Kernels of Hankel and Toeplitz Operators

Jim-Felix Lobsien
Date: August 4, 2011
1
CONTENTS 2
Contents
1 List of Symbols 3
Introduction 4
1.1 Foreword . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Preliminaries 5
2.1 Preliminary Introduction into the Topic . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 An Introduction to H
p
Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 The Poisson kernel and its mollier properties . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 A digression on harmonic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.5 Boundary Behaviour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.6 Back to Analytic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.7 Projections from L
p
to H
p
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3 Necessary Tools 18
3.1 Jensens Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Factorization for functions in H
p
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.3 The Nevanlinna class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.4 The forward and backward shift on H
p
, for p [1, ) . . . . . . . . . . . . . . . . . . . 23
4 Kernels of Hankel operators 24
4.1 The symbols of the Hankel operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2 Description of the kernels of Hankel operators . . . . . . . . . . . . . . . . . . . . . . . . 25
5 Kernels of Toeplitz operators 26
5.1 The symbols of Toeplitz operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.2 Description of the kernels of Toeplitz 0perators . . . . . . . . . . . . . . . . . . . . . . . 28
6 Appendix 36
6.1 Factorization for functions in H
p
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.2 Theorems used in this paper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
1 LIST OF SYMBOLS 3
1 List of Symbols
1. T denotes the unit circle T := z C : [z[ = 1.
2. D denotes the unit disk D := z C : [z[ < 1.
3. N
0
are the natural numbers including the zero.
4. N are the natural numbers without the zero.
5. L
p
(T) is the set of all measurable functions from T to C whose absolute value raised to the p-th
power has nite integral.
6. w

is the weak* topology.

7. H
p
(D) is the Hardy Space in the disc, see Denition 2.2.1.
8. H

(D) is the space of bounded analytic functions in D, see Denition 2.2.2.

9. {
r
() is the Poisson kernel, see Denition 2.2.3.
10. H
p
(T) is the Hardy Space on the unit circle, see Denition 2.16.1.
11. h
p
(D) is the harmonic Hardy Space in the disc, see Denition 2.3.1.
12. dm :=
|dz|
2
denotes the normalized Lebesgue measure on the unit circle.
13. P
+
denotes the analytic Riesz projection and P

the antianalytic Riesz projectioin, see Denition

2.17.1.
1 LIST OF SYMBOLS 4
Introduction
1.1 Foreword
In this paper, we aim to discuss the kernels of Hankel and Toeplitz operators.
In order to introduce the topic we start with the Hilbert case L
2
, where we dene the Hankel and
Toeplitz operators in a relatively straightforward way.
We use the L
2
case to give a preliminary illustration of the topic, however we have to resort back to
the actual denition of the Hardy space in the disc in order to proceed further with our investigations.
Hereafter we need to talk in terms of harmonic functions in order to make a connection with the L
p
-
space for 0 p on the unit circle. By further investigation while returning to analytic functions
we arrive at a point where we have an isomorphism between the analytic functions in the disc and L
p
functions on the unit circle. Therefore we can view the Hardy space as a closed subspace of L
p
and we
nd ourselves with a natural projection from L
p
to H
p
. This gives rise to the denition of the famous
Hankel and Toeplitz operators.
In order to look at the kernels we require some further components, which are dealt with in Section
3. There we explain briey the factorization of H
p
functions, Jensens inequality, the denition of the
Smirnov and Nevanlinna class and the forward and backward shift operator.
In Section 4 we look at the kernels of the Hankel operator, where we focus on the two questions:
1. For which symbols are the kernels non-trivial?
2. What are the kernels?
We solve both questions (almost) completely. In fact, the answer provided to Question 1 in the Toeplitz
case is not quite explicit, but we shall explain where the diculties lie when looking for a complete
solution; see Section 5. We then follow the theory of Konstantin Dyakonov and Stephan Ramon Garcia
to give a precise set of the kernels of Toeplitz operators.
In Section 6 we go back to the factorization of H
p
functions, nally giving a complete derivation of
this complex topic. To conclude, we list without proofs a few theorems that are used in this paper.
2 PRELIMINARIES 5
2 Preliminaries
2.1 Preliminary Introduction into the Topic
With the aim to do a preliminary introduction into the topic we are beginning with the space L
2
=
L
2
(T, dm) on the unit circle with respect to the normalized Lebesgue measure.
As it is customary, we abuse the terminology and view L
p
, p [1, ] as a space of functions rather
than a space of equivalence classes of functions. Therefore, two functions in L
p
are equal, when they
are equal almost everywhere.
Since L
2
is a Hilbert space, an orthogonal basis exists and it is proved in [1], Proposition 2.1. that the
set
e
in
: n Z with [0, 2]
forms an orthonormal basis for L
2
. Let us recall the denition of L
2
and the corresponding Fourier
series:
Denition 2.0.1 1. A measurable function f on T with respect to dm() from T to C belongs to
L
2
if
|f|
L
2 :=

f(e
i
)

2
dm()
1
2
< .
2. The corresponding Fourier series of f is given by
f(e
i
) =

nZ

f(n)e
in
with

f(n) =

T
f(e
i
)e
in
dm().
Remark
Due to the Riesz-Fischer theorem the series converges in L
2
-norm if and only if the function is square
integrable and Lennart Carleson proved that the Fourier expansion of any function in L
2
converges
almost everywhere.
Therefore we have a new but equivalent denition of L
2
.
L
2
:=

f(e
i
) =

nZ
a
n
e
in
: a
n
C,

nZ
[a
n
[
2
<

.
With the aid of this denition we are able to dene a closed subspace of L
2
.
Denition 2.0.2 The space
H
2
:=

f L
2
:

f(n) = 0 for n < 0

is called the Hardy-Hilbert space on the circle.

Remark The space H
2
is closed since it is the closed linear span of combinations of e
in

nN
0
, where
the coecients are square summable. Hence it is a Hilbert space.
We are going to use now the most useful feature of a Hilbert space; the fact that each closed subspace
of L
2
has a natural complementary closed subspace.
Denition 2.0.3 The orthogonal complement of H
2
is dened as

H
2

= f L
2
: (f, g) = 0 g H
2
0

where ( , ) is the inner product dened as

(f, g) :=

nZ
a
n
b
n
with f(e
i
) =

nZ
a
n
e
in
and g(e
i
) =

nZ
b
n
e
in
.
2 PRELIMINARIES 6
Consequently the space L
2
is the direct sum of the two subspaces (L
2
= H
2
H

).
Later on we will concentrate on the space L
p
:= L
p
(T, dm) on the circle with respect to normalized
Lebesgue measure which is unfortunately not a Hilbert space anymore for p = 2. Therefore the
orthogonal compliment does not exist and we will deal with the complex conjugate of H
2
, where the
zero Fourier coecient vanishes.
H
2
0
:=

f : f H
2
,

f(0) = 0

.
To illustrate that both denitions are equivalent, we state the following
Lemma 2.1

H
2

= H
2
0
.
Proof "" Let f

H
2

with f(e
i
) =

nZ
a
n
e
in
, then
(f, g) =

nZ
a
n
b
n
= 0 f H
2
with g(e
i
) =

nN
0
b
n
e
in
.
But since b
n
= 0n N
0
it follows that a
n
= 0n N
0
and hence f(e
i
) =

nN
a
n
e
in
. Let
c
n
:= a
n
we have
f(e
i
) =

nN
c
n
e
in
=

nN
c
n
e
in
= g with g H
2
0
.
"" Let f H
2
0
with f =

nN
c
n
e
in
. If we dene a
n
:= c
n
it is obvious that (g, f) = 0 g H
2
,
since f has no coecients for n 0.
Remark The proof above illustrates that another denition holds for the complex conjugate of H
2
,
where the zero Fourier coecient vanish.
H
2
0
:=

f L
2
:

f(n) = 0 for n 0

.
As is customary, we are going to use H
2
0
with the above denition instead of

H
2

.
The direct sum decomposition gives rise to natural projections, the so-called Riesz projections.
Denition 2.1.1 Let f be in L
2
with the Fourier series f(e
i
) =

nZ

f(n)e
in
, then
1. the operator P

: L
2
H
2
0
P

(f) :=

nN

f(n)e
in
is called the Riesz projection onto H
2
0
.
2. the operator P
+
: L
2
H
2
P
+
(f) :=

nN
0

f(n)e
in
is called the Riesz projection onto H
2
.
Remark The operators are well-dened because cutting o the positive or negative part still guarantees
the convergence of the Fourier series, and since the coecients are still square summable, the image of
the projections is still in L
2
and hence due to the denitions in H
2
or H
2
0
.
Both projections are bounded since L
2
is the sum of two closed subspaces. Therefore it is quite natural
to dene the following two operators.
2 PRELIMINARIES 7
Denition 2.1.2 Let f H
2
and L

.
1. The operator H

: H
2
H
2
0
H

f := P

(f)
is called the Hankel operator with symbol .
2. The operator T

: H
2
H
2
T

f := P
+
(f)
is called the Toeplitz operator with symbol .
Remark The Hankel and Toeplitz operator are bounded due to the boundedness of the Riesz projec-
tions.
2.2 An Introduction to H
p
Spaces
The Hankel and Toeplitz operators in Section 4 and 5 are dened for H
p
with p [1, ]. Thus our
rst step is to dene them, but the question arises is how? Introducing them using the same route like
we did for H
2
fails due to the following problems:
1. Although Carlesons theorem which he published in 1966 was later generalized by Richard Hunt
for p > 1, in 1922 already Andrey Kolmogorov constructed an example of a function in L
1
whose
Fourier series diverges everywhere.
2. Leaving aside the convergence of the Fourier series, cutting o the negative parts of a function in
L
1
or L

may not lead to a function belonging to the space H

1
or H

we are trying to dene

(More about that later).
3. Since we are no longer in a Hilbert space, we do not have an orthogonal compliment anymore,
which means that L
p
may not be the direct sum of H
p
and another closed space. Hence we would
be far away from dening natural projections and consequently the operators.
The remedy we suggest is to dene H
p
on the unit disk and work from there to H
p
on the circle which
we can then interpret as a subspace of L
p
. In fact it will be closed and on these grounds we are then
able to dene the Riesz projections which leads to the Hankel and Toeplitz operators.
To illustrate this a bit more and in order to give an impression of what the H
p
spaces are, let us
further investigate the case p = 2.
We arrived at the point that the functions in H
2
on the circle are of the form:
f(e
i
) =

nN
0

f(n)e
in
with

f(n) =

T
f(e
i
)e
in
dm().
Let us have a look at the functions
f(z) =

nN
0

f(n)z
n
with

f(n) =

T
f(e
i
)e
in
dm() and [z[ < 1.
Since [z[ < 1 the pointwise convergence of f(z) follows from the pointwise convergence for [z[ = 1 and
hence f is analytic in D. We arrive at the following denition.
Denition 2.1.3 The space

H
2
:= f analytic in D : the coecients are square summable
is called the Hardy-Hilbert space on D with norm
|f|

H
2
=

nN
0
[a
n
[
2
1
2
.
2 PRELIMINARIES 8
There is a natural identication between H
2
and

