Professional Documents
Culture Documents
Hardy Spaces in One Complex Variable
Hardy Spaces in One Complex Variable
Hardy Spaces in One Complex Variable
A.A. 2004-2005
1
2
Contents
Chapter III Pointwise convergence to the boundary and conjugate harmonic func-
tions in hp
1. The Hardy-Littlewood maximal function
2. Poisson integrals and maximal function
3. Pointwise convergence to the boundary
4. Poisson integrals of singular measures
5. Lp -estimates for the conjugate harmonic function
THE UNIT DISC 3
CHAPTER I
HARDY SPACES ON THE UNIT DISC
We begin by presenting the main properties of Hardy spaces on the unit disc
D = {z ∈ C : |z| < 1} .
The reason is that emphasis will be put on the group structure of T. The natural
identification between T and R/2πZ (both algebraic and topological) will be always
assumed. Hence functions defined on T will be identified with functions on R/2πZ,
i.e. with functions on the line, periodic of period 2π.
Integrals on T will be understood with respect to the normalized Lebesgue mea-
1
sure 2π dt. We shall use as alternative notation for an integral on T any of the
following1 :
Z Z π Z Z π
it 1 it 1
f (e ) dt , f (e ) dt , f (t) dt , f (t) dt
T 2π −π T 2π −π
The spaces Lp (T) must be intended w.r. to the normalized Lebesgue measure.
The 2-dimensional Lebesgue measure on C will be denoted by dz. Hence, in the
polar coordinates z = reit ,
dz = r dr dt .
This may cause some confusion in the occasions where we shall use line integrals
in C, Z I
f (z) dz or f (z) dz ,
γ γ
in which case the symbol dz denotes a linear differential form. However, the meaning
of the symbol dz will be revealed in each case by the domain of integration.
1 Tobe precise, in the first two integrals T is identified with the unit circle, in the last two with
R/2πZ. We may switch from one notation to the other, omitting explicit reference to composition
with the map t 7→ eit .
Typeset by AMS-TEX
4 CHAPTER I
and also
If we set
fr (eit ) = f (reit )
for 0 ≤ r < 1, we can then say that
Mp (f, r) = kfr kp .
kf kH p ≤ kf kH q .
Proof. The inequality kfr kp ≤ kfr kq follows easily from Hölder’s inequality if q <
∞, and from the trivial majorization if q = ∞. Taking suprema in r, the inequality
is preserved.
All these inclusions are proper. Interesting examples in this respect are the
functions
1
fα (z) = ,
(1 − z)α
with3 α > 0. Given p, we want to determine the values of α for which fα is in
H p (D). It is obvious that fα 6∈ H ∞ (D) for α > 0, so we take p < ∞.
The answer is based on the following lemma4 .
2 We define Hardy spaces only for p ≥ 1. The definition makes perfect good sense also for
p < 1, except that in this case (1.4) does not define a norm. H p -spaces with p < 1 have many
intereseting features, that we will not discuss in this course.
3 It is possible to choose a determination of the α-power of 1 − z on D for every α ∈ C,
bacause 1 − z does not vanish on D and D is simply connected. The “principal” determination is
(1 − z)α = eα log(1−z) , with arg(1 − z) ∈ (−π/2, π/2) (observe that <e(1 − z) > 0 for z ∈ D).
4 We write f g for r → 1 to denote that the ratio |f /g| is bounded from above and from
Proof. Take r > 12 . Considering that the triangle with vertices 1, r and reit is obtuse
in r, and that | sin θ| ≥ π2 |θ| for θ ∈ [−π/2, π/2], we have that, for t ∈ [−π, π],
|1 − reit | > max r|1 − eit |, 1 − r
11
≥ |1 − eit | + 1 − r
2 2
1 t
(1.5) = sin + 1 − r
2 2
1 1
≥ |t| + 1 − r
2 π
1
≥ (|t| + 1 − r) .
2π
We also have
so that Z π
1
Is (r) dt ,
−π (|t| + 1 − r)s
for every s > 0. Then,
Z π Z π
1 1
s
dt = 2 s
dt
−π (|t| + 1 − r) 0 (t + 1 − r)
2 log(1 − r + π) − log(1 − r) if s = 1 ,
= 2
1−s
(1 − r + π)1−s − (1 − r)1−s if s 6= 1 .
lim Mp (fα , r) = +∞
r→1
if αp ≥ 1.
We show now that H p (D) is complete. A couple of preliminary facts have their
own independent importance.
6 CHAPTER I
By Hölder’s inequality,
Z π 1/p0
1 1
|f (z)| ≤ Mp (f, r)
|z| i(θ−t) p0
dt ,
2π 1 −
r e
−π
≤ Cp kf kH p 1 −
r
|z| −1/p
= Cp kf kH p 1 − .
r
Letting r → 1, we conclude the proof.
Corollary 1.5. Convergence in H p (D) implies uniform convergence on compact
subsets of D.
Proof. Let K ⊂ D be compact. Then there is r < 1 such that K ⊂ Dr , the closed
disc of radius r centered at the origin. By Lemma 1.4, if fn → f in H p (D),
kfn − f kH p
kfn − f k∞,K = max |fn (z) − f (z)| ≤ Cp ,
z∈K (1 − r)1/p
also tends to zero.
THE UNIT DISC 7
Since ∆ is real, if u is harmonic, so are ū, <eu and =mu. In particular, anti-
holomorphic functions are also harmonic.
Harmonic functions are characterized by the mean value property. Let S n−1 be
the unit sphere in Rn , and dσ the surface measure on it.
8 CHAPTER I
Definition. A continuous function u satisfies the mean value property on Ω if, for
any x ∈ Ω and any r > 0 such that the closed ball B(x, r) is contained in Ω,
Z
1
(2.1) u(x) = u(x + ry) dσ(y) .
σ(S n−1 ) S n−1
defined for r ∈ [0, r0 ), where r0 is the radius of the largest ball centered at x and
contained in Ω. It is easy to verify that ϕ is continuous, ϕ(0) = u(x) and that
Z
0 1 ∂
ϕ (r) = u(x + ry) dσ(y) .
σ(S n−1 ) S n−1 ∂r
For any r, the last integral can be written as an integral on the boundary of the
ball B(x, r), in terms of the surface measure dσr on it:
Z Z
∂ 1
u(x + ry) dσ(y) = n−1 ν · ∇u(y) dσr (y) ,
S n−1 ∂r r ∂B(x,r)
which is well-defined for x ∈ B(x0 , r/2). Since ψ is radial, we can consider its
“profile” ψ0 , defined on the positive half-line so that ψ(x) = ψ0 (|x|). We then
have, integrating in polar coordinates:
Z r
2
Z
v(x) = u(x + ρw)ψ0 (ρ)ρn−1 dσ(w) dρ
0 S n−1
Z r Z
2
n−1
= ψ0 (ρ)ρ u(x + ρw) dσ(w) dρ
0 S n−1
Z r
2
n−1 n−1
= σ(S ) ψ0 (ρ)ρ dρ u(x) .
0
(the last equality takes into account the fact that ψ = 0 in the extra part of
the domain of integration). This last expression shows, by differentiating under
integral sign, that v is C 2 on B(x0 , r/2). The same is then true for u, and, since
x0 is arbitrary, u is C 2 on Ω.
Corollary 2.2. Harmonic functions satisfy the maximum modulus principle: if u
is harmonic on a connected open set Ω, and it attains its maximum modulus at a
point in Ω, then it is constant.
Proof. Suppose that x0 ∈ Ω is such that |u(x0 )| = maxx∈Ω |u(x)|. By replacing u
by eiθ u for an appropriate θ, we can assume that u(x0 ) is real and non-negative.
Let B(x0 , r) ⊂ Ω. Since <eu is harmonic,
u(x0 ) = <eu(x0 )
Z
1
= <eu(x0 + ry) dσ(y) .
σ(S n−1 ) S n−1
Therefore
Z
1
u(x0 ) − <eu(x0 + ry) dσ(y) = 0 ,
σ(S n−1 ) S n−1
We shall now restrict ourselves to n = 2, and focus our attention on the relations
between harmonic and holomorphic functions. We shall need the following lemma
of Fourier analysis.
10 CHAPTER I
Therefore,
n2 |fˆ(n)| ≤ kf 00 k∞ ≤ kf kC 2 .
By the Weierstrass test5 , it follows that the series
X
fˆ(n)eint
n∈Z
3. Poisson integrals
It follows from Lemma 2.3 that this series converges uniformly on D̄. Since each
summand is harmonic in D, so is the sum (verify the mean value property). Then
this is the solution of (3.1).
We use some Fourier analysis to derive an integral formula giving u. By (3.2),
the Fourier series of ur (eit ) = u(reit ) is, for r < 1,
X
ur (eit ) = fˆ(n)r |n| eint .
n∈Z
We set
X
(3.3) Pr (eit ) = r |n| eint ,
n∈Z
7 Tosimplify the notation, here we think of the ϕr as periodic functions on the line. We then
write ϕr (t) instead of ϕr (eit ). The same for f in the next proposition.
14 CHAPTER I
(3.5) lim f ∗ ϕr = f
r→1
uniformly on T. If f ∈ Lp (T), with 1 ≤ p < ∞, the limit (3.5) holds in the Lp -norm.
Proof. Suppose f is continuous. By uniform continuity, given ε > 0, there exists
δ > 0 such that |f (t − u) − f (t)| < ε for |u| < δ. By condition (3), there is r0 < 1
such that for r0 < r < 1, Z
|ϕr (t)| dt < ε .
