LECTURES ON METHODS OF PARTIAL DIFFERENTIAL

EQUATIONS IN SEVERAL COMPLEX VARIABLES

1
MEI-CHI SHAW
Department of Mathematics, University of Notre Dame
255 Hurley Hall, Notre Dame, IN 46556-4618, (USA), e-mail: [email protected]
Contents
1. The Levi problem in several complex variables 2
1.1. The Cauchy-Riemann Equation in C 2
1.2. The Cauchy-Riemann equations in C
n
3
1.3. Domains of holomorphy 5
1.4. The Levi Problem and the

Equation 6
2. L
2
theory for

on pseudoconvex domains in C
n
7
2.1. The

-Neumann problem 8
2.2. L
2
existence for

and

-Neumann operator 10
2.3. Global Regularity for

on pseudoconvex domains 14
3. Boundary regularity for the

-Neumann problem 16
3.1.

-Neumann problem on strictly pseudoconvex Lipschitz
domains 16
3.2. Regularity for the

-Neumann problem on pseudoconvex
domains 20
4. Strong Okas lemma and the

-Neumann problem 21
4.1. Strong Okas Lemma and bounded plurisubharmonic exhaustion
functions 21
4.2. Strong Okas lemma and nite type conditions 25
5. The

-Neumann problem on Kahler manifolds with
nonnegative curvature 26
5.1. The

-Neumann problem on domains in CP
n
29
6. Tangential Cauchy-Riemann equations and the Bochner
Martinelli kernel 31
7. CR manifolds and tangential Cauchy-Riemann equations 31
7.1. L
2
theory for
b
on strongly pseudoconvex CR manifolds 31
1
These notes are results from short courses given by the author in Hongzhou, China,
2004, Trento, Italy, 2005 and Luminy, France 2007, Academia Sinica, Taipei, Taiwan,
2008 and Fudan, China 2008. She would like to thank all the students and colleagues who
participated in the courses. The author would especially like to thank Dr. Donato Ciampa
who took notes and put it into the typed form for the summer school on July, 17-29, 2005,
at I.T.C.-I.R.S.T. of Povo-Trento on Real PDEs for Complex and CR Geometry. These
notes were the bases of the present lecture notes which have been substantially expanded
and revised by the author.
1
2
7.2. L
2
theory for

b
on boundary of pseudoconvex domains 33
7.3. Nonisotropic Sobolev and Holder spaces 36
7.4. Campanato Space 38
7.5. The Tangential Cauchy-Riemann Complex 39
7.6. Folland-Stein-Tanaka formula for
b
43
8. Applications of

and

b
46
8.1. The Levi problem 47
8.2. Regularity for the Bergman projection 48
8.3. The Newlander-Nirenberg theorem 49
8.4. Embedding strongly pseudoconvex compact CR manifolds 51
8.5. Nonexistence of Levi-at hypersurfaces in CP
n
53
9. Open problems 54
10. Appendix 55
10.1. Preliminaries for Lipschitz domains 55
10.2. Preliminaries for Lipschitz domains 57
References 60
1. The Levi problem in several complex variables
1.1. The Cauchy-Riemann Equation in C. Let C be the complex Eu-
clidian space with coordinate z = x+iy. We can dene the Cauchy-Riemann
operator

z
:=
1
2
_

x
+i

y
_
,
and we say that a function h is holomorphic in a domain D of C, and we
write h O(D), if and only if
(1.1)
h
z
= 0.
Equation (1.1) is called the homogeneous Cauchy-Riemann equation in C.
The behavior of holomorphic function and the study of the homogeneous
Cauchy-Riemann equation is closely related to the solvability and regularity
of the inhomogeneous Cauchy-Riemann equation
(1.2)
u
z
= f,
with f a given function.
We have the following result.
Theorem 1.1. Let D be a bounded domain in C and let f C
k
(

D), for
k 1. Then the function dened by
u(z) :=
1
2i
__
D
f()
z
d d

,
3
is in C
k
(D) and satises (1.2). Moreover, if f is only in C(

D), then u(z)
dened as before satises (1.2) in the distribution sense.
Remark 1.2. Observe that the function
u(z) =
1

1
z
,
is a fundamental solution to (1.1).
This can be derived by dierentiating the fundamental solution for .
Since
1
2
log [z[ =
2

2
zz
log [z[ =
0
,
where
0
is the Dirac delta function centered at 0, we have
2

z
log [z[ =
1

z
1
z
=
0
.
This implies that 1/z is a fundamental solution for / z.
Let D be a bounded domain in C with C
1
boundary. Let f be a continous
function f on D. We dene u = Gf to be the unique solution to the Dirichlet
problem with u = f with u = 0 on bD. It follows that
2
u/ zz = f in
C in the distribution sense. Thus we have f =

(Gf) and the function
Gf = u satises

u = f.
1.2. The Cauchy-Riemann equations in C
n
. Let C
n
be the n-dim-
ensional complex Euclidian space. We denote coordinates by z = (z
1
, . . . , z
n
),
where z
j
= x
j
+iy
j
, 1 j n. We can dene the Cauchy-Riemann operator

z
j
:=
1
2
_

x
j
+i

y
j
_
,
for 1 j n. It is easy to see that the foolowing decomposition holds
CTC
n
= T
1,0
(C
n
) T
0,1
(C
n
),
where
T
1,0
(C
n
) = span
_

z
j
_
, T
0,1
(C
n
) = span
_

z
j
_
.
Denition 1.3. Let f a C
1
function dened on an open subset D of C
n
. We
say that f is holomorphic in D and we write f O(D) if f(z) is holomorphic
in each variable z
j
, when the others are xed, i.e. f satises
f
z
j
= 0,
for all 1 j n.
The Cauchy-Riemann equations are
(1.3)
u
z
j
= f
j
, j = 1, . . . , n, n 2.
4
This system is overdetermined (one unknown function with n equations).
In order for (1.3) to be solvable, f must satisfy the following compatibility
conditions
(1.4)
f
k
z
j
=
f
j
z
k
, 1 j, k n.
As in the 1-dimensional case, we can state the following for the solvability
of Cauchy Riemann equation.
Theorem 1.4. Let f
j
C

0
(D), 1 j n, where D is a relatively compact
subset of C
n
, n 2. Suppose that f
j
satisfy (1.4). Then there is a function
u C

0
(D) which satises (1.3).
Proof. Set
u(z) =
1
2i
__
D
f
1
(, z
2
, . . . , z
n
)
z
1
d d

.
Then u has compact support and we have
u
z
1
= f
1
,
u
z
j
= f
j
, j = 2, . . . , n.

One of the most striking dierence between the 1-dimensional and the
n-dimensional case is the so called Hartogs phenomenon, which states that,
if a bounded domain D in C
n
, n 2, has connected boundary, then any
holomorphic function f dened in an open neighborhood of the boundary
bD can be holomorphically extended to all D. This is not true in C, as
one can check easily with the function f(z) = 1/z, which is holomorphic in
C 0 and cannot be extended in any way as an entire function.
More precisely, we can state the following.
Theorem 1.5 (Hartogs). Let D be an open set in C
n
and let K be a compact
subset in D, such that D K is connected. If f O(D K), then f can be
extended holomorphically to all D.
Proof. Let C

0
(D) a cut-o function such that 1 on some open
neighborhood of K. Then

((1 )f) =

f,
has compact support and satises compatibility condition in C
n
.
Then, from the previous theorem, there is u with compact support such
that

u =

(f).
Finally, the function
h = (1 )f u,
is the desired extension.
5
1.3. Domains of holomorphy.
Denition 1.6. A domain D in C
n
is called a domain of holomorphy if is
not possible to nd two nonempty open sets D
1
, D
2
in C
n
such that
(i) D
1
is connected, D
1
D, D
2
D D
1
;
(ii) for any f O(D) there is

f O(D
1
) such that f[
D
2
=

f.
Equivalently, D is a domain of holomorphy if and only if for any p bD,
there exists a holomorphic function f O(D) such that f is singular at p.
An important notion related to domain in C
n
is that of pseudoconvexity.
Denition 1.7. Let D be a domain of C
n
with C
2
boundary, i.e. there is
a neighborhood U of D and a function : D R of class C
2
, such that
D U = z C
n
[(z) < 0, bD U = z C
n
[(z) = 0,
[(z)[ , = 0, z bD.
We say that D is pseudoconvex if
L
p
(; a) : (=

[
p
(a, a)) =
n

j,k=1

z
j
z
k
(p)a
j
a
k
0, p bD,
for all vector a = (a
1
, . . . , a
n
) C
n
(a T
0,1
p
(bD)) such that
n

j=1
a
j

z
j
= 0.
L
p
(; a) is called the Levi form of at p. A domain D is called strictly
pseudoconvex if the Levi form is positive denite.
A function is called (strictly) plurisubharmonic at p if L
p
(; a) 0 (> 0)
for all vectors a C
n
.
A domain D is pseudoconvex if and only if it is union of strictly pseudocon-
vex domains D
j
with C
2
boundary such that D =
j
D
j
and D
j
D
j+1
for each j.
Theorem 1.8. Let D C
n
be a strongly pseudoconvex domain with C
2
boundary bD. Then there exists a C
2
dening function for D which is strictly
plurisubharmonic in a neighborhood of bD.
It is easy to see that convex domains are domains of holomorphy.
It is also easy to show that every strictly pseudoconvex domain is locally
biholomorphic to a strictly convex domain. Thus locally for each p bD,
one can nd a neighborhood U
p
and a holomorphic function f
p
: U
p
D
so that f
p
cannot be extended holomorphically past p. Can one nd a f
p
holomorphic in D?
6
1.4. The Levi Problem and the

Equation. Let D 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 D is a domain of holomorphy. Near
each boundary point p bD, one must nd a holomorphic function f(z) on
D which cannot be continued holomorphically near p. This problem is called
the Levi problem for D at p. It involves the construction of a holomorphic
function with certain specic local properties.
If the domain D is strongly pseudoconvex with C

boundary bD and
p bD, one can construct a local holomorphic function f in an open neigh-
borhood U of p, such that f is holomorphic in U D, f C(D U p)
and f(z) as z D approaches p. In fact f can be easily obtained as
follows: let r be a strictly plurisubharmonic dening function for D and we
assume that p = 0. Let
F(z) = 2
n

i=1
r
z
i
(0)z
i

i,j=1

2
r
z
i
z
j
(0)z
i
z
j
.
F(z) is holomorphic, and it is called the Levi polynomial of r at 0. Using
Taylors expansion at 0, there exists a suciently small neighborhood U of
0 and C > 0 such that for any z D U,
ReF(z) = r(z) +
n

i,j=1

2
r
z
i
z
j
(0)z
i
z
j
+O([z[
3
) C[z[
2
.
Thus, F(z) ,= 0 when z D U 0. Setting
f =
1
F
,
it is easily seen that f 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
(U) and = 1 in a neighborhood of 0. We note that
f is not holomorphic in D. However, if f can be corrected by solving a

-equation, then the Levi problem will be solved.

Let us consider the (0,1)-form g dened by
g =

(f) = (

)f.
This form g can obviously be extended smoothly up to the boundary. It is
easy to see that g is a

-closed form in D and g C

(0,1)
(D). If we can nd
a solution u C

(D) such that

u = g in D,
then we dene for z D,
h(z) = (z)f(z) u(z).
7
It follows that h is holomorphic in D, h C

(D 0) and h 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.
It has been known that domains of holomorphy are pseudoconvex. The
converse whether a pseudoconvex domain is a domain of holomorphy is the
so called Levi problem: given a domain D is it possible, for each point
p bD, to nd a holomorphic function f over D such that it cannot be
continued holomorphically near p? The answer to this question is given in
the following theorem.
Theorem 1.9. Let D be a domain in C
n
. The following are equivalent:
(i) D is pseudoconvex;
(ii) D is a domain of holomorphy;
(iii) for each f C

(D) with

f = 0 there is u C

(D) such that

u = f.
It is well known that (ii) implies (i).
To show that (i) implies (ii) is the so-called Levi problem.
We will show that (i) implies (iii) and (iii) implies (ii).
The Levi problem has been solved by
(1) (1942) Oka for n = 2 and later for all n 2.
(2) (1958) Grauert using Sheaf methods.
(3) (1962) Kohn by the

-Neumann problem.
(4) (1965) Hormander by the weighted

-Neumann problem.
2. L
2
theory for

on pseudoconvex domains in C
n
By the Hilbert space theory of unbounded linear operator, if H
1
and H
2
are two Hilbert spaces, T : H
1
H
2
is a closed, densely dened linear
operator, then
H
2
= (T) Ker(T

),
and
H
1
= (T

) Ker(T).
If we can show that (

) is closed and then, if

u = f can be solved, we
must have fKer(

), because if g is such that

g = 0, then
(f, g) = (

u, g) = (u,

g) = 0.
For a closed densely dened linear operator T, (T) is closed if and only if
|Tu| C|u|, u Dom(T) (Ker(T))

or if and only if
|T

u| C|u|, u Dom(T

) (Ker(T

))

.
8
2.1. The

-Neumann problem. Let D be a domain in C
n
. Let f be a
function of class C
1
and Let (z
1
, . . . , z
n
) be the complex coordinates for C
n
,
we get
df =
n

j=1
_
f
z
j
dz
j
+
f
z
j
d z
j
_
.
We dene

f :=
n

j=1
f
z
j
d z
j
, f :=
n

j=1
f
z
j
dz
j
so that d = +

. This means that the dierential of a function can be
decomposed in the sum of a (1, 0)-form f and a (0, 1)-form

f. A function
f is holomorphic if and only if

f = 0.
We dene the space of (p, q)-forms
p,q
(D) as the subspace of (p +q)-forms
which can be written as
f =

|I|=p,|J|=q
f
IJ
dz
I
d z
J
,
where I = (i
1
, . . . , i
p
) and J = (j
1
, . . . , j
q
) are multiindices of length p and
q respectively, dz
I
= dz
i
1
. . . dz
ip
, d z
J
= d z
j
1
. . . d z
jq
.
Since d is a complex, it is easy to see that the sequence
. . .

p,q1
(D)

p,q
(D)

p,q+1
(D)

. . .
is a complex, i.e.