H
2
. Namely we identify the function f H
2
as
having Fourier series

nN
0
a
n
e
in
with the analytic function f(z) =

nN
0
a
n
z
n
. This identication
is clearly an isomorphism between H
2
and

H
2
.
Of course, this identication, although natural, does not describe the relationship between f H
2
and

f

H
2
as functions (We come back to this later in the next subsection).
Theorem 1.1.12 in [2] (Pg. 7) provides an insight into the denition of H
p
functions since one just has
to replace 2 with p. It states the following:
Theorem 2.2 Let f be analytic on D. Then f

H
2
if and only if
sup
0<r<1

T
[f(re
i
)[
2
dm() < .
Moreover for f

H
2
|f|
2

H
2
= sup
0<r<1

T
[f(re
i
)[
2
dm().
Of course this was not the reason why Frigyes Riesz introduced the H
p
spaces in [4] (1923) which he
named after Godfrey Harold Hardy because of his paper [3] published in 1915.
The maximum modulus principle says that for f analytic in D the maximum modulus function
M

(r, f) := max

[f(re
i
)[ : [0, 2]

, r (0, 1)
is an increasing function of r. For a xed p (0, ) G.H. Hardy proved in [3] the same result for the
integral means
M
p
(r, f) :=

T
[f(re
i
)[
p
dm()
1
p
, r (0, 1)
and from here it is quite natural to consider the following class of analytic functions.
Denition 2.2.1 The space
H
p
(D) :=

f analytic in D : sup
0<r<1
M
p
(r, f) <

is called the Hardy space H

p
.
Remark Since M
p
(r, f) is an increasing function of r it is clear that
sup
0<r<1
M
p
(r, f) = lim
r1

M
p
(r, f).
Denition 2.2.2 When p = , we dene H

to be the class of analytic functions f in D for which

sup
0<r<1
M

(r, f) < .
These functions are precisely the bounded analytic functions on D.
From the basic inequality
(a +b)
p
2
p
(a
p
+b
p
) a 0, b 0
it follows that H
p
is a complex linear space and with
1 p < : |f|
H
p := sup
0<r<1
M
p
(r, f)
p = : |f|
H
:= sup
0<r<1
M

(r, f)
2 PRELIMINARIES 9
it is a normed space. One easily checks, that H
p
for 1 p is a Banach space (Remark 17.8 in [8]).
Furthermore a simple application of Hoelders inequality shows that for p, q [1, ], p < q, the
inequality
M
p
(r, f) M
q
(r, f)
holds for all r (0, 1). Therefore H
q
H
p
.
2.3 The Poisson kernel and its mollier properties
Before we continue our introduction let us have a look at the Poisson kernel, since it plays a major part
in this theory.
Denition 2.2.3 The function
{
r
() =
1 r
2
1 +r
2
2r cos
with (, ] is called the Poisson kernel.
It also has the following form
{
r
() = {
r
( ) = Re

1 +re
i
1 re
i

= Re

e
i
+re
i
e
i
re
i

= Re

e
i
+z
e
i
z

= {
z
()
with = and z = re
i
.
Theorem 2.3 The Poisson kernel has the following properties also referred to as mollier properties
1. {
z
() > 0 for z D and (, ].
2. {
z
( + 2) = {
z
().
3. for each z D

T
{
z
()dm() = 1.
4. Given a > 0
{
z
() 0 uniformly for [[ as r 1.
2.4 A digression on harmonic functions
We arrived at a point where we dened a class of analytic functions in the unit disc, but the connection
to the space L
p
is still missing. The connection does not seem to be natural since H
p
is living on the
unit disk and L
p
on the unit circle. In fact we need to talk in terms of harmonic functions, since the
theorems we need were proved for this class of functions. Therefore we dene the class of harmonic
functions with bounded integral means.
Denition 2.3.1 1. Let p [1, ), then the space
h
p
(D) := U harmonic in D : sup
0<r<1
M
p
(r, U) <
is called the harmonic Hardy space h
p
in the disc.
2 PRELIMINARIES 10
2. If p = , we dene h

(D) to be the class of harmonic functions U in D for which

sup
0<r<1
M

(r, U) < .
These functions are precisely the bounded harmonic functions in D.
The following theorem gives us the connection between harmonic functions in the disk and integrable
functions on the unit circle. It was published by Pierre Fatou in his famous paper [9] in 1906 and is
the starting point of the whole subject dealt with here.
Theorem 2.4 Let 1 < p and U(z) h
p
. Then there is a F L
p
such that
U(z) =

T
{
z
()F()dm().
The proof of this theorem uses strongly the fact, that L
q
is the dual of L
p
with
1
q
+
1
p
= 1. But the
problem is that L
1
itself is not the dual of any Banach space. A way out of this dilemma is to use
the space of nite complex measures on T, because it is the dual of C(T) ,i.e. the space of continuous
functions on T.
If p L
1
, we can associate to p a complex measure
p
, by putting

T
G()d
p
() =

T
G()p()d.
With this argument one can obtain by using the same route of the proof of Theorem 2.4 the following
Theorem 2.5 Let U(z) h
1
. Then there is a nite complex measure on T such that
U(z) =

T
{
z
()
d()
2
.
Conversely there is a theorem which for a given function in L
p
provides the corresponding harmonic
function in D. It states the following
Theorem 2.6 If 1 p and f L
p
and
U(z) =

T
{
z
()F()dm()
then U(z) h
p
.
The same result holds for complex measures.
Theorem 2.7 Let be a nite complex measure on T. Then
U(z) =

T
{
z
()
d()
2
is in h
1
.
2.5 Boundary Behaviour
We are now able to assign an harmonic function in the unit disc with an integrable function on the unit
circle and conversely. It would be convenient if the integrable function was the limit of the harmonic
function, if z tended to the boundary, since consequently a natural isomorphism would arise between
the functions in D and on T. Indeed with a few restrictions this is the case but let us rst examine the
useful L
p
and w

convergence.
2 PRELIMINARIES 11
Theorem 2.8 Let F L
p
, 1 p < and let
U(z) =

T
{
z
()F()dm()
then

T

U(re
i
) F()

p
d 0 as r 1
,i.e. U(re
i
) approaches F() in L
p
-norm, as r 1.
If p = or is a complex measure all we have is weak-star convergence (w

).
Theorem 2.9 1. If F L

and
U(z) =

T
{
z
()F()dm()
then U(re
i
) F() in w

, as r 1.
2. Let
U(z) =

T
{
z
()
d()
2
with a nite complex measure on T. Then U(re
i
)d d() in w

, as r 1.
Remark The proof of the above theorem needs in particular the mollier properties of {
z
()
1
2
.
Therefore all kinds of other kernels besides the Poisson kernel would work to yield similar results.
Since we now have the radial convergence in L
p
-norm and w

the following questions remain

1. Do pointwise limits exist?
2. If yes, are non-radial limits possible?
Both questions cannot be answered on the basis of the mollier properties alone and need a more
detailed examination of the topic.
A theorem which answers both of the questions is called the Fatou theorem, which was published as
well in 1906 in the paper [9], but before we state it, we need a little preparation.
It is clear that the case
U(z) =

T
{
z
()F()dm() with F L
p
can be subsumed to the case
U(z) =

T
{
z
()
d()
2
since d() := F()d is a nite complex measure on T. The theorem of Fatou, which deals with the
non-tangential limits of the Poisson integral says the following.
Theorem 2.10 Let
0
T and suppose that the derivative

(
0
) exists and is nite. Then
U(z) =

T
{
z
()
d()
2
tends to

(
0
) for z = re
i
tending to e
i
0
, from within any region of the form [
0
[ c(1 r), the
so-called Stolz angle.
Remark Strictly speaking it is the Stolz angle not a sector with straight sides and vertex at e
i
0
.
However it becomes asymptotic to such a sector near the point e
i
0
.
2 PRELIMINARIES 12
Let F L
p
, p 1 and let
U(z) =

T
{
z
()F()dm().
A classical theorem of Lebesgue says that
d
d

0
F(t)dt exists a.e. and equals F().
In conjunction with Fatous theorem (Theorem 2.10) we see, that a.e. in T
U(re
i
) F(
0
) as re
i

e
i
0
where

means within the Stolz angle. Combined with Theorem 2.4 we obtain the following
Theorem 2.11 Let 1 < p and let U(z) h
p
. Then for almost all
0
, U(z) tends to a nite limit,
say U(e
i
0
), as z = re
i

e
i
0
. The function
0
U(e
i
0
) L
p
and for all z D we have
U(z) =

T
{
z
()U(e
i
)dm().
Remark U(e
i
) is called the (non-tangential) limit of U(z). We frequently write
U(e
i
) = lim
z

e
i
U(z) a.e..
In case p = 1 the theorem is not completely true, since we are dealing in that case with a measure
d(). The decomposition theorem of Lebesgue says that then the derivative

() still exists and is

nite a.e. and that

() L
1
but that d() is not in general

()d. Instead
d() =

()d +d
s
()
where
s
is a singular measure.
Thus if we barely know that U(z) L
1
, we still have a.e. existence of the nite non-tangential limit
lim
z

e
i
U(z) =

()
but we cannot recover U(z) from this boundary value function. Instead we have
U(z) =

T
{
z
()

()dm() +

T
{
z
()
d
s
()
2
with some singular
s
.
Example The ordinary Poisson kernel shows that a representation with nonzero
s
can actually occur.
U(re
i
) =
1 r
2
1 +r
2
2r cos
.
It follows from this representation that
lim
z

e
i
U(z) = 0
save for = 0 and
U(z) =
1
2

T
{
z
()2d
s
()
where
s
is the unit point mass at 0.
This distinction between the cases p = 1 and p > 1 is one of the fundamental complications of the
theory and will be seen to have deep and far-reaching implications in its further developments.
2 PRELIMINARIES 13
2.6 Back to Analytic Functions
It is a well known fact that analytic functions are harmonic and since the only dierence between f h
p
and f H
p
is the analyticity, all the theorems in the last subsection apply for f H
p
.
Therefore we have for f H
p
, p > 1
f(z) =

T
{
z
()f(e
i
0
)dm()
and
f(e
i
) = lim
z

e
i
f(z) a.e..
Hence we have a natural isomorphism between the function f H
p
(D) and f H
p
(T). Since f on the
circle is in L
p
, we can view H
p
as a subspace of L
p
.
If p = 1, we saw in the previous subsection that we utterly failed to construct an isomorphism be-
tween the function on T and the function in D.
But since f is now analytic the situation changed after the proceedings of the fourth scandinavian
mathematical congress in 1917. There Frigyes and Marcel Riesz published the celebrated
Theorem 2.12 (F. and M.Riesz) If

T
e
in
d() = 0 for n = 1, 2, 3, . . .
then is absolutely continuous with respect to Lebesgue measure.
This theorem is the key in the following context.
If f H
1
, then we have since f is in particular harmonic for [z[ < 1 by Theorem 2.5
f(z) =

T
{
z
()
d()
2
for some complex measure (here complex valued) on T. By Theorem 2.7 we obtain
f(re
i
)d d() in w

