Banach Spaces of analytic functions

(MAGIC 082)
Jonathan R. Partington
University of Leeds, School of Mathematics
J.R.Partington@leeds.ac.uk
March 25, 2014

1
LECTURE 1
Books:
K. Hoffman, Banach spaces of analytic functions.
P. Koosis, Introduction to Hp spaces.
W. Rudin, Real and complex analysis.
N. Nikolski, Operators, functions and systems, an easy reading, Vol. 1.

1 Introduction
Recall that a Banach space is a complete normed space, and a Hilbert space
is a special kind of Banach space, where the norm is given by an inner prod-
uct. You should have met Lp and `p for 1 ≤ p ≤ ∞ and C(K) (the space of
continuous functions on K).

We are going to work with complex Banach and Hilbert spaces whose ele-
ments are functions.

1.1 Examples (treated informally for the moment)

1. The Hardy spaces H p (1 ≤ p ≤ ∞) are Banach spaces consisting of
analytic functions in the unit disc

D = {z ∈ C : |z| < 1},

whose boundary values are in Lp (T), where

T = {z ∈ C : |z| = 1}

is the unit circle. Thus ∞

X
f (z) = an z n
n=0

for |z| < 1, and on |z| = 1, f corresponds to an Lp function (pointwise radial

limits almost everywhere, in fact).
Recall that functions of period 2π have Fourier series, that is,
∞
X
g(t) ∼ cn eint ,
n=−∞

2
where Z 2π
1
cn = g(t)e−int dt.
2π 0

If g ∈ Lp (0, 2π) for some 1 ≤ p ≤ ∞, then g ∈ L1 (0, 2π), and

Z 2π
1
|cn | ≤ |g(t)||eint | dt = kgkL1 ,
2π 0
so the sequence (cn ) is bounded. Now if cn = 0 for all n < 0, we can define
∞
X
f (z) = cn z n ,
n=0

and it converges for |z| < 1.

So we think of H p either as a space of functions defined on the disc, or as
functions on the circle, putting f (eit ) = g(t), or indeed as functions of period
2π.

In the special case p = ∞, H ∞ consists of all bounded analytic functions on

D, and it forms a Banach algebra.

p
2. The harmonic Hardy spaces. We may proceed similarly for the set
P∞of all L
int
functions on the circle. Now the natural extension to the disc of −∞ cn e
is going to be harmonic rather than analytic, i.e., we take
∞
X ∞
X
cn z n + c−n z n .
n=0 n=1

(Recall that harmonic functions satisfy

∂ 2f ∂ 2f
+ 2 =0
∂x2 ∂y
and include analytic functions such as z n and anti-analytic ones such as z n .)
We define hp to be the space of harmonic functions in the disc with Lp bound-
ary values.

3. The disc algebra A(D). These are the functions continuous on the closed
unit disc D, and analytic on the open unit disc. We use the supremum norm.

3
This is a closed subalgebra of H ∞ .

4. The Bergman spaces. These are again spaces of analytic functions in the
disc, but now in Lp on the disc, i.e., f ∈ Ap for 1 ≤ p < ∞ if and only if f
is analytic in D and
 Z 1/p
1 p
kf kp = |f (z)| dA(z) < ∞,
π D

where dA(z) is two-dimensional Lebesgue measure, i.e.,

1 1
Z Z 2π
p
kf kp = |f (reiθ )|p r dr dθ.
π r=0 θ=0
In the case p = 2, if f has a Taylor series
∞
X
f (z) = an z n ,
n=0

then ∞
2
X |an |2
kf k = .
n=0
n+1
5. The Wiener algebra, W . Take all functions of period 2π that can be
written as ∞
X
g(t) = cn eint ,
n=−∞
P∞
with kgkW = −∞ |cn | < ∞. Here the series converges uniformly so g is
actually continuous. Transferring to the circle we can write
∞
X
f (z) = cn z n ,
−∞
P∞
with −∞ |cn | < ∞.
The positive
P∞ or analytic Wiener algebra W+ consists of all analytic functions
n
f (z) = n=0 cn z in the disc with
∞
X
kf kW = |cn | < ∞.
0

4
These functions all lie in A(D).
As the name suggests, both W and W+ are Banach algebras.

Elementary inclusions:

W+ ⊂ A(D) ⊂ H ∞ ⊂ H 2 ⊂ H 1

(and many others).

6. The Paley–Wiener space P W (K), where K ⊂ R is compact. Let f be a

function on R which can be written as
Z ∞
1
f (t) = fˆ(w)eiwt dw,
2π −∞

where fˆ(w) = 0 for w 6∈ K, and fˆ|K ∈ L2 (K). Here fˆ is the Fourier transform
of f , and can be defined (for f ∈ L1 (R)) by
Z ∞
ˆ
f (w) = f (t)e−iwt dt.
−∞

It turns out that such f can be extended to be analytic on the whole of C,

and that Z ∞ Z ∞
1
2
kf k = 2
|f (t)| dt = |fˆ(w)|2 dw,
−∞ 2π −∞

makes P W (K) into a Banach space, indeed even a Hilbert space.

An important special case is K = [−a, a], where a > 0. Then P W (K) is the
space of band-limited signals (no high-frequency components).

7. The Segal–Bargmann (or Fock) space SB2 consisting of all functions f

analytic on the whole of C such that
Z
1 2
2
kf k = |f (z)|2 e−|z| dA(z) < ∞.
π C
This has applications in quantum mechanics.

5
2 Hardy spaces on the disc
We begin with H 2 , which is a Hilbert space. Write

T = {z ∈ C : |z| = 1} = {eiθ : 0 ≤ θ ≤ 2π}.

dθ
Normalize it to have measure 1, so all integrals will be with respect to .
2π
2.1 Proposition
The set of functions

{z n : n ∈ Z} = {einθ : n ∈ Z}

forms an orthonormal basis for L2 (T), and its linear span is dense in Lp (T)
for all 1 ≤ p < ∞.
Here we use the inner product on L2 (T) given by
Z 2π
1
hf, gi = f (eiθ )g(eiθ ) dθ.
2π 0

Proof: We see that

(
2π
0 if m 6= n,
Z
1
hz n , z m i = einθ e−imθ dθ =
2π 0 1 if m = n,

and hence we have the orthonormality.

Recall now that if f ∈ Lp (0, 2π) and ε > 0, then there is a continuous
P∞ periodic
function g with kf − gkLp < ε/2. Next, g has a Fourier series n=−∞ cn einθ ,
whose Fejér sums
m k
1 X X
Gm (θ) = cn einθ
m + 1 k=0 n=−k
converge uniformly to g on [0, 2π]. Hence there is a trigonometric polynomial
Gm with
kg − Gm kLp < ε/2,
since uniform convergence implies Lp convergence. Now

kf − Gm k ≤ kf − gk + kg − Gm k < ε.

6
Translating this to the circle with z = eiθ we see that for f ∈ Lp (T) we have
N
X
kf − an z n k p < ε
−N

for suitable a−N , . . . , aN ∈ C.



7
LECTURE 2

2.2 Definition
The Hardy space H 2 consists of all functions
∞
X
f (z) = an z n
n=0

analytic in the unit disc D such that

∞
X
2
kf k = |an |2 < ∞.
n=0

It can naturally
P∞be identified with the closed subspace of L2 (T) consisting of
all functions n=0 an einθ (convergence in the L2 sense) with
∞
X
|an |2 < ∞,
n=0

i.e., the closed linear span of {einθ : n ≥ 0}.

Note that if |z| < 1 then
X X 1/2 X 1/2
|an z n | ≤ |an |2 |z 2n | < ∞,

by Cauchy–Schwarz, so all these power series have radius of convergence at

least 1, and define analytic functions in D.

2.3 Proposition
H 2 is a Hilbert space with inner product
*∞ ∞
+
X X X
an z n , bn z n = an b n ,
0 0

or equivalently Z 2π
1
hf, gi = f (eiθ )g(eiθ ) dθ.
2π 0

8
 Proof: This is clear, since it is a closed subspace of L2 (T).
Alternatively, it’s clear since the correspondence
∞
X
(an )∞
n=0 ←→ an z n
n=0
2 2
identifies H with ` (Z+ ).

Analytic and harmonic extensions

Recall the Dirichlet problem: given f continuous on the unit circle, find g
harmonic on the interior which extends it to a continuous function on the
closed disc. These can be real or complex functions.
Recall also that if g is analytic, then its real and imaginary parts are both
harmonic, so g itself is. but if g = u + iv, then g = u − iv is also harmonic.
So if f (eiθ ) = einθ for some n ∈ Z, then a solution is
(
rn einθ = z n if n ≥ 0,
g(reiθ ) = m −imθ m
r e =z if n = −m < 0,

so that if f (einθ ) ∼ ∞ inθ

P
n=−∞ an e , then
∞
X
iθ
g(re ) = an r|n| einθ
n=−∞

is the ‘natural’ extension to D.

2.4 Definition
The Poisson kernel P (z, w), defined for 0 ≤ |w| < |z| ≤ 1, is the function
|z|2 − |w|2
P (z, w) = .
|z − w|2
1−|w|2
When |z| = 1, this equals |1−wz|2
, since |z − w| = |zz − wz| = |1 − wz|. This
equals
1 − ww 1 1
= + −1
(1 − wz)(1 − wz) 1 − wz 1 − wz
X∞ ∞
X
= wn z n + wn z n .
n=0 n=1

9
Thus ∞
X
it iθ
P (e , re ) = r|n| ein(t−θ) = Pr (t − θ),
n=−∞
say.
Clearly the Poisson kernel is a positive continuous function of z and w for
0 ≤ |w| < |z| ≤ 1.
Note that ∞ ∞
X X
|n| in(t−θ)
r e = r|n| ein(θ−t) ,
n=−∞ n=−∞

so Pr (t − θ) = Pr (θ − t).

2.5 Theorem
If f ∈ L1 (T), then the function F defined on D by
Z 2π
iθ 1
F (re ) = Pr (θ − t)f (eit ) dt,
2π 0
is a harmonic function, the harmonic extension of f .

Proof: If f is real, then since for z = eit and w = reiθ we have

1 1
Pr (θ − t) = + −1
1 − wz
 1 − zw 
1 1
= 2 Re −
1 − zw 2
 
1 + zw
= Re ,
1 − zw
we see that
2π
1 + e−it w
Z  
iθ 1
F (w) = F (re ) = Re f (t) dt,
2π 0 1 − e−it w
which is the real part of an anlytic function of w, hence harmonic.

Now, a general f is g + ih with f, g real, and F = G + iH, where G, H are

the corresponding harmonic functions. Hence F is harmonic.


10
2.6 Proposition
If f (eint ) = eint and 0 ≤ r < 1, then its harmonic extension to D satisfies
F (reiθ ) = r|n| einθ ,
i.e., F (w) = wn for n ≥ 0 and w−n for n ≤ 0.

Proof: We have
Z 2π ∞
1 X
r|k| eik(θ−t) eint dt = r|n| einθ ,
2π 0 k=−∞

since we can integrate term-by-term as we have uniform convergence for

r < 1, and (
Z 2π
1 |k| ik(θ−t) int 0 if n 6= k
r e e dt = |n| inθ
2π 0 r e if n = k.