δ≤|t|≤π
kf ∗ ϕr − f kp ≤ k(f − g) ∗ ϕr kp + kg ∗ ϕr − gkp + kg − f kp .
Then
lim f ∗ ϕr = cf
r→1
where we have made the change of variable t = (1 − r)u. Since (|u| + 1)−2 is
integrable on the line,
Z
1
lim du = 0 ,
r→1 |u|≥ δ (|u| + 1)2
1−r
Theorem 3.5. For f ∈ C(T) the unique solution to the Dirichlet problem (3.1) is
u(reit ) = f ∗ Pr (eit ).
Proof. It follows from the uniform convergence of ur to f and the maximum prin-
ciple.
The same approach that we have used to obtain the Poisson integral u of a
continuous function f can be used to give a formula for its conjugate harmonic
function ũ. Using the Fourier series expansion of f , (3.2) and (2.7), we find that
∞
X ∞
X
ũ(z) = −i fˆ(n)z n + i fˆ(−n)z̄ n .
n=1 n=1
Then
X
(3.7) ũr (eit ) = −1 fˆ(n)sgn n r |n| eint = f ∗ P̃r (eit ) ,
n6=0
where
X
P̃r (eit ) = −i sgn n r |n| eint
n6=0
∞
X
= 2=m r n eint
(3.8) n=0
1
= 2=m
1 − reit
2r sin t
= .
1 + r 2 − 2r cos t
The statements of Proposition 1.1, Lemma 1.4, Corollary 1.5 and Theorem 1.6
remain valid with H p replaced by hp . Only Lemma 1.4 requires a modification in
the proof, which goes as follows.
Lemma 4.1. Let u ∈ hp (D). Then, for every z ∈ D,
kukhp
|u(z)| ≤ Cp .
(1 − |z|)1/p
THE UNIT DISC 17
Proof. Take ρ so that |z| = r < ρ < 1, and set v(z) = u(ρz). Then v is harmonic
in D and continuous on D̄. By Theorem 3.5,
By Young’s inequality,
where p0 is the dual exponent of p. We then estimate the Lq -norm of Pr for any q.
We know that kPr k1 = 1. Moreover
1 − r2 1 − r2 1+r
kPr k∞ = max = = .
|t|≤π 1 + r 2 − 2r cos t (1 − r)2 1−r
If 1 < q < ∞,
Z q1
kPr kq = Pr (eit )q dt
T
q−1 1
≤ kPr k∞q kPr k1q
1 + r q−1
q
≤ .
1−r
Therefore
kukhp
|u(z)| ≤ Cp 1/p ,
1 − ρr
u(reit ) = f ∗ Pr (eit ) .
The usual verification of the mean value property shows that u is harmonic in
D, so that u ∈ h1 (D) and kukh1 ≤ kµkM .
The question we will discuss now is if every hp -function can be obtained in this
way, i.e. if every hp -function is the Poisson integral of an Lp -function on T if
1 < p ≤ ∞, and of a regular Borel measure10 if p = 1.
Theorem 4.3. Consider the operator P mapping f (understood as either a func-
tion or a Borel measure on T) into the harmonic function Pf on D given by
(4.3) (Pf )r = f ∗ Pr .
Then P maps Lp (T) isometrically onto hp (D) for 1 < p ≤ ∞, and it maps M (T)
isometrically onto h1 (D).
The limit limr→1 (Pf )r exists in Lp if and only if one of the following holds
(1) f ∈ Lp (T) and 1 ≤ p < ∞;
(2) p = ∞ and f ∈ C(T).
In each of these cases, (Pf )r → f in the Lp -norm.
For general elements f of M (T) or L∞ (T), (Pf )r tends to f in the corresponding
weak*-topology 11 .
Proof. Suppose first that 1 < p < ∞. Take f ∈ Lp (T) and set u = Pf . By
Propositions 3.2 and 3.4, ur → f in the Lp -norm. In particular,
so that kPf khp = kf kp . We prove next that P : Lp (T) → hp (D) is onto. Take
u ∈ hp (D) and let {rj }j∈N be a sequence of radii tending to 1. Since kurj k ≤ kukhp ,
10 We
regard L1 (T) as a subspace of M (T), identifying the function f with the measure µ such
that dµ(t) = f (t) dt.
11 We recall that M (T) is the dual space of C(T) and L∞ (T) is the dual space of L1 (T). The
weak*-topology on the dual space X 0 of a Banach space X is the weakest topology induced by
the elements of X as linear functionals on X 0 .
THE UNIT DISC 19
it follows from the Banach-Alaoglu theorem12 that some subsequence {urjk } has a
limit f ∈ Lp (T) in the weak* topology of Lp (T). Set v = Pf and take r < 1. Since
0
Pr ∈ Lp (T), and by (4.3),
v(reit ) = f ∗ Pr (eit )
Z
0
= f (eit )Pr (t − t0 ) dt0
T
Z
0
= lim urjk (eit )Pr (t − t0 ) dt0
(4.5) k→∞ T
= lim urjk ∗ Pr (eit )
k→∞
= lim u(rrjk eit )
k→∞
it
= u(re )
Therefore Z
kµkM = sup g(t) dµ(t) ≤ kukh1 .
kgk∞ =1 T
12 The Banach-Alaoglu theorem says that, if X is a separable Banach space and X 0 is its dual
space, the closed balls in X 0 are sequentially compact in the weak* topology.
20 CHAPTER I
ZT T
= (g ∗ Pr )(t0 ) dµ(t0 ) .
T
showing that µ is the weak* limit of (Pµ)r . If the (Pµ)r have a strong limit
as r → 1, this limit is in L1 (T), and at the same time it must coincide with µ,
because the norm topology is stronger than the weak* topology. This implies that
µ ∈ L1 (T).
The same argument, replacing g ∈ C(T) with g ∈ L1 (T), shows that, if f ∈
∞
L (T), then (Pf )r → f in the weak* topology, and that the convergence is in
norm if and only if f is continuous.
Observe that the inclusions hp (D) ⊂ hq (D) if q < p also give other consequences
for convergence to the boundary. For instance, if u is bounded on D, i.e. u ∈
h∞ (D), and f is its boundary function, then ur → f in any Lq -norm for q < ∞.
Corollary 4.4. For r < 1, let u(r) (z) = u(rz). If u ∈ hp (D), with 1 < p < ∞,
u(r) → u in the hp -norm as r → 1. The same is true for p = 1, provided u] ∈ L1 (T)
and for p = ∞, provided u] is continuous. For 1 < p < ∞, harmonic polynomials
are dense in hp (D).
Proof. The first part is a direct consequence of Theorem 4.3. Given ε > 0, take
r < 1 so that ku] − ur kp < ε. Since each half of the Taylor series of u,
∞
X ∞
X
u(z) = an z n + a−n z̄ n ,
n=0 n=1
converges to u uniformly on compact sets, there is N such that the sum sN (z) of
the two partial sums of order N satisfies |u(z) − sN (z)| < ε for |z| = r. Then
ku] − (uN )r kp < 2ε.
The (positive) answer for the other values of p will be given in Chapter III. In
the last part of this section we discuss under which assumptions the congiugate
harmonic function of a function u continuous up to the boundary can be extended
continuously to the boundary.
The case p = 2 is special in the sense that h2 (D), being isometric with L2 (T), is
a Hilbert space. To be precise, given u, v ∈ h2 (D), there are f, g ∈ L2 (T) such that
u = Pf , v = Pg. If we set
hu, vih2 = hf, giL2 ,
this inner product induces the h2 -norm13 . It is also clear that P transforms or-
thonormal bases of L2 (T) into orthonormal bases of h2 (D). In particular, from the
basis {eint }n∈Z of L2 (T) we derive the orthonormal basis {z n }n∈N ∪ {z̄ n }n≥1 of
h2 (D).
Therefore, if X X
u= an z n + a−n z̄ n ,
n≥0 n≥1
then X
kuk2h2 = |an |2 .
n∈Z
1 − r2
P (reit ) = ∈ h1 (D) .
1 + r 2 − 2r cos t
2r sin t
P̃ (reit ) = .
1 + r 2 − 2r cos t
13 It follows from Lemma 4.2 and the polarization identity
1`
kx + yk2 + kx + iyk2 − kx − yk2 − kx − iyk2
´
hx, yi =
4
2
F (z) = P (z) + iP̃ (z) = −1
1−z
Passing to the case p = ∞, the following example shows that the harmonic
conjugate of a bounded function need not be bounded.
Consider an open vertical strip S = {z : a < <ez < b} in the complex plane.
By the Riemann mapping theorem, there is a conformal map ϕ from D onto S.
Adding, if necessary, an imaginary constant to ϕ, we can assume that ϕ(0) ∈ R.
If u = <eϕ, it follows that a < u(z) < b, so that u ∈ h∞ (D). But its harmonic
conjugate ũ is =mϕ, which is not bounded14 .
kũkh∞ ≤ Ckukh∞
14 Another
example is u(z) = =m log(1 − z), which is bounded, but whose harmonic conjugate,
−<e log(1 − z) = − log |1 − z|, is not.
THE UNIT DISC 23
for every u ∈ h∞
c (D). Given r < 1, consider the linear functional on C(T)
Z
ψr (f ) = f ∗ P̃r (1) = f (t)P̃r (−t) dt .
T
= sup |ψr (f )|
f ∈C(T) , kf k∞ ≤1
≤C ,
We prove that this limit equals the expression formally obtained by taking the
pointwise limit of the integrand, i.e.
Z π
1 sin t0
˜
f (t) = f (t − t0 ) − f (t) dt0
2π −π 1 − cos t 0
(5.4) Z π
1 t0
= f (t − t0 ) − f (t) cot dt0 .