(2.6.1)
as r 1.
Since f H
1
, the following holds.
Lemma 2.13 If f H
1
, then

T
e
in
d() = 0 for n = 1, 2, 3, . . . .
Proof Let = e
i
, then
F
z
() :=
+z
z
=
1 +z
1 z
= 1 + 2
z
1 z
= 1 + 2

kN
z
k

k
since [z
k

k
[ = [z[
k
< 1.
2 PRELIMINARIES 14
Therefore
{
z
() = Re [F
z
()] =
1
2

F
z
() +F
z
()

=
1
2
+

kN
z
k

k
+
1
2
+

kN
z
k

k
=

kN
0
z
k

k
+

kN
z
k

k
=
1
1 z
+

kN
z
k

k
.
By Theorem 2.5 we obtain
f(z) =

T
{
z
()
d()
2
=

T
1
1 z
d()
2
+

kN
z
k

k
d()
2
.
With [z[ < r it follows that [z
k

k
[ < r
k
and therefore the sum uniformly converges on compact subsets
which gives us the possibility to arrive at the following
f(z) =

T
1
1 z
d()
2
+
1
2

kN
z
k

k
d().
Since f and the left part of the sum is due to the Cauchy integral formula analytic, the anti-analytic
part on the right has to vanish. Therefore
0 =

n
d() =

T
e
in
d() for n = 1, 2, 3, . . . .
Now the theorem of the Riesz brothers (Theorem 2.12) guarantees that is absolutely continuous and
therefore due to the Radon-Nikodym theorem
d() = h()d for some h L
1
.
So in this case we really have
f(z) =

T
{
z
()h()dm()
with a function h.
The distinction between this case and the more general case (f merely harmonic) is very important for
the whole development of the theory.
By Fatous theorem (Theorem 2.10), we now have f(z) h() a.e. for z

e
i
, so if we call
f(e
i
) = lim
z

e
i
f(z)
we have
f(z) =

T
{
z
()f(e
i
)dm()
for f H
1
. Due to our observation before H
1
L
1
and consequently the case p = 1 is not special
anymore for analytic functions and we can state the following theorem.
2 PRELIMINARIES 15
Theorem 2.14 Let 1 p and let f(z) H
p
. Then for almost all
0
, f(z) tends to a nite limit,
say f(e
i
0
), as z = re
i

e
i
0
. The function
0
f(e
i
0
) L
p
and for all z D we have
f(z) =

T
{
z
()f(e
i
)dm().
Due to the following theorem the boundary functions are unique since H
1
contains all other H
p
spaces
for p > 1.
Theorem 2.15 Let f H
1
and suppose, for a set of E of positive measure, that f(e
i
) = 0 for E.
Then f 0.
2.7 Projections from L
p
to H
p
If X is a Banach space and A is a closed subspace of X, then there is not usually any closed subspace
B such that X = AB.
If nevertheless X = AB, we have a unique projection from X onto A with null space B. The aim is
in our case to compute a bounded projection from L
p
to H
p
, if possible. We have so far that H
p
is a
closed subspace of L
p
. The question now arises is, how do we dene a projection onto H
p
? In our rst
chapter we cut the negative Fourier coecients and due to our denition of H
2
and the fact that the
function without the negative part is still in L
2
, this was a projection. It was bounded, since L
2
has a
natural decomposition in two closed subspaces.
The core of the denition of the projection was our preliminary denition of H
2
, since we dened it as
functions in L
2
, where the Fourier coecients vanish for n < 0.
Our situation now is completely dierent. For p [1, ] we are dealing with special classes of analytic
functions in D which are isomorphic to functions on T which belong to a much larger space L
p
on T.
For the last class, there exists a Fourier representation (excluding p = 1) but where is the connection
to H
p
(T)? The answer is contained in the following two theorems proved by the Riesz brothers in [5].
Theorem 2.16 (F. and M. Riesz) 1. For p [1, ] a function f L
p
belongs to H
p
(T) if and
only if the Fourier coecients

f(n) =

T
f(e
i
)e
in
dm()
vanish for n < 0.
2. If f H
p
(T) has Fourier series f(e
i
) =

nN
0

f(n)e
in
, then the Taylor series of the corre-
sponding H
p
(D) is f(z) =

nN
0

f(n)z
n
. Conversely, if f H
p
(D) has Taylor series f(z) =

nN
0
a
n
z
n
, then f H
p
(T) has Fourier series f(e
i
) =

nN
0
a
n
e
in
.
Consequently we have a new denition of H
p
(T).
Denition 2.16.1 The space
H
p
(T) := f L
p
:

f(n) = 0 for n < 0
is called the Hardy space H
p
on T.
From here we have a similarly to the case p = 2 a natural candidate for a closed subspace complementary
to H
p
,
B := f L
p
:

f(n) = 0 for n 0.
The functions in B live on the unit circle and the following lemma gives the connection to anti-analytic
functions in D.
2 PRELIMINARIES 16
Lemma 2.17
B = H
p
0
(T) := f : f H
p
,

f(0) = 0.
Proof "" Let f B. It follows from the denition of B that

f(0) = 0. Then
f(e
i
) =

nN

f(n)e
in
=

nN
a
n
e
in
with a
n
:=

f(n).
Therefore f = g with g H
p
0
.
"" similarly
The extension of H
p
0
(T) to the disk makes use of the Theorem 2.16 (part 2) so we have, along with H
p
the space H
p
0
(D) of anti-analytic functions in D.
Therefore it is quite natural to introduce the following denition.
Denition 2.17.1 If f L
p
has Fourier series
f

nZ

f(n)e
in
then the operator
1.
P
+
(f) :=

nN
0

f(n)e
in
is called the analytic Riesz projection.
2.
P

(f) :=

nN

f(n)e
in
is called the anti-analytic Riesz projection.
Remark We are now using the symbol , because we are including the case p = 1, where Fourier
series must not converge. Therefore we are just associating f with its Fourier series.
Some care must be taken with our previous result in order for it not to be misunderstood. One might
be tempted to say that the image of the analytic Riesz projection P
+
belongs to H
p
for p [1, ].
Unfortunately this is not always the case as the following well known examples show.
Example 1. f() =

nN
sin(n)
n
= arg(1e
i
) is the Fourier series of a bounded function. However
with

nN
sin(n)
n
=
1
2i

nN
z
n
z
n
n
=
1
2i

nZ\{0}
z
n
n
with z = e
i
we obtain with the power series representation of the logarithm
P
+
(f) =
1
2i

nN
z
n
n
=
1
2i
log(1 z)
and the image does not belong to H

, since it is not bounded anymore.

2. f() =

n=2
cos(n)
log n
is the Fourier series of an L
1
function ([10] pg. 64) but
P
+
(f) =
1
2

n=2
z
n
log n
does not belong to H
1
due to Hardys inequality.
The last example shows that the image under P
+
of a function in L
1
must not be in L
1
anymore.
2 PRELIMINARIES 17
Therefore the hope that for an f L
p
it holds
f = g +h with g H
p
and h H
p
0
is utterly lost for the case p = 1 and p = .
For p (1, ) we do have the following positive result due to Marcel Riesz which he published in [6].
Theorem 2.18 (M.Riesz) For 1 < p < the operator P
+
maps L
p
boundedly onto H
p
.
Remark Due to this theorem, the operators P
+
and P

are called Riesz projections.

Consequently we have the following decomposition for p (1, )
L
p
= P
+
L
p
+P

L
p
which implies
L
p
= H
p
H
p
0
.
Therefore we have for a given f L
p
, p (1, ) a g H
p
, and an h H
p
0
such that
f = g +h. (2.7.1)
For p = 1 we gave an example such that
L
1
= H
1
H
1
0
but it holds that the functions of the from 2.7.1 are dense in L
1
.
Our example showed that the projection P
+
is unbounded but even more holds due to a theorem of
Donald J. Newman which he published in [7].
Theorem 2.19 (J.D. Newman) There is no closed subspace of L
1
complementary to H
1
. Equiva-
lently there is no bounded projection from L
1
to H
1
.
For p = the situation is (in a certain sense) even worse. As we noted
L

= H

0
but furthermore, the functions of the form 2.7.1 are not even dense.
Theorem 2.20 The closed linear span of the functions in H

and their complex conjugates is not all

of L

.
Remark The theorem is proved in [14], page 151.
It turns out that P
+
L
1
H
p
p < 1 and P
+
L

is equal to the space of analytic functions of "bounded

mean oscillation", abbreviated as BMOA. This space is just slightly bigger then H

and it holds
H
p
BMOA H

for all p < .

3 NECESSARY TOOLS 18
3 Necessary Tools
3.1 Jensens Inequality
For analytic functions across D, the inequality (3.2) below was rst noticed by Jensen [1899] and for
this reason the inequality is in this paper called Jensens inequality. The proof goes over subharmonic
functions and is therefore another topic and too vast to t into this paper. The result is very important
as we are going to use it quite often.
Theorem 3.1 If 0 < p and if f(z) H
p
, f 0, then

T
log [f(e
i
)[dm() > . (3.1)
If f(0) = 0, then
log [f(0)[

T
log [f(e
i
)[dm() (Jensens inequality) (3.2)
and more generally, if f(z
0
) = 0
log [f(z
0
)[

T
{
z
0
() log [f(e
i
)[dm().
Remark A fundamental result of H
p
theory is that the condition of (3.1) characterizes the moduli of
H
p
functions [f(t)[ among the positive L
p
functions, i.e.
= [f[, for some f H
p
, f 0

L
p
(T)
log L
1
(T).
Jensens inequality shows the direction "" but the other is contained in Section 3.2 where I compute
a more detailed factorization of H
p
functions.
A useful result which follows from Jensens inequality is the following
Corollary 3.2 Let 0 < p, r . If f(z) H
p
and if the boundary function f(e
i
) L
r
, then
f(z) H
r
. This is often written as
H
p
L
r
= H
r
.
3.2 Factorization for functions in H
p
The next result plays an important role in H
p
theory and especially with respect to our observations
regarding the kernels of the Hankel and Toeplitz operator. It is known as the "factorization theorem"
and is owed to F.Riesz [4] and Vladimir Ivanovich Smirnov [13].
Theorem 3.3 (factorization theorem) 1. Let f H
p
, f 0 with 1 p , then f has a
unique factorization of the form
f = BSF
where B is the Blaschke product
B(z) = z
m

|
k
| =0

k
[
k
[

k
z
1
k
z
,
n
: n N D
whose zeros, repeated according to multiplicity, satisfy

nN
(1 [
n
[) < .
3 NECESSARY TOOLS 19
S is a singular function of the form
S(z) = exp

T
e
i
+z
e
i
z
d()

with positive singular measure (with respect to dm) , and F is an outer function (later dened
as the associated outer function of f)
F(z) = exp

T
e
i
+z
e
i
z
log [f(e
i
)[dm()

where is a complex number of modulus 1.

