∂
∗
t
),
(6.1.1) t¡  ) 
2
(t)
≤  ∂) 
2
(t)
+  ∂
∗
t
) 
2
(t)
.
Let
t
= ∂∂
∗
t
+∂
∗
t
∂. If ) ∈ Dom(
t
), from (6.1.1), we have that
(6.1.2)
t¡  ) 
2
(t)
≤  ∂) 
2
(t)
+  ∂
∗
t
) 
2
(t)
= (
t
), ))
(t)
≤ 
t
) 
(t)
 ) 
(t)
.
Applying Lemma 4.1.1, (6.1.2) implies that the range of
t
is closed and
t
is one
toone. Thus,
t
has a bounded inverse ·
t
, the
¯
∂Neumann operator with weight
φ
t
. We can also show the following existence theorem of the weighted
¯
∂Neumann
operator on any bounded pseudoconvex domain by repeating the same argument
as in Theorem 4.4.1:
Theorem 6.1.2. Let 1 be a bounded pseudoconvex domain in C
n
, n ≥ 2. For each
0 ≤ j ≤ n, 1 ≤ ¡ ≤ n and t 0, there exists a bounded operator ·
t
: 1
2
(p,q)
(1) →
1
2
(p,q)
(1), such that
(1) Range(·
t
) ⊂ Dom(
t
). ·
t
t
=
t
·
t
= 1 on Dom(
t
).
(2) For any ) ∈ 1
2
(p,q)
(1), ) = ∂∂
∗
t
·
t
) ⊕∂
∗
t
∂·
t
).
(3) ∂·
t
= ·
t
∂ on Dom(∂), 1 ≤ ¡ ≤ n −1,
∂
∗
t
·
t
= ·
t
∂
∗
t
on Dom(∂
∗
t
), 2 ≤ ¡ ≤ n.
(4) The following estimates hold: For any ) ∈ 1
2
(p,q)
(1),
t¡  ·
t
) 
(t)
≤  ) 
(t)
,
122 Boundary Regularity for ∂ on Pseudoconvex Domains
√
t¡  ∂·
t
) 
(t)
≤  ) 
(t)
,
√
t¡  ∂
∗
·
t
) 
(t)
≤  ) 
(t)
.
(5) If ) ∈ 1
2
(p,q)
(1) and ∂) = 0 in 1, then for each t 0, there exists a solution
n
t
= ∂
∗
t
·
t
) satisfying ∂n
t
= ) and the estimate
t¡  n
t

2
(t)
≤  ) 
2
(t)
.
In Chapter 4, we have chosen t = δ
−2
, where δ is the diameter of 1, to obtain
the best constant for the bound of the
¯
∂Neumann operator without weights. Our
next theorem gives the regularity for ·
t
in the Sobolev spaces when t is large.
Theorem 6.1.3. Let 1 be a smooth bounded pseudoconvex domain in C
n
, n ≥
2. For every nonnegative integer /, there exists a constant o
k
0 such that the
weighted
¯
∂Neumann operator ·
t
maps \
k
(p,q)
(1) boundedly into itself whenever
t o
k
, where 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n.
Proof. We ﬁrst prove the a priori estimates for ·
t
when t is large. Let ϑ
t
be the
formal adjoint of
¯
∂ with respect to the weighted norm 1
2
(1, φ
t
). Note that for any
) ∈ C
∞
(p,q)
(1),
ϑ
t
) = c
φt
ϑ(c
−φt
)) = ϑ) +t¹
0
)
for some zeroth order operator ¹
0
. Hence, we have that for any ) ∈ C
∞
(p,q)
(1) with
compact support in 1,
Q
t
(), )) =  ∂) 
2
(t)
+  ϑ
t
) 
2
(t)
≥  ∂) 
2
(t)
+
1
2
 ϑ) 
2
(t)
− C
t
 ) 
2
(t)
≥  ) 
1
− C
t
 ) 
2
(t)
,
where  ) 
k
=
0≤]α]≤k
 1
α
) 
(t)
.
Thus the G˚arding inequality holds for compactly supported forms and the esti
mates in the interior are the same as before. We only need to estimate the solution
near the boundary. In the following, C and C
k
will always denote a constant inde
pendent of t.
Since the normal diﬀerentiation is controlled by ∂, ∂
∗
and the tangential deriva
tives, we shall only consider the action of tangential diﬀerentiations. Let l be a
special boundary chart near the boundary and n
1
, , n
n
= ∂: be a special bound
ary frame as before, where : is a deﬁning function normalized such that [d:[ = 1
on /1. We let T
k
be a tangential operator of order / and η ∈ C
∞
0
(l) as deﬁned in
Proposition 5.2.5. We use induction on / to prove the following estimate:
(6.1.3)  ) 
2
k
≤ C
k,t

t
) 
2
k
, ) ∈ Dom(
t
) ∩ C
∞
(p,q)
(1).
When / = 0, (6.1.3) holds for any t 0 by (6.1.2). We assume that (6.1.3) is
true for / −1 where / ≥ 1.
6.1 Global Regularity for ∂ on Pseudoconvex Domains 123
From the same argument as in Lemma 5.2.2, writing ) in the special frame, we
see that if ) ∈ Dom(
¯
∂
∗
t
) ∩ C
∞
(p,q)
(1), then ηT
k
) ∈ Dom(∂) ∩ Dom(∂
∗
t
). We obtain
from (6.1.1) that
(6.1.4) t  ηT
k
) 
2
(t)
≤ ( ∂(ηT
k
)) 
2
(t)
+  ϑ
t
(ηT
k
)) 
2
(t)
).
We note that the commutator [ϑ
t
, ηT
k
] = ¹
k
+ ¹
t
k−1
, where ¹
k
is a /th order
diﬀerential operator independent of t and ¹
t
k−1
is of order / − 1. Thus, using
arguments similar to (5.2.6) and (5.2.7), keeping track of the dependence on t, we
have
(6.1.5)
 ∂(ηT
k
)) 
2
(t)
+  (ϑ
t
ηT
k
)) 
2
(t)
≤ C
k
( ηT
k
∂) 
2
(t)
+  ηT
k
ϑ
t
) 
2
(t)
+  ) 
2
k
)
+C
k,t
 ) 
2
k−1
≤ C
k
([(ηT
k
), ηT
k
t
))
(t)
[ +  ) 
2
k
) +C
k,t
 ) 
2
k−1
≤ C
k
( ηT
k
) 
(t)
 ηT
k
t
) 
(t)
+  ) 
2
k
) +C
k,t
 ) 
2
k−1
.
Combining (6.1.4) and (6.1.5), we get
(6.1.6) t  ηT
k
) 
2
(t)
≤ C
k
(
t
) 
2
k
+  ) 
2
k
) +C
k,t
 ) 
2
k−1
,
where the constant C
k
is independent of t.
Repeating the arguments of Lemma 5.2.4, we observe that
 ) 
2
k
≤ C
k
(
t
) 
2
k−1
+ [[[)[[[
2
k
) +C
k,t
 ) 
2
k−1
.
Summing up all the tangential derivatives of the form ηT
k
in (6.1.6) and using a
partition of unity ¦η
i
¦
N
i=1
, such that
N
i=1
η
2
i
= 1 on 1, there exists a constant C
k
such that
(6.1.7)
t  ) 
2
k
≤ C
k
(
t
) 
2
k
+  ) 
2
k
)
+C
k,t

t
) 
2
k−1
+ C
k,t
 ) 
2
k−1
.
Choosing t C
k
+ 1, it follows, using the induction hypothesis, that
 ) 
2
k
≤ C
k

t
) 
2
k
+ C
k,t

t
) 
2
k−1
+ C
k,t
 ) 
2
k−1
≤ C
k,t

t
) 
2
k
.
This proves the a priori estimates (6.1.3) for the weighted ∂Neumann operators ·
t
when t is suﬃciently large. Using the elliptic regularization method as in the proof
of Theorem 5.2.1, one can pass from the a priori estimates to actual estimates and
the theorem is proved.
Arguing as in Theorem 4.4.3, we can prove that the weighted
¯
∂Neumann opera
tor ·
t,(p,0)
also exists for ¡ = 0. Let 1
t,(p,0)
denote the weighted Bergman projection
from 1
2
(p,0)
(1) onto the closed subspace H
(p,0)
(1) = ¦) ∈ 1
2
(p,0)
(1) [
¯
∂) = 0¦ with
124 Boundary Regularity for ∂ on Pseudoconvex Domains
respect to the weighted norm 1
2
(1, φ
t
). We have ·
t,(p,0)
: 1
2
(p,0)
(1) → 1
2
(p,0)
(1)
such that
t,(p,0)
·
t,(p,0)
= 1 −1
t(p,0)
and
·
t,(p,0)
=
¯
∂
∗
t
·
2
t,(p,1)
¯
∂.
As in the proof of Corollary 4.4.4, the weighted Bergman projection is given by
1
t,(p,0)
= 1 −
¯
∂
∗
t
·
t,(p,1)
¯
∂.
An operator is called exactly regular on \
k
(p,q)
(1), / ≥ 0, if it maps the Sobolev
space \
k
(p,q)
(1) continuously into forms with \
k
(1) coeﬃcients. The following
theorem shows that all the related operators of ·
t
are also exactly regular if ·
t
is
exactly regular.
Theorem 6.1.4. Let 1 be a smooth bounded pseudoconvex domain in C
n
, n ≥ 2.
For every nonnegative integer /, there exists a constant o
k
0 such that for every
t o
k
the operators ∂·
t
, ∂
∗
t
·
t
, ∂∂
∗
t
·
t
and ∂
∗
t
∂·
t
are exactly regular on \
k
(p,q)
(1),
where 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n. Furthermore, there exists a constant o
t
k
0 such that
for t o
t
k
, the weighted Bergman projection 1
t,(p,0)
maps \
k
(p,0)
(1) boundedly into
itself.
Proof. Let o
k
be as in Theorem 6.1.3 and t o
k
. From Theorem 6.1.3, ·
t
is a
bounded map from \
k
(p,q)
(1) into itself. We shall prove that ∂·
t
and ∂
∗
t
·
t
are
exactly regular simultaneously. As before, since
¯
∂ ⊕ ϑ
t
is elliptic, we only need to
prove a priori estimates for the tangential derivatives of
¯
∂·
t
) and
¯
∂
∗
t
·
t
) for any
) ∈ C
∞
(p,q)
(1). Let η and T be as in Theorem 6.1.3 and let O
t
( ) ) denote terms
which can be bounded by C
t
 ) . We have
 ηT
k
∂·
t
) 
2
(t)
+  ηT
k
∂
∗
t
·
t
) 
2
(t)
= (ηT
k
∂·
t
), ∂ηT
k
·
t
))
(t)
+ (ηT
k
∂
∗
t
·
t
), ∂
∗
t
ηT
k
·
t
))
(t)
+O
t
(( ηT
k
∂·
t
) 
(t)
+  ηT
k
∂
∗
t
·
t
) 
(t)
)  ·
t
) 
k
)
= (ηT
k
∂
∗
t
∂·
t
), ηT
k
·
t
))
(t)
+ (ηT
k
∂∂
∗
t
·
t
), ηT
k
·
t
))
(t)
+O
t
(( ηT
k
∂·
t
) 
(t)
+  ηT
k
∂
∗
t
·
t
) 
(t)
)  ·
t
) 
k
+  ·
t
) 
2
k
)
= (ηT
k
t
·
t
), ηT
k
·
t
))
(t)
+O
t
(( ηT
k
∂·
t
) 
(t)
+  ηT
k
∂
∗
t
·
t
) 
k
)  ·
t
) 
k
+  ·
t
) 
2
k
)
≤ C  ) 
k
 ·
t
) 
k
+C
t
(( ηT
k
∂·
t
) 
(t)
+  ηT
k
∂
∗
t
·
t
) 
k
)  ·
t
) 
k
+  ·
t
) 
2
k
).
Using small and large constants in o/ ≤ co
2
+
1
/
2
, it follows from (6.1.3) that
 ηT
k
∂·
t
) 
(t)
+  ηT
k
∂
∗
t
·
t
) 
(t)
≤ C
k,t
 ) 
k
.
By a partition of unity and arguments as before, we obtain the desired a priori
estimates
 ∂·
t
) 
k
+  ∂
∗
t
·
t
) 
k
≤ C
k,t
 ) 
k
.
6.1 Global Regularity for ∂ on Pseudoconvex Domains 125
For the operators ∂∂
∗
t
·
t
and ∂
∗
t
∂·
t
, 1 ≤ ¡ ≤ n, we have
 ηT
k
∂
∗
t
∂·
t
) 
2
(t)
+  ηT
k
∂∂
∗
t
·
t
) 
2
(t)
= (ηT
k
∂
∗
t
∂·
t
), ηT
k
∂
∗
t
∂·
t
))
(t)
+ (ηT
k
∂∂
∗
t
·
t
), ηT
k
∂∂
∗
t
·
t
))
(t)
= (ηT
k
t
·
t
), ηT
k
t
·
t
))
(t)
−(ηT
k
∂∂
∗
t
·
t
), ηT
k
∂
∗
t
∂·
t
))
(t)
−(ηT
k
∂
∗
t
∂·
t
), ηT
k
∂∂
∗
t
·
t
))
(t)
= (ηT
k
), ηT
k
))
(t)
−([∂, ηT
k
]∂∂
∗
t
·
t
), ηT
k
∂·
t
))
(t)
−([∂
∗
t
, ηT
k
]∂
∗
t
∂·
t
), ηT
k
∂
∗
t
·
t
))
(t)
+1
= (ηT
k
), ηT
k
))
(t)
−([∂, ηT
k
]∂
∗
t
·
t
), ηT
k
∂
∗
t
∂·
t
))
(t)
−([∂
∗
t
, ηT
k
]∂·
t
), ηT
k
∂∂
∗
t
·
t
))
(t)
+1,
where the error term 1 can be estimated by
( ηT
k
∂∂
∗
t
·
t
) 
(t)
+  ηT
k
∂
∗
t
∂·
t
) 
(t)
)( ∂·
t
) 
k
+  ∂
∗
t
·
t
) 
k
)
+  ∂·
t
) 
2
k
+  ∂
∗
t
·
t
) 
2
k
.
Since ∂·
t
and ∂
∗
t
·
t
are already known to be exactly regular on \
k
(p,q)
(1), we
obtain using small and large constants that
 ηT
k
∂
∗
t
∂·
t
) 
2
(t)
+  ηT
k
∂∂
∗
t
·
t
) 
2
(t)
≤ C
k,t
 ) 
2
k
.
Summing over a partition of unity, we have proved
 ∂
∗
t
∂·
t
) 
2
k
+  ∂∂
∗
t
·
t
) 
2
k
≤ C
k,t
 ) 
2
k
when t o
k
.
It remains to prove the exact regularity for 1
t,(p,0)
= 1 −
¯
∂
∗
t
·
t,(p,1)
¯
∂. We use
induction on / to prove the a priori estimates for ∂
∗
t
·
t,(p,1)
∂. The case when / = 0
is obvious. Denoting ·
t,(p,1)
by ·
t
, we have
 ηT
k
∂
∗
t
·
t
∂) 
2
(t)
= (ηT
k
∂
∗
t
·
t
∂), ηT
k
∂
∗
t
·
t
∂))
(t)
= (ηT
k
·
t
∂), ∂ηT
k
∂
∗
t
·
t
∂))
(t)
+O( ·
t
∂) 
k
 ∂
∗
t
·
t
∂) 
k
)
= (ηT
k
·
t
∂), ηT
k
∂))
(t)
+O( ·
t
∂) 
k
 ∂
∗
t
·
t
∂) 
k
)
= (ηT
k
∂
∗
t
·
t
∂), ηT
k
))
(t)
+O( ·
t
∂) 
k
( ) 
k
+  ∂
∗
t
·
t
∂) 
k
)).
Summing over a partition of unity and using the fact that
¯
∂ ⊕ ϑ
t
is elliptic, the
small and large constants technique gives
(6.1.8)  ∂
∗
t
·
t
∂) 
2
k
≤ C
k
( ) 
2
k
+  ·
t
∂) 
2
k
).
126 Boundary Regularity for ∂ on Pseudoconvex Domains
Using estimate (6.1.1) to ηT
k
·
t
∂), we obtain
(6.1.9) t  ηT
k
·
t
∂) 
2
(t)
≤  ∂ηT
k
·
t
∂) 
2
(t)
+  ∂
∗
t
ηT
k
·
t
∂) 
2
(t)
.
Since [
¯
∂, ηT
k
] +[ϑ
t
, ηT
k
] = 1
k
+1
t
k−1
where 1
k
is a /th order diﬀerential operator
independent of t and 1
t
k−1
is a diﬀerential operator of order / −1, it follows that
∂ηT
k
·
t
∂) 
2
(t)
+  ∂
∗
t
ηT
k
·
t
∂) 
2
(t)
≤ C
k
(  ∂
∗
t
·
t
∂) 
2
k
+  ·
t
∂) 
2
k
) +C
k,t
 ·
t
¯
∂) 
2
k−1
.
We see, using induction, that
¯
∂
∗
t
·
t
¯
∂ is bounded on \
k−1
(1). The above argument
implies that ·
t
¯
∂ is also bounded on \
k−1
(1). Thus, after summing over a partition
of unity, if t is chosen to be suﬃciently large in (6.1.9), we obtain that
(6.1.10)  ·
t
∂) 
2
k
≤ c  ∂
∗
t
·
t
∂) 
2
k
+ C
k,t
 ) 
2
k−1
,
for some small c 0. Letting c be suﬃciently small, (6.1.8) and (6.1.10) together
show
 ∂
∗
t
·
t
∂) 
2
k
≤ C
k,t
 ) 
2
k
.
This proves the a priori estimate for ∂
∗
t
·
t
∂ and Theorem 6.1.4.
We remark that the positive number o
k
in Theorem 6.1.4 can be chosen to be
the same as in Theorem 6.1.3.
Corollary 6.1.5. Let 1 be a smooth bounded pseudoconvex domain in C
n
, n ≥ 2.
If ) ∈ \
k
(p,q)
(1), / ≥ 0, such that ∂) = 0, where 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n, then
there exists n ∈ \
k
(p,q−1)
(1) such that ∂n = ) on 1.
Proof. Since ∂) = 0, using Theorem 6.1.2 n = ∂
∗
t
·
t
) is a solution to the ∂ equation
for any t 0. If t is suﬃciently large,
¯
∂
∗
t
·
t
is bounded on \
k
(p,q)
(1) by Theorem
6.1.4. This proves the corollary.
Proof of Theorem 6.1.1. From Corollary 6.1.5, there is n
k
∈ \
k
(p,q−1)
(1) satisfying
∂n
k
= ) for each positive integer /. We shall modify n
k
to generate a new sequence
that converges to a smooth solution.
We claim that \
m
(p,q)
(1) ∩ Ker(∂) is dense in \
s
(p,q)
(1) ∩ Ker(∂) for any :
: ≥ 0.
Let p
n
∈ C
∞
(p,q)
(1) be any sequence such that p
n
→ p in \
s
(p,q)
(1). Using
Theorem 6.1.4, for suﬃciently large t, the Bergman projection with weight 1
t
=
1
t,(p,q)
is bounded on \
m
(p,q)
(1). Since ∂p = 0, we have p − 1
t
p =
¯
∂
∗
t
·
t
¯
∂p = 0.
Thus, 1
t
p
n
= p
t
n
∈ \
m
(p,q)
(1), ∂p
t
n
= 0 and p
t
n
→ p in \
s
(p,q)
(1) since 1
t
is also
bounded on \
s
(p,q)
(1). This proves the claim.
Since n
k
−n
k+1
is in \
k
(p,q−1)
(1) ∩ Ker(∂), there exists a ·
k+1
∈ \
k+1
(p,q−1)
(1) ∩
Ker(∂) such that
 n
k
−n
k+1
−·
k+1

k
≤ 2
−k
, / = 1, 2, .
6.2 Sobolev Estimates for the ∂Neumann Operator 127
Setting ˜ n
k+1
= n
k+1
+·
k+1
, then ˜ n
k+1
∈ \
k+1
(p,q−1)
(1) and ∂˜ n
k+1
= ). Inductively,
we can choose a new sequence ˜ n
k
∈ \
k
(p,q−1)
(1) such that ∂˜ n
k
= ) and
 ˜ n
k+1
− ˜ n
k

k
≤ 2
−k
, / = 1, 2, .
Set
n
∞
= ˜ n
N
+
∞
k=N
(˜ n
k+1
− ˜ n
k
), · ∈ N.
Then n
∞
is well deﬁned and is in \
N
(p,q−1)
(1) for every ·. Thus n
∞
∈ C
∞
(p,q−1)
(1)
from the Sobolev embedding theorem and ∂n
∞
= ). This proves Theorem 6.1.1.
We also obtain the following result in the proof:
Corollary 6.1.6. Let 1 be a smooth bounded pseudoconvex domain in C
n
, n ≥ 2.
Then C
∞
(p,q)
(1) ∩ Ker(∂) is dense in \
s
(p,q)
(1) ∩ Ker(∂) in the \
s
(p,q)
(1) norm,
where 0 ≤ j ≤ n and 0 ≤ ¡ ≤ n. In particular, C
∞
(1) ∩ O(1) is dense in H(1)
in 1
2
(1), where H(1) is the space of all square integrable holomorphic functions.
6.2 Sobolev Estimates for the ∂Neumann Operator
In this section, using the vector ﬁeld method we shall give a suﬃcient condi
tion for verifying global regularity of the ∂Neumann problem on a certain class of
smooth bounded pseudoconvex domains. In particular, this method can be applied
to convex domains and circular domains with transverse symmetries.
Let 1 ⊆ C
n
, n ≥ 2, be a smooth bounded pseudoconvex domain, and let : be a
smooth deﬁning function for 1. Set
1
n
=
4
[∇:[
2
n
j=1
∂:
∂.
j
∂
∂.
j
,
if [∇:[ , = 0, and
1
jk
=
∂:
∂.
j
∂
∂.
k
−
∂:
∂.
k
∂
∂.
j
, for 1 ≤ , < / ≤ n.
We have 1
n
: = 1 in a neighborhood of the boundary and the 1
jk
’s are tangent to
the level sets of :. Also, the 1
jk
’s span the space of tangential type (1, 0) vector
ﬁelds at every boundary point of 1.
Denote by A
n
= ([∇:[,
√
2)1
n
the globally deﬁned type (1, 0) vector ﬁeld which
is transversal to the boundary everywhere. Obviously, we have  A
n
 = 1
in some open neighborhood of the boundary. Thus, near every boundary point
j ∈ /1, we may choose tangential type (1, 0) vector ﬁelds A
1
, , A
n−1
so that
A
1
, , A
n−1
together with A
n
form an orthonormal basis of the space of type
(1,0) vector ﬁelds in some open neighborhood of j. We shall also denote by
ω
1
, , ω
n
the orthonormal frame of (1,0)forms dual to A
1
, , A
n
near j. Note
that ω
n
= (
√
2,[∇:[)
n
j=1
(∂:,∂.
j
)d.
j
is a globally deﬁned (1,0)form in some open
neighborhood of the boundary.
The main idea of this method is to construct a real tangential vector ﬁeld T on
some open neighborhood of the boundary such that the commutators of T with
A
1
, , A
n
, A
1
, , A
n
have small modulus in A
n
direction on the boundary. We
formulate the required properties of T in the following condition:
128 Boundary Regularity for ∂ on Pseudoconvex Domains
Condition (T). For any given c 0 there exists a smooth real vector ﬁeld T = T
,
depending on c, deﬁned in some open neighborhood of 1 and tangent to the boundary
with the following properties:
(1) On the boundary, T can be expressed as
T = o
(.)(1
n
−1
n
), mod (T
1,0
(/1) ⊕T
0,1
(/1)),
for some smooth function o
(.) with [o
(.)[ ≥ δ 0 for all . ∈ /1, where δ
is a positive constant independent of c.
(2) If o is any one of the vector ﬁelds 1
n
, 1
n
, 1
jk
and 1
jk
, 1 ≤ , < / ≤ n,
then
[T, o][
bD
= ¹
S
(.)1
n
, mod (T
1,0
(/1) ⊕T
0,1
(/1), 1
n
),
for some smooth function ¹
S
(.) with sup
bD
[¹
S
(.)[ < c.
Here is a simple observation: Near a boundary point j, we have, say, ∂:,∂.
n
(j) ,=
0. Thus, for each , = 1, , n −1, we may write
A
j
=
n−1
k=1
c
jk
1
kn
,
for some smooth functions c
jk
. It follows that if condition (T) holds on 1, then
property (2) of condition (T) is still valid with o being taken to be A
j
’s or A
j
’s for
, = 1, , n −1, where A
j
’s are deﬁned as above in some small open neighborhood
of j.
Now we are in a position to prove the main theorem of this section.
Theorem 6.2.1. Let 1 be a smooth bounded pseudoconvex domain in C
n
, n ≥ 2,
with a smooth deﬁning function :. Suppose that condition (T) holds on 1. Then
the ∂Neumann operator · maps \
s
(p,q)
(1), 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n, boundedly into
itself for each nonnegative real :.
Proof. We shall prove the theorem only for nonnegative integers. For any nonneg
ative real :, the assertion will follow immediately from interpolation (see Theorem
B.3 in the Appendix).
In view of the elliptic regularization method employed in Chapter 5, it suﬃces
to prove a priori estimates for the ∂Neumann operator. The proof will be done by
induction on the order of diﬀerentiation. Let us assume that the given (j, ¡)form
) ∈ C
∞
(p,q)
(1) and the solution n = ·
q
) to the equation n = (∂∂
∗
+∂
∗
∂)n = ) is
also in C
∞
(p,q)
(1).
The initial step : = 0 is obvious, since · is a bounded operator by Theorem
4.4.1. To illustrate the idea, we prove the case : = 1 in detail. First, we choose
boundary coordinate charts ¦l
α
¦
m
α=1
such that ¦l
α
¦
m
α=1
and l
0
= 1 form an open
cover of 1. We shall assume that [∇:[ c 0 on ∪
m
α=1
l
α
for some positive
constant c. Let ¦ζ
α
¦
m
α=0
be a ﬁxed partition of unity subordinate to ¦l
α
¦
m
α=0
. On
each l
α
, 1 ≤ α ≤ :, let ω
αk
, / = 1, , n be an orthonormal frame of (1,0)forms
6.2 Sobolev Estimates for the ∂Neumann Operator 129
dual to A
αk
, / = 1, , n. We note that ω
αn
= ω
n
, α = 1, , :, is a globally
deﬁned (1,0)form dual to A
n
= ([∇:[,
√
2)1
n
. Similarly, A
αn
= A
n
, α = 1, , :,
is also a globally deﬁned type (1,0) vector ﬁeld.
On each boundary chart l
α
, α = 1, , :, we may write
n = n
α
=
t
n
α
I,J
ω
α,I
∧ ω
α,J
,
where ω
α,I
= ω
αi1
∧ ∧ ω
αip
and ω
α,J
= ω
αj1
∧ ∧ ω
αjq
. Let T be the smooth
real vector ﬁeld on 1 satisfying the hypothesis of condition (T). For each : ∈ N, we
deﬁne
(6.2.1) T
s
n = T
s
(ζ
0
n) +
m
α=1
t
I,J
T
s
(ζ
α
n
α
I,J
) ω
α,I
∧ ω
α,J
.
Thus, T
s
n ∈ T
(p,q)
using Lemma 5.2.2.
Set
 An 
2
=
m
α=1
t
I,J
n
k=1
 A
αk
(ζ
α
n
α
I,J
) 
2
,
and
 A
t
n 
2
=
m
α=1
t
I,J
n−1
k=1
 A
αk
(ζ
α
n
α
I,J
) 
2
.
From estimate (4.3.1), with φ = 0, and (4.4.6) we obtain, using Theorem 4.4.1,
(6.2.2)
 An 
2
+  n 
2
 ∂n 
2
+  ∂
∗
n 
2
=  ∂·) 
2
+  ∂
∗
·) 
2
 ) 
2
.
Since, by integration by parts, for 1 ≤ α ≤ : and 1 ≤ / ≤ n −1,
 A
αk
(ζ
α
n
α
I,J
) 
2
=  A
αk
(ζ
α
n
α
I,J
) 
2
+ O( An  +  Tn  +  n )  n ,
using small and large constants, we have
(6.2.3)  A
t
n 
2
 An 
2
+  n 
2
+ (:c)  Tn 
2
.
Here (:c) denotes a small constant that can be made as small as we wish. Esti
mates (6.2.2) and (6.2.3) together with the interior estimate indicate that if one
can control  Tn , then  n 
1
can be estimated. For this reason we shall call
A
α1
, , A
αn−1
, A
α1
, , A
αn−1
and A
αn
, 1 ≤ α ≤ :, “good” directions.
Our aim thus becomes to estimate  Tn . First, the basic estimate shows
 Tn 
2
 ∂Tn 
2
+  ∂
∗
Tn 
2
.
130 Boundary Regularity for ∂ on Pseudoconvex Domains
We estimate the righthand side as follows.
 ∂Tn 
2
= (∂Tn, ∂Tn)
= (T∂n, ∂Tn) + ([∂, T]n, ∂Tn)
= (∂n, −∂T
2
n) + (∂n, [∂, T]Tn) + ([∂, T]n, ∂Tn)
+ O( ∂Tn  ∂n )
= (∂n, −∂T
2
n) + (∂n, [[∂, T], T]n) + (−T∂n, [∂, T]n) + ([∂, T]n, ∂Tn)
+O(( ∂Tn  +  n 
1
)  ∂n )
= (∂
∗
∂n, −T
2
n) + (∂n, [[∂, T], T]n) + (−∂Tn, [∂, T]n)
+  [∂, T]n 
2
+ ([∂, T]n, ∂Tn) +O(( ∂Tn  +  n 
1
)  ∂n ).
Note that
Re¦(−∂Tn, [∂, T]n) + ([∂, T]n, ∂Tn)¦ = 0.
With similar estimates for  ∂
∗
Tn 
2
, we obtain
 ∂Tn 
2
+  ∂
∗
Tn 
2
 ) 
2
1
+ (:c)  Tn 
2
+ (:c)  n 
2
1
+  [∂, T]n 
2
+  [∂
∗
, T]n 
2
.
In order to estimate the crucial commutator terms  [∂, T]n  and  [∂
∗
, T]n ,
we shall use the hypothesis on T. First, from our observation right before Theorem
6.2.1, it is easy to see that on each boundary coordinate chart the commutators
between T and A
1
, , A
n−1
, A
1
, , A
n−1
can be controlled using the hypothesis
on T. Thus, we need to consider the commutator between T and A
n
(or A
n
) which
occurs, when commuting T with ∂ (or ∂
∗
), only for those multiindices (1, J) with
n , ∈ J (or n ∈ J). Such terms can be handled as follows:
[A
n
, T](ζ
α
n
α
I,J
) = [([∇:[,
√
2)1
n
, T](ζ
α
n
α
I,J
)
= ([∇:[,
√
2)[1
n
, T](ζ
α
n
α
I,J
) −(T([∇:[,
√
2))1
n
(ζ
α
n
α
I,J
)
= ([∇:[,
√
2)[1
n
, T](ζ
α
n
α
I,J
) −(T([∇:[),[∇:[)A
n
(ζ
α
n
α
I,J
),
for α = 1, , :. Using the basic estimate, we obtain
m
α=1
t
n/ ∈J
 (T([∇:[),[∇:[)A
n
(ζ
α
n
α
I,J
) 
2
 ∂n 
2
+  ∂
∗
n 
2
 ) 
2
.
The remaining commutator terms can be estimated directly using the hypothesis
on T. Thus,
 [∂, T]n 
2
m
α=1
t

¹
o
T(ζ
α
n
α
I,J
) 
2
+  An 
2
+  A
t
n 
2
+  ) 
2
1
_
c
δ
_
2
 Tn 
2
+  An 
2
+  A
t
n 
2
+  ) 
2
1
.
6.2 Sobolev Estimates for the ∂Neumann Operator 131
For  [∂
∗
, T]n  we commute T with A
n
if n ∈ J. Hence,
[A
n
, T](ζ
α
n
α
I,J
) = ([∇:[,
√
2)[1
n
, T](ζ
α
n
α
I,J
) −(T([∇:[),[∇:[)A
n
(ζ
α
n
α
I,J
),
for α = 1, , :. Observe that ±A
n
(ζ
α
n
α
I,J
) appears in the coeﬃcient of ω
α,I
∧ω
α,H
with ¦n¦ ∪ H = J in ∂
∗
n. Meanwhile, all the other terms in the coeﬃcient of
ω
α,I
∧ ω
α,H
are diﬀerentiated by A
1
, , A
n−1
only. Thus we have
m
α=1
t
n∈J
 (T([∇:[),[∇:[)A
n
(ζ
α
n
α
I,J
) 
2
 ∂
∗
n 
2
+  A
t
n 
2
+  ) 
2
1
,
and get the estimates as before. Here we use  ) 
2
1
to control the interior term.
Now we ﬁrst choose (:c) to be small enough, and then c to be suﬃciently small.
From (6.2.3) we obtain
 [∂, T]n 
2
+  [∂
∗
, T]n 
2
 An 
2
+  ) 
2
1
+ γ  Tn 
2
,
where γ 0 is a constant that can be made as small as we wish. Combining these
estimates, if we let γ be small enough, we get
 Tn 
2
 ∂Tn 
2
+  ∂
∗
Tn 
2
 ) 
2
1
+ (:c)  n 
2
1
.
This implies
 n 
1
≤ C  ) 
1
,
for some constant C 0 independent of ), and the proof for : = 1 is thus complete.
Assume that the ∂Neumann operator · is bounded on \
s−1
(p,q)
(1) for some
integer : ≥ 2, i.e.,
 n 
s−1
≤ C
s−1
 ) 
s−1
,
where C
s−1
0 is a constant independent of ). The strategy here is the same as
before. Using basic estimate and the induction hypothesis, we ﬁrst establish the
following a priori estimate
(6.2.4)
 T
s
n 
2
 ∂T
s
n 
2
+  ∂
∗
T
s
n 
2
 ) 
2
s
+ (:c)  n 
2
s
.
The next step is to consider the action of an arbitrary tangential diﬀerential op
erator of order :. Using (6.2.4), it suﬃces to consider the estimate near a boundary
point j. Let l be a boundary coordinate chart near j, and let ω
1
, , ω
n
be an
orthonormal basis for (1,0)forms on l dual to A
1
, , A
n
deﬁned as before. De
note by Op(:, ,), 1 ≤ , ≤ :, a tangential diﬀerential operator of order : formed out
of A
1
, , A
n−1
, A
1
, , A
n−1
and T with precisely : − , factors of T. Let ζ be
a cutoﬀ function supported in l such that ζ ≡ 1 in some open neighborhood of j.
We claim that
(6.2.5)  Op(:, ,)ζn 
2
 ) 
2
s
+ (:c)  n 
2
s
,
132 Boundary Regularity for ∂ on Pseudoconvex Domains
for all 0 ≤ , ≤ :. Estimate (6.2.5) will be proved by induction on ,. The initial
step , = 0 is done by (6.2.4). Hence we assume Estimate (6.2.5) holds up to , −1
for some 1 ≤ , ≤ :. We need to show that (6.2.5) is also true for ,. Denote by A
t
any one of the vector ﬁelds A
1
, , A
n−1
. Then, by commuting one A
t
or A
t
to
the left and applying integration by parts, it is easily seen that
 Op(:, ,)ζn 
2
 ) 
2
s
+  A
t
Op(: −1, , −1)ζn 
2
+  Op(:, , −1)ζn 
2
.
Thus, (6.2.5) is proved inductively for all 0 ≤ , ≤ :.
Estimate (6.2.5) shows that all the tangential derivatives of order : can be con
trolled. Finally, using the noncharacteristic nature of the ∂Neumann problem, we
also control the diﬀerentiation in the normal direction. Therefore, a partition of
unity argument gives
 n 
s
 ) 
s
+ (:c)  n 
s
which implies, by choosing (:c) suﬃciently small,
 n 
s
 ) 
s
.
Hence, by an induction argument the proof of Theorem 6.2.1 is now complete.
Theorem 6.2.1 provides us with a method for verifying the regularity of the ∂
Neumann operator ·. Once the regularity of · is known, we may also obtain the
regularity of other operators related to the ∂Neumann operator · as shown in the
next theorem.
Theorem 6.2.2. Let 1 be a smooth bounded pseudoconvex domain in C
n
, n ≥ 2.
If the ∂Neumann operator · is exactly regular on \
s
(p,q)
(1) for 0 ≤ j ≤ n,
1 ≤ ¡ ≤ n and : ≥ 0, then so are the operators ∂·, ∂
∗
·, ∂∂
∗
·, ∂
∗
∂· and the
Bergman projection 1
(p,0)
.
Proof. The exact regularity of the operators ∂·, ∂
∗
·, ∂∂
∗
· and ∂
∗
∂· can be
proved as in Theorem 6.1.4.
For the regularity of the Bergman projections 1
(p,0)
, we may assume j = 0.
Denote as before by 1 the Bergman projection on functions and by 1
t
the weighted
Bergman projection. Let Φ
t
be the multiplication operator by the weight c
−t]z]
2
.
Then for any square integrable holomorphic function p and any square integrable
function ), we have
(1), p) = (), p) = (Φ
−t
), p)
t
= (1
t
Φ
−t
), p)
t
= (Φ
t
1
t
(Φ
−t
)), p).
Hence, we get 1 = 1Φ
t
1
t
Φ
−t
. Recall that 1 = 1 −∂
∗
·∂ on smooth bounded pseu
doconvex domains (Theorem 4.4.5). Thus, the Bergman projection 1 on functions
can be expressed as
1 = Φ
t
1
t
Φ
−t
−∂
∗
·
1
((∂Φ
t
)1
t
Φ
−t
).
Since it has been proved that ∂
∗
·
1
preserves \
s
(0,1)
(1) for each nonnegative real
:, if we choose t to be suﬃciently large, Theorem 6.1.4 implies that 1 maps \
s
(1)
boundedly into itself. This proves the theorem.
6.2 Sobolev Estimates for the ∂Neumann Operator 133
Now, we will construct the vector ﬁeld T when the domain 1 has a plurisub
harmonic deﬁning function or transverse circular symmetry. We say that a smooth
bounded domain 1 has a plurisubharmonic deﬁning function if there exists some
smooth deﬁning function :(.) for 1 satisfying
(6.2.6)
n
j,k=1
∂
2
:
∂.
j
∂.
k
(.)t
j
t
k
≥ 0 for . ∈ /1 and t ∈ C
n
.
Note that (6.2.6) is required to hold only on the boundary, and not in a neighbor
hood of the boundary. Also, (6.2.6) implies 1 must be pseudoconvex.
Theorem 6.2.3. Let 1 ⊆ C
n
, n ≥ 2, be a smooth bounded pseudoconvex domain
admitting a plurisubharmonic deﬁning function :(.). Then the ∂Neumann operator
· is exactly regular on \
s
(p,q)
(1) for 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n and all real : ≥ 0.
Proof. The proof is based on the observation that if :(.) is a deﬁning function for 1
that is plurisubharmonic on the boundary, then for each ,, derivatives of (∂:,∂.
j
) of
type (0, 1) in directions that lie in the null space of the Levi form must vanish. For
instance, if coordinates are chosen so that (∂,∂.
1
)(j) lies in the null space of the Levi
form at j ∈ /1, then (∂
2
:,∂.
1
∂.
1
)(j) = 0. Since :(.) is plurisubharmonic on the
boundary, applying the complex Hessian ((∂
2
:,∂.
j
∂.
k
)(j))
n
j,k=1
to (α, 1, 0, , 0) ∈
C
n
for any α ∈ C, we obtain
2Re
_
α
∂
2
:
∂.
2
∂.
1
(j)
_
+
∂
2
:
∂.
2
∂.
2
(j) ≥ 0,
which forces (∂
2
:,∂.
2
∂.
1
)(j) = 0. Similarly, we get (∂
2
:,∂.
j
∂.
1
)(j) = 0 for 1 ≤
, ≤ n.
Fix a point j in the boundary, then (∂:,∂.
j
)(j) ,= 0 for some 1 ≤ , ≤ n.
Choose such a , and set A
p
= (∂:,∂.
j
)
−1
(∂,∂.
j
) in a neighborhood of j. Let
1
1
, , 1
n−1
be a local basis for the space of type (1, 0) tangential vector ﬁelds near
j. We may assume that 1
1
, , 1
n−1
are tangent to the level sets of : and that
the Levi form is diagonal at j in this basis, namely, ¸[1
k
, 1
j
], 1
n
)(j) is a diagonal
(n − 1) (n − 1) matrix, where ¸ , ) is the standard Hermitian inner product.
Now if the /th eigenvalue is zero, then by the above observation 1
k
annihilates the
coeﬃcients of A
p
at j. It follows that the system of linear equations
n−1
j=1
¸[1
k
, 1
j
], 1
n
)(j)o
j
= ¸[1
k
, A
p
], 1
n
)(j), 1 ≤ / ≤ n −1,
is solvable for o
1
, , o
n−1
∈ C. This implies that
__
A
p
−
n−1
j=1
o
j
1
j
, 1
k
_
, 1
n
_
(j) = 0
for / = 1, , n − 1. On the other hand, since A
p
(:) = 1 = 1
n
(:), there exist
scalars /
1
, , /
n−1
such that
1
n
−
n−1
j=1
/
j
1
j
= A
p
−
n−1
j=1
o
j
1
j
.
134 Boundary Regularity for ∂ on Pseudoconvex Domains
Therefore, given c 0, it is easily veriﬁed by using a partition of unity ¦ζ
l
¦
m
l=1
with small support that one may patch 1
n
−
n−1
j=1
/
j
1
j
together to form a globally
deﬁned type (1, 0) vector ﬁeld A = 1
n
− Y , where Y =
m
l=1
ζ
l
(
n−1
j=1
/
j
1
j
) is a
globally deﬁned tangential type (1, 0) vector ﬁeld, such that
(6.2.7) sup
bD
[¸[A, 1
k
], 1
n
)[ < c,
for 1 ≤ / ≤ n −1. Since A − A is a purely imaginary tangential vector ﬁeld, the
commutator [A −A, 1
k
], 1 ≤ / ≤ n −1, is also a tangential vector ﬁeld. Thus, we
may write
[A −A, 1
k
] = α
k
(1
n
−1
n
), mod (T
1,0
(/1) ⊕T
0,1
(/1)),
for 1 ≤ / ≤ n −1. By (6.2.7), we have [α
k
[ < c on the boundary. It follows that
(6.2.8) sup
bD
[¸[A −A, o], 1
n
)[ < c,
when o is any one of the tangential vector ﬁelds 1
1
, , 1
n−1
, 1
1
, , 1
n−1
. In
particular, we may achieve that
(6.2.9) sup
bD
[¸[A −A, Y −Y ], 1
n
)[ < c,
where c might be diﬀerent from that in (6.2.8). Since 1
n
: = 1 in a neighborhood
of the boundary, we have
[A −A, 1
n
]: = (A −A)1
n
(:) −1
n
(A −A): = 0,
which implies the vector ﬁeld [A−A, 1
n
] is tangent to the level sets of :. It follows
that [A −A, Re1
n
] is also a tangential vector ﬁeld and we can write
(6.2.10) [A −A, Re1
n
] = β(1
n
−1
n
), mod (T
1,0
(/1) ⊕T
0,1
(/1)),
where β is a realvalued function deﬁned in some neighborhood of the boundary.
Now, we set
(6.2.11) T = c
rβ
(A −A).
Obviously, T depends on c and satisﬁes property (1) of condition (T) on the bound
ary. If o is any one of the vector ﬁelds 1
1
, , 1
n−1
, 1
1
, , 1
n−1
, we have
¸[T, o], 1
n
)[
bD
= ¸[A −A, o], 1
n
)[
bD
.
For 1
n
, using (6.2.10) we get
¸[T, 1
n
], 1
n
)[
bD
= ¸[c
rβ
(A −A), Re1
n
], 1
n
)[
bD
+
1
2
¸[c
rβ
(A −A), 1
n
−1
n
], 1
n
)[
bD
= 0 +
1
2
¸[A −A, Y −Y ], 1
n
)[
bD
.
6.2 Sobolev Estimates for the ∂Neumann Operator 135
Similarly, we obtain
¸[T, 1
n
], 1
n
)[
bD
= −
1
2
¸[A −A, Y −Y ], 1
n
)[
bD
.
Thus, by (6.2.8) and (6.2.9) we see that the vector ﬁeld T satisﬁes all the hypotheses
of condition (T). Hence, by Theorem 6.2.1, we have proved Theorem 6.2.3.
Theorem 6.2.3 gives a suﬃcient condition for verifying the exact regularity of the
∂Neumann operator. However, this condition in general is not satisﬁed by every
smooth bounded pseudoconvex domain. For instance, by Theorem 6.4.2, the worm
domain constructed in Section 6.4 does not enjoy this property. Next, we show that
this condition indeed holds on any smooth bounded convex domain. Hence, the
∂Neumann problem is exactly regular on any convex domain.
Let 1 ⊆ 1
N
, · ≥ 2, be a smooth bounded convex domain, and let the origin be
contained in 1. For any r ∈ 1
N
, the Minkowski functional j(r) is deﬁned by
(6.2.12) j(r) = inf¦λ 0[ r ∈ λ1¦,
where λ1 = ¦λj[ j ∈ 1¦. Since the boundary of 1 is smooth, j(r) is smooth on
1
N
¸ ¦0¦. If r ,= 0, then the ray
−→
or will intersect the boundary /1 at exactly one
point, named r
t
. It is easy to see that the Minkowski functional j(r) is equal to
the ratio between d(0, r) and d(0, r
t
), where d(j, ¡) = dist(j, ¡). Hence, we have
r = j(r)r
t
.
Lemma 6.2.4. Let 1 be a smooth bounded convex domain in 1
N
containing the
origin, and let the Minkowski functional j(r) be deﬁned as in (6.2.12). Then j(r)
is a smooth, realvalued function on 1
N
¸ ¦0¦ satisfying the following properties:
(1) j(r) is a deﬁning function for 1, i.e., j(r) = 1 and ∇j(r) ,= 0 for r ∈ /1,
(2) j(r +j) ≤ j(r) +j(j) for r, j ∈ 1
N
,
(3) j(or) = oj(r) for r ∈ 1
N
and o 0.
Proof. Obviously, j(r) is a smooth, realvalued function on 1
N
¸ ¦0¦. (1) and (3)
are also clear. To prove (2), let r, j ,= 0 be two points in 1
N
, and let r
t
, j
t
be the
intersections with the boundary of the rays
−→
or and
−→
oj respectively. Then we have
j(r +j) = j(j(r)r
t
+j(j)j
t
)
= (j(r) +j(j))j
_
j(r)
j(r) +j(j)
r
t
+
j(j)
j(r) +j(j)
j
t
_
≤ j(r) +j(j).
Here we have used the fact that 1 is convex so that the point . = (j(r),(j(r) +
j(j)))r
t
+ (j(j),(j(r) + j(j)))j
t
lies in the closure of 1. Hence, j(.) ≤ 1. This
proves the lemma.
It follows from (2) and (3) of Lemma 6.2.4 that j(r) is convex. Consequently,
(6.2.13)
N
j,k=1
∂
2
j
∂r
j
∂r
k
(r)o
j
o
k
≥ 0,
for r ∈ 1
N
¸ ¦0¦ and any o = (o
1
, , o
N
) ∈ 1
N
. In particular, we have proved
the following theorem:
136 Boundary Regularity for ∂ on Pseudoconvex Domains
Theorem 6.2.5. Let 1 be a smooth bounded convex domain in C
n
, n ≥ 2. Then
the ∂Neumann problem is exactly regular on \
s
(p,q)
(1) for 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n
and : ≥ 0.
Another important class of pseudoconvex domains that satisfy the hypotheses of
condition (T) are circular domains with transverse symmetries. A domain 1 in C
n
is called circular if c
iθ
. = (c
iθ
.
1
, , c
iθ
.
n
) ∈ 1 for any . ∈ 1 and θ ∈ 1. 1 is
called Reinhardt if (c
iθ1
.
1
, , c
iθn
.
n
) ∈ 1 for any . ∈ 1 and θ
1
, , θ
n
∈ 1, and
1 is called complete Reinhardt if . = (.
1
, , .
n
) ∈ 1 implies (n
1
, , n
n
) ∈ 1
for all [n
j
[ ≤ [.
j
[, 1 ≤ , ≤ n. Thus, a Reinhardt domain is automatically circular.
Let 1 be a smooth bounded circular domain in C
n
, n ≥ 2, and let :(.) be deﬁned
as follows
(6.2.14) :(.) =
_
d(., /1), for . , ∈ 1
−d(., /1), for . ∈ 1,
where d(., /1) denotes the distance from . to the boundary /1. Then it is easy to
see that : is a deﬁning function for 1 such that :(.) = :(c
iθ
.) and that [∇:[ = 1
on the boundary. Denote by Λ the map of the o
1
action on 1 from o
1
1 to 1
deﬁned by
Λ : o
1
1 →1
(c
iθ
, .) →c
iθ
. = (c
iθ
.
1
, , c
iθ
.
n
).
For each ﬁxed θ, Λ is an automorphism of 1 and Λ can be extended smoothly to
a map from o
1
1 to 1. Hence, for each ﬁxed . ∈ 1, we consider the orbit of .,
namely, the map
π
z
: o
1
→1
c
iθ
→c
iθ
..
Then, π
z
induces a vector ﬁeld T on 1, in fact on C
n
, by
(6.2.15) T
z
= π
z,∗
_
∂
∂θ
¸
¸
¸
¸
θ=0
_
= i
n
j=1
.
j
∂
∂.
j
−i
n
j=1
.
j
∂
∂.
j
,
where π
z,∗
is the diﬀerential map induced by π
z
. Note that T is tangent to the level
sets of :. In particular, T is tangent to the boundary of 1.
Deﬁnition 6.2.6. Let 1 be a compact subset of the boundary of a smooth bounded
circular domain 1. 1 is said to have transverse circular symmetry on 1 if for each
point . ∈ 1 the vector ﬁeld T deﬁned in (6.2.15) is not contained in T
1,0
z
(/1) ⊕
T
0,1
z
(/1).
It is obvious from (6.2.15) and Deﬁnition 6.2.6 that 1 has transverse circular
symmetry on the whole boundary if and only if
n
j=1
.
j
(∂:,∂.
j
)(.) ,= 0 on /1.
Then, we prove
6.2 Sobolev Estimates for the ∂Neumann Operator 137
Theorem 6.2.7. Let 1 ⊆ C
n
, n ≥ 2, be a smooth bounded circular pseudoconvex
domain and let : be deﬁned by (6.2.14). Suppose that
n
j=1
.
j
(∂:,∂.
j
)(.) ,= 0 on
the boundary. Then the ∂Neumann problem is exactly regular on \
s
(p,q)
(1) for
0 ≤ j ≤ n, 1 ≤ ¡ ≤ n and : ≥ 0.
Proof. Let T be the vector ﬁeld deﬁned in (6.2.15). By assumption T is transversal
to T
1,0
(/1) ⊕T
0,1
(/1) everywhere on the boundary. Let 1
t
n
= ([∇:[
2
,4)1
n
and
1
jk
=
∂:
∂.
j
∂
∂.
k
−
∂:
∂.
k
∂
∂.
j
, for 1 ≤ , < / ≤ n.
It is easy to verify that [T, ∂,∂.
j
] = i∂,∂.
j
and [T, ∂,∂.
j
] = −i∂,∂.
j
. Then, we
have
[T, 1
jk
] =
_
T,
∂:
∂.
j
∂
∂.
k
−
∂:
∂.
k
∂
∂.
j
_
= T
_
∂:
∂.
j
_
∂
∂.
k
+
∂:
∂.
j
_
T,
∂
∂.
k
_
−T
_
∂:
∂.
k
_
∂
∂.
j
−
∂:
∂.
k
_
T,
∂
∂.
j
_
=
_
∂
∂.
j
(T:)
_
∂
∂.
k
+
__
T,
∂
∂.
j
_
:
_
∂
∂.
k
−i
∂:
∂.
j
∂
∂.
k
−
_
∂
∂.
k
(T:)
_
∂
∂.
j
−
__
T,
∂
∂.
k
_
:
_
∂
∂.
j
+i
∂:
∂.
k
∂
∂.
j
= −2i
_
∂:
∂.
j
∂
∂.
k
−
∂:
∂.
k
∂
∂.
j
_
= −2i1
jk
,
for all 1 ≤ , < / ≤ n, since T: ≡ 0, and
[T, 1
t
n
] =
_
T,
n
j=1
∂:
∂.
j
∂
∂.
j
_
=
n
j=1
_
T
_
∂:
∂.
j
_
∂
∂.
j
+
∂:
∂.
j
_
T,
∂
∂.
j
__
=
n
j=1
_
i
∂:
∂.
j
∂
∂.
j
−i
∂:
∂.
j
∂
∂.
j
_
≡ 0.
Similarly, we have [T, 1
t
n
] ≡ 0. Since 1
n
= (4,[∇:[
2
)1
t
n
and [∇:[ = 1 on /1, it is
easily seen that
[T, 1
n
][
bD
= [T, 1
n
][
bD
= 0.
Hence, condition (T) holds on 1. By Theorem 6.2.1 this proves Theorem 6.2.7.
The next result shows that a complete Reinhardt domain always enjoys transverse
circular symmetry.
138 Boundary Regularity for ∂ on Pseudoconvex Domains
Theorem 6.2.8. Let 1 ⊆ C
n
, n ≥ 2, be a smooth bounded complete Reinhardt
pseudoconvex domain with a smooth deﬁning function :(.) = :(c
iθ1
.
1
, , c
iθn
.
n
)
for all θ
1
, , θ
n
∈ 1. Then, we have
n
j=1
.
j
(∂:,∂.
j
) ,= 0 on /1. In particular,
the ∂Neumann problem is exactly regular on \
s
(p,q)
(1) for 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n
and : ≥ 0.
Proof. Let T be deﬁned by (6.2.15). Put 1 =
n
j=1
.
j
(∂,∂.
j
). Then, by our
construction, we have
T: = −2Im1(:) = 0
on the boundary. Hence, it suﬃces to show that Re1(:) ,= 0 on the boundary.
This is, in turn, equivalent to showing that the real vector ﬁeld
n
j=1
(r
j
(∂,∂r
j
) +
j
j
(∂,∂j
j
)) is transversal to the boundary everywhere.
Suppose now that for some point j ∈ /1 we have
n
j=1
_
r
j
∂:
∂r
j
+j
j
∂:
∂j
j
_
(j) = 0.
This implies that the point j = (r
1
(j), j
1
(j), , r
n
(j), j
n
(j)) is perpendicular
to the normal (∂:,∂r
1
, ∂:,∂j
1
, , ∂:,∂r
n
, ∂:,∂j
n
)(j). We may assume, by
rotation, that r
j
(j) 0 and j
j
(j) 0 for 1 ≤ , ≤ n. Hence, by elementary
tangent approximation, there exists a point ¡ ∈ 1 such that [r
j
(¡)[ [r
j
(j)[ and
[j
j
(¡)[ [j
j
(j)[ for 1 ≤ , ≤ n which in turn shows [.
j
(j)[ < [.
j
(¡)[ for 1 ≤ , ≤ n.
Since 1 is a complete Reinhardt domain, we must have j ∈ 1. This contradicts the
fact that j is a boundary point. In view of Theorem 6.2.7, the proof of Theorem
6.2.8 is now complete.
6.3 The Bergman Projection and Boundary Regularity of Biholomorphic
Maps
As an application of the regularity theorem proved earlier for the ∂Neumann
operator, we shall investigate the boundary regularity of a biholomorphic map in
this section. Recall that a holomorphic map ) between two domains 1
1
and 1
2
is called biholomorphic if ) is onetoone, onto and the inverse map )
−1
is also
holomorphic.
Let 1 be a domain in C
n
. we denote by H(1) the space of square integrable
holomorphic functions on 1 as before. Obviously, H(1) is a closed subspace of
1
2
(1), and hence is itself a Hilbert space. If 1 = C
n
, then H(C
n
) = ¦0¦. Thus,
we are interested in the case when H(1) is nontrivial, in particular, when 1 is
bounded. For any n ∈ 1, it is easily veriﬁed that the point evaluation map
Λ
w
: H(1) →C
) →)(n),
by Cauchy’s estimate, satisﬁes
(6.3.1) [)(n)[ ≤ cd(n)
−n
 ) 
L
2
(D)
,
6.3 The Bergman Projection 139
where d(n) is the distance from n to the complement of 1, and the constant c
depends only on the space dimension n. Hence, by the Riesz representation theorem,
there is a unique element, denoted by 1
D
(, n), in H(1) such that
)(n) = Λ
w
()) = (), 1
D
(, n)) =
_
D
)(.)1
D
(., n)d\
z
,
for all ) ∈ H(1). The function 1
D
(., n) thus deﬁned is called the Bergman kernel
function for 1. By (6.3.1) the Bergman kernel function clearly satisﬁes
(6.3.2)  1
D
(, n) 
L
2
(D)
≤ cd(n)
−n
,
for any n ∈ 1.
Next we verify a fundamental symmetry property of 1
D
(., n). We shall some
times omit the subscript 1 if there is no ambiguity.
Lemma 6.3.1. The Bergman kernel function 1(., n) satisﬁes
1(., n) = 1(n, .), for all ., n ∈ 1,
and hence 1(., n) is antiholomorphic in n.
Proof. For each n ∈ 1, 1(, n) ∈ H(1). Hence, by the reproducing property of
the kernel function, we obtain
1(., n) = (1(, n), 1(, .))
= (1(, .), 1(, n))
= 1(n, .).
This proves the lemma.
Since H(1) is a separable Hilbert space, the Bergman kernel function can also
be represented in terms of any orthonormal basis for H(1).
Theorem 6.3.2. Let ¦φ
j
(.)¦
∞
j=1
be an orthonormal basis for H(1). Then
(6.3.3) 1(., n) =
∞
j=1
φ
j
(.)φ
j
(n), for (., n) ∈ 1 1,
where the series (6.3.3) converges uniformly on any compact subset of 1 1. In
particular, 1(., n) is holomorphic in (., n) ∈ 1 1
∗
, where 1
∗
= ¦n [ n ∈ 1¦,
and hence 1(., n) ∈ C
∞
(1 1).
Proof. For any ﬁxed n ∈ 1, from general Hilbert space theory we have
1(., n) =
∞
j=1
(1(, n), φ
j
())φ
j
(.)
=
∞
j=1
(φ
j
(), 1(, n))φ
j
(.)
=
∞
j=1
φ
j
(n)φ
j
(.),
140 Boundary Regularity for ∂ on Pseudoconvex Domains
where the series converges in the 1
2
norm, and
(6.3.4)  1(, n) 
2
L
2
(D)
=
∞
j=1
[(1(, n), φ
j
())[
2
=
∞
j=1
[φ
j
(n)[
2
.
Since pointwise convergence is dominated by 1
2
convergence in H(1), we obtain the
pointwise convergence of (6.3.3). Therefore, to ﬁnish the proof, it suﬃces to show
by a normal family argument that [
m
j=1
φ
j
(.)φ
j
(n)[, for any : ∈ N, is uniformly
bounded on any compact subset of 1 1. Thus, letting ` be a compact subset
of 1, for any (., n) ∈ ` `, then (6.3.4) together with (6.3.2) shows
¸
¸
¸
¸
m
j=1
φ
j
(.)φ
j
(n)
¸
¸
¸
¸
≤
∞
j=1
[φ
j
(.)[[φ
j
(n)[
≤
_
_
∞
j=1
[φ
j
(.)[
2
_
_
1
2
_
_
∞
j=1
[φ
j
(n)[
2
_
_
1
2
≤ C
M
,
for some constant C
M
0 independent of :. This completes the proof of the
theorem.
The Bergman kernel function in general is not computable except for special
domains. When 1 is the unit ball 1
n
in C
n
, we shall apply Theorem 6.3.2 to
obtain an explicit formula for the Bergman kernel function on 1
n
. Obviously, ¦.
α
¦
is an orthogonal basis for H(1
n
), where the index α = (α
1
, , α
n
) runs over the
multiindices. We shall normalize it using the fact, for :, t ∈ N and 0 ≤ o < 1,
_
√
1−a
2
0
r
2s+1
_
1 −
r
2
1 −o
2
_
t+1
dr =
1
2
(1 −o
2
)
s+1
_
1
0
j
s
(1 −j)
t+1
dj
=
1
2
(1 −o
2
)
s+1
1(: + 1, t + 2)
=
1
2
(1 −o
2
)
s+1
Γ(: + 1)Γ(t + 2)
Γ(: +t + 3)
,
6.3 The Bergman Projection 141
where 1(, ) is the Beta function and Γ() is the Gamma function. Hence
 .
α