2π −π 2
Observe that this integral is absolutely convergent, due to the fact that f is
α-Lipschitz. In fact,
t0 C
(5.5) 0
f (t − t ) − f (t) cot ≤ 0 1−α ,
2 |t |
which is integrable on [−π, π].
Using (1.5), we find that
2r| sin t0 | 2|t0 | 2π 2
(5.6) |P̃r (t0 )| = ≤ π 2
≤ .
|1 − reit0 |2 (|t0 | + 1 − r)2 |t0 |
Hence
C
(5.7) f (t − t0 ) − f (t)P̃r (t0 ) ≤ ,
|t0 |1−α
and dominated convergence can be applied to show that limr→1 f ∗ P̃r (t) = f˜(t)
pointwise. Moreover,
Z π
1 0
kf˜ − f ∗ P̃r k∞ ≤ max f (t − t0 ) − f (t) cot t − P̃r (t0 ) dt0
t∈[−π,π] 2π −π 2
Z π
C t0
≤ |t0 |α cot − P̃r (t0 ) dt0 .
2π −π 2
We can apply dominated convergence again, because the integrand tends to zero
pointwise and
t0 t0 C
cot − P̃r (t0 ) ≤ cot + P̃r (t0 ) ≤ 0 .
2 2 |t |
Therefore,
lim kf˜ − f ∗ P̃r k∞ = 0 .
r→1
The Hardy space H p (D) is a closed subspace of hp (D). It is clear from Section 3
that the Poisson integral of a function (or measure) f on T is holomorphic in D if
and only if fˆ(n) = 0 for n < 0. If we define
and similarly
M+ (T) = {µ ∈ M (T) : µ̂(n) = 0 for n < 0} ,
the statement of Theorem 4.3 can be repeated word by word, replacing hp with
H p , Lp with Lp+ , and M with M+ .
In the case p = 2 we are working with Hilbert spaces, and we want to describe
the orthogonal projection from h2 (D) to H 2 (D), that we shall denote by C (for
Cauchy).
Lemma 6.1. For u ∈ h2 (D), let ũ be its harmonic conjugate. Then
1 1
(6.1) Cu = (u + iũ) + u(0) .
2 2
Denoting by u] ∈ L2 (T) the boundary function of u, i.e. u] = limr→1 ur , then
Z
u] (eit )
(6.2) Cu(z) = −it
dt .
T 1 − ze
Proof. Since {z n }n≥0 ∪ {z̄ n }n≥1 is an orthonormal basis of h2 (D), and {z n }n≥0
spans H 2 (D), if we write
∞
X ∞
X
n
u(z) = an z + a−n z̄ n ,
n=0 n=1
it follows that
∞
X
Cu(z) = an z n .
n=0
Then (6.1) follows from (2.6). Consider now the Fourier series of u] ,
X
u] (eit ) = an eint ,
n∈Z
one obtains (Cu)r from u] by multiplying each Fourier coefficient ub] (n) = an by r n
if n ≥ 0 and by 0 if n < 0. Consider therefore the function
∞
X
it 1
(6.3) Cr (e ) = r n eint = ,
n=0
1 − reit
THE UNIT DISC 27
So
C ] f = lim (CPf )r ,
r→1
2
where the limit is meant in the L -norm.
We want to give an expression of C ] that does not involve the harmonic extension
to the interior. Because of Lemma 6.1, this problem can be reduced to that of
finding a more direct formula for the operator
H : L2 (T) −→ L2 (T) ,
mapping f into
Hf = lim P̃r ∗ f ,
r→1
2
again in the L -norm. This operator is bounded, by Proposition 5.1, and
Hf (eit ) = lim f ∗ P̃r (eit )
r→1
(6.4) Z
0 2r sin t0
= lim f (ei(t−t ) ) dt0 .
r→1 T 1 + r 2 − 2r cos t0
28 CHAPTER I
If we prove that
where
0 P̃1−ε (t0 ) if |t0 | < ε
ϕε (t ) = 0
P̃1−ε (t0 ) − cot t2 if ε ≤ |t0 | ≤ π .
By (5.8),
t0 t0 ε2
0
P̃1−ε (t ) − cot = cot
2 2 1 + (1 − ε)2 − 2(1 − ε) cos t0
t0 ε2
= cot 2
2 ε + 4(1 − ε) sin2 t2
0
t0 ε2
≤ cot 2 .
2 ε + π42 (1 − ε)t0 2
We are interested in this quantity for ε ≤ |t0 | ≤ π, where we can use the inequality
t0 1 π π
cot ≤ t 0 ≤ 0 ≤ .
2 | sin 2 | |t | ε
4
We can also assume that ε is small enough so that π 2 (1 − ε) > 1. For ε and t0
subject to these restrictions,
t0 πε
0
cot − P̃1−ε (t ) ≤ 2 .
2 ε + t0 2
THE UNIT DISC 29
πε
|ϕε (t0 )| ≤ ,
ε2 + t0 2
for t0 ∈ [−π, π]. We can then apply Corollary 3.3 with c = 0. We leave the
verification of conditions (1) and (3) to the reader, and observe that ϕε has integral
0 because it is odd on [−π, π]. This gives (6.6).
We cannot eliminate the limit in (6.5) and write the integral over all of T,
because the integrand would not be absolutely convergent in general. For the same
reason, we cannot replace the integrals outside of the simmetric intervals [−ε, ε]
with integrals outside of arbitrary intervals containing 0, such as, for instance,
[−ε, 2ε], because the result would not be the same.
does not exist, or is infinite, for a generic continuous function f on T (take for
sgn t
instance f (t) = log |t|
near t = 0).
If, however, f is a C 1 -function16 , we have, using the oddness of cot 2t ,
Z Z
t t
lim f (t) cot dt = lim f (t) − f (0) cot dt
ε→0 ε<|t|<π 2 ε→0 ε<|t|<π 2
Z π
t
= f (t) − f (0) cot dt ,
−π 2
where this last integral is absolutely convergent. Moreover the linear functional
Z
t
(6.8) f 7−→ Φ(f ) = lim f (t) cot dt
ε→0 ε<|t|<π 2
kf kC 1 = kf k∞ + kf 0 k∞
on C 1 (T). In fact
Z
π t
Φ(f ) = f (t) − f (0) cot dt
2
−π
Z π
t
≤ f (t) − f (0) cot dt
−π 2
Z π
t
≤ kf 0 k∞ |t| cot dt
−π 2
≤ Ckf kC 1 .
t t
p.v. cot ∗f , or p.v. cot ∗f
2 2
for Φ ∗ f .
As a consequence of Proposition 6.2, we have the following extension result.
it |eit − α|
|ψα (e )| = =1,
|ᾱ − e−it |
32 CHAPTER I
Given ε > 0, it is possible to find r0 > |α| such that if |z| > r0 , then |ψα (z)| >
1 − ε. Then, if r > r0 ,
Passing to the limit as r → 1 and using the arbitrarity of ε, we have the conclu-
sion.
This suggests a way to “remove zeroes” from a function f ∈ H p (D) which is
not identically 0, without modifying its norm. If the zeroes of f in D are a finite
number, α0 , α1 , . . . , αn (counted with their multiplicities), it is sufficient to repeat
the above construction n + 1 times to obtain a factorization of f (z) as g(z)B(z),
with
n
Y Yn
z − αj
(7.2) B(z) = ψαj (z) = .
j=0 j=0
ᾱ j z − 1
Proof. The last equality being obvious, we can assume that k = 0 and a = 1, i.e.
that f = h.
We initially suppose that f is holomorphic on a larger disc DR = {z : |z| < R}
with R > 1, and also that f does not vanish for |z| = 1. We then have only finitely
18 More
precisely, ψα is a conformal map of D onto itself (i.e. a Möbius transformation). It can
be shown that any Möbius transformation has the form eiθ ψα for some α ∈ D and some eiθ ∈ T.
THE UNIT DISC 33
ZT T
= log |f (eit )| dt .
T
But
n
X
log |g(0)| = log |f (0)| − log ψαj (0)
j=0
n
X
=− log |αj | ,
j=0
so that
n
X Z
(7.3) − log |αj | = log |f (eit )| dt .
j=0 T
|αj | |αj |
1− < − log ,
r r
and Z
X |αj |
1− < log |f (reit )| dt .
r T
j:|αj |<r
19 This g0
is true since D 0 is simply connected. In fact g
is holomorphic in D, and it admits a
primitive h. Then (eh /g)0 = 0, so that eh = cg. Adding an appropriate constant to h, we obtain
c = 1.
34 CHAPTER I
This limit is then controlled by the supremum over r on the right-hand side, i.e.
∞
X Z
1 − |αj | ≤ sup log |f (reit )| dt .
r<1 T
j=0
Z Z p1
it it p
exp log |f (re )| dt = exp log |f (re )| dt
T T
Z p1
it p
≤ |f (re )| dt
T
≤ kf kH p ,
so that
∞
X
1 − |αj | ≤ log kf kH p .
j=0
Aj
lim aj = lim =1.
j→∞ j→∞ Aj−1
Consider therefore a point z ∈ D different from the αj , i.e. such that f (z) 6= 0.
In general, we cannot expect that limj→∞ ψαj (z) = 1. However, this difficulty is
bypassed if we by multiply each ψαj by a contant factor of modulus 1.
So, for α ∈ D, define
ᾱ ᾱ z − α
(7.4) ψ̃α (z) = ψα (z) = if α 6= 0, and ψ̃0 (z) = z .
|α| |α| ᾱz − 1
20 Let
m be a probability measure on a set X, and let f : X → R be measurable. If ϕ : R → R
is convex, then
“Z ” Z ` ´
ϕ f (x) dm(x) ≤ ϕ f (x) dm(x) .
