}

Introduction to Harmonic Analysis

} LECTURE NOTES

Chengchun Hao

Institute of Mathematics, AMSS, CAS

Updated: April 19, 2016
Contents

1 The Fourier Transform and Tempered Distributions . . . . . . . . . . . 1

1.1 The L1 theory of the Fourier transform . . . . . . . . . . . . . . . . . . . . . 1
1.2 The L2 theory and the Plancherel theorem . . . . . . . . . . . . . . . . . . 15
1.3 Schwartz spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4 The class of tempered distributions . . . . . . . . . . . . . . . . . . . . . . . . 21
1.5 Characterization of operators commuting with translations . . . . . 26

2 Interpolation of Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.1 Riesz-Thorin’s and Stein’s interpolation theorems . . . . . . . . . . . . 33
2.2 The distribution function and weak Lp spaces . . . . . . . . . . . . . . . 41
2.3 The decreasing rearrangement and Lorentz spaces . . . . . . . . . . . . 45
2.4 Marcinkiewicz’ interpolation theorem . . . . . . . . . . . . . . . . . . . . . . 51

3 The Maximal Function and Calderón-Zygmund Decomposition . 57

3.1 Two covering lemmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.2 Hardy-Littlewood maximal function . . . . . . . . . . . . . . . . . . . . . . . 59
3.3 Calderón-Zygmund decomposition . . . . . . . . . . . . . . . . . . . . . . . . 70

4 Singular Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.1 Harmonic functions and Poisson equation . . . . . . . . . . . . . . . . . . . 77
4.2 Poisson kernel and Hilbert transform . . . . . . . . . . . . . . . . . . . . . . . 82
4.3 The Calderón-Zygmund theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.4 Truncated integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.5 Singular integral operators commuted with dilations . . . . . . . . . . 101
4.6 The maximal singular integral operator . . . . . . . . . . . . . . . . . . . . . 107
4.7 *Vector-valued analogues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

5 Riesz Transforms and Spherical Harmonics . . . . . . . . . . . . . . . . . . . 115

5.1 The Riesz transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.2 Spherical harmonics and higher Riesz transforms . . . . . . . . . . . . 120
v
- vi - Contents

5.3 Equivalence between two classes of transforms . . . . . . . . . . . . . . 129

6 The Littlewood-Paley g-function and Multipliers . . . . . . . . . . . . . . 133

6.1 The Littlewood-Paley g-function . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.2 Fourier multipliers on Lp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
6.3 The partial sums operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
6.4 The dyadic decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
6.5 The Marcinkiewicz multiplier theorem . . . . . . . . . . . . . . . . . . . . . 164

7 Sobolev Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

7.1 Riesz potentials and fractional integrals . . . . . . . . . . . . . . . . . . . . . 169
7.2 Bessel potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.3 Sobolev spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
L ECTURE N OTES ON c 2016 by Chengchun Hao
Introduction to Harmonic Analysis Email: hcc@amss.ac.cn

Chapter 1
The Fourier Transform and Tempered
Distributions

In this chapter, we introduce the Fourier transform and study its more el-
ementary properties, and extend the definition to the space of tempered dis-
tributions. We also give some characterizations of operators commuting with
translations.

1.1 The L1 theory of the Fourier transform

We begin by introducing some notation that will be used throughout this

work. Rn denotes n-dimensional real Euclidean space. We consistently write
x = (x1 , x2 , · · · , xn ), ξ = (ξ1 , ξ2 , · · · , ξn ), · · · for the elements of Rn . The
inner product of x, ξ ∈ Rn is the number x·ξ = nj=1 xj ξj , the norm of x ∈ Rn
P
√
is the nonnegative number |x| = x · x. Furthermore, dx = dx1 dx2 · · · dxn
denotes the element of ordinary Lebesgue measure.
We will deal with various spaces of functions defined on Rn . The simplest of
these are the Lp = Lp (Rn ) spaces, 1 6 p < ∞, of all measurable functions f

1/p
such that kf kp = Rn |f (x)|p dx < ∞. The number kf kp is called the Lp
R

norm of f . The space L∞ (Rn ) consists of all essentially bounded functions on

Rn and, for f ∈ L∞ (Rn ), we let kf k∞ be the essential supremum of |f (x)|,
x ∈ Rn . Often, the space C0 (Rn ) of all continuous functions vanishing at
infinity, with the L∞ norm just described, arises more naturally than L∞ =
L∞ (Rn ). Unless otherwise specified, all functions are assumed to be complex
valued; it will be assumed, throughout the note, that all functions are (Borel)
measurable.
In addition to the vector-space operations, L1 (Rn ) is endowed with a “mul-
tiplication” making this space a Banach algebra. This operation, called convo-
lution, is defined in the following way: If both f and g belong to L1 (Rn ), then
their convolution h = f ∗ g is the function whose value at x ∈ Rn is

1
-2- 1. The Fourier Transform and Tempered Distributions
Z
h(x) = f (x − y)g(y)dy.
Rn
One can show by an elementary argument that f (x − y)g(y) is a measurable
function of the two variables x and y. It then follows immediately from Fib-
ini’s theorem on the interchange of the order of integration that h ∈ L1 (Rn )
and khk1 6 kf k1 kgk1 . Furthermore, this operation is commutative and as-
sociative. More generally, weRhave, with the help of Minkowski’s integral in-
equality k F (x, y)dykLpx 6 kF (x, y)kLpx dy, the following result:
R

Theorem 1.1. If f ∈ Lp (Rn ), p ∈ [1, ∞], and g ∈ L1 (Rn ) then h = f ∗ g is

well defined and belongs to Lp (Rn ). Moreover,
khkp 6 kf kp kgk1 .
Now, we first consider the Fourier1 transform of L1 functions.
Definition 1.2. Let ω ∈ R \ {0} be a constant. If f ∈ L1 (Rn ), then its
Fourier transform is F f or fˆ : Rn → C defined by
Z
F f (ξ) = e−ωix·ξ f (x)dx (1.1)
Rn
n
for all ξ ∈ R .
We now continue with some properties of the Fourier transform. Before
doing this, we shall introduce some notations. For a measurable function f on
Rn , x ∈ Rn and a 6= 0 we define the translation and dilation of f by
τy f (x) =f (x − y), (1.2)
δa f (x) =f (ax). (1.3)
Proposition 1.3. Given f, g ∈ L1 (Rn ), x, y, ξ ∈ Rn , α multiindex, a, b ∈ C,
ε ∈ R and ε 6= 0, we have
(i) Linearity: F (af + bg) = aF f + bF g.
(ii) Translation: F τy f (ξ) = e−ωiy·ξ fˆ(ξ).
(iii) Modulation: F (eωix·y f (x))(ξ) = τy fˆ(ξ).
(iv) Scaling: F δε f (ξ) = |ε|−n δε−1 fˆ(ξ).
(v) Differentiation: F ∂ α f (ξ) = (ωiξ)α fˆ(ξ), ∂ α fˆ(ξ) =
α
F ((−ωix) f (x))(ξ).
(vi) Convolution: F (f ∗ g)(ξ) = fˆ(ξ)ĝ(ξ).
(vii) Transformation: F (f ◦ A)(ξ) = fˆ(Aξ), where A is an orthogonal
matrix and ξ is a column vector.
(viii) Conjugation: f (x) = fˆ(−ξ).
d

1
Jean Baptiste Joseph Fourier (21 March 1768 – 16 May 1830) was a French mathematician and physi-
cist best known for initiating the investigation of Fourier series and their applications to problems of
heat transfer and vibrations. The Fourier transform and Fourier’s Law are also named in his honor.
Fourier is also generally credited with the discovery of the greenhouse effect.
1.1. The L1 theory of the Fourier transform -3-

Proof. These results are easy to be verified. We only prove (vii). In fact,
Z Z
−ωix·ξ −1
F (f ◦ A)(ξ) = e f (Ax)dx = e−ωiA y·ξ f (y)dy
n n
ZR ZR
>
= e−ωiA y·ξ f (y)dy = e−ωiy·Aξ f (y)dy = fˆ(Aξ),
Rn Rn
where we used the change of variables y = Ax and the fact that A−1 = A>
and | det A| = 1. 
Corollary 1.4. The Fourier transform of a radial function is radial.
Proof. Let ξ, η ∈ Rn with |ξ| = |η|. Then there exists some orthogonal matrix
A such that Aξ = η. Since f is radial, we have f = f ◦ A. Then, it holds
F f (η) = F f (Aξ) = F (f ◦ A)(ξ) = F f (ξ),
by (vii) in Proposition 1.3. 
It is easy to establish the following results:
Theorem 1.5 (Uniform continuity). (i) The mapping F is a bounded linear
transformation from L1 (Rn ) into L∞ (Rn ). In fact, kF f k∞ 6 kf k1 .
(ii) If f ∈ L1 (Rn ), then F f is uniformly continuous.
Proof. (i) is obvious. We now prove (ii). By
Z
ˆ ˆ
f (ξ + h) − f (ξ) = e−ωix·ξ [e−ωix·h − 1]f (x)dx,
Rn
we have Z
|fˆ(ξ + h) − fˆ(ξ)| 6 |e−ωix·h − 1||f (x)|dx
n
ZR Z
−ωix·h
6 |e − 1||f (x)|dx + 2 |f (x)|dx
|x|6r |x|>r
Z Z
6 |ω|r|h||f (x)|dx + 2 |f (x)|dx
|x|6r |x|>r

=:I1 + I2 ,
since for any θ > 0
iθ
q √
|e − 1| = (cos θ − 1)2 + sin2 θ = 2 − 2 cos θ = 2| sin(θ/2)| 6 |θ|.
Given any ε > 0, we can take r so large that I2 < ε/2. Then, we fix this r
and take |h| small enough such that I1 < ε/2. In other words, for given ε > 0,
there exists a sufficiently small δ > 0 such that |fˆ(ξ + h) − fˆ(ξ)| < ε when
|h| 6 δ, where ε is independent of ξ. 
Ex. 1.6. Suppose that a signal consists of a single rectangular pulse of
width 1 and height 1. Let’s say that it gets turned on at x = − 21 and
turned off at x = 12 . The standard name for this “normalized” rectan-
gular pulse is
-4- 1. The Fourier Transform and Tempered Distributions

1
1, if − 21 < x < 12 ,

Π(x) ≡ rect(x) :=
0, otherwise. − 21 1 x
2
It is also called, variously, the normalized boxcar function, the top hat
function, the indicator function, or the characteristic function for the
interval (−1/2, 1/2). The Fourier transform of this signal is
Z 1/2 1/2
e−ωixξ 2 ωξ
Z
−ωixξ
Π(ξ)
b = e Π(x)dx = e−ωixξ dx = = sin
R −1/2 −ωiξ −1/2 ωξ 2
R 1/2
when ξ 6= 0. When ξ = 0, Π(0)b = −1/2 dx = 1. By l’Hôpital’s rule,
sin ωξ
2
ω
cos ωξ
lim Π(ξ)
b = lim 2 = lim 2 2 2
= 1 = Π(0),
b
ξ→0 ξ→0 ωξ ξ→0 ω
so Π(ξ)
b is continuous at ξ = 0. There is a standard function called
“sinc” that is defined by sinc(ξ) = sinξ ξ . In this notation Π(ξ)
2 b = sinc ωξ
2
.
Here is the graph of Π(ξ).
b

− 2π
ω
2π
ω
ξ

Remark 1.7. The above definition of the Fourier transform in (1.1) ex-
tends immediately to finite Borel measures: if µ is such a measure on
Rn , we define F µ by letting
Z
F µ(ξ) = e−ωix·ξ dµ(x).
Rn
Theorem 1.5 is valid for this Fourier transform if we replace the L1
norm by the total variation of µ.
The following theorem plays a central role in Fourier Analysis. It takes its
name from the fact that it holds even for functions that are integrable according
to the definition of Lebesgue. We prove it for functions that are absolutely
integrable in the Riemann sense.3

2
The term “sinc” (English pronunciation:["sINk]) is a contraction, first introduced by Phillip M.
Woodward in 1953, of the function’s full Latin name, the sinus cardinalis (cardinal sine).
3
Let us very briefly recall what this means. A bounded function f on a finite interval [a, b] is
integrable if it can be approximated by Riemann sums from above and below in such a way that the
difference of the integrals of these sums can be made as small as we wish. This definition is then
extended to unbounded functions and infinite intervals by taking limits; these cases are often called
improperR integrals. If I is any interval and f is a function on I such that the (possibly improper)
integral I |f (x)|dx has a finite value, then f is said to be absolutely integrable on I.
1.1. The L1 theory of the Fourier transform -5-

Theorem 1.8 (Riemann-Lebesgue lemma). If f ∈ L1 (Rn ) then F f → 0

as |ξ| → ∞; thus, in view of the last result, we can conclude that F f ∈
C0 (Rn ).
The Riemann-Lebesgue lemma states that the integral of a function like the
left is small. The integral will approach zero as the number of oscillations
increases.

Proof. First, for n = 1, suppose that f (x) = χ(a,b) (x), the characteristic func-
tion of an interval. Then
Z b
ˆ e−ωiaξ − e−ωibξ
f (ξ) = e−ωixξ dx = → 0, as |ξ| → ∞.
a ωiξ
Similarly, the result holds when f is the characteristic function of the n-
dimensional rectangle I = {x ∈ Rn : a1 6 x1 6 b1 , · · · , an 6 xn 6 bn }
since we can calculate F f explicitly as an iterated integral. The same is there-
fore true for a finite linear combination of such characteristic functions (i.e.,
simple functions). Since all such simple functions are dense in L1 , the result
for a general f ∈ L1 (Rn ) follows easily by approximating f in the L1 norm by
such a simple function g, then f = g + (f − g), where F f − F g is uniformly
small by Theorem 1.5, while F g(ξ) → 0 as |ξ| → ∞. 
Theorem 1.8 gives a necessary condition for a function to be a Fourier trans-
form. However, that belonging to C0 is not a sufficient condition for being the
Fourier transform of an integrable function. See the following example.
Ex. 1.9. Suppose, for simplicity, that n = 1. Let
1


 , ξ > e,
ln ξ

g(ξ) =
 ξ
 ,
 0 6 ξ 6 e,
e
g(ξ) = − g(−ξ), ξ < 0.
It is clear that g(ξ) is uniformly continuous on R and g(ξ) → 0 as |ξ| →
∞.
Assume that there exists an f ∈ L1 (R) such that fˆ(ξ) = g(ξ), i.e.,
Z ∞
g(ξ) = e−ωixξ f (x)dx.
−∞
Since g(ξ) is an odd function, we have
Z ∞ Z ∞ Z ∞
−ωixξ
g(ξ) = e f (x)dx = −i sin(ωxξ)f (x)dx = sin(ωxξ)F (x)dx,
−∞ −∞ 0

where F (x) = i[f (−x) − f (x)] ∈ L1 (R). Integrating g(ξ)

ξ
over (0, N )
yields
Z N Z ∞ Z N  Z ∞ Z ωxN 
g(ξ) sin(ωxξ) sin t
dξ = F (x) dξ dx = F (x) dt dx.
0 ξ 0 0 ξ 0 0 t
Noticing that
-6- 1. The Fourier Transform and Tempered Distributions

N
sin t π
Z
lim dt = ,
N →∞ 0 t 2
and by Lebesgue dominated convergence theorem,we get that the inte-
gral of r.h.s. is convergent as N → ∞. That is,
π ∞
Z N
g(ξ)
Z
lim dξ = F (x)dx < ∞,
N →∞ 0 ξ 2 0
R∞ Re
which yields e g(ξ) ξ
dξ < ∞ since 0 g(ξ) ξ
dξ = 1. However,
Z N Z N
g(ξ) dξ
lim dξ = lim = ∞.
N →∞ e ξ N →∞ e ξ ln ξ
This contradiction indicates that the assumption was invalid.
We now turn to the problem of inverting the Fourier transform. That is,
we shall consider the question: Given the Fourier transform fˆ of an integrable
function f , how do we obtain f back again from fˆ ? The reader, who is familiar
with the elementary theory of Fourier series and integrals, would expect f (x)
to be equal to the integral
Z
C eωix·ξ fˆ(ξ)dξ. (1.4)
Rn
Unfortunately, fˆ need not be integrable (for example, let n = 1 and f be the
characteristic function of a finite interval). In order to get around this difficulty,
we shall use certain summability methods for integrals. We first introduce the
Abel method of summability, whose analog for series is very well-known. For
each ε > 0, we define the Abel mean Aε = Aε (f ) to be the integral
Z
Aε (f ) = Aε = e−ε|x| f (x)dx. (1.5)
Rn

It is clear that if f ∈ L1 (Rn ) then lim Aε (f ) = Rn f (x)dx. On the other

R
ε→0
hand, these Abel means are well-defined even when f is not integrable (e.g.,
if we only assume that f is bounded, then Aε (f ) is defined for all ε > 0).
Moreover, their limit Z
lim Aε (f ) = lim e−ε|x| f (x)dx (1.6)
ε→0 ε→0 Rn
may exist even when f is not integrable. A classical example of such a case is
obtained by letting f (x) = sinc(x) when n = 1. Whenever the limit in (1.6)
exists and is finite we say that Rn f dx is Abel summable to this limit.
R

A somewhat similar method of summability is Gauss summability. This

method is defined by the Gauss (sometimes called Gauss-Weierstrass) means
Z
2
Gε (f ) = e−ε|x| f (x)dx. (1.7)
Rn

We say that Rn f dx is Gauss summable (to l) if

R
1.1. The L1 theory of the Fourier transform -7-
Z
2
lim Gε (f ) = lim e−ε|x| f (x)dx (1.6’)
ε→0 ε→0 Rn
exists and equals the number `.
We see that both (1.6) and (1.6’) can be put in the form
Z
Mε,Φ (f ) = Mε (f ) = Φ(εx)f (x)dx, (1.8)
RnR
where Φ ∈ C0 and Φ(0) = 1. Then Rn f (x)dx is summable to ` if
limε→0 Mε (f ) = `. We shall call Mε (f ) the Φ means of this integral.
2
We shall need the Fourier transforms of the functions e−ε|x| and e−ε|x| . The
first one is easy to calculate.
Theorem 1.10. For all a > 0, we have
 −n
−a|ωx|2 |ω| |ξ|2
Fe (ξ) = (4πa)−n/2 e− 4a . (1.9)
2π
Proof. The integral in question is
Z
2
e−ωix·ξ e−a|ωx| dx.
Rn
Notice that this factors as a product of one variable integrals. Thus it is suffi-
cient to prove
R the case n = 1. For this we use the formula for the integral of a
−πx2
Gaussian: R e dx = 1. It follows that
Z ∞ Z ∞
ξ2
−ωixξ −aω 2 x2 2
e e dx = e−a(ωx+iξ/(2a)) e− 4a dx
−∞ −∞
2
Z ∞+iξ/(2a)
−1 − ξ4a 2
=|ω| e e−ax dx
−∞+iξ/(2a)
Z ∞
p ξ2 2
−1 − 4a
=|ω| e π/a e−πy dy
−∞
−1
|ω|

ξ2
= (4πa)−1/2 e− 4a ,
2π
where we used contour integration at the next to last one. 
The second one is somewhat harder to obtain:
Theorem 1.11. For all a > 0, we have
 −n
−a|ωx| |ω| cn a Γ ((n + 1)/2)
F (e )= , cn = . (1.10)
2π (a2 + |ξ|2 )(n+1)/2 π (n+1)/2
Proof. By a change of variables, i.e.,
Z Z
−a|ωx| −ωix·ξ −a|ωx| −n
F (e )= e e dx = (a|ω|) e−ix·ξ/a e−|x| dx,
Rn Rn
we see that it suffices to show this result when a = 1. In order to show this, we
need to express the decaying exponential as a superposition of Gaussians, i.e.,
-8- 1. The Fourier Transform and Tempered Distributions
Z ∞ −η
−γ 1 e 2
e =√ √ e−γ /4η dη, γ > 0. (1.11)
π 0 η
Then, using (1.9) to establish the third equality,
 Z ∞ −η 
1 e
Z Z
−ix·t −|x| −ix·t −|x|2 /4η
e e dx = e √ √ e dη dx
Rn Rn π 0 η
Z ∞ −η Z 
1 e −ix·t −|x|2 /4η
=√ √ e e dx dη
π 0 η Rn
Z ∞ −η 
1 e 2

=√ √ (4πη)n/2 e−η|t| dη
π 0 η
Z ∞
2 n−1
=2n π (n−1)/2 e−η(1+|t| ) η 2 dη
0

n+1 ∞ −ζ n+1 −1
Z
n (n−1)/2 2 − 2
=2 π 1 + |t| e ζ 2 dζ
0
 
n (n−1)/2 n+1 1
=2 π Γ .
2 (1 + |t|2 )(n+1)/2
Thus,
 −n
−a|ωx| (a|ω|)−n (2π)n cn |ω| cn a
F (e )= 2 (n+1)/2
= .
(1 + |ξ/a| ) 2π (a + |ξ|2 )(n+1)/2
2

Consequently, the theorem will be established once we show (1.11). In fact,

by changes of variables, we have
1 γ ∞ e−η −γ 2 /4η
Z
√ e √ e dη
π 0 η
√ Z
2 γ ∞ −γ(σ− 1 )2
=√ e 2σ dσ (by η = γσ 2 )
π 0
√ Z
2 γ ∞ −γ(σ− 1 )2 1 1
=√ e 2σ
2
dσ (by σ 7→ )
π 0 2σ 2σ
√ Z ∞
γ
 
−γ(σ− 2σ 1 2 1
=√ e )
1 + 2 dσ (by averaging the last two formula)
π 0 2σ
√ Z ∞
γ 2 1
=√ e−γu du (by u = σ − )
π −∞ 2σ
Z
2
=1, (by e−πx dx = 1)
R
which yields the desired identity (1.11). 
 n  n
|ω| 2 |ω|
We shall denote the Fourier transform of 2π
e−a|ωx| and 2π
e−a|ωx| ,
a > 0, by W and P , respectively. That is,
|ξ|2 cn a
W (ξ, a) = (4πa)−n/2 e− 4a , P (ξ, a) = . (1.12)
(a2 + |ξ|2 )(n+1)/2
1.1. The L1 theory of the Fourier transform -9-

The first of these two functions is called the Weierstrass (or Gauss-
Weierstrass) kernel while the second is called the Poisson kernel.
Theorem 1.12 (The multiplication formula). If f, g ∈ L1 (Rn ), then
Z Z
fˆ(ξ)g(ξ)dξ = f (x)ĝ(x)dx.
Rn Rn

Proof. Using Fubini’s theorem to interchange the order of the integration on

R2n , we obtain the identity. 

Theorem 1.13. If f and Φ belong to L1 (Rn ), ϕ = Φ̂ and ϕε (x) = ε−n ϕ(x/ε),

then Z Z
e ωix·ξ ˆ
Φ(εξ)f (ξ)dξ = ϕε (y − x)f (y)dy
Rn Rn
for all ε > 0. In particular,
 n Z
|ω|
Z
ωix·ξ −ε|ωξ| ˆ
e e f (ξ)dξ = P (y − x, ε)f (y)dy,
2π Rn Rn
and  n Z
|ω|
Z
ωix·ξ −ε|ωξ|2 ˆ
e e f (ξ)dξ = W (y − x, ε)f (y)dy.
2π Rn Rn

Proof. From (iii) and (iv) in Proposition 1.3, it implies (F eωix·ξ Φ(εξ))(y) =
ϕε (y − x). The first result holds immediately with the help of Theorem 1.12.
The last two follow from (1.9), (1.10) and (1.12). 
R
Lemma R 1.14. (i) Rn
W (x, ε)dx = 1 for all ε > 0.
(ii) Rn P (x, ε)dx = 1 for all ε > 0.

Proof. By a change of variable, we first note that

Z Z 2
Z
−n/2 − |x|
W (x, ε)dx = (4πε) e 4ε dx = W (x, 1)dx,
Rn Rn Rn
and
cn ε
Z Z Z
P (x, ε)dx = 2 2 (n+1)/2
dx = P (x, 1)dx.
Rn Rn (ε + |x| ) Rn
Thus, it suffices to prove the lemma when ε = 1. For the first one, we
use a change of variables and the formula for the integral of a Gaussian:
2
dx = 1 to get
R −πx
R
e
Z Z 2
Z
−n/2 − |x| 2
W (x, 1)dx = (4π) e 4 dx = (4π)−n/2 e−π|y| 2n π n/2 dy = 1.
Rn Rn Rn
For the second one,
Z we have
1
Z
P (x, 1)dx = cn dx.
Rn Rn (1 + |x|2 )(n+1)/2
- 10 - 1. The Fourier Transform and Tempered Distributions

Letting r = |x|, x0 = x/r (when x 6= 0), S n−1 = {x ∈ Rn : |x| = 1}, dx0 the
element of surface area on S n−1 whose surface area4 is denoted by ωn−1 and,
finally, putting r = tan θ, we have
Z ∞Z
1 1
Z
2 (n+1)/2
dx = 2 (n+1)/2
dx0 rn−1 dr
Rn (1 + |x| ) 0 n−1 (1 + r )
ZS ∞
rn−1
=ωn−1 dr
0 (1 + r2 )(n+1)/2
Z π/2
=ωn−1 sinn−1 θdθ.
0
But ωn−1 sin θ is clearly the surface area of the
n−1 x
n
n+1

S
sphere of radius sin θ obtained by intersecting S n

with the hyperplane x1 = cos θ. Thus, the area

sin θ
of the upper half of S n is obtained by summing O θ
cos θ
1
x 1

these (n − 1) dimensional areas as θ ranges from S n−1

0 to π/2, that is,

Z π/2
ωn
ωn−1 sinn−1 θdθ = ,
0 2
which is the desired result by noting that 1/cn = ωn /2. 

Theorem 1.15. Suppose ϕ ∈ L1 (Rn ) with Rn ϕ(x)dx = 1 and let ϕε (x) =

R

ε−n ϕ(x/ε) for ε > 0. If f ∈ Lp (Rn ), 1 6 p < ∞, or f ∈ C0 (Rn ) ⊂ L∞ (Rn ),

then for 1 6 p 6 ∞
kf ∗ ϕε − f kp → 0, as ε → 0.
In particular, the Poisson integral of f :
Z
u(x, ε) = P (x − y, ε)f (y)dy
Rn
and the Gauss-Weierstrass integral of f :
Z
s(x, ε) = W (x − y, ε)f (y)dy
Rn
converge to f in the Lp norm as ε → 0.