2
L
2
(Bn)
=
_
Bn
[.
1
[
2α1
[.
n
[
2αn
d\
2n
=
π
(α
n
+ 1)
_
Bn−1
[.
1
[
2α1
[.
n−1
[
2αn−1
(1 −[.
1
[
2
− −[.
n−1
[
2
)
αn+1
d\
2n−2
=
π
(α
n
+ 1)
_
Bn−1
[.
1
[
2α1
[.
n−2
[
2αn−2
(1 −[.
1
[
2
− −[.
n−2
[
2
)
αn+1
[.
n−1
[
2αn−1
_
1 −
[.
n−1
[
2
1 −[.
1
[
2
− −[.
n−2
[
2
_
αn+1
d\
2n−2
=
π
(α
n
+ 1)
πΓ(α
n−1
+ 1)Γ(α
n
+ 2)
Γ(α
n
+α
n−1
+ 3)
_
Bn−2
[.
1
[
2α1
[.
n−2
[
2αn−2
(1 −[.
1
[
2
− −[.
n−2
[
2
)
αn+αn−1+2
d\
2n−4
=
π
(α
n
+ 1)
πΓ(α
n−1
+ 1)Γ(α
n
+ 2)
Γ(α
n
+α
n−1
+ 3)
πΓ(α
1
+ 1)Γ(α
n
+ +α
2
+n)
Γ(α
n
+ +α
1
+n + 1)
=
π
n
α
1
! α
n
!
(α
n
+ +α
1
+n)!
.
It follows that the Bergman kernel function on the unit ball 1
n
is given by
1(., n) =
α
(α
n
+ +α
1
+n)!
π
n
α
1
! α
n
!
.
α
n
α
=
1
π
n
∞
k=0
]α]=k
(α
n
+ +α
1
+n)!
α
1
! α
n
!
.
α
n
α
=
1
π
n
∞
k=0
(/ +n)(/ +n −1) (/ + 1)(.
1
n
1
+ +.
n
n
n
)
k
=
1
π
n
d
n
dt
n
_
1
1 −t
_¸
¸
¸
¸
t=z1w1++znwn
=
n!
π
n
1
(1 −. n)
n+1
,
where . n = .
1
n
1
+ +.
n
n
n
.
Theorem 6.3.3. The Bergman kernel function on the unit ball 1
n
is given by
(6.3.5) 1(., n) =
n!
π
n
1
(1 −. n)
n+1
,
where . n = .
1
n
1
+ +.
n
n
n
.
For any ) ∈ 1
2
(1), one may write ) = )
1
+ )
2
, where )
1
∈ H(1) and
)
2
∈ H(1)
⊥
. It follows now from the reproducing property of the Bergman kernel
function that one has
1)(.) = )
1
(.) = ()
1
(), 1(, .))
= ()
1
(), 1(, .)) + ()
2
(), 1(, .))
= ()(), 1(, .)).
This proves the following theorem:
142 Boundary Regularity for ∂ on Pseudoconvex Domains
Theorem 6.3.4. The Bergman projection 1
D
: 1
2
(1) →H(1) is represented by
(6.3.6) 1
D
)(.) =
_
D
1(., n))(n)d\
w
,
for all ) ∈ 1
2
(1) and . ∈ 1.
The following result shows how the Bergman kernel function behaves under a
biholomorphic map:
Theorem 6.3.5. Let ) : 1
1
→ 1
2
be a biholomorphic map between two domains
1
1
and 1
2
in C
n
. Then
(6.3.7) 1
D1
(., n) = det)
t
(.)1
D2
()(.), )(n))det)
t
(n)
for all ., n ∈ 1
1
, where )
t
(.) is the complex Jacobian of ).
Proof. From an elementary calculation, we observe that
detJ
R
)(.) = [det)
t
(.)[
2
,
where J
R
)(.) is the real Jacobian of ) via the standard identiﬁcation between C
n
and 1
2n
. Hence, a change of variables shows / → (/ ◦ ))det)
t
is an isometry
between 1
2
(1
2
) and 1
2
(1
1
). Thus, for each n ∈ 1
1
, we have
det)
t
()1
D2
()(), )(n))det)
t
(n) ∈ H(1
1
),
and for any / ∈ H(1
1
), using the reproducing property of 1
D2
, we obtain
(/, det)
t
()1
D2
()(), )(n))det)
t
(n))
D1
= /(n).
Therefore, from the uniqueness of the kernel function, we must have
1
D1
(., n) = det)
t
(.)1
D2
()(.), )(n))det)
t
(n).
This proves the theorem.
Corollary 6.3.6. Let ) : 1
1
→ 1
2
be a biholomorphic map between two domains
1
1
and 1
2
in C
n
, and let 1
1
, 1
2
be the Bergman projection operator on 1
1
, 1
2
respectively. Then
(6.3.8) 1
1
(n (p ◦ ))) = n (1
2
(p) ◦ ))
for all p ∈ 1
2
(1
2
), where n = det()
t
(.)) is the determinant of the complex Jacobian
of ).
Proof. The proof follows directly from the transformation law of the Bergman kernel
functions. For p ∈ 1
2
(1
2
), n (p ◦ )) ∈ 1
2
(1
1
). Hence, from Theorem 6.3.5,
1
1
(n (p ◦ ))) =
_
D1
1
D1
(., n)det()
t
(n))p()(n)) d\
w
=
_
D1
n(.)1
D2
()(.), )(n))[n(n)[
2
p()(n)) d\
w
= n(.)
_
D2
1
D2
()(.), η)p(η) d\
η
= n(.) (1
2
(p) ◦ )).
This proves the corollary.
Now we introduce a condition concerning the regularity of the Bergman projec
tion operator which is useful in proving the regularity of a biholomorphic map near
the boundary.
6.3 The Bergman Projection 143
Deﬁnition 6.3.7. A smooth bounded domain 1 in C
n
is said to satisfy condition
1 if the Bergman projection 1 associated with 1 maps C
∞
(1) into C
∞
(1)∩O(1).
Denote by \
s
0
(1) the closure of C
∞
0
(1) in \
s
(1), and by H
s
(1) = \
s
(1) ∩
O(1). The next theorem gives various conditions equivalent to condition 1.
Theorem 6.3.8. Let 1 be a smooth bounded domain in C
n
with Bergman projec
tion 1 and Bergman kernel function 1(., n). The following conditions are equiva
lent:
(1) 1 satisﬁes condition 1.
(2) For each positive integer :, there is a nonnegative integer : = :
s
such that
1 is bounded from \
s+m
0
(1) to H
s
(1).
(3) For each multiindex α, there are constants c = c
α
and : = :
α
such that
sup
z∈D
¸
¸
¸
¸
∂
α
∂.
α
1(., n)
¸
¸
¸
¸
≤ cd(n)
−m
,
where d(n) is the distance from the point n to the boundary /1.
Before proving Theorem 6.3.8, we shall ﬁrst prove the following lemma:
Lemma 6.3.9. Let 1 be a smooth bounded domain in C
n
. Then, for each : ∈ N,
there is a linear diﬀerential operator Φ
s
of order n
s
= :(:+1),2 with coeﬃcients in
C
∞
(1) such that Φ
s
maps \
s+ns
(1) boundedly into \
s
0
(1) and that 1Φ
s
= 1.
In other words, for each p ∈ C
∞
(1) and : ∈ N, Lemma 6.3.9 allows us to
construct a / = p − Φ
s
p ∈ C
∞
(1) such that 1/ ≡ 0 and that / agrees with p up
to order : −1 on the boundary.
Proof. Let ρ be a smooth deﬁning function for 1, and let δ 0 be so small that
∇ρ ,= 0 on l
δ
= ¦.[ [ρ(.)[ < δ¦. Choose a partition of unity ¦φ
i
¦
m
i=1
and, for each
i, a complex coordinate .
i
in some neighborhood of the support of φ
i
such that
(1)
m
i=1
φ
i
≡ 1 on l
for some c < δ,
(2) suppφ
i
⊂ l
δ
and suppφ
i
∩ /1 ,= ∅, and
(3) ∂ρ,∂.
i
,= 0 on suppφ
i
.
To deﬁne the operator Φ
s
inductively on :, we need the fact that if p is in C
∞
(1)
and vanishes up to order : − 1 on the boundary, then p ∈ \
s
0
(1). For the initial
step : = 1, if / ∈ C
∞
(1), deﬁne
Φ
1
/ = / −
m
i=1
∂
∂.
i
(θ
i
0
ρ),
where θ
i
0
= (φ
i
/)(∂ρ,∂.
i
)
−1
. It is easy to see that Φ
1
/ = 0 on /1, and hence
Φ
1
/ ∈ \
1
0
(1).
Suppose θ
i
0
, , θ
i
s−1
have been chosen so that
Φ
s
i
/ = φ
i
/ −
∂
∂.
i
_
s−1
k=0
θ
i
k
ρ
k+1
_
144 Boundary Regularity for ∂ on Pseudoconvex Domains
vanishes to order : −1 on the boundary. We deﬁne
Φ
s
/ = / −
m
i=1
∂
∂.
i
_
s−1
k=0
θ
i
k
ρ
k+1
_
.
Since ρ vanishes on the boundary, it is easily veriﬁed by integration by parts that
m
i=1
∂
∂zi
(
s−1
k=0
θ
i
k
ρ
k+1
) is orthogonal to H(1). Hence Φ
s
/ ∈ \
s
0
(1) and 1Φ
s
/ =
1/. Put ∂,∂ν = ∇ρ ∇,[∇ρ[
2
, the normal diﬀerentiation, such that ∂ρ,∂ν = 1.
Let
θ
i
s
=
(
∂
∂ν
)
s
Φ
s
i
/
(: + 1)!
∂ρ
∂zi
.
Then the functions
Φ
s+1
i
/ = Φ
s
i
/ −
∂
∂.
i
(θ
i
s
ρ
s+1
)
vanish to order : on the boundary, so does the function
Φ
s+1
/ = Φ
s
/ −
m
i=1
∂
∂.
i
(θ
i
s
ρ
s+1
)
= / −
m
i=1
∂
∂.
i
_
s
k=0
θ
i
k
ρ
k+1
_
.
Hence, Φ
s+1
/ ∈ \
s+1
0
(1) and 1Φ
s+1
= 1. This completes the induction.
It is also easily veriﬁed by a simple induction argument that Φ
s
/ can be written
as
Φ
s
/ =
]α]≤k≤ns
/
α,k
ρ
k
1
α
/,
where n
s
= :(: + 1),2 and the /
α,k
’s are in C
∞
(1) and 1
α
is the real diﬀerential
operator of order [α[ associated to the multiindex α = (α
1
, α
2
, , α
2n
). This
proves the lemma.
We also need negative Sobolev norms for holomorphic functions. For p ∈ O(1)
and : a positive integer, we deﬁne
(6.3.9)  p 
−s
= sup
¸
¸
¸
¸
_
D
pφ
¸
¸
¸
¸
,
where the supremum is taken over all φ ∈ C
∞
0
(1) with  φ 
s
= 1.
Lemma 6.3.10. Let 1 be a smooth bounded domain in C
n
, n ≥ 2. For p ∈ O(1)
and any positive integer :, we have
(1) sup
z∈D
[p(.)[d(.)
s+n
≤ c
1
 p 
−s
,
(2)  p 
−s−n−1
≤ c
2
sup
z∈D
[p(.)[d(.)
s
,
6.3 The Bergman Projection 145
for some constants c
1
and c
2
independent of p.
Proof. Let χ be a smooth, nonnegative, radially symmetric function supported in
the unit ball 1
n
in C
n
with
_
Bn
χ(n)d\
w
= 1. For . ∈ 1, let χ
z
(n) = c
−2n
χ((. −
n),c), where c = d(.). Clearly,
 χ
z

k
≤ c(/)d(.)
−(k+n)
,
for some constant c(/) 0 depending on /. Then, using polar coordinates and the
mean value property of p ∈ O(1), we obtain, for  n,
[p(.)[ =
¸
¸
¸
¸
_
D
p(n)χ
z
(n) d\
w
¸
¸
¸
¸
≤  p 
−l+n
 χ
z

l−n
≤ c
1
 p 
−l+n
d(.)
−l
.
Setting  = : +n, this proves (1).
For (2), notice that if φ ∈ C
∞
0
(1), then by Taylor’s expansion and the Sobolev
embedding theorem, we have
[φ(.)[ ≤ c  φ 
s+n+1
d(.)
s
.
Hence,
 p 
−s−n−1
= sup
φ∈C
∞
0
(D)
φs+n+1=1
¸
¸
¸
¸
_
D
pφd\
z
¸
¸
¸
¸
≤ c
2
sup
z∈D
[p(.)[d(.)
s
.
This completes the proof of Lemma 6.3.10.
Proof of Theorem 6.3.8. If 1 satisﬁes condition 1, from the topology on C
∞
(1),
for each positive integer : there is a nonnegative integer : = :
s
such that 1 maps
\
s+m
(1) boundedly into H
s
(1). In particular, (1) implies (2).
Te see that (2) implies (1), by assumption, for each : ∈ N, there is a nonnegative
integer : = :
s
such that 1 is a bounded operator from \
s+ms
0
(1) into H
s
(1). For
this ﬁxed :+:
s
, Lemma 6.3.9 shows that there exists a positive integer n
t
s
= n
s+ms
such that Φ
s+ms
maps \
s+ms+n
s
(1) boundedly into \
s+ms
0
(1). It follows that,
for each p ∈ \
s+ms+n
s
(1), we have
 1p 
s
=  1Φ
s+ms
p 
s
 Φ
s+ms
p 
s+ms
 p 
s+ms+n
s
.
Hence, (2) implies (1).
Next, we prove the equivalence of (1) and (3). Suppose (3) holds. Then, for each
multiindex α and each . ∈ 1, we have by Lemma 6.3.10

∂
α
∂.
α
1(., ) 
−s
≤ c
2
sup
w∈D
¸
¸
¸
¸
∂
α
∂.
α
1(., n)
¸
¸
¸
¸
d(n)
mα
≤ C,
146 Boundary Regularity for ∂ on Pseudoconvex Domains
where : = :
α
= :
α
+n+1. Hence, by using the operator Φ
s
constructed in Lemma
6.3.9, for p ∈ \
s+ns
(1) and . ∈ 1, we have
¸
¸
¸
¸
∂
α
∂.
α
1p(.)
¸
¸
¸
¸
=
¸
¸
¸
¸
∂
α
∂.
α
_
D
1(., n)Φ
s
p(n)d\
w
¸
¸
¸
¸
=
¸
¸
¸
¸
_
D
∂
α
∂.
α
1(., n)Φ
s
p(n)d\
w
¸
¸
¸
¸
≤ 
∂
α
∂.
α
1(., ) 
−s
 Φ
s
p 
s
≤ C  p 
s+ns
.
The diﬀerentiation under the integral sign is justiﬁed, since
¸
¸
¸
¸
∂
α
∂.
α
1(., n)Φ
s
p(n)
¸
¸
¸
¸
≤ C
1
¸
¸
¸
¸
∂
α
∂.
α
1(., n)
¸
¸
¸
¸
 Φ
s
p 
s
d(n)
mα
≤ C
2
 Φ
s
p 
s
,
for some constant C
2
0 independent of . and n. Thus, condition 1 holds on 1.
On the other hand, if condition 1 holds on 1, then by the Sobolev embedding
theorem, for each nonnegative integer : there is an integer /(:) such that
sup
z∈D
¸
¸
¸
¸
∂
α
∂.
α
1)(.)
¸
¸
¸
¸
≤ C  ) 
k(s)
,
for all multiindices α with [α[ ≤ :. Therefore,

∂
α
∂.
α
1(., ) 
−k(s)
= sup
φ∈C
∞
0
(D)
φ
k(s)
=1
¸
¸
¸
¸
_
D
∂
α
∂.
α
1(., n)φ(n)d\
w
¸
¸
¸
¸
= sup
φ∈C
∞
0
(D)
φ
k(s)
=1
¸
¸
¸
¸
∂
α
∂.
α
1φ(.)
¸
¸
¸
¸
≤ C,
uniformly as . ranges over 1. Hence, by (1) of Lemma 6.3.10, condition (3) holds.
This completes the proof of Theorem 6.3.8.
Here are some consequences of condition 1.
Corollary 6.3.11. Let 1 be a smooth bounded domain in C
n
, n ≥ 2. Suppose that
condition 1 holds on 1. Then 1(, n) ∈ C
∞
(1) for each n ∈ 1.
Proof. For each ﬁxed n ∈ 1, let φ
w
(.) ∈ C
∞
0
(1) be a smooth realvalued function
such that φ
w
(.) is radially symmetric with respect to the center n and
_
D
φ
w
(.) d\
z
= 1. Since φ
w
is constant on the sphere centered at n, applying polar coordinates
and using the mean value property of 1(., ), we have
1(., n) =
_
D
1(., η)φ
w
(η) d\
η
.
Hence, 1(, n) ∈ C
∞
(1) by condition 1 on 1.
6.3 The Bergman Projection 147
Corollary 6.3.12. Let 1 be a smooth bounded domain in C
n
, n ≥ 2. Suppose
that condition 1 holds on 1. Then the linear span of ¦1(, n)[ n ∈ 1¦ is dense in
H
∞
(1) in C
∞
topology.
Proof. First, for each positive integer : and p ∈ H
∞
(1), Lemma 6.3.9 shows that
p = 1p = 1Φ
s
p.
Hence, we have H
∞
(1) ⊂ 1(\
s
0
(1)) for all real : ≥ 0.
Let Λ be the set of functions φ ∈ C
∞
0
(1) which are radially symmetric about
some point in 1 with
_
D
φd\ = 1. Thus, 1Λ = ¦1(, n)[ n ∈ 1¦. We claim that
the linear span of Λ is dense in \
s
0
(1) for each : ≥ 0. Let ) ∈ C
∞
0
(1). Choose
a smooth nonnegative function χ from C
∞
0
(C
n
) which is radially symmetric about
the origin with support contained in the unit ball and satisfying
_
C
n
χ(.)d\
z
= 1.
For c 0, set χ
(.) = c
−2n
χ(.,c). Then )
= ) ∗ χ
will converge to ) in \
s
0
(1).
Since
)
(.) = c
−2n
_
C
n
)(n)χ
_
. −n
c
_
d\
w
can be approximated by ﬁnite Riemann sums, this yields the density of span¦Λ¦ in
\
s
0
(1).
Now, condition 1 implies the H
s
(1) closure of the span of ¦1(, n)[ n ∈ 1¦
contains H
∞
(1) for every : ≥ 0. Hence, by the Sobolev embedding theorem,
span¦1(, n)[ n ∈ 1¦ is dense in H
∞
(1) in the C
∞
topology. This proves the
corollary.
To end this section, we prove the following important consequence of condition
1 concerning the boundary regularity of a biholomorphic map between two smooth
bounded pseudoconvex domains in C
n
.
Theorem 6.3.13. Let 1
1
and 1
2
be two smooth bounded pseudoconvex domains
in C
n
, n ≥ 2, and let ) be a biholomorphic map from 1
1
onto 1
2
. Suppose that
condition 1 holds on both 1
1
and 1
2
, then ) extends smoothly to the boundary.
We note from Theorem 1.7.1 that an analog of the Riemann mapping theorem
in the complex plane does not hold in C
n
for n ≥ 2. Theorem 6.3.13 provides an
important approach to the classiﬁcation of domains in higher dimensional spaces.
Therefore, given a domain 1, it is fundamental to verify whether condition 1 holds
on 1 or not. When 1 is a smooth bounded pseudoconvex domain, the Bergman
projection 1 can be expressed in terms of the ∂Neumann operator · by the formula
(4.4.14), 1 = 1 −∂
∗
·∂. We know from previous discussions that condition 1 holds
on the following classes of smooth bounded domains:
(1) 1 is strongly pseudoconvex (Theorem 5.2.1 and Corollary 5.2.7).
(2) 1 admits a plurisubharmonic deﬁning function. In particular, if 1 is convex
(Theorems 6.2.3 and 6.2.5).
(3) 1 is a circular pseudoconvex domain with transverse circular symmetry
(Theorem 6.2.7).
In fact, the Bergman projection 1 is exactly regular on all of the above three classes
of pseudoconvex domains.
Theorem 6.3.13 will be proved later. We ﬁrst prove the following theorem:
148 Boundary Regularity for ∂ on Pseudoconvex Domains
Theorem 6.3.14. Let ) : 1
1
→ 1
2
be a biholomorphic map between two smooth
bounded pseudoconvex domains 1
1
and 1
2
in C
n
. Then there is a positive integer
: such that
d(., /1
1
)
m
d()(.), /1
2
) d(., /1
1
)
1
m
for all . ∈ 1
1
.
Proof. By Theorem 3.4.12, there are continuous functions ρ
j
: 1
j
→ 1, , = 1, 2,
satisfying
(1) ρ
j
is smooth and plurisubharmonic on 1
j
,
(2) ρ
j
< 0 on 1
j
and ρ
j
vanishes on /1
j
,
(3) (−ρ
j
)
m
= −:
j
c
−K]z]
2
is smooth on 1
j
for some positive integer :, where
:
j
is a smooth deﬁning function for 1
j
.
Property (3) immediately implies that
[ρ
j
(.)[ d(., /1
j
)
1
m
, for . ∈ 1
j
.
Since ) : 1
1
→1
2
is a biholomorphic map, both ρ
2
◦ ) and ρ
1
◦ )
−1
satisfy (1)
and (2). Thus, an application of the classical Hopf lemma (see [GiTr 1]) shows that
d(., /1
1
) [ρ
2
◦ )(.)[ d()(.), /1
2
)
1
m
and
d(n = )(.), /1
2
) [ρ
1
◦ )
−1
(n)[ d(. = )
−1
(n), /1
1
)
1
m
.
This proves the theorem.
Lemma 6.3.15. Let ) : 1
1
→ 1
2
be a biholomorphic map between two smooth
bounded pseudoconvex domains 1
1
and 1
2
in C
n
. Let n(.) = det()
t
(.)) be the
determinant of the complex Jacobian of ). Then, for any positive integer :, there
is an integer , = ,(:) such that the mapping φ → n (φ ◦ )) is bounded from
\
s+j(s)
0
(1
2
) to \
s
0
(1
1
).
Proof. It suﬃces to show the estimate
 n (φ ◦ )) 
W
s
0
(D1)
≤ C  φ 
W
s+j(s)
0
(D2)
for all φ ∈ C
∞
0
(1
2
). Write ) = ()
1
, , )
n
). For any multiindex α with [α[ ≤ :, we
have
1
α
(n (φ ◦ ))) =
1
β
n 1
γ
φ()) 1
δ1
)
i1
1
δp
)
ip
,
where 1 ≤ i
1
, , i
p
≤ n, and β, γ, δ
1
, , δ
p
are multiindices with [β[ ≤ [α[,
[γ[ ≤ [α[ −[β[ and
p
j=1
[δ
j
[ = [α[ −[β[.
Since ) is a map between two bounded domains 1
1
and 1
2
in C
n
, the Cauchy
estimate implies that
¸
¸
¸
¸
∂
β
n
∂.
β
(.)
¸
¸
¸
¸
≤ C
β
d
1
(.)
−(]β]+1)
,
and
¸
¸
¸
¸
∂
δj
)
ij
∂.
δj
(.)
¸
¸
¸
¸
≤ C
j
d
1
(.)
−]δj]
,
6.3 The Bergman Projection 149
where the constant C
β
(C
j
) depends on 1
2
and the multiindex β (δ
j
), and d
1
(.) =
d(., /1
1
). Hence, for . ∈ 1
1
,
[1
β
n 1
δ1
)
i1
1
δp
)
ip
(.)[ ≤ Cd
1
(.)
−(]α]+1)
.
Also, for any φ ∈ C
∞
0
(1
2
) and every / ∈ N, it follows from Taylor’s expansion and
the Sobolev embedding theorem that
[1
γ
φ(n)[ ≤ C  φ 
k+]γ]+n+1
d
2
(n)
k
.
Thus, combining the preceding inequalities with Theorem 6.3.14, we obtain
[1
α
(n (φ ◦ )))(.)[ d
1
(.)
−(]α]+1)
 φ 
k+]α]+n+1
d
2
()(.))
k
 φ 
k+s+n+1
d
1
(.)
−s−1+
k
m
.
It is now clear by taking / = :(: +1) that the mapping φ →n (φ ◦ )) is bounded
from \
s+j(s)
0
(1
2
) to \
s
0
(1
1
) with ,(:) = :(:+1) +n+1. This proves the lemma.
We now return to the proof of Theorem 6.3.13.
Proof of Theorem 6.3.13. Let ) be a biholomorphic map from 1
1
onto 1
2
. From
Corollary 6.3.6 we obtain
(6.3.10) 1
1
(n (p ◦ ))) = n (1
2
(p) ◦ )),
for all p ∈ 1
2
(1
2
), where n = det()
t
(.)) is the determinant of the complex Jacobian
of ) and 1
ν
, ν = 1, 2, is the Bergman projection on 1
ν
.
Since condition 1 holds on 1
1
, for each positive integer :, there is an integer
:(:) such that 1
1
maps \
s+m(s)
0
(1
1
) boundedly into H
s
(1
1
). On the other hand,
by condition 1 on 1
2
we may choose a p ∈ \
s+m(s)+j(s)
0
(1
2
), as in the proof of
Corollary 6.3.12, such that 1
2
p ≡ 1, where ,(:) is determined in Lemma 6.3.15 for
the integer : +:(:). Now, Lemma 6.3.15 implies n (p ◦)) ∈ \
s+m(s)
0
(1
1
). Hence,
from (6.3.10) and condition 1 on 1
1
,
n = 1
1
(n (p ◦ )))
is in H
s
(1
1
). This shows that n ∈ C
∞
(1
1
).
Similarly, the determinant l(n) of the complex Jacobian of )
−1
is also in
C
∞
(1
2
). It follows that n(.) is nonvanishing on 1
1
.
Repeating the above arguments, for each : ∈ N, choose p
k
∈ \
s+m(s)+j(s)
0
(1
2
),
for / = 1, , n, such that 1
2
p
k
≡ n
k
, the /th coordinate function on 1
2
. Hence,
n )
k
= 1
1
(n (p
k
◦ )))
is in H
s
(1
1
), where ) = ()
1
, , )
n
). Since n does not vanish on 1
1
, this implies
)
k
∈ C
∞
(1
1
) for / = 1, , n. It follows that ) ∈ C
∞
(1
1
). Similarly, we have
)
−1
∈ C
∞
(1
2
). The proof of Theorem 6.3.13 is now complete.
150 Boundary Regularity for ∂ on Pseudoconvex Domains
6.4 Worm Domains
In this section we shall construct the socalled worm domains. Such domains
possess many pathological properties in complex analysis. We shall prove that such
domains do not always have plurisubharmonic deﬁning functions on the boundaries
nor do they always have pseudoconvex neighborhood bases.
A Hartogs domain in C
2
is a domain which is invariant under rotation in one of
the coordinates. Let 1
β
be the unbounded worm domain deﬁned by
1
β
= ¦(.
1
, .
2
) ∈ C
2
[ Re(.
1
c
−ilog]z2]
2
) 0, [log[.
2
[
2
[ < β −
π
2
¦,
for β π,2. Clearly, 1
β
is a Hartogs domain. Geometrically, if we use Re.
1
, Im.
1
and [.
2
[ as axes, then 1
β
can be visualized in 1
3
as an open half space in .
1
revolving
along the [.
2
[axis when [.
2
[ ranges from exp(−β,2 +π,4) to exp(β,2 −π,4).
To see that 1
β
is pseudoconvex, we note that locally, we can substitute the
inequality Re(.
1
c
−ilog]z2]
2
) 0 by
Re(.
1
c
−ilog]z2]
2
+argz
2
2
) = Re(.
1
c
−ilogz
2
2
) 0.
Since .
1
c
−ilogz
2
2
is locally holomorphic, its real part is a pluriharmonic function, with
vanishing complex Hessian. 1
β
is the intersection of two pseudoconvex domains.
Thus, it is pseudoconvex. As log[.
2
[ changes by a length of π, we see that the half
plane Re(.
1
c
−ilog]z2]
2
) rotates by an angle of 2π.
To construct a bounded worm domain we shall rotate discs instead of half planes.
We deﬁne
Ω
t
β
= ¦(.
1
, .
2
) ∈ C
2
[ [.
1
+c
ilog]z2]
2
[
2
< 1, [log[.
2
[
2
[ < β −
π
2
¦.
Since Ω
t
β
is deﬁned by [.
1
[
2
+ 2Re(.
1
c
−ilog]z2]
2
) < 0, locally, we can view Ω
t
β
as deﬁned by [.
1
[
2
c
argz
2
2
+ 2Re(.
1
c
−ilogz
2
2
) < 0. Since the function [.
1
[
2
c
argz
2
2
=
c
log]z1]
2
+argz
2
2
is plurisubharmonic, it is easy to see that Ω
t
β
is pseudoconvex and
bounded. But it is not smooth at [log[.
2
[
2
[ = β−
π
2
. For each ﬁxed [log[.
2
[
2
[ < β−
π
2
,
Ω
t
β
is a disc of radius 1 centered at −c
ilog]z2]
2
and (0, .
2
) ∈ /Ω
t
β
.
To construct a smooth worm domain, we have to modify Ω
t
β
. Let η : 1 →1 be
a ﬁxed smooth function with the following properties:
(1) η(r) ≥ 0, η is even and convex.
(2) η
−1
(0) = 1
β−π/2
, where 1
β−π/2
= [−β +π,2, β −π,2].
(3) there exists an o 0 such that η(r) 1 if r < −o or r o.
(4) η
t
(r) ,= 0 if η(r) = 1.
We note that (4) follows from (1) and (2). The existence of such a function is
obvious. For each β π,2, deﬁne
(6.4.1) Ω
β
= ¦(.
1
, .
2
) ∈ C
2
[ [.
1
+c
ilog]z2]
2
[
2
< 1 −η(log[.
2
[
2
)¦.
Then, we have:
6.4 Worm Domains 151
Proposition 6.4.1. For each ﬁxed β π,2, Ω
β
is a smooth bounded pseudoconvex
domain in C
2
.
Proof. Clearly, by (3), Ω
β
is bounded. For the smoothness of Ω
β
, we need to show
that ∇ρ(.) ,= 0 at every boundary point ., where ρ(.) = [.
1
+c
ilog]z2]
2
[
2
− 1 +
η(log[.
2
[
2
) is the deﬁning function for Ω
β
. If (∂ρ,∂.
1
)(.) = 0 at some boundary
point ., we get
∂ρ
∂.
1
(.) = .
1
+c
−ilog]z2]
2
= 0.
Since ρ(.) = 0, it follows that η(log[.
2
[
2
) = 1 whenever (∂ρ,∂.
1
)(.) = 0 at a
boundary point .. Now it is easy to see that (∂ρ,∂.
2
)(.) ,= 0 by (4) at such points.
This proves the smoothness of Ω
β
.
To see that Ω
β
is pseudoconvex, we write
ρ(.) = [.
1
[
2
+ 2Re(.
1
c
−ilog]z2]
2
) +η(log[.
2
[
2
).
Again locally, Ω
β
can be deﬁned by
[.
1
[
2
c
argz
2
2
+ 2Re(.
1
c
−ilogz
2
2
) +η(log[.
2
[
2
)c
argz
2
2
< 0.
The ﬁrst two terms are plurisubharmonic as before. We only need to show that the
last term η(log([.
2
[
2
)c
argz
2
2
is plurisubharmonic. A direct calculation shows that
´(η(log[.
2
[
2
)c
argz
2
2
) =
_
´η(log[.
2
[
2
)
_
c
argz
2
2
+η(log[.
2
[
2
)´c
argz
2
2
≥ 0,
since η is convex and nonnegative from (1). Ω
β
is deﬁned locally by a plurisubhar
monic function. Thus it is pseudoconvex with smooth boundary.
The following result shows that for each ﬁxed β π,2 there is no C
2
global
deﬁning function which is plurisubharmonic on the boundary of Ω
β
.
Theorem 6.4.2. For any β π,2, there is no C
2
deﬁning function ˜ ρ(.) for Ω
β
such that ˜ ρ(.) is plurisubharmonic on the boundary of Ω
β
.
Proof. Let ˜ ρ(.) be such a C
2
deﬁning function for Ω
β
that is plurisubharmonic
on the boundary /Ω
β
. Then there is a C
1
positive function / deﬁned in some
neighborhood of /Ω
β
such that ˜ ρ(.) = /ρ. Let ¹ = ¦(0, .
2
) ∈ C
2
[ [log[.
2
[
2
[ <
β − π,2¦. A direct calculation shows that the complex Hessian of ˜ ρ(.) acting on
any (α, β) ∈ C
2
for any point j ∈ ¹ ⊂ /Ω
β
is given by
(6.4.2)
L
˜ ρ(z)
(j; (α, β)) = 2Re
_
αβ
_
i/
.
2
+
∂/
∂.
2
_
c
ilog]z2]
2
_
+
_
/ + 2Re
_
∂/
∂.
1
c
ilog]z2]
2
__
[α[
2
.
Since, by assumption, (6.4.2) is always nonnegative, we must have
_
i/
.
2
+
∂/
∂.
2
_
c
ilog]z2]
2
≡ 0
152 Boundary Regularity for ∂ on Pseudoconvex Domains
on ¹, or equivalently,
∂
∂.
2
(/c
−ilog]z2]
2
) ≡ 0
on ¹. Consequently,
p(.
2
) = /(0, .
2
)c
−ilog]z2]
2
is a holomorphic function on the annulus ¹. It follows that
p(.
2
)c
ilogz
2
2
= /(0, .
2
)c
−2argz2
= c,
is also locally a holomorphic function on ¹, and hence it must be a constant c, since
the right hand side is real. This implies that
/(0, .
2
) = cc
2argz2
is a well deﬁned, C
1
positive function on ¹, which is impossible. This proves
Theorem 6.4.2.
In particular, Theorem 6.2.3 cannot be applied to worm domains. In fact we
will prove in the next section that the Bergman projection is not regular on worm
domains.
Another peculiar phenomenon about worm domains is that they do not have
pseudoconvex neighborhood bases if β is suﬃciently large. To illustrate this, we
ﬁrst examine the Hartogs triangle
G = ¦(.
1
, .
2
) [ [.
1
[ < [.
2
[ < 1¦.
By Cauchy’s integral formula, any function holomorphic in a neighborhood of G
extends holomorphically to the bidisc 1
2
= ¦(.
1
, .
2
) [ [.
1
[ < 1, [.
2
[ < 1¦. Thus if Ω
is any pseudoconvex domain containing G, then Ω contains the larger set 1
2
since
pseudoconvex domains are domains of holomorphy. This implies that we cannot
approximate G by a sequence of pseudoconvex domains ¦Ω
k
¦ such that G ⊂ Ω
k
and G = ∩
k
Ω
k
. However, the Hartogs triangle is not smooth.
We next show that Ω
β
does not have a pseudoconvex neighborhood base if β ≥
3π,2. When β ≥ 3π,2, Ω
β
contains the set
1 =¦(0, .
2
) [ −π ≤ log[.
2
[
2
≤ π¦
∪ ¦(.
1
, .
2
) [ log[.
2
[
2
= π or −π and [.
1
−1[ < 1¦.
Any holomorphic function in a neighborhood of 1 extends holomorphically to the
set
ˆ
1 = ¦(.
1
, .
2
) [ −π ≤ log[.
2
[
2
≤ π and [.
1
−1[ < 1¦.
Thus any holomorphic function in a neighborhood of Ω
β
extends holomorphically
to Ω
β
∪
ˆ
1. This implies that any pseudoconvex domain containing Ω
β
contains
ˆ
1.
6.5 Irregularity of the Bergman Projection on Worm Domains 153
Theorem 6.4.3. For β ≥ 3π,2, there does not exist a sequence ¦Ω
k
¦ of pseudo
convex domains in C
2
with Ω
β
⊂ Ω
k
and Ω
β
= ∩
k
Ω
k
.
Thus pseudoconvex domains do not always have a pseudoconvex neighborhood
base. We note that any pseudoconvex domain can always be exhausted by pseudo
convex domains from inside.
6.5 Irregularity of the Bergman Projection on Worm Domains
The purpose of this section is to prove that the Bergman projection 1 is irregular
on the worm domain Ω
β
in the Sobolev spaces. We ﬁrst study the Bergman kernel
function 1(., n) on the unbounded worm domain 1
β
where 1
β
is deﬁned in 6.4.
For each ﬁxed .
1
variable, the domain is a union of annuli in .
2
. Any holomorphic
function in 1
β
admits a Laurent expansion in .
2
. Using Fourier expansion, for any
) ∈ H(1
β
), we write
)(.) =
1
2π
j∈Z
_
2π
0
)(.
1
, c
iθ
.
2
)c
−ijθ
dθ.
Let )
j
(.
1
, .
2
) =
1
2π
_
2π
0
)(.
1
, c
iθ
.
2
)c
−ijθ
dθ. Then )
j
is holomorphic and )
j
(.
1
, c
iθ
.
2
)
= c
ijθ
)
j
(.
1
, .
2
). Such )
j
are necessarily of the form )
j
(.
1
, .
2
) = p(.
1
, [.
2
[).
j
2
, where
p(.
1
, [.
2
[) is holomorphic in 1
β
and locally constant in [.
2
[. The Bergman space
H(1
β
) admits an orthogonal decomposition
H(1
β
) = ⊕
j∈Z
H
j
(1
β
).
Any ) in H
j
(1
β
) satisﬁes )(.
1
, c
iθ
.
2
) = c
ijθ
)(.
1
, .
2
). Denote by 1
j
the orthogonal
projection from 1
2
(1
β
) onto H
j
(1
β
). It follows that if ) ∈ H(1
β
), we have
1
j
)(.) = )
j
(.) =
1
2π
_
2π
0
)(.
1
, c
iθ
.
2
)c
−ijθ
dθ,
and the Bergman kernel function 1
D
β
(., n) associated with 1
β
satisﬁes
1
D
β
(., n) =
j∈Z
1
j
(., n),
where 1
j
(., n) is the reproducing kernel for H
j
(1
β
). Each 1
j
(., n) is locally of
the form 1
j
(., n) = /
j
(.
1
, n
1
).
j
2
¯ n
j
2
. It turns out that for the unbounded worm
domain, the kernel 1
−1
can be computed explicitly.
To facilitate the calculation, we introduce the following domain:
1
t
β
= ¦(.
1
, .
2
) ∈ C
2
[ [Im.
1
−log[.
2
[
2
[ < π,2, [log[.
2
[
2
[ < β −π,2¦.
For each ﬁxed .
2
, 1
t
β
is an inﬁnite strip in .
1
. Thus, 1
t
β
is biholomorphically
equivalent to 1
β
via the mapping
ϕ : 1
t
β
→ 1
β
(.
1
, .
2
) →(c
z1
, .
2
).
154 Boundary Regularity for ∂ on Pseudoconvex Domains
Also from the transformation formula (6.3.3) for the Bergman kernel functions, we
have
1
D
β
(., n) =
1
.
1
n
1
1
D
β
(ϕ
−1
(.), ϕ
−1
(n)).
Since ϕ commutes with rotation in the .
2
variable, we have an analogous transfor
mation law on each component
(6.5.1) 1
j
(., n) =
1
.
1
n
1
1
t
j
(ϕ
−1
(.), ϕ
−1
(n)),
where 1
t
j
is the reproducing kernel for the square integrable holomorphic functions
H on 1
t
β
satisfying H(.
1
, c
iθ
.
2
) = c
ijθ
H(.
1
, .
2
). The kernel 1
t
−1
can be calculated
explicitly as follows:
For any H ∈ H
j
(1
t
β
), we may write H(.
1
, .
2
) = /(.
1
).
j
2
, where /(.
1
) is holo
morphic in .
1
. For each β 0, let o
β
be the strip on the complex plane deﬁned
by
o
β
= ¦. = r +ij ∈ C[ [j[ < β¦.
It follows that
(6.5.2)
 H 
2
L
2
(D
β
)
=
_
D
β
[/(.
1
)[
2
[.
2
[
2j
dr
1
dj
1
dr
2
dj
2
= 2π
_
]2logr]<β−
π
2
_
]y1−2logr]<
π
2
[/(.
1
)[
2
:
2j+1
dr
1
dj
1
d:
= π
_
]s]<β−
π
2
_
]y1−s]<
π
2
[/(.
1
)[
2
c
(j+1)s
dr
1
dj
1
d:
= π
_
∞
−∞
_
S
β
[/(.
1
)[
2
c
(j+1)s
χ
π
2
(j
1
−:)χ
β−
π
2
(:) dr
1
dj
1
d:
=
_
S
β
[/(.)[
2
λ
j
(j) drdj,
where λ
j
(j) = π(c
(j+1)s
χ
β−
π
2
)∗χ
π
2
(j), β π,2 and χ
α
is the characteristic function
on 1
α
= (−α, α). Let λ(j) be a continuous positive bounded function on the interval
1
β
= ¦j ∈ 1 [ [j[ < β¦. Denote by H(o
β
, λ) the weighted Bergman space on o
β
deﬁned by
H(o
β
, λ) = ¦) ∈ O(o
β
)[  ) 
2
λ
=
_
S
β
[)(.)[
2
λ(j) drdj < ∞¦.
To compute the kernel 1
t
−1
, it suﬃces to compute the Bergman kernel in one
variable on a strip o
β
with weight λ = πχ
β−
π
2
∗ χ
π
2
(j) and the kernel 1
t
−1
(., n)
is given by 1
t
−1
= 1
λ
(.
1
, n
1
),.
2
¯ n
2
. The next lemma allows us to compute the
weighted Bergman kernel function 1
λ
(., n) on o
β
.
6.5 Irregularity of the Bergman Projection on Worm Domains 155
Lemma 6.5.1. For each β 0, let λ(j) be a continuous positive bounded function
on the interval 1
β
= ¦j ∈ 1 [ [j[ < β¦. Then the weighted Bergman kernel function
1
λ
(., n) on o
β
is given by
(6.5.3) 1
λ
(., n) =
1
2π
_
R
c
i(z−w)ξ
ˆ
λ(−2ξi)
dξ,
where
ˆ
λ is the Fourier transform of λ if λ(j) is viewed as a function on 1 that
vanishes outside 1
β
.
Proof. For ) ∈ H(o
β
, λ) we deﬁne the partial Fourier transform
˜
) of ) with respect
to r by
˜
)(ξ, j) =
_
R
)(r +ij)c
−ixξ
dr.
It is easily veriﬁed by Cauchy’s theorem that
˜
)(ξ, j) = c
−yξ ˜
)
0
(ξ), where
˜
)
0
(ξ) =
˜
)(ξ, 0). Thus from Plancherel’s theorem,
(6.5.4)
 ) 
2
λ
= (2π)
−1
_
R·I
β
c
−2yξ
[
˜
)
0
(ξ) [
2
λ(j)dξdj
= (2π)
−1
_
R
[
˜
)
0
(ξ) [
2
ˆ
λ(−2iξ)dξ.
For ) ∈ H(o
β
, λ) and . ∈ o
β
, we have
_
R
˜
)
0
(ξ)c
izξ
dξ = 2π)(.)
= 2π
_
S
β
)(n)1
λ
(n, .)λ(j) drdj
=
_
β
−β
_
R
c
−2yξ
˜
)
0
(ξ)
¯
1
λ
((ξ, 0), .)λ(j) dξdj,
where n = r +ij. It follows that
c
−izξ
=
¯
1
λ
((ξ, 0), .)
_
β
−β
c
−2yξ
λ(j) dj,
and
¯
1
λ
((ξ, 0), .) =
c
−izξ
ˆ
λ(−2ξi)
.
Finally, by the Fourier inversion formula, we obtain
1
λ
(., n) =
1
2π
_
R
c
i(z−w)ξ
ˆ
λ(−2ξi)
dξ.
Here we note that
ˆ
λ(−2ξi) is real. This proves the lemma.
We next apply Lemma 6.5.1 to the piecewise linear weight
λ(j) = πχ
β−
π
2
∗ χ
π
2
(j).
156 Boundary Regularity for ∂ on Pseudoconvex Domains
Lemma 6.5.2. For β
π
2
, if λ(j) = πχ
β−
π
2
∗ χ
π
2
(j), then
ˆ
λ(−2ξi) =
πsinh
_
(2β −π)ξ
_
sinh(πξ)
ξ
2
,
and
(6.5.5) 1
λ
(., n) =
1
2π
2
_
R
ξ
2
c
i(z−w)ξ
sinh
_
(2β −π)ξ
_
sinh(πξ)
dξ.
Proof. If λ(j) = χ
α
(j) for some α 0, then
ˆ
λ(−2ξi) =
_
α
−α
c
−ix(−2ξi)
dr =
_
α
−α
c
−2xξ
dr =
sinh(2αξ)
ξ
.
Hence, for the piecewise linear weight λ(j) = πχ
β−
π
2
∗ χ
π
2
(j), we have
ˆ
λ(−2ξi) =
πsinh
_
(2β −π)ξ
_
sinh(πξ)
ξ
2
,
and (6.5.5) now follows from Lemma 6.5.1.
We observe that ξ
2
,sinh
_
(2β −π)ξ
_
sinhπξ has poles at nonzero integer multiples
of πi,(2β−π) and i. Let us ﬁrst assume that β π, and set ν
β
= π,(2β−π) so that
ν
β
< 1. Then, via a standard contour integration, one can obtain the asymptotic
expansion of the weighted Bergman kernel function 1
λ
(., n) and see that it is in
fact dominated by the residue of p(ξ) = ξ
2
c
i(z−w)ξ
,sinh
_
(2β −π)ξ
_
sinh(πξ) at the
ﬁrst pole ν
β
i.
Lemma 6.5.3. Let β π and λ(j) = πχ
β−
π
2
∗ χ
π
2
(j). Then
(6.5.6) 1
λ
(., n) = c
β
c
−ν
β
(z−w)
+O(c
−µ
β
(z−w)
)
for Re(. −n) 0 and
(6.5.7) 1
λ
(., n) = −c
β
c
ν
β
(z−w)
+O(c
µ
β
(z−w)
)
for Re(. −n) < 0, where c
β
= ν
β
3
,(π
2
sinν
β
π) and j
β
= min(2ν
β
, 1). Furthermore,
given any small positive c, the expansion in (6.5.6) or (6.5.7) is uniform for any
., n ∈ o
β−
.
Proof. Fix / 0 so that /i is the midpoint between the second and third poles of
p(ξ). Denote by Γ
N
the rectangular contour with vertices ±· and ±· + i/. Let
us ﬁrst assume that 2ν
β
< 1. Then, we have 2ν
β
< / < 1 and
_
N
−N
ξ
2
c
i(z−w)ξ
sinh
_
(2β −π)ξ
_
sinh(πξ)
dξ + 1
N
+ 1
−N
+ J
N
= 2πi(Res p(ξ) at ν
β
i and 2ν
β
i),
6.5 Irregularity of the Bergman Projection on Worm Domains 157
where Res p(ξ) denotes the residues of p(ξ) and
1
N
= i
_
h
0
(· +ij)
2
c
i(z−w)(N+iy)
sinh
_
(2β −π)(· +ij)
_
sinhπ(· +ij)
dj,
1
−N
= −i
_
h
0
(−· +ij)
2
c
i(z−w)(−N+iy)
sinh
_
(2β −π)(−· +ij)
_
sinhπ(−· +ij)
dj,
and
J
N
= −
_
N
−N
(r +i/)
2
c
i(z−w)(x+ih)
sinh
_
(2β −π)(r +i/)
_
sinhπ(r +i/)
dr.
A direct calculation shows that
Res
ξ=ν
β
i
p(ξ) =
ν
β
3
c
−ν
β
(z−w)
iπsin(ν
β
π)
,
and
Res
ξ=2ν
β
i
p(ξ) =
−4ν
β
3
c
−2ν
β
(z−w)
iπsin(2ν
β
π)
.
For any ., n ∈ o
β−
, we write . − n = n + i· with n 0. Hence, we have
[·[ ≤ 2(β −c) and
[1
N
[
_
h
0
(·
2
+j
2
)c
−vN−uy
c
2βN
dj
·
2
+ 1
c
2N
.
It follows that 1
N
converges to zero uniformly for any ., n ∈ o
β−
such that Re(. −
n) 0. Similarly, we get the uniform convergence to zero for 1
−N
. For J
N
, we
have
[J
N
[
_
1
−1
(r
2
+/
2
)c
−vx−uh
dr +
_
N
1
(r
2
+/
2
)c
−vx−uh
c
2βx
dr
+
_
−1
−N
(r
2
+/
2
)c
−vx−uh
c
−2βx
dr
c
−2ν
β
u
_
1 +
_
N
1
r
2
+/
2
c
2x
dr
_
.
It follows by letting · tend to inﬁnity that
_
∞
−∞
ξ
2
c
i(z−w)ξ
sinh
_
(2β −π)ξ
_
sinh(πξ)
dξ
=
2ν
β
3
c
−ν
β
(z−w)
sinν
β
π
−
8ν
β
3
c
−2ν
β
(z−w)
sin(2ν
β
π)
+
_
∞
−∞
(r +i/)
2
c
i(z−w)(x+ih)
sinh
_
(2β −π)(r +i/)
_
sinhπ(r +i/)
dr
=
2ν
β
3
sinν
β
π
c
−ν
β
(z−w)
+O(c
−2ν
β
(z−w)
).
158 Boundary Regularity for ∂ on Pseudoconvex Domains
Clearly, the estimate is uniform for all ., n ∈ o
β−
with Re(. −n) 0. This proves
the case for 2ν
β
< 1.
For cases 2ν
β
= 1 and 2ν
β
1, a similar argument applies.
If Re(. − n) < 0, we take the rectangular contour Γ
N
on the lower half space
with vertices ±·, and ±·−i/, and (6.5.7) can be proved similarly. This completes
the proof of the lemma.
From (6.5.2), the kernel 1
t
−1
(., n) is given by
1
t
−1
(., n) = 1
λ
(.
1
, n
1
),.
2
n
2
,
where 1
λ
(.
1
, n
1
) is calculated in Lemma 6.5.3. If β π, then (6.5.7) shows that
1
t
−1
(., n) = −c
β
c
ν
β
(z1−w1)
.
2
n
2
+O(c
µ
β
(z1−w1)
)
for Re(.
1
−n
1
) < 0. Hence
(6.5.8) 1
−1
(., n) = −c
β
.
ν
β
−1
1
n
−ν
β
−1
1
.
−1
2
n
−1
2
+
1
.
1
n
1
O
_
_
.
1
n
1
_
µ
β
_
for [.
1
[ < [n
1
[. The expansion in (6.5.7) is uniform on o
β−
for any small positive
c. Thus, for ﬁxed n, we have for any : ∈ N,
[Re(.
1
c
−ilog]z2]
2
)[
s
_
∂
∂.
1
_
m
1
−1
(., n) , ∈ 1
2
(1
β
), for : ≤ :−ν
β
.
It follows that
(6.5.9) [Re(.
1
c
−ilog]z2]
2
)[
s
_
∂
∂.
1
_
m
1
D
β
(., n) , ∈ 1
2
(1
β
),
for : ≤ :−ν
β
. Estimate (6.5.9) also holds for π,2 < β ≤ π. When π,2 <
β < π, (6.5.9) can be obtained by examining higher order terms in the asymptotic
expansion of 1
λ
.
When β = π, we compute the residue at the double pole −i of (6.5.5) to obtain
1
λ
(.
1
, n
1
) = π
−2
(−.
1
+ ¯ n
1
−2)c
(z1−¯ w1)
+O(c
2(z1−¯ w1)
)
for Re(.
1
− ¯ n
1
) < 0 and
1
−1
(., n) = π
−2
(−log(.
1
, ¯ n
1
) −2) ¯ n
−2
1
.
−1
2
¯ n
−1
2
+O((.
1
, ¯ n
1
)
2
)
for [.
1
[ < [n
1
[. Thus (6.5.9) holds for β = π also. Estimate (6.5.9) is crucial in
proving the irregularity of the Bergman projection 1 measured in the Sobolev norm
on the worm domain. For our purpose we need the following fact (See Lemma C.4
in the Appendix).
6.5 Irregularity of the Bergman Projection on Worm Domains 159
Lemma 6.5.4. Let 1 be a smooth bounded domain in 1
N
with a smooth deﬁning
function ρ(r). Then, for each : ≥ 0, the \
−s
norm of a harmonic function ) is
equivalent to the 1
2
norm of [ρ[
s
) on 1.
We ﬁrst observe the following result for the Bergman projection on the un
bounded worm domain 1
β
.
Proposition 6.5.5. For each β π,2, condition 1 does not hold on 1
β
. Fur
thermore, the Bergman projection 1
∞
on 1
β
does not map C
∞
0
(1
β
) into \
k
(1
β
)
when / ≥ π,(2β −π).
Proof. Let n ∈ 1
β
. We choose a realvalued function ) ∈ C
∞
0
(1
β
) such that )
depends on [. −n[ and
_
D
β
) = 1. Using the same argument as in Corollary 6.3.11,
we obtain
1
∞
) = 1
D
β
(, n).
Since 1
D
β
(, n) , ∈ C
∞
(1
β
), condition 1 fails on 1
β
.
Let Γ
A
= ¦. ∈ C
n
[ [.[ < ¹¦ be a large ball for ¹ 0. From (6.5.9), we have
(6.5.10) [ρ
∞
[
s
_
∂
∂.
1
_
m
1
∞
) , ∈ 1
2
(1
β
∩ Γ
A
)
for : ≤ :−ν
β
, where ρ
∞
= Re(.
1
c
−ilog]z2]
2
). If 1
∞
) ∈ \
k
(1
β
∩ Γ
A
), choose a
positive integer : / and let : = :− / ≤ :−ν
β
. Using Lemma 6.5.4, we have
[ρ
∞
[
s
∇
m
1
∞
) ∈ 1
2
(1
β
∩ Γ
A
), a contradiction. Thus 1
∞
) , ∈ \
k
(1
β
∩Γ
A
) and the
proposition is proved.
We prove the main result of this section on the irregularity of the Bergman
projection for the smooth worm domain Ω
β
.
Theorem 6.5.6. For each β π,2, the Bergman projection 1 on Ω
β
does not
map \
k
(Ω
β
) into \
k
(Ω
β
) when / ≥ π,(2β −π).
Proof. Assume on the contrary that the Bergman projection 1 maps \
k
(Ω
β
) into
\
k
(Ω
β
) with the estimate
(6.5.11)  1) 
W
k
(Ω
β
)
≤ C
k
 ) 
W
k
(Ω
β
)
for ) ∈ \
k
(Ω
β
) and / ≥ π,(2β −π) = ν
β
.
For any j ≥ 1, let τ
µ
be the dilation deﬁned by
τ
µ
: C
2
→ C
2
(.
1
, .
2
) →(j.
1
, .
2
).
Denote by Ω
β,µ
= τ
µ
(Ω
β
), Ω
t
β,µ
= τ
µ
(Ω
t
β
) where Ω
t
β
is deﬁned in Section 6.4. Then
Ω
t
β,µ
⊂ Ω
β,µ
and Ω
t
β,µ
¸ 1
β
. Let T
µ
be the pullback of the 1
2
functions on Ω
β,µ
,
i.e.,
T
µ
: 1
2
(Ω
β,µ
) →1
2
(Ω
β
)
) → ) ◦ τ
µ
.
160 Boundary Regularity for ∂ on Pseudoconvex Domains
A direct calculation shows that

_
∂
∂.
_
α
_
∂
∂.
_
γ
T
µ
) 
L
2
(Ω
β
)
= j
α1+γ1−1

_
∂
∂.
_
α
_
∂
∂.
_
γ
) 
L
2
(Ω
β,µ
)
,
where α = (α
1
, α
2
) and γ = (γ
1
, γ
2
) are multindices. Thus we have
(6.5.12)  T
µ
) 
W
l
(Ω
β
)
≤ j
l−1
 ) 
W
l
(Ω
β,µ
)
,
when  is a nonnegative integer. Then, by interpolation, it holds for all real  ≥ 0.
Let 1
µ
be the Bergman projection associated with Ω
β,µ
. Then
(6.5.13) 1
µ
= T
−1
µ
1T
µ
.
From the deﬁnition (6.4.1) of Ω
β
we see that the deﬁning function ρ(.) coincides
with [.
1
[
2
+ 2Re(.
1
c
−ilog]z2]
2
) when log[.
2
[
2
∈ 1
β−
π
2
. Let ρ
µ
(.) = jρ ◦ (τ
µ
)
−1
so
that ρ
µ
→ ρ
∞
= 2Re(.
1
c
−ilog]z2]
2
) as j → ∞, where ρ
∞
is a deﬁning function for
1
β
. Write / = :− :, where : is an integer and : ≥ 0. For any ) ∈ C
∞
0
(Ω
t
β,µ
) ⊂
C
∞
0
(Ω
β,µ
), we have using (6.5.11)(6.5.13) and Lemma 6.5.4,
(6.5.14)
 [ρ
µ
[
s
_
∂
∂.
1
_
m
1
µ
) 
L
2
(Ω
β,µ
)
=  [ρ
µ
[
s
_
∂
∂.
1
_
m
(T
µ
)
−1
1T
µ
) 
L
2
(Ω
β,µ
)
= j
s−m+1
 [ρ[
s
_
∂
∂.
1
_
m
1T
µ
) 
L
2
(Ω
β
)
≤ Cj
1−k

