Sunteți pe pagina 1din 365

CHAPTER 1

REAL AND COMPLEX MANIFOLDS


We shall begin by dening holomorphic functions and the Cauchy-Riemann equa-
tions in C
n
. In Sections 1.2-1.4 of this chapter we will review the denitions 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 n-dimensional 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 identied with 1
2n
in a natural manner,
. (r
1
, j
1
, , r
n
, j
n
).
Denition 1.1.1. A complex-valued C
1
function )(.) dened 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 xed. In other words, )(.) satises
(1.1.1)
)
.
j
= 0,
for each , = 1, , n, where
(1.1.2)

.
j
=
1
2
_

r
j
+i

j
j
_
is the so-called Cauchy-Riemann 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 dier, 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 dened on this domain can be extended holo-
morphically to a xed larger set, a feature that does not exist in one variable. In
Chapter 3 we will discuss this phenomenon in detail. Another main dierence 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 dierences will be further investigated in Chapters 4-6 using
solutions of the inhomogeneous Cauchy-Riemann equations (1.1.3).
There is yet another major dierence in solving (1.1.3) in one and several vari-
ables. When n 2, a compatibility condition must be satised 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 denition concerning the dierentiability of the boundary of
a domain.
Denition 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 real-valued C
k
function : dened
in some open neighborhood l of j such that 1l = r l[ :(r) < 0, /1l =
r l[ :(r) = 0 and d:(r) ,= 0 on /1 l. The function : is called a C
k
local
dening function for 1 near j. If l is an open neighborhood of 1, then : is called
a global dening function for 1, or simply a dening function for 1.
The relationship between two dening functions is claried in the next lemma.
Lemma 1.1.3. Let :
1
and :
2
be two local dening functions for 1 of class C
k
(1 / ) in a neighborhood l of j /1. Then there exists a positive C
k1
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
n1
). 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
k1
function / on l. For / 2, we clearly have
(2) and / 0 on l. When / = 1, (2) also follows directly from the denition of
dierentiation at 0. This proves the lemma.
1.2 Real and Complex Manifolds
Let ` be a Hausdor 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
p
: l
p
\
p
. We call the pair
(l
p
,
p
) a coordinate neighborhood of ` near j. Since, for any l
p
,
p
() is
a point in 1
n
, we have the usual Euclidean coordinates (r
1
(
p
()), , r
n
(
p
()))
for
p
(). We shall call the set (r
1
(
p
()), , r
n
(
p
())) the local coordinates for
the points in l
p
with respect to the coordinate neighborhood (l
p
,
p
), and it will
be abbreviated by (r
1
(), , r
n
()), and the n-tuple (r
1
, , r
n
) of functions on
l
p
will be called the local coordinate system on (l
p
,
p
).
Let ` be a topological manifold, then ` is covered by a family of such coordinate
neighborhoods (l

, where is an index set. If for some , in we


have l

= l

,= , then there is a well-dened homeomorphism


)

(l

(l

).
These will be called the transition functions with respect to the coordinate neigh-
borhood system (l

. Obviously, we have )

= )
1

. Now we give the


denition of a dierentiable manifold.
Denition 1.2.1. Let ` be a topological manifold together with a coordinate neigh-
borhood system (l

. We call ` an n-dimensional dierentiable 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

manifold.
Next we dene complex manifolds.
Denition 1.2.2. Let ` be a topological manifold together with a coordinate neigh-
borhood system (l

, where

(l

) = \

are open sets in C


n
. ` is called
a complex manifold of complex dimension n if the transition function )

is holomorphic on

(l

) C
n
, whenever l

= l

,= for all , .
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 (`, ) is simply the induced one given by the identity mapping
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
). Dene 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 1

= 1 0 such that r = j. 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 identied with the space
of all lines passing through the origin in 1
n+1
. The mapping from 1
n+1
0
onto 1P
n
so that (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
1
(l) is open in 1
n+1
0. Since also maps the
compact set o
n
onto 1P
n
, we see that 1P
n
is compact.
The coordinate neighborhood system (l
j
,
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 (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
j
from l
j
onto 1
n
is given by

j
([r
1
: : r
j
: : r
n+1
])
= (r
1
,r
j
, , r
j1
,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) =
i

1
j
(j)
=
i
([j
1
: : j
j1
: 1 : j
j
: : j
n
])
=
_
j
1
j
i
, ,
j
i1
j
i
,
j
i+1
j
i
, ,
j
j1
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 denition 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 denition 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 complex-valued function dened
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

for some local coordinate neighborhood l

, such that )

1
is holomorphic on the open subset

(\ ) in C
n
. Clearly, the above denition of
holomorphic functions at a point j l is independent of the choice of the local
coordinate system (l

). 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

of a complex manifold are holomorphic.


Let ` and be two complex manifolds of complex dimensions : and n with
local coordinate systems (l

and (\

respectively, and let ) be


1.3 Tangent Spaces and the Hermitian Metric 5
a continuous mapping from ` into . We shall say that ) denes a holomorphic
mapping at j `, if there exists an open neighborhood l
p
of j, contained in
a local coordinate neighborhood l

, with )(l
p
) contained in a local coordinate
neighborhood \

such that

)
1

denes a holomorphic mapping from

(l
p
)
into

(\

). The denition 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 one-to-one, 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 identied 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
.
Dene an 1-linear 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 complexied tangent
space CT
p
(C
n
) = T
p
(C
n
)
R
C. First we extend J to the whole complexied tangent
space by J(r) = (Jr). It follows that J is a C-linear 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 veried 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
) =
1,0
p
(C
n
)
0,1
p
(C
n
),
where
1,0
p
(C
n
) and
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
1,0
p
(C
n
)
and the space
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
.
Dene as before an 1-linear 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 denition 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
(`) =
1,0
p
(`)
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 dened 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 complex-valued function on l, and (/
ij
(j))
n
i,j=1
is a positive denite Hermitian
matrix dened 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
complexied 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.
Denition 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 , called the projection
map, from 1 onto ` such that the following conditions are satised:
(1) For each j `, 1
p
=
1
(j) is a vector space over 1 of dimension /. 1
p
is called the bre space over j.
(2) For each j `, there exists an open neighborhood l containing j and a
dieomorphism
/ :
1
(l) l 1
k
,
such that /(
1
()) = 1
k
and the restriction /
q
:
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 : 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

, /

) and (l

, /

) have nonempty intersection, i.e.,


l

,=, then a map p

is induced on l

:
p

: l

G1(/, 1),
such that
p

(j) = (/

)
p
(/

)
1
p
: 1
k
1
k
.
The matrices p

s are called transition matrices. Clearly, they are smooth and


satisfy the following conditions:
(1) p

= p
1

,
(2) p

= 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 : = 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 (l, 1).
Notice that the concept of vector bundle can obviously be dened for other cate-
gories. For instance, if 1 and ` are complex manifolds and the bers are complex
vector spaces, then one can dene 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(`) =
pM
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 from
1
(l) onto l 1
n
by
:
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 dened by the Jacobian matrix, with respect to these two local
8 Real and Complex Manifolds
coordinate systems, which by denition is smooth. It is also clear that any global
section : in (`, T(`)) is a smooth vector eld A dened 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 bre of the cotangent bundle, T

(`), at each point j ` is


the 1-linear 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 1-form over l. We also have the complexied cotangent bundle,
denoted by
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
r
(`) whose bers at each point .
0
` are the wedge product of degree
: of
1
p
(`), and
(`) =
2n

r=0

r
(`).
Any smooth section : of
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
1,0
(`)
and copies of
0,1
(`), where j n and n, we obtain the vector bundle of
bidegree (j, ), denoted by
p,q
(`), and we have

r
(`) =
p+q=r

p,q
(`).
Smooth sections of
p,q
(`), denoted by C

(p,q)
(`), are called (j, )-forms on `.
1.5 Exterior Derivatives and the Cauchy-Riemann 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
complex-valued function dened 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 Cauchy-Riemann Complex 9
and
d.
j
= dr
j
+idj
j
, d.
j
= dr
j
idj
j
,
for 1 , n. Dene 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 dierential 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 veried that the denitions of and are invariant under holomorphic
change of coordinates. Hence, the operators and are well dened for functions
on a complex manifold. A C
1
complex-valued function on a complex manifold is
holomorphic if and only if
) = 0.
Next we extend the denition of the operators and to dierential 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 dened by
d) =

I,J
d)
I,J
d.
I
d.
J
= ) +),
where ) and ) are dened 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 dened for (j, )-forms on `, and we have
d = +.
Since
0 = d
2
) =
2
) + ( +)) +
2
)
10 Real and Complex Manifolds
and all terms are of dierent types, we obtain
(1.5.2)
2
= 0, + = 0,
2
= 0.
Since
(1.5.3)
2
= 0,
It follows that the sequence
0
p,0
(`)

p,1
(`)


p,n1
(`)

p,n
(`) 0,
for 0 j n, is a complex. This is called the Cauchy-Riemann complex. Denote

p,q
= :
p,q
(`)
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
nk
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
cR
nk
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 elds on l such that they are
tangent to some
c
, c 1
nk
, everywhere. Since the restriction of the vector eld
A
i
, 1 i / to each
c
denes a vector eld 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 eld
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 sucient 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 elds
dened 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 eld A
1
(r) = (o
1
(r), , o
n
(r)). From the
basic existence theorem for a system of rst order ordinary dierential equation,
through every point j in a small open neighborhood of the origin, there exists
exactly one integral curve (t) = (
1
(t), ,
n
(t)), where t (, ) for some real
number 0, such that
d
i
dt
(t) = o
i
((t)), i = 1, , n.
It follows that for any smooth function ) in a small neighborhood \ of 0,
A
1
)((t)) =
k

i=1
o
i
((t))
)((t))
r
i
=
)((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 elds 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
n1
. Hence, by the induction hypotheses, there exist
new local coordinates near the origin in 1
n1
, 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

=2
o
1i
(r)A

(r
l
)
=
k

=2
o
1i
(r)c
l
(r).
12 Real and Complex Manifolds
Hence, the uniqueness part of the Cauchy problem for a system of rst order ordi-
nary dierential 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 elds A
1
, , A
k
dened in some open neighborhood of the origin in
1
n
such that conditions (1) and (2) of Theorem 1.6.1 are satised. Let us consider
a system of overdetermined partial dierential 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 sucient 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 dened 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 satised. 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 dierential 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
dened by the vector elds A
1
, , A
k
in some open neighborhood of the origin.
Proof. Notice rst 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
nk
, namely,
1.7 Inequivalence between the Ball and the Polydisc 13
H is dened by r
t
= /(r
tt
) where /(r
tt
) is a smooth function on \ . Now, for each
xed r
tt
\ ,
k

j=1
)
j
(r
t
, r
tt
)dr
j
is a dierential 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 well-dened, i.e., independent of the paths in r
t
-space from/(r
tt
) to r
t
. Obviously,
(1.6.5) denes 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 elds dened 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
nk
for c = (c
1
, , c
nk
) C
nk
, 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
nk
for c = (c
1
, , c
nk
) C
nk
,
and E = T
1,0
(
c
). This is the so-called complex Frobenius theorem which can be
deduced from the Newlander-Nirenberg 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 dierent 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 (Poincare). 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 Montels 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 xed 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 classication problem in several variables
is considerably more complicated than in one variable. An approach towards the
classication 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 [Hor 5]. The inequivalence between the polydisc and the unit ball was rst
discovered by H. Poincare 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
Cauchy-Riemann operator on the complex plane. We will rst show that the so-
lution to the Cauchy-Riemann 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 Bochner-Martinelli 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 Bochner-Martinelli
transform.
In Section 2.3, we will determine when a rst-order partial dierential equation
in two real variables is locally equivalent to the Cauchy-Riemann equation.
2.1 The Cauchy Integral Formula
All functions in this chapter are complex-valued unless otherwise stated. Then
the following formula, known as Cauchys 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
2i
__
bD
n()
.
d +
__
D
u

.
d d
_
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()d,( .) on the punctured domain 1

= 1 1

(.), we obtain
_
bD
n()
.
d i
_
2
0
n(. +cc
i
) d =
__
D
u

.
d d.
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.
Dene
(2.1.2) n(.) =
1
2i
__
D
)()
.
d d.
Then n(.) is in C
k
(1) and satises
(2.1.3)
n
.
= )(.)
on 1. When / = 0, n dened by (2.1.2) is in C(1) and satises (2.1.3) in the
distribution sense.
Proof. For the case / 1, we rst assume ) C
k
0
(C). Setting = ., we have
n(.) =
1
2i
__
C
)(. )

d d.
Dierentiation under the integral sign gives that n C
k
(C). Using Theorem 2.1.1
we obtain
n
.
(.) =
1
2i
__ f

.
d d = )(.).
For the general situation, let .
0
1, and let be a cut-o function, 0 1,
1 in some neighborhood \ of .
0
and supp 1. Thus,
n(.) =
1
2i
__
D
)()
.
d d
=
1
2i
__
D
())()
.
d d +
1
2i
__
D
(1 ()))()
.
d d
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
.
= (.))(.) = )(.),
for . \ .
To prove the case for / = 0, we observe that 1,( .) is an integrable kernel
after changing to polar coordinates. The following estimate holds for n dened by
(2.1.2):
| n |

C | ) |

.
Approximate ) by )
n
C
1
(1) in the sup norm on 1. Dene 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 innity.
This proves the theorem.
2.1 The Cauchy Integral Formula 17
We recall that a function ) dened in some domain 1 contained in 1
n
is said to
be Holder continuous of order , 0 < < 1, denoted by ) C

(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
[

,
where the constant 1 is independent of r
1
and r
2
. When is equal to 1, ) is called
Lipschitz. The space of Lipschitz continuous function on 1 is denoted by
1
(1).
Notice that C
1
(1)
1
(1) C

(1) C(1). A function ) is said to be Holder


continuous of order / + with / N and 0 < < 1, if all the partial derivatives of
) of order / are Holder continuous of order .
For any continuous function ) on the boundary, the Cauchy transform of ), i.e.,
(2.1.4) 1(.) =
1
2i
_
bD
)()
.
d for . C /1,
denes a holomorphic function 1(.) o the boundary. The Cauchy transform 1(.)
and the given data ) on the boundary are related by the so-called 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 dened
on the boundary. Dene 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
2i
_
bD
)() )
e
(.)
.
d,
and, for . C 1,
(2.1.7) 1
+
(.) =
1
2i
_
bD
)() )
e
(.)
.
d.
Since ) is, at least, of class C
1
on the boundary, the integral on the right-hand sides
of (2.1.6) and (2.1.7) denes 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
Hardy-Littlewood 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
dening function for 1. When / = 1, using (2.1.6), we have
(2.1.8)
[d(1

(.) )
e
(.))[
_
bD
[d)
e
(.)[
[. [
d:() +
_
bD
[)() )
e
(.)[
[. [
2
d:()

_
bD
1
[. [
d:().
Here, 1 means there is an universal constant C, independent of and 1,
such that C1. For any given c 0, to show 1

(.) C
1
(1), it suces
to estimate (2.1.8) over a small neighborhood l of (.) on the boundary, where
(.) is the projection of . on the boundary. Let d(.) be the distance from . to the
boundary. If . is suciently close to the boundary, it is easily seen that, for l,
[. [ is equivalent to d(.) +:(), where :() is the distance from to (.) along
the boundary. It follows that
_
U
d:()
[. [
d(.)

_
U
d:()
[. [
1
d(.)

_
1
0
d:()
(d(.) +:())
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

.
satises T
z
(:) = 0. Hence, T
z
is a tangential vector eld with C
k
coecients along
the level sets of :. Then, integration by parts shows, for . 1,

.
1

(.) =
1
2i
_
bD
)()

.
_
1
.
_
d
=
1
2i
_
bD
)()

_
1
.
_
d
=
1
2i
_
bD
)()T

_
1
.
_
d
=
1
2i
_
bD
T

)()
_
1
.
_
d,
where T

is a rst order dierential operator with C


k1
coecients on the boundary.
It follows that
(2.1.10)

k
.
k
1

(.) =
1
2i
_
bD
(T

)
k
)()
_
1
.
_
d.
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 Hardy-Littlewood 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 . satises
[.[ [[ for all /1, we have
1
+
(.) =
1
2i
_
bD
)()
.
d
=
1
2i

m=0
__
bD
)()
m
d
_
.
m1
= 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
)()
m
d
_
.
m1
= 0.
This implies
_
bD
)()
m
d = 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 sucient 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. Dene n(.) by (2.1.2). Then n(.) satises n,. = ) in C and is supported
in 1 if and only if
__
D
)()
m
d d = 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(; :) centered at =
(
1
, ,
n
) with multiradii : = (:
1
, , :
n
) in C
n
, we mean 1(; :) =

n
j=1
1
rj
(
j
)
where 1
rj
(
j
) = . C[ [.
j
[ < :
j
. Let 1
t
(; :
t
) =

n
j=1
1
r

j
(
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 dened as above with = 0, and let ) be a
smooth (j, +1)-form, j 0, 0, dened on 1, which satises 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.
k1
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
k1
for some / with /1 , and we proceed
to prove the statement o
k
. Write
) = d.
k
+,
where is a (j, )-form and is a (j, + 1)-form, and both and involve only
(0, 1)-forms from d.
1
, , d.
k1
. Express
=

t
]I]=p,]J]=q

IJ
d.
I
d.
J
.
It is easy to see by type consideration that
IJ
,.
j
= 0 for , / and all 1, J.
Now choose a cut-o function (.
k
) C

0
(1
r
k
) such that 0 1 and 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
2i
__
C
(
k
)
IJ
(.
1
, , .
k1
,
k
, .
k+1
, , .
n
)

k
.
k
d
k
d
k
is smooth and solves the -equation
1
IJ
.
k
= (.
k
)
IJ
(.) =
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
+
0
,
where
0
is a (j, + 1)-form that involves only (0, 1)-forms from d.
1
, , d.
k1
.
Hence,
) 1 =
0
is a smooth (j, + 1)-form which is -closed and involves only (0, 1)-forms from
d.
1
, , d.
k1
. It follows now from the induction hypotheses that there exists a
smooth (j, )-form n
0
that satises
n
0
=
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(; :) be a
polydisc in C
n
, n 2. Suppose that ) is continuous on 1(; :) and holomorphic in
1(; :). Then for any . 1(; :),
)(.) =
1
(2i)
n
_
]nn]=rn

_
]11]=r1
)(
1
, ,
n
)
(
1
.
1
) (
n
.
n
)
d
1
d
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(; :). Then

_

.
_

)()

`!
:
1
1
:
n
n
,
where = (
1
, ,
n
) is a multiindex with
j
0 N, ! =
1
!
n
! and
_

z
_

=
_

z1
_
1

_

zn
_
n
.
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
)(.) =

(. n)

such that the series converges absolutely in some open neighborhood of n. Here
the summation is over multiindices and (. n)

= (.
1
n
1
)
1
(.
n
n
n
)
n
.
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
dened 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 Bochner-Martinelli Formula
In this section, we shall extend the Cauchy kernel from the complex plane to
higher dimensional space. Dene the Bochner-Martinelli kernel by
(2.2.1) 1(, .) =
(n 1)!
(2i)
n
1
[ .[
2n
n

j=1
(
j
.
j
)d
j
(
k,=j
d
k
d
k
),
for = (
1
, ,
n
), . = (.
1
, , .
n
) C
n
and ,= .. 1(, .) is a form of type
(n, n 1) in . It is clear that when n = 1,
1(, .) =
1
2i
1
.
d
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
)()1(, .)
_
D
) 1(, .) for . 1,
and
(2.2.3) 0 =
_
bD
)()1(, .)
_
D
) 1(, .) for . , 1.
Proof. A direct calculation shows that

1(, .) = 0 for ,= .. Since 1(, .) is of


type (n, n 1) in , by Stokes theorem we have, for . 1,
_
bD
)()1(, .) =
_
D(z)
d()()1(, .)) +
_
bB(z)
)()1(, .)
=
_
D(z)
)() 1(, .) +
_
bB(z)
)()1(, .),
2.2 The Bochner-Martinelli Formula 23
where 1

(.) = C
n
[ [ .[ < c for small c 0 and 1

(.) = 1 1

(.). Using
homogeneity of the kernel and Stokes theorem, we easily get
_
bB(z)
1(, .) =
(n 1)!
(2i)
n
n

j=1
_
bB(0)

j
[[
2n
d
j
(
k,=j
d
k
d
k
)
=
(n 1)!
(2i)
n
1
c
2n
n

j=1
_
B(0)
d
j
d
j
(
k,=j
d
k
d
k
)
= 1,
for all c 0. Now, letting c 0, we obtain
)(.) =
_
bD
)()1(, .)
_
D
)() 1(, .).
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
)()1(, .) =
_
D
)() 1(, .).
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 Bochner-Martinelli 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
)()1(, .) 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 Bochner-Martinelli kernel also enjoys the reproducing property for
holomorphic functions, although 1(, .) 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 dened on the boundary. Dene
1(.) =
_
bD
)()1(, .) 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 so-called 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
()() )
e
(.))1(, .).
Since
[1(, .)[ [. [
12n
from the denition of Bochner-Martinelli kernel, the right-hand side of (2.2.6) de-
nes a continuous function on the whole space. Thus, we have 1

(.) C(1). For


. , 1, we get
(2.2.7) 1
+
(.) =
_
bD
()() )
e
(.))1(, .).
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 Hardy-Littlewood lemma
(see Theorem C.1 in the Appendix). Thus, we need to estimate the dierential of
1

(.). An easy exercise shows that


[d(1

(.) )
e
(.))[

_
bD
[d)
e
(.)[[1(, .)[ +
_
bD
[)() )
e
(.)[[d1(, .)[

_
bD
1
[. [
2n1
d()
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 Cauchy-Riemann Operator in C 25
2.3 The Cauchy-Riemann Operator in C
Let
(2.3.1) A = A
1
+iA
2
be a rst order partial dierential operator dened 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 real-valued 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 rst order operator A
0
with real coecients
in some neighborhood of the origin. It follows that A is reduced to
(2.3.3) A = (.)A
0
,
and the solvability of (2.3.3) will then follow from the basic theory of the ordinary
dierential 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 Cauchy-Riemann 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
Cauchy-Riemann equation via the Cauchy integral formula.
As another application of the Cauchy integral formula, we shall show that, under
certain regularity hypotheses on the coecients 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
Cauchy-Riemann operator in n. Hence, one can deduce the solvability of A from
the knowledge of the Cauchy-Riemann 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 coecients o
j
(r, j) and /
j
(r, j),
, = 1, 2, are Holder continuous of order , 0 < < 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 coecients 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 Holder continuous of order . 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 dierential 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 Holder regularity for
the coecients of A. We shall rst 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 = (o, /) 1
R
. Put : = [. [. Then, for any 0, we
have
(2.3.7)
__
D
R
:

:
2
drdj
2

(21)

.
Proof. This is obvious if we apply polar coordinates to the disc centered at 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
1
and
2
contained in 1
R
, we have
i
2
__
D
R
d. d.
[.
1
[
21
[.
2
[
22
c(c
1
, c
2
)
1
[
1

2
[
212
.
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
1
and
2
.
Proof. By changing to polar coordinates it is easy to see that the integral exists.
For the estimate of the integral, let 2 = [
1

2
[ and

1
= . 1
R
[ [.
1
[ < ,

2
= . 1
R
[ [.
2
[ < ,

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.
[.
1
[
21
[.
2
[
22
2
22
_

0
:
11
d: =
2
c
1

1+22
.
Similarly, we have the estimate over
2
,
i
2
__
Z2
d. d.
[.
1
[
21
[.
2
[
22

2
c
2

1+22
.
2.3 The Cauchy-Riemann Operator in C 27
Both estimates are of the desired form. For the estimate over
3
, note that the
function (.
1
),(.
2
) is smooth on
3
. Hence, we obtain
1
3

.
1
.
2

3, for .
3
.
It follows we have
i
2
__
Z3
d. d.
[.
1
[
21
[.
2
[
22
3
22
__
Z3
drdj
[.
1
[
412
3
22
__
C\Z1
drdj
[.
1
[
412
23
22
_

:
1+23
d:
=
23
22
2 c
1
c
2

1+22
.
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 complex-valued continuous function dened on 1
R
which satises
(2.3.8) [)(.
1
) )(.
2
)[ 1[.
1
.
2
[

,
for any two points .
1
, .
2
1
R
, where , 1 are positive constants with 0 < < 1.
Dene the function 1() for 1
R
by
1() =
1
2i
__
D
R
)(.)
.
d. d..
Then 1 C
1+
(1
R
). If [)(.)[ for all . 1
R
, then we have
(1) 1

and 1

exist, and 1

() = )(), 1
R
.
(2) [1()[ 41, 1
R
.
(3) [1

()[ (
2
+1

)1

1, 1
R
.
(4) [1(
1
) 1(
2
)[ 2(+ (
2
+1

)1

1)[
1

2
[,
1
,
2
1
R
.
(5) [1

(
1
) 1

(
2
)[ j()1[
1

2
[

,
1
,
2
1
R
.
where j() 0 is independent of
1
and
2
.
Proof. The existence of 1

and the equality 1

() = )() are guaranteed by Theo-


rem 2.1.2. To prove the existence of 1

(), we write
1() =
1
2i
__
D
R
)(.) )()
.
d. d. +
)()
2i
__
D
R
1
.
d. d..
Note that by Cauchys integral formula (2.1.1), we get
=
1
2i
__
D
R
1
.
d. d..
28 The Cauchy Integral Formula and its Applications
Hence, if ) C
1
(1), we clearly have
(2.3.9) 1

() =
1
2i
__
D
R
)(.) )()
(. )
2
d. d..
In general, if ) is only Holder continuous of order , 0 < < 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 = 1 since
[1()[

__
D
R
1
[. [
drdj 41.
Similarly, from (2.3.8) and (2.3.9) we have
[1

()[
1

__
D
R
[. [

[. [
2
drdj
_
2
+1

_
1

1.
This gives (3). Now (4) follows immediately from the Mean Value Theorem and
estimate (3).
Finally, we estimate (5). Let
1
and
2
be two xed distinct points in 1
R
, and
set
=
)(
2
) )(
1
)

2
.
By the assumption (2.3.8) on )(), we have
[[ 1[
1

2
[
1
.
Set

)(.) = )(.) +., then we have

1()
1
2i
__
D
R

)(.)
.
d. d. = 1() 1
2
+[[
2
.
It follows that

() = 1

() +.
Note also that the denition of gives the following:
(1)

)(
1
) =

)(
2
),
(2)

)(.)

)(
1
) = )(.) )(
1
) +(.
1
),
(3)

)(.)

)(
2
) = )(.) )(
2
) +(.
2
),
Based on these observations, we obtain
2i(

(
1
)

1

(
2
))
=
__
D
R
_

)(.)

)(
1
)
(.
1
)
2

)(.)

)(
2
)
(.
2
)
2
_
d. d.
=
__
D
R
(

)(.)

)(
1
))(
1

2
)((.
1
) + (.
2
))
(.
1
)
2
(.
2
)
2
d. d.
2.3 The Cauchy-Riemann Operator in C 29
= (
1

2
)
__
D
R
)(.) )(
2
)
(.
1
)(.
2
)
2
d. d.
+ (
1

2
)
__
D
R
)(.) )(
1
)
(.
1
)
2
(.
2
)
d. d.
+ 2(
1

2
)
__
D
R
1
(.
1
)(.
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[
1

2
[
__
D
R
drdj
[.
2
[
2
[.
1
[
c()1[
1

2
[

,
where the constant c() depends only on . A similar estimate holds for 1
2
. The
term 1
3
can be written as
1
3
= 2(
1

2
)
__
D
R
1
(.
1
)(.
2
)
d. d.
= 2
__
D
R
_
1
.
1

1
.
2
_
d. d.
= (4i)(
1

2
).
Hence, we have
[1
3
[ 41[
1

2
[

.
These estimates together show that
[1

(
1
) 1

(
2
)[ [

(
1
)

1

(
2
)[ +[[[
1

2
[
j()1[
1

2
[

.
The constant j() is obviously independent of
1
and
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 integro-dierential equation from which Theorem 2.3.1 will follow.
Proposition 2.3.5. Let
(2.3.10) 7n = o(.)n
z
,
be a partial dierential operator whose coecient o(.) is Holder continuous of order
, 0 < < 1, and vanishes at . = 0. Then the equation
(2.3.11) 2in()
__
D
R
7n(.)
.
d. d. = (), 1
R
,
where () is a holomorphic function with (0) = 0, has exactly one solution n(.)
C
1+
(1
R
), provided that 1 is suciently 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
rst 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 < `, [()[ `,
(2.3.13) [/(
1
) /(
2
)[ `[
1

2
[

,
(2.3.14) [7(
1
) 7(
2
)[
`
2
2

[
1

2
[

,
where ,
1
,
2
denote any three points of 1
R
and /(.) stands for the functions o(.)
and (.). Since o(.) and 7(.) both vanish at . = 0, we have
(2.3.15) [o()[ `[[

`1

,
and
(2.3.16) [7()[ `
2
1

.
Now we consider the function 1() dened in Lemma 2.3.4 and using the notation
of the lemma, we obtain
(2.3.17) [71()[ = [o()1

[
_
2
+1

_
`1
2
1,
and
(2.3.18)
[71(
1
) 71(
2
)[
= [o(
1
)1

(
1
) o(
2
)1

(
2
)[
`[
1

2
[

_
2
+1

_
1

1 +`1

j()1[
1

2
[

`[
1

2
[

p(1)1,
where
p(1) =
__
2
+1

_
+j()
_
1

,
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 rst
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 suciently small
so that it satises
(2.3.19) 41

1 and 2
2
_
1 +
2

_
1
1
1.
2.3 The Cauchy-Riemann Operator in C 31
Denote by c 0 another universal constant such that
(2.3.20) j() +
2
+1

c.
Now we construct a sequence of functions n
j
()

j=0
to generate a solution of
(2.3.11). We rst set
(2.3.21) 2in
0
() = (),
and inductively dene for 1
R
,
(2.3.22) 2in
n+1
() =
__
D
R
7n
n
(.)
.
d. d., n = 0, 1, 2, .
Claim 2.3.6. The functions n
j
()

j=0
satisfy the following estimates:
(1) [n
n
()[ `(c`1

)
n
, 1
R
.
(2) [7n
n
()[ `(c`1

)
n+1
, 1
R
.
(3) [n
n
(
1
) n
n
(
2
)[ `(c`1

)
n
[
1

2
[

,
1
,
2
1
R
.
(4) [7n
n
(
1
) 7n
n
(
2
)[ (
cM
2
2

)(c`1

)
n
[
1

2
[

.
1
,
2
1
R
.
In particular, estimate (4) of the claim implies that the function under the integral
sign in (2.3.22) is Holder continuous of order , thus allowing the denition 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
()[ 41`(c`1

)
n+1
`(c`1

)
n+1
,
and
[n
n+1
(
1
) n
n+1
(
2
)[
2[
1

2
[
_
`(c`1

)
n+1
+
_
2
+1

_
1

_
c`
2
2

_
(c`1

)
n
_
= `(c`1

)
n+1
[
1

2
[

_
1 +
2

_
2[
1

2
[
1
`(c`1

)
n+1
[
1

2
[

_
2
2
_
1 +
2

_
1
1
_
`(c`1

)
n+1
[
1

2
[

.
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
()[
_
2
+1

_
`1
2
(
c`
2
2

)
_
c`1

_
n
= `(c`1

)
n+2
_
2
c
_
`(c`1

)
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
(
1
) 7n
n+1
(
2
)[
_
c`
2
2

_
(c`1

)
n+1
[
1

2
[

_
1
c1

_
p(1)

_
c`
2
2

_
(c`1

)
n+1
[
1

2
[

.
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 satises c`1

< 1. It follows that


the series
(2.3.23)

j=0
n
j
(.)
converges absolutely and uniformly in 1
R
, and denes a solution n(.) C
1+
(1
R
)
which satises (2.3.11).
For the uniqueness of the solution when 1 is suciently small, let (.) be another
solution of (2.3.11) such that 7(.) satises a Holder condition of order . Then
the function
n(.) = n(.) (.)
satises the equation
(2.3.24) 2i n() =
__
D
R
7 n(.)
.
d. d..
Put
= (1) = sup
D
R
[7 n()[,
and
1 = 1(1) = sup
1,2D
R
1,=2
[7 n(
1
) 7 n(
2
)[
[
1

2
[

.
Obviously, 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

_
2
+1

_
`1
2
1,
33
and
1 `p(1)1.
Since p(1) approaches zero when 1 tends to zero, we must have 1 = 0 for su-
ciently small 1. This implies = 0 when 1 is suciently small. Hence n(.) = (.)
on 1
R
for some suciently 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 (.) = . in the statement of Proposition 2.3.5. Then there
exists a unique solution n(.) C
1+
(1
R
) to the equation
2in()
__
D
R
7n(.)
.
d. d. = , for 1
R
,
with n

(0) ,= 0 if 1 is suciently small. Since 7n(.) satises a Holder condition


of order , we see by Lemma 2.3.4 that n(.) satises 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 Poincare lemma for the operator
is often known as the Dolbeault-Grothendieck lemma (see [Dol 1,2]).
Theorem 2.2.1 is a special case of the so-called Bochner-Martinelli-Koppelman
formula due to W. Koppelman [Kop 1]. Corollary 2.2.2 concerning the reproducing
property of the Bochner-Martinelli 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 coecient functions
p
ij
(r, j), 1 i, , 2, are Holder continuous of order , 0 < < 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
= (n, )(dn
2
+d
2
),
for some (n, ) 0. If the coecient 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
real-valued dening function and an open
neighborhood l of j such that ` l = . l[(.) = 0 and d(.) ,= 0 on
`l. ` divides l into two sides, l
+
and l

, where l
+
= . l[(.) 0 and
l

= . l[(.) < 0. Dene by 1, a type (0, 1) vector eld on `, such that


1 =
n

j=1
o
j
(.)

.
j
on ` l,
where the o
t
j
: satisfy
n

j=1
o
j
(.)

.
j
(.) = 0 for all . ` l.
Any such vector eld 1 on ` is called a tangential Cauchy-Riemann 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 Cauchy-Riemann equations.
Denition 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 ) satises 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
(,.
j
)(.) = 0, . `, where (.) is
a C
1
dening 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 so-called holomorphic extension of C1 functions. In
general, the converse part is not true.
For instance, let ` be dened by j
1
= 0 in C
n
, where .
j
= r
j
+ ij
j
, 1 ,
n. Consider a real-valued 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 elds ,.
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 rst consider the problem of global holomorphic extension
of a C1 function on a compact hypersurface. We then study the local one-sided
holomorphic extension of a C1 function. In Sections 4 and 5, we dene plurisub-
harmonic functions, pseudoconvex domains and domains of holomorphy. We study
their relations with each other, and give several equivalent denitions 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 dierences between one and several complex variables is the
so-called Hartogs extension theorem, which states that if a bounded domain 1 in
C
n
, n 2, has connected boundary, then any holomorphic function )(.) dened
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 Cauchy-Riemann 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 satised. Then there is a function n C
k
0
(C
n
) satisfying
(3.1.1). When / = 0, if (3.1.2) is satised in the distribution sense, then there exists
a function n C
0
(C
n
) such that n satises (3.1.1) in the distribution sense.
Proof. For / 1, set
n(.) =
1
2i
_
C
)
1
(, .
2
, , .
n
)
.
1
d d
=
1
2i
_
C
)
1
( +.
1
, .
2
, , .
n
)

d d.
It is easily seen that n C
k
(C
n
) from dierentiation under the integral sign. We
also have n(.) = 0 when [.
2
[ + +[.
n
[ is suciently 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
2i
_
C
fj

(, .
2
, , .
n
)
.
1
d d = )
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 suciently 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, dene n(.) by the same equation. We see that n C
0
(C
n
), and
that n(.) = 0 when [.
2
[ + +[.
n
[ is suciently large.
By Theorem 2.1.2,
n
.
1
= )
1
(.).
For , 1, let C

0
(C
n
). Then, using ( , )
C
n to denote the pairing between
distributions and test functions, we have
_
n
.
j
,
_
C
n
=
_
n,

.
j
_
C
n
=
_
C
n
_
1
2i
_
C
)
1
( +.
1
, .
2
, , .
n
)

d d
_

.
j
d(.)
=
1
2i
_
C
1

__
C
n
)
1
( +.
1
, .
2
, , .
n
)

.
j
d(.)
_
d d
=
1
2i
_
C
1

_
)
1
.
j
( +.
1
, .
2
, , .
n
),
_
C
n
d d
=
1
2i
_
C
1

_
)
j
.
1
( +.
1
, .
2
, , .
n
),
_
C
n
d d
=
1
2i
_
C
1

__
C
n
)
j
( +.
1
, .
2
, , .
n
)

.
1
d(.)
_
d d
=
_
C
n
_
1
2i
_
C
)
j
( +.
1
, .
2
, , .
n
)

d d
_

.
1
d(.)
=
_

.
1
_
1
2i
_
C
)
j
(, .
2
, , .
n
)
.
1
d d
_
,
_
C
n
=
_
)
j
,
_
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 ) dened on 1 1 can be extended holomorphically to 1.
Proof. Choose a cut-o function C

0
(1) such that = 1 in some open neigh-
borhood of 1. Then )() C

(0,1)
(C
n
) satises the compatibility conditions,
and it has compact support. By Theorem 3.1.1 there is a n C

0
(C
n
) such that
n = ),
and that n = 0 in some open neighborhood of C
n
1. Then, it is easily seen that
1 = (1 )) 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 suces to show ) is holomorphic near each j 1 o. Let us x 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
n1
[ j = (r, .
t
),
for some smooth function such that (0) = 0 and d(0) = 0.
Hence, for any small 0, there exists a

0 and a polydisc l

in C
n1
centered at the origin, such that
[(r, .
t
)[ <
for all [r[ <

and .
t
l

. Let
1
0 be suciently small, and let
2

1
be a
positive number suciently close to
1
, then we may assume that \
0
l
1
, where
\
0
= .
1
C[ [r[ <
1
and
1
< j <
2

is contained in 1 o. Thus, ) O(\


0
l
1
).
Next, for each .
t
l
1
, )(.
1
, .
t
) is continuous on
\ = .
1
C[ [r[ <
1
and [j[ <
2

and holomorphic on \ except for the smooth curve j = (r, .


t
). By Moreras
theorem in one complex variable, )(.
1
, .
t
) is holomorphic in .
1
on \ . Now choose
38 Holomorphic Extension and Pseudoconvexity
a contour of integration in l
1
. Namely, let =
2

n
, where
j
= .
j

C[ [.
j
[ = :
j
, for 2 , n, so that l
1
.
Dene
1(.
1
, .
t
) =
1
(2i)
n1
_

)(.
1
,
t
)
(
2
.
2
) (
n
.
n
)
d
2
d
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
dening function :, / 1.
Then any C1 function of class C
k
on ` can be extended to a C
k1
function

) in
some open neighborhood of ` such that


) = 0 on `.
Proof. We rst 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
n1
forms a basis of the tangential Cauchy Riemann equations. Let
: be a dening 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
eld transversal to the boundary everywhere. When / 2, we have that :(1
n
)) is
a C
k1
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
denition, 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
dened 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 dene 1

(.) and 1
+
(.) as the Bochner-Martinelli 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 rst show that 1

(.) O(1) and 1


+
(.) O(C
n
1). Dene 1(, .) by
(2.2.1) and
1
1
(, .)
=
n 1
(2i)
n
_
n

j=1

j
.
j
[ .[
2n
d
j
_

_
n

j=1
d
j
d
j
_
n2

_
n

j=1
d.
j
d
j
_
.
Then we have
(3.2.2)
z
1(, .) +

1
1
(, .) = 0, for ,= ..
Identity (3.2.2) is proved by a straightforward calculation as follows: For ,= .,

z
1(, .)
=
(n 1)!
(2i)
n
n

j=1
d.
j
d
j
[ .[
2n
(
k,=j
d
k
d
k
)

n!
(2i)
n
n

j=1
_
n

l=1
(
j
.
j
)(
l
.
l
)
[ .[
2n+2
d.
l
d
j
_
(
k,=j
d
k
d
k
)
=
n 1
(2i)
n
1
[ .[
2n
_
n

j=1
d.
j
d
j
_

_
n

j=1
d
j
d
j
_
n1

n(n 1)
(2i)
n
_
_

j,=l
(
j
.
j
)(
l
.
l
)
[ .[
2n+2
d.
l
d
j
d
l
d
l
_
_

_
n

j=1
d
j
d
j
_
n2
+
n!
(2i)
n
n

j=1
_
1
[ .[
2n

[
j
.
j
[
2
[ .[
2n+2
_
d.
j
d
j
(
k,=j
d
k
d
k
)
40 Holomorphic Extension and Pseudoconvexity
=
n 1
(2i)
n
1
[ .[
2n
_
n

j=1
d.
j
d
j
_
(
n

j=1
d
j
d
j
)
n1
+
n(n 1)
(2i)
n
_
_

j,=l
(
j
.
j
)(
l
.
l
)
[ .[
2n+2
d
l
d
j
d.
l
d
l
_
_

_
n

j=1
d
j
d
j
_
n2
+
n(n 1)
(2i)
n
_
_
n

j=1
[
j
.
j
[
2
[ .[
2n+2
d
j
d
j
_
_

_
n

j=1
d.
j
d
j
_

_
n

j=1
d
j
d
j
_
n2
=
n 1
(2i)
n
1
[ .[
2n
_
n

j=1
d
j
d
j
_
n1

_
n

j=1
d.
j
d
j
_
+
n(n 1)
(2i)
n
_
_
n

j=1
n

l=1
(
j
.
j
)(
l
.
l
)
[ .[
2n+2
d
l
d
j
_
_

_
n

j=1
d
j
d
j
_
n2

_
n

j=1
d.
j
d
j
_
=

1
1
(, .).
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 dierentiable at /1 and ) = 0 on /1. It follows now from (3.2.2) that for
. 1, we have

z
1

(.) =
_
bD
)()
z
1(, .)
=
_
bD
)()

1
1
(, .)
=
_
bD

()()1
1
(, .))
=
_
bD
d

()()1
1
(, .))
= 0.
Thus, 1

(.) O(1) and, similarly, 1


+
(.) O(C
n
1).
Finally, we claim 1
+
(.) 0. We let [.
2
[ + + [.
n
[ be suciently large, then
1
+
(, .
2
, , .
n
) is an entire function in .
1
which tends to zero as [.
1
[ tends to
innity. Thus, by Liouvilles 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 one-sided holomorphic extension of
a smooth C1 function dened 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 so-called Levi form of the domain.
Denition 3.3.1. Let 1 be a bounded domain in C
n
with n 2, and let : be a C
2
dening 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,
dened 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 is another C
2
dening function for 1, then = /: 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

.
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 dening
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 dening 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 elds which are tangent to the
boundary at the point j. Smooth sections in T
0,1
(/1) are the tangential Cauchy-
Riemann operators dened in the introduction. By denition, the Levi form is
applied only to the tangential type (1, 0) vector elds. 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
dening 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 nd 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
+(.) +O([.[
3
),
where .
n
= r
n
+ij
n
and
(.) =
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
)
n1
j,k=1
is not positive semidenite. 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 rst choose 0, and then c 0 so that

2
:
.
1
.
1
(.) < 0
on
l
t
= . C
n
[ [.
1
[ < and [.
2
[ + +[.
n
[ < c l,
3.4 Pseudoconvexity 43
and :(.) < 0 on the part of the boundary where [.
1
[ = . For any xed .
t
=
(.
2
, , .
n
) with [.
2
[ + + [.
n
[ < c, the set of all .
1
with [.
1
[ < such that
:(.
1
, .
t
) < 0 must be connected. Otherwise, :(.
1
, .
t
) will attain a local minimum
at some point [.
1
[ < 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 satises ) = 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 dened on
\ = C \ , where \ = .
t
= (.
2
, , .
n
) C
n1
[[.
2
[ + + [.
n
[ < c. For any
.
t
\ , dene
(3.3.2) G(.
1
, .
t
) =
1
2i
_
C
p
1
(, .
t
)
.
1
d d.
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 suciently small positive number
0 < << 1, there is a small open neighborhood \
0
of (0, , 0, i) in C
n1
=
.
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 satises 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 one-sided 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 dened 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 dene the
concept of pseudoconvexity. We also discuss the relations between pseudoconvexity
and plurisubharmonic functions.
Denition 3.4.1. Let 1 be a bounded domain in C
n
with n 2, and let : be a C
2
dening 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 Denition 3.4.1 is clearly independent of the choice of the dening
function :.
Denition 3.4.2. A function dened on an open set 1 C
n
, n 2, with values
in [, +) is called plurisubharmonic if
(1) is upper semicontinuous,
(2) for any . 1 and n C
n
, (. + n) is subharmonic in C whenever
. +n[ C 1.
Theorem 3.4.3. A C
2
real-valued function on 1 is plurisubharmonic if and only
if
(3.4.2)
n

j,k=1

.
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 (. +n) in C whenever it is dened.
If (3.4.2) is strictly positive, we shall call a strictly plurisubharmonic function.
It is obvious from Denition 3.4.2 that any plurisubharmonic function satises the
submean value property on each complex line where it is dened. The following
theorem shows that there always exists a strictly plurisubharmonic dening 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 / ) dening function :(.). Then there exists a C
k
(2 / )
strictly plurisubharmonic dening function for 1.
Proof. For any 0, set
(.) = c
r
1 for . 1.
We will show that (.) is the desired strictly plurisubharmonic dening function
for 1 if is chosen to be suciently large.
First, is a C
k
dening function for 1 since
(.) = :(.) ,= 0 for . /1.
3.4 Pseudoconvexity 45
Next, we calculate the Levi form of (.) for . /1 to get that
n

j,k=1

.
j
.
k
(.)t
j
t
k
=
n

j,k=1

2
:
.
j
.
k
(.)t
j
t
k
+
2

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

.
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 satises [

n
j=1
(:,.
j
)(.)t
j
[

2
for . /1,
we may also achieve

n
j,k=1
(
2
,.
j
.
k
)(.)t
j
t
k
0 simply by choosing to be
suciently large. This shows that is strictly plurisubharmonic near the boundary
by continuity if 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
dening function for 1 such that
N

j,k=1

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
,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 rst that 1 is strongly pseudoconvex. By Theorem 3.4.4 there is
a C
2
strictly plurisubharmonic dening 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 dening 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.
Denition 3.4.6. A function : 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[ (r) < c is relatively
compact in 1.
Clearly, if is an exhaustion function for 1, then (r) as r /1. This
condition is also sucient 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
dening function
for 1 in some small open neighborhood of the boundary. Hence the Levi form
dened by : is positive semidenite.
If log(d
D
(.)) is not plurisubharmonic near the boundary, then

logd
D
(. +n)[
=0
0
for some n C
n
and . close to the boundary where d
D
(.) is C
2
. Expand logd
D
(. +
n) at = 0 to get
logd
D
(. +n) = log(d
D
(.)) + Re( +
2
) +[[
2
+O([[
3
),
for small . Here , C, 0 are constants. Choose C
n
such that .+ /1
and [[ = d
D
(.). Then consider the analytic disc

= .() = . +n +c
+
2
[ [[
for some suciently small 0. Using Taylors expansion, for [[ , ,= 0, we
get
d
D
(.()) d
D
(. +n) [[[c
+
2
[
[[(c

2
]]
2
1)[c
+
2
[
0,
if is small enough. Since .(0) = . + /1, this implies that

is tangent to
the boundary at .(0). Hence,

d
D
(.())[
=0
= 0 and

2

d
D
(.())[
=0
0.
From the denition of :, this means
n

k=1
:
.
k
(. +).
t
k
(0) = 0 and
n

j,k=1

2
:
.
j
.
k
(. +).
t
j
(0).
t
k
(0) < 0.
This contradicts the nonnegativeness of the Levi form at .(0) = . + . 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 (.) be a C
2
function on 1 such that (.) = 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
(.) = (.) +`[.[
2
is a C
2
strictly plurisubharmonic exhaustion function on 1 if ` is chosen large
enough.
The next step is to regularize (.). For each , N, we set 1
j
= . 1[(.) <
,, then 1
j
1. Choose a function (.) = ([.[) C

0
(1(0; 1)) such that
(.) 0 and
_
(.)d\ = 1. Set

(.) = c
2n
(.,c). For . 1
j
, the function

(.) =
_
(. )

()d\ () =
_
(. c)()d\ ()
is dened and smooth on 1
j
if c is suciently small. Since is strictly plurisub-
harmonic of class C
2
, it is clear that

(.) is strictly plurisubharmonic and, by the


submean value property,
1

2
if c
1
< c
2
, and

(.) converges uniformly to (.)


on any compact subset of 1.
Therefore, by extending

(.) in a smooth manner to 1, we see that there are


functions
j
(.) C

(1) for , N such that


j
(.) is strictly plurisubharmonic
on 1
j+2
, (.) <
1
(.) < (.) + 1 on 1
2
and (.) <
j
(.) < (.) + 1 on 1
j
for
, 2. It follows that

j
(.) , + 1 < 0 on 1
j2
for , 3,
and

j
(.) , + 1 0 on 1
j
1
j1
for , 3.
Now choose a (r) C

(1) with (r) = 0 for r 0 and (r),


t
(r),
tt
(r)
positive for r 0. Then, (
j
(.) , +1) 0 and (
j
(.) , +1) 0 on 1
j2
.
A direct computation shows that (
j
(.), +1) is plurisubharmonic on 1
j+2
and
strictly plurisubharmonic on 1
j
1
j1
. Thus, one may choose inductively :
j
N
so that, for / 3,

k
(.) =
1
(.) +
k

j=3
:
j
(
j
(.) , + 1)
is strictly plurisubharmonic and
k
(.) (.) on 1
k
. Clearly,
k
(.) =
k1
(.) on
1
k2
. Thus, (.) = lim
k

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 (.) on 1. Dene 1
c
= . 1[ (.) < c for every c 1. It follows
from Sards 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
dene pseudoconvexity by the following
Denition 3.4.9. An open domain 1 in C
n
is called pseudoconvex if there exists
a smooth strictly plurisubharmonic exhaustion function (.) on 1.
Theorem 3.4.10. Let 1 be a pseudoconvex domain in C
n
, n 2, in the sense of
Denition 3.4.9. Then log(d
D
(.)) is plurisubharmonic and continuous on 1.
Proof. Let 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
+n)
is subharmonic in C whenever .
0
+n 1. Choose 0 so that

0
= .
0
+n[ [[ 1,
and let )() be a holomorphic polynomial such that
(3.4.4) logd
D
(.
0
+n) Re)() for [[ = .
We want to show that
logd
D
(.
0
+n) Re)() for [[ .
Equation (3.4.4) is equivalent to
(3.4.5) d
D
(.
0
+n) [c
f()
[ for [[ = .
Now, for any C
n
with [[ < 1, we consider the mapping with 0 t 1,
(3.4.6) .
0
+n +tc
f()
for [[ .
The image of (3.4.6) is an analytic disc. Let
t
= .
0
+n +tc
f()
[ [[ .
Set 1 = t [0, 1][
t
1. Clearly, 0 1 and 1 is open. To show that 1 is
closed, set 1 =
0t1
/
t
. Estimate (3.4.5) implies that 1 is a compact subset of
1. Now, if
t
1 for some t, (.
0
+n +tc
f()
) would dene 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 , we see that
t
must be contained in . 1[ (.) sup
K
, a compact
subset of 1. It follows that 1 is closed, and hence 1 = [0, 1]. This implies, for any
C
n
with [[ < 1 and [[ , that
.
0
+n +c
f()
1.
Thus, we have
d
D
(.
0
+n) [c
f()
[ for [[ ,
or equivalently,
logd
D
(.
0
+n) Re)() for [[ .
Hence, logd
D
(.
0
+ n) is subharmonic in C whenever .
0
+ n 1. This
proves the theorem.
The equivalence between Denitions 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
Denition 3.4.9.
Proof. If 1 is Levi pseudoconvex, then by Corollary 3.4.8 1 is pseudoconvex in the
sense of Denition 3.4.9.
On the other hand, assume 1 is pseudoconvex according to Denition 3.4.9.
Dene : by(3.4.3). Then :(.) is a C
2
dening 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 suciently 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

j=1
d
D
.
j
o
j

2
0,
for any o C
n
and . 1 suciently 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 Denition 3.4.9 every pseudoconvex domain 1 can be exhausted
by strictly pseudoconvex domains, i.e.,
1 = 1

,
where 1

1
+1
1 and each 1

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 dening function for 1. Then there exist constants 1 0 and
0 <
0
< 1, such that for any with 0 <
0
, = (:c
K]z]
2
)

is a smooth
bounded strictly plurisubharmonic exhaustion function on 1.
Note that is continuous on 1 and vanishes on /1.
Proof. We rst assume that . 1 with [:(.)[ c for some small c 0. With
= (:c
K]z]
2
)

, a direct calculation shows, for t C


n
,
L
z
(; t) = (:)
2
c
K]z]
2
_
1:
2
_
[t[
2
1

j=1
.
j
t
j

2
_
+ (:)
_
L
z
(:; t) 21Re
_
n

i=1
:
.
i
t
i
__
n

j=1
.
j
t
j
__
+ (1 )

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

+ t

, where
t

= (t

1
, , t

n
) with
t

k
=
_

n
j=1
t
j
r
zj
(.)

n
j=1
[
r
zj
(.)[
2
_
:
.
k
(.),
and t

= (t

1
, , t

n
) T

z
= o C
n
[

n
j=1
(:,.
j
)(.)o
j
= 0. Such a decompo-
sition is clearly smooth when c is suciently small. Also, let (.) be the projection
of . along the normal on the boundary. Obviously, is smooth for small c. Then
the Levi form of : at . is
L
z
(:; t

) =
n

i,j=1

2
:
.
i
.
j
(.)t

i
(.)t

j
(.)
=
n

i,j=1

2
:
.
i
.
j
(.)(t
i
t

i
(.))(t
j
t

j
(.))
=
n

i,j=1
/
ij
(.)t
i
t
j
,
where /
ij
(.) is dened by the last equality. Hence, by pseudoconvexity of the
domain, we have
(3.4.7)
L
z
(:; t

(.)) L
z
(:; t

(.)) L
(z)
(:; t

((.)))
=
n

i,j=1
(/
ij
(.) /
ij
((.)))t
i
t
j
C[:(.)[[t[
2
,
for some constant C 0. Since
(3.4.8) [t

[ = O
_

i=1
:
.
i
t
i

_
,
(3.4.7) and (3.4.8) together imply
L
z
(:; t) C[:(.)[[t[
2
C[t[

j=1
:
.
j
t
j

.
Hence
L
z
(; t) (:)
2
c
K]z]
2
_
1:
2
(1 C1)[t[
2
C:
2
[t[
2
+C:[t[

j=1
:
.
j
t
j

+ (1 )

j=1
:
.
j
t
j

2
_
,
3.5 Domains of Holomorphy 51
for some constant C 0. Since
C[:[[t[

j=1
:
.
j
t
j

1
4

j=1
:
.
j
t
j

2
+C
1
:
2
[t[
2
,
we have
L
z
(; t) (:)
2
c
K]z]
2
_
1:
2
(1 C1)[t[
2
(C +C
1
):
2
[t[
2
+
_
3
4

_

j=1
:
.
j
t
j

2
_
.
Now, if we rst choose 1 2(C + C
1
) + 10 and then to be suciently small so
that < 1,4 and C1 < 1,2, we have
L
z
(; t) 0, for t C
n
0.
For this case we may take
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
denition of a domain of holomorphy.
Denition 3.5.1. A domain 1 in C
n
is called a domain of holomorphy if we
cannot nd 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. Dene 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 lling 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 dene:
52 Holomorphic Extension and Pseudoconvexity
Denition 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 0[ . +1(0; :) 1.
To prove (1) (2), we rst 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 be a xed point in

1
D
. Then any / O(1) extends holomorphically to
1 +[)()[1(0; :).
Proof. For each 0 < t < 1, the union of the polydiscs with centers at . 1
(3.5.2) 1
t
=
zK
. +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 Cauchys estimates of /, we obtain
(3.5.3)
[

h
z

(.)[t
]]
[)(.)[
]]
:

!
`
t
for . 1 and all multiindices = (
1
, ,
n
) with [[ =
1
+ +
n
. Since
(

/,.

)(.))(.)
]]
is holomorphic on 1, by denition, (3.5.3) also holds for .

1
D
. Letting t tend to one, we see that /(.) extends holomorphically to 1 +
[)()[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 suciently 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)
() (1 +c)d
w
(), for all

1
D
.
Letting c tend to zero, we get
dist(1, 1
c
) inf

b
K
D
( inf
]w]1
d
w
())
= inf

b
K
D
dist(, 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
i

i=1
be a sequence of points in 1 such that every point
belonging to T appears innitely 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
i
the largest polydisc of the form 1
i
=
i
+ 1(0; 1) that is contained in
1, where 0 and 1(0; 1) is the polydisc centered at the origin with multiradii
: = (1, , 1). Then, inductively for each ,, pick a .
j
(1
j
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
)[
j1

i=1
[)
i
(.
j
)[ +, + 1.
It follows that
/(.) =

j=1
)
j
(.)
54 Holomorphic Extension and Pseudoconvexity
denes a holomorphic function on 1 and that
[/(.
j
)[ [)
j
(.
j
)[
j1

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
i
for some
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 Denition 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
j
(.) =

mj
k=1
[)
jk
(.)[
2
satises
(3.5.5)
j
(.) <
1
2
j
for . 1
j
,
and

j
(.) , for . 1
j+2

o
1
j+1
.
Hence,
(.) =

j=1

j
(.)
is a continuous exhaustion function dened on 1. In fact, (.) 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 denes a holomorphic function on
11

. By substituting . for n in the above series, we obtain the real analyticity


of (.) on 1, and that one can dierentiate (.) term by term. Obviously, (.)
is plurisubharmonic on 1. It follows that [.[
2
+(.) is a smooth strictly plurisub-
harmonic exhaustion function on 1, and by denition, 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 nd 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 specic 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 dening function for 1 and we assume that j = 0. Let
1(.) = 2
n

i=1
:
.
i
(0).
i

i,j=1

2
:
.
i
.
j
(0).
i
.
j
.
1(.) is holomorphic, and it is called the Levi polynomial of : at 0. Using Taylors
expansion at 0, there exists a suciently 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
cut-o functions to patch together the local holomorphic data, since the cut-o
functions are no longer holomorphic. Let be a cut-o function such that
C

0
(l) and = 1 in a neighborhood of 0. We note that ) is not holomorphic in
1. However, if ) can be corrected by solving a

-equation, then the Levi problem
will be solved.
Let us consider the (0,1)-form p dened by
p =

()) = (

)).
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 nd a solution
n C

(1) such that


(3.6.1)

n = p in 1,
56 Holomorphic Extension and Pseudoconvexity
then we dene for . 1,
/(.) = (.))(.) n(.).
It follows that / is holomorphic in 1, / C

(10) 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 dicult 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

(Hormanders 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 (Kohns 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 Cauchy-Riemann
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 one-sided 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 so-called 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 (Denition 3.4.2) was rst 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 rst 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 Holder bounded strictly plurisubharmonic exhaustion
function on a C
2
bounded pseudoconvex domain was rst 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. Hormander [Hor 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 Hormanders 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
Morrey-Kohn-Hormander. 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 dened operator from
one Hilbert space to another. This will be done in the next section. We rst 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 dened operator. We recall that T is closed if and only if the graph of
T is closed. The domain of denition 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 dened operator and T

= T.
(See [RiNa 1].)
59
We use Ker(T) and (T) to denote the kernel and the range of T respectively.
Since T is a closed operator, Ker(T) is closed. Let (T) denote the closure of the
range of T. By the denition of the adjoint operator, it is easy to see that
(4.1.1) H
1
= Ker(T) (T

)
and
(4.1.2) H
2
= Ker(T

) (T).
In later applications, the operator T will be a system of dierential operators
associated with the Cauchy-Riemann equations and H
1
, H
2
will be spaces of forms
with 1
2
coecients. 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 dened operator.
The following conditions on T are equivalent:
(1) (T) is closed.
(2) There is a constant C such that
(4.1.3) | ) |
1
C | T) |
2
for all ) Dom(T) (T

).
(3) (T

) is closed.
(4) There is a constant C such that
(4.1.4) | ) |
2
C | T

) |
1
for all ) Dom(T

) (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) (T

) (T)
is one-to-one, and its inverse
T
1
: (T) Dom(T) (T

)
is well-dened 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) (T

). Thus
[(p, /)
2
[ C | T

p |
1
| / |
2
, for p Dom(T

) and / (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 dened for arbitrary 1 and J so that they
are antisymmetric. The operator

(p,q)
: C

(p,q)
(1) C

(p,q+1)
(1)
is dened 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 diers 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 coecients 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 dene
), 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 is a continuous
function in 1, then 1
2
(1, ) is the space of functions in 1 which are square inte-
grable with respect to the weight function c

. The norm in 1
2
(p,q)
(1, ) is dened
by
|)|
2

=
_
D
[)[
2
c

d\, ) 1
2
(p,q)
(1, ).
4.2 The -Neumann Problem 61
The inner product in 1
2
(p,q)
(1, ) is denoted by ( , )

. Notice that the space


1
2
(p,q)
(1, ) is equal to 1
2
(p,q)
(1) if 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 dened 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,q1)
(1) C

(p,q)
(1), 1 n, under the usual
1
2
norm is denoted by , where
=
(p,q)
: C

(p,q)
(1) C

(p,q1)
(1).
The operator is dened by the requirement that
(4.2.2) (), p) = (),

p)
for all smooth p C

(p,q1)
(1) with compact support in 1. If ) is expressed by
(4.2.0) and p =

t
]I]=p,]K]=q1
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
_
= (), p).
Therefore, can be expressed explicitly by
(4.2.3) ) = (1)
p1

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
=
2
= 0. Let

be the formal adjoint of



under the 1
2
(1, ) norm, i.e.,
(4.2.4) (

), p)

= (),

p)

for every compactly supported p C

(p,q1)
(1). We have the following relation
between and

: for any ) C

(p,q)
(1),
(4.2.5)

) = c

(c

)).
Thus

, and

are systems of rst order dierential operators.


We take the (weak) 1
2
closure of the unbounded dierential operator

, still
denoted by

. Let

p,q1
: 1
2
(p,q1)
(1, ) 1
2
(p,q)
(1, )
be the maximal closure of the Cauchy-Riemannn operator dened as follows: an
element n 1
2
(p,q1)
(1, ) is in the domain of

if

n, dened in the distribution
62 L
2
Theory for on Pseudoconvex Domains
sense, belongs to 1
2
(p,q)
(1, ). Then

denes a linear, closed, densely dened
operator.

is closed since dierentiation is a continuous operation in distribution
theory. It is densely dened since Dom(

) contains all the compactly supported


smooth (j, 1)-forms. If 1 is bounded, any ) C

(p,q1)
(1) is in Dom(

).
The Hilbert space adjoint of

, denoted by

, is a linear, closed, densely dened


operator and

: 1
2
(p,q)
(1, ) 1
2
(p,q1)
(1, ).
When = 0, we denote the adjoint by

. Let Dom(

) and Dom(

) denote the
domains for

and

, respectively. An element ) belongs to Dom(

) if there
exists a p 1
2
(p,q1)
(1, ) such that for every Dom(

) 1
2
(p,q1)
(1, ), we
have
(),

)

= (p, )

.
We then dene

) = p. Note that if ) Dom(

), then it follows from (4.2.4)


that

) =

) where

is dened in the distribution sense in 1.


If 1 is bounded, we have C

(p,q1)
(1) Dom(

). However, not every element


in C

(p,q)
(1) is in Dom(

). Any element in Dom(

) (or Dom(

)) must satisfy
certain boundary conditions in the weak sense. If 1 has C
1
boundary /1, then any
) Dom(

) 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 be a C
1
dening function for 1. For any ) Dom(

) C
1
(p,q)
(1), where C
1
(1), )
must satisfy the boundary condition
(4.2.6) (, d))(.) = 0, . /1,
where (, d))(.) = ())(.) denotes the symbol of in the d direction evaluated
at .. More explicitly, if ) is expressed as in (4.2.0), then ) must satisfy
(4.2.6
t
)

k
)
I,kK

.
k
= 0 on /1 for all 1, 1,
where [1[ = j and [1[ = 1.
Proof. We rst assume that = 0. Note that (4.2.6) and (4.2.6
t
) are independent
of the dening function . We normalize such that [d[ = 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 C

(1) Dom(

),
(), ) =
n

i=1
_

)
i
.
i
,
_
=
n

i=1
_
)
i
,

.
i
_

i=1
_
bD
)
i

.
i

do
= (),

) +
_
bD
(, d)), )do,
4.2 The -Neumann Problem 63
where do is the surface measure of /1. Similarly, for a (j, )-form ) and
C

(p,q1)
(1) Dom(

), using integration by parts, we obtain


(4.2.7) (), ) = (),

) +
_
bD
(, d)), )do.
If, in addition, has compact support in 1, we have
(

), ) = (), ) = (),

),
where the rst equality follows from ) Dom(

) C
1
(p,q)
(1). Since compactly
supported smooth (j, 1)-forms are dense in 1
2
(p,q1)
(1), we must have
_
bD
(, d)), )do = 0, for any C

(p,q1)
(1).
This implies that (, d))(.) = 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 ,= 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 dened as the dual of the wedge product. For any (j, )-form ),

) is dened as the (j, 1)-form satisfying


p

, )) = p,

)), p C

(p,q1)
(C
n
).
Using this notation, condition (4.2.6) or (4.2.6
t
) can be expressed as
(4.2.6
tt
)

) = 0 on /1.
It is also easy to see that ) C
1
(p,q)
(1) Dom(

) if and only if ) satises one of


the three equivalent conditions (4.2.6), (4.2.6
t
) or (4.2.6
tt
).
For a xed 0 j n, 1 n, we dene the Laplacian of the

complex
1
2
(p,q1)
(1)

(p,q1)

(p,q)
1
2
(p,q)
(1)

(p,q)

(p,q+1)
1
2
(p,q+1)
(1).
Denition 4.2.2. Let
(p,q)
=

(p,q1)

(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,q1)
).
Proposition 4.2.3.
(p,q)
is a linear, closed, densely dened self-adjoint operator.
Proof.
(p,q)
is densely dened 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,q1)
(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
) = (

)
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-
ne
1
1
=

+1 =
(p,q)
+1 on Dom(
(p,q)
).
We shall prove that 1
1
1
is self-adjoint. By a theorem of Von Neumann [RiNa 1],
(1 +

)
1
and (1 +

)
1
are bounded self-adjoint operators. We dene
Q
1
= (1 +

)
1
+ (1 +

)
1
1.
Then Q
1
is bounded and self-adjoint. We claim that Q
1
= 1
1
1
. Since
(1 +

)
1
1 = (1 (1 +

))(1 +

)
1
=

(1 +

)
1
,
we have that (1 +

)
1
Dom(

). Similarly, we have (1 +


)
1

Dom(


) and
Q
1
= (1 +

)
1

(1 +

)
1
.
Since

2
= 0, we have (Q
1
) Dom(


) and

Q
1
=

(1 +

)
1
.
Similarly, we have (Q
1
) Dom(

) and

Q
1
=

(1 +

)
1
.
Thus, (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 self-adjoint
which implies
(p,q)
= 1
1
1 is self-adjoint. 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 be a C
1
dening function. If ) C
2
(p,q)
(1), then
) Dom(
(p,q)
)
if and only if
(, d)) = 0 and (, d)) = 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 (, d)) =
(, d)) = 0 on /1. Conversely, if (, d)) = (, d)) = 0, then )
Dom(

) and ) Dom(

) from integration by parts. Also, it is easy to see that


) and

) = ) are in Dom(

). Thus ) Dom(
(p,q)
).
If ) C
2
(p,q)
(1) Dom(
(p,q)
), we have

(p,q)
) = (

)).
A direct calculation, using (4.2.1) and (4.2.3), gives us that
(4.2.9)

) =

I,J
t

2
)
I,J
.
k
.
k
d.
I
d .
J
+ (1)
p

I,K
t

2
)
I,jK
.
j
.
k
d .
k
d.
I
d .
K
,
and
(4.2.10)

) = (1)
p1

I,K
t

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

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 ) satises
)
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 rst 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 Lopatinskis 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 be a C
2
dening function in
a neighborhood of 1 such that 1 = . [ (.) < 0 and [d[ = 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 C
2
(1) be a xed function. Let
T

(p,q)
= Dom(

) 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

(p,q)
if
and only if (, d))(.) = 0 for any . /1, a condition independent of . Thus
we have
T

(p,q)
= T
(p,q)
,
which is also independent of . Similarly, we also have
Dom(

) C

(p,q)
(1) = T

(p,q)
.
Let Q

be the form on T
(p,q)
dened by
Q

(), )) = |

)|
2

+|

)|
2

.
We shall rst prove the following basic a priori identity:
Proposition 4.3.1 (Morrey-Kohn-Hormander). Let 1 C
n
be a domain
with C
2
boundary /1 and be a C
2
dening function for 1 such that [d[ = 1 on
/1. Let C
2
(1). For any ) =

t
]I]=p,]J]=q
)
I,J
d.
I
d .
J
T
1
(p,q)
,
(4.3.1)
Q

(), )) = |

) |
2

+ |

) |
2

t
]I]=p,]K]=q1

i,j
_
D

.
i
.
j
)
I,iK

)
I,jK
c

d\
+

t
]I]=p,]J]=q

k
_
D

)
I,J
.
k

2
c

d\
+

t
]I]=p,]K]=q1

i,j
_
bD

.
i
.
j
)
I,iK

)
I,jK
c

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)

) = (1)
p1

I,K
t

j
)
I,jK
d.
I
d.
K
,
where

j
n = c


.
j
(c

n). Thus, setting 1


j
= , .
j
, we get
(4.3.4)
|)|
2

+ |

)|
2

I,J,L
t

j,
c
jJ
L
(1
j
()
I,J
), 1

()
I,L
))

I,K
t

j,k
(

j
)
I,jK
,

k
)
I,kK
)

,
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

+ |

)|
2

I,J
t

j
|1
j
)
I,J
|
2

I,K
t

j,k
(1
k
)
I,jK
, 1
j
)
I,kK
)

I,K
t

j,k
(

j
)
I,jK
,

k
)
I,kK
)

.
We now apply integration by parts to the last term in (4.3.4). Note that for each
n, C
2
(1),
(n,

j
)

= (1
j
n, )

+
_
bD

.
j
nc

do
and
[

j
, 1
k
]n =

j
1
k
n 1
k

j
n = n

2

.
j
.
k
.
Thus, we nd that
(4.3.6)
(

j
n,

k
)

= (1
k

j
n, )

+
_
bD

.
k
(

j
n)c

do
= (

j
1
k
n, )

+ ([

j
, 1
k
]n, )

+
_
bD

.
k
(

j
n)c

do
= (1
k
n, 1
j
)

+ (

2

.
j
.
k
n, )

+
_
bD

.
k
(

j
n)c

do
_
bD

.
j
(1
k
n)c

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 xed 1, 1, it follows that
(4.3.7)

j,k
(

j
)
I,jK
,

k
)
I,kK
)

j,k
(1
k
)
I,jK
, 1
j
)
I,kK
)

j,k
(

2

.
j
.
k
)
I,jK
, )
I,kK
)

j,k
_
bD

.
k
(

j
)
I,jK
)

)
I,kK
c

do

j,k
_
bD

.
j
_

.
k
)
I,jK
_

)
I,kK
c

do.
If ) T
1
(p,q)
, Lemma 4.2.1 and (4.2.6
t
) show that
(4.3.8)

.
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

)
I,kK

.
k
_
_

.
j
)
I,jK
_
_
= 0 on /1 for each 1, 1.
This implies
(4.3.9)

)
I,kK

.
j
)
I,jK
.
k
+

j
)
I,jK

)
I,kK

.
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 C
2
(1). Then T

(p,q)
is dense in Dom(

) Dom(

) in the
graph norm
) |)|

+|

)|

+|

)|

.
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(

) in the graph norm.