Proof. By a change of variables, we have

Z Z Z
−n
ϕε (y)dy = ε ϕ(y/ε)dy = ϕ(y)dy = 1.
Rn Rn Rn
Hence, Z
(f ∗ ϕε )(x) − f (x) = [f (x − y) − f (x)]ϕε (y)dy.
Rn
Therefore, by Minkowski’s inequality for integrals and a change of variables,
we get
4
ωn−1 = 2π n/2 /Γ (n/2).
1.1. The L1 theory of the Fourier transform - 11 -
Z
kf ∗ ϕε − f kp 6 kf (x − y) − f (x)kp ε−n |ϕ(y/ε)|dy
ZRn

= kf (x − εy) − f (x)kp |ϕ(y)|dy.

Rn
We point out that if f ∈ L (Rn ), 1 6 p < ∞, and denote kf (x − t) −
p

f (x)kp = ∆f (t), then ∆f (t) → 0, as t → 0.5 In fact, if f1 ∈ D(Rn ) :=

C0∞ (Rn ) of all C ∞ functions with compact support, the assertion in that case
is an immediate consequence of the uniform convergence f1 (x − t) → f1 (x),
as t → 0. In general, for any σ > 0, we can write f = f1 + f2 , such that f1 is
as described and kf2 kp 6 σ, since D(Rn ) is dense in Lp (Rn ) for 1 6 p < ∞.
Then, ∆f (t) 6 ∆f1 (t) + ∆f2 (t), with ∆f1 (t) → 0 as t → 0, and ∆f2 (t) 6 2σ.
This shows that ∆f (t) → 0 as t → 0 for general f ∈ Lp (Rn ), 1 6 p < ∞.
For the case p = ∞ and f ∈ C0 (Rn ), the same argument gives us the result
since D(Rn ) is dense in C0 (Rn ) (cf. [Rud87, p.70, Proof of Theorem 3.17]).
Thus, by the Lebesgue dominated convergence theorem (due to ϕ ∈ L1 and
the fact ∆f (εy)|ϕ(y)| 6 2kf kp |ϕ(y)|) and the fact ∆f (εy) → 0 as ε → 0, we
have Z Z
lim kf ∗ ϕε − f kp 6 lim ∆f (εy)|ϕ(y)|dy = lim ∆f (εy)|ϕ(y)|dy = 0.
ε→0 ε→0 Rn Rn ε→0
This completes the proof. 
With the same argument, we have
Corollary 1.16. Let 1 6 p 6 ∞. Suppose ϕ ∈ L1 (Rn ) and Rn ϕ(x)dx = 0,
R

then kf ∗ ϕε kp → 0 as ε → 0 whenever f ∈ Lp (Rn ), 1 6 p < ∞, or

f ∈ C0 (Rn ) ⊂ L∞ (Rn ).
Proof. Once we observe that
Z
(f ∗ ϕε )(x) =(f ∗ ϕε )(x) − f (x) · 0 = (f ∗ ϕε )(x) − f (x) ϕε (y)dy
Z Rn

= [f (x − y) − f (x)]ϕε (y)dy,
Rn
the rest of the argument is precisely that used in the last proof. 
In particular, we also have
Corollary 1.17. Suppose ϕ ∈ L1 (Rn ) with Rn ϕ(x)dx = 1 and let ϕε (x) =
R

ε−n ϕ(x/ε) for ε > 0. Let f (x) ∈ L∞ (Rn ) be continuous at {0}. Then,
Z
lim f (x)ϕε (x)dx = f (0).
ε→0 Rn

Proof. Since Rn f (x)ϕε (x)dx − f (0) = Rn (f (x) − f (0))ϕε (x)dx, then we

R R

may assume without loss of generality that f (0) = 0. Since f is continuous at

{0}, then for any η > 0, there exists a δ > 0 such that
5
This statement is the continuity of the mapping t → f (x − t) of Rn to Lp (Rn ).
- 12 - 1. The Fourier Transform and Tempered Distributions

η
|f (x)| < ,
kϕk1
whenever |x| < δ. Noticing that | Rn ϕ(x)dx| 6 kϕk1 , we have
R

η
Z Z Z
f (x)ϕε (x)dx 6 |ϕε (x)|dx + kf k∞ |ϕε (x)|dx
Rn kϕk1 |x|<δ |x|>δ
η
Z
6 kϕk1 + kf k∞ |ϕ(y)|dy
kϕk1 |y|>δ/ε

=η + kf k∞ Iε .
But Iε → 0 as ε → 0. This proves the result. 
From Theorems 1.13 and 1.15, we obtain the following solution to the
Fourier inversion problem:

RTheorem 1.18. If both Φ and its Fourier transform ϕ = Φ̂ are integrable and
Rn
ϕ(x)dx = 1, then the Φ means of the integral (|ω|/2π) Rn eωix·ξ fˆ(ξ)dξ
nR

converges to f (x) in the L1 norm. In particular, the Abel and Gauss means of
this integral converge to f (x) in the L1 norm.
We have singled out the Gauss-Weierstrass and the Abel methods of summa-
bility. The former is probably the simplest and is connected with the solution
of the heat equation; the latter is intimately connected with harmonic functions
and provides us with veryn powerful tools in Fourier analysis.
|ω| ωix·ξ −ε|ωξ|2 ˆ
Since s(x, ε) = 2π f (ξ)dξ converges in L1 to f (x) as
R
Rn
e e
ε > 0 tends to 0, we can find a sequence εk → 0 such that s(x, εk ) → f (x)
for a.e. x. If we further assume that fˆ ∈ L1 (Rn ), the Lebesgue dominated
convergence theorem gives us the following pointwise equality:
Theorem 1.19 (Fourier inversion theorem). If both f and fˆ are integrable,
then  n Z
|ω|
f (x) = eωix·ξ fˆ(ξ)dξ,
2π Rn

for almost every x.

Remark 1.20. We know R from Theorem 1.5 that fˆ is continuous. If fˆ is in-
ωix·ξ ˆ
tegrable, the integral Rn e f (ξ)dξ also defines a continuous function
ˆˆ
(in fact, it equals f (−x)). Thus, by changing f on a set of measure 0, we
can obtain equality in Theorem 1.19 for all x.
It is clear from Theorem 1.18 that if fˆ(ξ) = 0 for all ξ then f (x) = 0
for almost every x. Applying this to f = f1 − f2 , we obtain the following
uniqueness result for the Fourier transform:
Corollary 1.21 (Uniqueness). If f1 and f2 belong to L1 (Rn ) and fˆ1 (ξ) =
fˆ2 (ξ) for ξ ∈ Rn , then f1 (x) = f2 (x) for almost every x ∈ Rn .
1.1. The L1 theory of the Fourier transform - 13 -

We will denote the inverse operation to the Fourier transform by F −1 or ˇ·.

If f ∈ L1 , then we have
 n Z
ˇ |ω|
f (x) = eωix·ξ f (ξ)dξ. (1.13)
2π Rn

We give a very useful result.

Theorem 1.22. Suppose f ∈ L1 (Rn ) and fˆ > 0. If f is continuous at 0, then
 n Z
|ω|
f (0) = fˆ(ξ)dξ.
2π Rn

ˆ 1 n
Moreover, we have f ∈ L (R ) and
 n Z
|ω|
f (x) = eωix·ξ fˆ(ξ)dξ,
2π Rn

for almost every x.

Proof. By Theorem 1.13, we have
 n Z
|ω|
Z
−ε|ωξ| ˆ
e f (ξ)dξ = P (y, ε)f (y)dy.
2π Rn Rn
From Lemma 1.14, we get, for any δ > 0,
Z Z
P (y, ε)f (y)dy − f (0) = P (y, ε)[f (y) − f (0)]dy
Rn Rn
Z Z
6 P (y, ε)[f (y) − f (0)]dy + P (y, ε)[f (y) − f (0)]dy
|y|<δ |y|>δ

=I1 + I2 .
Since f is continuous at 0, for any given σ > 0, we can choose δ small enough
such that |f (y) − f (0)| 6 σ when |y| < δ. Thus, I1 6 σ by Lemma 1.14. For
the second term, we have, by a change of variables, that
Z
I2 6kf k1 sup P (y, ε) + |f (0)| P (y, ε)dy
|y|>δ |y|>δ
cn ε
Z
=kf k1 2 + |f (0)| P (y, 1)dy → 0,
(ε + δ 2 )(n+1)/2 |y|>δ/ε
 n R
as ε → 0. Thus, |ω| 2π Rn
e−ε|ωξ| fˆ(ξ)dξ → f (0) as ε → 0. On the other
hand, by Lebesgue dominated convergence theorem, we obtain
 n Z  n
|ω| |ω|
Z
ˆ
f (ξ)dξ = lim e−ε|ωξ| fˆ(ξ)dξ = f (0),
2π R n 2π ε→0 Rn

ˆ ˆ
which implies f ∈ L (R ) due to f > 0. Therefore, from Theorem 1.19, it
1 n

follows the desired result. 

An immediate consequence is
ωix·ξ −ε|ωx|2
R
Corollary 1.23. i) Rn e W (ξ, ε)dξ = e .
ωix·ξ −ε|ωx|
R
ii) Rn e P (ξ, ε)dξ = e .
- 14 - 1. The Fourier Transform and Tempered Distributions

Proof. Noticing that

 n  n
|ω| |ω|
 
−ε|ωx|2 −ε|ωx|
W (ξ, ε) = F e , and P (ξ, ε) = F e ,
2π 2π
we have the desired results by Theorem 1.22. 
We also have the semigroup properties of the Weierstrass and Poisson ker-
nels.
Corollary 1.24. If α1 and
R α2 are positive real numbers, then
i) W (ξ, α1 + α2 ) = RRn W (ξ − η, α1 )W (η, α2 )dη.
ii) P (ξ, α1 + α2 ) = Rn P (ξ − η, α1 )P (η, α2 )dη.

Proof. It follows, from Corollary 1.23, that

 n
|ω| 2
W (ξ, α1 + α2 ) = (F e−(α1 +α2 )|ωx| )(ξ)
2π
 n
|ω| 2 2
= F (e−α1 |ωx| e−α2 |ωx| )(ξ)
2π
 n 
|ω|
Z 
−α1 |ωx|2 ωix·η
= F e e W (η, α2 )dη (ξ)
2π Rn
 n Z
|ω|
Z
−ωix·ξ −α1 |ωx|2
= e e eωix·η W (η, α2 )dηdx
2π n n
Z Z R  n R
|ω|

−ωix·(ξ−η) −α1 |ωx|2
= e e dx W (η, α2 )dη
n Rn 2π
ZR
= W (ξ − η, α1 )W (η, α2 )dη.
Rn
A similar argument can give the other equality. 
Finally, we give an example of the semigroup about the heat equation.
Ex. 1.25. Consider the Cauchy problem to the heat equation
ut − ∆u = 0, u(0) = u0 (x), t > 0, x ∈ Rn .
Taking the Fourier transform, we have
ût + |ωξ|2 û = 0, û(0) = û0 (ξ).
Thus, it follows, from Theorem 1.10, that
2 2 2 /4t
u =F −1 e−|ωξ| t F u0 = (F −1 e−|ωξ| t ) ∗ u0 = (4πt)−n/2 e−|x| ∗ u0
=W (x, t) ∗ u0 =: H(t)u0 .
Then, we obtain
H(t1 + t2 )u0 =W (x, t1 + t2 ) ∗ u0 = W (x, t1 ) ∗ W (x, t2 ) ∗ u0
=W (x, t1 ) ∗ (W (x, t2 ) ∗ u0 ) = W (x, t1 ) ∗ H(t2 )u0
=H(t1 )H(t2 )u0 ,
i.e., H(t1 + t2 ) = H(t1 )H(t2 ).
1.2. The L2 theory and the Plancherel theorem - 15 -

1.2 The L2 theory and the Plancherel theorem

The integral defining the Fourier transform is not defined in the Lebesgue
sense for the general function in L2 (Rn ); nevertheless, the Fourier transform
has a natural definition on this space and a particularly elegant theory.
If, in addition to being integrable, we assume f to be square-integrable then
fˆ will also be square-integrable. In fact, we have the following basic result:
Theorem 1.26 (Plancherel theorem). If f ∈ L1 (Rn )∩L2 (Rn ), then kfˆk2 =
 −n/2
|ω|
2π
kf k2 .

Proof. Let g(x) = f (−x). Then, by Theorem 1.1, h = f ∗ g ∈ L1 (Rn ) and,

by Proposition 1.3, ĥ = fˆĝ. But ĝ = fˆ,thusĥ = |fˆ|2 > 0. Applying Theorem
nR
1.22, we have ĥ ∈ L1 (Rn ) and h(0) = |ω| 2π Rn
ĥ(ξ)dξ. Thus, we get
 −n
|ω|
Z Z
ˆ 2
|f (ξ)| dξ = ĥ(ξ)dξ = h(0)
Rn Rn 2π
 −n Z
|ω|
= f (x)g(0 − x)dx
2π Rn
 −n Z  −n Z
|ω| |ω|
= f (x)f (x)dx = |f (x)|2 dx,
2π R n 2π Rn

which completes the proof. 

Since L1 ∩ L2 is dense in L2 , there exists a unique bounded extension, F ,
of this operator to all of L2 . F will be called the Fourier transform on L2 ; we
shall also use the notation fˆ = F f whenever f ∈ L2 (Rn ).
A linear operator on L2 (Rn ) that is an isometry and maps onto L2 (Rn ) is
called a unitary operator. It is an immediate consequence of Theorem 1.26
 n/2
that |ω|
2π
F is an isometry. Moreover, we have the additional property that
 n/2
|ω|
2π
F is onto:
 n/2
|ω|
Theorem 1.27. 2π
F is a unitary operator on L2 (Rn ).
 n/2
|ω|
Proof. Since 2π
F is an isometry, its range is a closed subspace of
L2 (Rn ). IfR this subspace were not all of L2 (Rn ), we could find a function g
such that Rn fˆgdx = 0 for all f ∈ L2 and kgk2 6= 0. Theorem 1.12 obvi-
ously extends to L2 ; consequently, Rn f ĝdx = Rn fˆgdx = 0 for all f ∈ L2 .
R R

But this implies that ĝ(x) = 0 for almost every x, contradicting the fact that
 −n/2
kĝk2 = |ω|2π
kgk2 6= 0. 
- 16 - 1. The Fourier Transform and Tempered Distributions

Theorem 1.27 is a major part of the basic theorem in the L2 theory of the
Fourier transform:
Theorem 1.28. The inverse of the Fourier transform, F −1 , can be obtained
by letting
 n
−1 |ω|
(F f )(x) = (F f )(−x)
2π
for all f ∈ L2 (Rn ).
We can also extend the definition of the Fourier transform to other spaces,
such as Schwartz space, tempered distributions and so on.

1.3 Schwartz spaces

Distributions (generalized functions) aroused mostly due to Paul Dirac and

his delta function δ. The Dirac delta gives a description of a point of unit mass
(placed at the origin). The mass density function is such that if its integrated on
a set not containing the origin it vanishes, but if the set does contain the origin
it is 1. No function (in the traditional sense) can have this property because
we know that the value of a function at a particular point does not change the
value of the integral.
In mathematical analysis, distributions are objects which generalize func-
tions and probability distributions. They extend the concept of derivative to all
integrable functions and beyond, and are used to formulate generalized solu-
tions of partial differential equations. They are important in physics and en-
gineering where many non-continuous problems naturally lead to differential
equations whose solutions are distributions, such as the Dirac delta distribu-
tion.
“Generalized functions” were introduced by Sergei Sobolev in 1935. They
were independently introduced in late 1940s by Laurent Schwartz, who devel-
oped a comprehensive theory of distributions.
The basic idea in the theory of distributions is to consider them as lin-
ear functionals on some space of “regular” functions — the so-called “test-
ing functions”. The space of testing functions is assumed to be well-behaved
with respect to the operations (differentiation, Fourier transform, convolution,
translation, etc.) we have been studying, and this is then reflected in the prop-
erties of distributions.
We are naturally led to the definition of such a space of testing functions by
the following considerations. Suppose we want these operations to be defined
on a function space, S , and to preserve it. Then, it would certainly have to
consist of functions that are indefinitely differentiable; this, in view of part (v)
1.3. Schwartz spaces - 17 -

in Proposition 1.3, indicates that each function in S , after being multiplied by

a polynomial, must still be in S . We therefore make the following definition:
Definition 1.29. The Schwartz space S (Rn ) of rapidly decaying func-
tions is defined
 as 
n ∞ n α β n
S (R ) = ϕ ∈ C (R ) : |ϕ|α,β := sup |x (∂ ϕ)(x)| < ∞, ∀α, β ∈ N0 ,
x∈Rn
(1.14)
where N0 = N ∪ {0}.
If ϕ ∈ S , then |ϕ(x)| 6 Cm (1 + |x|)−m for any m ∈ N0 . The second part
of next example shows that the converse is not true.
2
Ex. 1.30. ϕ(x) = e−ε|x| , ε > 0, belongs to S ; on the other hand, ϕ(x) =
e−ε|x| fails to be differential at the origin and, therefore, does not belong
to S .
2 )γ
Ex. 1.31. ϕ(x) = e−ε(1+|x| belongs to S for any ε, γ > 0.
Ex. 1.32. S contains the space D(Rn ).
But it is not immediately clear that D is nonempty. To find a function in D,
consider the function  −1/t
e , t > 0,
f (t) =
0, t 6 0.
Then, f ∈ C , is bounded and so are all its derivatives. Let ϕ(t) = f (1 +
∞
2
t)f (1 − t), then ϕ(t) = e−2/(1−t ) if |t| < 1, is zero otherwise. It clearly
belongs to D = D(R1 ). We can easily obtain n-dimensional variants from ϕ.
For examples,
(i) For x ∈ Rn , define ψ(x) = ϕ(x1 )ϕ(x2 ) · · · ϕ(xn ), then ψ ∈ D(Rn );
2
(ii) For x ∈ Rn , define ψ(x) = e−2/(1−|x| ) for |x| < 1 and 0 otherwise, then
ψ ∈ D(Rn );
(iii) If η ∈ C ∞ and ψ is the function in (ii), then ψ(εx)η(x) defines a func-
tion in D(Rn ); moreover, e2 ψ(εx)η(x) → η(x) as ε → 0.
Ex. 1.33. We observe that the order of multiplication by powers of
x1 , · · · , xn and differentiation, in (1.14), could have been reversed. That
is, ϕ ∈ S if and only if ϕ ∈ C ∞ and supx∈Rn |∂ β (xα ϕ(x))| < ∞ for all
multi-indices α and β of nonnegative integers. This shows that if P is a
polynomial in n variables and ϕ ∈ S then P (x)ϕ(x) and P (∂)ϕ(x) are
again in S , where P (∂) is the associated differential operator (i.e., we
replace xα by ∂ α in P (x)).
Ex. 1.34. Sometimes S (Rn ) is called the space of rapidly decaying func-
2 x
tions. But observe that the function ϕ(x) = e−x eie is not in S (R).
Hence, rapid decay of the value of the function alone does not assure
the membership in S (R).
- 18 - 1. The Fourier Transform and Tempered Distributions

Theorem 1.35. The spaces C0 (Rn ) and Lp (Rn ), 1 6 p 6 ∞, contain S (Rn ).

Moreover, both S and D are dense in C0 (Rn ) and Lp (Rn ) for 1 6 p < ∞.

Proof. S ⊂ C0 ⊂ L∞ is obvious by (1.14). The Lp norm of ϕ ∈ S is

bounded by a finite linear combination of L∞ norms of terms of the form
xα ϕ(x). In fact, by (1.14), we have
Z 1/p
p
|ϕ(x)| dx
Rn
Z 1/p Z 1/p
p p
6 |ϕ(x)| dx + |ϕ(x)| dx
|x|61 |x|>1
Z 1/p Z 1/p
2n −2np
6kϕk∞ dx + k|x| |ϕ(x)|k∞ |x| dx
|x|61 |x|>1
ω  1/p
n−1
1/p ωn−1
= kϕk∞ + |x|2n |ϕ| ∞
n (2p − 1)n
<∞.
For the proof of the density, we only need to prove the case of D since
D ⊂ S . We will use the fact that the set of finite linear combinations of
characteristic functions of bounded measurable sets in Rn is dense in Lp (Rn ),
1 6 p < ∞. This is a well-known fact from functional analysis.
Now, let E ⊂ Rn be a bounded measurable set and let ε > 0. Then, there
exists a closed set F and an open set Q such that F ⊂ E ⊂ Q and (Q \ m
m
F ) < εp (or only (Q) < εp if there is no closed set F ⊂ E). Here m
is the Lebesgue measure in Rn . Next, let ϕ be a function from D such that
supp ϕ ⊂ Q, ϕ|F ≡ 1 and 0 6 ϕ 6 1. Then,
m
Z Z
p p
kϕ − χE kp = |ϕ(x) − χE (x)| dx 6 dx = (Q \ F ) < εp
Rn Q\F
or
kϕ − χE kp < ε,
where χE denotes the characteristic function of E. Thus, we may conclude
that D(Rn ) = Lp (Rn ) with respect to Lp measure for 1 6 p < ∞.
For the case of C0 , we leave it to the interested reader. 
Remark 1.36. The density is not valid for p = ∞. Indeed, for a nonzero
constant function f ≡ c0 6= 0 and for any function ϕ ∈ D(Rn ), we have
kf − ϕk∞ > |c0 | > 0.
Hence we cannot approximate any function from L∞ (Rn ) by functions
from D(Rn ). This example also indicates that S is not dense in L∞
since lim |ϕ(x)| = 0 for all ϕ ∈ S .
|x|→∞
From part (v) in Proposition 1.3, we immediately have
1.3. Schwartz spaces - 19 -

Theorem 1.37. If ϕ ∈ S , then ϕ̂ ∈ S .

If ϕ, ψ ∈ S , then Theorem 1.37 implies that ϕ̂, ψ̂ ∈ S . Therefore, ϕ̂ψ̂ ∈
S . By part (vi) in Proposition 1.3, i.e., F (ϕ ∗ ψ) = ϕ̂ψ̂, an application of the
inverse Fourier transform shows that
Theorem 1.38. If ϕ, ψ ∈ S , then ϕ ∗ ψ ∈ S .
The space S (Rn ) is not a normed space because |ϕ|α,β is only a semi-norm
for multi-indices α and β, i.e., the condition
|ϕ|α,β = 0 if and only if ϕ = 0
fails to hold, for example, for constant function ϕ. But the space (S , ρ) is a
metric space if the metric ρ is defined by
X |ϕ − ψ|α,β
ρ(ϕ, ψ) = 2−|α|−|β| .
α,β∈N n
1 + |ϕ − ψ|α,β
0

Theorem 1.39 (Completeness). The space (S , ρ) is a complete metric space,

i.e., every Cauchy sequence converges.

Proof. Let {ϕk }∞ k=1 ⊂ S be a Cauchy sequence. For any σ > 0 and any
−|γ|
γ ∈ Nn0 , let ε = 21+2σσ , then there exists an N0 (ε) ∈ N such that ρ(ϕk , ϕm ) < ε
when k, m > N0 (ε) since {ϕk }∞ k=1 is a Cauchy sequence. Thus, we have
|ϕk − ϕm |0,γ σ
< ,
1 + |ϕk − ϕm |0,γ 1+σ
and then
sup |∂ γ (ϕk − ϕm )| < σ
x∈K
for any compact set K ⊂ R . It means that {ϕk }∞
n
k=1 is a Cauchy sequence in
the Banach space C (K). Hence, there exists a function ϕ ∈ C |γ| (K) such
|γ|

that
lim ϕk = ϕ, in C |γ| (K).
k→∞
Thus, we can conclude that ϕ ∈ C ∞ (Rn ). It only remains to prove that ϕ ∈ S .
It is clear that for any α, β ∈ Nn0
sup |xα ∂ β ϕ| 6 sup |xα ∂ β (ϕk − ϕ)| + sup |xα ∂ β ϕk |
x∈K x∈K x∈K

6Cα (K) sup |∂ (ϕk − ϕ)| + sup |xα ∂ β ϕk |.

β
x∈K x∈K
Taking k → ∞, we obtain
sup |xα ∂ β ϕ| 6 lim sup |ϕk |α,β < ∞.
x∈K k→∞
The last inequality is valid since {ϕk }k=1 is
∞
a Cauchy sequence, so that |ϕk |α,β
is bounded. The last inequality doesn’t depend on K either. Thus, |ϕ|α,β < ∞
and then ϕ ∈ S . 
- 20 - 1. The Fourier Transform and Tempered Distributions

Moreover, some easily established properties of S and its topology, are the
following:
Proposition 1.40. i) The mapping ϕ(x) 7→ xα ∂ β ϕ(x) is continuous.
ii) If ϕ ∈ S , then limh→0 τh ϕ = ϕ.
iii) Suppose ϕ ∈ S and h = (0, · · · , hi , · · · , 0) lies on the i-th coordinate
axis of Rn , then the difference quotient [ϕ − τh ϕ]/hi tends to ∂ϕ/∂xi as |h| →
0.
iv) The Fourier transform is a homeomorphism of S onto itself.
v) S is separable.
Finally, we describe and prove a fundamental result of Fourier analysis that
is known as the uncertainty principle. In fact this theorem was "discovered" by
W. Heisenberg in the context of quantum mechanics. Expressed colloquially,
the uncertainty principle says that it is not possible to know both the position
and the momentum of a particle at the same time. Expressed more precisely,
the uncertainty principle says that the position and the momentum cannot be
simultaneously localized.
In the context of harmonic analysis, the uncertainty principle implies that
one cannot at the same time localize the value of a function and its Fourier
transform. The exact statement is as follows.
Theorem 1.41 (The Heisenberg uncertainty principle). Suppose ψ is a
function in S (R). Then
 −1/2
|ω| kψk22
kxψk2 kξ ψ̂k2 > ,
2π 2|ω|
2
and equality holds if and only if ψ(x) = Ae−Bx where B > 0 and A ∈ R.
Moreover, we have
 −1/2
|ω| kψk22
k(x − x0 )ψk2 k(ξ − ξ0 )ψ̂k2 >
2π 2|ω|
for every x0 , ξ0 ∈ R.