_
∂
∂.
1
_
m
1T
µ
) 
W
−s
(Ω
β
)
≤ Cj
1−k
 1T
µ
) 
W
k
(Ω
β
)
≤ Cj
1−k
 T
µ
) 
W
k
(Ω
β
)
≤ C  ) 
W
k
(Ω
β,µ
)
,
where the constant C is independent of j. We claim that
1
µ
) ÷ 1
∞
) weakly in 1
2
(C
2
),
where 1
∞
) is the Bergman projection of ) on 1
β
, 1
∞
) = 0 outside 1
β
and we
have set 1
µ
) = 0 outside Ω
β,µ
. Assuming the claim, It follows from (6.5.14) that
(6.5.15)  [ρ
∞
[
s
_
∂
∂.
1
_
m
1
∞
) 
L
2
(D
β
)
≤ C  ) 
W
k
(D
β
)
for any ) ∈ C
∞
0
(1
β
). This contradicts (6.5.10) and the theorem is proved.
Thus, it remains to prove the claim. Since
 1
µ
) 
L
2
(C
2
)
≤  ) 
L
2
(C
2
)
,
there exists a subsequence of 1
µ
) that converges weakly to / ∈ 1
2
(C
2
). Since
Ω
t
β,µ
¸ 1
β
, / is holomorphic in 1
β
. Also / vanishes outside 1
β
since every
compact subset outside 1
β
is outside Ω
β,µ
for suﬃciently large j. To prove that
/ = 1
∞
), we need to show that ) −/ ⊥ H(1
β
). Choose ` 1 so that Ω
β,µ
⊂ 1
Mβ
for all j. Obviously ) − 1
µ
) ⊥ H(1
Mβ
). Therefore, by passing to the limit, we
obtain that ) − / ⊥ H(1
Mβ
). The claim will be proved by the following density
result:
Notes 161
Lemma 6.5.7. For each ` 1, the space H(1
Mβ
) is dense in H(1
β
).
Proof. It suﬃces to show that each H
j
(1
Mβ
) is dense in H
j
(1
β
), or equivalently,
H
j
(1
t
Mβ
) is dense in H
j
(1
t
β
). From (6.5.2) and (6.5.4) we have for any ) ∈ H
j
(1
t
β
),
 ) 
2
L
2
(D
β
)
= (2π)
−1
_
R
[
˜
)
0
(ξ) [
2
ˆ
λ
j
(−2iξ)dξ,
where
˜
)
0
(ξ) is the partial Fourier transform of ) evaluated at j = 0 and λ
j
(j) =
π(c
(j+1)()
χ
β−
π
2
) ∗ χ
π
2
(j). Thus, the space H
j
(1
t
β
) is isometric via the Fourier
transform to the space of functions on 1 which are square integrable with respect
to the weight
ˆ
λ
j
(−2ξi) =
πsinh
_
(2β −π)
_
ξ −(
j+1
2
)
_¸
sinh(πξ)
ξ
_
ξ −(
j+1
2
)
_ .
Since C
∞
0
(1) is dense in the latter space for any value of β, the lemma follows.
Using Theorem 6.2.2 and Theorem 6.5.6, we also obtain that the
¯
∂Neumann
operator is irregular on the worm domain.
Corollary 6.5.8. For each β π,2, the ∂Neumann operator on Ω
β
does not map
\
k
(0,1)
(Ω
β
) into \
k
(0,1)
(Ω
β
) when / ≥ π,(2β −π).
NOTES
The existence of a smooth solution up to the boundary, using the weighted ∂
Neumann problem, for the ∂ equation was proved by J. J. Kohn in [Koh 6]. The
equivalence between the Bergman projections and the ∂Neumann operators was
proved by H. P. Boas and E. J. Straube in [BoSt 2]. Theorem 6.2.1 provides a suﬃ
cient condition for verifying the exact regularity of the ∂Neumann operators, and
the idea has been used in [Che 4] and [BoSt 3,4,5]. The use of a smooth plurisubhar
monic deﬁning function (Theorem 6.2.3), based on an observation by A. Noell [Noe
1], was originated in [BoSt 3] where they treated directly the exact regularity of the
Bergman projections under the existence of such a deﬁning function. For a convex
domain in dimension two, a diﬀerent proof, using related ideas, was obtained inde
pendently in [Che 5]. The use of transverse symmetries for verifying the regularity
of the Bergman projection was ﬁrst initiated by D. Barrett [Bar 1]. The regularity
of the ∂Neumann problem on circular domains with symmetry (Theorems 6.2.7
and 6.2.8) was proved by S.C. Chen [Che 3]. See also [BCS 1]. Another suﬃcient
condition related to the De Rham cohomology on the set of inﬁnite type points for
the regularity of the ∂Neumann operators was also introduced by H. P. Boas and
E. J. Straube in [BoSt 5].
Another important class of smooth bounded pseudoconvex domains which is
beyond the scope of this book is the class of domains of ﬁnite type. The concept
of ﬁnite type on a pseudoconvex domain in C
2
, using the Lie brackets of complex
tangential vector ﬁelds, was ﬁrst introduced by J. J. Kohn [Koh 4]. Subsequently,
162 Boundary Regularity for ∂ on Pseudoconvex Domains
J. J. Kohn introduced subelliptic multipliers and ﬁnite ideal type in [Koh 8] and he
proved that ﬁnite ideal type condition is suﬃcient for the subelliptic estimates of the
∂Neumann operators. By measuring the order of contact of complex varieties with
a hypersurface at the reference point, J. D’Angelo [DAn 1,2] proposed a deﬁnition
of ﬁnite type in C
n
. The necessity of ﬁnite order of contact of complex varieties for
the subelliptic estimates was proved by D. Catlin in [Cat 1].
When the boundary is real analytic near a boundary point, Kohn’s theory of
ideals of subelliptic multipliers [Koh 8], together with a theorem of Diederich and
Fornaess [DiFo 3], showed that a subelliptic estimate on (j, ¡)forms for the ∂
Neumann problem is equivalent to the absence of germs of ¡ dimensional complex
varieties in the boundary near the point. In particular, subelliptic estimates always
hold on any bounded pseudoconvex domain with real analytic boundary.
D. Catlin also deﬁned in [Cat 4] his own notion of ﬁnite type. His theory of mul
titypes developed in [Cat 2] leads to the construction of a family of smooth bounded
plurisubharmonic functions with large Hessian on the boundary. This property is
now known as property (P) (see [Cat 3]). Property (P) implies the existence of a
compactness estimate for the ∂Neumann problem. Therefore, together with a the
orem of Kohn and Nirenberg [KoNi 1], global regularity of the ∂Neumann problem
will follow from property (P). See also the papers by N. Sibony [Sib 2,3] and S. Fu
and E. J. Straube [FuSt 1] for related results.
When a smooth bounded pseudoconvex domain has real analytic boundary, it is
also important to know the real analytic regularity of the ∂Neumann operator near
the boundary. Real analytic regularity of a holomorphic function near the boundary
is equivalent to holomorphic extension of the function across the boundary. For
strongly pseudoconvex domains, an aﬃrmative result of global analytic regularity
of the ∂Neumann problem had been obtained by M. Derridj and D. S. Tartakoﬀ
[DeTa 1] and G. Komatsu [Kom 1]. Local analytic hypoellipticity of the ∂Neumann
problem on strongly pseudoconvex domains was proved by D. S. Tartakoﬀ [Tar 1,2]
and F. Treves [Tre 2]. When the domains are weakly pseudoconvex of special type,
some positive results concerning global analytic hypoellipticity of the ∂Neumann
problem are also available by S.C. Chen [Che 1,2,6] and M. Derridj [Der 1] using
the vector ﬁeld technique. For local analytic regularity of the ∂Neumann problem
on certain weakly pseudoconvex domains, see [DeTa 2,3].
For introductory materials on the Bergman kernel function, the reader may con
sult the survey paper by S. Bell [Bel 4] or the texts by S. G. Krantz [Kra 2] and R.
M. Range [Ran 6]. See also the papers by S. Bell [Bel 3], H. P. Boas [Boa 3], S.C.
Chen [Che 7] and N. Kerzman [Ker 2] for the diﬀerentiability of the Bergman kernel
function near the boundaries of the domains. Theorem 6.3.7 on various equivalent
statements of condition 1 can be found in [BeBo 1]. The operator Φ
s
in Lemma
6.3.8 was ﬁrst constructed by S. Bell in [Bel 2]. Corollary 6.3.12 is the density
lemma due to S. Bell [Bel 1]. The smooth extension of a biholomorphic mapping
between two smooth bounded domains in C
n
, n ≥ 2, was ﬁrst achieved by C. Fef
ferman in his paper [Fef 1] when the domains are strongly pseudoconvex. Later,
condition 1 was proposed by S. Bell and E. Ligocka in [BeLi 1]. They showed using
condition 1 that, near a boundary point, one may choose special holomorphic local
coordinates resulting from the Bergman kernel functions so that any biholomorphic
map between these two smooth bounded domains becomes linear (Theorem 6.3.13).
163
Hence, the biholomorphism extends smoothly up to the boundaries. The present
proof of Theorem 6.3.13 was adopted from [Bel 2]. A smooth bounded nonpseudo
convex domain in C
2
which does not satisfy condition 1 was discovered in [Bar 2].
In contrast to Barrett’s counterexample, H. P. Boas and E. J. Straube showed in
[BoSt 1] that condition 1 always holds on any smooth bounded complete Hartogs
domain in C
2
regardless of whether it is pseudoconvex or not. Theorem 6.3.14 was
proved by R. M. Range [Ran 2].
The construction of worm domains is due to K. Diederich and J. E. Fornaess
in [DiFo 1] where Theorem 6.4.3 is proved (see also [FoSte 1]). Our exposition
follows that of C. O. Kiselman [Kis 1]. Most of the Section 6.5 is based on [Bar 3].
Recently, based on D. Barrett’s result, it was proved by M. Christ in [Chr 2] that
condition 1 does not hold for the Bergman projection on the worm domain. For
more about the regularity of the ∂Neumann problem and its related questions, the
reader may consult the survey paper by H. P. Boas and E. J. Straube [BoSt 6]. We
also refer the reader to the book by J. E. Fornaess and B. Stensønes [FoSte 1] for
counterexamples on pseudoconvex domains. For recent results on the ∂Neumann
problem on Lipschitz pseudoconvex domains, see the papers by BonamiCharpentier
[BoCh 1], HenkinIordan [HeIo 1], HenkinIordanKohn [HIK 1], MichelShaw [MiSh
1] and Straube [Str 2]. H¨older and 1
p
estimates of the ∂Neumann problem on
pseudoconvex domains of ﬁnite type in C
2
have been discussed in ChangNagel
Stein [CNS 1], FeﬀermanKohn [FeKo 1]. H¨older or 1
p
estimates for the ∂Neumann
problem on ﬁnite type pseudoconvex domains in C
n
for n ≥ 3 are still unknown.
164 Boundary Regularity for ∂ on Pseudoconvex Domains
CHAPTER 7
CAUCHYRIEMANN MANIFOLDS
AND THE TANGENTIAL CAUCHYRIEMANN COMPLEX
Let ` be a smooth hypersurface in a complex manifold. The restriction of the
∂ complex to ` naturally induces a new diﬀerential complex. This complex is
called the tangential CauchyRiemann complex or the ∂
b
complex. The tangential
CauchyRiemann complex, unlike the de Rham or the ∂ complex, is not elliptic. In
general, it is an overdetermined system with variable coeﬃcients.
We have seen in Chapter 3 that the tangential CauchyRiemann equations are
closely related to the holomorphic extension of a C1 function on the hypersurface.
The ∂
b
complex is also important in its own right in the theory of partial diﬀeren
tial equations. The tangential CauchyRiemann equation associated with a strongly
pseudoconvex hypersurface in C
2
provides a nonsolvable ﬁrst order partial diﬀeren
tial equation with variable coeﬃcients. It also serves as a prototype of subelliptic
operators.
In the next few chapters, we shall study the solvability and regularity of the ∂
b
complex. First, in Chapters 8 and 9 the subellipticity and the closed range property
of ∂
b
will be investigated using 1
2
method. Then, in Chapter 10 we construct an
explicit fundamental solution for
b
on the Heisenberg group. Next, the integral
representation is used to construct a solution operator for the ∂
b
operator on a
strictly convex hypersurface in Chapter 11. The C1 embedding problem will be
discussed in Chapter 12.
In this chapter, we shall ﬁrst deﬁne CauchyRiemann manifolds and the tangen
tial CauchyRiemann complex both extrinsically and intrinsically. The Levi form of
a CauchyRiemann manifold is introduced. In Section 7.3, we present the famous
nonsolvable Lewy operator. In contrast with the Lewy operator, we prove that
any linear partial diﬀerential operator with constant coeﬃcients is always locally
solvable.
7.1 C1 Manifolds
Let ` be a real smooth manifold of dimension 2n − 1 for n ≥ 2, and let T(`)
be the tangent bundle associated with `. Let CT(`) = T(`) ⊗
R
C be the
complexiﬁed tangent bundle over `. A C1 structure on ` is deﬁned as follows.
7.1 CR Manifolds 165
Deﬁnition 7.1.1. Let ` be a real smooth manifold of dimension 2n − 1, n ≥ 2,
and let T
1,0
(`) be a subbundle of CT(`). We say that (`, T
1,0
(`)) is a Cauchy
Riemann manifold, abbreviated as CR manifold, with the CauchyRiemann structure
T
1,0
(`) if the following conditions are satisﬁed:
(1) di:
C
T
1,0
(`) = n −1,
(2) T
1,0
(`) ∩ T
0,1
(`) = ¦0¦, where T
0,1
= T
1,0
(`),
(3) (Integrability condition) For any A
1
, A
2
∈ Γ(l, T
1,0
(`)), the Lie bracket
[A
1
, A
2
] is still in Γ(l, T
1,0
(`)), where l is any open subset of ` and
Γ(l, T
1,0
(`)) denotes the space of all smooth sections of T
1,0
(`) over l.
Here T
1,0
(`) in (2) means the complex conjugation of T
1,0
(`). Note also that
condition (3) in Deﬁnition 7.1.1 is void when n = 2. The most natural C1 manifolds
are those deﬁned by smooth hypersurfaces in C
n
.
Example 7.1.2. Let ρ : C
n
→ 1 be a realvalued smooth function. Suppose that
the diﬀerential dρ does not vanish on the hypersurface ` = ¦. ∈ C
n
[ ρ(.) = 0¦.
Then ` is a smooth manifold with dim
R
` = 2n −1. Deﬁne a subbundle T
1,0
(`)
of CT(`) by T
1,0
(`) = T
1,0
(C
n
) ∩ CT(`). It is easily seen that (`, T
1,0
(`))
is a C1 manifold with the C1 structure T
1,0
(`) induced from the ambient space
C
n
.
Hence, it is natural to ask whether a given abstract C1 structure (`, T
1,0
(`))
on ` can be C1 embedded into some C
N
so that the given C1 structure coincides
with the induced C1 structure from the ambient space. The embedding problem
of an abstract C1 structure will make up the main course of Chapter 12.
Let (`, T
1,0
(`)) and (·, T
1,0
(·)) be two C1 manifolds. A smooth mapping ϕ
from ` to · is called a C1 mapping if ϕ
∗
1 is a smooth section of T
1,0
(·) for any
smooth section 1 in T
1,0
(`). Furthermore, if ϕ has a smooth C1 inverse mapping
ϕ
−1
, then we say that (`, T
1,0
(`)) is C1 diﬀeomorphic to (·, T
1,0
(·)).
We have the following lemma:
Lemma 7.1.3. Let (`, T
1,0
(`)) be a C1 manifold, and let · be a manifold.
Suppose that ` is diﬀeomorphic to · via a mapping ϕ. Then ϕ induces a C1
structure on ·, namely, ϕ
∗
T
1,0
(`), so that ϕ becomes a C1 diﬀeomorphism from
(`, T
1,0
(`)) onto (·, ϕ
∗
T
1,0
(`)), where ϕ
∗
is the diﬀerential map induced by ϕ.
Proof. We need to check the integrability condition on ϕ
∗
T
1,0
(`). However, this
follows immediately from the integrability condition on T
1,0
(`) and the fact that
[ϕ
∗
A
1
, ϕ
∗
A
2
] = ϕ
∗
[A
1
, A
2
] for any smooth vector ﬁelds A
1
, A
2
deﬁned on `.
Deﬁnition 7.1.4. A smooth function p deﬁned on a C1 manifold (`, T
1,0
(`))
is called a C1 function if 1p = 0 for any smooth section 1 in T
0,1
(`).
When ` is the boundary of a smooth domain in C
n
, this deﬁnition coincides
with Deﬁnition 3.0.1. If, in Deﬁnition 7.1.4, p is just a distribution, then 1p should
be interpreted in the sense of distribution.
166 CauchyRiemann Manifolds
7.2 The Tangential CauchyRiemann Complex
Let ` be a hypersurface in a complex manifold. The ∂ complex restricted to `
induces the tangential CauchyRiemann complex, or the ∂
b
complex. In fact, the
tangential CauchyRiemann complex can be formulated on any C1 manifold. There
are two diﬀerent approaches in this setting. One way is to deﬁne the tangential
CauchyRiemann complex intrinsically on any abstract C1 manifold itself without
referring to the ambient space. On the other hand, if the C1 manifold is sitting in
C
n
, or more generally, a complex manifold, the tangential CauchyRiemann complex
can also be deﬁned extrinsically via the ambient complex structure.
First, we assume that ` is a smooth hypersurface in C
n
, and let : be a deﬁning
function for `. In some open neighborhood l of `, let 1
p,q
, 0 ≤ j, ¡ ≤ n, be the
ideal in Λ
p,q
(C
n
) such that at each point . ∈ l the ﬁber 1
p,q
z
is generated by : and
∂:, namely, each element in the ﬁber 1
p,q
z
can be expressed in the form
:H
1
+∂: ∧ H
2
,
where H
1
is a smooth (j, ¡)form and H
2
is a smooth (j, ¡ − 1)form. Denote by
Λ
p,q
(C
n
)[
M
and 1
p,q
[
M
the restriction of Λ
p,q
(C
n
) and 1
p,q
respectively to `. Then,
we deﬁne
Λ
p,q
(`) = ¦the orthogonal complement of 1
p,q
[
M
in Λ
p,q
(C
n
)[
M
¦.
We denote by c
p,q
the space of smooth sections of Λ
p,q
(`) over `, i.e., c
p,q
(`) =
Γ(`, Λ
p,q
(`)). Let τ denote the following map
(7.2.1) τ : Λ
p,q
(C
n
) →Λ
p,q
(`),
where τ is obtained by ﬁrst restricting a (j, ¡)form φ in C
n
to `, then projecting
the restriction to Λ
p,q
(`). One should note that Λ
p,q
(`) is not intrinsic to `,
i.e., Λ
p,q
(`) is not a subspace of the exterior algebra generated by the complexiﬁed
cotangent bundle of `. This is due to the fact that ∂: is not orthogonal to the
cotangent bundle of `. Note also that c
p,n
= 0.
The tangential CauchyRiemann operator
∂
b
: c
p,q
(`) →c
p,q+1
(`)
is now deﬁned as follows: For any φ ∈ c
p,q
(`), pick a smooth (j, ¡)form φ
1
in
C
n
that satisﬁes τφ
1
= φ. Then, ∂
b
φ is deﬁned to be τ∂φ
1
in c
p,q+1
(`). If φ
2
is
another (j, ¡)form in C
n
such that τφ
2
= φ, then
φ
1
−φ
2
= :p +∂: ∧ /,
for some (j, ¡)form p and (j, ¡ −1)form /. It follows that
∂(φ
1
−φ
2
) = :∂p +∂: ∧ p −∂: ∧ ∂/,
and hence,
τ∂(φ
1
−φ
2
) = 0.
7.2 The Tangential CauchyRiemann Complex 167
Thus, the deﬁnition of ∂
b
is independent of the choice of φ
1
. Since ∂
2
= 0, we have
∂
2
b
= 0 and the following boundary complex
0 → c
p,0
(`)
∂
b
−→ c
p,1
(`)
∂
b
−→
∂
b
−→ c
p,n−1
(`) → 0.
For the intrinsic approach, we will assume that (`, T
1,0
(`)) is an orientable C1
manifold of real dimension 2n−1 with n ≥ 2. A real smooth manifold ` is said to be
orientable if there exists a nonvanishing top degree form on `. We shall assume that
` is equipped with a Hermitian metric on CT(`) so that T
1,0
(`) is orthogonal
to T
0,1
(`). Denote by η(`) the orthogonal complement of T
1,0
(`) ⊕ T
0,1
(`).
It is easily seen that η(`) is a line bundle over `. Now denote by T
∗1,0
(`) and
T
∗0,1
(`) the dual bundles of T
1,0
(`) and T
0,1
(`) respectively. By deﬁnition it
means that forms in T
∗1,0
(`) annihilate vectors in T
0,1
(`) ⊕η(`) and forms in
T
∗0,1
(`) annihilate vectors in T
1,0
(`) ⊕η(`). Deﬁne the vector bundle Λ
p,q
(`),
0 ≤ j, ¡ ≤ n −1, by
Λ
p,q
(`) = Λ
p
T
∗1,0
(`) ⊗Λ
q
T
∗0,1
(`).
This can be identiﬁed with a subbundle of Λ
p+q
CT
∗
(`). It follows that Λ
p,q
(`)
deﬁned in this way is intrinsic to `. Denote by c
p,q
the space of smooth sections
of Λ
p,q
(`) over `, i.e., c
p,q
(`) = Γ(`, Λ
p,q
(`)). We deﬁne the operator
∂
b
: c
p,q
(`) →c
p,q+1
(`)
as follows: If φ ∈ c
p,0
, ∂
b
φ is deﬁned by
¸∂
b
φ, (\
1
∧ ∧ \
p
) ⊗1) = 1¸φ, \
1
∧ ∧ \
p
)
for all sections \
1
, , \
p
of T
1,0
(`) and 1 of T
0,1
(`). Then ∂
b
is extended to
c
p,q
(`) for ¡ 0 as a derivation. Namely, if φ ∈ c
p,q
(`), we deﬁne
¸∂
b
φ, (\
1
∧ ∧\
p
) ⊗(1
1
∧ ∧1
q+1
))
=
1
¡ + 1
_
q+1
j=1
(−1)
j+1
1
j
¸φ, (\
1
∧ ∧\
p
) ⊗(1
1
∧ ∧
´
1
j
∧ ∧1
q+1
))
+
i<j
(−1)
i+j
¸φ, (\
1
∧ ∧\
p
) ⊗([1
i
, 1
j
]∧1
1
∧ ∧
´
1
i
∧ ∧
´
1
j
∧ ∧1
q+1
))
_
.
Here by
´
1 we mean that the term 1 is omitted from the expression. If we let π
p,q
be the projection from Λ
p+q
CT
∗
(`) onto Λ
p,q
(`), then ∂
b
= π
p,q+1
◦ d, where d
is the exterior derivative on `.
One should note how the integrability condition of the C1 structure T
1,0
(`)
comes into play in the deﬁnition of ∂
b
, and it is standard to see that the following
sequence
0 → c
p,0
(`)
∂
b
−→ c
p,1
(`)
∂
b
−→
∂
b
−→ c
p,n−1
(`) → 0,
168 CauchyRiemann Manifolds
forms a complex, i.e., ∂
2
b
= 0.
Notice that j plays no role in the formulation of the tangential CauchyRiemann
operators. Thus, it suﬃces to consider the action of ∂
b
on type (0, ¡)forms,
0 ≤ ¡ ≤ n −1. When the C1 manifold (`, T
1,0
(`)) is embedded as a smooth
hypersurface in C
n
with the C1 structure T
1,0
(`) induced from the ambient space,
the tangential CauchyRiemann complex on ` can be deﬁned either extrinsically or
intrinsically. These two complexes are diﬀerent, but one can easily show that they
are isomorphic. Thus, if the C1 manifold is embedded, we shall not distinguish
the extrinsic or intrinsic deﬁnitions of the tangential CauchyRiemann complex.
The operator ∂
b
is a ﬁrst order diﬀerential operator, and one may consider the
inhomogeneous ∂
b
equation
(7.2.2) ∂
b
n = ),
where ) is a (0, ¡)form on `. Equation (7.2.2) is overdetermined when 0 < ¡ <
n −1. Since ∂
2
b
= 0, for equation (7.2.2) to be solvable, it is necessary that
(7.2.3) ∂
b
) = 0.
Condition (7.2.3) is called the compatibility condition for the ∂
b
equation. We shall
discuss the solvability and regularity of the ∂
b
operator in detail in the next few
chapters.
Let 1
1
, , 1
n−1
be a local basis for smooth sections of T
1,0
(`) over some open
subset l ⊂ `, so 1
1
, , 1
n−1
is a local basis for T
0,1
(`) over l. Next we
choose a local section T of CT(`) such that 1
1
, , 1
n−1
, 1
1
, , 1
n−1
and T
span CT(`) over l. We may assume that T is purely imaginary.
Deﬁnition 7.2.1. The Hermitian matrix (c
ij
)
n−1
i,j=1
deﬁned by
(7.2.4) [1
i
, 1
j
] = c
ij
T, mod (T
1,0
(l) ⊕T
0,1
(l))
is called the Levi form associated with the given C1 structure.
The Levi matrix (c
ij
) clearly depends on the choices of 1
1
, , 1
n−1
and T.
However, the number of nonzero eigenvalues and the absolute value of the signature
of (c
ij
) at each point are independent of the choices of 1
1
, , 1
n−1
and T. Hence,
after changing T to −T, it makes sense to consider positive deﬁniteness of the matrix
(c
ij
).
Deﬁnition 7.2.2. The C1 structure is called pseudoconvex at j ∈ ` if the matrix
(c
ij
(j)) is positive semideﬁnite after an appropriate choice of T. It is called strictly
pseudoconvex at j ∈ ` if the matrix (c
ij
(j)) is positive deﬁnite. If the C1 struc
ture is (strictly) pseudoconvex at every point of `, then ` is called a (strictly)
pseudoconvex C1 manifold. If the Levi form vanishes completely on an open set
l ⊂ `, i.e., c
ij
= 0 on l for 1 ≤ i, , ≤ n −1, ` is called Levi ﬂat.
7.2 The Tangential CauchyRiemann Complex 169
Theorem 7.2.3. Let 1 ⊂ C
n
, n ≥ 2, be a bounded domain with C
∞
boundary.
Then 1 is (strictly) pseudoconvex if and only if ` = /1 is a (strictly) pseudoconvex
C1 manifold.
Proof. Let : be a C
∞
deﬁning function for 1, and let j ∈ /1. We may assume
that (∂:,∂.
n
)(j) ,= 0. Hence,
1
k
=
∂:
∂.
n
∂
∂.
k
−
∂:
∂.
k
∂
∂.
n
, for / = 1, , n −1,
form a local basis for the tangential type (1, 0) vector ﬁelds near j on the boundary.
If 1 =
n
j=1
o
j
(∂,∂.
j
) is a tangential type (1, 0) vector ﬁeld near j, then we have
n
j=1
o
j
(∂:,∂.
j
) = 0 on /1 and 1 = (∂:,∂.
n
)
−1
n−1
j=1
o
j
1
j
on /1. Hence, if we
let η = ∂: −∂:, we obtain
n−1
i,j=1
c
ij
o
i
o
j
=
n−1
i,j=1
¸η, [1
i
, 1
j
])o
i
o
j
=
n−1
i,j=1
(1
i
¸η, 1
j
) −1
j
¸η, 1
i
) −2¸dη, 1
i
∧ 1
j
))o
i
o
j
=
n−1
i,j=1
4¸∂∂:, 1
i
∧ 1
j
)o
i
o
j
= 4
¸
¸
¸
¸
∂:
∂.
n
¸
¸
¸
¸
2
¸∂∂:, 1 ∧ 1)
= 4
¸
¸
¸
¸
∂:
∂.
n
¸
¸
¸
¸
2 n
i,j=1
∂
2
:
∂.
i
∂.
j
o
i
o
j
,
which gives the desired equivalence between these two deﬁnitions. This proves the
theorem.
We note that, locally, a C1 manifold in C
n
is pseudoconvex if and only if it is
the boundary of a smooth pseudoconvex domain from one side.
Lemma 7.2.4. Any compact strongly pseudoconvex C1 manifold (`, T
1,0
(`)) is
orientable.
Proof. Locally, let η, ω
1
, , ω
n−1
be the one forms dual to T, 1
1
, , 1
n−1
which
are deﬁned as above. The vector ﬁeld T is chosen so that the Levi form is positive
deﬁnite. Then we consider the following 2n −1 form
(7.2.5) η ∧ ω
1
∧ ω
1
∧ ∧ ω
n−1
∧ ω
n−1
.
It is not hard to see that the 2n−1 form (7.2.5) generated by other bases will diﬀer
from (7.2.5) only by a positive function. Hence, a partition of unity argument will
give the desired nowhere vanishing 2n −1 form on `, and the lemma is proved.
170 CauchyRiemann Manifolds
7.3 Lewy’s Equation
In this section, we shall present a partial diﬀerential operator of order one with
variable coeﬃcients that, in general, does not possess a solution for a given smooth
function. This discovery destroys all hope for the existence of solutions to a rea
sonably smooth partial diﬀerential operator. Since this operator arises from the
tangential CauchyRiemann operator on the boundary of a strongly pseudocon
vex domain, this discovery also inspires an intensive investigation of the tangential
CauchyRiemann operator.
Let Ω
n
be the Siegel upper half space deﬁned by
(7.3.1) Ω
n
= ¦(.
t
, .
n
) ∈ C
n
[ Im.
n
[.
t
[
2
¦,
where .
t
= (.
1
, , .
n−1
) and [.
t
[
2
= [.
1
[
2
+ + [.
n−1
[
2
. When n = 1, Ω
1
is
reduced to the upper half space of the complex plane which is conformally equivalent
to the unit disc. For n 1, the Cayley transform also maps the unit ball 1
n
biholomorphically onto the Siegel upper half space Ω
n
, i.e.,
(7.3.2)
Φ : 1
n
→ Ω
n
. → n = Φ(.)
= i
_
c
n
+.
1 −.
n
_
=
_
i.
1
1 −.
n
, ,
i.
n−1
1 −.
n
, i
1 +.
n
1 −.
n
_
,
where c
n
= (0, , 0, 1).
For n ≥ 2, a simple calculation shows
(7.3.3) 1
k
=
∂
∂.
k
+ 2i.
k
∂
∂.
n
for / = 1, , n −1,
forms a global basis for the space of tangential (1, 0) vector ﬁelds on the boundary
/Ω
n
and
[1
j
, 1
k
] = −2iδ
jk
∂
∂t
,
where .
n
= t + i: and δ
jk
is the Kronecker delta. It follows that if we choose
T = −2i(∂,∂t), the Levi matrix (c
ij
) is the identity matrix which implies that
Ω
n
is a strongly pseudoconvex domain. Furthermore, the boundary /Ω
n
can be
identiﬁed with H
n
= C
n−1
1 via the map
π : (.
t
, t +i[.
t
[
2
) →(.
t
, t).
Therefore, a C1 structure can be induced via π on H
n
by
(7.3.4) 7
k
= π
∗
1
k
=
∂
∂.
k
+i.
k
∂
∂t
for / = 1, , n−1. Thus, if ) =
n−1
j=1
)
j
ω
j
is a ∂
b
closed (0, 1)form on H
n
, where
ω
j
is the (1, 0)form dual to 7
j
, the solvability of the equation ∂
b
n = ) is equivalent
to the existence of a function n satisfying the following system of equations:
7
k
n = )
k
, 1 ≤ / ≤ n −1.
7.3 Lewy’s Equation 171
When n = 2, the Siegel upper half space is given by
¦(., n) ∈ C
2
[ [.[
2
−
1
2i
(n −n) < 0¦.
Hence, the tangential CauchyRiemann operator is generated by
(7.3.5) 1 =
∂
∂.
−2i.
∂
∂n
,
with n = t+i:, and the corresponding operator, denoted by 7, via the identiﬁcation
on H
2
is
(7.3.6) 7 =
∂
∂.
−i.
∂
∂t
,
where (., t) with . = r +ij are the coordinates on H
2
= C1. The coeﬃcients of
the operator 7 deﬁned in (7.3.6) are real analytic. Hence, for a given real analytic
function ), the equation
(7.3.7) 7n = )
always has a real analytic solution n locally as is guaranteed by the Cauchy
Kowalevski theorem. However, the next theorem shows that equation (7.3.7) does
not possess a solution in general even when ) is smooth.
Theorem 7.3.1 (Lewy). Let ) be a continuous realvalued function depending
only on t. If there is a C
1
solution n(r, j, t) to the equation (7.3.7), then ) must
be real analytic in some neighborhood of t = 0.
Proof. Locally, near the origin any point can be expressed in terms of the polar
coordinates as
(r, j, t) = (:c
iθ
, t)
with : < 1 and [t[ < 1 for some 1 0. Set : = :
2
. Consider the function \ (:, t)
deﬁned by
\ (:, t) =
_
¦]z]=r]
n(., t) d..
Then, by Stokes’ theorem we have
\ (:, t) = −
__
¦]z]<r]
∂n
∂.
d. ∧ d.
= −
__
¦]z]<r]
_
) +i.
∂n
∂t
_
d. ∧ d.
= 2πi:
2
)(t) −2
∂
∂t
_
2π
0
_
r
0
ρc
iθ
n(ρc
iθ
, t)ρdρdθ.
172 CauchyRiemann Manifolds
Hence,
∂\
∂:
(:, t) = 2πi)(t) −2
∂
∂t
_
1
2:
∂
∂:
_
2π
0
_
r
0
ρc
iθ
n(ρc
iθ
, t)ρdρdθ
_
= 2πi)(t) −
1
:
∂
∂t
_
2π
0
:c
iθ
n(:c
iθ
, t):dθ
= 2πi)(t) +i
∂
∂t
\ (:, t).
Set
1(t) =
_
t
0
)(η) dη.
Then, we obtain
_
∂
∂t
+i
∂
∂:
_
(\ (:, t) + 2π1(t)) = 0,
which implies that the function l(:, t) = \ (:, t) + 2π1(t) is holomorphic on the
set ¦t + i: ∈ C[ [t[ < 1, 0 < : < 1
2
¦, and l(:, t) is continuous up to the real
axis ¦: = 0¦ with realvalued boundary value 2π1(t). Hence, by the Schwarz
reﬂection principle, l(:, t) can be extended holomorphically across the boundary
to the domain ¦t + i: ∈ C[[t[ < 1, [:[ < 1
2
¦. In particular, 1(t), and hence )(t),
must be real analytic on (−1, 1). This proves Theorem 7.3.1.
In Section 10.3 we shall give a complete characterization of the local solvability
of the Lewy operator (7.3.6).
7.4 Linear Partial Diﬀerential Operators with Constant Coeﬃcients
In contrast to the nonsolvable operator (7.3.6), we shall present in this section
a fundamental positive result in the theory of partial diﬀerential equations which
asserts the existence of a distribution fundamental solution to any linear partial
diﬀerential operator with constant coeﬃcients. It follows by convolution that every
partial diﬀerential operator with constant coeﬃcients is locally solvable.
Theorem 7.4.1 (Malgrange, Ehrenpreis). Let
1 =
]α]≤k
o
α
1
α
x
be a partial diﬀerential operator with constant coeﬃcients on 1
n
, where 1
α
x
=
(∂,∂r
1
)
α1
(∂,∂r
n
)
αn
for any multiindex α = (α
1
, , α
n
) with nonnegative
integer components. If ) ∈ C
∞
0
(1
n
), then there exists a C
∞
function /(r) satisfying
1/ = ) on 1
n
.
Proof. The proof will be done via the Fourier transform. For any p ∈ 1
1
(1
n
), the
Fourier transform ˆ p(ξ) of p is deﬁned by
ˆ p(ξ) =
_
R
n
c
−ixξ
p(r) dr,
7.4 Linear Partial Diﬀerential Operators with Constant Coeﬃcients 173
where r ξ = r
1
ξ
1
+ + r
n
ξ
n
. Note ﬁrst that, via a rotation of coordinates
and multiplying 1 by a constant, we may assume that the corresponding Fourier
transform j(ξ) of the operator 1 is
j(ξ) = ξ
k
n
+
k−1
j=0
o
j
(ξ
t
)ξ
j
n
,
where ξ = (ξ
t
, ξ
n
) with ξ
t
∈ 1
n−1
, and o
j
(ξ
t
) is a polynomial in ξ
t
for 0 ≤ , ≤ / −1.
Next, we complexify j(ξ); namely, we view ξ as a variable in C
n
. Hence, for each
ξ
t
∈ 1
n−1
, j(ξ
t
, ξ
n
) is a polynomial of degree / in ξ
n
. Let λ
1
(ξ
t
), , λ
k
(ξ
t
) be its
zeros, arranged so that if i ≤ ,, Imλ
i
(ξ
t
) ≤ Imλ
j
(ξ
t
), and Reλ
i
(ξ
t
) ≤ Reλ
j
(ξ
t
) if
Imλ
i
(ξ
t
) = Imλ
j
(ξ
t
). One sees easily that these / functions Imλ
j
(ξ
t
) are continuous
in ξ
t
.
We then need to construct a measurable function
φ : 1
n−1
→[−/ −1, / + 1]
such that for all ξ
t
∈ 1
n−1
, we have
min¦[φ(ξ
t
) −Imλ
j
(ξ
t
)[ : 1 ≤ , ≤ /¦ ≥ 1.
Set n
0
(ξ
t
) = −/ −1 and n
k+1
(ξ
t
) = / + 1. For 1 ≤ , ≤ /, deﬁne
n
j
(ξ
t
) = max¦min¦Imλ
j
(ξ
t
), / + 1¦, −/ −1¦.
The functions n
j
(ξ
t
) are continuous in ξ
t
, so the sets
\
j
= ¦ξ
t
: n
j+1
(ξ
t
) −n
j
(ξ
t
) ≥ 2¦,
for , = 0, , /, are measurable. It is clear that ∪
k
j=0
\
j
is a covering of 1
n−1
. Thus,
we can construct disjoint measurable subsets \
j
⊂ \
j
which still cover 1
n−1
. Deﬁne
φ(ξ
t
) =
1
2
(n
j+1
(ξ
t
) +n
j
(ξ
t
)),
if ξ
t
∈ \
j
. This completes the construction of φ(ξ
t
).
We deﬁne /(r) by
/(r) =
1
(2π)
n
_
R
n−1
_
¦Imξn=φ(ξ
)]
c
ixξ
_
ˆ
)(ξ)
j(ξ)
_
dξ
n
dξ
t
.
The key is to observe that, as [Reξ[ → ∞,
ˆ
)(ξ) is rapidly decreasing whereas Imξ
remains bounded, and to see that the line Imξ
n
= φ(ξ
t
) in the ξ
n
plane has distance
at least one from any zero of j(ξ) and at most / +1 from the real axis. Hence, the
integrand is bounded and rapidly decreasing at inﬁnity, so the integral is absolutely
convergent. The same reasoning shows that we can diﬀerentiate under the integral
174 CauchyRiemann Manifolds
sign as often as we please. It follows that /(r) is smooth. Finally, we apply 1 to
/(r) and get
1/(r) =
1
(2π)
n
_
R
n−1
_
¦Imξn=φ(ξ
)]
c
ixξ
ˆ
)(ξ) dξ
n
dξ
t
.
The integrand on the righthand side is an entire function which is rapidly decreasing
as [Reξ
n
[ → ∞. Therefore, by Cauchy’s theorem, the contour of integration can
be deformed back to the real axis. By invoking the inverse Fourier transform we
obtain 1/ = ). The proof is now complete.
As an easy consequence of Theorem 7.4.1, if 1 is a linear partial diﬀerential
operator with constant coeﬃcients, then for any given function )(r) which is smooth
near some point r
0
, we can ﬁnd locally a smooth solution /(r) such that 1/ = )
near r
0
.
We now return to the solvability of the ∂
b
equation in a very special case. Let
the C1 manifold (`, T
1,0
(`)) of real dimension 2n − 1, n ≥ 2, be Levi ﬂat in a
neighborhood l of the reference point j, then we can apply the Frobenius theorem
(Theorem 1.6.1) to Re1
1
, Im1
1
, , Re1
n−1
, Im1
n−1
, where 1
1
, , 1
n−1
is a local
basis of T
1,0
(`) near j. Thus there exist local coordinates (r
1
, , r
2n−2
, t) such
that the vector ﬁelds Re1
1
, Im1
1
, , Re1
n−1
, Im1
n−1
span the tangent space of
each leaf ¦t = c¦ for some constant c. Therefore, on each leaf we may apply Theorem
2.3.1, for n = 2, or the NewlanderNirenberg theorem (Theorem 5.4.4), for n ≥ 3,
to show that ` is locally foliated by complex submanifolds of complex dimension
n−1. In this case, the local solvability of the tangential CauchyRiemann equation,
∂
b
n = ), where ) is a ∂
b
closed (0, 1)form, can be reduced to a ∂ problem with a
parameter. In the next few chapters, the global and local solvability of ∂
b
will be
discussed in detail.
NOTES
The tangential CauchyRiemann complex was ﬁrst introduced by J. J. Kohn and
H. Rossi [KoRo 1]. See also the books by A. Boggess [Bog 1], G. B. Folland and
J. J. Kohn [FoKo 1] and H. Jacobowitz [Jac 1]. The nonsolvability theorem for the
operator (7.3.6) was proved by H. Lewy [Lew 2]. A more general theorem due to L.
H¨ormander [H¨or 1,7] states that the tangential CauchyRiemann equation on a real
three dimensional C1 manifold is not locally solvable if it is not Levi ﬂat. For a
proof of the CauchyKowalevski theorem, the reader is referred to [Joh 1]. Theorem
7.4.1 was originally proved by B. Malgrange [Mal 1] and L. Ehrenpreis [Ehr 1]. The
proof we present here is due to L. Nirenberg [Nir 2].
175
176
CHAPTER 8
SUBELLIPTIC ESTIMATES FOR
SECOND ORDER DIFFERENTIAL EQUATIONS AND
b
In this chapter, we study subelliptic operators which are not elliptic. We analyze
two types of operators in detail. One is a real second order diﬀerential equation
which is a sum of squares of vector ﬁelds. The other is the ∂
b
Laplacian on a C1
manifold. We use pseudodiﬀerential operators to study both operators.
For this purpose, we shall brieﬂy review the deﬁnitions and basic properties
of the simplest pseudodiﬀerential operators. Using pseudodiﬀerential operators,
H¨ormander’s theorem on the hypoellipticity of sums of squares of vector ﬁelds will
be discussed in Section 8.2.
The ∂
b
Laplacian,
b
, is not elliptic. There is a onedimensional characteristic
set. However, under certain conditions, one is able to establish the 1,2estimate for
the
b
operator via potentialtheoretic methods and pseudodiﬀerential operators. In
the last two sections of this chapter, the 1,2estimate for the
b
operator on compact
strongly pseudoconvex C1 manifolds is proved, which leads to the existence and
regularity theorems of the ∂
b
equation. Global existence theorems for ∂
b
on the
boundary of a pseudoconvex domain in C
n
will be discussed in Chapter 9.
8.1 Pseudodiﬀerential Operators
We ﬁrst introduce some simple pseudodiﬀerential operators. Let o be the Schwar
tz space in 1
n
. For the deﬁnitions of the space o and the Sobolev space \
s
(1
n
),
the reader is referred to the Appendix A. We begin this section with the following
deﬁnition:
Deﬁnition 8.1.1. A linear operator T : o →o is said to be of order : if for each
: ∈ 1 we have
 Tn 
s
≤ C
s
 n 
s+m
, for all n ∈ o,
with the constant C
s
independent of n.
We note that, by deﬁnition, any linear operator of order : from o into itself
extends to a bounded linear operator from \
s+m
(1
n
) to \
s
(1
n
) for every : ∈ 1.
It is obvious that any diﬀerential operator 1
α
with [α[ = : is of order :, where
α = (α
1
, , α
n
) and 1
α
= 1
α1
x1
1
αn
xn
.
For any : ∈ 1, we deﬁne Λ
s
: o →o by
Λ
s
n(r) =
1
(2π)
n
_
R
n
c
ixξ
(1 +[ξ[
2
)
s
2
ˆ n(ξ) dξ
8.1 Pseudodiﬀerential Operators 177
where ˆ n is the Fourier transform of n. Here, σ(Λ
s
) = (1+[ξ[
2
)
s
2
is called the symbol
of Λ
s
. Obviously, the operator Λ
s
is of order :. Λ
s
is called a pseudodiﬀerential op
erator of order :. We should think of Λ
s
as a generalization of diﬀerential operators
to fractional and negative order.
The purpose of this section is to prove some basic properties of the commutators
between Λ
s
and functions in o. We need the following lemma:
Lemma 8.1.2.
_
(1 +[r[
2
),(1 +[j[
2
)
_
s
≤ 2
]s]
(1 +[r −j[
2
)
]s]
for all r, j ∈ 1
n
and
every : ∈ 1.
Proof. From the triangle inequality [r[ ≤ [r −j[ +[j[, we obtain [r[
2
≤ 2([r −j[
2
+
[j[
2
) and hence 1 + [r[
2
≤ 2(1 + [r −j[
2
)(1 + [j[
2
). Thus, if : ≥ 0, the lemma is
proved. For : < 0, the same arguments can be applied with r and j reversed and
: replaced by −:. This proves the lemma.
We employ Plancherel’s theorem to study the commutators of Λ
s
and functions
in o. We ﬁrst show that multiplication by a function in o is of order zero.
Lemma 8.1.3. For any p ∈ o and : ∈ 1,  pn 
s
 n 
s
uniformly for all n ∈ o.
Proof. By the Fourier transform formula for convolution, we obtain
(1 +[ξ[
2
)
s
2
´ pn(ξ) =
1
(2π)
n
_ _
1 +[ξ[
2
1 +[η[
2
_
s
2
ˆ p(ξ −η)(1 +[η[
2
)
s
2
ˆ n(η) dη.
Now we view
_
(1 +[ξ[
2
),(1 +[η[
2
)
_
s/2
ˆ p(ξ − η) as a kernel 1(ξ, η) and set )(η) =
(1 +[η[
2
)
s/2
ˆ n(η). Lemma 8.1.2 shows that
[1(ξ, η)[ (1 +[ξ −η[
2
)
s
2
[ˆ p(ξ −η)[.
Since p ∈ o, it follows that ˆ p ∈ o and that the hypotheses of Theorem B.10 in the
Appendix are satisﬁed by this kernel. Therefore, we have
 pn 
s
 )  =  n 
s
.
This proves the lemma.
Theorem 8.1.4. If p, / ∈ o, then for any :, : ∈ 1, we have
(1) [Λ
s
, p] is of order : −1,
(2) [Λ
r
, [Λ
s
, p]] is of order : +: −2,
(3) [[Λ
s
, p], /] is of order : −2.
Proof. The proof of the theorem will proceed exactly as in Lemma 8.1.3. For (1) it
suﬃces to show that Λ
r
[Λ
s
, p]Λ
1−r−s
is of order zero for any : ∈ 1. Let n ∈ o, and
set ) = Λ
r
[Λ
s
, p]Λ
1−r−s
n. A direct calculation shows that
ˆ
)(ξ) =
1
(2π)
n
_
1(ξ, η)ˆ n(η) dη,
where
1(ξ, η) =
(1 +[ξ[
2
)
r
2
(1 +[η[
2
)
r+s−1
2
_
(1 +[ξ[
2
)
s
2
−(1 +[η[
2
)
s
2
_
ˆ p(ξ −η).
178 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
From the mean value theorem, we have
[(1 +[ξ[
2
)
s
2
−(1 +[η[
2
)
s
2
[ [ξ −η[
_
(1 +[ξ[
2
)
s−1
2
+ (1 +[η[
2
)
s−1
2
_
for all ξ, η ∈ 1
n
, hence, by Lemma 8.1.2 we obtain
[1(ξ, η)[ [ξ −η[
_
_
1 +[ξ[
2
1 +[η[
2
_
r+s−1
2
+
_
1 +[ξ[
2
1 +[η[
2
_
r
2
_
[ˆ p(ξ −η)[
[ξ −η[
_
(1 +[ξ −η[
2
)
r+s−1
2
+ (1 +[ξ −η[
2
)
r
2
_
[ˆ p(ξ −η)[.
Since p ∈ o, so is ˆ p ∈ o. Therefore, by Theorem B.10 in the Appendix, (1) is
proved.
To prove (2), for n ∈ o, a similar calculation shows
([Λ
r
, [Λ
s
, p]]n)
ˆ
(ξ) =
1
(2π)
n
_
1(ξ, η)ˆ n(η) dη,
where
1(ξ, η) =
_
(1 +[ξ[
2
)
r
2
−(1 +[η[
2
)
r
2
_ _
(1 +[ξ[
2
)
s
2
−(1 +[η[
2
)
s
2
_
ˆ p(ξ −η).
Here ([Λ
r
, [Λ
s
, p]]n)
ˆ
(ξ) denotes the Fourier transform of [Λ
r
, [Λ
s
, p]]n. It follows
now from the same estimates as in (1) that
[1(ξ, η)[ [ξ −η[
2
_
(1 +[ξ[
2
)
r−1
2
+ (1 +[η[
2
)
r−1
2
_
_
(1 +[ξ[
2
)
s−1
2
+ (1 +[η[
2
)
s−1
2
_
[ˆ p(ξ −η)[,
which implies easily that [Λ
r
, [Λ
s
, p]] is of order : +: −2. This proves (2).
For (3), let n ∈ o. It is easy to get from Taylor’s expansion that
([Λ
s
, p]n)
ˆ
(ξ) =
1
(2π)
n
_
_
(1 +[ξ[
2
)
s
2
−(1 +[η[
2
)
s
2
_
ˆ p(ξ −η)ˆ n(η) dη
=
n
j=1
1
(2π)
n
_
(ξ
j
−η
j
)
∂
∂ξ
j
(1 +[ξ[
2
)
s
2
ˆ p(ξ −η)ˆ n(η) dη
+
1
(2π)
n
_
O(ξ, η)ˆ p(ξ −η)ˆ n(η) dη
=
¯
T
1
n +
¯
T
2
n,
where O(ξ, η) can be estimated by
[O(ξ, η)[ [ξ −η[
2
_
(1 +[ξ[
2
)
s−2
2
+ (1 +[η[
2
)
s−2
2
_
.
8.1 Pseudodiﬀerential Operators 179
Now, as in (1), it is easily seen that the operator T
2
is of order : − 2. Thus, to
ﬁnish the proof of (3), it suﬃces to show that [T
1
, /] is of order : −2. Write
/(ξ, η) =
n
j=1
(ξ
j
−η
j
)
∂
∂ξ
j
(1 +[ξ[
2
)
s
2
ˆ p(ξ −η)
= :
n
j=1
ξ
j
(ξ
j
−η
j
)(1 +[ξ[
2
)
s−2
2
ˆ p(ξ −η).
Thus, from a direct calculation we obtain
([T
1
, /]n)
ˆ
(ξ) =
1
(2π)
n
_
1(ξ, η)ˆ n(η) dη,
where
1(ξ, η) =
1
(2π)
n
_
_
/(ξ, ζ)
ˆ
/(ζ −η) −/(ζ, η)
ˆ
/(ξ −ζ)
_
dζ
=
1
(2π)
n
_
ˆ
/(ξ −ζ)
_
/(ξ, ξ +η −ζ) −/(ζ, η)
_
dζ.
Since
[/(ξ, ξ +η −ζ) −/(ζ, η)[
[ξ −ζ[[ζ −η[
_
(1 +[ξ[
2
)
s−2
2
+ (1 +[ζ[
2
)
s−2
2
_
[ˆ p(ζ −η)[,
we can estimate 1(ξ, η) as follows,
[1(ξ, η)[
_
[ξ −ζ[[ζ −η[(1 +[ξ[
2
)
s−2
2
[
ˆ
/(ξ −ζ)[[ˆ p(ζ −η)[ dζ
+
_
[ξ −ζ[[ζ −η[(1 +[ζ[
2
)
s−2
2
[
ˆ
/(ξ −ζ)[[ˆ p(ζ −η)[ dζ
(1 +[ξ[
2
)
s−2
2
__
[ζ[[
ˆ
/(ζ)[[ξ −η −ζ[[ˆ p(ξ −η −ζ)[ dζ
+
_
[ζ[(1 +[ζ[
2
)
s−2
2
[
ˆ
/(ζ)[[ξ −η −ζ[[ˆ p(ξ −η −ζ)[ dζ
_
≤ C
m
(1 +[ξ[
2
)
s−2
2
(1 +[ξ −η[
2
)
−m
,
for any : ∈ N, where C
m
is a constant depending on :. Here we have used the
fact that both
ˆ
/ and ˆ p are in o. Choosing : to be suﬃciently large and applying
arguments similar to those used in the proof of (1), (3) is proved. This completes
the proof of Theorem 8.1.4.
Theorem 8.1.5. Let 1 and Q be two diﬀerential operators of order / and :
respectively with coeﬃcients in o. Then [Λ
s
, 1] is of order : + / − 1, [Λ
r
, [Λ
s
, 1]]
is of order : +: +/ −2, and [[Λ
s
, 1], Q] is of order : +/ +:−2.
180 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
Proof. Write
1 =
]α]≤k
o
α
(r)1
α
,
with o
α
(r) ∈ o. Since 1
α
commutes with Λ
s
, the commutator of Λ
s
with 1 is
reduced to commutators of Λ
s
with o
α
(r), composed with 1
α
. This proves the
theorem.
We shall use operators of the form generated by Λ
s
, 1
α
and multiplication by
functions in o plus their Lie brackets. All these are pseudodiﬀerential operators
and the computation of their orders is similar to that for diﬀerential operators.
8.2 Hypoellipticity of Sum of Squares of Vector Fields
Let Ω be an open neighborhood of the origin in 1
n
. Let A
i
=
n
j=1
o
ij
(∂,∂r
j
),
0 ≤ i ≤ / with / ≤ n, be vector ﬁelds with smooth realvalued coeﬃcients o
ij
(r)
on Ω. Deﬁne the second order partial diﬀerential operator
(8.2.1) 1 =
k
i=1
A
2
i
+A
0
+/(r),
where /(r) is a smooth realvalued function on Ω.
Denote by L
1
the collection of the A
i
’s, 0 ≤ i ≤ /. Then, inductively for an
integer : ≥ 2 we deﬁne L
m
to be the collection of L
m−1
and the vector ﬁelds of
the form [A, Y ] with A ∈ L
1
and Y ∈ L
m−1
.
Deﬁnition 8.2.1. The partial diﬀerential operator 1 deﬁned as in (8.2.1) is said
to be of ﬁnite type at point j ∈ Ω if there exists an : such that L
m
spans the whole
tangent space at j.
Deﬁnition 8.2.2. A partial diﬀerential operator 1 is said to be hypoelliptic in Ω
if it satisﬁes the following property: let n and ) be distributions satisfying 1n = )
in Ω, then u is smooth on l if ) is smooth on l for any open subset l of Ω
The task of this section is to prove the following main theorem:
Theorem 8.2.3. Let 1 be the partial diﬀerential operator deﬁned as in (8.2.1).
Suppose that 1 is of ﬁnite type at every point in Ω. Then 1 is hypoelliptic in Ω.
The heat operator on 1
n+1
deﬁned by
1 =
_
∂
∂r
1
_
2
+ +
_
∂
∂r
n
_
2
−
∂
∂t
is a typical example of such an operator, where the coordinates in 1
n+1
are denoted
by (r, t) = (r
1
, , r
n
, t). Another simple example with variable coeﬃcients is the
Grushin operator on 1
2
deﬁned by
1 =
∂
2
∂r
2
+r
2
∂
2
∂j
2
.
According to Theorem 8.2.3, both operators are hypoelliptic.
To prove Theorem 8.2.3, we begin with the following a priori estimate. By
shrinking the domain Ω, if necessary, we may assume that o
ij
(r) and /(r) are in
C
∞
(Ω) for all i, ,.
8.2 Hypoellipticity of Sum of Squares of Vector Fields 181
Lemma 8.2.4. Let 1 be deﬁned as in (8.2.1). There exists C 0 such that
(8.2.2)
k
i=1
 A
i
n 
2
≤ C([(1n, n)[ +  n 
2
), n ∈ C
∞
0
(Ω).
Proof. Let A
∗
i
be the adjoint operator for A
i
. Then A
∗
i
= −A
i
+ /
i
, where /
i
=
−
n
j=1
(∂o
ij
,∂r
j
). Integration by parts shows
−(A
2
i
n, n) =  A
i
n 
2
+ O( A
i
n  n )
and
(A
0
n, n) = −(n, A
0
n) +O( n 
2
).
It follows that
Re(A
0
n, n) = O( n 
2
).
Adding up these estimates, we obtain
k
i=1
 A
i
n 
2
= −Re(1n, n) +O
_
k
i=1
 A
i
n  n  +  n 
2
_
.
Using small and large constants, this gives the desired estimate (8.2.2), and the
proof is complete.
We ﬁrst prove the following general theorem.
Theorem 8.2.5. If L
m
spans the tangent space of Ω for some : ∈ N, then there
exist c 0 and C 0 such that
(8.2.3)  n 
2
≤ C
_
k
i=0
 A
i
n 
2
+  n 
2
_
, n ∈ C
∞
0
(Ω).
Here we may take c = 2
1−m
.
Proof. We shall denote an element in L
j
by 7
j
. By the hypotheses of the theorem,
we get
 n 
2
n
j=1
 1
j
n 
2
−1
+  n 
2
Zm∈1m
 7
m
n 
2
−1
+  n 
2
,
where the last summation is a ﬁnite sum and 1
j
= (∂,∂r
j
).
Therefore, to prove the theorem, it suﬃces to bound each term  7
m
n 
−1
by
the right hand side of (8.2.3) for some c 0. If : = 1, clearly we can take c = 1.
For : ≥ 2, let c ≤ 1,2 for the time being. We shall make the choice of c later. We
may also assume that 7
m
= A7
m−1
−7
m−1
A with A ∈ L
1
. Thus, we see that
(8.2.4)
 7
m
n 
2
−1
= (7
m
n, Λ
2−2
7
m
n)
= (A7
m−1
n, Λ
2−2
7
m
n) −(7
m−1
An, Λ
2−2
7
m
n).
182 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
Using Theorem 8.1.5, we have
[(A7
m−1
n, Λ
2−2
7
m
n)[
= [(7
m−1
n, Λ
2−2
7
m
An)[ +O( n  7
m−1
n 
2−1
)
≤ C( An 
2
+  7
m−1
n 
2
2−1
+  n 
2
).
Also
[(7
m−1
An, Λ
2−2
7
m
n)[ = [(An, 7
m−1
Λ
2−2
7
m
n)[ +O( n  An )
≤ C( An 
2
+  7
m−1
n 
2
2−1
+  n 
2
).
Hence, substituting the above into (8.2.4), by induction, we get
 7
m
n 
2
−1
≤ C
_
k
i=0
 A
i
n 
2
+  7
m−1
n 
2
2−1
+  n 
2
_
≤ C
_
k
i=0
( A
i
n 
2
+  A
i
n 
2
2
m−1
−1
) +  n 
2
_
.
Now, for : ≥ 2, if we take c ≤ 2
1−m
, we obtain
 7
m
n 
2
−1
≤ C
_
k
i=0
 A
i
n 
2
+  n 
2
_
.
This proves the theorem.
For our purpose, we shall modify the proof of Theorem 8.2.5 to obtain:
Theorem 8.2.6. Under the hypotheses of Theorem 8.2.3, there exist c 0 and
C 0 such that
(8.2.5)  n 
2
≤ C( 1n 
2
+  n 
2
), n ∈ C
∞
0
(Ω).
Proof. As in the proof of Theorem 8.2.5, it suﬃces to control each term
(8.2.6)  7
m
n 
2
−1
= (A7
m−1
n, Λ
2−2
7
m
n) −(7
m−1
An, Λ
2−2
7
m
n)
by the righthand side of (8.2.5) for some c 0.
Let Q
s
be some pseudodiﬀerential operator of order : of the type discussed in
Section 8.1. To estimate (8.2.6), we shall distinguish A
0
from the other vector ﬁelds
A
1
, , A
k
.
Case (i). A = A
i
with 1 ≤ i ≤ /. By Lemma 8.2.4 and the proof of Theorem
8.2.5, we get
(8.2.7)  7
m
n 
2
−1
≤ C( 1n 
2
+  7
m−1
n 
2
2−1
+  n 
2
).
8.2 Hypoellipticity of Sum of Squares of Vector Fields 183
Case (ii). A = A
0
. We ﬁrst write A
0
= −1
∗
+
k
i=1
A
2
i
+
k
i=1
c
i
(r)A
i
+p(r)
with c
i
(r), 1 ≤ i ≤ /, and p(r) belonging to C
∞
(Ω). Hence, we have
(8.2.8)
(A
0
7
m−1
n, Λ
2−2
7
m
n)
= −(1
∗
7
m−1
n, Λ
2−2
7
m
n) +
k
i=1
(A
2
i
7
m−1
n, Λ
2−2
7
m
n)
+
k
i=1
(c
i
A
i
7
m−1
n, Λ
2−2
7
m
n) + (p7
m−1
n, Λ
2−2
7
m
n).
Obviously, the last two terms in (8.2.8) are bounded by the righthand side of
(8.2.7). Since
(1
∗
7
m−1
n, Λ
2−2
7
m
n) = (7
m−1
n, 1Q
2−1
n)
= (7
m−1
n, Q
2−1
1n) +
k
i=1
(7
m−1
n, Q
2−1
A
i
n) + (7
m−1
n, Q
2−1
n),
we conclude that the ﬁrst term on the righthand side in (8.2.8) is also bounded by
the righthand side of (8.2.7). The second term on the righthand side of (8.2.8)
can be estimated as follows:
(A
2
i
7
m−1
n, Λ
2−2
7
m
n)
= −(A
i
7
m−1
n, A
i
Q
2−1
n) +O( 1n 
2
+  7
m−1
n 
2
2−1
+  n 
2
)
= −(A
i
Q
2−1
7
m−1
n, A
i
n) +O( 1n 
2
+  7
m−1
n 
2
2−1
+  n 
2
).
It follows from Lemma 8.2.4 that
(8.2.9)
[(A
2
i
7
m−1
n, Λ
2−2
7
m
n)[
≤  A
i
Q
2−1
7
m−1
n 
2
+ O( 1n 
2
+  7
m−1
n 
2
2−1
+  n 
2
)
≤ C[(1Q
2−1
7
m−1
n, Q
2−1
7
m−1
n)[
+O( 1n 
2
+  7
m−1
n 
2
2−1
+  n 
2
)
≤ C([(1n, Q
4−1
7
m−1
n)[ +
k
i=1
[(A
i
n, Q
4−1
7
m−1
n)[
+[(n, Q
4−1
7
m−1
n)[) +O( 1n 
2
+  7
m−1
n 
2
2−1
+  n 
2
)
≤ C( 1n 
2
+  7
m−1
n 
2
4−1
+  n 
2
).
This completes the estimate of the ﬁrst term on the righthand side of (8.2.6).
For the second term on the righthand side of (8.2.6), we write
A
0
= 1 −
k
i=1
A
2
i
−/(r).
184 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
Thus, we have
(8.2.10)
(7
m−1
A
0
n, Λ
2−2
7
m
n)
= (7
m−1
1n, Λ
2−2
7
m
n) −
k
i=1
(7
m−1
A
2
i
n, Λ
2−2
7
m
n)
−(7
m−1
/n, Λ
2−2
7
m
n).
The ﬁrst term on the righthand side of (8.2.10) can be written as
(7
m−1
1n, Λ
2−2
7
m
n)
= −(1n, 7
m−1
Q
2−1
n) + (1n, p
m
Q
2−1
n)
= −(1n, Q
2−1
7
m−1
n) −(1n, [7
m−1
, Q
2−1
]n) + (1n, p
m
Q
2−1
n),
for some p
m
∈ C
∞
(Ω). Also
(7
m−1
/n, Λ
2−2
7
m
n) = (/7
m−1
n, Q
2−1
n) + ([7
m−1
, /]n, Q
2−1
n).
Hence, if c ≤ 1,2, the ﬁrst and third terms of (8.2.10) can be estimated by the
righthand side of (8.2.7).
To deal with the second term on the righthand side of (8.2.10), we have
(7
m−1
A
2
i
n, Λ
2−2
7
m
n)
= (A
2
i
7
m−1
n, Q
2−1
n) + ([7
m−1
, A
2
i
]n, Q
2−1
n)
= (A
2
i
7
m−1
n, Q
2−1
n) + ([7
m−1
, A
i
]A
i
n, Q
2−1
n)
+ (A
i
[7
m−1
, A
i
]n, Q
2−1
n)
= (A
2
i
7
m−1
n, Q
2−1
n) + (
´
7
m
A
i
n, Q
2−1
n) + (A
i
´
7
m
n, Q
2−1
n).
Note that
´
7
m
is a commutator of 7
m−1
with A
i
for some 1 ≤ i ≤ /. Thus, one
may apply the proof of Case (i) and (8.2.9) to get
[(7
m−1
A
2
i
n, Λ
2−2
7
m
n)[
≤ C( 1n 
2
+  7
m−1
n 
2
4−1
+  n 
2
+ 
´
7
m
n 
2
2−1
)
≤ C( 1n 
2
+  7
m−1
n 
2
4−1
+ jo:ocn 
2
).
This completes the estimate of the second term in (8.2.6).
Consequently, by induction, we obtain
 7
m
n 
2
−1
≤ C( 1n 
2
+  7
m−1
n 
2
4−1
+  n 
2
)
≤ C( 1n 
2
+
k
i=0
 A
i
n 
2
4
m−1
−1
+  n 
2
).
Thus, if we take c ≤ 2 4
−m
, we see that
 7
m
n 
2
−1
≤ C( 1n 
2
+  A
0
n 
2
−
1
2
+  n 
2
).
8.2 Hypoellipticity of Sum of Squares of Vector Fields 185
Since
 A
0
n 
2
−
1
2
= (A
0
n, Λ
−1
A
0
n)
= (1n, Q
0
n) −
k
i=1
(A
2
i
n, Q
0
n) −(/n, Q
0
n)
≤ C( 1n 
2
+  n 
2
),
the proof of Theorem 8.2.6 is now complete.
The next result shows that estimate (8.2.5) is localizable:
Theorem 8.2.7. Let ζ, ζ
1
∈ C
∞
0
(Ω) be two realvalued cutoﬀ functions with ζ
1
≡ 1
on the support of ζ. Then there is a constant C 0 such that
(8.2.11)  ζn 
≤ C( ζ
1
1n  +  ζ
1
n ),
for all n ∈ C
∞
(Ω).
Proof. From (8.2.5) it suﬃces to estimate  [1, ζ]n  by the righthand side of
(8.2.11). Since
[1, ζ]n = 2
k
i=1
[A
i
, ζ]A
i
n +
k
i=1
[A
i
, [A
i
, ζ]]n + [A
0
, ζ]n,
we have
 [1, ζ]n 
2
≤ C
_
k
i=1
 A
i
ζ
2
1
n 
2
+  ζ
1
n 
2
_
.
Now, as in the proof of Lemma 8.2.4, we get
k
i=1
 A
i
ζ
2
1
n 
2
= −(1ζ
2
1
n, ζ
2
1
n) + Re(A
0
ζ
2
1
n, ζ
2
1
n)
+O
_
k
i=1
 A
i
ζ
2
1
n  ζ
1
n  +  ζ
1
n 
2
_
= −(ζ
2
1
1n, ζ
2
1
n) −2
k
i=1
([A
i
, ζ
2
1
]A
i
n, ζ
2
1
n) + Re(A
0
ζ
2
1
n, ζ
2
1
n)
+O
_
k
i=1
 A
i
ζ
2
1
n  ζ
1
n  +  ζ
1
n 
2
_
= −(ζ
2
1
1n, ζ
2
1
n) + 4
k
i=1
([A
i
, ζ
1
]ζ
1
n, A
i
ζ
2
1
n) + Re(A
0
ζ
2
1
n, ζ
2
1
n)
+O
_
k
i=1
 A
i
ζ
2
1
n  ζ
1
n  +  ζ
1
n 
2
_
.
186 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
Hence, using small and large constants, we obtain
k
i=1
 A
i
ζ
2
1
n 
2
≤ C( ζ
1
1n 
2
+  ζ
1
n 
2
).
This proves the theorem.
The next step is to iterate the estimate (8.2.5) to obtain the following “bootstrap”
a priori estimate:
Theorem 8.2.8. For any : ∈ 1 and : 0, there exists a constant C
s,m
such that
(8.2.12)  n 
s+
≤ C
s,m
( 1n 
s
+  n 
−m
),
for all n ∈ C
∞
0
(Ω).
Proof. Let n ∈ C
∞
0
(Ω). We wish to apply (8.2.5) to Λ
s
n. Hence, we shall assume
that estimate (8.2.5) holds for smooth functions that are supported in a ﬁxed open
neighborhood l of Ω. Let η(r) and p(r) be two smooth cutoﬀ functions supported
in l such that η ≡ 1 on the support of p and that p ≡ 1 on Ω. We claim that the
operator (1 −η)Λ
s
p acting on o, : ∈ 1, is of negative inﬁnite order, namely, it is a
smoothing operator and for any : 0 there is a constant C
s,m
such that
(8.2.13)  (1 −η)Λ
s
pn  ≤ C
s,m
 n 
−m
,
for any n ∈ o. We ﬁrst prove the claim (8.2.13). Observe ﬁrst that (1 − η)p ≡ 0.
Hence, we have
(8.2.14)
1
(2π)
n
_
R
n
(1 −η(r))c
ixξ
j(ξ)1
α
ξ
ˆ p(ξ) dξ = 0,
for any polynomial j(ξ) and any multiindex α. A direct calculation shows
(1 −η)Λ
s
pn(r) =
1
(2π)
n
_
R
n
1(r, ξ)ˆ n(ξ) dξ,
where
1(r, ξ) =
1
(2π)
n
_
R
n
(1 −η(r))c
ixζ
(1 +[ζ[
2
)
s
2
ˆ p(ζ −ξ) dζ.
Thus, to prove the claim it suﬃces to show that for any o, / ∈ 1, the kernel
(8.2.15)
´
1(r, ξ) =
_
R
n
(1 −η(r))c
ixζ
(1 +[ζ[
2
)
a
(1 +[ξ[
2
)
b
ˆ p(ζ −ξ) dζ
satisﬁes the hypotheses of Theorem B.10 in the Appendix. Using (8.2.14), Lemma
8.1.2 and Taylor’s expansion of (1 +[ζ[
2
)
a
at ξ, for any c ∈ 1, we obtain
[
´
1(r, ξ)[
_
R
n
(1 +[ξ +t(ζ −ξ)[
2
)
a−c
[ζ −ξ[
c
(1 +[ξ[
2
)
b
[ˆ p(ζ −ξ)[ dζ
_
R
n
(1 +[t(ζ −ξ)[
2
)
]a−c]
(1 +[ξ[
2
)
a−c
[ζ −ξ[
c
(1 +[ξ[
2
)
b
[ˆ p(ζ −ξ)[ dζ,
8.2 Hypoellipticity of Sum of Squares of Vector Fields 187
where 0 < t < 1. Thus, if we choose c to be large enough so that o +/ −c ≤ 0, it
is easily veriﬁed by integration by parts that the kernel
´
1(r, ξ) deﬁned in (8.2.15)
satisﬁes
[r
δ
´
1(r, ξ)[ ≤ C
a,b,δ
uniformly in r and ξ, where δ is a multiindex. It follows that all the hypotheses of
Theorem B.10 are satisﬁed by the kernel
´
1(r, ξ). This proves the claim.
Now we return to the estimate of Λ
s
n for n ∈ C
∞
0
(Ω). Since ηΛ
s
n is supported
in l, we may apply estimate (8.2.5) to get
 n 
s+
=  Λ
s
n 
≤  ηΛ
s
n 
+  (1 −η)Λ
s
pn 
≤ C( 1ηΛ
s
n  +  ηΛ
s
n  +  n 
−m
)
≤ C( [1, ηΛ
s
]n  +  1n 
s
+  n 
s
+  n 
−m
).
To handle the term  [1, ηΛ
s
]n  we write
[1, ηΛ
s
] =
k
i=1
Q
s
i
A
i
+Q
s
k+1
.
Thus,
 [1, ηΛ
s
]n  ≤ C
_
k
i=1
 A
i
n 
s
+  n 
s
_
.
Lemma 8.2.4 then shows that
 A
i
n 
2
s
=  Λ
s
A
i
n 
2
≤ C( ηΛ
s
A
i
n 
2
+  (1 −η)Λ
s
pA
i
n 
2
)
≤ C( A
i
ηΛ
s
n 
2
+  [ηΛ
s
, A
i
]n 
2
+  n 
2
−m
)
≤ C([(1ηΛ
s
n, ηΛ
s
n)[ +  n 
2
s
+  n 
2
−m
)
≤ C([([1, ηΛ
s
]n, ηΛ
s
n)[ +  1n 
2
s
+  n 
2
s
+  n 
2
−m
)
≤ C( n 
s
_
k
i=1
 A
i
n 
s
)+  1n 
2
s
+  n 
2
s
+  n 
2
−m
_
.
Using small and large constants, we obtain
 n 
s+
≤ C( 1n 
s
+  n 
s
+  n 
−m
).
Finally, observe that for any δ 0, using the interpolation inequality for Sobolev
spaces (Theorem B.2 in the Appendix), there is a constant C
s,δ,m
such that
 n 
s
≤ δ  n 
s+
+ C
s,δ,m
 n 
−m
,
for all n ∈ C
∞
0
(Ω). Therefore, letting δ be suﬃciently small, Theorem 8.2.8 is
proved.
Estimate (8.2.12) can be localized as before to get the following theorem.
188 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
Theorem 8.2.9. Let ζ, ζ
1
be two smooth realvalued cutoﬀ functions supported in
Ω with ζ
1
≡ 1 on the support of ζ. For any : ∈ 1 and : 0 there is a constant
C
s,m
such that
(8.2.16)  ζn 
s+
≤ C
s,m
( ζ
1
1n 
s
+  ζ
1
n 
−m
),
for all n ∈ C
∞
(Ω).
We are now ready to prove the main result of this section.
Proof of Theorem 8.2.3. Suppose that n is a distribution with 1n = ), where 1
given by (8.2.1) is of ﬁnite type, and that ) is smooth on Ω. We wish to show that
n is also smooth on Ω. Without loss of generality, we shall show that n is smooth
in some open neighborhood \ of the origin.
Since n is a distribution, we may assume that locally near the origin n is in \
s
for some : = −: with : 0. Let ζ and ζ
k
, / ∈ N, be a sequence of smooth
realvalued cutoﬀ functions supported in some open neighborhood of \ such that
ζ
1
n ∈ \
−m
, ζ
k
≡ 1 on supp ζ
k+1
, ζ
j
≡ 1 on supp ζ for all , and ζ ≡ 1 on \ .
Let ϕ be a smooth nonnegative realvalued function supported in the unit open
ball of 1
n
such that ϕ(r) = ϕ([r[), 0 ≤ ϕ ≤ 1 and
_
R
n
ϕdr = 1. For any δ 0, set
ϕ
δ
(r) = δ
−n
ϕ(r,δ) and
o
δ
ζ
k
n(r) = ζ
k
n ∗ ϕ
δ
(r) =
_
ζ
k
n(j)ϕ
δ
(r −j) dj.
Clearly, o
δ
ζ
k
n is a smooth function supported in some neighborhood of \ . Hence,
we obtain from (8.2.16) with : = −: that
 o
δ
ζ
k
n 
−m+
≤ C
m
( 1o
δ
ζ
k
n 
−m
+  o
δ
ζ
k
n 
−m
).
To ﬁnish the proof we need the following two key observations. For any : ∈ 1,
(1) ζ
k
n ∈ \
s
with  ζ
k
n 
s
≤ C
k
if and only if  o
δ
ζ
k
n 
s
≤ C
k
for all small
δ 0.
(2) If ζ
k
n is in \
s
, then
 [o
δ
ζ
k
, A
i
]n 
s
=  [o
δ
ζ
k
, A
i
]ζ
k−1
n 
s
≤ C
k
 ζ
k−1
n 
s
,
where the constant C
k
is independent of δ.
These two facts can be veriﬁed directly, so we omit the proofs.
Now, using (1) and (2), by commuting 1 with o
δ
ζ and applying arguments
similar to those above, we obtain
 o
δ
ζ
2
n 
−m+
≤ C
1,m
( o
δ
ζ
1
) 
−m
+  o
δ
ζ
1
n 
−m
)
with C
1,m
independent of δ. This implies ζn ∈ \
−m+
. Inductively, for any / ∈ N,
we obtain
 o
δ
ζ
k+1
n 
−m+k
≤ C
k,m
( o
δ
ζ
k
) 
−m+(k−1)
+  o
δ
ζ
k
n 
−m+(k−1)
).
8.3 Subelliptic Estimates for the Tangential CauchyRiemann Complex 189
Choosing / to be suﬃciently large, we have ζn ∈ \
s
(\ ) for all : ≥ 0. Hence,
n ∈ C
∞
(\ ). This completes the proof of Theorem 8.2.3.
8.3 Subelliptic Estimates for the Tangential CauchyRiemann Complex
Let (`, T
1,0
(`)) be a compact orientable C1 manifold of real dimension 2n−1
with n ≥ 3. Let Λ
p,q
(`), 0 ≤ j, ¡ ≤ n −1, denote the subbundle of Λ
p+q
CT
∗
(`)
such that Λ
p,q
(`) = Λ
p
T
∗1,0
(`) ⊗ Λ
q
T
∗0,1
(`). Let c
p,q
(`) be the space of
smooth sections of Λ
p,q
(`) over `. Then, we deﬁne the tangential Cauchy
Riemann operator ∂
b
as in Section 7.2, and form the tangential CauchyRiemann
complex
(8.3.1) 0 →c
p,0
(`)
∂
b
−→ c
p,1
(`)
∂
b
−→
∂
b
−→ c
p,n−1
(`) →0.
For any r
0
∈ `, let 1
1
, , 1
n−1
be a local basis for (1,0) vector ﬁelds near r
0
,
and choose a globally deﬁned vector ﬁeld T which may be assumed to be purely
imaginary. This is ﬁrst done locally, then by restriction to those coordinate trans
formations which preserve T
1,0
(`) and orientation, a local choice of sign in the
direction of T will extend T to a global one. Fix a Hermitian metric on CT(`) so
that T
1,0
(`), T
0,1
(`) and T are mutually orthogonal. We may then assume that
1
1
, , 1
n−1
, 1
1
, , 1
n−1
and T is an orthonormal basis in some neighborhood of
a reference point r
0
∈ `. Let ω
1
, , ω
n−1
be an orthonormal basis for (1,0)forms
which is dual to the basis 1
1
, , 1
n−1
.
Denote by \
s
(p,q)
(`) the Sobolev space of order :, : ∈ 1, for (j, ¡)forms on
`. Extend ∂
b
to 1
2
(p,q)
(`) = \
0
(p,q)
(`) in the sense of distribution. Thus, the
domain of ∂
b
, denoted by Dom(∂
b
), will consist of all φ ∈ 1
2
(p,q)
(`) such that
∂
b
φ ∈ 1
2
(p,q+1)
(`), and we have the complex
(8.3.2) 0 →1
2
(p,0)
(`)
∂
b
−→ 1
2
(p,1)
(`)
∂
b
−→
∂
b
−→ 1
2
(p,n−1)
(`) →0.
Therefore, ∂
b
is a linear, closed, densely deﬁned operator on the Hilbert space
1
2
(p,q)
(`).
Now one can deﬁne the adjoint operator ∂
∗
b
of ∂
b
in the standard way. A (j, ¡)
form φ is in Dom(∂
∗
b
) if there exists a (j, ¡ − 1)form p ∈ 1
2
(p,q−1)
(`) such that
(φ, ∂
b
ψ) = (p, ψ) for every (j, ¡ − 1)form ψ ∈ Dom(∂
b
). In this case we deﬁne
∂
∗
b
φ = p. Let
(8.3.3)
b
= ∂
b
∂
∗
b
+∂
∗
b
∂
b
,
be the ∂
b
Laplacian deﬁned on
Dom(
b
) = ¦α ∈ 1
2
(p,q)
(`)[ α ∈ Dom(∂
b
) ∩ Dom(∂
∗
b
),
∂
b
α ∈ Dom(∂
∗
b
) and ∂
∗
b
α ∈ Dom(∂
b
)¦.
190 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
It follows from the same arguments as in Proposition 4.2.3 that
b
is a linear, closed,
densely deﬁned selfadjoint operator from 1
2
(p,q)
(`) into itself. Deﬁne a Hermitian
form Q
b
on c
p,q
(`) by
(8.3.4) Q
b
(φ, ψ) = (∂
b
φ, ∂
b
ψ) + (∂
∗
b
φ, ∂
∗
b
ψ) + (φ, ψ) = ((
b
+1)φ, ψ),
for φ, ψ ∈ c
p,q
(`).
Locally on a coordinate neighborhood l, we can express a smooth (j, ¡)form φ
as
(8.3.5) φ =
t
]I]=p,]J]=q
φ
I,J
ω
I
∧ ω
J
,
where 1 = (i
1
, , i
p
) and J = (,
1
, , ,
q
) are multiindices, n
I
= n
i1
∧ ∧ n
ip
,
¯ n
J
= ¯ n
j1
∧ ∧ ¯ n
jq
and the prime means that we sum over only increasing multi
indices. Here φ
I,J
’s are deﬁned for arbitrary 1 and J so that they are antisymmetric.
Then, a direct computation and integration by parts yield
(8.3.6) ∂
b
φ =
t
I,J
j
1
j
(φ
I,J
) ω
j
∧ ω
I
∧ ω
K
+ terms of order zero,
and
(8.3.7) ∂
∗
b
φ = (−1)
p−1
t
I,K
j
1
j
(φ
I,jK
) ω
I
∧ ω
K
+ terms of order zero.
We shall use (8.3.6) and (8.3.7) to obtain the desired estimates. We also abbreviate
k,I,J
 1
k
φ
I,J