2
= 0. It is called the Cauchy-Riemann complex.
Let L
2
(p,q)
(D) and C

(p,q)
(D) denote the (p, q)-form with L
2
or C

coe-
cients respectively, and let ( , ) denotes the usual L
2
inner product. We can
dene the formal adjoint of

under the L
2
-norm as
: C

(p,q)
(D) C

(p,q1)
(D),
(f, g) = (f,

g),
for every g C

(p,q)
(D) with compact support.
We denote the (maximal) L
2
closure of

with the same symbol. Then
we can dene its L
2
adjoint

as Hilbert space adjoint by

: L
2
(p,q)
(D) L
2
(p,q1)
(D),
in the following sense: f is in Dom(

), the domain of

, if there is a
g L
2
(p,q1)
(D) such that, for every Dom(

) L
2
(p,q1)
(D) we have
(f,

) = (g, ).
Then we dene

f = g.
We have that

is an unbounded operator and Dom(

) L
2
(p,q)
(D). More-
over, if D is a bounded domain, then C

(p,q)
(

D) Dom(

). Finally,

is a
9
closed, densely dened linear operator. The same is true for

too.
If f C
1
(p,q)
(

D) Dom(

), then
(f,

u) = (

f, u), u Dom(

).
If u C
1
(p,q1)
(

D) and D is a C
1
domain with dening function , then
(f,

u) = (

f, u) = (f, u) +
_
b
f

, u)d.
Since compactly supported forms in C

(p,q)
(

D) are dense in L
2
(p,q)
(D), it fol-
lows that if f C
1
(p,q)
(

D) Dom(

), we must have

f = f,
and
f

= 0on bD.
More explicitly, when f C
1
(0,1)
(

D), this condition becomes
n

j=1
f
j
()

j
() = 0, bD.
We can now dene the Laplacian of the

-complex. It is the operator
:=

: L
2
(p,q)
(D) L
2
(p,q)
(D),
with the domain of denition
Dom() := f Dom(

) Dom(

) :

f Dom(

),

f Dom(

).
The operator is a linear, closed, densely dened, self adjoint operator.
As in the case to be in Dom(

), in order to belong to Dom(), a (p, q)-

form f must satises the

-Neumann boundary conditions
f

= 0,

f

= 0,
on bD. The boundary value problem
_
_
_
u = f in D
u

= 0 in bD

= 0 in bD
is the

-Neumann problem.
Example.
Let D be a bounded domain with 0 bD and assume that there is a
neighborhood U of 0 such that
D U = (z
1
, . . . , z
n
) U : (z
n
) < 0.
10
If we take u C
2
(0,1)
(

D), then we can write u =

n
j=1
u
j
d z
j
and u Dom()
if and only if
(1) u
n
= 0, on bD U,
(2)
u
j
z
n
= 0, on bD U, j = 1, . . . , n 1.
In fact the condition u

= 0 is equivalent to
n

j=1
u
j

z
j
= 0 u
n
= 0,
on bD, while the condition

u

can be rewritten as
u
j
z
n

u
n
z
j
= 0
u
j
z
n
= 0, j = 1, . . . , n 1,
on bD.
Actually, to obtain L
2
existence results, we need to work in the weighted
L
2
spaces. If C
2
(

D) we dene
L
2
(D, ) :=
_
f : |f|
2

:=
_
D
[f[
2
e

dV <
_
,
the weighted L
2
space with weight . Observe that

, the Hilbert space

adjoint of

in L
2
(D, ), is related to

by

= e

_
=

+A
0
(),
where A
0
() is a zeroth order operator which depends on . We set up the
weighted

-Neumann problem

as before.
2.2. L
2
existence for

and

-Neumann operator. From the prelimi-
nary lemma, () is closed if and only if
|u| C|u|, u Dom() (Ker())

.
In order to prove the L
2
existence both for

and

-Neumann operator,
we need the following a priori identity.
Proposition 2.1 (Morrey-Kohn). Let D C
n
be a domain with C
2
boundary and dening function . For each f C
1
(p,q)
(

D) Dom(

) we
have
(2.1) |

f|
2
+|

f|
2
=
n

j=1
_
_
_
_
f
z
j
_
_
_
_
2
+
n

j,k=1
_
b

z
j
z
k
f
I,jK

f
J,kK
d.
11
Under this assumptions, we can rewrite a more general form for the iden-
tity of Morrey-Kohn-Hormander:
|

f|
2

+|

f|
2

=
n

j=1
_
_
_
_
f
z
j
_
_
_
_
2

+
n

j,k=1
_
b

z
j
z
k
f
I,jK

f
J,kK
e

d
+
n

j,k=1
_

z
j
z
k
f
I,jK

f
J,kK
e

dV. (2.2)
Since D is pseudoconvex, we have
(2.3)
n

j,k=1
_
b

z
j
z
k
f
I,jK

f
J,kK
e

d 0.
If we choose = [z[
2
, the last term becomes
_

[f[
2
e

dV = |f|
2

.
Then, if D is pseudoconvex, we get
(f, f)

= |

f|
2

+|

f|
2

|f|
2

,
which is the a priori estimate.
A crucial fact to prove the actual estimate from the a priori estimate is
the following lemma.
Lemma 2.2 (A Density Lemma). Let D be a bounded domain in C
n
with
C
2
boundary. Then C
1
(p,q)
(

D) Dom(

) is dense in Dom(

) Dom(

) in
the graph norm f |f|

+|

f|

+|

f|

.
Proof. We shall give only the main steps of the proof, which is essentially a
variation of the Friedrichs Lemma
2
. By a partition of unity, we may assume
that the domain is star-shaped and 0 D.
(1) The space C

(p,q)
(

D) is dense in Dom(

) in the graph norm f |f|

+
|

f|

.
2
Friedrichs lemma: Let C

0
(R
N
) a function with support in the unit ball and
such that
Z
R
N
dV = 1,
and dene (x) =
N
(x/). If v L
2
(R
N
) is with compact support and u is a C
1
function in a neighborhood of the support of v, we have
uDj(v ) (uDjv) 0, as 0,
where Dj = /xj and uDjv is dened in the sense of distribution.
12
Using convolution, we have that if f L
2
p,q
(D) Dom(

) and f

=
f(
1
1+
z), then
f

(f

)

f
_
_
_
0.
The rst sequence converges since it follows easily from the Youngs inequal-
ity:
|f

| |f

| |f|.
The second sequence converges from

(f

) =
1
1 +
(

f)(
1
1 +
z).
In other words, the weak maximal closure of

is equal to the strong maximal
closure.
(2) C

(p,q)
(

D) with compact support in D is dense in Dom(

).
We assume that = 0. This follows from the fact that

is the maximal
closure, its adjoint

is minimal . We extend f to

f by setting

f = 0 outside
D . Then we claim that

f =

f L
2
(C
n
) in the distribution sense. To see
this, we have for any v C

(p,q1)
(D),
(

f,

v)
C
n = (f,

v)
D
= (

f, v)
D
= (f, v)
D
= (

f, v)
C
n.
From (1), C

(p,q1)
(D) is dense in Dom(

), we have proved the claim.

We can approximate f by f

=

f(
z
1
) and then regularize, we get
f

0
(D). Using again Friedrichs lemma,
f

(f

f
_
_
_
0.
This shows that the Hilbert space adjoint

is the same as the strong

minimal extension of .
(3) C
1
(p,q)
(

D) Dom(

) is dense in Dom(

) Dom(

).
This last part is essentially proved using the fact that f C
1
(p,q)
(

D)
Dom(

) if and only if f

= 0 on bD, being a dening function for D.

By this and the fact that
(

f, g) = (f, g) +
_
b
f

, g)d = (f, g) +
_
b
f, g

)d.
We rst decompose f by f = f

+f

where f

= f

is the complex
normal part and f

is the complex tangent component. We rst approximate

13
f by
(2.4) f
()
= f

+f

= f

(
1
1 +
z) +

f

(
1
1
z).
Then we regularize again as in Friederichs lemma, we obtain

(f
()

) (

+

f

0
since the complex normal component is not in the Cauchy data of

.
On the other hand, we get
(f
()

) (f

+f

0
since f Dom(

). This gives that

f
()

(f
()

)

f

(f
()

f
_

_
0.
This proves the density lemma.
Using the density lemma, we have, for any f Dom(

) Dom(

)
|f|
2

f|
2

+|

f|
2

= (

f, f)

f|

|f|

,
so that
|f|

f|,
which implies that

has closed range and Ker(

) = 0.
Now we can prove the L
2
weighted existence of

-Neumann weighted
operator. We have the following.
Proposition 2.3. Let D a bounded pseudoconvex domain with C
2
boundary
in C
n
. For each 1 q n there is a bounded operator N

: L
2
(p,q)
(D, )
L
2
(p,q)
(D, ) such that:
(1) f =

f, for any f L
2
(p,q)
(D, );
(2) if

f = 0 then u

:=

f is the solution for

u = f such that
|u

1
q
|f|

.
Theorem 2.4 (L
2
existence theorem for

). Let D C
n
a bounded pseu-
doconvex domain. For every f L
2
(p,q)
(D), 1 q n, with

f = 0, one
can nd u L
2
(p,q1)
(D) such that

u = f and
|u|
_
e
q
|f|,
where = diam(D).
14
Choosing = t[z[
2
, we get
(2.5) te
t
2
|u|
2

1
q
|f|
2
.
This proves Hormanders theorem with t =
2
.
Summing up the previous results and passing to the unweighted L
2
spaces,
we have the following result due to Hormander.
Theorem 2.5 (Hormander). Let D C
n
a bounded pseudoconvex do-
main. Then has closed range and () = L
2
(p,q)
(D), 1 q n. More-
over, there is a bounded operatore N : L
2
(p,q)
(D) L
2
(p,q)
(D) such that
(i) N = N = I on Dom();
(ii) =

N, for any L
2
(p,q)
(D, );
(iii) if L
2
(p,q)
(D) such that

= 0 then u =

N is the canonical
solution to

u = and
|u|
_
e
q
||;
(iv) we have the following estimates for N:
|Nf|
e
q

2
|f|,
|

Nf|
_
e
q
|f|,
|

Nf|
_
e
q
|f|,
for any f L
2
(p,q)
(D).
Theorem 2.6. Let D be a pseudoconvex domain in C
n
. For any f
C

(p,q)
(D), where 0 p n and 1 q < n, such that

f = 0 in D,
there exists u C

(p,q1)
(D) satisfying

u = f.
From Hormanders theorem, we have
Corollary 2.7. Let D be a pseudoconvex domain in C
n
. Then D is a
domain of holomorphy.
2.3. Global Regularity for

on pseudoconvex domains. If the bound-
ary bD is smooth, we also have the following global boundary regularity
results for

.
Theorem 2.8 (Kohn). Let D C
n
be a pseudoconvex domain with
smooth boundary bD. For any f C

(p,q)
(D), where 0 p n and
1 q < n, such that

f = 0 in D, there exists u C

(p,q1)
(D) satisfy-
ing

u = f.
15
Proof. For any s > 0, there exists T
s
>> 1 such that N
t
W
s
(D) for all
t > T
s
. From the Sobolev embedding theorem, there exists u
k
C
k
(D)
for each k N. By a Mittag-Leer procedure, one can extract a solution
u

(D).
Theorem 2.9. Let D be a bounded pseudoconvex domain in C
n
, n 2.
For each 0 p n, 1 q n and t > 0, there exists a bounded operator
N
t
: L
2
(p,q)
(D) L
2
(p,q)
(D), such that
Range(N
t
) Dom(
t
). N
t

t
=
t
N
t
= I on Dom(
t
).
For any f L
2
(p,q)
(D), f =

t
N
t
f

t
N
t
f.
N
t
= N
t
on Dom(), 1 q n 1,

t
N
t
= N
t

t
on Dom(

t
), 2 q n.
The following estimates hold: For any f L
2
(p,q)
(D),
tq | N
t
f |
(t)
| f |
(t)
,

tq | N
t
f |
(t)
| f |
(t)
,

tq |

N
t
f |
(t)
| f |
(t)
.
If f L
2
(p,q)
(D) and f = 0 in D, then for each t > 0, there exists a solution
u
t
=

t
N
t
f satisfying u
t
= f and the estimate
tq | u
t
|
2
(t)
| f |
2
(t)
.
We have chosen t =
2
, where is the diameter of D, to obtain the best
constant for the bound of the

-Neumann operator without weights. Our
next theorem gives the regularity for N
t
in the Sobolev spaces when t is
large.
Theorem 2.10. Let D be a smooth bounded pseudoconvex domain in C
n
,
n 2. For every nonnegative integer k, there exists a constant S
k
> 0 such
that the weighted

-Neumann operator N
t
maps W
k
(p,q)
(D) boundedly into
itself whenever t > S
k
, where 0 p n, 1 q n.
An operator is called exactly regular on W
k
(p,q)
(D), k 0, if it maps the
Sobolev space W
k
(p,q)
(D) continuously into forms with W
k
(D) coecients.
The following theorem shows that all the related operators of N
t
are also
exactly regular if N
t
is exactly regular.
Theorem 2.11. Let D be a smooth bounded pseudoconvex domain in C
n
,
n 2. For every nonnegative integer k, there exists a constant S
k
> 0
such that for every t > S
k
the operators N
t
,

t
N
t
,

t
N
t
and

t
N
t
are
exactly regular on W
k
(p,q)
(D), where 0 p n, 1 q n. Furthermore,
there exists a constant S

k
> 0 such that for t > S

k
, the weighted Bergman
projection P
t,(p,0)
maps W
k
(p,0)
(D) boundedly into itself.
16
3. Boundary regularity for the

-Neumann problem
3.1.

-Neumann problem on strictly pseudoconvex Lipschitz do-
mains. We are interested in the following questions:
1. Can one solve equation (6.0.1) with a smooth solution u C

(p,q1)
(D) if
f is in C

(p,q)
(D)?
2. Does the canonical solution

Nf belong to W
s
(p,q1)
(D) if f is in
W
s
(p,q)
(D)?
We derive the 1/2-subelliptic estimate for the

-Neumann problem oper-
ator when is a strictly pseudoconvex Lipschitz domain. We use W
s
() to
denote the Sobolev space, s R, and by | |
s
its norm.
A bounded pseudoconvex Lipschitz domain in C
n
is said to have a
plurisubharmonic Lipschitz dening function if is a global dening func-
tion in (or on the boundary of b and is plurisubharmonic in . Recall
that a continuous function is plurisubharmonic if it is subharmonic in every
complex line. Plurisubharmonicity is well-dened for continuous functions
or even upper semicontinuous functions. We next dene strictly (or strongly)
pseudoconvex domains with Lipschitz boundaries.
Denition 3.1. A bounded Lipschitz domain in C
n
is called strictly
pseudoconvex (with Lipschitz boundary b) if there exists an exhaustion

of such that
(1) The sequence

is an increasing sequence of relatively compact sub-

sets of and =

.
(2) Each

has a C

plurisubharmonic dening function

such that

is uniformly bounded in and

n

i,j=1

z
i
z
j
a
i
a
j
c
0
[a[
2
for z

U and a C
n
,
where U is a neighborhood of bD and c
0
> 0 is a constant independent of .
(3) There exist positive constants c
1
, c
2
such that c
1
[

[ c
2
on

U,
where c
1
, c
2
are independent of .
We also have the following denition.
Denition 3.2. A bounded Lipschitz domain D in C
n
is called strictly
pseudoconvex if it has a strictly plurisubharmonic Lipschitz dening func-
tion, i.e., there exists a Lipschitz function : C
n
R such that is locally
a Lipschitz graph and
(1) < 0 in D, > 0 outside D and C
1
< [d[ < C
2
on bD almost
everywhere, where C
1
, C
2
are positive constants,
(2) C[z[
2
is plurisubharmonic for some C > 0 in U where U is a
neighborhood of b.
17
Lemma 3.3. Let D be a bounded Lipschitz domain in C
n
. Then D is strictly
pseudoconvex in the sense of Denition ***2.1.1 and Denition ***2.1.2 are
equivalent.
Proof. We rst assume that D is pseudoconvex in the sense of Denition
2.1.2. We will construct a sequence of subdomains D

and

. The proof is
similar to the proof of Lemma 0.3. Let D

and

be similarly dened as
in Lemma 0.3. It remains to prove that

satises (2) in Denition 2.1.1.

But this follows easily from the the fact that suitable regularizations of a
plurisubharmonic function are plurisubharmonic.
Let D be strictly pseudoconvex in the sense of Denition 2.1.1. Since
the sequence

is uniformly bounded in
1
on each compact subset of D,
we have from the Ascoli Theorem that there is a subsequence convergent
to some limit function
1
(D). Then is a Lipschitz dening function
which satises (1) and (2) in Denition 2.1.2. Thus the two denitions are
equivalent.