2. In the above factorization, the function I
f
:= BS is called the inner part of f and has the property
that [I
f
(z)[ 1 on D and [I
f
(e
i
)[ = 1 almost everywhere. Functions with these properties are
called inner functions. The function O
f
:= F, called the outer function of f, has no zeros in the
disk and belongs to H
p
.
3. Every product of the form
B(z)S(z)exp

T
e
i
+z
e
i
z
log [g(e
i
)[dm()

= BSF (3.1)
where B is a Blaschke product , S is a singular function and g is an L
p
function with log[g[ L
1
,
belongs to the space H
p
. Furthermore [F(e
i
)[ = [g(e
i
)[ almost everywhere.
The whole derivation is quite interesting which is why it is included in this paper. However since it is
not directly related to our topic it is stated in the back in Section 6.1.
An important property of outer and inner functions is the following
Lemma 3.4 Outer and inner functions are multiplicative.
Proof For inner functions this follows directly from the denition.
Let f, g be outer functions (The general denition of outer functions is stated in (3.2), where the only
dierence is that log [f(e
i
) is just a real valued integrable function k on T). Therefore
fg =
1
exp

T
e
i
+z
e
i
z
k
1
()dm()

2
exp

T
e
i
+z
e
i
z
k
2
()dm()

= exp

T
e
i
+z
e
i
z
k()dm()

where :=

2
a complex number of modulus 1 and k := k
1
+ k
2
is a real valued integrable function
on the unit circle.
Remark For f, g the quotient
f
g
is as well an outer function, since k := k
1
k
2
is as well a real valued
integrable function on T and :=

1

2
a complex number of modulus 1. For inner functions this is not
true, since they contain a Blaschke product with zeros.
The converse is also true for outer functions in H
p
.
Lemma 3.5 Let F be an outer function in H
1
, then F =
g
h
, with g, h H

and both outer functions

Proof Without loss of generality we assume = 1. Then with Lemma 6.1 we have
F(z) = exp

T
e
i
+z
e
i
z
log [F(e
i
)[dm()

(3.2)
3 NECESSARY TOOLS 20
with log [F[ L
1
(T). Let us dene
(e
i
) :=

[F(e
i
)[, [F(e
i
)[ 1
1, [F(e
i
)[ > 1
(e
i
) :=

1, [F(e
i
)[ 1
1
|F(e
i
)|
, [F(e
i
)[ > 1.
Then , are bounded functions in D, hence O

, O

are in H

and
O

(z)
O

(z)
= exp

T
e
i
+z
e
i
z
log
(e
i
)
(e
i
)
dm()

= exp

T
e
i
+z
e
i
z
log [F(e
i
)[dm()

= F.
Another representation of outer functions we are going to use later is the following theorem.
Theorem 3.6 Let F be an outer function with the well known form for z D
F(z) = exp

T
e
i
+z
e
i
z
k()dm()

.
Then F has on the boundary T the form
F(e
i
) = exp [k() +iH(k())] a.e. (3.3)
where H is the harmonic conjugation operator.
Proof Without loss of generality we assume = 1. Then
log F(z) =

T
e
i
+z
e
i
z
k()dm()
=

T
Re

e
i
+z
e
i
z

k()dm() +

T
Im

e
i
+z
e
i
z

k()dm()
=

T
{
z
()k()dm() +

T
O
z
()k()dm().
with {
z
() the Poisson kernel and O
z
() the conjugate Poisson kernel. Then with Theorem 2.14 we
know that the non-tangential limits of the Poisson integral exists a.e. and is in this case equal to k. On
the other hand Theorem 3.1.1. in [11] states, that the same is true for the conjugate Poisson integral
and that the limit is equal to the conjugate function in this case the conjugate function of k. Therefore
log F(e
i
) = k() +iH(k()) almost everywhere.
Another factorization of functions in H
1
which is proved in [14], pg. 52 is the following
Theorem 3.7 Every function in H
1
is the product of two functions in H
2
.
In our observation of the kernels of Hankel and Toeplitz operators, we will often deal with a space
IH
p
:= If : f H
p

where I is an inner function and p [1, ]. The following is true.

Lemma 3.8 Let I be an inner function and p [1, ], then IH
p
H
p
.
Proof Let If IH
p
, then the lemma follows from the inequality
[If[ [f[.
3 NECESSARY TOOLS 21
3.3 The Nevanlinna class
The Nevanlinna class is named after the famous nish mathematician Rolf Herman Nevanlinna. It
contains analytic functions in the disk with the following property.
Denition 3.8.1 An analytic function f(z) on D is in the Nevanlinna class, f N, if
sup
0<r<1

T
log
+

f(re
i
)

dm() < .
It is clear that H
p
N for all p > 0, since log
+
[f(z)[
1
p
[f(z)[
p
.
Before we mention the factorization of functions in N and the resulting connection to functions in H
p
,
let us rst introduce an important subclass of N.
If f N, f 0 then it can be shown (stated in [11] Theorem 2.5.3.) that

T
log

f(e
i
)

dm() > .
However the sharper inequality
log [f(z
0
)[
1
2

{
z
0
() log

f(e
i
)

d (3.1)
which was proven for H
p
functions (Jensens inequality Theorem 3.1), can fail for f(z) N which
shows the following example.
Example Let
g(z) = exp

1 +z
1 z

then log [g(z)[ = Re

1+z
1z

= {
z
(1). The function g has the harmonic majorant {
z
(1) is therefore in N
(It can be shown, that this is equivalent to the denition of functions in N) and the measure determined
by log [g(z)[ is the unit charge at e
i
= 1. Since
log [g(0)[ = 1 > 0 =

log [g(e
i
)[d
,(3.1) fails for g(z).
This counterexample contains the only thing that can go wrong with (3.1) for a function in N. The
functions in N for which (3.1) holds form a subclass of the Nevanlinna class the so called Smirnov class
which is denoted by N
+
.
In this paper we abandon the original denition of functions in N
+
, because it is not simple to see the
connection to N and H
p
and requires a more detailed examination of the topic. Hence we are going to
work with an equivalent denition of N
+
, which was proven in [11].
From the factorization theorem for H
p
functions, it holds, that any H
p
function can be written in
the form BS

F
f
, where B is the Blaschke product, S

is a singular function with positive singular

measure , and F
f
is an outer function from the L
p
function f on T, with log[f[ L
1
. Writing the real
singular measure as
1

2
as the dierence of two positive measures, one obtains the factorization
for functions in N, which states the following
Theorem 3.9 Let f(z) N, f 0. Then
f(z) = CB(z)
S
1
(z)
S
2
(z)
F(z) [C[ = 1 (3.2)
where B(z) is a Blaschke product, F(z) is an outer function and S
1
(z) and S
2
(z) are singular functions.
Except for the choice of the constant C, [C[ = 1, the factorization (3.2) is unique. Every function of
the form (3.2) is in N.
3 NECESSARY TOOLS 22
Remark The main dierence here, is that we have a quotient of two singular functions and that the
outer function is not necessarily in H
p
.
The factorization of functions in N
+
is similar to the factorization of H
p
functions with one important
dierence.
Theorem 3.10 (and Denition) Let f(z) N, f 0. Then in (3.2) the singular factor S
2
1 if
and only if f(z) N
+
. Therefore we have the unique factorization
f(z) = CB(z)S(z)F(z) [C[ = 1.
From here it is easy to see the following connection.
H
p
N
+
N
for p > 1. It also follows from a theorem in [11], that f(z) H
p
if and only if f(z) N
+
and
f(e
i
) L
p
. This fact can be written as
N
+
L
p
= H
p
.
The example we gave above shows that
N L
p
= H
p
.
The denition of functions in N
+
and the representation in Subsection 3.2 equation (3.1) of functions
in H
p
, 0 p is quite similar. The only dierence is that the outer part of functions in N
+
must
not be integrable which illustrates the following example.
Example The function h(z) = 1 z is outer since Corollary 3.4.8. in [11] which says, that every
function in H
p
with non-negative real values is outer. Therefore the function
f(z) =
1
1 z
is due to (3.4) in N
+
but is not integrable.
An equivalent denition of functions in N and N
+
is the following corollary
Corollary 3.11
N =
f
g
: f, g H

, g zero-free (3.3)
N
+
=
f
g
: f, g H

, g an outer function. (3.4)

Proof We are giving the proof for functions in N, since the proof for N
+
, is nearly the same.
"" Let f N, with factorization f = B
S
1
S
2
F dened like (3.2), we can assume without loss of
generality that C = 1.
We know that with Lemma 3.5, we obtain F =
F
1
F
2
, where F
1
, F
2
H

. Now we use the multiplicativity

of bounded functions and obtain
F =
BS
1
F
1
S
2
F
2
=:
g
h
with g, h H

.
"" Let f be of the form (3.3), then since g, h H

, they can be factorized into

g
h
=
O
g
B
g
S
g
O
h
S
h
.
Since h is zero-free, it does not contain the Blaschke product factor and we complete the proof by
means of the denition of the outer function F :=
O
g
O
h
.
3 NECESSARY TOOLS 23
3.4 The forward and backward shift on H
p
, for p [1, )
Denition 3.11.1 For a function f(z) =

nN
0
a
n
z
n
H
p
the operator
1.
(Sf)(z) := zf(z) =

nN
0
a
n
z
n+1
= a
0
z +a
1
z
2
+a
2
z
3
. . .
is called the forward shift operator.
(Bf)(z) :=
f(z) f(0)
z
= a
1
+a
2
z +a
3
z
2
. . .
is called the backward shift operator.
Remark It is clear from the denition of H
p
, that S and B are linear transformations from H
p
to H
p
A theorem we will use in conjunction with the kernels of Hankel operators is the following
Theorem 3.12 (Beurlings theorem) Let p (0, ) and K be a non-zero S-invariant subspace of
H
p
. Then the following statements are true.
1. K = IH
p
for some inner function I.
2. I
1
and I
2
are inner functions, then I
1
H
p
I
2
H
p
if and only if I
1
/I
2
H

, i.e. I
2
divides I
1
.
Remark 1. Beurlings theorem was originally proven by Beurling for the Hilbert space H
2
and then
generalized by others to H
p
, (0 < p < ).
2. From the Nevanlinna factorization theory, one can show that for any family M of inner functions,
there is an inner function I
M
with the property, that
(a) I
M
divides every I M, specically, I/I
M
H

, and
(b) if J is an inner function which divides every I M, then J divides I
M
.
The inner function I
M
is called the greatest common divisor of M. If K is an S-invariant subspace
of H
p
and I is the greatest common divisor of the inner factors of non-zero functions from K,
then K = IH
p
.
Conversely we will use the B-invariant subspace for our examination of the kernels of Toeplitz operators.
The following theorem explains the structure of this set.
Theorem 3.13 Let p (1, ) then the B-invariant subspace is
K
p
I
:= H
p
IH
p
0
where I ranges over the inner functions.
Remark 1. The operator B is often called S

since B is the adjoint operator of S.