2.7 Corollary
If f ∈ L1 (T) and Z 2π
1
cn = f (t)e−int dt,
2π 0
for n ∈ Z, are its Fourier coefficients, then the harmonic extension of f is
given by
X ∞ X∞
F (w) = cn w n + c−n wn ,
n=0 n=1
converging uniformly on |w| ≤ r for 0 ≤ r < 1.

Proof: Now
∞
Z 2π X ∞
1 X |k| ikθ 2π
Z
1 |k| ik(θ−t)
r e f (t) dt = r e f (t)e−ikt dt,
2π 0 k=−∞ 2π k=−∞ 0

as required, since the partial sums converge in L1 and we can integrate term-
by-term.

∞
Note that if f ∈ L1 , then its Fourier series n=−∞ cn eint need not converge
P
anywhere to f (t), so letting |w| → 1 is tricky.

11
2.8 Definition
The Hardy space H p for 1 ≤ p ≤ ∞ is the subspace of Lp (T) consisting of
all functions f such that
Z 2π
1
f (t)e−int dt = 0
2π 0
for all n < 0 (i.e., hf (z), z n i = 0). Such functions have an analytic extension
to D and can be represented by
∞
X
F (z) = cn z n ,
n=0

with Z 2π
1
cn = f (t)e−int dt (n ≥ 0).
2π 0
Also Z 2π
iθ 1
F (re ) = Pr (θ − t)f (eit ) dt.
2π 0
Note that H p equals ∞ −1
T
n=1 ϕn {0}, where
Z 2π
1
ϕn (f ) = f (t)eint dt.
2π 0
Since
|ϕn (f )| ≤ kf kp keint kq = kf kp ,
1 1
where + = 1 (Hölder’s inequality), we see that each ϕn is continuous and
p q
so ϕ−1 p
n {0} is closed. Hence H is a closed subspace of L .
p

Normally we don’t distinguish between f and F .

We normally think of H p as the class of analytic functions in the disc with
Lp boundary values (i.e., they are Poisson integrals of Lp functions).

2.9 Corollary
H p is a Banach space.

Proof: This is immediate since it is a closed subspace of Lp (T), which is

complete.

12
 
We have defined H ∞ to be the space of analytic functions corresponding to
L∞ boundary values – we now want to show that in fact we can define it to
be just the space of bounded analytic functions in D.

2.10 Theorem
If f ∈ L∞ (T) with
Z 2π
1
f (t)e−int dt = 0 for n < 0, (1)
2π 0

then F (z) as given by Theorem 2.5 satisfies kF k∞ = kf k∞ . Conversely,

any bounded analytic function F in the disc is the harmonic extension of a
function f satisfying (1).

Proof: We have
Z 2π
1
F (z) = Pr (θ − t)f (t) dt,
2π 0

with z = reiθ , so
 Z 2π 
1
|F (z)| ≤ Pr (θ − t) dt kf k∞ ,
2π 0

and ∞
Z 2π Z 2π
1 1 X
Pr (θ − t) dt = r|n| ein(t−θ) dt = 1, (2)
2π 0 2π 0 n=−∞

as only n = 0 gives a non-zero value.

P∞ n
Conversely, if F (z) = n=0 an z is bounded in the disc, then, writing
iθ iθ
Fr (e ) = F (re ) we have that
∞
X
iθ
Fr (e ) = an rn einθ .
n=0

Also,
∞ Z 2π
X
n 21
|an r | = |Fr (eiθ )|2 dθ ≤ kF k2∞ .
n=0
2π 0

13
 P∞
Letting r → 1 we have n=0 |an |2 ≤ kF k2∞ , so that the function
∞
X
iθ
f (e ) = an einθ
n=0

exists in L2 (T) and

kFr − f k2 → 0
as r → 1. Now Fr → f in measure, and hence a subsequence (Frk ) converges
pointwise almost everywhere. Then

|f (eiθ )| ≤ lim sup |Frk (eiθ )| a.e.,

i.e., kf k∞ ≤ kF k∞ .

Since F (z) = ∞ n
P
n=0 an z and f has Fourier coefficients (an ) we see that F is
the harmonic extension of f and kF k∞ = kf k∞ .

Note that kFr k∞ % kf k∞ by the maximum modulus theorem. By working
harder one can prove Fatou’s theorem that Fr → f a.e. as r → 1 (i.e., not
just a subsequence).

14
LECTURE 3
So we now have H 2 as a space of power series
∞
X
f (z) = an z n
n=0

with ∞
X
2
kf k = |an |2 < ∞,
n=0
∞
and H as all power series with f bounded in D: alternatively, as analytic
functions that are the Poisson extensions of functions in L2 (T) (respectively,
L∞ (T)), their “boundary values”, whose only non-zero Fourier coefficients
are the (an ).

The other important case is p = 1, but we proceed indirectly.

2.11 Proposition
If f ∈ H 1 , then Z 2π
1
sup |f (reiθ )| dθ ≤ kf kH 1 .
0<r<1 2π 0

Proof: Recall that

Z 2π
iθ 1
f (re ) = Pr (θ − t)f (eit ) dt,
2π 0

so
2π Z 2π  Z 2π
Z 
1 iθ 1 1 it

|f (re )| dθ =  P r (θ − t)f (e ) dt dθ
2π 0 2π 0  2π 0 
Z 2π Z 2π
1 1
≤ |f (eit )| Pr (θ − t)dθ dt
2π 0 2π 0
= kf kH 1 ,

using (2).

We later prove that in fact we have equality above.

15
 2.12 Definition
A function of the form
n
Y
iϕ z − zj
B(z) = e ,
j=1
1 − zj z

where ϕ ∈ R and |zj | < 1 for j = 1, . . . , n, is called a finite Blaschke product.

2.13 Proposition
Let B be a finite Blaschke product as above. Then
(i) B is analytic in D and continuous in D.
(ii) B has zeroes at z1 , . . . , zn only, and poles at 1/z1 , . . . , 1/zn only.
(iii) |B(eiθ )| = 1 for θ ∈ R.

Proof: This is elementary. For the last part, note that

 e − zj   1 − e−iθ zj 
 iθ   
 1 − zj eiθ  =  1 − eiθ zj  = 1
   

since it is of the form |w/w|.



2.14 Definition
An H ∞ function that has unit modulus almost everywhere on T is called an
inner function.

Examples are finite Blaschke products and the ‘singular inner function’
 
z−1
exp .
z+1

If we have an H p function, we can use Blaschke products to factor out its

zeroes, i.e., we can write f (z) = B(z)g(z), where B is an infinite Blaschke
product and g(z) 6= 0 for z ∈ D. We now show this.

2.15 Lemma

16
If f ∈ H p , f 6≡ 0, then the zeroes (zn ) of f in D are at most countable in
number and satisfy X
(1 − |zn |) < ∞.
n

Proof: It is sufficient to prove this for H 1 since H p ⊂ H 1 for all p ≥ 1.

Without loss of generality f (0) 6= 0, as we can divide out a power of z, which

doesn’t change the H 1 norm.

Take r < 1 and let z1 , . . . , zm be the zeroes in {|z| < r}. Since they can’t
accumulate in D there are only finitely many and we will also assume that
there are no zeroes on |z| = r.

Write fr (z) = f (rz); this has zeroes at z1 /r, . . . , zm /r in {|z| ≤ 1}.

So m
Y z − zk /r
f (rz) = g(z),
k=1
1 − zk z/r

with g analytic and nonzero in D. Now log g is analytic, so by Cauchy’s

formula
I
1 dz
log g(0) = log g(z)
2πi |z|=1 z
Z 2π
1
= log g(eiθ ) dθ.
2π 0
Take real parts. Then
Z 2π
1
log |g(0)| = log |g(eiθ )| dθ
2π 0
Z 2π
1
= log |f (reiθ )| dθ.
2π 0
Now m
X
log |f (0)| = log |zk /r| + log |g(0)|,
k=1

17
so
m Z 2π
X 1
log |f (0)| + log |r/zk | = log |f (reiθ )| dθ
k=1
2π 0
Z 2π
1
≤ log |f (reiθ )| dθ
2π 0
≤ log kf kH 1 ,
by Proposition (2.11). Here we have used Jensen’s inequality in the form
that Z Z
log f ≤ log f,
Ω Ω
for a concave function f such as log.
Now let r → 1. We see that
∞
X
log |f (0)| + log |1/zk | ≤ kf kH 1 < ∞,
k=1

Now log(1 + x) ∼ x for x small, so log |1/zk | ∼ 1 − |zk |, and from this we can
deduce that ∞
X
(1 − |zk |) < ∞.
k=1


2.16 Theorem
Let f ∈ H p . Then the infinite (or finite) Blaschke product
Y z n zn − z
B(z) = z m ,
z 6=0
|zn| 1 − znz
n

where (zn ) are the zeroes of f , with m of them at 0, converges uniformly

on compact subsets of D to define an H ∞ function whose only zeroes are
at the (zn ), with the correct multiplicities. Moreover, |B(z)| ≤ 1 in D and
|B(eiθ )| = 1 a.e. for eiθ ∈ T.

Proof: It’s sufficient to consider f (z)/z m and ignore zeroes at 0. Write

zn zn − z
bn (z) = ,
|zn | 1 − z n z

18
noting that bn (0) = |zn | > 0. Also

|bn (eiθ )| = 1,

as in Proposition 2.13.
Q
To get bn (z)Pconverging to an analytic function with zeroes at the (zk ) only,
we show that |1 − bn (z)| converges uniformly on discs {|z| ≤ R} for R < 1.

Again this follows from the fact that log(1 + x) ∼ x when x is small.

Now
 
1 + z n z − zn 
 
|1 − bn (z)| =  |zn | 1 − zn z 
 
 |zn | − |zn |zn z + zn z − zn zn 
=  
 zn (1 − zn z) 
(1 − |zn |)(zn z + |zn |)
=
|zn ||1 − zn z|
 
1 + |z|
≤ (1 − |zn |) ,
1 − |z|

giving convergence.
So B(z) ∈ H ∞ and iθ
Qn|B(e )| ≤ 1 a.e.
Now write Bn = k=1 bk . Then B/Bn is another Blaschke product, and so
Z 2π 
 B(eiθ ) 
  
 B(0) 
 Bn (0)  ≤  Bn (eiθ )  dθ
   
0
Z 2π
1
= |B(eiθ )| dθ.
2π 0
Let n → ∞ and we see that
Z 2π
1
|B(eiθ )| dθ ≥ 1,
2π 0

and we conclude that |B(eiθ )| = 1 a.e.



19
Now for f ∈ H ∞ we have

kf k∞ = lim sup |f (z)|,

r→1 |z|=r

and for f ∈ H 2 we have

 Z 2π 1/2
1 iθ 2
kf k2 = lim |f (re )| dθ ,
r→1 2π 0

i.e.,
∞
!1/2
X
lim |an |2 r2n .
r→1
n=0

We shall now get the same result for H 1 , i.e., kf k1 = limr→1 kfr k1 , where
fr (z) = f (rz). For this we need the Poisson kernel again. Recall that

|z|2 − |w|2
P (z, w) =
|z − w|2

for 0 ≤ |w| < |z| ≤ 1.