2
+  φ 
2
by  φ 
2
L
, and
k,I,J
 1
k
φ
I,J

2
+  φ 
2
by  φ 
2
L
.
The main eﬀort of this section is to derive the subelliptic 1,2estimate of the form
Q
b
. Before proceeding to do so, we shall digress for the moment to the regularity
theorem for the ∂ operator. Suppose now that ` is the boundary of a smooth
bounded strongly pseudoconvex domain 1 in C
n
, n ≥ 2. In Chapter 5 (Theorem
5.1.2 and Theorem 5.3.7), we prove the following subelliptic estimates
 ) 
2
1/2(D)
≤ C( ∂) 
2
+  ∂
∗
) 
2
+  ) 
2
),
for ) ∈ Dom(∂) ∩ Dom(∂
∗
), on a smooth bounded strongly pseudoconvex domain
1. The proof is based on the a priori estimate
(8.3.8)
_
bD
[)[
2
do ≤ C( ∂) 
2
+  ∂
∗
) 
2
+  ) 
2
) = CQ(), )),
for ) ∈ T
1
(p,q)
= C
1
(p,q)
(1) ∩ Dom(∂
∗
).
In fact, to prove (8.3.8) for a ﬁxed ¡, 1 ≤ ¡ ≤ n −1, one actually does not
need strong pseudoconvexity of the domain. The main ingredient is the socalled
condition 7(¡) deﬁned as follows:
8.3 Subelliptic Estimates for the Tangential CauchyRiemann Complex 191
Deﬁnition 8.3.1. Let 1 be a relatively compact subset with C
∞
boundary in a
complex Hermitian manifold of complex dimension n ≥ 2. 1 is said to satisfy
condition 7(¡), 1 ≤ ¡ ≤ n −1, if the Levi form associated with 1 has at least n−¡
positive eigenvalues or at least ¡ + 1 negative eigenvalues at every boundary point.
Obviously, condition 7(¡) is satisﬁed for all ¡ with 1 ≤ ¡ ≤ n −1 on any strongly
pseudoconvex domain.
Let r
0
∈ ` be a boundary point and let l be an open neighborhood of r
0
. For
any ) ∈ T
(p,q)
with support in l, the proof of Proposition 5.3.3 shows (with φ ≡ 0)
that
Q(), )) =
t
I,J
k
 1
k
)
I,J

2
+
t
I,K
j,k
_
bD∩U
ρ
jk
)
I,jK
)
I,kK
do
+O(( ∂)  +  ∂
∗
) )  )  +  ) 
L
 ) ).
We may assume that the Levi form is diagonal at r
0
, namely, ρ
jk
(r) = λ
j
δ
jk
+/
jk
(r)
for 1 ≤ ,, / ≤ n −1, where the λ
j
’s are the eigenvalues of the Levi form at r
0
, δ
jk
denotes the Kronecker delta and /
jk
(r
0
) = 0. It follows that
t
I,K
j,k
_
bD∩U
ρ
jk
)
I,jK
)
I,kK
do
=
t
I,J
_
k∈J
λ
k
__
bD
[)
I,J
[
2
do +δO
_
t
I,J
_
bD
[)
I,J
[
2
do
_
,
where δ 0 can be made arbitrarily small if l is chosen suﬃciently small. Integra
tion by parts also shows
 1
k
)
I,J

2
= −([1
k
, 1
k
])
I,J
, )
I,J
) +  1
k
)
I,J

2
+ O( ) 
L
 ) )
≥ −λ
k
_
bD
[)
I,J
[
2
do −δ
_
bD
[)
I,J
[
2
do +O( ) 
L
 )  +  ) 
2
).
Hence, if condition 7(¡) holds on /1, then for each ﬁxed J either there is a /
1
∈ J
with λ
k1
0 or there is a /
2
, ∈ J with λ
k2
< 0. For the former case and any c 0,
we have
Q(), )) ≥ c
t
I,J
k
 1
k
)
I,J

2
+ c
t
I,J
_
k∈J,λ
k
<0
λ
k
__
bD
[)
I,J
[
2
do
+
t
I,J
_
λ
k1
_
bD
[)
I,J
[
2
do
_
−δ
_
bD
[)
I,J
[
2
do
+O(( ∂)  +  ∂
∗
) )  )  +  ) 
L
 )  +  ) 
2
).
For the latter case, we see that
Q(), )) ≥ c
t
I,J
k
 1
k
)
I,J

2
+ c
t
I,J
_
k∈J,λ
k
<0
λ
k
__
bD
[)
I,J
[
2
do
+ (1 −c)
t
I,J
(−λ
k2
)
_
bD
[)
I,J
[
2
do −δ
_
bD
[)
I,J
[
2
do
+O(( ∂)  +  ∂
∗
) )  )  +  ) 
L
 )  +  ) 
2
).
192 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
Thus, choosing c, δ to be small enough and using small and large constants, we
obtain (8.3.8). Now, by a partition of unity argument, the next theorem follows
immediately from Theorem 5.1.2.
Theorem 8.3.2. Let 1 be a relatively compact subset with C
∞
boundary in a
complex Hermitian manifold of complex dimension n ≥ 2. Suppose that condition
7(¡) holds for some ¡, 1 ≤ ¡ ≤ n −1. Then we have
_
bD
[)[
2
do ≤ C( ∂) 
2
+  ∂
∗
) 
2
+  ) 
2
),
for ) ∈ T
(p,q)
. Furthermore, we have
(8.3.9)  ) 1
2
≤ C( ∂)  +  ∂
∗
)  +  ) ),
for ) ∈ Dom(∂) ∩ Dom(∂
∗
), where the constant C 0 is independent of ).
With Theorem 8.3.2 on hand, we now return to the subelliptic estimate for
b
on
(j, ¡)forms on `. If the C1 manifold ` is embedded as the boundary of a complex
manifold 1, topologically one can not distinguish whether ` is the boundary of
1 or ` is the boundary of the complement of 1. Thus, in order to obtain a
similar subelliptic estimate for (j, ¡)forms on `, we shall assume that condition
7(¡) holds on both 1 and its complement 1
c
as motivated by Theorem 8.3.2. Note
that condition 7(¡) on 1
c
is equivalent to condition 7(n −¡ −1) on 1. ` is said
to satisfy condition Y (¡), 1 ≤ ¡ ≤ n −1, if both conditions 7(¡) and 7(n −¡ −1)
are satisﬁed on 1.
In terms of the eigenvalues of the Levi form condition Y (¡) means that the
Levi form has at least either max(n − ¡, ¡ + 1) eigenvalues of the same sign or
min(n −¡, ¡ + 1) pairs of eigenvalues of opposite signs at every point on `. Since
condition Y (¡) will be used extensively in what follows, we now make a formal
deﬁnition for any C1 manifold.
Deﬁnition 8.3.3. Let ` be an oriented C1 manifold of real dimension 2n − 1
with n ≥ 2. ` is said to satisfy condition Y (¡), 1 ≤ ¡ ≤ n −1, if the Levi form has
at least either max(n − ¡, ¡ + 1) eigenvalues of the same sign or min(n − ¡, ¡ + 1)
pairs of eigenvalues of opposite signs at every point on `.
It follows that condition Y (¡) for 1 ≤ ¡ ≤ n −2 holds on any strongly pseudocon
vex C1 manifold ` of real dimension 2n−1 with n ≥ 3. Also, it should be pointed
out that condition Y (n − 1) is violated on any pseudoconvex C1 manifold ` of
real dimension 2n −1 with n ≥ 2. In particular, condition Y (1) is not satisﬁed on
any strongly pseudoconvex C1 manifold of real dimension three. This phenomenon
is related to the nonsolvable Lewy operator which we have discussed in Section 7.3.
Another example of a noncompact C1 manifold satisfying condition Y (¡) will be
given in Section 10.1.
Theorem 8.3.4. Suppose that condition Y (¡), for some ¡ with 1 ≤ ¡ ≤ n −1,
holds on a compact, oriented, C1 manifold (`, T
1,0
(`)) of real dimension 2n −1
with n ≥ 3. Then we have
(8.3.10)  φ 
2
1
2
Q
b
(φ, φ),
8.3 Subelliptic Estimates for the Tangential CauchyRiemann Complex 193
uniformly for all φ ∈ c
p,q
(`).
Proof. Let 1
1
, , 1
n−1
be an orthonormal basis for T
1,0
(`) locally. Since con
dition Y (¡) implies that the vector ﬁelds 1
1
, , 1
n−1
, 1
1
, , 1
n−1
and their Lie
brackets span the whole complex tangent space, using a partition of unity, the proof
is an easy consequence of Theorem 8.2.5 when : = 2, and the following theorem:
Theorem 8.3.5. Under the same hypotheses as in Theorem 8.3.4, for any r
0
∈ `,
there is a neighborhood \
x0
of r
0
such that
(8.3.11)  φ 
2
L
+  φ 
2
L
+
IJ
[1c(Tφ
I,J
, φ
I,J
)[ Q
b
(φ, φ),
uniformly for all φ ∈ c
p,q
(`) with support contained in \
x0
.
Proof. We start with (8.3.6) to obtain
 ∂
b
φ 
2
=
t
I,J
j / ∈J
 1
j
φ
I,J

2
+
t
I,J,L
j,l
c
jJ
lL
(1
j
φ
I,J
, 1
l
φ
I,L
) +O( φ 
L
 φ ),
where c
jJ
lL
= 0 unless , , ∈ J,  , ∈ 1 and ¦,¦∪J = ¦¦∪1, in which case c
jJ
lL
is the sign
of permutation (
jJ
lL
). Using the fact that φ
I,J
is antisymmetric in J, we rearrange
the terms in the above estimate to get
 ∂
b
φ 
2
=
t
I,J
j
 1
j
φ
I,J

2
−
t
I,K
j,k
(1
j
φ
I,kK
, 1
k
φ
I,jK
) +O( φ 
L
 φ ).
Using integration by parts, we have
(1
j
φ
I,kK
, 1
k
φ
I,jK
) = (−1
k
1
j
φ
I,kK
, φ
I,jK
) +O( φ 
L
 φ )
= (1
k
φ
I,kK
, 1
j
φ
I,jK
) + ([1
j
, 1
k
]φ
I,kK
, φ
I,jK
)
+O(( φ 
L
+  φ 
L
)  φ ).
Hence, from (8.3.7) we obtain
 ∂
b
φ 
2
=
t
I,J
j
 1
j
φ
I,J

2
−  ∂
∗
b
φ 
2
+
t
I,K
j,k
([1
j
, 1
k
]φ
I,jK
, φ
I,kK
)
+O(( φ 
L
+  φ 
L
)  φ ).
To handle the commutator term, we may assume that the Levi form is diagonal
at r
0
and that c
11
(r
0
) ,= 0, since condition Y (¡) holds. It follows that [c
11
(r)[
1,C 0 for r ∈ \
x0
if \
x0
is chosen to be small enough. For any smooth function
) with )(r
0
) = 0 on `, we have
[Re(Tφ
I,J
, )φ
I,L
)[ ≤
¸
¸
¸
¸
Re
_
1
c
11
[1
1
, 1
1
]φ
I,J
, )φ
I,L
_¸
¸
¸
¸
+O( φ 
L
 φ )
≤ C(sup
Vx
0
[)[)( φ 
2
L
+  φ 
2
L
) +O( φ 
L
 φ ).
194 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
Thus, if we denote the eigenvalues of the Levi form at r
0
by λ
1
, , λ
n−1
, we obtain
(8.3.12)
Q
b
(φ, φ) =
t
I,J
j
 1
j
φ
I,J

2
+
t
I,J
j∈J
λ
j
Re(Tφ
I,J
, φ
I,J
)
+δO( φ 
2
L
+  φ 
2
L
) +O( φ 
L
 φ ),
where δ 0 can be made arbitrarily small, if necessary, by shrinking \
x0
.
To ﬁnish the proof, we need to control the second term on the righthand side of
(8.3.12). This will be done by using a fraction of the ﬁrst term on the righthand
side of (8.3.12). Thus, we ﬁrst use integration by parts to get
 1
j
φ
I,J

2
=  1
j
φ
I,J

2
−λ
j
Re(Tφ
I,J
, φ
I,J
)
+δO( φ 
2
L
+  φ 
2
L
) +O(( φ 
L
+  φ 
L
)  φ ).
Next, for each multiindex pair (1, J), set
σ(1, J) = ¦,[ λ
j
< 0 if Re(Tφ
I,J
, φ
I,J
) 0 or λ
j
< 0 if Re(Tφ
I,J
, φ
I,J
) < 0¦.
It follows that, for any small c 0, we have
 φ 
2
L
≥ c  φ 
2
L
+ (1 −c)
t
I,J
j∈σ(I,J)
 1
j
φ
I,J

2
≥ c  φ 
2
L
− (1 −c)
t
I,J
j∈σ(I,J)
λ
j
Re(Tφ
I,J
, φ
I,J
)
−δ( φ 
2
L
+  φ 
2
L
) −C  φ 
2
.
Substituting the above into (8.3.12) we obtain
(8.3.13)
Q
b
(φ, φ) ≥ c  φ 
2
L
− (1 −c)
t
I,J
j∈σ(I,J)
λ
j
Re(Tφ
I,J
, φ
I,J
)
+
t
I,J
j∈J
λ
j
Re(Tφ
I,J
, φ
I,J
) −δ( φ 
2
L
+  φ 
2
L
)
−O( φ 
L
 φ )
= c  φ 
2
L
+
t
I,J
o
I,J
Re(Tφ
I,J
, φ
I,J
) −δ( φ 
2
L
+  φ 
2
L
)
−O( φ 
L
 φ ),
where
o
I,J
=
j∈J\σ(I,J)
λ
j
−(1 −c)
_
_
j∈σ(I,J)\J
λ
j
_
_
+c
_
_
j∈J∩σ(I,J)
λ
j
_
_
.
Since condition Y (¡) holds at r
0
, for each multiindex J with [J[ = ¡, one of the
following three cases must hold:
(1) If the Levi form has max(n − ¡, ¡ + 1) eigenvalues of the same sign, then
there exists a , ∈ J and a / , ∈ J so that λ
j
and λ
k
are of the same sign
which may be assumed to be positive, if necessary, by replacing T by −T.
8.4 Local Regularity and the Hodge Theorem for
b
195
(2) If the Levi form has min(n −¡, ¡ +1) pairs of eigenvalues of opposite signs,
then there are ,, / , ∈ J so that λ
j
0 and λ
k
< 0.
(3) Under the same hypothesis as in (2), there are ,, / ∈ J so that λ
j
0 and
λ
k
< 0.
Then it is not hard to verify that by choosing c 0 to be small enough we can
achieve o
I,J
0 if Re(Tφ
I,J
, φ
I,J
) 0, and o
I,J
< 0 if Re(Tφ
I,J
, φ
I,J
) < 0. Since
the second term on the righthand side of (8.3.13) is a ﬁnite sum, by letting δ 0
be suﬃciently small, we get
Q
b
(φ, φ)  φ 
2
L
+
I,J
[Re(Tφ
I,J
, φ
I,J
)[ −(:c)  φ 
2
L
−(c)  φ 
2
.
Since
 1
j
φ
I,J

2
 1
j
φ
I,J

2
+ [Re(Tφ
I,J
, φ
I,J
)[
+δ
1
O( φ 
2
L
+  φ 
2
L
) +O( φ 
L
 φ ),
where δ
1
0 can be made arbitrarily small, by choosing δ
1
and (:c) to be suﬃciently
small, we obtain
 φ 
2
L
+  φ 
2
L
+
I,J
[Re(Tφ
I,J
, φ
I,J
)[ Q
b
(φ, φ).
This completes the proof of the theorem.
Corollary 8.3.6. Q
b
is compact with respect to 1
2
(p,q)
(`).
Proof. Using Friedrichs’ lemma (see Appendix D) and Theorem 8.3.4, we obtain
Q
b
(φ, φ) ≥ C  φ 
2
1
2
, for φ ∈ Dom(∂
b
) ∩ Dom(∂
∗
b
).
In particular, Q
b
is compact with respect to 1
2
(p,q)
(`).
It is easy to see that (
b
+1)
−1
is injective on 1
2
(p,q)
(`). Corollary 8.3.6 implies
that (
b
+1)
−1
is compact using Rellich’s lemma. We will discuss this in detail in
the next section.
The a priori estimate obtained in Theorem 8.3.4 is the main ingredient for han
dling the local regularity problem of the operator
b
on compact strongly pseudo
convex C1 manifolds of real dimension 2n − 1 with n ≥ 3. It is Estimate (8.3.10)
that enables us to deduce the existence and regularity theorems for the ∂
b
complex.
8.4 Local Regularity and the Hodge Theorem for
b
The main task of this section is to prove the local regularity theorem for the
operator
b
and its related consequences. This will be done by ﬁrst proving a
priori estimates for the operator
b
+1 on a local coordinate neighborhood l.
196 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
Lemma 8.4.1. Under the same hypotheses as in Theorem 8.3.4, let l be a local
coordinate neighborhood, and let ¦ζ
k
¦
∞
k=1
be a sequence of real smooth functions
supported in l such that ζ
k
= 1 on the support of ζ
k+1
for all /. Then, for each
positive integer /, we have
 ζ
k
φ 
2
k
2
 ζ
1
(
b
+1)φ 
2
k−2
2
+  (
b
+1)φ 
2
uniformly for all φ ∈ c
p,q
(`) supported in l.
Proof. The lemma will be proved by induction. For / = 1, by Theorem 8.3.4, we
have
 ζ
1
φ 
2
1
2
Q
b
(ζ
1
φ, ζ
1
φ) =  ∂
b
ζ
1
φ 
2
+  ∂
∗
b
ζ
1
φ 
2
+  ζ
1
φ 
2
.
We estimate the righthand side as follows:
 ∂
b
ζ
1
φ 
2
= (∂
b
ζ
1
φ, ∂
b
ζ
1
φ)
= (ζ
1
∂
b
φ, ∂
b
ζ
1
φ) + ([∂
b
, ζ
1
]φ, ∂
b
ζ
1
φ)
= (∂
b
φ, ∂
b
ζ
2
1
φ) + (∂
b
φ, [ζ
1
, ∂
b
]ζ
1
φ) + ([∂
b
, ζ
1
]φ, ∂
b
ζ
1
φ)
= (∂
∗
b
∂
b
φ, ζ
2
1
φ) + (∂
b
ζ
1
φ, [ζ
1
, ∂
b
]φ)
+ ([ζ
1
, ∂
b
]φ, [ζ
1
, ∂
b
]φ) + ([∂
b
, ζ
1
]φ, ∂
b
ζ
1
φ).
Note that
Re
_
(∂
b
ζ
1
φ, [ζ
1
, ∂
b
]φ) + ([∂
b
, ζ
1
]φ, ∂
b
ζ
1
φ)
_
= 0.
A similar argument holds for  ∂
∗
b
ζ
1
φ 
2
. Thus, we get
 ζ
1
φ 
2
1
2
Q
b
(ζ
1
φ, ζ
1
φ)
Re((
b
+1)φ, ζ
2
1
φ) +O( φ 
2
)
 (
b
+1)φ  φ  +O( φ 
2
)
 (
b
+1)φ 
2
,
since  φ   (
b
+1)φ . This establishes the initial step.
Let us assume that the assertion is true for all integers up to / − 1 for some
/ 1, then we prove it for /. For simplicity we write the standard pseudodiﬀerential
operator Λ
k−1
2
= ¹
k
for short, and denote ζ
1
¹
k
ζ
k
by 1
k
. Then, we have
 ζ
k
φ 
2
k
2
 ¹
k
ζ
k
φ 
2
1
2
=  ¹
k
ζ
1
ζ
k
φ 
2
1
2
 ζ
1
¹
k
ζ
k
φ 
2
1
2
+  [¹
k
, ζ
1
]ζ
k
ζ
k−1
φ 
2
1
2
Q
b
(1
k
φ, 1
k
φ) +  ζ
k−1
φ 
2
k−2
2
.
Let 1
∗
k
be the adjoint operator of 1
k
, then the ﬁrst term can be estimated as
8.4 Local Regularity and the Hodge Theorem for
b
197
follows:
Q
b
(1
k
φ, 1
k
φ)
=  ∂
b
1
k
ζ
k−1
φ 
2
+  ∂
∗
b
1
k
ζ
k−1
φ 
2
+  1
k
ζ
k−1
φ 
2
= (∂
b
ζ
k−1
φ, 1
∗
k
∂
b
1
k
ζ
k−1
φ) + (∂
∗
b
ζ
k−1
φ, 1
∗
k
∂
∗
b
1
k
ζ
k−1
φ) + (ζ
k−1
φ, 1
∗
k
1
k
ζ
k−1
φ)
+O( ζ
k−1
φ k−1
2
( ∂
b
1
k
ζ
k−1
φ  +  ∂
∗
b
1
k
ζ
k−1
φ ))
= (∂
b
ζ
k−1
φ, ∂
b
1
∗
k
1
k
ζ
k−1
φ) + (∂
∗
b
ζ
k−1
φ, ∂
∗
b
1
∗
k
1
k
ζ
k−1
φ) + (ζ
k−1
φ, 1
∗
k
1
k
ζ
k−1
φ)
+O( ζ
k−1
φ 
2
k−1
2
+  ζ
k−1
φ k−1
2
( ∂
b
1
k
ζ
k−1
φ  +  ∂
∗
b
1
k
ζ
k−1
φ ))
= (∂
b
φ, ∂
b
1
∗
k
1
k
ζ
k−1
φ) + (∂
∗
b
φ, ∂
∗
b
1
∗
k
1
k
ζ
k−1
φ) + (φ, 1
∗
k
1
k
ζ
k−1
φ) +O( )
= ((
b
+1)φ, 1
∗
k
1
k
φ) +O( )
= (1
k
ζ
1
(
b
+1)φ, 1
k
φ) +O( )
 1
k
ζ
1
(
b
+1)φ 
−
1
2
 1
k
φ 1
2
+ O( )
(c)  ζ
1
(
b
+1)φ 
2
k−2
2
+ (:c)  ζ
k
φ 
2
k
2
+ (c)  ζ
k−1
φ 
2
k−1
2
+ (:c)( ∂
b
1
k
ζ
k−1
φ 
2
+  ∂
∗
b
1
k
ζ
k−1
φ 
2
).
Hence, by choosing (:c) small enough and using the induction hypothesis, we
obtain
 ζ
k
φ 
2
k
2
 ζ
1
(
b
+1)φ 
2
k−2
2
+  ζ
k−1
φ 
2
k−1
2
 ζ
1
(
b
+1)φ 
2
k−2
2
+  ζ
1
(
b
+1)φ 
2
k−3
2
+  (
b
+1)φ 
2
 ζ
1
(
b
+1)φ 
2
k−2
2
+  (
b
+1)φ 
2
.
This completes the proof of the lemma.
Theorem 8.4.2. Under the same hypotheses as in Theorem 8.3.4, given α ∈
1
2
(p,q)
(`), let φ ∈ Dom(
b
) be the unique solution of (
b
+ 1)φ = α. If l is
a subregion of ` and α[
U
∈ c
p,q
(l), then φ[
U
∈ c
p,q
(l). Moreover, if ζ and ζ
1
are two cutoﬀ functions supported in l such that ζ
1
= 1 on the support of ζ, then
for each integer : ≥ 0, there is a constant C
s
such that
(8.4.1)  ζφ 
2
s+1
≤ C
s
( ζ
1
α 
2
s
+  α 
2
).
Proof. If φ[
U
is smooth, then the estimate (8.4.1) follows immediately from Lemma
8.4.1. Therefore, it remains only to show that φ[
U
∈ c
p,q
(l). Since the Hermitian
form Q
b
is not elliptic, namely, G˚arding’s type inequality does not hold for Q
b
, we
shall apply the technique of elliptic regularization to Q
b
.
The elliptic regularization method has been used in the proof of Theorem 5.2.1
to deduce the regularity of the ∂Neumann operator on strongly pseudoconvex
domains. Thus, we shall only sketch the idea here and omit the details. Let
¦(l
i
, ϕ
i
)¦
m
i=1
be an open covering of ` formed out of the local coordinate neigh
borhood systems with local coordinates r
j
’s on l
i
, and let ¦η
i
¦
m
i=1
be a partition of
198 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
unity subordinate to ¦l
i
¦
m
i=1
. Deﬁne the form Q
b
, for each c with 0 < c << 1, by
Q
b
(φ, ψ) = Q
b
(φ, ψ) +c
m
i=1
2n−1
j=1
(1
j
η
i
φ, 1
j
η
i
ψ)
for all φ, ψ ∈ c
p,q
(`), where 1
j
= ∂,∂r
j
. Denote by
´
\
(p,q)
(`) the completion of
c
p,q
(`) under Q
b
.
Let φ
∈
´
\
(p,q)
(`) be the unique solution to the equation
Q
b
(φ
, ψ) = (α, ψ), for ψ ∈
´
\
(p,q)
(`).
Then, we have
(8.4.2)  ζφ

2
s+1
 ζ
1
α 
2
s
+  α 
2
and the estimate is uniform for all c with 0 < c << 1. Also, as in the proof of
Theorem 5.2.1, ¦φ
¦ converges to φ in 1
2
(p,q)
(`).
The sequence ¦ζφ
¦, by (8.4.2), is uniformly bounded in \
s+1
(p,q)
(`) for each :.
Hence, by Rellich’s lemma we can extract a subsequence ¦ζφ
j
¦ that converges in
\
s
(p,q)
(`) as c
j
→0. Since ¦φ
¦ converges to φ in 1
2
(p,q)
(`), ¦ζφ
j
¦ must converge
to ζφ in \
s
(p,q)
(`) for each :. Finally, by invoking the Sobolev embedding theorem,
we have ζφ ∈ c
p,q
(`). This completes the proof of the theorem.
A few consequences now follow immediately from Theorem 8.4.2.
Theorem 8.4.3. Let α, φ, l, ζ and ζ
1
be as in Theorem 8.4.2. If α[
U
∈ \
s
(p,q)
(l)
for some : ≥ 0, then ζφ ∈ \
s+1
(p,q)
(`) and
 ζφ 
2
s+1
 ζ
1
α 
2
s
+  α 
2
.
Proof. Let ζ
0
be a cutoﬀ function supported in l such that ζ
0
= 1 on the support
of ζ
1
. Choose sequences of smooth (j, ¡)forms ¦β
n
¦ and ¦γ
n
¦ with supp¦β
n
¦ ⊂
supp¦ζ
0
¦ and supp¦γ
n
¦ ⊂ supp¦(1 − ζ
0
)¦ such that β
n
→ ζ
0
α in \
s
(p,q)
(`) and
γ
n
→(1 −ζ
0
)α in 1
2
(p,q)
(`).
Hence, α
n
= β
n
+ γ
n
→ α in 1
2
(p,q)
(`) and ζ
1
α
n
→ ζ
1
α in \
s
(p,q)
(`). Let
φ
n
∈ Dom(
b
) be the solution to (
b
+1)φ
n
= α
n
, so φ
n
→φ in 1
2
(p,q)
(`). Then,
Theorem 8.4.2 shows
 ζ(φ
n
−φ
m
) 
s+1
 ζ
1
(α
n
−α
m
) 
s
+  α
n
−α
m
 .
It follows that ζφ
n
is Cauchy in \
s+1
(p,q)
(`), and lim
n→∞
ζφ
n
= ζφ is in \
s+1
(p,q)
(`).
Hence, we have
 ζφ 
s+1
 ζ
1
α 
s
+  α  .
This proves the theorem.
8.4 Local Regularity and the Hodge Theorem for
b
199
Theorem 8.4.4. Let α, φ, l, ζ and ζ
1
be as in Theorem 8.4.2. If ζ
1
α ∈ \
s
(p,q)
(`)
for some : ≥ 0, and if φ satisﬁes (
b
+ λ1)φ = α for some constant λ, then
ζφ ∈ \
s+1
(p,q)
(`). In other words,
b
+λ1 is hypoelliptic for every λ. Moreover, all
the eigenforms of
b
are smooth.
Proof. Let α
t
= α+(1 −λ)φ, then (
b
+1)φ = α
t
. The assertion now follows from
Theorem 8.4.3 and an induction argument. This proves the theorem.
If we patch up the local estimates, we obtain the following global estimate:
Theorem 8.4.5. Let ` be a compact, oriented, C1 manifold satisfying condition
Y (¡). Let φ ∈ Dom(
b
). If (
b
+ 1)φ = α with α ∈ \
s
(p,q)
(`), : ≥ 0, then
φ ∈ \
s+1
(p,q)
(`) and
 φ 
s+1
≤ C  α 
s
,
where the constant C is independent of α.
Here are some important consequences:
Corollary 8.4.6. Let ` be as in Theorem 8.4.5. The operator (
b
+ 1)
−1
is
compact.
Proof. Since (
b
+1)
−1
is a bounded operator from 1
2
(p,q)
(`) into \
1
(p,q)
(`), the
assertion follows from Rellich’s lemma (see Theorem A.8 in the Appendix).
Corollary 8.4.7. Let ` be as in Theorem 8.4.5. The operator
b
+ 1 has a
discrete spectrum with no ﬁnite limit point, and each eigenvalue occurs with ﬁnite
multiplicity. All eigenfunctions are smooth. In particular, Ker(
b
) is of ﬁnite
dimension and consists of smooth forms.
Proof. By Corollary 8.4.6 (
b
+ 1)
−1
is a compact operator from 1
2
(p,q)
(`) into
itself. Hence the spectrum of (
b
+1)
−1
is compact and countable with zero as its
only possible limit point. Since (
b
+ 1)
−1
is injective, zero is not an eigenvalue
of (
b
+ 1)
−1
. Each eigenvalue of (
b
+ 1)
−1
has ﬁnite multiplicity. Also λ is an
eigenvalue of
b
+ 1 if and only if λ
−1
is an eigenvalue of (
b
+ 1)
−1
. This proves
the corollary.
Proposition 8.4.8. Let ` be as in Theorem 8.4.5.
b
is hypoelliptic. Moreover,
if
b
φ = α with α ∈ \
s
(p,q)
(`), : ≥ 0, we have
 φ 
2
s+1
≤ C( α 
2
s
+  φ 
2
),
where the constant C 0 is independent of α.
Proof. We show the estimate by an induction on :. If : = 0, Theorem 8.4.5 implies
 φ 
2
1
 (
b
+1)φ 
2
 α 
2
+  φ 
2
.
In general, if we assume the assertion holds up to step :−1, we have φ ∈ \
s
(p,q)
(`).
For the case :, we apply Theorem 8.4.5 again and get
 φ 
2
s+1
 (
b
+1)φ 
2
s

b
φ 
2
s
+  φ 
2
s
 α 
2
s
+  φ 
2
,
200 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
where the ﬁnal step is accomplished by the induction hypothesis. This proves the
proposition.
Let H
b
(p,q)
(`) denote the space of harmonic forms on `, i.e., H
b
(p,q)
(`) =
Ker(
b
). Thus H
b
(p,q)
(`) consists of smooth harmonic (j, ¡)forms and is of ﬁnite
dimension. Using Corollary 8.4.7,
b
is bounded away from zero on the orthogonal
complement (H
b
(p,q)
(`))
⊥
, namely,
(8.4.3) 
b
φ  ≥ λ
1
 φ 
for all φ ∈ Dom(
b
) ∩ (H
b
(p,q)
(`))
⊥
, where λ
1
is the smallest positive eigenvalue
of
b
. It follows from Estimate (8.4.3) and Lemma 4.1.1 that the range of
b
is
closed. Also, the following strong Hodge type decomposition holds on 1
2
(p,q)
(`):
Proposition 8.4.9. Let ` be as in Theorem 8.4.5. 1
2
(p,q)
(`) admits the strong
orthogonal decomposition,
1
2
(p,q)
(`) = ¹(
b
) ⊕H
b
(p,q)
(`)
= ∂
b
∂
∗
b
(Dom
b
) ⊕∂
∗
b
∂
b
(Dom
b
) ⊕H
b
(p,q)
(`),
where ¹(
b
) denotes the range of
b
.
Proof. Since ¹(
b
)= (H
b
(p,q)
(`))
⊥
and ¹(∂
b
∂
∗
b
) ⊥ ¹(∂
∗
b
∂
b
), the decomposition
follows.
We can thus deﬁne the boundary operator, ·
b
: 1
2
(p,q)
(`) → Dom(
b
), as
follows: If α ∈ H
b
(p,q)
(`), set ·
b
α = 0. If α ∈ ¹(
b
), deﬁne ·
b
α = φ, where φ is
the unique solution of
b
φ = α with φ ⊥ H
b
(p,q)
(`), and we extend ·
b
by linearity.
It is easily seen that ·
b
is a bounded operator.
Let H
b
(p,q)
denote the orthogonal projection from 1
2
(p,q)
(`) onto H
b
(p,q)
(`).
Next, we prove the main result of this section.
Theorem 8.4.10. Suppose that condition Y (¡), for some ¡ with 1 ≤ ¡ ≤ n −1,
holds on a compact, oriented, C1 manifold (`, T
1,0
(`)) with n ≥ 3. Then there
exists an operator
·
b
: 1
2
(p,q)
(`) →1
2
(p,q)
(`)
such that:
(1) ·
b
is a compact operator,
(2) for any α ∈ 1
2
(p,q)
(`), α = ∂
b
∂
∗
b
·
b
α +∂
∗
b
∂
b
·
b
α +H
b
α,
(3) ·
b
H
b
= H
b
·
b
= 0.
·
b
b
=
b
·
b
= 1 −H
b
on Dom(
b
).
(4) If ·
b
is also deﬁned on 1
2
(p,q+1)
(`), then ·
b
∂
b
= ∂
b
·
b
on Dom(∂
b
).
If ·
b
is also deﬁned on 1
2
(p,q−1)
(`), then ·
b
∂
∗
b
= ∂
∗
b
·
b
on Dom(∂
∗
b
).
(5) ·
b
(c
p,q
(`)) ⊂ c
p,q
(`), and for each positive integer :, the estimate
(8.4.4)  ·
b
α 
s+1
 α 
s
holds uniformly for all α ∈ \
s
(p,q)
(`).
8.4 Local Regularity and the Hodge Theorem for
b
201
Proof. (1) follows from Proposition 8.4.8 and the Rellich lemma. (2) is just a
restatement of Proposition 8.4.9. The assertions in (3) follow immediately from the
deﬁnition of ·
b
. For (4), if α ∈ Dom(∂
b
), then
·
b
∂
b
α = ·
b
∂
b
∂
∗
b
∂
b
·
b
α
= ·
b
b
∂
b
·
b
α
= ∂
b
·
b
α.
A similar equation holds for ∂
∗
b
. For (5), if α ∈ c
p,q
(`), then α −H
b
α ∈ c
p,q
(`),
we have
b
·
b
α = α −H
b
α.
Since
b
is hypoelliptic by Theorem 8.4.4, ·
b
α ∈ c
p,q
(`). Estimate (8.4.4) now
follows from Proposition 8.4.8 since
 ·
b
α 
s+1

b
·
b
α 
s
+  ·
b
α 
 α 
s
+  H
b
α 
s
+  α 
 α 
s
.
Here we have used the fact that H
b
(p,q)
(`) is of ﬁnite dimension to conclude the
estimate:  H
b
α 
s
≤ C
s
 H
b
α  ≤ C
s
 α  for some constant C
s
. This proves
the theorem.
Corollary 8.4.11. Let ` be as in Theorem 8.4.10. The range of ∂
b
on Dom(∂
b
)∩
1
2
(p,q−1)
(`) is closed.
Proof. Since ¹(∂
b
) ⊥ Ker(∂
∗
b
), we have ¹(∂
b
) = ∂
b
∂
∗
b
(Dom
b
).
Deﬁnition 8.4.12. Let ` be a compact orientable C1 manifold. The Szeg¨o pro
jection o on ` is deﬁned to be the orthogonal projection o = H
b
(0,0)
from 1
2
(`)
onto H
b
(`) = H
b
(0,0)
(`).
If condition Y (1) holds on `, according to Theorem 8.4.10, there exists an
operator ·
b
on 1
2
(0,1)
(`). Then it is easy to obtain the following formula for the
Szeg¨o projection.
Theorem 8.4.13. Let ` be a compact orientable C1 manifold. Suppose that `
satisﬁes condition Y (1). Then the Szeg¨o projection o on ` is given by
o = 1 −∂
∗
b
·
b
∂
b
.
Theorem 8.4.10 gives the following solvability and regularity theorem for the ∂
b
equation.
202 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
Theorem 8.4.14. Under the same hypotheses as in Theorem 8.4.10, for any α ∈
1
2
(p,q)
(`) with ∂
b
α = 0 and H
b
α = 0, there is a unique solution φ of ∂
b
φ = α with
φ ⊥ Ker(∂
b
). If α ∈ c
p,q
(`), then φ ∈ c
p,q−1
(`). Furthermore, for each : ≥ 0, if
α ∈ \
s
(p,q)
(`), then φ ∈ \
s+
1
2
(p,q−1)
(`) and
(8.4.5)  φ 
s+
1
2
 α 
s
.
Proof. By (2) of Theorem 8.4.10, clearly we have α = ∂
b
∂
∗
b
·
b
α. We simply take
φ = ∂
∗
b
·
b
α, and φ is unique by the condition φ ⊥ Ker(∂
b
). The smoothness of φ
follows from (5) of Theorem 8.4.10. For the estimate (8.4.5), let ¦l
i
¦
m
i=1
be an open
cover of ` formed by the coordinate charts ¦l
i
¦
m
i=1
, and let ¦ζ
i
¦
m
i=1
be a partition
of unity subordinate to ¦l
i
¦
m
i=1
. Then we have
 φ 
2
s+
1
2
=  ∂
∗
b
·
b
α 
2
s+
1
2
≤  ∂
∗
b
·
b
α 
2
s+
1
2
+  ∂
b
·
b
α 
2
s+
1
2
m
i=1
((∂
∗
b
ζ
i
·
b
α, Λ
2s+1
∂
∗
b
ζ
i
·
b
α) + (∂
b
ζ
i
·
b
α, Λ
2s+1
∂
b
ζ
i
·
b
α))
=
m
i=1
((∂
∗
b
·
b
α, ζ
i
Λ
2s+1
∂
∗
b
ζ
i
·
b
α) + (∂
b
·
b
α, ζ
i
Λ
2s+1
∂
b
ζ
i
·
b
α)) +O( ·
b
α 
2
s+1
)
=
m
i=1
((∂
∗
b
·
b
α, ∂
∗
b
ζ
i
Λ
2s+1
ζ
i
·
b
α) + (∂
b
·
b
α, ∂
b
ζ
i
Λ
2s+1
ζ
i
·
b
α)) +O( ·
b
α 
2
s+1
)
=
m
i=1
(α, ζ
i
Λ
2s+1
ζ
i
·
b
α) +O( ·
b
α 
2
s+1
)
 α 
s
 ·
b
α 
s+1
+  ·
b
α 
2
s+1
.
By (5) of Theorem 8.4.10, we thus obtain
 φ 
s+
1
2
 α 
s
.
This completes the proof of Theorem 8.4.14.
A ﬁnal remark is in order. If 1 is a relatively compact complex manifold with
boundary /1 satisfying condition 7(¡), 1 ≤ ¡ ≤ n, analogous results to Theorems
8.4.10 and 8.4.14 can be obtained for the ∂Neumann operator. In particular,
=
¯
∂
¯
∂
∗
+
¯
∂
∗ ¯
∂ is hypoelliptic on 1.
NOTES
Pseudodiﬀerential operators were introduced by J. J. Kohn and L. Nirenberg
[KoNi 2] and L. H¨ormander [H¨or 4] as a generalization of singular integral operators
developed by A. P. Calder´on and A. Zygmund [CaZy 1]. These operators have
played an important role in the study of linear partial diﬀerential equations. We
Notes 203
refer the reader to the books by H¨ormander [H¨or 8], NagelStein [NaSt 1] and Treves
[Tre 3] for detailed discussions and applications of pseudodiﬀerential operators. The
pseudodiﬀerential operators used in section 8.1 are of the simplest kind.
Theorem 8.2.3 and Theorem 8.2.6 were ﬁrst proved by L. H¨ormander in [H¨or
6] where the argument is based on careful analysis of the oneparameter groups
generated by the given vector ﬁelds. H¨ormander’s original proof gives very precise
c in Theorem 8.2.6. He also showed that ﬁnite type condition is necessary for
subellipticity of the sum of squares operator. L. P. Rothschild and E. M. Stein
[RoSt 1] showed that sharp estimates in 1
p
spaces and Lipschitz spaces can be
achieved. The proof of Theorem 8.2.3 using pseudodiﬀerential operators that we
present in Section 8.2 follows the paper by J. J. Kohn [Koh 3] (see also Oleinik
Radkeviˇc [OlRa 1]).
However, it is well known that the ﬁnite type condition is not necessary for the
hypoellipticity of a sum of squares operator. In [OlRa 1], Theorem 2.5.3, Oleinik and
Radkeviˇc showed hypoellipticity when the ﬁnite type condition fails on a compact
set which is contained in a ﬁnite union of hypersurfaces. In [KuSt 1] Kusuoka and
Strook showed, by using probabilistic methods, that the operator
1
a
=
_
∂
∂t
_
2
+
_
∂
∂r
_
2
+
_
o(t)
∂
∂j
_
2
is hypoelliptic in 1
3
if and only if lim
t¸0
tlogo(t) = 0, where o ∈ C
∞
(1) is even,
realvalued with derivatives of all orders bounded, nondecreasing on [0, ∞), and
vanishing to inﬁnite order at t = 0. For instance, such an o is given by o(t) =
c
−(1/]t]
p
)
, 0 < j < 1. The new interesting phenomenon in this example is that
hypoellipticity depends on the exponential order of vanishing of o at zero. V. S.
Fedii [Fed 1] has shown that the operator
1
b
=
_
∂
∂t
_
2
+
_
/(t)
∂
∂r
_
2
is hypoelliptic in 1
2
for any realvalued / ∈ C
∞
(1) with /(t) ,= 0, for t ,= 0. Unlike
the previous example, / may vanish at 0 to any exponential order and still 1
b
is
hypoelliptic (see also recent related results in [Koh 12]).
It follows from the work of C. Feﬀerman and D. H. Phong [FePh 1] that both
operators 1
a
and 1
b
, when o and / are vanishing to inﬁnite order at t = 0, do not
satisfy local subelliptic estimates near t = 0. The proof of their hypoellipticity does
not use such estimates in contrast to the classical proof of H¨ormander’s Theorem
which is based on local subelliptic estimates. The formulation of a necessary and
suﬃcient condition for the hypoellipticity of a sum of squares operator is an open
problem.
The materials presented in Sections 8.3 and 8.4 are developed by J. J. Kohn
[Koh 2] where condition Y (¡) was ﬁrst introduced. See also [FoKo 1]. Theorem
8.3.2 shows that condition 7(¡) is suﬃcient for the subelliptic 1,2estimate for the
∂Neumann operator on (j, ¡)forms. The characterization of when the Morrey
Kohn estimate, (8.3.8), holds was proved by L. H¨ormander in [H¨or 2]. This lead to
a new proof of existence results for ∂ by A. Andreotti and H. Grauert [AnGr 1] where
204 Subelliptic Estimates for Second Order Diﬀerential Equations and
b
condition 7(¡) was ﬁrst introduced. The techniques from subelliptic cestimate to
regularity of the solution were treated along the lines of Kohn and Nirenberg [KoNi
1].
There is a vast amount of works concerning the ∂
b
complex or
b
on strongly
pseudoconvex C1 manifolds (or C1 manifolds satisfying condition Y (¡)). We refer
the reader to the papers by FollandStein [FoSt 1], RothschildStein [RoSt 1], Beals
GreinerStanton [BGS 1] and Boutet de MonvelSj¨ostrand [BdSj 1] as well as the
books by BealsGreiner [BeGr 1], Stein [Ste 4] and Treves [Tre 3].
205
206
CHAPTER 9
THE TANGENTIAL
CAUCHYRIEMANN COMPLEX
ON PSEUDOCONVEX C1 MANIFOLDS
We study existence theorems for the tangential CauchyRiemann complex on
a smooth pseudoconvex C1 manifold ` in this chapter. When ` is a strongly
pseudoconvex C1 manifold, 1
2
existence theorems and subelliptic estimates for
¯
∂
b
and
b
have been proved in Chapter 8 using pseudodiﬀerential operators. We shall
establish here the existence theorems for
¯
∂
b
in the C
∞
and 1
2
categories when ` is
the boundary of a smooth bounded pseudoconvex domain Ω in C
n
. One purpose of
this chapter is to study the relationship between
¯
∂ and
¯
∂
b
. To solve
¯
∂
b
, we construct
¯
∂closed extensions of forms from ` to Ω and use the solvability for
¯
∂ in Ω. The
extension problem can be converted into a
¯
∂Cauchy problem, which is to solve
¯
∂
with prescribed support.
The Cauchy problem for a bounded pseudoconvex domain in C
n
is formulated
and solved in the 1
2
sense in Section 9.1. In Section 9.2 we discuss C
∞
extensions
of smooth forms from the boundary and obtain the C
∞
solvability for
¯
∂
b
on pseu
doconvex boundaries. 1
2
existence theorems for
¯
∂
b
and estimates in Sobolev spaces
are proved in Section 9.3. The closedrange property of
b
and the strong Hodge
decomposition theorem for
¯
∂
b
are proved in Section 9.4.
9.1 The 1
2
Cauchy Problem for
¯
∂
Let Ω be a bounded pseudoconvex domain in C
n
, n ≥ 2, not necessarily with
smooth boundary. We study the question of solving
¯
∂ with prescribed support. The
1
2
or C
∞
Cauchy problem is the following question:
Given a
¯
∂closed (j, ¡)form ) with 1
2
(or C
∞
) coeﬃcients in C
n
such that )
is supported in Ω, can one ﬁnd n ∈ 1
2
(p,q−1)
(C
n
) (or n ∈ C
∞
(p,q−1)
(C
n
)) such that
¯
∂n = ) in C
n
and n is supported in Ω?
When ¡ = 1, it has been proved in Theorem 3.1.1 that one can solve
¯
∂ with com
pact support for smooth (0, 1)forms as long as C
n
¸Ω has no compact components.
There are no other restrictions on the boundary of Ω. When ¡ 1, we shall use
the duality of the
¯
∂Neumann problem to solve the Cauchy problem on bounded
pseudoconvex domains.
Let × : C
∞
(p,q)
(Ω) →C
∞
(n−p,n−q)
(Ω) be the Hodge star operator deﬁned by
¸φ, ψ)d\ = φ ∧ ×ψ,
9.1 The L
2
Cauchy Problem for
¯
∂ 207
where φ, ψ ∈ C
∞
(p,q)
(Ω) and d\ = i
n
d.
1
∧ d¯ .
1
∧ ∧ d.
n
∧ d¯ .
n
denotes the volume
element in C
n
as before. We can extend × from 1
2
(p,q)
(Ω) to 1
2
(n−p,n−q)
(Ω) in a
natural way. This is an algebraic operator given explicitly as follows: if we write
ψ =
t
I,J
ψ
I,J
d.
I
∧ d¯ .
J
,
then
×ψ =
t
I,J
i
n
c
IJ
¯
ψ
I,J
d.
[
ˆ
I]
∧ d¯ .
[
ˆ
J]
,
where [
ˆ
1] denotes the increasing (n−j)tuple consisting of elements in ¦1, , n¦¸1
and c
IJ
is the sign of the permutation from (1, J, [
ˆ
1], [
ˆ
J]) to (1, 1
t
, , n, n
t
).
Lemma 9.1.1. For every φ ∈ C
∞
(p,q)
(Ω), we have
(9.1.1) × × φ = (−1)
p+q
φ,
and
(9.1.2) ϑ = −×
¯
∂×,
where ϑ and
¯
∂ are viewed as diﬀerential operators.
Proof. (9.1.1) is easy to check. To prove (9.1.2), we have, for any ψ ∈ C
∞
(p,q−1)
(Ω)
and η ∈ C
∞
(p,q)
(Ω) such that η has compact support in Ω,
(
¯
∂ψ, η) =
_
Ω
¯
∂ψ ∧ ×η = (−1)
p+q
_
Ω
ψ ∧
¯
∂ × η +
_
Ω
d(ψ ∧ ×η).
Since
_
Ω
d(ψ ∧ ×η) = 0, using (9.1.1),
(
¯
∂ψ, η) = −
_
Ω
ψ ∧ × ×
¯
∂ × η = −(ψ, ×
¯
∂ × η) = (ψ, ϑη).
This proves (9.1.2).
Theorem 9.1.2. Let Ω be a bounded pseudoconvex domain with C
1
boundary in
C
n
, n ≥ 2. Let δ = sup
z,z
∈Ω
[. − .
t
[ be the diameter of Ω. For every ) ∈ 1
2
(p,q)
(C
n
),
where 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n − 1, with
¯
∂) = 0 in the distribution sense in
C
n
and ) supported in Ω, one can ﬁnd n ∈ 1
2
(p,q−1)
(C
n
) such that
¯
∂n = ) in the
distribution sense in C
n
with n supported in Ω and
(n −¡)
_
Ω
[ n [
2
d\ ≤ cδ
2
_
Ω
[ ) [
2
d\.
Proof. From Theorem 4.4.1, the
¯
∂Neumann operator of degree (n −j, n −¡) in Ω,
denoted by ·
(n−p,n−q)
, exists. We deﬁne
(9.1.3) n = −×
¯
∂·
(n−p,n−q)
×),
208 The Tangential CauchyRiemann Complex
then n ∈ 1
2
(p,q−1)
(Ω) and ×n ∈ Dom(
¯
∂
∗
).
Extending n to C
n
by deﬁning n = 0 in C
n
¸ Ω, we claim that
¯
∂n = ) in the
distribution sense in C
n
. First we prove that
¯
∂n = ) in the distribution sense in Ω.
From (9.1.1) and (9.1.2),
(9.1.4)
¯
∂n = −
¯
∂ ×
¯
∂·
(n−p,n−q)
× )
= (−1)
p+q+1
× ×
¯
∂ ×
¯
∂·
(n−p,n−q)
× )
= (−1)
p+q
×ϑ
¯
∂·
(n−p,n−q)
× )
= (−1)
p+q
×
¯
∂
∗
¯
∂·
(n−p,n−q)
× ),
where ϑ acts in the distribution sense in Ω. On the other hand, for any φ ∈
C
∞
(n−p,n−q−1)
(Ω),
(9.1.5)
(
¯
∂φ, ×))
Ω
= (−1)
p+q
_
Ω
¯
∂φ ∧ ) = (−1)
p+q
_
Ω
) ∧ × ×
¯
∂φ
= (−1)
p+q
(), ×
¯
∂φ)
Ω
= (), ϑ × φ)
Ω
= (
¯
∂), ×φ)
C
n
= 0
since supp ) ⊂ Ω and
¯
∂) = 0 in the distribution sense in C
n
. If Ω has C
1
boundary,
using the ﬁrst part of the proof of Lemma 4.3.2, the set C
∞
(n−p,n−q−1)
(Ω) is dense in
Dom(
¯
∂) in the graph norm. It follows from the deﬁnition of
¯
∂
∗
that ×) ∈ Dom(
¯
∂
∗
)
and
¯
∂
∗
(×)) = 0. Using Theorem 4.4.1 when 1 ≤ ¡ < n−1 and Theorem 4.4.3 when
¡ = n −1, we have
(9.1.6)
¯
∂
∗
·
(n−p,n−q)
× ) = ·
(n−p,n−q−1)
¯
∂
∗
(×)) = 0.
Combining (9.1.4) and (9.1.6),
¯
∂n = (−1)
p+q
×
¯
∂
∗
¯
∂·
(n−p,n−q)
× )
= (−1)
p+q
× (
¯
∂
∗
¯
∂ +
¯
∂
¯
∂
∗
)·
(n−p,n−q)
× )
= (−1)
p+q
× ×)
= )
in the distribution sense in Ω. Furthermore, we note that ×n ∈ Dom(
¯
∂
∗
) and this
additional condition implies that
¯
∂n = ) in the distribution sense in C
n
. Using
¯
∂
∗
(×n) = ϑ × n = (−1)
p+q
×
¯
∂n = (−1)
p+q
× ),
where ϑ × n is taken in the distribution sense in Ω, we have for any ψ ∈ C
∞
(p,q)
(C
n
),
(9.1.7)
(n, ϑψ)
C
n = (×ϑψ, ×n)
Ω
= (−1)
p+q
(
¯
∂ × ψ, ×n)
Ω
= (−1)
p+q
(×ψ,
¯
∂
∗
(×n))
Ω
= (×ψ, ×))
Ω
= (), ψ)
C
n,
9.1 The L
2
Cauchy Problem for
¯
∂ 209
where the third equality holds since ×n ∈ Dom(
¯
∂
∗
). Thus
¯
∂n = ) in the distribution
sense in C
n
. The estimate for n follows from Theorem 4.4.1. Theorem 9.1.2 is
proved.
The assumption that Ω has C
1
boundary is used to show that C
∞
(n−p,n−q−1)
(Ω)
is dense in Dom(
¯
∂) in the graph norm so that (9.1.6) holds. Using the proof of
Lemma 4.3.2, Theorem 9.1.2 also holds when the domain is starshaped or locally
starshaped.
When ¡ ≤ n, including the top degree case, we also have the following result.
Theorem 9.1.3. Let Ω be a bounded pseudoconvex domain in C
n
, n ≥ 2. For any
) ∈ 1
2
(p,q)
(C
n
), 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n, such that ) is supported in Ω and
(9.1.8)
_
Ω
) ∧ p = 0 for every p ∈ 1
2
(n−p,n−q)
(Ω) ∩ Ker(
¯
∂),
one can ﬁnd n ∈ 1
2
(p,q−1)
(C
n
) such that
¯
∂n = ) in the distribution sense in C
n
with
n supported in Ω and
_
Ω
[ n [
2
d\ ≤ cδ
2
_
Ω
[ ) [
2
d\,
where δ = sup
z,z
∈Ω
[. −.
t
[ is the diameter of Ω.
If Ω is a bounded pseudoconvex domain with smooth boundary, (9.1.8) can be
replaced by the condition
(9.1.9)
_
Ω
) ∧ p = 0 for every p ∈ C
∞
(n−p,n−q)
(Ω) ∩ Ker(
¯
∂),
and the same conclusion holds.
Proof. Using Theorem 4.4.1 and Theorem 4.4.3, the
¯
∂Neumann operator ·
(p,q)
exists for any 0 ≤ j ≤ n and 0 ≤ ¡ ≤ n. When ¡ = 0, we have
(9.1.10) ·
(p,0)
=
¯
∂
∗
·
2
(p,1)
¯
∂.
For every 0 ≤ ¡ ≤ n, the Bergman projection operator 1
(p,q)
is given by
(9.1.11)
¯
∂
∗
¯
∂·
(p,q)
= 1 −1
(p,q)
.
We deﬁne n by
(9.1.12) n = −×
¯
∂·
(n−p,n−q)
×).
Using Lemma 9.1.1, we have from (9.1.11),
(9.1.13)
¯
∂n = (−1)
p+q
×
¯
∂
∗
¯
∂·
(n−p,n−q)
× )
= ) −(−1)
p+q
×1
(n−p,n−q)
× ).
210 The Tangential CauchyRiemann Complex
From (9.1.8), we get for any p ∈ 1
2
(n−p,n−q)
(Ω) ∩ Ker(
¯
∂),
(9.1.14) (×), p) = (−1)
p+q
_
Ω
p ∧ ) = 0.
Thus 1
(n−p,n−q)
(×)) = 0 and
¯
∂n = ) in Ω from (9.1.13).
Using ×n ∈ Dom(
¯
∂
∗
) and extending n to be zero outside Ω, we can repeat the
arguments of (9.1.7) to show that
¯
∂n = ) in C
n
in the distribution sense. The
estimate holds from Theorems 4.4.1 and 4.4.3. Thus, the proposition is proved
when ) satisﬁes (9.1.8).
When /Ω is smooth, we have C
∞
(n−p,n−q)
(Ω)∩Ker(
¯
∂) is dense in H
(n−p,n−q)
(Ω) =
1
2
(n−p,n−q)
(Ω) ∩Ker(
¯
∂) in the 1
2
(Ω) norm, using Corollary 6.1.6. Thus if ) satisﬁes
condition (9.1.9), it also satisﬁes condition (9.1.8). Theorem 9.1.3 is proved.
Remark: When ¡ < n, condition (9.1.9) implies that
¯
∂) = 0 in the distribution
sense in C
n
. To see this, we take p =
¯
∂ × · for some · ∈ C
∞
(p,q+1)
(C
n
) in (9.1.9).
Then we have
(), ϑ·)
C
n =
_
D
) ∧ ×ϑ· = (−1)
p+q+1
_
D
) ∧
¯
∂ × · = 0
for any · ∈ C
∞
(p,q+1)
(C
n
). This implies that
¯
∂) = 0 in the distribution sense
in C
n
. From the proof of Theorem 9.1.2, the two conditions are equivalent if
C
∞
(n−p,n−q−1)
(Ω) is dense in Dom(
¯
∂) in the graph norm.
9.2
¯
∂Closed Extensions of Forms and C
∞
Solvability of
¯
∂
b
Let Ω be a bounded pseudoconvex domain in C
n
with smooth boundary ` = /Ω
and ρ be a smooth deﬁning function for Ω such that [dρ[ = 1 on `. We use c
(p,q)
(`)
to denote the smooth (j, ¡)forms on `, where 0 ≤ j ≤ n, 0 ≤ ¡ ≤ n−1. Here the
extrinsic deﬁnition for c
(p,q)
(`) is used.
We consider the following two kinds of
¯
∂closed extension problems:
Given α ∈ c
(p,q)
(`),
(1) can one ﬁnd an extension ˜ α of α such that τ ˜ α = α on ` and
¯
∂˜ α = 0
in Ω? (We recall that τ is the projection of smooth (j, ¡)forms in C
n
to
(j, ¡)forms on ` which are pointwise orthogonal to the ideal generated by
¯
∂ρ.)
(2) can one ﬁnd an extension ˜ α of α such that ˜ α = α on ` and
¯
∂˜ α = 0 in Ω?
When ¡ < n − 1, it is necessary that
¯
∂
b
α = 0 on ` in order to have a
¯
∂closed
extension. When α is a function (j = ¡ = 0), this is the question of holomorphic
extension of C1 functions. In this case problems (1) and (2) are the same. It
was proved in Theorem 3.2.2 that any C1 function of class C
1
on the boundary
of a C
1
bounded domain Ω has a holomorphic extension as long as C
n
¸ Ω has no
bounded component. When α is a (j, ¡)form with ¡ ≥ 1, (2) seems to be a stronger
problem than (1). It will be shown in the next two theorems that these two kinds
of extension problems are equivalent for smooth forms also.
9.2
¯
∂Closed Extensions of Forms and C
∞
Solvability of
¯
∂
b
211
Theorem 9.2.1. Let Ω be a bounded pseudoconvex domain in C
n
with smooth
boundary `. Let α ∈ c
p,q
(`), where 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n − 1. Then there
exists ˜ α ∈ C
∞
(p,q)
(Ω) such that τ ˜ α = α on ` and
¯
∂˜ α = 0 in Ω if and only if
(9.2.1)
_
M
α ∧ ψ = 0 for every ψ ∈ C
∞
(n−p,n−q−1)
(Ω) ∩ Ker(
¯
∂).
Furthermore, when 1 ≤ ¡ < n −1, (9.2.1) holds if and only if
(9.2.2)
¯
∂
b
α = 0 on `.
Theorem 9.2.2. Let ` and α be the same as in Theorem 9.2.1. There exists ˜ α
such that ˜ α ∈ C
∞
(p,q)
(Ω), ˜ α = α on ` and
¯
∂˜ α = 0 in Ω if and only if (9.2.1) (for
1 ≤ ¡ ≤ n −1) or (9.2.2) (for 1 ≤ ¡ < n −1) holds.
Proof of Theorem 9.2.1. It is easy to see that (9.2.1) and (9.2.2) are necessary
conditions for the existence of
¯
∂closed extensions. We assume that α satisﬁes
(9.2.1).
Let α
t
be a smooth extension of α by extending α componentwise from the
boundary to Ω. Thus α
t
∈ C
∞
(p,q)
(Ω) and α
t
= α on `. We set ) =
¯
∂α
t
in Ω, then
) ∈ C
∞
(p,q+1)
(Ω) and ) ∧
¯
∂ρ =
¯
∂
b
α ∧
¯
∂ρ = 0 on `. Using (9.2.1), we have for any
¯
∂closed ψ ∈ C
∞
(n−p,n−q−1)
(Ω),
_
Ω
) ∧ ψ =
_
Ω
¯
∂(α
t
∧ ψ) =
_
M
α ∧ ψ = 0.
Thus, ) satisﬁes condition (9.1.9) in Theorem 9.1.3. We ﬁrst assume that the
¯
∂
Neumann operator ·
(n−p,n−q−1)
for (n−j, n−¡ −1)forms on Ω is C
∞
regular up
to the boundary and deﬁne
(9.2.3) n = −×
¯
∂·
(n−p,n−q−1)
×).
Then n ∈ C
∞
(p,q)
(Ω). As in Theorem 9.1.3, it follows that
¯
∂n = ) in Ω and ×n ∈
Dom(
¯
∂
∗
). Since the boundary is smooth and n is smooth up to the boundary, we
can write n = τn +νn. Using Lemma 4.2.1, we have
ν(×n) = 0 on `.
However, this is equivalent to
τn = 0 on `.
Setting ˜ α = α
t
−n, we have
¯
∂˜ α = 0 in Ω and τ ˜ α = τα
t
= α on `. This proves the
theorem assuming that ·
(n−p,n−q−1)
is C
∞
regular.
In general, instead of (9.2.3), we deﬁne
n
t
= −×
¯
∂·
t
(n−p,n−q−1)
×),
212 The Tangential CauchyRiemann Complex
where ·
t
(n−p,n−q−1)
is the weighted
¯
∂Neumann operator introduced in Theorem
6.1.2 with weight function φ
t
= t[.[
2
, t 0 and × is taken with respect to the
metric 1
2
(1, φ
t
). Choosing t suﬃciently large, from the proof of Theorem 6.1.4
and Corollary 6.1.5, for each large integer / n+2, there exists a solution n
k
such
that n
k
∈ \
k
(p,q)
(Ω) ⊂ C
1
(p,q)
(Ω),
¯
∂n
k
= ) in Ω, τn
k
= 0 on ` and
 n
k

k(Ω)
≤ C
k
 ) 
k(Ω)
.
To construct a solution n ∈ C
∞
(p,q)
(Ω), we set /
k
= n
k
− n
k+1
. Each /
k
is
¯
∂closed
and τ/
k
= 0 on `. This implies that
¯
∂/
k
= 0 in C
n
in the distribution sense.
Using Friedrichs’ lemma and the arguments in the proof of Lemma 4.3.2, we can
ﬁnd a sequence /
k
n
∈ C
∞
(p,q)
(Ω) such that /
k
n
has compact support in Ω, /
k
n
→/
k
in
\
k
(p,q)
(Ω) and
¯
∂/
k
n
→ 0 in \
k
(p,q+1)
(Ω). For each arbitrarily large : ∈ N, one can
ﬁnd ·
k
n
∈ \
m
(p,q)
(Ω) with τ·
k
n
= 0 on ` and
¯
∂·
k
n
=
¯
∂/
k
n
in Ω. Setting
˜
/
k
n
= /
k
n
−·
k
n
,
we have
¯
∂
˜
/
k
n
= 0 in Ω,
˜
/
k
n
→/
k
in \
k
(p,q)
(Ω) with
˜
/
k
n
∈ \
m
(p,q)
(Ω) and τ
˜
/
k
n
= 0 on
`. This implies that inductively one can construct a new sequence n
t
k
∈ \
k
(p,q)
(Ω)
such that
¯
∂n
t
k
= ) in Ω, τn
t
k
= 0 on ` and
 n
t
k
−n
t
k+1