By this we mean that if ) Dom(

) Dom(

), one can construct a sequence


)
n
C

(p,q)
(1) such that )
n
),

)
n


) and

)
n

) in 1
2
(1, ). We rst
show that this can be done on a compact subset in 1 from the usual regularization
by convolution.
Let C

0
(C
n
) be a function such that 0,
_
d\ = 1, (.) depends only on
[.[ and vanishes when [.[ 1. We dene

(.) =
2n
(.,) for 0. Extending
) to be 0 outside 1, we dene for 0 and . C
n
,
)

(.) = )

(.)
=
_
)(.
t
)

(. .
t
)d\ (.
t
) =
_
)(. .
t
)(.
t
)d\ (.
t
),
where the convolution is performed on each component of ). In the rst integral
dening )

, we can dierentiate under the integral sign to show that )

is C

(C
n
).
From Youngs inequality for convolution, we have
| )

| | ) | .
Since )

) uniformly when ) C

0
(C
n
), a dense subset of 1
2
(C
n
), we have that
)

) in 1
2
(C
n
) for every ) 1
2
(C
n
).
Obviously, this implies that )

) in 1
2
(1, ).
70 L
2
Theory for on Pseudoconvex Domains
Let

be a sequence of small numbers with

0. For each

, we dene
1

= . 1[ (.) <

. Then 1

is a sequence of relatively compact open


subsets of 1 with union equal to 1. Using similar arguments as before, for any
rst order dierential operator 1
i
with constant coecients, if 1
i
) 1
2
(1, ), we
have
1
i
)

= 1
i
()

) = 1
i
)

1
i
) in 1
2
(1

, )
as 0. Since

= +
0
where
0
is an operator of degree 0, we have

)



)
and

) in 1
2
(1

, ) on 1

, where )

(p,q)
(1

).
To see that this can be done up to the boundary, we rst assume that the
domain 1 is star-shaped and 0 1 is a center. We approximate ) rst 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

) 1
2
(1). By regularizing
)

componentwise as before, we can nd a family of )


()
C

(p,q)
(1) dened by
(4.3.10) )
()
= )
_
.
1 +c
_

,
where

0 as c 0 and

is chosen suciently small. We have )


()
) in
1
2
(1, ),

)
()


) and

)
()

) in 1
2
(1, ). Thus, C

(p,q)
(1) is dense in the
graph norm when 1 is star-shaped. 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 suce in this step).
(ii). Compactly supported smooth forms (i.e., forms with coecients in C

0
(1))
are dense in Dom(

) in the graph norm


) |)|

+|

)|

.
We rst assume that = 0. Since

is the maximal closure (i.e., the domain of

contains all elements in C

(p,q)
(1)) of the Cauchy-Riemann 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

) 1
2
(C
n
) in the distribution sense.
In fact, for ) Dom(

), we have

) =

)
where

) = ) in 1 and

) = 0 in 1
c
. This can be checked from the denition of

, since for any C

(p,q1)
(C
n
),
(

),

)
C
n = (),

)
D
= (

), )
D
= (), )
D
= (

), )
C
n.
4.3 L
2
Existence Theorems for in Pseudoconvex Domains 71
Again, we can assume that 1 is star-shaped and 0 is a center. The general case can
be proved using a partition of unity. We rst approximate

) by

=

)
_
.
1 c
_
.
Then

)

has compact support in 1 and

) in 1
2
(C
n
). Regularizing

)

by convolution as before, we dene


(4.3.11) )
()
=

)
_
.
1 c
_

.
Then the )
()
are (j, )-forms with coecients in C

0
(1) such that )
()
) in
1
2
(1) and )
()
) in 1
2
(1). This proves (ii) when = 0. Again, in this step
we only require that the boundary be C
1
. The case for

can be proved similarly.


However, compactly supported smooth forms are not dense in Dom (

) in the
graph norm ) |)|

+|

)|

. Nevertheless, we have:
(iii). T

(p,q)
is dense in Dom(

) in the graph norm


) |)|

+|

)|

.
To prove (iii), we must use Friedrichs lemma in a more subtle way. From (i), it
suces to show that for any ) C

(p,q)
(1) that one can nd a sequence )
n
T

(p,q)
such that )
n
) in 1
2
(1, ) and

)
n


) in 1
2
(1, ). We may assume = 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 1l is star-shaped and ) is sup-
ported in l1. Let be a C
+1
dening function such that [d[ = 1 on /1. Shrink-
ing l if necessary, we can choose a special boundary chart (t
1
, t
2
, , t
2n1
, )
where (t
1
, t
2
, , t
2n1
), 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

= 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 are rst order dierential operators with variable coecients which
are in C

(1) when computed in the special frame n


1
, , n
n
. We write
) = )

+)

,
where
)

t
]I]=p,]J]=q, n/ J
)
I,J
n
I
n
J
,
)

t
]I]=p,]J]=q, nJ
)
I,J
n
I
n
J
.
72 L
2
Theory for on Pseudoconvex Domains
)

is the complex tangential part of ), and )

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 )

= 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) = (), p) +
_
bD
(

, d)), p)do,
where do is the surface measure of /1 and
(

, d)) =

()) =

) =

)

on /1.
Thus, when we do integration by parts for

), only the tangential part )

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 )

(From (4.2.7), it is
easy to see that the Cauchy data of ) for is the complex normal part )

).
We regularize only the complex normal part of ) and leave the complex tangential
part )

unchanged. Let

)

be the extension of )

to C
n
by setting

)

equal to zero
outside 1. We approximate

)

by the dilation and regularization by convolution


as in (4.3.11),
)

()
=

)

_
.
1 c
_

.
Thus, )

()
is smooth and supported in a compact subset in 1 l. By this, we
approximate )

by )

()
C

0
(1l) in the 1
2
norm. Furthermore, by extending

to be zero outside 1 l and denoting the extension by


, we have

in 1
2
(C
n
)
in the distribution sense. This follows from (4.3.12) since )

(1) and for any


p C

(p,q+1)
(C
n
),
(

)

, p)
C
n = (

, p)
D

_
bD

, p)do = (

, p)
C
n.
Since

is a rst order dierential operator with variable coecients, using the
arguments for (ii), but now applying Friedrichs lemma (see Appendix D), we have
(4.3.13)

)

()

in 1
2
(C
n
).
We set
)
()
= )

+)

()
.
It follows )
()
T

(p,q)
since each coecient 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 nd that

)
()
=

)

+

)

()


) in 1
2
(1) as c 0.
Thus, T

(p,q)
is dense in Dom(

) in the graph norm ) |)|

+|

)|

. This proves
(iii).
To nish the proof of the lemma, we assume that = 0. For any ) Dom(

)
Dom(

), we use a partition of unity and the same notation as before to regularize


) in each small star-shaped neighborhood near the boundary. We regularize the
complex tangential and normal part separately by setting

)
()
= )

()
+)

()
,
where )

()
is the regularization dened by (4.3.10) for each coecient in the com-
plex tangent component and )

()
is the regularization dened by (4.3.11) for each
coecient in the complex normal component. It follows that for suciently small
c 0, )

()
has coecients in C

0
(1) and )

()
has coecients in C

(1). Thus we
see that

)
()
T

(p,q)
,

)
()
) in 1
2
(1).
It follows from steps (i), (iii) and Friedrichss lemma that


)
()


) in 1
2
(1).
Also, from steps (i) and (ii), it follows that

)
()

) 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 = 0. For ,= 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 C
2
(1). We have for every ) 1o:() 1o:(

),
(4.3.14)

I,K
t

j,k
_
D

.
j
.
k
)
I,jK

)
I,kK
c

d\ | ) |
2

+ |

) |
2

.
Proof. From the assumption that 1 is pseudoconvex and has C
2
boundary, we have
for any ) T
1
(p,q)
,

i,j
_
bD

.
i
.
j
)
I,iK

)
I,jK
c

do 0,
since ) satises (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 nd n 1
2
(p,q1)
(1) such that n = ) and
(4.3.15)
_
D
[ n [
2
d\ c
2
_
D
[ ) [
2
d\,
where = sup
z,z

D
[. .
t
[ is the diameter of 1.
Proof. We rst prove the theorem for 1 with C
2
boundary. Without loss of gen-
erality, we may assume that 0 1. We shall choose = t[.[
2
for some positive
number t. From Proposition 4.3.3, we have for any p Dom() Dom(

),
(4.3.16) t
_
D
[ p [
2
c
t]z]
2
d\ | p |
2

+ |

p |
2

.
Since

2
= 0, we have
(4.3.17) (

) Ker(

) and (

) Ker(

).
It follows from (4.3.16) that for any p Dom(

) Ker(),
(4.3.18) t
_
D
[ p [
2
c
t]z]
2
d\ |

p |
2

.
Using Lemma 4.1.1, we have that (

) is closed in 1
2
(p,q)
(1, ). To show that
(4.3.19) (

) = 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)

[ C |

p |

, for all p Dom(

).
Using Lemma 4.1.1, (

) is also closed. From (4.1.1), we have


1
2
(p,q)
(1, ) = Ker(

) Ker(

= Ker(

) (

).
For any p
1
Dom(

) Ker(), using (4.3.18),


[ (), p
1
)

[ | ) |

| p
1
|

t
| ) |

p
1
|

.
If p
2
Dom(

) Ker()

, we have
(), p
2
)

= 0,
4.3 L
2
Existence Theorems for in Pseudoconvex Domains 75
since ) Ker(). For any p Dom(

), we write p = p
1
+ p
2
where p
1
Ker()
and p
2
Ker()

= (

) Ker(

). Thus, p
2
Dom(

) and

p
2
= 0. This
implies that p
1
Dom(

) and

p =

p
1
. Hence, we have for any p Dom(

),
[ (), p)

[ = [ (), p
1
)

t
| ) |

p
1
|

=
1

t
| ) |

p |

.
This proves the claim (4.3.20). Using the Hahn-Banach theorem and the Riesz
representation theorem applied to the antilinear functional

p (), p)

, there
exists n 1
2
(p,q1)
(1, ) such that for every p Dom(

),
(), p)

= (n,

p)

,
and
| n |

t
| ) |

.
This implies that n = ) in the distribution sense and n satises

_
D
[ n [
2
d\ c
t
2
_
D
[ n [
2
c
t]z]
2
d\

1
t
c
t
2
_
D
[ ) [
2
c
t]z]
2
d\

1
t
c
t
2
_
D
[ ) [
2
d\.
Since the function
1
t
c
t
2
achieves its minimum when t =
2
, we have

_
D
[ n [
2
d\ c
2
_
D
[ ) [
2
d\.
This proves the theorem when the boundary /1 is C
2
.
For a general pseudoconvex domain, from Denition 3.4.9, one can exhaust 1
by a sequence of pseudoconvex domains with C

boundary 1

. We write
1 =

=1
1

,
where each 1

is a bounded pseudoconvex domain with C

boundary and 1


1
+1
1 for each . Let

denote the diameter for 1

. On each 1

, there exists
a n

1
2
(p,q1)
(1

) such that n

= ) in 1

and

_
D
[ n

[
2
d\ c
2

_
D
[ ) [
2
d\ c
2
_
D
[ ) [
2
d\.
76 L
2
Theory for on Pseudoconvex Domains
We can choose a subsequence of n

, still denoted by n

, such that n

n weakly
in 1
2
(p,q1)
(1). Furthermore, n satises the estimate

_
D
[ n [
2
d\ liminf c
2

_
D
[ ) [
2
d\ c
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 nd n 1
2
(p,q1)
(1, loc)
such that n = ).
Proof. Since 1 is pseudoconvex, from Denition 3.4.9, there exists a C

strictly
plurisubharmonic exhaustion function for 1. For any ) 1
2
(p,q)
(1, loc), we can
choose a rapidly increasing convex function (t), t 1 such that (t) = 0 when
t 0 and ) 1
2
(p,q)
(1, ()). Let 1

= . 1[ (.) < , then


1 =

=1
1

,
where each 1

is a bounded pseudoconvex domain with C

boundary and 1


1
+1
1 for each . Since () is plurisubharmonic, the function = () +[.[
2
is strictly plurisubharmonic with
n

j,k=1

.
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

we
have for any p Dom() Dom(

),
| p |
2
(D)

_
D

I,K
t

j,k

.
j
.
k
p
I,jK
p
I,kK
c

d\
| p |
2
(D)
+ |

p |
2
(D)
.
Repeating the same argument as in Theorem 4.3.4, there exists a n

1
2
(p,q1)
(1

,
) such that

n

= ) in 1

and

_
D
[n

[
2
c

d\
_
D
[)[
2
c

d\
_
D
[)[
2
c

d\ < .
Taking a weak limit n of n

as , we have shown that there exists n such that

n = ) in 1 and

_
D
[n[
2
c

d\
_
D
[)[
2
c

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 self-adjoint.
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) = (
(p,q)
) Ker(
(p,q)
),
where (
(p,q)
) denotes the range of
(p,q)
. We shall show that (
(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 Ker(
(p,q)
), we have Dom(

) Dom(

) and
(,
(p,q)
) = |

|
2
+|

|
2
= 0.
Thus, Ker(
(p,q)
) Ker(

) Ker(

). On the other hand, if Ker(

) Ker(

),
then Dom(
(p,q)
) and
(p,q)
= 0. Thus, Ker(
(p,q)
) Ker(

) Ker(

) and
the rst equality in (4.4.2) is proved. To see that Ker(

) Ker(

) = 0, we note
that if Ker(

), from Theorem 4.3.4, there exists n 1


2
(p,q1)
(1) such that
=

n. If is also in Ker(

), we have
0 = (

n, n) = |

n|
2
and = 0. This proves (4.4.2).
We shall show that (
(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) (
(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,q1)

on Dom(

), 2 n.
(5) Let be the diameter of 1. The following estimates hold for any )
1
2
(p,q)
(1):
|
(p,q)
) |
c
2

| ) |,
|

(p,q)
) |

c
2

| ) |,
|

(p,q)
) |

c
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,q1)
(1) such that

n = ) and n satises the estimate (4.3.15). Thus, (

)
is closed in every degree and is equal to Ker(

). It follows from Lemma 4.1.1 that


(

) is closed also for every , and we have the following orthogonal decomposition:
(4.4.3) 1
2
(p,q)
(1) = Ker(

) (

) = (

) (

).
For every ) Dom(

) Dom(

), we have
) = )
1
)
2
where )
1
(

) and )
2
(

). 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
= c
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) = (
(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,q1)

(p,q)
on Dom(

) similarly.
To prove (5), we see from (4.4.8) that
|
(p,q)
) |
c
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)
)|

c
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 1
2
(p,q)
(1) such that

= 0, the (j, 1)-form
(4.4.9) n =

(p,q)

satises the equation



n = and the estimate
|n|
2

c
2

||
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,
=

(p,q)
+

(p,q)
.
Using (3) in Theorem 4.4.1, we have

(p,q)
=
(p,q+1)

= 0,
since

= 0. Thus we have =

(p,q)
. 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 dene
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) (
(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 be the diameter of 1. For any ) 1
2
(p,0)
(1),
|
(p,0)
) | c
2
| ) |,
|

(p,0)
) |

c | ) | .
Proof. Note that H
(p,0)
(1) = H
(p,0)
= Ker(
(p,0)
). We rst 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 ) (

).
Let =

), then 1
2
(p,1)
(1) since ) Dom(
(p,0)
). Using (4) in Theorem
4.4.1, we have that

(p,1)
is the unique solution satisfying

= and
Ker(

). Thus, = ). Applying Corollary 4.4.2, we have


|)|
2
c||
2
= c|

)|
2
= c(
(p,0)
), )) c|
(p,0)
)||)|,
where c = c
2
. This implies that
|)| c
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) = (
(p,0)
) H
(p,0)
=

(Dom(
(p,0)
)) H
(p,0)
.
For any (
(p,0)
), there is a unique
(p,0)
H
(p,0)
such that
(p,0)

(p,0)
=
. Extending
(p,0)
to 1
2
(p,0)
(1) by requiring
(p,0)
H
(p,0)
= 0,
(p,0)
satises (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)
| c
2
||.
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)

)|
c
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)
)| c
2
|)|.
Thus,
(p,0)
dened 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)
)| c
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 rst 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 nd n C

(p,q1)
(1) such that
n = ).
Proof. Let ) C

(p,q)
(1). From Theorem 4.3.5, there exists a strictly plurisub-
harmonic function C

(1) such that ) is in 1


2
(p,q)
(1, ) and there exists
1
2
(p,q1)
(1, ) with

= ) and
| |

| ) |

.
Repeating the same arguments as in Section 4.4, there exists a weighted

-
Neumann operator

such that for any ) 1


2
(p,q)
(1, ), we have
) =

) +

).
Since

) = 0, we have that ) =

). Setting n =

), we shall show that


n C

(p,q1)
(1). Since

n = n +
0
n = 0 for some zeroth order operator
0
, we
have
_

n = ),
n =
0
n 1
2
(1, loc).
However,

is an elliptic system. By this we mean that for any C

(p,q)
(1)
such that has compact support in 1, the following inequality holds:
(4.5.1) | |
1
C(|

| + | | + | |).
Inequality (4.5.1) is called Gardings inequality. To prove (4.5.1), we use Proposition
4.2.4 and (4.2.8) to get
(4.5.2) 4(|

|
2
+ | |
2
) = 4(, ) = (, ) = | |
2
,
where is the real Laplacian and is the gradient, both act on componentwise.
When = 0, (4.5.2) also holds since =

is also equal to
1
4
. Thus (4.5.1)
holds for any compactly supported smooth form . Let n = n where C

0
(1)
and dene n

= n

where and

are the same as in Lemma 4.3.2. It follows


4.5 Pseudoconvexity and the Levi Problem 83
that | n

| | n |,

n

=

n

and n

= n

. Substituting n

into (4.5.1),
we have
(4.5.3)
| n

|
1
C(|

n

| + | n

| + | n

|)
C(|

n | + | 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 dierential operator with constant
coecients, 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 unies domains of holomorphy, pseudoconvexity and ex-
istence theorems for the Cauchy-Riemann 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
nd n C

(p,q1)
(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 suces to prove this in a dense subset of /1.
Let .
0
be a boundary point such that there exists a complex (n1)-dimensional
hyperplane and .
0
/( 1). Such boundary points are dense in /1. To see
this, we note that for almost every boundary point .
0
, one can nd a ball 1 1
such that .
0
/(11). At such a .
0
, we obviously can nd a complex hyperplane
passing through the center of the ball and .
0
. It is easy to see that .
0
/(1).
By a linear transformation, we may assume that .
0
= 0 and
0
= 1 .
n
= 0 is
nonempty.
We shall show that on
0
, (3) is fullled. Let ) be a smooth

-closed (j, )-form
on
0
, where 0 j n1, 1 n1. We claim that ) can be extended to be a
smooth

-closed form in 1. We rst extend ) to

) in 1 such that

) C

(1) and

)(.) =

)(.
1
, , .
n
) = )(.
t
, 0) in an open neighborhood of
0
. This can be done
as follows: Let : 1 C
n1
be the projection such that (.) = (.
1
, , .
n1
, 0).
Then the set 1
0
= 1
1
(
0
) is a closed subset of 1. Since
0
and 1
0
are closed
(with respect to 1) disjoint subsets of 1, using Urysohns lemma, we see that there
exists a function C

(1) such that = 1 in a neighborhood of


0
and = 0
in a neighborhood of 1
0
. Then we can choose our

) =

)(.
t
), where

is the
pull-back 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 right-hand side of (4.5.4) is

-closed and is in C

(p,q+1)
(1), since

) is

-closed in a neighborhood of
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

0
. Thus, any

-closed form ) on
0
can be extended to a

closed form 1 on 1.
This is also true for = 0.
From (3), we can nd l(.) C

(p,q1)
(1) such that

l = 1 in 1. Restricting
l to
0
, we have shown that (3) is fullled on
0
.
By the induction hypothesis,
0
is a domain of holomorphy. Hence, there exists
a holomorphic function )(.
t
) = )(.
1
, , .
n1
) such that ) is singular at 0. Since

) = 0 in
0
, repeating the same argument above for = 0, there exists 1 in 1
such that

1 = 0 in 1 and 1 = ) on
0
. 1(.) is holomorphic in 1 and is equal to
)(.
t
, 0) on
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 Cauchy-Riemann equations. This approach generalizes the
Hodge-de Rham theorem from compact manifolds to complex manifolds with bound-
aries. The basic a priori estimates were rst 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 Kohns 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. Hormander [Hor
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. Hormander [Hor 3]. Theorem 4.3.4 is a
special case of Theorem 2.2.3 in [Hor 3] where the precise bounds are given. The
density lemma 4.3.2 was also proved in [Hor 3] in a much more general setting. Using
three dierent weight functions which are singular near the boundary, L. Hormander
[Hor 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 [Hor 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 . Let be a C
2
dening function for . The

-Neumann problem for (j, )-forms, 0 j n, 1 n, in is the boundary


value problem:
(5.0.1)
_

_
n = ) in ,

n = 0 on /,



n = 0 on /,
where n, ) are (j, )-forms, =

+

, denotes the interior product of forms.


On any bounded pseudoconvex domain in C
n
, we have derived the following
estimates: for any ) Dom(

) Dom(

),
(5.0.2) | ) |
2


c
2

(|

) |
2

+ |

) |
2

),
where is the diameter of (see (4.4.6)). It follows from Theorem 4.4.1 that the

-Neumann operator
(p,q)
exists in , 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 satises Gardings inequality in the interior, but
not near the boundary. However, under the assumption of strong pseudoconvexity,
we will show that it satises subelliptic 1,2-estimates near the boundary.
In Section 5.1, we prove that the following subelliptic 1,2-estimate holds on a
strongly pseudoconvex domain in C
n
: for any ) Dom(

) Dom(

),
(5.0.3) | ) |
2
1
2
()
C(|

) |
2

+ |

) |
2

),
where | |1
2
()
is the Sobolev norm in the Sobolev space \
1/2
(p,q)
().
The regularity of the

-Neumann operator in other Sobolev spaces when the
boundary / 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 Newlander-Nirenberg 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 1,2-estimate for the

-Neumann
operator when is strongly pseudoconvex with C
2
boundary. We shall use
instead of
(p,q)
to simplify the notation. We also use 1
2
(), \
s
() and \
s
(, loc)
to denote the spaces 1
2
(p,q)
(), \
s
(p,q)
() and \
s
(p,q)
(, loc) respectively, where
\
s
() is the Sobolev space, : 1. (See Appendix A for its denition and basic
properties.) The norm in \
s
is denoted by | |
s()
.
We rst observe that the rst order system

: C

(p,q)
() C

(p,q+1)
() C

(p,q1)
()
is elliptic in the interior. This means that we have Gardings inequality in the
interior.
Proposition 5.1.1. Let be a bounded domain in C
n
and C

0
(). For any
(j, )-form ) 1
2
(p,q)
() such that

) 1
2
(p,q+1)
() and ) 1
2
(p,q1)
(), where
0 j n and 0 n, we have the following estimates:
(5.1.1) | ) |
2
1()
C(|

) |
2

+ | ) |
2

+ | ) |
2

),
where C is a constant depending only on but not on ).
Proof. Using the basic estimates proved in Proposition 4.3.1, when = 0, we have
for any C

0
() and ) C

(p,q)
(), 0 j n and 1 n,
(5.1.2)

I,J
t

k
|
()
I,J
)
.
k
|
2

= (|

()) |
2

+ | ()) |
2

)
C(|

) |
2

+ | ) |
2

+ | ) |
2

).
(5.1.2) also holds trivially for = 0 (in this case, ) = 0). But
(5.1.3) |
()
I,J
)
.
k
|
2

= |
()
I,J
)
.
k
|
2

from integration by parts. We have for any smooth ),


| ) |
2
1()
C
_
_

I,J
t

k
|
()
I,J
)
.
k
|
2

I,J
t

k
|
()
I,J
)
.
k
|
2

+ | ) |
2

_
_
C(|

) |
2

+ | ) |
2

+ | ) |
2

).
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
(, loc). The diculty for the

-Neumann
problem is only on the boundary. The following subelliptic 1,2-estimates for a
strongly pseudoconvex domain are of fundamental importance:
88 The -Neumann Problem on Strongly Pseudoconvex Manifolds
Theorem 5.1.2. Let 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
()
C(|

) |
2

+ |

) |
2

),
where C is independent of ).
Theorem 5.1.3. Let 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
satises the estimates:
(5.1.5) | ) |
2
1
2
()
C | ) |
2

1
2
()
, ) 1
2
(p,q)
(),
where C is independent of ). can be extended as a bounded operator from
\
1/2
(p,q)
() into \
1/2
(p,q)
(). In particular, is a compact operator on 1
2
(p,q)
().
We divide the proof of Theorem 5.1.2 into several lemmas. From Lemma 4.3.2,
we have that C
1
(p,q)
() 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 Morrey-Kohn:
Lemma 5.1.4. Let be a bounded strongly pseudoconvex domain in C
n
with C
2
boundary /. There exists a constant C 0, such that for any ) C
1
(p,q)
()
Dom(

) = T
1
(p,q)
,
(5.1.6)
_
b
[)[
2
do C(|

) |
2

+ | ) |
2

),
where do is the surface element on / 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 be a C
2
dening function for normalized
such that [d[ = 1 on /. Following Proposition 4.3.1 with = 0, we have for each
) C
1
(p,q)
() Dom(

),
(5.1.7)
|

) |
2

+ | ) |
2

I,J
t

k
|
)
I,J
.
k
|
2

I,K
t
n

i,j=1
_
b

.
i
.
j
)
I,iK

)
I,jK
do,
where 1 is an increasing multiindex and [1[ = 1.
Since / is strongly pseudoconvex with C
2
boundary, there exists C
0
0 such
that for any . /,
(5.1.8)
n

i,j=1

.
i
.
j
o
i
o
j
C
0
[o[
2
, if
n

i=1
o
i

.
i
= 0.
5.1 Subelliptic Estimates for the

-Neumann Operator 89
Since ) Dom(

) C
1
(p,q)
(), we have from Lemma 4.2.1,
n

j=1
)
I,jK

.
j
= 0 on / for each 1, 1.
Substituting (5.1.8) into (5.1.7), we have
|

) |
2

+ | ) |
2

I,J
t

k
|
)
I,J
.
k
|
2

+ C
0
_
b
[)[
2
do.
This proves the lemma.
Lemma 5.1.5. Let be a bounded domain in C
n
with C
2
boundary / and let
be a C
2
dening function for . There exists a constant C 0 such that for any
) C
2
()
(5.1.9)
_

[[[)[
2
d\ C
__
b
[)[
2
do +
_

[)[
2
d\
_
+ Re
_

())

)d\,
where do is the surface element on / and C 0 is independent of ).
Proof. By Greens theorem, for any n, C
2
(),
(5.1.10)
_

(n n)d\ =
_
b
_
n

n

n
n
_
do,
where ,n is the directional derivative along the outward normal. Let n =
and = [)[
2
,2, then = Re(

))) +[)[
2
. We have from (5.1.10)
(5.1.11)
_

[)[
2
d\ + Re
_

())

)d\ +
_

[)[
2
2
d\ =
_
b
[)[
2
2

n
do.
Since is in C
2
(), there exists C 0 such that [[ C in and [,n[ C
on /. This implies that
(5.1.12)
_

[[[)[
2
d\ C
__
b
[)[
2
do +
_

[)[
2
d\
_
+ Re
_

())

)d\,
where C is independent of ). This proves the Lemma.
Lemma 5.1.6. Let be a bounded domain in C
n
with C
2
boundary /. There
exists a constant C 0 such that for any ) C
1
(p,q)
(),
(5.1.13) | ) |
2
1
2
()
C
__
b
[)[
2
do+ |

) |
2

+ | ) |
2

+ | ) |
2

_
,
where C is independent of ).
90 The -Neumann Problem on Strongly Pseudoconvex Manifolds
Proof. When ) C
2
(p,q)
(), one sees from Proposition 4.2.4 that ) =
1
4
),
where =

+

and is dened componentwise. Applying Lemma 5.1.5 to


each component of ), we get
(5.1.14)
_

[[[)[
2
d\ C
__
b
[)[
2
do +
_

[)[
2
d\
_
+ 4[(), ))

[.
Since ) C
1
(p,q)
(), we have
(5.1.15) | ) |
2
1
2
()
C
__

[[[)[
2
d\ +
_

[)[
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)
[ (), ))

[ [ (

()), ))

[ + [ ([

, ]), ))

[
+ [ ((

)), ))

[ + [ ([, ]

), ))

[
[ (), ))

[ + [ (

),

))

[
+ sup

[[(|

) |

+ | ) |

) | ) |

C(|

) |
2

+ | ) |
2

+ | ) |
2

).
Combining (5.1.14)-(5.1.16), we see that (5.1.13) holds for any ) C
2
(p,q)
(). An
approximation argument shows that (5.1.13) holds for any ) C
1
(p,q)
() since C
2
()
is dense in C
1
().
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)
() Dom(

),
(5.1.17)
_
b
[)[
2
do C(|

) |
2

+ | ) |
2

).
Using (4.4.6), we also get
(5.1.18) | ) |
2


c
2

(|

) |
2

+ | ) |
2

).
Thus, Lemmas 5.1.4 and 5.1.6 show that (5.1.4) holds for all forms ) C
1
(p,q)
()
Dom(

) = T
1
(p,q)
. Since T
1
(p,q)
is dense in Dom(

) Dom(

) in the graph norm


|

) |

+ |

) |

from Lemma 4.3.2, Theorem 5.1.2 is proved.


Proof of Theorem 5.1.3. By the denition of the space \
1/2
() (see Appendix
A), we have
(5.1.19) [(/, p)

[ | / |1
2
()
| p |

1
2
()
5.1 Subelliptic Estimates for the

-Neumann Operator 91
for any / \
1/2
() and p \
1/2
(). There exists a constant C 0 such that
for any ) 1
2
(p,q)
() Dom(), 0 j n and 1 n 1,
(5.1.20)
| ) |
2
1
2
()
C(|

) |
2

+ |

) |
2

)
= C(), ))

C | ) |

1
2
()
| ) |1
2
()
,
where C is independent of ). Substituting ) into (5.1.20), we have
| ) |1
2
()
C | ) |

1
2
()
= C | ) |

1
2
()
.
Thus, can be extended as a bounded operator from \
1/2
() to \
1/2
(). Using
the Rellich lemma (see Theorem A.8 in the Appendix), is a compact operator on
\
1/2
() and 1
2
(p,q)
(). This proves Theorem 5.1.3.
Corollary 5.1.7. Let and ) be as in Theorem 5.1.2. Then

and

are
compact operators on 1
2
(p,q)
(). Moreover, the following estimates hold: there exists
a C 0 such that for any ) 1
2
(p,q)
(),
(5.1.21) |

) |1
2
()
C | ) |

, 1 n,
(5.1.22) |

) |1
2
()
C | ) |

, 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
()
C(|

) |
2

+ |

) |
2

) C | ) |
2

.
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 ,
n = 0 on /,
where n and ) are (j, n)-forms. In this case, we have Gardings inequality in place
of (5.1.4):
| ) |
2
1()
C |

) |
2

, ) Dom(

),
where C is independent of ). Since for any ) 1
2
(p,n)
(), ) Dom(

) and we
obtain
| ) |
2
1()
C |

) |
2

= C(

), ))

= C(), ))

C | ) |
1()
| ) |
1()
.
92 The -Neumann Problem on Strongly Pseudoconvex Manifolds

(p,n)
can be extended as a bounded operator from \
1
() into \
1
0
() (see Ap-
pendix A for its denition) for any bounded domain with C
2
boundary.
5.2 Boundary Regularity for and

In this section, we assume that is strongly pseudoconvex and its boundary


/ is C

; i.e., there exists a C

dening function . We normalize such that


[d[ = 1 on /. First we derive estimates for the

-Neumann operator in Sobolev
spaces for : 0. As observed in Proposition 5.1.1, the system

is elliptic in
the interior. The interior regularity follows from the usual elliptic theory (see e.g.
Lions-Magenes [LiMa 1], Bers-John-Schechter [BJS 1] or Treves [Tre 1]). Based on
Gardings inequality (5.1.1), we have for every C

0
() and : 0,
) \
s+2
(p,q)
(), for any ) \
s
(p,q)
().
The main result in this section is to prove the following estimates on boundary
regularity for .
Theorem 5.2.1. Let C
n
be a bounded strongly pseudoconvex domain with
C

boundary. The

-Neumann operator is a bounded operator from \
s
(p,q)
()
to \
s+1
(p,q)
() where :
1
2
, 0 j n and 1 n 1, and satises the
estimate: there exists a constant C
s
such that for any ) \
s
(p,q)
(),
(5.2.1) | ) |
2
s+1()
C
s
| ) |
2
s()
,
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
2n1
,
such that t
1
, , t
2n1
restricted to / are coordinates for /. Let 1
tj
= ,t
j
,
, = 1, , 2n 1, and 1

= ,. Thus 1
tj
s are the tangential derivatives
on /, and 1

is the normal derivative. For a multiindex = (


1
, ,
2n1
),
where each
i
is a nonnegative integer, 1

t
denotes the product of 1
tj
s with order
[[ =
1
+ +
2n1
, i.e.,
1

t
= 1
1
t1
1
2n1
t2n1
.
For any n C

0
(l ), we dene the tangential Fourier transform for n in a
special boundary chart by
n(, ) =
_
R
2n1
c
it,)
n(t, )dt,
where = (
1
, ,
2n1
) and t, ) = t
1

1
+ + t
2n1

2n1
. For each xed
c < 0, we dene

= . [ (.) = c and set


| n(, c) |
2
s()
=
_
R
2n1
(1 +[[
2
)
s
[ n(, c)[
2
d.
5.2 Boundary Regularity for N and

N 93
We dene the tangential Sobolev norms [[[ [[[
s
by
[[[n[[[
2
s
=
_
R
2n1
_
0

(1 +[[
2
)
s
[ n(, )[
2
dd
=
_
0

| n(, ) |
2
s()
d.
As usual, tangential norms for forms are dened as the sum of the norms of the
components. The operator
s
t
for any : 1 is given by

s
t
n(t, ) = (2)
12n
_
R
2n1
c
it,)
(1 +[[
2
)
s
2
n(, )d.
Using this notation and Plancherels theorem (see Theorem A.5 in the Appendix),
we have
[[[n[[[
s
= |
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 ) under the [[[ [[[
s
norm. Then the following hold:
(1) When : is a nonnegative integer, we have
(5.2.2 i) [[[n[[[
s

0]]s
| 1

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

t
)[[[
s
C[[[)[[[
k+s
for [[ = /,
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 = (2)
12n
.
(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
]

= \
s1(1)+s2
t
for any 0 < < 1.
Properties (1) and (2) are easily checked from Plancherels Theorem (see Theorem
A.5 in the Appendix). (3) follows immediately from the denition. 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 dierential operators and 1
of order /
1
and /
2
, /
1
and /
2
are nonnegative integers, the commutator [, 1] =
1 1 is an operator of order /
1
+/
2
1.
The proof follows directly from the denition. We shall denote
1
t
n =
t
n and
dene for any real :,
[[[1n[[[
2
s
= [[[1

n[[[
2
s
+[[[
t
n[[[
2
s
= [[[1

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 xing a partition of unity, we can also dene [[[ [[[
s(

)
for some tubular
neighborhood

= . [ < (.) < 0. Let l


i
, i = 1, 2, , 1, be boundary
coordinate patches such that each l
i
/ ,= ,
K
i=1
l
i
covers

and there exists a


special boundary chart on each l
i
. We choose
i
C

0
(l
i
), i = 1, , 1, such
that
K

i=1

2
i
= 1 in

.
For each xed c, the norm | |
s()
is dened by a partition of unity.
We set
[[[n[[[
2
s(

)
=
_
0

| n(, ) |
2
s()
d.
We choose a special boundary frame such that n
1
, , n
n
is an orthonormal basis
for (1,0)-forms with n
n
= in a boundary patch l as before. Written in this
basis, a smooth (j, )-form supported in l 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 /, whenever n J,
where T
(p,q)
= C

(p,q)
()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 and ) is
expressed in the special boundary frame. Let T be a rst order tangential dierential
operator with smooth coecients acting componentwise. Then T) T
(p,q)
.
In order to obtain estimates for the

-Neumann operator on the boundary, our
rst observation is that when ) satises 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 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 . The following conditions are equivalent:
(a) | ) |1
2
()
< .
(b) [[[)[[[ 1
2
()
< .
(c) [[[1)[[[

1
2
()
< .
(d) | ) |
L
2
(b)
< .
Proof. It is obvious from the denition that (a) implies (b).
From the density lemma 4.3.2, we can assume ) C
1
(l ) and the other cases
follow by approximation. To show that (b) implies (c), we note that

is elliptic.
We can express 1

) by the sum of the components of



), ) and the tangential
derivatives of ). Thus
[[[1

)[[[

1
2
C([[[

)[[[

1
2
+[[[)[[[

1
2
+[[[
t
)[[[

1
2
)
C(|

) | + | ) | + [[[)[[[ 1
2
).
To see that (c) implies (d), we use
[

)(, 0) [
2
=
_
0

[

)(, ) [
2
d = Re
_
0

21


)(, )

)(, )d

_
0

[

)(, ) [
2
d +
1

_
0

[ 1


)(, ) [
2
d,
for any 0. Choosing = (1 + [[
2
)
1/2
and integrating over 1
2n1
, we have
from Plancherels theorem,
| ) |
2
L
2
(b)
=
_
1
2
_
2n1
_
R
2n1
[

)(, 0) [
2
d
C
__
R
2n1
_
0

(1 +[[
2
)
1
2
[

)(, ) [
2
dd
_
+C
__
R
2n1
_
0

(1 +[[
2
)

1
2
[ 1


)(, ) [
2
dd
_
C [[[1)[[[
2

1
2
()
.
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
()
C
__
b
[)[
2
do+ |

) |
2

+ | ) |
2

+ | ) |
2

_
.
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 rst 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 be a bounded pseudoconvex domain with C

boundary and
let be a C

dening function. Choose 0, such that the tangential norms are


dened on the xed tubular neighborhood

= . [ (.) . Then there


exists a constant C
k
, / = 1, 2, such that for any ) Dom() C

(p,q)
(),
(5.2.3) | ) |
k()
C
k
(| ) |
k1()
+ [[[)[[[
k(

)
),
where C
k
is independent of ).
Proof. We use induction on / for / = 1, 2, . Since
k

m=0
[[[1
m

)[[[
km
| ) |
k
,
(5.2.3) is proved if we can show
(5.2.4)
k

m=0
[[[1
m

)[[[
km
C
k
(| ) |
k1
+ [[[)[[[
k
).
We rst prove (5.2.4) under the assumption that ) is supported in 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

) by
the components of

), ) and the tangential derivatives of ). We have
| 1

) |
2
C
_
|

) |
2
+ | ) |
2
+
2n1

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

)[[[
2
k1
C( [[[

)[[[
2
k1
+[[[)[[[
2
k1
+
2n1

i=1
[[[1
ti
)[[[
2
k1
)
C( [[[

)[[[
2
k1
+[[[)[[[
2
k1
+[[[)[[[
2
k
).
Notice that
[[[

)[[[
2
k1
+[[[)[[[
2
k1
= |
k1
t

) |
2
+ |
k1
t
) |
2
C
_
_

0]]k1
| 1

t

) |
2
+

0]]k1
| 1

t
) |
2
_
_
.
For any nonnegative integer /, let T
k
denote any tangential dierential operator of
the form 1

t
, where [[ = /. Using Lemma 5.2.2, we nd 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
k1

), T
k1

)) + (T
k1
), T
k1
))
= (

T
k1
), T
k1

)) + (
_
T
k1
,

), T
k1

))
+ (T
k1
), T
k1
)) + (
_
T
k1
,

), T
k1
))
= (T
k1
), T
k1

)) + (
_
T
k1
,

), T
k1

))
+ (T
k1
),

T
k1
)) + (
_
T
k1
,

), T
k1
))
= (T
k1
), T
k1

)) + (
_
T
k1
,

), T
k1

))
+ (T
k1
), T
k1

)) + (
_
T
k1
,

), T
k1
))
+ (T
k1
),
_
, T
k1

)) + (T
k1
),
_

, T
k1

))
= (T
k1
), T
k1
)) ++O(| ) |
k1
([[[

)[[[
k1
+[[[)[[[
k1
)),
where
= (T
k1
),
_
, T
k1

)) + (T
k1
),
_

, T
k1

)).
The term (T
k1
),
_
, T
k1

)) in can be estimated by
(5.2.7)
(T
k1
),
_
, T
k1

))
= (T
k1
),

_
, T
k1

)) + (T
k1
),
__
, T
k1

))
= (T
k1
),
_
, T
k1

)) + (T
k1
),
__
, T
k1

))
= (T
k1
),
_
, T
k1

)) + (
_
, T
k1

),
_
, T
k1

))
+ (T
k1
),
__
, T
k1

)).
Similarly, one can estimate the term (T
k1
),
_

, T
k1

)) in . Thus [[ can be
estimated by
(5.2.8) C(| ) |
k1
[[[

)[[[
k1
+ | ) |
k1
[[[)[[[
k1
+ | ) |
2
k1
).
If we apply (5.2.6), (5.2.7) to each term of the form T
]]
= 1

t
, where 0 [[ /1,
we can conclude that

0]]k1
(| 1

t

) |
2
+ | 1

t
) |
2
)
C(|
k1
t
) ||
k1
t
) | + | ) |
k1
([[[

)[[[
k1
+[[[)[[[
k1
) + | ) |
2
k1
).
Using the inequality that o/ co
2
+(1,c)/
2
for any c 0, we see from (5.2.5) that
[[[

)[[[
2
k1
+[[[)[[[
2
k1
C(|
k1
t
) ||
k1
t
) |
+ | ) |
k1
([[[

)[[[
k1
+[[[)[[[
k1
) + | ) |
2
k1
)
C[[[)[[[
k1
[[[)[[[
k1
+c([[[

)[[[
2
k1
+[[[)[[[
2
k1
) +C

| ) |
2
k1
.
98 The -Neumann Problem on Strongly Pseudoconvex Manifolds
Choosing c suciently small, the induction hypothesis yields
(5.2.9) [[[

)[[[
2
k1
+[[[)[[[
2
k1
C(| ) |
2
k1
+ [[[)[[[
2
k1
).
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

) by the sum of the components of terms of the form


), 1

1
tj
), 1
ti
1
tj
), i, , = 1, , 2n 1,
and lower order terms. If : 2, dierentiation shows that one can express 1
m

)
by the sum of terms of the form
1
m2

), 1
m1

1
tj
), 1
m2

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

)[[[
km
C( [[[1
m2

)[[[
km
+
2n1

i=1
[[[1
m1

1
ti
)[[[
km
+
2n1

i,j=1
[[[1
m2

1
ti
1
tj
)[[[
km
+ | ) |
k1
)
C(| ) |
k2
+ [[[1
m1

)[[[
km+1
+ [[[1
m2

)[[[
km+2
+ | ) |
k1
)
C(| ) |
k2
+ | ) |
k1
+ [[[)[[[
k
+ | ) |
k1
)
C(| ) |
k1
+ [[[)[[[
k
).
Thus, (5.2.4) holds for all : /. This nishes 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 C

( l),
we have that
| ) |
2
1
C( |

()) |
2
+ | ()) |
2
+ [[[)[[[
2
1(

)
)
C( |

) |
2
+ | ) |
2
+ | ) |
2
+ [[[)[[[
2
1(

)
)
C((
2
), ))

+c(|

()) |
2
+ | ()) |
2
)
+C

| ) |
2
+ [[[)[[[
2
1(

)
).
If we choose c suciently small, then the term
c(|

()) |
2
+ | ()) |
2
) c | ) |
2
1
can be absorbed by the left-hand side. It follows that
| ) |
2
1
C(| ) || ) | + | ) |
2
+ [[[)[[[
2
1(

)
)
C(| ) |
2
+ | ) |
2
+ [[[)[[[
2
1(

)
)
C(| ) |
2
+ [[[)[[[
2
1(

)
).
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
0
C

0
(), using Proposition 5.1.1, one sees that
|
0
) |
2
1
C | ) |
2
.
Summing over a partition of unity
i
, i = 0, , 1 such that
0
C

0
(), each
i
is supported in a boundary coordinate patch and

K
i=0

2
i
= 1, we have proved
| ) |
2
1
C(| ) |
2
+ [[[)[[[
2
1(

)
).
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 C
n
be a bounded strongly pseudoconvex domain with
C

dening function. Choose 0, such that the tangential norms are dened on
the xed tubular neighborhood

= . [ (.) . For each / = 0, 1, 2, ,


there exists a constant C
k
such that for any ) Dom() C

(p,q)
(),
(5.2.11) [[[)[[[
2
k+
1
2
(

)
C
k
| ) |
2
k
1
2
()
,
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 rst assume that
) is supported in l and written in the special frame as before. Let T
k
be a /th
order tangential dierential operator of the form 1

t
where [[ = /. 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
+ | T
k
) |
2
).
Using arguments similar to those in (5.2.6) and (5.2.7), it follows that
(5.2.14)
|

T
k
) |
2
+ | T
k
) |
2
= | T
k

) |
2
+ | T
k
) |
2
+ O(| ) |
2
k
)
= (T
k
), T
k
)) +O(| ) |
k
[[[

)[[[
k
+ | ) |
k
[[[)[[[
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
+[[[)[[[
k
C

]]k
(| 1

t

) | + | 1

t
) |)
C

]]k
(|

1

t
) | + | 1

t
) |) +O(| ) |
k
),
it follows from from (5.2.14) that
(5.2.16)

]]k
(|

1

t
) |
2
+ | 1

t
) |
2
) C
_

]]k
(1

t
), 1

t
)) + | ) |
2
k
_
.
Combining (5.2.13)-(5.2.16) and summing up all the tangential derivatives of the
form 1

t
, where [[ /, we deduce that
(5.2.17)
[[[)[[[
2
k+
1
2
C
_

]]k
[[[1

t
)[[[ 1
2
[[[1

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
+ | ) |
k1
)
C(c
t
[[[)[[[
k+
1
2
+C

[[[)[[[ 1
2
+ | ) |
k1
).
Choosing rst c and then c
t
suciently 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 ,
t
C

0
(l)
and
t
= 1 on the support of . We have
[[[T
k
)[[[
2
1
2
C( [[[T
k
)[[[
2
1
2
+[[[
t
)[[[
2
k
1
2
)
C(|

T
k
) |
2
+ | T
k
) |
2
+ [[[
t
)[[[
2
k
1
2
).
Repeating the previous argument with T
k
substituted for T
k
, we see that
[[[)[[[
2
k+
1
2
c[[[
t
)[[[
2
k+
1
2
+C

[[[)[[[
2
k
1
2
+C |
t
) |
2
k
.
5.2 Boundary Regularity for N and

N 101
Now for any
0
C

0
(), we already know that
|
0
) |
k+
3
2
C | ) |
k
1
2
.
Summing over a partition of unity
i
, i = 1, , 1 for the tubular neighborhood

, yields that for some


0
C

0
(),
[[[)[[[
2
k+
1
2
(

)
C(| ) |
2
k
1
2
+ |
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
()
C | ) |

1
2
()
, for any ) \

1
2
(p,q)
().
We shall prove the theorem for : = / when / N. Since C

(p,q)
() is dense in
\
s
(p,q)
(), it suces to prove the following estimates:
(5.2.21) | ) |
s+
1
2
()
C | ) |
s
1
2
()
, for any ) C

(p,q)
().
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 dened by
Q(p, p) = |

p|
2
+|

p|
2
, p Dom(

) Dom(

).
We dene
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 c
2
, where is the diameter
of ). Thus, for any ) 1
2
(p,q)
(), 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

satises Gardings
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)
() if
) C

(p,q)
(). Applying the a priori estimates (5.2.11) to the form

), we get
(5.2.24) [[[

)[[[
k+
1
2
(

)
C
k
| ) |
k
1
2
()
,
where C
k
is independent of c. The interpolation theorem for the operator

on
Sobolev spaces \
s
() and tangential Sobolev spaces \
s
t
(

) (see Theorem B.3 in


the Appendix) gives
[[[

)[[[
k(

)
C | ) |
k1()
for / = 1, 2, .
Repeating the argument of Lemma 5.2.4, we obtain
|

) |
k()
C | ) |
k1()
for / = 1, 2, .
Thus, a subsequence of

) converges weakly in \
k
(p,q)
() to some element
\
k
(p,q)
(). We claim that = ).
For any p T
(p,q)
,
(), p) = Q(), p) = Q

), p).
It follows from the denition 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 Q-norm. But (a subsequence of)

)
weakly in the \
k
norm. Therefore, we must have
) =
and
| ) |
k
liminf |

) |
k
C | ) |
k1
, / = 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)
().
This proves Theorem 5.2.1.
5.2 Boundary Regularity for N and

N 103
Theorem 5.2.6. Let C
n
be a bounded strongly pseudoconvex domain with
C

boundary. Then

and

are bounded operators from \
s
(p,q)
() into
\
s+1/2
(p,q1)
() and \
s+1/2
(p,q+1)
() respectively, where : 1,2, 0 j n and 1
n 1. There exists a constant C
s
such that for any ) \
s
(p,q)
(),
(5.2.25) |

) |
2
s+
1
2
()
+ |

) |
2
s+
1
2
()
C
s
| ) |
2
s()
,
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 be a cut-o function. If is supported in a boundary
coordinate patch, let T
k
denote the same /th order tangential dierential operator
of the form 1

t
, [[ = /. 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
k1
+ [[[)[[[
2
k
1
2
),
if c in (5.2.28) is chosen suciently small.
Again, using the fact

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 C
n
be a bounded strongly pseudoconvex domain with
C

boundary and 0 j n and 0 n 1. Then maps C

(p,q)
() into
C

(p,q)
(). In particular, the Bergman projection 1 maps C

() into C

(). Also
if ) C

(p,q)
() and

)=0, the canonical solution n =

) C

(p,q1)
().
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) = )


) = )

).
In Chapter 6, we prove a more precise result for the Bergman projection. In fact,
1 preserve \
s
() for all : 0 (see Theorem 6.2.2).
5.3 Function Theory on Manifolds
Let ` be a complex manifold of dimension n (For the denition of complex
manifolds, see Chapter 1). The decomposition of dierential forms into forms
of type (j, ), the denition of the

operator and the denition 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 denite Hermitian matrix with C

coecients. The existence


of a Hermitian metric is trivial locally and is proved globally by a partition of unity.
We x a Hermitian metric in all that follows. This induces an inner product in
C

(p,q)
(`) for each j `. If , C

(p,q)
(`), this inner product is denoted by
, ). We have the following denition:
Denition 5.3.1. Let j ` and C
2
(`). If 1 T
1,0
p
(`), the complex
Hessian of at j is dened to be the Hermitian form
1 (

)
p
(1 1).
The function is called plurisubharmonic at j if the complex Hessian is positive
semi-denite. is called strictly plurisubharmonic at j if the complex Hessian is
positive denite.
Let be an open subset in ` whose closure is compact in `, i.e., is relatively
compact in ` and denoted by `. is called a complex manifold with
C
k
boundary / if there exists a neighborhood \ of and a real-valued function
C
k
(\ ) such that = . \ [ < 0, 0 in \ and [d[ ,= 0 on /. Let
CT
p
(/) be the complexied tangent bundle of / at j.
5.3 Function Theory on Manifolds 105
Denition 5.3.2. Let be a complex manifold with C
2
boundary and be a C
2
dening function. is called pseudoconvex (strictly pseudoconvex) if for each j
/, the restriction of the complex Hessian of to T
1,0
p
(`) CT
p
(/) is positive
semi-denite (positive denite).
In local coordinates, by the usual Gram-Schmidt orthogonalization process we
can choose an orthonormal basis n
1
, , n
n
for (1,0)-forms locally on a suciently
small neighborhood l such that n
i
, n
k
) =
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 rst order linear dierential 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 coecients 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 C
2
is plurisubharmonic (strictly plurisubharmonic) if the form
n

j,k=1

jk
o
j
o
k
, o = (o
1
, , o
n
) C
n
,
106 The -Neumann Problem on Strongly Pseudoconvex Manifolds
is positive semi-denite (positive denite).
We shall normalize such that [d[ = 1 on /. is pseudoconvex at a point
. on / 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

jk
o
j
o
k
0 if
n

j=1
o
j
(.)
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 denitions are independent of the choice of the dening function
and are independent of the choice of n
1
, , n
n
. If we choose a special boundary
chart such that n
n
= , then ,n
i
, i = 1, , n 1 are tangential operators.
We have, substituting for n in (5.3.3),
(5.3.5)

=

n
n
=

j,k

jk
n
j
n
k
,
where (
jk
) = ( c
n
jk
) = (c
n
kj
) is the Levi matrix. In this case, / is pseudoconvex if
and only if (
jk
)
n1
j,k=1
is positive semi-denite.
We shall use the same Hilbert space theory as that in Chapter 4 to study the
function theory on pseudoconvex manifolds. We x a function C
2
(). Let

: 1
2
(p,q)
(, ) 1
2
(p,q+1)
(, ) be the closure of the Cauchy-Riemann operator
and we dene the Hilbert space adjoint for

as before. Let .
0
/ be a boundary
point and l be an open neighborhood of .
0
. We shall x a special orthonormal
boundary frame n
1
, , n
n
= . 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
() = 0, when . / l, i = 1, , n 1,
1
n
() = 1, when . / l.
We compute

) and ) in this special coordinate chart. We can write any )
C

(p,q)
( l) as ) =

t
]I]=p,]J]=q
)
I,J
n
I
n
J
. Then
(5.3.7)
) C

0
( l) Dom(

)
if and only if )
I,J
= 0 whenever n J,
where C

0
(l ) denotes the space of functions in C

() which are supported in


l .
We denote the space C

(p,q)
() Dom(

) by T
(p,q)
and C

(p,q)
() Dom(

) 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
+ = ) + ,
5.3 Function Theory on Manifolds 107
and
(5.3.9)

) = (1)
p1

I,K
t

j
)
I,jK
n
I
n
K
+ = 1) + ,
where

j
n = c

1
j
(c

n) and dots indicate terms where no derivatives of )


I,J
occur and which do not involve . The second equalities in (5.3.8) and (5.3.9) are
denitions of and 1 which are rst order dierential operators.
Thus, we have
(5.3.10)
|)|
2

+ |

)|
2

I,J,L
t

j,
c
jJ
L
(1
j
()
I,J
), 1

()
I,L
))

I,K
t

j,k
(

j
)
I,jK
,

k
)
I,kK
)

+1()),
where 1()) involves terms that can be controlled by O((| ) |

+ | 1) |

) | ) |

)
and c
jJ
L
is dened as before. Rearranging the terms in (5.3.10), we have
(5.3.11)
|)|
2

+ |

)|
2

I,J
t

j
|1
j
)
I,J
|
2

I,K
t

j,k
(1
k
)
I,jK
, 1
j
)
I,kK
)

I,K
t

j,k
(

j
)
I,jK
,

k
)
I,kK
)

+1()).
We apply integration by parts to the terms (

j
)
I,jK
,

k
)
I,kK
)

. For each n,
C
1
0
( l), Greens formula gives
(5.3.12) (n,

j
)

= (1
j
n, )

+ (
j
n, )

+
_
b
(1
j
)nc

do,
where do is the surface element on / and
j
is in C
1
(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 (

j
)
I,jK
,

k
)
I,kK
)

, no boundary terms arise since


(5.3.13) )
I,J
= 0 on /, if n J,
and if , < n and n J,
(5.3.14) 1
j
() = 1
j
() = 1
j
()
I,J
) = 1
j
()
I,J
) = 0 on /.
In order to calculate the commutator [

j
, 1
k
], we use (5.3.2) and (5.3.3),
(5.3.15)
[

j
, 1
k
] n = [1
j
1
j
(), 1
k
]n = [1
j
, 1
k
]n +1
k
1
j
()n
=

i
c
i
kj
1
i
(n)

i
c
i
jk
1
i
(n) +1
k
1
j
()n
=
jk
n +

i
c
i
kj

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 xed 1, 1, ,, /, we have
(5.3.16)
(

j
)
I,jK
,

k
)
I,kK
)

= (1
k

j
)
I,jK
, )
I,kK
)

+ (

j
)
I,jK
,
k
)
I,kK
)

= (1
k
)
I,jK
, 1
j
)
I,kK
)

+ ([

j
, 1
k
])
I,jK
, )
I,kK
)

(1
k
)
I,jK
,
j
)
I,kK
)

+ (

j
)
I,jK
,
k
)
I,kK
)

.
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

I,J
t

j
| 1
j
)
I,J
|
2

+ | ) |
2

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) [(

j
)
I,jK
,
k
)
I,kK
)

[ C | 1) |

| ) |

,
where C is a constant independent of . We shall use O(| 1) |

| ) |

) to denote
terms which are bounded by C | 1) |

| ) |

where C is a constant independent


of . Thus, (5.3.16) reads
(5.3.18)
(

j
)
I,jK
,

k
)
I,kK
)
= (1
k
)
I,jK
, 1
j
)
I,kK
)

+ ([

j
, 1
k
])
I,jK
, )
I,kK
)

+O(| 1) |

| ) |

)
= (1
k
)
I,jK
, 1
j
)
I,kK
)

+ (
jk
)
I,jK
, )
I,kK
)

+
_

i
c
i
kj

i
)
I,jK
, )
I,kK
_

+O(| 1) |

| ) |

).
If i < n, integration by parts gives
[(c
i
kj

i
)
I,jK
, )
I,kK
)

[ C | 1) |

| ) |

.
If i = n, we get, using (5.3.5),
(5.3.19)
(c
n
kj

n
)
I,jK
, )
I,kK
)

=
_
bU
c
n
kj
)
I,jK

)
I,kK
c

do +O(| 1) |

| ) |

)
=
_
bU

jk
)
I,jK

)
I,kK
c

do +O(| 1) |

| ) |

).
Combining (5.3.11), (5.3.18) and (5.3.19), we obtain
(5.3.20)
|)|
2

+|

)|
2

I,J
t

j
|1
j
)
I,J
|
2

I,K
t

j,k
(
jk
)
I,jK
, )
I,kK
)

I,K
t

j,k
_
bU

jk
)
I,jK

)
I,kK
c

do +1()) +1()),
5.3 Function Theory on Manifolds 109
where [1())[ C(| 1) |

| ) |

). Also for any c 0, there exists a C

0 such
that
(5.3.21) [1())[ c(|)|
2

+ |

)|
2

) +C

|)|
2

,
where C

is independent of . Thus combining (5.3.20) and (5.3.21), we have proved


the following proposition (notice only three derivatives of are required).
Proposition 5.3.3. Let ` be a complex manifold with C
3
boundary and
be a C
3
dening function for . For any ) T
2
(p,q)
such that ) vanishes outside a
coordinate patch l near a boundary point in / and C
2
(), we have
(5.3.22)
|)|
2

+ |

)|
2

(1 c)

I,K
t
n

j,k=1
(
jk
)
I,jK
, )
I,kK
)

+ (1 c)

I,K
t
n

j,k=1
_
bU

jk
)
I,jK

)
I,kK
c

do
+ (1 c)

I,J
t
n

j=1
|1
j
)
I,J
|
2

+O(c)(| 1) |

| ) |

),
where c 0 can be chosen arbitrarily small and O(c)(| 1) |

| ) |

) denotes terms
which can be bounded by C

| 1) |

| ) |

for some constant C

independent of .
Let be the smallest eigenvalue of the Hermitian symmetric form
(5.3.23)
n

j,k=1

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

jk
o
j
o
k
, where
n

i=1
o
i

n
i
= 0.
Note that 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 ` be a complex manifold with C
3
boundary / and
C
2
(). We have the following estimates: for every ) T
(p,q)
,
(5.3.25) |)|
2

+ |

)|
2


1
2
__

( C)[)[
2
c

d\ +
_
b
j [)[
2
c

do
_
,
where 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 .
110 The -Neumann Problem on Strongly Pseudoconvex Manifolds
Proof. Let
i

N
i=0
be a partition of unity such that
0
C

0
() and each
i
,
1 i , is supported in a coordinate patch l
i
,
i
C

0
(l
i
), (
i
l
i
),
and
N

i=0

2
i
= 1 on .
Since )
I,nK
= 0 on /, we have, for 1 i ,

I,K
t

j,k
_
bUi

jk
)
I,jK

)
I,kK
c

do
_
bUi
j
2
i
[)[
2
c

do.
Applying Proposition 5.3.3 to each
i
), choosing c suciently small, we have
(5.3.26)
_
bUi
j
2
i
[)[
2
c

do +
_
Ui

2
i
[)[
2
c

d\
2(|
i

) |
2

+ |
i

) |
2

) +C
i
_
Ui
[)[
2
c

d\.
The constant C
i
depends only on
i
but not on . 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 .
Theorem 5.3.5. Let ` be a pseudoconvex manifold with C
3
boundary /
such that there exists a strictly plurisubharmonic function on . Then for any
) 1
2
(p,q)
() with

) = 0, there exists n 1
2
(p,q1)
() such that

n = ).
Proof. From pseudoconvexity, j 0 on / where j is the smallest eigenvalue of
the Levi form. We have that the last term in (5.3.25) is nonnegative. Since is
strictly plurisubharmonic in which is relatively compact in `, it follows that
0 where is the smallest eigenvalue of the form (5.3.23). If we choose t 0
such that t 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
t
|p|
2
t
+|

t
p|
2
t
.
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(

t
) in the graph norm |p|
t
+
|p|
t
+|

t
p|
t
and (5.3.27) holds for any p Dom(

)Dom(

t
). Using Lemma
4.1.1, this implies that (

) and (

t
) are closed. To show that (

) = ker(

),
we repeat the arguments of the proof of Theorem 4.3.4. For any p 1
2
(p,q)
()
Dom(

t
), one has
(5.3.28) [ (), p)
t
[ | ) |
t
| p |
t
| ) |
t
|

t
p |
t
.
Thus, there exists n 1
2
(p,q1)
() such that

n = ) in and
| n |
t
| ) |
t
.
5.3 Function Theory on Manifolds 111
This proves the theorem.
We note that if ` = C
n
, one can take = [.[
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 dened as in Deni-
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 dened self-adjoint operator on 1
2
(p,q)
(). 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 ` be a pseudoconvex Hermitian manifold with C
3
boundary / such that there exists a strictly plurisubharmonic function on .
For each j, such that 0 j n, 1 n, there exists a bounded operator

(p,q)
: 1
2
(p,q)
() 1
2
(p,q)
() such that
(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)
(), ) =

(p,q)
)

(p,q)
).
(3)

(p,q)
=
(p,q+1)

on Dom(

), 1 n 1.
(4)

(p,q)
=
(p,q1)

on Dom(

), 2 n.
(5) For any ) 1
2
(p,q)
() 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 / is strongly pseudoconvex, there exists a c 0 such that the smallest
eigenvalue of the Levi form j c 0 on /, and we have from (5.3.25) (setting
= 0),
(5.3.29) |)|
2
+ |)|
2

c
2
_
b
[)[
2
do C|)|
2
.
When / 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
() can be dened on a manifold 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 ` be a strongly pseudoconvex Hermitian manifold with
C
3
boundary /. 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
()
C(|

) |
2

+ |

) |
2

+ | ) |
2

),
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 given by (5.3.8) and (5.3.9) dier 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 modied 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)
() = ) 1
2
(p,q)
() Dom(

) Dom(

)[

) =

) = 0
= Ker(

) Ker(

)
= Ker(
(p,q)
).
The last equality can be veried in the same way as in the proof of (4.4.2). However,
on a strongly pseudoconvex complex manifold, H
(p,q)
() = H
(p,q)
is not always triv-
ial for 1. The following theorem shows that H
(p,q)
is always nite dimensional
when 1.
Theorem 5.3.8. Let ` be a strongly pseudoconvex Hermitian manifold with
a C
3
boundary /. For any 0 j n and 1 n, the space H
(p,q)
is nite
dimensional. Furthermore, the following estimate holds: for any ) Dom(

)
Dom(

) H

(p,q)
,
(5.3.31) | ) |
2

C(|

) |
2

+ |

) |
2

).
Proof. We have from (5.3.30),
(5.3.32) | ) |
2
1
2
()
C | ) |
2

, ) H
(p,q)
.
Since \
1/2
is compact in 1
2
() 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 nite
dimensional.
If (5.3.31) does not hold, there exists a sequence )
n
such that )
n
Dom(

)
Dom(

) H

(p,q)
,
| )
n
|
2

n(|

)
n
|
2

+ |

)
n
|
2

).
Let
n
= )
n
,| )
n
|

. Then |

n
|

+ |

n
|

0 and |
n
|

= 1. From
(5.3.30) we have |
n
|1
2
()
C. Using the Rellich lemma, there exists a subsequence
of
n
which converges to some element 1
2
(p,q)
H

(p,q)
. However,

=

= 0
and | |

= 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 coecients. The following theorem gives the
existence and regularity of the

-Neumann operator on any strongly pseudoconvex
Hermitian manifold.
Theorem 5.3.9. Let ` be a strongly pseudoconvex Hermitian manifold with
C

boundary /. For each j, such that 0 j n, 0 n, there exists a


compact operator
(p,q)
: 1
2
(p,q)
() 1
2
(p,q)
() such that
(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)
(), ) =

(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,q1)

on Dom(

).
(5)
(p,q)
(C

(p,q)
()) C

(p,q)
(), 0.
H
(p,q)
(C

(p,q)
()) C

(p,q)
(), 0.
Proof. We rst prove that (
(p,q)
) is closed for any 1 n. From (5.3.31), we
know that for any ) Dom(
(p,q)
) H

(p,q)
,
| ) |
2

C(|

) |
2

+ |

) |
2

)
= C(
(p,q)
), ))

C |
(p,q)
) |

| ) |

.
Now applying Lemma 4.1.1, we see that (
(p,q)
) is closed and
1
2
(p,q)
() = (
(p,q)
) Ker(
(p,q)
) = (
(p,q)
) H
(p,q)
.
Let
(p,q)
to be the bounded inverse operator of
(p,q)
on (
(p,q)
). We extend

(p,q)
to 1
2
(p,q)
() 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)
(),
(5.3.33)
|
(p,q)
) |
2
1
2
()
C(|
(p,q)

(p,q)
) |
2

+ |
(p,q)
) |
2

)
C | ) |
2

.
The Rellich lemma implies that is a compact operator. (3) and (4) can be veried
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)
=
2
(p,1)

.
Also,
(p,0)
is bounded. The compactness of
(p,0)
follows since
(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 ` be a strongly pseudoconvex Hermitian manifold
with C

boundary /. For 1 and each / = 0, 1, 2, , there exists a constant


C
k
0 such that for any ) \
k
(p,q)
(),
(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 ` be a strongly pseudoconvex Hermitian manifold
with C

boundary /. For any ) \


k
(p,q)
(), 1 and / 0, such that

) = 0
in and H
(p,q)
) = 0, we have n =

) is a solution of

n = ) in and
|n|
k+
1
2
C|)|
k
,
where C is a constant independent of ). In particular, if ) C

(p,q)
(),

) = 0 and
H
(p,q)
) = 0, there exists a solution n C

(p,q1)
() such that

n = ) in .
The solution n is called the canonical solution (or Kohns 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
with smooth boundary / is called a domain of holomorphy if for every j
/ there is a holomorphic function on which is singular at j (c.f. Denition
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 ` be a strongly pseudoconvex manifold with C

boundary /. Then is a domain of holomorphy.