2. An important point needs to be made here to avoid confusion. The space H
p
IH
p
0
is understood
on the circle as the space H
p
(T) IH
p
0
(T).
4 KERNELS OF HANKEL OPERATORS 24
4 Kernels of Hankel operators
Denition 4.0.1 1. Let p [1, ], L

and let P

be the Riesz-projection from L

p
onto H
p
0
.
Then the operator H

: H
p
H
p
0
,
H

f := P

(f)
is called Hankel operator with symbol .
2. The kernels of the Hankel operator are denoted by
Ker
p
H

:= f H
p
: H

f = 0.
Main questions:
1. For which symbols are the kernels non trivial?
2. What are the kernels?
4.1 The symbols of the Hankel operator
In this subsection we are going to compute the set of symbols for which the kernels are non trivial. The
set is precisely the following:
Denition 4.0.2 The space
N
mer
:=

u
v
: u, v H

, v 0

is called the meromorphic Nevanlinna class.

Let us begin to answer the rst question. We know the following:
Lemma 4.1 If H

, then H

is the zero operator.

Proof Let f H

then consequently with f H

p
it holds f H
p
and therefore P

(f) = 0.
Since this answer is not quite sucient let us assume L

` H

.
Now if f H
p
and H

f = P

(f) = 0 this is equivalent to

f = g H
p
=
g
f
.
Since f, g H
p
N, they can be factorized in g =
u
1
u
2
and f =
v
1
v
2
with u
1
, u
2
, v
1
, v
2
H

and u
2
and
v
2
zero free. Let us dene u := v
1
u
2
and v := v
2
u
1
to conclude
=
g
f
=
v
1
u
2
v
2
u
1
=
u
v
with u, v H

and v 0. We have proved the following:

Proposition 4.2 If Ker
p
H

is non trivial, then N

mer
.
Let us go into more detail. If N
mer
and H

then with N
+
L

= H

this is equivalent to
N
mer
` N
+
.
Recalling the denition of N
+
=
f
g
: f, g H

, g outer we obtain
N
mer
` N
+
:=

u
v
: u, v H

, v 0 and not outer

.
4 KERNELS OF HANKEL OPERATORS 25
When v is not outer and in H

we can decompose it in v = IG with I an inner function and G an

outer function. Therefore
=
F
I
with F =
u
G
. (4.1)
Since G is outer, F N
+
but actually F H

because it has the same boundary values like . Now

we are able to answer the second question and prove the following.
4.2 Description of the kernels of Hankel operators
Theorem 4.3 If Ker
p
H

is non trivial, then Ker

p
H

= IH
p
, with I the inner function we obtained
in (4.1).
Proof "" Let f = Ig for some g H
p
. Then
H

f = P

(f) = P

(Ig)
4.1
= P

(Fg) = 0
since Fg H
p
.
"" If f Ker
p
H

, then P

(f) = 0. Therefore
f = g H
p
f =
g

and because Ker

p
H

is not trivial we can use equation (4.1) to obtain

f = I
g
F
.
Since f H
p
it follows that I
g
F
H
p
and with I H

yield
g
F
H
p
which proves this direction.
This proof is very straight and needs the results we obtained before. Conversely we could obtain the
same result by using Beurlings theorem (Theorem 3.12), but in this situation just for some inner
function not as specialized as we did it before.
Proof Let M = Ker
p
H

, then f M zf M, since [z[ = 1 on T and [z[ < 1 on D it is an inner

function and therefore zH
p
H
p
. M is an invariant subspace of the forward shift operator and hence
representable in the form
M = IH
p
for some inner function I.
Remark The inner function I has a very nice property. It is the greatest common divisor of the inner
factors of non zero functions from M.
5 KERNELS OF TOEPLITZ OPERATORS 26
5 Kernels of Toeplitz operators
Denition 5.0.1 1. Let p [1, ], L

and let P
+
be the Riesz projection from L
p
onto H
p
.
Then the operator T

: H
p
H
p
,
T

f := P
+
(f)
is called Toeplitz operator with symbol .
2. The kernels of the Toeplitz operator are denoted by
Ker
p
T

:= f H
p
: T

f = 0.
Before we start with our examination of the kernels let us prove a useful lemma.
Lemma 5.1 Let L

, 0. Then one can nd an L

with [[ = 1 such that

Ker
p
T

= Ker
p
T

. (5.1)
Proof Let f H
p
, then (5.1) reduces to
T

f = 0 T

f = 0.
Let us dene F := O
||
and :=

F
, then [[ = 1, since
[[ =
[[
exp[Re(log [[ +iH(log h))]
=

= 1 almost everywhere.
"" If T

f = 0, this is equivalent to f = g with g H

p
0
. If we can prove that
g
F
H
p
, then f H
p
0
and f Ker
p
T

.
Since g H
p
N
+
, we have the decomposition
g =
g
1
g
2
with g
1
, g
2
H

and g an outer function.

If we use the decomposition of outer functions in Lemma 3.5, we obtain
F =
F
1
F
2
with F
1
, F
2
H

and outer functions.

Hence
g
F
=
g
1
F
2
g
2
F
1
and due to the multiplicativity of outer and bounded functions is the quotient of g and F in N
+
.
On the other hand

g
F

=
[[[f[
[F[
= [f[ L
p
and consequently
g
F
N
+
L
p
= H
p
.
"" If T

f = 0, this is equivalent to
f
F
= g with g H
p
0
, hence
f = Fg. (5.2)
Since f L
p
it holds with (5.2) that Fg L
p
. Conversely by use of Lemma 6.3 is F in H
1
N
+
and due to the multiplicativity of functions in N
+
it holds Fg N
+
and following the same route like
before we proved that f H
p
0
.
The Main questions are still the same:
1. For which symbols are the kernels non trivial?
2. What are the kernels?
5 KERNELS OF TOEPLITZ OPERATORS 27
5.1 The symbols of Toeplitz operators
Let us start with the rst question. We have seen for the Hankel operator, that the space of symbols
for which the kernels are non trivial is easy to nd and quite sucient. For the Toeplitz operator the
situation is completely dierent. The problem is reducible to the following. Let f H
p
, then
T
f
= 0 f = g with g H
p
0
=
g
f
where the diculties are to describe the space of functions
g
f
. One description is the following
Lemma 5.2 If Ker
p
T

is non trivial, then has the form =

F
IF
with F an outer function in H
p
and J is the greatest common divisor of the inner factors of IH
p
, with I an inner function.
Proof Let f H
p
then
T

f = 0 =
g
f
with g H
p
0
.
Since g, f H
p
, we have the following factorization
g = I
g
F
g
f = I
f
F
f
with I an inner function and F an outer function.
Due to Lemma 5.1, we can say that [[ = 1 without loss of generality. Therefore
1 = [[ =

F
g
I
g
I
f
F
f

F
g
F
f

F
g
= F
f
=: F
with J := I
g
I
f
the the greatest common devisor of I
f
H
p
. We arrive at our result
=
F
JF
. (5.1)
Remark The question now becomes, given an L

, does there exist an F in H

p
, satisfying the
condition (5.1). The answer depends on p and is therefore not quite sucient whereas the answer for
the Hankel operator did not depend on p.
Nevertheless we can state some further observations.
Lemma 5.3 (Coburns lemma) If L

, 0, then either Ker

2
T

= 0 or Ker
2
T

= 0.
Proof Suppose 0 f Ker
2
T

and 0 g Ker
2
T

, then on the one hand

f H
2
0
fg H
1
0
for a certain g H
2
due to Theorem 3.7. On the other hand
g H
2
0
g H
2
0
fg H
2
0
.
Therefore, with H
1
H
1
0
= 0 it follows that fg = 0 and hence due to our assumptions is one of the
kernels trivial.
5 KERNELS OF TOEPLITZ OPERATORS 28
Another interesting observation is if we choose =
zf
f
for a given f H
p
, then L

and
T

f = P
+

zf
f
f

= P
+

zf

= 0.
Remark The question for which symbols are the kernels non trivial is still unsolved and it does not
seem to get solved suciently in the nearby future.
5.2 Description of the kernels of Toeplitz 0perators
Let us start with some preliminary observations. If you choose H

,then f H
p
for a given
f H
p
and hence there is no need of a projection onto H
p
anymore which leads to a trivial kernel.
Conversely it holds the following:
Lemma 5.4 If you choose = I with I an inner function then
Ker
p
T
I
= K
p
I
where K
p
I
is the backward shift invariant subspace.
Proof Let f H
p
. Then
T
I
f = 0 fI H
p
0
f IH
p
0
and hence f H
p
IH
p
0
= K
p
I
.
Due to Theorem 2.20, we know that not only L

= H

0
but the closure of H

0
is not
all of L

. We have seen, that the functions which are neither in H

nor in H

0
are the functions
which complicate matters. That this is already the case for p = 2 shows the following theorem by Eric
Hayashi which he published in [16].
Theorem 5.5 (Hayashi) Let L

, 0 and Ker
2
T

not be trivial, then

Ker
2
T

= G K
2

(5.1)
where G H
2
is such that G
2
is exposed in H
1
and is an inner function with (0) = 0.
Remark We are not giving a denition of exposed points, since we are not dealing with them later.
Furthermore, they make up a chapter on their own thus are not of interest. On top of that there does
not seem to be a comfortable characterization of them which is the reason Konstantin Dyakonov (my
supervisor) felt it would be wise to devise a set without them.
Before we concentrate on the work of Konstantin Dyakonov, it is useful to state a few denitions and
useful lemmas.
Denition 5.5.1 1. Given an integer n 1, we denote P(n) the set of (analytic) polynomials of
degree < n.
2. (H

)
1
:= f H

:
1
f
H

.
3. Elements of the form B B (H

)
1
, where B is the set of all Blaschke products are called
triple.
Lemma 5.6 Let F H

and p [1, ], then

1. H
p
0
F
1
H
p
0
.
5 KERNELS OF TOEPLITZ OPERATORS 29
2. F
1
H
p
0
N
+
0
.
Proof
1. If f H
p
0
, then Ff H
p
0
since F is bounded. Therefore f =
1
F
Ff ,i.e. f is expressible by a
product of an H
p
0
function and a function
1
F
with F H

.
2. Let f H
p
0
N
+
0
and the factorization functions in H
p
yields
f
F
=
O
|f|
I
f
F
.
O
|f|
F
is with Lemma 3.4 outer and therefore
f
F
N
+
0
follows from the denition of N
+
0
.
Corollary 5.7 Let f L
p
, F H

. Then f H
p
0
f F
1
H
p
0
.
Proof Follows by the previous lemma with f N
+
0
L
p
= H
p
0
.
A more detailed description of the kernels is the following lemma.
Lemma 5.8 Let T

be the Toeplitz operator with symbol L

and p [1, ], then

f Ker
p
T

f H
p
0
.
Proof The case 1 < p < is obviously true since L
p
= H
p
H
p
0
.
If p = we have f H

H
2
and it holds
T

f = 0 f H
2
0
and since f L

this is equivalent to
f L

H
2
0
= H

0
.
If p = 1 then f L
1
and since T

f = P
+
(f) = 0, f has to be in H
1
0
due to the denition (recall:
H
1
0
= g L
1
: P
+
(g) = 0).
Remark The Subsection 2.7 deals with the question; what happens if you take a function in L
1
and
you truncate o the negative parts? It is possible that the new function is not even in L
1
anymore.
A necessary description of the symbols, for which the kernels are G H
p
multiplied by a special case
of a backward shift invariant subspace P(n) = H
p
z
n
H
p
0
= K
p
z
n.
Proposition 5.9 Let 1 p and n N. Suppose that G H
p
,
1
G
H

and set := z
n G
G
. Then
Ker
p
T

= G P(n).
Proof Fix f H
p
. In order that f Ker
p
T
u
, Lemma 5.8 says it is necessary and sucient that
fu H
p
0
. (5.2)
Recalling the denition of and setting h :=
f
G
, we rewrite (5.2) as
fz
n
G
G
= Ghz
n
H
p
0
. (5.3)
5 KERNELS OF TOEPLITZ OPERATORS 30
Since we have