k(Ω)
≤ 1,2
k
, / ∈ N.
Writing
n = n
t
N
+
∞
k=N+1
(n
t
k
−n
t
k−1
),
we have n ∈ C
∞
(p,q)
(Ω) such that
¯
∂n = ) in Ω, τn = 0 on `. Setting ˜ α = α
t
− n,
the ﬁrst part of the theorem is proved.
When 1 ≤ ¡ < n − 1, setting ψ =
¯
∂n for some n ∈ C
∞
(n−p,n−q−2)
(Ω) in (9.2.1),
we have
(9.2.4)
_
M
α ∧ ψ =
_
M
α ∧
¯
∂n = (−1)
p+q+1
_
M
¯
∂
b
α ∧ n = 0.
Thus, (9.2.1) implies (9.2.2). We see from (9.2.4) that (9.2.2) also implies (9.2.1),
since any
¯
∂closed form ψ in C
∞
(n−p,n−q−1)
(Ω) can be written as ψ =
¯
∂n for some
n ∈ C
∞
(n−p,n−q−2)
(Ω) using Theorem 6.1.1. Thus, (9.2.1) and (9.2.2) are equivalent
when ¡ < n −1. The proof of Theorem 9.2.1 is complete.
In order to prove Theorem 9.2.2, we need the following lemma:
Lemma 9.2.3. Let Ω be a bounded domain in C
n
with smooth boundary ` and let
ρ be a smooth deﬁning function for Ω. If α ∈ c
p,q
(`) and ∂
b
α = 0 on `, where
0 ≤ j ≤ n, 0 ≤ ¡ ≤ n − 1, then there exists 1
∞
α such that 1
∞
α ∈ C
∞
(p,q)
(C
n
),
1
∞
α = α on ` and
¯
∂1
∞
α = O(ρ
k
) at ` for every positive integer /.
9.2
¯
∂Closed Extensions of Forms and C
∞
Solvability of
¯
∂
b
213
Proof. We ﬁrst extend α componentwise and smoothly from ` to 1α in C
n
. We
claim that for every positive integer /, there exist smooth (j, ¡)forms α
1
, , α
k
and (j, ¡ + 1)forms γ
1
, , γ
k
such that
(9.2.5 a) 1
k
α = 1α −ρα
1
−
ρ
2
2
α
2
− −
ρ
k
/
α
k
,
and
(9.2.5 b)
¯
∂1
k
α = ρ
k
_
γ
k
−
1
/
¯
∂α
k
_
= O(ρ
k
).
Since
¯
∂
b
α = 0 on `,
¯
∂1α ∧
¯
∂ρ = 0 on `. We can ﬁnd α
1
and γ
1
such that
¯
∂1α =
¯
∂ρ ∧ α
1
+ργ
1
=
¯
∂(ρα
1
) +ρ(γ
1
−
¯
∂α
1
).
Setting 1
1
α = 1α − ρα
1
, it follows that
¯
∂1
1
α = ρ(γ
1
−
¯
∂α
1
) = O(ρ) at `. This
proves (9.2.5 a) and (9.2.5 b) for / = 1. We also note that α
1
is obtained from the
ﬁrst order derivatives of 1α and γ
1
is obtained from the second derivatives of 1α.
Assuming (9.2.5 a) and (9.2.5 b) have been proved for some / ∈ N, we apply
¯
∂
to both sides of (9.2.5 b) to obtain
0 =
¯
∂
2
1
k
α = /ρ
k−1
¯
∂ρ ∧
_
γ
k
−
1
/
¯
∂α
k
_
+ρ
k
¯
∂γ
k
.
Hence
¯
∂ρ ∧
_
γ
k
−(1,/)
¯
∂α
k
_
= 0 on `. Thus, we can ﬁnd a (j, ¡)form α
k+1
and
a (j, ¡ + 1)form γ
k+1
such that γ
k
−(1,/)
¯
∂α
k
=
¯
∂ρ ∧ α
k+1
+ργ
k+1
. We deﬁne
1
k+1
α = 1α −ρα
1
−
ρ
2
2
α
2
− −
ρ
k
/
α
k
−
ρ
k+1
/ + 1
α
k+1
= 1
k
α −
ρ
k+1
/ + 1
α
k+1
,
then
¯
∂1
k+1
α = ρ
k
(
¯
∂ρ ∧ α
k+1
+ργ
k+1
) −
¯
∂
_
ρ
k+1
/ + 1
α
k+1
_
= ρ
k+1
_
γ
k+1
−
1
/ + 1
¯
∂α
k+1
_
= O(ρ
k+1
).
This proves (9.2.5 a) and (9.2.5 b) for / + 1. Using induction, (9.2.5 a) and (9.2.5
b) hold for any positive integer /.
To ﬁnd an extension 1
∞
α such that 1
∞
α = α on ` and
¯
∂1
∞
α = O(ρ
k
) at
` for every positive integer /, we modify the construction as follows: Let Ω
δ
=
¦. ∈ Ω[ − δ < ρ(.) < δ¦ be a small tubular neighborhood of ` and η(.) be a
cutoﬀ function such that η ∈ C
∞
0
(Ω
δ
) and η = 1 on Ω
δ/2
. Let π(.) denote the
projection from Ω
δ
onto ` along the normal direction and n(.) denote the unit
outward normal at . ∈ `. We deﬁne ˜ η
j
(.) = η
_
π(.) +
ρ(z)
j
n(π(.))
_
and
1
∞
α = 1α −
∞
j=1
˜ η
j
(.)
ρ
j
,
α
j
,
214 The Tangential CauchyRiemann Complex
where c
j
is chosen to be suﬃciently small and c
j
¸ 0. One can choose c
j
so small
(depending on α
j
) such that, for each multiindex : = (:
1
, , :
2n
), we have
¸
¸
¸
¸
1
m
_
˜ η
j
(.)
ρ
j
,
α
j
_¸
¸
¸
¸
≤ C
m,j
c
j
≤
1
2
j
, for every : with [:[ ≤ , −1.
The series converges in C
k
for every / ∈ N to some element 1
∞
α ∈ C
∞
(p,q)
(C
n
) and
1
∞
α = α on `. Furthermore, we have
¯
∂1
∞
α =
¯
∂
_
_
1α −
k
j=1
˜ η
j
(.)
ρ
j
,
α
j
_
_
−
¯
∂
_
_
∞
j=k+1
˜ η
j
(.)
ρ
j
,
α
j
_
_
= O(ρ
k
) at `,
for every positive integer /. This proves the lemma.
Proof of Theorem 9.2.2. Let α
t
= 1
∞
α where 1
∞
α is as in Lemma 9.2.3. Using
the proof of Theorem 9.2.1, there exists n ∈ C
∞
(p,q)
(Ω) such that
¯
∂n =
¯
∂α
t
in Ω and
τn = 0 on `. Setting 1
0
α = n, we have
(9.2.6 a)
¯
∂1
0
α =
¯
∂α
t
in Ω,
(9.2.6 b) 1
0
α ∧
¯
∂ρ = 0 on `.
We shall prove that for any nonnegative integer /, there exist (j, ¡ − 1)forms
β
0
, β
1
, , β
k
and (j, ¡)form η
k
such that
(9.2.7)
1
0
α =
¯
∂(ρβ
0
) +
1
2
¯
∂(ρ
2
β
1
) + +
1
/ + 1
¯
∂(ρ
k+1
β
k
)
+ρ
k+1
_
η
k
−
1
/ + 1
¯
∂β
k
_
.
From (9.2.6 a) and (9.2.6 b), we can write 1
0
α =
¯
∂ρ∧β
0
+ρη
0
=
¯
∂(ρβ
0
)+ρ(η
0
−
¯
∂β
0
)
for some (j, ¡ −1)form β
0
and (j, ¡)form η
0
. This proves (9.2.7) for / = 0.
Assuming (9.2.7) is proved for / ≥ 0, from (9.2.6 a),
¯
∂1
0
α = (/ + 1)ρ
k
¯
∂ρ ∧
_
η
k
−
1
/ + 1
¯
∂β
k
_
+ρ
k+1
¯
∂η
k
=
¯
∂α
t
.
Since
¯
∂α
t
vanishes to arbitrarily high order at the boundary `, we have
¯
∂ρ ∧(η
k
−
1,(/ + 1)
¯
∂β
k
) = 0 on ` and there exist a (j, ¡ − 1)form β
k+1
and a (j, ¡)form
η
k+1
such that η
k
−1,(/+1)
¯
∂β
k
=
¯
∂ρ∧β
k+1
+ρη
k+1
. Substituting this into (9.2.7),
we obtain
1
0
α =
¯
∂(ρβ
0
) +
1
2
¯
∂(ρ
2
β
1
) + +
1
/ + 1
¯
∂(ρ
k+1
β
k
)
+ρ
k+1
(
¯
∂ρ ∧ β
k+1
+ρη
k+1
)
=
¯
∂(ρβ
0
) +
1
2
¯
∂(ρ
2
β
1
) + +
1
/ + 1
¯
∂(ρ
k+1
β
k
) +
1
/ + 2
¯
∂(ρ
k+2
β
k+1
)
+ρ
k+2
_
η
k+1
−
1
/ + 2
¯
∂β
k+1
_
.
9.2
¯
∂Closed Extensions of Forms and C
∞
Solvability of
¯
∂
b
215
Thus, (9.2.7) holds for /+1 and by induction, for any nonnegative integer /. Setting
(9.2.8) 1
k+1
α = 1
0
α −
k
i=0
1
i + 1
¯
∂(ρ
i+1
β
i
),
we have
(9.2.9a)
¯
∂1
k+1
α =
¯
∂α
t
in Ω,
(9.2.9b) 1
k+1
α = O(ρ
k+1
) at `.
Also each β
i
is obtained by taking ith derivatives of the components of 1
0
α. Thus
each β
i
is smooth and 1
k
α ∈ C
∞
(p,q)
(Ω).
Let η(.) and ˜ η
j
(.) be the same as in Lemma 9.2.3, we deﬁne
(9.2.10) 1
∞
α = 1
0
α −
∞
i=0
1
i + 1
¯
∂
_
˜ η
j
(.)ρ
i+1
β
i
_
.
As in the proof of Lemma 9.2.3, we can choose c
i
suﬃciently small such that the
series converges in every C
k
norm to some element 1
∞
α. 1
∞
α satisﬁes
¯
∂1
∞
α =
¯
∂α
t
in Ω and 1
∞
α = O(ρ
k
) at ` for every / = 1, 2, . Setting ˜ α = 1
∞
α −1
∞
α, we
have ˜ α = α on ` and
¯
∂˜ α = 0 in Ω. This proves the theorem.
The extension result proved in Theorem 9.2.1 can be used to study the global
solvability of the equation
(9.2.11)
¯
∂
b
n = α on `,
where α is a (j, ¡)form with smooth coeﬃcients, 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n − 1. it
is easy to see that if (9.2.11) is solvable, then α must satisfy
(9.2.12)
¯
∂
b
α = 0, when 1 ≤ ¡ < n −1.
Also using Stokes’ theorem, it is easy to see that if (9.2.11) is solvable for some
n ∈ c
p,q−1
(`), then α must satisfy
(9.2.12 a)
_
M
α ∧ φ = 0, φ ∈ c
n−p,n−q−1
(`) ∩ Ker(
¯
∂
b
)
where 1 ≤ ¡ ≤ n − 1. We note that using Theorem 9.2.1, we can substitute φ in
(9.2.12 a) by φ ∈ C
∞
(n−p,n−q−1)
(Ω) ∩ Ker(
¯
∂).
When 1 ≤ ¡ < n − 1, condition (9.2.12 a) always implies condition (9.2.12)
(regardless of pseudoconvexity). This can be seen easily if we take φ in (9.2.12 a)
to be of the form
¯
∂
b
), where ) is any smooth (n −j, n −¡ −2)form on `.
216 The Tangential CauchyRiemann Complex
Theorem 9.2.4. Let Ω be a bounded pseudoconvex domain in C
n
with smooth
boundary `. For any α ∈ c
p,q
(`), where 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n − 1, there
exists n ∈ c
p,q−1
(`) satisfying
¯
∂
b
n = α on ` if and only if the following conditions
hold:
¯
∂
b
α = 0 on `, when 1 ≤ ¡ < n −1,
and _
M
α ∧ ψ = 0, φ ∈ c
n−p,0
(`) ∩ Ker(
¯
∂
b
), when ¡ = n −1.
Proof. From Theorem 9.2.1, we can extend α to ˜ α such that ˜ α ∈ C
∞
(p,q)
(Ω),
¯
∂˜ α = 0, in Ω
and
˜ α ∧
¯
∂ρ = α ∧
¯
∂ρ, on `.
Using Theorem 6.1.1, we can ﬁnd a ˜ n ∈ C
∞
(p,q−1)
(Ω) such that
¯
∂˜ n = ˜ α in Ω.
Denoting the restriction of ˜ n to ` by n, we have
¯
∂˜ n ∧
¯
∂ρ = ˜ α ∧
¯
∂ρ on `, or
equivalently
¯
∂
b
n = α on `. This proves the theorem.
We conclude this chapter with the following theorem:
Theorem 9.2.5. Let Ω be a bounded pseudoconvex domain in C
n
with smooth
boundary `. For any α ∈ c
p,q
(`), where 0 ≤ j ≤ n and 1 ≤ ¡ ≤ n − 1, the
following conditions are equivalent:
(1) There exists n ∈ c
p,q−1
(`) satisfying
¯
∂
b
n = α on `.
(2) There exists ˜ α ∈ C
∞
(p,q)
(Ω) with τ ˜ α = α (or ˜ α = α) on ` and
¯
∂˜ α = 0 in Ω.
(3)
_
M
α ∧ ψ = 0, φ ∈ c
n−p,n−q−1
(`) ∩ Ker(
¯
∂
b
).
When 1 ≤ ¡ < n −1, the above conditions are equivalent to
(4)
¯
∂
b
α = 0 on `.
9.3 1
2
Existence Theorems and Sobolev Estimates for
¯
∂
b
Let ` be the boundary of a smooth domain Ω in C
n
. We impose the induced
metric from C
n
on ` and denote square integrable functions on ` by 1
2
(`).
The set of (j, ¡)forms on ` with 1
2
coeﬃcients, denoted by 1
2
(p,q)
(`), is the
completion of c
p,q
(`) under the sum of 1
2
norms of the coeﬃcients. We deﬁne the
space of (j, ¡)forms with C
k
(`) coeﬃcients by C
k
(p,q)
(`). In particular, c
p,q
(`) =
C
∞
(p,q)
(`). By using a partition of unity and the tangential Fourier transform, we
can deﬁne the Sobolev space \
s
(`) for any real number :. Let \
s
(p,q)
(`) be the
subspace of 1
2
(p,q)
(`) with \
s
(`) coeﬃcients for : ≥ 0 and the norm in \
s
(p,q)
(`)
is denoted by  
s(M)
. It is clear that \
0
(p,q)
(`) = 1
2
(p,q)
(`) and  
0(M)
= 
M
.
9.3 L
2
Existence Theorems and Sobolev Estimates for
¯
∂
b
217
The 1
2
closure of
¯
∂
b
, still denoted by
¯
∂
b
, is a linear, closed, densely deﬁned operator
such that
(9.3.1)
¯
∂
b
: 1
2
(p,q−1)
(`) →1
2
(p,q)
(`).
An element n ∈ 1
2
(p,q−1)
(`) belongs to Dom(
¯
∂
b
) if and only if
¯
∂
b
n, deﬁned in the
distribution sense, is in 1
2
(p,q)
(`).
Our main result in this section is the following theorem.
Theorem 9.3.1. Let Ω be a bounded pseudoconvex domain in C
n
with smooth
boundary `. For every α ∈ \
s
(p,q)
(`), where 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n −2 and : is a
nonnegative integer, such that
(9.3.2)
¯
∂
b
α = 0 on `,
there exists n ∈ \
s
(p,q−1)
(`) satisfying
¯
∂
b
n = α on `.
When ¡ = n −1, α ∈ 1
2
(p,n−1)
(`) and α satisﬁes
(9.3.3)
_
M
α ∧ φ = 0, φ ∈ C
∞
(n−p,0)
(`) ∩ Ker(
¯
∂
b
),
there exists n ∈ 1
2
(p,n−2)
(`) satisfying
¯
∂
b
n = α on `.
Corollary 9.3.2. Let Ω be a bounded pseudoconvex domain in C
n
with smooth
boundary `. Then
¯
∂
b
: 1
2
(p,q−1)
(`) → 1
2
(p,q)
(`), 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n − 1, has
closed range in 1
2
.
It is easy to see that (9.3.2) and (9.3.3) are necessary conditions for
¯
∂
b
to be
solvable in 1
2
. To prove Theorem 9.3.1, we shall ﬁrst assume that α is smooth. We
then show that there exists a constant C
s
independent of α such that
(9.3.4)  n 
s(M)
≤ C
s
 α 
s(M)
.
Using Theorem 9.2.4, we can ﬁnd a solution for any smooth α satisfying (9.3.2)
or (9.3.3). However, it is not easy to obtain estimates from this construction. We
shall use a diﬀerent method to solve α by exploiting the relationship between the
norms on the boundary ` and the tangential Sobolev norms. We also introduce
the weighted tangential Sobolev norms.
Let ρ be a deﬁning function for Ω. Let
˜
Ω be a large ball such that Ω ⊂⊂
˜
Ω. We
set
Ω
+
=
˜
Ω ¸ Ω, Ω
−
= Ω.
For a small δ 0, we set
Ω
−
δ
= ¦. ∈ Ω
−
[ −δ < ρ(.) < 0¦,
Ω
+
δ
= ¦. ∈ Ω
+
[ 0 < ρ(.) < δ¦,
Ω
δ
= ¦. ∈
˜
Ω [ −δ < ρ(.) < δ¦,
218 The Tangential CauchyRiemann Complex
and
Γ
= ¦. ∈ C
n
[ ρ(.) = c¦.
The special tangential norms in a tubular neighborhood Ω
−
δ
, Ω
+
δ
and Ω
δ
are deﬁned,
as in Section 5.2, by
[[[)[[[
2
s(Ω
δ
)
=
_
δ
−δ
)
2
s(Γρ)
dρ,
[[[)[[[
2
s(Ω
−
δ
)
=
_
0
−δ
)
2
s(Γρ)
dρ, [[[)[[[
2
s(Ω
+
δ
)
=
_
δ
0
)
2
s(Γρ)
dρ.
For each : ∈ N, : ∈ 1, we set
(9.3.5) [[[1
m
)[[[
s(Ω
δ
)
=
0≤k≤m
[[[1
k
ρ
)[[[
s+m−k(Ω
δ
)
,
where 1
ρ
= ∂,∂ρ. We also deﬁne the weighted tangential Sobolev norms by
(9.3.6)
[[[Θ
m
)[[[
s(Ω
δ
)
= [[[ρ
m
1
m
)[[[
s(Ω
δ
)
=
0≤k≤m
[[[ρ
m
1
k
ρ
)[[[
s+m−k(Ω
δ
)
,
and similarly
[[[1Θ
m
)[[[
s(Ω
δ
)
=
0≤k≤m
[[[1ρ
m
1
k
ρ
)[[[
s+m−k(Ω
δ
)
.
Thus Θ can be viewed as a ﬁrst order diﬀerential operator weighted with ρ. Cor
responding norms are also deﬁned on Ω
−
δ
and Ω
+
δ
similarly. We always assume
that δ 0 to be suﬃciently small without specifying so explicitly in the following
lemmas.
The next lemma on the extension of smooth functions from the boundary is the
key to the proof of (9.3.4).
Lemma 9.3.3. Let ` be the boundary of a smooth domain Ω in C
n
. For arbitrary
smooth functions n
j
on `, , = 0, 1, , /
0
, there exists a function 1n ∈ C
∞
0
(Ω
δ
)
such that 1
j
ρ
1n = n
j
on `, , = 0, 1, , /
0
. Furthermore, for every real number :
and nonnegative integer :, there exists a positive constant C depending on : and
: but independent of the n
j
’s such that
(9.3.7 i) [[[1
m
1n[[[
s−m+
1
2
(Ω
δ
)
≤ C
k0
j=0
n
j

s−j(M)
,
(9.3.7 ii) [[[Θ
m
1n[[[
s+
1
2
(Ω
δ
)
≤ C
k0
j=0
n
j

s−j(M)
.
9.3 L
2
Existence Theorems and Sobolev Estimates for
¯
∂
b
219
Proof. Using a partition of unity, it suﬃces to prove the lemma assuming that n
j
is supported in a small neighborhood l ∩ `, where l ⊂ C
n
and that there exists
a special boundary coordinate chart on l with coordinates t
1
, , t
2n−1
, ρ.
The Fourier transform for n in the special boundary chart is deﬁned by
ˆ n(τ) =
_
R
2n−1
c
−i¸t,τ)
n(t)dt,
where τ = (τ
1
, , τ
2n−1
) and ¸t, τ) = t
1
τ
1
+ +t
2n−1
τ
2n−1
.
Let ψ be a function in C
∞
0
(1) which is equal to 1 in a neighborhood of 0 and let
the partial Fourier transform of 1n be
(9.3.8) (1n)˜(τ, ρ) = ψ(λρ)
k0
j=0
ˆ n
j
(τ)
ρ
j
,!
,
where λ = (1 +[τ[
2
)
1/2
. It is easy to see that 1
j
ρ
1n = n
j
on `.
To prove (9.3.7 i) and (9.3.7 ii), we note that for every nonnegative integer i, by
a change of variables, there exists some C 0 such that
_
∞
−∞
[1
i
ρ
(ψ(λρ)ρ
j
)[
2
dρ = λ
2(i−j)−1
_
∞
−∞
[1
i
ρ
(ψ(ρ)ρ
j
)[
2
dρ ≤ Cλ
2(i−j)−1
.
Thus we have
[[[1
m
1n[[[
2
s−m+
1
2
(Ω
δ
)
=
0≤k≤m
[[[1
k
ρ
1n[[[
2
s−k+
1
2
(Ω
δ
)
≤ C
k0
j=0
_
R
2n−1
λ
2s−2j
[ˆ n
j
(τ)[
2
dτ
≤ C
k0
j=0
n
j

2
s−j(M)
,
which proves (9.3.7 i). Since
_
∞
−∞
ρ
2m
[1
i
ρ
(ψ(λρ)ρ
j
)[
2
dρ = λ
2(i−j−m)−1
_
∞
−∞
ρ
2m
[1
i
ρ
(ψ(ρ)ρ
j
)[
2
dρ
≤ Cλ
2(i−j−m)−1
,
we see that
[[[Θ
m
1n[[[
2
s+
1
2
(Ω
δ
)
=
0≤k≤m
[[[ρ
m
1
k
ρ
1n[[[
2
s+m−k+
1
2
(Ω
δ
)
≤ C
k0
j=0
_
R
2n−1
λ
2s−2j
[ˆ n
j
(τ)[
2
dτ
≤ C
k0
j=0
n
j

2
s−j(M)
.
220 The Tangential CauchyRiemann Complex
This proves (9.3.7 ii) and the lemma.
Estimate (9.3.7 i) shows that when one extends a function from a smooth bound
ary, one can have a “gain” of one half derivative. Estimate (9.3.7 ii) shows that the
operator ρ1 for any ﬁrst order derivative 1 should be treated as an operator of
order zero in view of extension of functions from the boundary. This fact is crucial
in the proof of Theorem 9.3.1. We also remark that Lemma 9.3.3 also holds for
/
0
= ∞ using arguments similar to Lemma 9.2.3.
Lemma 9.3.4. Let ` be the boundary of a smooth bounded domain Ω in C
n
.
Let α ∈ C
∞
(p,q)
(`), 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n − 1, and ∂
b
α = 0 on `. For every
positive integer /, there exists a smooth extension 1
k
α with support in a tubular
neighborhood Ω
δ
such that 1
k
α ∈ C
∞
(p,q)
(Ω
δ
), 1
k
α = α on ` and
(9.3.9)
¯
∂1
k
α = O(ρ
k
) at `.
Furthermore, for every real number : and nonnegative integer :, there exists a
positive constant C
k
depending on : and : but independent of α such that
(9.3.10 i) [[[1
m
1
k
α[[[
s−m+
1
2
(Ω
δ
)
≤ C
k
α
s(M)
,
(9.3.10 ii) [[[Θ
m
1
k
α[[[
s+
1
2
(Ω
δ
)
≤ C
k
α
s(M)
.
Proof. Using Lemma 9.3.3 with /
0
= 0, we ﬁrst extend α componentwise and
smoothly from ` to 1α in C
n
such that 1α has compact support in Ω
δ
and
satisﬁes the estimates
(9.3.11 i) [[[1
m
1α[[[
s−m+
1
2
(Ω
δ
)
≤ Cα
s(M)
,
(9.3.11 ii) [[[Θ
m
1α[[[
s+
1
2
(Ω
δ
)
≤ Cα
s(M)
,
where C depends on : and : but is independent of α. Using Lemma 9.2.3, for every
positive integer /, there exist smooth (j, ¡)forms α
1
, , α
k
and (j, ¡ + 1)forms
γ
1
, , γ
k
such that
(9.3.12 a) 1
k
α = 1α −ρα
1
−
ρ
2
2
α
2
− −
ρ
k
/
α
k
,
and
(9.3.12 b)
¯
∂1
k
α = ρ
k
_
γ
k
−
1
/
¯
∂α
k
_
= O(ρ
k
) at `.
From the proof of Lemma 9.2.3, each component of α
i
is a linear combination of
the ith derivatives of 1α. To show that 1
k
α satisﬁes the estimates, it suﬃces to
estimate each ρ
i
α
i
. Using (9.3.11 i) and (9.3.11 ii), we have for any : ∈ 1,
[[[1
k
α[[[
s+
1
2
(Ω
δ
)
≤ [[[1α[[[
s+
1
2
(Ω
δ
)
+
1≤i≤k
[[[ρ
i
α
i
[[[
s+
1
2
(Ω
δ
)
≤ C
0≤i≤k
[[[Θ
i
1α[[[
s+
1
2
(Ω
δ
)
≤ Cα
s(M)
.
9.3 L
2
Existence Theorems and Sobolev Estimates for
¯
∂
b
221
Again using (9.3.11 i) and (9.3.11 ii), we have for any : ∈ 1, : ∈ N,
[[[Θ
m
1
k
α[[[
s+
1
2
(Ω
δ
)
≤ C
0≤i≤k
[[[Θ
m+i
1α[[[
s+
1
2
(Ω
δ
)
≤ C
k
α
s(M)
,
and
[[[1
m
1
k
α[[[
s−m+
1
2
(Ω
δ
)
≤ C
0≤i≤k
[[[Θ
i
1
m
1α[[[
s−m+
1
2
(Ω
δ
)
≤ C
k
α
s(M)
,
where we have used (9.3.7 i) and (9.3.7 ii). This proves Lemma 9.3.4.
The following decomposition of ∂
b
closed forms on ` as the diﬀerence between
two
¯
∂closed forms is an analog of the jump formula for C1 functions discussed in
Theorem 2.2.3.
Lemma 9.3.5. Let ` be the boundary of a smooth bounded domain Ω in C
n
. Let
α ∈ C
∞
(p,q)
(`) with ∂
b
α = 0, 0 ≤ j ≤ n, 0 ≤ ¡ ≤ n−1. For each positive integer /,
there exist α
+
∈ C
k
(p,q)
(Ω
+
) and α
−
∈ C
k
(p,q)
(Ω) such that ∂α
+
= 0 in Ω
+
, ∂α
−
= 0
in Ω and the following decomposition holds:
(9.3.13) α
+
−α
−
= α on `.
Furthermore, we have the following estimates: for every integer 0 ≤ : ≤ / − 1,
0 ≤ : ≤ :,
(9.3.14 i) [[[1
m
α
+
[[[
s−m+
1
2
(Ω
+
δ
)
≤ Cα
s(M)
,
(9.3.14 ii) [[[1
m
α
−
[[[
s−m+
1
2
(Ω
−
δ
)
≤ Cα
s(M)
,
where the constant C depends only on :, :, but is independent of α.
Proof. Let /
0
be a positive integer to be determined later. Using Lemma 9.3.4, we
extend α from ` to 1
k0
α in C
n
smoothly such that 1
k0
α = α on `, 1
k0
α has
compact support in Ω
δ
and 1
k0
α satisﬁes (9.3.9) and (9.3.10) with / = /
0
.
We deﬁne a (j, ¡ + 1)form
˜
l
k0
in
˜
Ω by
˜
l
k0
=
_
¸
_
¸
_
−∂1
k0
α, if . ∈ Ω
−
,
0, if . ∈ `,
∂1
k0
α, if . ∈ Ω
+
.
From (9.3.9), we have
˜
l
k0
∈ C
k0−1
(
˜
Ω) and ∂
˜
l
k0
= 0 in
˜
Ω (in the distribution
sense if /
0
= 1). It follows from (9.3.10 i) that for any nonnegative integer :,
0 ≤ : ≤ /
0
−1,
(9.3.15)
[[[1
m
˜
l
k0
[[[
s−m−
1
2
(Ω
δ
)
≤ C[[[1
m+1
1
k0
α[[[
s−m−
1
2
(Ω
δ
)
≤ Cα
s(M)
.
222 The Tangential CauchyRiemann Complex
We deﬁne \
k0
=
¯
∂
∗
·
˜
Ω
(p,q+1)
˜
l
k0
. It follows from Theorem 4.4.1 that ∂\
k0
=
˜
l
k0
in
˜
Ω. Since
(p,q+1)
is elliptic in the interior of the domain
˜
Ω,
¯
∂
∗
·
˜
Ω
(p,q+1)
gains
one derivative in the interior. Since
˜
l
k0
has compact support in Ω
δ
, we get from
(9.3.15) that
(9.3.16)
[[[1
m
\
k0
[[[
s−m+
1
2
(Ω
δ
)
≤ C[[[1
m+1
1
k0
α[[[
s−m−
1
2
(Ω
δ
)
≤ Cα
s(M)
,
for some C 0 independent of α.
Setting
α
+
=
1
2
(1
k0
α −\
k0
), . ∈ Ω
+
,
α
−
= −
1
2
(1
k0
α +\
k0
), . ∈ Ω,
we see that
α = 1
k0
α = (α
+
−α
−
) on `.
We also have
∂α
+
=
1
2
(∂1
k0
α −∂\
k0
) =
1
2
(∂1
k0
α −
˜
l
k0
) = 0 in Ω
+
,
and
∂α
−
= −
1
2
(∂1
k0
α +∂\
k0
) = −
1
2
(∂1
k0
α +
˜
l
k0
) = 0 in Ω.
If we choose /
0
suﬃciently large (/
0
≥ n + / + 1), then α
+
∈ C
k
(p,q)
(Ω
+
∪ `) and
α
−
∈ C
k
(p,q)
(Ω) by the Sobolev embedding theorem. The estimates (9.3.14 i) and
(9.3.14 ii) follow easily from (9.3.16) and (9.3.10). Since this is true for an arbitrarily
large ball
˜
Ω, the lemma is proved.
Using the weighted
¯
∂Neumann operator on Ω
−
, we can solve
¯
∂n
−
= α
−
in Ω
−
with good estimates up to the boundary. To solve
¯
∂ for α
+
in Ω
+
, we use the
following lemma to extend α
+
to be
¯
∂closed in
˜
Ω with good estimates.
Lemma 9.3.6. Let Ω be a bounded pseudoconvex domain in C
n
with C
∞
boundary
`. Let α ∈ C
∞
(p,q)
(`) such that ∂
b
α = 0, where 0 ≤ j ≤ n, 0 ≤ ¡ < n − 1. For
every nonnegative integer /
1
, there exists ˜ α
+
in C
k1
(p,q)
(
˜
Ω) and α
−
∈ C
k1
(p,q)
(Ω) such
that ∂˜ α
+
= 0 in
˜
Ω, ∂α
−
= 0 in Ω
−
and ˜ α
+
− α
−
= α on `. Furthermore, for
every 0 ≤ : ≤ /
1
, there exists a constant C depending only on : but independent of
α such that
(9.3.17 i)  ˜ α
+

s−
1
2
(
˜
Ω)
≤ Cα
s(M)
.
(9.3.17 ii)  α
−

s+
1
2
(Ω)
≤ Cα
s(M)
.
When ¡ = n −1, α ∈ C
∞
(p,n−1)
(`) and α satisﬁes
(9.3.18)
_
M
α ∧ φ = 0, φ ∈ C
∞
(n−p,0)
(`) ∩ Ker(
¯
∂
b
),
9.3 L
2
Existence Theorems and Sobolev Estimates for
¯
∂
b
223
the same conclusion holds.
Proof. Let / be an integer with / 2(/
1
+n) and let α
+
, α
−
be deﬁned as in Lemma
9.3.5. For any : ≥ 0 and 0 ≤ : ≤ :, using arguments similar to those in the proof
of Lemma 5.2.3, the norm [[[1
m
α
−
[[[
s−m+
1
2
(Ω
−
δ
)
is equivalent to  α
−

s+
1
2
(Ω
−
δ
)
.
Thus (9.3.17 ii) follows immediately from (9.3.14 ii).
We next extend α
+
to
˜
Ω. By the trace theorem for Sobolev spaces and inequality
(9.3.14 i), we have for any integer 0 ≤ , ≤ : −1, 0 ≤ : ≤ /,
1
j
ρ
α
+

s−j(M)
≤ C1
j+1
ρ
α
+

s−j−
1
2
(Ω
+
δ
)
≤ Cα
s(M)
.
Using the proof of Lemma 9.3.3, we can extend α
+
from Ω
+
to α
t
in
˜
Ω such that
α
t
∈ C
k
(
˜
Ω), ∂α
t
= 0 in Ω
+
and the following estimates hold: for any integer
0 ≤ : ≤ / −1, 0 ≤ : ≤ :,
(9.3.19 i) [[[1
m
α
t
[[[
s−m+
1
2
(Ω
−
δ
)
≤ Cα
s(M)
,
(9.3.19 ii) [[[Θ
m
α
t
[[[
s+
1
2
(Ω
−
δ
)
≤ Cα
s(M)
.
We deﬁne
1α
t
= −×∂·
t
(n−p,n−q−1)
×
¯
∂α
t
in Ω,
where ·
t
(n−p,n−q−1)
is the weighted
¯
∂Neumann operator on (n−j, n−¡−1)forms.
Using Theorem 6.1.4, we can choose t suﬃciently large such that 1α
t
∈ \
k−1
(p,q)
(Ω) ⊂
C
2k1+1
(p,q)
(Ω). We set 1α
t
= 0 outside Ω.
When ¡ = n − 1, using the deﬁnition of α
+
, for every
¯
∂closed form φ ∈
C
∞
(n−p,0)
(Ω),
_
Ω
¯
∂α
t
∧ φ =
_
Ω
¯
∂(α
t
∧ φ) =
_
M
α
t
∧ φ = −
1
2
_
M
\
k0
∧ φ
= −
1
2
_
Ω
˜
l
k0
∧ φ =
1
2
_
M
α ∧ φ = 0,
by (9.3.18). Thus
¯
∂α
t
satisﬁes condition (9.1.9). Using Theorem 9.1.2 (1 ≤ ¡ ≤
n −2) and Theorem 9.1.3 (¡ = n −1) , it follows that
_
¯
∂1α
t
=
¯
∂α
t
in
˜
Ω,
1α
t
= 0 in Ω
+
.
We modify 1α
t
to make it smooth at `. Setting 1
0
α
t
= 1α
t
and using argu
ments similar to those in Lemma 9.2.3, one can choose (j, ¡ −1)forms β
0
, , β
k1
and deﬁne
1
k1+1
α
t
= 1
0
α
t
−
k1
i=0
1
i + 1
¯
∂(ρ
i+1
β
i
),
such that
¯
∂1
k1+1
α
t
=
¯
∂α
t
in Ω,
224 The Tangential CauchyRiemann Complex
and
1
k1+1
α
t
= O(ρ
k1+1
) at `.
Each β
i
is obtained by taking ith derivatives of the components of 1
0
α
t
. Thus
1
k1+1
α
t
∈ C
k1
(p,q)
(Ω). If we set 1
k1+1
α
t
= 0 in
˜
Ω ¸ Ω, then 1
k1+1
α
t
∈ C
k1
(p,q)
(
˜
Ω). We
deﬁne
˜ α
+
=
_
α
t
−1
k1+1
α
t
, in Ω,
α
t
−1
k1+1
α
t
= α
+
, in
˜
Ω ¸ Ω,
then ˜ α
+
∈ C
k1
(p,q)
(
˜
Ω) and ∂˜ α
+
= 0 in
˜
Ω. It remains to show that ˜ α
+
satisﬁes (9.3.17
i). Since α
+
satisﬁes (9.3.14 i) and α
t
satisﬁes (9.3.19 i), to estimate ˜ α
+
, we only
need to estimate 1
k1+1
α
t
in Ω. To prove (9.3.17 i), it suﬃces to show
(9.3.20)  1
k1+1
α
t

s−
1
2
(Ω)
≤ C  α 
s(M)
.
From Theorem 6.1.4, we have the estimates
(9.3.21)  1α
t

s−
1
2
(Ω)
≤ C 
¯
∂α
t

s−
1
2
(Ω)
≤ C  α 
s(M)
.
We claim that for each positive integer 0 ≤ : ≤ /
1
+ 1, 0 ≤ : ≤ /
1
,
(9.3.22) [[[ρ
m
1
m
1α
t
[[[
s−
1
2
(Ω
−
δ
)
≤ C  α 
s(M)
.
If the claim is true, then (9.3.20) holds from our construction of 1
k1+1
α
t
, since
1
k1+1
α
t
can be written as combinations of terms in 1α
t
and ρ
m
1
m
1α
t
. Thus it
remains to prove (9.3.22).
(9.3.21) implies that (9.3.22) holds when : = 0. To prove the claim for : 0,
it suﬃces to show for each positive integer 0 ≤ : ≤ /
1
+ 1, 0 ≤ : ≤ /
1
,
(9.3.23) [[[1Θ
m
1α
t
[[[
s−
3
2
(Ω
−
δ
)
≤ C  α 
s(M)
since 1α
t
satisﬁes an elliptic system
¯
∂ ⊕ ϑ
t
. Decompose Ω
−
δ
into subdomains Ω
j
such that
Ω
j
= ¦. ∈ Ω
−
δ
[ δ
j+1
< −ρ(.) < δ
j
¦,
where δ
j
= δ,2
j
. This is a Whitney type decomposition where the thickness of
each Ω
j
is comparable to the distance of Ω
j
to the boundary. We deﬁne Ω
∗
j
=
Ω
j−1
∪ Ω
j
∪ Ω
j+1
. Let φ
j
be a function in C
∞
0
(Ω
∗
j
) such that 0 ≤ φ
j
≤ 1, φ
j
= 1
on Ω
j
. Moreover,
(9.3.24) sup [gradφ
j
[ ≤ Cδ
−1
j
,
where C is independent of ,. Since
¯
∂ ⊕ϑ
t
is elliptic and φ
j
Θ
m
1α
t
is supported in
Ω
∗
j
, applying G˚arding’s inequality, we have
(9.3.25)
[[[1(φ
j
Θ
m
1α
t
)[[[
s−
3
2
(Ω
∗
j
)
≤ C
_
[[[
¯
∂(φ
j
Θ
m
1α
t
)[[[
s−
3
2
(Ω
∗
j
)
+[[[ϑ
t
(φ
j
Θ
m
1α
t
)[[[
s−
3
2
(Ω
∗
j
)
+[[[φ
j
Θ
m
1α
t
[[[
s−
3
2
(Ω
∗
j
)
_
,
9.3 L
2
Existence Theorems and Sobolev Estimates for
¯
∂
b
225
where C is independent of ,. We also know that
(9.3.26)
[[[
¯
∂(φ
j
Θ
m
1α
t
)[[[
s−
3
2
(Ω
∗
j
)
≤ [[[
¯
∂(φ
j
)Θ
m
1α
t
[[[
s−
3
2
(Ω
∗
j
)
+[[[φ
j
Θ
m
¯
∂1α
t
[[[
s−
3
2
(Ω
∗
j
)
+[[[φ
j
[
¯
∂, Θ
m
]1α
t
[[[
s−
3
2
(Ω
∗
j
)
,
and using (9.3.24),
(9.3.27)
[[[
¯
∂(φ
j
)Θ
m
1α
t
[[[
s−
3
2
(Ω
∗
j
)
+[[[φ
j
[
¯
∂, Θ
m
]1α
t
[[[
s−
3
2
(Ω
∗
j
)
≤ C[[[1Θ
m−1
1α
t
[[[
s−
3
2
(Ω
∗
j
)
.
Substituting (9.3.27) into (9.3.26), we obtain
(9.3.28)
[[[
¯
∂(φ
j
Θ
m
1α
t
)[[[
s−
3
2
(Ω
∗
j
)
≤ C([[[1Θ
m−1
1α
t
[[[
s−
3
2
(Ω
∗
j
)
+[[[Θ
m
¯
∂α
t
[[[
s−
3
2
(Ω
∗
j
)
)
≤ C([[[1Θ
m−1
1α
t
[[[
s−
3
2
(Ω
∗
j
)
+[[[1Θ
m
α
t
[[[
s−
3
2
(Ω
∗
j
)
).
Similarly,
(9.3.29)
[[[ϑ
t
(φ
j
Θ
m
1α
t
)[[[
s−
3
2
(Ω
∗
j
)
≤ [[[ϑ
t
(φ
j
)Θ
m
1α
t
[[[
s−
3
2
(Ω
∗
j
)
+[[[φ
j
Θ
m
ϑ
t
1α
t
[[[
s−
3
2
(Ω
∗
j
)
+[[[φ
j
[ϑ
t
, Θ
m
]1α
t
[[[
s−
3
2
(Ω
∗
j
)
≤ C[[[1Θ
m−1
1α
t
[[[
s−
3
2
(Ω
∗
j
)
.
Substituting (9.3.28) and (9.3.29) into (9.3.25) and summing over ,, we have using
induction,
[[[1Θ
m
1α
t
[[[
s−
3
2
(Ω
−
δ
2
)
≤ C([[[1Θ
m−1
1α
t
[[[
s−
3
2
(Ω
−
δ
)
+[[[Θ
m
¯
∂α
t
[[[
s−
3
2
(Ω
−
δ
)
)
≤ C  α 
s(M)
.
This proves (9.3.23) for a smaller δ. Thus (9.3.17 i) holds. The proof of Lemma
9.3.6 is complete.
We note that both Lemma 9.3.5 and Lemma 9.3.6 hold for ¡ = 0, i.e., when α is
C1. When ¡ = 0 and n = 1, Lemma 9.3.5 corresponds to the Plemelj jump formula
in one complex variable (see Theorem 2.1.3). In this case, there is no condition
on α since
¯
∂
b
α = 0 on ` is always satisﬁed. In contrast, there is a compatibility
condition (9.3.18) for ¡ = n−1 in Lemma 9.3.6 for n 1. One should compare this
case with Corollary 2.1.4.
Using Lemma 9.3.6, we have the following lemma for smooth
¯
∂
b
closed forms:
226 The Tangential CauchyRiemann Complex
Lemma 9.3.7. Let Ω be a bounded pseudoconvex domain in C
n
with smooth bound
ary `. Let α ∈ C
∞
(p,q)
(`), where 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n − 2, such that
¯
∂
b
α = 0
on `. For every nonnegative integer :, there exists n
s
∈ \
s
(p,q−1)
(`) satisfying
¯
∂
b
n
s
= α on `. Furthermore, there exists a constant C
s
independent of α such
that
(9.3.30) n
s

s(M)
≤ C
s
α
s(M)
.
When ¡ = n − 1, α ∈ C
∞
(p,n−1)
(`) and α satisﬁes (9.3.3), the same conclusion
holds.
Proof. Let : be a ﬁxed nonnegative integer. From Lemma 9.3.6, there exists a
decomposition α = (˜ α
+
− α
−
) such that ˜ α
+
∈ C
s
(p,q)
(
˜
Ω),
¯
∂˜ α
+
= 0 in
˜
Ω, α
−
∈
C
s
(p,q)
(Ω) and
¯
∂α
−
= 0 in Ω. We also have the estimates:
(9.3.31 i) ˜ α
+

s−
1
2
(
˜
Ω)
≤ Cα
s(M)
,
(9.3.31 ii) α
−

s+
1
2
(Ω)
≤ Cα
s(M)
.
(9.3.31 i) and (9.3.31 ii) follow from (9.3.17 i) and (9.3.17 ii).
Since Ω is pseudoconvex, we deﬁne
n
−
= ∂
∗
t
·
t
(p,q)
α
−
,
where ·
t
(p,q)
is the weighted
¯
∂Neumann operator in Ω. If we choose t 0 suf
ﬁciently large, using Theorem 6.1.4, it follows that n
−
∈ \
s+
1
2
(p,q−1)
(Ω), ∂n
−
= α
−
and
[[[1n
−
[[[
s−
1
2
(Ω
−
δ
)
≤ Cα
−

s+
1
2
(Ω)
≤ Cα
s(M)
for some constant C independent of α. Restricting n
−
to the boundary we have
¯
∂
b
n
−
= τα
−
on ` and using the trace theorem for Sobolev spaces, we obtain
(9.3.32) n
−

s(M)
≤ C
s
[[[1n
−
[[[
s−
1
2
(Ω
−
δ
)
≤ C
s
α
s(M)
.
Deﬁning
˜ n
+
= ∂
∗
·
˜
Ω
(p,q)
˜ α
+
,
where ·
˜
Ω
(p,q)
is the
¯
∂Neumann operator on
˜
Ω, we have
¯
∂˜ n
+
= ˜ α
+
in
˜
Ω and ˜ n
+
is
one derivative smoother than ˜ α
+
in the interior of
˜
Ω. From (9.3.31 i),
[[[1˜ n
+
[[[
s−
1
2
(Ω
δ
)
≤ C˜ α
+

s−
1
2
(
˜
Ω)
≤ Cα
s(M)
for some constant C independent of α. Restricting ˜ n
+
to `, using the trace theorem
again,
(9.3.33) ˜ n
+

s(M)
≤ C[[[1˜ n
+
[[[
s−
1
2
(Ω
δ
)
≤ Cα
s(M)
for some constant C independent of α. Letting
n
s
= ˜ n
+
−n
−
on `,
we get ∂
b
n
s
= α on `. We also have from (9.3.32) and (9.3.33),
n
s

s(M)
≤ C
s
α
s(M)
,
where C
s
is independent of α. This proves (9.3.30) and the lemma.
To prove Theorem 9.3.1, we need the following density lemmas which are of
interest independently.
9.3 L
2
Existence Theorems and Sobolev Estimates for
¯
∂
b
227
Lemma 9.3.8. Let ` be the boundary of a smooth bounded pseudoconvex domain
Ω in C
n
, n ≥ 2. For each 0 ≤ j ≤ n, 0 ≤ ¡ < n − 1 and : ≥ 0, the space
C
∞
(p,q)
(`) ∩ ker(
¯
∂
b
) is dense in \
s
(p,q)
(`) ∩ ker(
¯
∂
b
) in the \
s
(p,q)
(`) norm.
Proof. We deﬁne Z
∞
= C
∞
(p,q)
(`) ∩ ker(
¯
∂
b
) and Z
s
= \
s
(p,q)
(`) ∩ ker(
¯
∂
b
). For
any α ∈ Z
s
, using Friedrichs’ Lemma (see Appendix D), there exists a sequence of
smooth forms α
m
∈ C
∞
(p,q)
(`) such that α
m
→ α in \
s
(p,q)
(`) and
¯
∂
b
α
m
→ 0 in
\
s
(p,q+1)
(`). Since
¯
∂
b
α
m
is a smooth form satisfying the compatibility condition
(9.3.2) (when ¡ < n − 2) and (9.3.3) (when ¡ = n − 2), Lemma 9.3.7 implies that
there exists a suﬃciently smooth form ·
m
such that
¯
∂
b
·
m
=
¯
∂
b
α
m
in ` with
·
m

s(M)
≤ C
s

¯
∂
b
α
m

s(M)
→0.
We set
α
t
m
= α
m
−·
m
.
Then
¯
∂
b
α
t
m
= 0 and α
t
m
converges to α in \
s
(p,q)
(`). Thus, Z
k
is dense in Z
s
in
the \
s
(p,q)
(`) norm where / is an arbitrarily large integer.
For any c 0 and each positive integer / :, there exists an α
k
∈ Z
k
such that
α
k
−α
s(M)
< c.
Furthermore, we can require that
α
k
−α
k+1

k(M)
<
c
2
k
since Z
k+1
is dense in Z
k
. The series
α
k
+
∞
N=k+1
(α
N
−α
N−1
)
converges in every \
k
(p,q)
(`) norm to some element α
∞
. The Sobolev embedding
theorem then assures that α
∞
is in C
∞
(p,q)
(`) ∩ Ker(
¯
∂
b
). We also have
α
∞
−α
s(M)
< 2c.
This proves the lemma.
When ¡ = n −1, we have the density lemma in the 1
2
norm.
Lemma 9.3.9. Let ` be the boundary of a smooth bounded pseudoconvex domain
Ω in C
n
, n ≥ 2 and 0 ≤ j ≤ n. Let Z denote the space of all forms in 1
2
(p,n−1)
(`)
satisfying (9.3.3) and Z
∞
be the subspace of all forms in C
∞
(p,n−1)
(`) satisfying
(9.3.3). Then Z
∞
is dense in Z in the 1
2
norm.
Proof. Since the holomorphic degree j plays no role, for simplicity we assume j = n.
If α ∈ 1
2
(n,n−1)
(`), we can write α = )(×
¯
∂ρ) for some ) ∈ 1
2
(`). Using the
HahnBanach theorem, it suﬃces to show that any bounded linear functional / on
228 The Tangential CauchyRiemann Complex
1
2
(n,n−1)
(`) that vanishes on Z
∞
also vanishes on Z. From the Riesz representation
theorem, there exists a p ∈ 1
2
(`) such that / can be written as
/(α) =
_
M
α ∧ p, α ∈ 1
2
(n,n−1)
(`).
For any n ∈ C
∞
(n,n−2)
(`), it is easy to see that
¯
∂
b
n ∈ Z
∞
. If / vanishes on Z
∞
, we
have _
M
¯
∂
b
n ∧ p = 0, for any n ∈ C
∞
(n,n−2)
(`).
This implies that
¯
∂
b
p = 0 in the distribution sense. Using Lemma 9.3.8 when
j = ¡ = 0, there exists a sequence of smooth functions p
m
such that
¯
∂
b
p
m
= 0 and
p
m
→p in 1
2
(`). For any α ∈ Z, we have
/(α) =
_
M
α ∧ p = lim
m→∞
_
M
α ∧ p
m
= 0.
This proves the lemma.
We can now ﬁnish the proof of Theorem 9.3.1.
Proof of Theorem 9.3.1. When 1 ≤ ¡ < n − 1, α can be approximated by smooth
¯
∂
b
closed forms α
m
in \
s
(p,q)
(`) according to Lemma 9.3.8. For each α
m
, we apply
Lemma 9.3.7 to obtain a (j, ¡ −1)form n
m
such that
¯
∂
b
n
m
= α
m
and
n
m