X X
For αj 6= 0,
ᾱj z − |αj |2 − |αj |ᾱj z + |αj |
ψ̃αj (z) − 1 =
|αj |(ᾱj z − 1)
1 − |αj | ᾱj z + |αj |
= .
|αj | ᾱj z − 1
Since
|ᾱj z − 1| ≥ 1 − |αj ||z| > 1 − |z| ,
we have
(7.5) ψ̃α (z) − 1 < 2 1 − |αj | .
j
1 − |z|
It follows that
Proposition
P∞ 7.6.
Let {αj } be a sequence of complex numbers αj ∈ D, such that
j=0 1 − |α j | < ∞. Then the Blaschke product (7.7) converges unconditionally 21
and uniformly on compact sets to a function B(z) ∈ H ∞ (D) vanishing only at the
points αj . Moreover, |B(eit )| = 1 almost everywhere on T.
Qn
Proof. Let Bn (z) = j=0 ψ̃αj (z) be the partial products of (7.4). We discuss
unconditional convergence first.
If z is one of the αj , then Bn (z) = 0 for n large, and this holds independently of
ordering.
If z is not one of the αj , it follows from (7.6) that for j ≥ j0 = j0 (z), the points
ψ̃αj (z) are contained in the disc D̃ centered at 1 and radius 1/2. Let log z be the
principal determination of the logarithm in D̃, i.e. such that log 1 = 0. Then, if
n > j0 ,
n
Y
Bn (z) = Bj0 −1 (z) elog ψ̃αj (z)
j=j0
X
n
= Bj0 −1 (z) exp log ψ̃αj (z) .
j=j0
log w
is unconditionally (i.e. absolutely) convergent. Since w−1 is holomorphic in D̃ and
continuous on D̃, | log w| ≤ C|w − 1| for w ∈ D̃. By (7.5),
∞
X ∞
X
log ψ̃αj (z) ≤ C 1 − ψ̃αj (z)
j=j0 j=j0
(7.8) ∞
C X
≤ 1 − |αj | .
1 − |z|
j=j0
By (7.8),
m m
X X
log ψ̃ (z) ≤ log ψ̃α (z)
αj j
j=n+1 j=n+1
m
C 1+r X
≤ 1 − |αj | .
r 1 − r j=n+1
Given ε, 0 < ε < 1, we can then take n0 large enough so that, if n0 ≤ n < m,
m
X
log ψ̃ (z) <ε,
α j
j=n+1
for all z ∈ K. Let a > 0 be such that |1 − ew | < a|w| for |w| < 1. Then
X
m
Bn (z) − Bm (z) ≤ 1 − exp log ψ̃αj (z)
j=n+1
X
m
< a log ψ̃αj (z)
j=n+1
< aε ,
for all z ∈ K.
Finally, we prove that |B(eit )| = 1 a.e. on T. Since |B(z)| < 1 for z ∈ D,
kBkH ∞ ≤ 1, so that |B(eit )| ≤ 1 a.e. on T.
THE UNIT DISC 37
Let
Y∞
B(z)
Rn (z) = = ψ̃α (z) .
Bn (z) j=n+1 j
by Lemma 7.1. Moreover, |Bn (eit )| = 1 for every t, so that |B(eit )| = |Rn (eit )|.
Observe that
Z Z
it
|Rn (0)| = M1 (Rn , 0) ≤ |Rn (e )| dt = |B(eit )| dt .
T T
But
∞
Y P∞
log |αj |
lim Rn (0) = lim |αj | = lim e j=n+1 =1.
n→∞ n→∞ n→∞
j=n+1
Therefore, Z
|B(eit )| dt ≥ 1 ,
T
as before. Therefore,
We can prove now the F. and M. Riesz theorem.
Proof of Theorem 7.2. Let f (z) be the Poisson integral of µ. Since µ̂(n) = 0 for
n < 0, then
∞
X
f (z) = µ̂(n)z n
n=0
It follows that dµ(t) = f (eit ) dt, i.e. µ is absolutely continuous w.r. to the Lebesgue
measure.
8. Dual spaces
is well defined as a linear map from the space of all harmonic functions on D, with
values in the space of homorphic functions on D. It follows from (6.1) that C maps
hp (D) into H p (D) if and only if the conjugate function operator maps hp (D) into
itself.
Therefore we know from Section 5 that this does not happen if p = 1 or ∞, and
we still have to see that it happens instead if 1 < p < ∞. The scope of this section
is to remark that this issue is relevant for determining the dual space of H p (D).
Consider the inner product in h2 (D):
Z Z X
] it ] it
(8.1) hu, vih2 = u (e )v (e ) dt = lim u(reit )v(reit ) dt = a n bn ,
T r→1 T n∈Z
if u] , v ] are the boundary functions and an , bn their Fourier coefficients (i.e. the
Taylor coefficients of u and v respectively). By the general theory of Hilbert spaces,
the map
v 7−→ ϕv (u) = hu, vih2
THE UNIT DISC 39
Proof. The first statement follows from Lemma 8.1. For the second, it is sufficient
to observe that Z
X √ √
|n|
a n bn r = u( reit )v( reit ) dt .
n∈Z T
0
Proposition 8.3. If g ∈ H p (D), the linear functional
ϕg (f ) = B(f, g)
Φ : g 7→ ϕg
0
is a norm-decreasing conjugate-linear immersion of H p (D) into H p (D)∗ .
Proof. The only fact that does not follows P
directly from Corollary 8.2 is the injec-
∞ 0
tivity of Φ. But if ϕg = 0 for some g(z) = n=0 bn z n ∈ H p (D), then
bn = B(z n , g) = 0 ,
and
∞
X
g(z) = bn z n ,
n=0
then ψ(z n ) gives ub] (n) = an for n ≥ 0, but at the same time it equals ϕg (z n ) = bn .
Therefore,
kCukH p0 ≤ Akukhp0 .
THE UNIT DISC 41
0 0
Conversely, suppose that the Cauchy projection maps hp (D) into H p (D), and
let ψ be a continuous linear functional on H p (D). Define
ψ ] : Lp+ (T) −→ C
and Z
ψ(f ) = f ] (eit )h(eit ) dt ,
T
p
for every f ∈ H (D).
0
If u ∈ hp (D) is the Poisson integral of h, so that h = u] ,
ψ(f ) = B(f, u) .
Let
∞
X ∞
X
u(z) = bn z n + b−n z̄ n ,
n=0 n=1
and
∞
X
f (z) = an z n
n=0
0
Hence ψ = Φ(Cu), and Cu ∈ H p (D).
As we have anticipated, we shall see in Chapter III that C is bounded from hp (D)
to H p (D) for 1 < p < ∞, so that, for these values of p, Φ provides an identification
0
between H p (D)∗ and H p (D).
Since C does not map bounded harmonic functions into bounded holomorphic
functions (see Proposition 5.3), not all the continuous linear functionals on H 1 (D)
can be represented as ϕg for some g ∈ H ∞ (D).
Consider, for example, v(z) = arg(1 − z) = =m log(1 − z) ∈ h∞ (D). Then
ϕv is bounded on H 1 (D), and, for f ∈ H 1 (D), B(f, v) = B(f, g) with g(z) =
1 ∞
2i log(1 − z) 6∈ H (D).
More general examples arise from the following Hardy’s inequality.
P∞
Theorem 8.5. Let f (z) = n=0 an z n be in H 1 (D). Then
X∞
|an |
≤ kf kH 1 .
n=0
n + 1
42 CHAPTER I
Proof. We assume first that f (0) = a0 = 0, and prove in this case the stronger
inequality
X∞
|an |
≤ kf kH 1 .
n=1
n
The function Z t
g(t) = f ] (eis ) ds
0
R
is continuous on the real line, and, since T f ] (eit ) dt = f (0) = 0, it is periodic of
period 2π. Hence it defines a continuous function on T, whose derivative is f ] , and
whose Fourier coefficients are
Z 2π Z t
1
ĝ(0) = f ] (eis ) ds dt
2π 0 0
Z 2π Z 2π
1 ] is
= f (e ) dt ds
2π 0 s
Z 2π
1
= f ] (eis )(2π − s) ds
2π 0
Z 2π
1
=− sf ] (eis ) ds ,
2π 0
and
Z 2π Z t
1 −int
ĝ(n) = e f ] (eis ) ds dt
2π 0 0
Z 2π Z 2π
1 ] is
= f (e ) e−int dt ds
2π 0 s
Z 2π
1 1 −ins
= f ] (eis ) (e − 1) ds
2π 0 in
fb] (n)
= ,
in
if n 6= 0. Hence, an
if n > 0
ĝ(n) = in
0 if n < 0 .
It follows that the function
X∞
an n
F (z) = iĝ(0) + z ,
n=1
n
let
∞
X ∞
X
n
g1 (z) = |bn |z , g2 (z) = |cn |z n .
n=1 n=0
One argument is based on the following remark. Let HC(D) the space of holo-
morphic functions that admit a continuous extension to D̄. Then HC(D) is a
closed subspace of H ∞ (D), and proper, because it does not contain the infinite
Blaschke products. By the Hahn-Banach theorem, there is a non-zero continuous
linear functional ϕ on H ∞ (D) which vanishes identically on HC(D). Suppose that
ϕ(f ) = B(f, g) for some g ∈ H 1 (D), and let
∞
X
g(z) = an z n .
n=0
0 = B(z n , g) = an ,
so that g = 0. But this contradicts the fact that ϕ is not identically zero.