2.17 Proposition
Let P (z, w) be the Poisson kernel. Then:
(i) If f is analytic in D, then we have
Z 2π
1
f (w) = P (reiθ , w)f (reiθ ) dθ
2π 0

for |w| < r < 1;

Z 2π
1
(ii) P (reiθ , w) dθ = 1 for |w| < r;
2π Z0
2π
1
(iii) P (z, seit ) dt = 1 for |z| > s.
2π 0

Proof: (i)
Z 2π
r2 − ww
I
iθ iθ 1
P (re , w)f (re ) dθ = f (z) dz,
0 2πi |z|=r z(z − w)(r2 /z − w)

20
where z = eiθ and dz = izdθ. This is
(r2 − ww)f (z) dz
I
1
= f (w),
2πi |z|=r (z − w)(r2 − zw)

since the only pole is at w with residue f (w).

(ii) Put f (z) = 1 in (i).

(iii)
2π
(zz − s2 ) dw
Z I
1 it 1
P (z, se )dt =
2π 0 2πi |w|=s w(z − w)(z − s2 /w)
(zz − s2 ) dw
I
1
=
2πi |w|=s (z − w)(zw − s2 )
(zz − s2 ) dw
I
1
= =1
2πi |w|=s (zz − wz)(w − s2 /z)

the only pole inside the contour being at w = s2 /z.



2.18 Corollary
If 0 < s < r < 1 and f is analytic in D, then
Z 2π Z 2π
1 it 1
|f (se )| dt ≤ |f (reiθ )| dθ.
2π 0 2π 0

Proof: From Proposition 2.17 we have

Z 2π Z 2π  Z 2π 
1 it 1 1 iθ it iθ

|f (se )| dt =  P (re , se )f (re ) dθ  dt
2π 0 2π t=0  2π θ=0 
Z 2π Z 2π
1 1
≤ P (reiθ , seit )|f (reiθ )| dθ dt
2π t=0 2π θ=0
Z 2π
1
= |f (reiθ )| dθ.
2π 0


21
2.19 Corollary
For f ∈ H 1 , the limit Z 2π
1
lim |f (reiθ )| dθ
r→1 2π 0

exists and equals kf kH 1 . Moreover kfr − f k1 → 0, where fr (eiθ ) = f (reiθ ).

Proof: By Corollary 2.18 the integral increases with r. Also, for any
h ∈ H 1 and r < 1 we have
Z 2π Z 2π  Z 2π 
1 iθ 1 1 it

|h(re )| dθ ≤  Pr (θ − t)h(e ) dt
2π 0 2π 0  2π 0
Z 2π
1
≤ |h(eit )| dt.
2π 0

We adopt the notation hr (z) = h(rz), interpreted for h ∈ L1 (T) using the
Poisson extension to D.
Now, given f ∈ H 1 , and ε > 0, we may choose a trigonometric polynomial g
with
N
X
g(eiθ ) = an einθ ,
n=−N

such that kf − gk1 < ε/3.

Now

kfr − f k1 ≤ kfr − gr k1 + kgr − gk1 + kg − f k1

≤ ε/3 + kgr − gk1 + ε/3 < ε

for r close to 1, since

N
X
iθ
(gr − g)(e ) = an (r|n| − 1)einθ ,
n=−N

which tends to zero uniformly as r → 1.

Thus kfr − f k1 → 0 and kfr k1 % kf k1 , as required.


22
 LECTURE 4
For f ∈ H 1 or H 2 , we have kfr − f kp → 0 (as we have seen), while for
f ∈ H ∞ we don’t always have kf − fr k∞ → 0, since the fr are continuous
on T and f needn’t be. However, kfr k∞ → kf k∞ .

2.20 Theorem (Riesz)

Let f ∈ H p with f 6≡ 0, and let B be the Blaschke product formed using the
zeroes (zn ) of f . Then f = gB, where kf kp = kgkp .

We prove this in the course just for p = 1, 2 and ∞.

Proof: Let g(z) = f (z)/B(z) and gn (z) = f (z)/Bn (z), where Bn is the
Blaschke product corresponding to the first n zeroes of f .
If r < 1, then
Z 2π Z 2π
1 iθ p 1 |f (Reiθ )|p
|gn (re )| dθ ≤ lim dθ
2π 0 R→1 2π 0 |Bn (Reiθ )|p
Z 2π
1
= lim |f (Reiθ )|p dθ,
R→1 2π 0

since |Bn (Reiθ )| → 1 uniformly as R → 1. So

Z 2π
1
|gn (reiθ )|p dθ ≤ kf kpH p .
2π 0

But |gn | % |g| as n → ∞, so by the Monotone Convergence Theorem we

have Z 2π
1
|g(reiθ )|p dθ ≤ kf kpH p .
2π 0
But |g(z)| ≥ |f (z)| for z ∈ D, so we have equality.
There is a similar (easier) argument for p = ∞.


2.21 Theorem (Riesz factorization theorem)

23
A function f lies in H 1 if and only if there exist g, h ∈ H 2 such that f = gh
(pointwise product). We can choose g and h such that kf k1 = kgk2 khk2 .

Proof: Note first that if g, h ∈ H 2 , then gh ∈ H 1 and

kghkH 1 ≤ kgkH 2 khkH 2 ,
by Cauchy–Schwarz.

Conversely, given f ∈ H 1 , write f (z) = f1 (z)B(z), where B is as in Theo-

rem 2.20, kf k = kf1 k and f1 is never 0 in D.

Since f1 is never 0 we can define an analytic square root, say f1 (z) = g(z)2
in D. For
√ we write f1 (z) = Reit , where t is chosen continuously, and then
g(z) = Reit/2 .
Now
f (z) = g(z)(g(z)B(z)),
and
kf kH 1 = kgk2H 2 = kgkH 2 kgBkH 2 ,
since kgBkH 2 = kgkH 2 .


2.22 Definition
The disc algebra A(D) is the space
Z 2π
1
{f ∈ C(T) : f (eit )e−int dt = 0 for all n < 0}.
2π 0
It is a closed subspace of C(T), hence a Banach space. All functions in A(D)
lie in H 2 so have analytic extensions to D given by
∞
X
f (eit ) = cn eint
n=0

(convergence in L2 ), and
∞
X
f (reit ) = cn rn eint
n=0

(convergence locally uniformly).

24
2.23 Theorem
The closure of the space of polynomials in Lp (T) is the space H p for 1 ≤ p <
∞ and A(D) for p = ∞.

Proof: Since the polynomials lie in A(D) ⊂ H 2 ⊂ H 1 , we see that in

each case the closure is contained in the relevant Banach space.

Given f ∈ H 1 , H 2 or A(D) we write fr (z) = f (rz) as usual and a calculation

using the Poisson kernel shows that kfr − f k → 0 in the appropriate norm
as r → 1.

But ∞
X
iθ
fr (e ) = cn rn einθ ,
n=0

where the (cn ) are bounded: this can in turn be approximated arbitrarily
closely in norm by a finite sum
N
X
cn rn einθ
n=0

since
∞
X X
n n
cn r z ≤ |cn |rn → 0



n=N +1 n=N +1

as N → ∞.


2.24 Example
Here is a function in H ∞ that does not lie in the subalgebra A(D).
Let  
z−1
f (z) = exp .
z+1
This is bounded in modulus by 1 in the unit disc, since the Möbius mapping
z−1
M : z 7→
z+1

25
takes D into the left half-plane

C− = {x + iy ∈ C : x < 0},

and then |f (z)| = ex < 1.

However, f is discontinuous on the circle, as M maps the circle to the imag-
inary axis, and f (eiθ ) = es , where s → ±i∞ as θ → π.

26
3 Operators on H 2 and L2(T)
We look at three important classes of operators: multiplication (Laurent)
operators, Toeplitz operators, and Hankel operators.

Notation: P : L2 (T) → H 2 denotes the orthogonal projection,

∞
X ∞
X
int
an e 7→ an eint .
n=−∞ n=0

Clearly kP f k2 ≤ kf k2 .

3.1 Proposition
Let ϕ ∈ L∞ (T). Then the transformation Mϕ : L2 (T) → L2 (T) defined by
Mϕ f (eit ) = ϕ(eit )f (eit )
is a bounded operator of norm kMϕ k = kϕk∞ .
Moreover,
sup{kMϕ f k2 : f ∈ H 2 , kf k2 = 1} = kϕk∞ .

Proof: Clearly kMϕ k ≤ kϕk∞ , since

Z 2π
2 1
kMϕ f k2 = |ϕ(eit )f (eit )|2 dt
2π 0
Z 2π
2 1
≤ kϕk∞ |f (eit )|2 dt = kϕk2∞ kf k22 .
2π 0
Conversely, given ε > 0 there is a set Aε ⊂ T of positive measure with
|ϕ(eit )| > kϕk∞ − ε on Aε .
We consider the function
(
1 on Aε ,
χAε =
0 on T \ Aε .

Now χAε ∈ L2 (T), with

Z 2π
2 1
kχAε k = χAε (eit )2 dt = µ(Aε ),
2π 0

27
where µ is the normalized Lebesgue measure on the circle. Also
Z 2π
2 1
kMϕ χAε k = |ϕ(eit )|2 χAε (eit )2 dt
2π 0
> (kϕk∞ − ε)2 µ(Aε ).

Hence
kMϕ χAε k/kχAε k > kϕk∞ − ε,
and so kMϕ k ≥ kϕk∞ − ε. Since this is true for all ε > 0 we conclude that

kMϕ k = kϕk∞ .

The last part is harder, but will be needed. Now χAε will not lie in H 2 , but
if its Fourier series is ∞
X
it
χAε (e ) = cn eint ,
n=−∞

then the functions fm defined by

∞
X ∞
X
it imt int
fm (e ) = e cn e = cn ei(n+m)t
n=−∞ n=−∞

also satisfy
kfm k2 = kχAε k2 = µ(Aε )1/2
and
kMϕ fm k = keimt Mϕ χAε k > (kϕk∞ − ε)µ(Aε )1/2 ,
since multiplying by an inner function such as eimt doesn’t change the L2
norm.
Now

X ∞ X∞
kP fm − fm k = cn ei(m+n)t − cn ei(m+n)t

n=−m n=−∞

−m−1
!1/2
X
= |cn |2 ,
n=−∞

which tends to 0 as m → ∞.

28
Hence
kP fm k → µ(Aε )1/2
and

kMϕ P fm k = kMϕ fm + Mϕ (P fm − fm )k
≥ kMϕ fm k − kMϕ (P fm − fm )k
> (kϕk∞ − )µ(Aε )1/2

for m sufficiently large. Thus

sup{kMϕ f k2 : f ∈ H 2 , kf k2 = 1} = kϕk∞ .

3.2 Corollary
If ϕ ∈ H ∞ then Mϕ : H 2 → H 2 satisfies kMϕ k = kϕk∞ .

Proof: This follows from the previous result, with the exception of show-
ing that ϕf ∈ H 2 if ϕ ∈ H ∞ and f ∈ H 2 .
We know it is in L2 and the analyticity in D is easily seen to follow by
multiplying the Taylor series together.