Some examples of strictly pseudoconvex domains with Lipschitz bound-

aries are given below.
1. Piecewise smooth strictly pseudoconvex domains.
A bounded domain in C
n
is said to have a piecewise smooth strictly
pseudoconvex boundary dened by C
2
-dierentiable functions if there
exists a nite open covering U
1
, . . . , U
k
of an open neighborhood U of
and C
2
strictly plurisubharmonic functions
j
: U
j
R, j = 1, . . . , k such
that
(i) U = x U[ for each 1 i k, either x , U
i
or
i
(x) < 0,
(ii) for 1 i
1
< i
2
< < i

k, the 1-forms d
i
1
, . . . , d
i
l
are linearly
independent over R at every point of

v=1
U
i
.
Then is a strongly pseudoconvex domain with Lipschitz boundary in
the sense of Denition 2.1.2. To see this, rst note that is a Lipschitz
domain from the assumption of transversal intersection. Let
i
be a C
2
strictly plurisubharmonic dening function for
i
and = z C
n
[
i
(z) <
0, i = 1, . . . , k. Set (z) = max
1
(z), . . . ,
k
(z). If
i
is only locally
dened, we dene for z U,
(z) =
i,zU
i
max
i
(z)
and extend to by (z) = max (z),
0
where
0
> 0 is suciently
small. Then is a plurisubharmonic function. Let W = z [
0
<
(z) < 0.
Since each
i
is strictly plurisubharmonic, there exists c
0
> 0 such that
n

i,j=1

z
i
z
j
a
i
a
j
c
0
[a[
2
for z U

and a C
n
, 1 k.
18
It follows that r
0
(z) = (z) c
0
[z[
2
is a plurisubharmonic function on W
since for each 1 k,

c
0
[z[
2
is a plurisubharmonic function on U

.
Thus (z) c
0
[z[
2
=
i,zU
i
max
i
(z) c
0
[z[
2
= r
0
(z) is plurisubharmonic
on W.
Since [[ = [
i
[ for some i on the smooth part of , we have [[ C
2
a.e. on . To show that [[ is bounded from below, we note that
is Lipschitz and satises the exterior cone condition. There exist a nite
covering V

1k
of , a nite set of unit vectors

1k
and c
1
> 0
such that the inner product , )

c
1
> 0 a.e. for z V

, 1 k.
t

> 0 are pseudoconvex Thus has a strictly plurisubharmonic Lipschitz

dening function and D is a strongly pseudoconvex Lipschitz domain.
2. Strictly convex domains.
Let be a strictly convex domain. By this we mean that there exists a
Lipschitz dening function such that C[z[
2
is convex for some C > 0 and
C
1
[[ C
2
a.e. on b. Since is convex, has Lipschitz boundary.
It is easy to see that is strongly pseudoconvex with Lipschitz boundary.
Let be a bounded Lischitz domain. We observe, rst of all, that
the rst order system

is elliptic in the interior of . More, if

f Dom(

) Dom(

), the f W
1
(, loc). Then the problem about
the regularity of

-Neumann operator is only on the boundary.
Theorem 3.4. Let a bounded strictly pseudoconvex domain in C
n
with
Lipschitz boundary. Then there exists C > 0 such that
3
(3.1) |f|
2
1/2
C
_
|

f|
2
+|

f|
2
_
, f Dom(

) Dom(

)
Now, we can state and prove the main result
4
.
Theorem 3.5. Let be a bounded strictly pseudoconvex domain with Lips-
chitz boundary. For any 0 p n, 1 q n1, the

-Neumann operator
N can be extended as a bounded operator N : W
1/2
(p,q)
() W
1/2
(p,q)
() and
we have
(3.2) |Nf|
2
1/2
C|f|
2
1/2
, f W

1
2
(p,q)
(),
where C is independent on f. Finally, we have
(3.3) |

Nf|
1/2
+|

Nf|
1/2
C|f|, f L
2
(p,q)
(),
3
In fact, we can prove before that, under the same assumptions
Z
b
|f|
2
d C
`

f
2
+

f
2

.
From this, the result of theorem follows.
4
A rst estimate in this sense was given by Kohn [KO], who showed that Nfs+1
Csfs, under the assumption of C

boundary.
19
Proof. We rst prove a priori estimates: Suppose that is a strictly pseu-
doconvex domain with C
2
boundary. For any f C
2
(p,q)
() Dom(

), we
will prove the estimate holds.
Let be a C
2
strictly plurisubharmonic dening function with [[ = 1
on b. By Greens Theorem
5
with u = and v = [f[
2
/2, we have
_

[f[
2
dV +'
_

ffdV +
_

[f[
2
2
dV =
_
b

n
[f[
2
2
d,
where v = '(

ff) +[f[
2
and = 0 on b. Since > 0 by the strictly
pseudoconvexity we have
_

[[[f[
2
dV C
__
b
[f[
2
d +
_

[f[
2
dV
_
+'
_

ffdV.
Using the fact that f = f and the Hardy-Littlewood Lemma
6
which give
the following estimate
|f|
2
1/2
C
__

[f[
2
dV +
_

[f[
2
dV
_
,
we have,
|f|
2
1/2
CQ(f, f) = C(f, f) C|f|
1/2
|f|
1/2
which implies
|f|
1/2
C|f|
1/2
.
Here Q(f, f) = (f, f) is the energy functional. Finally, substituting f with
Nf, we get
|Nf|
1/2
C|Nf|
1/2
= C|f|
1/2
.
This proves the a priori estimates.
5
Let u, v C
2
(), then
Z

(uv vu) dV =
Z
b

u
v
n
v
u
n

d.
6
We give here a modied version of Hardy-Littlewood Lemma for Sobolev spaces. Let
a bounded Lipschitz domain in R
N
and let (x) the distance function from x to
the boundary b. If u L
2
() W
1
(, loc) and there is a constant 0 < < 1 such that
Z

(x)
22
|u|
2
dV < ,
then u W

(). Furthermore, there is a constant C depending only on such that

u
2

C
Z

(x)
22
|u|
2
dV +
Z

|u|
2
dV

.
20
From the assumption of strong pseudoconvexity for , there exists a se-
quence of strongly pseudoconvex smooth subdomains

satisfying condi-
tions (1)-(3) in Denition (***). We have for any f C
2
(p,q)
(

) Dom(

),
_
b
[f[
2
dS C(|

f |
2

+ | f |
2

),
where C is independent of . From the a priori estimate,
| f |
2
1
2
()
C(|

f |
2

+ | f |
2

),
where C is independent of .
Since C
1
(p,q)
(

) Dom(

) is dense in Dom(

) Dom(

) in the graph
norm |

f |

+ |

f |

. The theorem is proved by an approximation

argument.

Assume that is a strictly pseudoconvex bounded domain with C

boundary. Then we have the following theorem of Kohn.

Theorem 3.6. Let be a bounded strictly pseudoconvex domain in C
n
with
C

boundary b. Then the

-Neumann operator N : W
s
() W
s+1
(),
for every s 0.
Proof. We only sketch the proof for s =
1
2
. If T is a tangential operator,
then Tf Dom(

), and we have
|f|
2
3/2

_

[[[Tf[
2
dV C
_
|

Tf|
2
+|

Tf|
2
_

C|Tf|
1/2
|Tf|
1/2
= C|f|
1/2
|f|
3/2
.
Then N : W
1/2
() W
3/2
() is bounded.
Furthermore, form the Sobolev embedding theorem
7
we have N : C

()
C

().
3.2. Regularity for the

-Neumann problem on pseudoconvex do-
mains. When is still with C

boundary, but only pseudoconvex, the

-Neumann operator is regular for the following type of domains:

1. If there exists a smoth dening function plurisubharmonic on the bound-
ary b. The regularity of N from W
s
() to itself was proved by Boas and
Straube.
2. If is of nite type in the sense of Kohn or DAngelo. The regularity of
N from W
s
() to W
s+
for some > 0 was proved by Kohn and Catlin.
When is still with C

boundary, but only pseudoconvex, in general,

N is not regular from W
s
() in itself. This was proved by Barrett for the
7
Sobolev Embedding Theorem: If is a bounded domain in R
N
with Lipschitz
boundary, then there is an embedding W
k
() C
m
() for an integer 0 m < k N/2.
21
worm domain of Diederich and Fornss [DF].
4. Strong Okas lemma and the

-Neumann problem
Denition 4.1. An open complex manifold is called Stein if there exists
a strictly plurisubharmonic exhaustion function : (, ) such that
(1) is strictly plurisubharmonic on ,
(2) For each c R,
c
= < c is relatively compact in .
A well known result in several complex variables is the classical Okas
Lemma.
Lemma 4.2 (Oka). Let be a pseudoconvex domain in C
n
, n 2, then
log is plurisubharmonic where is some distance function from z
to the boundary. Every pseudoconvex domain in C
n
is a Stein manifold.
Proof. Let be a distance function from z to the boundary b. Then
log is plurisubharmonic (near the boundary). We can modify to make
by adding C[z[
2
to be strictly plurisubharmonic in . The theorem is proved
by regularization.
If is a relatively compact domain in a Stein manifold, Hormanders L
2
existence theorem and Kohns results can all be applied to by using as
the weight function.
Denition 4.3. Let M be a complex hermitian manifold with the metric
form . Let be relatively compact pseudoconvex domain in M. We say
that a distance function to the boundary b satises the strong Okas
condition if it can be extended from a neighborhood of b to such that
satises
(4.1) i

(log ) c
0
in
for some constant c
0
> 0.
4.1. Strong Okas Lemma and bounded plurisubharmonic exhaus-
tion functions. We rst study the relation between the strong Okas Lemma
and the existence of bounded strictly plurisubharmonic functions on a pseu-
doconvex domain in a complex manifold with C
2
boundary.
For a bounded pseudoconvex domain with C
2
boundary in C
n
or in
a Stein manifold, a well known result by Diederich-Fornaess [DF] shows
that there exists a Holder continuous strictly plurisubharmonic exhaustion
function with Holder exponent 0 < < 1. We rst examine some equivalent
conditions for the existence of such bounded plurisubharmonic functions
based on the following simple observation.
Proposition 4.4. Let M be a complex hermitian manifold with metric
and let M be a pseudoconvex domain with Lipschitz boundary. Then
the following two conditions are equivalent:
22
(1) There exists some 0 < t
0
1 such that
(4.2) i

(log ) it
0

2
.
(2) There exists some 0 < t
0
1 such that
(4.3) i

(
t
0
) 0.
Furthermore, suppose that the distance function satises the strong Oka
condition (4.1). Then (2) implies
(3) There exists some 0 < t
0
1 such that for any 0 < t < t
0
, there exists
some constant C
t
> 0
(4.4) i

(
t
) C
t

t
( +i

2
).
In particular,
t
is a Holder continuous strictly plurisubharmonic exhaus-
tion function for .
Proof. If (1) holds, we have
(4.5) i

(log ) = i

()

+
i

2
it
0

2
.
The above equation is obviously true for C
2
distance function. It is also true
for Lipschitz function since the term

is well dened. Condition (2)

is equivalent to
i

()

+ (1 t
0
)
i

2
0.
Comparing this with (4.5), it is easy to see that (1) and (2) are equivalent.
Assume that the distance function satises the strong Oka condition
(4.1). To see that (2) implies (3), we multiply (4.5) by (1 ) and (4.1) by
. Adding the two inequalities, we conclude that, for any 0 1, the
inequality
i

(log ) = i

()

+
i

2
c
0
+ (1 )t
0
i

2
holds. Hence, for any 0 < t < t
0
, we choose =
t
such that (1
t
)t
0
> t.
Then
i

(
t
) = it
t
(

()

+ (1 t)

2
)
= it
t
(

(log ) t

2
)
C
t
t
t
( +
i

2
) (4.6)
where C
t
= min(c
0

t
, (1
t
)t
0
t). This gives that (2) implies (3).

23
Theorem 4.5. Let M be a complex hermitian manifold and let M
be a pseudoconvex domain with C
2
boundary b. Let (x) = d(x, b) be
the distance function to b with respect to the hermitian metric such that
satises the strong Oka condition (4.1). Then there exists 0 < t
0
1 such
that
i

(log ) it
0

2
.
Proof. Near a boundary point, we choose a special orthonormal basis w
1
, , w
n
for (1, 0)-forms such that w
n
=

2. Let L
1
, , L
n
be its dual. Let a
be any (1, 0)-vector. We decompose a = a

+ a

where a

= a, L
n
) is the
complex normal component and a

is the complex tangential component.

We have

(log ), a a)
=

()

, a

) + 2'

()

, a

)
+

()

, a

) +
[a

[
2

2
. (4.7)
From (4.1), we have

(log ), a

) =

()

, a

) c
0
[a

[
2
.
If is C
2
up to the boundary, we have for any > 0,
[

()

, a

)[ C(
1

[a

[
2
+
[a

[
2

2
).
Also near the boundary when (z) < , we have
[

()

, a

)[
C

[a

[
2
C

2
[a

[
2
.
Choosing suciently small, we have

(log ), a a)
1
2
[a

[
2

2
K[a

[
2
for some large constant K. Multiplying (4.1) by
K
c
0
and adding it to the
above inequality, we have
(
K
c
0
+ 1)

(log ), a a)
1
2
[a

[
2

2
.
This proves the theorem with t
0
=
1
2(
K
c
0
+1)
.

Corollary 4.6. Let , (x) and t

0
be the same as in Theorem 1.2. Then for
any 0 < t < t
0
, the function

=
t
is a strictly plurisubharmonic bounded
exhaustion function on .
24
The corollary follows immediately from the equivalence of (1) and (3) in
Lemma *** We have the following Diederich-Fornaess Theorem [DF] and
the recent generalization by Harrington [?].
Theorem 4.7. Let M be a pseudoconvex domain with Lipschitz
boundary in a Stein manifold M. Then there exists a Lipschitz dening
function and some number 0 < t < 1 such that

= ()
t
is a strictly
plurisubharmonic bounded exhaustion function on .
Proof. We only prove the case when b is of class C
2
. Since M is Stein,
M can be embedded in C
N
for some large N. Let (x) = d(x, b) be the
distance function to b with respect to metric induced by the Euclidean
metric in C
N
. From Okas Lemma, we have i

(log ) 0 in a neighbor-
hood U of b.
Let be a smooth strictly plurisubharmonic function on M. For any
c
0
> 0, we can choose some large > 0 such that
i

(log(e

)) = i

log + c
0

where is the metric form induced by the Euclidean metric. Thus the
strong Oka condition (4.1) holds. The theorem follows from Corollary ***.
The case for Lipschitz domain is much more involved and we omit the
details.

Remark. In the proof of Theorem ***, if is in C

n
, we can choose = [z[
2
.
Theorem 4.8. Let be a bounded pseudoconvex domain with Lipschitz
boundary in C
n
. Then the

-Neumann operator N (as well as

N,

N
and the Bergman projection P) is bounded on W
s
() to W
s
() for any
0 s <
1
2
t
0
.
Remark. 1. If t
0
= 1 in Theorem***, the domain has a plurisubhar-
monic dening function in . It is proved in Bonami-Charpentier [?] that
we can take s =
1
2
. if has smooth boundary and there exists a plurisub-
harmonic dening function on b, it follows from Boas-Straube [?] that the

-Neumann operator is bounded in the Sobolev space W

s
for all s > 0. We
also mention that for any > 0, there exists a smooth bounded pseudocon-
vex domain

in C
n
such that the

-Neumann operator is not bounded on
W
s
(see the paper by Barrett [?]).
When the complex manifold is Kahler with positive curvature, we have
also the following result of Ohsawa-Sibony [?].
Theorem 4.9. (Ohsawa-Sibony) Let CP
n
be a pseudoconvex domain
with C
2
boundary b and let (x) = d(x, b) be the distance function to b
with the Fubini-Study metric . Then there exists t
0
= t
0
() with 0 < t
0
1
such that
i

(
t
0
) 0.
25
This follows easily from Takeuchis Theorem (see [?], also [?])
i

(log ) c
0

where c
0
can be chosen to be equal to
1
2
.
Remarks:
1. The theorems above show that if either the complex manifold has
positive curvature or there is a positive line bundle, then the strong Okas
lemma holds. Both theorems do not hold without the positivity condition
on the metric (see the counterexample in [DF] and Theorem 1.2 in [?]).
2. If is a Lipschitz bounded pseudoconvex domain in a Stein manifold,
it is proved in Demailly [?] that there exists a bounded strictly plurisub-
harmonic exhaustion function in (see also Kerzman-Rosay [?] for the C
1
case). A more rened arguments yield a Holder continuous strictly plurisub-
harmonic exhaustion function in (see Harrington [?]).
It is not known if this is true for pseudoconvex domains with Lipschitz
boundary in CP
n
. We also remark that strictly plurisubharmonic bounded
exhaustion functions might not exist if the Lipschitz boundary (as a graph)
condition is dropped (see [DF]).
4.2. Strong Okas lemma and nite type conditions. Let be a
bounded pseudoconvex domain in C
n
and let be the Kahler form. The
Okas lemma characterizes pseudoconvex domains with the condition that
the distance function satisifes i

(log )
x
c(x) where c(x) 0 for
every x . If we can take c(x) as a postive constant, it is called the strong
Okas condition. In this section, we will relate the growth rate of c(x) to
more ner properties of psuedoconvexity and the