Proof. The last inequality actually follows from the first by replacing ψ(x) by
e−ωixξ0 ψ(x + x0 ) (whose Fourier transform is eωix0 (ξ+ξ0 ) ψ̂(ξ + ξ0 ) by parts (ii)
and (iii) in Proposition 1.3) and changing variables. To prove the first inequal-
ity, we argue as follows.
Since ψ ∈ S , we know that ψ and ψ 0 are rapidly decreasing. Thus, an
integration by parts gives
Z ∞ Z ∞
2 2 d
kψk2 = |ψ(x)| dx = − x |ψ(x)|2 dx
−∞ −∞ dx
Z ∞ 
=− xψ 0 (x)ψ(x) + xψ 0 (x)ψ(x) dx.
−∞
1.4. The class of tempered distributions - 21 -

The last identity follows because |ψ|2 = ψψ. Therefore,

Z ∞
2
kψk2 6 2 |x||ψ(x)||ψ 0 (x)|dx 6 2kxψk2 kψ 0 k2 ,
−∞
where we have used the Cauchy-Schwarz inequality. By part (v) in Proposition
1.3, we have F (ψ 0 )(ξ) = ωiξ ψ̂(ξ). It follows, from the Plancherel theorem,
that
 1/2  1/2
0 |ω| 0 |ω|
kψ k2 = kF (ψ )k2 = |ω|kξ ψ̂k2 .
2π 2π
Thus, we conclude the proof of the inequality in the theorem.
If equality holds, then we must also have equality where we applied the
Cauchy-Schwarz inequality, and as a result, we find that ψ 0 (x) = βxψ(x) for
2
some constant β. The solutions to this equation are ψ(x) = Aeβx /2 , where
A is a constant. Since we want ψ to be a Schwartz function, we must take
β = −2B < 0. 

1.4 The class of tempered distributions

The collection S 0 of all continuous linear functionals on S is called the

space of tempered distributions. That is
Definition 1.42. The functional T : S → C is a tempered distribution if
i) T is linear, i.e., hT, αϕ + βψi = αhT, ϕi + βhT, ψi for all α, β ∈ C and
ϕ, ψ ∈ S .
ii) T is continuous on S , i.e., there exist n0 ∈ N0 and a constant c0 > 0
such that
X
|hT, ϕi| 6 c0 |ϕ|α,β
|α|,|β|6n0
for any ϕ ∈ S .
In addition, for Tk , T ∈ S 0 , the convergence Tk → T in S 0 means that
hTk , ϕi → hT, ϕi in C for all ϕ ∈ S .
Remark 1.43. Since D ⊂ S , the space of tempered distributions S 0 is
more narrow than the space of distributions D 0 , i.e., S 0 ⊂ D 0 . Another
more narrow distribution space E 0 which consists of continuous linear
functionals on the (widest test function) space E := C ∞ (Rn ). In short,
D ⊂ S ⊂ E implies that
E 0 ⊂ S 0 ⊂ D 0.
Ex. 1.44. Let f ∈ Lp (Rn ), 1 6 p 6 ∞, and define T = Tf by letting
Z
hT, ϕi = hTf , ϕi = f (x)ϕ(x)dx
Rn
- 22 - 1. The Fourier Transform and Tempered Distributions

for ϕ ∈ S . It is clear that Tf is a linear functional on S . To show that it

is continuous, therefore, it suffices to show that it is continuous at the
origin. Then, suppose ϕk → 0 in S as k → ∞. From the proof of Theo-
rem 1.35, we have seen that for any q > 1, kϕk kq is dominated by a finite
linear combination of L∞ norms of terms of the form xα ϕk (x). That is,
kϕk kq is dominated by a finite linear combination of semi-norms |ϕk |α,0 .
Thus, kϕk kq → 0 as k → ∞. Choosing q = p0 , i.e., 1/p+1/q = 1, Hölder’s
inequality shows that |hT, ϕk i| 6 kf kp kϕk kp0 → 0 as k → ∞. Thus,
T ∈ S 0.
Ex. 1.45. We consider the case n = 1. Let f (x) = m k
P
k=0 ak x be a poly-
nomial, then f ∈ S 0 since
Z X m
|hTf , ϕi| = ak xk ϕ(x)dx
R k=0
m
X Z
6 |ak | (1 + |x|)−1−ε (1 + |x|)1+ε |x|k |ϕ(x)|dx
k=0 R
m
X Z
6C |ak ||ϕ|k+1+ε,0 (1 + |x|)−1−ε dx,
k=0 R

so that the condition ii) of the definition is satisfied for ε = 1 and n0 =

m + 2.
Ex. 1.46. Fix x0 ∈ Rn and a multi-index β ∈ Nn0 . By the continuity of
the semi-norm | · |α,β in S , we have that hT, ϕi = ∂ β ϕ(x0 ), for ϕ ∈ S ,
defines a tempered distribution. A special case is the Dirac δ-function:
hTδ , ϕi = ϕ(0).
The tempered distributions of Examples 1.44-1.46 are called functions or
measures. We shall write, in these cases, f and δ instead of Tf and Tδ . These
functions and measures may be considered as embedded in S 0 . If we put on
S 0 the weakest topology such that the linear functionals T → hT, ϕi (ϕ ∈ S )
are continuous, it is easy to see that the spaces Lp (Rn ), 1 6 p 6 ∞, are
continuously embedded in S 0 . The same is true for the space of all finite
Borel measures on Rn , i.e., B(Rn ).
There exists a simple and important characterization of tempered distribu-
tions:
Theorem 1.47. A linear functional T on S is a tempered distribution if and
only if there exists a constant C > 0 and integers ` and m such that
X
|hT, ϕi| 6 C |ϕ|α,β
|α|6`,|β|6m
for all ϕ ∈ S .
1.4. The class of tempered distributions - 23 -

Proof. It is clear that the existence of C, `, m implies the continuity of T .

Suppose T is continuous. It follows from the definition of the metric that a
basis for the neighborhoods of the origin in S is the collection of sets Nε,`,m =
{ϕ : |α|6`,|β|6m |ϕ|α,β < ε}, where ε > 0 and ` and m are integers, because
P

ϕk → ϕ as k → ∞ if and only if |ϕk − ϕ|α,β → 0 for all (α, β) in the

topology induced by this system of neighborhoods and their translates. Thus,
there exists such a set Nε,`,m satisfying |hT, ϕi| 6 1 whenever ϕ ∈ Nε,`,m .
Let kϕk = |α|6`,|β|6m |ϕ|α,β for all ϕ ∈ S . If σ ∈ (0, ε), then ψ =
P

σϕ/kϕk ∈ Nε,`,m if ϕ 6= 0. From the linearity of T , we obtain

σ
|hT, ϕi| = |hT, ψi| 6 1.
kϕk
But this is the desired inequality with C = 1/σ. 
Ex. 1.48. Let T ∈ S 0 and ϕ ∈ D(Rn ) with ϕ(0) = 1. Then the product
ϕ(x/k)T is well-defined in S 0 by
hϕ(x/k)T, ψi := hT, ϕ(x/k)ψi,
for all ψ ∈ S . If we consider the sequence Tk := ϕ(x/k)T , then
hTk , ψi ≡ hT, ϕ(x/k)ψi → hT, ψi
as k → ∞ since ϕ(x/k)ψ → ψ in S . Thus, Tk → T in S 0 as k → ∞.
Moreover, Tk has compact support as a tempered distribution in view
of the compactness of ϕk = ϕ(x/k).
Now we are ready to prove more serious and more useful fact.
Theorem 1.49. Let T ∈ S 0 , then there exists a sequence {Tk }∞
k=0 ⊂ S such
that Z
hTk , ϕi = Tk (x)ϕ(x)dx → hT, ϕi, as k → ∞,
Rn
where ϕ ∈ S . In short, S is dense in S 0 with respect to the topology on S 0 .
Proof. If h and g are integrable functions and ϕ ∈ S , then it follows, from
Fubini’s theorem,
Z that Z Z Z
hh ∗ g, ϕi = ϕ(x) h(x − y)g(y)dydx = g(y) h(x − y)ϕ(x)dxdy
Rn Rn Rn R n
Z Z
= g(y) Rh(y − x)ϕ(x)dxdy = hg, Rh ∗ ϕi,
Rn Rn
where Rh(x) := h(−x) is the Rreflection of h.
Let now ψ ∈ D(Rn ) with Rn ψ(x)dx = 1 and ψ(−x) = ψ(x). Let ζ ∈
D(Rn ) with ζ(0) = 1. Denote ψk (x) := k n ψ(kx). For any T ∈ S 0 , denote
Tk := ψk ∗ T̃k , where T̃k = ζ(x/k)T . From above considerations, we know
that hψk ∗ T̃k , ϕi = hT̃k , Rψk ∗ ϕi.
Let us prove that these Tk meet the requirements of the theorem. In fact, we
have
- 24 - 1. The Fourier Transform and Tempered Distributions

hTk , ϕi ≡hψk ∗ T̃k , ϕi = hT̃k , Rψk ∗ ϕi = hζ(x/k)T, ψk ∗ ϕi

=hT, ζ(x/k)(ψk ∗ ϕ)i → hT, ϕi, as k → ∞,
by the fact ψk ∗ ϕ → ϕ in S as k → ∞ in view of Theorem 1.15, and the fact
ζ(x/k) → 1 pointwise as k → ∞ since ζ(0) = 1 and ζ(x/k)ϕ → ϕ in S as
k → ∞. Finally, since ψk , ζ ∈ D(Rn ), it follows that Tk ∈ D(Rn ) ⊂ S (Rn ).


Definition 1.50. Let L : S → S be a linear continuous mapping. Then,

the dual/conjugate mapping L0 : S 0 → S 0 is defined by
hL0 T, ϕi := hT, Lϕi, T ∈ S 0, ϕ ∈ S .
Clearly, L0 is also a linear continuous mapping.
Corollary 1.51. Any linear continuous mapping (or operator) L : S → S
admits a linear continuous extension L̃ : S 0 → S 0 .

Proof. If T ∈ S 0 , then by Theorem 1.49, there exists a sequence {Tk }∞

k=0 ⊂
S such that Tk → T in S as k → ∞. Hence,
0

hLTk , ϕi = hTk , L0 ϕi → hT, L0 ϕi := hL̃T, ϕi, as k → ∞,

for any ϕ ∈ S . 
Now, we can list the properties of tempered distributions about the multipli-
cation, differentiation, translation, dilation and Fourier transform.
Theorem 1.52. The following linear continuous operators from S into S
admit unique linear continuous extensions as maps from S 0 into S 0 : For
T ∈ S 0 and ϕ ∈ S ,
i) hψT, ϕi := hT, ψϕi, ψ ∈ S .
ii) h∂ α T, ϕi := hT, (−1)|α| ∂ α ϕi, α ∈ Nn0 .
iii) hτh T, ϕi := hT, τ−h ϕi, h ∈ Rn .
iv) hδλ T, ϕi := hT, |λ|−n δ1/λ ϕi, 0 6= λ ∈ R.
v) hF T, ϕi := hT, F ϕi.

Proof. See the previous definition, Theorem 1.49 and its corollary. 
Remark 1.53. Since hF −1 F T, ϕi = hF T, F −1 ϕi = hT, F F −1 ϕi =
hT, ϕi, we get F −1 F = F F −1 = I in S 0 .
Ex. 1.54. Since for any ϕ ∈ S ,
Z
hF 1, ϕi =h1, F ϕi = (F ϕ)(ξ)dξ
Rn
 −n  n Z
|ω| |ω|
= eωi0·ξ (F ϕ)(ξ)dξ
2π 2π Rn
 −n  −n
|ω| −1 |ω|
= F F ϕ(0) = ϕ(0)
2π 2π
1.4. The class of tempered distributions - 25 -
−n
|ω|

= hδ, ϕi,
2π
we have
−n
|ω|

1̂ = δ, in S 0 .
2π
 n
|ω|
Moreover, δ̌ = 2π
· 1.
Ex. 1.55. For ϕ ∈ S , we have
Z
hδ̂, ϕi = hδ, F ϕi = ϕ̂(0) = e−ωix·0 ϕ(x)dx = h1, ϕi.
Rn
0
Thus, δ̂ = 1 in S .
Ex. 1.56. Since
α δ, ϕi =h∂ α δ, ϕ̂i = (−1)|α| hδ, ∂ α ϕ̂i = hδ, F [(ωiξ)α ϕ]i
h∂d
=hδ̂, (ωiξ)α ϕi = h(ωiξ)α , ϕi,
α δ = (ωiξ)α .
we have ∂d
Now, we shall show that the convolution can be defined on the class S 0 .
We first recall a notation we have used: If g is any function on Rn , we define
its reflection, Rg, by letting Rg(x) = g(−x). A direct application of Fubini’s
theorem shows that if u, ϕ and ψ are all in S , then
Z Z
(u ∗ ϕ)(x)ψ(x)dx = u(x)(Rϕ ∗ ψ)(x)dx.
Rn Rn

The mappings ψ 7→ Rn (u ∗ ϕ)(x)ψ(x)dx and θ 7→ Rn u(x)θ(x)dx are linear

R R

functionals on S . If we denote these functionals by u ∗ ϕ and u, the last

equality can be written in the form:
hu ∗ ϕ, ψi = hu, Rϕ ∗ ψi. (1.15)
If u ∈ S 0 and ϕ, ψ ∈ S , the right side of (1.15) is well-defined since Rϕ∗ψ ∈
S . Furthermore, the mapping ψ 7→ hu, Rϕ ∗ ψi, being the composition of two
continuous functions, is continuous. Thus, we can define the convolution of the
distribution u with the testing function ϕ, u ∗ ϕ, by means of equality (1.15).
It is easy to show that this convolution is associative in the sense that (u ∗
ϕ) ∗ ψ = u ∗ (ϕ ∗ ψ) whenever u ∈ S 0 and ϕ, ψ ∈ S . The following result
is a characterization of the convolution we have just described.
Theorem 1.57. If u ∈ S 0 and ϕ ∈ S , then the convolution u ∗ ϕ is the
function f , whose value at x ∈ Rn is f (x) = hu, τx Rϕi, where τx denotes the
translation by x operator. Moreover, f belongs to the class C ∞ and it, as well
as all its derivatives, are slowly increasing.

Proof. We first show that f is C ∞ slowly increasing. Let h =

(0, · · · , hj , · · · , 0), then by part iii) in Proposition 1.40,
- 26 - 1. The Fourier Transform and Tempered Distributions

τx+h Rϕ − τx Rϕ ∂Rϕ
→ −τx ,
hj ∂yj
as |h| → 0, in the topology of S . Thus, since u is continuous, we have
f (x + h) − f (x) τx+h Rϕ − τx Rϕ ∂Rϕ
= hu, i → hu, −τx i
hj hj ∂yj
as hj → 0. This, together with ii) in Proposition 1.40, shows that f has con-
tinuous first-order partial derivatives. Since ∂Rϕ/∂yj ∈ S , we can iterate
this argument and show that ∂ β f exists and is continuous for all multi-index
β ∈ Nn0 . We observe that ∂ β f (x) = hu, (−1)|β| τx ∂ β Rϕi. Consequently, since
∂ β Rϕ ∈ S , if f were slowly increasing, then the same would hold for all the
derivatives of f . In fact, that f is slowly increasing is an easy consequence of
Theorem 1.47: There exist C > 0 and integers ` and m such that
X
|f (x)| = |hu, τx Rϕi| 6 C |τx Rϕ|α,β .
|α|6`,|β|6m

But |τx Rϕ|α,β = supy∈Rn |y ∂ Rϕ(y −x)| = supy∈Rn |(y +x)α ∂ β Rϕ(y)| and
α β

the latter is clearly bounded by a polynomial in x.

R In order to show that u ∗ ϕ is the function f , we must show that hu ∗ ϕ, ψi =
Rn
f (x)ψ(x)dx. But,
Z
hu ∗ ϕ, ψi =hu, Rϕ ∗ ψi = hu, Rϕ(· − x)ψ(x)dxi
Z Rn

=hu, τx Rϕ(·)ψ(x)dxi
n
Z R Z
= hu, τx Rϕiψ(x)dx = f (x)ψ(x)dx,
Rn Rn
since u is continuous and linear and the fact that the integral
τ Rϕ(y)ψ(x)dx converges in S , which is the desired equality.
R
Rn x


1.5 Characterization of operators commuting with translations

Having set down these facts of distribution theory, we shall now apply them
to the study of the basic class of linear operators that occur in Fourier analysis:
the class of operators that commute with translations.
Definition 1.58. A vector space X of measurable functions on Rn is
called closed under translations if for f ∈ X we have τy f ∈ X for all
y ∈ Rn . Let X and Y be vector spaces of measurable functions on Rn
that are closed under translations. Let also T be an operator from X to
Y . We say that T commutes with translations or is translation invariant if
T (τy f ) = τy (T f )
1.5. Characterization of operators commuting with translations - 27 -

for all f ∈ X and all y ∈ Rn .

It is automatic to see that convolution operators commute with translations.
One of the main goals of this section is to prove the converse, i.e., every
bounded linear operator that commutes with translations is of convolution
type. We have the following:
Theorem 1.59. Let 1 6 p, q 6 ∞. Suppose T is a bounded linear operator
from Lp (Rn ) into Lq (Rn ) that commutes with translations. Then there exists
a unique tempered distribution u such that
T f = u ∗ f, ∀f ∈ S .
The theorem will be a consequence of the following lemma.
Lemma 1.60. Let 1 6 p 6 ∞. If f ∈ Lp (Rn ) has derivatives in the Lp norm
of all orders 6 n + 1, then f equals almost everywhere a continuous function
g satisfying
X
|g(0)| 6 C k∂ α f kp ,
|α|6n+1
where C depends only on the dimension n and the exponent p.

Proof. Let ξ ∈ Rn . Then there exists a Cn0 such that

X
(1 + |ξ|2 )(n+1)/2 6 (1 + |ξ1 | + · · · + |ξn |)n+1 6 Cn0 |ξ α |.
|α|6n+1

Let us first suppose p = 1, we shall show fˆ ∈ L . By part (v) in Proposition

1

1.3 and part (i) in Theorem 1.5, we have

|fˆ(ξ)| 6C 0 (1 + |ξ|2 )−(n+1)/2 |ξ α ||fˆ(ξ)|
X
n
|α|6n+1
X
=Cn0 (1 + |ξ|2 )−(n+1)/2 |ω|−|α| |F (∂ α f )(ξ)|
|α|6n+1
X
6C 00 (1 + |ξ|2 )−(n+1)/2 k∂ α f k1 .
|α|6n+1

Since (1 + |ξ| )
2 −(n+1)/2
defines anR integrable function on Rn , it follows that
fˆ ∈ L1 (Rn ) and, letting C 000 = C 00 Rn (1 + |ξ|2 )−(n+1)/2 dξ, we get
kfˆk1 6 C 000
X
k∂ α f k1 .
|α|6n+1
Thus, by Theorem 1.19, f equals almost everywhere a continuous function g
and by Theorem 1.5,
 n
|ω|
kfˆk1 6 C
X
|g(0)| 6 kf k∞ 6 k∂ α f k1 .
2π
|α|6n+1
- 28 - 1. The Fourier Transform and Tempered Distributions

Suppose now that p > 1. Choose ϕ ∈ D(Rn ) such that ϕ(x) = 1 if |x| 6 1
and ϕ(x) = 0 if |x| > 2. Then, it is clear that f ϕ ∈ L1 (Rn ). Thus, f ϕ equals
almost everywhere a continuous function h such that
X
|h(0)| 6 C k∂ α (f ϕ)k1 .
|α|6n+1

By Leibniz’ rule for differentiation, we have ∂ α (f ϕ) = µ+ν=α α! µ

∂ f ∂ ν ϕ,
P
µ!ν!
and then Z X α!
α
k∂ (f ϕ)k1 6 |∂ µ f ||∂ ν ϕ|dx
|x|62 µ+ν=α µ!ν!
X Z
ν
6 C sup |∂ ϕ(x)| |∂ µ f (x)|dx
µ+ν=α |x|62 |x|62
X Z X
6A |∂ µ f (x)|dx 6 AB k∂ µ f kp ,
|µ|6|α| |x|62 |µ|6|α|

where A > k∂ ϕk∞ , |ν| 6 |α|, and B depends only on p and n. Thus, we can
ν

find a constant K such that

X
|h(0)| 6 K k∂ α f kp .
|α|6n+1

Since ϕ(x) = 1 if |x| 6 1, we see that f is equal almost everywhere to a

continuous function g in the sphere of radius 1 centered at 0, moreover,
X
|g(0)| = |h(0)| 6 K k∂ α f kp .
|α|6n+1
But, by choosing ϕ appropriately, the argument clearly shows that f equals
almost everywhere a continuous function on any sphere centered at 0. This
proves the lemma. 
Now, we turn to the proof of the previous theorem.
Proof of Theorem 1.59. We first prove that
∂ β T f = T ∂ β f, ∀f ∈ S (Rn ). (1.16)
In fact, if h = (0, · · · , hj , · · · , 0) lies on the j-th coordinate axis, we have
τh (T f ) − T f T (τh f ) − T f τh f − f
 
= =T ,
hj hj hj
since T is linear and commuting with translations. By part iii) in Proposition
1.40, τhhf j−f → − ∂x ∂f
j
in S as |h| → 0 and also in Lp norm due to the den-
sity of S in Lp . Since T is bounded operator from Lp to Lq , it follows that
τh (T f )−T f
hj
→ − ∂T f
∂xj
in Lq as |h| → 0. By induction, we get (1.16). By Lemma
1.60, T f equals almost everywhere a continuous function gf satisfying
X X
|gf (0)| 6C k∂ β (T f )kq = C kT (∂ β f )kq
|β|6n+1 |β|6n+1
1.5. Characterization of operators commuting with translations - 29 -
X
6kT kC k∂ β f kp .
|β|6n+1
From the proof of Theorem 1.35, we know that the Lp norm of f ∈
S is bounded by a finite linear combination of L∞ norms of terms of
the form xα f (x). Thus, there exists an m ∈ N such that |gf (0)| 6
C |α|6m,|β|6n+1 kxα ∂ β f k∞ = C |α|6m,|β|6n+1 |f |α,β . Then, by Theorem
P P

1.47, the mapping f 7→ gf (0) is a continuous linear functional on S , denoted

by u1 . We claim that u = Ru1 is the linear functional we are seeking. Indeed,
if f ∈ S , using Theorem 1.57, we obtain
(u ∗ f )(x) =hu, τx Rf i = hu, R(τ−x f )i = hRu, τ−x f i = hu1 , τ−x f i
=(T (τ−x f ))(0) = (τ−x T f )(0) = T f (x).
We note that it follows from this construction that u is unique. The theorem
is therefore proved. 
Combining this result with Theorem 1.57, we obtain the fact that T f , for
f ∈ S , is almost everywhere equal to a C ∞ function which, together with all
its derivatives, is slowly increasing.
Now, we give a characterization of operators commuting with translations
in L1 (Rn ).
Theorem 1.61. Let T be a bounded linear operator mapping L1 (Rn ) to itself.
Then a necessary and sufficient condition that T commutes with translations
is that there exists a measure µ in B(Rn ) such that T f = µ ∗ f , for all
f ∈ L1 (Rn ). One has then kT k = kµk.

Proof. We first prove the sufficiency. Suppose that T f = µ ∗ f for a measure

µ ∈ B(Rn ) and all f ∈ L1 (Rn ). Since B ⊂ S 0 , by Theorem 1.57, we have
τh (T f )(x) =(T f )(x − h) = hµ, τx−h Rf i = hµ(y), f (−y − x + h)i
=hµ, τx Rτh f i = µ ∗ τh f = T τh f,
i.e., τh T = T τh . On the other hand, we have kT f k1 = kµ ∗ f k1 6 kµkkf k1
which implies kT k = kµk.
Now, we prove the necessariness. Suppose that T commutes with transla-
tions and kT f k1 6 kT kkf k1 for all f ∈ L1 (Rn ). Then, by Theorem 1.59,
there exists a unique tempered distribution µ such that T f = µ ∗ f for all
f ∈ S . The remainder is to prove µ ∈ B(Rn ).
We consider the family of L1 functions µε = µ∗W (·, ε) = T W (·, ε), ε > 0.
Then by assumption and Lemma 1.14, we get
kµε k1 6 kT kkW (·, ε)k1 = kT k.
That is, the family {µε } is uniformly bounded in the L1 norm. Let us consider
L1 (Rn ) as embedded in the Banach space B(Rn ). B(Rn ) can be identified
with the dual of C0 (Rn ) by making each ν ∈ B corresponding to the linear
- 30 - 1. The Fourier Transform and Tempered Distributions

functional assigning to ϕ ∈ C0 the value Rn ϕ(x)dν(x). Thus, the unit sphere

R

of B is compact in the weak* topology. In particular, we can find a ν ∈ B

and a null sequence {εk } such that µεk → ν as k → ∞ in this topology. That
is, for each ϕ ∈ C0 ,
Z Z
lim ϕ(x)µεk (x)dx = ϕ(x)dν(x). (1.17)
k→∞ Rn Rn
We now claim that ν, consider as a distribution, equals µ.
Therefore, we must show that hµ, ψi = Rn ψ(x)dν(x) for all ψ ∈ S . Let
R

ψε = W (·, ε) ∗ ψ. Then, for all α ∈ Nn0 , we have ∂ α ψε = W (·, ε) ∗ ∂ α ψ. It

follows from Theorem 1.15 that ∂ α ψε (x) converges to ∂ α ψ(x) uniformly in x.
Thus, ψε → ψ in S as ε → 0 and this implies that hµ, ψε i → hµ, ψi. But,
since W (·, ε) = RW (·, ε),
Z
hµ, ψε i = hµ, W (·, ε) ∗ ψi = hµ ∗ W (·, ε), ψi = µε (x)ψ(x)dx.
Rn
Thus, putting ε = εk , letting k → ∞ Rand applying (1.17) with ϕ = ψ, we
obtain the desired equality hµ, ψi = Rn ψ(x)dν(x). Hence, µ ∈ B. This
completes the proof. 
For L2 , we can also give a very simple characterization of these operators.
Theorem 1.62. Let T be a bounded linear transformation mapping L2 (Rn ) to
itself. Then a necessary and sufficient condition that T commutes with trans-
lation is that there exists an m ∈ L∞ (Rn ) such that T f = u ∗ f with û = m,
for all f ∈ L2 (Rn ). One has then kT k = kmk∞ .