Matrix notation.

Write (en )∞ 2
n=−∞ for the orthonormal basis of H given by en (z) = z for
n
∞ 2
z ∈ T. Then (en )n=0 is an
P orthonormal basis of H .
If ϕ ∈ H ∞ with ϕ(z) = ∞ k=0 kd z k
, then calculating Mϕ en we get the infinite
matrix  
d0 0 0 0 . . .
d1 d0 0 0 . . . 
 
d2 d1 d0 0 . . .
 ,
d3 d2 d1 d0 . . .
 
.. .. .. .. . .
. . . . .
representing the operator Mϕ acting on H 2 . It is constant on the main
diagonals, i.e., a Toeplitz matrix. It is also lower triangular.

29
3.3 Definition
If ϕ ∈ L∞ (T), then the Toeplitz operator with symbol ϕ, Tϕ , is the operator
Tϕ : H 2 → H 2 defined by Tϕ f = P (Mϕ f ), for f ∈ H 2 .

Clearly, kTϕ k ≤ kϕk∞ , and we have equality if ϕ ∈ H ∞ . We’ll see that in

fact we have equality always.

Matrix of Tϕ .
P∞
Let ϕ(eit ) = k=−∞ dk eikt . Then
∞
X ∞
X
ikt int
Tϕ en = P dk e e = dp−n eipt ,
k=−∞ p=0

with p = n + k.
We then have the (full) Toeplitz matrix.
 
d0 d−1 d−2 d−3 ...
d1 d0 d−1 d−2 . . .
 
d2 d1 d0 d−1 . . .
 ,
d3 d2 d1 d0 . . .
 
.. .. .. .. ..
. . . . .
Special cases: if ϕ ≡ 1, then Tϕ is the identity. If ϕ(z) = z, then Tϕ is the
shift, z n 7→ z n+1 .

30
LECTURE 5

3.4 Theorem
For ϕ ∈ L∞ (T), kTϕ k = kϕk∞ , that is,
sup kTϕ f k = kϕk∞ .
f ∈H 2 ,kf k=1

Proof: Clearly,
kTϕ k = kP Mϕ k ≤ kMϕ k = kϕk∞ ,
so we need to show “≥”.
Given ε > 0 there is an f ∈ H 2 of norm 1 with kMϕ f k > kϕk∞ − ε, by
Proposition 3.1.
Without loss of generality, f is a polynomial, say, p(z) = N k
P
P∞ k=0 ck z , since if
f (z) = k=0 ak z k , then truncating it to fN (z) = N k
P
k=0 ak z we have fN → f
and kfN k → kf k = 1, so
fN fN
→f and Tϕ → Tϕ f.
kfN k kfN k
So take p = fN /kfN k for suitably large N .
Now we claim that as m → ∞,
kTϕ z m pk → kMϕ z m pk = kz m ϕpk > kϕk∞ − ε.
It is enough to show that kTϕ z m z k − Mϕ z m z k k → 0 for each k ≥ 0 and use
the triangle inequality to deduce that
kTϕ z m p − Mϕ z m pk → 0.
But with z = eit we have
∞

X
imt ikt imt ikt irt imt ikt
kTϕ e e − Mϕ e e k = (I − P ) dr e e e


r=−∞

−1−m−k
X
= dr ei(r+m+k)t

r=−∞
−1−m−k
!1/2
X
= |dr |2 → 0,
r=−∞

31
as m → ∞, and the result follows.

Convolution operators

Let (dk )∞
k=−∞ be a complex sequence, and consider the convolution operator
on `2 (Z+ ) defined by
k
X
(Td u)k = dj uk−j (k = 0, 1, 2, . . .).
j=−∞

Such operators are sometimes called discrete-time linear systems, with u be-
ing the input and y = Td u the output.

They are called causal if dj = 0 for j < 0, in which case yk only depends on
the “past”, i.e., u0 , . . . , uk .

If we consider the standard basis for `2 (Z+ ), namely e0 = (1, 0, 0, . . .), e1 =

(0, 1, 0, 0, . . .), e2 = (0, 0, 1, 0, 0, . . .), etc., then Td is represented by the
Toeplitz matrix  
d0 d−1 d−2 d−3 . . .
d1 d0 d−1 d−2 . . .
 
d2 d1 d0 d−1 . . .
 ,
d3 d2 d1 d0 . . .
 
.. .. .. .. . .
. . . . .
and so we know that, when Td is bounded, its norm given by

sup kTd uk2 = kGk∞ ,

kuk2 ≤1

where G is the L∞ (T) function with Fourier coefficients (dk ).

For Td to be causal, that is the same as saying that G is in H ∞ , and Td

is equivalent to an analytic Toeplitz operator (or Laurent operator), repre-
sented by a lower-triangular matrix.

32
If instead of working with `2 (Z+ ) we work with `∞ (Z+ ) (“bounded-input to
bounded-output”), then
k
X ∞
X
|(Td u)k | ≤ |dj ||uk−j | ≤ |dj |kuk∞ ,
j=−∞ j=−∞

so that, taking the sup over k, we have kTd k ≤ kdk`1 (Z) . By choosing suitable
u one can show that we get equality.

Thus Td is a bounded operator on `∞ (Z+ ) if and only if G lies in the Wiener

algebra, that is,
∞
X ∞
X
it ijt
G(e ) = dj e , with |dj | < ∞.
j=−∞ j=−∞

The causal operators correspond to the analytic Wiener algebra (i.e., dj = 0

for j < 0), and these are the ones with an analytic extension to D.

We now look at another class of operators in some detail.

3.5 Definition
We write (H 2 )⊥ for the orthogonal complement of H 2 in L2 (T), i.e., the
closed subspace spanned by the negative basis vectors en (z) = z n , for z ∈ T,
with n < 0.

If ϕ ∈ L∞ (T) then the Hankel operator Γϕ : H 2 → (H 2 )⊥ is defined by

Γϕ f = (I − P )Mϕ f , i.e., Mϕ = Tϕ + Γϕ . There are several equivalent
definitions, but we shall use this one.

3.6 Proposition
We have kΓϕ k ≤ kϕk∞ , and if ϕ(eit ) = ∞ int
P
n=−∞ dn e , then the matrix of Γϕ
with respect to the bases {e0 , e1 , e2 , . . .} of H 2 and {e−1 , e−2 , . . .} of (H 2 )⊥ is
 
d−1 d−2 d−3 . . .
d−2 d−3 d−4 . . .
 
d−3 d−4 d−5 . . .
 ,
d−4 d−5 d−6 . . .
 
.. .. .. ..
. . . .

33
which is constant on ‘minor’ diagonals. Also Γϕ = 0 if and only if ϕ ∈ H ∞
(whereas Tϕ = 0 if and only if ϕ = 0).

Proof:

kΓϕ f k2 = k(I − P )Mϕ f k2 ≤ kMϕ f k2 ≤ kϕk∞ kf k2 .

Also
∞
X
Γϕ ek = (I − P ) dn eint eikt
n=−∞
−1−k
X ∞
X
= dn ei(n+k)t = d−r−k e−irt ,
n=−∞ r=1

letting n + k = −r.
Clearly, Γϕ = 0 if and only if dn = 0 for each n < 0, which is the same as
saying that ϕ ∈ H ∞ .
Similarly, looking at the matrix, we see that Tϕ = 0 if and only if ϕ = 0 (and
we have already seen that kTϕ k = kϕk∞ ).


3.7 Theorem (Nehari)

If Γϕ : H 2 → (H 2 )⊥ is a bounded Hankel operator, then there exists a symbol
ψ ∈ L∞ (T) such that Γϕ = Γψ and kΓψ k = kψk∞ . Hence

kΓϕ k = inf{kϕ + F k : F ∈ H ∞ } = dist(ϕ, H ∞ ).

P∞
Proof: Observe that if ϕ(eit ) = k=−∞ dk eikt , then

hΓeint , e−imt e−it i = d−n−m−1 (n, m ≥ 0)

= hΓ(e e ), e−it i.
int imt

Hence
* N
! M
+ * N M
! +
X X X X
Γ bn eint , cm e−imt e−it = Γ bn eint cm eimt , e−it .
n=0 m=0 n=0 m=0

34
Thus we can define a linear functional on products of polynomials, by

α(f1 f2 ) = hΓ(f1 f2 ), e−it i = hΓf1 , f2 e−it i,

We see that
|α(f1 f2 )| ≤ kΓkkf1 k2 kf2 k2 .
But we have seen that products of polynomials are dense in H 1 , and so α
has a unique extension by

α(f1 f2 ) = hΓf1 , f2 e−it i, (f1 , f2 ∈ H 2 ),

with
|α(f1 f2 )| ≤ kΓkkf1 k2 kf2 k2 .
and hence, by Theorem 2.21 we have defined α as a linear functional on H 1
with
|α(g)| ≤ kΓkkgk1 (g ∈ H 1 ).
We may now use the Hahn–Banach theorem to extend α, so that we have
α : L1 (T) → C with the same norm.
This implies that we have a representation
Z 2π
1
α(g) = g(eit )h(eit ) dt
2π 0

for some h ∈ L∞ (T) with

khk∞ = kαk ≤ kΓk.

More calculations:

hΓeikt , e−it i = d−k−1 (k ≥ 0)

Z 2π
1
= eikt h(eit ) dt = hh, e−ikt i.
2π 0
Hence if ∞
X
it
h(e ) = h` ei`t ,
l=−∞

then we have h−k = d−k−1 .

35
We write ψ(eit ) = e−it h(eit ) and so ψ−k−1 = d−k−1 , and

kψk∞ = khk∞ ≤ kΓk,

as claimed.
Finally, since one has Γϕ = Γψ if and only if F := ψ − ϕ ∈ H ∞ , we have that

kΓϕ k ≤ inf{kϕ + F k∞ : F ∈ H ∞ },

and there exists a ψ such that ψ = ϕ + F and kψk∞ ≤ kΓϕ k ≤ kψk∞ .

Hence kψk∞ = kΓkψ , and we have the result.


3.8 Example
Hilbert’s Hankel matrix
 1 1

1 2 3
...
1 1 1
. . .
Γ =  21 3 4
 
1 1
3 4 5
. . .
.. .. .. . .
. . . .

has norm π as an operator on H 2 (take as symbol ψ(eiθ ) = i(θ − π) for

0 ≤ θ < 2π). We thus have Hilbert’s inequality

|hΓx, yi| ≤ πkxk kyk,

or   !1/2 !1/2
∞ X ∞ ∞ ∞
X xn y m  X X
≤π |xn |2 |ym |2 .

 n=0 m=0 n + m + 1 

n=0 m=0

36
LECTURE 6
To get a more explicit solution to the Nehari problem we need a technical
lemma on H 2 functions.

We write
log x = log+ x + log− x,
where log+ x = max(log x, 0) and log− x = min(0, log x).

3.9 Lemma
If f ∈ H 2 and f 6≡ 0, then
Z 2π
1
log− |f (eiθ )| dθ > −∞,
2π 0
and hence f 6= 0 a.e. on T.

The result can be extended to H 1 , since every H 1 function is the product of

two H 2 functions.