-Neumnann problem.
If is a strongly pseudoconvex Lipschitz domain, there exists a strictly
plurisubharmonic dening function such that
(4.8) i

(log ) i

()

C
1
,
in the sense of currents. In fact, we can use (4.12) as denition for strong
pseudoconvexity.
Denition 4.10. Let be a bounded pseudoconvex domain in C
n
with
Lipschitz boundary b. The domain is of nite type if there exist a
distance function and c > 0, 0 < 10 such that
i

(log ) c
2
.
Denition 4.11. Let be a bounded pseudoconvex domain in C
n
with
Lipschitz boundary b. The domain is said to satisfy condition (C) (for
compactness) if there exist a distance function and c(z) > 0 such that
i

(log ) c(z)
where where c(z) + as z b
26
Theorem 4.12. Let be a bounded pseudoconvex domain in C
n
with Lips-
chitz boundary b. Suppose that is of nite type, in the sense of *** then
the

-Neumann operator N on L
2
(p,q)
(), where 1 q n 1, satises
N : W
s
() W
s+2

(),

<
and
|f|

C(|

f| +|

f|), f Dom(

) Dom(

).
If satises condition (C), then N is compact.
5. The

-Neumann problem on Kahler manifolds with
nonnegative curvature
Let M be a complex manifold with a hermitian metric and let M
be a pseudoconvex domain. If M is Stein, the L
2
existence theorems for

and the

-Neumann problem follow from Hormanders theory. In the case
when the manifold M is K ahler with negative curvature, the distance func-
tion to a xed point is a strictly plurisubharmonic function and hence, M is
Stein (see [?]). In this section we study the L
2

theory when the manifold is
not Stein and there is no strictly plurisubharmonic weight function smooth
up to the boundary. One has to modify Hormanders weighted method to
establish the L
2
theory for

.
Suppose that the manifold has positive curvature, like CP
n
. Then there
exists some distance function for which satises the strong Okas con-
dition (0.1). Then we can use = log to be the weight function in
Hormanders theory and study the weighted

-Neumann problem (see [?] or
[CS]). However, is not continuous up to the boundary. To establish the
L
2
theory without weights, we use an idea by Berndtsson-Charpentier [?]
and streamlined in [CSW].
Let t be any real number and C
2
(). Let L
2
(
t
) denote the L
2
space
with respect to the weight function e
t
=
t
and
|f|
(t)
|f|
2
L
2
(
t
)
=
_

[f[
2
e
t
=
_

t
[f[
2
.
We use

t
to denote the adjoint of

with respect to the weighted space.
Then

t
=
t

t
whenever it is dened, where denotes the formal adjoint
with respect to the unweighted L
2
-norm. The norm|.|
2
(t)
is equivalent to the
Sobolev norm on a sub-space of W

t
2
() for harmonic functions or solutions
to elliptic equations.
Let be a distance function which satises the strong Oka condition (0.1).
Introduce the following two asymmetric weighted norms. These new norms
will be used to obtain more rened L
2
estimates.
For any (p, q)-form f on , we decompose f into complex normal and
tangential parts by setting f

= (f
(

)
#
)

and f

= f f

..
The above decomposition is well-dened for any (p, q)-form f supported
near the boundary and can be extended to the whole domain.
27
We dene the asymmetric weighted norm
[f[
2
A
= [f

[
2
+
[f

[
2
[[
2
and its dual norm
[f[
2
A
= [f

[
2
+[f

[
2
[[
2
.
For any t > 0, let L
2
A
(
t
) and L
2
A

(
t
) denote the weighted L
2
spaces on
(p, q)-forms dened by the norm
|u|
2
L
2
A
(
t
)
=
_

t
[f[
2
A
=
_

t
([f

[
2
+
[f

[
2
[[
2
)
and
|u|
2
L
2
A

(
t
)
=
_

t
[f[
2
A
=
_

t
([f

[
2
+[f

[
2
[[
2
).
Theorem 5.1. Let M be a complex Kahler manifold with nonnegative sec-
tional curvature. Let M be a pseudoconvex domain with C
2
-smooth
boundary b. Let (x) = d(x, b) be the distance function such that
satises the strong Oka condition (0.1). For any f L
2
A

(p,q)
(), where
0 p n and 1 q n, such that

f = 0 in , there exists u L
2
(p,q1)
()
satisfying

u = f and
_

[u[
2
C
_

[f[
2
A
.
Proof. We rst show that for any t > 0 and any (p, q)-form f L
2
A

(
t
),
0 p n and 1 q n, such that

f = 0 in , there exists u L
2
(p,q1)
(
t
)
satisfying

u = f and
(5.1) |u|
2
L
2
(
t
)

C
t
|f|
2
L
2
A

(
t
)
.
Let = t log , where t > 0. By the Bochner-Hormander-Kohn-Morrey
formula with weight function = t log , for any (p, q)-form g Dom(

)
Dom(

) with q 1 on , we have
|

g|
2

+|

g|
2

= |g|
2

+ (g, g)

+ ((i

)g, g)

+
_
b
(i

)g, g)e

,
where |u|
2

=
_

n
j=1
[D
L
j
u[
2
e

, L
1
, ..., L
n
is a local unitary frame of
T
(1,0)
() and is a curvature form. From our assumption, we have
(u, u)

0.
Since is pseudoconvex, we have that for any (p, q)-form g Dom(

)
Dom(

t
),
|

g|
2
(t)
+|

t
g|
2
(t)
((i

)g, g)
(t)
Ct|g|
L
2
A
(
t
)
.
28
Let (i

denote the dual norm for (p, q)-forms induced by i

. It
follows that for any f L
2
A

(
t
), there exist u L
2
(
t
) satisfying

u = f
and
(5.2)
_

[u[
2

u[
2
(i

1
Ct
_

u[
2
A

t
.
This proves (4.12). To get rid of t, we use an argument used in [?]. Let
f L
2
(p,q)
(). For any t > 0 , there exists u L
2
(p,q1)
(
t
) satisfying

u = f,
such that u is perpendicular to Ker(

) in L
2
(
t
) and u satises (4.13).
Consider v = u
t
. Then v Ker(

) in L
2
(
2t
). It follows from (4.13)
that the following holds:
(5.3)
_

[u[
2
=
_

[v[
2

2t

1
2C
0
t
_

v[
2
A

2t
.
Since
(5.4) [

v[
2
A

2t
C([

u[
2
A
+ 2t
2
[u[
2
),
choosing t suciently small and substituting (4.15) into (4.14), one obtains
_

[u[
2
C
t
_

u[
2
A
.
This proves the theorem.
Theorem 5.2. Let M be a complex Kahler manifold with nonnegative sec-
tional curvature. Let M be a pseudoconvex domain with C
2
-smooth
boundary b. Let (x) = d(x, b) be the distance function such that sat-
ises the strong Oka condition (0.1). Then
(p,q)
has closed range and the

-Neumann operator N
(p,q)
: L
2
(p,q)
() L
2
(p,q)
() exists for every p, q such
that 0 p n, 0 q n. Moreover, for any f L
2
(p,q)
(), we have
f =

N
(p,q)
f

N
(p,q)
f, 1 q n 1.
f =

N
(p,0)
f Pf, q = 0,
where P is the orthogonal projection from L
2
(p,0)
() onto L
2
(p,0)
() Ker(

)
and
N
(p,0)
=

N
2
(p,1)

.
Furthermore, there exists 0 < t
0
1 such that the

-Neumann operator
N,

N,

N and the Bergman projection P are exactly regular on W

s
(p,q)
()
for 0 s <
1
2
t
0
with respect to the W
s
()-Sobolev norms.
The L
2
-existence theorem for the

-Neumann operator N on follows
from the L
2
-existence of the solution u for the

-equation proved in Theorem
2.1.
To show that N is regular in W
s
for s <
1
2
t
0
, let t = 2s in the Bochner-
Hormander-Kohn-Morrey formula with weight function = t log . After
29
rearranging terms, we have for any (p, q)-form g Dom(

) Dom(

),
|

g|
2
(t)
+|

g|
2
(t)
2'(

g, g
(

t
)

)
= |g|
2
(t)
+ (g, g)
(t)
((i

(
t
)g, g).
(see Proposition 3.1 in [CSW]). Since for any > 0 we have
[2'(

g, g
(

t
)

)[
1

g|
2
(t)
+t
2
|
g

|
2
(t)
,
choosing small, we have from Corollary 4.6 that
|

g|
2
(t)
+|

g|
2
(t)
C
t
(|g|
2
(t)
+|
g

|
2
(t)
)
C
t
|g|
2
(t)
.
The rest of the proof is similar to the proof of Theorem 2 in [CSW], and
we omit the details. When C
n
, the Sobolev regularity for

N and
the Bergman projection have been obtained earlier in [?] (see also [?]).
Remarks: It is still not known if Theorems ***2.1 or 2.2 hold for pseudo-
convex domains with C
1
or Lipschitz boundary.
5.1. The

-Neumann problem on domains in CP
n
. We want to study
regularity and solvability of

on a complex manifold and not only a bounded
domain in C
n
. First of all, observe that if we consider all C
n
(as a complex
manifold), because it is unbounded we can have a L
2
theory as before. How-
ever C
n
can be covered by balls of radius R, for R which tends to innity. So
it is possible to have a L
2
theory locally on each in C
n
suciently large.
Then, if f L
2
(p,q)
(, loc) and

f = 0, for q 1, there is u L
2
(p,q1)
(, loc)
solution to

u = f in C
n
. Moreover, when f C

(p,q)
(, loc),

f = 0 and
is pseudoconvex, then there is u C

(p,q1)
(, loc) solution to

u = f (and
in particular is a domain of holomorphy).
We racall that X is a Stein manifold if and only if it is a complex sub-
manifold of C
N
, for some N. This is the same as to have a strictly plurisub-
harmonic exhaustion function . Then, on a Stein manifold existence of

-Neumann operator is guaranteed by a result of Boas and Straube [BS2].

If X = CP
n
, then it is not Stein. In fact, on a compact complex manifold
there arent no constant strictly plurisubhramonic exhaustion functions. We
are interested to study this particular case.
Let CP
n
be a pseudoconvex domain, i.e. for any p b there is
U p such that U is pseudoconvex. Let be a C
2
dening function for
. For any (0, q)-form f we have
|

f|
2
+|

f|
2
= |f|
2
+ (R
0,q
f, f) +
_
b
i

f, f)d.
30
Here, R
0,q
is the curvature form. In CP
n
, this coincides with the Ricci
curvature, so
R
0,q
= (n + 1)q > 0
Then we have
|

f|
2
+|

f|
2
> (n + 1)q|f|
2
.
When f is a (n, q)-form, R
n,q
0, then
|

f|
2
+|

f|
2
= |f|
2
+
_
b
i

f, f)d.
If we set (z) = dist(z, b) we have i

(log ) c, where c > 0, the

Fubini-Study metric of CP
n
. Then is Stein.
As in the case of bounded domain in C
n
, we want to start from a weighted
L
2
theory. Set then
t
= t log , t > 0. Then i

t
Ct, where
c > 0, t > 0. In particular we can study the L
2
weighted space with weight

t
(in this case e
t
=
t
) and we obtain, for all t > 0 the following estimate
|

f|
2
t
+|

t
f|
2
t

1
Ct
|f|
2
t
.
We have the following result.
Lemma 5.3. Let CP
n
a bounded doamin with C
2
boundary. Then
there is a 0 < t
0
1 such that
t
0
is plurisubharmonic, i.e. i

(
t
0
) 0.
Furthermore, for any 0 < t < t
0
we have
(5.5) i

(
t
) Ct
t
_
+

2
_
.
Theorem 5.4. Let CP
n
a bounded doamin with C
2
boundary. Then
there is a bounded operator N
p,q
: L
2
(p,q)
() L
2
(p,q)
() for any q 1, such
that:
(1) N = N = I on Dom();
(2) N,

N,

N and B = I

N

, the Bergman projection, are all bounded
from W
s
() to itself, for any 0 < s < t
0
/2.
Proof. We already know that
_

t
_
[

f[
2
+[

f[
2
_
dV C
_

t
[f[
2
dV.
By the estimate (5.5) we get
_

t
0
[f[
2
dV |f|
2
W
t
0
/2
()
.
Applying this two fact with

Nf and

Nf instead of f we obtain the

desired regularities.
31
6. Tangential Cauchy-Riemann equations and the Bochner
Martinelli kernel
7. CR manifolds and tangential Cauchy-Riemann equations
7.1. L
2
theory for
b
on strongly pseudoconvex CR manifolds. As
an example of CR manifolds, we may consider real orientable hypersurfaces
in C
n
. Let C
n
be a bounded domain with dening function . Its
boundary M = b = z C
n
: (z) = 0 with condition [(z)[ , = 0,
z M, is a CR manifold of dimension 2n1. Its CR structure is generated
by
L
j
:=
n

h=1
a
hj

z
h
, j = 1, . . . , n 1,
purely tangential, i.e. L
j
= 0, j = 1, . . . , n1. The systemoverlineL
j
1
j n 1 are the tangential Cauchy-Riemann equations on M.
If f is a holomoprhic function on and f C
1
(), then f satises
(7.1) L
j
f = 0, on M, j = 1, , n 1
Any function satisfying (5.1) is called a CR function.
For n = 2, let M = (z
1
, z
2
) C
2
: z
2
= [z[
2
, is the boundary of the
unbounded ball. If we parametrize M by (z
1
, t) and z
2
= t +i[z[
2
, we have

L =

z
1
iz
1

t
.
This is the Lewy operator. In [LE] he proved that the tangential Cauchy-
Riemann equation Lu = f is not solvable for a large class of functions f.
When n > 2, we have the system
L
j
u = f
j
, j = 1, . . . , n 1
with the compatibility conditions
L
j
f
h
L
h
f
j
= L
j
L
h
u L
h
L
j
u = [L
j
, L
h
]u = c

jh
L

u = c

jh
f

,
where [L
j
, L
h
] = c

jh
L

, which is given by condition (iii) in the denition of

CR manifolds. Dening

b
=

[
M
the previous system can be rewritten as
_

b
u = f

b
f = 0
.
More precisely, if we denote

b
= d[
T
0,1
(M)
, then the sequence
. . .
0,q1
(M)

0,q
(M)

0,q+1
(M)

b
. . .
is a complex, i.e.