This argument motivates a new question: can Φ be surjective from H 1 (D) onto
the dual space of HC(D)? The answer is negative again. It is possible to modify
the first part of the proof of Theorem 8.4 to show that, if this is the case, then the
Cauchy projection maps h1 (D) into H 1 (D), which we know to be false.
THE HALF-PLANE 45
CHAPTER II
HARDY SPACES ON THE HALF-PLANE
We keep the notation Mp (f, r) for kfr kp , and kf kH p (resp. kf khp ) for the supre-
mum in (1.1).
kf khp
|f (z)| ≤ Cp 1 .
yp
Typeset by AMS-TEX
46 CHAPTER II
As for the corresponding spaces on the unit disc, we then have the following
immediate consequence.
Theorem 1.2. H p (D+ ) and hp (D+ ) are Banach spaces, and convergence in their
norm implies uniform convergence on compact sets.
The information given by Lemma 1.1 concerns the behaviour of an hp -function f
both for y → 0 and for y → ∞. It does not say anything, however, on the behaviour
of f (z) when z tends to infinity within a horizontal strip. The best one can say is
the following.
Lemma 1.3. For 0 < a < b, let Sa,b = {x + iy : a ≤ y ≤ b}. If f ∈ hp (D+ ), with
p < ∞, , then
lim f (z) = 0 .
z→∞, z∈Sa,b
If |<ez| > M + a, Bz ⊂ {x + iy : |x| > M, 0 < y < b + a}. Putting (1.3) and
(1.4) together, we obtain that |f (z)| < ε.
In contrast with the corresponding spaces on the unit disc, no Hardy space on
D+ is contained in any other, as it can be seen with appropriate examples. For
instance, if α > 0,
1
fα (z) =
(z + i)α
is in H p (D+ ) if and only if pα > 1, and
1
gα (z) =
z α (z + i)2
2. Poisson integrals
Before discussing Poisson integrals in D+ , we must recall some facts about con-
volution on R on one side, and about conformal mappings on the other side.
Recall now that if A and B are two connected, simply connected, proper open
subsets of C, there exists a conformal mapping23 ϕ from A onto B. For A = D,
the unit disc, and B = D+ , one can explicitely write such mappings. One of them
is
1+z
(2.7) ϕ(z) = i ,
1−z
and it is called the Cayley transform.
Lemma 2.1. The Cayley transform ϕ maps D onto D+ , it is invertible, and
w−i
(2.8) ϕ−1 (w) = .
w+i
Moreover ϕ has a continuous extension to D̄ \ {1}, and
θ
(2.9) ϕ(eiθ ) = − cot ∈ R = ∂D+ .
2
Proof. One has
(1 + z)(1 − z̄) 1 − |z|2 + 2i=mz
ϕ(z) = i = i ,
|1 − z|2 |1 − z|2
so that
1 − |z|2
=mϕ(z) = >0,
|1 − z|2
if z ∈ D . Hence ϕ maps D into D+ .
It is easy to verify that ϕ is injective and that (2.8) gives its inverse function. If
w ∈ D+ , |w −i| < |w +i| by a simple geometric consideration, so that |ϕ−1 (w)| < 1.
This shows that ϕ is onto.
The extension of ϕ to the boundary is easy to derive.
Since v(0) = u(i) and v(eiθ ) = u(− cot 2θ ) by (2.9), the change of variable − cot θ
2 =t
gives
Z 2π Z
1 iθ 1 +∞ 1
u(i) = v(e ) dθ = u(t) 2 dt ,
2π 0 π −∞ t +1
which is (2.10) for x + iy = i.
For a general point x0 + iy0 ∈ D+ , consider the function ũ(z) = u(x0 + y0 z),
which satisfies the same hypotheses as u. Then, with simple changes of variables,
1 y
Py (x) =
π x + y2
2
(2.11) u y = u0 ∗ Py .
50 CHAPTER II
(2.12) Py+y 0 = Py ∗ Py 0 .
Proof. Let v(z) = u(z + iy1 ). By Lemma 1.1, v satisfies the assumptions of Lemma
2.3. Therefore
uy2 = vy2 −y1 = v0 ∗ Py2 −y1 = uy1 ∗ Py2 −y1 .
Applying this identity to
and Z
1 1
kP1 k = dx = 1 .
π R x2 +1
We are now in a position to extend to the spaces hp (D) the results proved about
norm or weak* convergence to the boundary. We summarize them in the following
statement.
THE HALF-PLANE 51
We can apply Fubini’s theorem and change the order of integration for the fol-
lowing reason. For fixed θ, write z0 + reiθ = xθ + iyθ . Then
P (z0 + reiθ − t)f (t) = P (xθ + iyθ − t)f (t)
0
is integrable in t, because it is the product of an Lp - and an Lp -function. By (2.13),
Z q1
1 x q Cq
kPy kq = q
P 1 dx = 1 .
R y y y 1− q
if 1 ≤ q < ∞, and similarly for q = ∞.
There is a δ > 0 such that =myθ ≥ δ for all θ. Therefore
Z
P (z0 + reiθ − t)f (t) dt ≤ Cp,δ kf kp
R
= f (t)P (z0 − t) dt
R
= u(z0 ) .
The following corollary can be seen as a replacement for the (non-existing) in-
clusion relations among the hp (D+ ).
52 CHAPTER II
Mp (u, y) = ku] ∗ Py kp
≤ k(u] − ϕ) ∗ Py kp + kϕ ∗ Py kp
< ε + kϕk1 kPy kp
− p10
≤ε+y kϕk1 .
We begin this section by recalling some of the basic facts about the Fourier
transform on R.
If f ∈ L1 (R) and ξ ∈ R, one defines the Fourier transform of f at ξ as
Z
(3.1) fˆ(ξ) = f (x)e−iξx dx .
R
Then |µ̂(ξ)| ≤ kµk1 , µ̂ is continuous, but we can no longer say that it vanishes
at infinity.
THE HALF-PLANE 53
√
(8) because of (3.2), F extends to every f ∈ L2 (R), and kfˆk2 = 2πkf k2 ;
(9) since Lp (R) ⊂ L1 (R) + L2 (R), if 1 < p < 2, F is well defined on Lp (R),
fˆ ∈ Lp (R), and the Hausdorff-Young inequality holds:
0
(3.3) kfˆkp0 ≤ kf kp ;
cy (ξ) = e−y|ξ| .
P
cy (ξ) = P
P c1 (yξ) ,
This must be equal to P1 (x), hence a = 1. The rest of the proof is obvious.
The Fourier transform allows to describe for which functions f ∈ Lp (R), the
Poisson integral u(x + iy) = f ∗ Py (x) is holomorphic. The main result concerns
p = 2, and is called the Paley-Wiener theorem.
Theorem 3.2. A function u ∈ h2 (D+ ) is holomorphic if and only if the Fourier
transform ub] of its boundary function u] ∈ L2 (R) is zero on the negative half-
line. Hence a holomorphic function f on D+ is in H 2 (D+ ) if and only if f is the
Fourier-Laplace transform,
Z ∞
1
(3.6) f (z) = g(ξ)eizξ dξ ,
2π 0
Proof. Suppose that u is holomorphic. Calling γa the horizontal curve γa (t) = t+ia,
consider the line integral
Z Z ∞
−iξz
Ia = u(z)e dz = u(x + ia)e−iξ(x+ia) dx = eaξ u
ca (ξ) .
γa −∞
Now,
Z Z b
−iξz
u(z)e dz = i u(M + it)e−iξ(M +it) dt ,
+
γM a
THE HALF-PLANE 55
so that
Z Z b
u(z)e −iξz
dz ≤ |u(M + it)|eξt dt
+
γM a
If the support of γ is contained in the half-plane =mz ≥ a > 0, then |eizξ | ≤ e−aξ ,
and g(ξ)eizξ is integrable on γ × R+ . We can then change order of integration and
obtain that Z Z ∞ Z
1
f (z) dz = g(ξ) eizξ dz dξ = 0 ,
γ 2π 0 γ
Proof. Let v(ε) (z) = u(z + iε). Then v(ε) ∈ hp (D+ ) ∩ h∞ (D+ ), hence in h2 (D+ ).
If we assume that u ∈ H p (D+ ), then v(ε) ∈ H 2 (D+ ). Therefore,
d]
v(ε) cε (ξ) = e−ε|ξ| ub] (ξ)
(ξ) = u
56 CHAPTER II
By the Paley-Wiener theorem and Theorem 2.6, the map which assigns to a
harmonic function u ∈ h2 (D+ ) the Fourier transform ub] is, up to a constant, an
isometry of h2 (D+ ) onto L2 (R), and the image of H 2 (D+ ) is
Since the orthogonal projection of L2 (R) onto L2+ is obviously the map g 7→
gχR+ , we derive the following recipe for constructing the orthogonal projection
from h2 (D+ ) onto H 2 (D+ ): starting with u ∈ h2 (D+ ), take ub] , multiply it by χR+ ,
and apply (3.6). This gives the Cauchy projection on D+ ,
Z ∞
1
Cu(x + iy) = ub] (ξ)ei(x+iy)ξ dξ
2π 0
(4.1)
= F −1 ub] e−yξ χR+ (x)
= u] ∗ F −1 (e−yξ χR+ )(x) .
with z = x + iy.
If u is already in H 2 (D+ ), then Cu = u, and (4.3) formally coincides with the
ordinary Cauchy integral formula
Z
1 u] (w)
u(z) = dw ,
2πi γ0 w−z
THE HALF-PLANE 57
where we denote by γ0 the curve describing the real axis with its natural orientation.
Notice however that the curve is not bounded25 and it is not contained in the interior
of the domain where u is holomorphic.