(Of course, arbitrary Lp functions can vanish on any set of positive measure
that we like.)

Proof: Without loss of generality f has no zeroes in D, as we can divide

out a Blaschke product, whose values on the circle have modulus 1 almost
everywhere.
As in Lemma 2.15 we have the inequality
Z 2π
1
log |f (0)| ≤ log |f (eiθ )| dθ
2π 0

Now log+ x ≤ x2 for x > 0, because for y ≥ 0 we have

2
y ≤ ey = 1 + y 2 + . . . ,

since 1 + y 2 − y = (1 + y/2)2 + 3y 2 /4 ≥ 0.

37
So Z 2π Z 2π
1 + iθ 1
log |f (e )| dθ ≤ |f (eiθ )|2 dθ = kf k22 .
2π 0 2π 0
Hence
Z 2π Z 2π Z 2π
1 − 1 1
iθ
log |f (e )| dθ = iθ
log |f (e )| dθ − log+ |f (eiθ )| dθ
2π 0 2π 0 2π 0
≥ log |f (0)| − kf k22 > −∞,

giving the result.



3.10 Corollary
If f ∈ A(D), and f 6≡ 0, then, regarding f as a function on D, its zero set
Z = {z ∈ D : f (z) = 0} satisfies:
P
(i) Z ∩ D is countable and zn ∈Z∩D (1 − |zn |) < ∞;
(ii) Z ∩ T has measure 0;
(iii) Z is closed.

Proof: This follows (via Lemmas 2.15 and 3.9) from the fact that
2
A(D) ⊂ H and the observation that f is continuous on D.

Solution of the Nehari problem

3.11 Theorem (Sarason)

If Γ : H 2 → (H 2 )⊥ is a bounded Hankel operator and if f ∈ H 2 with f 6≡ 0
satisfies kΓf k = kΓk kf k, then there is a unique symbol ψ for Γ = Γψ of
minimal norm, i.e., kψk∞ = kΓψ k, and it is given by ψ = Γff
, i.e.,

(Γf )(eiθ )
ψ(eiθ ) = a.e.
f (eiθ )

Moreover |ψ(eiθ )| is constant almost everywhere.

38
Proof: We have that Γψ f = (I − P )Mψ f , so that

kΓψ kkf k2 = kΓψ f k2 ≤ kψf k2 ≤ kψk∞ kf k2 .

But if we have equality throughout then Γf = ψf and so

Γf
ψ= ,
f
since f 6= 0 a.e. on T. Moreover, since kψf k2 = kψk∞ kf k2 we must have
|ψ(eiθ )| = kψk∞ a.e.


3.12 Example
Take a rank-1 Hankel operator with matrix
 
1 α α2 . . .
 α α 2 α 3 . . .
α 2 α 3 α 4 . . .  ,
 
 
.. .. .. . .
. . . .

where |α| < 1. One can check that

kΓk = 1/(1 − |α|2 ).

Let f (z) = 1/(1 − αz), so that

α2
 
1 1 α
(Γf )(z) = + + 3 + ... ,
1 − |α|2 z z2 z
and thus
(Γf )(z) 1 1 − αz
ψ(z) = = 2
.
f (z) 1 − |α| z − α
Note that (z−α)/(1−αz) is a Blaschke product, so ψ has modulus 1/(1−|α|2 )
a.e.

Best approximation problems (Nehari problems)

3.13 Corollary

39
If ϕ ∈ L∞ (T) is such that Γϕ attains its norm (i.e., kΓϕ f k = kΓϕ k kf k for
some f ∈ H 2 with f 6≡ 0), then ϕ has a unique best approximant h in H ∞ ,
i.e., a closest element, so that

kϕ − hk∞ = dist(ϕ, H ∞ )),

given by
Γf
h=ϕ− , and kϕ − hk∞ = kΓϕ k.
f
Moreover |(ϕ − h)(eiθ )| = kΓϕ k a.e.

Proof: if g ∈ H ∞ then Γϕ−g = Γϕ , and hence

kϕ − gk∞ ≥ kΓϕ−g k = kΓϕ k.

But there is a ψ ∈ L∞ (T), namely ψ = (Γf )/f , such that Γϕ = Γψ and

kψk∞ = kΓψ k = kΓϕ k.

Hence h = ϕ − ψ ∈ H ∞ , because Γh = Γϕ − Γψ = 0. Also,

kϕ − hk∞ = kψk∞ = kΓϕ k,

so h is a best approximant. Since ψ is unique, so is h. Since ψ has constant

modulus a.e., so has ϕ − h.

Note. If ϕ ∈ C(T), then Γϕ is compact and hence attains its norm (see
later); so there is always a best approximant in H ∞ , but it does not always
lie in A(D).

3.14 Example
Take
3 2
ϕ(z) = + 2,
z z
which lies in C(T). Note that the closest element in H 2 (in the L2 (T) norm)
is just the orthogonal projection onto H 2 , which is 0. To get the closest

40
element in H ∞ (in the L∞ (T) norm), form the rank-two Hankel matrix
 
3 2 0 ...
 2 0 0 . . .
 0 0 0 . . . .
 
 
.. .. .. . .
. . . .

This is symmetric, eigenvalues 4 and −1, and orthogonal eigenvectors

(2, 1, 0, 0, . . .)T and (1, −2, 0, 0, . . .)T .

Hence a maximizing vector is f (z) = 2 + z, mapped into

8 4
(Γf )(z) = + 2.
z z
Γf 3
Then h = ϕ − f
, and h(z) = , and h ∈ H ∞ .
2+z
Indeed kϕk∞ = 5, so 0 is not the best uniform approximant, whereas
4 1 + 2z
(ϕ − h)(z) = ,
z2 z + 2
which has constant modulus 4.

Two classical problems

1. The Carathéodory–Fejér extension problem.

Given a polynomial a0 +a1 z+a2 z 2 +. . .+an z n , choose coefficients an+1 , an+2 , . . .

to minimize
X ∞
k
ak z ;



k=0 ∞
i.e., minimize

a0 + a1 z + a2 z 2 + . . . + an z n + z n+1 h(z)
∞

over all h analytic and bounded in D.

41
2. The Nevanlinna–Pick interpolation problem.

Given z1 , . . . , zn ∈ D and w1 , . . . , wn ∈ C, find a function g analytic in D

such that g(zk ) = wk for k = 1, . . . , n, and kgk∞ is minimized.

3.15 Theorem
The Carathéodory–Fejér problem can be solved using the solution to the Ne-
hari problem.

Proof: Let
g(z) = a0 + a1 z + . . . + an z n .
We are trying to minimize
kg − z n+1 hk∞
over h ∈ H ∞ . This is equivalent to
g
min∞ n+1 − h ,

h∈H z ∞

We have seen that the solution is

Γϕ f
h=ϕ− ,
f

g(z)
where ϕ(z) = .
z n+1
It is easily seen that Γϕ has finite rank, so a maximizing vector f exists.

Example. Let g(z) = 2 + 3z. Then kgk∞ = 5.
The recipe tells us to look at
2 + 3z 2 3
ϕ(z) = 2
= 2+ ,
z z z
which is the ϕ we looked at earlier. Then
3
h(z) = .
2+z

42
So the minimal-norm extension is
3z 2 4 + 8z z + 1/2
f (z) = 2 + 3z − = =4 ,
2+z 2+z 1 + z/2

i.e., a constant times an inner function, with kf k∞ = 4.

3.16 Theorem
The Nevanlinna–Pick problem can be solved using the solution to the Nehari
problem.

Proof: There certainly do exist functions f ∈ H ∞ with f (zk ) = wk for

k = 1, . . . , n; for example, a suitable polynomial.

If g is another such function then f − g vanishes at z1 , . . . , zn , we we can

write
f (z) − g(z) = B(z)h(z),
where n
Y z − zk
B(z) =
k=1
1 − zk z
is the appropriate Blaschke product, and h ∈ H ∞ with khk∞ = kf − gk∞ ,
as in Proposition 2.13.

Moreover, any h ∈ H ∞ gives an interpolant g = f − Bh.

Our problem therefore is to find

inf{kf − Bhk∞ : h ∈ H ∞ }.

This is
inf{kB −1 f − hk∞ : h ∈ H ∞ },
since B is inner. And this is dist(B −1 f, H ∞ ), which is the Nehari problem
again.


43
Example. Find a minimal-norm interpolant g such that g(0) = 1 and
g( 21 ) = 0.

We start with an interpolant f (z) = 1 − 2z of H ∞ norm 3.

Now
z(z − 12 )
(f − Bh)(z) = (1 − 2z) − h(z),
1 − 12 z
so we minimize
−1
z − 2
kB f − hk =
− h
.
z ∞

Writing h = 1 + e
h, we have to find

z − 2 −2
inf − (1 + h)
e = inf − h
e .
h z h
z
∞ ∞
e

Now the best H ∞ approximant to ϕ(z) = −2/z turns out to be the 0 function
(just as in H 2 norm), as the corresponding Hankel operator is
 
−2 0 . . .
Γ= 0 0 . . .
... ... ...

which attains its norm at f = 1, and then

ϕ(z) − (Γf (z))/f (z) = −2/z − (−2/z) = 0.

Now we work out g = f − Bh for this choice (h = 1), and and then the best
interpolant is
z − 21
g(z) = −2 ,
1 − 12 z
of norm 2. Since kf k∞ = 3, we did better.

Finite-rank Toeplitz and Hankel operators.

3.17 Proposition
The only finite-rank Toeplitz operator is T0 = 0.

44
We shall later use a similar proof to show that there are no non-zero compact
Toeplitz operators.

Proof: Let (en )n≥0 be the usual orthonormal basis of H 2 .

Any finite rank operator T on a Banach space has the form
Xr
Tx = fk (x)xk ,
k=1

where x1 , . . . , xr are a basis for the range of T and f1 , . . . , fr ∈ X ∗ .

In our case fk (x) = hx, yk i for some yk ∈ H 2 . So

Xr
kT en k = hen , yk ixk


k=1
r
X
≤ |hen , yk i| kxk k → 0,
k=1

noting that |hen , yk i| is the absolute value of the nth Taylor coefficient of yk
and these tend to zero as n → ∞, since they are square-summable. Recall
that the matrix of T is
 
d0 d−1 d−2 d−3 . . .
d1 d0 d−1 d−2 . . .
 
d2 d1 d0 d−1 . . .
 .
d3 d2 d1 d0 . . . 
 
.. .. .. .. . .
. . . . .
We see that kT en k ≥ |dm | when n + m ≥ 0, since dm appears in the nth
column. Hence for T finite-rank, on letting n → ∞, we have that dm = 0 for
each m.


3.18 Theorem (Kronecker)

The Hankel operator
 
d−1 d−2 d−3 ...
d−2 d−3 d−4 . . .
 
Γ = d−3 d−4 d−5 . . .


d−4 d−5 d−6 . . .
 
.. .. .. ..
. . . .

45
has finite rank if and only if f (z) := −1 k
P
k=−∞ dk z is a rational function of z,
and its rank is the number of poles of f (which must lie in D).