2
b
= 0.

b
is the tangential Cauchy-Riemann operator.
We may equip CTM by the induced metric such that T
1,0
(M)T
0,1
(M),
and dene
b
as the adjoint of

b
. So we have the

b
-Laplacian

b
:=

b
+
b

b
:
0,q
(M)
0,q
(M).
32
It is well know that
b
is not elliptic. If N

(M) denotes the dual of T

1,0
(M)
T
0,1
(M) in CT(M), we have
CT
z
M = T
1,0
z
(M) T
0,1
z
(M) N
z
(M).
In particular, dim
R
N(M) = k, it is spanned by a vector eld T which can
be chosen to be purely imaginary, i.e. T = T.
We give an important denition.
Denition 7.1. Let (M, T
1,0
(M)) be a CR manifold of dimension 2n 1.
Let T N(M) and its dual, i.e. (T) = 1, (T
1,0
(M) T
0,1
(M)) = 0.
Then

z,
(L, L

) =
z
(L, L

)() := [L, L

], )
z
,
for every L, L

T
1,0
(M) is the Levi form in the direction .
We say that the Levi form satises condition Z(q) (0 q n 1) in the
direction , if
z,
has at least n q positive eigenvalues or at least q + 1
negative eigenvalues.
We say that the Levi form satises condition Y (q) if it satises condition
Z(q) for all directions .
Since there are only two directions for , it is easy to see that condition
Y (q) is equivalent to the following one:
there are max(n q, q + 1) eigenvalues with the same sign
or
there are min(n q, q + 1) pairs of eigenvalues with opposite signs.
Furthermore, condition Z(q) is equivalent to the estimate
|f|
2
1/2
=
_
M
[f[
2
d C
_
|

b
f|
2
+|
b
f|
2
_
.
Theorem 7.2. Let (M, T
1,0
(M)) be a compact CR manifold of dimension
2n 1. Suppose that M satises condition Y (q) for some 0 q n 1.
Then we have
|f|
2
1/2
C
_
|

b
f|
2
+|
b
f|
2
+|f|
2
_
,
for every f C
1
(0,q)
(M). Furthermore, there is an operator G
b
: L
2
(0,q)
(M)
L
2
(0,q)
(M) such that
(i)
b
G
b
= G
b

b
= I H
b
, where H
b
: L
2
(0,q)
(M) H
b
(0,q)
= Ker(
b
) =
L
2
(0,q)
(M) :

b
=
b
= 0 is the projection;
(ii) H
b
(0,q)
is nite dimensional;
(iii) G
b
is compact and |G
b
|
s+1
C|f|
s
.
Proof. We give an idea of the proof only. Suppose that L
1
, . . . , L
n
, L
1
, . . . , L
n
, T
is an orthonormal basis for CTM. For any f C
1
(0,q)
(M) we have
|

b
f|
2
+|
b
f|
2
= |

f|
2
+ (c
jh
Tf
j
, f
h
) +O
_
|f||f| +|f|
2
_
,
33
where |

f|
2
=

n1
j,h=1
|L
j
f
h
|
2
and
[L
j
, L
h
] = c
jh
T mod T
1,0
(M) T
0,1
(M).
Now the Y (q) condition assures that, if we put L
j
= X
j
+iY
j
, then the set
of vector elds X
j
, Y
j
: 1 j n 1 satises a nite type condition of
type 2, i.e. X
j
, Y
j
, [X
j
, Y
j
] generate all the tangent space of M. So we
have
(7.2) |Xf|
2
+|f|
2
= |

f|
2
+|

f|
2
+|f|
2
C|f|
2
1/2
,
where |

f|
2
=

n1
j,h=1
|L
j
f
h
|
2
. Furthermore we may prove that
(7.3) |Xf|
2
+[(Tf, f)[ C
_
|

b
f|
2
+|
b
f|
2
+|f|
2
_
.
This implies that
|f|1
2
C
_
|

b
f|
2
+|
b
f|
2
+|f|
2
_
.

7.2. L
2
theory for

b
on boundary of pseudoconvex domains. Let
be a bounded pseudoconvex domain in C
n
with smooth boundary b. We
study the global L
2
existence theorem for

b
on b. Let f be a (p, q)-form
with L
2
coecients, where 0 p n and 1 q < n. We study the global
solvability of
(7.4)

b
u = f in b.
When q < n 1, it is easy to see that for equation (5.1) to be solvable, f
must satisfy the condition
(7.5)

b
f = 0 in b.
When q = n 1, condition (5.2) is void and the compatibility condition is
given by
(7.6)
_
b
f h = 0, h L
2
(np,0)
Ker(

b
).
In fact, for all 1 q n 1, f must satisfy the compatibility condition
(7.7)
_
b
f h = 0, h L
2
(np,n1q)
Ker(

b
).
Lemma 7.3. For 1 q < n 1, if f satises (5.4), then f is

b
-closed.
Proof. From the assumption that q n2, we set h =

v for some smooth
v C

(np,n2q)
() Ker(

), then it is easy to see that

_
b
f h =
_
b
f

v = (1)
q
_
b

b
f v = 0.
Since v is any smooth form in , we have that

b
f = 0 on b.
We have the following L
2
existence theorem.
34
Theorem 7.4. Let be a bounded pseudoconvex domain in C
n
with smooth
boundary M. For every W
s
(p,q)
(M), where 0 p n, 1 q n 2
and s 0, such that
(7.8)

b
= 0 on M,
there exists u W
s
(p,q1)
(M) satisfying

b
u = on M.
When q = n 1, W
s
(p,n1)
(M) and satises
(7.9)
_
M
= 0, C

(np,0)
(M) Ker(

b
),
there exists u W
s
(p,n2)
(M) satisfying

b
u = on M.
Corollary 7.5. Let be a bounded pseudoconvex domain in C
n
with smooth
boundary M. Then

b
: L
2
(p,q1)
(M) L
2
(p,q)
(M), 0 p n, 1 q n1,
has closed range in L
2
.
We have the following strong Hodge decomposition theorem for

b
. Let
S
(p,0)
: L
2
(p,0)
(M) Ker(

b
)

S
(p,n1)
: L
2
(p,0)
(M) Ker(

b
).
The projection S
(0,0)
is the Szego projection operator.
Corollary 7.6. Let M be the boundary of a smooth bounded pseudoconvex
domain in C
n
, n 2. Then for any 0 p n, 0 q n 1, there
exists a linear operator G
b
: L
2
(p,q)
(M) L
2
(p,q)
(M) such that
G
b
is bounded and R(G
b
) Dom(
b
).
For any L
2
(p,q)
(M), we have
=
b

b
G
b

b
G
b
, if 1 q n 2,
=

b
G
b
S
(p,0)
, if q = 0,
=
b

b
G
b


S
(p,n1)
, if q = n 1.
If L
2
(p,q)
(M) with
b
= 0, where 1 q n 2 or L
2
(p,n1)
(M)
with

S
(p,n1)
= 0, then =
b

b
G
b
.
The solution u =

b
G
b
in (4) is called the canonical solution, i.e., the
unique solution orthogonal to Ker(
b
).
Theorem 7.7. Let D be a bounded pseudoconvex Lipschitz domain in C
n
with boundary bD. We assume that D has a Lipschitz dening function
which is plurisubharmonic in D. For every L
2
(p,q)
(bD), where 0 p
n, 1 q < n 1, such that

b
= 0,
there exists u L
2
(p,q1)
(bD) satisfying

b
u = in bD.
35
Furthermore, there exists a constant C depending only on the diameter
and the Lipschitz constant of D but is independent of such that
| u |
L
2
(p,q1)
(bD)
C | |
L
2
(p,q)
(bD)
.
When q = n1, we assume that D has a Lipschitz dening function which
is plurisubharmonic in a neighborhood of D. For every L
2
(p,n1)
(bD)
such that satises
_
M
= 0, for every C

(np,0)
(D) Ker(

),
the same conclusion holds.
Corollary 7.8. Let D be a bounded Lipschitz domain with a dening func-
tion which is plurisubharmonic in a neighborhood of D. The

b
operator
has closed range in L
2
(p,q)
(bD) for every 0 p n, 1 q n 1.
Since

b
is a closed, densely dened operator from one Hilbert space to
another, it has a Hilbert space adjoint, denoted by

b
. The

b
operator is
also a closed, densely dened operator (see Riesz-Nagy [?]). Let Ker(

b
)
and Ker(

b
) denote the kernels of

b
and

b
respectively. Then Ker(

b
) and
Ker(

b
) are both closed subspaces. Let R(

b
) and R(

b
) denote the range
of

b
and

b
respectively. We also have the following consequence from the
Main Theorem. Let the space of harmonic forms H
p,q
b
(bD) be dened by
H
p,q
b
(bD) = Dom(

b
) Dom(

b
) [

b
=

b
= 0.
Corollary 7.9. Corollary 2 (Hodge theory for

b
)Let D be a bounded Lip-
schitz domain with a dening function which is plurisubharmonic in a
neighborhood of D. The harmonic forms
H
p,q
b
(bD) = 0 for 0 p n, 1 q < n 1.
We have the following strong Hodge decomposition:
L
2
(p,0)
(bD) = Ker(

b
) R(

b
), 0 p n,
L
2
(p,q)
(bD) = R(

b
) R(

b
), 0 p n, 1 q < n 1,
L
2
(p,n1)
(bD) = R(

b
) Ker(

b
), 0 p n.
In particular, our results can be applied to the boundary of a strongly
pseudoconvex Lipschitz domain or the boundary of any convex Lipschitz
domain. When D is a smooth strongly pseudoconvex CR manifold, the

b
complex has been studied in the work of Kohn [?], Kohn-Rossi [?], Folland-
Stein [?] and Rothschild-Stein [?] for 1 q < n 1. Henkin [?] has derived
the kernel solutions for

b
on strongly pseudoconvex boundaries in C
n
for
1 q n 1. There are also numerous other results on

b
on boundaries
of smooth bounded strongly pseudoconvex domains. We refer the reader to
the book by Boggess [?] or Chen-Shaw [?] and the references therein.
When D is the boundary of a pseudoconvex domain with smooth bound-
ary, the L
2
and Sobolev estimates for

b
have been obtained by Shaw in
36
[?] for 1 q < n 1 and for q = n 1 in Boas-Shaw [?] (See also Kohn
[?]). The proof of the L
2
existence for

b
in these papers depends on the (at
least C
4
) smoothness of the boundary. Our results are the rst for the

b
complex on C
1
or Lipschitz boundaries.
The

b
operator corresponds to a system of rst order dierential equa-
tions with variable coecients. Since it is not possible to dene the mul-
tiplication of two distributions in general, one cannot simply dene

b
on
nonsmooth manifolds in the sense of distributions. However,

is always
well dened on any domain D in C
n
, regardless of the smoothness of the
boundary bD. When D is a bounded Lipschitz domain, the

b
complex can
be dened via duality by integration by parts and

, since Stokes theorem
still holds on any bounded Lipschitz domain. Our method can obviously be
extended to the boundaries of Lipschitz domains in complex manifolds. For
the sake of simplicity and clarity in presentation, we will only stay in C
n
.
The L
2
estimates and existence theorems of

b
have many applications.
The closed range property is also closely related to the embeddability of
CR structures. We note that

b
does not have closed range locally in the
top degree case (when q = n 1) on any smooth strongly pseudoconvex
boundaries, since it corresponds to the Lewy operator. Moreover, it does
not always have closed range for abstract smooth strongly pseudoconvex CR
structures. Burns [?] has showed that, for Rossis example [?] of a nonem-
beddable strongly pseudoconvex CR structure of real dimension three, the

b
operator does not have closed range in either the L
2
or C

sense.
7.3. Nonisotropic Sobolev and Holder spaces. Let (M, T
1,0
(M)) a CR
manifolds of dimension 2n+k and let L
1
, . . . , L
n
, L
1
, . . . , L
n
, T
1
, . . . , T
k
an
horthonormal frame for CTM. Dene
|Xf|
2
:=
n

j=1
_
|L
j
f|
2
+|L
j
f|
2
_
.
The nonisotropic Sobolev space W
1

(M) is dened as the space of function

u L
2
(M) such that the W
1

(M) norm
|u|
2
W
1

:= |Xu|
2
+|u|
2
is nite. The space W
1,p

(M) is dened in the same way, only substituting

the L
2
by the L
p
norm.
For any integer m > 1, the space W
m,p

is dened inductively as the space

of function u such that L
j
u, L
j
u W
m1,p

(M), for any j = 1, . . . , n. Thus,

u W
m,p

(M) if and only if u L

p
(M) and X
1
X
m
u L
p
(M) for any
X
j
T
1,0
(M) T
0,1
(M), 1 j m. The space W
1,p

(M) is dened as
the dual space of W
1,p

(M), where 1/p + 1/p

= 1 or, equivalently, as the

space of function f =

N
j=1
X
j
u
j
, X
j
T
1,0
(M) T
0,1
(M), u
j
L
p
(M).
37
The nonisotropic Holder space C

(M), 0 < < 1, is dened as follow:

u C

(M) if and only if

sup
(t)
[u((t)) u((0))[
[t[

C < ,
where is any C
1
curve in M such that

(t) T
1,0
(M) T
0,1
(M) and
[

(t)[ 1. We dene for any m > 1 the space C

m,

(M) inductively as
before.
As in the previous section, we have the following result.
Theorem 7.10. Let (M, T
1,0
(M)) be a compact CR manifold of dimension
2n+k, k > 0. Suppose that M satises condition Y (q) for some 0 q n.
Then, there is a compact operator G
b
: W
1,p

(M) W
1,p

(M) such that

b
G
b
= G
b

b
= I H
b
,
where H
b
: L
2
(0,q)
(M) Ker(
b
) is the projection as before. Furthermore, if
f C
m,

(M) and u = G
b
f then u C
m+2,

(M), for any m = 1, 0, 1, . . .,

0 < < 1. Finally, when f W
m,p

(M) then u W
m+2,p

(M), 1 < p < .

The main consequence of the previous theorem is the fact that the

b
problem is solvable in nonisotropic Sobolev and Holder spaces, as stated by
the following corollary.
Corollary 7.11. Let f L
2
(0,q)
(M), such that

b
f = 0 and fKer(
b
).
Then the function u =

b
G
b
f is in L
2
(0,q+1)
and is a solution to

b
u = f.
Furthermore if f C
m,

(M) then the solution u is in C

m+1,

(M).
What we stated in the previous results, may be written in terms of L
2
estimates. In fact, let
Q
b
(f, f) := |

b
f|
2
+|
b
f|
2
+|f|
2
= ((
b
+I)f, f)
denoting the energy functional. Then
[(Tf, f)[ +|Xf|
2
CQ
b
(f, f),
from which
|f|
2
W
1

(M)
CQ
b
(f, f) = C((
b
+I)f, f)
for any fKer(
b
). Then
(7.10) |f|
W
1

(M)
C|
b
f|
W
1

(M)
.
In the same way for estimates in the other nonisotropic spaces.
38
7.4. Campanato Space. We want to dene another important class of
function spaces, the so called Campanatos spaces. Let 0 < < 1, 1 p
and x
0
M. Let a measure on M and let B

(x
0
) the ball of radius
and center in x
0
. We say that a function f is in the space L
p,

(x
0
) if and
only if there is
0
> 0 such that f L
p
(B

0
(x
0
)) and there is a constant
C > 0 such that
_
1
(B

(x
0
))
_
B(x
0
)
[f f[
B(x
0
)
[
p
_
1/p
C

, 0 < <
0
.
This is the pointwise Holder spaces in the L
p
sense (for p = , this is the
pointwise nonisotropic Holder spaces).
To dene the higher order spaces L
p,m+

(x
0
), m N 1, we need
to proceed in the following way. We may choose coordinates x = (z, t) =
(x
1
, y
1
, . . . , x
n
, y
n
, t
1
, . . . , t
k
) called normal coordinates, such that
L
j
=

z
j

x
0
, T
h
=

t
h

x
0
, 1 j n, 1 h k.
We say that a polynomial P(z, t) in (z, t) is of degree m if
P(z, t) =

|I|+|J|+2|L|m
a
IJL
z
I
z
J
t
L
,
where I, J, L are multiindices.
Under this assumption we say that f L
p,m+

(x
0
) if and only if f is a
L
p
function and there is
0
> 0 and a polynomial of degree m, P
x
0
m
, such
that
_
1
(B

(x
0
))
_
B(x
0
)
[f P
x
0
m
[
p
_
1/p
C
m+
, 0 < <
0
.
The relation between this space and the ones introduced in the previous
section is stated in the the following lemma.
Lemma 7.12 (Campanato Embedding). For any 0 < < 1 and 1 p
we have
L
p,m+

(M) = C
m,

(M).
Before proving regularity of
b
on Campanato spaces, we want rst to
scale the equation for the
b
to the unit ball B
1
(x
0
). For any > 0, we set
x = (z,
2
t), the dilation. Let L

j
the operator dened by
L

j
(f(x)) = (L
j
f)(x), x B
1
(x
0
).
Then we have
L

j
L
0
j
=

z
j
+
k

=1
a

.
39
Now, the

b
and

b
associated to the CR structure T
1,0

generated by L

j
are
such that

b
u

0
b
u

b
u
0
b
u
_
as 0.
So, if M satises condition Y (q), we can restate all the results of the section
for the scaled situation, i.e for the CR manifold M

which approaches M,
obtain again the same kind of estimates. This leads us to the following
result.
Proposition 7.13. If f L
2,

(x
0
) and u W
1

(B

0
(x
0
)) is a solution to

b
u = f in B

0
(x
0
)), then u L
2,2+

(x
0
).
Proof. We will sketch the idea of the proof.
(1) It sucees to show that there exists a polynomial P of degree 2 in
normal coordinates (z, t) such that
_
1
B