Proof. If v ∈ S 0 and ψ ∈ S , we define their product, vψ, to be the element

of S 0 such that hvψ, ϕi = hv, ψϕi for all ϕ ∈ S . With the product of a
distribution with a testing function so defined we first observe that whenever
u ∈ S 0 and ϕ ∈ S , then
F (u ∗ ϕ) = ûϕ̂. (1.18)
To see this, we must show that hF (u ∗ ϕ), ψi = hûϕ̂, ψi for all ψ ∈ S . It
follows immediately, from (1.15), part (vi) in Proposition 1.3 and the Fourier
inversion formula, that
hF (u ∗ ϕ), ψi =hu ∗ ϕ, ψ̂i = hu, Rϕ ∗ ψ̂i = hû, F −1 (Rϕ ∗ ψ̂)i
  n
|ω|

= û, (F (Rϕ ∗ ψ̂))(−ξ)
2π
  n
|ω|

= û, (F (Rϕ))(−ξ)(F ψ̂)(−ξ) = hû, ϕ̂(ξ)ψ(ξ)i
2π
=hûϕ̂, ψi.
Thus, (1.18) is established.
1.5. Characterization of operators commuting with translations - 31 -

Now, we prove the necessariness. Suppose that T commutes with transla-

tions and kT f k2 6 kT kkf k2 for all f ∈ L2 (Rn ). Then, by Theorem 1.59,
there exists a unique tempered distribution u such that T f = u ∗ f for all
f ∈ S . The remainder is to prove û ∈ L∞ (Rn ).
|ω|
 −n/2
Let ϕ0 = e− 2 |x| , then, we have ϕ0 ∈ S and ϕ̂0 = |ω|
2
2π
ϕ0 by
Theorem 1.10 with a = 1/2|ω|. Thus, T ϕ0 = u ∗ ϕ0 ∈ L2 and therefore
Φ0 := F (u ∗ ϕ0 ) = ûϕ̂0 ∈ L2 by (1.18) and the Plancherel theorem. Let
 n/2 |ω|
m(ξ) = |ω|
2
2π
e 2 |ξ| Φ0 (ξ) = Φ0 (ξ)/ϕ̂0 (ξ).
We claim that
F (u ∗ ϕ) = mϕ̂ (1.19)
for all ϕ ∈ S . By (1.18), it suffices to show that hûϕ̂, ψi = hmϕ̂, ψi for
all ψ ∈ D since D is dense in S . But, if ψ ∈ D, then (ψ/ϕ̂0 )(ξ) =
 n/2 |ω|
|ω| 2
2π
ψ(ξ)e 2 |ξ| ∈ D; thus,
hûϕ̂, ψi =hû, ϕ̂ψi = hû, ϕ̂ϕ̂0 ψ/ϕ̂0 i = hûϕ̂0 , ϕ̂ψ/ϕ̂0 i
 n/2
|ω|
Z
|ω| 2
= Φ0 (ξ)ϕ̂(ξ) ψ(ξ)e 2 |ξ| dξ
n 2π
ZR
= m(ξ)ϕ̂(ξ)ψ(ξ)dξ = hmϕ̂, ψi.
Rn
It follows immediately that û = m: We have just shown that hû, ϕ̂ψi =
hmϕ̂, ψi = hm, ϕ̂ψi for all ϕ ∈ S and ψ ∈ D. Selecting ϕ such that ϕ̂(ξ) = 1
for ξ ∈ supp ψ, this shows that hû, ψi = hm, ψi for all ψ ∈ D. Thus, û = m.
Due to
 −n/2
|ω|
kmϕ̂k2 =kF (u ∗ ϕ)k2 = ku ∗ ϕk2
2π
 −n/2
|ω|
6 kT kkϕk2 = kT kkϕ̂k2
2π
for all ϕ ∈ S , it follows that
Z
kT k2 − |m|2 |ϕ̂|2 dξ > 0,


Rn
for all ϕ ∈ S . This implies that kT k2 − |m|2 > 0 for almost all x ∈ Rn .
Hence, m ∈ L∞ (Rn ) and kmk∞ 6 kT k.
Finally, we can show the sufficiency easily. If û = m ∈ L∞ (Rn ), the
Plancherel theorem and (1.18) immediately imply that
 n/2
|ω|
kT f k2 = ku ∗ f k2 = kmfˆk2 6 kmk∞ kf k2
2π
which yields kT k 6 kmk∞ .
Thus, if m = û ∈ L∞ , then kT k = kmk∞ . 
- 32 - 1. The Fourier Transform and Tempered Distributions

For further results, one can see [SW71, p.30] and [Gra04, p.137-140].
L ECTURE N OTES ON c 2016 by Chengchun Hao
Introduction to Harmonic Analysis Email: hcc@amss.ac.cn

Chapter 2
Interpolation of Operators

2.1 Riesz-Thorin’s and Stein’s interpolation theorems

We first present a notion that is central to complex analysis, that is, the
holomorphic or analytic function.
Let Ω be an open set in C and f a complex-valued function on Ω. The
function f is holomorphic at the point z0 ∈ Ω if the quotient
f (z0 + h) − f (z0 )
h
converges to a limit when h → 0. Here h ∈ C and h 6= 0 with z0 + h ∈ Ω,
so that the quotient is well defined. The limit of the quotient, when it exists, is
denoted by f 0 (z0 ), and is called the derivative of f at z0 :
f (z0 + h) − f (z0 )
f 0 (z0 ) = lim . (2.1)
h→0 h
It should be emphasized that in the above limit, h is a complex number that
may approach 0 from any directions.
The function f is said to be holomorphic on Ω if f is holomorphic at every
point of Ω. If C is a closed subset of C, we say that f is holomorphic on C if
f is holomorphic in some open set containing C. Finally, if f is holomorphic
in all of C we say that f is entire.
Every holomorphic function is analytic, in the sense that it has a power
series expansion near every point, and for this reason we also use the term
analytic as a synonym for holomorphic. For more details, one can see [SS03,
pp.8-10].
Ex. 2.1. The function f (z) = z is holomorphic on any open set in C, and
f 0 (z) = 1. The function f (z) = z̄ is not holomorphic. Indeed, we have
f (z0 + h) − f (z0 ) h̄
=
h h
which has no limit as h → 0, as one can see by first taking h real and
then h purely imaginary.
33
- 34 - 2. Interpolation of Operators

Ex. 2.2. The function 1/z is holomorphic on any open set in C that does
not contain the origin, and f 0 (z) = −1/z 2 .
One can prove easily the following properties of holomorphic functions.
Proposition 2.3. If f and g are holomorphic in Ω, then
i) f + g is holomorphic in Ω and (f + g)0 = f 0 + g 0 .
ii) f g is holomorphic in Ω and (f g)0 = f 0 g + f g 0 .
iii) If g(z0 ) 6= 0, then f /g is holomorphic at z0 and
 0
f f 0g − f g0
= .
g g2
Moreover, if f : Ω → U and g : U → C are holomorphic, the chain rule
holds
(g ◦ f )0 (z) = g 0 (f (z))f 0 (z), for all z ∈ Ω.
The next result pertains to the size of a holomorphic function.
Theorem 2.4 (Maximum modulus principle). Suppose that Ω is a region
with compact closure Ω̄. If f is holomorphic on Ω and continuous on Ω̄, then
sup |f (z)| 6 sup |f (z)|.
z∈Ω z∈Ω̄\Ω

Proof. See [SS03, p.92]. 

For convenience, let S = {z ∈ C : 0 6 <z 6 1} be the closed strip, S ◦ =
{z ∈ C : 0 < <z < 1} be the open strip, and ∂S = {z ∈ C : <z ∈ {0, 1}}.
Theorem 2.5 (Phragmen-Lindelöf theorem/Maximum principle). As-
sume that f (z) is analytic on S ◦ and bounded and continuous on S. Then
 
sup |f (z)| 6 max sup |f (it)|, sup |f (1 + it)| .
z∈S t∈R t∈R

Proof. Assume that f (z) → 0 as |=z| → ∞. Consider the mapping h : S → C

defined by
eiπz − i
h(z) = iπz , z ∈ S. (2.2)
e +i
Then h is a bijective mapping from S onto U = {z ∈ C : |z| 6 1} \ {±1},
that is analytic in S ◦ and maps ∂S onto {|z| = 1} \ {±1}. Therefore, g(z) :=
f (h−1 (z)) is bounded and continuous on U and analytic in the interior U ◦ .
Moreover, because of lim|=z|→∞ f (z) = 0, limz→±1 g(z) = 0 and we can
extend g to a continuous function on {z ∈ C : |z| 6 1}. Hence, by the
maximum modulus principle (Theorem 2.4), we have
 
|g(z)| 6 max |g(ω)| = max sup |f (it)|, sup |f (1 + it)| ,
|ω|=1 t∈R t∈R
which implies the statement in this case.
2.1. Riesz-Thorin’s and Stein’s interpolation theorems - 35 -

Next, if f is a general function as in the assumption, then we consider

2
fδ,z0 (z) = eδ(z−z0 ) f (z), δ > 0, z0 ∈ S ◦ .
2 2 2
Since |eδ(z−z0 ) | 6 eδ(x −y ) with z − z0 = x + iy, −1 6 x 6 1 and y ∈ R, we
have fδ,z0 (z) → 0 as |=z| → ∞. Therefore
 
|f (z0 )| =|fδ,z0 (z0 )| 6 max sup |fδ,z0 (it)|, sup |fδ,z0 (1 + it)|
t∈R t∈R
 
δ
6e max sup |f (it)|, sup |f (1 + it)| .
t∈R t∈R
Passing to the limit δ → 0, we obtain the desired result since z0 ∈ S is arbi-
trary. 
As a corollary we obtain the following three lines theorem, which is the
basis for the proof of the Riesz-Thorin interpolation theorem and the complex
interpolation method.
Theorem 2.6 (Hadamard three lines theorem). Assume that f (z) is ana-
lytic on S ◦ and bounded and continuous on S. Then
 1−θ  θ
sup |f (θ + it)| 6 sup |f (it)| sup |f (1 + it)| ,
t∈R t∈R t∈R
for every θ ∈ [0, 1].

Proof. Denote
A0 := sup |f (it)|, A1 := sup |f (1 + it)|.
t∈R t∈R
Let λ ∈ R and define
Fλ (z) = eλz f (z).
Then by Theorem 2.5, it follows that
|Fλ (z)| 6 max(A0 , eλ A1 ).
Hence,
|f (θ + it)| 6 e−λθ max(A0 , eλ A1 )
for all t ∈ R. Choosing λ = ln A
A0
1
such that eλ A1 = A0 , we complete the
proof. 
In order to state the Riesz-Thorin theorem in a general version, we will state
and prove it in measurable spaces instead of Rn only.
Let (X, µ) be a measure space, µ always being a positive measure. We adopt
the usual convention that two functions are considered equal if they agree ex-
cept on a set of µ-measure zero. Then we denote by Lp (X, dµ) (or simply
Lp (dµ), Lp (X) or even Lp ) the Lebesgue-space of (all equivalence classes of)
scalar-valued µ-measurable functions f on X, such that
- 36 - 2. Interpolation of Operators
Z 1/p
p
kf kp = |f (x)| dµ
X
is finite. Here we have 1 6 p < ∞. In the limiting case, p = ∞, Lp consists
of all µ-measurable and bounded functions. Then we write
kf k∞ = sup |f (x)|.
X
In this section, scalars are supposed to be complex numbers.
Let T be a linear mapping from Lp = Lp (X, dµ) to Lq (Y, dν). This means
that T (αf + βg) = αT (f ) + βT (g). We shall write
T : Lp → Lq
if in addition T is bounded, i.e., if
kT f kq
A = sup
f 6=0 kf kp
is finite. The number A is called the norm of the mapping T .
It will also be necessary to treat operators T defined on several Lp spaces
simultaneously.
Definition 2.7. We define Lp1 + Lp2 to be the space of all functions f ,
such that f = f1 + f2 , with f1 ∈ Lp1 and f2 ∈ Lp2 .
Suppose now p1 < p2 . Then we observe that
Lp ⊂ Lp1 + Lp2 , ∀p ∈ [p1 , p2 ].
In fact, let f ∈ L and let γ be a fixed positive constant. Set
p

f (x), |f (x)| > γ,
f1 (x) =
0, |f (x)| 6 γ,
and f2 (x) = f (x) − f1 (x). Then
Z Z Z
p1 p p1 −p p1 −p
|f1 (x)| dx = |f1 (x)| |f1 (x)| dx 6 γ |f (x)|p dx,
since p1 − p 6 0. Similarly,
Z Z Z
p2 p p2 −p p2 −p
|f2 (x)| dx = |f2 (x)| |f2 (x)| dx 6 γ |f (x)|p dx,
so f1 ∈ Lp1 and f2 ∈ Lp2 , with f = f1 + f2 .
Now, we have the following well-known theorem.
Theorem 2.8 (The Riesz-Thorin interpolation theorem). Let T be a linear
operator with domain (Lp0 + Lp1 )(X, dµ), p0 , p1 , q0 , q1 ∈ [1, ∞]. Assume that
kT f kLq0 (Y,dν) 6 A0 kf kLp0 (X,dµ) , if f ∈ Lp0 (X, dµ),
and
kT f kLq1 (Y,dν) 6 A1 kf kLp1 (X,dµ) , if f ∈ Lp1 (X, dµ),
for some p0 6= p1 and q0 6= q1 . Suppose that for a certain 0 < θ < 1
2.1. Riesz-Thorin’s and Stein’s interpolation theorems - 37 -

1 1−θ θ 1 1−θ θ
= + , = + . (2.3)
p p0 p1 q q0 q1
Then
kT f kLq (Y,dν) 6 Aθ kf kLp (X,dµ) , if f ∈ Lp (X, dµ),
with
Aθ 6 A01−θ Aθ1 . (2.4)
Remark 2.9. 1) (2.4) means that Aθ is logarithmi-
cally convex, i.e., ln Aθ is convex. 1
q (1, 1)
2) The geometrical meaning of (2.3) is that the
( p10 , q10 )
points (1/p, 1/q) are the points on the line seg-
ment between (1/p0 , 1/q0 ) and (1/p1 , 1/q1 ).
3) The original proof of this theorem, published ( p1 , 1q )

in 1926 by Marcel Riesz, was a long and difficult

( p11 , q11 )
calculation. Riesz’ student G. Olof Thorin subse-
quently discovered a far more elegant proof and O 1
p
published it in 1939, which contains the idea be-
hind the complex interpolation method.
Proof. Denote
Z
hh, gi = h(y)g(y)dν(y)
Y
and 1/q 0 = 1 − 1/q. Then we have, by Hölder inequality,
khkq = sup |hh, gi|, and Aθ = sup |hT f, gi|.
kgkq0 =1 kf kp =kgkq0 =1

Noticing that Cc (X) is dense in Lp (X, µ) for 1 6 p < ∞, we can assume

that f and g are bounded with compact supports since p, q 0 < ∞.1 Thus, we
have |f (x)| 6 M < ∞ for all x ∈ X, and supp f = R{x ∈ X : f (x) 6= 0}
is
R compact, `i.e., µ( supp` f ) < ∞ which implies X |f (x)| dµ(x) =
`

supp f
|f (x)| dµ(x) 6 M µ( supp f ) < ∞ for any ` > 0. So g does.
For 0 6 <z 6 1, we put
1 1−z z 1 1−z z
= + , = 0
+ ,
p(z) p0 p1 q 0 (z) q0 q10
and
p f (x)
η(z) =η(x, z) = |f (x)| p(z) , x ∈ X;
|f (x)|
q0 g(y)
ζ(z) =ζ(y, z) = |g(y)| q0 (z) , y ∈ Y.
|g(y)|
Now, we prove η(z), η 0 (z) ∈ Lpj for j = 0, 1. Indeed, we have

1−1/q0
1
Otherwise, it will be p0 = p1 = ∞ if p = ∞, or θ = 1/q1 −1/q0
> 1 if q 0 = ∞.
- 38 - 2. Interpolation of Operators
p
p( 1−z + pz ) p( 1−<z + <z )+ip( =z − =z )
|η(z)| = |f (x)| p(z) = |f (x)| p
0 1 = |f (x)| p 0 p1 p1 p0

p( 1−<z + <z ) p
=|f (x)| p p 0 1 = |f (x)| p(<z) .
Thus, Z Z ppj
kη(z)kppjj = |η(x, z)| dµ(x) =pj
|f (x)| p(<z) dµ(x) < ∞.
X X
We have 0 
0 p f (x) p
η (z) =|f (x)| ln |f (x)|
p(z)
p(z) |f (x)|
 
1 1 p f (x)
=p − |f (x)| p(z) ln |f (x)|.
p1 p0 |f (x)|
On one hand, we have lim|f (x)|→0+ |f (x)|α ln |f (x)| = 0 for any α > 0, that
is, ∀ε > 0, ∃δ > 0 s.t. ||f (x)|α ln |f (x)|| < ε if |f (x)| < δ. On the other
hand, if |f (x)| > δ, then we have ||f (x)|α ln |f (x)|| 6 M α |ln |f (x)|| 6
M α max(| ln M |, | ln δ|) < ∞. Thus, ||f (x)|α ln |f (x)|| 6 C. Hence,
1 1 p
|η 0 (z)| =p − |f (x)| p(z) −α |f (x)|α |ln |f (x)||
p1 p0
p p
6C |f (x)| p(z) −α = C|f (x)| p(<z) −α ,
which yields
Z
p
kη 0
(z)kppjj 6C |f (x)|( p(<z) −α)pj dµ(x) < ∞.
X
0
Therefore, η(z), η (z) ∈ L for j = 0, 1. So ζ(z), ζ 0 (z) ∈ Lqj for j = 0, 1
0 pj

in the same way. By the linearity of T , it holds (T η)0 (z) = T η 0 (z) in view of
(2.1). It follows that T η(z) ∈ Lqj , and (T η)0 (z) ∈ Lqj with 0 < <z < 1, for
j = 0, 1. This implies the existence of
F (z) = hT η(z), ζ(z)i, 0 6 <z 6 1.
Since
dF (z) d d
Z
= hT η(z), ζ(z)i = (T η)(y, z)ζ(y, z)dν(y)
dz dz dz Y
Z Z
= (T η)z (y, z)ζ(y, z)dν(y) + (T η)(y, z)ζz (y, z)dν(y)
Y Y
0 0
=h(T η) (z), ζ(z)i + hT η(z), ζ (z)i,
F (z) is analytic on the open strip 0 < <z < 1. Moreover it is easy to see that
F (z) is bounded and continuous on the closed strip 0 6 <z 6 1.
Next, we note that for j = 0, 1
p
pj
kη(j + it)kpj = kf kp = 1.
Similarly, we also have kζ(j + it)kqj0 = 1 for j = 0, 1. Thus, for j = 0, 1
|F (j + it)| =|hT η(j + it), ζ(j + it)i| 6 kT η(j + it)kqj kζ(j + it)kqj0
2.1. Riesz-Thorin’s and Stein’s interpolation theorems - 39 -

6Aj kη(j + it)kpj kζ(j + it)kqj0 = Aj .

Using Hadamard three line theorem, reproduced as Theorem 2.6, we get the
conclusion
|F (θ + it)| 6 A1−θ θ
0 A1 , ∀t ∈ R.
Taking t = 0, we have |F (θ)| 6 A1−θ
0 A1 . We also note that η(θ) = f and
θ

ζ(θ) = g, thus F (θ) = hT f, gi. That is, |hT f, gi| 6 A01−θ Aθ1 . Therefore,
Aθ 6 A01−θ Aθ1 . 
Now, we shall give two rather simple applications of the Riesz-Thorin inter-
polation theorem.
Theorem 2.10 (Hausdorff-Young inequality). Let 1 6 p 6 2 and 1/p +
1/p0 = 1. Then the Fourier transform defined as in (1.1) satisfies
 −n/p0
|ω|
kF f kp0 6 kf kp .
2π
Proof. It follows by interpolation between the L1 -L∞ result kF f k∞ 6 kf k1
 −n/2
(cf. Theorem 1.5) and Plancherel’s theorem kF f k2 = |ω| 2π
kf k2 (cf.
Theorem 1.26). 

Theorem 2.11 (Young’s inequality for convolutions). If f ∈ Lp (Rn ) and

g ∈ Lq (Rn ), 1 6 p, q, r 6 ∞ and 1r = p1 + 1q − 1, then
kf ∗ gkr 6 kf kp kgkq .

Proof. We fix f ∈ Lp , p ∈ [1, ∞] and then will apply the Riesz-Thorin in-
terpolation theorem to the mapping g 7→ f ∗ g. Our endpoints are Hölder’s
inequality which gives
|f ∗ g(x)| 6 kf kp kgkp0
0
and thus g 7→ f ∗ g maps Lp (Rn ) to L∞ (Rn ) and the simpler version of
Young’s inequality (proved by Minkowski’s inequality) which tells us that if
g ∈ L1 , then
kf ∗ gkp 6 kf kp kgk1 .
Thus g 7→ f ∗ g also maps L1 to Lp . Thus, this map also takes Lq to Lr where
1 1−θ θ 1 1−θ θ
= + 0 , and = + .
q 1 p r p ∞
Eliminating θ, we have r = p + q − 1.
1 1 1

The condition q > 1 is equivalent with θ > 0 and r > 1 is equivalent with
the condition θ 6 1. Thus, we obtain the stated inequality for precisely the
exponents p, q and r in the hypothesis. 
- 40 - 2. Interpolation of Operators

Remark 2.12. The sharp form of Young’s inequality for convolutions can
be found in [Bec75, Theorem 3], we just state it as follows. Under the
assumption of Theorem 2.11, we have
kf ∗ gkr 6 (Ap Aq Ar0 )n kf kp kgkq ,
0
where Am = (m1/m /m01/m )1/2 for m ∈ (1, ∞), A1 = A∞ = 1 and primes
always denote dual exponents, 1/m + 1/m0 = 1.
The Riesz-Thorin interpolation theorem can be extended to the case where
the interpolated operators allowed to vary. In particular, if a family of operators
depends analytically on a parameter z, then the proof of this theorem can be
adapted to work in this setting.
We now describe the setup for this theorem. Suppose that for every z in the
closed strip S there is an associated linear operator Tz defined on the space of
simple functions on X and taking values in the space of measurable functions
on Y such that Z
|Tz (f )g|dν < ∞ (2.5)
Y
whenever f and g are simple functions on X and Y , respectively. The family
{Tz }z is said to be analytic if the function
Z
z→ Tz (f )gdν (2.6)
Y
is analytic in the open strip S ◦ and continuous on its closure S. Finally, the
analytic family is of admissible growth if there is a constant 0 < a < π and a
constant Cf,g such that
Z
−a|=z|
e ln Tz (f )gdν 6 Cf,g < ∞ (2.7)
Y
for all z ∈ S. The extension of the Riesz-Thorin interpolation theorem is now
stated.
Theorem 2.13 (Stein interpolation theorem). Let Tz be an analytic fam-
ily of linear operators of admissible growth. Let 1 6 p0 , p1 , q0 , q1 6 ∞ and
suppose that M0 and M1 are real-valued functions such that
sup e−b|t| ln Mj (t) < ∞ (2.8)
t∈R
for j = 0, 1 and some 0 < b < π. Let 0 < θ < 1 satisfy
1 1−θ θ 1 1−θ θ
= + , and = + . (2.9)
p p0 p1 q q0 q1
Suppose that
kTit (f )kq0 6 M0 (t)kf kp0 , kT1+it (f )kq1 6 M1 (t)kf kp1 (2.10)
for all simple functions f on X. Then
kTθ (f )kq 6 M (θ)kf kp , when 0 < θ < 1 (2.11)
2.2. The distribution function and weak Lp spaces - 41 -

for all simple functions f on X, where

 Z   
sin πθ ln M0 (t) ln M1 (t)
M (θ) = exp + dt .
2 R cosh πt − cos πθ cosh πt + cos πθ
By density, Tθ has a unique extension as a bounded operator from Lp (X, µ)
into Lq (Y, ν) for all p and q as in (2.9).
The proof of the Stein interpolation theorem can be obtained from that of
the Riesz-Thorin theorem simply “by adding a single letter of the alphabet”.
Indeed, the way the Riesz-Thorin theorem is proven is to study an expression
of the form
F (z) = hT η(z), ζ(z)i,
the Stein interpolation theorem proceeds by instead studying the expression
F (z) = hTz η(z), ζ(z)i.
One can then repeat the proof of the Riesz-Thorin theorem more or less verba-
tim to obtain the Stein interpolation theorem. Of course, the explicit expression
of M (θ) need an extension of the three lines theorem. For the detailed proof,
one can see [SW71, p. 205-209] or [Gra04, p.38-42].

2.2 The distribution function and weak Lp spaces

We shall now be interested in giving a concise expression for the relative

size of a function. Thus we give the following concept.
Definition 2.14. Let f (x) be a measurable function on Rn . Then the
function f∗ : [0, ∞) 7→ [0, ∞] defined by
f∗ (α) = m({x : |f (x)| > α})
is called to be the distribution function of f .
The distribution function f∗ provides information about the size of f but not
about the behavior of f itself near any given point. For instance, a function on
Rn and each of its translates have the same distribution function.
In particular, the decrease of f∗ (α) as α grows describes the relative large-
ness of the function; this is the main concern locally. The increase of f∗ (α)
as α tends to zero describes the relative smallness of the function “at infinity”;
this is its importance globally, and is of no interest if, for example, the function
is supported on a bounded set.
Now, we give some properties of distribution functions.
Proposition 2.15. For the distribution function, we have following funda-
mental properties.
(i) f∗ (α) is decreasing and continuous on the right.
- 42 - 2. Interpolation of Operators

(ii) If |f (x)| 6 |g(x)|, then f∗ (α) 6 g∗ (α).

(iii) If |f (x)| 6 lim inf k→∞ |fk (x)| for a.e. x, then f∗ (α) 6
lim inf k→∞ (fk )∗ (α) for any α > 0.
(iv) If |f (x)| 6 |g(x)| + |h(x)|, then f∗ (α1 + α2 ) 6 g∗ (α1 ) + h∗ (α2 ) for
any α1 , α2 > 0.
(v) (f g)∗ (α1 α2 ) 6 f∗ (α1 ) + g∗ (α2 ) for any α1 , α2 > 0.
R For any p p ∈ (0, ∞) and α > 0, it holds f∗ (α) 6
(vi)
−p
α {x:|f (x)|>α}
|f (x)| dx.
(vii) If fR ∈ L , p ∈ [1, ∞), then limα→+∞ αp f∗ (α) = 0 = limα→0 αp f∗ (α).
p
∞
(viii) If 0 αp−1 f∗ (α)dα < ∞, p ∈ [1, ∞), then αp f∗ (α) → 0 as α → +∞
and α → 0, respectively.

Proof. For simplicity, denote Ef (α) = {x : |f (x)| > α} for α > 0.