Proof: If rank(Γ) = r, then the first r+1 columns are linearly dependent,
so that there are scalars λ1 , . . . , λm such that
r+1
X
λi d−i−m = 0 (m = 0, 1, 2, . . .)
i=1

This implies that

 
r d1 d2
(λ1 + λ2 z + · · · + λr+1 z ) + 2 + ...
z z

is a polynomial of degree at most r − 1 in z, since the coefficient of a negative

power z −m−1 is

λ1 d−m−1 + λ2 d−m−2 + · · · + λr+1 d−m−r−1 = 0.

Thus f , as above, is rational of degree at most r.

Conversely, if  
d1 d2
P (z) + 2 + ... = Q(z)
z z
with P , Q polynomial and deg P ≤ r, then working backwards we see that
the rank of Γ is at most r.

The poles of f are the roots of

λ1 + λ2 z + · · · + λr+1 z r = 0.

They must lie in D since otherwise we could split f into partial fractions
f = g + h, where deg g < deg f and h ∈ H ∞ , so Γf = Γg and rank(Γf ) =
deg g < deg f , a contradiction.


3.19 Example

46
Rank 1 Hankel matrices giving bounded operators are of the form
 
1 α α2 . . .
 α α 2 α 3 . . .
c
 α 2 α 3 α 4 . . . ,


... ... ... ...

c
where c ∈ C and α ∈ D. The corresponding symbol is .
z−α
Now we look at compactness of Toeplitz and Hankel operators.

3.20 Theorem
There are no compact Toeplitz operators except 0.

Proof: Recall that if (en ) is a sequence tending weakly to zero, and T is

compact, then (T en ) tends in norm to zero.

For if kT en k 6→ 0, then for some ε > 0 and a suitable subsequence we have

kT enk k ≥ ε and kT enk − wk → 0, so w 6= 0. But
hT enk , wi = henk , T ∗ wi → 0,
implying that hw, wi = 0, which is a contradiction.

We apply this with en (z) = z n , for n ≥ 0.

As in Proposition 3.17, recall the matrix of T is

 
d0 d−1 d−2 d−3 ...
d1 d0 d−1 d−2 . . .
 
d2 d1 d0 d−1 . . .
 .
d3 d2 d1 d0 . . .
 
.. .. .. .. ..
. . . . .

We see that kT en k ≥ |dm | when n + m ≥ 0, since dm appears in the nth

column. Hence for T compact, on letting n → ∞, we have that dm = 0 for
each m.


47
LECTURE 7

3.21 Theorem
The Hankel operator Γϕ is compact if and only if ϕ ∈ H ∞ + C(T), which is
a closed subalgebra of L∞ (T).

Proof: We prove just the easy part of this. It is enough to show that Γϕ
is compact for ϕ ∈ C(T), since Γϕ = 0 for ϕ ∈ H ∞ .

But if ϕ ∈ C(T), there is a sequence (ϕn ) of rational functions tending uni-

formly to ϕ on T.

Now
kΓϕ − Γϕn k ≤ kϕ − ϕn k∞ → 0.
Since each Γϕn has finite rank we deduce that Γϕ is compact.

The converse (due to Sarason) is harder, and we omit details. The basic idea
is that a compact Hankel operator is the limit of a sequence of finite-rank
Hankel operators, whose symbols have the form hn + rn where hn ∈ H ∞ and
rn is rational (by Kronecker’s theorem). Then it is a matter of choosing the
hn so that we get a uniformly convergent sequence of symbols, with limit in
H ∞ + C(T).

3.22 Corollary
−1
X
Suppose that the sequence (dn )−1 1
n=−∞ lies in ` , i.e., |dn | < ∞. Then
n=−∞
this Hankel matrix defines a compact operator:
 
d−1 d−2 d−3 . . .
d−2 d−3 d−4 . . .
 
d−3 d−4 d−5 . . .
 .
d−4 d−5 d−6 . . .
 
.. .. .. ..
. . . .

P−1
Proof: This follows since the function n=−∞ dn z n lies in C(T). 

48
Recall that an operator T : H → K, where H and K are Hilbert spaces,
∞
X
is said to be Hilbert–Schmidt if kT en k2 < ∞ for one (and hence all) or-
n=0
thonormal bases (en ) of H. It is well known that Hilbert–Schmidt operators
are always compact.

3.23 Proposition
The Hankel matrix  
d−1 d−2 d−3 ...
d−2 d−3 d−4 . . .
 
d−3 d−4 d−5 . . .
 
d−4 d−5 d−6 . . .
 
.. .. .. ..
. . . .
∞
X
defines a Hilbert–Schmidt operator if and only if n|d−n |2 < ∞.
n=1

Proof: Taking en (z) = z n for n ≥ 0, an orthonormal basis of H 2 , we see

that the square of the Hilbert–Schmidt norm is just the sum of the squares
of the absolute values of the entries of the matrix. 

The Hilbert transform

3.24 Definition
The Hilbert transform is the operator defined on L2 (T) by
Z ∞
it 1 f (eiθ )
(Hf )(e ) = − PV dθ,
π −∞ θ − t

R∞ R t−δ
where P V denotes the Cauchy principal value, i.e., limδ→0 t+δ
+ −∞
.

Strictly speaking,we define it on the orthonormal basis (en )∞

n=−∞ , where
it int
en (e ) = e , and extend by linearity and continuity.

49
So H maps the function en with en (eiθ ) = einθ to
Z ∞ inθ Z ∞ int inϕ
it 1 e 1 e e
(Hen )(e ) = − PV dθ = − PV dϕ,
π −∞ θ − t π −∞ ϕ

where θ = t + ϕ. Now
Z ∞ inϕ Z ∞
e cos nϕ + i sin nϕ
PV dϕ = PV dϕ = 0 + iπ
−∞ ϕ −∞ ϕ

for n ≥ 0, and we see easily that


−ien
 if n > 0,
Hen = 0 if n = 0,

ien if n < 0.


Then H is an operator of norm 1 with iH self-adjoint and

X  X
H4 an en = an e n .
n6=0

Note cos nθ 7→ sin nθ and sin nθ 7→ − cos nθ for n 6= 0.

Thus f + iHf is analytic in D, that is, if u is real harmonic in D, then Hu

is the harmonic conjugate v such that u + iv is analytic in D, and v(0) = 0;
this is unique.

Indeed,
∞
X ∞
X ∞
X ∞
X
an rn cos nθ + bn rn sin nθ 7→ an rn sin nθ − bn rn cos nθ.
0 1 0 1

The corresponding analytic function is

∞
X
a0 + (an − ibn )z n .
n=1

50
4 The shift operator
The shift operator on H 2 is defined by

(Sf )(z) = zf (z), (f ∈ H 2 ),

and is thus an analytic Toeplitz operator. When considered on the space

`2 (Z+ ) of coefficients, it acts equivalently by

(a0 , a1 , a2 , . . .) 7→ (0, a0 , a1 , a2 , . . .),

the unilateral shift.

The invariant subspaces for S are the closed subspaces M such that SM ⊆
M , i.e.,
f ∈ M =⇒ Sf ∈ M.
We exclude the trivial case M = {0} in what follows.

4.1 Theorem (Beurling)

Let M be a nontrivial (closed) invariant subspace for S. Then there is an
inner function Θ ∈ H ∞ such that

M = ΘH 2 = {Θf : f ∈ H 2 }.

Also, Θ is unique to within a constant of modulus 1.

Proof: It is clear that ΘH 2 is a closed invariant subspace (closed, since

the mapping f 7→ Θf is an isometry on H 2 ).

Also if Θ1 H 2 = Θ2 H 2 , then Θ1 = f Θ2 and Θ2 = gΘ1 for f, g ∈ H 2 . f and g

must themselves be inner, and so |f | ≤ 1 and |g| ≤ 1 on D. But f g = 1 so
|f | = 1 on D and by the maximum principle f and g are constant.

Note that SM 6= M , as otherwise every function in M would be divisible by

an arbitrarily large power of z and M = {0}.

So let Θ ∈ M SM , and without loss of generality kΘk2 = 1.

51
Now Z 2π
n 1
0 = hS Θ, Θi = |Θ(eit )|2 eint dt
2π 0

for each n ≥ 1. By conjugation it is true for n ≤ −1 as well so |Θ|2 is a

constant, which must be 1.

Thus Θ is inner and ΘH 2 ⊆ M . Also if ϕ ∈ M with ϕ ⊥ ΘH 2 , then

Z 2π
n 1
0 = hϕ, S Θi = ϕ(eit )Θ(eit )eint dt (3)
2π 0
for all n ≥ 0. But Θ ⊥ SM so
Z 2π
n 1
0 = hΘ, S ϕi = Θ(eit )eint ϕ(eit ) dt
2π 0

for all n ≥ 1, which is equivalent to (3) for all n < 0.

Thus all the Fourier coefficients of ϕΘ vanish, and ϕ = 0. So M = ΘH 2 .


4.2 Definition
For u ∈ H 2 we write [u] = span{u, Su, S 2 u, . . .} (the notation span means
closed linear span). This is the smallest (closed) invariant subspace contain-
ing u.

We say that u is cyclic, or outer, if [u] = H 2 .

Obviously non-zero constant functions are cyclic. This implies easily that
invertible functions in H 2 , such as u(z) = z + 2, are outer.

Note that outer functions u cannot vanish at any point of the disc, since all
functions in [u] would vanish at the same point.

4.3 Theorem
Every nontrivial function in H 2 has a factorization f = Θu, where Θ is inner
and u is outer. This is unique to within unimodular constants.

52
 Proof: Consider [f ] which is ΘH 2 for some inner function Θ, by Beurling’s
theorem. So f = Θu, with u ∈ H 2 . But then u is outer, as

Θ[u] = [Θu] = [f ] = ΘH 2 ,

and so [u] = H 2 . The uniqueness of Θ also leads easily to the uniqueness of

the factorization.

In fact, outer functions have a representation
 Z 2π it 
1 e +z it
u(z) = α exp k(e ) dt (z ∈ D),
2π 0 eit − z

where |α| = 1 and k ∈ L1 (T) is real. Moreover, k = log |u| on T. We do not

prove this here.

53
5 Hardy spaces on the half-plane
5.1 Proposition
The Möbius map M : z 7→ 1−z 1+z
is a self-inverse bijection from the disc D to
the right half-plane C+ = {x + iy : x > 0}. Let 1 ≤ p < ∞. Then a function
g defined on T is in Lp (T) if and only if the function G : iR → C defined by

G(s) = π −1/p (1 + s)−2/p g(M s)

is in Lp (iR), and moreover kgkp = kGkp .

We also have
g(z) = 22/p π 1/p (1 + z)−2/p G(M z).

Proof: The fact that |z| < 1 if and only if Re M z > 0 is easy, and so is
checking that M ◦ M = I.
Now
Z ∞
p
kGkp = |G(iy)|p dy
−∞
Z
1 1
= −i |g(M s)|p ds
iR π |1 + s|2
1 |1 + z|2 −2 dz
Z
= − |g(z)|p ,
|z|=1 π 4 (1 + z)2

1−z ds 2
making the change of variable s = with =− and 1 + s =
1+z dz (1 + z)2
2
.
1+z
Now
|1 + z|2 1+z 1
2
= =
(1 + z) 1+z z
on T and we find that kGkpp = kgkpp .