Ghz
n

= [fz
n
[ = [f[ [z
n
[ = [f[ L
p
since [z[ = 1 we are able to use Corollary 5.7 and
thus (5.3) is equivalent to
hz
n
H
p
. (5.4)
Finally noting that h H
p
and therefore h H
p
z
n
H
p
0
, we see when keeping the power series
representation in mind that (5.4) reduces to the inclusion h P(n) , i.e.
f G P(n). (5.5)
The resulting equivalence relation (5.2) (5.5) proves the proposition.
A naive attempt to extend (5.1) to the L
p
-scale might be to replace it with something like
Ker
p
T

= G K
p

(5.6)
where is an inner and G is a suitable weight function. Unfortunately this plan fails due to the following
example.
Example Let us take the function h(z) := (1z)
p
0
with p
0
(1, ) and =
zh
h
. Then L

and
if z = e
it
then for t being small z is reducible to the rst two parts of its power series representation
and therefore e
it
1 +it and hence

[dz[
[1 z[
p
p
0

dt
[t[
p
p
0
.
Now it follows from basic calculus that h H
p
for 0 < p < p
0
. It holds h Ker
p
T

for 1 p p
0
since P
+
(h) = P
+
(
z

hh
h
) = P
+
( zh) = 0.
Furthermore since Lemma 5.9 the subspace Ker
p
T

has dimension 1 and is therefore spanned by h. If

p p
0
, h / H
p
and since Ker
p
T

Ker
1
T

= ch : c C it follows that Ker

p
T

= 0. Thus for the

current choice of , the left hand side of (5.6) is nontrivial if 1 p p
0
and trivial if p p
0
. However
such a jump never occurs on the right side. The subspace is nontrivial for all p [1, ], provided that
G is nonzero and non constant (otherwise it is trivial for all ps).
Konstantin Dyakonov extended the work of Hayashi (although the work is entirely independent) for
L
p
, p [1, ]. Moreover, before going into this, it is useful to calculate a few results. A major part
and the beginning of our preparation of the kernels of Toeplitz operators is the following:
Theorem 5.10 (J.Bourgains factorization theorem) Let f L

and log[f[ L
1
, then f = gh
for some g, h H

.
When restricted to unimodular functions, Bourgains result amounts to the following:
Theorem 5.11 Given a function L

with [[ = 1 a.e. there is a triple (B, b, g) such that

=
b
B
g
g
.
Proof Since is unimodular, it follows that log[[ = 0 is integrable. Thus we use J. Bourgains
factorization theorem to obtain = gh, g, h H

. We continue the factorization by using the

factorization for H
p
functions.
f = O
|g|
O
|h|
I
g
I
h
with g = O
|g|
I
g
, h = O
|h|
I
h
. Since [[ = [h[[g[ = 1 and since outer functions are multiplicative it
follows
O
|g|
O
|h|
= O
|gh|
= O
1
1.
5 KERNELS OF TOEPLITZ OPERATORS 31
Therefore we dene G := O
|h|
and obtain
=
G
G
I
g
I
h
with G outer and with O
|g|
=
1
G
it follows G (H

)
1
.
If I
g
and I
h
are constant this would be the proof since the constant inner function 1 is a Blaschke
product and G multiplied by a constant is still in H

(If conversely just one of them is constant, the

following gets reduced to one function).
Therefore we are now assuming that the inner functions I
g
and I
h
are non constant and hence we can
apply Theorem 6.11 (Frostmans theorem) to both of them. Since the theorem holds for all , [[ < 1,
except possibly for a set of capacity zero and D is not the sum of two disjoint sets of capacity zero,
there ecists a D such that
I
a
=
I
g
a
1 aI
g
= I
g
1 aI
g
1 aI
g
= I
g
E
E
J
a
=
I
h
a
1 aI
h
= I
h
1 aI
h
1 aI
h
= I
h
F
F
with E := 1 aI
g
and F := 1 aI
h
. Therefore
I
g
= I
a
E
E
I
h
= J
a
F
F
with I
a
, J
a
Blaschke products. If we now merge everything together, we obtain
f = I
a
E
E
J
a
F
F
G
G
=
I
a
J
a
EFG
EFG
=
b
B
EFG
EFG
with b := I
a
and B := J
a
.
To nish the proof, we just need to show, that E, F are elements of (H

)
1
and are outer functions,
since the group structure of outer functions yield to g := EFG an outer function. It holds

1 aI
g

1 +

aI
g

E H

[aI
g
[ [a[ < 1
1
E
H

.
Following the same route for F yields E, F (H

)
1
and by using Corollary 6.13 we see that E, F
are outer functions.
Remark The conjugate part of the factorization does not have to be the numerator. By looking at
the proof closely you can change the denitions of E, F and G to
1
E
,
1
F
and
1
G
and one obtains an
equivalent factorization just with the conjugate part as a divisor.
The following lemma is the reason why we can use Burgains factorization theorem in our next theorem
Lemma 5.12 Let L

0. If log [[ L
1
then Ker
p
T

= 0.
Proof Let us assume Ker
p
T

= 0. Then with Lemma 5.8 we obtain f Ker

p
T

f H
p
0
.
Therefore it exists g H
p
0
with
f = g. (5.7)
Because L

and 0 there exists a set of positive measure E such that

(z) = 0 z E.
5 KERNELS OF TOEPLITZ OPERATORS 32
Therefore g 0 and is integrable since f is analytic and therefore just zero on a set of measure zero.
Now from (5.7) it follows that
log [[ + log [f[ = log [g[
and by using Jensens inequality (Theorem 3.1) log [f[ and log [g[ are integrable and hence L
1
which proves the lemma.
The next theorem is the mayor result of Konstantin Dyakonovs paper [17] which got published in 2000
in [17] and is the core of this paper. This theorem called my attention to the topic explained here and
further parts concerning the theory of H
p
-functions.
Theorem 5.13 (K. Dyakonov) 1. For any L

, 0, there exists a triple (B, b, g) such that

Ker
p
T

=
g
b
(K
p
B
bH
p
) p [1, ]. (5.8)
2. Conversely, given a triple (B, b, g) one can nd a L

for which (5.8) holds true. In fact, it

suces to put =
bg
Bg
.
Proof
1. First of all, if log[[ L
1
, it follows from Lemma 5.12 that Ker
p
T

= 0. In this case, we can

make the right hand side of (5.8) also trivial by putting e.g. B = b = g = 1.
Assuming from now on that log[[ L
1
we dene the outer function F := O
||
and set =

F
.
Then [[ = 1 (shown in the proof of Lemma 5.1) and Theorem 5.11 provides a triple (B, b, g)
satisfying =
b
B
g
g
. Now let 1 p and x f H
p
, f 0. In order that
f Ker
p
T

. (5.9)
Lemma 5.8 says that it is necessary and sucient that f H
p
0
. This can be rewritten as
f
1
F
H
p
0
. (5.10)
With [f[ = [f[ L
p
and the result of Corollary 5.7, (5.10) reduces to
f H
p
0
. (5.11)
Next we invoke the factorization to restate (5.11) as
fb
g
BH
p
0
(5.12)
recalling that g (H

)
1
. Since
fb
g
belongs to bH
p
and hence to H
p
(Lemma 3.8), we may as
well write (5.12) in the form
fb
g
BH
p
0
H
p
bH
p
= K
p
B
bH
p
. (5.13)
Finally (5.13) amounts to the inclusion
f
g
b
(K
p
B
bH
p
) (5.14)
and the resulting equivalence relation (5.9) (5.14) proves (5.8).
2. Given triple (B, b, g) dene =
bg
Bg
and set = , F = O
||
= 1. This done, the above reasoning
shows that conditions (5.9) (5.14) are all equivalent, whenever f H
p
(1 p ). A
juxtaposition of the endpoint statements (5.9) and (5.14) yields to the equality, just as before.
5 KERNELS OF TOEPLITZ OPERATORS 33
Five years later Stephan Ramon Garcia extended the work of Dyakonov (which was entirely dierent
as well) but just for 0 < p < . He dened an easy malleable space which is equal to the kernels of
the Toeplitz operators. We start with some necessary denitions:
Denition 5.13.1 1. A function f belonging to the Smirnov class N
+
is called a real Smirnov
function, f R
+
if its boundary function is real valued a.e. on T.
2. Let C
+
denote the subspace R
+
+iR
+
of N
+
.
3. We dene A
p

to be C
+
H
p
with an outer function.
4. Let f = h A
p

for some outer function and h C

+
and p (1, ). Since the functions f
and

f := h have the same modulus on T, they share the same outer factor, say F. Therefore

f H
p
and since h C
+
it follows that

f A
p

. We refer to f and

f as a conjugate pair in A
p

.
A set like R
+
for functions in H
p
would not make any sense due to the following lemma.
Lemma 5.14 If f H
1
and f(e
i
) R T, then f constant.
Proof With the Poisson kernel representation
f(z) =

T
{
z
()f(e
i
)dm()
and the fact that {
z
() R it follows that f(z) R z D which implies that f has to be constant.
Before we state the results of Stephan Ramon Garcia, let us obtain some preliminary results.
Lemma 5.15 If f and

f are conjugate functions in A
p

, then

f =
f

.
Proof If f A
p

, then h C
+
such that f = h. Consequently

f = h =
h

=
f

.
Proposition 5.16 Two functions f and g in H
p
satisfy g =
f

a.e. on T for some outer function

if and only if f and g belong to A
p

and are conjugate pairs.