(x
0
)
_
B(x
0
)
[u P[
2
_
1/2
C
2+
, 0 < <
0
.
Furthermore, we can discretize the estimates, i.e., it suces to prove for
= r
m
, m = 0, 1, 2, . . ., we get
_
1
B
r
m(x
0
)
_
B
r
m(x
0
)
[u P[
2
_
1/2
Cr
m(2+)
, 0 < r < r
0
.
(2) Given > 0, if f is small, i.e. |f|
2
< , for a given , then there is a
function h such that
0
b
h = 0 and
_
1
B

(x
0
)
_
B(x
0
)
[u h[
2
_
1/2
< , 0 < <
0
.
Then, h is of the form h = P +O(3), where P is a polynomial of degree 2.
This is proved by the compactness arguments since we already know that

b
is subelliptic in the interior. Thus we have the estimates for m = 0. The
estimates for m = 1, 2, are obtained by scaling and induction.
7.5. The Tangential Cauchy-Riemann Complex. Let M be a hyper-
surface in a complex manifold. The complex on , when restricted to
M, induces the tangential Cauchy-Riemann complex, or the
b
complex. In
fact, the tangential Cauchy-Riemann complex can be formulated on any CR
manifold. There are two dierent approaches in this setting. One way is
to dene the tangential Cauchy-Riemann complex intrinsically on any ab-
stract CR manifold itself without referring to the ambient space. On the
other hand, if the CR manifold is sitting in a complex manifold, the tan-
gential Cauchy-Riemann complex can also be dened extrinsically via the
ambient complex structure.
First, we assume that M is a smooth hypersurface in , and let r be
a dening function for M. In some open neighborhood U of M, let I
p,q
,
40
0 p, q n, be the ideal in
p,q
() such that at each point z U the ber
I
p,q
z
is generated by r and r, namely, each element in the ber I
p,q
z
can be
expressed in the form
rH
1
+r H
2
,
where H
1
is a smooth (p, q)-form and H
2
is a smooth (p, q 1)-form. Denote
by
p,q
()[
M
and I
p,q
[
M
the restriction of
p,q
() and I
p,q
respectively to
M.
We dene the quotient bundle

p,q
(M) =
p,q
()[
M
/I
p,q
[
M
.
We denote by c
p,q
the space of smooth sections of
p,q
(M) over M, i.e.,
c
p,q
(M) = (M,
p,q
(M)). Let denote the following map
(7.11) :
p,q
()
p,q
(M),
where is obtained by rst restricting a (p, q)-form in to M, then
projecting the restriction to
p,q
(M). One should note that
p,q
(M) is
not intrinsic to M, i.e.,
p,q
(M) is not a subspace of the exterior algebra
generated by the complexied cotangent bundle of M. This is due to the
fact that r is not orthogonal to the cotangent bundle of M.
The tangential Cauchy-Riemann operator

b
: c
p,q
(M) c
p,q+1
(M)
is now dened as follows: For any c
p,q
(M), pick a smooth (p, q)-form
1
in that satises
1
= . Then,
b
is dened to be
1
in c
p,q+1
(M).
If
2
is another (p, q)-form in such that
2
= , then

2
= rg +r h,
for some (p, q)-form g and (p, q 1)-form h. It follows that
(
1

2
) = rg +r g r h,
and hence,
(
1

2
) = 0.
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
(M)

b
c
p,1
(M)

b


b
c
p,n1
(M) 0.
For the intrinsic approach, we will assume that (M, T
1,0
(M)) is an ori-
entable CR manifold of real dimension 2n 1 with n 2. A real smooth
manifold M is said to be orientable if there exists a nonvanishing top degree
form on M.
We dene the quotient bundle
E = CT(M)/T
0,1
(M).
41
Then E is a (holomorphic) vector bundle of complex dimension n. If M is a
real hypersurface in a complex manifold , then E is isomorphic to T
1,0
()
restricted to M.
We dene the holomorphic p-forms on E by
p
. Dene the vector bundle
C
p,q
(M), 0 p n, 0 q n 1, by
C
p,q
(M) =
p

q
T
0,1
(M) = C
q
(M,
p
).
It follows that C
p,q
(M) dened in this way is intrinsic to M. Denote
by (
p,q
the space of smooth sections of C
p,q
(M) over M, i.e., (
p,q
(M) =
(M, C
p,q
(M)). We dene the operator

b
: (
p,q
(M) (
p,q+1
(M)
as follows: If (
p,0
,
b
is dened by

b
, L) = L)
for any section L of T
0,1
(M). Then
b
is extended to (
p,q
(M) for q > 0 as
a derivation. Namely, if (
p,q
(M), we dene

b
, (L
1
L
q+1
))
=
1
q + 1
_
q+1

j=1
(1)
j+1
L
j
, (L
1

L
j
L
q+1
))
+
i<j

(1)
i+j
, ([L
i
, L
j
]L
1

L
i

L
j
L
q+1
))
_
.
Here by

L we mean that the term L is omitted from the expression.
One should note how the integrability condition of the CR structure
T
1,0
(M) comes into play in the denition of
b
, and it is standard to see
that the following sequence
0 (
p,0
(M)

b
(
p,1
(M)

b


b
(
p,n1
(M) 0,
forms a complex, i.e.,
2
b
= 0.
When the CR manifold (M, T
1,0
(M)) is embedded as a smooth hyper-
surface in a complex manifold with the CR structure T
1,0
(M) induced
from the ambient space, the tangential Cauchy-Riemann complex on M can
be dened either extrinsically or intrinsically. These two complexes are dif-
ferent, but one can easily show that they are isomorphic. Thus, if the CR
manifold is embedded, we shall not distinguish the extrinsic or intrinsic def-
initions of the tangential Cauchy-Riemann complex. The operator
b
is a
rst order dierential operator, and one may consider the inhomogeneous

b
equation
(7.12)
b
u = f,
where f is a (p, q)-form on M. Equation (***) is overdetermined when
0 < q < n 1. Since
2
b
= 0, for equation (***) to be solvable, it is
42
necessary that

b
f = 0.
Let L
1
, , L
n1
be a local basis for smooth sections of T
1,0
(M) over some
open subset U M, so L
1
, , L
n1
is a local basis for T
0,1
(M) over U.
Next we choose a local section T of CT(M) such that L
1
, , L
n1
, L
1
, , L
n1
and T span CT(M) over U. We may assume that T is purely imaginary.
Denition 7.14. The Hermitian matrix (c
ij
)
n1
i,j=1
dened by
(7.13) [L
i
, L
j
] = c
ij
T, mod (T
1,0
(U) T
0,1
(U))
is called the Levi form associated with the given CR structure.
The Levi matrix (c
ij
) clearly depends on the choices of L
1
, , L
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 L
1
, , L
n1
and T. Hence, after changing T to T, it makes sense to consider positive
deniteness of the matrix (c
ij
).
Denition 7.15. The CR structure is called pseudoconvex at p M if the
matrix (c
ij
(p)) is positive semidenite after an appropriate choice of T. It
is called strictly pseudoconvex at p M if the matrix (c
ij
(p)) is positive
denite. If the CR structure is (strictly) pseudoconvex at every point of
M, then M is called a (strictly) pseudoconvex CR manifold. If the Levi
form vanishes completely on an open set U M, i.e., c
ij
= 0 on U for
1 i, j n 1, M is called Levi at.
Theorem 7.16. Let D C
n
, n 2, be a bounded domain with C

bound-
ary. Then D is (strictly) pseudoconvex if and only if M = bD is a (strictly)
pseudoconvex CR manifold.
Proof. Let r be a C

dening function for D, and let p bD. We may

assume that (r/z
n
)(p) ,= 0. Hence,
L
k
=
r
z
n

z
k

r
z
k

z
n
, for k = 1, , n 1,
form a local basis for the tangential type (1, 0) vector elds near p on the
boundary. If L =

n
j=1
a
j
(/z
j
) is a tangential type (1, 0) vector eld near
p, then we have

n
j=1
a
j
(r/z
j
) = 0 on bD and L = (r/z
n
)
1

n1
j=1
a
j
L
j
43
on bD. Hence, if we let = r r, we obtain
n1

i,j=1
c
ij
a
i
a
j
=
n1

i,j=1
, [L
i
, L
j
])a
i
a
j
=
n1

i,j=1
(L
i
, L
j
) L
j
, L
i
) 2d, L
i
L
j
))a
i
a
j
=
n1

i,j=1
4r, L
i
L
j
)a
i
a
j
= 4

r
z
n

2
r, L L)
= 4

r
z
n

2 n

i,j=1

2
r
z
i
z
j
a
i
a
j
,
which gives the desired equivalence between these two denitions. This
proves the theorem.

We note that, locally, a CR manifold in C

n
is pseudoconvex if and only
if it is the boundary of a smooth pseudoconvex domain from one side.
Lemma 7.17. Any compact strongly pseudoconvex CR manifold (M, T
1,0
(M))
is orientable.
Proof. Locally, let ,
1
, ,
n1
be the one forms dual to T, L
1
, , L
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

1

1

n1

n1
.
It is not hard to see that the 2n1 form generated by other bases will dier
only by a positive function. Hence, a partition of unity argument will give
the desired nowhere vanishing 2n 1 form on M, and the lemma is proved.

7.6. Folland-Stein-Tanaka formula for

b
. Let M be a strongly pseudo-
convex CR manifold with the pseudohermitian Levi metric. The

b
complex

b
: (
0,q
(M) (
0,q+1
(M), 0 q n 1.
The formal adjoint of

b
under the L
2
norm, denoted by
b
, is dened by
the requirement that
(
b
f, g) = (f,

b
g) 1
for compactly supported smooth forms.
44
Proposition 7.18. Let e
1
, , e
n1
be an orthonormal frame eld for T
1,0
(M)
in an open neighborhood in M and let w
1
, , w
n1
be its dual. Then

b
=
n1

j=1
w
j

e
j
. 2
Proposition 7.19. Let M be a strongly pseudoconvex CR manifold with
the pseudohermitian Levi metric. Let e
1
, , e
n1
be an orthonormal frame
eld for T
1,0
(M) in an open neighborhood in M and let w
1
, , w
n1
be its
dual. Then

b
=
n1

j=1
( e
j
)
e
j
. 3
Proof. Let be a (0, q)-form and be a (0, q + 1)-form on M. We have

b
, ) = ,
b
) +

4
where is a (0,1)-form dened by =
j
w
j
with ( e
j
) = , e
j

) and

is the function dened to be

=
n1

e
j

j
.
Let be a (2n 2)-form dened by
=
n1

j
e
j
dV .
We rst prove the following claim:
d() =

dV. 5
To see this, let X =

j

j
e
j
be the dual of . Let A
X
be the (1,1) tensor
on M dened by
A
X
= L
X

X
.
Since dV = 0, we have
L
X
dV = A
X
dV.
Let
1
, ,
2n1
be a basis for T
x
M. Since A
X
is a derivative which maps
every function into zero, we have
L
X
dV (
1
, ,
2n1
) = A
X
dV (
1
, ,
2n1
)
= A
X
(dV (
1
, ,
2n1
))

j
dV (
1
, , A
X

j
, ,
2n1
)
=

j
dV (
1
, , A
X

j
, ,
2n1
)
= TraceA
X
dV (
1
, ,
2n1
)
Thus we have
L
X
dV = TraceA
X
dV. 6
45
Recall the torsion *** formula
A
X
Y = L
X
Y
X
Y =
Y
X T(X, Y ). 7
where T is the torsion tensor of . Now we choose a special frame for
CT
x
(M) with e
j
=
j
+ i
n+j1
, e
j
, where j = 1, , n 1 and
2n1
= ,
the basic vector eld,. Then we have
T(e
j
, e
k
) = (e
i
, e
k
) =
1
i

jk

and
T(Y, ) T
1,0
+T
0,1
for all Y T
1,0
+ T
0,1
. Thus the torsion term in (7) has no contribution
when taking trace of A
X
. Also note that
X
Y preserves types for X and
Y in T
1,0
+T
0,1
and = 0. We have
TraceA
X
=
e
j
X, e
j
) =

and
TraceA
X
dV =

dV. 8
For each vector eld X, the divergence formula is dened to be
(div)XdV = L
X
dV = d(XdV ), 9
we have proved from (6)-(9) that

dV = L
X
dV = d().
Integrating with respect to the volume element dV in (4), we have proved
the proposition.

We dene the

b
-Laplacian

b
=

b
+
b

b
.
Theorem 7.20. Folland-Stein-Tanaka formula for
b
Let (M, T
1,0
(M) be
a strongly pseudoconvex CR manifold and ba a basic eld. Let e
1
, , e
n1
be an orthonormal frame eld for T
1,0
(M) in an open neighborhood in M
and let w
1
, , w
n1
be its dual.

b
=
j

2
e
j
e
j
+
q
i

i,j

w
k
( e
j
)K
e
j
e
k

i,j

w
k
( e
j
)R
e
j
e
k
, 4

b
=
j

2
e
j
e
j

(n q 1)
i

i,j

w
k
( e
j
)K
e
j
e
k
, 5

b
=
n q 1
n 1

j

2
e
j
e
j

q
n 1

j

2
e
j
e
j

i,j

w
k
( e
j
)K
e
j
e
k

n q 1
n 1

i,j

w
k
( e
j
)R
e
j
e
k
6
where

2
XY
=
X

X
Y
46
is the second covariant dierential and
R
XY
=
X

Y
+
Y

X

[X,Y ]
=
2
Y X

2
XY
is the curvature tensor, both extended C-linearly to (p, q)-forms, K is the
curvature for the holomorphic vector bundle
p
.
Proof.

b
=
j

( e
j
)
e
j
(
k

w
k

e
k
)
=
j,k

( e
j
) w
k

e
j

e
k
=
j

e
j

e
j
+
j,k

w
k
( e
j
)
e
j

e
k
.

b
=
k

w
k

e
k
(
j

( e
j
)
e
j
) =
j,k

w
k
( e
j
)(
e
k

e
j
)
since
e
k
(e
j
) =(e
j
)
e
k
.
From the Riccis formula, we have
[
e
j
,
e
k
] = R
e
j
e
k
K
e
j
e
k
D
T(e
i
, e
k
)
where T is the torsion tensor of and
T(e
j
, e
k
) = (e
i
, e
k
) =
1
i

jk

.
Thus we obtain

b
=
j

e
j

e
j
+
j,k

w
k
( e
j
)[
e
j
,
e
k
]
=
j

e
j

e
j
+
1
i
q

i,j

w
k
( e
j
)K
e
j
e
k

i,j

w
k
( e
j
)R
e
j
e
k
which proves (4). (5) follows from (4) and the Ricci formula and

j
K
e
j
e
j
=
0. (6) follows from (4) and (5).

Pseudo-hermitian manifolds with constant coecients

1. The Heisenberg group, the at case.
2. The sphere in C
n
, positive curvature.
3. The hyperbolic case.
8. Applications of

and

b
There are many applications from the solutions of the Cauchy-Riemann
equations and the tangential Cauchy-Riemann equations. We will give ap-
plications in the following areas.
47
8.1. The Levi problem. Let D 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 D is a domain of holomorphy. Near each boundary
point p bD, one must nd a holomorphic function f(z) on D which cannot
be continued holomorphically near p. This problem is called the Levi prob-
lem for D at p. It involves the construction of a holomorphic function with
certain specic local properties. From a well-known results of Benhke and
Stein, it suces to show that a strongly pseudoconvex domain is a domain
of holomorphy.
If the domain D is strongly pseudoconvex with C

boundary bD and
p bD, one can construct a local holomorphic function f in an open neigh-
borhood U of p, such that f is holomorphic in U D, f C(D U p)
and f(z) as z D approaches p. In fact f can be easily obtained as
follows: let r be a strictly plurisubharmonic dening function for D and we
assume that p = 0. Let
F(z) = 2
n

i=1
r
z
i
(0)z
i

i,j=1

2
r
z
i
z
j
(0)z
i
z
j
.
F(z) is holomorphic, and it is called the Levi polynomial of r at 0. Using
Taylors expansion at 0, there exists a suciently small neighborhood U of
0 and C > 0 such that for any z D U,
ReF(z) = r(z) +
n

i,j=1

2
r
z
i
z
j
(0)z
i
z
j
+O([z[
3
) C[z[
2
.
Thus, F(z) ,= 0 when z D U 0. Setting
f =
1
F
N
for some positive integer N, it is easily seen that f is locally a holomorphic
function which cannot be extended holomorphically across 0. We can require
that f is not locally L
2
near 0 by choosing N large.
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
(U) and = 1 in a neighborhood of 0. We note that
f is not holomorphic in D. However, if f can be corrected by solving a

-equation, then the Levi problem will be solved.