For p = ∞, we take G(s) = g(M s) and g(z) = G(M z). Then kgk∞ = kGk∞ .

54
5.2 Corollary
The functions En defined by

1 (1 − s)n
En (s) = √ , (n ∈ Z),
π (1 + s)n+1

form an orthonormal basis for L2 (iR).

Proof: Recall that (z n )n∈Z is an orthonormal basis of L2 (T). Now a

norm-preserving map also preserves inner products, so, using the calculation
in Proposition 5.1 for p = 2, we see that (En )n∈Z form an o.n.b. for L2 (iR).


55
LECTURE 8

5.3 Definition
The space H ∞ (C+ ) consists of all functions analytic and bounded in C+ ,
with norm kGk = sups∈C+ |G(s)|.

Looking at boundary values, we regard H ∞ (C
+ ) as the closed subspace of

L∞ (iR) consisting of all G with G(s) = g 1+s
1−s
, where g ∈ H ∞ (D).

Likewise H 2 (C+ ) consists of those functions analytic in C+ with L2 boundary

values, i.e., the closed subspace of L2 (iR) consisting of all G with
 
1 1 1−s
G(s) = √ g , for g ∈ H 2 (D).
π1+s 1+s
5.4 Corollary
The functions En defined by
1 (1 − s)n
En (s) = √ ,
π (1 + s)n+1
for n ≥ 0 form an orthonormal basis of H 2 (C+ ).

Note that En is analytic in C+ if and only if n ≥ 0.

Proof: Immediate, since (z n )n≥0 form an o.n.b. for H 2 (D).



5.5 Remark
One can define the norm on H p (C+ ) more directly via
Z ∞
p
kf kp = sup |f (x + iy)|p dy.
x>0 −∞

Note we do need to take all x > 0, not just a limit as x → 0; for consider
ez
f (z) = .
z+1

56
This is not in H ∞ (C+ ) or H 2 (C+ ), even though it is analytic in C+ with
boundary values in L∞ ∩ L2 (iR).

Laplace and Fourier transforms

5.6 Definition
For f ∈ L1 (R) ∩ L2 (R) define the 2-sided Laplace transform of f by
Z ∞
(Lf )(s) = e−st f (t) dt,
−∞

and the Fourier transform of f by

Z ∞
(Ff )(w) = e−iwt f (t) dt. (4)
−∞

Thus (Ff )(w) = (Lf )(iw). We are most interested in w ∈ R and s ∈ iR,
when, since f ∈ L1 (R), the integrals converge absolutely.

We regard L2 (R) as L2 (−∞, 0) ⊕ L2 (0, ∞), decomposing a function into its

values on t < 0 and t > 0.

5.7 Theorem (Paley–Wiener–Plancherel)

The Laplace transform (defined initially on (L1 ∩ L2 )(R), then extended by
continuity) determines a linear bijection between L2 (R) and L2 (iR) such that
√
kLf kL2 (iR) = 2πkf kL2 (R) .

Also, L2 (0, ∞) is mapped onto H 2 (C+ ).

There is an inverse map defined by

Z
−1 1
(L G)(t) = G(s)est ds,
2πi iR

for G ∈ (L1 ∩ L2 )(iR), extended by continuity to all of L2 (iR).

57
Thus,

L2 (R)= L2 (0, ∞) ⊕ L2 (−∞, 0)

L↓
L (iR) = H 2 (C+ ) ⊕ H 2 (C− ).
2

Compare the discrete version:

`2 (Z) = `2 (Z+ ) ⊕ `2 (Z− )

↓
L (T) = H 2 (D) ⊕ H 2 (D)⊥ ,
2

where Z+ is the non-negative, and Z− the strictly negative,

P integers. Here
the mapping takes a sequence (an ) to the function eiθ 7→ an einθ .

We shall sketch the proof of this shortly.

5.8 Corollary
If f ∈ L2 (R), then Ff ∈ L2 (R) (defined by (4) for f ∈ L1 ∩ L2 and extended
by continuity), and √
kFf k2 = 2πkf k2 .
Moreover, the inverse map is defined by
Z ∞
−1 1
(F F )(t) = F (w)eiwt dw,
2π −∞

again for F ∈ L1 ∩ L2 , then extended by continuity to L2 (R).

Notes.

1. If f (t) and f 0 (t) are both O((1 + |t|)−2 ) and f is C 2 , then f and Ff
are in L1 (R), so there are no problems about convergence of integrals. Such
functions form a dense subspace of L2 (R).
ˆ
2. We write fˆ(w) = (Ff )(w) and F̌ (t) = (F −1 F )(t), and note that fˆ(t) =
2πf (−t), and thus F 4 = 4π 2 I.

58
We give a sketch of the proof of Theorem 5.7.

Proof: Consider the dense subspace of L2 (−∞, ∞) consisting of all func-

tions
N 2
X
f (t) = aj χ[jM/N,(j+1)M/N ] (t),
j=−N 2

where M > 0 and N > 1 are integers, and the aj are complex scalars.
Then
N2 −M s/N
−jM s/N 1 − e
X
(Lf )(s) = aj e ,
2
s
j=−N

which lies in L2 (iR), and direct calculation shows that its norm is
 1/2
r N 2
√ M X
kLf kL2 (iR) = 2π |aj |2  .
N
j=−N 2

Now there is a unique continuous extension from this subspace to the whole
of L2 (given by an integral for f ∈ L1 ∩ L2 ).

1 (1 − s)n
Note that L carries L2 (0, ∞) onto H 2 (C+ ), since H 2 (C+ ) has o.n.b. √
π (1 + s)n+1
for n ≥ 0 and
n!
L(tn e−t χ(0,∞) ) = ,
(1 + s)n+1
so the orthonormal basis functions in H 2 (C+ ) are linear combinations of
these.

Similarly the functions with n < 0 are the Laplace transforms of functions
supported on (−∞, 0).

An orthonormal basis for L2 (0, ∞)

A rational orthonormal basis for H 2 (C+ ) was given by Corollary 5.4. The
functions transform under L−1 to functions√ t 7→ pn (t)e
−t
in L2 (0, ∞), where
pn is a real polynomial of degree n; then 2πpn (t)e−t form an orthonormal

59
basis for L2 (0, ∞).

In fact √
pn (t) = ±Ln (2t)/ π,
where Ln denotes the Laguerre polynomial

et dn n −t
Ln (t) = (t e ),
n! dtn
and these satisfy Z ∞
Ln (t)Lm (t)e−t dt = δmn
0
√
(Kronecker delta). Similarly 2πpn (−t)et form an orthonormal basis for
L2 (−∞, 0).

So, by a series of transformations we have linked an o.n.b. of

`2 (Z+ ), namely (0, . . . , 0, 1, 0, . . .), to an o.n.b. of

H 2 , namely z n , to an o.n.b. of

1 (1 − s)n
H 2 (C+ ), namely √ , to an o.n.b. of
π (1 + s)n+1
√
L2 (0, ∞), namely ± 2Ln (2t)e−t .

60
6 Reproducing kernel Hilbert spaces
6.1 Definition
A reproducing kernel Hilbert space (RKHS) H is a Hilbert space of functions
defined on a non-empty set S such that for all w ∈ S the evaluation mapping
f 7→ f (w) is bounded, and thus there is a function kw (the reproducing
kernel ) such that
f (w) = hf, kw i (f ∈ H).
Note that if `w is another such reproducing kernel, then

`w (s) = hlw , ks i = ks (w)

for all s, w ∈ S, and hence kw is unique and

kw (s) = ks (w).

Another useful formula is that

kkw k2 = hkw , kw i = kw (w).

6.2 Theorem (Aronszajn)

In a RKHS it holds that for all α1 , . . . , αn ∈ C and w1 , . . . , wn ∈ S we have
n X
X n
αi kwj (wi )αj ≥ 0. (5)
i=1 j=1

Moreover, if (w, z) 7→ kw (z) is a function on S × S satisfying (5), then there

is a unique RKHS with kw as the reproducing kernel, namely the completion
of the space of all finite linear combinations of kw for w ∈ S, with norm
2
X n n X
X n
α k = αi kwj (wi )αj .

j wj

j=1 i=1 j=1

Proof: Evaluating hf, f i for

n
X
f= αj kwj
j=1

61
shows that (5) holds.

Conversely, if (5) holds, then (kwj (wi ))ni,j=1 is a Hermitian matrix and the
formula * n +
X n
X Xn X n
αj kwj , βi kwi = β i kwj (wi )αj
j=1 i=1 i=1 j=1

satisfies the axioms of an inner product on the space of finite linear com-
binations of the kwj . It is now possible to take its completion in the usual
way.

Note that reproducing kernels do not have to be linearly independent (for
example, if every function in the space takes the same value at two distinct
points).

Examples

0. The Hilbert space Cn can be considered as a function space on {1, 2, . . . , n},

with reproducing kernel km (j) = δmj , the Kronecker delta.

Clearly,
h(a1 , . . . , an ), km i = am .

1. The Hardy space H 2 considered as a function space on D.

Note that if f ∈ H 2 with

∞
X
f (z) = an z n ,
n=0

then ∞
X
f (w) = an wn = hf, kw i,
n=0

where ∞
X 1
kw (z) = wn z n = ,
n=0
1 − wz
the Cauchy (or Szegö) kernel.

62
2. The Bergman space A2 of functionals analytic in D with finite Bergman
norm given by
 Z 1/2  Z 1 Z 2π 1/2
1 2 1 iθ 2
kf k = |f (z)| dA(z) = |f (re )| r dr dθ .
π D π r=0 θ=0
It can be checked that the norm of a function f represented by a power series
∞
X
f (z) = an z n
n=0

is given by
∞
!1/2
X |an |2
kf k = ,
n=0
n+1
√
and the functions n + 1 z n for n ≥ 0 form an orthonormal basis. Hence
∞
X
f (w) = an wn = hf, kw i,
n=0

where ∞
X 1
kw (z) = (n + 1)wn z n = ,
n=0
(1 − wz)2
the Bergman kernel.

An analogous space can be defined on an arbitrary open subset U of C, the

R
1/2
norm being U |f (z)|2 dA(z) .

6.3 Proposition
Suppose that H is a reproducing kernel Hilbert space on a set S, with or-
thonormal basis (en ). Then the reproducing kernel is given by
X
kw = en (w)en (w ∈ S).

Proof:
P Clearly there exist constants (an ) depending on w such that
k w = an e n .

63
Then
an = hkw , en i = hen , kw i = en (w),
as claimed.


64
LECTURE 9
More examples.

3. More generally, if H is a Hilbert space of power series in the disc such

that ∞ 2
X X∞
n
an z = γn |an |2 ,



n=0 H n=0

i.e., (z ) is an orthogonal basis with norm kz n k2 = γn , then kw if it exists

n

must be given by
X∞
kw (z) = wn z n /γn ,
n=0
P∞
and H is a RKHS if and only if kw ∈ H, i.e., n=0 |w|2n /γn < ∞, for all
w ∈ D.