Proof "" If f, g belong to H
p
and satisfy g =
f

a.e. on T, then they have the same modulus on T

and hence the same outer factor, say F. Writing f = I
f
F and g = I
g
F, it follows that
I
f
I
g
F
F
=

and
combining it with
I
f
I
g
= I
f
I
g
I
f
+I
g
I
f
+I
g
=
I
f
+I
g
1
I
f
+
1
I
g
=
I
f
+I
g
I
f
+I
g
=
1
2
(I
f
+I
g
)
1
2
(I
f
+I
g
)
=
1
2i
(I
f
I
g
)
1
2i
(I
f
I
g
)
(5.15)
on T yields
1
2
(I
f
+I
g
)F

=
1
2
(I
f
+I
g
)F

and
1
2i
(I
f
I
g
)F

=
1
2i
(I
f
I
g
)F

. (5.16)
This means, that the functions
1
2
(I
f
+I
g
)F

and
1
2i
(I
f
I
g
)F

are real-valued and since outer functions are

multiplicative and N
+
is a vector space there are in R
+
. Hence we see a and b in R
+
such that
1
2
(I
f
+I
g
)F = a and
1
2i
(I
f
I
g
)F = b. Thus f = (a +ib) and g = (a ib) belong to A
p

and are
a conjugate pair.
"" Since f, g are conjugate pairs in A
p

, g =
f

follows from Lemma 5.15 and f, g H

p
follows from
the denition of A
p

.
5 KERNELS OF TOEPLITZ OPERATORS 34
Denition 5.16.1 Let f = I
f
F and

f = I

f
F are a conjugate pair in A
p

. Then with Lemma 5.15 we

obtain I
f
I

f
F =
F

which shows by using Proposition 5.16 that F belongs to A

p

. Therefore it exists a
conjugate function

F of F with

F =
F

which leads to
I
f
I

f
F =

F. (5.17)
Moreover the inner function J
F
:= I
f
I

f
=
F
F
depends only upon F and . We call this inner function
J
F
the associated inner function of F in A
p

.
Remark The functions f = I
f
F in A
p

with outer factor F are precisely those functions whose inner

factor I
f
divides J
F
.
We show that the nontrivial kernels of Toeplitz operators Ker
p
T

for p (1, ) coincide with the

spaces A
p

, where is an outer function in H

p
. We require the following result of Hayashi which got
published in [15]:
Theorem 5.17 (Hayashi) Let T

be a Toeplitz operator on H
p
for some p (1, ). If Ker
p
T

is
non trivial, then there exists an outer function in H
p
such that Ker
p
T

= Ker
p
T
z

.
Proposition 5.18 If p (1, ) and H
p
is outer, then Ker
p
T
z

= A
p

.
Proof "" Since f H
p
, f = I
f
F with I
f
inner and F outer. Let f belongs to Ker
p
T
z

which is
equivalent to I
f
F
z

H
p
0
. Therefore g H
p
such that l :=
I
f
Fz

= zg. Since l and g have the same

modulus on T, they share the same outer factor. Therefore
I
f
Fz

= zIF
for some inner function I. Consequently
IF =
I
f
F

which implies by using proposition 5.16 that f is in A

p

.
"" If f belongs to A
p

, then f = h for some h in C

+
. Since P
+
(
z

f) = P
+
(zh) it follows that
T
z

= 0, since the function

f = h is in H
p
.
If is an outer function in H
p
, then Bourgains factorization theorem for unimodular functions (The-
orem 5.11) provides a triple (B, b, g) such that

=
b
B
g
g
. (5.18)
Our nal proposition relates A
p

spaces, B-invariant subspaces, and the kernels of Toeplitz operators

for p (1, ).
Now we are able to state our nal result which was published in 2005 in [18].
Theorem 5.19 (S.R. Garcia) Let p (1, ) and let be an outer function in H
p
. If (

) =
b
B
g
g
where (B, b, g) is a triple, then the following subspaces of H
p
are identical:
1. A
p

.
2. Ker
p
T
z

.
5 KERNELS OF TOEPLITZ OPERATORS 35
3.
g
b
(bH
p
BH
p
).
4. f g(H
p
BH
p
) : bI
f
[J
F/g
.
Remark The notation J
F/g
in (4) refers to the associated inner function for the outer function F/g
in H
p
bH
p
.
Proof The equality of (1) and (2) follows from Proposition 5.18.
"(1)(4)" If f = I
f
F belongs to A
p

, then

f = I

f
F =
I
f
F

= I
f
F
B
b
g
g
.
Hence
bI
f
I

f
F/g = (F/g)B. (5.19)
Since bI
f
I

f
(F/g) H
p
, it follows with (5.19) that bI
f
I

f
(F/g) BH
p
. Therefore F/g belongs to
H
p
BH
p
and since this is a B-invariant subspace F/g has the associated inner function J
F/g
:= bI
f
I

f
.
"(1)(4)" Suppose f = I
f
F belongs to g(H
p
BH
p
). Then (F/g)B H
p
and consequently (F/g)B
H
p
. Since

(F/g)b H
p

= [F/g[, O
[(F/g)B[
= F, the canonical function of Fb is
FB = FI
for some inner function I. The function I is the associated inner function to (F/g) and therefore
I = J
F/g
. With bI
f
[J
F/g
we obtain J
F/g
= bI
f
J for some inner function J. If we merge everything
together we obtain
J
F/g
= bI
f
JF/g = (F/g)B.
This is equivalent to

F := I
f
JF = F
B
b
g
g
=
F

.
Hence from the denition of the associated inner function, f belongs to A
p

(with conjugate JF).

"(3)(4)" If f = I
f
F belongs to
g
b
(bH
p
BH
p
), then
f =
g
b
(bI(F/g))
and bI
f
(F/g) belongs to H
p
BH
p
since bH
p
H
p
. So bI
f
(F/g) belongs to a backward shift invariant
subspace and F/g is outer.
This means that bI
f
must divide the associated inner function J
F/g
corresponding to the outer function
F/g. Therefore bI
f
[J
F/g
.
"(3)(4)" If f = I
f
F belongs to g(H
p
BH
p
) and bI
f
[J
F/g
, then bI
f
(F/g) belongs to H
p
bH
p
by assumption. Since BI
f
F
g
belongs to H
p
bI
f
(F/g) belongs to BH
p
and therefore f =
g
b
(bI
f
(F/g))
belongs to
g
b
(bH
p
BH
p
) which proves this direction.
6 APPENDIX 36
6 Appendix
6.1 Factorization for functions in H
p
We are now giving in this subsection a detailed derivation of the factorization of functions in H
p
. We
start with a few results about outer functions.
Denition 6.0.1 (outer function) An analytic function F in D of the form
F(z) = exp

1
2

T
e
i
+z
e
i
z
k()d

where k is a real-valued integrable function on T and is a complex number of modulus 1 is called an

outer function.
If the outer function is in H
1
, then k is determined due to the following lemma.
Lemma 6.1 If F is outer in H
1
(D), we have necessarily k() = log [F(e
i
)[.
Proof If we take the modulus of F we obtain since k() is real valued
[F(z)[ = exp

Re

T
e
i
+z
e
i
z
k()dm()

= exp

T
{
z
()k()dm()

.
Then by Theorem 2.14 we obtain
log

F(e
i
)

= k() almost everywhere.

An useful lemma:
Lemma 6.2 Let F be an outer function, then F H
1
if and only if e
k
is integrable.
Proof "" follows from Lemma 6.1.
"" It holds for 0 < r < 1 and z = re
i

T
[F(z)[ d =

T
exp

T
{
z
()k()dm()

d
and let d() := {
z
()dm(), then d() is a probability measure. Since k is real-valued and the
exponential function is convex we can apply Jensens inequality for convex functions (Theorem 6.12)
to obtain

T
exp

T
k()d()

T
e
k()
d()d.
Now we can use Fubini-Tonelli to switch the integrals and since

T
{
z
()
d
2
= 1, we arrive at the nal
estimate

T
[F(z)[ d

T
e
k()
d < .
One can state the connection between an outer function and a function in H
1
in the following way.
Lemma 6.3 Let F be outer, f H
1
and k() = log [f(e
i
)[, then F H
1
and [F[ = [f[ a.e. on T.
6 APPENDIX 37
Proof By using Lemma 6.1 we have log [F(e
i
)[ = k() = log [f(e
i
)[ almost everywhere. The rest
follows from Lemma 6.2.
We now provide another useful denition we employ.
Denition 6.3.1 Let f be in H
1
. The function
F(z) = exp

T
e
i
+z
e
i
z
log [f(e
i
)[dm()

(6.1)
where is a complex number of modulus 1 is called the associated outer function. This is well dened
since log [f(e
i
)[ is real-valued on T and with Jensens inequality (Theorem 3.1) Lebesgue integrable.
A deeper result then Lemma 6.3 can be obtained.
Theorem 6.4 Let F H
1
and F 0. The following are equivalent:
1. F is an outer function.
2. If f is any function in H
1
such that [F[ = [f[ a.e. on T, then [F(z)[ [f(z)[ z D.
Proof "" We use Jensens inequality (Theorem 3.1) and get
log [f(z)[

T
{
z
(e
i
) log

f(e
i
)

dm() = log [F(z)[ . (6.2)

Thus [F(z)[ = [f(z)[ z D.
"" If (2) holds, we dene another outer function G:
G(z) = exp

T
e
i
+z
e
i
z
log

F(e
i
)

dm()

. (6.3)
Then we use Lemma 6.3 and nd that [F[ = [G[ a.e. on T. Now we can use step (1)(2) and obtain
[F(z)[ [G(z)[. On the other hand one easily checks that [G(z)[ [F(z)[.
Thus
F
G
is analytic and always absolute value 1. So F = G, where [[ = 1, which proves that F is an
outer function.
Another useful characterization is the following theorem.
Theorem 6.5 Let F be a non zero function in H
1
. The following are equivalent:
1. F is an outer function.
2. log [F(0)[ =

T
log

F(e
i
)

dm().
Proof "" This is obviously true for all outer functions since {
0
() = 1.
"" We dene G as in the theorem above and following the same route to obtain F(z) G(z) z D.
Thus
F(z)
G(z)
1 z D but since
F(0)
G(0)
=
exp

T
log

F(e
i
)

dm()

exp

log [F(e
i
)[ dm()
= 1
the maximum principle yields
F(z)
G(z)
= z D with [[ = 1.
Let us now have a look at the function g(z) :=
f(z)
F(z)
z D with f H
1
and F the associated outer
function. Since [F[ = [f[ a.e. on T it follows that g is a bounded function with
[g(z)[ =
f(z)
F(z)
1 z D.
It will be convenient for us to make the following denition.
6 APPENDIX 38
Denition 6.5.1 (inner function) An inner function is an analytic function g on D such that
[g(z)[ 1 z D and [g[ = 1 a.e. on T.
With this result we are able to do a preliminary factorization:
Theorem 6.6 Let f be a non zero function in H
1
. Then f can be written in the form f = gF, where
g is an inner function and F is the associated outer function in H
1
. This factorization is unique up to
a constant of modulus 1.
Proof By dening
F(z) = exp

T
e
i
+z
e
i
z
log

f(e
i
)

dm()

we obtain with Lemma 6.3 that F H

1
and with Theorem 6.4 that the function g(z) :=
f(z)
F(z)
z D is
well dened and also an inner function. If we also have f = g
1
F
1
with g
1
inner function and F
1
outer,
then [F[ = [F
1
[ on T. Thus F = F
1
with [[ = 1 and hence g
1
F
1
= g
1
F
1
and g
1
= g.
Our next step is to factorize the inner function into two more specialized inner functions called Blaschke
product and singular function. Since an outer function does not contain any zeros, they are contained
in the inner function. In the next factorization the Blaschke product will take care of the zeros when
our analytic functions satisfy certain Blaschke conditions.
Theorem 6.7 Let f be a bounded analytic function in D and suppose f(0) = 0. If
k
is the sequence
of zeros of f in D, each repeated as often as the multiplicity of the zero of f, then the product

kN
[
k
[
is convergent ,i.e.