s(M)
≤ C
s
α
m

s(M)
.
Thus, n
m
converges to some (j, ¡ −1)form n such that
¯
∂
b
n = α on ` and
n
s(M)
≤ C
s
α
s(M)
.
This proves the theorem when 1 ≤ ¡ < n −1.
When ¡ = n − 1, we approximate α by α
m
∈ Z
∞
in the 1
2
(p,n−1)
(`) norm
using Lemma 9.3.9. Repeating the arguments above with : = 0, we can construct
n ∈ 1
2
(p,n−2)
(`) with
¯
∂
b
n = α. This completes the proof of Theorem 9.3.1.
Corollary 9.3.2 follows easily from Theorem 9.3.1.
9.4 The Hodge Decomposition Theorem for
¯
∂
b
The 1
2
existence result proved in Theorem 9.3.1 can be applied to prove the
Hodge decomposition theorem for the
¯
∂
b
complex on pseudoconvex boundaries. We
use the notation
¯
∂
∗
b
to denote the Hilbert space adjoint of the operator
¯
∂
b
with
respect to the induced metric on `.
We deﬁne
b
as in Chapter 8. Let
b
= ∂
b
∂
∗
b
+ ∂
∗
b
∂
b
be deﬁned on Dom(
b
),
where
Dom(
b
) = ¦φ ∈ 1
2
(p,q)
(`) [ φ ∈ Dom(∂
b
) ∩ Dom(∂
∗
b
);
∂
b
φ ∈ Dom(∂
∗
b
) and ∂
∗
b
φ ∈ Dom(∂
b
)¦.
9.4 The Hodge Decomposition Theorem for
¯
∂
b
229
Repeating arguments in the proof of Proposition 4.2.3, we can show that
b
is a
closed, densely deﬁned selfadjoint operator.
We use H
b
(p,q)
to denote the projection
H
b
(p,q)
: 1
2
(p,q)
(`) →Ker(
b
) = Ker(
¯
∂
b
) ∩ Ker(
¯
∂
∗
b
).
When ` is the boundary of a smooth bounded pseudoconvex domain, we claim
that for 1 ≤ ¡ ≤ n −2,
(9.4.1) Ker(
b
) = Ker(
¯
∂
b
) ∩ Ker(
¯
∂
∗
b
) = 0.
To prove (9.4.1), let α ∈ Ker(
b
) = Ker(
¯
∂
b
) ∩ Ker(
¯
∂
∗
b
). Then α =
¯
∂
b
n for some
n ∈ 1
2
(p,q−1)
(`) by Theorem 9.3.1. Thus
(α, α) = (
¯
∂
b
n, α) = (n,
¯
∂
∗
b
α) = 0.
We have H
b
(p,q)
= 0 for all 0 ≤ j ≤ n, 1 ≤ ¡ ≤ n − 2. Only H
b
(p,0)
≡ o
(p,0)
(the
Szeg¨o projection) and H
b
(p,n−1)
≡
˜
o
(p,n−1)
are nontrivial where
o
(p,0)
: 1
2
(p,0)
(`) →Ker(
¯
∂
b
),
˜
o
(p,n−1)
: 1
2
(p,n−1)
(`) →Ker(
¯
∂
∗
b
).
We derive some equivalent conditions for (9.3.3) in Theorem 9.3.1. Again one can
assume j = n. Let o = o
(0,0)
denote the Szeg¨o projection on functions and
˜
o denote
the projection from 1
2
(n,n−1)
(`) onto 1
2
(n,n−1)
∩Ker(
¯
∂
∗
b
). For any α ∈ 1
2
(n,n−1)
(`),
we can write α = )(×
¯
∂ρ) for some ) ∈ 1
2
(`), where × is the Hodge star operator
with respect to the standard metric in C
n
. The following lemma links condition
(9.3.3) to the Szeg¨o projection:
Lemma 9.4.1. For any α ∈ 1
2
(n,n−1)
(`), the following conditions are equivalent:
(1) α satisﬁes condition (9.3.3).
(2) o
¯
) = 0, where α = )(×
¯
∂ρ).
(3)
˜
oα = 0.
Proof. Let θ = ×(∂ρ ∧
¯
∂ρ). Since dρ = (∂ρ +
¯
∂ρ) vanishes when restricted to `, we
have
×∂ρ = θ ∧
¯
∂ρ = θ ∧ dρ −θ ∧ ∂ρ
= θ ∧ dρ +×
¯
∂ρ = ×
¯
∂ρ
on `. If ψ ∈ 1
2
(`),
_
M
α ∧ ψ =
_
M
)ψ(×
¯
∂ρ) =
_
M
)ψ(×dρ) −
_
M
)ψ(×∂ρ)
=
_
M
)ψdσ −
_
M
)ψ(×
¯
∂ρ),
230 The Tangential CauchyRiemann Complex
where dσ = ×dρ is the surface measure on `. Hence, for any ψ ∈ 1
2
(`),
_
M
α ∧ ψ =
1
2
_
M
)ψdσ =
1
2
(),
¯
ψ)
M
.
Since Z
∞
is dense in Z from Lemma 9.3.8, (1) and (2) are equivalent.
To prove that (3) and (1) are equivalent, we write β = p(×
¯
∂ρ) for any β ∈
1
2
(n,n−1)
(`), where p ∈ 1
2
(`). It is easy to see that β ∈ Ker(
¯
∂
∗
b
) if and only if
¯
∂
b
¯ p = 0 on `. For any β ∈ 1
2
(n,n−1)
(`) ∩ Ker(
¯
∂
∗
b
),
(α, β)
M
=
_
M
)¯ pdσ = 2
_
M
α ∧ ¯ p = 0.
This proves that (1) and (3) are equivalent.
We have the following strong Hodge decomposition theorem for
¯
∂
b
.
Theorem 9.4.2. Let ` be the boundary of a smooth bounded pseudoconvex domain
Ω in C
n
, n ≥ 2. Then for any 0 ≤ j ≤ n, 0 ≤ ¡ ≤ n − 1, there exists a linear
operator ·
b
: 1
2
(p,q)
(`) →1
2
(p,q)
(`) such that
(1) ·
b
is bounded and ¹(·
b
) ⊂ Dom(
b
).
(2) For any α ∈ 1
2
(p,q)
(`), we have
α = ∂
b
∂
∗
b
·
b
α ⊕∂
∗
b
∂
b
·
b
α, if 1 ≤ ¡ ≤ n −2,
α = ∂
∗
b
∂
b
·
b
α ⊕o
(p,0)
α, if ¡ = 0,
α = ∂
b
∂
∗
b
·
b
α ⊕
˜
o
(p,n−1)
α, if ¡ = n −1.
(3) If 1 ≤ ¡ ≤ n −2, we have
·
b
b
=
b
·
b
= 1 on Dom(
b
),
∂
b
·
b
= ·
b
∂
b
on Dom(∂
b
),
∂
∗
b
·
b
= ·
b
∂
∗
b
on Dom(∂
∗
b
).
(4) If α ∈ 1
2
(p,q)
(`) with ∂
b
α = 0, where 1 ≤ ¡ ≤ n − 2 or α ∈ 1
2
(p,n−1)
(`)
with
˜
o
(p,n−1)
α = 0, then α = ∂
b
∂
∗
b
·
b
α.
The solution n = ∂
∗
b
·
b
α in (4) is called the canonical solution, i.e., the unique
solution orthogonal to Ker(∂
b
).
Proof. From Corollary 9.3.2, the range of
¯
∂
b
, denoted by ¹(
¯
∂
b
), is closed in every
degree. If 1 ≤ ¡ ≤ n − 2, we have from Theorem 9.3.1, Ker(
¯
∂
b
) = ¹(
¯
∂
b
) and the
following orthogonal decomposition:
(9.4.2) 1
2
(p,q)
(`) = Ker(
¯
∂
b
) ⊕¹(
¯
∂
∗
b
) = ¹(
¯
∂
b
) ⊕¹(
¯
∂
∗
b
).
Repeating the arguments of Theorem 4.4.1, we can prove that for every α ∈
Dom(∂
b
) ∩ Dom(∂
∗
b
),
(9.4.3) α
2
≤ c(∂
b
α
2
+∂
∗
b
α
2
),
Notes 231
and for any α ∈ Dom(
b
),
(9.4.4) α
2
≤ c
b
α, 
2
where the constant c is independent of α.
(9.4.4) implies that
b
is onetoone and, from Lemma 4.1.1, that the range of
b
is closed. It follows that the strong Hodge decomposition holds:
1
2
(p,q)
(`) = ¹(
b
) ⊕Ker(
b
)
=
¯
∂
b
¯
∂
∗
b
(Dom(
b
)) ⊕
¯
∂
∗
b
¯
∂
b
(Dom(
b
)).
Thus
b
: Dom(
b
) → 1
2
(p,q)
(`) is onetoone, onto, and it has a unique inverse
·
b
: 1
2
(p,q)
(`) → Dom(
b
). Note that ·
b
is bounded. Following the same argu
ment as in Theorem 4.4.1, we have that ·
b
satisﬁes all the conditions (1)(4) and
Theorem 9.4.2 is proved when 1 ≤ ¡ ≤ n −2.
When ¡ = 0,
1
2
(p,0)
(`) = ¹(
¯
∂
∗
b
) ⊕Ker
¯
∂
b
.
Thus for any α ⊥ Ker(
¯
∂
b
) and α ∈ Dom(
b
),
(9.4.5) α
2
≤ c
¯
∂
b
α
2
≤ c
b
αα.
Thus
b
has closed range on Ker(
¯
∂
b
)
⊥
= Ker(
b
)
⊥
and
1
2
(p,0)
(`) = ¹(
b
) ⊕Ker
¯
∂
b
.
In particular, there exists a bounded operator ·
b
: 1
2
(p,0)
(`) →1
2
(p,0)
(`) satisfying
b
·
b
= 1 − o
(p,0)
, ·
b
= 0 on Ker(
¯
∂
b
). This proves (1) and (2) when ¡ = 0.
Properties (3) and (4) also follow exactly as before. The case for ¡ = n−1 can also
be proved similarly.
Thus, the strong Hodge decomposition for ∂
b
holds on the boundary of a smooth
bounded pseudoconvex domain in C
n
for all (j, ¡)forms including ¡ = 0 and ¡ =
n −1.
NOTES
The
¯
∂closed extension of
¯
∂
b
closed functions or forms from the boundary of a
domain in a complex manifold was studied by J. J. Kohn and H. Rossi [KoRo 1] who
ﬁrst introduced the
¯
∂
b
complex. In [KoRo 1], they show that a
¯
∂closed extension
exists for any (j, ¡)form from the boundary ` to the domain Ω in a complex
manifold if Ω satisﬁes condition 7(n −¡ −1). Formula (9.1.3) was ﬁrst given there.
The 1
2
Cauchy problem on any pseudoconvex domain was used by M.C. Shaw
[Sha 6] to study the local solvability for
¯
∂
b
. Theorem 9.3.3 was ﬁrst observed by
J. P. Rosay in [Rosa 1] where it was pointed out that global smooth solutions can
be obtained by combining the results of Kohn [Koh 6] and KohnRossi [KoRo 1].
The
¯
∂ Cauchy problem was also discussed by M. Derridj in [Der 1,2]. A. Andreotti
232 The Tangential CauchyRiemann Complex
and C. D. Hill used reduction to vanishing cohomology arguments to study the
Cauchy problem and
¯
∂
b
in [AnHi 1]. Kernel methods were also used to obtain
¯
∂
closed extension from boundaries of domains satisfying condition 7(n −¡ −1) (see
HenkinLeiterer [HeLe 2]). We mention the papers of G. M. Henkin [Hen 3] and
H. Skoda [Sko 1] where solutions of
¯
∂
b
, including the top degree case, on strongly
pseudoconvex boundaries were studied using integral kernel methods.
Much of the material in Section 9.3 on the 1
2
theory of
¯
∂
b
on weakly pseudocon
vex boundaries was based on the work of M.C. Shaw [Sha 2] and H. P. BoasM.C.
Shaw [BoSh 1]. In [Sha 1], Kohn’s results of the weighted
¯
∂Neumann operator
on a pseudoconvex domain was extended to an annulus between two pseudoconvex
domains. Using the weighted
¯
∂Neumann operators constructed in [Koh 6] and [Sha
1], a twosided
¯
∂closed extension for
¯
∂
b
closed forms away from the top degree was
constructed in [Sha 2]. The jump formula proved in Lemma 9.3.5 was derived from
the BochnerMartinelliKoppelman kernel in [BoSh 1] (c.f. Theorem 11.3.1). Our
proof of Lemma 9.3.5 presented here uses an idea of [AnHi 1]. Sobolev estimates
for
¯
∂
b
were also obtained for the top degree case (¡ = n − 1) in [BoSh 1]. The
proof depends on the regularity of the weighted Szeg¨o projection in Sobolev spaces.
For more discussion on the Sobolev estimates for the Szeg¨o projection, see [Boa
1,2,4]. Another proof of Theorem 9.3.1 was given by J. J. Kohn in [Koh 11] using
pseudodiﬀerential operators and microlocal analysis.
We point out that all results discussed in this chapter can be generalized to any
C1 manifolds which are boundaries of domains in complex manifolds, as long as the
corresponding
¯
∂Neumann operators (or weighted
¯
∂Neumann operators) exist and
are regular (e.g, pseudoconvex domains in a Stein manifold). However, 1
2
existence
theorems and the closed range property for
¯
∂
b
might not be true for abstract C1
manifolds. It was observed by D. Burns [Bur 1] that the range of
¯
∂
b
is not closed in
1
2
on a nonembeddable strongly pseudoconvex C1 manifold of real dimension three
discovered by H. Rossi [Ros 1]. This example along with the interplay between the
closedrange property of
¯
∂
b
and the embedding problem of abstract C1 structures
will be discussed in Chapter 12.
There are also results on Sobolev estimates for
b
on pseudoconvex manifolds.
Using subelliptic multipliers combining with microlocal analysis, J. J. Kohn (see
[Koh 10,11]) has proved subelliptic estimates for
b
when the C1 manifold is pseu
doconvex and of ﬁnite ideal type. If the domain has a plurisubharmonic deﬁning
function, Sobolev estimates for
b
have been obtained by H. P. Boas and E. J.
Straube [BoSt 4]. In particular, the Szeg¨o projections are exactly regular in both
cases.
Much less is known for the regularity of
b
and the Szeg¨o projection on pseu
doconvex manifolds in other function spaces except when the C1 manifold is the
boundary of a pseudoconvex domain of ﬁnite type in C
2
. We mention the work
of FeﬀermanKohn [FeKo 1] and NagelRosaySteinWainger [NRSW 1] and Christ
[Chr 1]. When the C1 manifold is of ﬁnite type in C
n
, n ≥ 3, and the Levi form
is diagonalizable, H¨older estimates for
¯
∂
b
and
b
have been obtained in Feﬀerman
KohnMachedon [FKM 1]. H¨older and 1
p
estimates for the Szeg¨o projection on
convex domains of ﬁnite type have been obtained by J. McNeal and E. M. Stein
[McSt 2]. It is still an open question whether H¨older or 1
p
estimates hold for
b
and
¯
∂
b
on general pseudoconvex C1 manifolds of ﬁnite type.
233
234
CHAPTER 10
FUNDAMENTAL SOLUTIONS FOR
b
ON THE HEISENBERG GROUP
In Chapters 8 and 9, we have proved the global solvability and regularity for
the
b
operator on compact pseudoconvex C1 manifolds. Under condition Y (¡),
subelliptic 1,2estimates for
b
were obtained in Chapter 8. On the other hand,
it was shown in Section 7.3 that the Lewy operator, which arises from the tan
gential CauchyRiemann operator associated with the Siegel upper half space, does
not possess a solution locally in general. The main task of this chapter is to con
struct a fundamental solution for the
b
operator on the Heisenberg group H
n
.
The Heisenberg group serves as a model for strongly pseudoconvex C1 manifolds
(or nondegenerate C1 manifolds). Using the group structure, we can construct
explicitly a solution kernel for ∂
b
and obtain estimates for the solutions in H¨older
spaces. The CauchySzeg¨o kernel on H
n
is discussed in Section 10.2. We construct
a relative fundamental solution for
b
in the top degree case and deduce from it the
necessary and suﬃcient conditions for the local solvability of the Lewy operator.
10.1 Fundamental Solutions for
b
on the Heisenberg Group
Let us recall that the Siegel upper half space Ω
n
is deﬁned by
(10.1.1) Ω
n
= ¦(.
t
, .
n
) ∈ C
n
[ Im.
n
[.
t
[
2
¦,
where .
t
= (.
1
, , .
n−1
) and [.
t
[
2
= [.
1
[
2
+ +[.
n−1
[
2
. Denote by Aut(Ω
n
) the
group of all holomorphic mappings that are onetoone from Ω
n
onto itself. Let
H
n
⊂ Aut(Ω
n
) be the subgroup deﬁned by
(10.1.2) H
n
= ¦/
a
; o ∈ /Ω
n
[ /
a
(.) = (o
t
+.
t
, o
n
+.
n
+ 2i¸.
t
, o
t
))¦,
where ¸.
t
, o
t
) is the standard inner product in C
n−1
, i.e., ¸.
t
, o
t
) =
n−1
i=1
.
i
¯ o
i
. To see
H
n
actually forms a subgroup of Aut(Ω
n
), put /
a
(.) = (n
t
, n
n
). Since Imo
n
= [o
t
[
2
,
we have
Imn
n
−[n
t
[
2
= Im.
n
−[.
t
[
2
.
Hence, each /
a
maps Ω
n
into Ω
n
and /Ω
n
into /Ω
n
. It is easily veriﬁed that if
o, / ∈ /Ω
n
, then /
a
◦ /
b
= /
c
with c = /
a
(/). It follows that if / = (−o
t
, −o
n
),
then /
a
◦ /
b
= the identity mapping. Thus, H
n
is indeed a subgroup of Aut(Ω
n
)
10.1 Fundamental Solutions for
b
on the Heisenberg Group 235
and it induces a group structure on the boundary /Ω
n
. The boundary /Ω
n
can be
identiﬁed with H
n
= C
n−1
1 via the mapping
(10.1.3) π : (.
t
, t +i[.
t
[
2
) → (.
t
, t),
where .
n
= t +i:. We shall call H
n
= C
n−1
1 the Heisenberg group of order n−1
with the group structure induced from the automorphism subgroup H
n
of Aut(Ω
n
)
by
(10.1.4) (.
t
1
, t
1
) (.
t
2
, t
2
) = (.
t
1
+.
t
2
, t
1
+t
2
+ 2Im¸.
t
1
, .
t
2
)).
It is easily veriﬁed that
7
j
=
∂
∂.
j
+i.
j
∂
∂t
, , = 1, , n −1, and
T =
∂
∂t
,
are left invariant vector ﬁelds with respect to the Lie group structure on H
n
such
that
(10.1.5) [7
j
, 7
j
] = −2iT, for , = 1, , n −1,
and that all other commutators vanish. Hence, H
n
is a strongly pseudoconvex
C1 manifold with type (1, 0) vector ﬁelds spanned by 7
1
, , 7
n−1
. We ﬁx a left
invariant metric on H
n
so that 7
1
, , 7
n−1
, 7
1
, , 7
n−1
and T are orthonormal
with respect to this metric. Let the dual basis be ω
1
, , ω
n−1
, ω
1
, , ω
n−1
and
τ, where ω
j
= dr
j
+idj
j
, , = 1, , n −1 and τ is given by
τ = dt + 2
n−1
j=1
(r
j
dj
j
−j
j
dr
j
).
Hence, ¸dr
j
, dr
j
) = ¸dj
j
, dj
j
) = 1,2, for , = 1, , n −1, and the volume element
is
d\ = 2
1−n
dr
1
dr
n−1
dj
1
dj
n−1
dt.
Next we calculate
b
on the (j, ¡)forms of the Heisenberg group H
n
. Since j
plays no role in the formulation of the ∂
b
and ∂
∗
b
operators, we may assume that
j = 0. Let ) ∈ C
∞
(0,q)
(H
n
) be a smooth (0, ¡)form with compact support on H
n
.
Write ) as
) =
t
]J]=q
)
J
ω
J
,
where J = (,
1
, , ,
q
) is an increasing multiindex and ω
J
= ω
j1
∧ ∧ ω
jq
. Then,
we have
∂
b
) =
t
]J]=q
_
k
7
k
)
J
ω
k
_
∧ ω
J
=
t
]L]=q+1
_
t
k,J
c
L
kJ
7
k
)
J
_
ω
L
,
236 Fundamental Solutions for
b
on the Heisenberg Group
and
∂
∗
b
) = −
t
]H]=q−1
_
t
l,J
c
J
lH
7
l
)
J
_
ω
H
.
It follows that
∂
∗
b
∂
b
) = −
t
]Q]=q
_
t
l,L
c
L
lQ
7
l
_
t
k,J
c
L
kJ
7
k
)
J
__
ω
Q
,
and
∂
b
∂
∗
b
) = −
t
]Q]=q
_
t
k,H
c
Q
kH
7
k
_
t
l,J
c
J
lH
7
l
)
J
__
ω
Q
.
For ﬁxed Q and  ,= /, it is easily veriﬁed that
c
L
lQ
c
L
kJ
= −c
Q
kH
c
J
lH
,
and
[7
l
, 7
k
] = 0.
Hence, we obtain
b
) = (∂
b
∂
∗
b
+∂
∗
b
∂
b
)
_
t
]J]=q
)
J
ω
J
_
= −
t
]J]=q
__
k/ ∈J
7
k
7
k
+
k∈J
7
k
7
k
_
)
J
_
ω
J
.
The calculation shows that
b
acts on a (0, ¡)form ) diagonally. It is also easily
veriﬁed that
−
_
k/ ∈J
7
k
7
k
+
k∈J
7
k
7
k
_
= −
1
2
n−1
k=1
(7
k
7
k
+7
k
7
k
) +i(n −1 −2¡)T.
Therefore, to invert
b
it suﬃces to invert the operators
(10.1.6) −
1
2
n−1
k=1
(7
k
7
k
+7
k
7
k
) +i(n −1 −2¡)T.
In particular, when n = 2 and ¡ = 0,
b
acts on functions and
b
= −77
= −
1
2
(77 +77) +iT,
where 7 is the Lewy operator. Hence,
b
in general is not locally solvable.
10.1 Fundamental Solutions for
b
on the Heisenberg Group 237
However, we shall investigate the solvability and regularity of
b
via the following
more general operator L
α
deﬁned by
(10.1.7) L
α
= −
1
2
n−1
k=1
(7
k
7
k
+7
k
7
k
) +iαT,
for α ∈ C.
The second order term L
0
= −
1
2
n−1
k=1
(7
k
7
k
+7
k
7
k
) is usually called the sub
Laplacian on a stratiﬁed Lie group. By deﬁnition, a Lie group is stratiﬁed if it is
nilpotent and simply connected and its Lie algebra g admits a vector space decom
position g = \
1
⊕ ⊕\
m
such that [\
1
, \
j
] = \
j+1
for , < : and [\
1
, \
m
] = ¦0¦.
The Heisenberg group H
n
is a step two nilpotent Lie group, namely, the Lie algebra
is stratiﬁed with : = 2, where \
1
is generated by 7
1
, , 7
n−1
, 7
1
, , 7
n−1
and
\
2
is generated by T. By Theorems 8.2.3 and 8.2.5, L
0
satisﬁes a subelliptic esti
mate of order 1,2 and is hypoelliptic. We want to construct an explicit fundamental
solution ϕ
0
for L
0
. The group structure on the Heisenberg group suggests that one
can deﬁne a nonisotropic dilation in the following way: for o 0,
o(.
t
, t) = (o.
t
, o
2
t)
which forms an oneparameter subgroup of Aut(H
n
). We also deﬁne a norm on H
n
by [(.
t
, t)[ = ([.
t
[
4
+t
2
)
1
4
to make it homogeneous of degree one with respect to the
nonisotropic dilation.
By following the harmonic analysis on real Euclidean spaces, it is reasonable to
guess that a fundamental solution ϕ
0
for L
0
should be given by some negative power
of [(.
t
, t)[, and that the power should respect the nonisotropic dilation on H
n
. In
fact, we have the following theorem:
Theorem 10.1.1. Let ϕ
0
(.
t
, t) = [(.
t
, t)[
−2(n−1)
= ([.
t
[
4
+t
2
)
−
n−1
2
. Then L
0
ϕ
0
=
c
0
δ, where δ is the Dirac function at the origin and c
0
is given by
(10.1.8) c
0
= (n −1)
2
_
Hn
(([.
t
[
2
+ 1)
2
+t
2
)
−
n+1
2
d\.
Proof. Deﬁne, for c 0,
ϕ
0,
(.
t
, t) = (([.
t
[
2
+c
2
)
2
+t
2
)
−
n−1
2
.
Then, a simple calculation shows that
L
0
ϕ
0,
(.
t
, t) = (n −1)
2
c
2
(([.
t
[
2
+c
2
)
2
+t
2
)
−
n+1
2
= c
−2n
(n −1)
2
__¸
¸
¸
¸
.
t
c
¸
¸
¸
¸
2
+ 1
_
2
+
_
t
c
2
_
2
_
−
n+1
2
= c
−2n
φ
_
1
c
(.
t
, t)
_
,
238 Fundamental Solutions for
b
on the Heisenberg Group
where φ(.
t
, t) = (n − 1)
2
(([.
t
[
2
+ 1)
2
+ t
2
)
−
n+1
2
. Then, by integration on H
n
, we
obtain the following:
_
Hn
L
0
ϕ
0,
(.
t
, t) d\ =
_
Hn
c
−2n
φ
_
1
c
(.
t
, t)
_
d\
=
_
Hn
φ(.
t
, t) d\ = c
0
.
Hence, lim
→0
L
0
ϕ
0,
= c
0
δ in the distribution sense. On the other hand, L
0
ϕ
0,
also tends to L
0
ϕ
0
in the distribution sense. This proves the theorem.
It follows that c
−1
0
ϕ
0
is the fundamental solution for L
0
. We now proceed to
search for a fundamental solution ϕ
α
for L
α
with α ∈ C. Observe that L
α
has
the same homogeneity properties as L
0
with respect to the nonisotropic dilation on
H
n
, and that L
α
is invariant under unitary transformation in .
t
variable. Hence,
we can expect that certain ϕ
α
will have the same invariant properties. From these
observations we intend to look for a fundamental solution ϕ
α
of the form
ϕ
α
(.
t
, t) = [(.
t
, t)[
−2(n−1)
)(t[(.
t
, t)[
−2
).
After a routine, but lengthy, calculation, we see that ) must satisfy the following
ordinary second order diﬀerential equation:
(10.1.9)
(1 −n
2
)
3
2
)
tt
(n)−(nn(1 −n
2
)
1
2
+iα(1 −n
2
)))
t
(n)
+i(n −1)αn)(n) = 0
with n = t[(.
t
, t)[
−2
.
By setting n = cosθ, )(n) = p(θ), 0 ≤ θ ≤ π, (10.1.9) is reduced to
(10.1.10)
_
sinθ
d
dθ
+ (n −1)cosθ
__
d
dθ
+iα
_
p(θ) = 0.
Equation (10.1.10) has two linearly independent solutions
p
1
(θ) = c
−iαθ
,
and
p
2
(θ) = c
−iαθ
_
c
iαθ
(sinθ)
n−1
dθ.
Hence, the only bounded solutions for 0 ≤ θ ≤ π are p(θ) = cc
−iαθ
. It follows that
)(n) = c(n −i
_
1 −n
2
)
α
= c
_
t −i[.
t
[
2
[(.
t
, t)[
2
_
α
.
If we choose c = i
α
, then
(10.1.11) ϕ
α
(.
t
, t) = ([.
t
[
2
−it)
−
(n−1+α)
2
([.
t
[
2
+it)
−
(n−1−α)
2
.
Here we have used the principal branch for the power functions. Then we have the
following theorem.
10.1 Fundamental Solutions for
b
on the Heisenberg Group 239
Theorem 10.1.2. Let ϕ
α
(.
t
, t) be deﬁned as in (10.1.11) for α ∈ C. Then,
L
α
ϕ
α
= c
α
δ, where
c
α
=
2
4−2n
π
n
Γ(
n−1+α
2
)Γ(
n−1−α
2
)
.
Proof. For any c 0, set
ρ
(.
t
, t) = [.
t
[
2
+c
2
−it,
and deﬁne
(10.1.12) ϕ
α,
(.
t
, t) = ρ
−
(n−1+α)
2
ρ
−
(n−1−α)
2
.
Hence, ϕ
α,
is smooth and ϕ
α,
tends to ϕ
α
as distributions when c approaches zero.
It follows that L
α
ϕ
α,
tends to L
α
ϕ
α
as distributions. Thus, it suﬃces to show that
L
α
ϕ
α,
tends to c
α
δ in the distribution sense.
Recall that 7
k
= (∂,∂.
k
) +i.
k
(∂,∂t), for 1 ≤ / ≤ n −1. Then, for a ﬁxed c and
any o ∈ C, a direct computation shows
7
k
ρ
a
= 2o.
k
ρ
a−1
, and 7
k
ρ
a
= 2o.
k
ρ
a−1
,
7
k
ρ
a
= 7
k
ρ
a
= 0,
Tρ
a
= −ioρ
a−1
, and Tρ
a
= ioρ
a−1
.
It follows that
L
α
ϕ
α,
(.
t
, t) =
_
−
1
2
n−1
k=1
(7
k
7
k
+7
k
7
k
) +iαT
__
ρ
−
(n−1+α)
2
ρ
−
(n−1−α)
2
_
= c
2
((n −1)
2
−α
2
)ρ
−
(n+1+α)
2
ρ
−
(n+1−α)
2
= c
−2n
L
α
ϕ
α,1
_
1
c
(.
t
, t)
_
.
Hence,
_
Hn
L
α
ϕ
α,
(.
t
, t) d\ =
_
Hn
c
−2n
L
α
ϕ
α,1
_
1
c
(.
t
, t)
_
d\
=
_
Hn
L
α
ϕ
α,1
(.
t
, t) d\.
Since the mass of L
α
ϕ
α,
concentrates at zero as c → 0, it follows that L
α
ϕ
α,
tends to c
α
δ with c
α
=
_
L
α
ϕ
α,1
d\ as distributions. Therefore, it remains only to
compute the integral c
α
.
We set o =
1
2
(n + 1 +α) and / =
1
2
(n + 1 −α), then
c
α
=
1
2
n−1
_
Hn
((n −1)
2
−α
2
)([.
t
[
2
+ 1 −it)
−a
([.
t
[
2
+ 1 +it)
−b
drdjdt
=
((n −1)
2
−α
2
)
2
n−1
_
C
n−1
([.
t
[
2
+ 1)
−n
drdj
_
∞
−∞
(1 −it)
−a
(1 +it)
−b
dt.
240 Fundamental Solutions for
b
on the Heisenberg Group
Here, if there is no ambiguity, we shall write drdj for dr
1
dr
n−1
dj
1
dj
n−1
.
The ﬁrst integral is evaluated by
_
C
n−1
([.
t
[
2
+ 1)
−n
drdj =
2π
n−1
Γ(n −1)
_
∞
0
:
2n−3
(1 +:
2
)
n
d:
=
π
n−1
Γ(n −1)
_
∞
1
t
−n
(t −1)
n−2
dt, t = 1 +:
2
=
π
n−1
Γ(n −1)
_
1
0
(1 −:)
n−2
d:, : = t
−1
=
π
n−1
Γ(n)
,
where Γ() denotes the Gamma function. For the second integral we ﬁrst assume
that −(n −1) ≤ α ≤ n −1 so that o ≥ 1 and / ≥ 1. We start with the formula
_
∞
0
c
−xs
r
b−1
dr = Γ(/):
−b
which is valid if the real part of : is positive. Set : = 1 +it, then
Γ(/)(1 +it)
−b
=
_
∞
0
c
−ixt
c
−x
r
b−1
dr =
ˆ
)(t),
where
ˆ
)(t) is the Fourier transform of )(r) deﬁned by
)(r) =
_
c
−x
r
b−1
, for r 0
0, for r ≤ 0.
Similarly, we obtain
Γ(o)(1 −it)
−a
=
_
∞
0
c
ixt
c
−x
r
a−1
dr
=
_
0
−∞
c
−ixt
c
−]x]
[r[
a−1
dr
= ˆ p(t),
where p(r) is deﬁned by
p(r) =
_
0, for r ≥ 0
c
−]x]
[r[
a−1
, for r < 0.
Hence, by the Plancherel theorem, we have
Γ(o)Γ(/)
_
∞
−∞
(1 −it)
−a
(1 +it)
−b
dt =
_
∞
−∞
ˆ
)(t)ˆ p(t) dt
= 2π
_
∞
−∞
)(r)p(−r) dr
= 2π
_
∞
0
c
−2x
r
a+b−2
dr
=
πΓ(n)
2
n−1
.
10.1 Fundamental Solutions for
b
on the Heisenberg Group 241
This implies
(10.1.13)
_
∞
−∞
(1 −it)
−a
(1 +it)
−b
dt =
2
−(n−1)
πΓ(n)
Γ(o)Γ(/)
,
for −(n − 1) ≤ α ≤ n −1. In fact, the lefthand side of (10.1.13) deﬁnes an entire
function of α from the following equality:
_
∞
−∞
(1 −it)
−a
(1 +it)
−b
dt =
_
∞
−∞
(1 +t
2
)
−
n+1
2
c
iαtan
−1
t
dt.
Hence, (10.1.13) holds for all α ∈ C, and we obtain
c
α
=
((n −1)
2
−α
2
)
2
n−1
π
n−1
Γ(n)
2
−(n−1)
πΓ(n)
Γ(
n+1+α
2
)Γ(
n+1−α
2
)
=
2
4−2n
π
n
Γ(
n−1+α
2
)Γ(
n−1−α
2
)
.
This completes the proof of Theorem 10.1.2.
It follows from Theorem 10.1.2 that c
α
= 0 if α = ±(2/ +n−1) for any nonneg
ative integer /. Hence, if α ,= ±(2/ +n −1) for / ∈ N∪ ¦0¦, then Φ
α
= c
−1
α
ϕ
α
is a
fundamental solution for L
α
.
We now derive some consequences from Theorem 10.1.2. The convolution of two
functions ) and p on H
n
is deﬁned by
) ∗ p(n) =
_
Hn
)(·)p(·
−1
n) d\ (·) =
_
Hn
)(n·
−1
)p(·) d\ (·).
Set ˇ p(n) = p(n
−1
), then
_
Hn
() ∗ p)(n)/(n) d\ (n) =
_
Hn
)(n)(/ ∗ ˇ p)(n) d\ (n),
provided that both sides make sense.
If α ,= ±(2/ + n − 1) for / ∈ N ∪ ¦0¦, then for any ) ∈ C
∞
0
(H
n
), deﬁne
1
α
) = ) ∗ Φ
α
. It is clear that 1
α
) ∈ C
∞
(H
n
) since Φ
α
has singularity only
at zero. For the rest of this section / will always mean a nonnegative integer.
Theorem 10.1.3. If ) ∈ C
∞
0
(H
n
) and α ,= ±(2/+n−1), then L
α
1
α
) = 1
α
L
α
) =
).
Proof. Since L
α
is left invariant, clearly we have L
α
1
α
) = ). For the other equality,
let p ∈ C
∞
0
(H
n
). Notice that −α ,= ±(2/ + n − 1) whenever α ,= ±(2/ + n − 1).
Then
_
p(n))(n) d\ (n) =
_
L
−α
1
−α
p(n))(n) d\ (n)
=
_
(p ∗ Φ
−α
)(n)L
α
)(n) d\ (n)
=
_
p(n)(L
α
)∗
ˇ
Φ
−α
)(n) d\ (n)
=
_
p(n)1
α
L
α
)(n) d\ (n).
Hence, 1
α
L
α
) = ). This proves the theorem.
242 Fundamental Solutions for
b
on the Heisenberg Group
Theorem 10.1.4. The operator L
α
is hypoelliptic if and only if α ,= ±(2/+n−1).
In particular,
b
is hypoelliptic on H
n
for (0, ¡)forms when 1 ≤ ¡ < n −1.
Proof. If α = ±(2/ + n − 1) for some nonnegative integer /, then the function
ϕ
α
(.
t
, t) deﬁned in (10.1.11) is a nonsmooth solution to the equation L
α
ϕ
α
= 0.
Next, let α ,= ±(2/ +n −1), ) ∈ T
t
such that L
α
) = p is smooth on some open
set l. Let \ ⊂⊂ l be an open set which is relatively compact in l. Choose a
cutoﬀ function ζ ∈ C
∞
0
(l) with ζ = 1 in some open neighborhood of \ . Then,
by Theorem 10.1.3, 1
α
(ζp) is smooth and satisﬁes L
α
1
α
(ζp) = ζp. Hence, to
show that ) is smooth on \ , it suﬃces to show that / = ζ() −1
α
(ζp)) is smooth
on \ . Since / is a distribution with compact support, a standard argument from
functional analysis shows / = 1
α
L
α
/. But on \ we have
L
α
/ = L
α
) −L
α
1
α
(ζp) = p −ζp = 0.
The fact that Φ
α
(.
t
, t) is just singular at the origin will then guarantee that 1
α
L
α
/
is smooth on \ which in turns shows /, and hence ), is smooth on \ .
The hypoellipticity of
b
on (0, ¡)forms when 1 ≤ ¡ < n −1 follows immediately
from the expression of
b
in (10.1.6). This proves the theorem.
Theorem 10.1.3 can be used to obtain the following existence and regularity
theorem for the ∂
b
equation on H
n
:
Theorem 10.1.5. Let ) ∈ C
(0,q)
(H
n
), 1 ≤ ¡ < n − 1, with compact support. If
∂
b
) = 0 in the distribution sense, then n = ∂
∗
b
1) satisﬁes ∂
b
n = ) and n ∈
Λ
1/2
(0,q)
(H
n
, loc), where 1 = 1
α
with α = n − 1 − 2¡. Moreover, if ) ∈ C
k
(0,q)
(H
n
),
/ ∈ N, with compact support, then n ∈ C
k+
1
2
(0,q)
(H
n
, loc).
Proof. Since ) is a continuous (0, ¡)form with compact support on H
n
and 1 ≤ ¡ <
n −1, we obtain from Theorem 10.1.3 that
b
1) = (∂
b
∂
∗
b
+∂
∗
b
∂
b
)1) = ),
where 1 acts on ) componentwise. The hypothesis ∂
b
) = 0 implies
∂
b
1) = 1∂
b
) = 0.
Hence,
∂
b
∂
∗
b
1) = ).
For the regularity of n = ∂
∗
b
1), we write ) =
t
]J]=q
)
J
ω
J
. Then, we obtain
from the previous calculation that
∂
∗
b
1) = −
t
]H]=q−1
_
t
l,J
c
J
lH
7
l
(1)
J
)
_
ω
H
.
Hence, it suﬃces to estimate the following integral, for 1 ≤ , ≤ n −1:
(10.1.14) 7
j
_
Hn
)(ξ)Φ(ξ
−1
ζ) d\ (ξ),
10.1 Fundamental Solutions for
b
on the Heisenberg Group 243
where ) is a continuous function with compact support on H
n
and ζ = (.
t
, t),
ξ = (n
t
, n) and
Φ(.
t
, t) = ([.
t
[
2
−it)
−(n−1−q)
([.
t
[
2
+it)
−q
.
We can rewrite (10.1.14) as
(10.1.15) −2(n −1 −¡)) ∗ Ψ
j
(ζ) = −2(n −1 −¡)
_
Hn
)(ξ)Ψ
j
(ξ
−1
ζ) d\ (ξ),
where
Ψ
j
(.
t
, t) = .
j
([.
t
[
2
−it)
−(n−1−q)−1
([.
t
[
2
+it)
−q
.
Deﬁne a new kernel Ψ
j
y
(.
t
, t, j) on H
n
1
+
, where 1
+
= ¦j ∈ 1[ j 0¦, by
Ψ
j
y
(.
t
, t, j) = .
j
([.
t
[
2
−it +j)
−(n−1−q)−1
([.
t
[
2
+it +j)
−q
.
It is easily seen that ) ∗ Φ
j
y
(.
t
, t, j) is smooth on H
n
1
+
and
lim
y→0
+
) ∗ Ψ
j
y
(.
t
, t, j) = ) ∗ Ψ
j
(.
t
, t).
The assertion then follows from the HardyLittlewood lemma proved in Theorem
C.1 in the Appendix if one can show, for 0 < j ≤ 1,2, that
(10.1.16) [∇() ∗ Ψ
j
y
)[ ≤ cj
−
1
2
 ) 
L
∞
(Hn)
,
for some constant c 0 and 1 ≤ , ≤ n −1.
Since ξ
−1
ζ = (. − n, t − n − 2Imn .), we may introduce new coordinates
η
2j−1
= Re(.
j
−n
j
), η
2j
= Im(.
j
−n
j
) and δ = t −n−2Im(n .) for 1 ≤ , ≤ n −1.
A direct calculation shows that (10.1.16) will be proved if one can show that
(10.1.17) 1
1
=
_
](η,δ)]≤M
dη
1
dη
2n−2
dδ
([η[
2
+[δ[ +j)
n
≤ cj
−
1
2
,
(10.1.18) 1
2
=
_
](η,δ)]≤M
[η
j
[dη
1
dη
2n−2
dδ
([η[
2
+[δ[ +j)
n+1
≤ cj
−
1
2
,
and
(10.1.19) 1
3
=
_
](η,δ)]≤M
[η
j
η
k
[dη
1
dη
2n−2
dδ
([η[
2
+[δ[ +j)
n+1
≤ cj
−
1
2
,
where ` 0 is a positive constant. Notice ﬁrst that (10.1.19) follows immediately
from (10.1.17). For 1
1
we have
1
1
≤ c
_
M
0
_
M
0
:
2n−3
(:
2
+δ +j)
n
d:dδ
≤ c
_
M
0
:
2n−3
(:
2
+j)
n−1
d:
≤ c
_ M
√
y
0
r
2n−3
(1 +r
2
)
n−1
dr
≤ c(`)(−logj)
≤ c(`, o)j
−a
,
244 Fundamental Solutions for
b
on the Heisenberg Group
for any o 0. 1
2
can be estimated as follows:
1
2
≤ c
_
M
0
_
M
0
:
2n−2
(:
2
+δ +j)
n+1
d:dδ
≤ c
_
M
0
:
2n−2
(:
2
+j)
n
d:
≤ cj
−
1
2
_
∞
0
r
2n−2
(1 +r
2
)
n
dr
≤ cj
−
1
2
.
This proves the case when ) is continuous with compact support.
If ) ∈ C
k
(0,q)
(H
n
) with compact support, we let A be any one of the left invari
ant vector ﬁelds 7
1
, , 7
n−1
,7
1
, , 7
n−1
and T. Then, from (10.1.14) and the
convolution formula we get
A7
j
_
Hn
)(ξ)Φ(ξ
−1
ζ) d\ (ξ) =
_
Hn
A)(ξ)7
j
Φ(ξ
−1
ζ) d\ (ξ)
= 7
j
_
Hn
A)(ξ)Φ(ξ
−1
ζ) d\ (ξ).
Since the left invariant vector ﬁelds 7
j
, 7
j
, where , = 1, , n−1, and T span the
tangent space of H
n
, the assertion now follows from diﬀerentiating / times and the
ﬁrst part of the proof. This proves the theorem.
The construction developed in this section can be extended to the generalized
Heisenberg group H
n,k
which we now deﬁne. For each 1 ≤ / ≤ n −1, let
Ω
n,k
= ¦(.
t
, .
n
) ∈ C
n
[ Im.
n
[.
1
[
2
+ +[.
k
[
2
−[.
k+1
[
2
− −[.
n−1
[
2
¦.
The boundary of Ω
n,k
is identiﬁed with the generalized Heisenberg group H
n,k
=
C
n−1
1 by
π :
_
.
t
, t +i
_
k
j=1
[.
j
[
2
−
n−1
j=k+1
[.
j
[
2
__
→(.
t
, t),
where (.
t
, t) is the coordinates on H
n,k
.
The group structure on H
n,k
is deﬁned by
(10.1.20) (.
t
, t) (n
t
, n) =
_
.
t
+n
t
, t +n + 2Im
_
k
j=1
.
j
n
j
−
n−1
j=k+1
.
j
n
j
__
.
One veriﬁes immediately that
7
j
=
∂
∂.
j
+i.
j
∂
∂t
, 1 ≤ , ≤ /,
7
j
=
∂
∂.
j
−i.
j
∂
∂t
, / + 1 ≤ , ≤ n −1, and
T =
∂
∂t
,
10.1 Fundamental Solutions for
b
on the Heisenberg Group 245
are leftinvariant vector ﬁelds on H
n,k
such that
[7
j
, 7
j
] =
_
−2iT, for 1 ≤ , ≤ /,
2iT, for / + 1 ≤ , ≤ n −1,
and that all other commutators vanish. It follows that the 7
j
’s deﬁne a nondegen
erate C1 structure on H
n,k
such that the Levi matrix has / positive eigenvalues
and n−1 −/ negative eigenvalues. Without loss of generality, / can be assumed to
be at least (n −1),2. We shall call such a C1 structure /strongly pseudoconvex.
We ﬁx a leftinvariant metric on H
n,k
which makes 7
j
, 7
j
and T, 1 ≤ , ≤ n −1,
orthonormal. The dual basis is given by ω
1
, , ω
n−1
, ω
1
, , ω
n−1
and τ, where
ω
j
= dr
j
+idj
j
for 1 ≤ , ≤ n −1 and τ is given by
τ = dt + 2
k
j=1
(r
j
dj
j
−j
j
dr
j
) −2
n−1
j=k+1
(r
j
dj
j
−j
j
dr
j
).
The volume element is
d\ = 2
1−n
dr
1
dr
n−1
dj
1
dj
n−1
dt.
Next, we calculate
b
on the generalized Heisenberg group H
n,k
as before. Let
1 = ¦1, , /¦ and 1
t
= ¦/ + 1, , n −1¦. For each multiindex J with [J[ = ¡,
we set
α
J
= [1 ¸ J[ +[1
t
∩ J[ −[1 ∩ J[ −[1
t
¸ J[,
where [ [ denotes cardinality of the set. Hence, if ) =
t
]J]=q
)
J
ω
J
is a smooth
(0, ¡)form with compact support on H
n,k
, we get
b
) = (∂
b
∂
∗
b
+∂
∗
b
∂
b
)
_
t
]J]=q
)
J
ω
J
_
= −
t
]J]=q
__
m/ ∈J
7
m
7
m
+
m∈J
7
m
7
m
_
)
J
_
ω
J
=
t
]J]=q
__
−
1
2
n−1
m=1
(7
m
7
m
+7
m
7
m
) +iα
J
T
_
)
J
_
ω
J
.
Notice that −(n − 1) ≤ α
J
≤ (n − 1). The extreme case α
J
= n − 1 occurs if and
only if [J[ = n − 1 − / and J = 1
t
. On the other hand, α
J
= −(n − 1) occurs if
and only if [J[ = / and J = 1. Hence, we have
Theorem 10.1.6.
b
is hypoelliptic for (0, ¡)forms, 0 ≤ ¡ ≤ n −1, on H
n,k
if
¡ ,= / and ¡ ,= n −1 −/.
Proof. The assertion follows immediately from Theorem 10.1.4 if we change the
coordinates .
j
, / + 1 ≤ , ≤ n −1, to .
j
. This proves the theorem.
We note that Theorem 10.1.6 is a variant of Theorem 8.4.4 since condition Y (¡)
holds on H
n,k
when ¡ ,= / and ¡ ,= n −1 −/. The conclusion of Theorem 10.1.5
246 Fundamental Solutions for
b
on the Heisenberg Group
also holds on H
n,k
when ¡ ,= / and ¡ ,= n −1 −/. The proof is exactly the same
and we omit the details.
10.2 The CauchySzeg¨o Kernel on the Heisenberg Group
In this section we compute the Szeg¨o projection on H
n
. Let Ω
n
be the Siegel
upper half space in C
n
. Denote by H
2
(Ω
n
) the Hardy space of all holomorphic
functions ) deﬁned on Ω
n
such that
sup
s>0
 )
s
(.) 
L
2
(bΩn)
< ∞,
where )
s
(.) = )(.
t
, .
n
+ i:) for . = (.
t
, .
n
) ∈ /Ω
n
and : 0. It will be clear
later that H
2
(Ω
n
) forms a Hilbert space under the norm  ) 
H
2
(Ωn)
= sup
s>0
 )
s

L
2
(bΩn)
.
If )(.) ∈ H
2
(Ω
n
), then by deﬁnition )(.) satisﬁes
(10.2.1)
_
C
n−1
_
∞
−∞
[)(.
t
, t +i[.
t
[
2
+i:)[
2
dtdr
t
dj
t
< C,
where the constant C 0 is independent of : 0 and dr
t
dj
t
stands for dr
1
∧dj
1
∧
∧ dr
n−1
∧ dj
n−1
with .
j
= r
j
+ ij
j
for 1 ≤ , ≤ n −1. By using the mean
value property of a holomorphic function it is not hard to see from (10.2.1) that
for each .
t
∈ C
n−1
and : 0 the function )(.
t
, t + i[.
t
[
2
+ i:), when viewed as
a function in t on 1, is 1
2
integrable. Thus, we can form the Fourier transform
of )(.
t
, t + i[.
t
[
2
+ i:) with respect to t which will be denoted by
˜
)
s
(.
t
, λ). The
resulting function
˜
)
s
(.
t
, λ) is 1
2
integrable with respect to λ and satisﬁes
(10.2.2)
1
2π
_
C
n−1
_
∞
−∞
[
˜
)
s
(.
t
, λ)[
2
dλdr
t
dj
t
< C.
Since ) is holomorphic on Ω
n
, we have by Cauchy’s theorem
(10.2.3)
˜
)
s+s
(.
t
, λ) = c
−λs
˜
)
s
(.
t
, λ),
for :, :
t
0. It follows that, for ﬁxed : 0, we get that
(10.2.4)
_
C
n−1
_
∞
−∞
[)(.
t
, t +i[.
t
[
2
+i: +i:
t
)[
2
dtdr
t
dj
t
=
1
2π
_
C
n−1
_
∞
−∞
[
˜
)
s
(.
t
, λ)[
2
c
−2λs
dλdr
t
dj
t
,
which implies
˜
)
s
(.
t
, λ) = 0 a.e. for λ < 0. Therefore, we may assume that
˜
)
s
(.
t
, λ)
is concentrated on 1
+
= ¦r ∈ 1[ r 0¦ with respect to λ. It is also clear from
(10.2.3) that
˜
)
s
(.
t
, λ) =
˜
)
0
(.
t
, λ)c
−λs
,
10.2 The CauchySzeg¨o Kernel on the Heisenberg Group 247
for some measurable function
˜
)
0
(.
t
, λ). We set
˜
)
0
(.
t
, λ) =
˜
)(.
t
, λ)c
−λ]z
]
2
.
Since )(.) is holomorphic on Ω
n
, the homogeneous tangential CauchyRiemann
equation on each level set ¦.
n
= t +i([.
t
[
2
+:)¦ with : 0 must be satisﬁed by ),
namely,
_
∂
∂.
k
−i.
k
∂
∂t
_
)(.
t
, t +i([.
t
[
2
+:)) = 0, 1 ≤ / ≤ n −1.
It follows that
_
∂
∂.
k
+λ.
k
_
˜
)
s
(.
t
, λ) = 0, 1 ≤ / ≤ n −1,
for : 0. Hence, for 1 ≤ / ≤ n −1, we have
0 =
_
∂
∂.
k
+λ.
k
_
˜
)
0
(.
t
, λ)
=
_
∂
∂.
k
+λ.
k
_
(
˜
)(.
t
, λ)c
−λ]z
]
2
)
=
∂
˜
)
∂.
k
(.
t
, λ)c
−λ]z
]
2
.
This shows that
˜
)(.
t
, λ) is holomorphic in .
t
and measurable in λ. By substituting
˜
)(.
t
, λ) into (10.2.2), we obtain
1
2π
_
C
n−1
_
∞
0
[
˜
)(.
t
, λ)[
2
c
−2λ]z
]
2
c
−2λs
dλdr
t
dj
t
< C,
where the constant C 0 is independent of : 0. Letting : tend to zero, we see
that the function
˜
)(.
t
, λ) satisﬁes
(10.2.5)
1
2π
_
C
n−1
_
∞
0
[
˜
)(.
t
, λ)[
2
c
−2λ]z
]
2
dλdr
t
dj
t
< C,
and the function )(.) is recovered by
(10.2.6) )(.) = )(.
t
, .
n
) =
1
2π
_
∞
0
˜
)(.
t
, λ)c
iλzn
dλ,
for .
n
= t +i[.
t
[
2
+i: with : 0. Moreover, the Plancherel theorem shows that
(10.2.7)
lim
s,s
→0
_
C
n−1
_
∞
−∞
[)(.
t
, t +i[.
t
[
2
+i:) −)(.
t
, t +i[.
t
[
2
+i:
t
)[
2
dtdr
t
dj
t
= lim
s,s
→0
_
C
n−1
_
∞
0
[
˜
)(.
t
, λ)[
2
c
−2λ]z
]
2
(c
−λs
−c
−λs
)
2
dλdr
t
dj
t
= 0.
This means that )(.
t
, t +i[.
t
[
2
+i:) converges in the 1
2
norm to )(.
t
, t +i[.
t
[
2
) as
: →0.
The next theorem shows that the existence of the function
˜
)(.
t
, λ) with property
(10.2.5) is also suﬃcient for representing a function )(.) in the Hardy space H
2
(Ω
n
).
248 Fundamental Solutions for
b
on the Heisenberg Group
Theorem 10.2.1. A complexvalued function ) deﬁned on Ω
n
belongs to H
2
(Ω
n
) if
and only if there exists a function
˜
)(.
t
, λ), (.
t
, λ) ∈ C
n−1
1
+
, which is holomorphic
in .
t
and measurable in λ and satisﬁes
1
2π
_
C
n−1
_
∞
0
[
˜
)(.
t
, λ)[
2
c
−2λ]z
]
2
dλdr
t
dj
t
< ∞,
where dr
t
dj
t
stands for dr
1
∧ dj
1
∧ ∧ dr
n−1
∧ dj
n−1
and 1
+
= ¦r ∈ 1[ r 0¦.
The integral
)(.) = )(.
t
, .
n
) =
1
2π
_
∞
0
˜
)(.
t
, λ)c
iλzn
dλ,
for .
n
= t + i[.
t
[
2
+ i:, converges absolutely for all : 0 and deﬁnes a function
)(.) ∈ H
2
(Ω
n
).
Proof. Suppose that there is a function
˜
)(.
t
, λ) which is holomorphic in .
t
and
measurable in λ and
˜
)(.
t
, λ) satisﬁes (10.2.5). For any .
t
and .
n
= t + i[.
t
[
2
+ i:
with : 0, we may choose a polydisc 1(.
t
; :) in C
n−1
centered at .
t
with small
multiradii : = (:
1
, , :
n−1
) such that [n
t
[
2
≤ [.
t
[
2
+ (:,2) for all n
t
∈ 1(.
t
; :).
Since the value of a holomorphic function is dominated by its 1
1
norm, we obtain
by H¨older’s inequality that
_
∞
0
[
˜
)(.
t
, λ)c
iλzn
[ dλ
_
∞
0
__
D(z
;r)
[
˜
)(n
t
, λ)[ d\ (n
t
)
_
c
−λ]z
]
2
−λs
dλ
_
∞
0
__
D(z
;r)
[
˜
)(n
t
, λ)[c
−λ]w
]
2
d\ (n
t
)
_
c
−λs/2
dλ
__
D(z
;r)
_
∞
0
[
˜
)(n
t
, λ)[
2
c
−2λ]w
]
2
dλd\ (n
t
)
_1
2
__
D(z
;r)
_
∞
0
c
−λs
dλd\ (n
t
)
_1
2
< ∞.
This shows that the integral deﬁned by (10.2.6) converges absolutely and deﬁnes
a holomorphic function on Ω
n
. To see )(.) is actually in H
2
(Ω
n
), we apply the
Plancherel theorem to the λvariable and get
_
C
n−1
_
∞
−∞
[)(.
t
, t +i[.
t
[
2
+i:)[
2
dtdr
t
dj
t
=
1
2π
_
C
n−1
_
∞
0
[
˜
)(.
t
, λ)[
2
c
−2λ]z
]
2
c
−2λs
dλdr
t
dj
t
≤ C,
for all : 0. This completes the proof of Theorem 10.2.1.
It is clear from the proof of Theorem 10.2.1 that H
2
(Ω
n
) can be identiﬁed with
a closed subspace of 1
2
(/Ω
n
), namely, any )(.) ∈ H
2
(Ω
n
) is identiﬁed with its 1
2
limiting value )(.
t
, t +i[.
t
[
2
) on /Ω
n
with the norm
 ) 
H
2
(Ωn)
= sup
s>0
 )
s