Write now
1 x − iy 1 y i x 1
Cy (x) = − = + = P y (x) + i P̃ y (x) ,
2πi x2 + y 2 2π x2 + y 2 2π x2 + y 2 2
noticing that the real part of 2Cy is the Poisson kernel Py and setting
1 x
(4.4) P̃y (x) = .
π x + y2
2
1 i
(4.5) Cu(x + iy) = u(x + iy) + u] ∗ P̃y (x) .
2 2
Observe however that P̃y is not integrable for any y > 0, so that the convolution
u ∗ P̃y is not defined for u ∈ h∞ (D+ ). On the other hand, P̃y is in Lq (R) for any
]
q > 1 and in C0 (R), so that the convolution is well defined for u ∈ hp (D+ ) with
p < ∞.
Proposition 4.2. Let u ∈ hp (D+ ) with p < ∞. Then
lim ũ(z) = 0 ,
z→∞ , =mz≥a
for every a > 0. If u is real-valued, the ũ is the only real-valued function satisfying
the above properties.
Proof. We verify the mean value property for ũ. If B(z0 , r) ⊂ D+ , then
Z Z Z
iθ
(4.6) ũ(z0 + re ) dθ = P̃ (z0 + reiθ − t)u] (t) dt dθ .
T T R
We can apply Fubini’s theorem and change the order of integration for the fol-
lowing reason. For fixed θ, write z0 + reiθ = xθ + iyθ . Then
1 x
P̃y (x) = P̃1 ,
y y
25 It becomes a nice closed curve if embedded in the Riemann sphere.
58 CHAPTER II
and P̃1 ∈ Lq (R) for q > 1, we have, as in the proof of Theorem 2.6,
1
kP̃y kp0 ≤ Cp y − p ,
for p < ∞.
The proof continues as for Theorem 2.6.
Observe that in the proof we have implicitely obtained the inequality
1
(4.7) |ũ(x + iy)| ≤ Cp y − p kukhp .
(4.8) ub̃] (ξ) = iub] (ξ) − 2iχR+ (ξ)ub] (ξ) = −i(sgn ξ)ub] (ξ) ,
The operations described in (4.8) take place on the real line, and define a linear
operator H : L2 (R) → L2 (R),
Hf = F −1 − i sgn ξ fˆ(ξ) .
This operator is called the Hilbert transform. Proposition 6.2 in Chapter I has
the following analogue (with a very similar proof, that we omit).
Proposition 4.3. For f ∈ L2 (R),
Z
1 1
Hf (x) = lim f (x − t) dt ,
ε→0 π |t|>ε t
Much of the content of Sections 5 and 6 of Chapter I can be repeated here, with
the unit disc replaced by the upper half-plane.
The Cauchy projection is bounded on hp (D+ ) for p 6= 2 if and only if the same
is true for the conjugate function operator mapping u to ũ. As for the unit disc
(and with the same arguments) this turns out to be false if p = 1. Also for p = ∞,
when the formulas above do not hold, it is not true in general that the conjugate
functions of a bounded harmonic function are bounded26 . This is not even true if
u is bounded and continuous on D+ .
We shall see later that C is bounded from hp (D+ ) onto H p (D+ ) for 1 < p < ∞.
We shall now establish simple relations between H p -functions on the unit disc
and the upper half-plane. This will allow us to transfer to D+ many results obtained
in Chapter I for D.
The simplest situation occurs for p = ∞: if ϕ denotes the Cayley transform
(2.7), the map
is an isometry of H ∞ (D) onto H ∞ (D+ ), for the simple reason that f is bounded
if and only if f ◦ ϕ−1 is bounded, and the two ∞-norms are equal27 .
Clearly the same does not hold for p < ∞ (take f = 1). In order to see what
needs to be modified, let us consider the simple case where f is continuous on D̄
and holomorphic in the interior.
Then g = f ◦ ϕ−1 is continuous on D+ , and let us try to compute the Lp -norm
of g|R . By (2.9), with the change of variable x = − cot θ2 we have
Z ∞ p1
−1 p
kg|R kp = |f ◦ ϕ (x)| dx
−∞
Z 2π p1
1 iθ p dθ
= |f (e )| ,
2 0 sin2 θ2
and this differs from the Lp -norm of g by the factor sin−2 θ2 coming from the change
−2i
of variable. This suggest that we must multiply g by a power of (ϕ−1 )0 (z) = (z+i) 2.
p
Precisely, for p < ∞ and f ∈ H (D), we set
1
(5.2) Tp f (z) = 1 2 f ◦ ϕ−1 (z) .
π (z + i)
p p
kg|R kp = kf|T kp .
We have so proved that Tp is an isometry from a dense subspace of H p (D) (see
Corollaries 4.3 and 7.3 in Chapter I) onto its image, call it V , in H p (D+ ). Therefore
Tp extends to an isometry (also denoted by Tp ) of all H p (D) onto the closure of V
in H p (D+ ). The proof will be finished if we prove that Tp has a dense image in
H p (D+ ) if p > 1 or in H01 (D+ ) if p = 1.
Take g ∈ H p (D+ ), resp. in H01 (D+ ), ε > 0, and define
g(z + iε)
hε (z) = .
(1 − iεz)2
Observe that for z = x+iy ∈ D+ , |1−iεz| ≥ 1+εy ≥ 1, so that the denominator
does not vanish, and moreover
|hε (z)| ≤ |g(z + iε)| .
This implies that hε ∈ H p (D+ ). We prove that limε→0 hε = g in H p (D+ ). In
fact
Z p p1
gε (x) ]
khε − gkH p = 2
− g (x) dx
R (1 − iεx)
Z p1
|gε (x) − g ] (x)|p
≤ dx
R |1 − iεx|2p
Z p p1
] p 1
+ |g (x)| 2
− 1 dx
R (1 − iεx)
Z p1
] p
≤ |gε (x) − g (x)| dx
R
Z p p1
] p 1
+ |g (x)| 2
− 1 dx .
R (1 − iεx)
THE HALF-PLANE 61
As ε → 0, the first
term tends to 0 by Theorem 2.6, and the second by dominated
convergence, since (1−iεx)
1
2 − 1 ≤ 2.
Proof. For p > 1, we take g = Tp−1 f ∈ H p (D), and apply Lemma 7.5 in Chapter I.
The zeroes of g are the points
αj − i
βj = ϕ−1 (αj ) = .
αj + i
Then
∞
X ∞
X
2
(1 − |βj | ) (1 − |βj |) < ∞ ,
j=0 j=0
and, if αj = aj + ibj ,
1
gε = −1 ε T1−1 f ε .
T1 f (0)
=mαεj bj − ε bj − ε
1 − |βjε | ε = 2 2 .
|αj + i| 2 aj + (bj − ε + 1) 2 aj + b2j + 1
Therefore,
X bj − ε
≤ C log kf kH 1 (D+ ) − log 4πf (1 + ε)i .
ε
a2j + b2j + 1
j∈Eε
If ε decreases, the left-hand side increases for each j ∈ N (taking into account
that the sets Eε also increase). Passing to the limit by monotone convergence,
X bj
≤ C log kf k H 1 (D ) − log |4πf (i)| < ∞ .
a2j + b2j + 1 +
j∈N
β̄ w − β |α2 + 1| z − α
σα (z) = ψ̃β (w) = = 2
|β| β̄w − 1 α + 1 z − ᾱ
if α 6= i, and
z−i
σi (z) = ϕ−1 (z) = .
z+i
Once we are at this stage, we can simply state the analogues of the results of
Section 7 in Chapter I. They can be transfered directly from D via the maps Tp for
p > 1, or reproved in exactly the same way for p = 1.
THE HALF-PLANE 63
CHAPTER III
POINTWISE CONVERGENCE TO THE BOUNDARY
AND CONJUGATE HARMONIC FUNCTIONS IN hp
We attack now a question that has been in the background so far, i.e. the
pointwise behaviour of hp - or H p -functions. In particular, in the cases where f ] is
a function, we wnt to know (e.g. in the disc, to fix the notation) if
for almost every eit ∈ T. If f ∈ H ∞ (D) and f ] is continuous, the answer is positive,
since we know that the functions fr tend to f ] uniformly. Partial answers can be
given easily also in other cases. If f ∈ H p (D) with 1 < p < ∞, then fr → f ]
in norm, and therefore there is a sequence rn → 1 such that frn → f almost
everywhere. But this is much less than what we are looking for.
Sharp answers to our question follow from considerations about maximal func-
tions. This notion makes sense and is useful in many different situations, and it is
worth therefore to discuss their properties in a general context.
for all x, y, z ∈ X;
(5) the balls B(x, r) = {y : d(x, y) < r} form a fundamental neighborhood
system at each x ∈ X.
Let m be a positive regular Borel measure on X. One says that m is doubling
(more precisely, d-doubling) if there is a constant c0 such that
(1.2) m B(x, 2r) ≤ c0 m B(x, r)
Examples.
The following are spaces of homogeneous type29 :
0 0
(1) The unit circle T, with the distance d(eit , eit ) = |eit − eit | (or equivalently
with the quotient distance of R/2πZ) and the normalized Lebesgue measure.
(2) Rn , with Euclidean distance and the Lebesgue measure.
(3) R, with the Euclidean distance and the measure dm(x) = |x|α dx, with
α > −n.
(4) Rn , with the Lebesgue measure and the distance
n o
d(x, y) = max |x1 − y1 |1/d1 , . . . , |xn − yn |1/dn ,
where d1 , . . . , dn > 0.
(5) Z, with the distance d(n, m) = |n − m| and the counting measure m(E) =
cardE.