kN
(1 [
k
[) .
Proof Without loss of generality, we assume [f[ 1. Since for a nite number of zeros the case is
trivial, we assume that f has a countable number of zeros. Let B
n
(z) be the nite product:
B
n
(z) =
n

k=1

k
[
k
[

k
z
1
k
z
(6.4)
Now B
n
(z) is a rational complex valued function and since the poles are lying outside of D it is analytic
inside. It holds for all [0, 2]

k
[
k
[

k
e
i
1
k
e
i

k
e
i
1
k
e
i

k
e
i
e
i

k
e
i

e
i

= 1
and therefore

B
n
(e
i
)

= 1 n N. Furthermore
f
B
n
is a bounded analytic function in D. Since n N

f(e
i
)

[B
n
(e
i
)[
=

f(e
i
)

1 almost everywhere.
we have [f(z)[ [B
n
(z)[ on D. In particular
0 < [f(0)[ [B(0)[ =
n

k=1
[
k
[ .
Since [
k
[ < 1 k N and since each of the partial products

n
k=1
[
k
[ is not less then [f(0)[, the
innite product converges.
6 APPENDIX 39
Remark Let us assume that the
k
are not nite then
1. Since f 0, the identity theorem for holomorphic functions states that the accumulation point
cannot lie in D and lies therefore either on T or outside of D. Consequently the modulus of the

k
increases when k goes to innity.
2. It is impossible to have an with f() = 0 and innity multiplicity since with [[ < 1 it would
lead to

kN
[[
k
= 0.
3. The zeros of f are not allowed to lie too dense in D. This can be illustrated with the following
example:
Let
k
:= 1
1
k
k N. Since the convergency of

kN
[
k
[ is equivalent to

kN
(1[
k
[) <
the
k
cannot be the zeros of an bounded analytic function, since

kN
(1 [
k
[) =

kN
(1
1 +
1
k
) =

kN
(
1
k
) is the harmonic series.
4. If we drop the boundedness of the analytic function Remark (3) becomes false since the function
f(z) = sin(

1z
) vanishes at
k
k N and is analytic in D.
Theorem 6.8 (and Denition of the Blaschke product) Let
k
be a sequence of points in D
such that

kN
(1 [
k
[) < . Let m be the number of
k
equal to zero. Then the Blaschke product
B(z) = z
m

|
k
| =0

k
[
k
[

k
z
1
k
z
converges on D. The function B(z) is in H

and the zeros of B(z) are precisely the points

k
, each zero
having multiplicity equal to the number of times it occurs in the sequence
k
. Moreover a Blaschke
product is an inner function.
Proof We can suppose [
k
[ > 0 n N. Let
b
k
(z) :=

k
[
k
[

k
z
1
k
z
. (6.5)
With Theorem 15.6. in [8] the product

kN
b
k
converges on D to an analytic function having the

k
for zeros if and only if

kN
[1 b
k
(z)[ converges uniformly on each compact subset of D. By
calculation it follows
[1 b
k
(z)[ =

[
k
[(1
k
z)
k
(
k
z)
[
k
[(1
k
z)

[
k
[ [
k
[
k
z [
k
[
2

k
+z
k
[
k
[(1
k
z)

[
k
[ +z
k
[
k
[(1
k
z)

(1 [
k
[)

[
k
[ +[z[[
k
[
[
k
[[1
k
z[
(1 [
k
[)
=
1 +[z[
1 [
k
[[z[
(1 [
k
[)

1 +[z[
1 [z[
(1 [
k
[)
and the convergence follows from

kN
[1 b
k
(z)[ < . Since [b
k
(z)[ 1 z D, it is clear that
B(z) H

and [B(z)[ 1. The bounded harmonic function B(z) has non tangential limits [B(e
i
)[ 1
almost everywhere. To see that [B(e
i
)[ = 1 a.e., set B
n
(z) :=

n
k=1
b
k
(z). Then
B
B
n
is another Blaschke
product and

B(0)
B
n
(0)

B(e
i
)

[B
n
(e
i
)[
dm() =

B(e
i
)

dm().
6 APPENDIX 40
Letting n now yields

T

B(e
i
)

dm() = 1
so that

B(e
i
)

= 1 almost everywhere.
Remark We are now able to factorize every inner function into a Blaschke product B and a function
g without zeros since g =
f
B
is analytic and bounded in D. The factorization f = gB is unique since a
Blaschke product is uniquely determined by its zeros.
Theorem 6.9 Let g be an inner function without zeros and suppose that g(0) is positive, then there is
a unique singular positive measure on T such that
g(z) = exp

T
e
i
+z
e
i
z
d()

. (6.6)
Proof Since g is analytic in the disc and has no zeros, it follows from a well-known theorem that
g = e
h
, where h is an analytic function in the disc. Since g is bounded by 1, the real part of h must
be non negative in D. Let h = u + iv so that u 0. The non-negative harmonic function is uniquely
expressible in the form
u(z) =

T
{
z
()d()
where is a positive measure on T and z = re
i
. Since g(0) > 0, we have e
uiv
and consequently
v(0) = 0 or with e
2ni
= 1 n N, v(0) = 2ni. Without loss of generality, v(0) = 0 and thus
h(z) =

T
e
i
+z
e
i
z
d()
where is a positive measure on T. This is due to Herglotzs theorem (Theorem 6.14) since h takes
values in the right half plane. Now we have as an inner function [g[ = 1 a.e. on T. Since [g[ = e
u
,
this just means that the non-tangential limits of u must vanish almost everywhere on T. But by using
Fatous theorem these non-tangential limits are equal to 2
d
d
= 2f with f the Radon-Nikodym
derivative. By using Lebesgues decomposition theorem (Theorem 6.16) we obtain
d = fd +d
s
where d
s
is a singular measure with respect to d. Since the Radon-Nikodym derivative is zero a.e.
d is singular and this completes the proof.
Denition 6.9.1 (singular function) We dene an inner function without zeros which is positive at
the origin a singular function.
Theorem 6.10 Let f 0 be an H
1
function on D. Then f is uniquely expressible in the form
f = BSF, where B is a Blaschke product, S is a singular function and F is the associated outer
function (in H
1
).
Proof We know from our previous results (Theorem 6.6) that f = gF, where g is an inner function
and F is the associated outer function and that this factorization is unique up to a constant multiple
of modulus 1. If B is the Blaschke product formed from the zeros of g (i.e. the zeros of f), then
g = BS, where S is an inner function without zeros. By multiplying g by a constant of modulus 1, we
can arrange that S(0) > 0 ,i.e. that S is a singular function. We then absorb that constant into the
associated outer function F and we are done!
6 APPENDIX 41
6.2 Theorems used in this paper
Theorem 6.11 (Frostmans theorem) Let f(z) be a non-constant inner function on D. Then for
all , [[ < 1, except possibly for a set of capacity zero, the function
f

(z) =
f(z)
1 f(z)
(6.1)
is a Blaschke product.
Remark This is the key part of the proof that the set of Blaschke products is uniformly dense in the
set of inner functions.
Theorem 6.12 (Jensens inequality for convex functions) Let (, ) be a measure space such
that is a probability measure ,i.e. () = 1. Let L
1
() be a real-valued function and let (t) be
a convex function on R. Then
(

d)

()d. (6.2)
Corollary 6.13 If f(z) H
p
and if for some r > 0,
1
f(z)
H
r
, then f(z) is an outer function.
Theorem 6.14 (Herglotzs theorem) Every analytic function in the unit disc with values in the
right half-plane such that f(0) > 0 has the form
f(z) =

T
e
i
+z
e
i
z
d() (6.3)
where is a nite positive measure on T.
Theorem 6.15 (Radon-Nikodym theorem) Given a measurable space (, ), if a -nite on
(, ) is absolutely continuous with respect to a -nite measure on (, ), then there is a measurable
function f on and taking values in [0, ) such that
(A) =

A
fd (6.4)
for any measurable set A.
Remark The function f satisfying the above equality is uniquely dened up to -null set. The function
f is commonly written as
d
d
and is called the Radon-Nikodym derivative.
Theorem 6.16 (Lebesgues decomposition theorem) Given and , two -nite measures on a
measurable space (, ), there exist two -nite signed measures
a
and
s
such that
1. =
a
+
s
.
2.
a
<.
3.
s
.
6.3 Acknowledgments
I thank Prof. Konstantin Dyakonov for accepting my request of beeing my supervisor for this bachelor
thesis since it is not part of the Erasmus program, for his numerous private lectures which gave me
a more detailed understanding of the topic and for his support especially at the end of the project.
Also, it is my pleasure to thank Prof. A.I. Bobenko who gave me the possibility to write this paper in
Barcelona during my Erasmus year.
REFERENCES 42
References
[1] J.R. PARTINGTON An Introduction to Hankel Operators - London Mathematical Society Student
Texts
[2] R.A. MARTINEZ-AVENADO and P.ROSENTHAL An Introduction to Operators on the Hard-
Hilbert Space - Graduate Texts in Mathematics
[3] G.H. HARDY On the mean value of the modulus of an analytic function - Proceedings of the
London mathematical society series 2 14:269-277
[4] F. RIESZ Ueber die Randwerte einer analytischen Funktion - Math.Z. 18:87-95
[5] F. and M. RIESZ Ueber die Randwerte einer analytischen Funktion - Quatrime Congrs des
Math. Scand. Stockholm (1916) 27-44
[6] M. RIESZ Les fonctions conjugeres et les sries de Fourier - C.R.Acad. Sci. Paris, Sr. A-B 178
(1924), 1464-1467
[7] J.D. NEWMAN The non.existence of projections from L
1
to H
1
- Proc. Amer. Math.Soc. 12
(1961), 98-99
[8] W. RUDIN Real and Complex Analysis - International Series in Pure and Applied Mathematics
[9] P. FATOU Sries trigonomtriques et seris de Taylor - Acta Math. 30 (1906), no.1, 335-400
[10] P.L. DUREN Theory of H
p
spaces - Pure and Applied Math., Vol. 38, Academic Press, New
York-London, 1970
[11] JOHN B. GARNETT, Bounded Analytic Functions - Graduate Texts in Mathematics
[12] JOSEPH A. CIMA and WILLIAM T. ROSS, The Backward Shift on the Hardy Space - Mathe-
matical Surveys and Monographs
[13] V.I. SMIRNOV, Sur les valeurs limites de fonctions rgulires linterrieur dun cerle - Journal
de la Socit Phys.-Math. de Lningrade, 2(1929), no. 2, 22-37
[14] K. HOFFMAN, Banach Spaces of Analytic Functions - Dover Publications, Inc., New York, 1988.
[15] E.HAYASHI, The solution sets of extremal problems in H
1
- Proc. Amer. Math. Soc. 93(1985),
690-696.
[16] E.HAYASHI, Classication of the nearly invariant Subspaces of the Backward Shift - Proc. Amer.
Math. Soc. 110(1990), 441-448.
[17] K. DYAKONOV, Kernels of Toeplitz Operators via Bourgains factorization theorem - J. Funct.
Anal. 170 (2000), 93-106
[18] S.R. GARCIA, Conjugation, the backward shift, and Toeplitz kernels - J. Operator Theory 54
(2005), no. 2, 239-250.