L
2
(bΩn)
=  )(.
t
, t +i[.
t
[
2
) 
L
2
(bΩn)
.
Thus, following the procedure of the Bergman kernel function, we obtain the repro
ducing kernel, named CauchySzeg¨o kernel, o(., n) for the Hardy space H
2
(Ω
n
).
We make the following deﬁnition:
10.2 The CauchySzeg¨o Kernel on the Heisenberg Group 249
Deﬁnition 10.2.2. The CauchySzeg¨o kernel associated with Ω
n
is the unique func
tion o(., n) which is holomorphic in . and antiholomorphic in n with respect to
. ∈ Ω
n
and n ∈ Ω
n
such that
(10.2.8) )(.) =
_
bΩn
o(., n))(n) dσ
w
,
for any ) ∈ H
2
(Ω
n
) and any . ∈ Ω
n
, where dσ
w
is the surface element on /Ω
n
.
For each ﬁxed . ∈ Ω
n
, (10.2.8) deﬁnes a bounded linear functional on H
2
(Ω
n
).
It is also clear from general Hilbert space theory that o(., n) can be expressed in
terms of any orthonormal basis ¦φ
k
(.)¦
∞
k=1
of H
2
(Ω
n
), i.e.,
o(., n) =
∞
k=1
φ
k
(.)φ
k
(n).
Now we want to calculate the CauchySzeg¨o kernel o(., n) on the Siegel upper
half space Ω
n
. One way to achieve this goal is via the pullback of the CauchySzeg¨o
kernel on the unit ball in C
n
by the inverse Cayley transform. Recall that the
Cayley transform n = Φ(.) is a biholomorphic mapping from the unit ball 1
n
in
C
n
onto the Siegel upper half space Ω
n
deﬁned by (7.3.2). Thus, the inverse Cayley
transform φ = Φ
−1
is given by
φ : Ω
n
→1
n
n → . =
_
−2in
1
1 −in
n
, ,
−2in
n−1
1 −in
n
, −
1 +in
n
1 −in
n
_
.
First, by constructing an orthonormal basis for H
2
(1
n
) directly, the Cauchy
Szeg¨o kernel on the unit ball can be calculated as follows:
Proposition 10.2.3. The CauchySzeg¨o kernel o(ζ, η) on the unit ball in C
n
can
be expressed explicitly as
(10.2.9) o(ζ, η) =
(n −1)!
2π
n
1
(1 −ζ η)
n
,
where ζ η = ζ
1
η
1
+ +ζ
n
η
n
.
Proof. It is clear that ¦ζ
α
¦ forms an orthogonal basis for the Hardy space H
2
(1
n
),
where α = (α
1
, , α
n
) is any multiindex with α
j
∈ N ∪ ¦0¦ for 1 ≤ , ≤ n.
Therefore, to get the CauchySzeg¨o kernel, we need to normalize ¦ζ
α
¦.
We proceed as in Section 6.3 for the Bergman kernel function on the unit ball
1
n
in C
n
. Hence, we get
c
α
=
_
bBn
[ζ
α
[
2
dσ
2n−1
= 2([α[ +n)
_
Bn
[ζ
α
[
2
d\
2n
=
2π
n
α
n
! α
1
!
([α[ +n −1)!
.
250 Fundamental Solutions for
b
on the Heisenberg Group
It follows that the CauchySzeg¨o kernel o(ζ, η) on the unit ball is given by
o(ζ, η) =
α
1
c
α
ζ
α
η
α
=
(n −1)!
2π
n
_
1 +
α,=0
n(n + 1) (n +[α[ −1)
α
1
! α
n
!
ζ
α
η
α
_
=
(n −1)!
2π
n
_
1 +
∞
k=1
]α]=k
n(n + 1) (n +[α[ −1)
α
1
! α
n
!
ζ
α
η
α
_
=
(n −1)!
2π
n
_
1 +
∞
k=1
n(n + 1) (n +/ −1)
/!
(ζη)
k
_
=
(n −1)!
2π
n
1
(1 −ζη)
n
.
This proves the proposition.
Our next step is to pull the CauchySzeg¨o kernel o(ζ, η) on the unit ball back to
the Siegel upper half space. Denote by :(.) =
n
j=1
[.
j
[
2
−1 the deﬁning function
for the unit ball in C
n
, and ﬁx the standard metric on C
n
. Then, the surface element
d: on the boundary /1
n
is given by the interior product of the volume form d\
2n
with d:,[d:[, namely,
d: = Reι
∗
__
n
j=1
.
j
d.
j
_
∨
_
1
2i
_
n
_
n
k=1
d.
k
∧ d.
k
__
= Reι
∗
_
n
j=1
1
2
n−1
i
n
.
j
d.
j
∧
_
k,=j
d.
k
∧ d.
k
__
,
where ∨ denotes the interior product and ι : /1
n
→C
n
is the inclusion map. Hence,
the pullback of d: by the inverse Cayley transform φ is
φ
∗
(d:) =
1
2
n−1
i
n
_
2
2n−1
i
[1 −in
n
[
2n
1 −in
n
1 −in
n
dn
n
∧
_
n−1
j=1
dn
j
∧ dn
j
_
+
n−1
j=1
2
2n−1
n
j
(1 +in
n
)
[1 −in
n
[
2n+2
dn
j
∧
_
k,=j
dn
k
∧ dn
k
__
.
Since n
n
= t +i[n
t
[
2
on /Ω
n
, we have
dn
n
= dt +i
n−1
j=1
(n
j
dn
j
+n
j
dn
j
).
It follows that
dn
n
∧
_
n−1
j=1
dn
j
∧ dn
j
_
= dt ∧
_
n−1
j=1
dn
j
∧ dn
j
_
,
10.2 The CauchySzeg¨o Kernel on the Heisenberg Group 251
and, for 1 ≤ , ≤ n −1,
dn
j
∧
_
k,=j
dn
k
∧ dn
k
_
= 2in
j
dt ∧
_
n−1
j=1
dn
j
∧ dn
j
_
,
on the boundary of Ω
n
. Thus,
φ
∗
(d:)
= (−2i)
n
_
i
[1 −in
n
[
2n
1 −in
n
1 −in
n
+
n−1
k=1
2i[n
k
[
2
(1 +in
n
)
[1 −in
n
[
2n+2
_
dt ∧
_
n−1
j=1
dn
j
∧ dn
j
_
= (−2i)
n
_
i(1 +n
2
n
) + (n
n
−n
n
)(1 +in
n
)
[1 −in
n
[
2n+2
_
dt∧
_
n−1
j=1
dn
j
∧ dn
j
_
=
2
2n−1
[1 −in
n
[
2n
dt∧
_
n−1
j=1
dn
j
∧ d·
j
_
,
where n
j
= n
j
+i·
j
.
Since the surface element dσ on the boundary /Ω
n
is given by dσ = dt ∧
(∧
n−1
j=1
dn
j
∧d·
j
), the above calculation suggests that the CauchySzeg¨o kernel o(., n)
associated with the Siegel upper half space should be given by
(10.2.10)
o((.
t
, .
n
), (n
t
, n
n
))
=
2
n−
1
2
(1 −i.
n
)
n
2
n−
1
2
(1 +in
n
)
n
(n −1)!
2π
n
1
(1 −φ(.)φ(n))
n
=
(−1)
n
2
n−2
(n −1)!
π
n
_
i(.
n
−n
n
) + 2
n−1
j=1
.
j
n
j
_
−n
.
We must show that the function o(., n) obtained in (10.2.10) has the required
reproducing property for H
2
(Ω
n
) as stated in Deﬁnition 10.2.2. Theorem 10.2.1
suggests that one should check the Fourier transform of o(., n).
Notice ﬁrst that, for . ∈ Ω
n
and n ∈ Ω
n
, we have
Re
_
i(.
n
−n
n
) + 2
n−1
j=1
.
j
n
j
_
= −j
n
−·
n
+
n−1
j=1
(.
j
n
j
+.
j
n
j
)
= −(j
n
−[.
t
[
2
) −(·
n
−[n
t
[
2
) −[.
t
−n
t
[
2
,
which is always negative. Therefore, we can rewrite o(., n) as
(10.2.11) o(., n) =
2
n−2
π
n
_
∞
0
λ
n−1
c
(i(zn−wn)+2z
w
)λ
dλ,
and the above integral converges absolutely for . ∈ Ω
n
and n ∈ Ω
n
. Here we use
.
t
n
t
to denote the inner product
n−1
j=1
.
j
n
j
in C
n−1
.
252 Fundamental Solutions for
b
on the Heisenberg Group
Deﬁne
(10.2.12)
˜
o(.
t
, λ; n) =
_
2
π
_
n−1
λ
n−1
c
(−iwn+2z
w
)λ
.
We shall show that for each n ∈ Ω
n
, the integral
(10.2.13)
1
2π
_
C
n−1
_
∞
0
[
˜
o(.
t
, λ; n)[
2
c
−2λ]z
]
2
dλdr
t
dj
t
converges. Since n ∈ Ω
n
, n
n
= n +i[n
t
[
2
+i· with · 0. Hence, (10.2.13) can be
rewritten as
2
2n−3
π
2n−1
_
C
n−1
_
∞
0
λ
2n−2
c
−2λ]z
−w
]
2
c
−2λv
dλdr
t
dj
t
=
2
2n−3
π
2n−1
_
C
n−1
_
∞
0
λ
2n−2
c
−2λ]z
]
2
c
−2λv
dλdr
t
dj
t
=
2
2n−2
π
n
(n −2)!
_
∞
0
_
∞
0
λ
2n−2
c
−2λr
2
c
−2λv
:
2n−3
d:dλ
=
2
n−2
π
n
_
∞
0
λ
n−1
c
−2λv
dλ
=
(n −1)!
4π
n
·
n
.
It follows now from Theorem 10.2.1 and (10.2.11) that
o(., n) =
2
n−2
π
n
_
∞
0
λ
n−1
c
(i(zn−wn)+2z
w
)λ
dλ
=
1
2π
_
∞
0
˜
o(.
t
, λ; n)c
iλzn
dλ.
Hence, for each n ∈ Ω
n
, o(, n) ∈ H
2
(Ω
n
). Let n
j
= α
j
+ iβ
j
for 1 ≤ , ≤ n −1
and let dα
t
dβ
t
denote dα
1
∧dβ
1
∧ ∧dα
n−1
∧dβ
n−1
. Since o(., n) = o(n, .), for
any . ∈ Ω
n
and any ) ∈ H
2
(Ω
n
), we obtain
_
bΩn
o(., n))(n) dndα
t
dβ
t
=
1
2π
_
C
n−1
_
∞
0
˜
o(n
t
, λ; .)
˜
)(n
t
, λ)c
−2λ]w
]
2
dλdα
t
dβ
t
=
2
n−2
π
n
_
C
n−1
_
∞
0
λ
n−1
c
λ(izn+2z
w
−2]w
]
2
)
˜
)(n
t
, λ) dλdα
t
dβ
t
10.2 The CauchySzeg¨o Kernel on the Heisenberg Group 253
=
2
n−2
π
n
_
C
n−1
_
∞
0
λ
n−1
c
λ(izn+2]z
]
2
−2z
w
−2]w
−z
]
2
)
˜
)(n
t
, λ) dλdα
t
dβ
t
=
2
n−2
π
n
_
C
n−1
_
∞
0
λ
n−1
c
λ(izn+2]z
]
2
)
c
−2λ]η
]
2
c
−2λz
(z
+η
)
˜
)(.
t
+η
t
, λ) dλdζ
t
dξ
t
=
1
2π
2
n
(n −2)!
_
∞
0
λ
n−1
c
iλzn ˜
)(.
t
, λ)
__
∞
0
c
−2λρ
2
ρ
2n−3
dρ
_
dλ
=
1
2π
_
∞
0
˜
)(.
t
, λ)c
iλzn
dλ
= )(.),
where η
j
= n
j
−.
j
= ζ
j
+iξ
j
for 1 ≤ , ≤ n −1 and dζ
t
dξ
t
denotes dζ
1
∧dξ
1
∧ ∧
dζ
n−1
∧dξ
n−1
. The last equality is guaranteed by (10.2.6) in Theorem 10.2.1. Thus,
we have shown that the kernel function (10.2.10) reproduces the functions belonging
to the Hardy space H
2
(Ω
n
). Hence, by the uniqueness of the CauchySzeg¨o kernel
function, we obtain the following theorem.
Theorem 10.2.4. The CauchySzeg¨o kernel function o(., n) associated with the
Siegel upper half space Ω
n
is given by
o(., n) = o((.
t
, .
n
), (n
t
, n
n
))
=
(−1)
n
2
n−2
(n −1)!
π
n
_
i(.
n
−n
n
) + 2
n−1
j=1
.
j
n
j
_
−n
.
Hence, for any ) ∈ 1
2
(/Ω
n
), the integral
(10.2.14) o)(.) =
_
bΩn
o(., n))(n) dσ(n),
deﬁnes a function o)(.) in the Hardy space H
2
(Ω
n
) which has a welldeﬁned 1
2
integrable limiting value on /Ω
n
. We recall that the Szeg¨o projection on /Ω
n
is the
orthogonal projection from 1
2
(/Ω
n
) onto the closed subspace consisting of square
integrable C1 functions, which coincide with the limiting values of functions be
longing to the Hardy space H
2
(Ω
n
). We shall still use (10.2.14) to denote the Szeg¨o
projection on /Ω
n
.
Since, for each ) ∈ 1
2
(/Ω
n
), o) ∈ H
2
(Ω
n
), Theorem 10.2.1 shows that
lim
→0
o)(.
t
, t +i[.
t
[
2
+ic
2
) = o)(.
t
, t +i[.
t
[
2
)
in the 1
2
sense. Denote o)(.
t
, t +i[.
t
[
2
+ic
2
) by (o))
(.
t
, t +i[.
t
[
2
) which can be
regarded as an 1
2
integrable function on /Ω
n
. Let
ρ
(.
t
, t) = [.
t
[
2
+c
2
−it,
on /Ω
n
. Then we have
254 Fundamental Solutions for
b
on the Heisenberg Group
Proposition 10.2.5. For any ) ∈ 1
2
(/Ω
n
) and any c 0, (o))
is given by
(10.2.15) (o))
(.
t
, t +i[.
t
[
2
) =
2
n−2
(n −1)!
π
n
) ∗ ρ
−n
(.
t
, t),
where the convolution is taken with respect to the group structure on /Ω
n
, and the
coordinates on /Ω
n
are .
t
= (.
1
, , .
n−1
) and t.
Proof. Let β = (.
t
, t +i[.
t
[
2
) and α = (n
t
, n +i[n
t
[
2
). Hence,
2
n−2
(n −1)!
π
n
) ∗ ρ
−n
(.
t
, t) =
2
n−2
(n −1)!
π
n
_
bΩn
)(α)ρ
−n
(α
−1
β) dσ(α).
A direct calculation shows that
α
−1
β = (−n
t
, −n +i[n
t
[
2
) (.
t
, t +i[.
t
[
2
)
= (.
t
−n
t
, t −n +i[.
t
[
2
+i[n
t
[
2
−2i.
t
n
t
).
It follows that we have
ρ
−n
(α
−1
β) = ([.
t
−n
t
[
2
+c
2
−i(t −n) −.
t
n
t
+.
t
n
t
)
−n
= ([.
t
[
2
+[n
t
[
2
+c
2
−i(t −n) −2.
t
n
t
)
−n
= (−1)
n
(i(t +i[.
t
[
2
+ic
2
−(n −i[n
t
[
2
)) + 2.
t
n
t
)
−n
= (−1)
n
(i(.
n
+ic
2
−n
n
) + 2.
t
n
t
)
−n
.
Hence, we obtain
2
n−2
(n −1)!
π
n
) ∗ ρ
−n
(.
t
, t)
=
(−1)
n
2
n−2
(n −1)!
π
n
_
bΩn
)(n
t
, n)
(i(.
n
+ic
2
−n
n
) + 2.
t
n
t
)
n
dσ(α)
= o)(.
t
, t +i[.
t
[
2
+ic
2
).
This proves the proposition.
Finally, we can describe the Szeg¨o projection on the Heisenberg group H
n
as
follows:
Theorem 10.2.6. The Szeg¨o projection o) for any ) ∈ 1
2
(H
n
) is given by
(10.2.16) o)(.
t
, t) = lim
→0
2
2n−3
(n −1)!
π
n
) ∗ ρ
−n
(.
t
, t),
where the convolution is taken with respect to the group structure of H
n
.
The convergence of (10.2.16) is guaranteed by Theorem 10.2.1. Notice also that
the factor 2
n−1
that appears in the formulation of (10.2.16) is due to the fact that
the volume form d\ on the Heisenberg group H
n
has been taken to be
d\ = 2
1−n
dr
1
dj
1
dr
n−1
dj
n−1
dt.
10.3 Local Solvability of the Lewy Operator 255
10.3 Local Solvability of the Lewy Operator
We now return to the local solvability of L
n−1
=
0
b
on the Heisenberg group.
When
b
acts on functions,
0
b
is not hypoelliptic since it annihilates all C1 func
tions. However, we shall show that, modulo the Szeg¨o projection o, there exists a
relative fundamental solution for
0
b
. Rewrite
0
b
as
(10.3.1)
0
b
= L
α
−i(α −n + 1)T,
for α ∈ C. Recall that
ϕ
α
(.
t
, t) = ([.
t
[
2
−it)
−
(n−1+α)
2
([.
t
[
2
+it)
−
(n−1−α)
2
,
and
c
α
=
2
4−2n
π
n
Γ(
n−1+α
2
)Γ(
n−1−α
2
)
.
Then, we have
0
b
ϕ
α
= L
α
ϕ
α
−i(α −n + 1)Tϕ
α
= c
α
δ −i(α −n + 1)Tϕ
α
.
(10.3.2)
Now with the aid of the identity
Γ(n)Γ(1 −n) =
π
sinπn
,
we formally diﬀerentiate (10.3.2) with respect to α and evaluate it at α = n −1 to
get
(10.3.3)
0
b
_
(n −2)!
2
4−2n
π
n
log
_
[.
t
[
2
−it
[.
t
[
2
+it
_
([.
t
[
2
−it)
−n+1
_
= δ−
2(n −1)!
2
4−2n
π
n
([.
t
[
2
−it)
−n
.
Here the logarithm of the quotient means the diﬀerence of the corresponding loga
rithm. Set
(10.3.4) Φ =
(n −2)!
2
4−2n
π
n
log
_
[.
t
[
2
−it
[.
t
[
2
+it
_
([.
t
[
2
−it)
−n+1
,
and deﬁne the operator 1 by
(10.3.5) 1) = ) ∗ Φ,
where the convolution is taken with respect to the group structure on H
n
.
256 Fundamental Solutions for
b
on the Heisenberg Group
Theorem 10.3.1. Let the operator 1 be deﬁned as in (10.3.5), then we have
0
b
1 = 1
0
b
= 1 −o,
when acting on distributions with compact support.
Proof. It suﬃces to show only that
0
b
1 = 1 −o. The other identity then follows
immediately by transposition.
Set ρ
= [.
t
[
2
+c
2
−it, and deﬁne
Φ
(.
t
, t) =
(n −2)!
2
4−2n
π
n
log
_
[.
t
[
2
+c
2
−it
[.
t
[
2
+c
2
+it
_
([.
t
[
2
+c
2
−it)
−n+1
.
Then, by the calculations done in the proof of Theorem 10.1.2, we obtain
0
b
Φ
= −
n−1
k=1
7
k
7
k
Φ
=
(n −2)!
2
4−2n
π
n
_
−4(n −1)
[.
t
[
2
ρ
ρ
n
+ 2(n −1)
1
ρ
ρ
n−1
_
=
(n −1)!
2
4−2n
π
n
_
4c
2
ρ
ρ
n
−
2
ρ
n
_
.
Hence, as c → 0, we get by the integral evaluated in the proof of Theorem 10.1.2,
that
0
b
Φ =
_
4(n −1)!
2
4−2n
π
n
_
Hn
([.
t
[
2
+ 1 −it)
−n
([.
t
[
2
+ 1 +it)
−1
d\
_
δ
−
2(n −1)!
2
4−2n
π
n
1
([.
t
[
2
−it)
n
= δ −
(n −1)!
2
3−2n
π
n
1
([.
t
[
2
−it)
n
.
The assertion now follows from (10.2.16). This proves the theorem.
Theorem 10.3.1 shows that the operator 1 inverts
0
b
on the space of functions
that are orthogonal to the 1
2
integrable C1 functions. It is also clear that o
0
b
=
0
b
o = 0. Then, we have the following local solvability theorem for
0
b
.
Theorem 10.3.2. Let ) ∈ 1
2
(H
n
). The equation
0
b
n = ) is solvable in the 1
2
sense in some neighborhood of j ∈ H
n
if and only if o()) is real analytic in a
neighborhood of j.
Proof. We may assume that ) is an 1
2
integrable function of compact support.
Suppose that o()) is real analytic near j. Then, by the CauchyKowalevski theorem,
there is a real analytic solution n
1
locally such that
0
b
n
1
= o())
10.3 Local Solvability of the Lewy Operator 257
in some neighborhood of j. On the other hand, by Theorem 10.3.1, a solution
n
2
= 1) exists for
0
b
n
2
= (1 −o)).
Hence, n = n
1
+n
2
is a local solution of
0
b
n = ).
Conversely, let n be a local solution of
0
b
n = ). Choose a cutoﬀ function ζ
with ζ = 1 in some open neighborhood of j. Set
0
b
(ζn) = /.
Then, o/ = 0 and ) − / = 0 in some neighborhood of j. Now, from the explicit
formula (10.2.16) of the Szeg¨o projection o, it is easily seen that o()) = o() −/) is
real analytic in some neighborhood of j. This completes the proof of the theorem.
If n = 2, we can deduce the local solvability of the Lewy operator from Theorem
10.3.2.
Theorem 10.3.3. Let 7 = (∂,∂.) − i.(∂,∂t) and ) ∈ 1
2
(H
2
). The equation
7n = ) is locally solvable in the 1
2
sense in some open neighborhood of j ∈ H
2
if
and only if o()) is real analytic in a neighborhood of j.
Proof. ) is still assumed to be an 1
2
integrable function with compact support. If
o()) is real analytic in some neighborhood of j, then Theorem 10.3.2 assures the
existence of a solution · of the equation
0
b
· = −77· = )
which , by conjugation, gives a solution n = −7· of the Lewy equation.
On the other hand, if there exists locally a solution n to the equation 7n = ),
we may assume that n is of compact support. Hence, Theorem 10.3.1 guarantees a
solution · of the equation 7· = n−on. Now, by Theorem 10.3.2 again, we see that
o()) is real analytic in some open neighborhood of j. This proves the theorem.
We note that the Lewy’s example can be extended to any tangential Cauchy
Riemann equation 1 of a hypersurface in C
2
which is not Leviﬂat, i.e., its Levi
form c(r), where
[1, 1] = c(r)T, mod(1, 1),
does not vanish identically in a neighborhood of the reference point. Note also
that, from the discussion at the end of Chapter 7, when the Levi form vanishes
completely, the ∂
b
equation is reduced to a ∂equation with a parameter.
Since we have established in Chapter 9 that the range of the
b
operator on
the boundary of any smooth bounded pseudoconvex domain in C
n
with n ≥ 2 is
closed, the arguments for proving Theorems 10.3.2 and 10.3.3 can then be applied
verbatim to the boundary of any smooth bounded pseudoconvex domain with real
analytic boundary, provided that the following analyticity hypothesis on the Szeg¨o
projection is fulﬁlled:
Analyticity Hypothesis. Let 1 be a smooth bounded pseudoconvex domain in
C
n
, n ≥ 2, with real analytic boundary /1 and j ∈ /1. Let o be the corresponding
Szeg¨o projection on /1. If ) ∈ 1
2
(/1) vanishes on some open neighborhood l of
j ∈ /1, then o) is real analytic on l.
Now, with this hypothesis, we can state the following theorem:
258 Fundamental Solutions for
b
on the Heisenberg Group
Theorem 10.3.4. Let 1 be a smooth bounded pseudoconvex domain in C
2
with
real analytic boundary. Let : be a real analytic deﬁning function for 1 and 1 =
(∂:,∂.
2
)(∂,∂.
1
)−(∂:,∂.
1
)(∂,∂.
2
). Suppose that the Szeg¨o projection o associated
with /1 satisﬁes the analyticity hypothesis. Then the tangential CauchyRiemann
equation 1n = ) is locally solvable for some 1
2
function ) near j ∈ /1 if and only
if o()) is real analytic near j.
Finally, following the arguments of Theorem 10.1.5 we obtain the regularity the
orem of the Lewy operator 7 in the usual H¨older class.
Theorem 10.3.5. Let ) be a continuous function with compact support on H
2
, and
let j ∈ supp). Suppose that o()) is real analytic in some open neighborhood l of
j, then locally there exists a solution n ∈ Λ
1/2
(\ ) on some open neighborhood \ of
j with \ ⊂ l such that 7n = ). Furthermore, if ) ∈ C
k
(H
2
) for / ∈ N ∪ ¦0¦ with
compact support, then n ∈ C
k+
1
2
(\ ).
NOTES
Most of the materials in Sections 10.1 are essentially taken from G. B. Folland
and E. M. Stein [FoSt 1]. Theorem 10.1.1 was proved by G. B. Folland [Fol 1].
The kernel Φ
α
= c
−1
α
ϕ
α
deﬁned in (10.1.11) is homogeneous of order −2n + 2 with
respect to the nonisotropic dilation on H
n
. It follows that the regularity property of
the operator 1
α
) = ) ∗ Φ
α
in the nonisotropic normed spaces can be drawn from a
general theory described in [FoSt 1]. We refer the reader to the book by E. M. Stein
[Ste 4] for a systematic treatment on analysis on Heisenberg groups. The proof of
Theorem 10.1.5 follows that of M.C. Shaw [Sha 9]. The characterization via the
Fourier transform of the Hardy space H
2
(Ω
n
) on the Siegel upper half space was
proved by S. G. Gindikin [Gin 1] (Theorem 10.2.1). The CauchySzeg¨o kernel for
the ball 1
n
in C
n
, n ≥ 2, was found by L. K. Hua [Hua 1] (Proposition 10.2.3), and
for the Siegel upper half space Ω
n
by S. G. Gindikin [Gin 1]. The characterization
of the range of the Lewy operator was proved by Greiner, Kohn and Stein [GKS 1].
(See also [GrSt 1]). We also refer the reader to the books by R. Beals and P. C.
Greiner [BeGr 1] and F. Treves [Tre 3,6] for more discussions on Heisenberg group
and C1 manifolds.
The generalization of the nonsolvability of the Lewy operator to any tangential
CauchyRiemann equation on a hypersurface which is not Leviﬂat in C
2
was proved
by L. H¨ormander [H¨or 1,7]. It is known that the analyticity hypothesis holds on any
smooth bounded strongly pseudoconvex domain with real analytic boundary. For
instance, see [Tar 1,2] and [Tre 2] for n ≥ 3 and [Gel 1] for n = 2. Unfortunately,
there are no general theorems which would guarantee that the Szeg¨o projection
o on weakly pseudoconvex boundaries satisﬁes this hypothesis. One should also
note that, in general, the analytic pseudolocality of o is false on pseudoconvex
boundaries, as shown by M. Christ and D. Geller in [ChGe 1].
259
260
CHAPTER 11
INTEGRAL REPRESENTATIONS
FOR
¯
∂ AND
¯
∂
b
In this chapter the method of integral representation in several complex variables
is discussed. This method can be viewed as a generalization of the Cauchy inte
gral formula in one variable to several variables. The integral kernel method gives
solutions to
¯
∂ and
¯
∂
b
represented by integral formulas on strongly pseudoconvex
domains or boundaries. The representations are especially easy to construct on a
strictly convex domain where solution formulas can be written explicitly. It is in
this setting that we derive integral formulas for
¯
∂ and
¯
∂
b
in this chapter.
The 1
2
approach is fruitful for solving
¯
∂ and
¯
∂
b
in the Sobolev spaces on pseu
doconvex domains and their boundaries. In Chapters 46, the 1
2
method to solve
¯
∂
was discussed using the
¯
∂Neumann problem. In Chapters 8 and 9, we studied the
global solvability and regularity for the tangential CauchyRiemann operator in the
Sobolev spaces on compact C1 manifolds. However, H¨older and 1
p
estimates for
¯
∂
and
¯
∂
b
are not easy to obtain by the 1
2
method. An explicit kernel was computed
in Chapter 10 for
b
on the Heisenberg group and H¨older estimates were obtained
for solutions of
¯
∂
b
. Our goal here is to construct integral formulas for solutions of
¯
∂ and
¯
∂
b
with H¨older and 1
p
estimates on strictly convex domains.
In Section 11.1, some terminology necessary in developing the kernel formulas is
deﬁned. We derive the BochnerMartinelliKoppelman formula as a generalization
of the Cauchy integral. Unlike the Cauchy kernel in C
1
, the BochnerMartinelli
Koppelman kernel is only harmonic, but not holomorphic. Then we introduce
the Leray kernel and derive the homotopy formula for
¯
∂ on convex domains in
Section 11.2. H¨older estimates for the solutions of
¯
∂ on strictly convex domains are
obtained. In Section 11.3 the jump formula derived from the BochnerMartinelli
Koppelman formula is discussed and homotopy formulas for
¯
∂
b
on strictly convex
compact boundaries are constructed and estimated.
The kernel method is especially suitable for the local solvability of
¯
∂
b
on an open
subset with smooth boundary in a strictly convex boundary. It allows the derivation
of an explicit formula of a solution kernel on a domain with boundary in a strictly
pseudoconvex C1 manifold. This is discussed in Section 11.4. The 1
p
estimates for
the local solutions for
¯
∂
b
are proved in Section 11.5. We discuss the
¯
∂
b
Neumann
problem in Section 11.6, which is an analogue for
¯
∂
b
of the
¯
∂Neumann problem.
The 1
2
Hodge decomposition theorem for
¯
∂
b
on an open set with boundary in a
strictly pseudoconvex C1 manifold is proved in Theorem 11.6.4.
11.1 Integral Kernels in Several Complex Variables 261
11.1 Integral Kernels in Several Complex Variables
Our ﬁrst goal is to ﬁnd a fundamental solution of
¯
∂ for (j, ¡)forms in several
complex variables. Since j plays no role in the
¯
∂ equation, we shall assume that
j = 0. In C, the Cauchy kernel is a fundamental solution for
¯
∂. This can be derived
by diﬀerentiating the fundamental solution for ´. Since
1
2π
´log [.[ =
2
π
∂
2
∂¯ .∂.
log [.[ = δ
0
,
where δ
0
is the Dirac delta function centered at 0, we have
2
π
∂
∂¯ .
∂
∂.
log [.[ =
1
π
∂
∂¯ .
1
.
= δ
0
.
This implies that 1(.) = 1,π. is a fundamental solution for ∂,∂¯ .. For any bounded
function ) on C with compact support in 1, where 1 is a bounded domain in C,
we deﬁne
n(.) = ) ∗ 1(.) =
1
π
_
D
)(ζ)
. −ζ
d\ =
1
2πi
_
D
)(ζ)
ζ −.
dζ ∧ d
¯
ζ.
It follows that ∂n,∂¯ . = ) in C in the distribution sense. This can also be proved
directly as in Theorem 2.1.2.
In C
n
when n 1, we can also derive a fundamental solution for the
¯
∂ operator in
the top degree case similarly. Let α = )d¯ .
1
∧ ∧d¯ .
n
be a (0, n)form in C
n
where )
is a bounded function with compact support in C
n
. Since there is no compatibility
condition for α to be solvable, we can derive a solution for the equation
¯
∂n = α as
follows: Let c(.) be a fundamental solution for ´ in C
n
, n ≥ 2, deﬁned by
c(.) = c(:) =
−(n −2)!
4π
n
1
:
2n−2
, : = [.[.
Both c(:) and all the ﬁrst order derivatives of c(:) are locally integrable functions.
We deﬁne a (0, n)form c
n
= −4c(:)d¯ .
1
∧ ∧ d¯ .
n
. Then
¯
∂
¯
∂
∗
c
n
= 4
n
i=1
∂
2
c(:)
∂¯ .
i
∂.
i
d¯ .
1
∧ ∧ d¯ .
n
= δ
0
d¯ .
1
∧ ∧ d¯ .
n
.
We deﬁne
1(.) =
¯
∂
∗
c
n
=
n
j=1
(−1)
j
(n −1)!
π
n
¯ .
j
:
2n
d¯ .
1
∧
∧
d¯ .
j
∧ d¯ .
n
,
where
∧
d¯ .
j
denotes that the term d¯ .
j
is omitted. It follows that
¯
∂1(.) = δ
0
d¯ .
1
∧ ∧ d¯ .
n
.
262 Integral Representations for ∂ and ∂
b
Deﬁne
(11.1.1)
n(.) = ) ∗ 1(.)
=
n
j=1
_
(−1)
j
(n −1)!
π
n
_
D
¯ .
j
−
¯
ζ
j
[. −ζ[
2n
)(ζ)d\
_
[
∧
d¯ .
j
],
where [
∧
d¯ .
j
] = d¯ .
1
∧
∧
d¯ .
j
∧ d¯ .
n
. Then n satisﬁes
¯
∂n = α and 1(.) is a funda
mental solution for
¯
∂ when ¡ = n. For general 0 < ¡ < n, due to the compatibility
condition, the fundamental solution for
¯
∂ is more involved. We introduce some
notation ﬁrst.
Let (ζ − .) = (ζ
1
− .
1
, ζ
2
− .
2
, , ζ
n
− .
n
) ∈ C
n
and dζ = (dζ
1
, , dζ
n
). Let
¹ = (o
1
, , o
n
), 1 = (/
1
, , /
n
) be two vectors in C
n
. We deﬁne
< ¹, 1 =
n
i=1
o
i
/
i
, < ¹, dζ =
n
i=1
o
i
dζ
i
.
Thus <
¯
ζ − ¯ ., ζ −. = [ζ −.[
2
and <
¯
ζ − ¯ ., dζ =
n
i=1
(
¯
ζ
i
− ¯ .
i
)dζ
i
. Let \ be an
open subset of C
n
C
n
with coordinates (ζ, .) and let G(ζ, .) be a C
1
map from
\ into C
n
such that G(ζ, .) = (p
1
(ζ, .), , p
n
(ζ, .)). We deﬁne the (1, 0)form ω
G
by
ω
G
=
1
2πi
< G(ζ, .), dζ
< G(ζ, .), ζ −.
=
1
2πi
n
i=1
p
i
(ζ, .)dζ
i
n
i=1
p
i
(ζ, .)(ζ
i
−.
i
)
on the set of (ζ, .) ∈ C
n
C
n
where < G, ζ −. ,= 0.
When n = 1, ω
G
is independent of G and is the Cauchy kernel. We set the
CauchyRiemann operator on \ equal to
¯
∂
ζ,z
=
¯
∂
ζ
+
¯
∂
z
,
and
<
¯
∂
ζ,z
G(ζ, .), dζ =
n
i=1
¯
∂
ζ,z
p
i
(ζ, .) ∧ dζ
i
.
Let Ω(G) be an (n, n −1)form in (ζ, .) deﬁned by
Ω(G) = ω
G
∧ (
¯
∂
ζ,z
ω
G
)
n−1
= ω
G
∧
¯
∂
ζ,z
ω
G
∧ ∧
¯
∂
ζ,z
ω
G
. ¸¸ .
n−1 times
.
Given : maps G
i
: \ → C
n
, i = 1, , :, we abbreviate ω
G
i
by ω
i
and Ω
1m
is
the (n, n −:)form deﬁned by
Ω
1m
= Ω(G
1
, , G
m
)
= ω
1
∧ ∧ ω
m
∧
k1++km=n−m
(
¯
∂
ζ,z
ω
1
)
k1
∧ ∧ (
¯
∂
ζ,z
ω
m
)
km
on the set where all the denominators are nonvanishing. Since
(11.1.2)
¯
∂
ζ,z
ω
i
=
1
2πi
<
¯
∂
ζ,z
G
i
, dζ
< G
i
(ζ, .), ζ −.
−
1
2πi
<
¯
∂
ζ,z
G
i
, ζ −. ∧ < G
i
, dζ
(< G
i
, ζ −. )
2
,
11.1 Integral Kernels in Several Complex Variables 263
we have for / ≥ 0 that
(11.1.3) ω
i
∧ (
¯
∂
ζ,z
ω
i
)
k
=
_
1
2πi
_
k+1
< G
i
, dζ
< G
i
, ζ −.
∧
_
<
¯
∂
ζ,z
G
i
, dζ
< G
i
, ζ −.
_
k
.
This follows from the fact that ω
i
wedge the last term in (11.1.2) vanishes.
The following lemma is essential in the construction of the kernel formulas.
Lemma 11.1.1. Let G
i
(ζ, .) : \ ⊂ C
n
C
n
→ C
n
, i = 1, , :, be C
1
maps.
We have
(11.1.4)
¯
∂
ζ,z
Ω
1m
=
m
j=1
(−1)
j
Ω
1
∧
jm
on the set where < G
i
, ζ −. ,= 0 for every i = 1, , :, where
∧
, denotes that the
term , is omitted. In particular, we have
(11.1.4i)
¯
∂
ζ,z
Ω
1
= 0,
(11.1.4ii)
¯
∂
ζ,z
Ω
12
= Ω
1
−Ω
2
,
(11.1.4iii)
¯
∂
ζ,z
Ω
123
= −Ω
23
+ Ω
13
−Ω
12
,
on the set where the denominators are nonvanishing.
Proof. We use the notation
(
¯
∂
ζ,z
ω)
Km
= (
¯
∂
ζ,z
ω
1
)
k1
∧ ∧ (
¯
∂
ζ,z
ω
m
)
km
for each multiindex 1
m
= (/
1
, , /
m
) and [1
m
[ = /
1
+ +/
m
. It follows that
¯
∂
ζ,z
Ω
1m
=
m
j=1
(−1)
j−1
ω
1
∧ ∧
¯
∂
ζ,z
ω
j
∧ ∧ ω
m
∧
]Km]=n−m
(
¯
∂
ζ,z
ω)
Km
=
m
j=1
(−1)
j−1
ω
1
∧ ∧
∧
ω
j
∧ ∧ ω
m
∧
]Km]=n−m+1,
k
j
≥1
(
¯
∂
ζ,z
ω)
Km
=
m
j=1
(−1)
j
Ω
1
∧
jm
+
m
j=1
(−1)
j−1
ω
1
∧ ∧
∧
ω
j
∧ ∧ ω
m
∧
]Km]=n−m+1,
k
j
≥0
(
¯
∂
ζ,z
ω)
Km
.
We claim that for each 1
m
such that [1
m
[ = n −:+ 1,
m
j=1
(−1)
j−1
ω
1
∧ ∧
∧
ω
j
∧ ∧ ω
m
∧ (
¯
∂
ζ,z
ω)
Km
= 0.
264 Integral Representations for ∂ and ∂
b
We ﬁrst observe that
(11.1.5) ω
1
∧ ∧ ω
j
∧ ∧ ω
m
∧ (
¯
∂
ζ,z
ω)
Km
= 0,
since there are n + 1 dζ
t
:. Deﬁne
Θ = 2πi
n
j=1
(ζ
j
−.
j
)dζ
j
.
It is easy to see that
(11.1.6) Θ∨ ω
i
= 1, i = 1, , :,
where ∨ denotes the interior product.
Also from (11.1.2), we have
(11.1.7) Θ∨
¯
∂
ζ,z
ω
i
= 0, i = 1, , :.
Contraction of equation (11.1.5) with Θ, using (11.1.6) and (11.1.7), gives
0 = Θ∨ (ω
1
∧ ∧ ω
j
∧ ∧ ω
m
∧ (
¯
∂
ζ,z
ω)
Km
)
=
m
j=1
(−1)
j−1
ω
1
∧ ∧ (Θ∨ ω
j
) ∧ ∧ ω
m
∧ (
¯
∂
ζ,z
ω)
Km
=
m
j=1
(−1)
j−1
ω
1
∧ ∧
∧
ω
j
∧ ∧ ω
m
∧ (
¯
∂
ζ,z
ω)
Km
.
This proves the claim and the lemma.
We also write
Ω
1m
=
n−m
q=0
Ω
1m
q
,
where Ω
1m
q
denotes the piece in Ω
1m
which is of degree (0, ¡) in . and (n, n−:−¡)
degree in ζ. If ) is a (0, ¡
t
)form in C
n
, the form Ω
1m
q
(ζ, .)∧)(ζ) is a form of degree
(n, n−:−¡ +¡
t
) in ζ and of degree (0, ¡) in .. We write the form Ω
1m
q
(ζ, .)∧)(ζ)
uniquely as
t
]J]=q
¹
J
(ζ, .) ∧ d¯ .
J
where ¹
J
(ζ, .) is an (n, n −:−¡ +¡
t
)form in
ζ only with coeﬃcients depending on . and J is an increasing multiindex. We
shall deﬁne the integration of the the form Ω
1m
(ζ, .) ∧ )(ζ) with respect to the ζ
variables on a (2n −)dimensional real manifold ` as follows:
_
M
Ω
1m
(ζ, .) ∧ )(ζ) =
_
M
Ω
1m
q
(ζ, .) ∧ )(ζ)
=
]J]=q
t
__
M
¹
J
(ζ, .)
_
d¯ .
J
,
11.1 Integral Kernels in Several Complex Variables 265
where ¡ =  −:+¡
t
, provided the integral on the righthand side exists. Note that
from this deﬁnition, we have
¯
∂
z
_
M
Ω
1m
(ζ, .) ∧ )(ζ) = (−1)
2n−l
_
M
¯
∂
z
Ω
1m
(ζ, .) ∧ )(ζ)
= (−1)
l
_
M
¯
∂
z
Ω
1m
(ζ, .) ∧ )(ζ),
provided that the diﬀerentiation under the integral sign is allowed.
Let
G
0
(ζ, .) = (
¯
ζ − ¯ .) = (
¯
ζ
1
− ¯ .
1
, ,
¯
ζ
n
− ¯ .
n
).
The BochnerMartinelliKoppelman kernel 1(ζ, .) is deﬁned by
(11.1.8)
1(ζ, .) = Ω(G
0
) = Ω
0
=
1
(2πi)
n
<
¯
ζ − ¯ ., dζ
[ζ −.[
2
∧
_
< d
¯
ζ −d¯ ., dζ
[ζ −.[
2
_
n−1
=
n−1
q=0
1
q
(ζ, .),
where 1
q
is the summand which is of degree (0, ¡) in . and of degree (n, n −¡ −1)
in ζ. Using (11.1.4i), we have
(11.1.9)
¯
∂
ζ
1(ζ, .) +
¯
∂
z
1(ζ, .) = 0 for ζ ,= .,
or equivalently, for each 0 ≤ ¡ ≤ n,
(11.1.9q)
¯
∂
ζ
1
q
(ζ, .) +
¯
∂
z
1
q−1
(ζ, .) = 0 for ζ ,= .,
if we set 1
−1
(ζ, .) = 1
n
(ζ, .) = 0. In particular, 1
0
is the BochnerMartinelli
kernel deﬁned by (2.2.1) and (11.1.9q) was proved directly in (3.2.2) when ¡ = 1.
When n = 1, 1(ζ, .) = (2πi)
−1
dζ,(ζ −.) is the Cauchy kernel. The following
theorem shows that the BochnerMartinelliKoppelman kernel is indeed a general
ization of the Cauchy integral formula to several variables.
Theorem 11.1.2 (BochnerMartinelliKoppelman). Let 1 be a bounded do
main in C
n
with C
1
boundary. For ) ∈ C
1
(0,q)
(1), 0 ≤ ¡ ≤ n, the following formula
holds:
(11.1.10)
)(.) =
_
bD
1
q
(, .) ∧ ) +
_
D
1
q
(, .) ∧
¯
∂
ζ
)
+
¯
∂
z
_
D
1
q−1
(, .) ∧ ), . ∈ 1,
where 1(ζ, .) is deﬁned in (11.1.8).
Proof. For ¡ = 0, the BochnerMartinelli formula was proved in Theorem 2.2.1. We
ﬁrst assume that 1 ≤ ¡ < n.
266 Integral Representations for ∂ and ∂
b
Let .
0
∈ 1 and β
be a small ball of radius c centered at .
0
such that β
⊂ 1.
We shall prove the theorem at . = .
0
. Applying Stokes’ theorem to the form
d
ζ
(1
q
(ζ, .) ∧ )(ζ)) on 1
≡ 1 ¸ β
, we have, using (11.1.9), that
(11.1.11)
_
bD
1
q
(ζ, .) ∧ ) −
_
bβ
1
q
(ζ, .) ∧ )
=
_
D
¯
∂
ζ
1
q
(ζ, .) ∧ ) −
_
D
1
q
(ζ, .) ∧
¯
∂
ζ
)
= −
_
D
¯
∂
z
1
q−1
(ζ, .) ∧ ) −
_
D
1
q
(ζ, .) ∧
¯
∂
ζ
).
Since 1(ζ, .) = O([ζ − .[
−2n+1
), 1(ζ, .) is an integrable function for each ﬁxed ..
We see from the dominated convergence theorem that
(11.1.12)
_
D
1
q
(ζ, .) ∧
¯
∂
ζ
) →
_
D
1
q
(ζ, .) ∧
¯
∂
ζ
).
Note that
_
bβ
1
0
(ζ, .) =
1
(2πi)
n
1
c
2n
_
bβ
<
¯
ζ − ¯ ., dζ ∧ < d
¯
ζ, dζ
n−1
=
1
(2πi)
n
1
c
2n
_
β
< d
¯
ζ, dζ
n
=
n!
π
n
_
β1
d\ = 1.
For any increasing multiindex J = (,
1
, , ,
q
), we get that
_
bβ
1
q
(ζ, .) ∧ d
¯
ζ
J
=
1
(2πi)
n
1
c
2n
_
bβ
<
¯
ζ − ¯ ., dζ ∧ < d
¯
ζ −d¯ ., dζ
n−1
∧d
¯
ζ
J
=
1
(2πi)
n
1
c
2n
_
β
< d
¯
ζ, dζ ∧ < d
¯
ζ −d¯ ., dζ
n−1
∧d
¯
ζ
J
=
(n −¡)(n −1)!
π
n
__
β1
d\
_
∧ d¯ .
J
=
n −¡
n
d¯ .
J
.
Let )(.
0
) denote the (0, ¡)form whose coeﬃcients are equal to the values of the
coeﬃcients of ) at .
0
. It follows from the above calculation that
(11.1.13)
_
bβ
1
q
(ζ, .) ∧ ) =
n −¡
n
)(.
0
) +
_
bβ
1
q
(ζ, .) ∧
_
) −)(.
0
)
_
−→
n −¡
n
)(.
0
), as c →0.
The kernel
¯
∂
z
1
q−1
(ζ, .) = O([ζ −.[
2n
) is not integrable but the components are the
classical singular integrals (see e.g. Stein [Ste 2]). The Principalvalue limit deﬁned
by
P.V.
_
D
¯
∂
z
1
q−1
(ζ, .
0
) ∧ ) = lim
→0
_
D
¯
∂
z
1
q−1
(ζ, .
0
) ∧ )
= lim
→0
_
¯
∂
z
_
D
1
q−1
(ζ, .) ∧ )
_¸
¸
¸
¸
z=z0
11.1 Integral Kernels in Several Complex Variables 267
exists for each .
0
∈ 1. We claim that
(11.1.14) P.V.
_
D
¯
∂
z
1
q−1
(ζ, .
0
) ∧ ) =
_
¯
∂
z
_
D
1
q−1
(ζ, .) ∧ )
_¸
¸
¸
¸
z=z0
−
¡
n
)(.
0
).
We use the notation d
¯
ζ ∧ dζ = d
¯
ζ
1
∧ dζ
1
∧ ∧ d
¯
ζ
n
∧ dζ
n
and [
∧
dζ
j
]= d
¯
ζ
1
∧ dζ
1
∧
∧d
¯
ζ
j
∧
∧
dζ
j
∧ ∧d
¯
ζ
n
∧dζ
n
, where
∧
dζ
j
denotes that the term dζ
j
is omitted. Let
)(ζ) = )
J
d
¯
ζ
J
, where J = (1, , ¡). Using Stokes’ theorem, we obtain
_
β
1
q−1
(ζ, .) ∧ )(ζ)
=
(n −1)!
(2πi)
n
q
j=1
(−1)
j
__
β
¯
ζ
j
− ¯ .
j
[ζ −.[
2n
)
J
(ζ)d
¯
ζ ∧ dζ
_
∧ d¯ .
1
∧
jq
=
(n −1)!
(2πi)
n
q
j=1
(−1)
j
_
β
_
−1
n −1
_
∂
∂ζ
j
_
1
[ζ −.[
2n−2
_
)
J
(ζ) d
¯
ζ ∧ dζ ∧ d¯ .
1
∧
jq
=
(n −1)!
(2πi)
n
q
j=1
(−1)
j
n −1
__
bβ
1
[ζ −.[
2n−2
)
J
(ζ)[
∧
dζ
j
]
+
_
β
1
[ζ −.[
2n−2
∂)
J
∂ζ
j
(ζ)d
¯
ζ ∧ dζ
_
∧ d¯ .
1
∧
jq
.
Thus,
¯
∂
z
_
β
1
q−1
(ζ, .) ∧ )(ζ) =
(n −1)!
(2πi)
n
q
j=1
(−1)
j
n
k=1
__
bβ
ζ
k
−.
k
[ζ −.[
2n
)
J
(ζ)[
∧
dζ
j
]
+
_
β
ζ
k
−.
k
[ζ −.[
2n
∂)
J
∂ζ
j
(ζ)d
¯
ζ ∧ dζ
_
d¯ .
k
∧ d¯ .
1
∧
jq
.
Since
(n −1)!
(2πi)
n
(−1)
j
_
bβ
ζ
k
−.
k
[ζ −.[
2n
)
J
(ζ)[
∧
dζ
j
] −→δ
jk
(−1)
j−1
n
)
J
(.
0
),
and
_
β
ζ
k
−.
k
[ζ −.[
2n
∂)
J
∂ζ
j
(ζ)d
¯
ζ ∧ dζ −→0, as c →0,
we have
¯
∂
z
_
β
1
q−1
(ζ, .) ∧ )(ζ) −→
¡
n
)
J
(.
0
)d¯ .
J
.
268 Integral Representations for ∂ and ∂
b
Thus
¯
∂
z
__
D
1
q−1
(ζ, .) ∧ )(ζ)
_¸
¸
¸
¸
z=z0
= lim
→0
_
_
¯
∂
z
_
D
1
q−1
(ζ, .) ∧ )(ζ) +
¯
∂
z
_
β
1
q−1
(ζ, .) ∧ )(ζ)
_¸
¸
¸
¸
z=z0
_
= lim
→0
_
_
D
¯
∂
z
1
q−1
(ζ, .
0
) ∧ )(ζ) +
¯
∂
z
_
β
1
q−1
(ζ, .) ∧ )(ζ)
¸
¸
¸
¸
z=z0
_
= P.V.
_
D
¯
∂
z
1
q−1
(ζ, .
0
) ∧ )(ζ) +
¡
n
)
J
(.
0
)d¯ .
J
.
This proves the claim (11.1.14) for the special case of ). The proof for a gen
eral (0, ¡)form ) is the same. Combining (11.1.11)(11.1.14), we have proved the
theorem for 0 ≤ ¡ < n. When ¡ = n, it follows from (11.1.1) that
)(.) =
¯
∂
z
_
D
1
n−1
(, .) ∧ ).
Thus Theorem 11.1.2 holds for all 0 ≤ ¡ ≤ n.
Corollary 11.1.3. Let 1 be a bounded domain in C
n
with C
1
boundary. For
any ) ∈ C
(0,q)
(1), 1 ≤ ¡ ≤ n, such that ) = 0 on /1 and
¯
∂) = 0 in 1 in the
distribution sense, there exists n ∈ C
α
(0,q−1)
(1) with
¯
∂n = ) in the distribution
sense, where 0 < α < 1. Furthermore, there exists a C 0 such that
(11.1.15) n
C
α
(D)
≤ C)
L
∞
(D)
.
Proof. For . ∈ 1, deﬁne
n(.) =
_
D
1
q−1
(, .) ∧ ).
We ﬁrst prove (11.1.15). Since
¸
¸
¸
¸
_
D
1
q−1
(, .) ∧ ) −
_
D
1
q−1
(, .
t
) ∧ )
¸
¸
¸
¸
≤ C
_
_
n
j=1
_
D
¸
¸
¸
¸
¯
ζ
j
− ¯ .
j
[ζ −.[
2n
−
¯
ζ
j
− ¯ .
t
j
[ζ −.
t
[
2n
¸
¸
¸
¸
d\
_
_
)
L
∞
(D)
,
it suﬃces to show that for each 1 ≤ , ≤ n,
(11.1.16)
_
D
¸
¸
¸
¸
¯
ζ
j
− ¯ .
j
[ζ −.[
2n
−
¯
ζ
j
− ¯ .
t
j
[ζ −.
t
[
2n
¸
¸
¸
¸
d\ ≤ C[. −.
t
[
¸
¸
log [. −.
t
[
¸
¸
.
Let [. − .
t
[ = 2c. We divide 1 into three regions: β
(.), β
(.
t
) and 1
= 1 ¸
(β
(.) ∪ β
(.
t
)) where β
(.) is a ball of radius c centered at .. On β
(.), we have
_
β(z)
¸
¸
¸
¸
¯
ζ
j
− ¯ .
j
[ζ −.[
2n
−
¯
ζ
j
− ¯ .
t
j
[ζ −.
t
[
2n
¸
¸
¸
¸
d\ ≤ 2
_
β(z)
1
[ζ −.[
2n−1
d\ ≤ C[. −.
t
[.
11.2 The Homotopy Formula for
¯
∂ on Convex Domains 269
Similarly, we have the estimate on β
(.
t
). To estimate the integral on 1
, we note
that
1
3
[ζ −.
t
[ ≤ [ζ −.[ ≤ 3[ζ −.
t
[ for ζ ∈ 1
, thus there exists an ¹ 0 such that
_
D
¸
¸
¸
¸
¯
ζ
j
− ¯ .
j
[ζ −.[
2n
−
¯
ζ
j
− ¯ .
t
j
[ζ −.
t
[
2n
¸
¸
¸
¸
d\ ≤ C
_
≤]ζ−z]≤A
[. −.
t
[
[ζ −.[
2n
d\
≤ C[. −.
t
[
¸
¸
log [. −.
t
[
¸
¸
.
This proves (11.1.16) and (11.1.15) follows.
If ) ∈ C
1
(0,q)
(1), Theorem 11.1.2 implies that
¯
∂n = ) since ) = 0 on /1. For
) ∈ C
(0,q)
(1), we use an approximation argument. We ﬁrst assume that the domain
1 is starshaped and 0 ∈ 1. Let φ(.) = φ([.[) be a function supported in [.[ ≤ 1
and φ ≥ 0,
_
φ = 1. We set φ
δm
= δ
−2n
m
φ(.,δ
m
) for δ
m
¸ 0. Extending ) to be 0
outside 1, we deﬁne
)
m
(.) = )
_
.
1 −
1
m
_
∗ φ
δm
for suﬃciently small δ
m
. One can easily check that )
m
has coeﬃcients in C
∞
0
(1),
¯
∂)
m
= 0 in 1 and )
m
→ ) uniformly in 1. When the boundary is C
1
, we can
use a partition of unity ¦ζ
i
¦
N
i=1
, with each ζ
i
supported in an open set l
i
such that
l
i
∩ 1 is starshaped. We then regularize ζ
i
) in l
i
as before. It is easy to see
that there exists a sequence )
m
∈ C
∞
(0,q)
(1) with compact support in 1 such that
)
m
→) uniformly in 1 and
¯
∂)
m
→0 uniformly. Applying Theorem 11.1.2 to each
)
m
and letting : →∞, we have proved
¯
∂n = ) in the distribution sense.
Corollary 11.1.3 allows us to solve the equation
¯
∂n = ) for any
¯
∂closed form
) with compact support. Thus the BochnerMartinelliKoppelman kernel is a fun
damental solution for
¯
∂ in C
n
. In the next section we introduce new kernels and
derive a homotopy formula for
¯
∂ for forms which do not necessarily have compact
support.
11.2 The Homotopy Formula for
¯
∂ on Convex Domains
The BochnerMartinelliKoppelman kernel is independent of the domain 1. Next
we introduce another kernel, the Leray kernel, which in general depends on the
domain.
Deﬁnition 11.2.1. A C
n
valued C
1
function G(ζ, .) = (p
1
(ζ, .), , p
n
(ζ, .)) is
called a Leray map for 1 if it satisﬁes < G(ζ, .), ζ − . ,= 0 for every (ζ, .) ∈
/1 1.
In particular, the C
n
valued function G
0
(ζ, .) = (
¯
ζ − ¯ .) = (
¯
ζ
1
− ¯ .
1
, ,
¯
ζ
n
− ¯ .
n
)
is a Leray map for any domain 1. We use the same notation Ω
0
= Ω(G
0
) = 1(ζ, .)
to denote the BochnerMartinelliKoppelman kernel. If G
1
(ζ, .) is another Leray
map, we set
(11.2.1) Ω
1
= Ω(G
1
) =
_
1
2πi
_
n
< G
1
, dζ
< G
1
, ζ −.
∧
_
<
¯
∂
ζ,z
G
1
, dζ
< G
1
, ζ −.
_
n−1
270 Integral Representations for ∂ and ∂
b
and
(11.2.2)
Ω
01
= Ω(G
0
, G
1
) =
_
1
2πi
_
n
<
¯
ζ − ¯ ., dζ
[ζ −.[
2
∧
< G
1
, dζ
< G
1
, ζ −.
∧
k1+k2=n−2
_
< d
¯
ζ −d¯ ., dζ
[ζ −.[
2
_
k1
∧
_
<
¯
∂
ζ,z
G
1
, dζ
< G
1
, ζ −.
_
k2
.
Notice that Ω
1
and Ω
01
are well deﬁned for ζ ∈ /1 and . ∈ 1. Also we use the
notation Ω
1
q
, Ω
01
q
to denote the summand of forms with degree (0, ¡) in . in Ω
1
, Ω
01
respectively.
Theorem 11.2.2 (LerayKoppelman). Let 1 be a bounded domain in C
n
with
C
1
boundary. Let G
0
= (
¯
ζ − ¯ .) and G
1
be another Leray map for 1. For ) ∈
C
1
(0,q)
(1), 0 ≤ ¡ ≤ n, we have
(11.2.3) )(.) =
_
bD
Ω
1
q
∧ ) +
¯
∂
z
T
q
) +T
q+1
¯
∂), . ∈ 1,
where
T
q
)(.) =
_
D
Ω
0
q−1
(ζ, .) ∧ )(ζ) −
_
bD
Ω
01
q−1
(ζ, .) ∧ )(ζ).
Ω
0
, Ω
1
and Ω
01
are deﬁned in (11.1.8), (11.2.1) and (11.2.2) respectively.
Proof. From (11.1.4ii), we have
¯
∂
ζ,z
Ω
01
= Ω
0
−Ω
1
on the set where ζ ∈ /1 and . ∈ 1. Thus, for . ∈ 1,
_
bD
Ω
0
∧ ) =
_
bD
¯
∂
ζ,z
Ω
01
∧ ) +
_
bD
Ω
1
∧ ).
Applying Stokes’ theorem, we have
_
bD
¯
∂
ζ
Ω
01
∧ ) =
_
bD
d
ζ
(Ω
01
∧ )) −
_
bD
Ω
01
∧
¯
∂
ζ
) = −
_
bD
Ω
01
∧
¯
∂
ζ
).
Since Ω
01
∧ ) is of degree (n, n −1) in ζ, it follows that
_
bD
¯
∂
z
Ω
01
∧ ) = −
¯
∂
z
_
bD
Ω
01
∧ ).
Substituting the above formulas into (11.1.10), we have for . ∈ 1,
)(.) =
_
bD
Ω
0
∧ ) +
_
D
Ω
0
∧
¯
∂
ζ
) +
¯
∂
z
_
D
Ω
0
∧ )
=
_
bD
Ω
1
∧ ) +
¯
∂
z
__
D
Ω
0
∧ ) −
_
bD
Ω
01
∧ )
_
+
__
D
Ω
0
∧
¯
∂
ζ
) −
_
bD
Ω
01
∧
¯
∂
ζ
)
_
.
(11.2.3) follows from the degree consideration.
11.2 The Homotopy Formula for
¯
∂ on Convex Domains 271
Corollary 11.2.3 (Leray). Let 1 be a bounded domain in C
n
with C
1
boundary.
Let G
1
be any Leray map for 1. For any ) ∈ C
1
(1) ∩ O(1), we have
)(.) =
_
bD
Ω
1
0
(ζ, .) ∧ )(ζ), . ∈ 1,
where Ω
1
is deﬁned in (11.2.1) and Ω
1
0
is the piece in Ω
1
of degree (0,0) in ..
Corollary 11.2.3 shows that a holomorphic function in 1 is represented by its
boundary value through any Leray map for 1. So far we have not constructed any
Leray map other than the BochnerMartinelliKoppelman kernel. Our next goal is
to construct a Leray map which is holomorphic in the . variable when the domain
is convex. We recall the following deﬁnition:
Deﬁnition 11.2.4. Let 1 ⊂⊂ 1
N
be a domain with C
2
boundary and ρ is any C
2
deﬁning function. 1 is a convex (or strictly convex) domain with C
2
boundary if
N
i,j=1
∂
2
ρ
∂r
i
∂r
j
(r)o
i
o
j
≥ 0 (o: 0) on /1,
for every o = (o
1
, , o
N
) ,= 0 with
N
i=1
o
i
∂ρ
∂xi
(r) = 0 on /1. Here we use
(r
1
, , r
N
) to denote the real coordinates for 1
N
and o
i
∈ 1.
It is easy to check that the deﬁnition of convexity or strict convexity is indepen
dent of the choice of the deﬁning function ρ. In fact, for a strictly convex domain
1, we can choose a special deﬁning function such that its real Hessian is positive
deﬁnite without restricting to the tangent plane as the next proposition shows.
Proposition 11.2.5. Let 1 be a strictly convex domain with C
2
boundary in 1
N
.
Then there exists a C
2
deﬁning function ρ such that
(11.2.4)
N
i,j=1
∂
2
ρ
∂r
i
∂r
j
(r)o
i
o
j
≥ c[o[
2
, for all r ∈ /1and o ∈ 1
n
,
where c is a positive constant.
Proof. Let ρ
0
be any C
2
deﬁning function for 1. We set ρ = c
Aρ0
− 1 where ¹
is a positive constant. Then ρ is another C
2
deﬁning function. Arguments similar
to those in the proof of Theorem 3.4.4 show that ρ is strictly convex and satisﬁes
(11.2.4) if we choose ¹ suﬃciently large.
A deﬁning function ρ satisfying (11.2.4) is called a strictly convex deﬁning func
tion for 1. By continuity, ρ satisﬁes (11.2.4) in a small neighborhood of /1.
Lemma 11.2.6. Let 1 be a bounded convex domain in C
n
with C
2
boundary and
let ρ be a C
2
deﬁning function for 1. Then the map
(11.2.5) G
1
(ζ, .) =
_
∂ρ
∂ζ
_
=
_
∂ρ
∂ζ
1
, ,
∂ρ
∂ζ
n
_
is a Leray map.
272 Integral Representations for ∂ and ∂
b
Proof. Using convexity, we have for any . ∈ 1, ζ ∈ /1,
(11.2.6) Re
n
i=1
∂ρ
∂ζ
i
(ζ
i
−.
i
) 0.
Thus G
1
is a Leray map.
Note that G
1
is a Leray map which is independent of .. The importance of the
existence of a Leray map which is holomorphic in . (or independent of .) is shown
in the next theorem.
Theorem 11.2.7 (A homotopy formula for
¯
∂ on convex domains). Let 1
be a bounded convex domain in C
n
with C
2
boundary /1 and let ρ be a C
2
deﬁning
function for 1. Suppose that G
0
= (
¯
ζ − ¯ .) and G
1
is deﬁned by (11.2.5). For
) ∈ C
1
(0,q)
(1), 0 ≤ ¡ ≤ n, we have
(11.2.7) )(.) =
¯
∂
z
T
q
) +T
q+1
¯
∂), . ∈ 1, if 1 ≤ ¡ ≤ n,
(11.2.8) )(.) =
_
bD
Ω
1
0
∧ ) +T
1
¯
∂), . ∈ 1, if ¡ = 0,
where
(11.2.9) T
q
)(.) =
_
D
Ω
0
q−1
(ζ, .) ∧ )(ζ) −
_
bD
Ω
01
q−1
(ζ, .) ∧ )(ζ).
Ω
0
, Ω
1
and Ω
01
are deﬁned in (11.1.8), (11.2.1) and (11.2.2) respectively.
Proof. Since G
1
is a Leray map which does not depend on ., the kernel Ω
1
has no
d¯ .’s. Thus for any 1 ≤ ¡ ≤ n, Ω
1
q
= 0 and
_
bD
Ω
1
q
∧ ) = 0.
Thus (11.2.7) and (11.2.8) follow from (11.2.3).
Corollary 11.2.8 (A solution operator for
¯
∂ on convex domains). Let 1 be a
bounded convex domain in C
n
with C
2
boundary /1. Let ) ∈ C
1
(0,q)
(1), 1 ≤ ¡ ≤ n,
with
¯
∂) = 0 in 1. Then
n = T
q
)(.)
is a solution to the equation
¯
∂n = ), where T
q
is deﬁned in (11.2.9).
Formula (11.2.9) gives an explicit solution operator for
¯
∂ when the domain is
convex. Next we shall estimate the solution kernel in H¨older spaces when the
domain is strictly convex.
11.2 The Homotopy Formula for
¯
∂ on Convex Domains 273
Lemma 11.2.9. Let 1 be a bounded strictly convex domain in C
n
with C
2
boundary
/1 and let ρ be a strictly convex deﬁning function for 1. There exists a constant
C 0 such that for any ζ ∈ /1, . ∈ 1,
(11.2.10) Re < G
1
, ζ −. ≥ C(ρ(ζ) −ρ(.) +[ζ −.[
2
),
where G
1
is deﬁned by (11.2.5).
Proof. Since ρ is a strictly convex deﬁning function satisfying (11.2.4), we apply
Taylor’s expansion to ρ(.) at the point ζ ∈ /1, then
ρ(.) = ρ(ζ) −2Re
n
i=1
∂ρ
∂ζ
i
(ζ
i
−.
i
) +
n
i,j=1
∂
2
ρ
∂ζ
i
∂
¯
ζ
j
(ζ
i
−.
i
)(
¯
ζ
j
− ¯ .
j
)
+ Re
n
i,j=1
∂
2
ρ
∂ζ
i
∂ζ
j
(ζ
i
−.
i
)(ζ
j
−.
j
) +o([ζ −.[
2
).
Thus, for [ζ −.[ ≤ c, where c 0 is suﬃciently small,
Re
n
i=1
∂ρ
∂ζ
i
(ζ
i
−.
i
) ≥
1
2
ρ(ζ) −
1
2
ρ(.) +
c
2
[ζ −.[
2
,
where c 0 is the positive constant in (11.2.4). To show that (11.2.10) holds for
[ζ −.[ c, we set
˜ . =
_
1 −
c
[ζ −.[
_
ζ +
c
[ζ −.[
..
Then [ζ − ˜ .[ = c and ˜ . ∈ 1 since 1 is strictly convex. It follows that
Re < G
1
, ζ −. = Re
[ζ −.[
c
< G
1
, ζ − ˜ .
≥
[ζ −.[
2c
(ρ(ζ) −ρ(˜ .) +
c
2
[ζ − ˜ .[
2
)
≥
cc
2
4
≥ C(ρ(ζ) −ρ(.) +[ζ −.[
2
),
since (−ρ(.) +[ζ −.[
2
) ≤ ` for some constant ` 0.
Lemma 11.2.10. Let 1 be a bounded strictly convex domain in C
n
with C
2
bound
ary and let ρ be a strictly convex deﬁning function for 1. The kernel Ω
01
q−1
(ζ, .),
0 < ¡ < n, is absolutely integrable on /1 for any . ∈ 1. Furthermore, there exists
a constant C such that for any . ∈ 1,
(11.2.11)
_
bD
[Ω
01
q−1
(ζ, .)[ < C,
where C is independent of ..
Proof. Let
_
Φ(ζ, .) =< G
1
, ζ −. ,
Φ
0
(ζ, .) = [ζ −.[
2
.
274 Integral Representations for ∂ and ∂
b
Using (11.2.10), the kernel Ω
01
(ζ, .) has singularities only at ζ = . on /1.
We choose a special coordinate system in a neighborhood of a ﬁxed . near /1.
From the deﬁnition of Φ, we have d
ζ
Φ[
ζ=z
= ∂ρ and d
ζ
(Im Φ)[
ζ=z
=
1
2i
(∂ρ − ∂ρ).
Thus dρ and d
ζ
(Im Φ) are linearly independent at ζ = .. On a small neighborhood
l
= ¦ζ [ [ζ − .[ < c¦ of a ﬁxed . ∈ 1, Let (t, j) = (t
1
, , t
2n−1
, j) where
t = (t
t
, t
2n−1
) = (t
1
, , t
2n−1
) are tangential coordinates for l
∩ /1, t
i
(.) = 0
and
(11.2.12)
_
j = ρ(ζ),
t
2n−1
= ImΦ(ζ, .).
From (11.2.10) it follows that there exists a positive constant γ
0
such that
(11.2.13)
_
[Φ(ζ, .)[ ≥ γ
0
([ρ(.)[ +[t
t
[
2
+[t
2n−1
[),
[ζ −.[ ≥ γ
0
([ρ(.)[ +[t[).
Using (11.2.13), we have for some ¹ 0,
(11.2.14)
_
bD∩U
[Ω
01
q−1
(ζ, .)[ ≤ C
_
n−1
k=1
_
bD∩U
[ζ −.[
[Φ[
n−k
[Φ
0
[
k
do
_
≤ C
_
ζ∈bD∩U
1
[Φ[[ζ −.[
2n−3
do
≤ C
_
]t]≤A
dt
1
dt
2
dt
2n−1
([t
2n−1
[ +[t
t
[
2
)[t[
2n−3
≤ C
_
]t
]≤A
[ log [t
t
[[dt
1
dt
2n−2
[t
t
[
2n−3
≤ C
_
A
0
:
2n−3
[ log :[d:
:
2n−3
≤ C,
where do is the surface element on /1. This proves the lemma.
Thus the kernel Ω
01
is integrable uniformly on /1. We have the following H¨older
regularity result for
¯
∂:
Theorem 11.2.11 (
1
2
H¨older estimates for
¯
∂ on strictly convex domains).
Let 1 be a bounded strictly convex domain in C
n
with C
2
boundary. For any
) ∈ C
(0,q)
(1), 1 ≤ ¡ ≤ n, such that
¯
∂) = 0 in 1, there exists a n ∈ C
1/2
(0,q−1)
(1)
such that
¯
∂n = ) in 1 and
(11.2.15) n
C
1
2 (D)
≤ C)
L
∞
(D)
,
where C is a constant independent of ).
Proof. We ﬁrst assume that ) ∈ C
1
(0,q)
(1). Let
n = T
q
) = n
0
+n
1
,
11.2 The Homotopy Formula for
¯
∂ on Convex Domains 275
where
n
0
=
_
D
Ω
0
q−1
(ζ, .) ∧ )(ζ)
and
n
1
= −
_
bD
Ω
01
q−1
(ζ, .) ∧ )(ζ).
It follows from Corollary 11.2.8 that
¯
∂T
q
) = ). From Corollary 11.1.3, for any
., .
t
∈ 1,
[n
0
(.) −n
0
(.
t
)[ ≤ C
α
 ) 
∞
[. −.
t
[
α
for any α < 1. Also n
1
is smooth in 1. In order to estimate n
1
near the boundary,
we use the assumption of strict convexity on 1.
We may assume 1 ≤ ¡ ≤ n − 1 since n
1
= 0 if ¡ = n. From Lemma 11.2.10,
n
1
∈ C
(0,q−1)
(1). Since n
1
∈ C
∞
(0,q−1)
(1), using the HardyLittlewood lemma (see
Theorem C.1 in the Appendix), to prove that n
1
∈ C
1/2
(0,q−1)
(1), it suﬃces to show
that there exists a C such that
(11.2.16) [∇n
1
(.)[ ≤ C[ρ(.)[
−
1
2
, . ∈ 1.
Using the same notation as in Lemma 11.2.10, we have for 1 ≤ ¡ ≤ n −1,
¸
¸
¸
¸
∇
z
_
bD∩U
Ω
01
q−1
(ζ, .) ∧ )(ζ)
¸
¸
¸
¸
≤ C  ) 
∞
n−1
k=1
_
_
_
bD∩U
[ζ −.[
[Φ[
n−k+1
[Φ
0
[
k
do +
_
bD∩U
1
[Φ[
n−k
[ζ −.[
2k
do
_
_
.
To prove (11.2.16), using the change of coordinates (11.2.12) and estimates (11.2.13),
it suﬃces to show that for some ¹ 0, there exists a C 0 such that for δ 0,
1 ≤ ¡ ≤ n −1,
(11.2.17)
_
]t]≤A
dt
1
dt
2
dt
2n−1
(δ +[t
2n−1
[ +[t
t
[
2
)
n−q+1
[t[
2q−1
< Cδ
−
1
2
,
(11.2.18)
_
]t]≤A
dt
1
dt
2
dt
2n−1
(δ +[t
2n−1
[ +[t
t
[
2
)
n−q
([t[ +δ)
2q
< C
α
δ
−1+α
,
where 0 < α < 1 and C, C
α
are independent of δ. To prove (11.2.17), integrating
with respect to t
2n−1
and then using polar coordinates [t
t
[ = :, we have
_
]t]≤A
dt
1
dt
2
dt
2n−1
(δ +[t
2n−1
[ +[t
t
[
2
)
n−q+1
[t[
2q−1
≤ C
_
]t
]≤A
dt
1
dt
2
dt
2n−2
(δ +[t
t
[
2
)
n−q
[t
t
[
2q−1
≤ C
_
A
0
:
2n−3
d:
(δ +:
2
)
n−q
:
2q−1
≤ C
_
A
0
d:
δ +:
2
≤ Cδ
−
1
2
.
276 Integral Representations for ∂ and ∂
b
(11.2.18) is proved similarly. Thus n
1
∈ C
1/2
(0,q−1)
(1) and (11.2.15) is proved for
) ∈ C
1
(0,q)
(1).
When ) ∈ C
(0,q)
(1), we can ﬁnd a sequence )
∈ C
∞
(0,q)
(1) such that )
→ )
uniformly in 1 and
¯
∂)
= 0 in 1. The )
’s can be constructed easily by a dilation
(assuming that 0 ∈ 1) followed by regularization. Letting n
= T
q
)
, we get
n

C
1
2 (D)
≤ C)