(6) The unit sphere S n−1 ⊂ Rn , with the induced distance from Rn and the
Hausdorff measure.
(7) The unit sphere S 2n−1 ⊂ Cn , with the Hausdorff measure and the distance
d(ζ, ζ 0 ) = |1 − hζ, ζ 0 i|, where hζ, ζ 0 i is the Hermitean inner product in Cn .
i.e. limited to the averages of |f | over balls centered at x. Clearly M 0 f (x) ≤ M f (x);
however M 0 f is not necessarily lower-semicontinuous. The measurability of M 0 f
follows in this case from the fact that the map
Z
1
F (x, r) = |f (y)| dy
m B(x, r) B(x,r)
C
M f (x) ≥ ,
1 + |x|n
so that M f 6∈ L1 .
Nevertheless, the starting point of our proof is that, for p = 1, M satisfies a
weaker form of boundedness. We first recall some properties of the distribution
function, defined for α > 0,
(1.6) δf (α) = m {x : |f (x)| > α}
of an m-measurable function f on X.
Lemma 1.2. Let (X, m) be a measure space. If f ∈ Lp (X, m), 1 ≤ p < ∞, then
the following hold:
(1) the Chebishev inequality
kf k p
p
δf (α) ≤ ;
α
and, setting c0 = 0,
Z ∞ n Z
X cj n
X
p−1 p−1
p m(Eα )α dα = p α m(Ak ) dα
0 j=1 cj−1 k=j
n
X n
X
= (cpj − cpj−1 ) m(Ak )
j=1 k=j
n
X
= cpj m(Ak )
j=1
= kf kpp .
The expression “weak-type” comes from the fact that (1.7) is a weaker condition
than boundedness on Lp , as a consequence of Chebishev’s inequality. In fact, if T
is bounded on Lp (X),
kT f k p kf k p
p p
δT f (α) ≤ ≤ kT kpLp →Lp .
α α
One also says that an operator T is strong type (p, p) if it is bounded on Lp (X).
Going back to the maximal operator M , we shall see that it is weak-type (1, 1).
The proof is based on the following Vitali covering lemma.
Lemma 1.3. Inside a space X of homogeneous type, let {Bj }j∈J be a finite family
of balls covering a measurable set E. There exists a sub-family {B j }j∈J 0 such that
Bj ∩ Bk = ∅ for j, k ∈ J 0 , j 6= k, and
[
m Bj ≥ κm(E) ,
j∈J 0
POINTWISE CONVERGENCE 69
empty intersection and the radius of B 0 is not larger than the radius of B, then
B 0 ⊆ B ∗ by (1.1).
Let B 0 one of the excluded balls. It necessarily intersects one of the selected
ones. Let `¯ be the smallest integer ` such that B 0 ∩ Bj` 6= ∅. Then the radius Bj`¯
is greater than or equal to the radius of B 0 , so that
B 0 ⊆ Bj∗`¯ .
Therefore [ [
E⊆ Bj ⊆ Bj∗ ,
j∈J j∈J 0
1
so that, with κ = c0 ν ,
X X [
m(E) ≤ m(Bj∗ ) = κ−1 m(Bj ) = κ−1 m Bj .
j∈J 0 j∈J 0 j∈J 0
i.e.
Z
1
(1.8) m(Bx ) ≤ |f (y)| dy .
α Bx
(2.1) Mvert f (x) = sup Pf (x + iy) ,
y>0
Obviously,
Mvert f (x) ≤ Mnt,α f (x) ≤ Mnt,α0 f (x) ,
if α < α0 .
Lemma 2.1. For every α ∈ (0, π/2) there is a constant Cα > 0 such that, if
f ∈ Lp (R),
Mnt,α f (x) ≤ Cα M f (x) .
In particular, each Mnt,α and Mvert are weak-type (1, 1) and bounded on Lp for
1 < p < ∞.
Proof. We can assume that α > π/4, so that a = tan α > 1. Consider the dyadic
a
intervals Ija = [−a2j , a2j ] in the real line, with j ≥ 0. We set Eja = Ija \ Ij−1 for
72 CHAPTER III
It follows that
Z
f ∗ Py (x) ≤ |f (x − t)|Py (t) dt
R
∞
X Z
2−2j
≤C |f (x − t)| dt
(2.4) y Ijay
j=0
∞
X Z x+ay2j
2−2j
= Ca |f (t)| dt .
j=0
y x−ay2j
Therefore
∞
X
f ∗ Py (x) ≤ Ca2 2−j M f (x0 ) ,
j=0
and
Mnt,α f (x0 ) ≤ Ca2 M f (x0 ) .
POINTWISE CONVERGENCE 73
are in Lp (R) and ku∗α kp ≤ Cp,α kukhp . The same is true for
def
u∗ (x) = sup u(x + iy) .
y>0
Proof. For p > 1, u∗α ∈ Lp (R) because u] ∈ Lp (R) and by Lemma 2.1.
It remains to discuss the case u ∈ H 1 (D+ ). By Corollary 5.5 in Chapter II, we
1
can factorize u as u = vw, with v, w ∈ H 2 (D+ ) and kvkH 2 = kwkH 2 = kukH
2
1.
Then
u∗α (x) = sup v(z)w(z) ≤ vα∗ (x)wα∗ (x) .
z∈Γα (x)
Therefore,
We pass now to the unit disc. The substitute for the vertical maximal function
is the radial maximal function,
Mrad f (eit ) = sup Pf (reit ) .
r<1
Lemma 2.3. For every ρ ∈ (0, 1) there is a constant Cρ > 0 such that, if f ∈
L1 (T),
Mnt,ρ f (eit ) ≤ Cρ M f (eit ) .
In particular, each Mnt,ρ and Mrad are weak-type (1, 1) and bounded on Lp for
1 < p < ∞.
Proof. We use (3.6) in Chapter I,
1−r
Pr (eit ) ≤ C
t2 + (1 − r)2
with t ∈ [−π, π], and proceed as in the proof of Lemma 1.1, just replacing y by
a(1−r)
1 − r, and stopping the sums in j as soon as the intervals Ij do not intersect
[−π, π].
74 CHAPTER III
are in Lp (T) and ku∗α kp ≤ Cp,ρ kukhp . The same is true for
def
u∗ (x) = sup u(reit ) .
r<1
Therefore, using Lemma 2.3 and denoting by |E| the normalized Lebesgue mea-
sure of E ⊂ T, we have, for α > 0,
it
(3.1) {e : δu(eit ) > α} ≤ Cρ kf k1 .
α
Given ε > 0, there is g ∈ C(T) such that kf − gk1 < ε. By Theorem 4.3 in
Chapter I, v = Pg is continuous on D̄, so that
and therefore
δ(u − v)(eit ) = δu(eit ) .
Applying (3.1) to u − v, we obtain that
it
(3.2) {e : δu(eit ) > α} ≤ Cρ ε ,
α
and this holds for every ε > 0. Hence
it
{e : δu(eit ) > α} = 0 ,
almost everywhere.
Corollary 3.2. If ρ < 1 and u ∈ hp (D), 1 < p ≤ ∞, or in H p (D), 1 ≤ p ≤ ∞,
then for almost every eit ∈ T, limz→eit , z∈Sρ (eit ) u(z) = u] (eit ) for every ρ < 1.
In the proof of Theorem 3.1 we have used in a crucial way the density of con-
tinuous functions in L1 , in order to have (3.2). The same proof can be adapted to
prove the analogous result in the upper half-plane. In this case, we approximate
f ∈ L1 (R) by functions g ∈ C0 (R). This also works for f ∈ Lp (R) as long as p < ∞,
but it breaks down for p = ∞. For this case we use a different argument.
Theorem 3.3. If u = Pf , with f ∈ Lp (R) and 1 ≤ p ≤ ∞, then u converges to f
non-tangentially a.e.
Proof. If f ∈ L1 (R), we proceed as described above. Take now f ∈ L∞ (R). Then
u ∈ h∞ (D+ ). If ϕ is the Cayley transform in (2.7) of Chapter II, then v = u ◦ ϕ ∈
h∞ (D). We know by Corollary 3.2 that v converges to v ] (eit ) non-tangentially a.e.
Fix now x ∈ R and α > 0, let Γ0α (x) = Γα (x) ∩ {z : =mz < 1} and eit =
ϕ (x). Since ϕ0 (eit ) 6= 0, ϕ is a diffeomorphism
−1
of a neighborhood of eit onto a
neighborhood of x. Therefore ϕ−1 Γ0α (x) is contained in a Stolz angle Sρ (eit ).
76 CHAPTER III
In particular,
lim u(x + iy) = v ] ϕ−1 (x)
y→0
i.e. uy → v ] ◦ ϕ−1 in the weak* topology. Hence v ] ◦ ϕ−1 = u] = f , and (3.3) then
says that u converges to f non-tangentially a.e.
At this point, the simplest argument to prove the statement for 1 < p < ∞ is to
observe that any f ∈ Lp (R) decomposes as a sum f = f1 + f∞ , with f1 ∈ L1 (R)
and f∞ ∈ L∞ (R). To see this, take
f (x) if |f (x)| ≤ 1 ,
f∞ (x) =
0 if |f (x)| > 1 ,
and f1 = f − f∞ .
Corollary 3.4. If α < π/2 and u ∈ hp (D+ ), 1 < p ≤ ∞, or in H p (D+ ), 1 ≤ p ≤
∞, then limz→x , z∈Γα (x) u(z) = u] (x) almost everywhere.
The case p = ∞ in Corollary 3.2 and 3.4 is referred to as “Fatou’s theorem”.