Proof. For each boundary point j /, we will construct a function /(.) such that
/ C

( j), / is holomorphic in and


lim
zp
/(.) = .
Since is strongly pseudoconvex with C

boundary /, one can construct a


local holomorphic function ) on 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 dening function for
such that : is strictly plurisubharmonic near j. That such a dening function exists
follows from the same arguments as before (see Theorem 3.4.4). Let
1(.) = 2
n

i=1
:
.
i
(0).
i

i,j=1

2
:
.
i
.
j
(0).
i
.
j
.
1(.) is holomorphic in l. Using Taylors expansion at 0, there exists a suciently
small neighborhood \ l of 0 and C 0 such that for any . \ ,
Re1(.) = :(.) +
n

i,j=1

2
:
.
i
.
j
(0).
i
.
j
+O([.[
3
) C[.[
2
.
Thus, 1(.) ,= 0 when . \ 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 be a cut-o function such that C

0
(\ ) and = 1 in a neighborhood
of 0. We extend to be 0 on \ . Let p be the (0, 1)-form dened by
p =

()) = (

)) in .
Obviously p can be extended smoothly up to the boundary. Thus, p C

(0,1)
() and

p = 0 in . To show that H
(0,1)
p = 0, notice that when n 3, ) 1
2
(). Thus,
p (

) = Ker(

and p H
(0,1)
. When n = 2, we approximate ) by )

=
1
P+
for c 0. Then )

() and

()

) p in 1
2
. However, each

()

) is
in the (

) which is closed from Theorem 5.3.9. This implies that p (

) and
H
(0,1)
p = 0 for n = 2 also. We dene
n =

(0,1)
p.
It follows from Corollary 5.3.11 that n C

() and

n = p in . Let / be dened
by
/ = ) n.
Then, / is holomorphic in , / C

( 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 eld in 1
2
is actually a Cauchy-
Riemann equation in other coordinates. In this section we study the n-dimensional
analog of this problem.
Denition 5.4.1. Let ` be a real C

manifold of dimension 2n. An almost


complex structure T
1,0
(`) is a subbundle of the complexied 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
dened 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 right-hand side denotes the linear span by vector elds ,.
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
Denition 5.4.2. An almost complex structure T
1,0
(`) is called integrable if
(5.4.2) is satised.
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 rst note that on an almost complex manifold, there is a notion of
the Cauchy-Riemann equations and the

operator.
Let
1,0
,
0,1
denote the projection from CT(`) onto T
1,0
(`) and T
0,1
(`)
respectively. Then, one has

1,0
+
0,1
= 1,
1,0

0,1
= 0, and
0,1
=
1,0
.
The last equation means that
0,1
=
1,0

, for CT(`). Thus there is a


natural splitting of the dierential 1-forms
1
(`) into (1,0)-forms,
1,0
(`), and
(0,1)-forms,
0,1
(`), which are dened to be the dual of T
1,0
(`) and T
0,1
(`)
respectively. We shall still use
1,0
,
0,1
to denote the projection from
1
(`)
onto
1,0
(`) and
0,1
(`) respectively. For any smooth function n, we have dn =

1,0
dn +
0,1
dn. On an almost complex manifold, one can dene the Cauchy-
Riemann equation by

n =
0,1
dn and n =
1,0
dn,
where n is any smooth function on `. We can also extend this denition to (j, )-
forms and dene 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 1-forms, we


have
1
(`) =
1,0
(`)+
0,1
(`). To verify for 1-forms, it suces 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 (Newlander-Nirenberg). 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 nd complex-valued functions
1
, ,
n
such that
(5.4.3) 1
i

j
= 0, i, , = 1, , n,
where d
1
, , d
n
are linearly independent, then the theorem will be proved since
(5.4.4) 1
1
, , 1
n
) =
_

1
, ,

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 coecient 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
)
denes an almost complex structure that is integrable for each c < c
0
for some
suciently small c
0
0.
From Lemma 5.4.3, there is a Cauchy-Riemann 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 =

n
i=1
[.
i
[
2
= [r[
2
, then at 0 we see that
n

j,k=1

jk
o
j
o
k
=
n

i=1
[o
i
[
2
.
Thus, is a strictly plurisubharmonic function near 0. If we set
= r ` [ [r[
2
<
118 The -Neumann Problem on Strongly Pseudoconvex Manifolds
for some small 0, then 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
such that
(5.4.7)

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

.
i
= O(c)
for any 1

= (,r
1
)
1
(,r
2n
)
2n
, where the
i
s are nonnegative integers.
This implies that
(5.4.9) |

.
i
|
s
0 if c 0.
Let
(5.4.10)

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

i
= 0 in and also d

i
(0) = d.
i
dn

i
(0) are linearly
independent if c is suciently small. If we pull back

i
to c by setting
i
=

i
(r,c),
then we have that

i
= 0 and d
i
are linearly independent in c provided we choose
c suciently small. This proves the theorem.
NOTES
The subelliptic 1,2-estimates 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
simplication of the proof of the boundary regularity for subelliptic operators using
pseudodierential 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
Pseudodierential operators and subelliptic estimates will be discussed in Chapter
8.
The proof of subelliptic 1,2-estimates 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 dierential
operators are used but not pseudodierential operators. The discussion of function
theory on manifolds mainly follows that of L. Hormander [Hor 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 which lies in a Stein manifold. For a detailed discussion
of function theory on Stein manifolds, see Chapter 5 in L. Hormander [Hor 9].
The Levi problem on a strongly pseudoconvex manifold (Theorem 5.3.12) was
rst 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 denition 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 rst proved by A. Newlander and L. Nirenberg [NeNi 1].
B. Malgrange has given a totally dierent 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 Holder 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. Feerman 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,q1)
(1) if )
is in C

(p,q)
(1)?
(2) Does the canonical solution

) belong to \
s
(p,q1)
(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 rst question can be answered armatively.
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 dening
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 dening 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
nd n C

(p,q1)
(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,
t
) introduced in Section 4.2, where
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,
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 briey describe below.
From Proposition 4.3.3, we have for any (j, )-form ) Dom() Dom(

t
),
_
D

I,K
t

j,k

t
.
j
.
k
)
I,jK
)
I,kK
c
t
d\ | ) |
2
t
+ |

t
) |
2
t
.
Using the notation | |
(t)
= | |
t
and

t
=

t
, 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-
to-one. Thus,
t
has a bounded inverse
t
, the

-Neumann operator with weight

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
).
(3)
t
=
t
on Dom(), 1 n 1,

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
) satisfying n
t
= ) and the estimate
t | n
t
|
2
(t)
| ) |
2
(t)
.
In Chapter 4, we have chosen t =
2
, where 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 rst prove the a priori estimates for
t
when t is large. Let
t
be the
formal adjoint of

with respect to the weighted norm 1
2
(1,
t
). Note that for any
) C

(p,q)
(1),

t
) = c
t
(c
t
)) = ) +t
0
)
for some zeroth order operator
0
. Hence, we have that for any ) C

(p,q)
(1) with
compact support in 1,
Q
t
(), )) = | ) |
2
(t)
+ |
t
) |
2
(t)
| ) |
2
(t)
+
1
2
| ) |
2
(t)
C
t
| ) |
2
(t)
| ) |
1
C
t
| ) |
2
(t)
,
where | ) |
k
=

0]]k
| 1

) |
(t)
.
Thus the Garding 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 dierentiation is controlled by ,

and the tangential deriva-


tives, we shall only consider the action of tangential dierentiations. 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 dening function normalized such that [d:[ = 1
on /1. We let T
k
be a tangential operator of order / and C

0
(l) as dened 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 T
k
) Dom() Dom(

t
). We obtain
from (6.1.1) that
(6.1.4) t | T
k
) |
2
(t)
(| (T
k
)) |
2
(t)
+ |
t
(T
k
)) |
2
(t)
).
We note that the commutator [
t
, T
k
] =
k
+
t
k1
, where
k
is a /th order
dierential operator independent of t and
t
k1
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)
| (T
k
)) |
2
(t)
+ | (
t
T
k
)) |
2
(t)
C
k
(| T
k
) |
2
(t)
+ | T
k

t
) |
2
(t)
+ | ) |
2
k
)
+C
k,t
| ) |
2
k1
C
k
([(T
k
), T
k

t
))
(t)
[ + | ) |
2
k
) +C
k,t
| ) |
2
k1
C
k
(| T
k
) |
(t)
| T
k

t
) |
(t)
+ | ) |
2
k
) +C
k,t
| ) |
2
k1
.
Combining (6.1.4) and (6.1.5), we get
(6.1.6) t | T
k
) |
2
(t)
C
k
(|
t
) |
2
k
+ | ) |
2
k
) +C
k,t
| ) |
2
k1
,
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
k1
+ [[[)[[[
2
k
) +C
k,t
| ) |
2
k1
.
Summing up all the tangential derivatives of the form T
k
in (6.1.6) and using a
partition of unity
i

N
i=1
, such that

N
i=1

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
k1
+ C
k,t
| ) |
2
k1
.
Choosing t C
k
+ 1, it follows, using the induction hypothesis, that
| ) |
2
k
C
k
|
t
) |
2
k
+ C
k,t
|
t
) |
2
k1
+ C
k,t
| ) |
2
k1
C
k,t
|
t
) |
2
k
.
This proves the a priori estimates (6.1.3) for the weighted -Neumann operators
t
when t is suciently 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,
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)
=

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,(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) coecients. 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
and

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
are
exactly regular simultaneously. As before, since


t
is elliptic, we only need to
prove a priori estimates for the tangential derivatives of

t
) and

t
) for any
) C

(p,q)
(1). Let and T be as in Theorem 6.1.3 and let O
t
(| ) |) denote terms
which can be bounded by C
t
| ) |. We have
| T
k

t
) |
2
(t)
+ | T
k

t
) |
2
(t)
= (T
k

t
), T
k

t
))
(t)
+ (T
k

t
),

t
T
k

t
))
(t)
+O
t
((| T
k

t
) |
(t)
+ | T
k

t
) |
(t)
) |
t
) |
k
)
= (T
k

t
), T
k

t
))
(t)
+ (T
k

t
), T
k

t
))
(t)
+O
t
((| T
k

t
) |
(t)
+ | T
k

t
) |
(t)
) |
t
) |
k
+ |
t
) |
2
k
)
= (T
k

t
), T
k

t
))
(t)
+O
t
((| T
k

t
) |
(t)
+ | T
k

t
) |
k
) |
t
) |
k
+ |
t
) |
2
k
)
C | ) |
k
|
t
) |
k
+C
t
((| T
k

t
) |
(t)
+ | T
k

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
| T
k

t
) |
(t)
+ | T
k

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
) |
k
C
k,t
| ) |
k
.
6.1 Global Regularity for on Pseudoconvex Domains 125
For the operators

t
and

t
, 1 n, we have
| T
k

t
) |
2
(t)
+ | T
k

t
) |
2
(t)
= (T
k

t
), T
k

t
))
(t)
+ (T
k

t
), T
k

t
))
(t)
= (T
k

t
), T
k

t
))
(t)
(T
k

t
), T
k

t
))
(t)
(T
k

t
), T
k

t
))
(t)
= (T
k
), T
k
))
(t)
([, T
k
]

t
), T
k

t
))
(t)
([

t
, T
k
]

t
), T
k

t
))
(t)
+1
= (T
k
), T
k
))
(t)
([, T
k
]

t
), T
k

t
))
(t)
([

t
, T
k
]
t
), T
k

t
))
(t)
+1,
where the error term 1 can be estimated by
(| T
k

t
) |
(t)
+ | T
k

t
) |
(t)
)(|
t
) |
k
+ |

t
) |
k
)
+ |
t
) |
2
k
+ |

t
) |
2
k
.
Since
t
and

t
are already known to be exactly regular on \
k
(p,q)
(1), we
obtain using small and large constants that
| T
k

t
) |
2
(t)
+ | T
k

t
) |
2
(t)
C
k,t
| ) |
2
k
.
Summing over a partition of unity, we have proved
|

t
) |
2
k
+ |

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,(p,1)

. We use
induction on / to prove the a priori estimates for

t,(p,1)
. The case when / = 0
is obvious. Denoting
t,(p,1)
by
t
, we have
| T
k

t
) |
2
(t)
= (T
k

t
), T
k

t
))
(t)
= (T
k

t
), T
k

t
))
(t)
+O(|
t
) |
k
|

t
) |
k
)
= (T
k

t
), T
k
))
(t)
+O(|
t
) |
k
|

t
) |
k
)
= (T
k

t
), T
k
))
(t)
+O(|
t
) |
k
(| ) |
k
+ |

t
) |
k
)).
Summing over a partition of unity and using the fact that


t
is elliptic, the
small and large constants technique gives
(6.1.8) |

t
) |
2
k
C
k
(| ) |
2
k
+ |
t
) |
2
k
).
126 Boundary Regularity for on Pseudoconvex Domains
Using estimate (6.1.1) to T
k

t
), we obtain
(6.1.9) t | T
k

t
) |
2
(t)
| T
k

t
) |
2
(t)
+ |

t
T
k

t
) |
2
(t)
.
Since [

, T
k
] +[
t
, T
k
] = 1
k
+1
t
k1
where 1
k
is a /th order dierential operator
independent of t and 1
t
k1
is a dierential operator of order / 1, it follows that
|T
k

t
) |
2
(t)
+ |

t
T
k

t
) |
2
(t)
C
k
( |

t
) |
2
k
+ |
t
) |
2
k
) +C
k,t
|
t

) |
2
k1
.
We see, using induction, that

is bounded on \
k1
(1). The above argument
implies that
t

is also bounded on \
k1
(1). Thus, after summing over a partition
of unity, if t is chosen to be suciently large in (6.1.9), we obtain that
(6.1.10) |
t
) |
2
k
c |

t
) |
2
k
+ C
k,t
| ) |
2
k1
,
for some small c 0. Letting c be suciently small, (6.1.8) and (6.1.10) together
show
|

t
) |
2
k
C
k,t
| ) |
2
k
.
This proves the a priori estimate for

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,q1)
(1) such that n = ) on 1.
Proof. Since ) = 0, using Theorem 6.1.2 n =

t
) is a solution to the equation
for any t 0. If t is suciently large,

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,q1)
(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 suciently 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 =

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,q1)
(1) Ker(), there exists a
k+1
\
k+1
(p,q1)
(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,q1)
(1) and n
k+1
= ). Inductively,
we can choose a new sequence n
k
\
k
(p,q1)
(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 dened and is in \


N
(p,q1)
(1) for every . Thus n

(p,q1)
(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 eld method we shall give a sucient 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 dening 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
elds at every boundary point of 1.
Denote by A
n
= ([:[,

2)1
n
the globally dened type (1, 0) vector eld 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 elds A
1
, , A
n1
so that
A
1
, , A
n1
together with A
n
form an orthonormal basis of the space of type
(1,0) vector elds in some open neighborhood of j. We shall also denote by

1
, ,
n
the orthonormal frame of (1,0)-forms dual to A
1
, , A
n
near j. Note
that
n
= (

2,[:[)

n
j=1
(:,.
j
)d.
j
is a globally dened (1,0)-form in some open
neighborhood of the boundary.
The main idea of this method is to construct a real tangential vector eld 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 eld T = T

,
depending on c, dened 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

(.)[ 0 for all . /1, where


is a positive constant independent of c.
(2) If o is any one of the vector elds 1
n
, 1
n
, 1
jk
and 1
jk
, 1 , < / n,
then
[T, o][
bD
=
S
(.)1
n
, mod (T
1,0
(/1) T
0,1
(/1), 1
n
),
for some smooth function
S
(.) with sup
bD
[
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
=
n1

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 dened 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 dening 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 suces
to prove a priori estimates for the -Neumann operator. The proof will be done by
induction on the order of dierentiation. 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

m
=1
such that l

m
=1
and l
0
= 1 form an open
cover of 1. We shall assume that [:[ c 0 on
m
=1
l

for some positive


constant c. Let

m
=0
be a xed partition of unity subordinate to l

m
=0
. On
each l

, 1 :, let
k
, / = 1, , n be an orthonormal frame of (1,0)-forms
6.2 Sobolev Estimates for the -Neumann Operator 129
dual to A
k
, / = 1, , n. We note that
n
=
n
, = 1, , :, is a globally
dened (1,0)-form dual to A
n
= ([:[,

2)1
n
. Similarly, A
n
= A
n
, = 1, , :,
is also a globally dened type (1,0) vector eld.
On each boundary chart l

, = 1, , :, we may write
n = n

t
n

I,J

,I

,J
,
where
,I
=
i1

ip
and
,J
=
j1

jq
. Let T be the smooth
real vector eld on 1 satisfying the hypothesis of condition (T). For each : N, we
dene
(6.2.1) T
s
n = T
s
(
0
n) +
m

=1

t
I,J
T
s
(

I,J
)
,I

,J
.
Thus, T
s
n T
(p,q)
using Lemma 5.2.2.
Set
| An |
2
=
m

=1

t
I,J
n

k=1
| A
k
(

I,J
) |
2
,
and
| A
t
n |
2
=
m

=1

t
I,J
n1

k=1
| A
k
(

I,J
) |
2
.
From estimate (4.3.1), with = 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 : and 1 / n 1,
| A
k
(

I,J
) |
2
= | A
k
(

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
1
, , A
n1
, A
1
, , A
n1
and A
n
, 1 :, 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 right-hand side as follows.
| Tn |
2
= (Tn, Tn)
= (Tn, 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) + (Tn, [, 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
n1
, A
1
, , A
n1
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](

I,J
) = [([:[,

2)1
n
, T](

I,J
)
= ([:[,

2)[1
n
, T](

I,J
) (T([:[,

2))1
n
(

I,J
)
= ([:[,

2)[1
n
, T](

I,J
) (T([:[),[:[)A
n
(

I,J
),
for = 1, , :. Using the basic estimate, we obtain
m

=1

t
n/ J
| (T([:[),[:[)A
n
(

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

=1

t
|

o

T(

I,J
) |
2
+ | An |
2
+ | A
t
n |
2
+ | ) |
2
1

_
c

_
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](

I,J
) = ([:[,

2)[1
n
, T](

I,J
) (T([:[),[:[)A
n
(

I,J
),
for = 1, , :. Observe that A
n
(

I,J
) appears in the coecient of
,I

,H
with n H = J in

n. Meanwhile, all the other terms in the coecient of

,I

,H
are dierentiated by A
1
, , A
n1
only. Thus we have
m

=1

t
nJ
| (T([:[),[:[)A
n
(

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 rst choose (:c) to be small enough, and then c to be suciently small.
From (6.2.3) we obtain
| [, T]n |
2
+ | [

, T]n |
2
| An |
2
+ | ) |
2
1
+ | Tn |
2
,
where 0 is a constant that can be made as small as we wish. Combining these
estimates, if we let 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 \
s1
(p,q)
(1) for some
integer : 2, i.e.,
| n |
s1
C
s1
| ) |
s1
,
where C
s1
0 is a constant independent of ). The strategy here is the same as
before. Using basic estimate and the induction hypothesis, we rst 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 dierential op-
erator of order :. Using (6.2.4), it suces to consider the estimate near a boundary
point j. Let l be a boundary coordinate chart near j, and let
1
, ,
n
be an
orthonormal basis for (1,0)-forms on l dual to A
1
, , A
n
dened as before. De-
note by Op(:, ,), 1 , :, a tangential dierential operator of order : formed out
of A
1
, , A
n1
, A
1
, , A
n1
and T with precisely : , factors of T. Let be
a cut-o function supported in l such that 1 in some open neighborhood of j.
We claim that
(6.2.5) | Op(:, ,)n |
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 elds A
1
, , A
n1
. Then, by commuting one A
t
or A
t
to
the left and applying integration by parts, it is easily seen that
| Op(:, ,)n |
2
| ) |
2
s
+ | A
t
Op(: 1, , 1)n |
2
+ | Op(:, , 1)n |
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 dierentiation in the normal direction. Therefore, a partition of
unity argument gives
| n |
s
| ) |
s
+ (:c) | n |
s
which implies, by choosing (:c) suciently 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
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) = (
t
), p)
t
= (1
t

t
), p)
t
= (
t
1
t
(
t
)), p).
Hence, we get 1 = 1
t
1
t

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 =
t
1
t

1
((
t
)1
t

t
).
Since it has been proved that

1
preserves \
s
(0,1)
(1) for each nonnegative real
:, if we choose t to be suciently 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 eld T when the domain 1 has a plurisub-
harmonic dening function or transverse circular symmetry. We say that a smooth
bounded domain 1 has a plurisubharmonic dening function if there exists some
smooth dening 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 dening 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 dening 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 (, 1, 0, , 0)
C
n
for any C, we obtain
2Re
_


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
n1
be a local basis for the space of type (1, 0) tangential vector elds near
j. We may assume that 1
1
, , 1
n1
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
coecients of A
p
at j. It follows that the system of linear equations
n1

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
n1
C. This implies that
__
A
p

n1

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
, , /
n1
such that
1
n

n1

j=1
/
j
1
j
= A
p

n1

j=1
o
j
1
j
.
134 Boundary Regularity for on Pseudoconvex Domains
Therefore, given c 0, it is easily veried by using a partition of unity
l

m
l=1
with small support that one may patch 1
n

n1
j=1
/
j
1
j
together to form a globally
dened type (1, 0) vector eld A = 1
n
Y , where Y =

m
l=1

l
(

n1
j=1
/
j
1
j
) is a
globally dened tangential type (1, 0) vector eld, 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 eld, the
commutator [A A, 1
k
], 1 / n 1, is also a tangential vector eld. Thus, we
may write
[A A, 1
k
] =
k
(1
n
1
n
), mod (T
1,0
(/1) T
0,1
(/1)),
for 1 / n 1. By (6.2.7), we have [
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 elds 1
1
, , 1
n1
, 1
1
, , 1
n1
. In
particular, we may achieve that
(6.2.9) sup
bD
[[A A, Y Y ], 1
n
)[ < c,
where c might be dierent 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 eld [AA, 1
n
] is tangent to the level sets of :. It follows
that [A A, Re1
n
] is also a tangential vector eld and we can write
(6.2.10) [A A, Re1
n
] = (1
n
1
n
), mod (T
1,0
(/1) T
0,1
(/1)),
where is a real-valued function dened in some neighborhood of the boundary.
Now, we set
(6.2.11) T = c
r
(A A).
Obviously, T depends on c and satises property (1) of condition (T) on the bound-
ary. If o is any one of the vector elds 1
1
, , 1
n1
, 1
1
, , 1
n1
, 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
r
(A A), Re1
n
], 1
n
)[
bD
+
1
2
[c
r
(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 eld T satises 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 sucient condition for verifying the exact regularity of the
-Neumann operator. However, this condition in general is not satised 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 dened by
(6.2.12) j(r) = inf 0[ r 1,
where 1 = j[ 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 dened as in (6.2.12). Then j(r)
is a smooth, real-valued function on 1
N
0 satisfying the following properties:
(1) j(r) is a dening 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, real-valued 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
i
. = (c
i
.
1
, , c
i
.
n
) 1 for any . 1 and 1. 1 is
called Reinhardt if (c
i1
.
1
, , c
in
.
n
) 1 for any . 1 and
1
, ,
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 dened
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 dening function for 1 such that :(.) = :(c
i
.) and that [:[ = 1
on the boundary. Denote by the map of the o
1
-action on 1 from o
1
1 to 1
dened by
: o
1
1 1
(c
i
, .) c
i
. = (c
i
.
1
, , c
i
.
n
).
For each xed , is an automorphism of 1 and can be extended smoothly to
a map from o
1
1 to 1. Hence, for each xed . 1, we consider the orbit of .,
namely, the map

z
: o
1
1
c
i
c
i
..
Then,
z
induces a vector eld T on 1, in fact on C
n
, by
(6.2.15) T
z
=
z,
_

=0
_
= i
n

j=1
.
j

.
j
i
n

j=1
.
j

.
j
,
where
z,
is the dierential map induced by
z
. Note that T is tangent to the level
sets of :. In particular, T is tangent to the boundary of 1.
Denition 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 eld T dened 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 Denition 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 dened 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 eld dened 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 dening function :(.) = :(c
i1
.
1
, , c
in
.
n
)
for all
1
, ,
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 dened 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 suces to show that Re1(:) ,= 0 on the boundary.
This is, in turn, equivalent to showing that the real vector eld

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 one-to-one, 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 veried that the point evaluation map

w
: H(1) C
) )(n),
by Cauchys estimate, satises
(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) =
w
()) = (), 1
D
(, n)) =
_
D
)(.)1
D
(., n)d\
z
,
for all ) H(1). The function 1
D
(., n) thus dened is called the Bergman kernel
function for 1. By (6.3.1) the Bergman kernel function clearly satises
(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) satises
1(., n) = 1(n, .), for all ., n 1,
and hence 1(., n) is anti-holomorphic 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
j
(.)

j=1
be an orthonormal basis for H(1). Then
(6.3.3) 1(., n) =

j=1

j
(.)
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 xed n 1, from general Hilbert space theory we have
1(., n) =

j=1
(1(, n),
j
())
j
(.)
=

j=1
(
j
(), 1(, n))
j
(.)
=

j=1

j
(n)
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),
j
())[
2
=

j=1
[
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 nish the proof, it suces to show
by a normal family argument that [

m
j=1

j
(.)
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

j=1

j
(.)
j
(n)

j=1
[
j
(.)[[
j
(n)[

_
_

j=1
[
j
(.)[
2
_
_
1
2
_
_

j=1
[
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, .

is an orthogonal basis for H(1


n
), where the index = (
1
, ,
n
) runs over the
multiindices. We shall normalize it using the fact, for :, t N and 0 o < 1,
_

1a
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
(: + 1)(t + 2)
(: +t + 3)
,
6.3 The Bergman Projection 141
where 1(, ) is the Beta function and () is the Gamma function. Hence
| .

|
2
L
2
(Bn)
=
_
Bn
[.
1
[
21
[.
n
[
2n
d\
2n
=

(
n
+ 1)
_
Bn1
[.
1
[
21
[.
n1
[
2n1
(1 [.
1
[
2
[.
n1
[
2
)
n+1
d\
2n2
=

(
n
+ 1)
_
Bn1
[.
1
[
21
[.
n2
[
2n2
(1 [.
1
[
2
[.
n2
[
2
)
n+1
[.
n1
[
2n1
_
1
[.
n1
[
2
1 [.
1
[
2
[.
n2
[
2
_
n+1
d\
2n2
=

(
n
+ 1)
(
n1
+ 1)(
n
+ 2)
(
n
+
n1
+ 3)
_
Bn2
[.
1
[
21
[.
n2
[
2n2
(1 [.
1
[
2
[.
n2
[
2
)
n+n1+2
d\
2n4
=

(
n
+ 1)

(
n1
+ 1)(
n
+ 2)
(
n
+
n1
+ 3)

(
1
+ 1)(
n
+ +
2
+n)
(
n
+ +
1
+n + 1)
=

n

1
!
n
!
(
n
+ +
1
+n)!
.
It follows that the Bergman kernel function on the unit ball 1
n
is given by
1(., n) =

(
n
+ +
1
+n)!

n

1
!
n
!
.

=
1

k=0

]]=k
(
n
+ +
1
+n)!

1
!
n
!
.

=
1

k=0
(/ +n)(/ +n 1) (/ + 1)(.
1
n
1
+ +.
n
n
n
)
k
=
1

n
d
n
dt
n
_
1
1 t
_

t=z1w1++znwn
=
n!

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!

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 identication 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
()(.), )p() d\

= 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
Denition 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 satises 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 , there are constants c = c

and : = :

such that
sup
zD

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 rst 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 dierential operator
s
of order n
s
= :(:+1),2 with coecients in
C

(1) such that


s
maps \
s+ns
(1) boundedly into \
s
0
(1) and that 1
s
= 1.
In other words, for each p C

(1) and : N, Lemma 6.3.9 allows us to


construct a / = p
s
p C

(1) such that 1/ 0 and that / agrees with p up


to order : 1 on the boundary.
Proof. Let be a smooth dening function for 1, and let 0 be so small that
,= 0 on l

= .[ [(.)[ < . Choose a partition of unity


i

m
i=1
and, for each
i, a complex coordinate .
i
in some neighborhood of the support of
i
such that
(1)

m
i=1

i
1 on l

for some c < ,


(2) supp
i
l

and supp
i
/1 ,= , and
(3) ,.
i
,= 0 on supp
i
.
To dene the operator
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), dene

1
/ = /
m

i=1

.
i
(
i
0
),
where
i
0
= (
i
/)(,.
i
)
1
. It is easy to see that
1
/ = 0 on /1, and hence

1
/ \
1
0
(1).
Suppose
i
0
, ,
i
s1
have been chosen so that

s
i
/ =
i
/

.
i
_
s1

k=0

i
k

k+1
_
144 Boundary Regularity for on Pseudoconvex Domains
vanishes to order : 1 on the boundary. We dene

s
/ = /
m

i=1

.
i
_
s1

k=0

i
k

k+1
_
.
Since vanishes on the boundary, it is easily veried by integration by parts that

m
i=1

zi
(

s1
k=0

i
k

k+1
) is orthogonal to H(1). Hence
s
/ \
s
0
(1) and 1
s
/ =
1/. Put , = ,[[
2
, the normal dierentiation, such that , = 1.
Let

i
s
=
(

)
s

s
i
/
(: + 1)!

zi
.
Then the functions

s+1
i
/ =
s
i
/

.
i
(
i
s

s+1
)
vanish to order : on the boundary, so does the function

s+1
/ =
s
/
m

i=1

.
i
(
i
s

s+1
)
= /
m

i=1

.
i
_
s

k=0

i
k

k+1
_
.
Hence,
s+1
/ \
s+1
0
(1) and 1
s+1
= 1. This completes the induction.
It is also easily veried by a simple induction argument that
s
/ can be written
as

s
/ =

]]kns
/
,k

k
1

/,
where n
s
= :(: + 1),2 and the /
,k
s are in C

(1) and 1

is the real dierential


operator of order [[ associated to the multiindex = (
1
,
2
, ,
2n
). This
proves the lemma.
We also need negative Sobolev norms for holomorphic functions. For p O(1)
and : a positive integer, we dene
(6.3.9) | p |
s
= sup

_
D
p

,
where the supremum is taken over all C

0
(1) with | |
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
zD
[p(.)[d(.)
s+n
c
1
| p |
s
,
(2) | p |
sn1
c
2
sup
zD
[p(.)[d(.)
s
,
6.3 The Bergman Projection 145
for some constants c
1
and c
2
independent of p.
Proof. Let be a smooth, nonnegative, radially symmetric function supported in
the unit ball 1
n
in C
n
with
_
Bn
(n)d\
w
= 1. For . 1, let
z
(n) = c
2n
((.
n),c), where c = d(.). Clearly,
|
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)
z
(n) d\
w

| p |
l+n
|
z
|
ln
c
1
| p |
l+n
d(.)
l
.
Setting | = : +n, this proves (1).
For (2), notice that if C

0
(1), then by Taylors expansion and the Sobolev
embedding theorem, we have
[(.)[ c | |
s+n+1
d(.)
s
.
Hence,
| p |
sn1
= sup
C

0
(D)
||s+n+1=1

_
D
pd\
z

c
2
sup
zD
[p(.)[d(.)
s
.
This completes the proof of Lemma 6.3.10.
Proof of Theorem 6.3.8. If 1 satises 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 xed :+:
s
, Lemma 6.3.9 shows that there exists a positive integer n
t
s
= n
s+ms
such that
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
= | 1
s+ms
p |
s
|
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 and each . 1, we have by Lemma 6.3.10
|

1(., ) |
s
c
2
sup
wD

1(., n)

d(n)
m
C,
146 Boundary Regularity for on Pseudoconvex Domains
where : = :

= :

+n+1. Hence, by using the operator


s
constructed in Lemma
6.3.9, for p \
s+ns
(1) and . 1, we have

1p(.)

_
D
1(., n)
s
p(n)d\
w

_
D

1(., n)
s
p(n)d\
w

1(., ) |
s
|
s
p |
s
C | p |
s+ns
.
The dierentiation under the integral sign is justied, since

1(., n)
s
p(n)

C
1

1(., n)

|
s
p |
s
d(n)
m
C
2
|
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
zD

1)(.)

C | ) |
k(s)
,
for all multiindices with [[ :. Therefore,
|

1(., ) |
k(s)
= sup
C

0
(D)
||
k(s)
=1

_
D

1(., n)(n)d\
w

= sup
C

0
(D)
||
k(s)
=1

1(.)

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 xed n 1, let
w
(.) C

0
(1) be a smooth real-valued function
such that
w
(.) is radially symmetric with respect to the center n and
_
D

w
(.) d\
z
= 1. Since
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(., )
w
() d\

.
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 = 1
s
p.
Hence, we have H

(1) 1(\
s
0
(1)) for all real : 0.
Let be the set of functions C

0
(1) which are radially symmetric about
some point in 1 with
_
D
d\ = 1. Thus, 1 = 1(, n)[ n 1. We claim that
the linear span of is dense in \
s
0
(1) for each : 0. Let ) C

0
(1). Choose
a smooth nonnegative function from C

0
(C
n
) which is radially symmetric about
the origin with support contained in the unit ball and satisfying
_
C
n
(.)d\
z
= 1.
For c 0, set

(.) = c
2n
(.,c). Then )

= )

will converge to ) in \
s
0
(1).
Since
)

(.) = c
2n
_
C
n
)(n)
_
. n
c
_
d\
w
can be approximated by nite Riemann sums, this yields the density of span 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,


span1(, 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 classication 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 dening 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 rst 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
j
: 1
j
1, , = 1, 2,
satisfying
(1)
j
is smooth and plurisubharmonic on 1
j
,
(2)
j
< 0 on 1
j
and
j
vanishes on /1
j
,
(3) (
j
)
m
= :
j
c
K]z]
2
is smooth on 1
j
for some positive integer :, where
:
j
is a smooth dening function for 1
j
.
Property (3) immediately implies that
[
j
(.)[ d(., /1
j
)
1
m
, for . 1
j
.
Since ) : 1
1
1
2
is a biholomorphic map, both
2
) and
1
)
1
satisfy (1)
and (2). Thus, an application of the classical Hopf lemma (see [GiTr 1]) shows that
d(., /1
1
) [
2
)(.)[ d()(.), /1
2
)
1
m
and
d(n = )(.), /1
2
) [
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 n ( )) is bounded from
\
s+j(s)
0
(1
2
) to \
s
0
(1
1
).
Proof. It suces to show the estimate
| n ( )) |
W
s
0
(D1)
C | |
W
s+j(s)
0
(D2)
for all C

0
(1
2
). Write ) = ()
1
, , )
n
). For any multiindex with [[ :, we
have
1

(n ( ))) =

n 1

()) 1
1
)
i1
1
p
)
ip
,
where 1 i
1
, , i
p
n, and , ,
1
, ,
p
are multiindices with [[ [[,
[[ [[ [[ and

p
j=1
[
j
[ = [[ [[.
Since ) is a map between two bounded domains 1
1
and 1
2
in C
n
, the Cauchy
estimate implies that

n
.

(.)

d
1
(.)
(]]+1)
,
and

j
)
ij
.
j
(.)

C
j
d
1
(.)
]j]
,
6.3 The Bergman Projection 149
where the constant C

(C
j
) depends on 1
2
and the multiindex (
j
), and d
1
(.) =
d(., /1
1
). Hence, for . 1
1
,
[1

n 1
1
)
i1
1
p
)
ip
(.)[ Cd
1
(.)
(]]+1)
.
Also, for any C

0
(1
2
) and every / N, it follows from Taylors expansion and
the Sobolev embedding theorem that
[1

(n)[ C | |
k+]]+n+1
d
2
(n)
k
.
Thus, combining the preceding inequalities with Theorem 6.3.14, we obtain
[1

(n ( )))(.)[ d
1
(.)
(]]+1)
| |
k+]]+n+1
d
2
()(.))
k
| |
k+s+n+1
d
1
(.)
s1+
k
m
.
It is now clear by taking / = :(: +1) that the mapping n ( )) 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

, = 1, 2, is the Bergman projection on 1

.
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 so-called worm domains. Such domains
possess many pathological properties in complex analysis. We shall prove that such
domains do not always have plurisubharmonic dening 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

be the unbounded worm domain dened by


1

= (.
1
, .
2
) C
2
[ Re(.
1
c
ilog]z2]
2
) 0, [log[.
2
[
2
[ <

2
,
for ,2. Clearly, 1

is a Hartogs domain. Geometrically, if we use Re.


1
, Im.
1
and [.
2
[ as axes, then 1

can be visualized in 1
3
as an open half space in .
1
revolving
along the [.
2
[-axis when [.
2
[ ranges from exp(,2 +,4) to exp(,2 ,4).
To see that 1

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

is the intersection of two pseudoconvex domains.


Thus, it is pseudoconvex. As log[.
2
[ changes by a length of , we see that the half
plane Re(.
1
c
ilog]z2]
2
) rotates by an angle of 2.
To construct a bounded worm domain we shall rotate discs instead of half planes.
We dene

= (.
1
, .
2
) C
2
[ [.
1
+c
ilog]z2]
2
[
2
< 1, [log[.
2
[
2
[ <

2
.
Since
t

is dened by [.
1
[
2
+ 2Re(.
1
c
ilog]z2]
2
) < 0, locally, we can view
t

as dened 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
t

is pseudoconvex and
bounded. But it is not smooth at [log[.
2
[
2
[ =

2
. For each xed [log[.
2
[
2
[ <

2
,

is a disc of radius 1 centered at c


ilog]z2]
2
and (0, .
2
) /
t

.
To construct a smooth worm domain, we have to modify
t

. Let : 1 1 be
a xed smooth function with the following properties:
(1) (r) 0, is even and convex.
(2)
1
(0) = 1
/2
, where 1
/2
= [ +,2, ,2].
(3) there exists an o 0 such that (r) 1 if r < o or r o.
(4)
t
(r) ,= 0 if (r) = 1.
We note that (4) follows from (1) and (2). The existence of such a function is
obvious. For each ,2, dene
(6.4.1)

= (.
1
, .
2
) C
2
[ [.
1
+c
ilog]z2]
2
[
2
< 1 (log[.
2
[
2
).
Then, we have:
6.4 Worm Domains 151
Proposition 6.4.1. For each xed ,2,

is a smooth bounded pseudoconvex


domain in C
2
.
Proof. Clearly, by (3),

is bounded. For the smoothness of

, we need to show
that (.) ,= 0 at every boundary point ., where (.) = [.
1
+c
ilog]z2]
2
[
2
1 +
(log[.
2
[
2
) is the dening function for

. If (,.
1
)(.) = 0 at some boundary
point ., we get

.
1
(.) = .
1
+c
ilog]z2]
2
= 0.
Since (.) = 0, it follows that (log[.
2
[
2
) = 1 whenever (,.
1
)(.) = 0 at a
boundary point .. Now it is easy to see that (,.
2
)(.) ,= 0 by (4) at such points.
This proves the smoothness of

.
To see that

is pseudoconvex, we write
(.) = [.
1
[
2
+ 2Re(.
1
c
ilog]z2]
2
) +(log[.
2
[
2
).
Again locally,

can be dened by
[.
1
[
2
c
argz
2
2
+ 2Re(.
1
c
ilogz
2
2
) +(log[.
2
[
2
)c
argz
2
2
< 0.
The rst two terms are plurisubharmonic as before. We only need to show that the
last term (log([.
2
[
2
)c
argz
2
2
is plurisubharmonic. A direct calculation shows that
((log[.
2
[
2
)c
argz
2
2
) =
_
(log[.
2
[
2
)
_
c
argz
2
2
+(log[.
2
[
2
)c
argz
2
2
0,
since is convex and nonnegative from (1).

is dened locally by a plurisubhar-


monic function. Thus it is pseudoconvex with smooth boundary.
The following result shows that for each xed ,2 there is no C
2
global
dening function which is plurisubharmonic on the boundary of

.
Theorem 6.4.2. For any ,2, there is no C
2
dening function (.) for

such that (.) is plurisubharmonic on the boundary of

.
Proof. Let (.) be such a C
2
dening function for

that is plurisubharmonic
on the boundary /

. Then there is a C
1
positive function / dened in some
neighborhood of /

such that (.) = /. Let = (0, .


2
) C
2
[ [log[.
2
[
2
[ <
,2. A direct calculation shows that the complex Hessian of (.) acting on
any (, ) C
2
for any point j /

is given by
(6.4.2)
L
(z)
(j; (, )) = 2Re
_

_
i/
.
2
+
/
.
2
_
c
ilog]z2]
2
_
+
_
/ + 2Re
_
/
.
1
c
ilog]z2]
2
__
[[
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 , or equivalently,

.
2
(/c
ilog]z2]
2
) 0
on . Consequently,
p(.
2
) = /(0, .
2
)c
ilog]z2]
2
is a holomorphic function on the annulus . It follows that
p(.
2
)c
ilogz
2
2
= /(0, .
2
)c
2argz2
= c,
is also locally a holomorphic function on , 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 dened, C
1
positive function on , 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 is suciently large. To illustrate this, we
rst examine the Hartogs triangle
G = (.
1
, .
2
) [ [.
1
[ < [.
2
[ < 1.
By Cauchys 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
is any pseudoconvex domain containing G, then 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
k
such that G
k
and G =
k

k
. However, the Hartogs triangle is not smooth.
We next show that

does not have a pseudoconvex neighborhood base if


3,2. When 3,2,

contains the set


1 =(0, .
2
) [ log[.
2
[
2

(.
1
, .
2
) [ log[.
2
[
2
= or and [.
1
1[ < 1.
Any holomorphic function in a neighborhood of 1 extends holomorphically to the
set

1 = (.
1
, .
2
) [ log[.
2
[
2
and [.
1
1[ < 1.
Thus any holomorphic function in a neighborhood of

extends holomorphically
to


1. This implies that any pseudoconvex domain containing

contains

1.
6.5 Irregularity of the Bergman Projection on Worm Domains 153
Theorem 6.4.3. For 3,2, there does not exist a sequence
k
of pseudo-
convex domains in C
2
with


k
and

=
k

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

in the Sobolev spaces. We rst study the Bergman kernel


function 1(., n) on the unbounded worm domain 1

where 1

is dened in 6.4.
For each xed .
1
variable, the domain is a union of annuli in .
2
. Any holomorphic
function in 1

admits a Laurent expansion in .


2
. Using Fourier expansion, for any
) H(1

), we write
)(.) =
1
2

jZ
_
2
0
)(.
1
, c
i
.
2
)c
ij
d.
Let )
j
(.
1
, .
2
) =
1
2
_
2
0
)(.
1
, c
i
.
2
)c
ij
d. Then )
j
is holomorphic and )
j
(.
1
, c
i
.
2
)
= c
ij
)
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

and locally constant in [.


2
[. The Bergman space
H(1

) admits an orthogonal decomposition


H(1

) =
jZ
H
j
(1

).
Any ) in H
j
(1

) satises )(.
1
, c
i
.
2
) = c
ij
)(.
1
, .
2
). Denote by 1
j
the orthogonal
projection from 1
2
(1

) onto H
j
(1

). It follows that if ) H(1

), we have
1
j
)(.) = )
j
(.) =
1
2
_
2
0
)(.
1
, c
i
.
2
)c
ij
d,
and the Bergman kernel function 1
D

(., n) associated with 1

satises
1
D

(., n) =

jZ
1
j
(., n),
where 1
j
(., n) is the reproducing kernel for H
j
(1

). 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

= (.
1
, .
2
) C
2
[ [Im.
1
log[.
2
[
2
[ < ,2, [log[.
2
[
2
[ < ,2.
For each xed .
2
, 1
t

is an innite strip in .
1
. Thus, 1
t

is biholomorphically
equivalent to 1

via the mapping


: 1
t

(.
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

(., n) =
1
.
1
n
1
1
D

(
1
(.),
1
(n)).
Since 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
(
1
(.),
1
(n)),
where 1
t
j
is the reproducing kernel for the square integrable holomorphic functions
H on 1
t

satisfying H(.
1
, c
i
.
2
) = c
ij
H(.
1
, .
2
). The kernel 1
t
1
can be calculated
explicitly as follows:
For any H H
j
(1
t

), we may write H(.


1
, .
2
) = /(.
1
).
j
2
, where /(.
1
) is holo-
morphic in .
1
. For each 0, let o

be the strip on the complex plane dened


by
o

= . = r +ij C[ [j[ < .


It follows that
(6.5.2)
| H |
2
L
2
(D

)
=
_
D

[/(.
1
)[
2
[.
2
[
2j
dr
1
dj
1
dr
2
dj
2
= 2
_
]2logr]<

2
_
]y12logr]<

2
[/(.
1
)[
2
:
2j+1
dr
1
dj
1
d:
=
_
]s]<

2
_
]y1s]<

2
[/(.
1
)[
2
c
(j+1)s
dr
1
dj
1
d:
=
_

_
S

[/(.
1
)[
2
c
(j+1)s

2
(j
1
:)

2
(:) dr
1
dj
1
d:
=
_
S

[/(.)[
2

j
(j) drdj,
where
j
(j) = (c
(j+1)s

2
)

2
(j), ,2 and

is the characteristic function


on 1

= (, ). Let (j) be a continuous positive bounded function on the interval


1

= j 1 [ [j[ < . Denote by H(o

, ) the weighted Bergman space on o

dened by
H(o

, ) = ) O(o

)[ | ) |
2

=
_
S

[)(.)[
2
(j) drdj < .
To compute the kernel 1
t
1
, it suces to compute the Bergman kernel in one
variable on a strip o

with weight =

2
(j) and the kernel 1
t
1
(., n)
is given by 1
t
1
= 1

(.
1
, n
1
),.
2
n
2
. The next lemma allows us to compute the
weighted Bergman kernel function 1

(., n) on o

.
6.5 Irregularity of the Bergman Projection on Worm Domains 155
Lemma 6.5.1. For each 0, let (j) be a continuous positive bounded function
on the interval 1

= j 1 [ [j[ < . Then the weighted Bergman kernel function


1

(., n) on o

is given by
(6.5.3) 1

(., n) =
1
2
_
R
c
i(zw)

(2i)
d,
where

is the Fourier transform of if (j) is viewed as a function on 1 that
vanishes outside 1

.
Proof. For ) H(o

, ) we dene the partial Fourier transform



) of ) with respect
to r by

)(, j) =
_
R
)(r +ij)c
ix
dr.
It is easily veried by Cauchys theorem that

)(, j) = c
y
)
0
(), where

)
0
() =

)(, 0). Thus from Plancherels theorem,


(6.5.4)
| ) |
2

= (2)
1
_
RI

c
2y
[

)
0
() [
2
(j)ddj
= (2)
1
_
R
[

)
0
() [
2

(2i)d.
For ) H(o

, ) and . o

, we have
_
R

)
0
()c
iz
d = 2)(.)
= 2
_
S

)(n)1

(n, .)(j) drdj


=
_

_
R
c
2y

)
0
()

((, 0), .)(j) ddj,


where n = r +ij. It follows that
c
iz
=

1

((, 0), .)
_

c
2y
(j) dj,
and

((, 0), .) =
c
iz

(2i)
.
Finally, by the Fourier inversion formula, we obtain
1

(., n) =
1
2
_
R
c
i(zw)

(2i)
d.
Here we note that

(2i) is real. This proves the lemma.
We next apply Lemma 6.5.1 to the piecewise linear weight
(j) =

2
(j).
156 Boundary Regularity for on Pseudoconvex Domains
Lemma 6.5.2. For

2
, if (j) =

2
(j), then

(2i) =
sinh
_
(2 )
_
sinh()

2
,
and
(6.5.5) 1

(., n) =
1
2
2
_
R

2
c
i(zw)
sinh
_
(2 )
_
sinh()
d.
Proof. If (j) =

(j) for some 0, then

(2i) =
_

c
ix(2i)
dr =
_

c
2x
dr =
sinh(2)

.
Hence, for the piecewise linear weight (j) =

2
(j), we have

(2i) =
sinh
_
(2 )
_
sinh()

2
,
and (6.5.5) now follows from Lemma 6.5.1.
We observe that
2
,sinh
_
(2 )
_
sinh has poles at nonzero integer multiples
of i,(2) and i. Let us rst assume that , and set

= ,(2) so that

< 1. Then, via a standard contour integration, one can obtain the asymptotic
expansion of the weighted Bergman kernel function 1

(., n) and see that it is in


fact dominated by the residue of p() =
2
c
i(zw)
,sinh
_
(2 )
_
sinh() at the
rst pole

i.
Lemma 6.5.3. Let and (j) =

2
(j). Then
(6.5.6) 1

(., n) = c

(zw)
+O(c

(zw)
)
for Re(. n) 0 and
(6.5.7) 1

(., n) = c

(zw)
+O(c

(zw)
)
for Re(. n) < 0, where c

3
,(
2
sin

) and j

= min(2

, 1). Furthermore,
given any small positive c, the expansion in (6.5.6) or (6.5.7) is uniform for any
., n o

.
Proof. Fix / 0 so that /i is the midpoint between the second and third poles of
p(). Denote by
N
the rectangular contour with vertices and + i/. Let
us rst assume that 2

< 1. Then, we have 2

< / < 1 and


_
N
N

2
c
i(zw)
sinh
_
(2 )
_
sinh()
d + 1
N
+ 1
N
+ J
N
= 2i(Res p() at

i and 2

i),
6.5 Irregularity of the Bergman Projection on Worm Domains 157
where Res p() denotes the residues of p() and
1
N
= i
_
h
0
( +ij)
2
c
i(zw)(N+iy)
sinh
_
(2 )( +ij)
_
sinh( +ij)
dj,
1
N
= i
_
h
0
( +ij)
2
c
i(zw)(N+iy)
sinh
_
(2 )( +ij)
_
sinh( +ij)
dj,
and
J
N
=
_
N
N
(r +i/)
2
c
i(zw)(x+ih)
sinh
_
(2 )(r +i/)
_
sinh(r +i/)
dr.
A direct calculation shows that
Res
=

i
p() =

3
c

(zw)
isin(

)
,
and
Res
=2

i
p() =
4

3
c
2

(zw)
isin(2

)
.
For any ., n o

, we write . n = n + i with n 0. Hence, we have


[[ 2( c) and
[1
N
[
_
h
0
(
2
+j
2
)c
vNuy
c
2N
dj

2
+ 1
c
2N
.
It follows that 1
N
converges to zero uniformly for any ., n o

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
vxuh
dr +
_
N
1
(r
2
+/
2
)c
vxuh
c
2x
dr
+
_
1
N
(r
2
+/
2
)c
vxuh
c
2x
dr
c
2

u
_
1 +
_
N
1
r
2
+/
2
c
2x
dr
_
.
It follows by letting tend to innity that
_

2
c
i(zw)
sinh
_
(2 )
_
sinh()
d
=
2

3
c

(zw)
sin

3
c
2

(zw)
sin(2

)
+
_

(r +i/)
2
c
i(zw)(x+ih)
sinh
_
(2 )(r +i/)
_
sinh(r +i/)
dr
=
2

3
sin

(zw)
+O(c
2

(zw)
).
158 Boundary Regularity for on Pseudoconvex Domains
Clearly, the estimate is uniform for all ., n o

with Re(. n) 0. This proves


the case for 2

< 1.
For cases 2

= 1 and 2

1, a similar argument applies.


If Re(. n) < 0, we take the rectangular contour
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

(.
1
, n
1
),.
2
n
2
,
where 1

(.
1
, n
1
) is calculated in Lemma 6.5.3. If , then (6.5.7) shows that
1
t
1
(., n) = c

(z1w1)
.
2
n
2
+O(c

(z1w1)
)
for Re(.
1
n
1
) < 0. Hence
(6.5.8) 1
1
(., n) = c

1
1
n

1
1
.
1
2
n
1
2
+
1
.
1
n
1
O
_
_
.
1
n
1
_

_
for [.
1
[ < [n
1
[. The expansion in (6.5.7) is uniform on o

for any small positive


c. Thus, for xed n, we have for any : N,
[Re(.
1
c
ilog]z2]
2
)[
s
_

.
1
_
m
1
1
(., n) , 1
2
(1

), for : :

.
It follows that
(6.5.9) [Re(.
1
c
ilog]z2]
2
)[
s
_

.
1
_
m
1
D

(., n) , 1
2
(1

),
for : :

. Estimate (6.5.9) also holds for ,2 < . When ,2 <


< , (6.5.9) can be obtained by examining higher order terms in the asymptotic
expansion of 1

.
When = , we compute the residue at the double pole i of (6.5.5) to obtain
1

(.
1
, n
1
) =
2
(.
1
+ n
1
2)c
(z1 w1)
+O(c
2(z1 w1)
)
for Re(.
1
n
1
) < 0 and
1
1
(., n) =
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 = 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 dening
function (r). Then, for each : 0, the \
s
norm of a harmonic function ) is
equivalent to the 1
2
norm of [[
s
) on 1.
We rst observe the following result for the Bergman projection on the un-
bounded worm domain 1

.
Proposition 6.5.5. For each ,2, condition 1 does not hold on 1

. Fur-
thermore, the Bergman projection 1

on 1

does not map C

0
(1

) into \
k
(1

)
when / ,(2 ).
Proof. Let n 1

. We choose a real-valued function ) C

0
(1

) such that )
depends on [. n[ and
_
D

) = 1. Using the same argument as in Corollary 6.3.11,


we obtain
1

) = 1
D

(, n).
Since 1
D

(, n) , C

(1

), condition 1 fails on 1

.
Let
A
= . C
n
[ [.[ < be a large ball for 0. From (6.5.9), we have
(6.5.10) [

[
s
_

.
1
_
m
1

) , 1
2
(1


A
)
for : :

, where

= Re(.
1
c
ilog]z2]
2
). If 1

) \
k
(1


A
), choose a
positive integer : / and let : = : / :

. Using Lemma 6.5.4, we have


[

[
s

m
1

) 1
2
(1


A
), a contradiction. Thus 1

) , \
k
(1

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

.
Theorem 6.5.6. For each ,2, the Bergman projection 1 on

does not
map \
k
(

) into \
k
(

) when / ,(2 ).
Proof. Assume on the contrary that the Bergman projection 1 maps \
k
(

) into
\
k
(

) with the estimate


(6.5.11) | 1) |
W
k
(

)
C
k
| ) |
W
k
(

)
for ) \
k
(

) and / ,(2 ) =

.
For any j 1, let

be the dilation dened by

: C
2
C
2
(.
1
, .
2
) (j.
1
, .
2
).
Denote by
,
=

),
t
,
=

(
t

) where
t

is dened in Section 6.4. Then

t
,

,
and
t
,
1

. Let T

be the pullback of the 1


2
functions on
,
,
i.e.,
T

: 1
2
(
,
) 1
2
(

)
) )

.
160 Boundary Regularity for on Pseudoconvex Domains
A direct calculation shows that
|
_

.
_

_

.
_

) |
L
2
(

)
= j
1+11
|
_

.
_

_

.
_

) |
L
2
(
,
)
,
where = (
1
,
2
) and = (
1
,
2
) are multindices. Thus we have
(6.5.12) | T

) |
W
l
(

)
j
l1
| ) |
W
l
(
,
)
,
when | is a nonnegative integer. Then, by interpolation, it holds for all real | 0.
Let 1

be the Bergman projection associated with


,
. Then
(6.5.13) 1

= T
1

1T

.
From the denition (6.4.1) of

we see that the dening function (.) coincides


with [.
1
[
2
+ 2Re(.
1
c
ilog]z2]
2
) when log[.
2
[
2
1

2
. Let

(.) = j (

)
1
so
that

= 2Re(.
1
c
ilog]z2]
2
) as j , where

is a dening function for


1

. Write / = ::, where : is an integer and : 0. For any ) C

0
(
t
,
)
C

0
(
,
), we have using (6.5.11)-(6.5.13) and Lemma 6.5.4,
(6.5.14)
| [

[
s
_

.
1
_
m
1

) |
L
2
(
,
)
= | [

[
s
_

.
1
_
m
(T

)
1
1T

) |
L
2
(
,
)
= j
sm+1
| [[
s
_

.
1
_
m
1T

) |
L
2
(

)
Cj
1k
|
_

.
1
_
m
1T

) |
W
s
(

)
Cj
1k
| 1T

) |
W
k
(

)
Cj
1k
| T

) |
W
k
(

)
C | ) |
W
k
(
,
)
,
where the constant C is independent of j. We claim that
1

) 1

) weakly in 1
2
(C
2
),
where 1

) is the Bergman projection of ) on 1

, 1

) = 0 outside 1

and we
have set 1

) = 0 outside
,
. Assuming the claim, It follows from (6.5.14) that
(6.5.15) | [

[
s
_

.
1
_
m
1

) |
L
2
(D

)
C | ) |
W
k
(D

)
for any ) C

0
(1

). This contradicts (6.5.10) and the theorem is proved.


Thus, it remains to prove the claim. Since
| 1

) |
L
2
(C
2
)
| ) |
L
2
(C
2
)
,
there exists a subsequence of 1

) that converges weakly to / 1


2
(C
2
). Since

t
,
1

, / is holomorphic in 1

. Also / vanishes outside 1

since every
compact subset outside 1

is outside
,
for suciently large j. To prove that
/ = 1

), we need to show that ) / H(1

). Choose ` 1 so that
,
1
M
for all j. Obviously ) 1

) H(1
M
). Therefore, by passing to the limit, we
obtain that ) / H(1
M
). The claim will be proved by the following density
result:
Notes 161
Lemma 6.5.7. For each ` 1, the space H(1
M
) is dense in H(1

).
Proof. It suces to show that each H
j
(1
M
) is dense in H
j
(1

), or equivalently,
H
j
(1
t
M
) is dense in H
j
(1
t

). From (6.5.2) and (6.5.4) we have for any ) H


j
(1
t

),
| ) |
2
L
2
(D

)
= (2)
1
_
R
[

)
0
() [
2

j
(2i)d,
where

)
0
() is the partial Fourier transform of ) evaluated at j = 0 and
j
(j) =
(c
(j+1)()

2
)

2
(j). Thus, the space H
j
(1
t

) is isometric via the Fourier


transform to the space of functions on 1 which are square integrable with respect
to the weight

j
(2i) =
sinh
_
(2 )
_
(
j+1
2
)
_
sinh()

_
(
j+1
2
)
_ .
Since C

0
(1) is dense in the latter space for any value of , 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 ,2, the -Neumann operator on

does not map


\
k
(0,1)
(

) into \
k
(0,1)
(

) when / ,(2 ).
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 su-
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 dening 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 dening function. For a convex
domain in dimension two, a dierent 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 rst 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 sucient
condition related to the De Rham cohomology on the set of innite 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 nite type. The concept
of nite type on a pseudoconvex domain in C
2
, using the Lie brackets of complex
tangential vector elds, was rst introduced by J. J. Kohn [Koh 4]. Subsequently,
162 Boundary Regularity for on Pseudoconvex Domains
J. J. Kohn introduced subelliptic multipliers and nite ideal type in [Koh 8] and he
proved that nite ideal type condition is sucient 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. DAngelo [DAn 1,2] proposed a denition
of nite type in C
n
. The necessity of nite 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, Kohns 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 dened in [Cat 4] his own notion of nite 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 armative result of global analytic regularity
of the -Neumann problem had been obtained by M. Derridj and D. S. Tartako
[DeTa 1] and G. Komatsu [Kom 1]. Local analytic hypoellipticity of the -Neumann
problem on strongly pseudoconvex domains was proved by D. S. Tartako [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 eld 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 dierentiability 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
s
in Lemma
6.3.8 was rst 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 rst 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 Barretts 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. Barretts 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. Stensnes [FoSte 1] for
counterexamples on pseudoconvex domains. For recent results on the -Neumann
problem on Lipschitz pseudoconvex domains, see the papers by Bonami-Charpentier
[BoCh 1], Henkin-Iordan [HeIo 1], Henkin-Iordan-Kohn [HIK 1], Michel-Shaw [MiSh
1] and Straube [Str 2]. Holder and 1
p
estimates of the -Neumann problem on
pseudoconvex domains of nite type in C
2
have been discussed in Chang-Nagel-
Stein [CNS 1], Feerman-Kohn [FeKo 1]. Holder or 1
p
estimates for the -Neumann
problem on nite type pseudoconvex domains in C
n
for n 3 are still unknown.
164 Boundary Regularity for on Pseudoconvex Domains
CHAPTER 7
CAUCHY-RIEMANN MANIFOLDS
AND THE TANGENTIAL CAUCHY-RIEMANN COMPLEX
Let ` be a smooth hypersurface in a complex manifold. The restriction of the
complex to ` naturally induces a new dierential complex. This complex is
called the tangential Cauchy-Riemann complex or the
b
complex. The tangential
Cauchy-Riemann complex, unlike the de Rham or the complex, is not elliptic. In
general, it is an overdetermined system with variable coecients.
We have seen in Chapter 3 that the tangential Cauchy-Riemann 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 dieren-
tial equations. The tangential Cauchy-Riemann equation associated with a strongly
pseudoconvex hypersurface in C
2
provides a nonsolvable rst order partial dieren-
tial equation with variable coecients. 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 rst dene Cauchy-Riemann manifolds and the tangen-
tial Cauchy-Riemann complex both extrinsically and intrinsically. The Levi form of
a Cauchy-Riemann 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 dierential operator with constant coecients 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
complexied tangent bundle over `. A C1 structure on ` is dened as follows.
7.1 CR Manifolds 165
Denition 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 Cauchy-Riemann structure
T
1,0
(`) if the following conditions are satised:
(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
(l, T
1,0
(`)), the Lie bracket
[A
1
, A
2
] is still in (l, T
1,0
(`)), where l is any open subset of ` and
(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 Denition 7.1.1 is void when n = 2. The most natural C1 manifolds
are those dened by smooth hypersurfaces in C
n
.
Example 7.1.2. Let : C
n
1 be a real-valued smooth function. Suppose that
the dierential d does not vanish on the hypersurface ` = . C
n
[ (.) = 0.
Then ` is a smooth manifold with dim
R
` = 2n 1. Dene 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
from ` to is called a C1 mapping if

1 is a smooth section of T
1,0
() for any
smooth section 1 in T
1,0
(`). Furthermore, if has a smooth C1 inverse mapping

1
, then we say that (`, T
1,0
(`)) is C1 dieomorphic 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 dieomorphic to via a mapping . Then induces a C1
structure on , namely,

T
1,0
(`), so that becomes a C1 dieomorphism from
(`, T
1,0
(`)) onto (,

T
1,0
(`)), where

is the dierential map induced by .


Proof. We need to check the integrability condition on

T
1,0
(`). However, this
follows immediately from the integrability condition on T
1,0
(`) and the fact that
[

A
1
,

A
2
] =

[A
1
, A
2
] for any smooth vector elds A
1
, A
2
dened on `.
Denition 7.1.4. A smooth function p dened 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 denition coincides
with Denition 3.0.1. If, in Denition 7.1.4, p is just a distribution, then 1p should
be interpreted in the sense of distribution.
166 Cauchy-Riemann Manifolds
7.2 The Tangential Cauchy-Riemann Complex
Let ` be a hypersurface in a complex manifold. The complex restricted to `
induces the tangential Cauchy-Riemann complex, or the
b
complex. In fact, the
tangential Cauchy-Riemann complex can be formulated on any C1 manifold. There
are two dierent approaches in this setting. One way is to dene the tangential
Cauchy-Riemann 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 Cauchy-Riemann complex
can also be dened extrinsically via the ambient complex structure.
First, we assume that ` is a smooth hypersurface in C
n
, and let : be a dening
function for `. In some open neighborhood l of `, let 1
p,q
, 0 j, n, be the
ideal in
p,q
(C
n
) such that at each point . l the ber 1
p,q
z
is generated by : and
:, namely, each element in the ber 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

p,q
(C
n
)[
M
and 1
p,q
[
M
the restriction of
p,q
(C
n
) and 1
p,q
respectively to `. Then,
we dene

p,q
(`) = the orthogonal complement of 1
p,q
[
M
in
p,q
(C
n
)[
M
.
We denote by c
p,q
the space of smooth sections of
p,q
(`) over `, i.e., c
p,q
(`) =
(`,
p,q
(`)). Let denote the following map
(7.2.1) :
p,q
(C
n
)
p,q
(`),
where is obtained by rst restricting a (j, )-form in C
n
to `, then projecting
the restriction to
p,q
(`). One should note that
p,q
(`) is not intrinsic to `,
i.e.,
p,q
(`) is not a subspace of the exterior algebra generated by the complexied
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 Cauchy-Riemann operator

b
: c
p,q
(`) c
p,q+1
(`)
is now dened as follows: For any c
p,q
(`), pick a smooth (j, )-form
1
in
C
n
that satises
1
= . Then,
b
is dened to be
1
in c
p,q+1
(`). If
2
is
another (j, )-form in C
n
such that
2
= , then

2
= :p +: /,
for some (j, )-form p and (j, 1)-form /. It follows that
(
1

2
) = :p +: p : /,
and hence,
(
1

2
) = 0.
7.2 The Tangential Cauchy-Riemann Complex 167
Thus, the denition of
b
is independent of the choice of
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,n1
(`) 0.
For the intrinsic approach, we will assume that (`, T
1,0
(`)) is an orientable C1
manifold of real dimension 2n1 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 (`) the orthogonal complement of T
1,0
(`) T
0,1
(`).
It is easily seen that (`) 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 denition it
means that forms in T
1,0
(`) annihilate vectors in T
0,1
(`) (`) and forms in
T
0,1
(`) annihilate vectors in T
1,0
(`) (`). Dene the vector bundle
p,q
(`),
0 j, n 1, by

p,q
(`) =
p
T
1,0
(`)
q
T
0,1
(`).
This can be identied with a subbundle of
p+q
CT

(`). It follows that


p,q
(`)
dened in this way is intrinsic to `. Denote by c
p,q
the space of smooth sections
of
p,q
(`) over `, i.e., c
p,q
(`) = (`,
p,q
(`)). We dene the operator

b
: c
p,q
(`) c
p,q+1
(`)
as follows: If c
p,0
,
b
is dened by

b
, (\
1
\
p
) 1) = 1, \
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 c
p,q
(`), we dene

b
, (\
1
\
p
) (1
1
1
q+1
))
=
1
+ 1
_
q+1

j=1
(1)
j+1
1
j
, (\
1
\
p
) (1
1

1
j
1
q+1
))
+

i<j
(1)
i+j
, (\
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
p,q
be the projection from
p+q
CT

(`) onto
p,q
(`), then
b
=
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 denition of
b
, and it is standard to see that the following
sequence
0 c
p,0
(`)

b
c
p,1
(`)

b


b
c
p,n1
(`) 0,
168 Cauchy-Riemann Manifolds
forms a complex, i.e.,
2
b
= 0.
Notice that j plays no role in the formulation of the tangential Cauchy-Riemann
operators. Thus, it suces 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 Cauchy-Riemann complex on ` can be dened either extrinsically or
intrinsically. These two complexes are dierent, but one can easily show that they
are isomorphic. Thus, if the C1 manifold is embedded, we shall not distinguish
the extrinsic or intrinsic denitions of the tangential Cauchy-Riemann complex.
The operator
b
is a rst order dierential 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
n1
be a local basis for smooth sections of T
1,0
(`) over some open
subset l `, so 1
1
, , 1
n1
is a local basis for T
0,1
(`) over l. Next we
choose a local section T of CT(`) such that 1
1
, , 1
n1
, 1
1
, , 1
n1
and T
span CT(`) over l. We may assume that T is purely imaginary.
Denition 7.2.1. The Hermitian matrix (c
ij
)
n1
i,j=1
dened 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
n1
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
n1
and T. Hence,
after changing T to T, it makes sense to consider positive deniteness of the matrix
(c
ij
).
Denition 7.2.2. The C1 structure is called pseudoconvex at j ` if the matrix
(c
ij
(j)) is positive semidenite after an appropriate choice of T. It is called strictly
pseudoconvex at j ` if the matrix (c
ij
(j)) is positive denite. 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 at.
7.2 The Tangential Cauchy-Riemann 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

dening 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 elds near j on the boundary.
If 1 =

n
j=1
o
j
(,.
j
) is a tangential type (1, 0) vector eld near j, then we have

n
j=1
o
j
(:,.
j
) = 0 on /1 and 1 = (:,.
n
)
1

n1
j=1
o
j
1
j
on /1. Hence, if we
let = : :, we obtain
n1

i,j=1
c
ij
o
i
o
j
=
n1

i,j=1
, [1
i
, 1
j
])o
i
o
j
=
n1

i,j=1
(1
i
, 1
j
) 1
j
, 1
i
) 2d, 1
i
1
j
))o
i
o
j
=
n1

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 denitions. 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 ,
1
, ,
n1
be the one forms dual to T, 1
1
, , 1
n1
which
are dened as above. The vector eld T is chosen so that the Levi form is positive
denite. Then we consider the following 2n 1 form
(7.2.5)
1

1

n1

n1
.
It is not hard to see that the 2n1 form (7.2.5) generated by other bases will dier
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 Cauchy-Riemann Manifolds
7.3 Lewys Equation
In this section, we shall present a partial dierential operator of order one with
variable coecients 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 dierential operator. Since this operator arises from the
tangential Cauchy-Riemann operator on the boundary of a strongly pseudocon-
vex domain, this discovery also inspires an intensive investigation of the tangential
Cauchy-Riemann operator.
Let
n
be the Siegel upper half space dened by
(7.3.1)
n
= (.
t
, .
n
) C
n
[ Im.
n
[.
t
[
2
,
where .
t
= (.
1
, , .
n1
) and [.
t
[
2
= [.
1
[
2
+ + [.
n1
[
2
. When n = 1,
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
n
, i.e.,
(7.3.2)
: 1
n

n
. n = (.)
= i
_
c
n
+.
1 .
n
_
=
_
i.
1
1 .
n
, ,
i.
n1
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 elds on the boundary
/
n
and
[1
j
, 1
k
] = 2i
jk

t
,
where .
n
= t + i: and
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

n
is a strongly pseudoconvex domain. Furthermore, the boundary /
n
can be
identied with H
n
= C
n1
1 via the map
: (.
t
, t +i[.
t
[
2
) (.
t
, t).
Therefore, a C1 structure can be induced via on H
n
by
(7.3.4) 7
k
=

1
k
=

.
k
+i.
k

t
for / = 1, , n1. Thus, if ) =

n1
j=1
)
j

j
is a
b
-closed (0, 1)-form on H
n
, where

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 Lewys 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 Cauchy-Riemann operator is generated by
(7.3.5) 1 =

.
2i.

n
,
with n = t+i:, and the corresponding operator, denoted by 7, via the identication
on H
2
is
(7.3.6) 7 =

.
i.

t
,
where (., t) with . = r +ij are the coordinates on H
2
= C1. The coecients of
the operator 7 dened 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 real-valued 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
i
, t)
with : < 1 and [t[ < 1 for some 1 0. Set : = :
2
. Consider the function \ (:, t)
dened 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.
= 2i:
2
)(t) 2

t
_
2
0
_
r
0
c
i
n(c
i
, t)dd.
172 Cauchy-Riemann Manifolds
Hence,
\
:
(:, t) = 2i)(t) 2

t
_
1
2:

:
_
2
0
_
r
0
c
i
n(c
i
, t)dd
_
= 2i)(t)
1
:

t
_
2
0
:c
i
n(:c
i
, t):d
= 2i)(t) +i

t
\ (:, t).
Set
1(t) =
_
t
0
)() d.
Then, we obtain
_

t
+i

:
_
(\ (:, t) + 21(t)) = 0,
which implies that the function l(:, t) = \ (:, t) + 21(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 real-valued boundary value 21(t). Hence, by the Schwarz
reection 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 Dierential Operators with Constant Coecients
In contrast to the nonsolvable operator (7.3.6), we shall present in this section
a fundamental positive result in the theory of partial dierential equations which
asserts the existence of a distribution fundamental solution to any linear partial
dierential operator with constant coecients. It follows by convolution that every
partial dierential operator with constant coecients is locally solvable.
Theorem 7.4.1 (Malgrange, Ehrenpreis). Let
1 =

]]k
o

x
be a partial dierential operator with constant coecients on 1
n
, where 1

x
=
(,r
1
)
1
(,r
n
)
n
for any multiindex = (
1
, ,
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 p() of p is dened by
p() =
_
R
n
c
ix
p(r) dr,
7.4 Linear Partial Dierential Operators with Constant Coecients 173
where r = r
1

1
+ + r
n

n
. Note rst that, via a rotation of coordinates
and multiplying 1 by a constant, we may assume that the corresponding Fourier
transform j() of the operator 1 is
j() =
k
n
+
k1

j=0
o
j
(
t
)
j
n
,
where = (
t
,
n
) with
t
1
n1
, and o
j
(
t
) is a polynomial in
t
for 0 , / 1.
Next, we complexify j(); namely, we view as a variable in C
n
. Hence, for each

t
1
n1
, j(
t
,
n
) is a polynomial of degree / in
n
. Let
1
(
t
), ,
k
(
t
) be its
zeros, arranged so that if i ,, Im
i
(
t
) Im
j
(
t
), and Re
i
(
t
) Re
j
(
t
) if
Im
i
(
t
) = Im
j
(
t
). One sees easily that these / functions Im
j
(
t
) are continuous
in
t
.
We then need to construct a measurable function
: 1
n1
[/ 1, / + 1]
such that for all
t
1
n1
, we have
min[(
t
) Im
j
(
t
)[ : 1 , / 1.
Set n
0
(
t
) = / 1 and n
k+1
(
t
) = / + 1. For 1 , /, dene
n
j
(
t
) = maxminIm
j
(
t
), / + 1, / 1.
The functions n
j
(
t
) are continuous in
t
, so the sets
\
j
=
t
: n
j+1
(
t
) n
j
(
t
) 2,
for , = 0, , /, are measurable. It is clear that
k
j=0
\
j
is a covering of 1
n1
. Thus,
we can construct disjoint measurable subsets \
j
\
j
which still cover 1
n1
. Dene
(
t
) =
1
2
(n
j+1
(
t
) +n
j
(
t
)),
if
t
\
j
. This completes the construction of (
t
).
We dene /(r) by
/(r) =
1
(2)
n
_
R
n1
_
Imn=(

)]
c
ix
_

)()
j()
_
d
n
d
t
.
The key is to observe that, as [Re[ ,

)() is rapidly decreasing whereas Im
remains bounded, and to see that the line Im
n
= (
t
) in the
n
-plane has distance
at least one from any zero of j() and at most / +1 from the real axis. Hence, the
integrand is bounded and rapidly decreasing at innity, so the integral is absolutely
convergent. The same reasoning shows that we can dierentiate under the integral
174 Cauchy-Riemann Manifolds
sign as often as we please. It follows that /(r) is smooth. Finally, we apply 1 to
/(r) and get
1/(r) =
1
(2)
n
_
R
n1
_
Imn=(

)]
c
ix

)() d
n
d
t
.
The integrand on the right-hand side is an entire function which is rapidly decreasing
as [Re
n
[ . Therefore, by Cauchys 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 dierential
operator with constant coecients, then for any given function )(r) which is smooth
near some point r
0
, we can nd 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 at 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
n1
, Im1
n1
, where 1
1
, , 1
n1
is a local
basis of T
1,0
(`) near j. Thus there exist local coordinates (r
1
, , r
2n2
, t) such
that the vector elds Re1
1
, Im1
1
, , Re1
n1
, Im1
n1
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 Newlander-Nirenberg theorem (Theorem 5.4.4), for n 3,
to show that ` is locally foliated by complex submanifolds of complex dimension
n1. In this case, the local solvability of the tangential Cauchy-Riemann 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 Cauchy-Riemann complex was rst 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.
Hormander [Hor 1,7] states that the tangential Cauchy-Riemann equation on a real
three dimensional C1 manifold is not locally solvable if it is not Levi at. For a
proof of the Cauchy-Kowalevski 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 dierential equation
which is a sum of squares of vector elds. The other is the
b
-Laplacian on a C1
manifold. We use pseudodierential operators to study both operators.
For this purpose, we shall briey review the denitions and basic properties
of the simplest pseudodierential operators. Using pseudodierential operators,
Hormanders theorem on the hypoellipticity of sums of squares of vector elds will
be discussed in Section 8.2.
The
b
-Laplacian,
b
, is not elliptic. There is a one-dimensional characteristic
set. However, under certain conditions, one is able to establish the 1,2-estimate for
the
b
operator via potential-theoretic methods and pseudodierential operators. In
the last two sections of this chapter, the 1,2-estimate 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 Pseudodierential Operators
We rst introduce some simple pseudodierential operators. Let o be the Schwar-
tz space in 1
n
. For the denitions 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
denition:
Denition 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 denition, 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 dierential operator 1

with [[ = : is of order :, where


= (
1
, ,
n
) and 1

= 1
1
x1
1
n
xn
.
For any : 1, we dene
s
: o o by

s
n(r) =
1
(2)
n
_
R
n
c
ix
(1 +[[
2
)
s
2
n() d
8.1 Pseudodierential Operators 177
where n is the Fourier transform of n. Here, (
s
) = (1+[[
2
)
s
2
is called the symbol
of
s
. Obviously, the operator
s
is of order :.
s
is called a pseudodierential op-
erator of order :. We should think of
s
as a generalization of dierential operators
to fractional and negative order.
The purpose of this section is to prove some basic properties of the commutators
between
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 Plancherels theorem to study the commutators of
s
and functions
in o. We rst 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 +[[
2
)
s
2
pn() =
1
(2)
n
_ _
1 +[[
2
1 +[[
2
_
s
2
p( )(1 +[[
2
)
s
2
n() d.
Now we view
_
(1 +[[
2
),(1 +[[
2
)
_
s/2
p( ) as a kernel 1(, ) and set )() =
(1 +[[
2
)
s/2
n(). Lemma 8.1.2 shows that
[1(, )[ (1 +[ [
2
)
|s|
2
[ p( )[.
Since p o, it follows that p o and that the hypotheses of Theorem B.10 in the
Appendix are satised 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) [
s
, p] is of order : 1,
(2) [
r
, [
s
, p]] is of order : +: 2,
(3) [[
s
, p], /] is of order : 2.
Proof. The proof of the theorem will proceed exactly as in Lemma 8.1.3. For (1) it
suces to show that
r
[
s
, p]
1rs
is of order zero for any : 1. Let n o, and
set ) =
r
[
s
, p]
1rs
n. A direct calculation shows that