L
∞
(D)
.
It is easy to see that n
converges C
1/2
(1) to n = T
q
) ∈ C
1/2
(0,q−1)
(1) and
¯
∂n = )
in the distribution sense.
Remark. In Chapter 5, we have proved that for any
¯
∂closed (0, ¡)form with
\
s
(1) coeﬃcients in a strictly pseudoconvex domain 1, the canonical solution n
given by
¯
∂
∗
·) is in \
s+
1
2
(1) where · is the
¯
∂Neumann operator (see Theo
rem 5.2.6). Theorem 11.2.11 gives a solution operator which has a “gain” of 1,2
derivatives in H¨older spaces on strictly convex domains. Near a boundary point
of a strictly pseudoconvex domain, locally there exists a holomorphic change of
coordinates such that it is strictly convex (see Corollary 3.4.5). Globally, one can
also embed strongly pseudoconvex domains in C
n
into strictly convex domains in
C
N
for some large · (see, e.g., Fornaess [For 2]). The H¨older 1,2estimates proved
in Theorem 11.2.11 can be extended to any strictly pseudoconvex domain, but we
omit the details here.
11.3 Homotopy Formulas for
¯
∂
b
on Strictly Convex Boundaries
Let 1 be a bounded domain in C
n
with C
2
boundary and let ρ be a C
2
deﬁning
function for 1, normalized such that [dρ[ = 1 on /1. ) is a (0, ¡)form on /1 with
continuous coeﬃcients, denoted by ) ∈ C
(0,q)
(/1), if and only if
(11.3.1) ) = τp,
where p is a continuous (0, ¡)form in C
n
and τ is the projection operator from
(0, ¡)forms in C
n
onto (0, ¡)forms on /1 ( i.e., (0, ¡)forms which are pointwise
orthogonal to
¯
∂ρ ). (11.3.1) is also equivalent to the following condition: for any
continuous (n, n −¡ −1)form φ deﬁned in a neighborhood of /1, we have
(11.3.2)
_
bD
) ∧ φ =
_
bD
p ∧ φ.
To see that (11.3.1) and (11.3.2) are equivalent, we note that for any (0, ¡ −1)form
/ with continuous coeﬃcients in C
n
,
_
bD
¯
∂ρ ∧ / ∧ φ =
_
bD
(dρ −∂ρ) ∧ / ∧ φ = 0.
The space of (0, ¡)forms with H¨older or 1
p
coeﬃcients are denoted by C
α
(0,q)
(/1)
or 1
p
(0,q)
(/1), where 0 < α < 1 and 1 ≤ j ≤ ∞. If n ∈ 1
p
(0,q−1)
(/1), u satisﬁes
11.3 Homotopy Formulas for
¯
∂
b
on Strictly Convex Boundaries 277
¯
∂
b
n = ) for some ) ∈ 1
p
(0,q)
(/1) in the distribution sense if and only if for any
φ ∈ C
∞
(n,n−1−q)
(C
n
),
(11.3.3)
_
bD
n ∧
¯
∂φ = (−1)
q
_
bD
) ∧ φ.
Let 1
−
= 1 and 1
+
= C
n
¸ 1. We deﬁne the BochnerMartinelliKoppelman
transform for any ) ∈ C
(0,q)
(/1) as follows:
(11.3.4)
_
bD
1
q
(ζ, .) ∧ )(ζ) =
_
1
−
(.), if . ∈ 1
−
,
1
+
(.), if . ∈ 1
+
.
It is easy to see that 1
−
∈ C
∞
(0,q)
(1
−
) and 1
+
∈ C
∞
(0,q)
(1
+
). In fact, 1
−
and
1
+
are continuous up to the boundary if ) is H¨older continuous and we have the
following jump formula:
Theorem 11.3.1. (BochnerMartinelliKoppelman jump formula). Let 1
be a bounded domain in C
n
with C
2
boundary. Let ) ∈ C
1
(0,q)
(/1), where 0 ≤ ¡ ≤
n −1. Then
(11.3.5) 1
−
∈ C
α
(0,q)
(1
−
), 1
+
∈ C
α
(0,q)
(1
+
),
for every α with 0 < α < 1 and
(11.3.6) ) = τ(1
−
−1
+
), . ∈ /1.
Proof. We ﬁrst assume that the boundary /1 is ﬂat with /1 = ¦(.
1
, , .
n
) ∈ C
n
[
Im.
n
= 0¦ and ) has compact support in /1. The coeﬃcients of 1(ζ, .) are of the
form
¯
ζ
j
− ¯ .
j
[ζ −.[
2n
, , = 1, , n.
We rename the real coordinates .
j
= r
j
+ij
j
by setting r
j+n
= j
j
, , = 1, , n−1,
and j = j
n
. Similarly we set ξ
j+n
= η
j
, , = 1, , n −1, where ζ
j
= ξ
j
+ iη
j
. Set
r = (r
1
, , r
2n−1
), ξ = (ξ
1
, , ξ
2n−1
) and . = (r, j). We deﬁne
_
j
y
(ξ) =
j
([ξ[
2
+j
2
)
n
, j 0,
¡
j
y
(ξ) =
ξ
j
([ξ[
2
+j
2
)
n
, j 0, , = 1, , 2n −1.
Then j
y
is a constant multiple of the Poisson kernel for the upper half space
1
2n
+
= ¦.[ j 0¦ and the ¡
j
y
’s are the conjugate Poisson kernels. If we write
) =
t
]I]=q
)
I
d¯ .
I
, then each summand in
_
bD
1(ζ, .) ∧ ) is a constant multiple of
the following form:
1)
I
(.) =
_
ξ∈R
2n−1
j
([ξ −r[
2
+j
2
)
n
)
I
(ξ) dξ
1
∧ ∧ dξ
2n−1
= j
y
∗ )
I
,
278 Integral Representations for ∂ and ∂
b
or
Q
j
)
I
(.) =
_
ξ∈R
2n−1
ξ
j
−r
j
([ξ −r[
2
+j
2
)
n
)
I
(ξ) dξ
1
∧ ∧ dξ
2n−1
= ¡
j
y
∗ )
I
,
where , = 1, , 2n − 1. The Poisson integral 1 is bounded from C
0
(1
2n−1
) to
C(1
2n
+
). Since it ts a convolution operator, it is bounded from C
1
0
(1
2n−1
) to
C
1
(1
2n
+
).. The integral Q
j
)
I
is the Poisson integral of the Riesz transform of )
I
. To
see that Q
j
)
I
is bounded from C
1
0
(1
2n−1
) to C
α
(1
2n
+
), we use integration by parts
and arguments similar to those used in Corollary 11.1.3. This proves (11.3.5) when
the boundary is ﬂat. For the general case, we note that the BochnerMartinelli
Koppelman kernel is obtained by diﬀerentiation of the fundamental solution c(.)
for ´(c.f. 11.1). Using integration by parts and arguments in the proof of Corollary
11.1.3, one can also prove similarly that 1
−
∈ C
α
(0,q)
(1
−
) and 1
+
∈ C
α
(0,q)
(1
+
).
To prove (11.3.6), we ﬁrst extend ) to 1 such that the extension, still denoted
by ), is in C
1
(0,q)
(1). From Theorem 11.1.2, we have
(11.3.7)
_
bD
1
q
(, .) ∧ ) +
_
D
1
q
(, .) ∧
¯
∂
ζ
) +
¯
∂
z
_
D
1
q−1
(, .) ∧ )
=
_
)(.), . ∈ 1,
0, . ∈ C
n
¸ 1.
When . ∈ 1, (11.3.7) was proved in (11.1.10). From the proof of (11.1.10), it is
easy to see that (11.3.7) holds for . ∈ C
n
¸ 1. Since 1(ζ, .) is an integrable kernel
in C
n
, the term
_
D
1(, .) ∧
¯
∂
ζ
) is continuous up to the boundary /1. We denote
by ν
z
the outward unit normal to /1 at .. Then for . ∈ /1,
lim
→0
+
__
D
1
q
(, . −cν
z
) ∧
¯
∂
ζ
) −
_
D
1
q
(, . +cν
z
) ∧
¯
∂
ζ
)
_
= 0.
It remains to see that the term
¯
∂
z
_
D
1(, .) ∧ ) when restricted to the boundary
has no jump in the complex tangential component. For any φ ∈ C
∞
(n,n−q−1)
(C
n
),
we have
(11.3.8)
lim
→0
+
_
bD
¯
∂
z
__
D
1
q−1
(ζ, . −cν
z
) ∧ )(ζ)
_
∧ φ(.)
= (−1)
q
lim
→0
+
_
bD
__
D
1
q−1
(ζ, . −cν
z
) ∧ )(ζ)
_
∧
¯
∂
z
φ(.)
= (−1)
q
_
bD
__
D
1
q−1
(ζ, .) ∧ )(ζ)
_
∧
¯
∂
z
φ(.).
Similarly, we obtain
(11.3.9)
lim
→0
+
_
bD
¯
∂
z
__
D
1
q−1
(ζ, . +cν
z
) ∧ )(ζ)
_
∧ φ(.)
= (−1)
q
_
bD
__
D
1
q−1
(ζ, .) ∧ )(ζ)
_
∧
¯
∂
z
φ(.).
11.3 Homotopy Formulas for
¯
∂
b
on Strictly Convex Boundaries 279
Thus from (11.3.7)(11.3.9), we get for any φ ∈ C
∞
(n,n−q−1)
(C
n
),
_
bD
)(.) ∧ φ(.) = lim
→0
+
_
bD
__
bD
[1
q
(ζ, . −cν
z
) −1
q
(ζ, . +cν
z
)] ∧ )(ζ)
_
∧ φ(.)
=
_
bD
[1
−
(.) −1
+
(.)] ∧ φ(.).
Using (11.3.2), we have proved (11.3.6). This proves the theorem.
Corollary 11.3.2. Let 1 be a bounded domain in C
n
with C
2
boundary. For any
) ∈ C
1
(0,q)
(/1) with
¯
∂
b
) = 0 on /1, we have
) = τ(1
−
−1
+
), . ∈ /1,
and
¯
∂1
−
= 0 in 1
−
,
¯
∂1
+
= 0 in 1
+
. Furthermore, we have 1
−
∈ C
α
(0,q)
(1
−
),
1
+
∈ C
α
(0,q)
(1
+
) for any 0 < α < 1.
Proof. Since 1
−
∈ C
∞
(0,q)
(1), diﬀerentiation under the integral sign and Stoke’s
theorem imply that for . ∈ 1,
¯
∂
z
1
−
(.) = −
_
bD
¯
∂
z
1
q
(ζ, .) ∧ )(ζ)
=
_
bD
¯
∂
ζ
1
q+1
(ζ, .) ∧ )(ζ)
=
_
bD
d
ζ
(1(ζ, .) ∧ )(ζ)) +
_
bD
1
q+1
(ζ, .) ∧
¯
∂
ζ
)(ζ)
= 0.
Here we have used (11.1.9). Similarly,
¯
∂1
+
= 0 in 1
+
. Using Theorem 11.3.1, the
corollary is proved.
One should compare Corollary 11.3.2 with Lemma 9.3.5. When ¡ = 0, Corollary
11.3.2 implies that any C1 function ) can be written as the diﬀerence of two
holomorphic functions. Thus Corollary 11.3.2 generalizes the Plemelj jump formula
in C proved in Theorem 2.1.3.
From Corollary 11.3.2, every
¯
∂
b
closed form can be written as the jump of two
¯
∂closed forms. Solving
¯
∂
b
is reduced to solving the
¯
∂ problem on both 1
−
and
1
+
. When 1 is strictly convex, we have already discussed how to solve
¯
∂ on 1 by
integral formulas. We shall use Theorem 11.3.1 to derive homotopy formulas for
¯
∂
b
when 1 is a strictly convex domain with C
2
boundary.
Let ρ be a strictly convex deﬁning function for 1. Deﬁne C
1
functions G
−
and
G
+
in C
n
C
n
by
G
−
(ζ, .) =
_
∂ρ
∂ζ
1
, ,
∂ρ
∂ζ
n
_
, (11.3.10)
G
+
(ζ, .) =
_
−
∂ρ
∂.
1
, , −
∂ρ
∂.
n
_
. (11.3.11)
Using Lemma 11.2.6, G
−
(ζ, .) is a Leray map for 1. Let
G
0
(ζ, .) = (
¯
ζ
1
− ¯ .
1
, ,
¯
ζ
n
− ¯ .
n
)
be the same as before.
280 Integral Representations for ∂ and ∂
b
Lemma 11.3.3. Let 1 be a strictly convex domain in C
n
with C
2
boundary and ρ
be a C
2
strictly convex deﬁning function for 1. For ζ, . ∈ /1, the kernels
(11.3.12) Ω
−
= Ω(G
−
), Ω
+
= Ω(G
+
),
(11.3.13) Ω
−0
= Ω(G
−
, G
0
), Ω
+0
= Ω(G
+
, G
0
)
have singularities only when ζ = .. Furthermore, there exists a constant C 0
independent of . such that
(11.3.14)
_
bD
([Ω
−0
(ζ, .)[ +[Ω
+0
(ζ, .)[) < C, . ∈ /1.
Proof. Set
_
Φ(ζ, .) = < G
−
(ζ, .), ζ −. ,
Ψ(ζ, .) = < G
+
(ζ, .), ζ −. .
Note that Φ(ζ, .) = Ψ(., ζ). Using Lemma 11.2.9, there exists a constant C 0
such that for any ζ ∈ /1, . ∈ 1
−
,
(11.3.15) ReΦ(ζ, .) ≥ C(ρ(ζ) −ρ(.) +[ζ −.[
2
).
Let l be some small tubular neighborhood of /1. Again the proof of Lemma 11.2.9
shows that (11.3.15) holds for ζ ∈ 1
+
∩ l if l is suﬃciently small. Reversing the
role of ζ and ., we have for any . ∈ 1
+
∩ l and ζ ∈ 1,
(11.3.16)
ReΨ(ζ, .) = ReΦ(., ζ) = Re
n
i=1
−
∂ρ
∂.
i
(ζ
i
−.
i
)
≥ C(ρ(.) −ρ(ζ) +[ζ −.[
2
).
Inequality (11.3.16) holds for . ∈ 1
+
∩ l since ρ is strictly convex in a neigh
borhood of /1. Thus Ω
+
and Ω
0+
have singularities only at ζ = . ∈ /1. Using
estimate (11.3.15), we have already proved that Ω
−0
is absolutely integrable in
Lemma 11.2.10. Since Ψ satisﬁes a similar estimate (11.3.16), the proof for Ω
+0
follows from the arguments of Lemma 11.2.10. This proves (11.3.14) and the lemma.
For ζ ,= ., we set
Γ = Ω
−0
−Ω
+0
=
n−2
q=0
Γ
q
(ζ, .),
where Γ
q
= Ω
−0
q
−Ω
+0
q
is the summand which is of degree (0, ¡) in .. Using Lemma
11.3.3, Γ is an integrable kernel on /1. If ) ∈ C
(0,q)
(/1), the form
(11.3.17) H
q
) =
_
bD
Γ
q−1
(, .) ∧ ) =
_
bD
(Ω
−0
q−1
−Ω
+0
q−1
) ∧ )
is a well deﬁned (0, ¡−1)form on /1 with continuous coeﬃcients. The next theorem
shows that Γ(ζ, .) is a fundamental solution for
¯
∂
b
on strictly convex boundaries.
11.3 Homotopy Formulas for
¯
∂
b
on Strictly Convex Boundaries 281
Theorem 11.3.4 (First homotopy formula for
¯
∂
b
on strictly convex bound
aries). Let 1 be a strictly convex domain in C
n
with C
2
boundary and let ρ be a
C
2
deﬁning function for 1. Then for any ) ∈ C
1
(0,q)
(/1), 0 < ¡ < n −1, we have
(11.3.18) )(.) =
¯
∂
b
H
q
) +τH
q+1
¯
∂
b
), . ∈ /1,
where H
q
is deﬁned in (11.3.17).
Proof. For any ) ∈ C
1
(0,q)
(/1), using Theorem 11.3.1, the BochnerMartinelli
Koppelman transform 1
−
and 1
+
deﬁned by (11.3.4) are continuous up to the
boundary. We denote the boundary value of 1
−
and 1
+
by (
_
bD
1(ζ, .) ∧ )(ζ))
−
and (
_
bD
1(ζ, .) ∧ )(ζ))
+
respectively. From (11.3.6), we have for any . ∈ /1,
(11.3.19)
)(.) = τ(1
−
(.) −1
+
(.))
= τ
__
bD
1(ζ, .) ∧ )(ζ)
_
−
−τ
__
bD
1(ζ, .) ∧ )(ζ)
_
+
.
Applying (11.1.4ii), we have for any ζ ∈ /1,
¯
∂
ζ,z
Ω
−0
= −Ω
0
+ Ω
−
= −1(ζ, .) + Ω
−
, . ∈ 1
−
,
¯
∂
ζ,z
Ω
+0
= −Ω
0
+ Ω
+
= −1(ζ, .) + Ω
+
, . ∈ 1
+
.
Thus, for . ∈ 1
−
,
(11.3.20)
_
bD
1(, .) ∧ )
= −
_
bD
¯
∂
ζ,z
Ω
−0
(, .) ∧ ) +
_
bD
Ω
−
(, .) ∧ )
=
¯
∂
z
_
bD
Ω
−0
(, .) ∧ ) +
_
bD
Ω
−0
(, .) ∧
¯
∂
b
) +
_
bD
Ω
−
(, .) ∧ ).
Similarly, for . ∈ 1
+
,
(11.3.21)
_
bD
1(, .) ∧ )
= −
_
bD
¯
∂
ζ,z
Ω
+0
(, .) ∧ ) +
_
bD
Ω
+
(, .) ∧ )
=
¯
∂
z
_
bD
Ω
+0
(, .) ∧ ) +
_
bD
Ω
+0
(, .) ∧
¯
∂
b
) +
_
bD
Ω
+
(, .) ∧ ).
Since G
−
is independent of ., Ω
−
(ζ, .) = Ω
−
0
(ζ, .). It follows that
(11.3.22)
_
bD
Ω
−
(, .) ∧ ) = 0, when ¡ ,= 0.
Also since G
+
is independent of ζ, we have Ω
+
(ζ, .) = Ω
+
n−1
(ζ, .) and
(11.3.23)
_
bD
Ω
+
(, .) ∧ ) = 0, when 0 ≤ ¡ < n −1.
From Lemma 11.3.3, Ω
−0
and Ω
+0
are absolutely integrable kernels. Substituting
(11.3.20)(11.3.23) into (11.3.19) and letting . →/1, we have proved (11.3.18). This
proves the theorem.
282 Integral Representations for ∂ and ∂
b
Corollary 11.3.5 (A solution operator for
¯
∂
b
on strictly convex bound
aries). Let 1 be a bounded strictly convex domain in C
n
with C
2
boundary /1.
For ) ∈ C
(0,q)
(/1), 1 ≤ ¡ ≤ n −2, such that
¯
∂
b
) = 0 on /1, deﬁne
(11.3.24) n(.) = H
q
) =
_
bD
(Ω
−0
q−1
−Ω
+0
q−1
) ∧ ), . ∈ /1.
Then n ∈ C
(0,q−1)
(/1) and n satisﬁes
¯
∂
b
n = ).
Proof. Using Lemma 11.3.3, we have
 n 
L
∞
(bD)
≤ C  ) 
L
∞
(bD)
.
Thus n ∈ C
(0,q−1)
(/1). From Theorem 11.3.4, it follows that
¯
∂
b
n = ) in the
distribution sense.
Remark. Under the same assumption as in Theorem 11.3.4, we also have the
following formula when ¡ = 0 () is a function) and ¡ = n −1 (the top degree case):
When ¡ = n −1, for any ) ∈ C
1
(0,n−1)
(/1),
)(.) = −τ
_
bD
Ω
+
n−1
(, .) ∧ ) +
¯
∂
b
_
bD
Γ
n−2
(, .) ∧ ).
The kernel Ω
+
= Ω
+
n−1
is a holomorphic function in ζ. If ) is a (0, n − 1)form
satisfying the compatibility condition (9.2.12 a), then
_
bD
Ω
+
n−1
(, .) ∧ ) = 0, . ∈ 1 and . →/1.
Thus, we have
)(.) =
¯
∂
b
_
bD
Γ
n−2
(, .) ∧ ), . ∈ /1.
This gives us an explicit solution formula for the
¯
∂
b
operator on strictly convex
boundaries for ¡ = n −1.
On the other hand, for any ) ∈ C
1
(/1),
)(.) =
_
bD
Ω
−
0
(, .) ∧ ) +
_
bD
Γ
0
(, .) ∧
¯
∂
b
).
If ) is a C1 function, we have
)(.) =
_
bD
Ω
−
0
(, .) ∧ ), . ∈ /1.
Thus Ω
−
0
is another reproducing kernel for holomorphic functions in O(1) ∩C
1
(1).
We have already proved in Corollary 2.2.2 that the BochnerMartinelli kernel is
a reproducing kernel. However, Ω
−
can be viewed as a true generalization of the
Cauchy kernel to C
n
since Ω
−
is holomorphic in ..
We shall derive another homotopy formula for
¯
∂
b
on strictly convex boundaries.
Let Ω
−+
be deﬁned by
(11.3.25) Ω
−+
= Ω(G
−
, G
+
),
where G
−
and G
+
are deﬁned by (11.3.10) and (11.3.11). We ﬁrst show that Ω
−+
is integrable.
11.3 Homotopy Formulas for
¯
∂
b
on Strictly Convex Boundaries 283
Lemma 11.3.6. Let 1 be a bounded strictly convex domain in C
n
with C
2
boundary
and let ρ be a strictly convex deﬁning function for 1. The kernel Ω
−+
q−1
(ζ, .),
1 ≤ ¡ ≤ n − 1 has singularities only at ζ = . for ζ, . ∈ /1. Furthermore, there
exists a constant C such that for any . ∈ /1,
(11.3.26)
_
bD
[Ω
−+
q−1
(ζ, .)[ < C,
where C is independent of ..
Proof. Since
Ω
−+
q−1
=
_
1
2πi
_
n
< G
−
, dζ
< G
−
, ζ −.
∧
< G
+
, dζ
< G
+
, ζ −.
∧
_
<
¯
∂
ζ,z
G
−
, dζ
< G
−
, ζ −.
_
n−q−1
∧
_
<
¯
∂
ζ,z
G
+
, dζ
< G
+
, ζ −.
_
q−1
,
it follows from (11.3.15) and (11.3.16) that Ω
−+
q−1
has singularities only at ζ = ..
Thus we only need to estimate the kernel when ζ is close to .. For a ﬁxed ., let l
=
¦ζ[ [ζ −.[ < c¦ be a suﬃciently small neighborhood of ., Φ(ζ, .) = < G
−
, ζ −.
and Ψ(ζ, .) = < G
+
, ζ −. the same as before. Using the same change of variables
t = (t
1
, , t
2n−1
) = (t
t
, t
2n−1
) as in Lemma 11.2.10 with t
2n−1
= ImΦ(ζ, .) and
t
i
(.) = 0 for i = 1, , 2n−1, there exists a constant γ
0
0 such that for ζ, . ∈ /1,
(11.3.27)
_[Φ(ζ, .)[ ≥ γ
0
([t
t
[
2
+[t
2n−1
[),
[Ψ(ζ, .)[ ≥ γ
0
([t
t
[
2
+[t
2n−1
[
2
),
γ
0
[t[ ≤ [ζ −.[ ≤ (1,γ
0
)[t[.
We note that
(11.3.28)
[ < G
−
, dζ ∧ < G
+
, dζ [ = [ < G
−
, dζ ∧ < G
+
−G
−
, dζ [
= O([ζ −.[).
Let do denote the surface element of /1. Repeating the arguments of (11.2.14),
using (11.3.27) and (11.3.28), there exists an ¹ 0 such that
_
bD∩U
[Ω
−+
q−1
(ζ, .)[ ≤ C
__
bD∩U
[ζ −.[
[Φ[
n−q
[Ψ[
q
do
_
≤ C
_
ζ∈bD∩¦]ζ−z]<]
1
[Φ[
n−q
[ζ −.[
2q−1
do
≤ C
_
]t]≤A
dt
1
dt
2
dt
2n−1
([t
2n−1
[ +[t
t
[
2
)
n−q
[t[
2q−1
≤ C.
Thus, the kernel Ω
−+
is absolutely integrable and (11.3.26) is proved.
284 Integral Representations for ∂ and ∂
b
Theorem 11.3.7 (Second homotopy formula for
¯
∂
b
on strictly convex
boundaries). Let 1 be a strictly convex domain in C
n
with C
2
boundary and let ρ
be a C
2
deﬁning function for 1. Then for any ) ∈ C
1
(0,q)
(/1), where 0 < ¡ < n−1,
we have
(11.3.29) )(.) =
¯
∂
b
1
q
) +1
q+1
¯
∂
b
), . ∈ /1,
where
1
q
) = τ
_
bD
Ω
−+
q−1
(, .) ∧ )
and Ω
−+
is deﬁned by (11.3.25).
Proof. Using Lemma 11.3.6, the kernel Ω
−+
is absolutely integrable. From Theorem
11.3.4, we have when 0 < ¡ < n −1,
)(.) =
¯
∂
b
_
ζ∈bD
Γ(ζ, .) ∧ ) +τ
_
ζ∈bD
Γ(ζ, .) ∧
¯
∂
b
).
Using (11.1.4iii), we have for ζ, . ∈ /1 and ζ ,= .,
(11.3.30)
¯
∂
ζ,z
Ω
−+0
=
¯
∂
ζ,z
Ω(G
−
, G
+
, G
0
)
= Ω
−0
−Ω
+0
−Ω
−+
= Γ −Ω
−+
.
For each ﬁxed . ∈ /1, we claim that Ω
−+0
and
¯
∂
ζ,z
Ω
−+0
are absolutely integrable
kernels and
(11.3.31)
¸
¸
¸
¸
_
bD
Ω(G
−
, G
+
, G
0
)
¸
¸
¸
¸
< C,
(11.3.32)
¸
¸
¸
¸
_
bD
¯
∂
ζ,z
Ω(G
−
, G
+
, G
0
)
¸
¸
¸
¸
< C,
where C is independent of .. Let Φ
0
= [ζ − .[
2
as before. Using (11.3.27) and
(11.3.28), (11.3.31) can be estimated by
¸
¸
¸
¸
_
bD∩U
Ω(G
−
, G
+
, G
0
)
¸
¸
¸
¸
≤ C
k1+k2+k3=n−3
_
bD∩U
[ζ −.[
2
[Φ
k1+1
Ψ
k2+1
Φ
k3+1
0
[
do
≤ C
n−2
k=2
_
]t]≤A
[t[
2
dt
1
dt
2
dt
2n−1
([t
2n−1
[ +[t
t
[
2
)
n−k
[t[
2k
< ∞.
Since
¯
∂
ζ,z
Φ =
¯
∂
ζ
Φ = O([ζ −.[) and
¯
∂
ζ,z
Ψ =
¯
∂
z
Ψ = O([ζ −.[), we can use (11.3.27),
11.3 Homotopy Formulas for
¯
∂
b
on Strictly Convex Boundaries 285
(11.3.28) and diﬀerentiation term by term to get
¸
¸
¸
¸
_
bD∩U
¯
∂
ζ,z
Ω(G
−
, G
+
, G
0
)
¸
¸
¸
¸
≤ C
k1+k2+k3=n−3
_
bD∩U
[ζ −.[
[Φ
k1+1
Ψ
k2+1
Φ
k3+1
0
[
do
+C
k1+k2+k3=n−2
_
bD∩U
[ζ −.[
3
[Φ
k1+1
Ψ
k2+1
Φ
k3+1
0
[
do
≤ C
n−2
k=1
_
]t]≤A
[t[dt
1
dt
2
dt
2n−1
([t
2n−1
[ +[t
t
[
2
)
n−k
[t[
2k
+C
n−1
k=1
_
]t]≤A
[t[
3
dt
1
dt
2
dt
2n−1
([t
2n−1
[ +[t
t
[
2
)
n−k
[t[
2k+2
< ∞,
where do is the surface element of /1. This proves (11.3.32). From (11.3.31) and
(11.3.32), we can interchange the order of integration and diﬀerentiation to obtain
(11.3.33)
_
ζ∈bD
¯
∂
ζ,z
Ω
−+0
∧ ) =
_
ζ∈bD
¯
∂
ζ
Ω
−+0
∧ ) −
¯
∂
z
_
ζ∈bD
Ω
−+0
∧ )
=
_
ζ∈bD
Ω
−+0
∧
¯
∂
b
) −
¯
∂
z
_
ζ∈bD
Ω
−+0
∧ ),
where the last equality follows from Stokes’ theorem. The Stokes’ theorem can be
used here by ﬁrst substituting Φ
= Φ + c, Ψ
= Ψ + c and Φ
0
= Φ
0
+ c for Φ, Ψ
and Φ
0
respectively in the kernel Ω
−+0
and then letting c ¸0. Similarly,
(11.3.34)
_
ζ∈bD
¯
∂
ζ,z
Ω
−+0
∧
¯
∂
b
) =
_
ζ∈bD
¯
∂
z
Ω
−+0
∧
¯
∂
b
)
= −
¯
∂
z
_
ζ∈bD
Ω
−+0
∧
¯
∂
b
).
From (11.3.33) and (11.3.34), we have
¯
∂
b
τ
_
ζ∈bD
¯
∂
ζ,z
Ω
−+0
∧ ) +τ
_
ζ∈bD
¯
∂
ζ,z
Ω
−+0
∧
¯
∂
b
) = 0, . ∈ /1.
Thus using (11.3.30), we obtain
)(.) =
¯
∂
b
_
ζ∈bD
Γ(ζ, .) ∧ ) +τ
_
ζ∈bD
Γ(ζ, .) ∧
¯
∂
b
)
=
_
¯
∂
b
_
ζ∈bD
¯
∂
ζ,z
Ω
−+0
∧ ) +τ
_
ζ∈bD
¯
∂
ζ,z
Ω
−+0
∧
¯
∂
b
)
_
+
_
¯
∂
b
_
ζ∈bD
Ω
−+
∧ ) +τ
_
ζ∈bD
Ω
−+
∧
¯
∂
b
)
_
=
¯
∂
b
_
ζ∈bD
Ω
−+
∧ ) +τ
_
ζ∈bD
Ω
−+
∧
¯
∂
b
)
for every . ∈ /1. This proves Theorem 11.3.7.
From Lemma 11.3.6 and Theorem 11.3.7, we have derived another solution for
mula for
¯
∂
b
.
286 Integral Representations for ∂ and ∂
b
Corollary 11.3.8 (Second solution operator for
¯
∂
b
on strictly convex boun
daries). Let 1 be a strictly convex domain in C
n
with C
2
boundary and let ρ be
a C
2
deﬁning function for 1. Let ) ∈ C
(0,q)
(/1), where 0 < ¡ < n − 1 such that
¯
∂
b
) = 0 on /1. Setting n = 1
q
), then u is in C
(0,q−1)
(/1) and
¯
∂
b
n = ) on /1.
Proof. That n is in C
(0,q−1)
(/1) follows from Lemma 11.3.6. Using Theorem 11.3.7,
we have
¯
∂
b
n = ) on /1 in the distribution sense.
Next we shall estimate 1
q
) in the H¨older and 1
p
spaces. We use  
L
p to denote
the 1
p
(0,q)
(/1) norms for (0, ¡)forms.
Theorem 11.3.9 (H¨older and 1
p
estimates for
¯
∂
b
on strictly convex bound
aries). Let 1 be a strictly convex domain in C
n
with C
3
boundary and ρ be a C
3
deﬁning function for 1. For any ) ∈ 1
p
(0,q)
(/1), 1 ≤ j ≤ ∞ and 1 ≤ ¡ < n − 1,
1
q
) satisﬁes the following estimates:
(1) 1
q
)
L
2n
2n−1
−
≤ C)
L
1, for any small c 0.
(2) 1
q
)
L
p
≤ C)
L
p, where
1
p
=
1
p
−
1
2n
for 1 < j < 2n.
(3) 1
q
)
L
p
≤ C)
L
p, where j = 2n and j < j
t
< ∞.
(4) 1
q
)
C
α ≤ C)
L
p, where 2n < j < ∞ and α =
1
2
−
n
p
, C
α
is the H¨older
space of exponent α on /1.
(5) 1
q
)
C
1
2
≤ C)
L
∞.
Proof. We shall prove that the kernel Ω
−+
(ζ, .) is of weak type
2n
2n−1
on /1 uni
formly in ζ and in .. (For deﬁnition of weak type, see Deﬁnition B.5 in the Ap
pendix). Since Ω
−+
only has singularities when ζ = ., following the change of
coordinates ζ →t and (11.3.27), it suﬃces to show that the function
(11.3.35) 1(t) =
1
([t
2n−1
[ +[t
t
[
2
)
n−q
[t
t
[
2q−1
is of weak type
2n
2n−1
, where t = (t
1
, , t
2n−2
, t
2n−1
) = (t
t
, t
2n−1
) and [t[ < 1. Let
¹
λ
be the subset
¹
λ
= ¦t ∈ 1
2n−1
, [t[ < 1 [ 1(t) λ¦, λ 0,
and let : be the Lebesgue measure in 1
2n−1
. We shall show that there exists a
constant ˜ c 0 such that
(11.3.36) :(¹
λ
) ≤
_
˜ c
λ
_ 2n
2n−1
, for all λ 0.
By a change of variables t →
˜
t such that t
i
= λ
−
1
2n−1 ˜
t
i
, i = 1, , 2n − 2 and
t
2n−1
= λ
−
2
2n−1 ˜
t
2n−1
, we have for some c 0,
:(¹
λ
) = cλ
−
2n
2n−1
:(¹
1
) ≤
_
˜ c
λ
_ 2n
2n−1
,
since :(¹
1
) < ∞. This proves (11.3.36). It follows from Theorem B.11 in the
Appendix that the estimates (1), (2) and (3) hold.
11.3 Homotopy Formulas for
¯
∂
b
on Strictly Convex Boundaries 287
To prove (4) and (5), we deﬁne
˜
Ω
−+
by
˜
Ω
−+
=
_
1
2πi
_
n
< G
−
, dζ
Φ(ζ, .)
∧
< G
+
, dζ
Ψ(ζ, .) −jρ(.)
∧
k1+k2=n−2
_
<
¯
∂
ζ,z
G
−
, dζ
Φ(ζ, .)
_
k1
∧
_
<
¯
∂
ζ,z
G
+
, dζ
Ψ(ζ, .) −jρ(.)
_
k2
,
where j 0 is suﬃciently large. We ﬁrst note that the kernel Ω
−+
is the same as
˜
Ω
−+
when ζ, . ∈ /1. It follows from (11.3.15) and (11.3.16) that there exists a
C 0 such that for any ζ ∈ /1 and . ∈ 1,
(11.3.37) ReΦ(ζ, .) ≥ C([ρ(.)[ +[ζ −.[
2
),
and
(11.3.38) Re
˜
Ψ(ζ, .) ≡ ReΨ(ζ, .) −jρ(.) ≥ C([ρ(.)[ +[ζ −.[
2
),
if j is chosen suﬃciently large. Let 1
δ
= ¦. ∈ 1 [ ρ(.) < −δ¦ for some δ 0.
From the HardyLittlewood lemma (see Theorem C.1 in the Appendix), to prove
(4) and (5), it suﬃces to show that for some small δ
0
0 and all 0 < δ < δ
0
,
(11.3.39) sup
z∈bD
δ
¸
¸
¸
¸
grad
z
_
bD
˜
Ω
−+
(, .) ∧ )
¸
¸
¸
¸
≤ Cδ
−
1
2
−
n
p
)
L
p,
where 2n ≤ j ≤ ∞. After the same change of variables in a small neighborhood
[ζ − .[ < c that of (11.2.12), applying (11.3.37) and (11.3.38), (11.3.39) is proved
for j = ∞ if the following holds:
(11.3.40)
_
]t]≤A
dt
1
dt
2
dt
2n−1
(δ +[t
2n−1
[ +[t
t
[
2
)
n−q+1
[t
t
[
2q−1
< Cδ
−
1
2
,
(11.3.41)
_
]t]≤A
dt
1
dt
2
dt
2n−1
(δ +[t
2n−1
[ +[t
t
[
2
)
n−q
[t
t
[
2q+1
< Cδ
−
1
2
,
where C is independent of δ. Inequality (11.3.40) is proved in (11.2.17) and (11.3.41)
is proved similarly since 1 ≤ ¡ < n −1.
To prove (11.3.39) when j = 2n, it suﬃces to show that
(11.3.42)
_
]t]≤A
[)[dt
1
dt
2
dt
2n−1
(δ +[t
2n−1
[ +[t
t
[
2
)
n−q+1
[t
t
[
2q−1
< Cδ
−1
)
L
2n,
(11.3.43)
_
]t]≤A
[)[dt
1
dt
2
dt
2n−1
(δ +[t
2n−1
[ +[t
t
[
2
)
n−q
[t
t
[
2q+1
< Cδ
−1
)
L
2n,
288 Integral Representations for ∂ and ∂
b
where C is independent of δ. To prove (11.3.42), we set n
∗
=
2n
2n−1
and use H¨older’s
inequality to obtain
_
]t]≤A
[)[dt
1
dt
2
dt
2n−1
(δ +[t
2n−1
[ +[t
t
[
2
)
n−q+1
[t
t
[
2q−1
≤ )
L
2n
_
_
]t]≤A
dt
1
dt
2
dt
2n−1
(δ +[t
2n−1
[ +[t
t
[
2
)
n
∗
(n−q+1)
[t
t
[
n
∗
(2q−1)
_ 1
n
∗
≤ )
L
2n
_
_
]t
]≤A
dt
1
dt
2
dt
2n−2
(δ +[t
t
[
2
)
n
∗
(n−q+1)−1
[t
t
[
n
∗
(2q−1)
_ 1
n
∗
≤ C)
L
2n
_
_
A
0
:
2n−3
d:
(δ +:
2
)
n
∗
(n−q+1)−1
:
n
∗
(2q−1)
_ 1
n
∗
≤ C)
L
2n
_
_
A
0
d:
(δ +:
2
)
2n+1
2n−1
:
2n−3
2n−1
_ 1
n
∗
, (· = :,
√
δ)
≤ Cδ
−1
)
L
2n
_
_
∞
0
d·
(1 +·
2
)
2n+1
2n−1
·
2n−3
2n−1
_ 1
n
∗
≤ Cδ
−1
)
L
2n.
This proves (11.3.42) and (11.3.43) can be proved similarly. Inequality (11.3.39) is
proved for j = 2n and j = ∞. The other cases are proved by interpolation (see
Theorem B.6 in the Appendix). This completes the proof of Theorem 11.3.9.
Remark. In Chapter 8, we have proved that when /1 is strictly pseudoconvex
or more generally, /1 satisﬁes condition Y(¡), the canonical solution gains 1,2
derivatives in the Sobolev spaces (see Theorem 8.4.14). Theorem 11.3.9 gives a
solution operator which gains 1,2derivatives in the H¨older space on strictly con
vex boundaries. This result again can be generalized to any strictly pseudoconvex
boundary by a partition of unity since the boundary can be convexiﬁed locally. We
note that the solution for
¯
∂
b
deﬁned by (11.3.24) has the same properties as the
solution given by 1
q
) by a similar proof. It is interesting to note that when /1
is the boundary of the Siegel upper half space, 1
q
) obtained by the integral ker
nel method agrees with the solution obtained in Theorem 10.1.5 using a completely
diﬀerent method. The reader should compare Theorem 11.3.9 with Theorem 10.1.5.
11.4 Solvability for
¯
∂
b
on C1 Manifolds with Boundaries
Let 1 be a strictly convex domain in C
n
with C
2
boundary /1 and ω ⊂⊂ /1
be a connected open C1 manifold with smooth boundary /ω. We consider the
¯
∂
b
equation
(11.4.1) ∂
b
n = α on ω,
11.4 Solvability for
¯
∂
b
on CR Manifolds with Boundaries 289
where α is a (0, ¡)form on ω, 1 ≤ ¡ ≤ n −2. In order for (11.4.1) to be solvable, it
is necessary that α satisﬁes
(11.4.2) ∂
b
α = 0 on ω.
Note that when ¡ = n − 1, (11.4.2) is void and (11.4.1) is related to the local
nonsolvable phenomenon of Lewy’s equation. Due to the fact that the compatibility
condition (11.4.2) is satisﬁed only on ω instead of the whole boundary /1, this
question cannot be answered from the global solvability results obtained in the
previous section. The solvability of (11.4.1) depends on the special geometry of the
boundary /ω.
In Chapter 9, we have seen that when ¡ = n−1 with an additional compatibility
condition (9.2.12 a), one still can solve
¯
∂
b
globally on /1. In fact, we have proved
that
¯
∂
b
has closed range in 1
2
(0,q)
(/1) on any pseudoconvex boundary /1 for any
1 ≤ ¡ ≤ n −1. When we discuss the local solvability of (11.4.1), we must avoid the
top degree case (when ¡ = n −1) due to the Lewy example.
In this section we study the solvability of (11.4.1) on ω for any α satisfying
(11.4.2) on ω. When ¡ = n−2, there is another compatibility condition for (11.4.1)
to be solvable without shrinking. This compatibility condition can be derived as
follows:
Let 1 be a compact set in C
n
and O(1) be the set of functions which are deﬁned
and holomorphic in some open neighborhood of 1. Let α be a form in C
(0,n−2)
(ω)
such that there exists n ∈ C
(0,n−3)
(ω) with
¯
∂
b
n = α on ω. Then for any p ∈ O(/ω),
we have
_
bω
α ∧ p ∧ d.
1
∧ ∧ d.
n
=
_
bω
¯
∂
b
n ∧ p ∧ d.
1
∧ ∧ d.
n
=
_
bω
¯
∂(n ∧ p ∧ d.
1
∧ ∧ d.
n
)
=
_
bω
d(n ∧ p ∧ d.
1
∧ ∧ d.
n
) = 0.
Thus, another necessary condition for (11.4.1) to be solvable for some n ∈ C
(0,n−3)
(ω) is that
(11.4.2 a)
_
bω
α ∧ p ∧ d.
1
∧ ∧ d.
n
= 0 for all p ∈ O(/ω).
It is easy to see that (11.4.2 a) is also necessary for the existence of a solution
n ∈ C
(0,n−3)
(ω) by approximation. This additional condition makes it necessary to
diﬀerentiate between 1 ≤ ¡ < n − 2 and ¡ = n − 2 when considering (11.4.1). At
the end of this section, we will discuss when condition (11.4.2) implies (11.4.2 a)
and give an example to show that (11.4.2 a) is indeed an additional compatibility
condition.
We ﬁrst describe the geometry of the boundary /ω of ω on which one can con
struct a solution kernel for
¯
∂
b
on ω. Let ρ be a strictly convex deﬁning function
for 1. The set ω is a domain in /1 deﬁned by some C
2
function : deﬁned in a
neighborhood of /1 such that
(11.4.3) ω = ¦. ∈ C
n
[ ρ(.) = 0, :(.) < 0¦.
290 Integral Representations for ∂ and ∂
b
We require that : be a C
2
smooth function depending on only one complex variable.
Without loss of generality, we may assume that : depends on .
n
only. This implies
that the hypersurface `
0
= ¦. ∈ C
n
[ :(.
n
) = 0¦ is a Leviﬂat hypersurface. The
boundary /ω is deﬁned by
/ω = /1 ∩ `
0
= ¦. ∈ C
n
[ ρ(.) = 0, :(.
n
) = 0¦.
On /ω, we assume
(11.4.4) dρ ∧ d: ,= 0 on /ω.
Thus, the hypersurfaces /1 and `
0
intersect transversally over 1. If
(11.4.5) ∂ρ ∧ ∂: ,= 0,
we say that /1 and `
0
intersect transversally over C. The points in
Σ = ¦. ∈ /ω [ ∂ρ ∧ ∂: = 0¦
are called characteristic points. Any point in /ω ¸ Σ is called a noncharacteristic
point or a generic point.
If j is a characteristic point on /ω, the space T
1,0
p
(/1) ∩ T
1,0
p
(`
0
) has complex
dimension n − 1. Near the noncharacteristic point j ∈ /ω, the set T
1,0
p
(/1) ∩
T
1,0
p
(`
0
) has complex dimension n−2. This jump in the dimension of the tangential
(1,0) vector ﬁelds at the characteristic points makes it diﬃcult to study (11.4.1) by
imitating the 1
2
techniques used in Chapter 4. We shall study the solvability of
(11.4.1) by integral kernels.
The following example shows that in general, an open C1 manifold with smooth
boundary has characteristic points.
Example. If 1 = ¦. ∈ C
n
[ [.[ < 1¦ is the unit ball and :(.
n
) = Im .
n
, then the
boundary /ω of the set
ω = ¦. ∈ C
n
[ [.[ = 1, Im.
n
< 0¦
has two characteristic points at Σ = ¦(0, , +1), (0, , −1)¦, since ∂ρ ∧ ∂: = 0
if and only if .
1
= = .
n−1
= 0.
If :(.
n
) = [.
n
[
2
and /ω is the boundary of
ω
1
= ¦. ∈ C
n
[ [.[ = 1, [.
n
[
2
<
1
2
¦,
then /ω
1
has no characteristic points.
Notice that ω is simply connected but ω
1
is not.
To use the integral kernels to solve (11.4.1), our starting point is the homotopy
formula derived in Theorem 11.3.7. From (11.3.29), we have that Ω
−+
is a funda
mental solution for
¯
∂
b
on the compact hypersurface /1. Thus, it gives a solution
kernel for (11.4.1) if α has compact support in ω. To solve
¯
∂
b
for forms which do
11.4 Solvability for
¯
∂
b
on CR Manifolds with Boundaries 291
not vanish on /ω, we introduce new kernels constructed from the special deﬁning
function : for ω. Set
(11.4.6) G
(ζ, .) = G
(ζ) =
_
0, , 0,
∂:(ζ
n
)
∂ζ
n
_
and
(11.4.7) η(ζ, .) =
∂:(ζ
n
)
∂ζ
n
(ζ
n
−.
n
) = < G
, ζ −. .
Let
(11.4.8) ω
=
1
2πi
< G
, dζ
< G
, ζ −.
=
1
2πi
dζ
n
ζ
n
−.
n
,
where the notation . is used to indicate that the hypersurface `
0
deﬁned by : is
Levi ﬂat. Note that ω
is independent of : and
¯
∂
ζ,z
ω
= 0, ζ
n
,= .
n
.
In other words, it is holomorphic both in the ζ and . variables away from singular
ities. Setting
(11.4.9) Ω
−+
= Ω(G
, G
−
, G
+
),
we see that Ω
−+
is an (n, n −3)form. We write
(11.4.10) Ω
−+
(ζ, .) =
n−3
q=0
Ω
−+
q
(ζ, .),
where
Ω
−+
q
=
1
(2πi)
n
dζ
n
ζ
n
−.
n
∧
< G
−
(ζ), dζ
Φ(ζ, .)
∧
< G
+
(.), dζ
Ψ(ζ, .)
∧
_
< ∂
ζ
G
−
(ζ), dζ
Φ(ζ, .)
_n−3−q
∧
_
< ∂
z
G
+
(.), dζ
Ψ(ζ, .)
_q
away from the singularities. Thus Ω
−+
q
has exactly ¡ d¯ .’s.
If ζ ∈ /ω and . ∈ ω, we have
:(ζ
n
) −:(.
n
) 0.
It follows that ζ
n
,= .
n
when ζ ∈ /ω and . ∈ ω. The kernel Ω
−+
is well deﬁned
and smooth when ζ ∈ /ω and . ∈ ω.
For α ∈ C
(0,q)
(ω), 1 ≤ ¡ ≤ n −2, we deﬁne
(11.4.11) o
q
α =
_
ζ∈ω
Ω
−+
q−1
(ζ, .) ∧ α(ζ) +
_
ζ∈bω
Ω
−+
q−1
(ζ, .) ∧ α(ζ),
where Ω
−+
(ζ, .) and Ω
−+
(ζ, .) are kernels deﬁned by (11.3.25) and (11.4.10) re
spectively. The following two theorems are the main results of this section.
292 Integral Representations for ∂ and ∂
b
Theorem 11.4.1 (A homotopy formula for
¯
∂
b
on C1 manifolds with bound
aries). Let 1 be a strictly convex domain in C
n
with C
2
boundary and let ρ be a
C
2
strictly convex deﬁning function for 1. Let ω ⊂⊂ /1 be an open connected
C1 manifold with smooth boundary deﬁned by (11.4.3) where :(.) = :(.
n
) is a C
2
function. We assume that dρ∧d: ,= 0 on /ω. For any α ∈ C
1
(0,q)
(ω), 1 ≤ ¡ < n−2,
(11.4.12) α =
¯
∂
b
o
q
α +o
q+1
¯
∂
b
α, . ∈ ω,
where o
q
is the integral operator deﬁned by (11.4.11).
Theorem 11.4.2 (A solution operator for
¯
∂
b
on C1 manifolds with bound
aries). Let ω be as in Theorem 11.4.1. For any α ∈ C
(0,q)
(ω), 1 ≤ ¡ < n −2, with
∂
b
α = 0 on ω, the form n = o
q
α is in C
(0,q−1)
(ω) and
¯
∂
b
n = α on ω, where o
q
is
the integral operator deﬁned by (11.4.11).
When ¡ = n −2, we assume furthermore that α satisﬁes the additional compati
bility condition
_
bω
α ∧ p ∧ d.
1
∧ ∧ d.
n
= 0 for all p ∈ O(/ω),
the same conclusion holds.
To prove Theorems 11.4.1 and 11.4.2, we start with the following proposition:
Proposition 11.4.3. Let /1 and ω be as in Theorem 11.4.1. For every ) ∈
C
1
(0,q)
(ω), 1 ≤ ¡ < n −1,
)(.) =
¯
∂
b
_
ω
Ω
−+
q−1
∧ )(ζ) +τ
_
ω
Ω
−+
q
∧
¯
∂
b
)(ζ) −τ
_
bω
Ω
−+
q
∧ )(ζ)
for every . ∈ ω, where Ω
−+
is deﬁned by (11.3.25).
Proof. We ﬁrst extend ) to
˜
) on an open set ˜ ω ⊃⊃ ω such that
˜
) ∈ C
1
(0,q)
(˜ ω). Let
χ
∈ C
∞
0
(˜ ω) be cutoﬀ functions such that χ
≡ 1 on ω for every c and χ
converges
to the characteristic function of ω as c →0. Applying the homotopy formula proved
in Theorem 11.3.7 to χ
˜
), we have for . ∈ ω,
)(.) = χ
˜
) =
¯
∂
b
_
bD
Ω
−+
q−1
∧ χ
˜
) +τ
_
bD
Ω
−+
q
∧
¯
∂
b
(χ
˜
))
=
¯
∂
b
_
bD
Ω
−+
q−1
∧ χ
˜
) +τ
_
bD
Ω
−+
q
∧ χ
¯
∂
b
˜
)
+τ
_
˜ ω\ω
Ω
−+
q
∧ (
¯
∂
b
χ
) ∧
˜
).
For . ∈ ω, we note that Ω
−+
is smooth for ζ ∈ (˜ ω¸ω) and we can apply Stokes’
theorem to the third term on the righthand side to obtain
lim
→0
_
˜ ω\ω
Ω
−+
q
∧
¯
∂
b
χ
∧
˜
) = lim
→0
_
˜ ω\ω
Ω
−+
q
∧ dχ
∧
˜
)
= lim
→0
_
˜ ω\ω
d(χ
Ω
−+
q
∧
˜
)) − lim
→0
_
˜ ω\ω
χ
d(Ω
−+
q
∧
˜
))
= −
_
bω
Ω
−+
q
∧ ).
11.4 Solvability for
¯
∂
b
on CR Manifolds with Boundaries 293
Thus for any . ∈ ω, letting c → 0, since Ω
−+
is an absolutely integrable kernel by
Lemma 11.3.6 , we have
)(.) = lim
→0
_
¯
∂
b
_
bD
Ω
−+
q−1
∧ χ
˜
) +τ
_
bD
Ω
−+
q
∧ χ
¯
∂
b
˜
)
_
+ lim
→0
τ
_
˜ ω\ω
Ω
−+
q
∧ (
¯
∂
b
χ
) ∧
˜
)
=
¯
∂
b
_
ω
Ω
−+
q−1
∧ ) +τ
_
ω
Ω
−+
q
∧
¯
∂
b
) −τ
_
bω
Ω
−+
q
∧ ).
This proves the proposition.
Proof of Theorem 11.4.1. We deﬁne
Ω
−
= Ω(G
, G
−
) and Ω
+
= Ω(G
, G
+
),
where G
−
and G
+
are deﬁned by (11.3.10) and (11.3.11) respectively. The kernels
Ω
−
(ζ, .) and Ω
+
(ζ, .) are smooth for (ζ, .) ∈ /ω ω. Using (11.1.4iii), we have
(11.4.13) ∂
ζ,z
Ω
−+
= −Ω
−+
+ Ω
+
−Ω
−
for any ζ ∈ /ω and . ∈ ω. Applying Proposition 11.4.3 and (11.4.13), we obtain for
. ∈ ω,
(11.4.14)
α(.) = ∂
b
_
ω
Ω
−+
q−1
(ζ, .) ∧ α(ζ) +τ
_
ω
Ω
−+
q
(ζ, .) ∧
¯
∂
b
α(ζ)
+τ
_
bω
∂
ζ
Ω
−+
q
(ζ, .) ∧ α(ζ) + ∂
z
_
bω
Ω
−+
q−1
(ζ, .) ∧ α(ζ)
−τ
_
bω
Ω
+
q
(ζ, .) ∧ α(ζ) +τ
_
bω
Ω
−
q
(ζ, .) ∧ α(ζ).
We claim that for any α ∈ C
1
(0,q)
(ω), the following three equalities hold for . ∈ ω:
(i)
_
bω
∂
ζ
Ω
−+
q
(ζ, .) ∧ α(ζ) =
_
bω
Ω
−+
q
(ζ, .) ∧
¯
∂
b
α(ζ),
(ii)
_
bω
Ω
−
q
(ζ, .) ∧ α(ζ) = 0, 1 ≤ ¡ ≤ n −2,
(iii)
_
bω
Ω
+
q
(ζ, .) ∧ α(ζ) = 0, 1 ≤ ¡ < n −2.
Since the kernel Ω
−+
(ζ, .) has the factor dζ
1
∧ ∧dζ
n
, applying Stokes’ theorem,
we have
_
bω
∂
ζ
Ω
−+
(ζ, .) ∧ α(ζ) =
_
bω
d
ζ
(Ω
−+
(ζ, .) ∧ α(ζ))
+
_
bω
Ω
−+
(ζ, .) ∧
¯
∂
b
α(ζ),
294 Integral Representations for ∂ and ∂
b
which proves (i). Since n
is holomorphic in both the ζ and . variables, for any
ζ ∈ /ω and . ∈ ω, we have
(11.4.15) Ω
−
= ω
∧ ω
−
∧ (∂
ζ
ω
−
)
n−2
,
(11.4.16) Ω
+
= ω
∧ ω
+
∧ (∂
z
ω
+
)
n−2
.
(ii) follows from (11.4.15) and the fact that integration of an (n, n −2 +¡)form
on /ω is zero. Similarly when 1 ≤ ¡ < n − 2, (iii) follows from type consideration
since each component in (11.4.16) has (n − 2) d.’s and no dζ’s. Substituting (i),
(ii) and (iii) into (11.4.14), we have proved Theorem 11.4.1.
Proof of Theorem 11.4.2. When ¡ < n−2, Theorem 11.4.1 implies Theorem 11.4.2
if α ∈ C
1
(0,q)
(ω). If α is only in C
(0,q)
(ω), we approximate α by a sequence α
n
∈
C
∞
(0,q)
(ω) such that α
n
→α and
¯
∂
b
α
n
→0 uniformly on ω. This is possible from the
proof of Friedrichs’ lemma (see Appendix D). It is easy to see that o
q+1
¯
∂
b
α
n
→ 0
in the distribution sense in ω and o
q
α
n
→o
q
α uniformly on compact subset of ω.
Thus n = o
q
α is in C
(0,q−1)
(ω) and
¯
∂
b
n = α in the distribution sense in ω.
To show that the theorem holds when ¡ = n −2, we use (11.4.14) to obtain
(11.4.17) α =
¯
∂
b
o
n−2
α −
_
bω
Ω
+
n−2
(ζ, .) ∧ α(ζ), . ∈ ω.
To show that the last integral in (11.4.17) vanishes when ¡ = n−2, notice that the
kernel Ω
+
(ζ, .) is holomorphic in ζ in a neighborhood of /ω for each ﬁxed . ∈ ω.
Thus from our assumption on α, we have for . ∈ ω,
_
bω
Ω
+
n−2
(ζ, .) ∧ α(ζ) = 0.
This proves the theorem.
In general, the additional assumption (11.4.2 a) on α when ¡ = n−2 is necessary.
The next proposition characterizes all domains ω such that condition (11.4.2) will
imply condition (11.4.2 a). At the end of this section we shall give an example of a
¯
∂
b
closed form which does not satisfy condition (11.4.2 a).
Proposition 11.4.4. Suppose that O(¯ ω) is dense in O(/ω) (in the C(/ω) norm).
Then any (0, n − 2)form α ∈ C
(0,n−2)
(ω) satisfying condition (11.4.2) also satis
ﬁes condition (11.4.2 a). In particular, if polynomials are dense in O(/ω), then
condition (11.4.2) implies condition (11.4.2 a).
Proof. From the assumption, for any p ∈ O(/ω), there exists a sequence of holo
morphic functions p
n
∈ O(¯ ω) such that p
n
converges to p in C(/ω). We have, for
any
¯
∂
b
closed α,
_
bω
α ∧ p ∧ d.
1
∧ ∧ d.
n
= lim
n→∞
_
bω
α ∧ p
n
∧ d.
1
∧ ∧ d.
n
= lim
n→∞
_
ω
¯
∂(α ∧ p
n
∧ d.
1
∧ ∧ d.
n
)
= lim
n→∞
_
ω
¯
∂
b
α ∧ p
n
∧ d.
1
∧ ∧ d.
n
= 0.
11.4 Solvability for
¯
∂
b
on CR Manifolds with Boundaries 295
Thus condition (11.4.2) implies condition (11.4.2 a). The proposition is proved.
Using Proposition 11.4.4, we have the following:
Corollary 11.4.5. Let ω be as in Theorem 11.4.1. We assume that the set C¸ o,
where
o = ¦.
n
∈ C [ . = (.
1
, , .
n
) ∈ /ω¦,
is connected. For any α ∈ C
(0,q)
(ω), 1 ≤ ¡ ≤ n − 2, with ∂
b
α = 0 on ω, the form
n = o
q
α is in C
(0,q−1)
(ω) and
¯
∂
b
n = α on ω, where o
q
is the integral operator
deﬁned by (11.4.11).
Proof. Using Theorem 11.4.2, we only need to prove the assertion for ¡ = n − 2.
From (11.4.17), it suﬃces to show that
_
bω
Ω
+
n−2
(ζ, .) ∧ α(ζ) = 0, . ∈ ω.
An approximation argument can be applied using the additional assumption on the
set o.
Since the set C ¸ o is connected by assumption, by the Runge approximation
theorem, the function
/(ζ
n
) =
1
ζ
n
−.
n
can be approximated by polynomials 1
ν
(ζ
n
, .
n
) for each ﬁxed .
n
in the sup norm
on o. We approximate Ψ by Ψ
(ζ, .) = Ψ(ζ, .) +c for some c and let c →0
+
. Then
Ψ
(ζ, .) is smooth when . ∈ ω and ζ ∈ ω. Also Ψ
is holomorphic in ζ ∈ ω. Deﬁne
ω
+
=
1
2πi
< G
+
(ζ), dζ
< G
+
(ζ), ζ −. +c
=
1
2πi
< G
+
(ζ), dζ
Ψ(ζ, .) +c
.
We can apply Stokes’ theorem ﬁrst to the modiﬁed kernel with Ψ substituted by
Ψ
, letting c →0, to obtain for . ∈ ω,
_
bω
Ω
+
n−2
(ζ, .) ∧ α(ζ)
= lim
ν→∞
lim
→0
+
_
bω
1
ν
(ζ
n
, .
n
)dζ
n
∧ ω
+
∧ (∂
z
ω
+
)
n−2
∧ α(ζ)
= lim
ν→∞
lim
→0
+
_
ω
∂
ζ
_
1
ν
(ζ
n
, .
n
)dζ
n
∧ ω
+
∧ (∂
z
ω
+
)
n−2
∧ α(ζ)
_
= 0
since every term in the integrand is ∂
ζ
closed. This proves the corollary.
Example. We note that the additional assumption on α or ω when ¡ = n − 2
cannot be removed. Let /1 = ¦. [ [.
1
[
2
+ [.
2
[
2
+ [.
3
[
2
= 1¦ be the unit sphere in
C
3
. Let
ω = /1 ∩ ¦. ∈ C
3
[ [.
3
[
2
< 1,2¦.
296 Integral Representations for ∂ and ∂
b
Then o = ¦.
3
∈ C [ [.
3
[
2
= 1,2¦ does not satisﬁes the hypothesis imposed on o in
Corollary 11.4.5 since C¸ o is not connected. We shall show that equation
¯
∂
b
n = α
is not solvable for ¡ = 1 in ω. Let
α =
¯ .
1
d¯ .
2
− ¯ .
2
d¯ .
1
([.
1
[
2
+[.
2
[
2
)
2
.
Then α is a constant multiple of the BochnerMartinelliKoppelman kernel in C
2
and
¯
∂α = 0 for [.
1
[
2
+[.
2
[
2
,= 0. Thus α ∈ C
∞
(0,1)
(ω) and
¯
∂
b
α = 0 on ω.
If α =
¯
∂
b
n for some n ∈ C(ω), then α must satisfy
_
bω
α ∧
1
i.
3
d.
1
∧ d.
2
∧ d.
3
=
_
bω
α ∧ d.
1
∧ d.
2
∧ dθ
3
= 0,
where dθ
3
= d.
3
,(i.
3
). On the other hand, we have that
_
bω
α ∧ d.
1
∧ d.
2
∧ dθ
3
= 8π
_
¦]z1]
2
+]z2]
2
=
1
2
]
(¯ .
1
d¯ .
2
− ¯ .
2
d¯ .
1
) ∧ d.
1
∧ d.
2
= 16π
_
¦]z1]
2
+]z2]
2
<
1
2
]
d¯ .
1
∧ d¯ .
2
∧ d.
1
∧ d.
2
,= 0.
Thus there does not exist any solution n ∈ C(ω). There does not exist any n ∈ C(ω)
satisfying
¯
∂
b
n = α either, by an approximation argument. Thus the assumption on
α in Theorem 11.4.2 cannot be removed. We note that O(ω) is not dense in O(/ω)
here.
On the other hand, if
ω = /1 ∩ ¦. ∈ C
n
[ Im .
3
< 0¦,
then o = ¦.
3
∈ C [ −1 < Re.
3
< 1, Im .
3
= 0¦ and C ¸ o is connected. Thus it
satisﬁes the hypothesis imposed in Corollary 11.4.5 and we can solve (11.4.1) for all
¯
∂
b
closed form α ∈ C
(0,q)
(ω) when 1 ≤ ¡ ≤ n −2.
11.5 1
p
Estimates for Local Solutions of
¯
∂
b
Let 1 be a strongly pseudoconvex domain in C
n
with smooth boundary `.
we shall study the local solvability of the tangential CauchyRiemann equations ∂
b
near a point .
0
in `. After a quadratic change of coordinates we may assume that
.
0
= 0 and there exists a strictly plurisubharmonic deﬁning function ρ(.) for `
which has the following form near the origin:
(11.5.1) ρ(.) = −Im .
n
+
n
j,k=1
¹
jk
.
j
.
k
+O([.[
3
),
11.5 L
p
Estimates for Local Solutions of
¯
∂
b
297
where (¹
jk
) is a positive deﬁnite hermitian matrix (see the proof of Corollary 3.4.5).
The function ρ is strictly convex near the origin. Let l be a small neighborhood of
0 and δ
0
0 be suﬃciently small. We deﬁne ¦ω
δ
¦ by
(11.5.2) ω
δ
= ¦. ∈ ` ∩ l [ Im .
n
< δ¦, 0 < δ < δ
0
.
We can always choose l and δ
0
0 suﬃciently small such that each ω
δ
is an open
neighborhood in a connected strictly convex hypersurface whose boundary lies in a
ﬂat surface. It is easy to see that ∩
δ
ω
δ
= ¦0¦. Thus the ¦ω
δ
¦ forms a neighborhood
base at 0.
Using Theorem 11.4.2 and Corollary 11.4.5, there is a solution operator o
q
α
satisfying
¯
∂
b
o
q
α = α on ω
δ
for any
¯
∂
b
closed α ∈ C
(0,q)
(ω
δ
). Our main goal is to
prove that there exists a solution operator satisfying 1
p
estimates on ω
δ
.
Theorem 11.5.1 (1
p
existence and estimates for local solutions of
¯
∂
b
). Let
` be a strongly pseudoconvex hypersurface in C
n
and .
0
∈ `. There exists a
neighborhood base ¦ω
δ
¦ of .
0
such that for any α ∈ 1
p
(0,q)
(ω
δ
), 1 ≤ ¡ ≤ n − 2 and
1 < j < ∞, satisfying ∂
b
α = 0, there exists n ∈ 1
p
(0,q−1)
(ω
δ
) satisfying ∂
b
n = α.
Furthermore, there exists a positive constant c such that the following estimate holds:
(11.5.3) n
L
p
(0,q−1)
(ω
δ
)
≤ cα
L
p
(0,q)
(ω
δ
)
,
where c depends on j, ω
δ
but is independent of α.
Corollary 11.5.2. Let ` and ω
δ
be as in Theorem 11.5.1. The range of ∂
b
is
closed in the 1
p
(0,q)
(ω
δ
) space, where 1 < j < ∞ and 1 ≤ ¡ ≤ n −2.
Corollary 11.5.3. Let ` and ω
δ
be as in Theorem 11.5.1. For each 1 ≤ ¡ < n−2,
there exists a solution operator
˜
o
q
given by integral kernels such that for any ∂
b

closed α ∈ 1
p
(0,q)
(ω
δ
), 1 < j < ∞, we have
¯
∂
b
˜
o
q
α = α and

˜
o
q
α
L
p
(0,q−1)
(ω
δ
)
≤ cα
L
p
(0,q)
(ω
δ
)
,
where c depends on j, ω
δ
but is independent of α.
The rest of this section is to prove Theorem 11.5.1. Let ω
δ
be deﬁned by (11.5.2).
To prove Theorem 11.5.1, we ﬁrst prove the 1
p
estimates for the solution constructed
for ∂
b
closed forms with C(ω
δ
) coeﬃcients in Theorem 11.4.2.
Proposition 11.5.4. Let ` be a strongly pseudoconvex C1 manifold deﬁned by
(11.5.1) and ω
δ
be deﬁned by (11.5.2). For any α ∈ C
(0,q)
(ω
δ
) such that ∂
b
α = 0,
1 ≤ ¡ ≤ n − 2, there exists a solution n ∈ C
(0,q−1)
(ω
δ
) satisfying ∂
b
n = α on ω
δ
.
Furthermore, for every 1 < j < ∞, there exists a constant C
p
0 such that
(11.5.4) n
L
p
(0,q−1)
(ω
δ
)
≤ C
p
α
L
p
(0,q)
(ω
δ
)
,
where C
p
is independent of α and small perturbation of δ.
Proof. Let
(11.5.5) n(.) ≡ o
q
α(.) = 1
1
(α) +1
2
(α),
298 Integral Representations for ∂ and ∂
b
where
1
1
(α) =
_
ω
δ
Ω
−+
q−1
(ζ, .) ∧ α(ζ)
and
1
2
(α) =
_
bω
δ
Ω
−+
q−1
(ζ, .) ∧ α(ζ).
Since the set C¸o is connected. it follows from Theorem 11.4.2 and Corollary 11.4.5
that for every 1 ≤ ¡ ≤ n −2,
¯
∂
b
n = α on ω
δ
and n ∈ C
(0,q−1)
(ω
δ
).
To prove Proposition 11.5.4, we only need to prove that n satisﬁes (11.5.4). Using
Theorem 11.3.9, there exists a C 0 such that the integral 1
1
(α) satisﬁes
(11.5.6)  1
1
(α) 
L
p
(ω
δ
)
≤ C  α 
L
p
(ω
δ
)
.
We only need to estimate 1
2
(α).
Since 1
2
(α) is an integral on /ω
δ
, we rewrite 1
2
(α) to be an integral on ω
δ
to
facilitate the 1
p
estimates. Since the kernel Ω
−+
(ζ, .) has singularities at ζ
n
= .
n
for any ζ, . ∈ ω
δ
, we shall modify the kernel ﬁrst so that Stokes’ theorem can be
applied.
Let :(.) = :(.
n
) = Im .
n
. Then for any ζ, . ∈ ω
δ
,
(11.5.7) :(.) −:(ζ) −2Re
∂:(ζ)
∂ζ
n
(.
n
−ζ
n
) = 0.
We set
˜ η(ζ, .) =
∂:(ζ)
∂ζ
n
(ζ
n
−.
n
) −(:(ζ) −δ)
= η(ζ, .) −(:(ζ) −δ).
It follows from (11.5.7) that
(11.5.8)
Re ˜ η(ζ, .) =
1
2
_
−:(ζ) −:(.)
_
+δ
=
1
2
_
−
_
:(ζ) −δ
_
−
_
:(.) −δ
_
_
0
for all ζ, . ∈ ω
δ
. Thus Re ˜ η(ζ, .) vanishes only when ζ and . are both in /ω
δ
. Also
we have
˜ η(ζ, .) = η(ζ, .), when ζ ∈ /ω
δ
and . ∈ ω
δ
.
We deﬁne the kernel
˜
Ω
−+
(ζ, .) by modifying Ω
−+
with ˜ η substitute for η. Set
(11.5.9)
˜
Ω
−+
(ζ, .) =
n−3
q=0
˜
Ω
−+
q
(ζ, .),
where
˜
Ω
−+
q
(ζ, .) =
1
(2πi)
n
∂r
∂ζn
dζ
n
˜ η(ζ, .)
∧
< G
−
(ζ), dζ
Φ(ζ, .)
∧
< G
+
(.), dζ
Ψ(ζ, .)
∧
_
< ∂
ζ
G
−
(ζ), dζ
Φ(ζ, .)
_n−3−q
∧
_
< ∂
z
G
+
(.), dζ
Ψ(ζ, .)
_q
11.5 L
p
Estimates for Local Solutions of
¯
∂
b
299
away from the singularities. The kernel
˜
Ω
−+
q
has exactly ¡ d¯ .’s. Since
(11.5.10)
˜
Ω
−+
(ζ, .) = Ω
−+
(ζ, .), when ζ ∈ /ω
δ
and . ∈ ω
δ
,
we shall substitute
˜
Ω
−+
in 1
2
(α) for Ω
−+
. The advantage is that
˜
Ω
−+
is integrable
for each ﬁxed . ∈ ω
δ
since ˜ η satisﬁes (11.5.8). Thus for any . ∈ ω
δ
, by Stokes’
theorem and a limiting argument (substituting Φ
= Φ + c and Ψ
= Ψ + c for
Φ and Ψ, approximating α by smooth forms α
such that α
→ α and ∂
b
α
→ 0
uniformly on ω
δ
, then letting c ¸0), we can write
(11.5.11)
1
2
(α)(.) =
_
ζ∈bω
δ
˜
Ω
−+
q−1
∧ α(ζ) =
_
ζ∈ω
δ
∂
ζ
_
˜
Ω
−+
q−1
∧ α(ζ)
_
=
_
ζ∈ω
δ
∂
ζ
˜
Ω
−+
q−1
∧ α(ζ).
From (11.3.28), we have
[ < G
−
(ζ), dζ ∧ < G
+
(.), dζ [ = O([ζ −.[).
Thus for every 1 ≤ ¡ < n −1,
(11.5.12)
˜
Ω
−+
q−1
(ζ, .) =
O([ζ −.[)
Φ(ζ, .)
n−q−1
Ψ(ζ, .)
q
˜ η(ζ
n
, .
n
)
.
We write
∂
ζ
˜
Ω
−+
q−1
(ζ, .) =
1
(2πi)
n
∂
ζ
_
∂r
∂ζn
dζ
n
˜ η(ζ, .)
∧
< G
−
, dζ
Φ(ζ, .)
∧
< G
+
−G
−
, dζ
Ψ(ζ, .)
∧
_
< ∂
ζ
G
−
, dζ
Φ(ζ, .)
_n−2−q
∧
_
< ∂
z
G
+
, dζ
Ψ(ζ, .)
_q−1_
.
It follows from the deﬁnition of Φ and Ψ that
_
∂
ζ
Φ(ζ, .) = O([ζ −.[),
∂
ζ
Ψ(ζ, .) = 0.
Using < G
−
, dζ = ∂
ζ
ρ,
∂r
∂ζn
dζ
n
= ∂
ζ
: and estimate (11.3.15), after grouping terms
of the same form together, we have
[∂
ζ
˜
Ω
−+
q−1
(ζ, .)[ ≤ C
_
[ζ −.[
[˜ η(ζ, .)[[Φ(ζ, .)[
n−1−q
[Ψ(ζ, .)[
q
+
[∂
ζ
ρ ∧ ∂
ζ
: ∧ \
2n−3
(ζ)[
[˜ η(ζ, .)[[Φ(ζ, .)[
n−1−q
[Ψ(ζ, .)[
q
+
[ζ −.[[∂
ζ
ρ ∧ ∂
ζ
: ∧ ∂
ζ
˜ η ∧ \
2n−4
(ζ)[
[˜ η(ζ, .)[
2
[Φ(ζ, .)[
n−1−q
[Ψ(ζ, .)[
q
_
,
300 Integral Representations for ∂ and ∂
b
where
ranges over all possible monomials \
2n−3
(ζ) and \
2n−4
(ζ) of degree 2n−3
and 2n −4 respectively in dζ
1
, dζ
1
, , dζ
n
, dζ
n
. Let
1
1
(ζ, .) =
[ζ −.[
[˜ η(ζ, .)[[Φ(ζ, .)[
n−1−q
[Ψ(ζ, .)[
q
, (11.5.13)
1
2
(ζ, .) =
[∂
ζ
ρ ∧ ∂
ζ
: ∧ \
2n−3
(ζ)[
[˜ η(ζ, .)[[Φ(ζ, .)[
n−1−q
[Ψ(ζ, .)[
q
, (11.5.14)
1
3
(ζ, .) =
[ζ −.[[∂
ζ
ρ ∧ ∂
ζ
: ∧ ∂
ζ
˜ η ∧ \
2n−4
(ζ)[
[˜ η(ζ, .)[
2
[Φ(ζ, .)[
n−1−q
[Ψ(ζ, .)[
q
. (11.5.15)
We deﬁne
J
i
(α)(.) =
_
ζ∈ω
δ
[α(ζ)[1
i
(ζ, .)d:
2n−1
(ζ), i = 1, 2, 3,
where :
2n−1
(ζ) is the surface measure of ω
δ
. For . ∈ ω
δ
, using [˜ η(ζ, .)[ 0,
_
ζ∈ω
δ
[1
i
(ζ, .)[d:
2n−1
(ζ) ≤ C
z
, i = 1, 2, 3,
where C
z
depends on .. Thus the operator J
i
is bounded from 1
p
(ω
δ
) to 1
p
(ω
δ
, loc),
i = 1, 2, 3. Near the boundary /ω
δ
, the singularities of 1
i
are not absolutely
integrable, but are of Hilbert integral type (see Theorem B.9 in the Appendix).
However, we shall show that there exists a constant c 0 such that
(11.5.16) J
i
(α)
L
p
(ω
δ
)
≤ cα
L
p
(ω
δ
)
i = 1, 2, 3.
Let :
δ
(ζ
n
) = :(ζ
n
) − δ be the deﬁning function for ω
δ
. To prove (11.5.16), we use
the following lemma:
Lemma 11.5.5. If for every 0 < c < 1, there exist a constant c
such that 1
i
(ζ, .)
satisﬁes
(11.5.17)
_
ζ∈ω
δ
[:
δ
(ζ
n
)[
−
1
i
(ζ, .) ≤ c
[:
δ
(.
n
)[
−
for all . ∈ ω
δ
,
(11.5.18)
_
z∈ω
δ
[:
δ
(.
n
)[
−
1
i
(ζ, .) ≤ c
[:
δ
(ζ
n
)[
−
for all ζ ∈ ω
δ
,
then for 1 < j < ∞, there exists c
p
0 such that
J
i
(α)
L
p
(ω
δ
)
≤ c
p
α
L
p
(ω
δ
)
for all α ∈ 1
p
(ω
δ
).
11.5 L
p
Estimates for Local Solutions of
¯
∂
b
301
Proof. By H¨older’s inequality and (11.5.17), we have
[J
i
(α)(.)[
p
≤
_
ζ∈ω
δ
1
i
(ζ, .)[α(ζ)[
p
[:
δ
(ζ
n
)[
p/p
d:
2n−1
(ζ)
__
ζ∈ω
δ
1
i
(ζ, .)[:
δ
(ζ
n
)[
−
d:
2n−1
(ζ)
_
p/p
≤ (c
)
p/p
[:
δ
(.
n
)[
−p/p
_
ζ∈ω
δ
1
i
(ζ, .)[α(ζ)[
p
[:
δ
(ζ
n
)[
p/p
d:
2n−1
(ζ),
where
1
p
+
1
p
= 1. Integrating with respect to . and interchanging the order of
integration we obtain, using (11.5.18),
_
z∈ω
δ
[J
i
(α)(.)[
p
d:
2n−1
≤ (c
)
p/p
_
ζ∈ω
δ
__
z∈ω
δ
[:
δ
(.
n
)[
−p
p
1
i
(ζ, .)d:(.)
_
[α(ζ)[
p
[:
δ
(ζ
n
)[
p
p
d:(ζ)
≤ (c
)
p/p
(c
p/p
)α
p
L
p
(ω
δ
)
.
This proves Lemma 11.5.5 with the constant c
p
= (c
)
1/p
(c
p/p
)
1/p
if one chooses
c so small such that 0 < c < 1 and 0 < cj,j
t
< 1.
Thus, to prove (11.5.16), it suﬃces to prove (11.5.17) and (11.5.18) for i = 1, 2, 3.
Using a partition of unity in both ζ and . variables, we can assume that ζ lies in
a coordinate patch l and . lies in a coordinate patch \. When l ∩ \ = ∅, then
[Φ(ζ, .)[ 0 and [Ψ(ζ, .)[ 0 for ζ ∈ l and . ∈ \ and the estimation will be
simpler. We assume l and \ are the same coordinate patch and omit the other
cases.
Let Σ denote the set of the characteristic points, i.e., points where ∂ρ ∧ ∂: = 0
on /ω
δ
. We ﬁrst assume that l ∩ Σ = ∅. We shall choose special coordinates for
ω
δ
∩ l.
Since dρ ∧ d: ,= 0 on /ω
δ
, we can choose :(ζ
n
) as a coordinate function near
l ∩ ω
δ
. Since d
ζ
Φ(ζ, .)[
ζ=z
= ∂ρ(ζ) and ∂ρ = −∂ρ on ω
δ
, it follows that ∂ρ(ζ) =
1
2
(∂ρ −∂ρ) = id
ζ
ImΦ(ζ, .)[
ζ=z
. Thus,
∂ρ(ζ) = id
ζ
ImΦ(ζ, .) +O([ζ −.[).
Similarly for ζ ∈ /ω
δ
, we have
∂
ζ
: = id
ζ
Im˜ η(ζ
n
, .
n
) +O([ζ
n
−.
n
[).
Thus, if ζ ∈ /ω
δ
¸ Σ,
d:(ζ
n
) ∧ d
ζ
ImΦ(ζ, .) ∧ d
ζ
Im˜ η(ζ
n
, .
n
) ∧ dρ(ζ)[
ζ=z
= −d:(ζ
n
) ∧ ∂
ζ
ρ(ζ) ∧ ∂
ζ
:(ζ
n
) ∧ dρ(ζ)
= −∂
ζ
: ∧ ∂
ζ
ρ ∧ ∂
ζ
: ∧ ∂
ζ
ρ
,= 0.
302 Integral Representations for ∂ and ∂
b
Let ImΦ(ζ, .) = t
1
and Im˜ η(ζ, .) = t
2
. We can choose coordinates (:(ζ
n
), t
1
, ,
t
2n−2
) with t
i
(.) = 0, i = 1, , 2n − 2. From (11.3.15), (11.3.16) and (11.5.8),
there exists c 0 with
[Φ(ζ, .)[ ≥ c([t[
2
+[t
1
[), (11.5.19)
[ReΨ(ζ, .)[ ≥ c[t[
2
, (11.5.20)
[Re˜ η(ζ, .)[ ≥ c([:(ζ) −δ[ +[:(.) −δ[). (11.5.21)
It follows that there exists C 0 independent of . such that
[1
1
(ζ, .)[ ≤
C
([:(ζ) −δ[ +[:(.) −δ[)[t[
2n−3
,
[1
2
(ζ, .)[ ≤
C
([:(ζ) −δ[ +[:(.) −δ[)([t
1
[ +[t[
2
)
n−1−q
[t[
2q
,
[1
3
(ζ, .)[ ≤
C
([:(ζ) −δ[ +[:(.) −δ[ +[t
2
[)
2
([t
1
[ +[t[
2
)
n−1−q
[t[
2q−1
.
Estimate (11.5.17) will be proved for i = 1, 2, 3 when ζ and . are away from the
characteristic points using the following lemma (letting j = [:(ζ) − δ[ and σ =
[:(.) −δ[).
Lemma 11.5.6. Let t = (t
1
, , t
2n−2
) and dt = dt
1
dt
2
dt
2n−2
. For any 0 <
c < 1, ¹ 0, there exists c
i
0, i = 0, 1, 2, 3, such that the following inequalities
hold: For any σ 0, 1 ≤ ¡ ≤ n −2,
(1)
_
∞
0
j
−
σ +j
dj ≤ c
0
σ
−
,
(2)
_
∞
0
_
]t]≤A
j
−
(σ +j)[t[
2n−3
dtdj ≤ c
1
σ
−
,
(3)
_
∞
0
_
]t]≤A
j
−
(σ +j)([t
1
[ +[t[
2
)
n−1−q
[t[
2q
dtdj ≤ c
2
σ
−
,
(4)
_
∞
0
_
]t]≤A
j
−
dtdj
(σ +j +[t
2
[)
2
([t
1
[ +[t[
2
)
n−1−q
[t[
2q−1
≤ c
3
σ
−
.
Proof. (1) follows from a change of variables to the case when σ = 1. In fact one
can show using contour integration that c
0
= π, sin πc.
Estimate (2) follows from (1) by using polar coordinates for t variables. To prove
(4), we integrate t
1
, t
2
ﬁrst and then use polar coordinates for t
tt
= (t
3
, , t
2n−2
)
and apply (1) to obtain
_
∞
0
_
]t]≤A
j
−
(σ +j +[t
2
[)
2
([t
1
[ +[t[
2
)
n−1−q
[t[
2q−1
dtdj
≤ C
_
∞
0
_
]t
]≤A
[ log [t
tt
[[j
−
(σ +j)([t
tt
[
2
)
n−2−q
[t
tt
[
2q−1
dt
tt
dj
≤ C
_
∞
0
__
0<v≤A
[ log ·[·
2n−5
·
2n−5
d·
_
j
−
(σ +j)
dj
≤ c
3
σ
−
.
11.5 L
p
Estimates for Local Solutions of
¯
∂
b
303
Estimate (3) can be similarly proved.
Thus (11.5.17) is proved when there are no characteristic points. Similarly one
can prove (11.5.18) by reversing the roles of Φ and Ψ, ζ and ..
Near any characteristic point . ∈ Σ, we cannot choose :(ζ
n
), ImΦ (ζ, .) and
Im˜ η(ζ, .) as coordinates since they are linearly dependent at ζ = .. The kernel 1
1
is less singular than 1
2
or 1
3
and can be estimated as before by choosing :(ζ
n
)
and Im Φ = t
1
as coordinates. To estimate (11.5.17) and (11.5.18) when i = 2, 3,
one observes that at characteristic points, the numerator of each 1
i
(ζ, .), i = 2, 3
also vanishes. We shall prove (11.5.17) for 1
3
and the case for 1
2
is similar.
We write
p
1
(ζ, .) = ImΦ(ζ, .),
p
2
(ζ, .) = Im˜ η(ζ, .) = −
1
2
Re(ζ
n
−.
n
).
It is easy to see that d
ζ
Im˜ η(ζ, .) ∧ dj =
−1
i
∂: ∧ ∂:. Since ∂ρ = (∂ρ −
¯
∂ρ),2 =
id
ζ
ImΦ(ζ, .) +O([ζ −.[), we have, setting j(ζ
n
) = δ −:(ζ
n
),
dp
1
(ζ, .) ∧ dp
2
(ζ, .) ∧ dj(ζ)
¸
¸
ζ=z
= −d
ζ
ImΦ(ζ, .) ∧ d
ζ
Im˜ η(ζ, .) ∧ d:(ζ)
¸
¸
ζ=z
= ∂
ζ
ρ ∧ ∂:(ζ) ∧ ∂:(ζ)
¸
¸
ζ=z
.
Thus (11.5.17) will be proved for i = 3 if we can prove that
(11.5.22)
˜
J
3
(.) ≤ Cσ
−
,
where
˜
J
3
(.) is the integral
_
ζ∈U∩ω
δ
j
−
[dp
1
(ζ, .) ∧ dp
2
(ζ, .) ∧ dj ∧ \
2n−4
(ζ)[
([p
2
(ζ, .)[ +j +σ)
2
([p
1
(ζ, .)[ +[ζ −.[
2
)
n−1−q
[ζ −.[
2q−1
.
The other terms are less singular and can be estimated as before.
Let r = (r
1
, , r
2n−2
, j) = (r
t
, j) be real coordinates on l ∩ ω
δ
such that
. = (0, , 0, j(.)) where r
i
= Re(ζ
j
−.
j
) or r
i
= Im(ζ
j
−.
j
) for some , = 1, , n.
In this coordinate system, we have, for some ¹ 0,
˜
J
3
is bounded by the integral