The discussion in the previous section does not say anything about non-tan-
gential limits of general h1 -functions. We complete the picture here, proving that
Corollary 3.4 can be extended to h1 (R) (the same can be done on T, but we omit
the proof). It must be noted however that no maximal function is involved in the
proof.
Every h1 -function on D+ is the Poisson integral of a measure µ ∈ M (R). We
recall the Lebesgue decomposition of µ as
µ = µa + µs ,
where µa is absolutely continuous with respect to Lebesgue measure m (or µa
m), i.e. dµa (x) = h(x) dx with h ∈ L1 (R), and µs is singular with respect to
the Lebesgue measure (or µs ⊥ m). This means that there is a set E such that
m(R \ E) = 0 and |µs |(E) = 0.
The function
ϕ(x) = |µs |(−∞, x)
is non-decreasing, hence differentiable at (Lebesgue-) almost every point, and its
being singular implies that ϕ0 (x) = 0 a.e. This means that at a.e. x ∈ R,
|µs |(x − h, x + h)
(4.1) lim =0.
h→0 h
POINTWISE CONVERGENCE 77
Lemma 4.1. Let µ be a singular measure on R, and u = Pµ. Then for almost
every x ∈ R,
lim u(z) = 0
z→x , z∈Γα (x)
We discuss now a crucial point in the theory of Hardy spaces, that has been left
aside in the previous chapters: the fact that for 1 < p < ∞ the conjugate harmonic
function of an hp -function is also in hp .
We begin with the unit disc, and recall that if
∞
X ∞
X
n
u(z) = an z + a−n z̄ n
n=0 n=1
Proof. We have
∂x2 (up ) = ∂x pup−1 ∂x u = p(p − 1)up−2 (∂x u)2 + pup−1 ∂x2 u .
summing with the corresponding formula for ∂y2 (up ), we have (5.2).
As to (5.3), setting the standard notation ∂z = 12 (∂x − i∂y ), ∂z̄ = 21 (∂x + i∂y ),
we have ∆ = 4∂z ∂z̄ . We fix a point z0 in the domain of f , and a determination
p
of log f in a neighborhood of z0 . We then set f p/2 (z) = e 2 log f (z) . Then, in this
neighbourhood of z0 ,
Teorema 5.2. If 1 < p < ∞, there is a constant Cp such that kũkhp ≤ Cp kukhp
for every u ∈ hp (D).
Proof. The case p = 2 is already known, with C2 = 1.
Consider first the case where 1 < p < 2, and u > 0 in D. Let f = u + iũ. Then
f is holomorphic and non-zero in D.
POINTWISE CONVERGENCE 79
But
(5.5) |∇u|2 = (∂x u)2 + (∂y u)2 = (∂x u)2 + (∂x ũ)2 = |∂x f |2 = |f 0 |2 .
In other words,
∂r Mp (f, r)p ≤ p0 ∂r Mp (u, r)p .
Since Mp (u, 0) = Mp (f, 0) = u(0), we conclude that Mp (f, r)p ≤ p0 Mp (u, r)p for
every r < 1, hence kf kH p ≤ (p0 )1/p kukhp .
Since |ũ| ≤ |f |, we also have
u = u1 − u2 + iu3 − iu4 ,
4
X
kuj khp ≤ Ckukhp .
j=1
Therefore
4
X
kũkhp ≤ kũj khp ≤ Cp kukhp ,
j=1
≤ Cp0 kukhp .
1
Mp (∇u, y) ≤ Mp P|u] |, y ;
y
POINTWISE CONVERGENCE 81
1 1
(5.8) |∇Py (x)| = 2
= Py (x) .
π|x + iy| y
By (5.7),
Z
1 1
|∇u(x + iy)| ≤ |u] (t)|Py (x − t) dt = P|u] | (x + iy) .
y R y
The first part of the statement then follows from Lemma 1.3 in Chapter II, since
P|u] | ∈ hp (D+ ), and the last part is now obvious.
Proof of Theorem 5.3. As in the proof of Theorem 5.2, we consider first the case
1 < p < 2. We further suppose that u = Pu] , with u] ≥ 0, continuous with compact
support, and non-identically zero. Then u is strictly positive on D+ .
In this hypotheses, we can say right now that ũ is “not far from” being in hp (D+ ).
In fact
− p10
(5.9) Mp (ũ, y) = ku] ∗ P̃y kp ≤ ku] k1 kP̃y kp ≤ Cp y ku] k1 ;
− p10
(5.9’) Mp (u, y) ≤ ku] k1 kPy kp ≤ Cp y ku] k1 ,
and
Z R Z b
p
− ∂y |f (x + ia)| dx + ∂x |f (R + iy)p | dy
−R a
Z R Z b
p
(5.12) + ∂y |f (x + ib)| dx − ∂x |f (−R + iy)p | dy
−R a
Z
= p2 |f (z)|p−2 |f 0 (z)|2 dz .
Qa,b,R
We first fix a and b and let R tend to infinity. By Lemma 5.4 and Lemma 1.3 in
Chapter II,
∂x u(±R + iy)p = pu(±R + iy)p−1 ∂x u(±R + iy)
tends to zero as R → +∞, uniformly in y ∈ [a, b].
To see that the same holds for ∂x |f (−R + iy)p |, observe that
p p p p p
∂x |f p | = <e∂z f 2 f 2 = <ef 2 −1 f 2 f 0 ,
2
so that
(5.13) ∂x |f p | ≤ p |f |p−1 |f 0 | .
2
By (5.5), f 0 (x + iy) tends to zero as x → ±∞ uniformly in y ∈ [a, b]. The same
is true for |f (z)|, as a consequence of (5.10).
We can then let R → +∞ in (5.11) and (5.12), to obtain
Z Z
p
− ∂y u(x + ia) dx + ∂y u(x + ib)p dx
R
(5.14) Z R
= p(p − 1) u(z)p−2 |∇u(z)|2 dz ,
x∈R , y∈[a,b]
and
Z Z
p
− ∂y |f (x + ia)| dx + ∂y |f (x + ib)|p dx
R R
(5.15) Z
2
=p |f (z)|p−2 |f 0 (z)|2 dz .
x∈R , y∈[a,b]
We now make tend b to +∞, and show that the integrals containg b tend to zero.
In (5.14) we use Hölder’s inequality to obtain
Z Z
∂y u(x + ib)p dx ≤ p u(x + ib)p−1 |∂y u(x + ib)| dx
R R
≤ pMp (u, b)p−1 Mp (∂y u, b) ,
which tends to 0 by Lemma 5.5. For the second integral in (5.15), we use the
analogue of (5.13) for ∂y (|f |p ) to obtain that
Z
∂y |f (x + ib)|p dx ≤ p Mp (f, b)p−1 Mp (f 0 , b)
2
R
p p−1
≤ Mp (u, b) + Mp (ũ, b) Mp (∇u, b) .
2
POINTWISE CONVERGENCE 83
This last quantity tends to zero by (5.10) and Lemma 5.4. So (5.14) and (5.15)
respectively imply that
Z Z
p
− ∂y u(x + ia) dx = p(p − 1) u(z)p−2 |∇u(z)|2 dz
R x∈R , y≥a
(5.16) Z Z
− ∂y |f (x + ia)|p dx = p2 |f (z)|p−2 |f 0 (z)|2 dz .
R x∈R , y≥a
We claim that the two left-hand sides in (5.16) are −∂y Mp (u, y)p | and
p
y=b
−∂y Mp (f, y) | respectively. We postpone this proof, and assume this fact
y=b
to be true for the moment.
Since Mp (u, y) and Mp (f, y) are decreasing in y, the left-hand sides in (5.16) are
in fact positive quantities. Since
by (5.9) and (5.9’), the same argument used in the proof of Theorem 5.2 shows that
for every y > 0. The rest of the proof will then proceed as in Theorem 5.2, taking
initially u as the Poisson integral of a continuous, complex-valued function u] on R
with compact support. Once we have the estimate
kũkhp ≤ Cp kukhp
In order to be allowed to move the limit inside the integral, we use dominated
convergence. We shall use vertical maximal functions for this purpose.
Fix ε > 0 and smaller than b/2, and call u(ε) (z), resp. f(ε) (z), the functions
u(z +iε), resp. f (z +iε). Notice that u(ε) , ∂y u(ε) ∈ hp (D+ ) and f(ε) , f(ε)
0
∈ H p (D+ ),
by (5.9), (5.9’) and Lemma 5.4.
We take h > −(b − ε), as we can, in order to stay above the level ε.
By the mean value theorem, for every x there is t = t(x) ∈ (b, b + h) such that
p
u x + i(b + h) − u(x + ib)p
= ∂y u(x + iy)p |y=t = pu(x + it)p−1 ∂y u(x + it) .
h
Therefore,
u x + i(b + h)p − u(x + ib)p p−1
≤ p Mvert u(ε) (x) Mvert (∂y u(ε) )(x) .
h
Applying Hölder’s inequality and Theorem 2.2, we can see that this last function
is integrable. The same applies to f .
84 CHAPTER III
We can now complete some statements left incomplete in the previous chapters.
We give a list of them.
(i) The Cauchy projection is bounded from hp onto H p for 1 < p < ∞, both
in D and in D+ .
(ii) The conjugate function operator on T (see Propsoition 6.2 in Chapter I) and
the Hilbert transform on R (see Proposition 4.3 in Chapter II) are bounded
on Lp for 1 < p < ∞.
(iii) Under the sesquilinear map B in Proposition 8.3 in Chapter I, the dual of
0
H p (D) is identified with H p (D) for 1 < p < ∞. The same can be verified
on D+ .