)() =
1
(2)
n
_
1(, ) n() d,
where
1(, ) =
(1 +[[
2
)
r
2
(1 +[[
2
)
r+s1
2
_
(1 +[[
2
)
s
2
(1 +[[
2
)
s
2
_
p( ).
178 Subelliptic Estimates for Second Order Dierential Equations and
b
From the mean value theorem, we have
[(1 +[[
2
)
s
2
(1 +[[
2
)
s
2
[ [ [
_
(1 +[[
2
)
s1
2
+ (1 +[[
2
)
s1
2
_
for all , 1
n
, hence, by Lemma 8.1.2 we obtain
[1(, )[ [ [
_
_
1 +[[
2
1 +[[
2
_
r+s1
2
+
_
1 +[[
2
1 +[[
2
_
r
2
_
[ p( )[
[ [
_
(1 +[ [
2
)
|r+s1|
2
+ (1 +[ [
2
)
|r|
2
_
[ p( )[.
Since p o, so is p o. Therefore, by Theorem B.10 in the Appendix, (1) is
proved.
To prove (2), for n o, a similar calculation shows
([
r
, [
s
, p]]n)

() =
1
(2)
n
_
1(, ) n() d,
where
1(, ) =
_
(1 +[[
2
)
r
2
(1 +[[
2
)
r
2
_ _
(1 +[[
2
)
s
2
(1 +[[
2
)
s
2
_
p( ).
Here ([
r
, [
s
, p]]n)

() denotes the Fourier transform of [


r
, [
s
, p]]n. It follows
now from the same estimates as in (1) that
[1(, )[ [ [
2
_
(1 +[[
2
)
r1
2
+ (1 +[[
2
)
r1
2
_

_
(1 +[[
2
)
s1
2
+ (1 +[[
2
)
s1
2
_
[ p( )[,
which implies easily that [
r
, [
s
, p]] is of order : +: 2. This proves (2).
For (3), let n o. It is easy to get from Taylors expansion that
([
s
, p]n)

() =
1
(2)
n
_
_
(1 +[[
2
)
s
2
(1 +[[
2
)
s
2
_
p( ) n() d
=
n

j=1
1
(2)
n
_
(
j

j
)

j
(1 +[[
2
)
s
2
p( ) n() d
+
1
(2)
n
_
O(, ) p( ) n() d
=

T
1
n +

T
2
n,
where O(, ) can be estimated by
[O(, )[ [ [
2
_
(1 +[[
2
)
s2
2
+ (1 +[[
2
)
s2
2
_
.
8.1 Pseudodierential Operators 179
Now, as in (1), it is easily seen that the operator T
2
is of order : 2. Thus, to
nish the proof of (3), it suces to show that [T
1
, /] is of order : 2. Write
/(, ) =
n

j=1
(
j

j
)

j
(1 +[[
2
)
s
2
p( )
= :
n

j=1

j
(
j

j
)(1 +[[
2
)
s2
2
p( ).
Thus, from a direct calculation we obtain
([T
1
, /]n)

() =
1
(2)
n
_
1(, ) n() d,
where
1(, ) =
1
(2)
n
_
_
/(, )

/( ) /(, )

/( )
_
d
=
1
(2)
n
_

/( )
_
/(, + ) /(, )
_
d.
Since
[/(, + ) /(, )[
[ [[ [
_
(1 +[[
2
)
s2
2
+ (1 +[[
2
)
s2
2
_
[ p( )[,
we can estimate 1(, ) as follows,
[1(, )[
_
[ [[ [(1 +[[
2
)
s2
2
[

/( )[[ p( )[ d
+
_
[ [[ [(1 +[[
2
)
s2
2
[

/( )[[ p( )[ d
(1 +[[
2
)
s2
2
__
[[[

/()[[ [[ p( )[ d
+
_
[[(1 +[[
2
)
|s2|
2
[

/()[[ [[ p( )[ d
_
C
m
(1 +[[
2
)
s2
2
(1 +[ [
2
)
m
,
for any : N, where C
m
is a constant depending on :. Here we have used the
fact that both

/ and p are in o. Choosing : to be suciently 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 dierential operators of order / and :
respectively with coecients in o. Then [
s
, 1] is of order : + / 1, [
r
, [
s
, 1]]
is of order : +: +/ 2, and [[
s
, 1], Q] is of order : +/ +:2.
180 Subelliptic Estimates for Second Order Dierential Equations and
b
Proof. Write
1 =

]]k
o

(r)1

,
with o

(r) o. Since 1

commutes with
s
, the commutator of
s
with 1 is
reduced to commutators of
s
with o

(r), composed with 1

. This proves the


theorem.
We shall use operators of the form generated by
s
, 1

and multiplication by
functions in o plus their Lie brackets. All these are pseudodierential operators
and the computation of their orders is similar to that for dierential operators.
8.2 Hypoellipticity of Sum of Squares of Vector Fields
Let 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 elds with smooth real-valued coecients o
ij
(r)
on . Dene the second order partial dierential operator
(8.2.1) 1 =
k

i=1
A
2
i
+A
0
+/(r),
where /(r) is a smooth real-valued function on .
Denote by L
1
the collection of the A
i
s, 0 i /. Then, inductively for an
integer : 2 we dene L
m
to be the collection of L
m1
and the vector elds of
the form [A, Y ] with A L
1
and Y L
m1
.
Denition 8.2.1. The partial dierential operator 1 dened as in (8.2.1) is said
to be of nite type at point j if there exists an : such that L
m
spans the whole
tangent space at j.
Denition 8.2.2. A partial dierential operator 1 is said to be hypoelliptic in
if it satises the following property: let n and ) be distributions satisfying 1n = )
in , then u is smooth on l if ) is smooth on l for any open subset l of
The task of this section is to prove the following main theorem:
Theorem 8.2.3. Let 1 be the partial dierential operator dened as in (8.2.1).
Suppose that 1 is of nite type at every point in . Then 1 is hypoelliptic in .
The heat operator on 1
n+1
dened 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 coecients is the
Grushin operator on 1
2
dened 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 , if necessary, we may assume that o
ij
(r) and /(r) are in
C

() for all i, ,.
8.2 Hypoellipticity of Sum of Squares of Vector Fields 181
Lemma 8.2.4. Let 1 be dened 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
().
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 rst prove the following general theorem.
Theorem 8.2.5. If L
m
spans the tangent space of 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
().
Here we may take c = 2
1m
.
Proof. We shall denote an element in L
j
by 7
j
. By the hypotheses of the theorem,
we get
| n |
2

j=1
| 1
j
n |
2
1
+ | n |
2


Zm1m
| 7
m
n |
2
1
+ | n |
2
,
where the last summation is a nite sum and 1
j
= (,r
j
).
Therefore, to prove the theorem, it suces 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
m1
7
m1
A with A L
1
. Thus, we see that
(8.2.4)
| 7
m
n |
2
1
= (7
m
n,
22
7
m
n)
= (A7
m1
n,
22
7
m
n) (7
m1
An,
22
7
m
n).
182 Subelliptic Estimates for Second Order Dierential Equations and
b
Using Theorem 8.1.5, we have
[(A7
m1
n,
22
7
m
n)[
= [(7
m1
n,
22
7
m
An)[ +O(| n || 7
m1
n |
21
)
C(| An |
2
+ | 7
m1
n |
2
21
+ | n |
2
).
Also
[(7
m1
An,
22
7
m
n)[ = [(An, 7
m1

22
7
m
n)[ +O(| n || An |)
C(| An |
2
+ | 7
m1
n |
2
21
+ | 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
m1
n |
2
21
+ | n |
2
_
C
_
k

i=0
(| A
i
n |
2
+ | A
i
n |
2
2
m1
1
) + | n |
2
_
.
Now, for : 2, if we take c 2
1m
, 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
().
Proof. As in the proof of Theorem 8.2.5, it suces to control each term
(8.2.6) | 7
m
n |
2
1
= (A7
m1
n,
22
7
m
n) (7
m1
An,
22
7
m
n)
by the right-hand side of (8.2.5) for some c 0.
Let Q
s
be some pseudodierential 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 elds
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
m1
n |
2
21
+ | n |
2
).
8.2 Hypoellipticity of Sum of Squares of Vector Fields 183
Case (ii). A = A
0
. We rst 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

(). Hence, we have


(8.2.8)
(A
0
7
m1
n,
22
7
m
n)
= (1

7
m1
n,
22
7
m
n) +
k

i=1
(A
2
i
7
m1
n,
22
7
m
n)
+
k

i=1
(c
i
A
i
7
m1
n,
22
7
m
n) + (p7
m1
n,
22
7
m
n).
Obviously, the last two terms in (8.2.8) are bounded by the right-hand side of
(8.2.7). Since
(1

7
m1
n,
22
7
m
n) = (7
m1
n, 1Q
21
n)
= (7
m1
n, Q
21
1n) +
k

i=1
(7
m1
n, Q
21
A
i
n) + (7
m1
n, Q
21
n),
we conclude that the rst term on the right-hand side in (8.2.8) is also bounded by
the right-hand side of (8.2.7). The second term on the right-hand side of (8.2.8)
can be estimated as follows:
(A
2
i
7
m1
n,
22
7
m
n)
= (A
i
7
m1
n, A
i
Q
21
n) +O(| 1n |
2
+ | 7
m1
n |
2
21
+ | n |
2
)
= (A
i
Q
21
7
m1
n, A
i
n) +O(| 1n |
2
+ | 7
m1
n |
2
21
+ | n |
2
).
It follows from Lemma 8.2.4 that
(8.2.9)
[(A
2
i
7
m1
n,
22
7
m
n)[
| A
i
Q
21
7
m1
n |
2
+ O(| 1n |
2
+ | 7
m1
n |
2
21
+ | n |
2
)
C[(1Q
21
7
m1
n, Q
21
7
m1
n)[
+O(| 1n |
2
+ | 7
m1
n |
2
21
+ | n |
2
)
C([(1n, Q
41
7
m1
n)[ +
k

i=1
[(A
i
n, Q
41
7
m1
n)[
+[(n, Q
41
7
m1
n)[) +O(| 1n |
2
+ | 7
m1
n |
2
21
+ | n |
2
)
C(| 1n |
2
+ | 7
m1
n |
2
41
+ | n |
2
).
This completes the estimate of the rst term on the right-hand side of (8.2.6).
For the second term on the right-hand side of (8.2.6), we write
A
0
= 1
k

i=1
A
2
i
/(r).
184 Subelliptic Estimates for Second Order Dierential Equations and
b
Thus, we have
(8.2.10)
(7
m1
A
0
n,
22
7
m
n)
= (7
m1
1n,
22
7
m
n)
k

i=1
(7
m1
A
2
i
n,
22
7
m
n)
(7
m1
/n,
22
7
m
n).
The rst term on the right-hand side of (8.2.10) can be written as
(7
m1
1n,
22
7
m
n)
= (1n, 7
m1
Q
21
n) + (1n, p
m
Q
21
n)
= (1n, Q
21
7
m1
n) (1n, [7
m1
, Q
21
]n) + (1n, p
m
Q
21
n),
for some p
m
C

(). Also
(7
m1
/n,
22
7
m
n) = (/7
m1
n, Q
21
n) + ([7
m1
, /]n, Q
21
n).
Hence, if c 1,2, the rst and third terms of (8.2.10) can be estimated by the
right-hand side of (8.2.7).
To deal with the second term on the right-hand side of (8.2.10), we have
(7
m1
A
2
i
n,
22
7
m
n)
= (A
2
i
7
m1
n, Q
21
n) + ([7
m1
, A
2
i
]n, Q
21
n)
= (A
2
i
7
m1
n, Q
21
n) + ([7
m1
, A
i
]A
i
n, Q
21
n)
+ (A
i
[7
m1
, A
i
]n, Q
21
n)
= (A
2
i
7
m1
n, Q
21
n) + (

7
m
A
i
n, Q
21
n) + (A
i

7
m
n, Q
21
n).
Note that

7
m
is a commutator of 7
m1
with A
i
for some 1 i /. Thus, one
may apply the proof of Case (i) and (8.2.9) to get
[(7
m1
A
2
i
n,
22
7
m
n)[
C(| 1n |
2
+ | 7
m1
n |
2
41
+ | n |
2
+ |

7
m
n |
2
21
)
C(| 1n |
2
+ | 7
m1
n |
2
41
+ jo:o||c|n |
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
m1
n |
2
41
+ | n |
2
)
C(| 1n |
2
+
k

i=0
| A
i
n |
2
4
m1
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,
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 ,
1
C

0
() be two real-valued cut-o functions with
1
1
on the support of . Then there is a constant C 0 such that
(8.2.11) | n |

C(|
1
1n | + |
1
n |),
for all n C

().
Proof. From (8.2.5) it suces to estimate | [1, ]n | by the right-hand side of
(8.2.11). Since
[1, ]n = 2
k

i=1
[A
i
, ]A
i
n +
k

i=1
[A
i
, [A
i
, ]]n + [A
0
, ]n,
we have
| [1, ]n |
2
C
_
k

i=1
| A
i

2
1
n |
2
+ |
1
n |
2
_
.
Now, as in the proof of Lemma 8.2.4, we get
k

i=1
| A
i

2
1
n |
2
= (1
2
1
n,
2
1
n) + Re(A
0

2
1
n,
2
1
n)
+O
_
k

i=1
| A
i

2
1
n ||
1
n | + |
1
n |
2
_
= (
2
1
1n,
2
1
n) 2
k

i=1
([A
i
,
2
1
]A
i
n,
2
1
n) + Re(A
0

2
1
n,
2
1
n)
+O
_
k

i=1
| A
i

2
1
n ||
1
n | + |
1
n |
2
_
= (
2
1
1n,
2
1
n) + 4
k

i=1
([A
i
,
1
]
1
n, A
i

2
1
n) + Re(A
0

2
1
n,
2
1
n)
+O
_
k

i=1
| A
i

2
1
n ||
1
n | + |
1
n |
2
_
.
186 Subelliptic Estimates for Second Order Dierential Equations and
b
Hence, using small and large constants, we obtain
k

i=1
| A
i

2
1
n |
2
C(|
1
1n |
2
+ |
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
().
Proof. Let n C

0
(). We wish to apply (8.2.5) to
s
n. Hence, we shall assume
that estimate (8.2.5) holds for smooth functions that are supported in a xed open
neighborhood l of . Let (r) and p(r) be two smooth cut-o functions supported
in l such that 1 on the support of p and that p 1 on . We claim that the
operator (1 )
s
p acting on o, : 1, is of negative innite order, namely, it is a
smoothing operator and for any : 0 there is a constant C
s,m
such that
(8.2.13) | (1 )
s
pn | C
s,m
| n |
m
,
for any n o. We rst prove the claim (8.2.13). Observe rst that (1 )p 0.
Hence, we have
(8.2.14)
1
(2)
n
_
R
n
(1 (r))c
ix
j()1

p() d = 0,
for any polynomial j() and any multiindex . A direct calculation shows
(1 )
s
pn(r) =
1
(2)
n
_
R
n
1(r, ) n() d,
where
1(r, ) =
1
(2)
n
_
R
n
(1 (r))c
ix
(1 +[[
2
)
s
2
p( ) d.
Thus, to prove the claim it suces to show that for any o, / 1, the kernel
(8.2.15)

1(r, ) =
_
R
n
(1 (r))c
ix
(1 +[[
2
)
a
(1 +[[
2
)
b
p( ) d
satises the hypotheses of Theorem B.10 in the Appendix. Using (8.2.14), Lemma
8.1.2 and Taylors expansion of (1 +[[
2
)
a
at , for any c 1, we obtain
[

1(r, )[

_
R
n
(1 +[ +t( )[
2
)
ac
[ [
c
(1 +[[
2
)
b
[ p( )[ d

_
R
n
(1 +[t( )[
2
)
]ac]
(1 +[[
2
)
ac
[ [
c
(1 +[[
2
)
b
[ p( )[ d,
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 veried by integration by parts that the kernel

1(r, ) dened in (8.2.15)
satises
[r


1(r, )[ C
a,b,
uniformly in r and , where is a multiindex. It follows that all the hypotheses of
Theorem B.10 are satised by the kernel

1(r, ). This proves the claim.
Now we return to the estimate of
s
n for n C

0
(). Since
s
n is supported
in l, we may apply estimate (8.2.5) to get
| n |
s+
= |
s
n |

|
s
n |

+ | (1 )
s
pn |

C(| 1
s
n | + |
s
n | + | n |
m
)
C(| [1,
s
]n | + | 1n |
s
+ | n |
s
+ | n |
m
).
To handle the term | [1,
s
]n | we write
[1,
s
] =
k

i=1
Q
s
i
A
i
+Q
s
k+1
.
Thus,
| [1,
s
]n | C
_
k

i=1
| A
i
n |
s
+ | n |
s
_
.
Lemma 8.2.4 then shows that
| A
i
n |
2
s
= |
s
A
i
n |
2
C(|
s
A
i
n |
2
+ | (1 )
s
pA
i
n |
2
)
C(| A
i

s
n |
2
+ | [
s
, A
i
]n |
2
+ | n |
2
m
)
C([(1
s
n,
s
n)[ + | n |
2
s
+ | n |
2
m
)
C([([1,
s
]n,
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 0, using the interpolation inequality for Sobolev
spaces (Theorem B.2 in the Appendix), there is a constant C
s,,m
such that
| n |
s
| n |
s+
+ C
s,,m
| n |
m
,
for all n C

0
(). Therefore, letting be suciently 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 Dierential Equations and
b
Theorem 8.2.9. Let ,
1
be two smooth real-valued cut-o functions supported in
with
1
1 on the support of . For any : 1 and : 0 there is a constant
C
s,m
such that
(8.2.16) | n |
s+
C
s,m
(|
1
1n |
s
+ |
1
n |
m
),
for all n C

().
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 nite type, and that ) is smooth on . We wish to show that
n is also smooth on . 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 and
k
, / N, be a sequence of smooth
real-valued cut-o functions supported in some open neighborhood of \ such that

1
n \
m
,
k
1 on supp
k+1
,
j
1 on supp for all , and 1 on \ .
Let be a smooth nonnegative real-valued function supported in the unit open
ball of 1
n
such that (r) = ([r[), 0 1 and
_
R
n
dr = 1. For any 0, set

(r) =
n
(r,) and
o

k
n(r) =
k
n

(r) =
_

k
n(j)

(r j) dj.
Clearly, o

k
n is a smooth function supported in some neighborhood of \ . Hence,
we obtain from (8.2.16) with : = : that
| o

k
n |
m+
C
m
(| 1o

k
n |
m
+ | o

k
n |
m
).
To nish the proof we need the following two key observations. For any : 1,
(1)
k
n \
s
with |
k
n |
s
C
k
if and only if | o

k
n |
s
C
k
for all small
0.
(2) If
k
n is in \
s
, then
| [o

k
, A
i
]n |
s
= | [o

k
, A
i
]
k1
n |
s
C
k
|
k1
n |
s
,
where the constant C
k
is independent of .
These two facts can be veried directly, so we omit the proofs.
Now, using (1) and (2), by commuting 1 with o

and applying arguments


similar to those above, we obtain
| o

2
n |
m+
C
1,m
(| o

1
) |
m
+ | o

1
n |
m
)
with C
1,m
independent of . This implies n \
m+
. Inductively, for any / N,
we obtain
| o

k+1
n |
m+k
C
k,m
(| o

k
) |
m+(k1)
+ | o

k
n |
m+(k1)
).
8.3 Subelliptic Estimates for the Tangential Cauchy-Riemann Complex 189
Choosing / to be suciently large, we have n \
s
(\ ) for all : 0. Hence,
n C

(\ ). This completes the proof of Theorem 8.2.3.


8.3 Subelliptic Estimates for the Tangential Cauchy-Riemann Complex
Let (`, T
1,0
(`)) be a compact orientable C1 manifold of real dimension 2n1
with n 3. Let
p,q
(`), 0 j, n 1, denote the subbundle of
p+q
CT

(`)
such that
p,q
(`) =
p
T
1,0
(`)
q
T
0,1
(`). Let c
p,q
(`) be the space of
smooth sections of
p,q
(`) over `. Then, we dene the tangential Cauchy-
Riemann operator
b
as in Section 7.2, and form the tangential Cauchy-Riemann
complex
(8.3.1) 0 c
p,0
(`)

b
c
p,1
(`)

b


b
c
p,n1
(`) 0.
For any r
0
`, let 1
1
, , 1
n1
be a local basis for (1,0) vector elds near r
0
,
and choose a globally dened vector eld T which may be assumed to be purely
imaginary. This is rst 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
n1
, 1
1
, , 1
n1
and T is an orthonormal basis in some neighborhood of
a reference point r
0
`. Let
1
, ,
n1
be an orthonormal basis for (1,0)-forms
which is dual to the basis 1
1
, , 1
n1
.
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 1
2
(p,q)
(`) such that

b
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,n1)
(`) 0.
Therefore,
b
is a linear, closed, densely dened operator on the Hilbert space
1
2
(p,q)
(`).
Now one can dene the adjoint operator

b
of
b
in the standard way. A (j, )-
form is in Dom(

b
) if there exists a (j, 1)-form p 1
2
(p,q1)
(`) such that
(,
b
) = (p, ) for every (j, 1)-form Dom(
b
). In this case we dene

b
= p. Let
(8.3.3)
b
=
b

b
+

b
,
be the
b
-Laplacian dened on
Dom(
b
) = 1
2
(p,q)
(`)[ Dom(
b
) Dom(

b
),

b
Dom(

b
) and

b
Dom(
b
).
190 Subelliptic Estimates for Second Order Dierential Equations and
b
It follows from the same arguments as in Proposition 4.2.3 that
b
is a linear, closed,
densely dened self-adjoint operator from 1
2
(p,q)
(`) into itself. Dene a Hermitian
form Q
b
on c
p,q
(`) by
(8.3.4) Q
b
(, ) = (
b
,
b
) + (

b
,

b
) + (, ) = ((
b
+1), ),
for , c
p,q
(`).
Locally on a coordinate neighborhood l, we can express a smooth (j, )-form
as
(8.3.5) =

t
]I]=p,]J]=q

I,J

I

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
I,J
s are dened for arbitrary 1 and J so that they are antisymmetric.
Then, a direct computation and integration by parts yield
(8.3.6)
b
=

t
I,J

j
1
j
(
I,J
)
j

I

K
+ terms of order zero,
and
(8.3.7)

b
= (1)
p1

t
I,K

j
1
j
(
I,jK
)
I

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

I,J
|
2
+ | |
2
by | |
2
L
, and

k,I,J
| 1
k

I,J
|
2
+ | |
2
by | |
2
L
.
The main eort of this section is to derive the subelliptic 1,2-estimate 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 xed , 1 n 1, one actually does not
need strong pseudoconvexity of the domain. The main ingredient is the so-called
condition 7() dened as follows:
8.3 Subelliptic Estimates for the Tangential Cauchy-Riemann Complex 191
Denition 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 satised 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 0)
that
Q(), )) =

t
I,J

k
| 1
k
)
I,J
|
2
+

t
I,K

j,k
_
bDU

jk
)
I,jK
)
I,kK
do
+O((| ) | + |

) |) | ) | + | ) |
L
| ) |).
We may assume that the Levi form is diagonal at r
0
, namely,
jk
(r) =
j

jk
+/
jk
(r)
for 1 ,, / n 1, where the
j
s are the eigenvalues of the Levi form at r
0
,
jk
denotes the Kronecker delta and /
jk
(r
0
) = 0. It follows that

t
I,K

j,k
_
bDU

jk
)
I,jK
)
I,kK
do
=

t
I,J
_

kJ

k
__
bD
[)
I,J
[
2
do +O
_

t
I,J
_
bD
[)
I,J
[
2
do
_
,
where 0 can be made arbitrarily small if l is chosen suciently 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
| ) |)

k
_
bD
[)
I,J
[
2
do
_
bD
[)
I,J
[
2
do +O(| ) |
L
| ) | + | ) |
2
).
Hence, if condition 7() holds on /1, then for each xed J either there is a /
1
J
with
k1
0 or there is a /
2
, J with
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
_

kJ,
k
<0

k
__
bD
[)
I,J
[
2
do
+

t
I,J
_

k1
_
bD
[)
I,J
[
2
do
_

_
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
_

kJ,
k
<0

k
__
bD
[)
I,J
[
2
do
+ (1 c)

t
I,J
(
k2
)
_
bD
[)
I,J
[
2
do
_
bD
[)
I,J
[
2
do
+O((| ) | + |

) |) | ) | + | ) |
L
| ) | + | ) |
2
).
192 Subelliptic Estimates for Second Order Dierential Equations and
b
Thus, choosing c, 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 satised 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
denition for any C1 manifold.
Denition 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 2n1 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 satised 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) | |
2
1
2
Q
b
(, ),
8.3 Subelliptic Estimates for the Tangential Cauchy-Riemann Complex 193
uniformly for all c
p,q
(`).
Proof. Let 1
1
, , 1
n1
be an orthonormal basis for T
1,0
(`) locally. Since con-
dition Y () implies that the vector elds 1
1
, , 1
n1
, 1
1
, , 1
n1
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) | |
2
L
+ | |
2
L
+

IJ
[1c(T
I,J
,
I,J
)[ Q
b
(, ),
uniformly for all c
p,q
(`) with support contained in \
x0
.
Proof. We start with (8.3.6) to obtain
|
b
|
2
=

t
I,J

j / J
| 1
j

I,J
|
2
+

t
I,J,L

j,l
c
jJ
lL
(1
j

I,J
, 1
l

I,L
) +O(| |
L
| |),
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
I,J
is antisymmetric in J, we rearrange
the terms in the above estimate to get
|
b
|
2
=

t
I,J

j
| 1
j

I,J
|
2

t
I,K

j,k
(1
j

I,kK
, 1
k

I,jK
) +O(| |
L
| |).
Using integration by parts, we have
(1
j

I,kK
, 1
k

I,jK
) = (1
k
1
j

I,kK
,
I,jK
) +O(| |
L
| |)
= (1
k

I,kK
, 1
j

I,jK
) + ([1
j
, 1
k
]
I,kK
,
I,jK
)
+O((| |
L
+ | |
L
) | |).
Hence, from (8.3.7) we obtain
|
b
|
2
=

t
I,J

j
| 1
j

I,J
|
2
|

b
|
2
+

t
I,K

j,k
([1
j
, 1
k
]
I,jK
,
I,kK
)
+O((| |
L
+ | |
L
) | |).
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(T
I,J
, )
I,L
)[

Re
_
1
c
11
[1
1
, 1
1
]
I,J
, )
I,L
_

+O(| |
L
| |)
C(sup
Vx
0
[)[)(| |
2
L
+ | |
2
L
) +O(| |
L
| |).
194 Subelliptic Estimates for Second Order Dierential Equations and
b
Thus, if we denote the eigenvalues of the Levi form at r
0
by
1
, ,
n1
, we obtain
(8.3.12)
Q
b
(, ) =

t
I,J

j
| 1
j

I,J
|
2
+

t
I,J

jJ

j
Re(T
I,J
,
I,J
)
+O(| |
2
L
+ | |
2
L
) +O(| |
L
| |),
where 0 can be made arbitrarily small, if necessary, by shrinking \
x0
.
To nish the proof, we need to control the second term on the right-hand side of
(8.3.12). This will be done by using a fraction of the rst term on the right-hand
side of (8.3.12). Thus, we rst use integration by parts to get
| 1
j

I,J
|
2
= | 1
j

I,J
|
2

j
Re(T
I,J
,
I,J
)
+O(| |
2
L
+ | |
2
L
) +O((| |
L
+ | |
L
) | |).
Next, for each multiindex pair (1, J), set
(1, J) = ,[
j
< 0 if Re(T
I,J
,
I,J
) 0 or
j
< 0 if Re(T
I,J
,
I,J
) < 0.
It follows that, for any small c 0, we have
| |
2
L
c | |
2
L
+ (1 c)

t
I,J

j(I,J)
| 1
j

I,J
|
2
c | |
2
L
(1 c)

t
I,J

j(I,J)

j
Re(T
I,J
,
I,J
)
(| |
2
L
+ | |
2
L
) C | |
2
.
Substituting the above into (8.3.12) we obtain
(8.3.13)
Q
b
(, ) c | |
2
L
(1 c)

t
I,J

j(I,J)

j
Re(T
I,J
,
I,J
)
+

t
I,J

jJ

j
Re(T
I,J
,
I,J
) (| |
2
L
+ | |
2
L
)
O(| |
L
| |)
= c | |
2
L
+

t
I,J
o
I,J
Re(T
I,J
,
I,J
) (| |
2
L
+ | |
2
L
)
O(| |
L
| |),
where
o
I,J
=

jJ\(I,J)

j
(1 c)
_
_

j(I,J)\J

j
_
_
+c
_
_

jJ(I,J)

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
j
and
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
j
0 and
k
< 0.
(3) Under the same hypothesis as in (2), there are ,, / J so that
j
0 and

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(T
I,J
,
I,J
) 0, and o
I,J
< 0 if Re(T
I,J
,
I,J
) < 0. Since
the second term on the right-hand side of (8.3.13) is a nite sum, by letting 0
be suciently small, we get
Q
b
(, ) | |
2
L
+

I,J
[Re(T
I,J
,
I,J
)[ (:c) | |
2
L
(|c) | |
2
.
Since
| 1
j

I,J
|
2
| 1
j

I,J
|
2
+ [Re(T
I,J
,
I,J
)[
+
1
O(| |
2
L
+ | |
2
L
) +O(| |
L
| |),
where
1
0 can be made arbitrarily small, by choosing
1
and (:c) to be suciently
small, we obtain
| |
2
L
+ | |
2
L
+

I,J
[Re(T
I,J
,
I,J
)[ Q
b
(, ).
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
(, ) C | |
2
1
2
, for 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 Rellichs 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 rst proving a
priori estimates for the operator
b
+1 on a local coordinate neighborhood l.
196 Subelliptic Estimates for Second Order Dierential 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
k

k=1
be a sequence of real smooth functions
supported in l such that
k
= 1 on the support of
k+1
for all /. Then, for each
positive integer /, we have
|
k
|
2
k
2
|
1
(
b
+1) |
2
k2
2
+ | (
b
+1) |
2
uniformly for all c
p,q
(`) supported in l.
Proof. The lemma will be proved by induction. For / = 1, by Theorem 8.3.4, we
have
|
1
|
2
1
2
Q
b
(
1
,
1
) = |
b

1
|
2
+ |

1
|
2
+ |
1
|
2
.
We estimate the right-hand side as follows:
|
b

1
|
2
= (
b

1
,
b

1
)
= (
1

b
,
b

1
) + ([
b
,
1
],
b

1
)
= (
b
,
b

2
1
) + (
b
, [
1
,
b
]
1
) + ([
b
,
1
],
b

1
)
= (

b
,
2
1
) + (
b

1
, [
1
,
b
])
+ ([
1
,
b
], [
1
,
b
]) + ([
b
,
1
],
b

1
).
Note that
Re
_
(
b

1
, [
1
,
b
]) + ([
b
,
1
],
b

1
)
_
= 0.
A similar argument holds for |

1
|
2
. Thus, we get
|
1
|
2
1
2
Q
b
(
1
,
1
)
Re((
b
+1),
2
1
) +O(| |
2
)
| (
b
+1) || | +O(| |
2
)
| (
b
+1) |
2
,
since | | | (
b
+1) |. 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 pseudodierential
operator
k1
2
=
k
for short, and denote
1

k
by 1
k
. Then, we have
|
k
|
2
k
2
|
k

k
|
2
1
2
= |
k

k
|
2
1
2
|
1

k
|
2
1
2
+ | [
k
,
1
]
k

k1
|
2
1
2
Q
b
(1
k
, 1
k
) + |
k1
|
2
k2
2
.
Let 1

k
be the adjoint operator of 1
k
, then the rst term can be estimated as
8.4 Local Regularity and the Hodge Theorem for
b
197
follows:
Q
b
(1
k
, 1
k
)
= |
b
1
k

k1
|
2
+ |

b
1
k

k1
|
2
+ | 1
k

k1
|
2
= (
b

k1
, 1

b
1
k

k1
) + (

k1
, 1

b
1
k

k1
) + (
k1
, 1

k
1
k

k1
)
+O(|
k1
|k1
2
(|
b
1
k

k1
| + |

b
1
k

k1
|))
= (
b

k1
,
b
1

k
1
k

k1
) + (

k1
,

b
1

k
1
k

k1
) + (
k1
, 1

k
1
k

k1
)
+O(|
k1
|
2
k1
2
+ |
k1
|k1
2
(|
b
1
k

k1
| + |

b
1
k

k1
|))
= (
b
,
b
1

k
1
k

k1
) + (

b
,

b
1

k
1
k

k1
) + (, 1

k
1
k

k1
) +O( )
= ((
b
+1), 1

k
1
k
) +O( )
= (1
k

1
(
b
+1), 1
k
) +O( )
| 1
k

1
(
b
+1) |

1
2
| 1
k
|1
2
+ O( )
(|c) |
1
(
b
+1) |
2
k2
2
+ (:c) |
k
|
2
k
2
+ (|c) |
k1
|
2
k1
2
+ (:c)(|
b
1
k

k1
|
2
+ |

b
1
k

k1
|
2
).
Hence, by choosing (:c) small enough and using the induction hypothesis, we
obtain
|
k
|
2
k
2
|
1
(
b
+1) |
2
k2
2
+ |
k1
|
2
k1
2
|
1
(
b
+1) |
2
k2
2
+ |
1
(
b
+1) |
2
k3
2
+ | (
b
+1) |
2
|
1
(
b
+1) |
2
k2
2
+ | (
b
+1) |
2
.
This completes the proof of the lemma.
Theorem 8.4.2. Under the same hypotheses as in Theorem 8.3.4, given
1
2
(p,q)
(`), let Dom(
b
) be the unique solution of (
b
+ 1) = . If l is
a subregion of ` and [
U
c
p,q
(l), then [
U
c
p,q
(l). Moreover, if and
1
are two cut-o functions supported in l such that
1
= 1 on the support of , then
for each integer : 0, there is a constant C
s
such that
(8.4.1) | |
2
s+1
C
s
(|
1
|
2
s
+ | |
2
).
Proof. If [
U
is smooth, then the estimate (8.4.1) follows immediately from Lemma
8.4.1. Therefore, it remains only to show that [
U
c
p,q
(l). Since the Hermitian
form Q
b
is not elliptic, namely, Gardings 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
,
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
i

m
i=1
be a partition of
198 Subelliptic Estimates for Second Order Dierential Equations and
b
unity subordinate to l
i

m
i=1
. Dene the form Q

b
, for each c with 0 < c << 1, by
Q

b
(, ) = Q
b
(, ) +c
m

i=1
2n1

j=1
(1
j

i
, 1
j

i
)
for all , c
p,q
(`), where 1
j
= ,r
j
. Denote by

\

(p,q)
(`) the completion of
c
p,q
(`) under Q

b
.
Let

(p,q)
(`) be the unique solution to the equation
Q

b
(

, ) = (, ), for

\

(p,q)
(`).
Then, we have
(8.4.2) |

|
2
s+1
|
1
|
2
s
+ | |
2
and the estimate is uniform for all c with 0 < c << 1. Also, as in the proof of
Theorem 5.2.1,

converges to in 1
2
(p,q)
(`).
The sequence

, by (8.4.2), is uniformly bounded in \


s+1
(p,q)
(`) for each :.
Hence, by Rellichs lemma we can extract a subsequence
j
that converges in
\
s
(p,q)
(`) as c
j
0. Since

converges to in 1
2
(p,q)
(`),
j
must converge
to in \
s
(p,q)
(`) for each :. Finally, by invoking the Sobolev embedding theorem,
we have 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 , , l, and
1
be as in Theorem 8.4.2. If [
U
\
s
(p,q)
(l)
for some : 0, then \
s+1
(p,q)
(`) and
| |
2
s+1
|
1
|
2
s
+ | |
2
.
Proof. Let
0
be a cut-o function supported in l such that
0
= 1 on the support
of
1
. Choose sequences of smooth (j, )-forms
n
and
n
with supp
n

supp
0
and supp
n
supp(1
0
) such that
n

0
in \
s
(p,q)
(`) and

n
(1
0
) in 1
2
(p,q)
(`).
Hence,
n
=
n
+
n
in 1
2
(p,q)
(`) and
1

n

1
in \
s
(p,q)
(`). Let

n
Dom(
b
) be the solution to (
b
+1)
n
=
n
, so
n
in 1
2
(p,q)
(`). Then,
Theorem 8.4.2 shows
| (
n

m
) |
s+1
|
1
(
n

m
) |
s
+ |
n

m
| .
It follows that
n
is Cauchy in \
s+1
(p,q)
(`), and lim
n

n
= is in \
s+1
(p,q)
(`).
Hence, we have
| |
s+1
|
1
|
s
+ | | .
This proves the theorem.
8.4 Local Regularity and the Hodge Theorem for
b
199
Theorem 8.4.4. Let , , l, and
1
be as in Theorem 8.4.2. If
1
\
s
(p,q)
(`)
for some : 0, and if satises (
b
+ 1) = for some constant , then
\
s+1
(p,q)
(`). In other words,
b
+1 is hypoelliptic for every . Moreover, all
the eigenforms of
b
are smooth.
Proof. Let
t
= +(1 ), then (
b
+1) =
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 Dom(
b
). If (
b
+ 1) = with \
s
(p,q)
(`), : 0, then
\
s+1
(p,q)
(`) and
| |
s+1
C | |
s
,
where the constant C is independent of .
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 Rellichs 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 nite limit point, and each eigenvalue occurs with nite
multiplicity. All eigenfunctions are smooth. In particular, Ker(
b
) is of nite
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 nite multiplicity. Also is an
eigenvalue of
b
+1 if and only if
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
= with \
s
(p,q)
(`), : 0, we have
| |
2
s+1
C(| |
2
s
+ | |
2
),
where the constant C 0 is independent of .
Proof. We show the estimate by an induction on :. If : = 0, Theorem 8.4.5 implies
| |
2
1
| (
b
+1) |
2
| |
2
+ | |
2
.
In general, if we assume the assertion holds up to step :1, we have \
s
(p,q)
(`).
For the case :, we apply Theorem 8.4.5 again and get
| |
2
s+1
| (
b
+1) |
2
s
|
b
|
2
s
+ | |
2
s
| |
2
s
+ | |
2
,
200 Subelliptic Estimates for Second Order Dierential Equations and
b
where the nal 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 nite
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
|
1
| |
for all Dom(
b
) (H
b
(p,q)
(`))

, where
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)
(`) = (
b
) H
b
(p,q)
(`)
=
b

b
(Dom
b
)

b
(Dom
b
) H
b
(p,q)
(`),
where (
b
) denotes the range of
b
.
Proof. Since (
b
)= (H
b
(p,q)
(`))

and (
b

b
) (

b
), the decomposition
follows.
We can thus dene the boundary operator,
b
: 1
2
(p,q)
(`) Dom(
b
), as
follows: If H
b
(p,q)
(`), set
b
= 0. If (
b
), dene
b
= , where is
the unique solution of
b
= with 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 1
2
(p,q)
(`), =
b

b
+

b
+H
b
,
(3)
b
H
b
= H
b

b
= 0.

b
=
b

b
= 1 H
b
on Dom(
b
).
(4) If
b
is also dened on 1
2
(p,q+1)
(`), then
b

b
=
b

b
on Dom(
b
).
If
b
is also dened on 1
2
(p,q1)
(`), then
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
|
s+1
| |
s
holds uniformly for all \
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
denition of
b
. For (4), if Dom(
b
), then

b
=
b

=
b

=
b

b
.
A similar equation holds for

b
. For (5), if c
p,q
(`), then H
b
c
p,q
(`),
we have

b
= H
b
.
Since
b
is hypoelliptic by Theorem 8.4.4,
b
c
p,q
(`). Estimate (8.4.4) now
follows from Proposition 8.4.8 since
|
b
|
s+1
|
b

b
|
s
+ |
b
|
| |
s
+ | H
b
|
s
+ | |
| |
s
.
Here we have used the fact that H
b
(p,q)
(`) is of nite dimension to conclude the
estimate: | H
b
|
s
C
s
| H
b
| C
s
| | 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,q1)
(`) is closed.
Proof. Since (
b
) Ker(

b
), we have (
b
) =
b

b
(Dom
b
).
Denition 8.4.12. Let ` be a compact orientable C1 manifold. The Szego pro-
jection o on ` is dened 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
Szego projection.
Theorem 8.4.13. Let ` be a compact orientable C1 manifold. Suppose that `
satises condition Y (1). Then the Szego projection o on ` is given by
o = 1

b
.
Theorem 8.4.10 gives the following solvability and regularity theorem for the
b
equation.
202 Subelliptic Estimates for Second Order Dierential Equations and
b
Theorem 8.4.14. Under the same hypotheses as in Theorem 8.4.10, for any
1
2
(p,q)
(`) with
b
= 0 and H
b
= 0, there is a unique solution of
b
= with
Ker(
b
). If c
p,q
(`), then c
p,q1
(`). Furthermore, for each : 0, if
\
s
(p,q)
(`), then \
s+
1
2
(p,q1)
(`) and
(8.4.5) | |
s+
1
2
| |
s
.
Proof. By (2) of Theorem 8.4.10, clearly we have =
b

b
. We simply take
=

b
, and is unique by the condition Ker(
b
). The smoothness of
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
i

m
i=1
be a partition
of unity subordinate to l
i

m
i=1
. Then we have
| |
2
s+
1
2
= |

b
|
2
s+
1
2
|

b
|
2
s+
1
2
+ |
b

b
|
2
s+
1
2

i=1
((

b
,
2s+1

b
) + (
b

b
,
2s+1

b
))
=
m

i=1
((

b
,
i

2s+1

b
) + (
b

b
,
i

2s+1

b
)) +O(|
b
|
2
s+1
)
=
m

i=1
((

b
,

2s+1

b
) + (
b

b
,
b

2s+1

b
)) +O(|
b
|
2
s+1
)
=
m

i=1
(,
i

2s+1

b
) +O(|
b
|
2
s+1
)
| |
s
|
b
|
s+1
+ |
b
|
2
s+1
.
By (5) of Theorem 8.4.10, we thus obtain
| |
s+
1
2
| |
s
.
This completes the proof of Theorem 8.4.14.
A nal 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
Pseudodierential operators were introduced by J. J. Kohn and L. Nirenberg
[KoNi 2] and L. Hormander [Hor 4] as a generalization of singular integral operators
developed by A. P. Calderon and A. Zygmund [CaZy 1]. These operators have
played an important role in the study of linear partial dierential equations. We
Notes 203
refer the reader to the books by Hormander [Hor 8], Nagel-Stein [NaSt 1] and Treves
[Tre 3] for detailed discussions and applications of pseudodierential operators. The
pseudodierential operators used in section 8.1 are of the simplest kind.
Theorem 8.2.3 and Theorem 8.2.6 were rst proved by L. Hormander in [Hor
6] where the argument is based on careful analysis of the one-parameter groups
generated by the given vector elds. Hormanders original proof gives very precise
c in Theorem 8.2.6. He also showed that nite 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 pseudodierential operators that we
present in Section 8.2 follows the paper by J. J. Kohn [Koh 3] (see also Oleinik-
Radkevic [OlRa 1]).
However, it is well known that the nite type condition is not necessary for the
hypoellipticity of a sum of squares operator. In [OlRa 1], Theorem 2.5.3, Oleinik and
Radkevic showed hypoellipticity when the nite type condition fails on a compact
set which is contained in a nite 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
t0
tlogo(t) = 0, where o C

(1) is even,
real-valued with derivatives of all orders bounded, non-decreasing on [0, ), and
vanishing to innite 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 real-valued / 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. Feerman and D. H. Phong [FePh 1] that both
operators 1
a
and 1
b
, when o and / are vanishing to innite 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 Hormanders Theorem
which is based on local subelliptic estimates. The formulation of a necessary and
sucient 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 rst introduced. See also [FoKo 1]. Theorem
8.3.2 shows that condition 7() is sucient for the subelliptic 1,2-estimate for the
-Neumann operator on (j, )-forms. The characterization of when the Morrey-
Kohn estimate, (8.3.8), holds was proved by L. Hormander in [Hor 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 Dierential Equations and
b
condition 7() was rst introduced. The techniques from subelliptic c-estimate 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 Folland-Stein [FoSt 1], Rothschild-Stein [RoSt 1], Beals-
Greiner-Stanton [BGS 1] and Boutet de Monvel-Sjostrand [BdSj 1] as well as the
books by Beals-Greiner [BeGr 1], Stein [Ste 4] and Treves [Tre 3].
205
206
CHAPTER 9
THE TANGENTIAL
CAUCHY-RIEMANN COMPLEX
ON PSEUDOCONVEX C1 MANIFOLDS
We study existence theorems for the tangential Cauchy-Riemann 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 pseudodierential 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 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 and use the solvability for



in . 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 closed-range 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 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

) coecients in C
n
such that )
is supported in , can one nd n 1
2
(p,q1)
(C
n
) (or n C

(p,q1)
(C
n
)) such that

n = ) in C
n
and n is supported in ?
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
has no compact components.
There are no other restrictions on the boundary of . When 1, we shall use
the duality of the

-Neumann problem to solve the Cauchy problem on bounded
pseudoconvex domains.
Let : C

(p,q)
() C

(np,nq)
() be the Hodge star operator dened by
, )d\ = ,
9.1 The L
2
Cauchy Problem for

207
where , C

(p,q)
() 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)
() to 1
2
(np,nq)
() in a
natural way. This is an algebraic operator given explicitly as follows: if we write
=

t
I,J

I,J
d.
I
d .
J
,
then
=

t
I,J
i
n
c
IJ

I,J
d.
[

I]
d .
[

J]
,
where [

1] denotes the increasing (nj)-tuple consisting of elements in 1, , n1


and c
IJ
is the sign of the permutation from (1, J, [

1], [

J]) to (1, 1
t
, , n, n
t
).
Lemma 9.1.1. For every C

(p,q)
(), we have
(9.1.1) = (1)
p+q
,
and
(9.1.2) =

,
where and

are viewed as dierential operators.
Proof. (9.1.1) is easy to check. To prove (9.1.2), we have, for any C

(p,q1)
()
and C

(p,q)
() such that has compact support in ,
(

, ) =
_

= (1)
p+q
_



+
_

d( ).
Since
_

d( ) = 0, using (9.1.1),
(

, ) =
_



= (,

) = (, ).
This proves (9.1.2).
Theorem 9.1.2. Let be a bounded pseudoconvex domain with C
1
boundary in
C
n
, n 2. Let = sup
z,z

[. .
t
[ be the diameter of . 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 , one can nd n 1
2
(p,q1)
(C
n
) such that

n = ) in the
distribution sense in C
n
with n supported in and
(n )
_

[ n [
2
d\ c
2
_

[ ) [
2
d\.
Proof. From Theorem 4.4.1, the

-Neumann operator of degree (n j, n ) in ,
denoted by
(np,nq)
, exists. We dene
(9.1.3) n =

(np,nq)
),
208 The Tangential Cauchy-Riemann Complex
then n 1
2
(p,q1)
() and n Dom(

).
Extending n to C
n
by dening n = 0 in C
n
, we claim that

n = ) in the
distribution sense in C
n
. First we prove that

n = ) in the distribution sense in .
From (9.1.1) and (9.1.2),
(9.1.4)

n =

(np,nq)
)
= (1)
p+q+1



(np,nq)
)
= (1)
p+q

(np,nq)
)
= (1)
p+q

(np,nq)
),
where acts in the distribution sense in . On the other hand, for any
C

(np,nq1)
(),
(9.1.5)
(

, ))

= (1)
p+q
_

) = (1)
p+q
_

= (1)
p+q
(),

= (), )

= (

), )
C
n
= 0
since supp ) and

) = 0 in the distribution sense in C
n
. If has C
1
boundary,
using the rst part of the proof of Lemma 4.3.2, the set C

(np,nq1)
() is dense in
Dom(

) in the graph norm. It follows from the denition of


that ) Dom(

)
and

()) = 0. Using Theorem 4.4.1 when 1 < n1 and Theorem 4.4.3 when
= n 1, we have
(9.1.6)

(np,nq)
) =
(np,nq1)

()) = 0.
Combining (9.1.4) and (9.1.6),

n = (1)
p+q

(np,nq)
)
= (1)
p+q
(

)
(np,nq)
)
= (1)
p+q
)
= )
in the distribution sense in . Furthermore, we note that n Dom(

) and this
additional condition implies that

n = ) in the distribution sense in C
n
. Using

(n) = n = (1)
p+q


n = (1)
p+q
),
where n is taken in the distribution sense in , we have for any C

(p,q)
(C
n
),
(9.1.7)
(n, )
C
n = (, n)

= (1)
p+q
(

, n)

= (1)
p+q
(,

(n))

= (, ))

= (), )
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 has C
1
boundary is used to show that C

(np,nq1)
()
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 star-shaped or locally
star-shaped.
When n, including the top degree case, we also have the following result.
Theorem 9.1.3. Let 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 and
(9.1.8)
_

) p = 0 for every p 1
2
(np,nq)
() Ker(

),
one can nd n 1
2
(p,q1)
(C
n
) such that

n = ) in the distribution sense in C
n
with
n supported in and
_

[ n [
2
d\ c
2
_

[ ) [
2
d\,
where = sup
z,z

[. .
t
[ is the diameter of .
If is a bounded pseudoconvex domain with smooth boundary, (9.1.8) can be
replaced by the condition
(9.1.9)
_

) p = 0 for every p C

(np,nq)
() 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 dene n by
(9.1.12) n =

(np,nq)
).
Using Lemma 9.1.1, we have from (9.1.11),
(9.1.13)

n = (1)
p+q

(np,nq)
)
= ) (1)
p+q
1
(np,nq)
).
210 The Tangential Cauchy-Riemann Complex
From (9.1.8), we get for any p 1
2
(np,nq)
() Ker(

),
(9.1.14) (), p) = (1)
p+q
_

p ) = 0.
Thus 1
(np,nq)
()) = 0 and

n = ) in from (9.1.13).
Using n Dom(

) and extending n to be zero outside , 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 ) satises (9.1.8).
When / is smooth, we have C

(np,nq)
()Ker(

) is dense in H
(np,nq)
() =
1
2
(np,nq)
() Ker(

) in the 1
2
() norm, using Corollary 6.1.6. Thus if ) satises
condition (9.1.9), it also satises 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
(), )
C
n =
_
D
) = (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

(np,nq1)
() is dense in Dom(

) in the graph norm.


9.2

-Closed Extensions of Forms and C

Solvability of

b
Let be a bounded pseudoconvex domain in C
n
with smooth boundary ` = /
and be a smooth dening function for such that [d[ = 1 on `. We use c
(p,q)
(`)
to denote the smooth (j, )-forms on `, where 0 j n, 0 n1. Here the
extrinsic denition for c
(p,q)
(`) is used.
We consider the following two kinds of

-closed extension problems:
Given c
(p,q)
(`),
(1) can one nd an extension of such that = on ` and

= 0
in ? (We recall that is the projection of smooth (j, )-forms in C
n
to
(j, )-forms on ` which are pointwise orthogonal to the ideal generated by

.)
(2) can one nd an extension of such that = on ` and

= 0 in ?
When < n 1, it is necessary that

b
= 0 on ` in order to have a

-closed
extension. When 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 has a holomorphic extension as long as C
n
has no
bounded component. When 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 be a bounded pseudoconvex domain in C
n
with smooth
boundary `. Let c
p,q
(`), where 0 j n and 1 n 1. Then there
exists C

(p,q)
() such that = on ` and

= 0 in if and only if
(9.2.1)
_
M
= 0 for every C

(np,nq1)
() Ker(

).
Furthermore, when 1 < n 1, (9.2.1) holds if and only if
(9.2.2)

b
= 0 on `.
Theorem 9.2.2. Let ` and be the same as in Theorem 9.2.1. There exists
such that C

(p,q)
(), = on ` and

= 0 in 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 satises
(9.2.1).
Let
t
be a smooth extension of by extending componentwise from the
boundary to . Thus
t
C

(p,q)
() and
t
= on `. We set ) =

t
in , then
) C

(p,q+1)
() and )

=

b


= 0 on `. Using (9.2.1), we have for any

-closed C

(np,nq1)
(),
_

) =
_

(
t
) =
_
M
= 0.
Thus, ) satises condition (9.1.9) in Theorem 9.1.3. We rst assume that the

-
Neumann operator
(np,nq1)
for (nj, n 1)-forms on is C

regular up
to the boundary and dene
(9.2.3) n =

(np,nq1)
).
Then n C

(p,q)
(). As in Theorem 9.1.3, it follows that

n = ) in and n
Dom(

). Since the boundary is smooth and n is smooth up to the boundary, we


can write n = n +n. Using Lemma 4.2.1, we have
(n) = 0 on `.
However, this is equivalent to
n = 0 on `.
Setting =
t
n, we have

= 0 in and =
t
= on `. This proves the
theorem assuming that
(np,nq1)
is C

regular.
In general, instead of (9.2.3), we dene
n
t
=

t
(np,nq1)
),
212 The Tangential Cauchy-Riemann Complex
where
t
(np,nq1)
is the weighted

-Neumann operator introduced in Theorem
6.1.2 with weight function
t
= t[.[
2
, t 0 and is taken with respect to the
metric 1
2
(1,
t
). Choosing t suciently 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)
() C
1
(p,q)
(),

n
k
= ) in , n
k
= 0 on ` and
| n
k
|
k()
C
k
| ) |
k()
.
To construct a solution n C

(p,q)
(), we set /
k
= n
k
n
k+1
. Each /
k
is

-closed
and /
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
nd a sequence /
k
n
C

(p,q)
() such that /
k
n
has compact support in , /
k
n
/
k
in
\
k
(p,q)
() and

/
k
n
0 in \
k
(p,q+1)
(). For each arbitrarily large : N, one can
nd
k
n
\
m
(p,q)
() with
k
n
= 0 on ` and

k
n
=

/
k
n
in . Setting

/
k
n
= /
k
n

k
n
,
we have

/
k
n
= 0 in ,

/
k
n
/
k
in \
k
(p,q)
() with

/
k
n
\
m
(p,q)
() and

/
k
n
= 0 on
`. This implies that inductively one can construct a new sequence n
t
k
\
k
(p,q)
()
such that

n
t
k
= ) in , n
t
k
= 0 on ` and
| n
t
k
n
t
k+1
|
k()
1,2
k
, / N.
Writing
n = n
t
N
+

k=N+1
(n
t
k
n
t
k1
),
we have n C

(p,q)
() such that

n = ) in , n = 0 on `. Setting =
t
n,
the rst part of the theorem is proved.
When 1 < n 1, setting =

n for some n C

(np,nq2)
() in (9.2.1),
we have
(9.2.4)
_
M
=
_
M


n = (1)
p+q+1
_
M

b
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 in C

(np,nq1)
() can be written as =

n for some
n C

(np,nq2)
() 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 be a bounded domain in C
n
with smooth boundary ` and let
be a smooth dening function for . If c
p,q
(`) and
b
= 0 on `, where
0 j n, 0 n 1, then there exists 1

such that 1

(p,q)
(C
n
),
1

= on ` and

= O(
k
) at ` for every positive integer /.
9.2

-Closed Extensions of Forms and C

Solvability of

b
213
Proof. We rst extend componentwise and smoothly from ` to 1 in C
n
. We
claim that for every positive integer /, there exist smooth (j, )-forms
1
, ,
k
and (j, + 1)-forms
1
, ,
k
such that
(9.2.5 a) 1
k
= 1
1


2
2

2


k
/

k
,
and
(9.2.5 b)

1
k
=
k
_

1
/

k
_
= O(
k
).
Since

b
= 0 on `,

1

= 0 on `. We can nd
1
and
1
such that

1 =


1
+
1
=

(
1
) +(
1

1
).
Setting 1
1
= 1
1
, it follows that

1
1
= (
1

1
) = O() at `. This
proves (9.2.5 a) and (9.2.5 b) for / = 1. We also note that
1
is obtained from the
rst order derivatives of 1 and
1
is obtained from the second derivatives of 1.
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
= /
k1

1
/

k
_
+
k

k
.
Hence


_

k
(1,/)

k
_
= 0 on `. Thus, we can nd a (j, )-form
k+1
and
a (j, + 1)-form
k+1
such that
k
(1,/)

k
=


k+1
+
k+1
. We dene
1
k+1
= 1
1


2
2

2


k
/

k


k+1
/ + 1

k+1
= 1
k


k+1
/ + 1

k+1
,
then

1
k+1
=
k
(


k+1
+
k+1
)

k+1
/ + 1

k+1
_
=
k+1
_

k+1

1
/ + 1

k+1
_
= O(
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 nd an extension 1

such that 1

= on ` and

1

= O(
k
) at
` for every positive integer /, we modify the construction as follows: Let

=
. [ < (.) < be a small tubular neighborhood of ` and (.) be a
cut-o function such that C

0
(

) and = 1 on
/2
. Let (.) denote the
projection from

onto ` along the normal direction and n(.) denote the unit
outward normal at . `. We dene
j
(.) =
_
(.) +
(z)
j
n((.))
_
and
1

= 1

j=1

j
(.)

j
,

j
,
214 The Tangential Cauchy-Riemann Complex
where c
j
is chosen to be suciently small and c
j
0. One can choose c
j
so small
(depending on
j
) such that, for each multiindex : = (:
1
, , :
2n
), we have

1
m
_

j
(.)

j
,

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

(p,q)
(C
n
) and
1

= on `. Furthermore, we have

_
_
1
k

j=1

j
(.)

j
,

j
_
_

_
_

j=k+1

j
(.)

j
,

j
_
_
= O(
k
) at `,
for every positive integer /. This proves the lemma.
Proof of Theorem 9.2.2. Let
t
= 1

where 1

is as in Lemma 9.2.3. Using


the proof of Theorem 9.2.1, there exists n C

(p,q)
() such that

n =

t
in and
n = 0 on `. Setting 1
0
= n, we have
(9.2.6 a)

1
0
=

t
in ,
(9.2.6 b) 1
0


= 0 on `.
We shall prove that for any nonnegative integer /, there exist (j, 1)-forms

0
,
1
, ,
k
and (j, )-form
k
such that
(9.2.7)
1
0
=

(
0
) +
1
2

(
2

1
) + +
1
/ + 1

(
k+1

k
)
+
k+1
_

1
/ + 1

k
_
.
From (9.2.6 a) and (9.2.6 b), we can write 1
0
=

0
+
0
=

(
0
)+(
0

0
)
for some (j, 1)-form
0
and (j, )-form
0
. This proves (9.2.7) for / = 0.
Assuming (9.2.7) is proved for / 0, from (9.2.6 a),

1
0
= (/ + 1)
k

1
/ + 1

k
_
+
k+1

k
=

t
.
Since

t
vanishes to arbitrarily high order at the boundary `, we have

(
k

1,(/ + 1)

k
) = 0 on ` and there exist a (j, 1)-form
k+1
and a (j, )-form

k+1
such that
k
1,(/+1)

k
=

k+1
+
k+1
. Substituting this into (9.2.7),
we obtain
1
0
=

(
0
) +
1
2

(
2

1
) + +
1
/ + 1

(
k+1

k
)
+
k+1
(


k+1
+
k+1
)
=

(
0
) +
1
2

(
2

1
) + +
1
/ + 1

(
k+1

k
) +
1
/ + 2

(
k+2

k+1
)
+
k+2
_

k+1

1
/ + 2

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
= 1
0

k

i=0
1
i + 1

(
i+1

i
),
we have
(9.2.9a)

1
k+1
=

t
in ,
(9.2.9b) 1
k+1
= O(
k+1
) at `.
Also each
i
is obtained by taking i-th derivatives of the components of 1
0
. Thus
each
i
is smooth and 1
k
C

(p,q)
().
Let (.) and
j
(.) be the same as in Lemma 9.2.3, we dene
(9.2.10) 1

= 1
0

i=0
1
i + 1

_

j
(.)
i+1

i
_
.
As in the proof of Lemma 9.2.3, we can choose c
i
suciently small such that the
series converges in every C
k
norm to some element 1

. 1

satises

1

t
in and 1

= O(
k
) at ` for every / = 1, 2, . Setting = 1

, we
have = on ` and

= 0 in . 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 = on `,
where is a (j, )-form with smooth coecients, 0 j n and 1 n 1. it
is easy to see that if (9.2.11) is solvable, then must satisfy
(9.2.12)

b
= 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,q1
(`), then must satisfy
(9.2.12 a)
_
M
= 0, c
np,nq1
(`) Ker(

b
)
where 1 n 1. We note that using Theorem 9.2.1, we can substitute in
(9.2.12 a) by C

(np,nq1)
() 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 in (9.2.12 a)
to be of the form

b
), where ) is any smooth (n j, n 2)-form on `.
216 The Tangential Cauchy-Riemann Complex
Theorem 9.2.4. Let be a bounded pseudoconvex domain in C
n
with smooth
boundary `. For any c
p,q
(`), where 0 j n and 1 n 1, there
exists n c
p,q1
(`) satisfying

b
n = on ` if and only if the following conditions
hold:

b
= 0 on `, when 1 < n 1,
and _
M
= 0, c
np,0
(`) Ker(

b
), when = n 1.
Proof. From Theorem 9.2.1, we can extend to such that C

(p,q)
(),

= 0, in
and


=

, on `.
Using Theorem 6.1.1, we can nd a n C

(p,q1)
() such that

n = in .
Denoting the restriction of n to ` by n, we have

n

=

on `, or
equivalently

b
n = on `. This proves the theorem.
We conclude this chapter with the following theorem:
Theorem 9.2.5. Let be a bounded pseudoconvex domain in C
n
with smooth
boundary `. For any c
p,q
(`), where 0 j n and 1 n 1, the
following conditions are equivalent:
(1) There exists n c
p,q1
(`) satisfying

b
n = on `.
(2) There exists C

(p,q)
() with = (or = ) on ` and

= 0 in .
(3)
_
M
= 0, c
np,nq1
(`) Ker(

b
).
When 1 < n 1, the above conditions are equivalent to
(4)

b
= 0 on `.
9.3 1
2
Existence Theorems and Sobolev Estimates for

b
Let ` be the boundary of a smooth domain 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
coecients, denoted by 1
2
(p,q)
(`), is the
completion of c
p,q
(`) under the sum of 1
2
norms of the coecients. We dene the
space of (j, )-forms with C
k
(`) coecients 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 dene the Sobolev space \
s
(`) for any real number :. Let \
s
(p,q)
(`) be the
subspace of 1
2
(p,q)
(`) with \
s
(`) coecients 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 dened operator
such that
(9.3.1)

b
: 1
2
(p,q1)
(`) 1
2
(p,q)
(`).
An element n 1
2
(p,q1)
(`) belongs to Dom(

b
) if and only if

b
n, dened 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 be a bounded pseudoconvex domain in C
n
with smooth
boundary `. For every \
s
(p,q)
(`), where 0 j n, 1 n 2 and : is a
nonnegative integer, such that
(9.3.2)

b
= 0 on `,
there exists n \
s
(p,q1)
(`) satisfying

b
n = on `.
When = n 1, 1
2
(p,n1)
(`) and satises
(9.3.3)
_
M
= 0, C

(np,0)
(`) Ker(

b
),
there exists n 1
2
(p,n2)
(`) satisfying

b
n = on `.
Corollary 9.3.2. Let be a bounded pseudoconvex domain in C
n
with smooth
boundary `. Then

b
: 1
2
(p,q1)
(`) 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 rst assume that is smooth. We
then show that there exists a constant C
s
independent of such that
(9.3.4) | n |
s(M)
C
s
| |
s(M)
.
Using Theorem 9.2.4, we can nd a solution for any smooth satisfying (9.3.2)
or (9.3.3). However, it is not easy to obtain estimates from this construction. We
shall use a dierent method to solve by exploiting the relationship between the
norms on the boundary ` and the tangential Sobolev norms. We also introduce
the weighted tangential Sobolev norms.
Let be a dening function for . Let

be a large ball such that

. We
set

+
=

,

= .
For a small 0, we set

= .

[ < (.) < 0,

= .
+
[ 0 < (.) < ,

= .

[ < (.) < ,
218 The Tangential Cauchy-Riemann Complex
and

= . C
n
[ (.) = c.
The special tangential norms in a tubular neighborhood

,
+

and

are dened,
as in Section 5.2, by
[[[)[[[
2
s(

)
=
_

|)|
2
s()
d,
[[[)[[[
2
s(

)
=
_
0

|)|
2
s()
d, [[[)[[[
2
s(
+

)
=
_

0
|)|
2
s()
d.
For each : N, : 1, we set
(9.3.5) [[[1
m
)[[[
s(

)
=

0km
[[[1
k

)[[[
s+mk(

)
,
where 1

= ,. We also dene the weighted tangential Sobolev norms by


(9.3.6)
[[[
m
)[[[
s(

)
= [[[
m
1
m
)[[[
s(

)
=

0km
[[[
m
1
k

)[[[
s+mk(

)
,
and similarly
[[[1
m
)[[[
s(

)
=

0km
[[[1
m
1
k

)[[[
s+mk(

)
.
Thus can be viewed as a rst order dierential operator weighted with . Cor-
responding norms are also dened on

and
+

similarly. We always assume


that 0 to be suciently 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 in C
n
. For arbitrary
smooth functions n
j
on `, , = 0, 1, , /
0
, there exists a function 1n C

0
(

)
such that 1
j

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[[[
sm+
1
2
(

)
C
k0

j=0
|n
j
|
sj(M)
,
(9.3.7 ii) [[[
m
1n[[[
s+
1
2
(

)
C
k0

j=0
|n
j
|
sj(M)
.
9.3 L
2
Existence Theorems and Sobolev Estimates for

b
219
Proof. Using a partition of unity, it suces 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
2n1
, .
The Fourier transform for n in the special boundary chart is dened by
n() =
_
R
2n1
c
it,)
n(t)dt,
where = (
1
, ,
2n1
) and t, ) = t
1

1
+ +t
2n1

2n1
.
Let 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)(, ) = ()
k0

j=0
n
j
()

j
,!
,
where = (1 +[[
2
)
1/2
. It is easy to see that 1
j

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

(()
j
)[
2
d =
2(ij)1
_

[1
i

(()
j
)[
2
d C
2(ij)1
.
Thus we have
[[[1
m
1n[[[
2
sm+
1
2
(

)
=

0km
[[[1
k

1n[[[
2
sk+
1
2
(

)
C
k0

j=0
_
R
2n1

2s2j
[ n
j
()[
2
d
C
k0

j=0
|n
j
|
2
sj(M)
,
which proves (9.3.7 i). Since
_

2m
[1
i

(()
j
)[
2
d =
2(ijm)1
_

2m
[1
i

(()
j
)[
2
d
C
2(ijm)1
,
we see that
[[[
m
1n[[[
2
s+
1
2
(

)
=

0km
[[[
m
1
k

1n[[[
2
s+mk+
1
2
(

)
C
k0

j=0
_
R
2n1

2s2j
[ n
j
()[
2
d
C
k0

j=0
|n
j
|
2
sj(M)
.
220 The Tangential Cauchy-Riemann 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 1 for any rst 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 in C
n
.
Let C

(p,q)
(`), 0 j n, 1 n 1, and
b
= 0 on `. For every
positive integer /, there exists a smooth extension 1
k
with support in a tubular
neighborhood

such that 1
k
C

(p,q)
(

), 1
k
= on ` and
(9.3.9)

1
k
= O(
k
) at `.
Furthermore, for every real number : and nonnegative integer :, there exists a
positive constant C
k
depending on : and : but independent of such that
(9.3.10 i) [[[1
m
1
k
[[[
sm+
1
2
(

)
C
k
||
s(M)
,
(9.3.10 ii) [[[
m
1
k
[[[
s+
1
2
(

)
C
k
||
s(M)
.
Proof. Using Lemma 9.3.3 with /
0
= 0, we rst extend componentwise and
smoothly from ` to 1 in C
n
such that 1 has compact support in

and
satises the estimates
(9.3.11 i) [[[1
m
1[[[
sm+
1
2
(

)
C||
s(M)
,
(9.3.11 ii) [[[
m
1[[[
s+
1
2
(

)
C||
s(M)
,
where C depends on : and : but is independent of . Using Lemma 9.2.3, for every
positive integer /, there exist smooth (j, )-forms
1
, ,
k
and (j, + 1)-forms

1
, ,
k
such that
(9.3.12 a) 1
k
= 1
1


2
2

2


k
/

k
,
and
(9.3.12 b)

1
k
=
k
_

1
/

k
_
= O(
k
) at `.
From the proof of Lemma 9.2.3, each component of
i
is a linear combination of
the i-th derivatives of 1. To show that 1
k
satises the estimates, it suces to
estimate each
i

i
. Using (9.3.11 i) and (9.3.11 ii), we have for any : 1,
[[[1
k
[[[
s+
1
2
(

)
[[[1[[[
s+
1
2
(

)
+

1ik
[[[
i

i
[[[
s+
1
2
(

)
C

0ik
[[[
i
1[[[
s+
1
2
(

)
C||
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,
[[[
m
1
k
[[[
s+
1
2
(

)
C

0ik
[[[
m+i
1[[[
s+
1
2
(

)
C
k
||
s(M)
,
and
[[[1
m
1
k
[[[
sm+
1
2
(

)
C

0ik
[[[
i
1
m
1[[[
sm+
1
2
(

)
C
k
||
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 dierence 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 in C
n
. Let
C

(p,q)
(`) with
b
= 0, 0 j n, 0 n1. For each positive integer /,
there exist
+
C
k
(p,q)
(
+
) and

C
k
(p,q)
() such that
+
= 0 in
+
,

= 0
in and the following decomposition holds:
(9.3.13)
+

= on `.
Furthermore, we have the following estimates: for every integer 0 : / 1,
0 : :,
(9.3.14 i) [[[1
m

+
[[[
sm+
1
2
(
+

)
C||
s(M)
,
(9.3.14 ii) [[[1
m

[[[
sm+
1
2
(

)
C||
s(M)
,
where the constant C depends only on :, :, but is independent of .
Proof. Let /
0
be a positive integer to be determined later. Using Lemma 9.3.4, we
extend from ` to 1
k0
in C
n
smoothly such that 1
k0
= on `, 1
k0
has
compact support in

and 1
k0
satises (9.3.9) and (9.3.10) with / = /
0
.
We dene a (j, + 1)-form

l
k0
in

by

l
k0
=
_

_
1
k0
, if .