For the Hardy space γn = 1 for each n, and for the Bergman space γn =
1/(n + 1).

4. The Hardy–Sobolev space of functions f such that f 0 ∈ H 2 . It is usual to

take
kf k2 = kf k2H 2 + kf 0 k2H 2 ,
which means that γn = 1 + n2 .

5. The Dirichlet space of functions f such that f 0 ∈ A2 , where we can take

γ0 = 1 and γn = n for n > 0.

The reproducing kernel for the Dirichlet space is given by

∞
X
kw (z) = 1 + wn z n /n = 1 − log(1 − wz).
n=1

6.4 Proposition
Let H be a reproducing kernel Hilbert space of analytic functions in the
disc, and let G ∈ H ∞ (D) such that MG is a bounded operator on H. Then
MG∗ kw = G(w)kw , and hence the spectrum of MG contains the closure of the
set
{G(w) : w ∈ D}.

65
In particular, kMG k ≥ kGk∞ .

Proof: For f ∈ H, we have

hf, MG∗ kw i = hMG f, kw i = G(w)f (w) = hf, G(w)kw i,

and so kw is an eigenvector of MG∗ with eigenvalue G(w).

The rest follows from standard results about the spectrum: in particular,
that the norm of MG is greater than or equal to the spectral radius of MG .

This is a simple way of proving that the shift S on H 2 has spectrum equal
to the closed unit disc, although it has no eigenvalues.

Here’s another class of operators that can be studied using reproducing ker-
nels.

6.5 Definition
Let ϕ : D → D be analytic. Then the composition operator Cϕ is defined on
H 2 by
Cϕ f (z) = (f ◦ ϕ)(z) = f (ϕ(z)), (f ∈ H 2 ).

6.6 Theorem (Littlewood’s subordination theorem)

The operator Cϕ maps H 2 boundedly into H 2 .

Proof: Suppose first that ϕ(0) = 0: we claim that in this case Cϕ is a

contraction.

Let f be a polynomial. We prove that kCϕ f k ≤ kf k by induction on the

degree of f . Clearly it is true for deg f = 0. In general,

f (z) = f (0) + zS ∗ f,

where S ∗ is the backward shift, T1/z .

66
Thus
Cϕ f = f (0) + ϕCϕ S ∗ f
and
kCϕ f k2 = |f (0)|2 + kϕCϕ S ∗ f k2 ≤ |f (0)|2 + kCϕ S ∗ f k2 ,
and by the induction hypothesis kCϕ S ∗ f k ≤ kS ∗ f k. So

kCϕ f k2 ≤ |f (0)|2 + kS ∗ f k2 = kf k2 .

For general f ∈ H 2 we have polynomials fn tending to f in H 2 (and locally

uniformly in D), so Cϕ fn → Cϕ f locally uniformly. Then for each 0 < r < 1
we have
Z 2π Z 2π
1 it 2 1
|Cϕ f (re )| dt = lim |Cϕ fn (reit )|2 dt
2π 0 n→∞ 2π 0

≤ lim sup kCϕ fn k2

≤ lim sup kfn k2 = kf k2 .

Then letting r → 1 we see that Cϕ is a contraction when ϕ(0) = 0.

For general ϕ we consider the Möbius map ψa defined by

a−z
ψa (z) = .
1 − az
Note that ψa ◦ ψa is the identity map.

It is easy to see by a change of variable on T (since the derivative of ψa is

bounded) that Cψa acts as a bounded operator on L2 (T); also, it preserves
the subspace H 2 . Indeed
 1/2
1 + |a|
kCψa k ≤ .
1 − |a|

Choosing a = ϕ(0) we have (ψa ◦ ϕ)(0) = 0. Also

Cψa ◦ϕ f (z) = f (ψa (ϕ(z))),

so
Cψa ◦ϕ = Cϕ Cψa = Cϕ Cψ−1
a
.

67
Thus  1/2
1 + |ϕ(0)|
kCϕ k ≤ kCψa ◦ϕ kkCψa k ≤ kCψa k ≤ .
1 − |ϕ(0)|


6.7 Theorem
Let ϕ : D → D be analytic, and let kw (for w ∈ D) denote the reproducing
kernels for H 2 . Then Cϕ∗ kw = kϕ(w) . Hence
1/2
1 − |w|2

kCϕ k ≥ sup .
w∈D 1 − |ϕ(w)|2

Proof: Clearly

hf, Cϕ∗ kw i = hCϕ f, kw i = f (ϕ(w)) = hf, kϕ(w) i

for f ∈ H 2 .

Also
kCϕ∗ kw k
kCϕ k = kCϕ∗ k ≥ sup ,
w∈D kkw k

which gives the estimate, recalling that

1
kkw k2 = hkw , kw i = kw (w) = .
1 − |w|2

In general the norms of many operators (e.g. Toeplitz, Hankel, composition,
. . . ) can be estimated very well by “testing” them on reproducing kernels.
This is sometimes called the reproducing kernel thesis.

For example a Hankel matrix Γ is bounded if and only if

sup kΓkw k/kkw k < ∞

(Bonsall’s theorem, not proved here).

68
6.8 Remark
In a RKHS H the subspace Zw = {f ∈ H : f (w) = 0} is just the orthogonal
complement of the span of kw , and moreover f (wj ) = g(wj ) for j = 1, . . . , n
if and only if (f − g) ⊥ kw1 , . . . , kwn .

For example, in H 2 , we have seen that f and g take the same values at
w1 , . . . , wn if and only if the Blaschke product B with zeroes w1 , . . . , wn di-
vides f − g.

Another way of expressing this is to say that the orthogonal complement of

BH 2 is span{kw1 , . . . , kwn }.

A similar result holds for infinite sequences (wk )∞

k=1 if they satisfy the Blaschke
condition ∞
X
(1 − |zk |) < ∞,
k=1

so that there is a Blaschke product with the (wk ) as its zeroes. See Theo-
rem 2.16.

6.9 Definition
Let K ⊂ R be compact and nonempty. Then the Paley–Wiener space
P W (K) consists of all f ∈ L2 (R) such that fˆ is supported on K.

So fˆ ∈ L2 (K), and so by Cauchy–Schwarz

Z Z 1/2 Z 1/2
|fˆ| ≤ |fˆ|2 2
1 < ∞.
K K K

Thus Z
1
f (t) = fˆ(w)eiwt dw,
2π K
the integral converging pointwise absolutely.

Functions in P W (K) are very well behaved.

69
6.10 Lemma (Riemann–Lebesgue)
If fˆ ∈ L1 (R), then f lies in C0 (R), i.e., it is continuous and tends to zero at
±∞.

Proof: If
N 2 −1
X
fˆ = aj χ[jM/N,(j+1)M/N ] ,
j=−N 2

a step function, then

N −1 2
1 X eM it/N − 1
f (t) = aj ejM it/N ,
2π 2
it
j=−N

which lies in C0 (R). For general g ∈ L1 (R) there is a sequence (gn ) of step
functions converging to g in L1 norm.

Now
1
kǧn − ǧk∞ ≤ kgn − gkL1 → 0,
2π
so ǧ ∈ C0 (R).


6.11 Theorem
P W (K) is a RKHS of continuous functions on R, with
Z
−1 −iwt 1
kt (s) = F (χK (w)e )(s) = eiw(s−t) dw.
2π K

Proof:
Z
1
f (t) = fˆ(w)eiwt dw
2π K
1 ˆ −iwt
= hf , e χK (w)iL2 (R)
2π
= hf, kt iL2 (R) ,

where k̂t (w) = e−iwt χK (w), so kt ∈ P W (K).

70
We used the identity
1 ˆ
hf, gi = hf , ĝi
2π
from Corollary 5.8.

In the special case K = [−b, b] for b > 0, we write P W (b), the space of
“band-limited signals”.

The reproducing kernel is

Z b
1
ks (t) = eiw(s−t) dw
2π −b
1
eib(s−t) − e−ib(s−t)


=
2πi(s − t)
1 sin b(s − t)
= (s 6= t),
π s−t
with ks (s) = b/π.

Note that ks ⊥ kt if ks (t) = 0, i.e., if (s − t) is a nonzero multiple of π/b.

6.12 Proposition
If f ∈ P W (b), then f extends to an entire function (i.e., analytic on the
whole of C) by
Z b
1
f (z) = fˆ(w)eiwz dw,
2π −b
and
1 ˆ
|f (z)| ≤ kf kL1 eb|z| ,
2π
that is, f is of exponential type.

Proof: With Z b
1
f (z) = fˆ(w)eiwz dw,
2π −b
we have Z b
1
f (n)
(z) = fˆ(w)(iw)n eiwz dw,
2π −b

71
since this integral converges absolutely; thus f is an entire function. In
particular we have
1 n ˆ
|f (n) (0)| ≤ b kf k1 .
2π
and ∞
X z n f (n) (0)
f (z) = ,
n=0
n!
so ∞
X |z|n 1 n ˆ 1 ˆ b|z|
|f (z)| ≤ b kf k1 = kf k1 e ,
n=0
n! 2π 2π
as required.

In fact there is a converse, which we do not prove.

6.13 Theorem
Let b > 0. If f is an entire function such that f|R ∈ L2 (R) and there exists
C > 0 such that |f (z)| ≤ Ceb|z| for all z ∈ C, then f|R ∈ P W (b).

72
LECTURE 10
We now look at an important orthonormal basis for P W (b).

6.14 Lemma
If F ∈ L2 [−b, b], then
∞
π X
F (w) = F̌ (nπ/b)e−inπw/b ,
b n=−∞

converging in L2 norm.

Proof: It is well known that the functions (en )n∈Z , defined by

1
en (w) = √ e−inπw/b
2b
form an orthonormal basis for L2 [−b, b]. Thus
∞
X
F = hF, en ien
n=−∞

which gives
∞ Z b
1 X
F (w) = F (y)einπy/b dye−inπw/b
2b n=−∞ −b
∞
1 X
= 2π F̌ (nπ/b)e−inπw/b .
2b n=−∞

6.15 Corollary (Whittaker–Kotel’nikov–Shannon sampling theorem)

If f ∈ P W (b), then
∞
π X
f= f (nπ/b)knπ/b ,
b n=−∞

73
where k denotes the reproducing kernel. The series converges in L2 (R) norm
and uniformly (i.e., in L∞ (R) norm). That is, f can be reconstructed from
samples spaced at intervals π/b.

Proof: By Lemma 6.14, with F = fˆ, we have

∞
π X
f (t) = f (nπ/b)F −1 (e−inπw/b χ[−b,b] )(t),
b n=−∞

converging in P W (b) and hence in L2 and uniformly. But

ks = F −1 (e−iws χ[−b,b] ),

which gives the result.


Notes. Note that an orthonormal basis for P W (b) is the set
r 
π
knπ/b : n ∈ Z .
b

Since P W (b) ⊂ P W (c) for 0 < b < c, we can also derive a formula for f in
terms of f (nπ/c), n ∈ Z. This is called “oversampling”.

THE END

74