(i) Let {αk } is a decreasing positive sequence which tends to α, then we
have Ef (α) = ∪∞ k=1 Ef (αk ). Since {Ef (αk )} is a increasing sequence of sets,
it follows limk→∞ f∗ (αk ) = f∗ (α). This implies the continuity of f∗ (α) on the
right.
(iii) Let E = {x : |f (x)| > α} and Ek = {x : |fk (x)| > α}, k ∈ N. By the
assumption and the definition of inferior limit, i.e.,
|f (x)| 6 lim inf |fk (x)| = sup inf |fk (x)|,
k→∞ `∈N k>`
for x ∈ E, there exists an integer M such that for all k > M , |fk (x)| > α.
Thus, E ⊂ ∞
T∞
Ek , and for any ` > 1,
S
M =1
∞
! k=M
m m m m
\
Ek 6 inf (Ek ) 6 sup inf (Ek ) = lim inf (Ek ).
k>` ` k>` k→∞
k=`
Since { ∞ k=M Ek }M =1 is an increasing sequence of sets, we obtain
∞
T
∞ \ ∞ ∞
! !
m m m
[ \
f∗ (α) = (E) 6 Ek = lim Ek 6 lim inf (fk )∗ (α).
M →∞ k→∞
M =1 k=M k=M
(v) Noticing that {x : |f (x)g(x)| > α1 α2 } ⊂ {x : |f (x)| > α1 } ∪ {x :
|g(x)| > α2 }, we have the desired result.
(vi) f∗ (α) = m ({x : |f (x)| > α}) =
R
{x:|f (x)|>α}
dx 6
R |f (x)| p
{x:|f (x)|>α}
( α ) dx
−p
|f (x)|p dx.
R
=α {x:|f (x)|>α}
(vii) From (vi), it follows αp f∗ (α) 6 |f (x)|p dx 6
R

m
{x:|f (x)|>α}
|f (x)|p dx. Thus, ({x : |f (x)| > α}) → 0 as α → +∞ and
R
Rn Z
lim |f (x)|p dx = 0.
α→+∞ {x:|f (x)|>α}
Hence, α f∗ (α) → 0 as α → +∞ since αp f∗ (α) > 0.
p

For any 0 < α < β, we have, by noticing that 1 6 p < ∞, that

2.2. The distribution function and weak Lp spaces - 43 -

α→0 α→0 α→0

m
lim αp f∗ (α) = lim αp (f∗ (α) − f∗ (β)) = lim αp ({x : α < |f (x)| 6 β})
Z
6 |f (x)|p dx.
{x:|f (x)|6β}
By the arbitrariness
R α of β, it follows αp f∗ (α) → 0 as α → 0.
(viii) Since α/2 (tp )0 dt = αp − (α/2)p and f∗ (α) 6 f∗ (t) for t 6 α, we
have Z α
f∗ (α)αp (1 − 2−p ) 6 p tp−1 f∗ (t)dt
α/2
which implies the desired result.
For other ones, they are easy to verify. 
From this proposition, we can prove the following equivalent norm of Lp
spaces.
Theorem 2.16 (The equivalent norm of Lp ). Let f (x) be a measurable func-
tion in Rn , then R
1/p
∞
i) kf kp = p 0 αp−1 f∗ (α)dα , if 1 6 p < ∞,
ii) kf k∞ = inf {α : f∗ (α) = 0}.

Proof. In order to prove i), we first prove the following conclusion: If f (x) is
finite and f∗ (α) < ∞ for any α > 0, then
Z Z ∞
p
|f (x)| dx = − αp df∗ (α). (2.12)
Rn 0
Indeed, the r.h.s. of the equality is well-defined from the conditions. For the
integral in the l.h.s., we can split it into Lebesgue integral summation. Let
0 < ε < 2ε < · · · < kε < · · · and
Ej = {x ∈ Rn : (j − 1)ε < |f (x)| 6 jε} , j = 1, 2, · · · ,
m
then, (Ej ) = f∗ ((j − 1)ε) − f∗ (jε), and
∞ ∞
m
Z X X
p p
|f (x)| dx = lim (jε) (Ej ) = − lim (jε)p [f∗ (jε) − f∗ ((j − 1)ε)]
Rn ε→0 ε→0
j=1 j=1
Z ∞
=− αp df∗ (α).
0
Now we return to prove i). If the values of both sides are infinite, then it is
clearly true. If one of the integral is finite, then it is clear that f∗ (α) < +∞
and f (x) is finite almost everywhere.
R ∞ p−1 Thus (2.12) is valid.
If either f ∈ L (R ) or 0 α f∗ (α)dα < ∞ for 1 6 p < ∞ , then we
p n

always have αp f∗ (α) → 0 as α → +∞ and α → 0 from the property (vii) and

(viii) in Proposition 2.15.
Therefore, integrating by part, we have
Z ∞ Z ∞ Z ∞
p p−1 p +∞
− α df∗ (α) =p α f∗ (α)dα − α f∗ (α)|0 = p αp−1 f∗ (α)dα.
0 0 0
- 44 - 2. Interpolation of Operators

Thus, i) is true.
For ii), we have
inf {α : f∗ (α) = 0} = inf {α : m({x : |f (x)| > α}) = 0}
= inf {α : |f (x)| 6 α, a.e.}
=ess supx∈Rn |f (x)| = kf kL∞ .
We complete the proofs. 
Using the distribution function f∗ , we now introduce the weak Lp -spaces
denoted by Lp∗ .
Definition 2.17. The space Lp∗ , 1 6 p < ∞, consists of all f such that
kf kLp∗ = sup αf∗1/p (α) < ∞.
α
In the limiting case p = ∞, we put L∞ ∞
∗ = L .

By the part (iv) in Proposition 2.15 and the triangle inequality of Lp norms,
we have
kf + gkLp∗ 6 2(kf kLp∗ + kgkLp∗ ).
Thus, one can verify that Lp∗ is a quasi-normed vector space. The weak Lp
spaces are larger than the usual Lp spaces. We have the following:
Theorem 2.18. For any 1 6 p < ∞, and any f ∈ Lp , we have kf kLp∗ 6 kf kp ,
hence Lp ⊂ Lp∗ .

Proof. From the part (vi) in Proposition 2.15, we have

Z 1/p
1/p p
αf∗ (α) 6 |f (x)| dx
{x:|f (x)|>α}
which yields the desired result. 
The inclusion Lp ⊂ Lp∗ is strict for 1 6 p < ∞. For example, let h(x) =
|x|−n/p . Obviously, h is not in Lp (Rn ) but h is in Lp∗ (Rn ) and we may check
easily that
khkLp∗ = sup αh1/p
α
m
∗ (α) = sup α( ({x : |x|
α
−n/p
> α}))1/p

α
m
= sup α( ({x : |x| < α−p/n }))1/p = sup α(α−p Vn )1/p
α

=Vn1/p ,
where Vn = π R/Γ (1 + n/2) is the volume of the unit ball in Rn and Γ -
n/2
∞
function Γ (z) = 0 tz−1 e−t dt for <z > 0.
It is not immediate from their definition that the weak Lp spaces are com-
plete with respect to the quasi-norm k·kLp∗ . For the completeness, we will state
it later as a special case of Lorentz spaces.
2.3. The decreasing rearrangement and Lorentz spaces - 45 -

2.3 The decreasing rearrangement and Lorentz spaces

The spaces Lp∗ are special cases of the more general Lorentz spaces Lp,q . In
their definition, we use yet another concept, i.e., the decreasing rearrangement
of functions.
Definition 2.19. If f is a measurable function on Rn , the decreasing rear-
rangement of f is the function f ∗ : [0, ∞) 7→ [0, ∞] defined by
f ∗ (t) = inf {α > 0 : f∗ (α) 6 t} ,
where we use the convention that inf ∅ = ∞.
Now, we first give some examples of distribution function and decreasing
rearrangement. The first example establish some important relations between
a simple function, its distribution function and decreasing rearrangement.
Ex. 2.20 (Decreasing rearrangement of a simple function). Let f be a
simple function of the following form
k
X
f (x) = aj χAj (x)
j=1
where a1 > a2 > · · · > ak > 0, Aj = {x ∈ R : f (x) = aj } and χA is the
characteristic function of the set A (see Figure (a)). Then
k k
m m
X X
f∗ (α) = ({x : |f (x)| > α}) = ({x : aj χAj (x) > α}) = bj χBj (α),
j=1 j=1
Pj
m
where bj = i=1 (Ai ), Bj = [aj+1 , aj ) for j = 1, 2, · · · , k and ak+1 = 0
which shows that the distribution function of a simple function is a
simple function (see Figure (b)). We can also find the decreasing rear-
rangement (by denoting b0 = 0)
k
X
f ∗ (t) = inf{α > 0 : f∗ (α) 6 t} = inf{α > 0 : bj χBj (α) 6 t}
j=1
k
X
= aj χ[bj−1 ,bj ) (t)
j=1
which is also a simple function (see Figure (c)).
- 46 - 2. Interpolation of Operators

f (x) f∗ (α) f ∗ (t)

a1 a1

a2 a2
a3 b5 a3
b4
a4 a4
b3
a5 b2 a5
b1
A3 A4 A1 A5 A2 x a5 a4 a3 a2 a1 α b 1 b2 b3 b4 b5 t
(a) (b) (c)

Ex. 2.21. Let f : [0, ∞) 7→ [0, ∞) be

1 − (x − 1)2 , 0 6 x 6 2,

f (x) =
0, x > 2.
It is clear that f∗ (α) = 0 for α > 1 since |f (x)| 6 1. For α ∈ [0, 1], we
have
m
f∗ (α) = ({x ∈ [0, ∞) : 1 − (x − 1)2 > α})
m √ √ √
= ({x ∈ [0, ∞) : 1 − 1 − α < x < 1 + 1 − α}) = 2 1 − α.
That is,
 √
2 1 − α, 0 6 α 6 1,
f∗ (α) =
0, α > 1.
∗
The decreasing rearrangement f (t) = 0 for t > 2 since f∗ (α) 6 2 for
any α > 0. For t 6 2, we have
√
f ∗ (t) = inf{α > 0 : 2 1 − α 6 t}
= inf{α > 0 : α > 1 − t2 /4} = 1 − t2 /4.
Thus,
1 − t2 /4,

∗ 0 6 t 6 2,
f (t) =
0, t > 2.

f f∗ f∗

2 2 2

1 1 1

1 2 x 1 2 α 1 2 t
(a) (b) (c)

Observe that the integral over f , f∗ and f ∗ are all the same, i.e.,
Z ∞
√
Z 2 Z 1 Z 2
2
f (x)dx = [1 − (x − 1) ]dx = 2 1 − αdα = (1 − t2 /4)dt = 4/3.
0 0 0 0
2.3. The decreasing rearrangement and Lorentz spaces - 47 -

Ex. 2.22. We define an extended function f : [0, ∞) 7→ [0, ∞] as



 0, x = 0,
1

 ln( 1−x ), 0 < x < 1,


f (x) = ∞, 1 6 x 6 2,
1
ln( ), 2 < x < 3,


x−2



 0, x > 3.
Even if f is infinite over some interval the distribution function and the
decreasing rearrangement are still defined and can be calculated, for
any α > 0
m
f∗ (α) = ({x ∈ [1, 2] : ∞ > α} ∪ {x ∈ (0, 1) : ln(
1
1−x
) > α}
1
∪ {x ∈ (2, 3) : ln( ) > α})
x−2
m
=1 + ((1 − e−α , 1)) + ((2, e−α + 2)) m
−α
=1 + 2e ,
and 
 ∞, 0 6 t 6 1,
∗ 2
f (t) = ln( t−1 ), 1 < t < 3,
0, t > 3.


f∗ f∗
f 5 5
5
4 4 4

3 3 3

2 2 2

1 1 1

1 2 3 x 1 2 3 α 1 2 3 t
(a) (b) (c)

Ex. 2.23. Consider the function f (x) = x for all x ∈ [0, ∞). Then f∗ (α) =
m ({x ∈ [0, ∞) : x > α}) = ∞ for all α > 0, which implies that f ∗ (t) =
inf{α > 0 : ∞ 6 t} = ∞ for all t > 0.
x
Ex. 2.24. Consider f (x) = 1+x for x > 0.
It is clear that f∗ (α) = 0 for α > 1 since
|f (x)| < 1. For α ∈ [0, 1), we have
m
f∗ (α) = ({x ∈ [0, ∞) :
x
1+x
> α})
f∗
m
= ({x ∈ [0, ∞) : x >
1−α
α
}) = ∞.
1

f
That is,

∞, 0 6 α < 1, 1 2
f∗ (α) =
0, α > 1.
Thus, f ∗ (t) = inf{α > 0 : f∗ (α) 6 t} = 1.
- 48 - 2. Interpolation of Operators

Proposition 2.25. The decreasing rearrangement f ∗ of the measurable func-

tion f on Rn has the following properties:
(i) f ∗ (t) is a non-negative and non-increasing function on [0, ∞).
(ii) f ∗ (t) is right continuous on [0, ∞).
(iii) (kf )∗ = |k|f ∗ for k ∈ C.
(iv) |f | 6 |g| a.e. implies that f ∗ 6 g ∗ .
(v) (f + g)∗ (t1 + t2 ) 6 f ∗ (t1 ) + g ∗ (t2 ).
(vi) (f g)∗ (t1 + t2 ) 6 f ∗ (t1 )g ∗ (t2 ).
(vii) |f | 6 lim inf k→∞ |fk | a.e. implies that f ∗ 6 lim inf k→∞ fk∗ .
(viii) |fk | ↑ |f | a.e. implies that fk∗ ↑ f ∗ .
(ix) f ∗ (f∗ (α)) 6 α whenever f∗ (α) < ∞.
m m
(x) f∗ (f ∗ (t)) = ({|f | > f ∗ (t)}) 6 t 6 ({|f | > f ∗ (t)}) if f ∗ (t) < ∞.
(xi) f ∗ (t) > α if and only if f∗ (α) > t.
(xii) f ∗ is equimeasurable with f , that is, (f ∗ )∗ (α) = f∗ (α) for any α > 0.
(xiii) (|f |p )∗ (t) = (f ∗ (t))p for 1 6 p < ∞.
(xiv) kf ∗ kp = kf kp for 1 6 p < ∞.
(xv) kf k∞ = f ∗ (0).
(xvi) supt>0 ts f ∗ (t) = supα>0 α(f∗ (α))s for 0 < s < ∞.

Proof. (v) Assume that f ∗ (t1 ) + g ∗ (t2 ) < ∞, otherwise, there is nothing to
prove. Then for α1 = f ∗ (t1 ) and α2 = g ∗ (t2 ), by (x), we have f∗ (α1 ) 6 t1 and
g∗ (α2 ) 6 t2 . From (iv) in Proposition 2.15, it holds
(f + g)∗ (α1 + α2 ) 6 f∗ (α1 ) + g∗ (α2 ) 6 t1 + t2 .
Using the definition of the decreasing rearrangement, we have
(f +g)∗ (t1 +t2 ) = inf{α : (f +g)∗ (α) 6 t1 +t2 } 6 α1 +α2 = f ∗ (t1 )+g ∗ (t2 ).
(vi) Similar to (v), by (v) in Proposition 2.15, it holds that (f g)∗ (α1 α2 ) 6
f∗ (α1 ) + g∗ (α2 ) 6 t1 + t2 . Then, we have
(f g)∗ (t1 + t2 ) = inf{α : (f g)∗ (α) 6 t1 + t2 } 6 α1 α2 = f ∗ (t1 )g ∗ (t2 ).
(xi) If f∗ (α) > t, then by the decreasing of f∗ , we have α < inf{β : f∗ (β) 6
t} = f ∗ (t). Conversely, if f ∗ (t) > α, i.e., inf{β : f∗ (β) 6 t} > α, we get
f∗ (α) > t by the decreasing of f∗ again.
(xii) By the definition and (xi), we have
(f ∗ )∗ (α) = m({t > 0 : f (t) > α}) = m({t > 0 : f (α) > t}) = f (α).
∗
∗ ∗

(xiii) For α ∈ [0, ∞), we have

(|f |p )∗ (t) = inf{α > 0 : m({x : |f (x)| > α}) 6 t}
p

= inf{σ p > 0 : m({x : |f (x)| > σ}) 6 t} = (f (t)) ,

∗ p

where σ = α1/p .
(xiv) From Theorem 2.16, we have
2.3. The decreasing rearrangement and Lorentz spaces - 49 -
Z ∞ Z ∞
∗ ∗
kf (t)kpp = p
|f (t)| dt = p αp−1 (f ∗ )∗ (α)dα
Z0 ∞
0

=p αp−1 f∗ (α)dα = kf kpp .

0
We remain the proofs of others to interested readers. 
Having disposed of the basic properties of the decreasing rearrangement of
functions, we proceed with the definition of the Lorentz spaces.
Definition 2.26. Given f a measurable function on Rn and 1 6 p, q 6
∞, define
q dt  1q
 Z
∞
1
∗


 t p f (t) , q < ∞,
kf kLp,q = 0 t
1
 sup t p f ∗ (t), q = ∞.


t>0
The set of all f with kf kLp,q < ∞ is denoted by Lp,q (Rn ) and is called
the Lorentz space with indices p and q.
As in Lp and in weak Lp , two functions in Lp,q will be considered equal if
they are equal almost everywhere. Observe that the previous definition implies
that Lp,∞ = Lp∗ in view of (xvi) in Proposition 2.25 and Lp,p = Lp in view
of (xiv) in Proposition 2.25 for 1 6 p < ∞. By (i) and (xv) in Proposition
2.25, we have kf kL∞,∞ = supt>0 f ∗ (t) = f ∗ (0) = kf k∞ which implies that
L∞,∞ = L∞ = L∞ ∗ . Thus, we have

Theorem 2.27. Let 1 6 p 6 ∞. Then it holds, with equality of norms, that

Lp,p = Lp , Lp,∞ = Lp∗ .
Remark 2.28. For the Lorentz space Lp,q , the case when p = ∞ and 1 6
q < ∞ is not of any interest. The reason is that kf kL∞,q < ∞ implies that
f = 0 a.e. on Rn . In fact, assume that L∞,q is a non-trivial space, there
exists a nonzero function f ∈ L∞,q on a nonzero measurable set, that is,
there exists a constant c > 0 and a set E of positive measure such that
|f (x)| > c for all x ∈ E. Then, by (iv) in Proposition 2.25, we have
Z ∞ Z ∞ Z m(E)
q ∗ q dt ∗ q dt dt
kf kL∞,q = (f (t)) > [(f χE ) (t)] > cq = ∞,
t t t
m
0 0 0
∗
since (f χE ) (t) = 0 for t > (E). Hence, we have a contradiction. Thus,
f = 0 a.e. on Rn .
The next result shows that for any fixed p, the Lorentz spaces Lp,q increase
as the exponent q increases.
Theorem 2.29. Let 1 6 p 6 ∞ and 1 6 q < r 6 ∞. Then, there exists some
constant Cp,q,r such that
kf kLp,r 6 Cp,q,r kf kLp,q , (2.13)
- 50 - 2. Interpolation of Operators

where Cp,q,r = (q/p)1/q−1/r . In other words, Lp,q ⊂ Lp,r .

Proof. We may assume p < ∞ since the case p = ∞ is trivial. Since f ∗ is

non-creasing, we have
 Z t 1/q  Z t 1/q
1/p ∗ q q/p−1 ∗ q 1/p ∗ q ds
t f (t) = s ds f (t) = [s f (t)]
p 0 p 0 s
 Z t 1/q  1/q
q ds q
6 [s1/p f ∗ (s)]q 6 kf kLp,q .
p 0 s p
Hence, taking the supremum over all t > 0, we obtain
 1/q
q
kf kLp,∞ 6 kf kLp,q . (2.14)
p
This establishes (2.13) in the case r = ∞. Finally, when r < ∞, we have by
(2.14)
Z ∞ 1/r
1/p ∗ r−q+q dt
kf kLp,r = [t f (t)]
0 t
Z ∞  1q · rq
1/p ∗ (r−q)/r 1/p ∗ q dt
6 sup[t f (t)] [t f (t)]
t>0 0 t
  r−q
(r−q)/r q/r q rq
=kf kLp,∞ kf kLp,q 6 kf kLp,q .
p
This completes the proof. 
In general, Lp,q is a quasi-normed space, since the functional k·kLp,q satisfies
the conditions of normed spaces except the triangle inequality. In fact, by (v)
in Proposition 2.25, it holds
kf + gkLp,q 6 21/p (kf kLp,q + kgkLp,q ). (2.15)
However, is this space complete with respect to its quasi-norm? The next the-
orem answers this question.
Theorem 2.30. Let 1 6 p, q 6 ∞. Then the spaces Lp,q (Rn ) are complete
with respect to their quasi-norms and they are therefore quasi-Banach spaces.

Proof. See [Gra04, p. 50, Theorem 1.4.11]. 

For the duals of Lorentz spaces, we have
Theorem 2.31. Let 1 < p, q < ∞, 1/p + 1/p0 = 1 and 1/q + 1/q 0 = 1. Then
we have
0 0
(L1,1 )0 = (L1 )0 = L∞ , (L1,q )0 = {0}, (Lp,q )0 = Lp ,q .

Proof. See [Gra04, p. 52-55, Theorem 1.4.17]. 

For more results, one can see [Gra04, Kri02].
2.4. Marcinkiewicz’ interpolation theorem - 51 -

2.4 Marcinkiewicz’ interpolation theorem

We first introduce the definition of quasi-linear operators.

Definition 2.32. An operator T mapping functions on a measure space
into functions on another measure space is called quasi-linear if T (f + g)
is defined whenever T f and T g are defined and if |T (λf )(x)| 6
κ|λ||T f (x)| and |T (f + g)(x)| 6 K(|T f (x)| + |T g(x)|) for a.e. x, where κ
and K is a positive constant independent of f and g.
The idea we have used, in Definition 2.7, of splitting f into two parts ac-
cording to their respective size, is the main idea of the proof of the theorem
that follows. There, we will also use two easily proved inequalities, which are
well-known results of Hardy’s (see [HLP88, p. 245–246]):
Lemma 2.33 (Hardy inequalities). If q > 1, r > 0 and g is a measurable,
non-negative function on (0, ∞), then
Z ∞ Z t q 1/q Z ∞ 1/q
−r dt q q −r dy
g(y)dy t 6 (yg(y)) y , (2.16)
0 0 t r 0 y
Z ∞ Z ∞ q 1/q Z ∞ 1/q
r dt q q r dy
g(y)dy t 6 (yg(y)) y . (2.17)
0 t t r 0 y
Proof. To prove (2.16), we use Jensen’s inequality2 with the convex function
ϕ(x) = xq on (0, ∞). Then
Z t q Z t !q Z q
t
1 1−r/q r/q−1 r/q−1
g(y)dy = R t g(y)y y dy y dy
y r/q−1 dy
0 0 0 0
Z t q−1 Z t
r/q−1

q
6 y dy g(y)y 1−r/q y r/q−1 dy
0 0
q q−1 Z t
= tr/q (yg(y))q y r/q−1−r dy.
r 0
By integrating both sides over (0, ∞) and use the Fubini theorem, we get that
Z ∞ Z t q
g(y)dy t−r−1 dt
0 0
 q q−1 Z ∞ Z t 
−1−r/q q r/q−1−r
6 t (yg(y)) y dy dt
r 0 0
 q q−1 Z ∞ Z ∞ 
q r/q−1−r −1−r/q
= (yg(y)) y t dt dy
r 0 y
2
Jensen’s inequality: If f is any real-valued measurable function on a set Ω and ϕ is convex over
the range of f , then  Z 
1 1
Z
ϕ f (x)g(x)dx 6 ϕ(f (x))g(x)dx,
G Ω G Ω
R
where g(x) > 0 satisfies G = Ω g(x)dx > 0.
- 52 - 2. Interpolation of Operators
 q q Z ∞
= (yg(y))q y −1−r dy,
r 0
which yields (2.16) immediately.
To prove (2.17), we denote f (x) = g(1/x)/x2 . Then by taking t = 1/s and
y = 1/x, and then applying (2.16) and changing variable again by x = 1/y,
we obtain
Z ∞ Z ∞ q 1/q Z ∞ Z ∞ q 1/q
r−1 −r−1
g(y)dy t dt = g(y)dy s ds
0 t 0 1/s
Z ∞ Z s q 1/q
−r−1
= g(1/x)/x2 dx s ds
0 0
Z ∞ Z s q 1/q
−r−1
= f (x)dx s ds
0 0
Z ∞ 1/q Z ∞ 1/q
q q −r−1 q q −r−1
6 (xf (x)) x dx = (g(1/x)/x) x dx
r 0 r 0
Z ∞ 1/q
q q r−1
= (g(y)y) y dy .
r 0
Thus, we complete the proofs. 
Now, we give the Marcinkiewicz3 interpolation theorem4 and its proof due
to Hunt and Weiss in [HW64].
Theorem 2.34 (Marcinkiewicz interpolation theorem). Assume that 1 6
pi 6 qi 6 ∞, p0 < p1 , q0 6= q1 and T is a quasi-linear mapping, defined on
Lp0 + Lp1 , which is simultaneously of weak types (p0 , q0 ) and (p1 , q1 ), i.e.,
kT f kLq0 ,∞ 6 A0 kf kp0 , kT f kLq1 ,∞ 6 A1 kf kp1 . (2.18)
If 0 < θ < 1, and
1 1−θ θ 1 1−θ θ
= + , = + ,
p p0 p1 q q0 q1
then T is of type (p, q), namely
kT f kq 6 Akf kp , f ∈ Lp .
Here A = A(Ai , pi , qi , θ), but it does not otherwise depend on either T or f .
Proof. Let σ be the slope of the line segment in R2 joining (1/p0 , 1/q0 ) with
(1/p1 , 1/q1 ). Since (1/p, 1/q) lies on this segment, we can denote the slope of
this segment by
3
Józef Marcinkiewicz (1910–1940) was a Polish mathematician. He was a student of Antoni Zyg-
mund; and later worked with Juliusz Schauder, and Stefan Kaczmarz.
4
The theorem was first announced by Marcinkiewicz (1939), who showed this result to Antoni Zyg-
mund shortly before he died in World War II. The theorem was almost forgotten by Zygmund, and
was absent from his original works on the theory of singular integral operators. Later Zygmund
(1956) realized that Marcinkiewicz’s result could greatly simplify his work, at which time he pub-
lished his former student’s theorem together with a generalization of his own.
2.4. Marcinkiewicz’ interpolation theorem - 53 -