,
0, if . `,
1
k0
, if .
+
.
From (9.3.9), we have

l
k0
C
k01
(

) and

l
k0
= 0 in

(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
[[[
sm
1
2
(

)
C[[[1
m+1
1
k0
[[[
sm
1
2
(

)
C||
s(M)
.
222 The Tangential Cauchy-Riemann Complex
We dene \
k0
=

(p,q+1)

l
k0
. It follows from Theorem 4.4.1 that \
k0
=

l
k0
in

. Since
(p,q+1)
is elliptic in the interior of the domain

,

(p,q+1)
gains
one derivative in the interior. Since

l
k0
has compact support in

, we get from
(9.3.15) that
(9.3.16)
[[[1
m
\
k0
[[[
sm+
1
2
(

)
C[[[1
m+1
1
k0
[[[
sm
1
2
(

)
C||
s(M)
,
for some C 0 independent of .
Setting

+
=
1
2
(1
k0
\
k0
), .
+
,

=
1
2
(1
k0
+\
k0
), . ,
we see that
= 1
k0
= (
+

) on `.
We also have

+
=
1
2
(1
k0
\
k0
) =
1
2
(1
k0


l
k0
) = 0 in
+
,
and

=
1
2
(1
k0
+\
k0
) =
1
2
(1
k0
+

l
k0
) = 0 in .
If we choose /
0
suciently large (/
0
n + / + 1), then
+
C
k
(p,q)
(
+
`) and

C
k
(p,q)
() 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

, the lemma is proved.
Using the weighted

-Neumann operator on

, we can solve

n

in

with good estimates up to the boundary. To solve



for
+
in
+
, we use the
following lemma to extend
+
to be

-closed in

with good estimates.
Lemma 9.3.6. Let be a bounded pseudoconvex domain in C
n
with C

boundary
`. Let C

(p,q)
(`) such that
b
= 0, where 0 j n, 0 < n 1. For
every nonnegative integer /
1
, there exists
+
in C
k1
(p,q)
(

) and

C
k1
(p,q)
() such
that
+
= 0 in

,

= 0 in

and
+

= on `. Furthermore, for
every 0 : /
1
, there exists a constant C depending only on : but independent of
such that
(9.3.17 i) |
+
|
s
1
2
(

)
C||
s(M)
.
(9.3.17 ii) |

|
s+
1
2
()
C||
s(M)
.
When = n 1, C

(p,n1)
(`) and satises
(9.3.18)
_
M
= 0, C

(np,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
+
,

be dened 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

[[[
sm+
1
2
(

)
is equivalent to |

|
s+
1
2
(

)
.
Thus (9.3.17 ii) follows immediately from (9.3.14 ii).
We next extend
+
to

. By the trace theorem for Sobolev spaces and inequality
(9.3.14 i), we have for any integer 0 , : 1, 0 : /,
|1
j

+
|
sj(M)
C|1
j+1


+
|
sj
1
2
(
+

)
C||
s(M)
.
Using the proof of Lemma 9.3.3, we can extend
+
from
+
to
t
in

such that

t
C
k
(

),
t
= 0 in
+
and the following estimates hold: for any integer
0 : / 1, 0 : :,
(9.3.19 i) [[[1
m

t
[[[
sm+
1
2
(

)
C||
s(M)
,
(9.3.19 ii) [[[
m

t
[[[
s+
1
2
(

)
C||
s(M)
.
We dene
1
t
=
t
(np,nq1)

t
in ,
where
t
(np,nq1)
is the weighted

-Neumann operator on (nj, n1)-forms.
Using Theorem 6.1.4, we can choose t suciently large such that 1
t
\
k1
(p,q)
()
C
2k1+1
(p,q)
(). We set 1
t
= 0 outside .
When = n 1, using the denition of
+
, for every

-closed form
C

(np,0)
(),
_

t
=
_

(
t
) =
_
M

t
=
1
2
_
M
\
k0

=
1
2
_

l
k0
=
1
2
_
M
= 0,
by (9.3.18). Thus

t
satises condition (9.1.9). Using Theorem 9.1.2 (1
n 2) and Theorem 9.1.3 ( = n 1) , it follows that
_

1
t
=

t
in

,
1
t
= 0 in
+
.
We modify 1
t
to make it smooth at `. Setting 1
0

t
= 1
t
and using argu-
ments similar to those in Lemma 9.2.3, one can choose (j, 1)-forms
0
, ,
k1
and dene
1
k1+1

t
= 1
0

k1

i=0
1
i + 1

(
i+1

i
),
such that

1
k1+1

t
=

t
in ,
224 The Tangential Cauchy-Riemann Complex
and
1
k1+1

t
= O(
k1+1
) at `.
Each
i
is obtained by taking i-th derivatives of the components of 1
0

t
. Thus
1
k1+1

t
C
k1
(p,q)
(). If we set 1
k1+1

t
= 0 in

, then 1
k1+1

t
C
k1
(p,q)
(

). We
dene

+
=
_

t
1
k1+1

t
, in ,

t
1
k1+1

t
=
+
, in

,
then
+
C
k1
(p,q)
(

) and
+
= 0 in

. It remains to show that
+
satises (9.3.17
i). Since
+
satises (9.3.14 i) and
t
satises (9.3.19 i), to estimate
+
, we only
need to estimate 1
k1+1

t
in . To prove (9.3.17 i), it suces to show
(9.3.20) | 1
k1+1

t
|
s
1
2
()
C | |
s(M)
.
From Theorem 6.1.4, we have the estimates
(9.3.21) | 1
t
|
s
1
2
()
C |

t
|
s
1
2
()
C | |
s(M)
.
We claim that for each positive integer 0 : /
1
+ 1, 0 : /
1
,
(9.3.22) [[[
m
1
m
1
t
[[[
s
1
2
(

)
C | |
s(M)
.
If the claim is true, then (9.3.20) holds from our construction of 1
k1+1

t
, since
1
k1+1

t
can be written as combinations of terms in 1
t
and
m
1
m
1
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 suces to show for each positive integer 0 : /
1
+ 1, 0 : /
1
,
(9.3.23) [[[1
m
1
t
[[[
s
3
2
(

)
C | |
s(M)
since 1
t
satises an elliptic system


t
. Decompose

into subdomains
j
such that

j
= .

[
j+1
< (.) <
j
,
where
j
= ,2
j
. This is a Whitney type decomposition where the thickness of
each
j
is comparable to the distance of
j
to the boundary. We dene

j
=

j1

j

j+1
. Let
j
be a function in C

0
(

j
) such that 0
j
1,
j
= 1
on
j
. Moreover,
(9.3.24) sup[grad
j
[ C
1
j
,
where C is independent of ,. Since


t
is elliptic and
j

m
1
t
is supported in

j
, applying Gardings inequality, we have
(9.3.25)
[[[1(
j

m
1
t
)[[[
s
3
2
(

j
)
C
_
[[[

(
j

m
1
t
)[[[
s
3
2
(

j
)
+[[[
t
(
j

m
1
t
)[[[
s
3
2
(

j
)
+[[[
j

m
1
t
[[[
s
3
2
(

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)
[[[

(
j

m
1
t
)[[[
s
3
2
(

j
)
[[[

(
j
)
m
1
t
[[[
s
3
2
(

j
)
+[[[
j

1
t
[[[
s
3
2
(

j
)
+[[[
j
[

,
m
]1
t
[[[
s
3
2
(

j
)
,
and using (9.3.24),
(9.3.27)
[[[

(
j
)
m
1
t
[[[
s
3
2
(

j
)
+[[[
j
[

,
m
]1
t
[[[
s
3
2
(

j
)
C[[[1
m1
1
t
[[[
s
3
2
(

j
)
.
Substituting (9.3.27) into (9.3.26), we obtain
(9.3.28)
[[[

(
j

m
1
t
)[[[
s
3
2
(

j
)
C([[[1
m1
1
t
[[[
s
3
2
(

j
)
+[[[
m

t
[[[
s
3
2
(

j
)
)
C([[[1
m1
1
t
[[[
s
3
2
(

j
)
+[[[1
m

t
[[[
s
3
2
(

j
)
).
Similarly,
(9.3.29)
[[[
t
(
j

m
1
t
)[[[
s
3
2
(

j
)
[[[
t
(
j
)
m
1
t
[[[
s
3
2
(

j
)
+[[[
j

t
1
t
[[[
s
3
2
(

j
)
+[[[
j
[
t
,
m
]1
t
[[[
s
3
2
(

j
)
C[[[1
m1
1
t
[[[
s
3
2
(

j
)
.
Substituting (9.3.28) and (9.3.29) into (9.3.25) and summing over ,, we have using
induction,
[[[1
m
1
t
[[[
s
3
2
(

2
)
C([[[1
m1
1
t
[[[
s
3
2
(

)
+[[[
m

t
[[[
s
3
2
(

)
)
C | |
s(M)
.
This proves (9.3.23) for a smaller . 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 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 since

b
= 0 on ` is always satised. In contrast, there is a compatibility
condition (9.3.18) for = n1 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 Cauchy-Riemann Complex
Lemma 9.3.7. Let be a bounded pseudoconvex domain in C
n
with smooth bound-
ary `. Let C

(p,q)
(`), where 0 j n, 1 n 2, such that

b
= 0
on `. For every nonnegative integer :, there exists n
s
\
s
(p,q1)
(`) satisfying

b
n
s
= on `. Furthermore, there exists a constant C
s
independent of such
that
(9.3.30) |n
s
|
s(M)
C
s
||
s(M)
.
When = n 1, C

(p,n1)
(`) and satises (9.3.3), the same conclusion
holds.
Proof. Let : be a xed nonnegative integer. From Lemma 9.3.6, there exists a
decomposition = (
+

) such that
+
C
s
(p,q)
(

),


+
= 0 in

,


C
s
(p,q)
() and

= 0 in . We also have the estimates:


(9.3.31 i) |
+
|
s
1
2
(

)
C||
s(M)
,
(9.3.31 ii) |

|
s+
1
2
()
C||
s(M)
.
(9.3.31 i) and (9.3.31 ii) follow from (9.3.17 i) and (9.3.17 ii).
Since is pseudoconvex, we dene
n

t
(p,q)

,
where
t
(p,q)
is the weighted

-Neumann operator in . If we choose t 0 suf-
ciently large, using Theorem 6.1.4, it follows that n

\
s+
1
2
(p,q1)
(), n

and
[[[1n

[[[
s
1
2
(

)
C|

|
s+
1
2
()
C||
s(M)
for some constant C independent of . Restricting n

to the boundary we have

b
n

on ` and using the trace theorem for Sobolev spaces, we obtain


(9.3.32) |n

|
s(M)
C
s
[[[1n

[[[
s
1
2
(

)
C
s
||
s(M)
.
Dening
n
+
=

(p,q)

+
,
where

(p,q)
is the

-Neumann operator on

, we have

n
+
=
+
in

and n
+
is
one derivative smoother than
+
in the interior of

. From (9.3.31 i),
[[[1 n
+
[[[
s
1
2
(

)
C|
+
|
s
1
2
(

)
C||
s(M)
for some constant C independent of . Restricting n
+
to `, using the trace theorem
again,
(9.3.33) | n
+
|
s(M)
C[[[1 n
+
[[[
s
1
2
(

)
C||
s(M)
for some constant C independent of . Letting
n
s
= n
+
n

on `,
we get
b
n
s
= on `. We also have from (9.3.32) and (9.3.33),
|n
s
|
s(M)
C
s
||
s(M)
,
where C
s
is independent of . 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
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 dene Z

= C

(p,q)
(`) ker(

b
) and Z
s
= \
s
(p,q)
(`) ker(

b
). For
any Z
s
, using Friedrichs Lemma (see Appendix D), there exists a sequence of
smooth forms
m
C

(p,q)
(`) such that
m
in \
s
(p,q)
(`) and

m
0 in
\
s
(p,q+1)
(`). Since

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 suciently smooth form
m
such that

m
=

m
in ` with
|
m
|
s(M)
C
s
|

m
|
s(M)
0.
We set

t
m
=
m

m
.
Then

t
m
= 0 and
t
m
converges to 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
k
Z
k
such that
|
k
|
s(M)
< c.
Furthermore, we can require that
|
k

k+1
|
k(M)
<
c
2
k
since Z
k+1
is dense in Z
k
. The series

k
+

N=k+1
(
N

N1
)
converges in every \
k
(p,q)
(`) norm to some element

. The Sobolev embedding


theorem then assures that

is in C

(p,q)
(`) Ker(

b
). We also have
|

|
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
in C
n
, n 2 and 0 j n. Let Z denote the space of all forms in 1
2
(p,n1)
(`)
satisfying (9.3.3) and Z

be the subspace of all forms in C

(p,n1)
(`) 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 1
2
(n,n1)
(`), we can write = )(

) for some ) 1
2
(`). Using the
Hahn-Banach theorem, it suces to show that any bounded linear functional / on
228 The Tangential Cauchy-Riemann Complex
1
2
(n,n1)
(`) 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
/() =
_
M
p, 1
2
(n,n1)
(`).
For any n C

(n,n2)
(`), 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,n2)
(`).
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 Z, we have
/() =
_
M
p = lim
m
_
M
p
m
= 0.
This proves the lemma.
We can now nish the proof of Theorem 9.3.1.
Proof of Theorem 9.3.1. When 1 < n 1, can be approximated by smooth

b
-closed forms
m
in \
s
(p,q)
(`) according to Lemma 9.3.8. For each
m
, we apply
Lemma 9.3.7 to obtain a (j, 1)-form n
m
such that

b
n
m
=
m
and
|n
m
|
s(M)
C
s
|
m
|
s(M)
.
Thus, n
m
converges to some (j, 1)-form n such that

b
n = on ` and
|n|
s(M)
C
s
||
s(M)
.
This proves the theorem when 1 < n 1.
When = n 1, we approximate by
m
Z

in the 1
2
(p,n1)
(`) norm
using Lemma 9.3.9. Repeating the arguments above with : = 0, we can construct
n 1
2
(p,n2)
(`) with

b
n = . 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 dene
b
as in Chapter 8. Let
b
=
b

b
+

b
be dened on Dom(
b
),
where
Dom(
b
) = 1
2
(p,q)
(`) [ Dom(
b
) Dom(

b
);

b
Dom(

b
) and

b
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 dened self-adjoint 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 Ker(
b
) = Ker(

b
) Ker(

b
). Then =

b
n for some
n 1
2
(p,q1)
(`) by Theorem 9.3.1. Thus
(, ) = (

b
n, ) = (n,

b
) = 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
Szego projection) and H
b
(p,n1)


o
(p,n1)
are nontrivial where
o
(p,0)
: 1
2
(p,0)
(`) Ker(

b
),

o
(p,n1)
: 1
2
(p,n1)
(`) 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 Szego projection on functions and

o denote
the projection from 1
2
(n,n1)
(`) onto 1
2
(n,n1)
Ker(

b
). For any 1
2
(n,n1)
(`),
we can write = )(

) 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 Szego projection:
Lemma 9.4.1. For any 1
2
(n,n1)
(`), the following conditions are equivalent:
(1) satises condition (9.3.3).
(2) o

) = 0, where = )(

).
(3)

o = 0.
Proof. Let = (

). Since d = ( +

) vanishes when restricted to `, we
have
=

= d
= d +

on `. If 1
2
(`),
_
M
=
_
M
)(

) =
_
M
)(d)
_
M
)()
=
_
M
)d
_
M
)(

),
230 The Tangential Cauchy-Riemann Complex
where d = d is the surface measure on `. Hence, for any 1
2
(`),
_
M
=
1
2
_
M
)d =
1
2
(),

)
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 = p(

) for any
1
2
(n,n1)
(`), where p 1
2
(`). It is easy to see that Ker(

b
) if and only if

b
p = 0 on `. For any 1
2
(n,n1)
(`) Ker(

b
),
(, )
M
=
_
M
) pd = 2
_
M
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
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 (
b
) Dom(
b
).
(2) For any 1
2
(p,q)
(`), we have
=
b

b
, if 1 n 2,
=

b
o
(p,0)
, if = 0,
=
b

b


o
(p,n1)
, if = n 1.
(3) If 1 n 2, we have

b
=
b

b
= 1 on Dom(
b
),

b
=
b

b
on Dom(
b
),

b
=
b

b
on Dom(

b
).
(4) If 1
2
(p,q)
(`) with
b
= 0, where 1 n 2 or 1
2
(p,n1)
(`)
with

o
(p,n1)
= 0, then =
b

b
.
The solution n =

b
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 (

b
), is closed in every
degree. If 1 n 2, we have from Theorem 9.3.1, Ker(

b
) = (

b
) and the
following orthogonal decomposition:
(9.4.2) 1
2
(p,q)
(`) = Ker(

b
) (

b
) = (

b
) (

b
).
Repeating the arguments of Theorem 4.4.1, we can prove that for every
Dom(
b
) Dom(

b
),
(9.4.3) ||
2
c(|
b
|
2
+|

b
|
2
),
Notes 231
and for any Dom(
b
),
(9.4.4) ||
2
c|
b
, |
2
where the constant c is independent of .
(9.4.4) implies that
b
is one-to-one 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)
(`) = (
b
) Ker(
b
)
=

b
(Dom(
b
))

b
(Dom(
b
)).
Thus
b
: Dom(
b
) 1
2
(p,q)
(`) is one-to-one, 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
satises all the conditions (1)-(4) and
Theorem 9.4.2 is proved when 1 n 2.
When = 0,
1
2
(p,0)
(`) = (

b
) Ker

b
.
Thus for any Ker(

b
) and Dom(
b
),
(9.4.5) ||
2
c|

b
|
2
c|
b
|||.
Thus
b
has closed range on Ker(

b
)

= Ker(
b
)

and
1
2
(p,0)
(`) = (
b
) Ker

b
.
In particular, there exists a bounded operator
b
: 1
2
(p,0)
(`) 1
2
(p,0)
(`) satisfying

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 = n1 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
rst introduced the

b
complex. In [KoRo 1], they show that a

-closed extension
exists for any (j, )-form from the boundary ` to the domain in a complex
manifold if satises condition 7(n 1). Formula (9.1.3) was rst 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 rst 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 Kohn-Rossi [KoRo 1].
The

Cauchy problem was also discussed by M. Derridj in [Der 1,2]. A. Andreotti
232 The Tangential Cauchy-Riemann 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
Henkin-Leiterer [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. Boas-M.-C.
Shaw [BoSh 1]. In [Sha 1], Kohns 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 two-sided

-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 Bochner-Martinelli-Koppelman 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 Szego projection in Sobolev spaces.
For more discussion on the Sobolev estimates for the Szego projection, see [Boa
1,2,4]. Another proof of Theorem 9.3.1 was given by J. J. Kohn in [Koh 11] using
pseudodierential 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
closed-range 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 nite ideal type. If the domain has a plurisubharmonic dening
function, Sobolev estimates for
b
have been obtained by H. P. Boas and E. J.
Straube [BoSt 4]. In particular, the Szego projections are exactly regular in both
cases.
Much less is known for the regularity of
b
and the Szego projection on pseu-
doconvex manifolds in other function spaces except when the C1 manifold is the
boundary of a pseudoconvex domain of nite type in C
2
. We mention the work
of Feerman-Kohn [FeKo 1] and Nagel-Rosay-Stein-Wainger [NRSW 1] and Christ
[Chr 1]. When the C1 manifold is of nite type in C
n
, n 3, and the Levi form
is diagonalizable, Holder estimates for

b
and
b
have been obtained in Feerman-
Kohn-Machedon [FKM 1]. Holder and 1
p
estimates for the Szego projection on
convex domains of nite type have been obtained by J. McNeal and E. M. Stein
[McSt 2]. It is still an open question whether Holder or 1
p
estimates hold for
b
and

b
on general pseudoconvex C1 manifolds of nite 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 1,2-estimates 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 Cauchy-Riemann 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 Holder
spaces. The Cauchy-Szego 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 sucient 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
n
is dened by
(10.1.1)
n
= (.
t
, .
n
) C
n
[ Im.
n
[.
t
[
2
,
where .
t
= (.
1
, , .
n1
) and [.
t
[
2
= [.
1
[
2
+ +[.
n1
[
2
. Denote by Aut(
n
) the
group of all holomorphic mappings that are one-to-one from
n
onto itself. Let
H
n
Aut(
n
) be the subgroup dened by
(10.1.2) H
n
= /
a
; o /
n
[ /
a
(.) = (o
t
+.
t
, o
n
+.
n
+ 2i.
t
, o
t
)),
where .
t
, o
t
) is the standard inner product in C
n1
, i.e., .
t
, o
t
) =

n1
i=1
.
i
o
i
. To see
H
n
actually forms a subgroup of Aut(
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
n
into
n
and /
n
into /
n
. It is easily veried that if
o, / /
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(
n
)
10.1 Fundamental Solutions for
b
on the Heisenberg Group 235
and it induces a group structure on the boundary /
n
. The boundary /
n
can be
identied with H
n
= C
n1
1 via the mapping
(10.1.3) : (.
t
, t +i[.
t
[
2
) (.
t
, t),
where .
n
= t +i:. We shall call H
n
= C
n1
1 the Heisenberg group of order n1
with the group structure induced from the automorphism subgroup H
n
of Aut(
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 veried that
7
j
=

.
j
+i.
j

t
, , = 1, , n 1, and
T =

t
,
are left invariant vector elds 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 elds spanned by 7
1
, , 7
n1
. We x a left
invariant metric on H
n
so that 7
1
, , 7
n1
, 7
1
, , 7
n1
and T are orthonormal
with respect to this metric. Let the dual basis be
1
, ,
n1
,
1
, ,
n1
and
, where
j
= dr
j
+idj
j
, , = 1, , n 1 and is given by
= dt + 2
n1

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
1n
dr
1
dr
n1
dj
1
dj
n1
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

J
,
where J = (,
1
, , ,
q
) is an increasing multiindex and
J
=
j1

jq
. Then,
we have

b
) =

t
]J]=q
_

k
7
k
)
J

k
_

J
=

t
]L]=q+1
_

t
k,J
c
L
kJ
7
k
)
J
_

L
,
236 Fundamental Solutions for
b
on the Heisenberg Group
and

b
) =

t
]H]=q1
_

t
l,J
c
J
lH
7
l
)
J
_

H
.
It follows that

b
) =

t
]Q]=q
_

t
l,L
c
L
lQ
7
l
_

t
k,J
c
L
kJ
7
k
)
J
__

Q
,
and

b
) =

t
]Q]=q
_

t
k,H
c
Q
kH
7
k
_

t
l,J
c
J
lH
7
l
)
J
__

Q
.
For xed Q and | ,= /, it is easily veried 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
)
_

t
]J]=q
)
J

J
_
=

t
]J]=q
__

k/ J
7
k
7
k
+

kJ
7
k
7
k
_
)
J
_

J
.
The calculation shows that
b
acts on a (0, )-form ) diagonally. It is also easily
veried that

k/ J
7
k
7
k
+

kJ
7
k
7
k
_
=
1
2
n1

k=1
(7
k
7
k
+7
k
7
k
) +i(n 1 2)T.
Therefore, to invert
b
it suces to invert the operators
(10.1.6)
1
2
n1

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

dened by
(10.1.7) L

=
1
2
n1

k=1
(7
k
7
k
+7
k
7
k
) +iT,
for C.
The second order term L
0
=
1
2

n1
k=1
(7
k
7
k
+7
k
7
k
) is usually called the sub-
Laplacian on a stratied Lie group. By denition, a Lie group is stratied 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 stratied with : = 2, where \
1
is generated by 7
1
, , 7
n1
, 7
1
, , 7
n1
and
\
2
is generated by T. By Theorems 8.2.3 and 8.2.5, L
0
satises a subelliptic esti-
mate of order 1,2 and is hypoelliptic. We want to construct an explicit fundamental
solution
0
for L
0
. The group structure on the Heisenberg group suggests that one
can dene a nonisotropic dilation in the following way: for o 0,
o(.
t
, t) = (o.
t
, o
2
t)
which forms an one-parameter subgroup of Aut(H
n
). We also dene 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
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
0
(.
t
, t) = [(.
t
, t)[
2(n1)
= ([.
t
[
4
+t
2
)

n1
2
. Then L
0

0
=
c
0
, where 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. Dene, for c 0,

0,
(.
t
, t) = (([.
t
[
2
+c
2
)
2
+t
2
)

n1
2
.
Then, a simple calculation shows that
L
0

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

_
1
c
(.
t
, t)
_
,
238 Fundamental Solutions for
b
on the Heisenberg Group
where (.
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

0,
(.
t
, t) d\ =
_
Hn
c
2n

_
1
c
(.
t
, t)
_
d\
=
_
Hn
(.
t
, t) d\ = c
0
.
Hence, lim
0
L
0

0,
= c
0
in the distribution sense. On the other hand, L
0

0,
also tends to L
0

0
in the distribution sense. This proves the theorem.
It follows that c
1
0

0
is the fundamental solution for L
0
. We now proceed to
search for a fundamental solution

for L

with C. Observe that L

has
the same homogeneity properties as L
0
with respect to the nonisotropic dilation on
H
n
, and that L

is invariant under unitary transformation in .


t
-variable. Hence,
we can expect that certain

will have the same invariant properties. From these


observations we intend to look for a fundamental solution

of the form

(.
t
, t) = [(.
t
, t)[
2(n1)
)(t[(.
t
, t)[
2
).
After a routine, but lengthy, calculation, we see that ) must satisfy the following
ordinary second order dierential equation:
(10.1.9)
(1 n
2
)
3
2
)
tt
(n)(nn(1 n
2
)
1
2
+i(1 n
2
)))
t
(n)
+i(n 1)n)(n) = 0
with n = t[(.
t
, t)[
2
.
By setting n = cos, )(n) = p(), 0 , (10.1.9) is reduced to
(10.1.10)
_
sin
d
d
+ (n 1)cos
__
d
d
+i
_
p() = 0.
Equation (10.1.10) has two linearly independent solutions
p
1
() = c
i
,
and
p
2
() = c
i
_
c
i
(sin)
n1
d.
Hence, the only bounded solutions for 0 are p() = cc
i
. It follows that
)(n) = c(n i
_
1 n
2
)

= c
_
t i[.
t
[
2
[(.
t
, t)[
2
_

.
If we choose c = i

, then
(10.1.11)

(.
t
, t) = ([.
t
[
2
it)

(n1+)
2
([.
t
[
2
+it)

(n1)
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

(.
t
, t) be dened as in (10.1.11) for C. Then,
L

= c

, where
c

=
2
42n

n
(
n1+
2
)(
n1
2
)
.
Proof. For any c 0, set

(.
t
, t) = [.
t
[
2
+c
2
it,
and dene
(10.1.12)
,
(.
t
, t) =

(n1+)
2

(n1)
2

.
Hence,
,
is smooth and
,
tends to

as distributions when c approaches zero.


It follows that L

,
tends to L

as distributions. Thus, it suces to show that


L

,
tends to c

in the distribution sense.


Recall that 7
k
= (,.
k
) +i.
k
(,t), for 1 / n 1. Then, for a xed c and
any o C, a direct computation shows
7
k

= 2o.
k

a1

, and 7
k

= 2o.
k

a1

,
7
k

= 7
k

= 0,
T
a

= io
a1

, and T
a

= io
a1

.
It follows that
L

,
(.
t
, t) =
_

1
2
n1

k=1
(7
k
7
k
+7
k
7
k
) +iT
__

(n1+)
2

(n1)
2

_
= c
2
((n 1)
2

2
)

(n+1+)
2

(n+1)
2

= c
2n
L

,1
_
1
c
(.
t
, t)
_
.
Hence,
_
Hn
L

,
(.
t
, t) d\ =
_
Hn
c
2n
L

,1
_
1
c
(.
t
, t)
_
d\
=
_
Hn
L

,1
(.
t
, t) d\.
Since the mass of L

,
concentrates at zero as c 0, it follows that L

,
tends to c

with c

=
_
L

,1
d\ as distributions. Therefore, it remains only to
compute the integral c

.
We set o =
1
2
(n + 1 +) and / =
1
2
(n + 1 ), then
c

=
1
2
n1
_
Hn
((n 1)
2

2
)([.
t
[
2
+ 1 it)
a
([.
t
[
2
+ 1 +it)
b
drdjdt
=
((n 1)
2

2
)
2
n1
_
C
n1
([.
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
n1
dj
1
dj
n1
.
The rst integral is evaluated by
_
C
n1
([.
t
[
2
+ 1)
n
drdj =
2
n1
(n 1)
_

0
:
2n3
(1 +:
2
)
n
d:
=

n1
(n 1)
_

1
t
n
(t 1)
n2
dt, t = 1 +:
2
=

n1
(n 1)
_
1
0
(1 :)
n2
d:, : = t
1
=

n1
(n)
,
where () denotes the Gamma function. For the second integral we rst assume
that (n 1) n 1 so that o 1 and / 1. We start with the formula
_

0
c
xs
r
b1
dr = (/):
b
which is valid if the real part of : is positive. Set : = 1 +it, then
(/)(1 +it)
b
=
_

0
c
ixt
c
x
r
b1
dr =

)(t),
where

)(t) is the Fourier transform of )(r) dened by
)(r) =
_
c
x
r
b1
, for r 0
0, for r 0.
Similarly, we obtain
(o)(1 it)
a
=
_

0
c
ixt
c
x
r
a1
dr
=
_
0

c
ixt
c
]x]
[r[
a1
dr
= p(t),
where p(r) is dened by
p(r) =
_
0, for r 0
c
]x]
[r[
a1
, for r < 0.
Hence, by the Plancherel theorem, we have
(o)(/)
_

(1 it)
a
(1 +it)
b
dt =
_

)(t) p(t) dt
= 2
_

)(r)p(r) dr
= 2
_

0
c
2x
r
a+b2
dr
=
(n)
2
n1
.
10.1 Fundamental Solutions for
b
on the Heisenberg Group 241
This implies
(10.1.13)
_

(1 it)
a
(1 +it)
b
dt =
2
(n1)
(n)
(o)(/)
,
for (n 1) n 1. In fact, the left-hand side of (10.1.13) denes an entire
function of from the following equality:
_

(1 it)
a
(1 +it)
b
dt =
_

(1 +t
2
)

n+1
2
c
itan
1
t
dt.
Hence, (10.1.13) holds for all C, and we obtain
c

=
((n 1)
2

2
)
2
n1


n1
(n)

2
(n1)
(n)
(
n+1+
2
)(
n+1
2
)
=
2
42n

n
(
n1+
2
)(
n1
2
)
.
This completes the proof of Theorem 10.1.2.
It follows from Theorem 10.1.2 that c

= 0 if = (2/ +n1) for any nonneg-


ative integer /. Hence, if ,= (2/ +n 1) for / N 0, then

= c
1

is a
fundamental solution for L

.
We now derive some consequences from Theorem 10.1.2. The convolution of two
functions ) and p on H
n
is dened by
) p(n) =
_
Hn
)()p(
1
n) d\ () =
_
Hn
)(n
1
)p() d\ ().
Set p(n) = p(n
1
), then
_
Hn
() p)(n)/(n) d\ (n) =
_
Hn
)(n)(/ p)(n) d\ (n),
provided that both sides make sense.
If ,= (2/ + n 1) for / N 0, then for any ) C

0
(H
n
), dene
1

) = )

. It is clear that 1

) C

(H
n
) since

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 ,= (2/+n1), then L

) = 1

) =
).
Proof. Since L

is left invariant, clearly we have L

) = ). For the other equality,


let p C

0
(H
n
). Notice that ,= (2/ + n 1) whenever ,= (2/ + n 1).
Then
_
p(n))(n) d\ (n) =
_
L

p(n))(n) d\ (n)
=
_
(p

)(n)L

)(n) d\ (n)
=
_
p(n)(L

)(n) d\ (n)
=
_
p(n)1

)(n) d\ (n).
Hence, 1

) = ). This proves the theorem.


242 Fundamental Solutions for
b
on the Heisenberg Group
Theorem 10.1.4. The operator L

is hypoelliptic if and only if ,= (2/+n1).


In particular,
b
is hypoelliptic on H
n
for (0, )-forms when 1 < n 1.
Proof. If = (2/ + n 1) for some nonnegative integer /, then the function

(.
t
, t) dened in (10.1.11) is a nonsmooth solution to the equation L

= 0.
Next, let ,= (2/ +n 1), ) T
t
such that L

) = p is smooth on some open


set l. Let \ l be an open set which is relatively compact in l. Choose a
cut-o function C

0
(l) with = 1 in some open neighborhood of \ . Then,
by Theorem 10.1.3, 1

(p) is smooth and satises L

(p) = p. Hence, to
show that ) is smooth on \ , it suces to show that / = () 1

(p)) is smooth
on \ . Since / is a distribution with compact support, a standard argument from
functional analysis shows / = 1

/. But on \ we have
L

/ = L

) L

(p) = p p = 0.
The fact that

(.
t
, t) is just singular at the origin will then guarantee that 1

/
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) satises
b
n = ) and n

1/2
(0,q)
(H
n
, loc), where 1 = 1

with = 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
)1) = ),
where 1 acts on ) componentwise. The hypothesis
b
) = 0 implies

b
1) = 1
b
) = 0.
Hence,

b
1) = ).
For the regularity of n =

b
1), we write ) =

t
]J]=q
)
J

J
. Then, we obtain
from the previous calculation that

b
1) =

t
]H]=q1
_

t
l,J
c
J
lH
7
l
(1)
J
)
_

H
.
Hence, it suces to estimate the following integral, for 1 , n 1:
(10.1.14) 7
j
_
Hn
)()(
1
) d\ (),
10.1 Fundamental Solutions for
b
on the Heisenberg Group 243
where ) is a continuous function with compact support on H
n
and = (.
t
, t),
= (n
t
, n) and
(.
t
, t) = ([.
t
[
2
it)
(n1q)
([.
t
[
2
+it)
q
.
We can rewrite (10.1.14) as
(10.1.15) 2(n 1 ))
j
() = 2(n 1 )
_
Hn
)()
j
(
1
) d\ (),
where

j
(.
t
, t) = .
j
([.
t
[
2
it)
(n1q)1
([.
t
[
2
+it)
q
.
Dene a new kernel
j
y
(.
t
, t, j) on H
n
1
+
, where 1
+
= j 1[ j 0, by

j
y
(.
t
, t, j) = .
j
([.
t
[
2
it +j)
(n1q)1
([.
t
[
2
+it +j)
q
.
It is easily seen that )
j
y
(.
t
, t, j) is smooth on H
n
1
+
and
lim
y0
+
)
j
y
(.
t
, t, j) = )
j
(.
t
, t).
The assertion then follows from the Hardy-Littlewood lemma proved in Theorem
C.1 in the Appendix if one can show, for 0 < j 1,2, that
(10.1.16) [()
j
y
)[ cj

1
2
| ) |
L

(Hn)
,
for some constant c 0 and 1 , n 1.
Since
1
= (. n, t n 2Imn .), we may introduce new coordinates

2j1
= Re(.
j
n
j
),
2j
= Im(.
j
n
j
) and = t n2Im(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
=
_
](,)]M
d
1
d
2n2
d
([[
2
+[[ +j)
n
cj

1
2
,
(10.1.18) 1
2
=
_
](,)]M
[
j
[d
1
d
2n2
d
([[
2
+[[ +j)
n+1
cj

1
2
,
and
(10.1.19) 1
3
=
_
](,)]M
[
j

k
[d
1
d
2n2
d
([[
2
+[[ +j)
n+1
cj

1
2
,
where ` 0 is a positive constant. Notice rst that (10.1.19) follows immediately
from (10.1.17). For 1
1
we have
1
1
c
_
M
0
_
M
0
:
2n3
(:
2
+ +j)
n
d:d
c
_
M
0
:
2n3
(:
2
+j)
n1
d:
c
_ M

y
0
r
2n3
(1 +r
2
)
n1
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
:
2n2
(:
2
+ +j)
n+1
d:d
c
_
M
0
:
2n2
(:
2
+j)
n
d:
cj

1
2
_

0
r
2n2
(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 elds 7
1
, , 7
n1
,7
1
, , 7
n1
and T. Then, from (10.1.14) and the
convolution formula we get
A7
j
_
Hn
)()(
1
) d\ () =
_
Hn
A)()7
j
(
1
) d\ ()
= 7
j
_
Hn
A)()(
1
) d\ ().
Since the left invariant vector elds 7
j
, 7
j
, where , = 1, , n1, and T span the
tangent space of H
n
, the assertion now follows from dierentiating / times and the
rst 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 dene. For each 1 / n 1, let

n,k
= (.
t
, .
n
) C
n
[ Im.
n
[.
1
[
2
+ +[.
k
[
2
[.
k+1
[
2
[.
n1
[
2
.
The boundary of
n,k
is identied with the generalized Heisenberg group H
n,k
=
C
n1
1 by
:
_
.
t
, t +i
_
k

j=1
[.
j
[
2

n1

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 dened by
(10.1.20) (.
t
, t) (n
t
, n) =
_
.
t
+n
t
, t +n + 2Im
_
k

j=1
.
j
n
j

n1

j=k+1
.
j
n
j
__
.
One veries 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 left-invariant vector elds 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 dene a nondegen-
erate C1 structure on H
n,k
such that the Levi matrix has / positive eigenvalues
and n1 / 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 x a left-invariant metric on H
n,k
which makes 7
j
, 7
j
and T, 1 , n 1,
orthonormal. The dual basis is given by
1
, ,
n1
,
1
, ,
n1
and , where

j
= dr
j
+idj
j
for 1 , n 1 and is given by
= dt + 2
k

j=1
(r
j
dj
j
j
j
dr
j
) 2
n1

j=k+1
(r
j
dj
j
j
j
dr
j
).
The volume element is
d\ = 2
1n
dr
1
dr
n1
dj
1
dj
n1
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

J
= [1 J[ +[1
t
J[ [1 J[ [1
t
J[,
where [ [ denotes cardinality of the set. Hence, if ) =

t
]J]=q
)
J

J
is a smooth
(0, )-form with compact support on H
n,k
, we get

b
) = (
b

b
+

b
)
_

t
]J]=q
)
J

J
_
=

t
]J]=q
__

m/ J
7
m
7
m
+

mJ
7
m
7
m
_
)
J
_

J
=

t
]J]=q
__

1
2
n1

m=1
(7
m
7
m
+7
m
7
m
) +i
J
T
_
)
J
_

J
.
Notice that (n 1)
J
(n 1). The extreme case
J
= n 1 occurs if and
only if [J[ = n 1 / and J = 1
t
. On the other hand,
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 Cauchy-Szego Kernel on the Heisenberg Group
In this section we compute the Szego projection on H
n
. Let
n
be the Siegel
upper half space in C
n
. Denote by H
2
(
n
) the Hardy space of all holomorphic
functions ) dened on
n
such that
sup
s>0
| )
s
(.) |
L
2
(bn)
< ,
where )
s
(.) = )(.
t
, .
n
+ i:) for . = (.
t
, .
n
) /
n
and : 0. It will be clear
later that H
2
(
n
) forms a Hilbert space under the norm | ) |
H
2
(n)
= sup
s>0
| )
s
|
L
2
(bn)
.
If )(.) H
2
(
n
), then by denition )(.) satises
(10.2.1)
_
C
n1
_

[)(.
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
n1
dj
n1
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
n1
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
, ). The
resulting function

)
s
(.
t
, ) is 1
2
integrable with respect to and satises
(10.2.2)
1
2
_
C
n1
_

)
s
(.
t
, )[
2
ddr
t
dj
t
< C.
Since ) is holomorphic on
n
, we have by Cauchys theorem
(10.2.3)

)
s+s
(.
t
, ) = c
s


)
s
(.
t
, ),
for :, :
t
0. It follows that, for xed : 0, we get that
(10.2.4)
_
C
n1
_

[)(.
t
, t +i[.
t
[
2
+i: +i:
t
)[
2
dtdr
t
dj
t
=
1
2
_
C
n1
_

)
s
(.
t
, )[
2
c
2s

ddr
t
dj
t
,
which implies

)
s
(.
t
, ) = 0 a.e. for < 0. Therefore, we may assume that

)
s
(.
t
, )
is concentrated on 1
+
= r 1[ r 0 with respect to . It is also clear from
(10.2.3) that

)
s
(.
t
, ) =

)
0
(.
t
, )c
s
,
10.2 The Cauchy-Szego Kernel on the Heisenberg Group 247
for some measurable function

)
0
(.
t
, ). We set

)
0
(.
t
, ) =

)(.
t
, )c
]z

]
2
.
Since )(.) is holomorphic on
n
, the homogeneous tangential Cauchy-Riemann
equation on each level set .
n
= t +i([.
t
[
2
+:) with : 0 must be satised by ),
namely,
_

.
k
i.
k

t
_
)(.
t
, t +i([.
t
[
2
+:)) = 0, 1 / n 1.
It follows that
_

.
k
+.
k
_

)
s
(.
t
, ) = 0, 1 / n 1,
for : 0. Hence, for 1 / n 1, we have
0 =
_

.
k
+.
k
_

)
0
(.
t
, )
=
_

.
k
+.
k
_
(

)(.
t
, )c
]z

]
2
)
=


)
.
k
(.
t
, )c
]z

]
2
.
This shows that

)(.
t
, ) is holomorphic in .
t
and measurable in . By substituting

)(.
t
, ) into (10.2.2), we obtain
1
2
_
C
n1
_

0
[

)(.
t
, )[
2
c
2]z

]
2
c
2s
ddr
t
dj
t
< C,
where the constant C 0 is independent of : 0. Letting : tend to zero, we see
that the function

)(.
t
, ) satises
(10.2.5)
1
2
_
C
n1
_

0
[

)(.
t
, )[
2
c
2]z

]
2
ddr
t
dj
t
< C,
and the function )(.) is recovered by
(10.2.6) )(.) = )(.
t
, .
n
) =
1
2
_

0

)(.
t
, )c
izn
d,
for .
n
= t +i[.
t
[
2
+i: with : 0. Moreover, the Plancherel theorem shows that
(10.2.7)
lim
s,s

0
_
C
n1
_

[)(.
t
, t +i[.
t
[
2
+i:) )(.
t
, t +i[.
t
[
2
+i:
t
)[
2
dtdr
t
dj
t
= lim
s,s

0
_
C
n1
_

0
[

)(.
t
, )[
2
c
2]z

]
2
(c
s
c
s

)
2
ddr
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
, ) with property
(10.2.5) is also sucient for representing a function )(.) in the Hardy space H
2
(
n
).
248 Fundamental Solutions for
b
on the Heisenberg Group
Theorem 10.2.1. A complex-valued function ) dened on
n
belongs to H
2
(
n
) if
and only if there exists a function

)(.
t
, ), (.
t
, ) C
n1
1
+
, which is holomorphic
in .
t
and measurable in and satises
1
2
_
C
n1
_

0
[

)(.
t
, )[
2
c
2]z

]
2
ddr
t
dj
t
< ,
where dr
t
dj
t
stands for dr
1
dj
1
dr
n1
dj
n1
and 1
+
= r 1[ r 0.
The integral
)(.) = )(.
t
, .
n
) =
1
2
_

0

)(.
t
, )c
izn
d,
for .
n
= t + i[.
t
[
2
+ i:, converges absolutely for all : 0 and denes a function
)(.) H
2
(
n
).
Proof. Suppose that there is a function

)(.
t
, ) which is holomorphic in .
t
and
measurable in and

)(.
t
, ) satises (10.2.5). For any .
t
and .
n
= t + i[.
t
[
2
+ i:
with : 0, we may choose a polydisc 1(.
t
; :) in C
n1
centered at .
t
with small
multiradii : = (:
1
, , :
n1
) 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 Holders inequality that
_

0
[

)(.
t
, )c
izn
[ d

_

0
__
D(z

;r)
[

)(n
t
, )[ d\ (n
t
)
_
c
]z

]
2
s
d

_

0
__
D(z

;r)
[

)(n
t
, )[c
]w

]
2
d\ (n
t
)
_
c
s/2
d

__
D(z

;r)
_

0
[

)(n
t
, )[
2
c
2]w

]
2
dd\ (n
t
)
_1
2
__
D(z

;r)
_

0
c
s
dd\ (n
t
)
_1
2
< .
This shows that the integral dened by (10.2.6) converges absolutely and denes
a holomorphic function on
n
. To see )(.) is actually in H
2
(
n
), we apply the
Plancherel theorem to the -variable and get
_
C
n1
_

[)(.
t
, t +i[.
t
[
2
+i:)[
2
dtdr
t
dj
t
=
1
2
_
C
n1
_

0
[

)(.
t
, )[
2
c
2]z

]
2
c
2s
ddr
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
(
n
) can be identied with
a closed subspace of 1
2
(/
n
), namely, any )(.) H
2
(
n
) is identied with its 1
2
limiting value )(.
t
, t +i[.
t
[
2
) on /
n
with the norm
| ) |
H
2
(n)
= sup
s>0
| )
s
|
L
2
(bn)
= | )(.
t
, t +i[.
t
[
2
) |
L
2
(bn)
.
Thus, following the procedure of the Bergman kernel function, we obtain the repro-
ducing kernel, named Cauchy-Szego kernel, o(., n) for the Hardy space H
2
(
n
).
We make the following denition:
10.2 The Cauchy-Szego Kernel on the Heisenberg Group 249
Denition 10.2.2. The Cauchy-Szego kernel associated with
n
is the unique func-
tion o(., n) which is holomorphic in . and antiholomorphic in n with respect to
.
n
and n
n
such that
(10.2.8) )(.) =
_
bn
o(., n))(n) d
w
,
for any ) H
2
(
n
) and any .
n
, where d
w
is the surface element on /
n
.
For each xed .
n
, (10.2.8) denes a bounded linear functional on H
2
(
n
).
It is also clear from general Hilbert space theory that o(., n) can be expressed in
terms of any orthonormal basis
k
(.)

k=1
of H
2
(
n
), i.e.,
o(., n) =

k=1

k
(.)
k
(n).
Now we want to calculate the Cauchy-Szego kernel o(., n) on the Siegel upper
half space
n
. One way to achieve this goal is via the pullback of the Cauchy-Szego
kernel on the unit ball in C
n
by the inverse Cayley transform. Recall that the
Cayley transform n = (.) is a biholomorphic mapping from the unit ball 1
n
in
C
n
onto the Siegel upper half space
n
dened by (7.3.2). Thus, the inverse Cayley
transform =
1
is given by
:
n
1
n
n . =
_
2in
1
1 in
n
, ,
2in
n1
1 in
n
,
1 +in
n
1 in
n
_
.
First, by constructing an orthonormal basis for H
2
(1
n
) directly, the Cauchy-
Szego kernel on the unit ball can be calculated as follows:
Proposition 10.2.3. The Cauchy-Szego kernel o(, ) on the unit ball in C
n
can
be expressed explicitly as
(10.2.9) o(, ) =
(n 1)!
2
n
1
(1 )
n
,
where =
1

1
+ +
n

n
.
Proof. It is clear that

forms an orthogonal basis for the Hardy space H


2
(1
n
),
where = (
1
, ,
n
) is any multiindex with
j
N 0 for 1 , n.
Therefore, to get the Cauchy-Szego kernel, we need to normalize

.
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

=
_
bBn
[

[
2
d
2n1
= 2([[ +n)
_
Bn
[

[
2
d\
2n
=
2
n

n
!
1
!
([[ +n 1)!
.
250 Fundamental Solutions for
b
on the Heisenberg Group
It follows that the Cauchy-Szego kernel o(, ) on the unit ball is given by
o(, ) =

1
c

=
(n 1)!
2
n
_
1 +

,=0
n(n + 1) (n +[[ 1)

1
!
n
!

_
=
(n 1)!
2
n
_
1 +

k=1

]]=k
n(n + 1) (n +[[ 1)

1
!
n
!

_
=
(n 1)!
2
n
_
1 +

k=1
n(n + 1) (n +/ 1)
/!
()
k
_
=
(n 1)!
2
n
1
(1 )
n
.
This proves the proposition.
Our next step is to pull the Cauchy-Szego kernel o(, ) on the unit ball back to
the Siegel upper half space. Denote by :(.) =

n
j=1
[.
j
[
2
1 the dening function
for the unit ball in C
n
, and x 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: = Re

__
n

j=1
.
j
d.
j
_

_
1
2i
_
n
_
n

k=1
d.
k
d.
k
__
= Re

_
n

j=1
1
2
n1
i
n
.
j
d.
j

_

k,=j
d.
k
d.
k
__
,
where denotes the interior product and : /1
n
C
n
is the inclusion map. Hence,
the pullback of d: by the inverse Cayley transform is

(d:) =
1
2
n1
i
n
_
2
2n1
i
[1 in
n
[
2n

1 in
n
1 in
n
dn
n

_
n1

j=1
dn
j
dn
j
_
+
n1

j=1
2
2n1
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 /
n
, we have
dn
n
= dt +i
n1

j=1
(n
j
dn
j
+n
j
dn
j
).
It follows that
dn
n

_
n1

j=1
dn
j
dn
j
_
= dt
_
n1

j=1
dn
j
dn
j
_
,
10.2 The Cauchy-Szego Kernel on the Heisenberg Group 251
and, for 1 , n 1,
dn
j

_

k,=j
dn
k
dn
k
_
= 2in
j
dt
_
n1

j=1
dn
j
dn
j
_
,
on the boundary of
n
. Thus,

(d:)
= (2i)
n
_
i
[1 in
n
[
2n

1 in
n
1 in
n
+
n1

k=1
2i[n
k
[
2
(1 +in
n
)
[1 in
n
[
2n+2
_
dt
_
n1

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
_
n1

j=1
dn
j
dn
j
_
=
2
2n1
[1 in
n
[
2n
dt
_
n1

j=1
dn
j
d
j
_
,
where n
j
= n
j
+i
j
.
Since the surface element d on the boundary /
n
is given by d = dt
(
n1
j=1
dn
j
d
j
), the above calculation suggests that the Cauchy-Szego 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)!
2
n

1
(1 (.)(n))
n
=
(1)
n
2
n2
(n 1)!

n

_
i(.
n
n
n
) + 2
n1

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
(
n
) as stated in Denition 10.2.2. Theorem 10.2.1
suggests that one should check the Fourier transform of o(., n).
Notice rst that, for .
n
and n
n
, we have
Re
_
i(.
n
n
n
) + 2
n1

j=1
.
j
n
j
_
= j
n

n
+
n1

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
n2

n
_

0

n1
c
(i(znwn)+2z

)
d,
and the above integral converges absolutely for .
n
and n
n
. Here we use
.
t
n
t
to denote the inner product

n1
j=1
.
j
n
j
in C
n1
.
252 Fundamental Solutions for
b
on the Heisenberg Group
Dene
(10.2.12)

o(.
t
, ; n) =
_
2

_
n1

n1
c
(iwn+2z

)
.
We shall show that for each n
n
, the integral
(10.2.13)
1
2
_
C
n1
_

0
[

o(.
t
, ; n)[
2
c
2]z

]
2
ddr
t
dj
t
converges. Since n
n
, n
n
= n +i[n
t
[
2
+i with 0. Hence, (10.2.13) can be
rewritten as
2
2n3

2n1
_
C
n1
_

0

2n2
c
2]z

]
2
c
2v
ddr
t
dj
t
=
2
2n3

2n1
_
C
n1
_

0

2n2
c
2]z

]
2
c
2v
ddr
t
dj
t
=
2
2n2

n
(n 2)!
_

0
_

0

2n2
c
2r
2
c
2v
:
2n3
d:d
=
2
n2

n
_

0

n1
c
2v
d
=
(n 1)!
4
n

n
.
It follows now from Theorem 10.2.1 and (10.2.11) that
o(., n) =
2
n2

n
_

0

n1
c
(i(znwn)+2z

)
d
=
1
2
_

0

o(.
t
, ; n)c
izn
d.
Hence, for each n
n
, o(, n) H
2
(
n
). Let n
j
=
j
+ i
j
for 1 , n 1
and let d
t
d
t
denote d
1
d
1
d
n1
d
n1
. Since o(., n) = o(n, .), for
any .
n
and any ) H
2
(
n
), we obtain
_
bn
o(., n))(n) dnd
t
d
t
=
1
2
_
C
n1
_

0

o(n
t
, ; .)

)(n
t
, )c
2]w

]
2
dd
t
d
t
=
2
n2

n
_
C
n1
_

0

n1
c
(izn+2z

2]w

]
2
)

)(n
t
, ) dd
t
d
t
10.2 The Cauchy-Szego Kernel on the Heisenberg Group 253
=
2
n2

n
_
C
n1
_

0

n1
c
(izn+2]z

]
2
2z

2]w

]
2
)

)(n
t
, ) dd
t
d
t
=
2
n2

n
_
C
n1
_

0

n1
c
(izn+2]z

]
2
)
c
2]

]
2
c
2z

(z


)(.
t
+
t
, ) dd
t
d
t
=
1
2
2
n
(n 2)!
_

0

n1
c
izn
)(.
t
, )
__

0
c
2
2

2n3
d
_
d
=
1
2
_

0

)(.
t
, )c
izn
d
= )(.),
where
j
= n
j
.
j
=
j
+i
j
for 1 , n 1 and d
t
d
t
denotes d
1
d
1

d
n1
d
n1
. 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
(
n
). Hence, by the uniqueness of the Cauchy-Szego kernel
function, we obtain the following theorem.
Theorem 10.2.4. The Cauchy-Szego kernel function o(., n) associated with the
Siegel upper half space
n
is given by
o(., n) = o((.
t
, .
n
), (n
t
, n
n
))
=
(1)
n
2
n2
(n 1)!

n

_
i(.
n
n
n
) + 2
n1

j=1
.
j
n
j
_
n
.
Hence, for any ) 1
2
(/
n
), the integral
(10.2.14) o)(.) =
_
bn
o(., n))(n) d(n),
denes a function o)(.) in the Hardy space H
2
(
n
) which has a well-dened 1
2
integrable limiting value on /
n
. We recall that the Szego projection on /
n
is the
orthogonal projection from 1
2
(/
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
(
n
). We shall still use (10.2.14) to denote the Szego
projection on /
n
.
Since, for each ) 1
2
(/
n
), o) H
2
(
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 /
n
. Let

(.
t
, t) = [.
t
[
2
+c
2
it,
on /
n
. Then we have
254 Fundamental Solutions for
b
on the Heisenberg Group
Proposition 10.2.5. For any ) 1
2
(/
n
) and any c 0, (o))

is given by
(10.2.15) (o))

(.
t
, t +i[.
t
[
2
) =
2
n2
(n 1)!

n
)
n

(.
t
, t),
where the convolution is taken with respect to the group structure on /
n
, and the
coordinates on /
n
are .
t
= (.
1
, , .
n1
) and t.
Proof. Let = (.
t
, t +i[.
t
[
2
) and = (n
t
, n +i[n
t
[
2
). Hence,
2
n2
(n 1)!

n
)
n

(.
t
, t) =
2
n2
(n 1)!

n
_
bn
)()
n

(
1
) d().
A direct calculation shows that

1
= (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

(
1
) = ([.
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
n2
(n 1)!

n
)
n

(.
t
, t)
=
(1)
n
2
n2
(n 1)!

n
_
bn
)(n
t
, n)
(i(.
n
+ic
2
n
n
) + 2.
t
n
t
)
n
d()
= o)(.
t
, t +i[.
t
[
2
+ic
2
).
This proves the proposition.
Finally, we can describe the Szego projection on the Heisenberg group H
n
as
follows:
Theorem 10.2.6. The Szego projection o) for any ) 1
2
(H
n
) is given by
(10.2.16) o)(.
t
, t) = lim
0
2
2n3
(n 1)!

n
)
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
n1
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
1n
dr
1
dj
1
dr
n1
dj
n1
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
n1
=
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 Szego projection o, there exists a
relative fundamental solution for
0
b
. Rewrite
0
b
as
(10.3.1)
0
b
= L

i( n + 1)T,
for C. Recall that

(.
t
, t) = ([.
t
[
2
it)

(n1+)
2
([.
t
[
2
+it)

(n1)
2
,
and
c

=
2
42n

n
(
n1+
2
)(
n1
2
)
.
Then, we have

0
b

= L

i( n + 1)T

= c

i( n + 1)T

.
(10.3.2)
Now with the aid of the identity
(n)(1 n) =

sinn
,
we formally dierentiate (10.3.2) with respect to and evaluate it at = n 1 to
get
(10.3.3)
0
b
_
(n 2)!
2
42n

n
log
_
[.
t
[
2
it
[.
t
[
2
+it
_
([.
t
[
2
it)
n+1
_
=
2(n 1)!
2
42n

n
([.
t
[
2
it)
n
.
Here the logarithm of the quotient means the dierence of the corresponding loga-
rithm. Set
(10.3.4) =
(n 2)!
2
42n

n
log
_
[.
t
[
2
it
[.
t
[
2
+it
_
([.
t
[
2
it)
n+1
,
and dene the operator 1 by
(10.3.5) 1) = ) ,
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 dened 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 suces to show only that
0
b
1 = 1 o. The other identity then follows
immediately by transposition.
Set

= [.
t
[
2
+c
2
it, and dene

(.
t
, t) =
(n 2)!
2
42n

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

=
n1

k=1
7
k
7
k

=
(n 2)!
2
42n

n
_
4(n 1)
[.
t
[
2

n
+ 2(n 1)
1

n1
_
=
(n 1)!
2
42n

n
_
4c
2

_
.
Hence, as c 0, we get by the integral evaluated in the proof of Theorem 10.1.2,
that

0
b
=
_
4(n 1)!
2
42n

n
_
Hn
([.
t
[
2
+ 1 it)
n
([.
t
[
2
+ 1 +it)
1
d\
_

2(n 1)!
2
42n

n
1
([.
t
[
2
it)
n
=
(n 1)!
2
32n

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 Cauchy-Kowalevski 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 cut-o function
with = 1 in some open neighborhood of j. Set

0
b
(n) = /.
Then, o/ = 0 and ) / = 0 in some neighborhood of j. Now, from the explicit
formula (10.2.16) of the Szego 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 = non. 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 Lewys example can be extended to any tangential Cauchy-
Riemann equation 1 of a hypersurface in C
2
which is not Levi-at, 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 Szego
projection is fullled:
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
Szego 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 dening function for 1 and 1 =
(:,.
2
)(,.
1
)(:,.
1
)(,.
2
). Suppose that the Szego projection o associated
with /1 satises the analyticity hypothesis. Then the tangential Cauchy-Riemann
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 Holder 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
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

= c
1

dened 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

) = )

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
(
n
) on the Siegel upper half space was
proved by S. G. Gindikin [Gin 1] (Theorem 10.2.1). The Cauchy-Szego 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
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
Cauchy-Riemann equation on a hypersurface which is not Levi-at in C
2
was proved
by L. Hormander [Hor 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 Szego projection
o on weakly pseudoconvex boundaries satises 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 4-6, 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 Cauchy-Riemann operator in the
Sobolev spaces on compact C1 manifolds. However, Holder 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 Holder estimates were obtained
for solutions of

b
. Our goal here is to construct integral formulas for solutions of

and

b
with Holder and 1
p
estimates on strictly convex domains.
In Section 11.1, some terminology necessary in developing the kernel formulas is
dened. We derive the Bochner-Martinelli-Koppelman formula as a generalization
of the Cauchy integral. Unlike the Cauchy kernel in C
1
, the Bochner-Martinelli-
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. Holder estimates for the solutions of

on strictly convex domains are
obtained. In Section 11.3 the jump formula derived from the Bochner-Martinelli-
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 rst goal is to nd 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 dierentiating the fundamental solution for . Since
1
2
log [.[ =
2

2
..
log [.[ =
0
,
where
0
is the Dirac delta function centered at 0, we have
2

.
log [.[ =
1

.
1
.
=
0
.
This implies that 1(.) = 1,. 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 dene
n(.) = ) 1(.) =
1

_
D
)()
.
d\ =
1
2i
_
D
)()
.
d d

.
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 = )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 to be solvable, we can derive a solution for the equation

n = as
follows: Let c(.) be a fundamental solution for in C
n
, n 2, dened by
c(.) = c(:) =
(n 2)!
4
n
1
:
2n2
, : = [.[.
Both c(:) and all the rst order derivatives of c(:) are locally integrable functions.
We dene 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
=
0
d .
1
d .
n
.
We dene
1(.) =

c
n
=
n

j=1
(1)
j
(n 1)!

n
.
j
:
2n
d .
1

d .
j
d .
n
,
where

d .
j
denotes that the term d .
j
is omitted. It follows that

1(.) =
0
d .
1
d .
n
.
262 Integral Representations for and
b
Dene
(11.1.1)
n(.) = ) 1(.)
=
n

j=1
_
(1)
j
(n 1)!

n
_
D
.
j

j
[. [
2n
)()d\
_
[

d .
j
],
where [

d .
j
] = d .
1

d .
j
d .
n
. Then n satises

n = 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 rst.
Let ( .) = (
1
.
1
,
2
.
2
, ,
n
.
n
) C
n
and d = (d
1
, , d
n
). Let
= (o
1
, , o
n
), 1 = (/
1
, , /
n
) be two vectors in C
n
. We dene
< , 1 =
n

i=1
o
i
/
i
, < , d =
n

i=1
o
i
d
i
.
Thus <

., . = [ .[
2
and <

., d =

n
i=1
(

i
.
i
)d
i
. Let \ be an
open subset of C
n
C
n
with coordinates (, .) and let G(, .) be a C
1
map from
\ into C
n
such that G(, .) = (p
1
(, .), , p
n
(, .)). We dene the (1, 0)-form
G
by

G
=
1
2i
< G(, .), d
< G(, .), .
=
1
2i

n
i=1
p
i
(, .)d
i

n
i=1
p
i
(, .)(
i
.
i
)
on the set of (, .) C
n
C
n
where < G, . ,= 0.
When n = 1,
G
is independent of G and is the Cauchy kernel. We set the
Cauchy-Riemann operator on \ equal to

,z
=

z
,
and
<

,z
G(, .), d =
n

i=1

,z
p
i
(, .) d
i
.
Let (G) be an (n, n 1)-form in (, .) dened by
(G) =
G
(

,z

G
)
n1
=
G

,z

,z

G
. .
n1 times
.
Given : maps G
i
: \ C
n
, i = 1, , :, we abbreviate
G
i
by
i
and
1m
is
the (n, n :)-form dened by

1m
= (G
1
, , G
m
)
=
1

m

k1++km=nm
(

,z

1
)
k1
(

,z

m
)
km
on the set where all the denominators are nonvanishing. Since
(11.1.2)

,z

i
=
1
2i
<

,z
G
i
, d
< G
i
(, .), .

1
2i
<

,z
G
i
, . < G
i
, d
(< G
i
, . )
2
,
11.1 Integral Kernels in Several Complex Variables 263
we have for / 0 that
(11.1.3)
i
(

,z

i
)
k
=
_
1
2i
_
k+1
< G
i
, d
< G
i
, .

_
<

,z
G
i
, d
< G
i
, .
_
k
.
This follows from the fact that
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
(, .) : \ C
n
C
n
C
n
, i = 1, , :, be C
1
maps.
We have
(11.1.4)

,z

1m
=
m

j=1
(1)
j

jm
on the set where < G
i
, . ,= 0 for every i = 1, , :, where

, denotes that the
term , is omitted. In particular, we have
(11.1.4-i)

,z

1
= 0,
(11.1.4-ii)

,z

12
=
1

2
,
(11.1.4-iii)

,z

123
=
23
+
13

12
,
on the set where the denominators are nonvanishing.
Proof. We use the notation
(

,z
)
Km
= (

,z

1
)
k1
(

,z

m
)
km
for each multiindex 1
m
= (/
1
, , /
m
) and [1
m
[ = /
1
+ +/
m
. It follows that

,z

1m
=
m

j=1
(1)
j1

,z

j

m

]Km]=nm
(

,z
)
Km
=
m

j=1
(1)
j1

j

m

]Km]=nm+1,
k
j
1
(

,z
)
Km
=
m

j=1
(1)
j

jm
+
m

j=1
(1)
j1

j

m

]Km]=nm+1,
k
j
0
(

,z
)
Km
.
We claim that for each 1
m
such that [1
m
[ = n :+ 1,
m

j=1
(1)
j1

j

m
(

,z
)
Km
= 0.
264 Integral Representations for and
b
We rst observe that
(11.1.5)
1

j

m
(

,z
)
Km
= 0,
since there are n + 1 d
t
:. Dene
= 2i
n

j=1
(
j
.
j
)d
j
.
It is easy to see that
(11.1.6)
i
= 1, i = 1, , :,
where denotes the interior product.
Also from (11.1.2), we have
(11.1.7)

,z

i
= 0, i = 1, , :.
Contraction of equation (11.1.5) with , using (11.1.6) and (11.1.7), gives
0 = (
1

j

m
(

,z
)
Km
)
=
m

j=1
(1)
j1

1
(
j
)
m
(

,z
)
Km
=
m

j=1
(1)
j1

j

m
(

,z
)
Km
.
This proves the claim and the lemma.
We also write

1m
=
nm

q=0

1m
q
,
where
1m
q
denotes the piece in
1m
which is of degree (0, ) in . and (n, n:)
degree in . If ) is a (0,
t
)-form in C
n
, the form
1m
q
(, .))() is a form of degree
(n, n: +
t
) in and of degree (0, ) in .. We write the form
1m
q
(, .))()
uniquely as

t
]J]=q

J
(, .) d .
J
where
J
(, .) is an (n, n : +
t
)-form in
only with coecients depending on . and J is an increasing multiindex. We
shall dene the integration of the the form
1m
(, .) )() with respect to the
variables on a (2n |)-dimensional real manifold ` as follows:
_
M

1m
(, .) )() =
_
M

1m
q
(, .) )()
=

]J]=q
t
__
M

J
(, .)
_
d .
J
,
11.1 Integral Kernels in Several Complex Variables 265
where = | :+
t
, provided the integral on the right-hand side exists. Note that
from this denition, we have

z
_
M

1m
(, .) )() = (1)
2nl
_
M

1m
(, .) )()
= (1)
l
_
M

1m
(, .) )(),
provided that the dierentiation under the integral sign is allowed.
Let
G
0
(, .) = (

.) = (

1
.
1
, ,

n
.
n
).
The Bochner-Martinelli-Koppelman kernel 1(, .) is dened by
(11.1.8)
1(, .) = (G
0
) =
0
=
1
(2i)
n
<

., d
[ .[
2

_
< d

d ., d
[ .[
2
_
n1
=
n1

q=0
1
q
(, .),
where 1
q
is the summand which is of degree (0, ) in . and of degree (n, n 1)
in . Using (11.1.4-i), we have
(11.1.9)

1(, .) +

z
1(, .) = 0 for ,= .,
or equivalently, for each 0 n,
(11.1.9-q)

1
q
(, .) +

z
1
q1
(, .) = 0 for ,= .,
if we set 1
1
(, .) = 1
n
(, .) = 0. In particular, 1
0
is the Bochner-Martinelli
kernel dened by (2.2.1) and (11.1.9-q) was proved directly in (3.2.2) when = 1.
When n = 1, 1(, .) = (2i)
1
d,( .) is the Cauchy kernel. The following
theorem shows that the Bochner-Martinelli-Koppelman kernel is indeed a general-
ization of the Cauchy integral formula to several variables.
Theorem 11.1.2 (Bochner-Martinelli-Koppelman). 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
(, .)

)
+

z
_
D
1
q1
(, .) ), . 1,
where 1(, .) is dened in (11.1.8).
Proof. For = 0, the Bochner-Martinelli formula was proved in Theorem 2.2.1. We
rst assume that 1 < n.
266 Integral Representations for and
b
Let .
0
1 and

be a small ball of radius c centered at .


0
such that

1.
We shall prove the theorem at . = .
0
. Applying Stokes theorem to the form
d

(1
q
(, .) )()) on 1

, we have, using (11.1.9), that


(11.1.11)
_
bD
1
q
(, .) )
_
b
1
q
(, .) )
=
_
D

1
q
(, .) )
_
D
1
q
(, .)

)
=
_
D

z
1
q1
(, .) )
_
D
1
q
(, .)

).
Since 1(, .) = O([ .[
2n+1
), 1(, .) is an integrable function for each xed ..
We see from the dominated convergence theorem that
(11.1.12)
_
D
1
q
(, .)

)
_
D
1
q
(, .)

).
Note that
_
b
1
0
(, .) =
1
(2i)
n
1
c
2n
_
b
<

., d < d

, d
n1
=
1
(2i)
n
1
c
2n
_

< d

, d
n
=
n!

n
_
1
d\ = 1.
For any increasing multiindex J = (,
1
, , ,
q
), we get that
_
b
1
q
(, .) d

J
=
1
(2i)
n
1
c
2n
_
b
<

., d < d

d ., d
n1
d

J
=
1
(2i)
n
1
c
2n
_

< d

, d < d

d ., d
n1
d

J
=
(n )(n 1)!

n
__
1
d\
_
d .
J
=
n
n
d .
J
.
Let )(.
0
) denote the (0, )-form whose coecients are equal to the values of the
coecients of ) at .
0
. It follows from the above calculation that
(11.1.13)
_
b
1
q
(, .) ) =
n
n
)(.
0
) +
_
b
1
q
(, .)
_
) )(.
0
)
_

n
n
)(.
0
), as c 0.
The kernel

z
1
q1
(, .) = O([ .[
2n
) is not integrable but the components are the
classical singular integrals (see e.g. Stein [Ste 2]). The Principal-value limit dened
by
P.V.
_
D

z
1
q1
(, .
0
) ) = lim
0
_
D

z
1
q1
(, .
0
) )
= lim
0
_

z
_
D
1
q1
(, .) )
_

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
q1
(, .
0
) ) =
_

z
_
D
1
q1
(, .) )
_

z=z0


n
)(.
0
).
We use the notation d

d = d

1
d
1
d

n
d
n
and [

d
j
]= d

1
d
1

d
j
d

n
d
n
, where

d
j
denotes that the term d
j
is omitted. Let
)() = )
J
d

J
, where J = (1, , ). Using Stokes theorem, we obtain
_

1
q1
(, .) )()
=
(n 1)!
(2i)
n
q

j=1
(1)
j
__

j
.
j
[ .[
2n
)
J
()d

d
_
d .
1

jq
=
(n 1)!
(2i)
n
q

j=1
(1)
j
_

_
1
n 1
_

j
_
1
[ .[
2n2
_
)
J
() d

d d .
1

jq
=
(n 1)!
(2i)
n
q

j=1
(1)
j
n 1
__
b
1
[ .[
2n2
)
J
()[

d
j
]
+
_

1
[ .[
2n2
)
J

j
()d

d
_
d .
1

jq
.
Thus,

z
_

1
q1
(, .) )() =
(n 1)!
(2i)
n
q

j=1
(1)
j
n

k=1
__
b

k
.
k
[ .[
2n
)
J
()[

d
j
]
+
_

k
.
k
[ .[
2n
)
J

j
()d

d
_
d .
k
d .
1

jq
.
Since
(n 1)!
(2i)
n
(1)
j
_
b

k
.
k
[ .[
2n
)
J
()[

d
j
]
jk
(1)
j1
n
)
J
(.
0
),
and
_

k
.
k
[ .[
2n
)
J

j
()d

d 0, as c 0,
we have

z
_

1
q1
(, .) )()

n
)
J
(.
0
)d .
J
.
268 Integral Representations for and
b
Thus

z
__
D
1
q1
(, .) )()
_

z=z0
= lim
0
_
_

z
_
D
1
q1
(, .) )() +

z
_

1
q1
(, .) )()
_

z=z0
_
= lim
0
_
_
D

z
1
q1
(, .
0
) )() +

z
_

1
q1
(, .) )()

z=z0
_
= P.V.
_
D

z
1
q1
(, .
0
) )() +

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
n1
(, .) ).
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

(0,q1)
(1) with

n = ) in the distribution
sense, where 0 < < 1. Furthermore, there exists a C 0 such that
(11.1.15) |n|
C

(D)
C|)|
L

(D)
.
Proof. For . 1, dene
n(.) =
_
D
1
q1
(, .) ).
We rst prove (11.1.15). Since

_
D
1
q1
(, .) )
_
D
1
q1
(, .
t
) )

C
_
_
n

j=1
_
D

j
.
j
[ .[
2n

j
.
t
j
[ .
t
[
2n

d\
_
_
|)|
L

(D)
,
it suces to show that for each 1 , n,
(11.1.16)
_
D

j
.
j
[ .[
2n

j
.
t
j
[ .
t
[
2n

d\ C[. .
t
[

log [. .
t
[

.
Let [. .
t
[ = 2c. We divide 1 into three regions:

(.),

(.
t
) and 1

= 1
(

(.)

(.
t
)) where

(.) is a ball of radius c centered at .. On

(.), we have
_
(z)

j
.
j
[ .[
2n

j
.
t
j
[ .
t
[
2n

d\ 2
_
(z)
1
[ .[
2n1
d\ C[. .
t
[.
11.2 The Homotopy Formula for

on Convex Domains 269
Similarly, we have the estimate on

(.
t
). To estimate the integral on 1

, we note
that
1
3
[ .
t
[ [ .[ 3[ .
t
[ for 1

, thus there exists an 0 such that


_
D

j
.
j
[ .[
2n

j
.
t
j
[ .
t
[
2n

d\ C
_
]z]A
[. .
t
[
[ .[
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 rst assume that the domain
1 is star-shaped and 0 1. Let (.) = ([.[) be a function supported in [.[ 1
and 0,
_
= 1. We set
m
=
2n
m
(.,
m
) for
m
0. Extending ) to be 0
outside 1, we dene
)
m
(.) = )
_
.
1
1
m
_

m
for suciently small
m
. One can easily check that )
m
has coecients 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
i

N
i=1
, with each
i
supported in an open set l
i
such that
l
i
1 is star-shaped. We then regularize
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 Bochner-Martinelli-Koppelman 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 Bochner-Martinelli-Koppelman kernel is independent of the domain 1. Next
we introduce another kernel, the Leray kernel, which in general depends on the
domain.
Denition 11.2.1. A C
n
-valued C
1
function G(, .) = (p
1
(, .), , p
n
(, .)) is
called a Leray map for 1 if it satises < G(, .), . ,= 0 for every (, .)
/1 1.
In particular, the C
n
-valued function G
0
(, .) = (

.) = (

1
.
1
, ,

n
.
n
)
is a Leray map for any domain 1. We use the same notation
0
= (G
0
) = 1(, .)
to denote the Bochner-Martinelli-Koppelman kernel. If G
1
(, .) is another Leray
map, we set
(11.2.1)
1
= (G
1
) =
_
1
2i
_
n
< G
1
, d
< G
1
, .