Let us consider the (0,1)-form g dened by
g =

(f) = (

)f.
This form g can obviously be extended smoothly up to the boundary. It is
easy to see that g is a

-closed form in D and g C

(0,1)
(D). If we can nd
a solution u C

(D) such that

(8.1)

u = g in D,
48
then we dene for z D,
h(z) = (z)f(z) u(z).
It follows that h is holomorphic in D, h C

(D 0) and h is singular
at 0. Thus one can solve the Levi problem for strongly pseudoconvex do-
mains provided one can solve equation (6.1) with solutions smooth up to the
boundary. From Kohns solution (or Hormanders L
2
solution), we can take
u =

N

(f) and u C

(D) and h = f u is a holomorphic function

with singularity at 0. Notice that h = f

N

(f) = Pf where P is the
Bergman projection on D.
This gives that any strictly pseudoconvex domain is a domain of holo-
morphy. From a well known results of Benhke and Stein, any pseudoconvex
domain is a domain of holomorphy since it can be exhausted by strictly
pseudoconvex domains. This gives solution of the Levi problem. Notice
that Kohns method works even for pseudoconvex domains in complex man-
ifolds.
8.2. Regularity for the Bergman projection. Let D be a bounded pseu-
doconvex domain in C
n
, we consider the Bergman projection
(8.2) P : L
2
(D) L
2
(D) Ker(

).
From the Hormanders L
2
theory for

, we have the strong Hodge decom-
postion for for L
2
(0,1)
(D). For any f L
2
(0,1)
(D), we have
f = Nf
where N is the

-Neumann operator.
This gives that =

on functions has closed range also. Thus we have

for any f L
2
(D), Notice that Ker() is the same as Ker(

) for functions.
We have
f = N
0
f Pf.
Since

f =

N
0
f
and
N
1

f = N
1
(

)N
0
f = N
1
(

N
0
f,
we have
N
1

f =

N
0
f
Thus N
0
=

N
0
=

N
1

, this give the Kohns formula for the Bergman

projection :
(8.3) Pf = f

N
1

f. f L
2
(D).
Similarly, we have that
(8.4) N
0
f = (I P)N
0
f =

N
1

N
0
f =

N
2
1

f.
49
Theorem 8.1 (Kohn). Let D be a bounded strongly pseudoconvex domain
in C
n
with smooth boundary. Then the Bergman projection P : C

(D)
C

(D).
Theorem 8.2 (Feerman). Let D
1
and D
2
be two bounded strongly pseu-
doconvex domains in C
n
with smooth boundary. Suppose that f is a biholo-
moprhic map from D
1
to D
2
. Then f extends smoothly to the boundary
bD
1
.
Denition 8.3 (Condition R). Let D be a bounded strongly pseudoconvex
domain in C
n
with smooth boundary. Then D is said to satisfy condition R
if the Bergman projection P : C

(D) C

(D).
Theorem 8.4 (Bell-Ligocka). Let D
1
and D
2
be two bounded pseudoconvex
domains in C
n
with smooth boundary. Suppose that f is a biholomoprhic
map from D
1
to D
2
and suppose that one of the domains satises condition
R. Then f extends smoothly to the boundary bD
1
.
It was already known to Poincare that the ball in C
2
is not biholomorphic
to the bidisc.
Corollary 8.5 (Bell-Ligocka). Let D
1
be a bounded strongly pseudoconvex
domains in C
n
with smooth boundary and D
2
be a bounded weakly pseudocon-
vex domain with smooth boundary. Then D
1
and D
2
are not biholomorphic.
We remark that condition R does not hold on Diederich-Fornaess worm
domains (see Barrett [?] and Christ [?]).
8.3. The Newlander-Nirenberg theorem.
Denition 8.6. Let M be a real (2n + k)-dimensional manifold. M is a
CR manifold if there is a subbundle T
1,0
(M) of CTM, the so called CR
structure, such that:
(i) dim
C
T
1,0
z
(M) = n, for any z M;
(ii) T
1,0
z
(M) T
0,1
z
(M) = (0), for any z M, where T
0,1
z
(M) = T
1,0
z
(M);
(iii) [L, L

(T
1,0
(M)), for every L, L

(T
1,0
(M)).
When k = 0 in the previous deniton, M is an almost complex manifold.
Theorem 8.7. (Newlander-Nirenberg) Let (M, T
1,0
) be an almost complex
manifold of class C
k
(U) C
k,
(x
0
) in a neighborhood U of a point x
0
in M
where k 1 and 0 < < 1. Assume that the almost complex structure T
1,0
is (formally) integrable. Then there exist a neighborhood V U of x
0
and
holomorphic coordinates of class C
k
(V ) C
k+1,
(x
0
) in V which embed V
into C
n
.
Proof. The proof is exactly as in [?] combined with the pointwise interior
Holder regularity for elliptic operators.
Let L
1
, , L
n
be a local basis for smooth sections of T
1,0
(M). Let
x
1
, , x
2n
be the real coordinates for M and we write z
j
= x
j
+ ix
n+j
.
50
We can using compatibility condition, after a quadratic change of coordi-
nates, assume that
L
i
=

z
i
+
n

j=1
a
ij

z
j
, i = 1, , n,
where the a
ij
s are C
1,
functions and a
ij
(0) = 0 for all i, j = 1, , n.
At the origin, L
i
is the constant coecient operator /z
i
. Let L

i
=

z
i
+

n
j=1
a
ij
(x)

z
j
, i = 1, , n, where > 0 is small. Then T
0,1

=
L

1
, , L

n
) denes an almost complex structure that is integrable for each
<
0
for some suciently small
0
> 0.
Let

denote the Cauchy-Riemann complex associated with the almost

complex structure T
0,1

equipped with a Hermitian metric. Then the exis-

tence and regularity theory developed for

in the previous section on any
complex manifold can be applied to M with

substituted by

. Let =

n
i=1
[z
i
[
2
= [x[
2
, is a strictly plurisubharmonic function near 0. Choosing

0
suciently small, we may assume that = x R
2n
[ [x[
2
< 1 = B
1
and is strongly pseudoconvex with respect to the almost complex structure
T
0,1

(M). Using the L

2
existence results for

(see [?] or [?]), the

-Neumann
operator N

exists on and there exists a solution u

i
=

z
i
on such
that

i
=

z
i
and
(8.5) | u

i
|

C |

z
i
|

where C can be chosen uniformly for <

0
. Since
L

i
z
j
= a
ij
(x),
we have
D

z
i
= O()
for any D

= (/x
1
)

1
(/x
2n
)

2n
, where the
i
s are nonnegative
integers with [[ k. It is easy to check that
[

z
i
[
C
k,
(0)
0 if 0.
Let

i
= z
i
u

i
. From the interior Holder regularity for N

and (**), we
have
[u

i
[
C
k+1,
(0)
C([

z
i
[
C
k,
(0)
+ | u

i
|) 0 if 0.
We have that

i
= 0 in and also d

i
(0) = dz
i
du

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

i
to by setting

i
=

i
(x/), then we have that

i
= 0 and d
i
are linearly independent in
provided we choose suciently small. This proves the theorem.

Remarks (1) In the Newlander-Nirenberg Theorem above, we only need

the assumption on Holder condition C
k,
to be just at one point x
0
. (2) If
(M, T
1,0
) is of real dimension 2, then we can relax the assumption by requir-
ing only k 0 since there is no compatibility condition to be satised. This
51
is the result obtained in Bers and Chern (see [?, ?]). Again, the assumption
on Holder condition C

is only required at one point.

8.4. Embedding strongly pseudoconvex compact CR manifolds. In
this section we rst extend the embedding theorem of Boutet De Monvel [?]
to CR manifolds of class C
3,
.
Theorem 8.8 (Boutet De Monvel). Let (M, T
1,0
) be a compact strongly
pseudoconvex CR manifold with real dimension 2n + 1, n 2, of class
C
3,
, where 0 < < 1. Then there exist global CR functions h
1
, , h
N

C
2,

(M) such that

b
h
i
= 0 for every i = 1, , N and = (h
1
, , h
N
) :
M C
N
is an embedding.
Proof. The arguments are similar to the smooth case used in [?]. Since
T
1,0
(M) is of class C
3,
, we equip M with a metric which is of class C
3,
.
For each point p M, using a polynomial change of coordinates, we can
choose C
3
coordinates x = (z, t) = (z
1
, , z
n1
, t) near a neighborhood U
of p = 0 such that a basis for T
1,0
vector elds is given by
L
j
=

z
j
+
n1

i=1
A
ji

z
i
+ (

1z
j
+B
j
)

t
, j = 1, , n 1
where both A
ji
= O([x[
2
) and B
j
= O([x[
2
). We choose
Z(z, t) = Z(x) = (z
1
, , z
n
),
where z = (z
1
, , z
n1
) and z
n
= t + i[z[
2
. Then Z : U C
n
Then we
have
(8.6) L
j
z
i
= O([x[
2
), j = 1, , n 1, i = 1, , n.
Thus the map Z gives an approximate embedding near p. Extending z
j
to be a C
3
function
j
on M, we have functions
1
, ,
n
C
3
(M)
such that
j
(p) = 0, d
1
(p), , d
n
(p) are linearly independent at p and
= (
1
, ,
n
) : M C
n
is a C
3
dieomorphism of a small neighborhood
of p on M into C
n
with (p) = 0 and (M) is strongly pseudoconvex at the
origin.
Let g be a polynomial g(z, t) = iz
n
+z
2
n
. Then g satises dg(p) ,= 0 and

b
g(p) = O([x[
2
). Choosing U small, we have
(8.7) Reg = [z[
2
+t
2
([z[
2
)
2
c[x[
2
,
where c is a positive constant. Extend g to M such that g C
3
(M) and g
satises Reg > 0 on M p.
Let be a cut-o function such that = 1 near a neighborhood V U
of p and is supported in U. To construct CR embedding functions, we
set z

j
=
j
e
g
for suciently large > 0. Notice that Theorems 3.1,
3.2 and Corollary 3.3 hold for any metric of class C
3
, not necessarily the
pseudo-hermitian Levi metric. Let G
b
be the inverse of
b
. Set
h
j
= o(z

j
) = z

j

b
G
b

b
z

j
for j = 1, , n,
52
where o is the Szego projection on M. We have
b
h
j
= 0 in M for j =
1, , n. Since z

j
is in C
3
(M) C
2,

(M), from Corollary 3.6 we have

h
j
C
2,

(M) for any 0 < < 1.

Using (4.1) and (4.2), it follows that in a neighborhood U near p
[
b
z

j
[ = [

b
(
j
)e
g

j
e
g
(
b
g)[
C([x[
2
+[x[
3
)e
|x|
2
C(
1
+

1
2
)
v>0
sup([v[
1
e
v
+[v[
1+
1
2
e
v
) C

1
2
0, .
This gives that |
b
z

j
| 0 as . Since

b
z

j
0 in C
1,
(0) as
, we have
b
G
b

b
z

j
in C
2,

(0) can be made arbitrarily small as

. This gives dh
1
(0), , dh
n
(0) are linearly independent for large .
The map = (h
1
, , h
n
) forms a local embedding of M into C
n
by global
C
2,

CR functions. From the same arguments as in [?], we also have that

global CR functions separating points. From the compactness of M and a
partition of unity, there exists a global embedding map consisting of CR
functions. This proves the theorem.
We remark that Boutet De Monvel s theorem does not hold for strongly
pseudoconvex CR manifold of real dimension 3.
We shall present in this section a compact real analytic three dimensional
CR manifold which can not be globally CR embedded into C
n
for any di-
mension n. Let S
3
= (z
1
, z
2
) C
2
[ [z
1
[
2
+ [z
2
[
2
= 1 be the boundary of
the unit ball in C
2
, and let the induced CR structure T
1,0
(S
3
) be gener-
ated by L = z
2
(/z
1
) z
1
(/z
2
). Thus, (S
3
, T
1,0
(S
3
)) forms a compact
strongly pseudoconvex CR manifold of real dimension three. For each t R,
[t[ < 1, dene a new CR structure T
1,0
t
(S
3
) on S
3
by letting T
1,0
t
(S
3
) be
generated by the vector eld L
t
= L +tL. If t = 0, T
1,0
0
(S
3
) coincides with
the induced standard CR structure T
1,0
(S
3
). It is easily veried that for
[t[ < 1, (S
3
, T
1,0
t
(S
3
)) is a compact real analytic strongly pseudoconvex CR
manifold of real dimension three.
The next theorem shows that any L
2
integrable CR function f on S
3
with respect to the CR structure (S
3
, T
1,0
t
(S
3
)) for 0 < [t[ < 1 must be
even. Obviously, this implies that, for 0 < [t[ < 1, (S
3
, T
1,0
t
(S
3
)) can not be
globally CR embedded into any C
n
.
Theorem 8.9 (Rossi). Any L
2
integrable CR function f(z) on S
3
with
respect to the CR structure T
1,0
t
(S
3
), 0 < [t[ < 1, is even, i.e., f(z) = f(z).
Theorem 8.10 (Burns-Kohn). Any strongly pseudoconvex CR structure
T
(0,1
(S
3
) of real dimension 3 is embeddable if and only if S
t
; C

(S
3
)
C

(S
3
). This is equivalent to that the range of

b
t is closed.
53
8.5. Nonexistence of Levi-at hypersurfaces in CP
n
.
Denition 8.11. Let M be a C
2
hypersurface in CP
n
. M is said to be Levi
at if for any C
2
dening function for M, the Levi form i

[ vanishes on
T
1,0
(M)
Denition 8.12. A Lipschitz hypersurface is a hypersurface which locally
is the graph of a Lipschitz function. A Lipschitz (or C
1
) hypersurface is
said to be Levi-at if it is locally foliated by complex manifolds of complex
dimension n 1.
From the implicit function theorem, any C
1
hypersurface locally is the
graph of some C
1
function. A C
2
hypersurface M is called Levi-at if its
Levi-form vanishes on M. Any C
k
Levi-at hypersurface, k 2 is locally
foliated by complex manifolds of complex dimension n 1. The foliation is
of class C
k
if the hypersurface is of class C
k
, k 2 (see Barrett-Fornaess
[?]). The proof in [?] also gives that if a real C
1
hypersurface admits a
continuous foliation by complex manifolds, then the foliation is actually C
1
.
Thus our denition is a natural generalization of Levi-atness to Lipschitz
or C
1
hypersurfaces.
Theorem 8.13. There exist no Lipschitz Levi-at hypersurfaces in CP
n
for
n 3.
For example in CP
2
, let M = [z
0
, z
1
, z
2
] CP
2
: [z
0
[ = [z
1
[. It is possi-
ble to show that CP
2
M =
+

, where both domains are pseudoconvex.