1/q0 − 1/q 1/q − 1/q1

σ= = ,
1/p0 − 1/p 1/p − 1/p1
which may be positive or negative, but is not either 0 or ∞ since q0 6= q1 and
p0 < p1 .
For any t > 0, we split an arbitrary function f ∈ Lp as follows:
f = f t + ft
where
|f (x)| > f ∗ (tσ ),

t f (x),
f (x) =
0, otherwise,
and ft = f − f t .
Then we can verify that
6 f ∗ (y), 0 6 y 6 tσ ,

t ∗
(f ) (y)
= 0, y > tσ ,
(2.19)
f ∗ (tσ ), 0 6 y 6 tσ ,

∗
(ft ) (y) 6
f ∗ (y), y > tσ .
In fact, by (iv) in Proposition 2.25, |f t | 6 |f | implies (f t )∗ (y) 6 f ∗ (y) for all
y > 0. Moreover, by the definition of f t and (x) in Proposition 2.25, we have
(f t )∗ (α) 6 (f t )∗ (f ∗ (tσ )) = f∗ (f ∗ (tσ )) 6 tσ for any α > 0, since (f t )∗ (α) =
m m
({x : |f t (x)| > α}) = ({x : |f (x)| > f ∗ (tσ ), and |f (x)| > α}) =
m m
({x : |f (x)| > f ∗ (tσ )}) = ({x : |f t (x)| > f ∗ (tσ )}) = (f t )∗ (f ∗ (tσ )) for
0 6 α 6 f ∗ (tσ ). Thus, for y > tσ , we get (f t )∗ (y) = 0. Similarly, by (iv) in
Proposition 2.25, we have (ft )∗ (y) 6 f ∗ (y) for any y > 0 since |ft | 6 |f |.
On the other hand, for y > 0, we have (ft )∗ (y) 6 (ft )∗ (0) = kft k∞ 6 f ∗ (tσ )
with the help of the non-increasing of (ft )∗ (y) and (xv) in Proposition 2.25.
Thus, (ft )∗ (y) 6 min(f ∗ (y), f ∗ (tσ )) for any y > 0 which implies (2.19).
Suppose p1 < ∞. Notice that p 6 q, because pi 6 qi . By Theorems 2.27
and 2.29, (iv) and (v) in Proposition 2.25, (2.18), and then by a change of
variables and Hardy’s inequalities (2.16) and (2.17), we get
kT f kq = kT f kLq,q 6 (p/q)1/p−1/q kT f kLq,p
 1/p−1/q Z ∞
p dt 1/p

p  1/q t ∗
6K (2t) (T f + T ft ) (2t)
q 0 t
 1/p−1/q (Z ∞ 1/p
p p dt
621/q K t (T f t )∗ (t)
 1/q
q 0 t
Z ∞ 1/p )
p dt
t (T ft )∗ (t)
 1/q 
+
0 t
 1/p−1/q (
 1/q−1/q0 t p dt 1/p
Z ∞ 
1/q p
62 K A0 t kf kp0
q 0 t
- 54 - 2. Interpolation of Operators
Z ∞ 1/p )
 1/q−1/q1 p dt
+ A1 t kft kp1
0 t
 " #p !1/p
 1/p−1/q  ∞  1−1/p0
p 1 dt
Z
621/q K A0 t1/q−1/q0 kf t kLp0 ,1
q  0 p0 t
" 1−1/p1 #p !1/p 
∞ 
1 dt
Z 
+A1 t1/q−1/q1 kft kLp1 ,1
0 p1 t 
 1/p−1/q (  1−1/p0
1/q p 1
=2 K A0
q p0
Z ∞  Z tσ p 1/p
1/q−1/q0 1/p0 ∗ dy dt
· t y f (y)
0 0 y t
 1−1/p1 Z ∞  Z ∞ p 1/p
1 1/q−1/q1 1/p1 ∗ dy dt
+ A1 t y f (y)
p1 0 tσ y t
 1−1/p1 Z ∞  Z tσ p 1/p )
1 dy dt
+ A1 t1/q−1/q1 y 1/p1 f ∗ (tσ )
p1 0 0 y t
 1/p−1/q (  1−1/p0
1/q p − p1 1
=2 K |σ| A0
q p0
Z ∞ Z s p 1/p
−p(1/p0 −1/p) 1/p0 ∗ dy ds
· s y f (y)
0 0 y s
 1−1/p1 Z ∞ Z ∞ p 1/p
1 p(1/p−1/p1 ) 1/p1 ∗ dy ds
+ A1 s y f (y)
p1 0 s y s
 1−1/p1 Z ∞ Z s p 1/p )
1 dy ds
+ A1 sp(1/p−1/p1 ) y 1/p1 f ∗ (s)
p1 0 0 y s
 1/p−1/q (  
1−1/p0

p dy 1/p
 Z ∞ 
1/q p − p1 1 1 1/p ∗
62 K |σ| A0 y f (y)
q p0 (1/p0 − 1/p) 0 y
 1−1/p1 Z ∞ 1/p
1 1
p dy
+ A1 y 1/p f ∗ (y)
p1 (1/p − 1/p1 ) 0 y
 1−1/p1 Z ∞ 1/p )
1 ds
+A1 s1−p/p1 (p1 s1/p1 f ∗ (s))p
p1 0 s
  1−1/p 1−1/p1

0
 
 A0 p 1 1
 1/p−1/q
p
 A1 p1 

1/q −1/p 0 1/p1
=2 K |σ| 1 1 + 1 1 + A p
1 1 kf kp
q 
 p0
− p p
− p1



=Akf kp .
2.4. Marcinkiewicz’ interpolation theorem - 55 -

For the case p1 = ∞ the proof is the same except for the use of the estimate
kft k∞ 6 f ∗ (tσ ), we can get
  1−1/p 
0
 1/p−1/q  A0 p
 1 
1/q p −1/p 0

A=2 K |σ| 1 + A1 .
q 
 p
− p1
0


Thus, we complete the proof. 
From the proof given above it is easy to see that the theorem can be extended
to the following situation: The underlying measure space Rn of the Lpi (Rn )
can be replaced by a general measurable space (and the measurable space oc-
curring in the domain of T need not be the same as the one entering in the
range of T ). A less superficial generalization of the theorem can be given in
terms of the notation of Lorentz spaces, which unify and generalize the usual
Lp spaces and the weak-type spaces. For a discussion of this more general
form of the Marcinkiewicz interpolation theorem see [SW71, Chapter V] and
[BL76, Chapter 5].
As an application of this powerful tool, we present a generalization of the
Hausdorff-Young inequality due to Paley. The main difference between the
theorems being that Paley introduced a weight function into his inequality and
resorted to the theorem of Marcinkiewicz. In what follows, we consider the
measure space (Rn , µ) where µ denotes the Lebesgue measure. Let w be a
weihgt function on Rn , i.e., a positive and measurable function on Rn . Then
we denote by Lp (w) the Lp -space with respect to wdx. The norm on Lp (w) is
Z 1/p
p
kf kLp (w) = |f (x)| w(x)dx .
Rn
With this notation we have the following theorem.
Theorem 2.35 (Hardy-Littlewood-Paley theorem on Rn ). Assume that
1 6 p 6 2. Then
kF f kLp (|ξ|−n(2−p) ) 6 Cp kf kp .

Proof. We considering the mapping (T f )(ξ) = |ξ|n fˆ(ξ). By Plancherel theo-

rem, we have
kT f kL2 (|ξ|−2n ) 6 kT f kL2 (|ξ|−2n ) = kfˆk2 6 Ckf k2 ,
∗

which implies that T is of weak type (2, 2). We now work towards showing
that T is of weak type (1, 1). Thus, the Marcinkiewicz interpolation theorem
implies the theorem.
Now, consider the set Eα = {ξ : |ξ|n fˆ(ξ) > α}. For simplicity, we let ν
denote the measure |ξ|−2n dξ and assume that kf k1 = 1. Then, |fˆ(ξ)| 6 1. For
ξ ∈ Eα , we therefore have α 6 |ξ|n . Consequently,
- 56 - 2. Interpolation of Operators
Z Z
−2n
(T f )∗ (α) = ν(Eα ) = |ξ| dξ 6 |ξ|−2n dξ 6 Cα−1 .
Eα |ξ|n >α
Thus, we proves that
α · (T f )∗ (α) 6 Ckf k1 ,
which implies T is of weak type (1, 1). Therefore, we complete the proof. 
L ECTURE N OTES ON c 2016 by Chengchun Hao
Introduction to Harmonic Analysis Email: hcc@amss.ac.cn

Chapter 3
The Maximal Function and Calderón-Zygmund
Decomposition

3.1 Two covering lemmas

Lemma 3.1 (Finite version of Vitali covering lemma). Suppose B =

{B1 , B2 , · · · , BN } is a finite collection of open balls in Rn . Then, there exists
a disjoint sub-collection Bj1 , Bj2 , · · · , Bjk of B such that
N
! k
m
[ X
n
m B` 6 3 (Bji ).
`=1 i=1

Proof. The argument we give is constructive and relies on the following simple
observation: Suppose B and B 0 are a pair of balls that intersect, with the
radius of B 0 being not greater than that of B. Then B 0 is contained in the ball
B̃ that is concentric with B but with 3 times its radius. (See Fig 3.1.)
As a first step, we pick a ball Bj1 in B with maximal
(i.e., largest) radius, and then delete from B the ball Bj1 as
B0
well as any balls that intersect Bj1 . Thus all the balls that
are deleted are contained in the ball B̃j1 concentric with B
Bj1 , but with 3 times its radius.
The remaining balls yield a new collection B 0 , for which B̃

we repeat the procedure. We pick Bj2 and any ball that

Fig. 3.1 The balls B
intersects Bj2 . Continuing this way, we find, after at mostFigure 1: The balls B and B̃
and B̃
N steps, a collection of disjoint balls Bj1 , Bj2 , · · · , Bjk .
Finally, to prove that this disjoint collection of balls sat-
isfies the inequality in the lemma, we use the observation made at the begin-
ning of the proof. Let B̃ji denote the ball concentric with Bji , but with 3 times
its radius. Since any ball B in B must intersect a ball Bji and have equal or
smaller radius than Bji , we must have ∪B∩Bji 6=∅ B ⊂ B̃ji , thus
N
! k
! k k
m m
[ [ X X
m B` 6 m B̃ji 6 (B̃ji ) = 3n (Bji ).
`=1 i=1 i=1 i=1

57
- 58 - 3. The Maximal Function and Calderón-Zygmund Decomposition

In the last step, we have used the fact that in Rn a dilation of a set by δ > 0
results in the multiplication by δ n of the Lebesgue measure of this set. 
For the infinite version of Vitali covering lemma, one can see the textbook
[Ste70, the lemma on p.9].
The decomposition of a given set into a disjoint union of cubes (or balls) is
a fundamental tool in the theory described in this chapter. By cubes we mean
closed cubes; by disjoint we mean that their interiors are disjoint. We have in
mind the idea first introduced by Whitney and formulated as follows.
Theorem 3.2 (Whitney covering lemma). Let F be a non-empty closed
set in Rn and Ω be its complement. Then there exists a collection of cubes
F = {Qk } whose sides are parallel to the axes, such that
(i) ∞ c
S
k=1 Qk = Ω = F ,
(ii) Q◦j ∩ Q◦k = ∅ if j 6= k, where Q◦ denotes the interior of Q,
(iii) there exist two constants c1 , c2 > 0 independent of F (In fact we may
take c1 = 1 and c2 = 4.), such that
c1 diam (Qk ) 6 dist (Qk , F ) 6 c2 diam (Qk ).

Proof. Consider the lattice of points in Rn whose coordinates are integers.

This lattice determines a mesh M0 , which is a collection of cubes: namely
all cubes of unit length, whose vertices are points of the above lattice. The
mesh M0 leads to a two-way infinite chain of such meshes {Mk }∞ −∞ , with
Mk = 2 M0 .
−k

Thus each cube in the mesh Mk 62 Chapter 4 Calderón-Zygmund Decomposition

gives rise to 2n cubes in the mesh

Mk+1 by bisecting the sides. The
M −1
cubes in the mesh Mk each have M M 0 k
MM k
sides of length 2 and are thus of di-
−k
Q
1

√ O 1 2 3
ameter n2−k .
F
In addition to the meshes Mk , we Ωk+1

consider the layers Ωk , defined by Ωk

Ωk = x : c2−k < dist (x, F ) 6 c2−k+1


Fig. , 3.2 Meshes and layers: M0 with dashed
Fig. 4.1 Meshes and layers: M0 with dashed (green) lines; M1 with dotted lines; M−1 with solid (blue) lines

where c is a positive constant which (green) lines; M 6 dist

diam (Q) 1
with dotted lines; M
(Q, F) 6 4 diam (Q), Q ∈ F . −1
with (4.1)
0
solid (blue) lines √
we shall fix momentarily. Obviously,Let us prove (4.1) first. Suppose Q ∈ M ; then diam (Q) = n2 . Since Q ∈ F , there exists an
x ∈ Q ∩ Ω . Thus dist (Q, F) 6 dist (x, F) 6 c2
k
k
−k+1
, and dist (Q, F) > dist (x, F) − diam (Q) >
−k
0

S∞ √ √
Ω = k=−∞ Ωk .
−k −k
c2 − n2 . If we choose c = 2 n we get (4.1).
Then by (4.1) the cubes Q ∈ F are disjoint from F and clearly cover Ω. Therefore, (i) is also
0

Now we make an initial choice of cubes, and denote the resulting collection
proved.
Notice that the collection F has all our required properties, except
0 that the cubes in it are not
necessarily disjoint. To finish the proof of the theorem, we need to refine our choice leading to F ,
by F0 . Our choice is made as follows. Wethose consider the
reallycubes
unnecessary. of the mesh Mk ,
0
eliminating cubes which were
We require the following simple observation. Suppose Q and Q are two cubes (taken respectively 1 2

(each such cube is of size approximately 2−k

from the mesh M ), andand
M ). Theninclude
if Q and Q a
k1
in the other. (In particular, Q ⊂ Q , if k > k .)
arecube
k2not disjoint,of
one this mesh
of the two
1 must be contained
2
1 2 1 2

in F0 if it intersects Ωk , (the points of the latter are all approximately atcontains

Start now with any cube Q ∈ F , and consider the maximal cube
0 in F which
0
view of the inequality (4.1), for any cube Q ∈ F which contains Q ∈ F , we have diam (Q ) 6
a it. In
0 0
0
0

distance 2−k from F ). Namely, 0

dist (Q , F) 6 dist(Q, F) 6 4 diam (Q). Moreover, any two cubes Q and Q which contain Q have
obviously a non-trivial intersection. Thus by the observation made above each cube Q ∈ F has
0 00

0
a unique maximal cube in F0 which contains it. By the same taken these maximal cubes are also
disjoint. We let F denote the collection of maximal cubes of F0 . Then obviously
S
(i) Q∈F Q = Ω,
(ii) The cubes of F are disjoint,
(iii) diam (Q) 6 dist (Q, F) 6 4 diam (Q), Q ∈ F .
Therefore, we complete the proof. t
u
3.2. Hardy-Littlewood maximal function - 59 -
[
F0 = {Q ∈ Mk : Q ∩ Ωk 6= ∅} .
k
We then have
[
Q = Ω.
Q∈F 0
For appropriate choice of c, we claim that
diam (Q) 6 dist (Q, F ) 6 4 diam (Q), Q ∈ F0 . (3.1)
√ −k
Let us prove (3.1) first. Suppose Q ∈ Mk ; then diam (Q) = n2 . Since
Q ∈ F0 , there exists an x ∈ Q ∩ Ωk . Thus dist (Q, F ) 6 dist (x, F ) 6
√
c2−k+1 , and dist (Q, F ) > dist (x, F ) − diam (Q) > c2−k − n2−k . If we
√
choose c = 2 n we get (3.1).
Then by (3.1) the cubes Q ∈ F0 are disjoint from F and clearly cover Ω.
Therefore, (i) is also proved.
Notice that the collection F0 has all our required properties, except that the
cubes in it are not necessarily disjoint. To finish the proof of the theorem, we
need to refine our choice leading to F0 , eliminating those cubes which were
really unnecessary.
We require the following simple observation. Suppose Q1 and Q2 are two
cubes (taken respectively from the mesh Mk1 and Mk2 ). Then if Q1 and Q2
are not disjoint, one of the two must be contained in the other. (In particular,
Q1 ⊂ Q2 , if k1 > k2 .)
Start now with any cube Q ∈ F0 , and consider the maximal cube in F0
which contains it. In view of the inequality (3.1), for any cube Q0 ∈ F0 which
contains Q ∈ F0 , we have diam (Q0 ) 6 dist (Q0 , F ) 6 dist (Q, F ) 6
4 diam (Q). Moreover, any two cubes Q0 and Q00 which contain Q have ob-
viously a non-trivial intersection. Thus by the observation made above each
cube Q ∈ F0 has a unique maximal cube in F0 which contains it. By the
same taken these maximal cubes are also disjoint. We let F denote the collec-
tion of maximal cubes of F0 . Then obviously
(i) Q∈F Q = Ω,
S

(ii) The cubes of F are disjoint,

(iii) diam (Q) 6 dist (Q, F ) 6 4 diam (Q), Q ∈ F .
Therefore, we complete the proof. 

3.2 Hardy-Littlewood maximal function

Maximal functions appear in many forms in harmonic analysis. One of the

most important of these is the Hardy-Littlewood maximal function. They play
an important role in understanding, for example, the differentiability properties
- 60 - 3. The Maximal Function and Calderón-Zygmund Decomposition

of functions, singular integrals and partial differential equations. They often

provide a deeper and more simplified approach to understanding problems in
these areas than other methods.
First, we consider the differentiation of the integral for one-dimensional
functions. If f is given on [a, b] and integrable on that interval, we let
Z x
F (x) = f (y)dy, x ∈ [a, b].
a
To deal with F 0 (x), we recall the definition of the derivative as the limit of the
quotient F (x+h)−F
h
(x)
when h tends to 0, i.e.,
F (x + h) − F (x)
F 0 (x) = lim .
h→0 h
We note that this quotient takes the form (say in the case h > 0)
1 x+h 1
Z Z
f (y)dy = f (y)dy,
h x |I| I
where we use the notation I = (x, x + h) and |I| for the length of this interval.
At this point, we pause to observe that the above expression in the “average”
value of f over I, and that in the limit as |I| → 0, we might expect that
these averages tend to f (x). Reformulating the question slightly, we may ask
whether
1
Z
lim f (y)dy = f (x)
|I|→0 |I|
x∈I I

holds for suitable points x. In higher dimensions we can pose a similar ques-
tion, where the averages of f are taken over appropriate sets that generalize
the intervals in one dimension.
In particular, we can take the sets involved as the ball B(x, r) of radius r,
m
centered at x, and denote its measure by (B(x, r)). It follows
1
Z
lim
r→0 m (B(x, r)) B(x,r)
f (y)dy = f (x), for a.e. x? (3.2)

Let us first consider a simple case, when f is continuous at x, the limit

does converge to f (x). Indeed, given ε > 0, there exists a δ > 0 such that
|f (x) − f (y)| < ε whenever |x − y| < δ. Since
1 1
Z Z
f (x) −
m (B(x, r)) B(x,r)
f (y)dy =
m (B(x, r)) B(x,r)
(f (x) − f (y))dy,

we find that whenever B(x, r) is a ball of radius r < δ, then

1 1
Z Z
f (x) −
m (B(x, r)) B(x,r)
f (y)dy 6
m(B(x, r)) B(x,r)
|f (x)−f (y)|dy < ε,

as desired.
In general, for this “averaging problem” (3.2), we shall have an affirmative
answer. In order to study the limit (3.2), we consider its quantitative analogue,
3.2. Hardy-Littlewood maximal function - 61 -

where “limr→0 ” is replaced by “supr>0 ”, this is the (centered) maximal func-

tion. Since the properties of this maximal function are expressed in term of
relative size and do not involve any cancelation of positive and negative val-
ues, we replace f by |f |.
Definition 3.3. If f is locally integrable1 on Rn , we define its maximal
function M f : Rn → [0, ∞] by
1
Z
M f (x) = sup
r>0 m
(B(x, r)) B(x,r)
|f (y)|dy, x ∈ Rn . (3.3)

Moreover, M is also called as the Hardy-Littlewood maximal operator.

The maximal function that we consider arose first in the one-dimensional
situation treated by Hardy and Littlewood.2 It is to be noticed that nothing
excludes the possibility that M f (x) is infinite for any given x.
It is immediate from the definition that
Theorem 3.4. If f ∈ L∞ (Rn ), then M f ∈ L∞ (Rn ) and
kM f k∞ 6 kf k∞ .
By the previous statements, if f is continuous at x, then we have
1
Z
|f (x)| = lim
r→0 m
(B(x, r)) B(x,r)
|f (y)|dy

1
Z
6 sup
r>0 m
(B(x, r)) B(x,r)
|f (y)|dy = M f (x).

Thus, we have proved

Theorem 3.5. If f ∈ C(Rn ), then
|f (x)| 6 M f (x)
n
for all x ∈ R .
Sometimes, we will define the maximal function with cubes in place of
balls. If Q(x, r) is the cube [xi − r, xi + r]n , define
1
Z
0
M f (x) = sup n
|f (y)|dy, x ∈ Rn . (3.4)
r>0 (2r) Q(x,r)
When n = 1, M and M 0 coincide. If n > 1, then there exist constants cn and
Cn , depending only on n, such that
cn M 0 f (x) 6 M f (x) 6 Cn M 0 f (x). (3.5)

1
A measurable function f on Rn is called to be locally integrable, if for every ball B the function
f (x)χB (x) is integrable. We shall denote by L1loc (Rn ) the space of all locally integrable functions.
Loosely speaking, the behavior at infinity does not affact the local integrability of a function. For
example, the functions e|x| and |x|−1/2 are both locally integrable, but not integrable on Rn .
2
The Hardy-Littlewood maximal operator appears in many places but some of its most notable uses
are in the proofs of the Lebesgue differentiation theorem and Fatou’s theorem and in the theory of
singular integral operators.
- 62 - 3. The Maximal Function and Calderón-Zygmund Decomposition

Thus, the two operators M and M 0 are essentially interchangeable, and we

will use whichever is more appropriate, depending on the circumstances. In
addition, we can define a more general maximal function
1
Z

m
00
M f (x) = sup |f (y)|dy, (3.6)
Q3x (Q) Q
where the supremum is taken over all cubes containing x. Again, M 00 is point-
wise equivalent to M . One sometimes distinguishes between M 0 and M 00 by
referring to the former as the centered and the latter as the non-centered max-
imal operator. Alternatively, we could define the non-centered maximal func-
tion with balls instead of cubes:
1
Z
M̃ f (x) = sup
B3x m
(B) B
|f (y)|dy
at each x ∈ Rn . Here, the supremum is taken over balls B in Rn which contain
m
the point x and (B) denotes the measure of B (in this case a multiple of the
radius of the ball raised to the power n).
Ex. 3.6. Let f : R → R, f (x) = χ(0,1) (x). Then
 1
 2x , x > 1,
0
M f (x) =M f (x) = 1, 0 6 x 6 1,
 1
2(1−x)
, x < 0,
1
 x , x > 1,
00
M̃ f (x) =M f (x) = 1, 0 6 x 6 1,
 1
1−x
, x < 0.
In fact, for x > 1, we get
Z x+h
0 1
M f (x) = M f (x) = sup χ(0,1) (y)dy
h>0 2h x−h
1−x+h
 
1 1
= max sup , sup = ,
x−h>0 2h x−h60 2h 2x
Z x+h2
00 1
M̃ f (x) = M f (x) = sup χ(0,1) (y)dy
h1 ,h2 >0 h1 + h2 x−h1

1 − x + h1
 
1 1
= max sup , sup = .
0<x−h1 <1 h1 x−h1 60 h1 x
For 0 6 x 6 1, it follows
Z x+h
0 1
M f (x) = M f (x) = sup χ(0,1) (y)dy
h>0 2h x−h
1−x+h

2h
= max sup , sup ,
0<x−h<x+h<1 2h 0<x−h<16x+h 2h

x+h 1
sup , sup
x−h60<x+h<1 2h x−h60<16x+h 2h
3.2. Hardy-Littlewood maximal function - 63 -
  
1 1 1
= max 1, 1, 1, min , = 1,
2 x 1−x
Z x+h2
00 1
M̃ f (x) = M f (x) = sup χ(0,1) (y)dy
h1 ,h2 >0 h1 + h2 x−h1

h1 + h2 x + h2
= max sup , sup ,
0<x−h1 <x+h2 <1 h1 + h2 x−h1 <0<x+h2 <1 h1 + h2
1 − x + h1

1
sup , sup
0<x−h1 <1<x+h2 h1 + h2 x−h1 <0<1<x+h2 h1 + h2

=1.
For x < 0, we have
 
0 x+h 1 1
M f (x) = M f (x) = max sup , sup = ,
0<x+h<1,h>0 2h x+h>1 2h 2(1 − x)
 
00 x + h2 1
M̃ f (x) = M f (x) = max sup , sup
h1 ,h2 >0,0<x+h2 <1 h1 + h2 h1 >0,x+h2 >1 h1 + h2
1
= .
1−x
Observe that f ∈ L1 (R), but M f, M 0 f, M 00 f, M̃ f ∈
/ L1 (R).
Remark 3.7. (i) M f is defined at every point x ∈ Rn and if f = g a.e.,
then M f (x) = M g(x) at every x ∈ Rn .
(ii) It may be well that M f = ∞ for every x ∈ Rn . For example, let
n = 1 and f (x) = x2 .
(iii) There are several definitions in the literature which are often
equivalent.
Next, we state some immediate properties of the maximal function. The
proofs are left to interested readers.
Proposition 3.8. Let f, g ∈ L1loc (Rn ). Then
(i) Positivity: M f (x) > 0 for all x ∈ Rn .
(ii) Sub-linearity: M (f + g)(x) 6 M f (x) + M g(x).
(iii) Homogeneity: M (αf )(x) = |α|M f (x), α ∈ R.
(iv) Translation invariance: M (τy f ) = (τy M f )(x) = M f (x − y).
With the Vitali covering lemma, we can state and prove the main results for
the maximal function.
Theorem 3.9 (The maximal function theorem). Let f be a given function
defined on Rn .
(i) If f ∈ Lp (Rn ), p ∈ [1, ∞], then the function M f is finite almost every-
where.
(ii) If f ∈ L1 (Rn ), then for every α > 0, M is of weak type (1, 1), i.e.,
- 64 - 3. The Maximal Function and Calderón-Zygmund Decomposition

m({x : M f (x) > α}) 6 3α kf k .

n
1

(iii) If f ∈ Lp (Rn ), p ∈ (1, ∞], then M f ∈ Lp (Rn ) and

kM f kp 6 Ap kf kp ,
n
where Ap = 3 p/(p − 1) + 1 for p ∈ (1, ∞) and A∞ = 1.
Proof. We first prove the second one, i.e., (ii). Denote
Eα = {x : M f (x) > α} ,
then from the definitions of M f and the supremum, for each x ∈ Eα and
0 < ε < M f (x) − α, there exists a r > 0 such that
1
Z

m
(B(x, r)) B(x,r)
|f (y)|dy > M f (x) − ε > α.