_
<

,z
G
1
, d
< G
1
, .
_
n1
270 Integral Representations for and
b
and
(11.2.2)

01
= (G
0
, G
1
) =
_
1
2i
_
n
<

., d
[ .[
2

< G
1
, d
< G
1
, .

k1+k2=n2
_
< d

d ., d
[ .[
2
_
k1

_
<

,z
G
1
, d
< G
1
, .
_
k2
.
Notice that
1
and
01
are well dened for /1 and . 1. Also we use the
notation
1
q
,
01
q
to denote the summand of forms with degree (0, ) in . in
1
,
01
respectively.
Theorem 11.2.2 (Leray-Koppelman). Let 1 be a bounded domain in C
n
with
C
1
boundary. Let G
0
= (

.) and G
1
be another Leray map for 1. For )
C
1
(0,q)
(1), 0 n, we have
(11.2.3) )(.) =
_
bD

1
q
) +

z
T
q
) +T
q+1

), . 1,
where
T
q
)(.) =
_
D

0
q1
(, .) )()
_
bD

01
q1
(, .) )().

0
,
1
and
01
are dened in (11.1.8), (11.2.1) and (11.2.2) respectively.
Proof. From (11.1.4-ii), we have

,z

01
=
0

1
on the set where /1 and . 1. Thus, for . 1,
_
bD

0
) =
_
bD

,z

01
) +
_
bD

1
).
Applying Stokes theorem, we have
_
bD

01
) =
_
bD
d

(
01
))
_
bD

01

) =
_
bD

01

).
Since
01
) is of degree (n, n 1) in , it follows that
_
bD

01
) =

z
_
bD

01
).
Substituting the above formulas into (11.1.10), we have for . 1,
)(.) =
_
bD

0
) +
_
D

) +

z
_
D

0
)
=
_
bD

1
) +

z
__
D

0
)
_
bD

01
)
_
+
__
D

)
_
bD

01

)
_
.
(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

1
0
(, .) )(), . 1,
where
1
is dened in (11.2.1) and
1
0
is the piece in
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 Bochner-Martinelli-Koppelman 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 denition:
Denition 11.2.4. Let 1 1
N
be a domain with C
2
boundary and is any C
2
dening function. 1 is a convex (or strictly convex) domain with C
2
boundary if
N

i,j=1

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

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 denition of convexity or strict convexity is indepen-
dent of the choice of the dening function . In fact, for a strictly convex domain
1, we can choose a special dening function such that its real Hessian is positive
denite 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
dening function such that
(11.2.4)
N

i,j=1

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
0
be any C
2
dening function for 1. We set = c
A0
1 where
is a positive constant. Then is another C
2
dening function. Arguments similar
to those in the proof of Theorem 3.4.4 show that is strictly convex and satises
(11.2.4) if we choose suciently large.
A dening function satisfying (11.2.4) is called a strictly convex dening func-
tion for 1. By continuity, satises (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 be a C
2
dening function for 1. Then the map
(11.2.5) G
1
(, .) =
_

_
=
_

1
, ,

n
_
is a Leray map.
272 Integral Representations for and
b
Proof. Using convexity, we have for any . 1, /1,
(11.2.6) Re
n

i=1

i
(
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 be a C
2
dening
function for 1. Suppose that G
0
= (

.) and G
1
is dened 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

1
0
) +T
1

), . 1, if = 0,
where
(11.2.9) T
q
)(.) =
_
D

0
q1
(, .) )()
_
bD

01
q1
(, .) )().

0
,
1
and
01
are dened 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
1
has no
d .s. Thus for any 1 n,
1
q
= 0 and
_
bD

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 dened 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 Holder 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 be a strictly convex dening function for 1. There exists a constant
C 0 such that for any /1, . 1,
(11.2.10) Re < G
1
, . C(() (.) +[ .[
2
),
where G
1
is dened by (11.2.5).
Proof. Since is a strictly convex dening function satisfying (11.2.4), we apply
Taylors expansion to (.) at the point /1, then
(.) = () 2Re
n

i=1

i
(
i
.
i
) +
n

i,j=1

j
(
i
.
i
)(

j
.
j
)
+ Re
n

i,j=1

j
(
i
.
i
)(
j
.
j
) +o([ .[
2
).
Thus, for [ .[ c, where c 0 is suciently small,
Re
n

i=1

i
(
i
.
i
)
1
2
()
1
2
(.) +
c
2
[ .[
2
,
where c 0 is the positive constant in (11.2.4). To show that (11.2.10) holds for
[ .[ c, we set
. =
_
1
c
[ .[
_
+
c
[ .[
..
Then [ .[ = c and . 1 since 1 is strictly convex. It follows that
Re < G
1
, . = Re
[ .[
c
< G
1
, .

[ .[
2c
(() ( .) +
c
2
[ .[
2
)

cc
2
4
C(() (.) +[ .[
2
),
since ((.) +[ .[
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 be a strictly convex dening function for 1. The kernel
01
q1
(, .),
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
[
01
q1
(, .)[ < C,
where C is independent of ..
Proof. Let
_
(, .) =< G
1
, . ,

0
(, .) = [ .[
2
.
274 Integral Representations for and
b
Using (11.2.10), the kernel
01
(, .) has singularities only at = . on /1.
We choose a special coordinate system in a neighborhood of a xed . near /1.
From the denition of , we have d

[
=z
= and d

(Im )[
=z
=
1
2i
( ).
Thus d and d

(Im ) are linearly independent at = .. On a small neighborhood


l

= [ [ .[ < c of a xed . 1, Let (t, j) = (t


1
, , t
2n1
, j) where
t = (t
t
, t
2n1
) = (t
1
, , t
2n1
) are tangential coordinates for l

/1, t
i
(.) = 0
and
(11.2.12)
_
j = (),
t
2n1
= Im(, .).
From (11.2.10) it follows that there exists a positive constant
0
such that
(11.2.13)
_
[(, .)[
0
([(.)[ +[t
t
[
2
+[t
2n1
[),
[ .[
0
([(.)[ +[t[).
Using (11.2.13), we have for some 0,
(11.2.14)
_
bDU
[
01
q1
(, .)[ C
_
n1

k=1
_
bDU
[ .[
[[
nk
[
0
[
k
do
_
C
_
bDU
1
[[[ .[
2n3
do
C
_
]t]A
dt
1
dt
2
dt
2n1
([t
2n1
[ +[t
t
[
2
)[t[
2n3
C
_
]t

]A
[ log [t
t
[[dt
1
dt
2n2
[t
t
[
2n3
C
_
A
0
:
2n3
[ log :[d:
:
2n3
C,
where do is the surface element on /1. This proves the lemma.
Thus the kernel
01
is integrable uniformly on /1. We have the following Holder
regularity result for

:
Theorem 11.2.11 (
1
2
-Holder 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,q1)
(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 rst 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

0
q1
(, .) )()
and
n
1
=
_
bD

01
q1
(, .) )().
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

| ) |

[. .
t
[

for any < 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,q1)
(1). Since n
1
C

(0,q1)
(1), using the Hardy-Littlewood lemma (see
Theorem C.1 in the Appendix), to prove that n
1
C
1/2
(0,q1)
(1), it suces to show
that there exists a C such that
(11.2.16) [n
1
(.)[ C[(.)[

1
2
, . 1.
Using the same notation as in Lemma 11.2.10, we have for 1 n 1,

z
_
bDU

01
q1
(, .) )()

C | ) |

n1

k=1
_
_
_
bDU
[ .[
[[
nk+1
[
0
[
k
do +
_
bDU
1
[[
nk
[ .[
2k
do
_
_
.
To prove (11.2.16), using the change of coordinates (11.2.12) and estimates (11.2.13),
it suces to show that for some 0, there exists a C 0 such that for 0,
1 n 1,
(11.2.17)
_
]t]A
dt
1
dt
2
dt
2n1
( +[t
2n1
[ +[t
t
[
2
)
nq+1
[t[
2q1
< C

1
2
,
(11.2.18)
_
]t]A
dt
1
dt
2
dt
2n1
( +[t
2n1
[ +[t
t
[
2
)
nq
([t[ +)
2q
< C

1+
,
where 0 < < 1 and C, C

are independent of . To prove (11.2.17), integrating


with respect to t
2n1
and then using polar coordinates [t
t
[ = :, we have
_
]t]A
dt
1
dt
2
dt
2n1
( +[t
2n1
[ +[t
t
[
2
)
nq+1
[t[
2q1
C
_
]t

]A
dt
1
dt
2
dt
2n2
( +[t
t
[
2
)
nq
[t
t
[
2q1
C
_
A
0
:
2n3
d:
( +:
2
)
nq
:
2q1
C
_
A
0
d:
+:
2
C

1
2
.
276 Integral Representations for and
b
(11.2.18) is proved similarly. Thus n
1
C
1/2
(0,q1)
(1) and (11.2.15) is proved for
) C
1
(0,q)
(1).
When ) C
(0,q)
(1), we can nd a sequence )

(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,q1)
(1) and

n = )
in the distribution sense.
Remark. In Chapter 5, we have proved that for any

-closed (0, )-form with
\
s
(1) coecients 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 Holder 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 Holder 1,2-estimates 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 be a C
2
dening
function for 1, normalized such that [d[ = 1 on /1. ) is a (0, )-form on /1 with
continuous coecients, denoted by ) C
(0,q)
(/1), if and only if
(11.3.1) ) = p,
where p is a continuous (0, )-form in C
n
and is the projection operator from
(0, )-forms in C
n
onto (0, )-forms on /1 ( i.e., (0, )-forms which are pointwise
orthogonal to

). (11.3.1) is also equivalent to the following condition: for any
continuous (n, n 1)-form dened in a neighborhood of /1, we have
(11.3.2)
_
bD
) =
_
bD
p .
To see that (11.3.1) and (11.3.2) are equivalent, we note that for any (0, 1)-form
/ with continuous coecients in C
n
,
_
bD

/ =
_
bD
(d ) / = 0.
The space of (0, )-forms with Holder or 1
p
coecients are denoted by C

(0,q)
(/1)
or 1
p
(0,q)
(/1), where 0 < < 1 and 1 j . If n 1
p
(0,q1)
(/1), u satises
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
C

(n,n1q)
(C
n
),
(11.3.3)
_
bD
n

= (1)
q
_
bD
) .
Let 1

= 1 and 1
+
= C
n
1. We dene the Bochner-Martinelli-Koppelman
transform for any ) C
(0,q)
(/1) as follows:
(11.3.4)
_
bD
1
q
(, .) )() =
_
1

(.), if . 1

,
1
+
(.), if . 1
+
.
It is easy to see that 1

(0,q)
(1

) and 1
+
C

(0,q)
(1
+
). In fact, 1

and
1
+
are continuous up to the boundary if ) is Holder continuous and we have the
following jump formula:
Theorem 11.3.1. (Bochner-Martinelli-Koppelman 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

(0,q)
(1

), 1
+
C

(0,q)
(1
+
),
for every with 0 < < 1 and
(11.3.6) ) = (1

1
+
), . /1.
Proof. We rst assume that the boundary /1 is at with /1 = (.
1
, , .
n
) C
n
[
Im.
n
= 0 and ) has compact support in /1. The coecients of 1(, .) are of the
form

j
.
j
[ .[
2n
, , = 1, , n.
We rename the real coordinates .
j
= r
j
+ij
j
by setting r
j+n
= j
j
, , = 1, , n1,
and j = j
n
. Similarly we set
j+n
=
j
, , = 1, , n 1, where
j
=
j
+ i
j
. Set
r = (r
1
, , r
2n1
), = (
1
, ,
2n1
) and . = (r, j). We dene
_
j
y
() =
j
([[
2
+j
2
)
n
, j 0,

j
y
() =

j
([[
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(, .) ) is a constant multiple of
the following form:
1)
I
(.) =
_
R
2n1
j
([ r[
2
+j
2
)
n
)
I
() d
1
d
2n1
= j
y
)
I
,
278 Integral Representations for and
b
or
Q
j
)
I
(.) =
_
R
2n1

j
r
j
([ r[
2
+j
2
)
n
)
I
() d
1
d
2n1
=
j
y
)
I
,
where , = 1, , 2n 1. The Poisson integral 1 is bounded from C
0
(1
2n1
) to
C(1
2n
+
). Since it ts a convolution operator, it is bounded from C
1
0
(1
2n1
) 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
2n1
) to C

(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 at. For the general case, we note that the Bochner-Martinelli-
Koppelman kernel is obtained by dierentiation 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

(0,q)
(1

) and 1
+
C

(0,q)
(1
+
).
To prove (11.3.6), we rst 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
(, .)

) +

z
_
D
1
q1
(, .) )
=
_
)(.), . 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(, .) is an integrable kernel
in C
n
, the term
_
D
1(, .)

) is continuous up to the boundary /1. We denote


by
z
the outward unit normal to /1 at .. Then for . /1,
lim
0
+
__
D
1
q
(, . c
z
)

)
_
D
1
q
(, . +c
z
)

)
_
= 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 C

(n,nq1)
(C
n
),
we have
(11.3.8)
lim
0
+
_
bD

z
__
D
1
q1
(, . c
z
) )()
_
(.)
= (1)
q
lim
0
+
_
bD
__
D
1
q1
(, . c
z
) )()
_

z
(.)
= (1)
q
_
bD
__
D
1
q1
(, .) )()
_

z
(.).
Similarly, we obtain
(11.3.9)
lim
0
+
_
bD

z
__
D
1
q1
(, . +c
z
) )()
_
(.)
= (1)
q
_
bD
__
D
1
q1
(, .) )()
_

z
(.).
11.3 Homotopy Formulas for

b
on Strictly Convex Boundaries 279
Thus from (11.3.7)-(11.3.9), we get for any C

(n,nq1)
(C
n
),
_
bD
)(.) (.) = lim
0
+
_
bD
__
bD
[1
q
(, . c
z
) 1
q
(, . +c
z
)] )()
_
(.)
=
_
bD
[1

(.) 1
+
(.)] (.).
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
) = (1

1
+
), . /1,
and

1

= 0 in 1

,

1
+
= 0 in 1
+
. Furthermore, we have 1

(0,q)
(1

),
1
+
C

(0,q)
(1
+
) for any 0 < < 1.
Proof. Since 1

(0,q)
(1), dierentiation under the integral sign and Stokes
theorem imply that for . 1,

z
1

(.) =
_
bD

z
1
q
(, .) )()
=
_
bD

1
q+1
(, .) )()
=
_
bD
d

(1(, .) )()) +
_
bD
1
q+1
(, .)

)()
= 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 dierence 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 be a strictly convex dening function for 1. Dene C
1
functions G

and
G
+
in C
n
C
n
by
G

(, .) =
_

1
, ,

n
_
, (11.3.10)
G
+
(, .) =
_


.
1
, ,

.
n
_
. (11.3.11)
Using Lemma 11.2.6, G

(, .) is a Leray map for 1. Let


G
0
(, .) = (

1
.
1
, ,

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
be a C
2
strictly convex dening function for 1. For , . /1, the kernels
(11.3.12)

= (G

),
+
= (G
+
),
(11.3.13)
0
= (G

, G
0
),
+0
= (G
+
, G
0
)
have singularities only when = .. Furthermore, there exists a constant C 0
independent of . such that
(11.3.14)
_
bD
([
0
(, .)[ +[
+0
(, .)[) < C, . /1.
Proof. Set
_
(, .) = < G

(, .), . ,
(, .) = < G
+
(, .), . .
Note that (, .) = (., ). Using Lemma 11.2.9, there exists a constant C 0
such that for any /1, . 1

,
(11.3.15) Re(, .) C(() (.) +[ .[
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 1
+
l if l is suciently small. Reversing the
role of and ., we have for any . 1
+
l and 1,
(11.3.16)
Re(, .) = Re(., ) = Re
n

i=1

.
i
(
i
.
i
)
C((.) () +[ .[
2
).
Inequality (11.3.16) holds for . 1
+
l since is strictly convex in a neigh-
borhood of /1. Thus
+
and
0+
have singularities only at = . /1. Using
estimate (11.3.15), we have already proved that
0
is absolutely integrable in
Lemma 11.2.10. Since satises a similar estimate (11.3.16), the proof for
+0
follows from the arguments of Lemma 11.2.10. This proves (11.3.14) and the lemma.
For ,= ., we set
=
0

+0
=
n2

q=0

q
(, .),
where
q
=
0
q

+0
q
is the summand which is of degree (0, ) in .. Using Lemma
11.3.3, is an integrable kernel on /1. If ) C
(0,q)
(/1), the form
(11.3.17) H
q
) =
_
bD

q1
(, .) ) =
_
bD
(
0
q1

+0
q1
) )
is a well dened (0, 1)-form on /1 with continuous coecients. The next theorem
shows that (, .) 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 be a
C
2
dening function for 1. Then for any ) C
1
(0,q)
(/1), 0 < < n 1, we have
(11.3.18) )(.) =

b
H
q
) +H
q+1

b
), . /1,
where H
q
is dened in (11.3.17).
Proof. For any ) C
1
(0,q)
(/1), using Theorem 11.3.1, the Bochner-Martinelli-
Koppelman transform 1

and 1
+
dened by (11.3.4) are continuous up to the
boundary. We denote the boundary value of 1

and 1
+
by (
_
bD
1(, .) )())

and (
_
bD
1(, .) )())
+
respectively. From (11.3.6), we have for any . /1,
(11.3.19)
)(.) = (1

(.) 1
+
(.))
=
__
bD
1(, .) )()
_

__
bD
1(, .) )()
_
+
.
Applying (11.1.4-ii), we have for any /1,

,z

0
=
0
+

= 1(, .) +

, . 1

,z

+0
=
0
+
+
= 1(, .) +
+
, . 1
+
.
Thus, for . 1

,
(11.3.20)
_
bD
1(, .) )
=
_
bD

,z

0
(, .) ) +
_
bD

(, .) )
=

z
_
bD

0
(, .) ) +
_
bD

0
(, .)

b
) +
_
bD

(, .) ).
Similarly, for . 1
+
,
(11.3.21)
_
bD
1(, .) )
=
_
bD

,z

+0
(, .) ) +
_
bD

+
(, .) )
=

z
_
bD

+0
(, .) ) +
_
bD

+0
(, .)

b
) +
_
bD

+
(, .) ).
Since G

is independent of .,

(, .) =

0
(, .). It follows that
(11.3.22)
_
bD

(, .) ) = 0, when ,= 0.
Also since G
+
is independent of , we have
+
(, .) =
+
n1
(, .) and
(11.3.23)
_
bD

+
(, .) ) = 0, when 0 < n 1.
From Lemma 11.3.3,
0
and
+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, dene
(11.3.24) n(.) = H
q
) =
_
bD
(
0
q1

+0
q1
) ), . /1.
Then n C
(0,q1)
(/1) and n satises

b
n = ).
Proof. Using Lemma 11.3.3, we have
| n |
L

(bD)
C | ) |
L

(bD)
.
Thus n C
(0,q1)
(/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,n1)
(/1),
)(.) =
_
bD

+
n1
(, .) ) +

b
_
bD

n2
(, .) ).
The kernel
+
=
+
n1
is a holomorphic function in . If ) is a (0, n 1)-form
satisfying the compatibility condition (9.2.12 a), then
_
bD

+
n1
(, .) ) = 0, . 1 and . /1.
Thus, we have
)(.) =

b
_
bD

n2
(, .) ), . /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

0
(, .) ) +
_
bD

0
(, .)

b
).
If ) is a C1 function, we have
)(.) =
_
bD

0
(, .) ), . /1.
Thus

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 Bochner-Martinelli kernel is
a reproducing kernel. However,

can be viewed as a true generalization of the


Cauchy kernel to C
n
since

is holomorphic in ..
We shall derive another homotopy formula for

b
on strictly convex boundaries.
Let
+
be dened by
(11.3.25)
+
= (G

, G
+
),
where G

and G
+
are dened by (11.3.10) and (11.3.11). We rst show that
+
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 be a strictly convex dening function for 1. The kernel
+
q1
(, .),
1 n 1 has singularities only at = . for , . /1. Furthermore, there
exists a constant C such that for any . /1,
(11.3.26)
_
bD
[
+
q1
(, .)[ < C,
where C is independent of ..
Proof. Since

+
q1
=
_
1
2i
_
n
< G

, d
< G

, .

< G
+
, d
< G
+
, .

_
<

,z
G

, d
< G

, .
_
nq1

_
<

,z
G
+
, d
< G
+
, .
_
q1
,
it follows from (11.3.15) and (11.3.16) that
+
q1
has singularities only at = ..
Thus we only need to estimate the kernel when is close to .. For a xed ., let l

=
[ [ .[ < c be a suciently small neighborhood of ., (, .) = < G

, .
and (, .) = < G
+
, . the same as before. Using the same change of variables
t = (t
1
, , t
2n1
) = (t
t
, t
2n1
) as in Lemma 11.2.10 with t
2n1
= Im(, .) and
t
i
(.) = 0 for i = 1, , 2n1, there exists a constant
0
0 such that for , . /1,
(11.3.27)
_[(, .)[
0
([t
t
[
2
+[t
2n1
[),
[(, .)[
0
([t
t
[
2
+[t
2n1
[
2
),

0
[t[ [ .[ (1,
0
)[t[.
We note that
(11.3.28)
[ < G

, d < G
+
, d [ = [ < G

, d < G
+
G

, d [
= O([ .[).
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 0 such that
_
bDU
[
+
q1
(, .)[ C
__
bDU
[ .[
[[
nq
[[
q
do
_
C
_
bD]z]<]
1
[[
nq
[ .[
2q1
do
C
_
]t]A
dt
1
dt
2
dt
2n1
([t
2n1
[ +[t
t
[
2
)
nq
[t[
2q1
C.
Thus, the kernel
+
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
be a C
2
dening function for 1. Then for any ) C
1
(0,q)
(/1), where 0 < < n1,
we have
(11.3.29) )(.) =

b
1
q
) +1
q+1

b
), . /1,
where
1
q
) =
_
bD

+
q1
(, .) )
and
+
is dened by (11.3.25).
Proof. Using Lemma 11.3.6, the kernel
+
is absolutely integrable. From Theorem
11.3.4, we have when 0 < < n 1,
)(.) =

b
_
bD
(, .) ) +
_
bD
(, .)

b
).
Using (11.1.4-iii), we have for , . /1 and ,= .,
(11.3.30)

,z

+0
=

,z
(G

, G
+
, G
0
)
=
0

+0

+
=
+
.
For each xed . /1, we claim that
+0
and

,z

+0
are absolutely integrable
kernels and
(11.3.31)

_
bD
(G

, G
+
, G
0
)

< C,
(11.3.32)

_
bD

,z
(G

, G
+
, G
0
)

< C,
where C is independent of .. Let
0
= [ .[
2
as before. Using (11.3.27) and
(11.3.28), (11.3.31) can be estimated by

_
bDU
(G

, G
+
, G
0
)

k1+k2+k3=n3
_
bDU
[ .[
2
[
k1+1

k2+1

k3+1
0
[
do
C
n2

k=2
_
]t]A
[t[
2
dt
1
dt
2
dt
2n1
([t
2n1
[ +[t
t
[
2
)
nk
[t[
2k
< .
Since

,z
=

= O([ .[) and


,z
=

z
= O([ .[), we can use (11.3.27),
11.3 Homotopy Formulas for

b
on Strictly Convex Boundaries 285
(11.3.28) and dierentiation term by term to get

_
bDU

,z
(G

, G
+
, G
0
)

k1+k2+k3=n3
_
bDU
[ .[
[
k1+1

k2+1

k3+1
0
[
do
+C

k1+k2+k3=n2
_
bDU
[ .[
3
[
k1+1

k2+1

k3+1
0
[
do
C
n2

k=1
_
]t]A
[t[dt
1
dt
2
dt
2n1
([t
2n1
[ +[t
t
[
2
)
nk
[t[
2k
+C
n1

k=1
_
]t]A
[t[
3
dt
1
dt
2
dt
2n1
([t
2n1
[ +[t
t
[
2
)
nk
[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 dierentiation to obtain
(11.3.33)
_
bD

,z

+0
) =
_
bD

+0
)

z
_
bD

+0
)
=
_
bD

+0

b
)

z
_
bD

+0
),
where the last equality follows from Stokes theorem. The Stokes theorem can be
used here by rst substituting

= + c,

= + c and

0
=
0
+ c for ,
and
0
respectively in the kernel
+0
and then letting c 0. Similarly,
(11.3.34)
_
bD

,z

+0

b
) =
_
bD

+0

b
)
=

z
_
bD

+0

b
).
From (11.3.33) and (11.3.34), we have

_
bD

,z

+0
) +
_
bD

,z

+0

b
) = 0, . /1.
Thus using (11.3.30), we obtain
)(.) =

b
_
bD
(, .) ) +
_
bD
(, .)

b
)
=
_

b
_
bD

,z

+0
) +
_
bD

,z

+0

b
)
_
+
_

b
_
bD

+
) +
_
bD

b
)
_
=

b
_
bD

+
) +
_
bD

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 be
a C
2
dening 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,q1)
(/1) and

b
n = ) on /1.
Proof. That n is in C
(0,q1)
(/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 Holder and 1
p
spaces. We use | |
L
p to denote
the 1
p
(0,q)
(/1) norms for (0, )-forms.
Theorem 11.3.9 (Holder 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 be a C
3
dening function for 1. For any ) 1
p
(0,q)
(/1), 1 j and 1 < n 1,
1
q
) satises the following estimates:
(1) |1
q
)|
L
2n
2n1

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
C|)|
L
p, where 2n < j < and =
1
2

n
p
, C

is the Holder
space of exponent on /1.
(5) |1
q
)|
C
1
2
C|)|
L
.
Proof. We shall prove that the kernel
+
(, .) is of weak type
2n
2n1
on /1 uni-
formly in and in .. (For denition of weak type, see Denition B.5 in the Ap-
pendix). Since
+
only has singularities when = ., following the change of
coordinates t and (11.3.27), it suces to show that the function
(11.3.35) 1(t) =
1
([t
2n1
[ +[t
t
[
2
)
nq
[t
t
[
2q1
is of weak type
2n
2n1
, where t = (t
1
, , t
2n2
, t
2n1
) = (t
t
, t
2n1
) and [t[ < 1. Let

be the subset

= t 1
2n1
, [t[ < 1 [ 1(t) , 0,
and let : be the Lebesgue measure in 1
2n1
. We shall show that there exists a
constant c 0 such that
(11.3.36) :(

)
_
c

_ 2n
2n1
, for all 0.
By a change of variables t

t such that t
i
=

1
2n1
t
i
, i = 1, , 2n 2 and
t
2n1
=

2
2n1
t
2n1
, we have for some c 0,
:(

) = c

2n
2n1
:(
1
)
_
c

_ 2n
2n1
,
since :(
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 dene

+
by

+
=
_
1
2i
_
n
< G

, d
(, .)

< G
+
, d
(, .) j(.)

k1+k2=n2
_
<

,z
G

, d
(, .)
_
k1

_
<

,z
G
+
, d
(, .) j(.)
_
k2
,
where j 0 is suciently large. We rst note that the kernel
+
is the same as

+
when , . /1. It follows from (11.3.15) and (11.3.16) that there exists a
C 0 such that for any /1 and . 1,
(11.3.37) Re(, .) C([(.)[ +[ .[
2
),
and
(11.3.38) Re

(, .) Re(, .) j(.) C([(.)[ +[ .[


2
),
if j is chosen suciently large. Let 1

= . 1 [ (.) < for some 0.


From the Hardy-Littlewood lemma (see Theorem C.1 in the Appendix), to prove
(4) and (5), it suces to show that for some small
0
0 and all 0 < <
0
,
(11.3.39) sup
zbD

grad
z
_
bD

+
(, .) )

1
2

n
p
|)|
L
p,
where 2n j . After the same change of variables in a small neighborhood
[ .[ < 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
2n1
( +[t
2n1
[ +[t
t
[
2
)
nq+1
[t
t
[
2q1
< C

1
2
,
(11.3.41)
_
]t]A
dt
1
dt
2
dt
2n1
( +[t
2n1
[ +[t
t
[
2
)
nq
[t
t
[
2q+1
< C

1
2
,
where C is independent of . 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 suces to show that
(11.3.42)
_
]t]A
[)[dt
1
dt
2
dt
2n1
( +[t
2n1
[ +[t
t
[
2
)
nq+1
[t
t
[
2q1
< C
1
|)|
L
2n,
(11.3.43)
_
]t]A
[)[dt
1
dt
2
dt
2n1
( +[t
2n1
[ +[t
t
[
2
)
nq
[t
t
[
2q+1
< C
1
|)|
L
2n,
288 Integral Representations for and
b
where C is independent of . To prove (11.3.42), we set n

=
2n
2n1
and use Holders
inequality to obtain
_
]t]A
[)[dt
1
dt
2
dt
2n1
( +[t
2n1
[ +[t
t
[
2
)
nq+1
[t
t
[
2q1
|)|
L
2n
_
_
]t]A
dt
1
dt
2
dt
2n1
( +[t
2n1
[ +[t
t
[
2
)
n

(nq+1)
[t
t
[
n

(2q1)
_ 1
n

|)|
L
2n
_
_
]t

]A
dt
1
dt
2
dt
2n2
( +[t
t
[
2
)
n

(nq+1)1
[t
t
[
n

(2q1)
_ 1
n

C|)|
L
2n
_
_
A
0
:
2n3
d:
( +:
2
)
n

(nq+1)1
:
n

(2q1)
_ 1
n

C|)|
L
2n
_
_
A
0
d:
( +:
2
)
2n+1
2n1
:
2n3
2n1
_ 1
n

, ( = :,

)
C
1
|)|
L
2n
_
_

0
d
(1 +
2
)
2n+1
2n1

2n3
2n1
_ 1
n

C
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 satises 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 1,2-derivatives in the Holder 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 convexied locally. We
note that the solution for

b
dened 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
dierent 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 /1
be a connected open C1 manifold with smooth boundary /. We consider the

b
equation
(11.4.1)
b
n = on ,
11.4 Solvability for

b
on CR Manifolds with Boundaries 289
where is a (0, )-form on , 1 n 2. In order for (11.4.1) to be solvable, it
is necessary that satises
(11.4.2)
b
= 0 on .
Note that when = n 1, (11.4.2) is void and (11.4.1) is related to the local
nonsolvable phenomenon of Lewys equation. Due to the fact that the compatibility
condition (11.4.2) is satised only on 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 /.
In Chapter 9, we have seen that when = n1 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 for any satisfying
(11.4.2) on . When = n2, 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 dened
and holomorphic in some open neighborhood of 1. Let be a form in C
(0,n2)
()
such that there exists n C
(0,n3)
() with

b
n = on . Then for any p O(/),
we have
_
b
p d.
1
d.
n
=
_
b

b
n p d.
1
d.
n
=
_
b

(n p d.
1
d.
n
)
=
_
b
d(n p d.
1
d.
n
) = 0.
Thus, another necessary condition for (11.4.1) to be solvable for some n C
(0,n3)
() is that
(11.4.2 a)
_
b
p d.
1
d.
n
= 0 for all p O(/).
It is easy to see that (11.4.2 a) is also necessary for the existence of a solution
n C
(0,n3)
() by approximation. This additional condition makes it necessary to
dierentiate 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 rst describe the geometry of the boundary / of on which one can con-
struct a solution kernel for

b
on . Let be a strictly convex dening function
for 1. The set is a domain in /1 dened by some C
2
function : dened in a
neighborhood of /1 such that
(11.4.3) = . C
n
[ (.) = 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 Levi-at hypersurface. The
boundary / is dened by
/ = /1 `
0
= . C
n
[ (.) = 0, :(.
n
) = 0.
On /, we assume
(11.4.4) d d: ,= 0 on /.
Thus, the hypersurfaces /1 and `
0
intersect transversally over 1. If
(11.4.5) : ,= 0,
we say that /1 and `
0
intersect transversally over C. The points in
= . / [ : = 0
are called characteristic points. Any point in / is called a noncharacteristic
point or a generic point.
If j is a characteristic point on /, the space T
1,0
p
(/1) T
1,0
p
(`
0
) has complex
dimension n 1. Near the noncharacteristic point j /, the set T
1,0
p
(/1)
T
1,0
p
(`
0
) has complex dimension n2. This jump in the dimension of the tangential
(1,0) vector elds at the characteristic points makes it dicult 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 / of the set
= . C
n
[ [.[ = 1, Im.
n
< 0
has two characteristic points at = (0, , +1), (0, , 1), since : = 0
if and only if .
1
= = .
n1
= 0.
If :(.
n
) = [.
n
[
2
and / is the boundary of

1
= . C
n
[ [.[ = 1, [.
n
[
2
<
1
2
,
then /
1
has no characteristic points.
Notice that is simply connected but
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
+
is a funda-
mental solution for

b
on the compact hypersurface /1. Thus, it gives a solution
kernel for (11.4.1) if has compact support in . To solve

b
for forms which do
11.4 Solvability for

b
on CR Manifolds with Boundaries 291
not vanish on /, we introduce new kernels constructed from the special dening
function : for . Set
(11.4.6) G

(, .) = G

() =
_
0, , 0,
:(
n
)

n
_
and
(11.4.7) (, .) =
:(
n
)

n
(
n
.
n
) = < G

, . .
Let
(11.4.8)

=
1
2i
< G

, d
< G

, .
=
1
2i
d
n

n
.
n
,
where the notation . is used to indicate that the hypersurface `
0
dened by : is
Levi at. Note that

is independent of : and

,z

= 0,
n
,= .
n
.
In other words, it is holomorphic both in the and . variables away from singular-
ities. Setting
(11.4.9)
+
= (G

, G

, G
+
),
we see that
+
is an (n, n 3)-form. We write
(11.4.10)
+
(, .) =
n3

q=0

+
q
(, .),
where

+
q
=
1
(2i)
n
d
n

n
.
n

< G

(), d
(, .)

< G
+
(.), d
(, .)

_
<

(), d
(, .)
_n3q

_
<
z
G
+
(.), d
(, .)
_q
away from the singularities. Thus
+
q
has exactly d .s.
If / and . , we have
:(
n
) :(.
n
) 0.
It follows that
n
,= .
n
when / and . . The kernel
+
is well dened
and smooth when / and . .
For C
(0,q)
(), 1 n 2, we dene
(11.4.11) o
q
=
_

+
q1
(, .) () +
_
b

+
q1
(, .) (),
where
+
(, .) and
+
(, .) are kernels dened 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 be a
C
2
strictly convex dening function for 1. Let /1 be an open connected
C1 manifold with smooth boundary dened by (11.4.3) where :(.) = :(.
n
) is a C
2
function. We assume that dd: ,= 0 on /. For any C
1
(0,q)
(), 1 < n2,
(11.4.12) =

b
o
q
+o
q+1

b
, . ,
where o
q
is the integral operator dened by (11.4.11).
Theorem 11.4.2 (A solution operator for

b
on C1 manifolds with bound-
aries). Let be as in Theorem 11.4.1. For any C
(0,q)
(), 1 < n 2, with

b
= 0 on , the form n = o
q
is in C
(0,q1)
() and

b
n = on , where o
q
is
the integral operator dened by (11.4.11).
When = n 2, we assume furthermore that satises the additional compati-
bility condition
_
b
p d.
1
d.
n
= 0 for all p O(/),
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 be as in Theorem 11.4.1. For every )
C
1
(0,q)
(), 1 < n 1,
)(.) =

b
_

+
q1
)() +
_

+
q

b
)()
_
b

+
q
)()
for every . , where
+
is dened by (11.3.25).
Proof. We rst extend ) to

) on an open set such that

) C
1
(0,q)
( ). Let

0
( ) be cut-o functions such that

1 on for every c and

converges
to the characteristic function of as c 0. Applying the homotopy formula proved
in Theorem 11.3.7 to


), we have for . ,
)(.) =


) =

b
_
bD

+
q1


) +
_
bD

+
q

b
(


))
=

b
_
bD

+
q1


) +
_
bD

+
q

b

)
+
_
\

+
q
(

)

).
For . , we note that
+
is smooth for ( ) and we can apply Stokes
theorem to the third term on the right-hand side to obtain
lim
0
_
\

+
q


) = lim
0
_
\

+
q
d


)
= lim
0
_
\
d(

+
q


)) lim
0
_
\

d(
+
q


))
=
_
b

+
q
).
11.4 Solvability for

b
on CR Manifolds with Boundaries 293
Thus for any . , letting c 0, since
+
is an absolutely integrable kernel by
Lemma 11.3.6 , we have
)(.) = lim
0
_

b
_
bD

+
q1


) +
_
bD

+
q

b

)
_
+ lim
0

_
\

+
q
(

)

)
=

b
_

+
q1
) +
_

+
q

b
)
_
b

+
q
).
This proves the proposition.
Proof of Theorem 11.4.1. We dene

= (G

, G

) and
+
= (G

, G
+
),
where G

and G
+
are dened by (11.3.10) and (11.3.11) respectively. The kernels

(, .) and
+
(, .) are smooth for (, .) / . Using (11.1.4-iii), we have
(11.4.13)
,z

+
=
+
+
+

for any / and . . Applying Proposition 11.4.3 and (11.4.13), we obtain for
. ,
(11.4.14)
(.) =
b
_

+
q1
(, .) () +
_

+
q
(, .)

b
()
+
_
b

+
q
(, .) () +
z
_
b

+
q1
(, .) ()

_
b

+
q
(, .) () +
_
b

q
(, .) ().
We claim that for any C
1
(0,q)
(), the following three equalities hold for . :
(i)
_
b

+
q
(, .) () =
_
b

+
q
(, .)

b
(),
(ii)
_
b

q
(, .) () = 0, 1 n 2,
(iii)
_
b

+
q
(, .) () = 0, 1 < n 2.
Since the kernel
+
(, .) has the factor d
1
d
n
, applying Stokes theorem,
we have
_
b

+
(, .) () =
_
b
d

(
+
(, .) ())
+
_
b

+
(, .)

b
(),
294 Integral Representations for and
b
which proves (i). Since n

is holomorphic in both the and . variables, for any


/ and . , we have
(11.4.15)

)
n2
,
(11.4.16)
+
=


+
(
z

+
)
n2
.
(ii) follows from (11.4.15) and the fact that integration of an (n, n 2 +)-form
on / 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 ds. Substituting (i),
(ii) and (iii) into (11.4.14), we have proved Theorem 11.4.1.
Proof of Theorem 11.4.2. When < n2, Theorem 11.4.1 implies Theorem 11.4.2
if C
1
(0,q)
(). If is only in C
(0,q)
(), we approximate by a sequence
n

C

(0,q)
() such that
n
and

n
0 uniformly on . This is possible from the
proof of Friedrichs lemma (see Appendix D). It is easy to see that o
q+1

n
0
in the distribution sense in and o
q

n
o
q
uniformly on compact subset of .
Thus n = o
q
is in C
(0,q1)
() and

b
n = in the distribution sense in .
To show that the theorem holds when = n 2, we use (11.4.14) to obtain
(11.4.17) =

b
o
n2

_
b

+
n2
(, .) (), . .
To show that the last integral in (11.4.17) vanishes when = n2, notice that the
kernel
+
(, .) is holomorphic in in a neighborhood of / for each xed . .
Thus from our assumption on , we have for . ,
_
b

+
n2
(, .) () = 0.
This proves the theorem.
In general, the additional assumption (11.4.2 a) on when = n2 is necessary.
The next proposition characterizes all domains 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( ) is dense in O(/) (in the C(/) norm).
Then any (0, n 2)-form C
(0,n2)
() satisfying condition (11.4.2) also satis-
es condition (11.4.2 a). In particular, if polynomials are dense in O(/), then
condition (11.4.2) implies condition (11.4.2 a).
Proof. From the assumption, for any p O(/), there exists a sequence of holo-
morphic functions p
n
O( ) such that p
n
converges to p in C(/). We have, for
any

b
-closed ,
_
b
p d.
1
d.
n
= lim
n
_
b
p
n
d.
1
d.
n
= lim
n
_

( p
n
d.
1
d.
n
)
= lim
n
_

b
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 be as in Theorem 11.4.1. We assume that the set C o,
where
o = .
n
C [ . = (.
1
, , .
n
) /,
is connected. For any C
(0,q)
(), 1 n 2, with
b
= 0 on , the form
n = o
q
is in C
(0,q1)
() and

b
n = on , where o
q
is the integral operator
dened 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 suces to show that
_
b

+
n2
(, .) () = 0, . .
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
/(
n
) =
1

n
.
n
can be approximated by polynomials 1

(
n
, .
n
) for each xed .
n
in the sup norm
on o. We approximate by

(, .) = (, .) +c for some c and let c 0


+
. Then

(, .) is smooth when . and . Also

is holomorphic in . Dene

=
1
2i
< G
+
(), d
< G
+
(), . +c
=
1
2i
< G
+
(), d
(, .) +c
.
We can apply Stokes theorem rst to the modied kernel with substituted by

, letting c 0, to obtain for . ,


_
b

+
n2
(, .) ()
= lim

lim
0
+
_
b
1

(
n
, .
n
)d
n

+

(
z

)
n2
()
= lim

lim
0
+
_

_
1

(
n
, .
n
)d
n

+

(
z

)
n2
()
_
= 0
since every term in the integrand is

-closed. This proves the corollary.


Example. We note that the additional assumption on or when = n 2
cannot be removed. Let /1 = . [ [.
1
[
2
+ [.
2
[
2
+ [.
3
[
2
= 1 be the unit sphere in
C
3
. Let
= /1 . C
3
[ [.
3
[
2
< 1,2.
296 Integral Representations for and
b
Then o = .
3
C [ [.
3
[
2
= 1,2 does not satises the hypothesis imposed on o in
Corollary 11.4.5 since C o is not connected. We shall show that equation

b
n =
is not solvable for = 1 in . Let
=
.
1
d .
2
.
2
d .
1
([.
1
[
2
+[.
2
[
2
)
2
.
Then is a constant multiple of the Bochner-Martinelli-Koppelman kernel in C
2
and

= 0 for [.
1
[
2
+[.
2
[
2
,= 0. Thus C

(0,1)
() and

b
= 0 on .
If =

b
n for some n C(), then must satisfy
_
b

1
i.
3
d.
1
d.
2
d.
3
=
_
b
d.
1
d.
2
d
3
= 0,
where d
3
= d.
3
,(i.
3
). On the other hand, we have that
_
b
d.
1
d.
2
d
3
= 8
_
]z1]
2
+]z2]
2
=
1
2
]
( .
1
d .
2
.
2
d .
1
) d.
1
d.
2
= 16
_
]z1]
2
+]z2]
2
<
1
2
]
d .
1
d .
2
d.
1
d.
2
,= 0.
Thus there does not exist any solution n C(). There does not exist any n C()
satisfying

b
n = either, by an approximation argument. Thus the assumption on
in Theorem 11.4.2 cannot be removed. We note that O() is not dense in O(/)
here.
On the other hand, if
= /1 . C
n
[ Im .
3
< 0,
then o = .
3
C [ 1 < Re.
3
< 1, Im .
3
= 0 and C o is connected. Thus it
satises the hypothesis imposed in Corollary 11.4.5 and we can solve (11.4.1) for all

b
-closed form C
(0,q)
() 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 Cauchy-Riemann 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 dening function (.) for `
which has the following form near the origin:
(11.5.1) (.) = Im .
n
+
n

j,k=1

jk
.
j
.
k
+O([.[
3
),
11.5 L
p
Estimates for Local Solutions of

b
297
where (
jk
) is a positive denite hermitian matrix (see the proof of Corollary 3.4.5).
The function is strictly convex near the origin. Let l be a small neighborhood of
0 and
0
0 be suciently small. We dene

by
(11.5.2)

= . ` l [ Im .
n
< , 0 < <
0
.
We can always choose l and
0
0 suciently small such that each

is an open
neighborhood in a connected strictly convex hypersurface whose boundary lies in a
at surface. It is easy to see that

= 0. Thus the

forms a neighborhood
base at 0.
Using Theorem 11.4.2 and Corollary 11.4.5, there is a solution operator o
q

satisfying

b
o
q
= on

for any

b
-closed C
(0,q)
(

). Our main goal is to


prove that there exists a solution operator satisfying 1
p
estimates on

.
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

of .
0
such that for any 1
p
(0,q)
(

), 1 n 2 and
1 < j < , satisfying
b
= 0, there exists n 1
p
(0,q1)
(

) satisfying
b
n = .
Furthermore, there exists a positive constant c such that the following estimate holds:
(11.5.3) |n|
L
p
(0,q1)
(

)
c||
L
p
(0,q)
(

)
,
where c depends on j,

but is independent of .
Corollary 11.5.2. Let ` and

be as in Theorem 11.5.1. The range of


b
is
closed in the 1
p
(0,q)
(

) space, where 1 < j < and 1 n 2.


Corollary 11.5.3. Let ` and

be as in Theorem 11.5.1. For each 1 < n2,


there exists a solution operator

o
q
given by integral kernels such that for any
b
-
closed 1
p
(0,q)
(

), 1 < j < , we have


b

o
q
= and
|

o
q
|
L
p
(0,q1)
(

)
c||
L
p
(0,q)
(

)
,
where c depends on j,

but is independent of .
The rest of this section is to prove Theorem 11.5.1. Let

be dened by (11.5.2).
To prove Theorem 11.5.1, we rst prove the 1
p
estimates for the solution constructed
for
b
-closed forms with C(

) coecients in Theorem 11.4.2.


Proposition 11.5.4. Let ` be a strongly pseudoconvex C1 manifold dened by
(11.5.1) and

be dened by (11.5.2). For any C


(0,q)
(

) such that
b
= 0,
1 n 2, there exists a solution n C
(0,q1)
(

) satisfying
b
n = on

.
Furthermore, for every 1 < j < , there exists a constant C
p
0 such that
(11.5.4) |n|
L
p
(0,q1)
(

)
C
p
||
L
p
(0,q)
(

)
,
where C
p
is independent of and small perturbation of .
Proof. Let
(11.5.5) n(.) o
q
(.) = 1
1
() +1
2
(),
298 Integral Representations for and
b
where
1
1
() =
_

+
q1
(, .) ()
and
1
2
() =
_
b

+
q1
(, .) ().
Since the set Co is connected. it follows from Theorem 11.4.2 and Corollary 11.4.5
that for every 1 n 2,

b
n = on

and n C
(0,q1)
(

).
To prove Proposition 11.5.4, we only need to prove that n satises (11.5.4). Using
Theorem 11.3.9, there exists a C 0 such that the integral 1
1
() satises
(11.5.6) | 1
1
() |
L
p
(

)
C | |
L
p
(

)
.
We only need to estimate 1
2
().
Since 1
2
() is an integral on /

, we rewrite 1
2
() to be an integral on

to
facilitate the 1
p
estimates. Since the kernel
+
(, .) has singularities at
n
= .
n
for any , .

, we shall modify the kernel rst so that Stokes theorem can be


applied.
Let :(.) = :(.
n
) = Im .
n
. Then for any , .

,
(11.5.7) :(.) :() 2Re
:()

n
(.
n

n
) = 0.
We set
(, .) =
:()

n
(
n
.
n
) (:() )
= (, .) (:() ).
It follows from (11.5.7) that
(11.5.8)
Re (, .) =
1
2
_
:() :(.)
_
+
=
1
2
_

_
:()
_

_
:(.)
_
_
0
for all , .

. Thus Re (, .) vanishes only when and . are both in /

. Also
we have
(, .) = (, .), when /

and .

.
We dene the kernel

+
(, .) by modifying
+
with substitute for . Set
(11.5.9)

+
(, .) =
n3

q=0

+
q
(, .),
where

+
q
(, .) =
1
(2i)
n
r
n
d
n
(, .)

< G

(), d
(, .)

< G
+
(.), d
(, .)

_
<

(), d
(, .)
_n3q

_
<
z
G
+
(.), d
(, .)
_q
11.5 L
p
Estimates for Local Solutions of

b
299
away from the singularities. The kernel

+
q
has exactly d .s. Since
(11.5.10)

+
(, .) =
+
(, .), when /

and .

,
we shall substitute

+
in 1
2
() for
+
. The advantage is that

+
is integrable
for each xed .

since satises (11.5.8). Thus for any .

, by Stokes
theorem and a limiting argument (substituting

= + c and

= + c for
and , approximating by smooth forms

such that

and
b

0
uniformly on

, then letting c 0), we can write


(11.5.11)
1
2
()(.) =
_
b

+
q1
() =
_

+
q1
()
_
=
_

+
q1
().
From (11.3.28), we have
[ < G

(), d < G
+
(.), d [ = O([ .[).
Thus for every 1 < n 1,
(11.5.12)

+
q1
(, .) =
O([ .[)
(, .)
nq1
(, .)
q
(
n
, .
n
)
.
We write

+
q1
(, .) =
1
(2i)
n

_
r
n
d
n
(, .)

< G

, d
(, .)

< G
+
G

, d
(, .)

_
<

, d
(, .)
_n2q

_
<
z
G
+
, d
(, .)
_q1_
.
It follows from the denition of and that
_

(, .) = O([ .[),

(, .) = 0.
Using < G

, d =

,
r
n
d
n
=

: and estimate (11.3.15), after grouping terms


of the same form together, we have
[

+
q1
(, .)[ C

_
[ .[
[ (, .)[[(, .)[
n1q
[(, .)[
q
+
[

: \
2n3
()[
[ (, .)[[(, .)[
n1q
[(, .)[
q
+
[ .[[

\
2n4
()[
[ (, .)[
2
[(, .)[
n1q
[(, .)[
q
_
,
300 Integral Representations for and
b
where

ranges over all possible monomials \


2n3
() and \
2n4
() of degree 2n3
and 2n 4 respectively in d
1
, d
1
, , d
n
, d
n
. Let
1
1
(, .) =
[ .[
[ (, .)[[(, .)[
n1q
[(, .)[
q
, (11.5.13)
1
2
(, .) =
[

: \
2n3
()[
[ (, .)[[(, .)[
n1q
[(, .)[
q
, (11.5.14)
1
3
(, .) =
[ .[[

\
2n4
()[
[ (, .)[
2
[(, .)[
n1q
[(, .)[
q
. (11.5.15)
We dene
J
i
()(.) =
_

[()[1
i
(, .)d:
2n1
(), i = 1, 2, 3,
where :
2n1
() is the surface measure of

. For .

, using [ (, .)[ 0,
_

[1
i
(, .)[d:
2n1
() C
z
, i = 1, 2, 3,
where C
z
depends on .. Thus the operator J
i
is bounded from 1
p
(

) to 1
p
(

, loc),
i = 1, 2, 3. Near the boundary /

, 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
()|
L
p
(

)
c||
L
p
(

)
i = 1, 2, 3.
Let :

(
n
) = :(
n
) be the dening function for

. 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
(, .)
satises
(11.5.17)
_

[:

(
n
)[

1
i
(, .) c

[:

(.
n
)[

for all .

,
(11.5.18)
_
z

[:

(.
n
)[

1
i
(, .) c

[:

(
n
)[

for all

,
then for 1 < j < , there exists c
p
0 such that
|J
i
()|
L
p
(

)
c
p
||
L
p
(

)
for all 1
p
(

).
11.5 L
p
Estimates for Local Solutions of

b
301
Proof. By Holders inequality and (11.5.17), we have
[J
i
()(.)[
p

1
i
(, .)[()[
p
[:

(
n
)[
p/p

d:
2n1
()

__

1
i
(, .)[:

(
n
)[

d:
2n1
()
_
p/p

(c

)
p/p

[:

(.
n
)[
p/p

1
i
(, .)[()[
p
[:

(
n
)[
p/p

d:
2n1
(),
where
1
p
+
1
p

= 1. Integrating with respect to . and interchanging the order of


integration we obtain, using (11.5.18),
_
z

[J
i
()(.)[
p
d:
2n1
(c

)
p/p

__
z

[:

(.
n
)[
p
p

1
i
(, .)d:(.)
_
[()[
p
[:

(
n
)[
p
p

d:()
(c

)
p/p

(c
p/p
)||
p
L
p
(

)
.
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 suces to prove (11.5.17) and (11.5.18) for i = 1, 2, 3.
Using a partition of unity in both and . variables, we can assume that lies in
a coordinate patch l and . lies in a coordinate patch \. When l \ = , then
[(, .)[ 0 and [(, .)[ 0 for l and . \ and the estimation will be
simpler. We assume l and \ are the same coordinate patch and omit the other
cases.
Let denote the set of the characteristic points, i.e., points where : = 0
on /

. We rst assume that l = . We shall choose special coordinates for

l.
Since d d: ,= 0 on /

, we can choose :(
n
) as a coordinate function near
l

. Since d

(, .)[
=z
= () and = on

, it follows that () =
1
2
( ) = id

Im(, .)[
=z
. Thus,
() = id

Im(, .) +O([ .[).


Similarly for /

, we have

: = id

Im (
n
, .
n
) +O([
n
.
n
[).
Thus, if /

,
d:(
n
) d

Im(, .) d

Im (
n
, .
n
) d()[
=z
= d:(
n
)

()

:(
n
) d()
=

,= 0.
302 Integral Representations for and
b
Let Im(, .) = t
1
and Im (, .) = t
2
. We can choose coordinates (:(
n
), t
1
, ,
t
2n2
) with t
i
(.) = 0, i = 1, , 2n 2. From (11.3.15), (11.3.16) and (11.5.8),
there exists c 0 with
[(, .)[ c([t[
2
+[t
1
[), (11.5.19)
[Re(, .)[ c[t[
2
, (11.5.20)
[Re (, .)[ c([:() [ +[:(.) [). (11.5.21)
It follows that there exists C 0 independent of . such that
[1
1
(, .)[
C
([:() [ +[:(.) [)[t[
2n3
,
[1
2
(, .)[
C
([:() [ +[:(.) [)([t
1
[ +[t[
2
)
n1q
[t[
2q
,
[1
3
(, .)[
C
([:() [ +[:(.) [ +[t
2
[)
2
([t
1
[ +[t[
2
)
n1q
[t[
2q1
.
Estimate (11.5.17) will be proved for i = 1, 2, 3 when and . are away from the
characteristic points using the following lemma (letting j = [:() [ and =
[:(.) [).
Lemma 11.5.6. Let t = (t
1
, , t
2n2
) and dt = dt
1
dt
2
dt
2n2
. For any 0 <
c < 1, 0, there exists c
i

0, i = 0, 1, 2, 3, such that the following inequalities


hold: For any 0, 1 n 2,
(1)
_

0
j

+j
dj c
0

,
(2)
_

0
_
]t]A
j

( +j)[t[
2n3
dtdj c
1

,
(3)
_

0
_
]t]A
j

( +j)([t
1
[ +[t[
2
)
n1q
[t[
2q
dtdj c
2

,
(4)
_

0
_
]t]A
j

dtdj
( +j +[t
2
[)
2
([t
1
[ +[t[
2
)
n1q
[t[
2q1
c
3

.
Proof. (1) follows from a change of variables to the case when = 1. In fact one
can show using contour integration that c
0

= , sinc.
Estimate (2) follows from (1) by using polar coordinates for t variables. To prove
(4), we integrate t
1
, t
2
rst and then use polar coordinates for t
tt
= (t
3
, , t
2n2
)
and apply (1) to obtain
_

0
_
]t]A
j

( +j +[t
2
[)
2
([t
1
[ +[t[
2
)
n1q
[t[
2q1
dtdj
C
_

0
_
]t

]A
[ log [t
tt
[[j

( +j)([t
tt
[
2
)
n2q
[t
tt
[
2q1
dt
tt
dj
C
_

0
__
0<vA
[ log [
2n5

2n5
d
_
j

( +j)
dj
c
3

.
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 and , and ..
Near any characteristic point . , we cannot choose :(
n
), Im (, .) and
Im (, .) as coordinates since they are linearly dependent at = .. The kernel 1
1
is less singular than 1
2
or 1
3
and can be estimated as before by choosing :(
n
)
and Im = 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
(, .), 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
(, .) = Im(, .),
p
2
(, .) = Im (, .) =
1
2
Re(
n
.
n
).
It is easy to see that d

Im (, .) dj =
1
i
: :. Since = (

),2 =
id

Im(, .) +O([ .[), we have, setting j(


n
) = :(
n
),
dp
1
(, .) dp
2
(, .) dj()

=z
= d

Im(, .) d

Im (, .) d:()

=z
=

:() :()

=z
.
Thus (11.5.17) will be proved for i = 3 if we can prove that
(11.5.22)

J
3
(.) C

,
where

J
3
(.) is the integral
_
U

[dp
1
(, .) dp
2
(, .) dj \
2n4
()[
([p
2
(, .)[ +j +)
2
([p
1
(, .)[ +[ .[
2
)
n1q
[ .[
2q1
.
The other terms are less singular and can be estimated as before.
Let r = (r
1
, , r
2n2
, j) = (r
t
, j) be real coordinates on l

such that
. = (0, , 0, j(.)) where r
i
= Re(
j
.
j
) or r
i
= Im(
j
.
j
) for some , = 1, , n.
In this coordinate system, we have, for some 0,

J
3
is bounded by the integral

3
(.) =
_
]x]<A
j

[dp
1
(r, .) dp
2
(r, .) dj \
2n4
(r)[
([p
2
(r, .)[ +j +)
2
([p
1
(r, .)[ +[r[
2
)
n1q
[r[
2q1
,
where \
2n4
is a monomial of degree 2n 4 in dr
1
, , dr
2n2
. Without loss of
generality, we can assume that \
2n4
= dr
3
dr
2n2
. Let 1
A
= r 1
2n1
[
[r[ < . The integral

3
(.) is the pull-back of the integral 1,
1 =
_
t=(t

,)G(B
A
)
j

dt
1
dt
2
dt
2n2
dj
( +j +[t
2
[)
2
([t
1
[ +[t[
2
)
n1q
[t[
2q1
,
by the map
G : r 1
A
G(r) = (p
1
, p
2
, r
3
, , r
2n2
, j).
304 Integral Representations for and
b
1 can be estimated by
1 =
_
t=(t

,)G(B
A
)
j

dt
1
dt
2
dt
2n2
dj
( +j +[t
2
[)
2
([t
1
[ +[t[
2
)
n1q
[t[
2q1

_

0
_
]t

]A

dt
1
dt
2
dt
2n2
dj
( +j +[t
2
[)
2
([t
1
[ +[t[
2
)
n1q
[t[
2q1
c
3

,
using (4) in Lemma 11.5.6. However, the set (p
1
, p
2
, r
3
, , r
2n2
, j)[[r[ <
may cover the image G(1
A
) innitely many times. We have to modify the function
p
1
to guarantee a nite covering. We approximate p
1
(r, .) by the polynomial p
1
(r, .)
where p
1
(r, .) is the second-order Taylor polynomial in r of p
1
(r, .) at the point ..
From (11.3.15) we have for [r[ suciently small,
(11.5.23) [(r, .)[ c([ p
1
(r, .)[ +[r[
2
)
where c is independent of ..
For each xed o
1
1 and xed r
2
, r
3
, , r
2n2
, j 0, the equation p
2
(r, .) =
o
1
has at most two solutions by the strict convexity of the dening function for

.
For each xed o G(1
A
), ( p
1
, p
2
, r
3
, , r
2n2
, j) = o has at most four solutions
from Bezouts theorem (see [GrHa 1]) since p
1
is a second-order polynomial in r.
With (11.5.23), one can estimate (11.5.22) by substituting p
1
(r, .) by p
1
(r, .) plus
remaining terms which are less singular. We have
_
]x]<A
j

[d p
1
(r, .) dp
2
(r, .) dj \
2n4
(r)[
([p
2
(r, .)[ +j +)
2
([ p
1
(r, .)[ +[r[
2
)
n1q
[r[
2q1
C1 C

for any 0 < c < 1. This proves (11.5.22). Reversing the roles of and , and and
., one can show (11.5.18) for i = 3 similarly. This proves (11.5.16) for i = 3 and
(11.5.24) | 1
2
() |
L
p
(

)
C | |
L
p
(

)
.
Combining (11.5.6) and (11.5.24), (11.5.4) follows with n = o
q
. We have proved
Proposition 11.5.4.
In order to prove Theorem 11.5.1, we need the following density lemma:
Lemma 11.5.7. Under the same assumption as in Theorem 11.5.1, the set of
b
-
closed forms in C
(0,q)
(

) is dense in the set of


b
-closed 1
p
(0,q)
(

) forms in the
1
p
(0,q)
(

) norm where 1 < j < , 0 < n 2.


Proof. Let 1
p
(0,q)
(

) and

b
= 0 on

. We approximate by smooth (0, )-


forms
k
C

(0,q)
(

) such that
k
in 1
p
(0,q)
(

) and
b

k
0 in 1
p
(0,q+1)
(

).
This is possible by Friedrichs Lemma (see Appendix D). Since
b

k
is a smooth

b
-closed form on a slightly larger set

where
k
and
k
is suciently
close to , we can apply Proposition 11.5.4 to
b

k
on

k
(since 1 +1 < n1)
to nd (0, )-forms
k
such that
(11.5.25)
_

k
=
b

k
on

k
,
|
k
|
L
p
(0,q)
(

k
)
c
p
|
b

k
|
L
p
(0,q+1)
(

k
)
,
11.5 L
p
Estimates for Local Solutions of

b
305
where c
p
is a constant independent of / and . This is true since the constant
proved in Proposition 11.5.4 is independent of small perturbation of . We set
(11.5.26)
t
k
=
k

k
,
then
t
k
C
(0,q)
(

). It follows from (11.5.25) that


t
k
is
b
-closed and
t
k
in
1
p
(0,q)
(

). This proves the lemma.


For the case when = n 2, we have the following density lemma:
Lemma 11.5.8. For every
b
-closed form 1
p
(0,n2)
(

), there exists a sequence


of
b
-closed forms
k
such that
k
C

(0,n2)
(

) and
k
converges to in
1
p
(0,n2)
(

, loc), 1 < j < .


Proof. Let 1 denote the space of all
b
-closed 1
p
(0,n2)
(

) forms. We note that the


dual of 1
p
(

, loc) in the Frechet norm is the space of compactly supported functions


in 1
p

), where
1
p
+
1
p

= 1. Consider the linear functional 1 on 1


p
(0,n2)
(

, loc)
dened by
(11.5.27) 1() =
_

p for 1 1
p
(0,n2)
(

, loc),
where p 1
p

) such that p has compact support in

. We assume that 1() =


_

p = 0 for every 1 C

(0,n2)
(

). From the Hahn-Banach theorem, the


lemma will be proved if one can show that
1() = 0 for every 1.
Let 1 be a strictly convex set in C
n
such that the boundary of 1, denoted by `,
contains

. Let 1 = supp p

. Since (11.5.27) holds for all =


b
for any
C

(0,n3)
(

),
b
p = 0 on

in the distribution sense. We extend p to be zero


on `

, then
b
p = 0 on ` in the distribution sense. Applying Theorem 11.3.9
for (n, 1) forms with 1
p

(`) coecients on `, we can nd n = 1


q
) 1
p

(n,0)
(`)
such that
b
n = p on `. It follows from (11.3.27) that n C

(n,0)
(` 1). Let
0 <
0
<
1
< be chosen such that 1
0

1

1

and let be
a cut-o function such that C

0
(
1
) and 1 on 1. We set n
1
= n and
n
2
= (1 )n, then n
1
,
b
n
1
and
b
n
2
have compact support. Thus, we can write
_

p =
_


b
n =
_


b
n
1
+
_


b
n
2
.
We shall prove that for every 1,
(11.5.28)
_


b
n
1
= 0
(11.5.29)
_


b
n
2
= 0.
306 Integral Representations for and
b
Since n
1
has compact support, we regularize n
1
and (11.5.28) follows easily from
Friedrichs lemma and integration by parts.
To prove (11.5.29), we note that the coecients of n
2
are C1 functions on `
1
since
b
n
2
= p = 0 on `
1
. It follows from Theorem 3.3.2 (Lewys extension)
that one can extend n
2
holomorphically into the set 1
1
= 1 . C
n
[ :(.
n
)

1
. The set 1
1
is convex. We can approximate n
2
C

(n,0)
(1
1
) by (n, 0)-forms
1
n
with polynomial coecients and the convergence is in C

(`
1
). Let
1
be a
cut-o function such that
1
C

0
(

) and
1
1 on
1
. Since
b
n
2
is supported
on
1
1, we have
(11.5.30)
_


b
n
2
=
_

b
(n
2
1
n
)
=
_


b
(
1
(n
2
1
n
))
_

(
b

1
) (n
2
1
n
).
The rst term on the right-hand side of (11.5.30) vanishes from the same arguments
of (11.5.28). Thus
_


b
n
2
=
_

(
b

1
) (n
2
1
n
) 0
as n , since n
2
1
n
converges to 0 uniformly on the supp(
b

1
). This proves
(11.5.29) and Lemma 11.5.8.
Proof of Theorem 11.5.1. Theorem 11.5.1 can be proved for any
b
-closed with
1
p
(

) coecients using an approximation argument. We rst assume 1 < n2.


Using Lemma 11.5.7, there exists a sequence of

b
-closed forms
t
m
C
(0,q)
(

)
such that
t
m
in 1
p
(0,q)
(

). We can apply Proposition 11.5.4 to


t
m
to nd
(0, 1)-form n
m
such that
(11.5.31)
b
n
m
=
t
m
on

,
and
(11.5.32) |n
m
|
L
p
(0,q1)
(

)
c
p
|
t
m
|
L
p
(0,q)
(

)
.
From (11.5.31) and (11.5.32), n
m
must converge to some (0, 1)-form n such that
n satises

b
n = on

and
(11.5.33) |n|
L
p
(0,q1)
(

)
c
p
||
L
p
(0,q)
(

)
.
Theorem 11.5.1 is proved for 1 < n 2.
When = n 2, from Lemma 11.5.8, there exists
k
C

(0,n2)
(

) such that

k
= 0 on

and
k
in 1
p
(0,n2)
(

) for any 0 <


t
< . Let
m
be an
increasing sequence such that 0 <
m
. Applying Proposition 11.5.4 to
k
on

m, there exists a solution n


m
k
C
(0,n3)
(

m) such that
(11.5.34)
b
n
m
k
=
k
on

m
11.6 The

b
-Neumann Problem 307
and
(11.5.35) |n
m
k
|
L
p
(0,n3)
(

m)
C|
k
|
L
p
(0,n2)
(

m)
,
where C is independent of : and /. It follows that n
m
k
converges strongly to an
element n
m
1
p
(0,n3)
(

m) for every : and


(11.5.36)
b
n
m
= on

m.
Furthermore, we have
(11.5.37) |n
m
|
L
p
(0,n3)
(

m)
C||
L
p
(0,n2)
(

)
,
where C is independent of :. There exists a subsequence of n
m
which converges
weakly in 1
p
(0,n3)
(

) to a limit n 1
p
(0,n3)
(

). It follows from (11.5.36) that

b
n = on

in the distribution sense. From Fatous lemma and (11.5.37) we have


|n|
L
p
(0,n3)
(

)
C||
L
p
(0,n2)
(

)
.
Theorem 11.5.1 is proved for = n 2. This proves Theorem 11.5.1.
Corollary 11.5.2 follows easily from Theorem 11.5.1. To prove Corollary 11.5.3,
we dene

o
q
= 1
1
() +

1
2
()
where
1
1
() =
_

+
q1
(, .) (),
and

1
2
() =
_

+
q1
().
Corollary 11.5.3 follows from the proof of Proposition 11.5.4 and Lemma 11.5.7.
11.6 The

b
-Neumann Problem
Let 1 be a strongly pseudoconvex domain in C
n
with smooth boundary /1. Let

be an open connected subset in /1 with smooth boundary /

dened in (11.5.2).
The 1
2
existence theory for

b
can be applied to obtain the Hodge theorem for

b
on

. We shall set up the


b
-Neumann problem along the lines of the -Neumann
problem for pseudoconvex complex manifolds. Let
b
be the linear, closed, densely
dened operator

b
: 1
2
(0,q1)
(

) 1
2
(0,q)
(

).
The formal adjoint of
b
is denoted by
b
and dened on smooth (0, )-forms by the
requirement that (
b
, ) = (,
b
) for all smooth with compact support in

.
The Hilbert space adjoint of
b
, denoted by

b
, is a linear, closed, densely dened
operator dened on Dom(

b
) 1
2
(0,q)
(

). An element belongs to Dom(

b
) if
308 Integral Representations for and
b
there exists a p 1
2
(0,q1)
(

) such that for every Dom(


b
) 1
2
(0,q1)
(

), we
have (,
b
) = (p, ). We then dene

b
= p. We have the following description
of the smooth forms in Dom(

b
):
For all c
(0,q)
(

), c
(0,q1)
(

), integration by parts gives


(11.6.1) (
b
, ) = (,
b
) +
_
b

(
b
, d:), )d:,
where d: is the surface measure of /

and (
b
, d:) denote the symbol of
b
in
the d: direction. More explicitly, for every r /

, (
b
, d:)

x
=
b
(:)

x
. The
following characterization of Dom(

b
) c
(0,q)
(

) uses arguments similar to those


in Lemma 4.2.1:
Proposition 11.6.1. Dom(

b
) c
(0,q)
(

) if and only if (
b
, d:) = 0 on
/

. If Dom(

b
) c
(0,q)
(

),

b
=
b
.
We next dene the

b
-Laplacian
b
=

b
+

b
from 1
2
(0,q)
(

) to 1
2
(0,q)
(

)
such that Dom(
b
) = ) Dom(

b
) Dom(

b
);

b
) Dom(

b
) and

b
)
Dom(

b
). Repeating the proof of Proposition 4.2.3, we have the following propo-
sition:
Proposition 11.6.2.
b
is a linear, closed, densely dened self-adjoint operator.
We note that the smooth forms in Dom(
b
) must satisfy two boundary condi-
tions, namely, the
b
-Neumann boundary conditions. To be precise, we have the
following:
Proposition 11.6.3. c
(0,q)
(

) is in Dom(
b
) if and only if (
b
, d:) = 0
on /

and (
b
, d:)
b
= 0 on /

.
The

b
-Neumann problem is formulated in exactly the same way as the

-
Neumann problem. However, due to the existence of characteristic points, it is
much harder to study this boundary value problem using a priori estimates by
imitating the

-Neumann problem. By applying the 1
2
existence result proved in
Theorem 11.5.1, we have the following 1
2
existence theorem for the

b
-Neumann
operator on

:
Theorem 11.6.4. Let ` be a strongly pseudoconvex hypersurface in C
n
, n 4,
and .
0
`. Let

`, be the neighborhood base of .