The, M can be foliated (locally) by compact manifolds, i.e. M U =

t
,
where
t
are complex manifolds.
Our proof follows the work by Siu [?]. Let M be a Lipschitz Levi at
hypersurface in CP
n
such that CP
n
M =
+

, where both domains

are pseudoconvex. As before M is locally foliated by complex manifolds.
Suppose that M is dened globally by a Lipschitz function with ,= 0
on M almost everywhere.
If we consider the curvature form
N
of the complex line bundle generated
by complex normal , then
N
[
t
> 0. This is due to the fact that on the
submanifold
t
, the curvature is no more than the curvature of the ambient
manifold CP
n
. The complex line bundle generated by is a quotient
bundle, its curvature is more positive then that of CP
n
. If we can nd a
real-valued h such that
(8.8)
N
= i
b

b
h on M
with a continuous function h, then h must assume its maximum at some
point p. But this implies that h is strictly plurisubharmonic in the interior
of one of the leaves
t
, which is not possible by the maximum principle.
This prove the nonexistence of Levi at hypersurfaces.
Since the curvature form
N
is already known to be d-exact, we can
reduced the problem to solving a continuous solution u such that
(8.9)

b
u = f,
54
when f C

(M) with >

1
2
.
Now come back to regularity. To obtain the regularity for the solution,
we need to extend rst to in CP
n
, to get W
1+
(CP
n
). Next, we
need to nd

v

0
=

on

such that

= v

0
on

. Then, we will
have

= 0,

[
b
= 0.
If we set v

0
=

we obtain what we request. Moreover,
v

0


= 0 on b

and v

0
W

). So we can solve

u = v

0
on all
CP
n
, to conclude that u W
1+
CP
n
. By the Trace theorem, this leads to
u = u[
M
W
1/2+
(M).
Let (z

, t) be the local coordinates on each leaf

t
. Notice that the tan-
gential Cauchy-Riemann operator

b
is nothing but the Cauchy-Riemann
equation

b
u =

z
u = on each leaf with W
1/2+
(M). Since u
W
1/2+
(M), the restriction of u on each leaf
t
is a function with W
s
(
t
)
for each t. Thus we can apply the interior regularity to obtain that u(z

, t) is
Holder continuous on each leaf. We also obtain that u is uniformly bounded
on M. It remains to show that u is continuous in the transversal direction
t. This amounts to a one-dimensional Sobolev embedding theorem. We
modify an argument of nite dierence methods in the Besov norms used
by Hormander to conclude that u actually is H older continuous.
9. Open problems
Theorem 9.1 (Boundary regularity for biholomorphic mappings). Let D
1
and D
2
be two bounded domains in C
n
with smooth boundary. Suppose that
f is a biholomoprhic map from D
1
to D
2
. Then f extends smoothly to the
boundary bD
1
.
Theorem 9.2 (Embedding strongly pseudoconvex CR structure of dimen-
sion 5). Let (M, T
1,0
) be a strongly pseudoconvex CR manifold with real
dimension 2n + 1, n 2. Then there exist local CR functions h
1
, , h
n

C

(M) such that

b
h
i
= 0 for every i = 1, , n and = (h
1
, , h
n
) :
M C
n
is an embedding.
Theorem 9.3 (C

regularity for

in CP
n
). Let CP
n
be a pseu-
doconvex domain with C

boundary. For any f C

(0,1)
() with

f = 0,
there exists u C

() such that

u = f.
When is in C
n
or a Stein manifold, this is proved by Kohn [?]. In fact,
one would like to know if there exists a solution u W
1
().
Theorem 9.4 (Bounded holomorphic functions on negatively curved man-
ifolds). Let (M, g) be a complete simply connected Kahler manifold of com-
plex dimension n 3 with negative sectional curvature a
2
< k < b
2
with
some nonzero a, b. Then there exist bounded holomorphic functions on M.
55
10. Appendix
10.1. Preliminaries for Lipschitz domains. We rst give some basic
properties of Lipschitz domains in Chapter 0. These simple and useful facts
do not seem to have been systematically treated.
A bounded domain R
n
is called Lipschitz if locally the boundary
b is the graph of a Lipschitz function. Let : R
n1
R be a function
which satises the Lipschitz condition
(10.1) [(y

) (x

)[ M[y

[, for all y

, x

R
n1
.
A bounded domain is called Lipschitz if near every boundary point p bD,
there exists a neighborhood U of p such that after a rotation,
U = (x

, x
n
) U [ x
n
> (x

)
for some Lipschitz function . The smallest M in which (3.1) holds will
be called the bound of the Lipschitz constant. By choosing nitely many
balls U
i
covering b, the Lipschitz constant for a Lipschitz domain is the
smallest M such that the Lipschitz constant is bounded by M in every ball
U
i
. A Lipschitz function is dierentiable almost everywhere (See Evans-
Gariepy [?] for a proof of this fact).
Let be a bounded Lipschitz domain. A Lipschitz function is called
a global dening function for if : R
n
R satises < 0 in D, > 0
outside D and
(10.2) C
1
< [d[ < C
2
a.e. on bD,
where C
1
, C
2
are positive constants.
Lemma 10.1. Let be a bounded Lipschitz domain in R
n
. Then D has a
global Lipschitz dening function .
Proof. We cover bD by nitely many boundary coordinate patches U
i
where
i = 1, , K. Let r
i
be a local dening function on U
i
which is locally
a Lipschitz graph. Let
i
C

0
(U
i
) be a partition of unity such that

i
= 1 in a neighborhood of bD. We dene =

i

i
r
i
. Then is a
dening function for D. To see that satises (0.2), it is easy to see that
[[ C
2
a.e. on bD. Since D is Lipschitz, for each p bD such that the
tangent plane of p for bD exists, one can nd such that r
i
, )
p
C
0
> 0
for every i such that p U
i
. Also we have =

i

i
r
i
on bD if r
i
exists. Thus
, )
p
=

i
r
i
, )
p

i
(p)C
0
= C
0
.
This proves that satises (3.2).
Lemma 10.2. If is any global Lipschitz dening function for D and r is
a local dening function in a neighborhood U of a boundary point p bD
56
such that r(x) = x
n
(x
1
, , x
n1
) for some Lipschitz function , then
there exists a positive function h(x) L

(D U) C(D U) such that

(10.3) (x) = h(x)r(x), x D U,
(10.4) d(x) = h(x)dr(x), a.e. x bD U
where

C
1
h(x)

C
2
, a.e. on bD
for some positive constants

C
1
and

C
2
.
Proof. Since r(x) = (, 1), we have
C
1
[dr(x)[ C
2
a.e. on U bD.
Dene
h(x) =
(x)
r(x)
, x U D.
Then h(x) is Lipschitz on D and h(x) > 0 in U D. Let p be any point
in U bD such that both d and dr exist at p. Then there exists a conic
neighborhood with vertex 0 R
n
such that for any unit vector ,
r, )
p
C
0
> 0. This implies that
(10.5)

[[
, )
p

C
0
[r[
> 0, p U bD.
Since r and are dierentiable at p, we have for any x +p U,
(10.6)
(x)
r(x)
=
(x) r(p)
r(x) (p)

D

(p)
D

r(p)
.
Thus h(x) has nontangential boundary value a.e. on U bD. Also from
(3.5) and (3.6), there exist

C
1
> 0 and

C
2
> 0 such that

C
1
h(x)

C
2
, a.e. for x U bD.
(3.4) follows from the denition of dierentiation.

Lemma 10.3. Let be a bounded Lipschitz domain in R

n
. There exists an
exhaustion

of such that

is an increasing sequence of relatively compact subsets of and

= .
Each

has a C

dening function

, i.e.,

= x R
n
;

(x) < 0.
There exist positive constants c
1
, c
2
such that c
1
[

[ c
2
on

, where
c
1
, c
2
are independent of .
Proof. Using Lemma 0.1, there exists a global Lipschitz dening function
for satisfying (0.2) and is obtained by a partition of unity of dening
functions r
i
which is a Lipschitz graph. Let C

0
(R
n
) be a function such
that 0,
_
dV = 1, (x) depends only on [x[ and vanishes when [x[ > 1.
57
We dene

(x) =
1

_
x

_
for > 0. Let

0 and we dene

=
x [ (x) <

. Then

is a sequence of relatively compact open

subsets of with union equal to . Each

is well dened if 0 <

<
+1
for x

+1
. Letting c
2
= sup

[[, then for

suciently small, we have

(x) <
(x)
< (x) +c
2

on

+1
. For each we choose

=
1
2c
2
(
1

)
and t

(
+1
,

). We dene

= x R
n
[

< t

.
Since (x) <

(x) < t

<
+1
, we have that

+1

. Also
if x

1
, then

(x) < (x) + c

2

<

< t

. Thus we have

+1

1
and (1) is satised. Each

is dened by

+t

.
That the subdomain

has smooth boundary will follow from (3).

To prove (3), it is easy to see that [

[ c
2
in bD

. To show that [

[
is uniformly bounded from below, we note bD satises the uniform interior
cone property. Then there exists a conic neighborhood with vertex 0 R
n
such that for any unit vector + p, , )
p
> C
0
a.e. in U bD,
where C
0
is a positive constant independent of p if U is suciently small.
There exist a nite covering V

1K
of , a nite set of unit vectors

1K
and c
1
> 0 such that the inner product ,

) c
1
> 0 a.e.
for x V

, 1 K. Since this is preserved by convolution, (3) is proved.

This proves Lemma 0.3.
10.2. Preliminaries for Lipschitz domains. We rst give some basic
denitions and properties of Lipschitz domains.
A bounded domain R
n
is called Lipschitz if locally the boundary
b is the graph of a Lipschitz function. Let : R
n1
R be a function
which satises the Lipschitz condition
(10.7) [(y

) (x

)[ M[y

[, for all y

, x

R
n1
.
A bounded domain is called Lipschitz if near every boundary point p bD,
there exists a neighborhood U of p such that after a rotation,
U = (x

, x
n
) U [ x
n
> (x

)
for some Lipschitz function . The smallest M in which (3.1) holds will
be called the bound of the Lipschitz constant. By choosing nitely many
balls U
i
covering b, the Lipschitz constant for a Lipschitz domain is the
smallest M such that the Lipschitz constant is bounded by M in every ball
U
i
. A Lipschitz function is dierentiable almost everywhere (See Evans-
Gariepy [?] for a proof of this fact).
58
Let be a bounded Lipschitz domain. A Lipschitz function is called
a global dening function for if : R
n
R satises < 0 in D, > 0
outside D and
(10.8) C
1
< [d[ < C
2
a.e. on bD,
where C
1
, C
2
are positive constants.
Lemma 10.4. Let be a bounded Lipschitz domain in R
n
. Then D has a
global Lipschitz dening function .
Proof. We cover bD by nitely many boundary coordinate patches U
i
where
i = 1, , K. Let r
i
be a local dening function on U
i
which is locally
a Lipschitz graph. Let
i
C

0
(U
i
) be a partition of unity such that

i
= 1 in a neighborhood of bD. We dene =

i

i
r
i
. Then is a
dening function for D. To see that satises (3.2), it is easy to see that
[[ C
2
a.e. on bD. Since D is Lipschitz, for each p bD such that the
tangent plane of p for bD exists, one can nd such that r
i
, )
p
C
0
> 0
for every i such that p U
i
. Also we have =

i

i
r
i
on bD if r
i
exists. Thus
, )
p
=

i
r
i
, )
p

i
(p)C
0
= C
0
.
This proves that satises (3.2).
Lemma 10.5. If is any global Lipschitz dening function for D and r is
a local dening function in a neighborhood U of a boundary point p bD
such that r(x) = x
n
(x
1
, , x
n1
) for some Lipschitz function , then
there exists a positive function h(x) L

(D U) C(D U) such that

(10.9) (x) = h(x)r(x), x D U,
(10.10) d(x) = h(x)dr(x), a.e. x bD U
where

C
1
h(x)

C
2
, a.e. on bD
for some positive constants

C
1
and

C
2
.
Proof. Since r(x) = (, 1), we have
C
1
[dr(x)[ C
2
a.e. on U bD.
Dene
h(x) =
(x)
r(x)
, x U D.
Then h(x) is Lipschitz on D and h(x) > 0 in U D. Let p be any point
in U bD such that both d and dr exist at p. Then there exists a conic
neighborhood with vertex 0 R
n
such that for any unit vector ,
r, )
p
C
0
> 0. This implies that
(10.11)

[[
, )
p

C
0
[r[
> 0, p U bD.
59
Since r and are dierentiable at p, we have for any x +p U,
(10.12)
(x)
r(x)
=
(x) r(p)
r(x) (p)

D

(p)
D

r(p)
.
Thus h(x) has nontangential boundary value a.e. on U bD. Also from
(3.5) and (3.6), there exist

C
1
> 0 and

C
2
> 0 such that

C
1
h(x)

C
2
, a.e. for x U bD.
(3.4) follows from the denition of dierentiation.

Lemma 10.6. Let be a bounded Lipschitz domain in R

n
. There exists an
exhaustion

of such that
(1)

is an increasing sequence of relatively compact subsets of and

= .
(2) Each

has a C

dening function

, i.e.,

= x R
n
;

(x) < 0.
(3) There exist positive constants c
1
, c
2
such that c
1
[

[ c
2
on

,
where c
1
, c
2
are independent of .
Proof. Using Lemma 3.1, there exists a global Lipschitz dening function
for satisfying (3.2) and is obtained by a partition of unity of dening
functions r
i
which is a Lipschitz graph. Let C

0
(R
n
) be a function such
that 0,
_
dV = 1, (x) depends only on [x[ and vanishes when [x[ > 1.
We dene

(x) =
1

_
x

_
for > 0. Let

0 and we dene

=
x [ (x) <

. Then

is a sequence of relatively compact open

subsets of with union equal to . Each

is well dened if 0 <

<
+1
for x

+1
. Letting c
2
= sup

[[, then for

suciently small, we have

(x) <
(x)
< (x) +c
2

on

+1
. For each we choose

=
1
2c
2
(
1

)
and t

(
+1
,

). We dene

= x R
n
[

< t

.
Since (x) <

(x) < t

<
+1
, we have that

+1

. Also
if x

1
, then

(x) < (x) + c

2

<

< t

. Thus we have

+1

1
and (1) is satised. Each

is dened by

+t

.
That the subdomain

has smooth boundary will follow from (3).

To prove (3), it is easy to see that [

[ c
2
in bD

. To show that [

[
is uniformly bounded from below, we note bD satises the uniform interior
cone property. Then there exists a conic neighborhood with vertex 0 R
n
such that for any unit vector + p, , )
p
> C
0
a.e. in U bD,
where C
0
is a positive constant independent of p if U is suciently small.
There exist a nite covering V

1K
of , a nite set of unit vectors
60

1K
and c
1
> 0 such that the inner product ,

) c
1
> 0 a.e.
for x V

, 1 K. Since this is preserved by convolution, (3) is proved.

This proves Lemma 0.3.
References
[BO] BOGGESS, A., CR manifolds and the tangential Cauchy-Riemann complex, Stud-
ies in Advanced Mathematics, CRC Press, Boca Raton, FL, 1991.
[BS1] BOAS, H.P. & STRAUBE, E.J., Equivalence of regularity for the Bergman pro-
jection and the -Neumann operator, Manuscripta Math. 67(1990), no. 1, 25-33.
[BS2] BOAS, H.P. & 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. Z. 206(1991), no. 1, 81-88.
[BS3] BOAS, H.P. & STRAUBE, E.J., Sobolev estimates for the complex Green op-
erator on a class of weakly pseudoconvex boundaries Comm. Partial Dierential
Equations 16(1991), no. 10, 1573-1582.
[CSW] CAO J., SHAW M.C. & WANG L., Estimates for

-Neumann problem and nonex-
istence of C
2
Levi at hypersurfaces in CP
n
,
[CS] CHEN, S.C. & SHAW, M.C., Partial dierential equations in several complex
variables, AMS-International Press., Studies in Advanced Mathematics, volume
19, Providence, Rhode Island, 2001.
[DF] DIEDERICH, K. & FORNSS, J.E., Pseudoconvex domains: an example with
nontrivial Nebenh ulle, Math. Ann. 225(1977), no. 3, 275-292.
[FS] FOLLAND, G.B. & STEIN, E.M., Estimates for the
b
complex and analysis on
the Heisenberg group, Comm. Pure Appl. Math. 27(1974), 429-522.
[H

O] H

ORMANDER, L., Hypoelliptic second-order dierential equations, Acta Math.,

119(1967),
[KO] KOHN, J.J., Boundaries of complex manifolds, Proc. Conf. Complex Manifolds
(Minneapolis, 1964), Springer-Verlag, New York, (1965), 81-94.
[LE] LEWY, H., An example of a smooth linear partial dierential equation without
solution, Ann. Math., 66(1957), 155-158.
[PS] PHONG, D.H. & STEIN, E.M., Hilbert integrals, singular integrals, and Radon
transforms, II, Invent. Math. 86(1986), no. 1, 75-113
[SW] SHAW, M.C. & WANG, L., H older and L
p
estimates for
b
on CR manifolds of
arbitrary codimension, Math. Ann. 331(2005), no. 2, 297-343.