We denote that ball B(x, r) by Bx that contains x. Therefore, for each Bx , we

have
m 1
Z
(Bx ) < |f (y)|dy. (3.7)
α Bx
Fix a compact subset K of Eα . Since K is covered by ∪x∈Eα Bx , by Heine-
Borel theorem,3 we may select a finite subcover of K, say K ⊂ N `=1 B` .
S

Lemma 3.1 guarantees the existence of a sub-collection Bj1 , · · · , Bjk of dis-

joint balls with
N k
m m
[ X
( B` ) 6 3n (Bji ). (3.8)
`=1 i=1
Since the balls Bj1 , · · · , Bjk are disjoint and satisfy (3.7) as well as (3.8), we
find that
N k k Z
m m m 3n X
[ X
n
(K) 6 ( B` ) 6 3 (Bji ) 6 |f (y)|dy
`=1 i=1
α i=1 B ji

3n 3n
Z Z
= |f (y)|dy 6 |f (y)|dy.
α Ski=1 Bji α Rn
Since this inequality is true for all compact subsets K of Eα , the proof of the
weak type inequality (ii) for the maximal operator is complete.
The above proof also gives the proof of (i) for the case when p = 1. For the
case p = ∞, by Theorem 3.4, (i) and (iii) is true with A∞ = 1.
Now, by using the Marcinkiewicz interpolation theorem between L1 →
L1,∞ and L∞ → L∞ , we can obtain simultaneously (i) and (iii) for the case
p ∈ (1, ∞). 
Now, we make some clarifying comments.
3
The Heine-Borel theorem reads as follows: A set K ⊂ Rn is closed and bounded if and only if K is a
compact set (i.e., every open cover of K has a finite subcover). In words, any covering of a compact
set by a collection of open sets contains a finite sub-covering. For the proof, one can see the wiki:
http://en.wikipedia.org/wiki/Heine%E2%80%93Borel_theorem.
3.2. Hardy-Littlewood maximal function - 65 -

Remark 3.10. (1) The weak type estimate (ii) is the best possible for the
distribution function of M f , where f is an arbitrary function in L1 (Rn ).
Indeed, we replace |f (y)|dy in the definition of (3.3) by a Dirac mea-
sure dµ whose
R total measure of one is concentrated at the origin. The
integral B(x,r) dµ = 1 only if the ball B(x, r) contains the origin; other-
wise, it will be zeros. Thus,
1
M (dµ)(x) = sup
r>0, 0∈B(x,r) m
(B(x, r))
= (Vn |x|n )−1 ,

i.e., it reaches the supremum when r = |x|. Hence, the distribution

function of M (dµ) is
m m
(M (dµ))∗ (α) = ({x : |M (dµ)(x)| > α}) = ( x : (Vn |x|n )−1 > α )


m m
= ( x : Vn |x|n < α−1 ) = (B(0, (Vn α)−1/n ))

−1
=Vn (Vn α) = 1/α.
But we can always find a sequence {fm (x)} of positive integrable func-
tions, whose L1 norm is each 1, and which converges weakly to the
measure dµ. So we cannot expect an estimate essentially stronger than
the estimate (ii) in Theorem 3.9, since, in the limit, a similar stronger
version would have to hold for M (dµ)(x).
(2) It is useful, for certain applications, to observe that
 
1
Ap = O , as p → 1.
p−1
In contrast with the case p > 1, when p = 1 the mapping f 7→ M f is not
bounded on L1 (Rn ). So the proof of the weak bound (ii) for M f requires a less
elementary arguments of geometric measure theory, like the Vitali covering
lemma. In fact, we have
Theorem 3.11. If f ∈ L1 (Rn ) is not identically zero, then M f is never inte-
grable on the whole of Rn , i.e., M f ∈
/ L1 (Rn ).

Proof. We can choose an N large enough such that

1
Z
|f (x)|dx > kf k1 .
B(0,N ) 2
Then, we take an x ∈ R such that |x| > N . Let r = 2(|x| + N ), we have
n

1 1
Z Z
M f (x) >
m (B(x, r)) B(x,r)
|f (y)|dy =
Vn (2(|x| + N ))n B(x,r)
|f (y)|dy

1 1
Z
> |f (y)|dy > kf k1
Vn (2(|x| + N ))n B(0,N ) 2Vn (2(|x| + N ))n
1
> kf k1 .
2Vn (4|x|)n
It follows that for sufficiently large |x|, we have
- 66 - 3. The Maximal Function and Calderón-Zygmund Decomposition

M f (x) > c|x|−n , c = (2Vn 4n )−1 kf k1 .

This implies that M f ∈
/ L1 (Rn ). 
Moreover, even if we limit our consideration to any bounded subset of Rn ,
then the integrability of M f holds only if stronger conditions than the integra-
bility of f are required. In fact, we have
Theorem 3.12. Let E be a bounded subset of Rn . If f ln+ |f | ∈ L1 (Rn ) and
supp f ⊂ E, then
m
Z Z
M f (x)dx 6 2 (E) + C |f (x)| ln+ |f (x)|dx,
E E
+
where ln t = max(ln t, 0).

Proof. By Theorem 2.16, it follows that

m
Z Z ∞
M f (x)dx =2 ({x ∈ E : M f (x) > 2α})dα
E 0

m
Z 1 Z ∞ 
=2 + ({x ∈ E : M f (x) > 2α})dα
0 1

m m
Z ∞
62 (E) + 2 ({x ∈ E : M f (x) > 2α})dα.
1
Decompose f as f1 + f2 , where f1 = f χ{x:|f (x)|>α} and f2 = f − f1 . Then, by
Theorem 3.4, it follows that
M f2 (x) 6 kM f2 k∞ 6 kf2 k∞ 6 α,
which yields
{x ∈ E : M f (x) > 2α} ⊂ {x ∈ E : M f1 (x) > α}.
Hence, by Theorem 3.9, we have
m m
Z ∞ Z ∞
({x ∈ E : M f (x) > 2α})dα 6 ({x ∈ E : M f1 (x) > α})dα
1 1
Z ∞ Z Z max(1,|f (x)|)
1 dα
Z
6C |f (x)|dxdα 6 C |f (x)| dx
1 α {x∈E:|f (x)|>α} E 1 α
Z
=C |f (x)| ln+ |f (x)|dx.
E
This completes the proof. 
As a corollary of Theorem 3.9, we have the differentiability almost every-
where of the integral, expressed in (3.2).
Theorem 3.13 (Lebesgue differentiation theorem). If f ∈ Lp (Rn ), p ∈
[1, ∞], or more generally if f is locally integrable (i.e., f ∈ L1loc (Rn )), then
1
Z
lim
r→0 m
(B(x, r)) B(x,r)
f (y)dy = f (x), for a.e. x. (3.9)
3.2. Hardy-Littlewood maximal function - 67 -

Proof. We first consider the case p = 1. It suffices to show that for each α > 0,
the set  
1
Z
Eα = x : lim sup
r→0 m
(B(x, r)) B(x,r)
f (y)dy − f (x) > 2α

has measure zero, because this assertion then guarantees that the set E =
S∞
k=1 E1/k has measure zero, and the limit in (3.9) holds at all points of E .
c

Fix α, since the continuous functions of compact support are dense in

L (Rn ), for each ε > 0 we may select a continuous function g of compact
1

support with kf − gk1 < ε. As we remarked earlier, the continuity of g implies

that
1
Z
lim
r→0 m
(B(x, r)) B(x,r)
g(y)dy = g(x), for all x.

Since we may write the difference m(B(x,r))

1
f (y)dy − f (x) as
R
B(x,r)
1
Z

m
(B(x, r)) B(x,r)
(f (y) − g(y))dy

1
Z
+
m
(B(x, r)) B(x,r)
g(y)dy − g(x) + g(x) − f (x),

we find that
1
Z
lim sup
r→0 m
(B(x, r)) B(x,r)
f (y)dy − f (x) 6 M (f − g)(x) + |g(x) − f (x)|.

Consequently, if
Fα = {x : M (f − g)(x) > α} and Gα = {x : |f (x) − g(x)| > α} ,
then Eα ⊂ Fα ∪ Gα , because if u1 and u2 are positive, then u1 + u2 > 2α only
if ui > α for at least one ui .
On the one hand, Tchebychev’s inequality4 yields
m 1
(Gα ) 6 kf − gk1 ,
α
and on the other hand, the weak type estimate for the maximal function gives
m 3n
(Fα ) 6 kf − gk1 .
α
Since the function g was selected so that kf − gk1 < ε, we get
m 3n
(Eα ) 6 ε + ε =
α
1
α
3n + 1
α
ε.
m
Since ε is arbitrary, we must have (Eα ) = 0, and the proof for p = 1 is
completed.

4
Tchebychev inequality (also spelled as Chebyshev’s inequality): Suppose f > 0, and f is integrable.
If α > 0 and Eα = {x ∈ Rn : f (x) > α}, then

m 1
Z
(Eα ) 6 f dx.
α Rn
- 68 - 3. The Maximal Function and Calderón-Zygmund Decomposition

Indeed, the limit in the theorem is taken over balls that shrink to the point
x, so the behavior of f far from x is irrelevant. Thus, we expect the result to
remain valid if we simply assume integrability of f on every ball. Clearly, the
conclusion holds under the weaker assumption that f is locally integrable.
For the remained cases p ∈ (1, ∞], we have by Hölder inequality, for any
ball B,
m(B)
Z
1/p0
|f (x)|dx 6 kf kLp (B) k1kLp0 (B) 6 kf kp .
B
Thus, f ∈ L1loc (Rn ) and then the conclusion is valid for p ∈ (1, ∞]. Therefore,
we complete the proof of the theorem. 
By the Lebesgue differentiation theorem, we have
Theorem 3.14. Let f ∈ L1loc (Rn ). Then
|f (x)| 6 M f (x), a.e. x ∈ Rn .
Combining with the maximal function theorem (i.e., Theorem 3.9), we get
Corollary 3.15. If f ∈ Lp (Rn ), p ∈ (1, ∞], then we have
kf kp 6 kM f kp 6 Ap kf kp .
As an application, we prove the (Gagliardo-Nirenberg-) Sobolev inequality
by using the maximal function theorem for the case 1 < p < n. We note that
the inequality also holds for the case p = 1 and one can see [Eva98, p.263-264]
for the proof.
Theorem 3.16 ((Gagliardo-Nirenberg-) Sobolev inequality). Let p ∈
(1, n) and its Sobolev conjugate p∗ = np/(n − p). Then for f ∈ D(Rn ),
we have
kf kp∗ 6 Ck∇f kp ,
where C depends only on n and p.

Proof. Since f ∈ D(Rn ), we have

∞
∂
Z
f (x) = − f (x + rz)dr,
0 ∂r
where z ∈ S n−1 . Integrating this over the whole unit sphere surface S n−1
yields
Z ∞
∂
Z Z
ωn−1 f (x) = f (x)dσ(z) = − f (x + rz)drdσ(z)
S n−1 S n−1 0 ∂r
Z Z ∞
=− ∇f (x + rz) · zdrdσ(z)
S n−1 0
Z ∞Z
=− ∇f (x + rz) · zdσ(z)dr.
0 S n−1
3.2. Hardy-Littlewood maximal function - 69 -

Changing variables y = x + rz, dσ(z) = r−(n−1) dσ(y), z = (y − x)/|y − x|

and r = |y − x|, we get
Z ∞Z
y−x
ωn−1 f (x) = − ∇f (y) · dσ(y)dr
0 ∂B(x,r) |y − x|n
y−x
Z
=− ∇f (y) · dy,
Rn |y − x|n
which implies that
1 |∇f (y)|
Z
|f (x)| 6 dy.
ωn−1 Rn |y − x|n−1
We split this integral into two parts as Rn = B(x,r) + Rn \B(x,r) . For the
R R R

first part, we have

1 |∇f (y)|
Z
dy
ωn−1 B(x,r) |x − y|n−1
∞ Z
1 X |∇f (y)|
= dy
ωn−1 k=0 B(x,2−k r)\B(x,2−k−1 r) |x − y|n−1
∞ Z
1 X |∇f (y)|
6 dy
ωn−1 k=0 B(x,2−k r)\B(x,2−k−1 r) (2−k−1 r)n−1
∞
2−k r n−1 |∇f (y)|
X Z
6 −k
2 −k r)n−1
dy
k=0
nV n 2 r B(x,2−k r) (2
∞
1 X −k+n−1 1
Z
6
n k=0
2 r
m
(B(x, 2−k r)) B(x,2−k r)
|∇f (y)|dy

∞
2n−1 X 2n
6 rM (∇f )(x) 2−k = rM (∇f )(x).
n k=0
n
For the second part, by Hölder inequality, we get for 1 < p < n
|∇f (y)|
Z
n−1
dy
Rn \B(x,r) |x − y|
Z 1/p Z 1/p0
p (1−n)p0
6 |∇f (y)| dy |x − y| dy
Rn \B(x,r) Rn \B(x,r)
 Z ∞ 1/p0
(1−n)p0 n−1
6 ωn−1 ρ ρ dρ k∇f kp
r
1/p0
(p − 1)ωn−1

= r1−n/p k∇f kp .
n−p
 p/n
(p−1)(p−1)/n nk∇f kp
Choose r = 1/n satisfying
M (∇f )(x)
(n−p)(p−1)/n ωn−1 2p
1/p0
(p − 1)ωn−1

n n
2 rM (∇f )(x) = r1−n/p k∇f kp ,
ωn−1 n−p
- 70 - 3. The Maximal Function and Calderón-Zygmund Decomposition

then we get
|f (x)| 6 Ck∇f kp/n
p (M (∇f )(x))
1−p/n
.
Thus, by part (iii) in Theorem 3.9, we obtain for 1 < p < n
1−p/n
kf kp∗ 6Ck∇f kp/n
p kM (∇f )kp∗ (1−p/n)

=Ck∇f kp/n 1−p/n

p kM (∇f )kp 6 Ck∇f kp .
This completes the proof. 

3.3 Calderón-Zygmund decomposition

Applying Lebesgue differentiation theorem, we give a decomposition of Rn ,

called Calderón-Zygmund decomposition, which is extremely useful in har-
monic analysis.
Theorem 3.17 (Calderón-Zygmund decomposition of Rn ). Let f ∈
L1 (Rn ) and α > 0. Then there exists a decomposition of Rn such that
(i) Rn = F ∪ Ω, F ∩ Ω = ∅.
(ii) |f (x)| 6 α for a.e. x ∈ F .
S
(iii) Ω is the union of cubes, Ω = k Qk , whose interiors are disjoint and
edges parallel to the coordinate axes, and such that for each Qk
1
Z
α<
m
(Qk ) Qk
|f (x)|dx 6 2n α. (3.10)

(0)
Proof. We decompose Rn into a mesh of equal cubes Qk (k = 1, 2, · · · ),
whose interiors are disjoint and edges parallel to the coordinate axes, and
whose common diameter is so large that
1
Z

m (0)
(Qk ) Q(0) k
|f (x)|dx 6 α, (3.11)

since f ∈ L1 .
(0) (1)
Split each Qk into 2n congruent cubes. These we denote by Qk , k =
1, 2, · · · . There are two possibilities:
1 1
Z Z
either |f (x)|dx 6 α, or
m (1)
(Qk ) Q(1)
k m (Q
(1)
k ) Q
(1)
k
|f (x)|dx > α.

(1) (2)
In the first case, we split Qk again into 2n congruent cubes to get Qk (k =
1, 2, · · · ). In the second case, we have
1 1
Z Z
|f (x)|dx 6 2n α
α<
m (1)
(Qk ) Qk (1)
|f (x)|dx 6
2 m
−n (0)
(Qk̃ ) Qk̃
(0)

(1) (0) (1)

in view of (3.11) where Qk is split from Qk̃ , and then we take Qk as one
of the cubes Qk .
3.3. Calderón-Zygmund decomposition - 71 -

(j)
/ Ω =: ∞
A repetition of this argument shows that if x ∈ k=1 Qk then x ∈ Qkj
S

(j = 0, 1, 2, · · · ) for which
m 1
Z
(j)
(Qkj ) → 0 as j → ∞, and
m (j)
(Qkj ) Qk(j)j
|f (x)|dx 6 α (j = 0, 1, · · · ).

Thus |f (x)| 6 α a.e. x ∈ F = Ω c by a variation of the Lebesgue differentia-

tion theorem. Thus, we complete the proof. 
We now state an immediate corollary.
Corollary 3.18. Suppose f , α, F , Ω and Qk have the same meaning as in
Theorem 3.17. Then there exists two constants A and B (depending only on
the dimension n), such that (i) and (ii) of Theorem 3.17 hold and
m A
(a) (Ω) 6 kf k1 ,
Zα
1
(b)
m(Qk ) Qk
|f |dx 6 Bα.

Proof. In fact, by (3.10) we can take B = 2n , and also because of (3.10)

m m 1 1
X Z
(Ω) = (Qk ) < |f (x)|dx 6 kf k1 .
k
α Ω α
This proves the corollary with A = 1 and B = 2n . 
It is possible however to give another proof of this corollary without using
Theorem 3.17 from which it was deduced, but by using the maximal function
theorem (Theorem 3.9) and also the theorem about the decomposition of an
arbitrary open set as a union of disjoint cubes. This more indirect method of
proof has the advantage of clarifying the roles of the sets F and Ω into which
Rn was divided.
Another proof of the corollary. We know that in F , |f (x)| 6 α, but this fact
does not determine F . The set F is however determined, in effect, by the fact
that the maximal function satisfies M f (x) 6 α on it. So we choose F =
{x : M f (x) 6 α} and Ω = Eα = {x : M f (x) > α}. Then by Theorem 3.9,
m n
part (ii) we know that (Ω) 6 3α kf k1 . Thus, we can take A = 3n .
Since by definition F is closed, we can choose cubes Qk according to The-
orem 3.2, such that Ω = k Qk , and whose diameters are approximately pro-
S

portional to their distances from F . Let Qk then be one of these cubes, and pk
a point of F such that
dist (F, Qk ) = dist (pk , Qk ).
Let Bk be the smallest ball whose center is pk and which contains the interior
of Qk . Let us set
m
(Bk )
γk =
m
(Qk )
.
- 72 - 3. The Maximal Function and Calderón-Zygmund Decomposition

We have, because pk ∈ {x : M f (x) 6 α}, that

1 1
Z Z
α > M f (pk ) >
m
(Bk ) Bk
|f (x)|dx >
m
γk (Qk ) Qk
|f (x)|dx.

Thus, we can take a upper bound of γk as the value of B.

The elementary geometry and the inequality (iii) of Theorem 3.2 then show
that
radius(Bk ) 6 dist (pk , Qk ) + diam (Qk ) = dist (F, Qk ) + diam (Qk )
6(c2 + 1) diam (Qk ),
and so
m(B ) =V (radius(B ))
k n k
n
6 Vn (c2 + 1)n ( diam (Qk ))n
m
=Vn (c2 + 1)n nn/2 (Qk ),
m √
since (Qk ) = ( diam (Qk )/ n)n . Thus, γk 6 Vn (c2 + 1)n nn/2 for all k.
Thus, we complete the proof with A = 3n and B = Vn (c2 + 1)n nn/2 . 
Remark 3.19. Theorem 3.17 may be used to give another proof of the
fundamental inequality for the maximal function in part (ii) of Theorem
3.9. (See [Ste70, §5.1, p.22–23] for more details.)
The Calderón-Zygmund decomposition is a key step in the real-variable
analysis of singular integrals. The idea behind this decomposition is that it is
often useful to split an arbitrary integrable function into its “small” and “large”
parts, and then use different techniques to analyze each part.
The scheme is roughly as follows. Given a function f and an altitude α, we
write f = g +b, where g is called the good function of the decomposition since
it is both integrable and bounded; hence the letter g. The function b is called
the bad function since it contains the singular part of f (hence the letter b), but
it is carefully chosen to have mean value zero. To obtain the decomposition
f = g + b, one might be tempted to “cut” f at the height α; however, this
is not what works. Instead, one bases the decomposition on the set where the
maximal function of f has height α.
Indeed, the Calderón-Zygmund decomposition on Rn may be used to de-
duce the Calderón-Zygmund decomposition on functions. The later is a very
important tool in harmonic analysis.
Theorem 3.20 (Calderón-Zygmund decomposition for functions). Let
f ∈ L1 (Rn ) and α > 0. Then there exist functions g and b on Rn such that
f = g + b and
(i) kgk1 6 kf k1 and kgk∞ 6 2n α.
P
(ii) b = j bj , where each bj is supported in a dyadic cube Qj satisfying
R
b
Qj j
(x)dx = m
0 and kbj k1 6 2n+1 α (Qj ). Furthermore, the cubes Qj and
Qk have disjoint interiors when j 6= k.
m
(iii) j (Qj ) 6 α−1 kf k1 .
P
3.3. Calderón-Zygmund decomposition - 73 -

Proof. Applying Corollary 3.18 (with A = 1 and B = 2n ), we have

1) Rn = F ∪ Ω, F ∩ Ω = ∅;
2) |f (x)| 6 α, a.e. x ∈ F ;
3) Ω = ∞ , with the interiors of the Qj mutually disjoint;
S
j=1 Qj R
4) m(Ω) 6 α −1
|f (x)|dx, and α < m(Qj ) Qj |f (x)|dx 6 2n α.
1
R
Rn
Now define !
1
Z
bj = f −
m (Qj ) Qj
f dx χQj ,

b = j bj and g = f − b. Consequently,
P

m 1
Z Z Z

Qj
|bj |dx 6
Qj
|f (x)|dx + (Qj )
m (Qj ) Qj
f (x)dx

m
Z
62 |f (x)|dx 6 2n+1 α (Qj ),
Qj
m
which proves kbj k1 6 2n+1 α (Qj ).
Next, we need to obtain the estimates on g. Write Rn = ∪j Qj ∪ F , where
F is the
R closed set obtained by Corollary 3.18. Since b = 0 on F and f − bj =
m(Qj ) Qj f (x)dx, we have
1

 f,
 on F,
g= 1
Z
(3.12)

m
 (Q )
j Qj
f (x)dx, on Qj .

On the cube Qj , g is equal to the constant m(Q 1

f (x)dx, and this is
R
j ) Qj
bounded by 2 α by 4). Then by 2), we can get kgk∞ 6 2 α. Finally, it follows
n n

from (3.12) that kgk1 6 kf k1 . This completes the proof. 

As an application of Calderón-Zygmund decomposition and Marcinkiewicz
interpolation theorem, we now prove the weighted estimates for the Hardy-
Littlewood maximal function.
Theorem 3.21 (Weighted inequality for Hardy-Littlewood maximal
function). For p ∈ (1, ∞), there exists a constant C = Cn,p such that, for
any nonnegtive measurable function ϕ(x) on Rn , we have the inequality
Z Z
p
(M f (x)) ϕ(x)dx 6 C |f (x)|p M ϕ(x)dx. (3.13)
Rn Rn

Proof. Except when M ϕ(x) = ∞ a.e., in which case (3.13) holds trivially, M ϕ
is the density of a positive measure µ. Thus, we may assume that M ϕ(x) < ∞
a.e. x ∈ Rn and M ϕ(x) > 0. If we denote
dµ(x) = M ϕ(x)dx and dν(x) = ϕ(x)dx,
- 74 - 3. The Maximal Function and Calderón-Zygmund Decomposition

then by the Marcinkiewicz interpolation theorem in order to get (3.13), it

suffices to prove that M is both of type (L∞ (µ), L∞ (ν)) and of weak type
(L1 (µ), L1 (ν)).
Let us first show that M is of type (L∞ (µ), L∞ (ν)). In fact, if kf kL∞ (µ) 6
α, then Z
M ϕ(x)dx = µ({x ∈ Rn : |f (x)| > α}) = 0.
{x∈Rn :|f (x)|>α}
m
Since M ϕ(x) > 0 for any x ∈ Rn , we have ({x ∈ Rn : |f (x)| > α}) = 0,
equivalently, |f (x)| 6 α a.e. x ∈ Rn . Thus, M f (x) 6 α a.e. x ∈ Rn and this
follows kM f kL∞ (ν) 6 α. Therefore, kM f kL∞ (ν) 6 kf kL∞ (µ) .
Before proving that M is also of weak type (L1 (µ), L1 (ν)), we give the
following lemma.
Lemma 3.22. Let f ∈ L1 (Rn ) and α > 0. If the sequence {Qk } of cubes is
chosen from the Calderón-Zygmund decomposition of Rn for f and α > 0,
then
[
{x ∈ Rn : M 0 f (x) > 7n α} ⊂ Q∗k ,
k
where Q∗k = 2Qk . Then we have
m({x ∈ R m(Q ).
X
n
: M 0 f (x) > 7n α}) 6 2n k
k

/ k Q∗k . Then there are two cases for any cube Q with
Proof. Suppose that x ∈
S

the center x. If Q ⊂ F := Rn \ k Qk , then

S

1
Z

(Q) Q m
|f (x)|dx 6 α.