0
obtained in Theorem
11.5.1. Then for each xed , 1 n 3, there exists a linear operator ^

:
1
2
(0,q)
(

) 1
2
(0,q)
(

) such that
(1) ^

is bounded and Range(^

) Dom(
b
).
(2) For any 1
2
(0,q)
(

), =
b

b
^

b
^

.
(3) ^

b
=
b
^

= 1 on Dom(
b
);

b
^

= ^

b
on Dom(
b
), 1 n 4;

b
^

= ^

b
on Dom(

b
), 2 n 3.
(4) If 1
2
(0,q)
(

) and
b
= 0, then =
b

b
^

. The form n =

b
^

gives the canonical solution (i.e., the unique solution which is orthogonal to
Ker(
b
)) to the equation

b
n = .
11.6 The

b
-Neumann Problem 309
Using Theorem 11.5.1 and Corollary 11.5.2, the

b
operator has closed range
in 1
2
(0,q)
(

) when 1 n 2. Theorem 11.6.4 can be proved by repeating


the arguments of the proof of Theorem 4.4.1, Thus the 1
2

b
-Neumann problem is
solved for 1 n 3 and ^

is called the
b
-Neumann operator.
Thus the Hodge decomposition theorem for compact strongly pseudoconvex C1
manifolds proved in Theorem 8.4.10 has been extended to strongly pseudoconvex
C1 manifolds with boundaries.
We next study the interior regularity of ^

with applications to the regularity


of the solutions of
b
and the related Szego projection. Let \
s
(

) denote the
Sobolev : space and \
s
(

, loc) denote the Frechet space of functions which are in


\
s
on every compact subset of

.
Theorem 11.6.5. Under the hypothesis of Theorem 11.6.4, given \
s
(0,q)
(

),
: 0, then = ^

satises the following estimates: for any ,


1
C

0
(

) such
that
1
= 1 on the support of , there exists a C
s
0 such that
||
2
s+1
C
s
(|
1
|
2
s
+||
2
).
Proof. Since
b
= (

b
+
b

b
) = in the distribution sense on

and

is
strongly pseudoconvex, the theorem follows from the interior regularity results for

b
on strongly pseudoconvex manifolds proved in Theorem 8.4.3.
Corollary 11.6.6. Let 1
2
(0,q)
(

) \
s
(0,q)
(

, loc) and
b
= 0, 1
n 3, then there exists n 1
2
(0,q1)
(

) \
s+1/2
(0,q1)
(

, loc) satisfying
b
n = .
In particular, if 1
2
(0,q)
(

) C

(0,q)
(

), then there exists n 1


2
(0,q1)
(

)
C

(0,q1)
(

) satisfying
b
n = .
Proof. Let n =

b
^

and let
2
C

0
(

) such that
2
= 1 on supp
1
, then from
Theorem 11.6.5, we have
|n|
2
s
= |

b
|
2
s
2|

b
()|
2
s
+ 2|[,

b
]|
2
s
c(||
2
s+1
+|
1
|
2
s
)
c(|
2
|
2
s
+||
2
).
Thus, n \
s
(0,q1)
(

, loc). To show that n 1


2
(0,q1)
(

), we note that
|n|
2
= (

b
^

b
^

)
= (
b

b
^

, ^

) = (, ^

)
|| |^

| c||
2
.
To show that n \
s+1/2
(0,q1)
(

, loc), we assume rst that : = 0. Let


k
be
the pseudodierential operator of order /. Then, from Theorem 11.6.5 and the
310 Integral Representations for and
b
discussion above
|n|
2
1/2
c(
1/2

b
^

,
1/2

b
^

)
c((

b
^

,
1

b
^

) +||
2
)
c((
b

b
^

,
1
^

) +||
2
)
c(|| |^

|
1
+||
2
)
c||
2
.
For general : N, one can prove that n \
s+1/2
(0,q1)
(

, loc) similarly by induction


and we omit the details. If 1
2
(0,q)
(

) C

(0,q)
(

), then \
s
(0,q)
(

, loc) for
every : N. Thus the solution n =

b
^

\
s+1/2
(0,q1)
(

, loc) for every : N. It


follows from the Sobolev embedding theorem that n C

(0,q1)
(

) and the corollary


is proved.
Denition 11.6.7. Let H
b
(

) = ) 1
2
(

) [
b
) = 0 and let o
b
denote
the orthogonal projection from 1
2
(

) onto H
b
(

). We shall call o
b
the Szego
projection on

.
o
b
is the natural analogue of the global Szego projection. We have the following
expression for o
b
which is an analogue of the formula for the Bergman projection
using the -Neumann operator:
Theorem 11.6.8. Let ) 1
2
(

). Then o
b
) = (1

b
^

b
)). In particular, if
) C

), then o
b
) C

).
Proof. Since

b
^

b
) = (
b

b
+

b
)^

b
) =
b
)
by (3) in Theorem 11.6.4, we have

b
()

b
^

b
)) =
b
)
b
) = 0.
This implies that (1

b
^

b
)) H
b
(

). On the other hand, for any / H


b
(

),
(

b
^

b
), /) = (^

b
),
b
/) = 0.
It follows that (1

b
^

b
)) = o
b
). The interior regularity proved in Theorem
11.6.5 implies o
b
) C

) if ) C

). In fact one can show that if )


\
s
(

), then o
b
) \
s
(

, loc) following the same argument as for the Bergman


projection and we omit the details.
NOTES
The use of explicit kernels to solve the Cauchy-Riemann equations in several
variables is a dierent approach parallel to the 1
2
method. It is an attempt to
Notes 311
generalize the Cauchy integral formula in one variable to several variables. Start-
ing from the Bochner-Martinelli formula, the integral formula stated in Corollary
11.2.3 for holomorphic functions was discovered by J. Leray in [Ler 1,2]. G. M.
Henkin [Hen 1] and E. Ramirez [Ram 1] introduced Cauchy-type integral formu-
las for strictly pseudoconvex domains. Subsequently, H. Grauert and I. Lieb [GrLi
1] and G. M. Henkin [Hen 2] constructed the integral solution formulas for

on
strictly pseudoconvex domains with uniform estimates. Our exposition of the rst
three sections in this chapter follows that of R. Harvey and J. Polking [HaPo 1,2]
(see also the book of A. Boggess [Bog 2]) without referring to currents. It is their
notation that we adopt here.
The so-called Bochner-Martinelli-Koppelman formula was proved by S. Bochner
[Boc 1], E. Martinelli [Mar 1] independently for functions (when =0) and W.
Koppelman [Kop 1] for forms. Our proof is due to N. N. Tarkhanov [Tark 1].
The jump formula of the Bochner-Martinelli-Koppelman formula was proved in R.
Harvey and B. Lawson [HaLa 1] for continuous functions. For more discussion on
the Bochner-Martinelli-Koppelman formula, see the book by A. M. Kytmanov [Kyt
1].
The Holder estimates for

in strongly pseudoconvex domains were proved in N.
Kerzman [Ker 1] using the integral solution operators for

constructed by Grauert
and Lieb [GrLi 1] and Henkin [Hen 2]. 1
p
estimates were obtained by N. Kerzman
for = 1 and by N. vrelid [vr 1]. Exact Holder 1,2-estimates for

were proved
by G. M. Henkin and A. V. Romanov [HeRo 1] for (0, 1)-forms and by R. M. Range
and Y.-T. Siu [RaSi 1] for the general case. Sup-norm and Holder estimates for
derivatives of solution for are obtained in Siu [Siu 1] and Lieb-Range [LiRa 1].
Holder estimates for on piecewise strongly pseudoconvex domains are discussed
in Michel-Perotti [MiPe 1], Polyakov [Poly 1], and Range-Siu [RaSi 1]. Optimal
Holder and 1
p
estimates for

was proved by S. G. Krantz [Kra 1], where a theorem
similar to Theorem 11.3.9 was proved for

. There are also many results on integral
kernels for on weakly pseudoconvex domains (see Chaumat-Chollet [ChCh 1],
Michel [Mic 1], and Range [Ran 1,3,7,8]). We refer the reader to the books by G.
M. Henkin and J. Leiterer [HeLe 1], S. G. Krantz [Kra 2] and R. M. Range [Ran 6]
for more discussion and references on integral representations for

.
The homotopy formula for

b
on compact strictly pseudoconvex boundaries was
constructed by G. M. Henkin [Hen 3], A. V. Romanov [Rom 1] and H. Skoda [Sko
1] where Holder and 1
p
estimates for

b
are obtained. Our proof of Theorem 11.3.9
was based on [Hen 3]. These estimates have also been obtained by a dierent
method by L. P. Rothschild and E. M. Stein [RoSt 1]. Using the estimates for

b
,
G. M. Henkin [Hen 3] and H. Skoda [Sko 1] have constructed holomorphic functions
in the Nevanlinna class with prescribed zeros in strongly pseudoconvex domains.
There is another proof of the Henkin-Skoda theorem using estimates for

directly
by R. Harvey and J. Polking [HaPo 1]. When the domain is a ball, this is treated
explicitly in the book of W. Rudin [Rud 2]. The Henkin-Skoda theorem has been
extended to nite type domains in C
2
by D.-C. Chang, A. Nagel and E. M. Stein
[CNS 1] and for convex domains of nite type recently by J. Bruna, P. Charpentier
and Y. Dupain [BCD 1].
There are many results on the Holder and 1
p
estimates for

and

b
on convex
boundaries using kernel methods. In particular, Holder estimates for

on convex
312 Integral Representations for and
b
domains in C
2
and for complex ellipsoids in C
n
are proved by R. M. Range [Ran
3,8]. Ranges results have been generalized by J. Bruna and J. del Castillo [BrCa
1]. Hilbert integrals were used by J. Polking [Pol 1] to prove 1
p
estimates for

on
convex domains in C
2
. Sharp Holder estimates for

on real ellipsoids are obtained
in Diederich-Fornaess-Wiegerinck [DFW 1]. Holder estimates for

on convex nite
type domains are proved in A. Cumenge [Cum 1](see also [DiFo 4] and [DFF 1] ).
Holder and 1
p
estimates for

b
on the boundaries of real ellipsoids are proved in
M.-C. Shaw [Sha 4] (for related results for

, see [CKM 1]). Holder estimates for

b
on convex boundaries in C
2
are proved in D. Wu [Wu 1].
J. E. Fornaess [For 4] rst obtained the sup-norm estimates for

on certain
nite type domains in C
2
which are not convexible, including the Kohn-Nirenberg
domains [KoNi 3]. Using pseudodierential operators, Holder estimates for

and

b
were obtained by C. Feerman and J. J. Kohn in [FeKo 1] for nite type domains
in C
2
(see also [CNS 1]) and for domains in C
n
with diagonalizable Levi forms (see
Feerman-Kohn-Machedon [FKM 1]). 1
p
estimates for

b
for nite type domains
in C
2
were obtained by M. Christ [Chr 1].
We also note that N. Sibony [Sib 1] has given an example to show that the sup-
norm estimates for

in general fail for smooth pseudoconvex domains of innite
type. The example in [Sib 1] is not convex and is strongly pseudoconvex except
at one boundary point. It is still unknown if sup-norm estimates hold for

on
convex domains in C
2
. J. E. Fornaess and N. Sibony [FoSi 1] also showed that 1
p
estimates, 1 < j , also do not hold in general for pseudoconvex domains in C
2
with smooth boundaries except for j = 2.
The local homotopy formula discussed in Section 11.4 was derived in G. M.
Henkin [Hen 3]. This homotopy formula is useful in proving the embeddability of
abstract C1 structures (see the notes in Chapter 12). When = n2, A. Nagel and
J. P. Rosay [NaRo 1]) showed that there does not exist any homotopy formula for

b
locally on a strictly convex hypersurface. The additional compatibility condition
(11.4.2 a) derived for = n 2 was observed in M.-C. Shaw [Sha 7]. The example
given at the end of Section 11.4 was due to J. P. Rosay [Rosa 2].
The 1
p
estimates for the local solution discussed in Section 11.5 was based on the
paper of M.-C. Shaw [Sha 3]. It is proved there that there does not exist any solution
operator which maps

b
-closed forms with 1
p
, j < 2, coecients to solutions with
1
2
coecients. It is also proved in [Sha 3] that the closed range property in 1
2
for

b
is equivalent to the local embeddability of abstract strongly pseudoconvex C1
structures. Lemma 11.5.5 was based on the work of D. H. Phong and E. M. Stein
[PhSt 1] on Hilbert integral operators. Theorem 11.5.1 is also true for j = (see
the paper by L. Ma and J. Michel [MaMi 1]).
The

b
-Neumann problem on strongly pseudoconvex C1 manifolds with bound-
aries follows the paper by M.-C. Shaw [Sha 5]. The

b
-Neumann problem with
weights was discussed earlier in M. Kuranishi [Kur 1] in order to prove the embed-
ding theorem for abstract C1 structures. The weight functions used in [Kur 1] are
singular in the interior. Boundary regularity for the Dirichlet problem for
b
is
discussed by D. Jerison [Jer 1].
Solvability of

b
on a weakly pseudoconvex C1 manifold near a point of nite
type is discussed in [Sha 6]. It is proved there that near a point of nite type,
there exists a neighborhood base

such that

b
is solvable on

with interior
313
Sobolev estimates. C

solvability for

b
on weakly pseudoconvex manifolds with
at boundaries were proved by J. Michel and M.-C. Shaw [MiSh 2] based on the
barrier functions constructed in [MiSh 3]. When the boundary is piecewise at,
solvability for

b
is discussed by J. Michel and M.-C. Shaw [MiSh 3,4]. Integral
kernels on a domain in a convex hypersurface with piecewise smooth boundary are
constructed by S. Vassiliadou in [Vas 1].
For integral formulas for

on domains which are not pseudoconvex, we refer the
reader to the paper by W. Fischer and I. Lieb [FiLi 1] and the book by G. M. Henkin
and J. Leiterer [HeLe 2]. There are also results on the local solvability for

b
when
the Levi form is not denite. For more discussion on the integral representation
for local solutions for

b
under condition Y(), we refer the reader to the papers
by R. A. Airapetyan and G. M. Henkin [AiHe 1], A. Boggess [Bog 1], A. Boggess
and M.-C. Shaw [BoSh 1] and M.-C. Shaw [Sha 8,9]. The reader should consult the
book by A. Boggess [Bog 2] for more discussions on integral representations for

b
and C1 manifolds.
314 Integral Representations for and
b
CHAPTER 12
EMBEDDABILITY OF ABSTRACT
C1 STRUCTURES
The purpose of this chapter is to discuss the embeddability of a given abstract
C1 structure. This includes local realization of any real analytic C1 structure. In
Section 12.2, using the subelliptic estimate for
b
obtained in Chapter 8, global C1
embeddability into complex Euclidean space of any compact strongly pseudoconvex
C1 manifold of real dimension 2n 1 with n 3 is proved. In Sections 12.4 and
12.5, we present three dimensional counterexamples to the C1 embedding either
locally or globally.
12.1 Introduction
Let (`, T
1,0
(`)) be a smooth C1 manifold of real dimension 2n 1, n 2,
as dened in Section 7.1. If ` is dieomorphic to another manifold `
1
of equal
dimension via a map , then clearly induces a C1 structure

(T
1,0
(`)) on `
1
.
Since the most natural C1 structures are those induced from complex Euclidean
spaces on a smooth hypersurface, it is of fundamental importance to see whether
a given abstract C1 structure T
1,0
(`) on ` can be C1 embedded into some
C
N
or not. Namely, can one nd a smooth embedding of ` into C
N
so that
the induced C1 structure

(T
1,0
(`)) on (`) coincides with the C1 structure
T
1,0
(C
N
) CT((`)) from the ambient space C
N
. More precisely, we make the
following denition:
Denition 12.1.1. Let (`, T
1,0
(`)) be a C1 manifold. A smooth mapping
from ` into C
N
is called a C1 embedding if
(1) is an embedding, namely, is a one-to-one mapping and the Jacobian of
is of full rank everywhere,
(2)

(T
1,0
(`)) = T
1,0
(C
N
) CT((`)).
The C1 embedding problem could be formulated either locally or globally. The
following lemma shows that condition (2) in Denition 12.1.1 is equivalent to the
fact that each component
j
of is a C1 function.
Lemma 12.1.2. Let (`, T
1,0
(`)) be a C1 manifold and let = (
1
, ,
N
) be
a smooth embedding of ` into C
N
. Then is a C1 embedding if and only if
j
is
a C1 function for 1 , .
12.1 Introduction 315
Proof. If is a C1 embedding, then for any type (0, 1) vector eld 1 on `, we
have 1(
j
) =

1(.
j
) = 0 for 1 , . Thus
j
is a C1 function. By reversing
the arguments we obtain the proof for the other direction.
To conclude this section, we prove that any real analytic C1 structure is locally
realizable. Let (`, T
1,0
(`)) be a C1 manifold of real dimension 2n 1, n 2,
and let j `. Locally near j, a basis for T
1,0
(`) can be described by
(12.1.1) 1
j
=
2n1

k=1
o
jk
(r)

r
k
for , = 1, , n 1,
and the integrability condition is then equivalent to
(12.1.2) [1
j
, 1
k
] =
n1

p=1
/
jkp
(r)1
p
,
for all 1 ,, / n 1. The real analyticity of the C1 structure means that the
coecient functions o
jk
(r) dened in (12.1.1) are real analytic. The real analyticity
of the /
jkp
s then follows.
Theorem 12.1.3. Any real analytic C1 manifold (`, T
1,0
(`)) of dimension 2n
1 with n 2 can locally be C1 embedded as a hypersurface in C
n
.
Proof. We may assume that

r
2n1
, T
1,0
(`) T
0,1
(`),
and that j is the origin. Choose a small neighborhood l
0
of the origin in 1
2n1
,
and a c 0 small enough so that, when the variable r
2n1
is complexied, i.e.,
replacing r
2n1
by r
2n1
+ it, the power series of the real analytic functions that
are involved in the expressions of (12.1.1) and (12.1.2) converge on l
0
(c, c).
Dene
A
j
=
2n1

k=1
o
jk
(r
1
, , r
2n2
, r
2n1
+it)

r
k
for 1 , n 1,
and
A
n
=

r
2n1
+i

t
.
Then we have
[A
j
, A
k
] =
n1

p=1
/
jkp
(r
1
, , r
2n2
, r
2n1
+it)A
p
,
and
[A
j
, A
n
] = 0 for 1 , n 1.
316 Embeddability of Abstract CR Structures
Hence, by the Newlander-Nirenberg theorem proved in Section 5.4, there is a com-
plex structure dened on l
0
(c, c), and ` is embedded as the hypersurface
t = 0 in this complex structure. This completes the proof of the theorem.
One should note that a compact real analytic C1 manifold of real dimension
2n 1, in general, can not be globally C1 embedded into C
N
for any . A
counterexample will be provided in Section 12.4.
12.2 Boutet de Monvels Global Embeddability Theorem
Let (`, T
1,0
(`)) be a compact strongly pseudoconvex C1 manifold of real di-
mension 2n 1 with n 2. Choose a purely imaginary vector eld T dened
on ` so that T
p
is complementary to T
1,0
p
(`) T
0,1
p
(`) at each point j `.
Fix a Hermitian metric on CT(`) so that T
1,0
(`), T
0,1
(`) and T are mutu-
ally orthogonal. Let o be the orthogonal projection, called the Szego projection,
from 1
2
(`) onto the closed subspace H(`), where H(`)=) 1
2
(`)[
b
) =
0 in the sense of distribution. Denote by c
p,q
(`) the space of smooth (j, )-forms
on `. Then we have the following global embeddability theorem of the C1 struc-
tures:
Theorem 12.2.1 (Boutet de Monvel). Let (`, T
1,0
(`)) be a compact strongly
pseudoconvex C1 manifold of real dimension 2n1 with n 3. Then (`, T
1,0
(`))
can be globally C1 embedded into C
k
for some / N.
Theorem 12.2.1 will follow from the next theorem.
Theorem 12.2.2. Let (`, T
1,0
(`)) be a compact strongly pseudoconvex C1 man-
ifold of real dimension 2n 1 with n 2. Suppose that
(1)
b
: c
0,0
(`) c
0,1
(`) has closed range in the C

topology, and that


(2) o maps C

(`) into C

(`) continuously in the C

topology.
Then (`, T
1,0
(`)) can be globally C1 embedded into complex Euclidean space.
Also, C1 functions separate points on `.
Proof. The rst step is to show that C1 functions separate points on `. By
assumption (2) we have the following orthogonal, topological direct sum decompo-
sition:
C

(`) = (Ker(o) C

(`)) (Range(o) C

(`)).
Let the range of
b
on c
0,0
be denoted by which is a closed subspace of c
0,1
in
the C

topology. Since both Ker(o) C

(`) and are Frechet spaces, the open


mapping theorem implies that the isomorphism
(12.2.1)
b
: Ker(o) C

(`)


and its inverse are continuous.
For each j ` we claim that there exists a
p
C

(`) satisfying
(a)
p
(j) = 0 and
b
vanishes to innite order at j,
(b) for some coordinate neighborhood system centered at j, we have
Re
p
(r) c[r[
2
,
12.2 Boutet de Monvels Global Embeddability Theorem 317
in some neighborhood of j, where c is a positive constant,
(c) Re
p
(r) 1 outside a small neighborhood of j on `.
Proof of the claim. If ` is the boundary of a smooth bounded strongly pseudo-
convex domain 1 in C
n
, and let : be a strictly plurisubharmonic dening function
for 1, then we may take (.) to be the Levi polynomial p
p
(.) at j in some small
neighborhood of j and extend it suitably to ` to satisfy (c). Namely, dene

p
(.) = p
p
(.)
=
n

j=1
:
.
j
(j)(j
j
.
j
)
1
2
n

j,k=1

2
:
.
j
.
k
(j)(j
j
.
j
)(j
k
.
k
),
in a small open neighborhood of j. Using Taylors expansion, it is easily veried
that for . ` near j we have
Re
p
(.) c[. j[
2
.
Thus
p
satises (b), and (c) is done by an appropriate extension to `. This proves
the claim for the embedded case.
If ` is an abstract C1 manifold, we can rst nd functions
1
, ,
n
C

(`)
such that
j
(j) = 0, d
1
(j), , d
n
(j) are linearly independent at j and
b

j
vanishes to innite order at j for , = 1, , n. Then
= (
1
, ,
n
) : ` C
n
is a smooth embedding of a small neighborhood of j on ` into C
n
with (j) = 0
and (`) is strongly pseudoconvex at the origin. Let p
0
(.) be the Levi polynomial
for (`) dened at the origin, then the pullback
p
(r) = p
0
(r) is dened in some
small neighborhood of j and satises conditions (a) and (b) on a small coordinate
neighborhood. Condition (c) is satised by a suitable extension of
p
to `. This
completes the proof of the claim.
Now if j, ` with j ,= , let
p
(r) be the function satisfying (a), (b) and (c)
so that Re
p
() 1. Consider the function
n
t
= c
tp
for t 0.
Then n
t
C

(`), n
t
(j) = 1 and n
t
() is close to 0 for large t 0. Write
n
t
= o(n
t
) + (1 o)(n
t
).
Applying
b
to n
t
we obtain

b
n
t
= tc
tp
(
b

p
).
We claim that
b
n
t
converges to zero in the C

topology as t +. First we
note that any /th derivative of
b
n
t
can be written in the following form:
(12.2.2) 1
k
= t
j
c
tp
1

(
b

p
)(r),
318 Embeddability of Abstract CR Structures
where (r) is a smooth function on ` and 1 , / + 1, [[ /. Hence, by (b),
(12.2.2) is bounded in some open neighborhood \
1
of j by
[1
k
[ C
k
(ct[r[
2
)
j
c
ct]x]
2
[r[
2j
[1

(
b

p
)(r)[,
for some positive constant C
k
0. Given any c 0, since
b

p
vanishes to innite
order at j and (ct[r[
2
)
j
c
ct]x]
2
is uniformly bounded for all r and t 0, one may
choose a suciently small neighborhood \
2
\
1
so that [1
k
[ < c on \
2
. For r , \
2
,
we have [r[ 0 for some constant . Letting t be suciently large, we see also
that [1
k
[ < c for r , \
2
. This proves the claim.
It follows that, by (12.2.1), (1 o)(n
t
) also converges to zero in the C

topology
and that the C1 function o(n
t
) for suciently large t 0 will separate j and .
By the same reasoning as above, we see that the functions
/
j
= o(
j
c
tp
) for , = 1, , n,
for suciently large t 0, satisfy
(1)
b
/
j
= 0 for , = 1, , n, and
(2) d/
1
(j), , d/
n
(j) are linearly independent, and
(3) /
j
(j) = 0 for , = 1, , n, if necessary, by a translation in C
n
.
Hence, for each j `, there exists an open neighborhood l
p
of j on ` and
smooth C1 functions /
1
p
, , /
n
p
such that d/
1
p
(r), , d/
n
p
(r) are linearly inde-
pendent for all r l
p
.
Now cover ` by a nite number of such l
pi
, i = 1, , /, and let p
1
, , p
s
be
the C1 functions that separate points o, / with distance d(o, /) 0 for some
constant . Then set
1 = (/
1
p1
, , /
n
p1
, /
1
p2
, , /
n
p2
, , /
1
p
k
, , /
n
p
k
, p
1
, , p
s
).
It is easily veried that 1 is a global C1 embedding of ` into C
nk+s
. The proof
of Theorem 12.2.2 is now complete.
We now return to the proof of Theorem 12.2.1.
Proof of Theorem 12.2.1. By the hypothesis of the theorem Condition Y (1) (see
Denition 8.3.3) holds on (`, T
1,0
(`)) if the real dimension of ` is at least
ve. Hence, Corollary 8.4.11 shows that the range of
b
on \
0
(0,0)
(`) is closed
in \
0
(0,1)
(`) in the 1
2
sense. The formula for the Szego projection o,
o = 1

b
together with Theorem 8.4.14 shows that o maps C

(`) continuously into itself


in the C

topology. Theorem 8.4.14 also shows that the range of


b
on c
0,0
(`)
is closed in c
0,1
(`) in the C

topology. It follows that conditions (1) and (2) in


Theorem 12.2.2 are established for any compact strongly pseudoconvex C1 manifold
(`, T
1,0
(`)) of real dimension 2n 1 with n 3. This proves Theorem 12.2.1.
12.3 Spherical Harmonics 319
12.3 Spherical Harmonics
In this section we will review the spherical harmonics in 1
n
. For any / N 0,
denote by T
k
the vector space of all homogeneous polynomials of degree / over the
complex number eld. A basis for T
k
is given by all monomials r

]]=k
of degree
/, and it is easily seen that the dimension d
k
of T
k
over C is equal to
d
k
=
_
n +/ 1
n 1
_
=
(n +/ 1)!
(n 1)!/!
.
We dene an inner product on T
k
as follows. For any 1(r) =

]]=k
o

, Q(r)
=

]]=k
/

, the inner product between 1(r) and Q(r) is dened by


(12.3.1) 1, Q) =

]]=k
o

!.
If 1(r) =

is any polynomial, set


1(1) =

.
Then the inner product (12.3.1) can be realized as a dierentiation
(12.3.2) 1, Q) = 1(1)(Q(r)).
Lemma 12.3.1. For any 1(r) T
k
, we can write
(12.3.3) 1(r) = 1
0
(r) +[r[
2
1
1
(r) + +[r[
2l
1
l
(r),
where each polynomial 1
j
(r) is homogeneous and harmonic of degree / 2, for
0 , | with | being the largest integer less than or equal to /,2.
Proof. We may assume that / 2. Dene a map
k

k
: T
k
T
k2
,
1(r) 1(r),
where is the classical Laplacian. The adjoint operator

k
of
k
is then dened
by
Q,
k
1) =

k
Q, 1),
where 1(r) T
k
and Q(r) T
k2
. A direct computation shows that
Q,
k
1) = Q(1)(1)
= Q(1)(1)
=

k
Q(1)(1).
320 Embeddability of Abstract CR Structures
This implies that

k
Q(r) = [r[
2
Q(r).
It follows that

k
is one-to-one, and the following decomposition holds:
T
k
Ker
k
Range

k
.
Hence for any 1(r) T
k
, we can write 1(r) as
1(r) = 1
0
(r) +[r[
2
Q(r),
where 1
0
(r) is a homogeneous harmonic polynomial of degree / and Q(r) T
k2
.
The proof of the lemma is then completed by an induction argument.
Lemma 12.3.1 shows that the restriction of any polynomial 1(r) to the unit
sphere o
n1
in 1
n
is given by a sum of restrictions of homogeneous harmonic
polynomials to o
n1
.
Denition 12.3.2. Denote by oH
k
the space of the restrictions to the unit sphere
o
n1
of all homogeneous harmonic polynomials of degree /, i.e., oH
k
= HT
k
[
S
n1,
where HT
k
= Ker
k
.
The restriction is clearly an isomorphism from HT
k
onto oH
k
, and
dimoH
k
= dimHT
k
= d
k
d
k2
=
_
n +/ 1
n 1
_

_
n +/ 3
n 1
_
,
for / 2. In particular, dimoH
0
= 1 and dimoH
1
= n.
The elements in HT
k
are called solid spherical harmonics and the elements in
oH
k
are called surface spherical harmonics, or simply spherical harmonics. As
an easy consequence of the Stone-Weierstrass theorem, we obtain the following
proposition:
Proposition 12.3.3. The nite linear combination of elements in

k=0
oH
k
is
uniformly dense in C(o
n1
), and 1
2
dense in 1
2
(o
n1
, d).
Proposition 12.3.4. If Y
(j)
oH
j
and Y
(k)
oH
k
with , ,= /, then
_
S
n1
Y
(j)
(r
t
)Y
(k)
(r
t
) d(r
t
) = 0.
Proof. The proof will rely on the following two facts:
(i) (Greens identity) Let 1 be a bounded domain with C
2
boundary. If ), p
C
2
(1), we have
_
D
_
)
p
n
p
)
n
_
d =
_
D
()p p)) d\,
12.3 Spherical Harmonics 321
where ,n is the unit outward normal derivative on the boundary 1.
(ii) If ) C
1
(1
n
) and is harmonic on 1
n
, then
_
Bn
)
n
d = 0.
Here 1
n
denotes the unit ball in 1
n
.
For r 1
n
, write r = :r
t
with : = [r[ and [r
t
[ = 1. If Y
(j)
oH
j
and
Y
(k)
oH
k
, dene
n
j
(r) = [r[
j
Y
(j)
(r
t
) = :
j
Y
(j)
(r
t
),
and
n
k
(r) = [r[
k
Y
(k)
(r
t
) = :
k
Y
(k)
(r
t
).
Case (I). If one of , or / is zero, say, , = 0, then n
j
(r) = c, a constant, and

n
n
k
(r
t
) =

:
(:
k
Y
(k)
(r
t
)) = /Y
(k)
(r
t
).
Thus, by fact (ii) we have
_
S
n1
Y
(j)
(r
t
)Y
(k)
(r
t
) d =
c
/
_
S
n1
n
k
n
(r
t
) d = 0.
Case (II). If both , and / are nonzero with , ,= /, then
(/ ,)
_
S
n1
Y
(j)
(r
t
)Y
(k)
(r
t
) d =
_
S
n1
_
n
j
n
k
n
n
k
n
j
n
_
d
=
_
Bn
(n
j
n
k
n
k
n
j
) d\
= 0.
This completes the proof of the proposition.
Let 1
2
(o
n1
, d) be equipped with the usual inner product. For each /
N 0, let Y
(k)
1
, , Y
(k)
m
k
be an orthonormal basis for oH
k
, where :
k
= d
k

d
k2
. It follows from Proposition 12.3.3 that

k=0
Y
(k)
1
, , Y
(k)
m
k

forms a complete orthonormal basis for 1


2
(o
n1
, d). Hence, for ) 1
2
(o
n1
, d),
we have a unique representation
) =

k=0
Y
(k)
322 Embeddability of Abstract CR Structures
such that the series converges to ) in the 1
2
norm, and Y
(k)
oH
k
can be expressed
in terms of the Fourier coecients
Y
(k)
=
m
k

p=1
Y
(k)
, Y
(k)
p
)Y
(k)
p
.
When n = 2, we have d
k
d
k2
= 2 for all / 2. It is easily seen that
HT
k
= .
k
, .
k
. This implies, by normalization,
_
1

2
_

k=0
_
1

cos/,
1

sin/
__
is a complete orthonormal basis for 1
2
(o
1
).
12.4 Rossis Global Nonembeddability Example
We shall present in this section a compact real analytic three dimensional C1
manifold which can not be globally C1 embedded into C
n
for any dimension n.
In view of Theorem 12.1.3 one sees that the nature of global embedding of a C1
structure is quite dierent from that of local embedding. Global properties of the
C1 structure should be taken into account in the set up of the global embedding
problem.
Let o
3
= (.
1
, .
2
) C
2
[ [.
1
[
2
+ [.
2
[
2
= 1 be the boundary of the unit ball in
C
2
, and let the induced C1 structure T
1,0
(o
3
) be generated by 1 = .
2
(,.
1
)
.
1
(,.
2
). Thus, (o
3
, T
1,0
(o
3
)) forms a compact strongly pseudoconvex C1 man-
ifold of real dimension three. For each t 1, [t[ < 1, dene a new C1 structure
T
1,0
t
(o
3
) on o
3
by letting T
1,0
t
(o
3
) be generated by the vector eld 1
t
= 1 + t1.
If t = 0, T
1,0
0
(o
3
) coincides with the induced standard C1 structure T
1,0
(o
3
). It
is easily veried that for [t[ < 1, (o
3
, T
1,0
t
(o
3
)) is a compact real analytic strongly
pseudoconvex C1 manifold of real dimension three.
The next theorem shows that any 1
2
integrable C1 function ) on o
3
with respect
to the C1 structure (o
3
, T
1,0
t
(o
3
)) for 0 < [t[ < 1 must be even. Obviously, this
implies that, for 0 < [t[ < 1, (o
3
, T
1,0
t
(o
3
)) can not be globally C1 embedded into
any C
n
.
Theorem 12.4.1. Any 1
2
integrable C1 function )(.) on o
3
with respect to the
C1 structure T
1,0
t
(o
3
), 0 < [t[ < 1, is even, i.e., )(.) = )(.).
Proof. Notice rst that we can decompose the space of homogeneous harmonic
polynomials of degree / as follows.
HT
k
=
p+q=k
HT
p,q
k
,
where HT
p,q
k
denotes the space of all homogeneous harmonic polynomials of degree
/ that can be expressed as a linear combination of terms .

with [[ = j, [[ =
and j + = /. We set oH
p,q
k
= HT
p,q
k
[
S
3, then we have
oH
k
=
p+q=k
oH
p,q
k
.
12.4 Rossis Global Nonembeddability Example 323
If / is a harmonic function on C
2
, then a simple computation shows that both
1/ and 1/ are also harmonic on C
2
. It follows that
1(oH
p,q
k
) oH
p1,q+1
k
,
and
1(oH
p,q
k
) oH
p+1,q1
k
.
Thus, if )(.) is a square integrable C1 function on o
3
with respect to the C1
structure T
1,0
t
(o
3
) for 0 < [t[ < 1, according to Proposition 12.3.3, there is a unique
representation
)(.) =

m0
)
m
(.),
where )
m
(.) oH
m
and the series converges to ) in the 1
2
norm. Since 1
t
)(.) = 0
in the distribution sense on o
3
, we obtain 1
t
)
m
(.) = 0 on o
3
for all : 0. For
: = 2/ + 1, we can write
)
2k+1
= )
2k+1,0
+)
2k,1
+ +)
k+1,k
+)
k,k+1
+ +)
1,2k
+)
0,2k+1
,
where )
p,q
oH
p,q
p+q
. Here we have identied )
p,q
with its preimage in HT
p,q
p+q
.
Since 1
t
)
2k+1
(.) = 0 on o
3
, we obtain
1)
2k,1
(.) = 0 on o
3
.
Hence, )
2k,1
(.) is a real analytic C1 function on o
3
. By Theorem 3.2.2, )
2k,1
[
S
3
extends smoothly to a holomorphic function 1
2k,1
(.) dened on 1
2
. Then, by
harmonicity of )
2k,1
(.) and the maximum modulus principle, we obtain )
2k,1
(.) =
1
2k,1
(.) on 1
2
. It follows that )
2k,1
(.) is holomorphic on C
2
and that no . terms
appear in )
2k,1
(.). This implies that )
2k,1
(.) 0 on C
2
.
Similarly, we obtain )
1,2k
(.) 0 on C
2
. Inductively, one can show
)
2k2,3
(.) = )
2k4,5
(.) = = )
2,2k1
(.) = )
0,2k+1
(.) 0,
and
)
3,2k2
(.) = )
5,2k4
(.) = = )
2k1,2
(.) = )
2k+1,0
(.) 0.
Therefore, )
m
(.) 0 for all odd indices :, and )(.) must be even. This completes
the proof of the theorem.
Theorem 12.4.1 indicates that a three dimensional compact strongly pseudocon-
vex C1 manifold in general can not be globally C1 embedded into a complex
Euclidean space. However, we shall show now for any 0 < [t[ < 1, (o
3
, T
1,0
t
(o
3
))
can always be C1 immersed into C
3
.
We have seen that the only possible solutions to the 1
t
equation on (o
3
, T
1,0
t
(o
3
))
are the even functions. By reasoning similarly, one can show that for / even, if
n oH
k
such that 1
t
n = 0 and n
k,0
= 0, then n = 0. It follows that the space of
solutions of 1
t
(n) = 0 in oH
2
is of dimension three which is spanned by
A =

2
2i
(.
2
1
+.
2
2
+t(.
1
2
+.
2
2
)),
Y =

2
2
(.
2
1
+.
2
2
+t(.
1
2
.
2
2
)),
7 =

2(.
1
.
2
t.
1
.
2
).
324 Embeddability of Abstract CR Structures
A direct computation shows that
(12.4.1) A
2
+Y
2
+7
2
= 2t,
and
(12.4.2) [A[
2
+[Y [
2
+[7[
2
= 1 +t
2
.
For each xed t, 0 < [t[ < 1, equation (12.4.1) denes a two dimensional complex
submanifold `
t
in C
3
. We claim that the map
(12.4.3)
: (o
3
, T
1,0
t
(o
3
)) `
t
C
3
. = (.
1
, .
2
) (A(.), Y (.), 7(.))
is a two-to-one C1 immersion.
Proof of the claim. First we show that is two-to-one. If . = (.
1
, .
2
), n = (n
1
, n
2
)
are two points on o
3
such that (.) = (n), then we have
(12.4.4) .
2
1
+.
2
2
+t(.
2
1
+.
2
2
) = n
2
1
+n
2
2
+t(n
2
1
+n
2
2
),
(12.4.5) .
2
1
+.
2
2
+t(.
2
1
.
2
2
) = n
2
1
+n
2
2
+t(n
2
1
n
2
2
),
and
(12.4.6) .
1
.
2
t.
1
.
2
= n
1
n
2
tn
1
n
2
.
From (12.4.4) and (12.4.5) we obtain
t
2
(n
2
1
.
2
1
) = n
2
1
.
2
1
.
Hence, n
1
= .
1
. If n
1
= .
1
and n
2
= .
2
, then n = .. Otherwise, we have n
1
= .
1
and n
2
= .
2
. From (12.4.6) this implies
.
1
.
2
t.
1
.
2
= 0.
Hence, .
1
.
2
= 0. If .
1
= 0 and .
2
,= 0, then n = .. If .
1
,= 0 and .
2
= 0, then
n = .. Similarly, if n
1
= .
1
, we have either n = . or n = .. Thus, is a
two-to-one mapping.
Next we show that the Jacobian of is of full rank at each point . o
3
. Since
CT(o
3
) is spanned by 1
t
, 1
t
and 1
2
1
2
, where 1
2
= .
1
(,.
1
) + .
2
(,.
2
),
it suces to show that the images

(1
t
),

(1
t
) and

(1
2
1
2
) are linearly
independent for each point . o
3
.
Let n = (n
1
, n
2
, n
3
) be the coordinates for C
3
. Suppose that we have
(12.4.7) o

(1
t
) +/

(1
t
) +c

(1
2
1
2
) = 0.
12.5 Nirenbergs Local Nonembeddability Example 325
Case (i). For [.
1
[ , = [.
2
[, we apply dn
2
and dn
3
respectively to (12.4.7) to get
(12.4.8) o(1 t
2
)(.
1
.
2
+.
1
.
2
) +c(.
2
1
+.
2
2
+t(.
2
1
+.
2
2
)) = 0,
and
(12.4.9) o(1 t
2
)([.
2
[
2
[.
1
[
2
) +c(2.
1
.
2
+t(2.
1
.
2
)) = 0.
A direct calculation shows the determinant of the coecient matrix given by (12.4.8)
and (12.4.9) is
(1 t
2
)([.
1
[
2
+[.
2
[
2
)((.
2
1
+.
2
2
) +t(.
2
1
+.
2
2
))
= (1 t
2
)((.
2
1
+.
2
2
) +t(.
2
1
+.
2
2
))
,= 0.
It follows that o = c = 0, and hence / = 0.
Case (ii). For .
2
= c
i
.
1
,= 0 with ,= ,2, 3,2, we obtain similarly from Case
(i) that o = c = 0. Then by applying dn
2
to (12.4.7) we get
0 = /(1 t
2
)(.
1
.
2
+.
1
.
2
)
= /(1 t
2
)[.
1
[
2
(c
i
+c
i
).
Hence, we have / = 0.
Case (iii). For .
2
= i.
1
,= 0, we have
dn
1
(

(1
t
)) = o(2

2)(1 t
2
)[.
1
[
2
= 0,
dn
1
(

(1
t
)) = /(2

2)(1 t
2
)[.
1
[
2
= 0,
dn
3
(

(1
2
1
2
)) = c(2

2i)(.
2
1
t.
2
1
) = 0.
Thus, o = / = c = 0. It shows that is a two-to-one C1 immersion of (o
3
, T
1,0
t
(o
3
))
into `
t
in C
3
.
12.5 Nirenbergs Local Nonembeddability Example
In this section we shall construct strongly pseudoconvex C1 structures which
are not locally embeddable.
As in Section 7.3, the following notation will be used: The Siegel upper half space

2
in C
2
is dened by

2
= (., n) C
2
[ : [.[
2
,
where . = r + ij and n = t + i:. The boundary of
2
will be denoted by `, and
will be identied with the Heisenberg group H
2
via the mapping
(12.5.1) : (., t +i[.[
2
) (., t).
326 Embeddability of Abstract CR Structures
Hence, the tangential Cauchy-Riemann operator on ` is generated by
(12.5.2) 1 =

.
2i.

n
,
and the corresponding Lewy operator on H
2
is
(12.5.3) 7 =

.
i.

t
.
We start working on H
2
= C1 with coordinates given by (r, j, t). Construct a
sequence of disjoint closed discs 1
k
on the rt-plane with centers (1,/, 0, 1,/). The
radii of these discs are chosen to be so small that 1
i
1
j
= if i ,= ,, and that
1
j
has no intersection with the t-axis for all ,. Denote by C
k
the boundary of 1
k
and by 1
k
the interior of 1
k
. Then denote by T
k
the open solid torus obtained by
sweeping 1
k
around the t-axis. The topological boundary of T
k
is denoted by o
k
which is given by sweeping C
k
around the t-axis.
Now lift these objects from H
2
via the mapping to `, namely, set

C
k
=
1
(C
k
),

o
k
=
1
(o
k
) and

T
k
=
1
(T
k
).
Next, let 1 be the projection from C
2
onto the second component, i.e., 1(., n) =
(0, n), and set
C
t
k
= 1(

C
k
), o
t
k
= 1(

o
k
) and T
t
k
= 1(

T
k
).
It is then easily seen that C
t
k
= o
t
k
is a sequence of disjoint simple closed curves
in the rst quadrant of the n-plane converging to the origin, and T
t
k
is exactly the
open region bounded by C
t
k
. Obviously, we have the following lemma:
Lemma 12.5.1. 1(`

k=1

T
k
) is a connected subset of n C[ : 0 which
contains the t-axis.
For any function ) : H
2
C, let

) be the lifting of ) to `, namely, ) =

)
1
.
Hence

) is a C1 function on `, i.e., 1

) = 0 on `, if and only if 7) = 0 on H
2
.
Then we have
Lemma 12.5.2. Let

) : ` C be a C
1
function.
(1) If 1

) = 0 on an open subset \ of `, then the function


1(n) =
_
(w)

) d.
is holomorphic on n C[ (n) \ , where (n) = ` 1
1
(n).
(2) If 1

) = 0 on `

k=1

T
k
, and (n) `

k=1

T
k
, then
_
(w)

) d. = 0.
(3) If 1

) = 0 on `

k=1

T
k
, then for each / 1, we have
__

S
k

) d. dn = 0.
12.5 Nirenbergs Local Nonembeddability Example 327
Proof. For (1), notice that

) is a C1 function of class C
1
. Hence,

) can be extended
to a C
0
function, denoted also by

), in an ambient neighborhood so that 1

) exists
and is continuous on \ and

) vanishes on \ . Also, the circle
(n = t +i:) = (., t +i:)[ : = [.[
2

can be parameterized by . =

:c
i
, 0 2. Then, we have

n
1(n) =

n
_
(w)

) d.
=

n
_
2
0

)(., n)
.

d
=
_
2
0
_


)
.
(., n)
.
n
.

+

)(., n)

2
.
n
_
d
=
_
2
0
d
d
_

)(., n)
.
n
_
d
= 0.
The assertion in (2) follows now from (1). First, 1(n) is holomorphic on the
interior of the set 1 = n C[ (n) `

k=1

T
k
and continuous up to the
boundary of 1. Observe that (n) degenerates to just a point on the t-axis, this
implies 1(n) = 0 on the t-axis, and hence 1(n) = 0 on 1.
For (3), we parameterize C
t
k
by n() for 0 2. Then

o
k
is parameter-
ized by (.(, ), n()) = (
_
Imn()c
i
, n()), and using the fact that (n())
`

k=1

T
k
, we obtain
__

S
k

) d. dn =
_
2
0
_
2
0

)(., n)
dn
d
.

dd
=
_
2
0
__
2
0

)(., n)
.

d
_
dn
d
d
=
_
2
0
_ _
(w())

) d.
_
dn
d
d
= 0.
This completes the proof of Lemma 12.5.2.
Lemma 12.5.3. Let 1 be a domain with C
1
boundary on `. If

) : 1 C is a
C
1
function, then
__
bD

) d. dn = 2i
___
D
(1

)) dtdrdj,
where 1 is dened in (12.5.2).
Proof. Notice that on `, n = t +i[.[
2
. Hence,
dn = dt +i.d. +i.d..
328 Embeddability of Abstract CR Structures
Then, by Stokes theorem we have
__
bD

) d. dn =
___
D
d

) d. dn
=
___
D
_


)
.
d. +


)
n
dn
_
d. dn
=
___
D
_


)
.
+ 2i.


)
n
_
dt d. d.
= 2i
___
D
(1

)) dtdrdj.
The proof of Lemma 12.5.3 is thus completed.
Now let p be a smooth function on H
2
with support contained in

k=1
T
k
such
that p is positive on

k=1
T
k
and vanishes to innite order at the origin. Dene the
operator 7
g
on H
2
by
(12.5.4) 7
g
= 7 +p

t
.
There exists a neighborhood l of the origin such that 7
g
and 7
g
are linearly
independent and (l, 7
g
) denes a strongly pseudoconvex C1 structure on l. The
next theorem shows that (l, 7
g
) can not be realized as a three dimensional C1
submanifold of C
n
for any n 2.
Theorem 12.5.4 (Nirenberg). Let 7
g
be dened as in (12.5.4). Suppose that
)
1
and )
2
are two C
1
functions on H
2
such that 7
g
)
1
= 7
g
)
2
= 0 on l. Then
d)
1
d)
2
= 0 at the origin. In particular, the C1 structure (l, 7
g
) is not embeddable.
Proof. The corresponding vector eld of 7
g
on ` is given by
1
g
= 1 + p

t
.
It follows that
1
g

)
1
= 1
g

)
2
= 0
on
1
(l). Hence, by the construction of p, 1

)
1
= p(

)
1
,t) vanishes on `

k=1

T
k
. Lemma 12.5.2 then implies that for all / 1,
0 =
__

S
k

)
1
d. dn = 2i
___

T
k
(1

)
1
) dtdrdj
= 2i
___

T
k
p


)
1
t
dtdrdj.
12.5 Nirenbergs Local Nonembeddability Example 329
Since p is positive on

T
k
, each of the functions Re(

)
1
,t) and Im(

)
1
,t) must
vanish at some point in

T
k
for all /. Equivalently, both Re()
1
,t) and Im()
1
,t)
vanish at some point in T
k
for all /. Hence, ()
1
,t)(0) = 0. The fact that )
1
is
a C1 function with respect to the C1 structure (l, 7
g
) implies ()
1
,.)(0) = 0.
Thus, we obtain
d)
1
(0) =
)
1
.
(0)d.[
0
.
A similar argument also holds for )
2
. Therefore, d)
1
(0) and d)
2
(0) are always linearly
dependent for any two C1 functions )
1
and )
2
of class C
1
on H
2
. This proves the
theorem.
We now extend the local nonembeddability example to higher dimensions. Let `
be a smooth nondegenerate C1 manifold in C
n+1
, n 2, with signature n2 near
a point j, namely, the Levi form at j ` has either n1 negative eigenvalues and
one positive eigenvalue or n 1 positive eigenvalues and one negative eigenvalue.
We may assume j is the origin. Let :(.) be a local dening function for `. As in
the proof of Theorem 3.3.2 we may write
:(.) = Im.
n+1
+
n

j,k=1
c
jk
.
j
.
k
+O([.
t
[[t[ +[t[
2
+[(.
t
, t)[
3
)
in local coordinates . = (.
t
, .
n+1
), where .
n+1
= t + i:. Another linear change of
coordinates will turn the dening function :(.) locally to the form
(12.5.5) :(.) = : [.
1
[
2
+
n

j=2
[.
j
[
2
(.
t
, .
t
, t),
where (.
t
, .
t
, t) = O([.
t
[[t[ + [t[
2
+ [(.
t
, t)[
3
). Then we show that a small pertur-
bation of the induced C1 structure will in general yield a nonembeddable new C1
structure on `.
Theorem 12.5.5 (Jacobowitz-Treves). Let ` be the nondegenerate C1 mani-
fold with signature n 2 dened locally near the origin in C
n+1
by (12.5.5). Then
there exists a new nonembeddable C1 structure on ` which agrees with the induced
C1 structure CT(`) T
1,0
(C
n+1
) to innite order at the origin.
Proof. We shall identify ` locally with an open subset l containing the origin in
C
n
1 via the map
: (.
t
, t +i(.
t
, .
t
, t)) (.
t
, t),
where
(12.5.6) (.
t
, .
t
, t) = [.
1
[
2

j=2
[.
j
[
2
+ (.
t
, .
t
, t).
It is easily veried that type (0, 1) vector elds on ` are spanned by
(12.5.7) 1
j
=

.
j
i
j

.
n+1
, , = 1, , n,
330 Embeddability of Abstract CR Structures
where

1
= 2
_
.
1
+

z1
1 +i

t
_
and
j
= 2
_
.
j
+

zj
1 +i

t
_
for , = 2, , n.
It follows that the corresponding embeddable C1 structure on l is spanned by
(12.5.8) 7
j
=

.
j
i
j

t
, , = 1, , n,
where

1
=
.
1
+

z1
1 +i

t
and
j
=
.
j
+

zj
1 +i

t
for , = 2, , n.
To get a nonembeddable C1 structure we shall perturb the induced C1 structure
on l. Let / be any smooth function in .
n+1
with support contained in t + i:
C[ [t[ :. Note that / must vanish to innite order at the origin. We set n =
t +i(.
t
, .
t
, t), the restriction of .
n+1
to `, where (.
t
, .
t
, t) is dened in (12.5.6).
Then, the composition function / n is supported in (.
t
, t) l[ [t[ (.
t
, .
t
, t).
Dene
(12.5.9) p =
/ n
.
1
w
t
/ n
and

j
=
j
(1 +p) for , = 1, , n,
and set
(12.5.10) 7
gj
=

.
j
i

t
= 7
j
i
j
p

t
.
We claim that 7
g1
, , 7
gn
denes a new C1 structure on an open neigh-
borhood, denoted still by l, containing the origin in C
n
1 which agrees with
7
1
, , 7
n
to innite order at the origin. We shall show that
(1) [7
gj
, 7
gk
] = 0 for 1 ,, / n, and
(2) the coecients of 7
gj
are smooth and agree with those of 7
j
to innite
order at the origin.
Proof of the claim. Since the problem is purely local, we may assume that by
shrinking the domain, if necessary, the open set l is suciently small,
l = (.
t
, t) C
n
1[ [.
1
[ +[.
tt
[ +[t[ < c,
for some suciently small c 0, where .
tt
= (.
2
, , .
n
).
First, we show that the function p is well dened and smooth. The constant c
that appears below may be dierent at each occurrence. We estimate the function
n = t +i: with : = (.
t
, .
t
, t) as follows:
: [.
1
[
2
[.
tt
[
2
+c([.
1
[[t[ +[.
tt
[[t[ +[t[
2
) +c([.
1
[
2
+[.
tt
[
2
+[t[
2
)
(1 +c)[.
1
[
2
(1 c)[.
tt
[
2
+c([.
1
[
2
+[t[
2
+[.
tt
[[t[)
(1 +c +c)[.
1
[
2
+ 2cc[t[.
12.5 Nirenbergs Local Nonembeddability Example 331
Since [t[ :, we obtain if c is suciently small,
: c[.
1
[
2
,
and
[n[ c[.
1
[
2
,
on the support of / n.
Since / = O([n[
k
) for any / N, we have .
1
,= 0 if / n ,= 0. Noting that
(n,t) = O(1), we see that the denominator of p is never zero if /n ,= 0. Hence,
we get p = O([.
1
[
j
[n[
k
) for any ,, / N which in turn implies that p is smooth and
vanishes to innite order at the origin. This proves (2).
To prove (1) we note that
[7
j
, 7
k
] = 0, for 1 ,, / n.
Hence, a direct calculation shows that
[7
gj
, 7
gk
] =
_

k
_
i7
j
(p) +

j
t
(p +p
2
)
_

j
_
i7
k
(p) +

k
t
(p +p
2
)
__

t
.
Thus, for the integrability of the new C1 structure it suces to show that
7
j
(p) =
j
i

j
t
(p +p
2
), , = 1, , n,
for some function independent of ,. Since 7
j
n = 0 for , = 1, , n, we get
7
j
_
n
t
_
= i

j
t
n
t
.
Note that
7
j
n = 7
j
(n +n) = 2i
j
,
hence,
7
j
(/ n) =
_
/
n
n
_
7
j
(n) = 2i
j
_
/
n
n
_
.
It follows that
7
j
(p) = 7
j
_
/ n
.
1
w
t
/ n
_
=
2i
j
(
h
w
n)
.
1
w
t
/ n

(/ n)7
j
(.
1
w
t
/ n)
(.
1
w
t
/ n)
2
=
j
i

j
t
(p +p
2
),
where
= 2i(1 +p)
h
w
n
.
1
w
t
/ n
332 Embeddability of Abstract CR Structures
is independent of ,. This proves (1), and hence the claim.
Thus, we have shown that, for each smooth function / in .
n+1
with support
contained in t +i: C[ [t[ :, Equations (12.5.9) and (12.5.10) dene a new C1
structure on l. With an appropriate choice of /, we shall show that this new C1
structure is not realizable locally near the origin. Let ) be a C1 function of class
C
1
with respect to the new C1 structure, namely, 7
gj
) = 0 for , = 1, , n. In
particular, we have
(12.5.11) 7
g1
) = 0.
We may set .
tt
= (.
2
, , .
n
) = 0 in (12.5.11), and reduces the problem to the
case when n = 1. Obviously, we have 7
1
n = 0 for n = t + i([.
1
[
2
+ (.
1
, .
1
, t))
with = O([.
1
[[t[ + [t[
2
+ [(.
1
, t)[
3
). Then, as in the three dimensional local
nonembeddability example we study the intersection of `
1
= `[
z

=0
with the
complex line .
n+1
= j. Writing j = +i, this intersection is given by
(j) = 1
1
(j) `
1
= (.
1
, j)[ = [.
1
[
2
+ (.
1
, .
1
, ),
where 1(.
1
, .
n+1
) = (0, .
n+1
) is the projection from C
2
onto the second component.
Then we have
Lemma 12.5.6. In the j-plane there is a smooth curve given by = () such
that
(1) for < (), ( +i) = ,
(2) for = (), ( +i) is a point which varies smoothly in ,
(3) for (), ( + i) is a simple closed curve which varies smoothly in
j.
Proof. Let .
1
= r +ij, we write
1(r, j, ) = [.
1
[
2
+ (.
1
, .
1
, )
= r
2
+j
2
+ (r, j, ).
Since vanishes at the origin to the order at least two, it is easily seen that for
each xed the minimum of 1 occurs at a point (r(), j()) which varies smoothly
with . Set () = 1(r(), j(), ). This proves (1) and (2). For (3), we write
1(r, j, ) = () +Q(r, j, ) + ,
where Q is a positive denite quadratic in r
1
= r r() and j
1
= j j(). It
follows that if (), then the level sets = 1 are smooth simple closed curves
which vary smoothly with and . This completes the proof of Lemma 12.5.6.
Note that t +i: C[ [t[ < : and : (t) is an open subset in the .
n+1
-plane
with piecewise smooth boundary passing through the origin. Therefore, as in the
three dimensional local nonembeddability example one may construct a sequence
of disjoint open discs T
t
k
in this open set which converges to the origin, and let T
k
be the corresponding solid open topological torus ((T
t
k
)) in l. Now let / be a
Notes 333
smooth nonnegative function in the .
n+1
-plane with support contained in

k=1
T
t
k
such that / is positive on

k=1
T
t
k
. Dene p by (12.5.9), and let the new C1 structure
be dened by (12.5.10). Obviously, when restricted to .
tt
= 0, p is supported in

k=1
T
k
. Thus, it follows from the same arguments that Lemma 12.5.2 and 12.5.3
hold in this setting. Hence, for any solution ) of class C
1
to 7
gj
) = 0, 1 , n,
we have _
(())
) d.
1
= 0,
provided that j ,

k=1
T
t
k
, and
(12.5.12)
___
T
k
(7
1
)) drdjdt = 0.
Since 7
1
) = i
1
p(),t) on T
k
, we get
(12.5.13)
___
T
k

1
p
)
t
drdjdt = 0.
On T
k
, the previous estimate shows that both
1
and .
1
(n,t) / n are given
by .
1
+O([.
1
[
2
). Thus, (12.5.13) becomes
(12.5.14)
0 =
___
T
k
(.
1
+O([.
1
[
2
))
/ n
.
1
+O([.
1
[
2
)
)
t
drdjdt
=
___
T
k
(1 +O([.
1
[))(/ n)
)
t
drdjdt.
Equation (12.5.14) holds for all /. Hence, we must have (),t)(0) = 0. Since )
is a C1 function of class C
1
with respect to this new C1 structure, we conclude
that (),.
1
)(0) = = (),.
n
)(0) = 0. This implies d)(0) = (),.
1
)(0)d.
1
+
+ (),.
n
)(0)d.
n
. Obviously, this new C1 structure locally can not be C1
embedded into C
N
for any n + 1. This proves Theorem 12.5.5.
NOTES
Boutet de Monvels global embeddability theorem 12.2.1 for compact strongly
pseudoconvex C1 manifolds with dimension at least ve is proved in [BdM 1]. Our
presentation here follows that of J. J. Kohn [Koh 7]. Based on the ideas of Boutet
de Monvel the formulation of Theorem 12.2.2 for n = 2 can be found in [Bur 1]. For
more details concerning various properties of the spherical harmonics the reader is
referred to [StWe 1].
The nonembeddable compact strongly pseudoconvex C1 manifold (o
3
, T
1,0
t
(o
3
))
of dimension three for 0 < [t[ < 1 is due to H. Rossi [Ros 1]. We proved in
Section 12.4 that Rossis nonembeddable example can be C1 immersed into a two
dimensional complex submanifold `
t
sitting in C
3
. The image of (o
3
, T
1,0
t
(o
3
)),
334 Embeddability of Abstract CR Structures
0 < [t[ < 1, under the map dened by (12.4.3) is precisely described by (12.4.1)
and (12.4.2). In particular, the image bounds a relatively compact domain
t
in
`
t
. Thus, by combining a theorem proved by L. Boutet de Monvel and J. Sjostrand
in [BdSj 1], one can show that the Szego projection o on (o
3
, T
1,0
t
(o
3
)) must map
C

(o
3
) into C

(o
3
) continuously in the C

topology (see also [Bur 1]). It follows


from Kohns work [Koh 10] that the nonclosedness of the range of
b
on 1
2
(o
3
) in the
1
2
sense is the only obstruction to the global C1 embeddability of (o
3
, T
1,0
t
(o
3
)).
The three dimensional local nonembeddable strongly pseudoconvex C1 structure
was discovered by L. Nirenberg [Nir 4]. Theorem 12.5.5 which generalizes Niren-
bergs local nonembeddability example to higher dimension is due to H. Jacobowitz
and F. Treves [JaTr 1].
The local C1 embedding problem for a strongly pseudoconvex C1 manifold of
dimension 2n1 with n 3 is more complicated. M. Kuranishi showed in [Kur 1,2,3]
that if n 5, the answer is armative. Later, it was proved by T. Akahori [Aka 1]
that the theorem remains true for n = 4. By employing Henkins homotopy formula
proved in Theorem 11.4.1 and using interior estimates of the solution operator, S.
Webster presents in [Web 2,3] a simplied proof of the theorem for the cases n 4
(see also [MaMi 2]). The remaining case n = 3 is still open. When the Levi form has
mixed signature, C1 embedding problems are discussed in [Cat 5]. Local homotopy
formulas for
b
on C1 manifolds with mixed Levi signatures have been obtained in
[Sha 8,9] and [Tre 5].
335
336
APPENDIX
A. Sobolev Spaces
We include a short summary of the basic properties of the Sobolev spaces for the
convenience of the reader. Our goal is to give precise denitions and statements of
all theorems or lemmas about the Sobolev spaces which have been used in this book.
Since most of the results are well-known and due to the vast amount of literature
on this subject, we will provide very few proofs.
Let ) 1
1
(1
N
), the Fourier transform

) of ) is dened by
(1.1)

)() =
_
R
N
c
ix
)(r) dr,
where r =

n
j=1
r
j

j
. The estimate
|

) |

| ) |
L
1
is clear from the denition. We now list some basic properties of the Fourier trans-
form whose proofs are left to the reader or can be found in any standard text. For
instance, see Stein-Weiss [StWe 1].
Theorem A.1 (Riemann-Lebesgue). Suppose that ) 1
1
(1
N
), then

)()
C
0
, where C
0
denotes the space of continuous functions on 1
N
that vanish at inn-
ity.
Theorem A.2 (Fourier Inversion). Suppose that ) 1
1
(1
N
) and that

)()
1
1
(1
N
). Then
)(r) = (2)
N
_
R
N
c
ix

)() d, a.e.
In other words, )(r) can be redened on a Lebesgue measure zero set so that )(r)
C
0
.
Theorem A.3 (Uniqueness). If ) 1
1
(1
N
) and

)() = 0 for all 1
N
, then
)(r) = 0 almost everywhere.
Denote by o the Schwartz space of rapidly decreasing smooth functions on 1
N
,
i.e., o consists of all smooth functions ) on 1
N
with
sup
R
N
[r

)(r)[ < ,
for all multiindices , , where = (
1
, ,
N
), r

= r
1
1
r

N
N
and 1

=
1
1
x1
1

N
x
N
, each
i
is a nonnegative integer. Obviously, any smooth function
with compact support belongs to o and we have the following formulas:
(1.2)
(

)) () = (i)


)().
1


)() = (

(ir)

))().
A. Sobolev Spaces 337
Theorem A.4. The Fourier transform is an isomorphism from o onto itself.
Since 1
2
(1
N
) _ 1
1
(1
N
), the Fourier transform dened by (1.1) in general cannot
be applied to 1
2
functions directly. Using the following fundamental theorem of the
Fourier transform, one can extend the denition to 1
2
functions easily:
Theorem A.5 (Plancherels Theorem). The Fourier transform can be extended
to be an automorphism of 1
2
(1
N
) with
(1.3) |

) |
2
= (2)
N
| ) |
2
for all ) 1
2
(1
N
).
Equation (1.3) is called the Parsevals identity.
We collect a few results about the Sobolev spaces. For a detailed treatment
of the Sobolev spaces \
s
() for any real :, we refer the reader to Chapter 1 in
Lions-Magenes [LiMa 1] for smooth domains or to Grisvard [Gri 1] for nonsmooth
domains.
We rst dene the Sobolev spaces in 1
N
. Let
j(1) =

]]m
o

be a dierential operator of order : with constant coecients. Then, by (1.2), it


is easy to see that for any ) o,
(1.4)

(j(1)))() = j(i)

)().
Here, the polynomial j(i) is obtained by replacing the operator 1 in j(1) by i.
For any : 1, we dene
s
: o o by
(1.5)
s
n(r) =
1
(2)
N
_
R
N
c
ix
(1 +[[
2
)
s
2
n() d.
Set (
s
) = (1 + [[
2
)
s/2
. (
s
) is called the symbol of
s
. Dene the scalar
product (n, )
s
on o o by
(n, )
s
= (
s
n,
s
)
and the norm
| n |
s
=
_
(n, n)
s
for n o.
The Sobolev space H
s
(1
N
) is the completion of o under the norm dened above. In
particular, 1
2
(1
N
) = H
0
(1
N
). The Sobolev norms | |
H
s
(R
N
)
for any n C

0
(1
N
)
is given by
(1.6) | n |
2
H
s
(R
N
)
=
_
R
N
(1 +[[
2
)
s
[ n()[
2
d.
Next, we dene the Sobolev spaces for domains in 1
N
. Let 1
N
be a domain
with C
k
boundary, / = 1, 2, . By this we mean that there exists a real-valued
C
k
function dened in 1
N
such that = . 1
N
[(.) < 0 and [[ , = 0 on /.
338 Appendix
The implicit function theorem shows that locally, / can always be expressed as a
graph of a C
k
function. If the boundary can be expressed locally as the graph of a
Lipschitz function, then it is called a Lipschitz domain or a domain with Lipschitz
boundary.
For any domain in 1
N
, let H
s
(), : 0, be dened as the space of the
restriction of all functions n H
s
(1
N
) to . We dene the norm of H
s
() by
(1.7) | n |
H
s
()
= inf
/H
s
(R
N
)
/]=u
| | |
s(R
N
)
.
When : is a positive integer, there is another way to dene the Sobolev spaces
by weak derivatives. For any domain 1
N
, we dene \
s
() to be the space of
all the distributions n in 1
2
() such that
1

n 1
2
(), [[ :,
where is a multiindex and [[ =
1
+ +
N
. We dene the norm | |
W
s
()
by
(1.8) | n |
2
W
s
()
=

]]s
| 1

n |
2
()
< .
The space C

() denotes the space of functions which are restrictions of func-


tions in C

(1
N
) to . If is a bounded Lipschitz domain, then C

() is dense in
\
s
() in the \
s
() norm (see Theorem 1.4.2.1 in Grisvard [Gri 1]). Thus \
s
()
can also be dened as the completion of the functions of C

() under the norm


(1.8) when has Lipschitz boundary.
When = 1
N
, we have H
s
(1
N
) = \
s
(1
N
) for any positive integer :. This
follows from Plancherels theorem and the inequality
1
C

]]s
[

[
2
(1 +[[
2
)
s
C

]]s
[
s
[
2
,
where C 0.
Obviously for any bounded domain , we have H
s
() \
s
() for any . If /
is Lipschitz, the following theorem shows that the two spaces are equal:
Theorem A.6 (Extension Theorem). Let be a bounded open subset of 1
N
with Lipschitz boundary. For any positive integer :, there exists a continuous linear
operator 1
s
from \
s
() into \
s
(1
N
) such that
1
s
n[

= n.
The extension operator 1
s
can be chosen to be independent of :. In particular, we
have
\
s
() = H
s
().
For a proof of Theorem A.6, see Chapter 6 in Stein [Ste 2] or Grisvard [Gri 1].
Thus when : is a positive integer and is bounded Lipschitz, the Sobolev spaces
will be denoted by \
s
() with norm | |
s()
, or simply | |
s
A. Sobolev Spaces 339
Theorem A.7 (Sobolev Embedding). If is a bounded domain in 1
N
with
Lipschitz boundary, then there is an embedding
\
k
() C
m
() for any interger :, 0 : < / ,2.
Theorem A.8 (Rellich Lemma). Let be a bounded domain in 1
N
with Lip-
schitz boundary. If : t 0, the inclusion \
s
() \
t
() is compact.
Theorem A.9 (Trace Theorem). Let be a bounded domain in 1
N
with smooth
boundary. For : 1,2, the restriction map ) )

b
for any ) C

() can be
extended as a bounded operator from \
s
() to \
s1/2
(/). For any ) \
s
(),
)

b
\
s1/2
(/) and there exists a constant C
s
independent of ) such that
| ) |
s
1
2
(b)
C
s
| ) |
s()
.
We remark that in general, the trace theorem does not hold for : = 1,2. However,
if ) \
1/2
() and ) is harmonic or ) satises some elliptic equations, then the
restriction of ) to / is in 1
2
(c.f. Lemma 5.2.3).
Let be a bounded domain in 1
N
. We introduce other Sobolev spaces. Let
\
s
0
() be the completion of C

0
() under \
s
() norm. When : = 0, since C

0
()
is dense in 1
2
(), it follows that \
0
0
() = \
0
() = 1
2
(). If : 1,2, we also
have C

0
() is dense in \
s
(). Thus
\
s
() = \
s
0
(), :
1
2
.
This implies that the trace theorem does not hold for : 1,2. When : 1,2,
\
s
0
() _ \
s
().
We dene \
s
() to be the dual of \
s
0
() when : 0 and the norm of \
s
()
is dened by
| ) |
s()
= sup
[(), p)[
| p |
s()
,
where the supremum is taken over all functions p C

0
(). We note that the
generalized Schwarz inequality for ) \
s
(), p \
s
(),
[(), p)

[ | ) |
s()
| p |
s()
holds only when : 1,2 for a bounded domain . The proof of these results can
be found in Lions-Magenes [LiMa 1] or Grisvard [Gri 1].
The Sobolev spaces can also be dened for functions or forms on manifolds. Let
` be a compact Riemannian manifold of real dimension . Choose a nite number
of coordinate neighborhood systems (l
i
,
i
)
m
i=1
, where

i
: l
i

\
i
1
N
is a homeomorphism from l
i
onto an open subset \
i
contained in 1
N
. For each i,
1 i :, let
i
j

N
j=1
be an orthonormal basis for CT

(`) on l
i
, and let
i

m
i=1
340 Appendix
be a partition of unity subordinate to l
i

m
i=1
. Thus, locally on each coordinate
chart l
i
, one may express a smooth :-form as
(1.9) =

t
]I]=r

i
I

i
I
,
where 1 = (i
1
, , i
r
) and
i
I
=
i
i1

i
ir
. Then, we dene the Sobolev : norm
of c
r
(`), for : 1, by
(1.10) | |
2
s
=
m

i=1

t
]I]=r
| (
i

i
I
)
1
i
|
2
s
.
Denote by \
s
r
(`) the completion of c
r
(`) under the norm| |
s
. The denition of
\
s
r
(`) is highly nonintrinsic. Obviously, it depends on the choice of the coordinate
neighborhood systems, the partition of unity and the local orthonormal basis
i
j
.
However, it is easily seen that dierent choices of these candidates will come up
with an equivalent norm. Therefore, \
s
r
(`) is a well-dened topological vector
space. If ` is a complex manifold of dimension n and is a relatively compact
subset in `, the space \
s
(p,q)
(), 0 j, n and : 1, are dened similarly.
The Sobolev embedding theorem and the Rellich lemma also hold for manifolds.
B. Interpolation Theorems and some Inequalities
There is yet another way to dene the Sobolev spaces \
s
() when : is not an
integer and : 0. Let /
1
and /
2
be two nonnegative integers and /
1
/
2
. On any
domain in 1
N
, we have \
k1
() \
k2
(). The space \
s
() for /
2
< : < /
1
can be dened by interpolation theory. We shall describe the procedure in detail
for the interpolation between \
1
and 1
2
(i.e., when /
1
= 1 and /
2
= 0).
For each \
1
() and n \
1
(),
(n, )
1
= (n, ) +
N

i=1
(1
i
n, 1
i
),
where 1
i
= ,r
i
. Let 1(L) denote the set of all functions n such that the linear
map
(n, )
1
, \
1
()
is continuous in 1
2
(). From the Hahn-Banach theorem and the Riesz representa-
tion theorem, there exists Ln 1
2
() such that
(2.1) (n, )
1
= (Ln, ), \
1
().
If n C

0
(), then n 1(L) and Ln = (+ 1)n. It is easy to see that L is a
densely dened, unbounded self-adjoint operator and L is strictly positive since
(Ln, n) = |n|
2
1
|n|
2
.
Using the spectral theory of positive self-adjoint operators (see e.g. Riesz-Nagy
[RiNa 1]), we can dene L

of L for 1. Let
= L
1/2
.
Then is self-adjoint and positive in 1
2
() with domain \
1
. From (2.1), we have
(n, )
1
= (n, ), for every n, \
1
().
B. Interpolation Theorems and some Inequalities 341
Denition B.1. Let \

() be the interpolation space between the spaces \


1
()
and 1
2
() dened by
\

() [\
1
(), 1
2
()]

= Dom(
1
), 0 1,
with norm
|n| +|
1
n| = the norm of the graph of
1
,
where Dom(
1
) denotes the domain of
1
.
From the denition, we have the following interpolation inequality:
(2.2) |
1
n| |n|
1
|n|

Thus
(2.3) |n|

C|n|
1
1
|n|

.
The general case for arbitrary integers /
1
and /
2
can be done similarly. Thus, this
gives another denition for the Sobolev spaces \
s
() when : is not an integer. If
/ is bounded Lipschitz, this space is the same Sobolev space as the one introduced
in Appendix A (see [LiMa 1] for details for the equivalence of these spaces). For a
bounded Lipschitz domain, we can use any of the denitions for \
s
(), : 0.
The following interpolation inequality holds for general Sobolev spaces:
Theorem B.2 (Interpolation Inequality). Let be a bounded domain in 1
N
with Lipschitz boundary. For any c 0, ) \
s1
(), :
1
: :
2
0, we have the
following inequality:
(2.4) | ) |
2
s
c | ) |
2
s1
+ C

| ) |
2
s2
,
where C

is independent of ).
Theorem B.3 (Interpolation Theorem). Let T be a bounded linear operator
from \
si
() into \
ti
(), i = 1, 2, and
:
1
:
2

1
2
, t
1
t
2

1
2
,
then T is bounded from [\
s1
(), \
s2
()]

into [\
t1
(), \
t2
()]

, 0 1.
We warn our reader of the danger of interpolation of spaces if the assumption :
i

1,2 and t
i
1,2 is dropped! (See [LiMa 1].) Next we discuss the interpolation
between 1
p
spaces and some applications.
Denition B.4. Let (A, j) and (Y, ) be two measure spaces and let T be a linear
operator from a linear subspace of measurable functions on (A, j) into measurable
functions dened on (Y, ). T is called an operator of type (j, ) if there exists a
constant ` 0 such that
(2.5) | T) |
L
q ` | ) |
L
p
for all ) 1
p
(A).
The least ` for which inequality (2.5) holds is called the (j, )-norm of T. If )
is a measurable function on (A, j), we dene its distribution function
f
: (0, )
[0, ] by

f
() = j(r [ [)(r)[ ).
342 Appendix
Denition B.5. Let (A, j) and (Y, ) be two measure spaces and let T be a linear
operator from a linear subspace of measurable functions on (A, j) into measurable
functions dened on (Y, ). T is a linear operator of weak type (j, ), 1 j
and 1 < , if there exists a constant / such that
(:)
_
/ | ) |
L
p
:
_
q
for every ) 1
p
(A),
where is the distribution function of T).
We have the following interpolation theorems:
Theorem B.6 (Riesz-Thorin). Let (A, j) and (Y, ) be two measure spaces and
j
0
, j
1
,
0
,
1
be numbers in [1, ]. If T is of type (j
i
,
i
) with (j
i
,
i
)-norm `
i
,
i = 0, 1, then T is of type (j
t
,
t
) and
(2.6) | T) |
L
q
t `
1t
0
`
t
1
| ) |
L
p
t ,
provided
1
j
t
=
1 t
j
0
+
t
j
1
and
1

t
=
1 t

0
+
t

1
with 0 < t < 1.
For proof of this fact, see Theorem 1.3 in Chapter 5 in Stein-Weiss [StWe 1].
Theorem B.7 (Marcinkiewicz). Let (A, j) and (Y, ) be two measure spaces
and j
0
, j
1
,
0
,
1
be numbers such that 1 j
i

i
for i = 0, 1 and
0
,=
1
.
If T is of weak type (j
i
,
i
), i = 0, 1, then T is of type (j
t
,
t
) provided
1
j
t
=
1 t
j
0
+
t
j
1
and
1

t
=
1 t

0
+
t

1
with 0 < t < 1.
For a proof of this theorem, see Appendix B in Stein [Ste 3].
Theorem B.8 (Hardys Inequality). If ) 1
p
(0, ), 1 < j and
T)(r) =
1
r
_
x
0
)(t)dt, r 0,
then
|T)|
L
p
j
j 1
|)|
L
p.
Proof. We use a change of variables and Minkowskis inequality for integrals,
|T)(r)|
L
p =
_
_
_
_
_
1
0
)(tr)dt
_
_
_
_
p

_
1
0
|)(tr)|
p
dt
=
_
1
0
|)|
p
1
t
1
p
dt =
j
j 1
|)|
L
p.
B. Interpolation Theorems and some Inequalities 343
Theorem B.9. Let
T)(r) =
_

0
1(r, j))(j)dj, r 0,
where 1(r, j) is homogeneous of degree 1, that is, 1(r, j) =
1
1(r, j), for
0. If for each 1 j ,
_
[ 1(1, j) [ j
1/p
dj =
p
< ,
then
|T)|
L
p
p
|)|
L
p, for every ) 1
p
(0, ).
In particular, the Hilbert integral dened by
T)(r) =
_

0
)(j)
r +j
dj, r 0,
is a bounded operator of type (j, j) for each 1 < j < .
Proof. Since
T)(r) =
_

0
1(1, j))(rj)dj,
using Minkowskis inequality for integrals, we get
|T)|
L
p
__
[ 1(1, j) [ j
1/p
dj
_
|)|
L
p =
p
|)|
L
p.
The Hilbert integral is of type (j, j) since, for 1 < j < , using contour integration,
we have
_
j
1/p
1 +j
dj =

sin(,j)
.
Theorem B.10. Let (A, j) and (Y, ) be two measure spaces and let 1(r, j) be a
measurable function on A Y such that
_
X
[1(r, j)[ dj C, for a.e. j,
and _
Y
[1(r, j)[ d C, for a.e. r,
where C 0 is a constant. Then, for 1 j , the operator T dened by
T)(r) =
_
Y
1(r, j))(j) d
is a bounded linear operator from 1
p
(Y, d) into 1
p
(A, dj) with
| T) |
L
p
(X)
C | ) |
L
p
(Y )
.
For a proof of Theorem B.10, we refer the reader to Theorem 6.18 in Folland [Fol
3].
344 Appendix
Theorem B.11. Let (A, j) and (Y, ) be two measure spaces and 1 < < . Let
1(r, j) be a measurable function on A Y such that
j Y [ 1(r, j) :
_
C
:
_
q
, for a.e. r A,
and
j r A [ 1(r, j) :
_
C
:
_
q
, for a.e. j Y,
where C 0 is a constant. Then the operator T dened by
T)(r) =
_
Y
1(r, j))(j) d
is a bounded linear operator from 1
p
(Y ) into 1
r
(A) provided
1 < j < : < and
1
j
+
1


1
:
= 1.
T is bounded from 1
1
(Y ) to 1
q
(A) for any c 0.
The proof of this theorem is based on the Marcinkiewicz Interpolation Theorem
B.7. We refer the reader to Theorem 15.3 in Folland-Stein [FoSt 1] or Theorem 6.35
in Folland [Fol 3].
C. Hardy-Littlewood Lemma and its Variations
We rst prove the Hardy-Littlewood lemma for bounded Lipschitz domains.
Theorem C.1 (Hardy-Littlewood Lemma). Let be a bounded Lipschitz do-
main in 1
N
and let (r) denote the distance function from r to the boundary of .
If n is a C
1
function in and there exists an 0 < < 1 and C 0 such that
(3.1) [ n(r) [ C (r)
1+
for every r ,
then n C

(), i.e., there exists some constant C


1
such that
[ n(r) n(j) [ C
1
[ r j [

for r, j .
Proof. Since n is C
1
in the interior of , we only need to prove the assertion when
r and j are near the boundary. Using a partition of unity, we can assume that n is
supported in l , where l is a neighborhood of a boundary point r
0
/. After
a linear change of coordinates, we may assume r
0
= 0 and for some 0,
l = r = (r
t
, r
N
) [ r
N
(r
t
), [ r
t
[< , [ r
N
[< ,
where (0) = 0 and is some Lipschitz function with Lipschitz constant `. The
distance function (r) is comparable to r
N
(r
t
), i.e., there exists a constant
C 0 such that
(3.2)
1
C
(r) r
N
(r
t
) C(r) for r .
C. Hardy-Littlewood Lemma and its Variations 345
We set r
t
= r
t
+ (1 )j
t
and r
N
= r
N
+ (1 )j
N
. Let d = [r j[. If
r = (r
t
, r
N
), j = (j
t
, j
N
) , then the line segment 1 dened by (r
t
, r
N
+`d)+
(1 )(j
t
, j
N
+`d) = ( r
t
, r
N
+`d), 0 1, lies in since
(r
N
+`d) + (1 )(j
N
+`d)
`d +(r
t
) + (1 )(j
t
)
`d +((r
t
) ( r
t
)) + (1 )((j
t
) ( r
t
)) +( r
t
)
( r
t
).
Since n is C
1
in , using the mean value theorem, there exists some ( r
t
, r
N
+
`d) 1 such that
[ n(r
t
, r
N
+`d) n(j
t
, j
N
+`d) [ [ n( r
t
, r
N
+`d) [ d.
From (3.1) and (3.2), it follows that
[ n(r
t
, r
N
+`d) n(j
t
, j
N
+`d) [ C( r
t
, r
N
+`d)
1+
d


C((` + 1)d)
1+
d C
M
d

.
Also we have
[ n(r) n(r
t
, r
N
+`d) [
=

_
Md
0
n(r
t
, r
N
+t)
t
dt

C
_
Md
0
(r
t
, r
N
+t)
1+
dt C
_
Md
0
(r
N
+t (r
t
))
1+
dt
C
_
Md
0
t
1+
dt C(`d)

.
Thus for any r, j ,
[ n(r) n(j) [ [ (n(r) n(r
t
, r
N
+`d) [ + [ n(j
t
, n
N
+`d) n(j) [
+ [ n(r
t
, r
N
+`d) n(j
t
, j
N
+`d) [
C
M
d

.
This proves the theorem.
The following is a variation of the Hardy-Littlewood lemma for Sobolev spaces.
Theorem C.2. Let be a bounded Lipschitz domain in 1
N
and let (r) be the
distance function from r in to the boundary /. If n 1
2
()\
1
loc
() and there
exists an 0 < < 1 such that
(3.3)
_

(r)
22
[ n [
2
d\ < ,
346 Appendix
then n \

(). Furthermore, there exists a constant C, depending only on ,


such that
| n |
2
()
C
__

(r)
22
[ n [
2
d\ +
_

[n[
2
d\
_
.
Proof. For 0 < < 1, \

() = [\
1
(), 1
2
()]
1
. The interpolation norm of a
function n in \

() (see Lions-Magenes [LiMa 1]) is comparable to the inmum


over all functions
) : [0, ) 1
2
() +\
1
() with )(0) = n
of the norm 1
f
where 1
f
is dened to be
(3.4) 1
f
=
__

0
|t
1
)(t)|
2
W
1
()
t
1
dt
_1
2
+
__

0
|t
1
)
t
(t)|
2
L
2
()
t
1
dt
_1
2
.
From (3.3), we have n \
1
(
t
) for any
t
. Thus we only need to estimate
n in a small neighborhood of the boundary. Using a partition of unity and a change
of coordinates as in Theorem C.1, we can assume l = r
N
(r
t
). Let
C

0
(, ) such that 0 1, 1 when [ t [< ,2. We dene
)(t) = n(r
t
, r
N
+t)(t).
Then )(0) = n(r) and )(t) \
1
() for t 0. To compute the norm dened by
(3.4), we have
(3.5)
[1
f
[
2
C
__

0
_
U
[ n(r
t
, r
N
+t) [
2
dr t
12
dt
+
_

0
_
U
t
12
[ n(r
t
, r
N
+t) [
2
drdt
_
.
Since 1 2 1, the rst integral on the right-hand side of (3.4) is bounded by
|n|
2
L
2
()
. To estimate the second integral on the right-hand side of (3.5), we rst
note that for r l, using (3.2), there exists C
1
0,
(r
t
, r
N
+t) C
1
(r
N
+t (r
t
))
C
1
t.
Thus, after changing variables and the order of integration, we have
_

0
_
U
t
12
[n(r
t
, r
N
+t)[
2
drdt

_
U
_
C(x)
0
t
12
[n(r)[
2
dtdr
C
_
U
(r)
22
[n(r)[
2
dr
< .
This implies that 1
f
< and n \

(). Theorem C.2 is proved.


C. Hardy-Littlewood Lemma and its Variations 347
Theorem C.3. Let be a bounded domain in 1
N
with C

boundary and let : be


a positive integer. If n \
s
0
(), then we have

s+]]
1

n 1
2
(), for every with [[ :,
where is the distance function to the boundary, is a multiindex and 1

is dened
as in Appendix A.
Proof. If ) C

0
(0, ), using Taylors theorem, we have
)(r) =
1
(: 1)!
_
x
0
)
(s)
(t)(r t)
s1
dt.
Applying Hardys inequality (Theorem B.8 in the Appendix), we see that
_
_
_
_
)(r)
r
s
_
_
_
_
L
2

_
_
_
_
1
(: 1)! r
_
x
0
[)
(s)
(t)[dt
_
_
_
_
L
2

2
(: 1)!
|)
(s)
(t)|
L
2.
Using localization and a partition of unity, we can assume that n is supported in
a compact set in the upper half space r = (r
t
, r
N
) [ r
N
0. Applying the
argument to the Taylor expansion in the r
N
variable, we have for any n C

0
(),
|
s+]]
1

n|
L
2
()
C|n|
W
s
0
()
.
The theorem follows by approximating n \
s
0
() by functions in C

0
().
Theorem C.4. Let be a bounded domain in 1
N
with C

boundary. Let : be
any positive number such that : ,= n 1,2 for any n N. If n 1
2
(, |oc) and
(3.6)
_

2s
[n[
2
d\ < ,
where is the distance function to the boundary, then n \
s
().
When : = n 1,2 for some positive integer n, if we assume in addition that n
is harmonic, the same statement also holds.
Proof. We rst assume that : is a positive integer. For any \
s
0
(), we have
from Theorem C.3,
[(n, )[ |
s
n||
s
|
C
s
|
s
n|||
W
s
0
.
Thus, n \
s
() from denition.
For other : when : ,= n 1,2, we use interpolation between \
s
(). For :
2

:
1
0, :
1
, :
2
integers, if (1 ):
1
+:
2
,= n 1,2, then
(3.7) [\
s1
(), \
s2
()]

= \
(1)s1s2
().
348 Appendix
When (1 ):
1
+:
2
= n1,2, (3.7) no longer holds (see Lions-Magenes [LiMa
1]) and we restrict ourselves to harmonic functions.
We rst prove for : = 1,2. Using a partition of unity, we may assume that is
star-shaped and 0 . Dene
(r) =
_
1
0
1
:
n(:r)d:.
Then is harmonic and
r, (r)) =
N

i=1
_
1
0
r
i
n
r
i
(:r)d: =
_
1
0

:
n(:r)d: = n(r) n(0).
Without loss of generality, we may assume that n(0) = 0. We have expressed n as
a linear combination of the derivatives of some harmonic function and, from our
assumption,
(3.8)
_

(r)[r, )[
2
d\ =
_

(r)[n[
2
d\ < ,
where C is some positive constant. We claim that
(3.9)
_

(r)[[
2
d\ C
__

(r)[r, )[
2
d\ +
_

(.)[(r)[
2
d\
_
.
To prove (3.9), we apply the Rellich identity to the harmonic function on the
boundary /

, where

= r [ (r) for small 0. We have


(3.10)
_
b
_
[[
2
r, n) 2r, )
n
n
( 2)

n
_
do = 0,
where n is the outward normal on /

and do is the surface element on /

. Identity
(3.10) follows from the equality
N

i=1

r
i
_
[[
2
r
i
2

r
i
r, ) ( 2)

r
i
_
= 2r, ) ( 2) = 0
and Stokes theorem. If is suciently small, we have r, n) C
0
0 for some
C
0
0 uniformly on /

, it follows from (3.10) that


(3.11)
C
0
_
b
[[
2
do
_
b

2r, )

n
+ ( 2)

do
c
_
b

2
do +C

_
_
b
[[
2
do + [r, )[
2
do
_
,
D. Friedrichs Lemma 349
where c 0. If c is suciently small, the rst term on the right-hand side of (3.11)
can be absorbed by the left-hand side and we obtain
(3.12)
_
b
[[
2
C
_
_
b
[[
2
do +[r, )[
2
do
_
.
Multiplying (3.12) by and integrating over , (3.9) is proved. Using (3.8) and
(3.9), we get
(3.13)
_

(r)[[
2
d\ C
_

(r)[n[
2
d\ < .
It follows from Theorem C.2 that \
1
2
(). Since for any rst order derivative
1 with constant coecients, we have
(3.14) 1 : H\
1
2
() H\

1
2
(),
where H\
s
() = \
s
() n C

() [ n = 0. This implies that n


\
1/2
(). The cases for other integers can be proved similarly and this completes
the proof of Theorem C.4.
We remark that (3.14) does not hold without restricting to the subspace of har-
monic functions (see [LiMa 1]). The technique used in the proof of Theorem C.2
involves real interpolation, while the proof of (3.14) uses complex interpolation. We
refer the reader to Jerison-Kenig [JeKe 1] and Kenig [Ken 3] for more discussion on
these matters.
D. Friedrichs Lemma
Let C

0
(1
N
) be a function with support in the unit ball such that 0
and
(4.1)
_
d\ = 1.
We dene

(r) =
N
(r,) for 0. Extending ) to be 0 outside 1, we dene
for 0 and r 1
N
,
)

(r) = )

(r) =
_
)(r
t
)

(r r
t
)d\ (r
t
)
=
_
)(r r
t
)(r
t
)d\ (r
t
).
In the rst integral dening )

we can dierentiate under the integral sign to show


that )

is C

(1
N
). From Youngs inequality for convolution, we have
(4.2) | )

| | ) | .
Since

is an approximation of the identity, we have )

) uniformly if )
C

0
(1
N
). Since C

0
(1
N
) is a dense subset of 1
2
(1
N
), this implies that
)

) in 1
2
(1
N
) for every ) 1
2
(1
N
).
A very useful fact concerning approximating solutions of a rst order dierential
operator by regularization using convolution is given by the following lemma (see
Friedrichs [Fri 1] or Hormander [Hor 2]):
350 Appendix
Lemma D.1 (Friedrichs Lemma). If 1
2
(1
N
) with compact support and o
is a C
1
function in a neighborhood of the support of , it follows that
(4.3) o1
i
(

) (o1
i
)

0 in 1
2
(1
N
) as c 0,
where 1
i
= ,r
i
and o1
i
is dened in the sense of distribution.
Corollary D.2. Let
1 =
N

i=1
o
i
1
i
+o
0
be a rst order dierential operator with variable coecients where o
i
C
1
(1
N
)
and o
0
C(1
N
). If 1
2
(1
N
) with compact support and 1 = ) 1
2
(1
N
) where
1 is dened in the distribution sense, the convolution

is in C

0
(1
N
)
and
(4.4)

, 1

) in 1
2
(1
N
) as c 0.
Proof of Friedrichs lemma. First note that if C

0
(1
N
), we have from the
discussion above,
1
i
(

) = (1
i
)

1
i
, (o1
i
)

o1
i
,
with uniform convergence. We claim that
(4.5) | o1
i
(

) (o1
i
)

| C | |, 1
2
(1
N
),
where C is some positive constant independent of c and . Since C

0
(1
N
) is dense
in 1
2
(1
N
), (4.3) will be proved if one can prove (4.5). To see this, we approximate
by a sequence
j
C

0
(1
N
) in 1
2
(1
N
) and observe that if (4.5) holds, we have
| o1
i
(

) (o1
i
)

|
C(|
j
| + | o1
i
(
j

) (o1
i

j
)

|).
Thus, it remains to prove (4.5). Without loss of generality, we may assume that
o C
1
0
(1
N
) since has compact support. We have for C

0
(1
N
),
o1
i
(

) (o1
i
)

= o(r)1
i
_
(r j)

(j)dj
_
o(r j)

r
i
(r j)

(j)dj
=
_
(o(r) o(r j))

r
i
(r j)

(j)dj
=
_
(o(r) o(r j))

j
i
(r j)

(j)dj
=
_
(o(r) o(r j)) (r j)

j
i

(j)dj

_ _

j
i
o(r j)
_
(r j)

(j)dj.
D. Friedrichs Lemma 351
Let ` be the Lipschitz constant for o such that [o(r) o(r j)[ `[j[ for all
r, j. We obtain
[o1
i
(

) (o1
i
)

[ `
_
[(r j)[ (

(j) +[j1
i

(j)[) dj.
Using Youngs inequality for convolution, we have
| o1
i
(

) (o1
i
)

| ` | |
_
(

(j) +[j1
i

(j)[) dj
= `(1 +:
i
) | |,
where
:
i
=
_
[j1
i

(j)[dj =
_
[j(1
i
)(j)[dj.
This proves (4.5) when C

0
(1
N
). Since C

0
(1
N
) is dense in 1
2
(1
N
), we have
proved (4.5) and the lemma.
Proof of the Corollary. Since o
0
1
2
(1
N
), we have
lim
0
o
0
(

) = lim
0
(o
0

) = o
0
in 1
2
(1
N
).
Using Friedrichs lemma, we have
1

= 1

0 in 1
2
(1
N
) as c 0.
The corollary follows easily since )

) in 1
2
(1
N
).
352
BIBLIOGRAPHY
[ADN 1] Agmon, S., Douglis, A., and Nirenberg, L., Estimates near the boundary for solutions
of elliptic partial dierential equations satisfying general boundary conditions I, comm.
Pure Appl. Math., 12 (1959), 623727; II 17 (1964), 35-92.
[AlWe 1] Alexander, H., and Wermer, J., Several Complex Variables and Banach Algebras, Third
Edition, Springer Verlag, New York, Berlin, Heidelberg, 1998.
[AiHe 1] Airapetyan, R. A., and Henkin, G. M., Integral representations of dierential forms
on Cauchy-Riemann manifolds and the theory of CR-functions, Uspekhi Mat. Nauk,
39 (1984), 39106.
[Aka 1] Akahori, T., A New Approach to the Local Embedding Theorem of CR-Structure for
n 4, Memoirs of the A.M.S., No.366, Providence, Rohde Island, 1987.
[AnGr 1] Andreotti, A., and Grauert, H., Theor`emes de nitude pour la colomologie des espaces
complexes, Bull. Soc. Math. France, 90 (1962), 193-259.
[AnHi 1] Andreotti, A., and Hill, C. D., Convexity and the H. Lewy problem. Part I: Reduction
to the vanishing theorems, Ann. Scuola Norm. Sup. Pisa, 26 (1972), 325363.
[AnVe 1] Andreotti, A., and Vesentini, E., Carleman estimates for the Laplace-Beltrami equation
on complex manifolds, Inst. Hautes Etudes Sci. Publ. Math., 25 (1965), 81130.
[Ash 1] Ash, M. E., The basic estimate of the

-Neumann problem in the non-Kahlerian case,
Amer. J. Math., 86 (1964), 247254.
[BER 1] Baouendi, M. S., Ebenfelt, P., and Rothschild, L. P., Real Submanifolds in Complex
Space and their Mappings, Math. Series 47, Princeton University Press, Princeton,
New Jersey, 1999.
[BaTr 1] Baouendi, M. S., and Treves, F., A property of the functions and distributions anni-
hilated by a locally integrable system ofcomplex vector elds, Ann. Math., 113 (1981),
387421.
[Bar 1] Barrett, D., Regularity of the Bergman projection on domains with transverse symme-
tries, Math. Ann., 258 (1982), 441446.
[Bar 2] Barrett, D., Irregularity of the Bergman projection on a smooth bounded domain in
C
2
, Ann. Math., 119 (1984), 431436.
[Bar 3] Barrett, D., Behavior of the Bergman projection on the Diederich-Fornaess worm,
Acta Math., 168 (1992), 110.
[BFG 1] Beals, M., Feerman, C., and Grossman, R., Strictly pseudoconvex domains in C
n
,
Bull. Amer. Math. Soc. 8 (1983), 125322.
[BeGr 1] Beals, R., and Greiner, P. C., Calculus on Heisenberg Manifolds, Ann. Math. Studies
119, Princeton Univerisity Press, Princeton, New Jersey, 1988.
[BGS 1] Beals, R., Greiner, P. C., and Stanton, N. K., L
p
and Lipschitz estimates for the
-equation and the -Neumann problem, Math. Ann., 277 (1987), 185196.
[BeSt 1] Behnke, H., and Stein, K., Konvergente Folgen von Regularitatsbereichen und die
Meromorphiekonvexitat, Math. Ann., 116 (1938), 204216.
[Bel 1] Bell. S., Non-vanishing of the Bergman kernel function at boundary points of certain
domains in C
n
, Math. Ann., 244 (1979), 6974.
[Bel 2] Bell. S., Biholomorphic mappings and the -problem, Ann. Math., 114 (1981), 103
113.
[Bel 3] Bell. S., Dierentiability of the Bergman kernel and pseudo-local estimates, Math.
Zeit., 192 (1986), 467472.
[Bel 4] Bell. S., Mapping problems in complex analysis and the -problem, Bull. Amer. Math.
Soc., 22 (1990), 233259.
[BeBo 1] Bell. S., and Boas, H. P., Regularity of the Bergman projection in weakly pseudoconvex
domains, Math. Ann., 257 (1981), 2330.
Bibliography 353
[BeLi 1] Bell. S., and Ligocka, E., A simplication and extension of Feermans theorem on
biholomorphic mappings, Invent. Math., 57 (1980), 283289.
[Bern 1] Berndtsson, B.,
b
and Carleson type inequalities, Complex Analysis II, Lecture Notes
in Math., 1276, Springer-Verlag, (1987), 4254.
[Ber 1] Bers, L., Riemann Surfaces, New York University, New York, 1957-1958.
[BJS 1] Bers, L., John, F., and Schechter, M., Partial Dierential Equations, Interscience, New
York, 1964.
[Boa 1] Boas, H. P., Regularity of the Szego projection in weakly pseudoconvex domains, Indi-
ana Univ. Math. J., 34 (1985), 217223.
[Boa 2] Boas, H. P., Sobolev space projections in strictly pseudoconvex domains, Trans. Amer.
Math. Soc., 288 (1985), 227-240.
[Boa 3] Boas, H. P., Extension of Kerzmans theorem on dierentiability of the Bergman kernel
function, Indiana Univ. Math. J., 36 (1987), 495499.
[Boa 4] Boas, H. P., The Szego projection: Sobolev estimates in regular domains, Trans. Amer.
Math. Soc., 300 (1987), 109132.
[BCS 1] Boas, H. P., Chen, S.-C., and Straube, E. J., Exact regularity of the Bergman and Szego
projections on domains with partially transverse symmetries, Manuscripta Mathemat-
ica, 62 (1988), 467475.
[BoSh 1] Boas, H. P., and Shaw, M.-C., Sobolev estimates for the Lewy operator on weakly
pseudo-convex boundaries, Math. Ann., 274 (1986), 221231.
[BoSt 1] Boas, H. P., and Straube, E. J., Complete Hartogs domains in C
2
have regular Bergman
and Szego projections, Math. Zeit., 201 (1989), 441454.
[BoSt 2] Boas, H. P., and Straube, E. J., Equivalence of regularity for the Bergman projection
and the -Neumann operator, Manuscripta Math., 67 (1990), 2533.
[BoSt 3] Boas, H. P., and Straube, E. J., Sobolev estimates for the -Neumann operator on
domains in C
n
admitting a dening function that is plurisubharmonic on the boundary,
Math. Zeit., 206 (1991), 8188.
[BoSt 4] Boas, H. P., and Straube, E. J., Sobolev estimates for the complex Green operator on a
class of weakly pseudoconvex boundaries, Commun. in Partial Dierential Equations,
16 (1991), 15731582.
[BoSt 5] Boas, H. P., and Straube, E. J., De Rham cohomology of manifolds containing the
points of innite type, and Sobolev estimates for the -Neumann problem, J. of Geo-
metric Analysis, 3 (1993), 225235.
[BoSt 6] Boas, H. P., and Straube, E. J., Global regularity of the -Neumann problem: A survey
of the L
2
-Sobolev theory, Several Complex Variables, Mathematical Sciences Research
Institute Publications (M. Schneider and Y-T Siu, eds.), vol. 37, Cambridge University
Press, 1999, pp. 79111.
[Boc 1] Bochner, S., Analytic and meromorphic continuation by means of Greens formula,
Ann. Math., 44 (1943), 652673.
[Bog 1] Boggess, A., Kernels for the tangential Cauchy-Riemann equations, Trans. of Amer.
Math. Soc., 262 (1980), 149.
[Bog 2] Boggess, A., CR Manifolds and the Tangential Cauchy-Riemann Complex, CRC Press,
1991.
[BoSh 1] Boggess, A., and Shaw, M.-C., A kernel approach to local solvability of the tangential
Cauchy-Riemann equations, Trans. Amer. Math. Soc., 289 (1985), 643659.
[BoCh 1] Bonami, A., and Charpentier, P., Une estimation Sobolev
1
2
pour le projecteur de
Bergman, C. R. Acad. Sci. Paris, 307 (1988), 173176.
[BdM 1] Boutet de Monvel, L., Integration des equations de Cauchy-Riemann induites formell-
es, Seminaire Goulaouic-Lions-Schwartz, Expose No. 9 (1974-1975).
[BdSj 1] Boutet de Monvel, L., and Sjostrand, J., Sur la singularite des noyaux de Bergman et
de Szego, Societe Mathematique de France, Asterisque, 34-35 (1976), 123164.
354 Bibliography
[Bre 1] Bremermann, H. J.,

Uber die

Aquivalenz der pseudokonvexen Gebiete und der Holo-
morphiegebiete im Raum von n komplexen Veranderlichen, Math. Ann., 128 (1954),
6391.
[Bre 2] Bremermann, H. J., Complex convexity, Trans. Amer. Math. Soc., 82 (1956), 1751.
[BCD 1] Bruna, J., Charpentier, P., and Dupain, Y., Zero varieties for the Nevanlinna class in
convex domains of nite type in C
n
, Ann. Math., 147 (1998), 391415.
[BrCa 1] Bruna, J., and del Castillo, J., Holder and L
p
-estimates for the

-equation in some
convex domains with real-analytic boundary, Math. Ann., 269 (1984), 527539.
[Bur 1] Burns, D., Jr., Global Behavior of Some Tangential Cauchy-Riemann Equations, Par-
tial Dierential Equations and Geometry(Proc. Conf., Park City, Utah, 1977), Dekker,
New York, (1979), 5156.
[CaZy 1] Calderon, A. P., and Zygmund, A., Singular integral operators and dierential equa-
tions, Amer. J. Math. 79 (1957), 901-921.
[CaTh 1] Cartan, H., and Thullen, P., Zur Theorie der Singularitaten der Funktionen mehrerer
Veranderlichen: Regularitats- und Konvergenzbereiche, Math. Ann., 106 (1932), 617
647.
[Cat 1] Catlin, D., Necessary conditions for subellipticity of the -Neumann problem, Ann.
Math., 117 (1983), 147171.
[Cat 2] Catlin, D., Boundary invariants of pseudoconvex domains, Ann. of Math., 120 (1984),
529586.
[Cat 3] Catlin, D., Global regularity of the -Neumann problem, Proc. Symposia Pure Math.,
Amer. Math. Soc., Providence, Rhode Island, 41 (1984), 3949.
[Cat 4] Catlin, D., Subelliptic estimates for the -Neumann problem on pseudoconvex domains,
Ann. Math., 126 (1987), 131191.
[Cat 5] Catlin, D., Sucient conditions for the extension of CR structures, J. of Geometric
Analysis, 4 (1994), 467538.
[CNS 1] Chang, D.-C., Nagel, A., and Stein, E. M., Estimates for the

-Neumann problem in
pseudoconvex domains of nite type in C
2
, Acta Math., 169 (1992), 153228.
[ChCh 1] Chaumat, J., Chollet, A.M., Noyaux pour resoudre lequation dans des classes ultra-
dierentiables sur des compacts irreguliers de C
n
, Several Complex Variables: Proc.
of the Mittag-Leer Institute 1987/1988, Math. Notes 38, Princeton University Press,
Princeton, New Jersey, (1993), 205226.
[Che 1] Chen, S.-C., Global analytic hypoellipticity of the -Neumann problem on circular
domains, Invent. Math., 92 (1988), 173185.
[Che 2] Chen, S.-C., Global real analyticity of solution to the -Neumann problem on Reinhardt
domains, Indiana Univ. Math. J., 37 (1988), 421430.
[Che 3] Chen, S.-C., Global regularity of the -Neumann problem on circular domains, Math.
Ann., 285 (1989), 112.
[Che 4] Chen, S.-C., Global regularity of the -Neumann problem: a sucient condition, Bull.
Inst. Math., Academia Sinica, 19 (1991), 191205.
[Che 5] Chen, S.-C., Global regularity of the -Neumann problem in dimension two, Proc.
Symposia Pure Math., Amer. Math. Soc., Providence, Rhode Island, 52, III (1991),
5561.
[Che 6] Chen, S.-C., Real analytic regularity of the Bergman and Szego projections on decoupled
domains, Math. Zeit., 213 (1993), 491508.
[Che 7] Chen, S.-C., A counterexample to the dierentiability of the Bergman kernel function,
Proc. Amer. Math. Soc., Providence, Rhode Island, 124 (1996), 18071810.
[CKM 1] Chen, Z., Krantz, S. G., and Ma, D., Optimal L
p
estimates for the

-equation on
complex ellipsoid in C
n
, Manu. Math., 80 (1993), 131-149.
[Cher 1] Chern, S. S., An elementary proof of the existence of isothermal parameters on a
surface, Proc. Amer. Math. Soc., 6 (1955), 771782.
[Cher 2] Chern, S. S., Complex Manifolds without Potential Theory, 2nd ed., Springer-Verlag,
New York, 1979.
Bibliography 355
[Chr 1] Christ, M., Regularity properties of the

b
equation on weakly pseudoconvex CR man-
ifolds of dimension 3, J. Amer. Math. Soc., 1 (1988), 587646.
[Chr 2] Christ, M., Global C

irregularity of the -Neumann problem for worm domains, J.


of the Amer. Math. Soc., 9 (1996), 11711185.
[ChGe 1] Christ, M., and Geller, D., Counterexamples to analytic hypoellipticity for domains of
nite type, Ann. Math., 135 (1992), 551566.
[CoHi 1] Courant, R., and Hilbert, D., Methods of Mathematical Physics, volume II, John Wiley
& Sons, New York, 1989.
[Cum 1] Cumenge, A., Estimees Lipschitz optimales dans les convexes de type ni, C. R. Acad.
Sci. Paris, Ser. I, Math., 325 (1997), 10771080.
[DAn 1] DAngelo, J., Real hypersurfaces, order of contact, and applications, Ann. Math., 115
(1982), 615637.
[DAn 2] DAngelo, J., Several Complex Variables and the Geometry of Real Hypersurfaces,
CRC Press, Boca Raton, 1993.
[Der 1] Derridj, M., Regularite pour dans quelques domaines faiblement pseudo-convexes, J.
Dierential Geometry, 13 (1978), 559576.
[Der 2] Derridj, M., Inegates de Carleman et extension locale des fonctions holomorphes, Ann
Scuola Norm. Sup., Pisa, 15 (1981), 645669.
[Der 3] Derridj, M., Le problem de Cauchy four

et application, Ann Scient.

EC. Norm. Sup.
17 (1984), 439449.
[DeTa 1] Derridj, M., and Tartako, D. S., On the global real analyticity of solutions to the
-Neumann problem, Commun. in Partial Dierential Equations, 1 (1976), 401435.
[DeTa 2] Derridj, M., and Tartako, D. S., Local analyticity for
b
and the -Neumann problem
at certain weakly pseudoconvex points, Commun. in Partial Dierential Equations, 13
(1988), 15211600.
[DeTa 3] Derridj, M., and Tartako, D. S., Local analyticity for the -Neumann problem and

b
-some model domains without maximal estimates, Duke Math. J., 64 (1991), 377
402.
[DFF 1] Diederich, K., Fischer, B., and Fornaess, J. E., Holder estimates on convex domains
of nite type, Math. Zeit., 232 (1999), 4361.
[DiFo 1] Diederich, K., and Fornaess, J. E., Pseudoconvex domains: An example with nontrivial
Nebenh ulle, Math. Ann., 225 (1977), 275292.
[DiFo 2] Diederich, K., and Fornaess, J. E., Pseudoconvex domains: Bounded strictly plurisub-
harmonic exhaustion functions, Invent. Math., 39 (1977), 129141.
[DiFo 3] Diederich, K., and Fornaess, J. E., Pseudoconvex domains with real analytic boundary,
Ann. Math., 107 (1978), 371384.
[DiFo 4] Diederich, K., and Fornaess, J. E., Support functions for convex domains of nite type,
Math. Zeit., 230 (1999), 145164.
[DFW 1] Diederich, K., Fornaess, J. E., and Wiegerinck, J., Sharp Holder estimates for

on
ellipsoids, Manu. Math., 56 (1986), 399417.
[Dol 1] Dolbeault, P., Sur la cohomologie des varietes analytiques complexes, C. R. Acad. Sci.
Paris, 236 (1953), 175177.
[Dol 2] Dolbeault, P., Formes dierentielles et cohomologie sur une variete analytique com-
plexe, I, Ann. Math., 64 (1956), 8130; II, Ann. Math., 65 (1957), 282330.
[Duf 1] Dufresnoy, A., Sur Loperateur d et les fonctions dierentiables au sens de Whitney,
Ann. Inst. Fourier, Grenoble, 29 (1979), 229238.
[Ehr 1] Ehrenpreis, L., Solutions of some problems of division I, Amer. J. Math., 76 (1954),
883903.
[Ehr 2] Ehrenpreis, L., A new proof and an extension of Hartogs theorem, Bull. Amer. Math.
Soc., 67 (1961), 507509.
[Fed 1] Fedii, V. S., Estimates in H
s
norms and hypoellipticity, Dokl. Akad. Nauk SSSR, 193
(1970), 940942.
356 Bibliography
[Fef 1] Feerman, C., The Bergman kernel and biholomorphic mappings of pseudoconvex do-
mains, Invent. Math., 26 (1974), 165.
[FeKo 1] Feerman, C., and Kohn, J. J., Holder estimates on domains of complex dimension
two and on three dimensional CR manifolds, Adv. in Math. 69 (1988), 233303.
[FKM 1] Feerman, C., Kohn, J. J., and Machedon, M., Holder estimates on CR manifolds
with a diagonalizable Levi form, Adv. in Math. 84 (1990), 190.
[FePh 1] Feerman, C., and Phong, D. H., Subelliptic eigenvalue problems,Conf. on Harmonic
Analysis in honor of A. Zygmund, ed. by Beckner et al., Wadsworth Intern. Math.
Series, Wadsworth, Belmont, (1981), 590606.
[FiLi 1] Fischer, W., and Lieb I., Locale Kerne und Beschrankte Losungen f ur den -Operator
auf q-konvexen Gebieten, Math. Ann., 208 (1974), 249265.
[Fol 1] Folland, G. B., A fundamental solution for a subelliptic operator, Bull. Amer. Math.
Soc., 79 (1973), 373376.
[Fol 2] Folland, G. B., Introduction to Partial Dierential Equations, Math. Notes 17, Prince-
ton University Press, Princeton, New Jersey, 1976.
[Fol 3] Folland, G. B., Real Analysis, John Wiley & Sons, New york, 1984.
[FoKo 1] Folland, G. B., and Kohn, J. J., The Neumann Problem for the Cauchy-Riemann
Complex, Ann. Math. Studies 75, Princeton University Press, Princeton, N.J., 1972.
[FoSt 1] Folland, G. B., and Stein, E. M., Estimates for the
b
complex and analysis on the
Heisenberg group, Comm. Pure and Applied Math., 27 (1974), 429522.
[For 1] Fornaess, J. E., An increasing sequence of Stein manifolds whose limit is not Stein,
Math. Ann., 223 (1976), 275277.
[For 2] Fornaess, J. E., Embedding strictly pseudoconvex domains in convex domains, Amer.
J. Math., 98 (1976), 529569.
[For 3] Fornaess, J. E., A counterexample for the Levi problem for branched Riemann domains
over C
n
, Math. Ann., 234 (1978), 275277.
[For 4] Fornaess, J. E., Sup-norm estimates for

in C
2
, Ann. Math., 123 (1986), 335345.
[FoSi 1] Fornaess, J. E., and Sibony, N., On L
p
estimates for , Several Complex Variables and
Complex Geometry, (Santa Cruz, CA, 1989), Proc. Sympos. Pure Math., 52, Part 3,
Amer. Math. Soc., Providence, Rhode Island, (1991), 129163.
[FoSte 1] Fornaess, J. E., and Stensnes, B., Lectures on Counterexamples in Several Complex
Variables, Math. Notes 33, Princeton University Press, Princeton, New Jersey, 1987.
[Fri 1] Friedrichs, K., The identity of weak and strong extensions of dierential operators,
Trans. Amer. Math. Soc., 55 (1944), 132141.
[FuSt 1] Fu, S., and Straube, E. J., Compactness of the -Neumann problem on convex do-
mains., J. Funct. Anal., 159 (1998), 629641.
[Gaf 1] Ganey, M. P., Hilbert space methods in the theory of harmonic integrals, Trans. Amer.
Math. Soc., 78 (1955), 426444.
[GaSp 1] Garabedian P. R., and Spencer, D. C., Complex boundary value problem, Trans. Amer.
Math. Soc., 73 (1952), 223242.
[Gel 1] Geller, D., Analytic Pseudodierential Operators for the Heisenberg Group and Local
Solvability, Math. Notes 37, Princeton University Press, Princeton, N.J., 1990.
[Gin 1] Gindikin, S. G., Analysis in homogeneous domains, Uspekhi Mat. Nauk, 19 (1964),
392; English transl, Russian Math. Surveys 19 (1964), 1-89.
[Gra 1] Grauert, H., On Levis problem and the imbedding of real analytic manifolds, Ann. of
Math., 68 (1958), 460472.
[GrLi 1] Grauert H., and Lieb, I., Das Ramirezsche integral und die Losung der Gleichung

f = im Bereich der beschrinkten Formen, Proc. Conf. Complex Analysis, Rice


Univ. Studies, Houston, Texas, 56 (1970), 2950.
[GKS 1] Greiner, P. C., Kohn, J. J., and Stein, E. M., Necessary and sucient conditions for
solvability of the Lewy equation, Proc. Nat. Acad. Sci., 72 (1975), 32873289.
Bibliography 357
[GrSt 1] Greiner, P. C., and Stein, E. M., Estimates for the -Neuman Problem, Math. Notes
19, Princeton University Press, Princeton, New Jersey, 1977.
[GrHa 1] Griths, P., and Harris, J., Principles of Algebraic Geometry, Wiley and Sons, Inc.,
New York, 1978.
[Gri 1] Grisvard, P., Elliptic Problems in Nonsmooth Domains, Pitman, Boston, 1985.
[GuRo 1] Gunning, R. C., and Rossi, H., Analytic Functions of Several Ccomplex Variables ,
Prentice-hall, Englewood Clis, New Jersey, 1965.
[HaWi 1] Hartman, P., and Wintner, A., On the existence of Riemannian manifolds which cannot
carry non-constant analytic or harmonic functions in the small, Amer. J. Math., 75
(1953), 260-276.
[Har 1] Hartogs, F., Einige Folgerungen aus der Cauchyschen Integralformel bei Funktionen
mehrerer Veranderlichen, M unch. Ber. 36 (1906), 223242.
[HaLa 1] Harvey, R., and Lawson, B., On boundaries of complex analytic varieties,I, Ann. Math.,
102 (1975), 233290.
[HaPo 1] Harvey, R., and Polking, J., Fundamental solutions in complex analysis, I. The Cauchy-
Riemann operator, Duke Math. J., 46 (1979), 253300.
[HaPo 2] Harvey, R., and Polking, J., Fundamental solutions in complex analysis, II. The in-
duced Cauchy-Riemann operator, Duke Math. J., 46 (1979), 301340.
[HaPo 3] Harvey, R., and Polking, J., The -Neumann kernel in the ball in C
n
, Proc. Sympos.
Pure Math., Amer. Math. Soc. Providence, R.I. 41 (1984), 117136.
[HaPo 4] Harvey, R., and Polking, J., The -Neumann solution to the inhomogeneous Cauchy-
Riemann equation in the ball in C
n
, Trans. Amer. Math. Soc., 281 (1984), 587613.
[Hen 1] Henkin, G. M., Integral representation of functions in strictly pseudoconvex domains
and applications to the

-problem, Math. USSR Sb., 7 (1969), 579-616.
[Hen 2] Henkin, G. M., Uniform estimates for solutions to the

-problem in Weil domains
(Russian), Uspehi Mat. Nauk, 26 (1971), 211-212.
[Hen 3] Henkin, G. M., The Lewy equation and analysis on pseudoconvex manifolds, (Russian
Math. Surveys,32(1977),59-130), Uspehi Mat. Nauk, 32 (1977), 57118.
[HeIo 1] Henkin, G. M., and Iordan, A., Compactness of the Neumann operator for hyperconvex
domains with non-smooth B-regular boundary, Math. Ann., 307 (1997), 151168.
[HIK 1] Henkin, G. M., Iordan, A., and Kohn, J. J., Estimations sous-elliptiques pour le
probl`eme

-Neumann dans un domaine strictement pseudoconvexe `a fronti`ere lisse
par morceaux, C. R. Acad. Sci. Paris, 323 (1996), 1722.
[HeLe 1] Henkin, G. M., and Leiterer, J., Theory of Functions on Complex Manifolds, Birkhaus-
er, Boston, Mass., 1984.
[HeLe 2] Henkin, G. M., and Leiterer, J., Andreotti-Grauert Theory by Integral Formulas,
Birkhauser, Boston, Mass., 1984.
[HeRo 1] Henkin, G. M., and Romanov, A. V., Exact Holder estimates for the solutions of the

-
equation, Izv. Akad. Nauk SSSR, (Translated in Math. USSR Izvestija, 5(1971),1180
1192), 35 (1971), 11711183.
[Hor 1] Hormander, L., Dierential equations without solutions, Math. Ann., 140 (1960), 169
173.
[Hor 2] Hormander, L., Weak and strong extensions of dierential operators, Comm. Pure.
Appl. Math., 14 (1961), 371379.
[Hor 3] Hormander, L., L
2
estimates and existence theorems for the operator, Acta Math.,
113 (1965), 89152.
[Hor 4] Hormander, L., Pseudo-dierential operators, Comm. Pure Appl. Math., 18 (1965),
501517.
[Hor 5] Hormander, L., The Frobenius-Nirenberg theorem, Ark. Mat., 5 (1965), 425432.
[Hor 6] Hormander, L., Hypoelliptic second-order dierential equations, Acta Math., 119 (19-
67), 147171.
[Hor 7] Hormander, L., Linear Partial Dierential Operators, 3rd Edition, Springer-Verlag,
BerlinHeidelberg, 1969.
358 Bibliography
[Hor 8] Hormander, L., The Analysis of Linear Partial Dierential Operators III, Pseudo-
Dierential Operators, Spring-Verlag, Berlin Heidelberg New York Tokyo, 1985.
[Hor 9] Hormander, L., An Introduction to Complex Analysis in Several Complex Variables,
Third Edition, Van Nostrand, Princeton, N.J., 1990.
[Hua 1] Hua, L. K., Harmonic Analysis of Functions of Several Complex Variables in the Clas-
sical Domains, Transl. of Math. Monographs 6, American Math. Society, 1963.
[Jac 1] Jacobowitz, H., An Introduction to CR Structures, vol. 32, Amer. Math. Soc., Provi-
dence, Rhode Island, 1990.
[JaTr 1] Jacobowitz, H., and Treves, F., Non-realizable CR structures, Invent. Math., 66 (1982),
231249.
[Jer 1] Jerison, D. S., Boundary regularity in the Dirichlet problem for
b
on CR manifolds,
Comm. Pure Appl. Math., 36 (1983), 143181.
[JeKe 1] Jerison, D., and Kenig, C. E., The inhomogeneous Dirichlet Problem in Lipschitz
Domains, Journal of Functional Anal., 130 (1995), 161219.
[Joh 1] John, F., Partial Dierential Equations, Springer-Verlag, 1982.
[Ken 1] Kenig, C., Elliptic boundary value problems on Lipschitz domains, Beijing Lectures in
Harmonic Analysis, Ann. Math. Studies, Princeton University Press, Princeton, New
Jersey, 112 (1986), 131-184.
[Ker 1] Kerzman, N., Holder and L
p
estimates for solutions of

u = f in strongly pseudoconvex
domains, Comm. Pure Appl. Math., 24 (1971), 301379.
[Ker 2] Kerzman, N., The Bergman kernel function. Dierentiability at the boundary, Math.
Ann., 195 (1972), 149158.
[KeSt 1] Kerzman, N., and Stein, E. M., The Szego kernel in terms of Cauchy-Fantappi`e kernels,
Duke Math. J., 45 (1978), 197224.
[Kis 1] Kiselman, C. O., A study of the Bergman projection in certain Hartogs domains, Proc.
Symposia Pure Math., Amer. Math. Soc., Providence, Rhode Island, 52, III (1991),
219231.
[Koh 1] Kohn, J. J., Harmonic integrals on strongly pseudoconvex manifolds, I, Ann. Math.,
78 (1963), 112148; II, Ann. Math., 79 (1964), 450472.
[Koh 2] Kohn, J. J., Boundaries of complex manifolds, Proc. Conf. Complex Manifolds (Min-
neapolis,1964), Springer-Verlag, New York, (1965), 8194.
[Koh 3] Kohn, J. J., Pseudo-dierential operators and non-elliptic problems, Pseudo-Dierenti-
al Operators (C.I.M.E., Stresa, 1968), Edizioni Cremonese, Rome, (1969), 157165.
[Koh 4] Kohn, J. J., Boundary behavior of on weakly pseudo-convex manifolds of dimension
two, J. Dierential Geom., 6 (1972), 523542.
[Koh 5] Kohn, J. J., Pseudo-dierential operators and hypoellipticity, Proc. Symposia Pure
Math., Amer. Math. Soc., Providence, Rhode Island, 23 (1973), 6169.
[Koh 6] Kohn, J. J., Global regularity for on weakly pseudoconvex manifolds, Trans. Amer.
Math. Soc., 181 (1973), 273292.
[Koh 7] Kohn, J. J., Methods of partial dierential equations in complex analysis, Proc. Sym-
posia Pure Math., Amer. Math. Soc., Providence, Rhode Island, 30, Part I (1977),
215237.
[Koh 8] Kohn, J. J., Subellipticity of the -Neumann problem on pseudoconvex domains: suf-
cient conditions, Acta Math., 142 (1979), 79122.
[Koh 9] Kohn, J. J., Microlocalization of CR structures, Several Complex Variables, (Proc. of
the 1981 Hangzhou conf.), Birkhauser, Boston, (1984), 2936.
[Koh 10] Kohn, J. J., Estimates for

b
on pseudoconvex CR manifolds, Proc. sympos. Pure
Math., Amer. Math. Soc., Providence, Rhode Island, 43 (1985), 207217.
[Koh 11] Kohn, J. J., The range of the tangential Cauchy-Riemann operator, Duke Math. J.,
53 (1986), 525545.
[Koh 12] Kohn, J. J., Hypoellipticity of some degenerate subelliptic operators, J. Funct. Anal.,
159 (1998), 203216.
[KoNi 1] Kohn, J. J., and Nirenberg, L., Noncoercive boundary value problems, Comm. Pure
Appl. Math., 18 (1965), 443492.
Bibliography 359
[KoNi 2] Kohn, J. J., and Nirenberg, L., An algebra of pseudo-dierential operators, Comm.
Pure Appl. Math., 18 (1965), 269305.
[KoNi 3] Kohn, J. J., and. Nirenberg, L., A pseudo-convex domain not admitting a holomorphic
support function, Math. Ann., 201 (1973), 265268.
[KoRo 1] Kohn, J. J., and Rossi, H., On the extension of holomorphic functions from the bound-
ary of a complex manifold, Ann. Math., 81 (1965), 451472.
[Kom 1] Komatsu, G., Global analytic-hypoellipticity of the -Neumann problem, Tohoku Math.
J.,II. Ser., 28 (1976), 145156.
[Kop 1] Koppelman, W., The Cauchy integral for dierential forms, Bull. Amer. Math. Soc.,
73 (1967), 554556.
[Kra 1] Krantz, S. G., Optimal Lipschitz and L
p
regularity for the equation

u = f on strongly
pseudoconvex domains, Math. Ann., 219 (1976), 233260.
[Kra 2] Krantz, S., Function Theory of Several Complex Variables, 2nd edition, Wadsworth,
Belmont, California, 1992.
[Kra 3] Krantz, S. G., Partial Dierential Equations and Complex Analysis, CRC press, Boca
Raton, 1992.
[Kur 1] Kuranishi, M., Strongly pseudoconvex CR structures over small balls, Part I, An a
priori estimate, Ann. Math., 115 (1982), 451500.
[Kur 2] Kuranishi, M., Strongly pseudoconvex CR structures over small balls, Part II, A reg-
ularity theorem, Ann. Math., 116 (1982), 164.
[Kur 3] Kuranishi, M., Strongly pseudoconvex CR structures over small balls, Part III, An
embedding theorem, Ann. Math., 116 (1982), 249330.
[KuSt 1] Kusuoka, S., and Strook, D., Applications of the Malliavin calculus, Part II, J. Fac.
Sci. Univ. Tokyo, Sect. IA, Math., 32 (1985), 176.
[Kyt 1] Kytmanov A. M., The Bochner-Martinelli Integral and its Applications, Birkhauser,
Basel, Boston, Berlin, 1995.
[Lau 1] Laurent-Thiebaut, C., Theorie des fonctions holomorphes de plusieurs variables, Savoi-
rs Actuels, Paris, 1997.
[Lel 1] Lelong, P., Les fonctions plurisousharmoniques, Ann.

Ec. Norm. Sup., 62 (1945), 301
338.
[Lel 2] Lelong, P., La convexite et les fonctions analytiques de plusieurs variables complexes,
J. Math. Pures Appl., 31 (1952), 191219.
[Ler 1] Leray, J., Fonction de variables complexes:sa representation comme somme de puis-
sances negatives de fonctions lineaires, Rend. Accad. Naz. Lincei., ser. 8, 20 (1956),
589590.
[Ler 2] Leray, J., Le calcul dierentiel et integral sur une variete analytique complexe: Probl`e-
me de Cauchy III, Bull. Soc. Math. France, 87 (1959), 81180.
[Lew 1] Lewy, H., On the local character of the solutions of an atypical linear dierential
equation in three variables and a related theorem for regular functions of two complex
variables, Ann. Math., 64 (1956), 514522.
[Lew 2] Lewy, H., An example of a smooth linear partial deerential equation without solution,
Ann. Math., 66 (1957), 155158.
[Lie 1] Lieb, I., Ein Approximationssatz auf streng pseudokonvexen Gebieten, Math. Ann.,
184 (1969), 5560.
[Lie 2] Lieb, I., Die Cauchy-Riemannschen Dierentialgleichungen auf streng pseudokonvexen
gebieten I, Math. Ann., 190 (1970), 644; II 199 (1972), 241256.
[LiRa 1] Lieb, I., and Range, R. M., Losungsoperatoren f ur den Cauchy-Riemann Komplex mit
C
k
-Abschatzungen, Math. Ann., 253 (1980), 145164.
[LiRa 2] Lieb, I., and Range, R. M., On integral representations and a priori Lipschitz estimates
for the canonical solution of the

-equation, Math. Ann., 265 (1983), 221251.
[LiRa 3] Lieb, I., and Range, R. M., Integral representations and estimates in the theory of the

-Neumann problem, Ann. of Math., 123 (1986), 265301.


360 Bibliography
[LiRa 4] Lieb, I., and Range, R. M., Estimates for a class of integral operators and applications
to the -Neumann problem, Invent. Math., 85 (1986), 415438.
[LiMa 1] Lions, J.-L., and Magenes, E., Non-Homogeneous Boundary Value Problems and Ap-
plications, Volume I, Springer-Verlag, New York, 1972.
[MaMi 1] Ma, L., and Michel, J., Local regularity for the tangential Cauchy-Riemann Complex,
J. Reine Angew. Math, 442 (1993), 6390.
[MaMi 2] Ma, L., and Michel, J., Regularity of local embeddings of strictly pseudoconvex CR
structures, J. Reine Angew. Math, 447 (1994), 147-164.
[Mal 1] Malgrange, B., Existence et approximation des solutions des equations aux derivees
partielles et des equations de convolution, Ann. Inst. Fourier (Grenoble), 6 (1955-56),
271355.
[Mal 2] Malgrange, B., Sur lint`egrabilit`e des structures presque-complexes, Symposia Math.,
(INDAM, Rome, 1968), Academic Press, London, II (1969), 289296.
[Mar 1] Martinelli, E., Alcuni teoremi integrali per le funzioni analitiche di pi` u variabili com-
plesse, Mem. della R. Accad. dItalia, 9 (1938), 269283.
[McSt 1] McNeal, J., and Stein, E. M., Mapping properties of the Bergman projection on convex
domains of nite type, Duke Math. J., 73 (1994), 177199.
[McSt 2] McNeal, J., and Stein, E. M., The Szego projection on convex domains, Math. Zeit.,
224 (1997), 519553.
[MeCo 1] Meyer, Y., and Coifman, R., Wavelets: Calderon-Zygmund and Multilinear Operators,
Cambridge University Press, Cambridge, United Kingdom, 1997.
[Mic 1] Michel, J., Integral representations on weakly pseudoconvex domains, Math. Zeit., 208
(1991), 437462.
[MiPe 1] Michel, J., and Perotti, A., C
k
-regularity for the -equation on strictly pseudoconvex
domains with piecewise smooth boundaries, Math. Zeit., 203 (1990), 415427.
[MiSh 1] Michel, J., and Shaw, M.-C., Subelliptic estimates for the

-Neumann operator on
piecewise smooth strictly pseudoconvex domains, Duke Math. J., 93 (1998), 115128.
[MiSh 2] Michel, J., and Shaw, M.-C., C

-regularity of solutions of the tangential CR-equations


on weakly pseudoconvex manifolds, Math. Ann., 311 (1998), 147162.
[MiSh 3] Michel, J., and Shaw, M.-C., and
b
problems on nonsmooth domains, Analysis and
Geometry in Several Complex Variables, Trends Math., Birkhuser, Boston, (1999),
159-192.
[MiSh 4] Michel, J., and Shaw, M.-C., The

problem on domains with piecewise smooth bound-
aries with applications, Trans. Amer. Math. Soc. 311 (1999), 4365-4380.
[MoKo 1] Morrow, J., and Kodaira, K., Complex Manifolds, Holt, Rinehart and Winston, Inc.,
New York, 1971.
[Mor 1] Morrey, C. B., The analysis embedding of abstract real analytic manifolds, Ann. of
Math., (2) 40 (1958), 6270.
[Mor 2] Morrey, C. B., Multiple Integrals in the Calculus of Variations, Springer-Verlag, New
York, 1966.
[NaRo 1] Nagel, A., and Rosay, J.-P., Non-existence of homotopy formula for (0, 1) forms on
hypersurfaces in C
3
, Duke Math. J., 58 (1989), 823827.
[NRSW 1] Nagel, A., Rosay, J. P., Stein, E. M., and Wainger, S., Estimates for the Bergman and
Szego kernels in C
2
, Ann. Math., 129 (1989), 113149.
[NaSt 1] Nagel, A., and Stein, E. M., Lectures on Pseudodierential Operators: Regularity The-
orems and Applications to Non-elliptic Problems, Math. Notes 24, Princeton University
Press, Princeton, New Jersey, 1979.
[Nar 1] Narasimhan, R., Several Complex Variables, Univ. of Chicago Press, Chicago, 1971.
[NeNi 1] Newlander, A., and Nirenberg, L., Complex coordinates in almost complex manifols,
Ann. Math., 64 (1957), 391404.
[Nir 1] Nirenberg, L., A complex Frobenius theorem, Seminar on analytic functions I, Prince-
ton, (1957), 172189.
Bibliography 361
[Nir 2] Nirenberg, L., On elliptic partial dierential equations, Ann. Scuola Norm. Sup. Pisa,
13 (1959), 115162.
[Nir 3] Nirenberg, L., Lectures on Linear Partial Dierential Equations, Amer. Math. Soc.,
CBMS Regional Conference, Providence, Rhode Island, 1973.
[Nir 4] Nirenberg, L., On a question of Hans Lewy, Russian Math. Surveys, 29 (1974), 251
262.
[Noe 1] Noell, A., Local versus global convexity of pseudoconvex domains, Proc. Symposia Pure
Math., Amer. Math. Soc., Providence, Rhode Island, 52, I (1991), 145150.
[Nor 1] Norguet, F., Sur les domains dholomorphie des fonctions uniformes de plusieurs vari-
able complexes (passage du local au global), Bull. Soc. Math. France, 82 (1954), 137
159.
[Oka 1] Oka, K., Sur les fonctions analytiques de plusieurs variables, VI, Domaines pseudo-
convexes, Tohoku Math. J., 49 (1942), 1552.
[Oka 2] Oka, K., Sur les fonctions analytiques de plusieurs variables, IX, Domaines nis sans
point critique interieur, Jap. J. Math., 23 (1953), 97-155.
[OlRa 1] Oleinik, O. A., and Radkevic, E. V., Second order equations with nonnegative charac-
teristic form, Itogi Nauk, Moscow, (1971).
[vr 1] vrelid, N., Integral representation formulas and L
p
-estimates for the

-equation,
Math. Scand., 29 (1971), 137-160.
[PhSt 1] Phong, D. H., and Stein, E. M., Estimates for the Bergman and Szego projections on
strongly pseudoconvex domains, Duke Math. J., 44 (1977), 659704.
[PhSt 2] Phong, D. H., and Stein, E. M., Hilbert integrals, singular integrals and Radon trans-
forms II, Invent. Math., 86 (1986), 75113.
[Ple 1] Plemelj, J., Ein Erganzungssatz zur Cauchyschen Integraldarstellung analytischer Fun-
ktionen, Randwerte betreend, Monatsh. Math. und Phys., 19 (1908), 205210.
[Pol 1] Polking, J. C., L
p
estimates for

on convex domains in C
2
, Proc. Amer. Math. Soc.,
Summer Institute, Santa Cruz, 92, I (1989), 309-322.
[Poly 1] Polyakov, P. L., The Cauchy-Weil formula for dierential forms, Mat. Sb., 85 (1971),
388-402; Engl. Transl.: Math. USSR Sb., 14(1971), 383-398.
[Rad 1] Radkevic, E., Hypoelliptic operators with multiple characteristics, Math. USSR Sb., 8
(1969), 181206.
[Ram 1] Ramirez, E., Ein Divisionsproblem in der komplexen analysis mit einer Anwendung
auf Randintegraldarstellung, Math. Ann., 184 (1970), 172187.
[Ran 1] Range, R. M., Holder estimates for on convex domains in C
2
with real analytic
boundary, Several Complex Variables: Proc. Sympos. Pure Math.,XXX, Part 2, Willi-
ams Coll., Williamstown, Mass., 1975, Amer. Math. Soc., Providence, R.I., 1977.,
pp. 3133.
[Ran 2] Range, R. M., The Carath`eodory metric and holomorphic maps on a class of weakly
pseudoconvex domains, Pacic J. Math, 78 (1978), 173189.
[Ran 3] Range, R. M., On Holder estimates for

u = f on weakly pseudoconvex domains, Proc.
Int. Conf. (Cortona, Italy, 1976-1977), Scuola Norm. Sup. Pisa, (1978), 247-267.
[Ran 4] Range, R. M., A remark on bounded strictly plurisubharmonic exhaustion functions,
Proc. Amer. Math. Soc., 81 (1981), 220222.
[Ran 5] Range, R. M., The

-Neumann operator on the unit ball in C
n
, Math. Ann., 266
(1984), 449456.
[Ran 6] Range, R. M., Holomorphic Functions and Integral Representations in Several Complex
Variables, Graduate Texts in Math.,Vol.108, Springer-Verlag, N.Y., 1986.
[Ran 7] Range, R. M., Integral kernels and Holder estimates for on pseudoconvex domains
of nite type in C
2
, Math. Ann., 288 (1990), 6374.
[Ran 8] Range, R. M., On Holder and BMO estimates for on convex domains in C
2
, J.
Geom. Anal., 2 (1992), 575584.
[RaSi 1] Range, R. M., and Siu, Y.-T., Uniform estimates for the

-equation on domains with
piecewise smooth strictly pseudo-convex boundaries, Math. Ann., 206 (1974), 325354.
362 Bibliography
[RaSi 2] Range, R. M., and Siu, Y.-T., C
k
approximation by holomorphic functions and -
closed forms on C
k
submanifolds of a complex manifold, Math. Ann., 210 (1974),
105122.
[ReSt 1] Remmert, R., and Stein, K., Eigentliche holomorphe Abbildungen, Math. Zeit., 73
(1960), 159189.
[RiNa 1] Riesz, F., and Sz.-Nagy, B., Functional Analysis, Frederick Ungar Publishing Co., New
York, 1955.
[Rom 1] Romanov, A. V., A formula and estimates for the solutions of the tangential Cauchy-
Riemann equation, Math. USSR Sb., 28 (1976), 4971.
[Rosa 1] Rosay, J.-P., Equation de Lewy-resolubilite globale de lequation
b
= f sur la frontiere
de domaines faiblement pseudo-convexes de C
2
(o` u C
n
), Duke Math. J., 49 (1982),
121128.
[Rosa 2] Rosay, J.-P., Some applications of Cauchy-Fantappie forms to (local) problems on

b
,
Ann. Scuola Normale Sup. Pisa, 13 (1986), 225243.
[Ros 1] Rossi, H., Attaching analytic spaces to an analytic space along a pseudoconcave bound-
ary, Proc. Conf. Complex Manifolds (Minneapolis), 1964, Springer-Verlag, New York,
(1965), 242256.
[RoSt 1] Rothschild, L., and Stein, E. M., Hypoelliptic dierential operators and nilpotent
groups, Acta Math., 137 (1976), 247320.
[Rud 1] Rudin, W., Functional Analysis, McGraw-Hill Book Co., New York, 1973.
[Rud 2] Rudin, W., Function Theory in the Unit Ball of C
n
, Springer-Verlag, New York, 1980.
[Ser 1] Serre, J.-P., Un theor`eme de dualite, Comm. Math. Helv., 29 (1955), 926.
[Sha 1] Shaw, M.-C., Global solvability and regularity for on an annulus between two weakly
pseudoconvex domains, Trans. Amer. Math. Soc., 291 (1985), 255267.
[Sha 2] Shaw, M.-C., L
2
estimates and existence theorems for the tangential Cauchy-Riemann
complex, Invent. Math., 82 (1985), 133150.
[Sha 3] Shaw, M.-C., L
p
estimates for local solutions of

b
on strongly pseudo-convex CR
manifolds, Math. Ann., 288 (1990), 3562.
[Sha 4] Shaw, M.-C., Optimal Holder and L
p
estimates for

b
on the boundaries of real ellip-
soids in C
n
, Trans. Amer. Math. Soc., 324 (1991), 213234.
[Sha 5] Shaw, M.-C., L
2
existence theorems for the

b
-Neumann problem on strongly pseudo-
convex CR manifolds, Jour. Geometric Analysis, 1 (1991), 139163.
[Sha 6] Shaw, M.-C., Local existence theorems with estimates for
b
on weakly pseudoconvex
boundaries, Math. Ann., 294 (1992), 677700.
[Sha 7] Shaw, M.-C., Semi-global existence theorems of
b
for (0, n-2) forms on pseudoconvex
boundaries in C
n
, Asterisque, Societe Mathematiqe de France, Colloque DAnalyse
Complexe et Geometrie, Marseille, (1993), 227240.
[Sha 8] Shaw, M.-C., Integral representation for

b
in CR manifolds, Proceedings of the Geo-
metric Complex Analysis conference, Hayama, Japan, (1996), 535549.
[Sha 9] Shaw, M.-C., Homotopy formulas for
b
in CR manifolds with mixed Levi signatures,
Math. Zeit., 224 (1997), 113135.
[Sib 1] Sibony, N., Un exemple de domain pseudoconvexe regulier ou lequation

u = f
nadmet pas de solution bornee pour f bornee, Invent. Math., 62 (1980), 235242.
[Sib 2] Sibony, N., Une classe de domaines pseudoconvexes, Duke Math. J., 55 (1987), 299
319.
[Sib 3] Sibony, N., Hypoellipticite pour loperateur , Math. Ann., 276 (1987), 279290.
[Sib 4] Sibony, N., On Holder estimates for , Several Complex Variables: Proc. of the Mittag-
Leer Institute 1987/1988, Math. Notes 38, Princeton University Press, Princeton,
New Jersey, (1993), 587599.
[Siu 1] Siu, Y.-T., The -problem with uniform bounds on derivatives, Math. Ann., 207 (1974),
163176.
[Siu 2] Siu, Y.-T., The Levi problem, Several Complex Variables: Proc. Sympos. Pure Math.,
XXX , Part 2, Williams Coll., Williamstown, Mass., 1975, Amer. Math. Soc., Provi-
dence, R.I., 1977., pp. 4548.
Bibliography 363
[Siu 3] Siu, Y.-T., Pseudoconvexity and the problem of Levi, Bull. Amer. Math. Soc., 84
(1978), 481512.
[Sko 1] Skoda, H., Application des techniques L
2
`a la theorie des ideaux dune alg`ebre de
fonctions holomorphes avec poids, Ann. Sci. Ec. Norm. Sup., 5 (1972), 545579.
[Sko 2] Skoda, H., Valeurs au bord pour les solutions de loperateur d, et caracterisation des
zeros des fonctions de la classe de Nevanlinna, Bull. Soc. Math., France, 104 (1976),
225299.
[Ste 1] Stein, E. M., Boundary values of holomorphic functions, Bull. Amer. Math. Soc., 76
(1970), 1292-1296.
[Ste 2] Stein, E. M., Singular Integrals and Dierentiability Properties of Functions, Math.
Series 30, Princeton University Press, Princeton, New Jersey, 1970.
[Ste 3] Stein, E. M., Boundary Behavior of Holomorphic Functions of Several Complex Vari-
ables, Math. Notes 11, Princeton University Press, Princeton, New Jersey, 1972.
[Ste 4] Stein, E. M., Harmonic Analysis: Real-Variable Methods, Orthogonality, and Oscil-
latory Integrals, Math. Series 43, Princeton University Press, Princeton, New Jersey,
1993.
[StWe 1] Stein, E. M., and Weiss, G., Introduction to Fourier Analysis on Euclidean Spaces,
Math. Series 32, Princeton University Press, Princeton, New Jersey, 1975.
[Str 1] Straube, E. J., Interpolation between Sobolev and between Lipschitz spaces of analytic
functions on starshaped domains, Trans. Amer. Math. Soc., 316 (1989), 653671.
[Str 2] Straube, E. J., Plurisubharmonic functions and subellipticity of the -Neumann prob-
lem on nonsmooth domains, Math. Research Letters, 4 (1997), 459467.
[Tark 1] Tarkhanov, N. N., Theorems on the jump and exterior product of currents from D

q
(M)
by means of their harmonic representations (Russian), Sibirsk. Otdel. Akad. Nauk SSS,
Krasnoyarsk, (1980), 147-160.
[Tar 1] Tartako, D. S., Local analytic hypoellipticity for
b
on non-degenerate Cauchy-
Riemann manifolds, Proc. Natl. Acad. Sci. USA, 75 (1978), 30273028.
[Tar 2] Tartako, D. S., The local real analyticity of solutions to
b
and the -Neumann
problem, Acta Math., 145 (1980), 177204.
[Tre 1] Treves, F., Basic Linear Partial Dierential Equations, Academic Press, New york,
1975.
[Tre 2] Treves, F., Analytic hypo-ellipticity of a class of pseudodierential operators with dou-
ble characteristics and applications to the -Neumann problem, Commun. in Partial
Dierential Equations, 3 (1978), 475642.
[Tre 3] Treves, F., Introduction to Pseudodierential and Fourier Integral Operators, Volume
I, Pseudodierential Operators; Volume II, Fourier Integral Operators, Plenum Press,
New York and London, 1980.
[Tre 4] Treves, F., Poincare lemma in analytic complex with nondegenerate Levi form, Comm.
in Partial Dierential Equations, 7 (1982), 1467-1482.
[Tre 5] Treves, F., Homotopy Formulas in the Tangential Cauchy-Riemann Complex, Memoirs
of the Amer. Math. Soc., Providence, Rhode Island, 1990.
[Tre 6] Treves, F., Hypo-Analytic Structures: Local Theory, Math. Series 40, Princeton Uni-
versity Press, Princeton, New Jersey, 1992.
[Vas 1] Vassiliadou, S., Semiglobal results for
b
on weakly convex hypersurfaces in C
n
, to
appear in Illinois Math. Journal.
[Web 1] Webster, S. M., A new proof of the Newlander-Nirenberg theorem, Math. Zeit., 201
(1989), 303316.
[Web 2] Webster, S. M., On the local solution of the tangential Cauchy-Riemann equations,
Annals Inst. H. Poincare(Anl), 6 (1989), 167182.
[Web 3] Webster, S. M., On the proof of Kuranishis Embedding Theorem, Annals Inst. H.
Poincare(Anl), 6 (1989), 183207.
[Wel 1] Wells, R. O., Dierential Analysis on Complex Manifolds, 2nd ed., Springer-Verlag,
New York, 1980.
364 Bibliography
[Wu 1] Wu, D., Sobolev and Holder estimates for

on bounded convex domains in C
2
, Illinois
J. Math., 42 (1998), 371388.
Bibliography 365

S-ar putea să vă placă și