If Q ∩ Qk 6= ∅ for some k, then it is easy to check that Qk ⊂ 3Q, and

[
{Qk : Qk ∩ Q 6= ∅} ⊂ 3Q.
k
Hence, we have
Z Z Z
X
|f (x)|dx 6 |f (x)|dx + |f (x)|dx
Q Q∩F Qk ∩Q6=∅ Qk

m 2 αm(Q )
X
n
6α (Q) + k
Qk ∩Q6=∅

6αm(Q) + 2 αm(3Q)
n

67 αm(Q).
n

Thus we know that M 0 f (x) 6 7n α for any x ∈

/ k Q∗k , and it yields that
S
!
m m m
[ X
n 0 n ∗
({x ∈ R : M f (x) > 7 α}) 6 Qk = 2n (Qk ).
k k
We complete the proof of the lemma. 
3.3. Calderón-Zygmund decomposition - 75 -

Let us return to the proof of weak type (L1 (µ), L1 (ν)). We need to prove
that there exists a constant C such that for any α > 0 and f ∈ L1 (µ)
Z
ϕ(x)dx =ν({x ∈ Rn : M f (x) > α})
n
{x∈R :M f (x)>α}
(3.14)
C
Z
6 |f (x)|M ϕ(x)dx.
α Rn
We may assume that f ∈ L1 (Rn ). In fact, if we take f` = |f |χB(0,`) , then
f` ∈ L1 (Rn ), 0 6 f` (x) 6 f`+1 (x) for x ∈ Rn and ` = 1, 2, · · · . Moreover,
lim`→∞ f` (x) = |f (x)| and
[
{x ∈ Rn : M f (x) > α} = {x ∈ Rn : M f` (x) > α}.
`
By the pointwise equivalence of M and M 0 , there exists cn > 0 such that
M f (x) 6 cn M 0 f (x) for all x ∈ Rn . Applying the Calderón-Zygmund de-
composition on Rn for f and α0 = α/(cn 7n ), we get a sequence {Qk } of
cubes satisfying
1
Z

m
0
α < |f (x)|dx 6 2n α0 .
(Qk ) Qk
By Lemma 3.22 and the pointwise equivalence of M and M 00 , we have that
Z
ϕ(x)dx
{x∈Rn :M f (x)>α}
Z
6 ϕ(x)dx
{x∈Rn :M 0 f (x)>7n α0 }
Z XZ
6 S ϕ(x)dx 6 ϕ(x)dx
k Q∗k k Q∗k
! Z 
1 1
Z

m
X
6 ϕ(x)dx 0
|f (y)|dy
k
(Q k ) Q∗k α Q k
!
cn 7n X 2n
Z Z
=
α k Qk
|f (y)|
m (Q∗k ) Q∗k
ϕ(x)dx dy

cn 14n X
Z
6 |f (y)|M 00 ϕ(y)dy
α k Qk

C
Z
6 |f (y)|M ϕ(y)dy.
α Rn
Thus, M is of weak type (L1 (µ), L1 (ν)), and the inequality can be obtained
by applying the Marcinkiewicz interpolation theorem. 
L ECTURE N OTES ON c 2016 by Chengchun Hao
Introduction to Harmonic Analysis Email: hcc@amss.ac.cn

Chapter 4
Singular Integrals

4.1 Harmonic functions and Poisson equation

Among the most important of all PDEs are undoubtedly Laplace equation
∆u = 0 (4.1)
and Poisson equation
−∆u = f. (4.2)
In both (4.1) and (4.2), x ∈ Ω and the unknown is u : Ω̄ → R, u = u(x),
where Ω ⊂ Rn is a given open set. In (4.2), the function f : Ω → R is also
given. Remember that the Laplacian of u is ∆u = nk=1 ∂x2k u.
P

Definition 4.1. A C 2 function u satisfying (4.1) is called a harmonic func-

tion.
Now, we derive a fundamental solution of Laplace’s equation. One good
strategy for investigating any PDEs is first to identify some explicit solutions
and then, provided the PDE is linear, to assemble more complicated solutions
out of the specific ones previously noted. Furthermore, in looking for explicit
solutions it is often wise to restrict attention to classes of functions with cer-
tain symmetry properties. Since Laplace equation is invariant under rotations,
it consequently seems advisable to search first for radial solutions, that is, func-
tions of r = |x|. Let us therefore attempt to find a solution u of Laplace equa-
tion (4.1) in Ω = Rn , having the form
u(x) = v(r),
where r = |x| and v is to be selected (if possible) so that ∆u = 0 holds. First
note for k = 1, · · · , n that
∂r xk
= , x 6= 0.
∂xk r
We thus have
x2k 1 x2k
 
0 xk 2 00 0
∂xk u = v (r) , ∂xk u = v (r) 2 + v (r) − 3
r r r r
77
- 78 - 4. Singular Integrals

for k = 1, · · · , n, and so
n−1 0
∆u = v 00 (r) + v (r).
r
Hence ∆u = 0 if and only if
n−1 0
v 00 + v = 0. (4.3)
r
If v 0 6= 0, we deduce
v 00
0 0 1−n
(ln v ) = 0 = ,
v r
and hence v 0 (r) = rn−1
a
for some constant a. Consequently, if r > 0, we have

 b ln r + c, n = 2,
v(r) =
 b + c, n > 3,
rn−2
where b and c are constants.
These considerations motivate the following
Definition 4.2. The function
 1
−
 ln |x|, n = 2,
2π
Φ(x) := 1 1 (4.4)

 , n > 3,
n(n − 2)Vn |x|n−2
defined for x ∈ Rn , x 6= 0, is the fundamental solution of Laplace equa-
tion.
The reason for the particular choices of the constants in (4.4) will be appar-
ent in a moment.
We will sometimes slightly abuse notation and write Φ(x) = Φ(|x|) to em-
phasize that the fundamental solution is radial. Observe also that we have the
estimates
C C
|∇Φ(x)| 6 n−1 , |∇2 Φ(x)| 6 n , (x 6= 0) (4.5)
|x| |x|
for some constant C > 0.
By construction, the function x 7→ Φ(x) is harmonic for x 6= 0. If we shift
the origin to a new point y, the PDE (4.1) is unchanged; and so x 7→ Φ(x − y)
is also harmonic as a function of x for x 6= y. Let us now take f : Rn → R
and note that the mapping x 7→ Φ(x − y)f (y) (x 6= y) is harmonic for each
point y ∈ Rn , and thus so is the sum of finitely many such expression built for
different points y. This reasoning might suggest that the convolution
1
 Z
 − 2π 2 (ln |x − y|)f (y)dy, n = 2,
Z 

R
u(x) = Φ(x − y)f (y)dy =
1 f (y)
Z
Rn
dy, n > 3


n(n − 2)Vn Rn |x − y|n−2

(4.6)
4.1. Harmonic functions and Poisson equation - 79 -

would solve Laplace equation (4.1). However, this is wrong: we cannot just
compute
Z
∆u(x) = ∆x Φ(x − y)f (y)dy = 0. (4.7)
Rn
Indeed, as intimated by estimate (4.5), ∆Φ(x − y) is not summable near the
singularity at y = x, and so the differentiation under the integral sign above
is unjustified (and incorrect). We must proceed more carefully in calculating
∆u.
Let us for simplicity now assume f ∈ Cc2 (Rn ), that is, f is twice continu-
ously differentiable, with compact support.
Theorem 4.3 (Solving Poisson equation). Let f ∈ Cc2 (Rn ), define u by
(4.6). Then u ∈ C 2 (Rn ) and −∆u = f in Rn .
We consequently see that (4.6) provides us with a formula for a solution of
Poisson’s equation (4.2) in Rn .
Proof. Step 1: To show u ∈ C 2 (Rn ). We have
Z Z
u(x) = Φ(x − y)f (y)dy = Φ(y)f (x − y)dy,
Rn Rn
hence
u(x + hek ) − u(x) f (x + hek − y) − f (x − y)
Z  
= Φ(y) dy,
h Rn h
where h 6= 0 and ek = (0, · · · , 1, · · · , 0), the 1 in the k th -slot. But
f (x + hek − y) − f (x − y) ∂f
→ (x − y)
h ∂xk
uniformly on Rn as h → 0, and thus
∂u ∂f
Z
(x) = Φ(y) (x − y)dy, k = 1, · · · , n.
∂xk Rn ∂xk
Similarly,
∂ 2u ∂ 2f
Z
(x) = Φ(y) (x − y)dy, k, j = 1, · · · , n. (4.8)
∂xk ∂xj Rn ∂xk ∂xj
As the expression on the r.h.s. of (4.8) is continuous in the variable x, we see
that u ∈ C 2 (Rn ).
Step 2: To prove the second part. Since Φ blows up at 0, we will need for
subsequent calculations to isolate this singularity inside a small ball. So fix
ε > 0. ThenZ Z
∆u(x) = Φ(y)∆x f (x − y)dy + Φ(y)∆x f (x − y)dy =: Iε + Jε .
B(0,ε) Rn \B(0,ε)
(4.9)
Now
- 80 - 4. Singular Integrals
(
Cε2 (1 + | ln ε|), n = 2,
Z
|Iε | 6 Ck∆f k∞ |Φ(y)|dy 6 (4.10)
B(0,ε) Cε2 , n > 3,
since Z Z ε  Z ε 
2 ε
| ln |y||dy = − 2π r ln rdr = −π r ln r|0 − rdr
B(0,ε) 0 0
2 2
= − π(ε ln ε − ε /2)
π
=πε2 | ln ε| + ε2 ,
2
for ε ∈ (0, 1] and n = 2 by an integration by parts.
An integration by parts yields
Z
Jε = Φ(y)∆x f (x − y)dy
Rn \B(0,ε)
∂f
Z Z
= Φ(y) (x − y)dσ(y) − ∇Φ(y) · ∇y f (x − y)dy
∂B(0,ε) ∂ν Rn \B(0,ε)

=:Kε + Lε ,
(4.11)
where ν denotes the inward pointing unit normal along ∂B(0, ε). We readily
check Z Z
|Kε | 6k∇f k∞ |Φ(y)|dσ(y) 6 C|Φ(ε)| dσ(y) = C|Φ(ε)|εn−1
∂B(0,ε) ∂B(0,ε)
(
Cε| ln ε|, n = 2,
6
Cε, n > 3,
(4.12)
since Φ(y) = Φ(|y|) = Φ(ε) on ∂B(0, ε) = {y ∈ R : |y| = ε}. n

We continue by integrating by parts once again in the term Lε , to discover

∂Φ
Z Z
Lε = − (y)f (x − y)dσ(y) + ∆Φ(y)f (x − y)dy
∂B(0,ε) ∂ν Rn \B(0,ε)
∂Φ
Z
=− (y)f (x − y)dσ(y),
∂B(0,ε) ∂ν
since Φ is harmonic away from the origin. Now, ∇Φ(y) = − nV1 n |y|yn for y 6= 0
and ν = −y|y|
= − yε on ∂B(0, ε). Consequently, ∂Φ
∂ν
(y) = ν · ∇Φ(y) = nVn1εn−1
on ∂B(0, ε). Since nVn εn−1 is the surface area of the sphere ∂B(0, ε), we have
1
Z
Lε = − f (x − y)dσ(y)
nVn εn−1 ∂B(0,ε)
(4.13)
1
Z
=−
m
(∂B(x, ε)) ∂B(x,ε)
f (y)dσ(y) → −f (x) as ε → 0.

by Lebesgue differentiation theorem.

4.1. Harmonic functions and Poisson equation - 81 -

Combining now (4.9)-(4.13) and letting ε → 0, we find that −∆u(x) =

f (x), as asserted. 
Remark 4.4. We sometimes write
−∆Φ = δ0 in Rn ,
where δ0 denotes the Dirac measure on Rn giving unit mass to the point
0. Adopting this notation, we may formally compute
Z Z
−∆u(x) = −∆x Φ(x − y)f (y)dy = δx f (y)dy = f (x), x ∈ Rn ,
Rn Rn
in accordance with Theorem 4.3. This corrects the erroneous calculation
(4.7).
Consider now an open set Ω ⊂ Rn and suppose u is a harmonic function
within Ω. We next derive the important mean-value formulas, which declare
that u(x) equals both the average of u over the sphere ∂B(x, r) and the average
of u over the entire ball B(x, r), provided B(x, r) ⊂ Ω.
Theorem 4.5 (Mean-value formula for harmonic functions). If u ∈
C 2 (Ω) is harmonic, then for each ball B(x, r) ⊂ Ω,
1 1
Z Z
u(x) =
m
(∂B(x, r)) ∂B(x,r)
u(y)dσ(y) =
m
(B(x, r)) B(x,r)
u(y)dy.

Proof. Denote
1 1
Z Z
f (r) =
m (∂B(x, r)) ∂B(x,r)
u(y)dσ(y) =
ωn−1 S n−1
u(x + rz)dσ(z).

Obviously,
n
1 1 ∂u
Z X Z
0
f (r) = ∂xj u(x+rz)zj dσ(z) = (x+rz)dσ(z),
ωn−1 S n−1 j=1 ωn−1 S n−1 ∂ν
where ∂ν ∂
denotes the differentiation w.r.t. the outward normal. Thus, by
changes of variable
1 ∂u
Z
0
f (r) = n−1
(y)dσ(y).
ωn−1 r ∂B(x,r) ∂ν
By Stokes theorem, we get
1
Z
0
f (r) = ∆u(y)dy = 0.
ωn−1 rn−1 B(x,r)
Thus f (r) = const. Since limr→0 f (r) = u(x), hence, f (r) = u(x).
Next, observe that our employing polar coordinates gives, by the first iden-
tity proved just now, that
m
Z Z r Z  Z r
u(y)dy = u(y)dσ(y) ds = (∂B(x, s))u(x)ds
B(x,r) 0 ∂B(x,s) 0
Z r
=u(x) nVn sn−1 ds = Vn rn u(x).
0
- 82 - 4. Singular Integrals

This completes the proof. 

Theorem 4.6 (Converse to mean-value property). If u ∈ C 2 (Ω) satisfies

1
Z
u(x) =
m(∂B(x, r)) ∂B(x,r)
u(y)dσ(y)

for each ball B(x, r) ⊂ Ω, then u is harmonic.

Proof. If ∆u 6≡ 0, then there exists some ball B(x, r) ⊂ Ω such that, say,
∆u > 0 within B(x, r). But then for f as above,
1
Z
0
0 = f (r) = n−1 ∆u(y)dy > 0,
r ωn−1 B(x,r)
is a contradiction. 

4.2 Poisson kernel and Hilbert transform

We shall now introduce a notation that will be indispensable in much of

our further work. Indeed, we have shown some properties of Poisson kernel
in Chapter 1. The setting for the application of this theory will be as follows.
We shall think of Rn as the boundary hyperplane of the (n + 1) dimensional
upper-half space Rn+1 . In coordinate notation,
Rn+1
+ = {(x, y) : x ∈ Rn , y > 0} .
We shall consider the Poisson integral of a function f given on Rn . This
Poisson integral is effectively the solution to the Dirichlet Problem for Rn+1
+ :
find a harmonic function u(x, y) on R+ , whose boundary values on R (in
n+1 n

the appropriate sense) are f (x), that is

(
∆x,y u(x, y) = 0, (x, y) ∈ Rn+1 + ,
n
(4.14)
u(x, 0) = f, x ∈ R .
The formal solution of this problem can be given neatly in the context of the
L theory.
2

In fact, let f ∈ L2 (Rn ), and consider

 n Z
|ω|
u(x, y) = eωiξ·x e−|ωξ|y fˆ(ξ)dξ, y > 0. (4.15)
2π Rn

This integral converges absolutely (cf. Theorem 1.15), because fˆ ∈ L2 (Rn ),

and e−|ωξ|y is rapidly decreasing in |ξ| for y > 0. For the same reason, the
integral above may be differentiated w.r.t. x and y any number of times by
carrying out the operation under the sign of integration. This gives
n
∂ 2u X ∂ 2u
∆x,y u = 2 + 2
= 0,
∂y k=1
∂x k
4.2. Poisson kernel and Hilbert transform - 83 -

because the factor eωiξ·x e−|ωξ|y satisfies this property for each fixed ξ. Thus,
u(x, y) is a harmonic function on Rn+1 + .
By Theorem 1.15, we get that u(x, y) → f (x) in L2 (Rn ) norm, as y →
0. That is, u(x, y) satisfies the boundary condition and so u(x, y) structured
above is a solution for the above Dirichlet problem.
This solution of the problem can also be written without explicit use of the
Fourier transform. For this purpose, we define the Poisson kernel Py (x) :=
P (x, y) by
 n Z
|ω|
Py (x) = eωiξ·x e−|ωξ|y dξ = (F −1 e−|ωξ|y )(x), y > 0. (4.16)
2π R n

Then the function u(x, y) obtained above can be written as a convolution

Z
u(x, y) = Py (z)f (x − z)dz, (4.17)
Rn
as the same as in Theorem 1.15. We shall say that u is the Poisson integral of
f.
For convenience, we recall (1.12) and (1.10) as follows.
Proposition 4.7. The Poisson kernel has the following explicit expression:
cn y Γ ((n + 1)/2)
Py (x) = n+1 , cn = n+1 . (4.18)
2 2
(|x| + y ) 2 π 2
Remark 4.8. We list the properties of the Poisson kernel that are now
more or less evident:
(i) Py (x) > 0 for y > 0.
cy (ξ) = e−|ωξ|y by
R
(ii) Rn Py (x)dx = P cy (0) = 1, y > 0; more generally, P
Lemma 1.14 and Corollary 1.23, respectively.
(iii) Py (x) is homogeneous of degree −n: Py (x) = y −n P1 (x/y), y > 0.
(iv) Py (x) is a decreasing function of |x|, and Py ∈ Lp (Rn ), 1 6 p 6 ∞.
Indeed, by changes of variables, we have for 1 6 p < ∞
Z  p
p p y
kPy kp =cn dx
Rn (|x|2 + y 2 )(n+1)/2
1
Z
x=yz p −n(p−1)
==cn y 2 p(n+1)/2
dz
Rn (1 + |z| )
Z ∞
z=rz 0 p −n(p−1) 1
== cn y ωn−1 2 p(n+1)/2
rn−1 dr
(1 + r )
0Z 1 Z ∞ 
p −n(p−1) n−1−p(n+1)
6cn y ωn−1 dr + r dr
0 1
 
p −n(p−1) 1
6cn y ωn−1 1 + .
p(n + 1) − n
For p = ∞, it is clear that kPy (x)k∞ = cn y −n .
- 84 - 4. Singular Integrals

(v) Suppose f ∈ Lp (Rn ), 1 6 p 6 ∞, then its Poisson integral u, given

by (4.17), is harmonic in Rn+1
+ . This is a simple consequence of the fact
that Py (x) is harmonic in Rn+1
+ ; the latter is immediately derived from
(4.16).
(vi) We have the “semi-group property” Py1 ∗Py2 = Py1 +y2 if y1 , y2 > 0
in view of Corollary 1.24.
The boundary behavior of Poisson integrals is already described to a signif-
icant extension by the following theorem.
Theorem 4.9. Suppose f ∈ Lp (Rn ), 1 6 p 6 ∞, and let u(x, y) be its
Poisson integral. Then
(a) supy>0 |u(x, y)| 6 M f (x), where M f is the maximal function.
(b) limy→0 u(x, y) = f (x), for almost every x.
(c) If p < ∞, u(x, y) converges to f (x) in Lp (Rn ) norm, as y → 0.
The theorem will now be proved in a more general setting, valid for a large
class of approximations to the identity.
Let ϕ be an integrable function on Rn , and set ϕε (x) = ε−n ϕ(x/ε), ε > 0.
Theorem 4.10. Suppose that the least decreasing radial majorant
R of ϕ is in-
tegrable; i.e., let ψ(x) = sup|y|>|x| |ϕ(y)|, and we suppose Rn ψ(x)dx = A <
∞. Then with the same A,
(a) supε>0 |(f ∗ ϕRε )(x)| 6 AM f (x), f ∈ Lp (Rn ), 1 6 p 6 ∞.
(b) If in addition Rn ϕ(x)dx = 1, then limε→0 (f ∗ ϕε )(x) = f (x) almost
everywhere.
(c) If p < ∞, then kf ∗ ϕε − f kp → 0, as ε → 0.

Proof. For the part (c), we have shown in Theorem 1.15.

Next, we prove assertion (a). We have already considered a special case of
(a) in Chapter 3, with ϕ = m(B) 1
χB . The point of the theorem is to reduce
matters to this fundamental special case.
With a slight abuse of notation, let us write ψ(r) = ψ(x), if |x| = r; it
should cause no confusion R since ψ(x) is anyway radial.
R Now observe that ψ(r)
is decreasing and then r/26|x|6r ψ(x)dx > ψ(r) r/26|x|6r dx = cψ(r)rn .
Therefore the assumption ψ ∈ L1 proves that rn ψ(r) → 0 as r → 0 or
r → ∞. To prove (a), we need to show that
(f ∗ ψε )(x) 6 AM f (x), (4.19)
where f > 0, f ∈ Lp (Rn ), ε > 0 and A = Rn ψ(x)dx.
R

Since (4.19) is clearly translation invariant w.r.t f and also dilation invariant
w.r.t. ψ and the maximal function, it suffices to show that
(f ∗ ψ)(0) 6 AM f (0). (4.20)
4.2. Poisson kernel and Hilbert transform - 85 -

In proving (4.20), we may clearly assume that M f (0) < ∞. Let us write
λ(r) = S n−1 f (rx )dσ(x ), and Λ(r) = |x|6r f (x)dx, so
0 0
R R
Z rZ Z r
0 0 n−1
Λ(r) = f (tx )dσ(x )t dt = λ(t)tn−1 dt, i.e., Λ0 (r) = λ(r)rn−1 .
0 S n−1 0
We have Z Z ∞ Z
(f ∗ ψ)(0) = f (x)ψ(x)dx = r n−1
f (rx0 )ψ(r)dσ(x0 )dr
Rn 0 S n−1
Z ∞ Z N
= rn−1 λ(r)ψ(r)dr = lim
ε→0
λ(r)ψ(r)rn−1 dr
0 N →∞ ε
Z N  Z N 
0
= lim
ε→0
Λ (r)ψ(r)dr = lim
ε→0
− [Λ(r)ψ(r)]N
ε Λ(r)dψ(r) .
N →∞ ε N →∞ ε

Since Λ(r) = |x|6r f (x)dx 6 Vn r M f (0), and the fact r ψ(r) → 0 as r → 0

n n
R

or r → ∞, we have
0 6 lim Λ(N )ψ(N ) 6 Vn M f (0) lim N n ψ(N ) = 0,
N →∞ N →∞
which implies limN →∞ Λ(N )ψ(N ) = 0 and similarly limε→0 Λ(ε)ψ(ε) = 0.
Thus, by integration by parts, we have
Z ∞ Z ∞
(f ∗ ψ)(0) = Λ(r)d(−ψ(r)) 6 Vn M f (0) rn d(−ψ(r))
0 0
Z ∞ Z
n−1
=nVn M f (0) ψ(r)r dr = M f (0) ψ(x)dx,
0 Rn
since ψ(r) is decreasing which implies ψ (r) 6 0, and nVn = ωn−1 . This
0

proves (4.20) and then (4.19).

Finally, we prove (b) in a familiar way as follows. First, we can verify that
if f1 ∈ Cc , then (f1 ∗ ϕε )(x) → f1 (x) uniformly as ε → 0 (cf. Theorem 1.15).
Next we can deal with the case f ∈ Lp (Rn ), 1 6 p < ∞, by writing f =
f1 + f2 with f1 as described and with kf2 kp small. The argument then follows
closely that given in the proof of Theorem 3.13 (the Lebesgue differentiation
theorem). Thus we get that limε→0 f ∗ ϕε (x) exists almost everywhere and
equals f (x).
To deal with the remaining case, that of bounded f , we fix any ball B =
B(x0 , r), and set ourselves the task of showing that
lim(f ∗ ϕε )(x) = f (x), for almost every x ∈ B.
ε→0
Let B1 be any other ball which strictly contains B and the origin {0} sat-
isfying δ > |x0 | + r where δ =  dist (B, B1c ) is the distance from B to the
f (x), x ∈ B1 ,
complement of B1 . Let f1 (x) = ; f (x) = f1 (x) + f2 (x).
0, x∈/ B1 ,
Then, f1 ∈ L1 (Rn ), and so the appropriate conclusion holds for it. However,
for x ∈ B,
- 86 - 4. Singular Integrals
Z Z
|(f2 ∗ ϕε )(x)| = f2 (x − y)ϕε (y)dy 6 |f2 (x − y)||ϕε (y)|dy
Rn |x−y|>δ>0
Z
6kf k∞ |ϕ(y)|dy → 0, as ε → 0.
|y|>(δ−|x|)/ε>0
Thus, we complete the proof. 
Proof of Theorem 4.9. Theorem 4.10 then applies directly to prove Theorem
4.9, because of properties (i)–(iv) of the Poisson kernel in the case ϕ(x) =
ψ(x) = P1 (x). 
There are also some variants of the result of Theorem 4.10, which apply
equally well to Poisson integrals. The first is an easy adaptation of the argu-
ment already given, and is stated without proof.
Corollary 4.11. Suppose f is continuous and bounded on Rn . Then (f ∗
ϕε )(x) → f (x) uniformly on compact subsets of Rn .
The second variant is somewhat more difficult. It is the analogue for finite
Borel measures in place of integrable functions, and is outlined in further result
of [Ste70, §4.1, p.77–78].
Now, we give the definition of harmonic conjugate functions as follows.
Definition 4.12. The harmonic conjugate to a given function u(x, y) is a
function v(x, y) such that
f (x, y) = u(x, y) + iv(x, y)
is analytic, i.e., satisfies the Cauchy-Riemann equations
ux = vy , uy = −vx ,
where ux ≡ ∂u/∂x, uy ≡ ∂u/∂y. It is given by
Z (x,y)
v(x, y) = ux dy − uy dx + C,
(x0 ,y0 )
along any path connecting (x0 , y0 ) and (x, y) in the domain, where C is
a constant of integration.
Given a function f in S (R), its harmonic extension to the upper half-plane
is given by u(x, y) = Py ∗ f (x), where Py is the Poisson kernel. We can also
write, in view of (4.15),
|ω|
Z
u(z) =u(x, y) = eωiξ·x e−|ωξ|y fˆ(ξ)dξ
2π R
Z ∞ Z 0
|ω|

ωiξ·x −|ω|ξy ˆ ωiξ·x |ω|ξy ˆ
= e e f (ξ)dξ + e e f (ξ)dξ
2π 0 −∞
Z ∞ Z 0
|ω|

ωiξ·(x+i sgn (ω)y) ˆ ωiξ·(x−i sgn (ω)y) ˆ
= e f (ξ)dξ + e f (ξ)dξ ,
2π 0 −∞
4.2. Poisson kernel and Hilbert transform - 87 -

where z = x + iy. If we now define

|ω| h ∞ ωiξ·(x+i sgn (ω)y) ˆ
Z
i sgn (ω)v(z) = e f (ξ)dξ
2π 0
Z 0 i
− eωiξ·(x−i sgn (ω)y) fˆ(ξ)dξ ,
−∞
then v is also harmonic in R2+ and both u and v are real if f is. Furthermore,
u + iv is analytic since it satisfies the Cauchy-Riemann equations ux = vy =
ωiξu(z) and uy = −vx = −ωiξv(z), so v is the harmonic conjugate of u.
Clearly, v can also be written as, by Theorem 1.12, Proposition 1.3 and
Theorem 1.28,
|ω|
Z
v(z) = −i sgn (ω) sgn (ξ)eωiξ·x e−|ωξ|y fˆ(ξ)dξ
2π R
|ω|
Z
= −i sgn (ω)Fξ [ sgn (ξ)eωiξ·x e−|ωξ|y ](η)f (η)dη
2π R
|ω|
Z
= −i sgn (ω)Fξ [ sgn (ξ)e−|ωξ|y ](η − x)f (η)dη
2π R
Z
= −i sgn (ω)Fξ−1 [ sgn (ξ)e−|ωξ|y ](x − η)f (η)dη,
R
which is equivalent to
v(x, y) = Qy ∗ f (x), (4.21)
where
Q̂y (ξ) = −i sgn (ω) sgn (ξ)e−|ωξ|y . (4.22)
Now we invert the Fourier transform, we get, by a change of variables and
integration by parts,
|ω|
Z
Qy (x) = −i sgn (ω) eωix·ξ sgn (ξ)e−|ωξ|y dξ
2π R
Z ∞ Z 0
|ω|

ωix·ξ −|ω|ξy ωix·ξ |ω|ξy
= − i sgn (ω) e e dξ − e e dξ
2π 0 −∞
Z ∞ Z ∞
|ω|

ωix·ξ −|ω|ξy −ωix·ξ −|ω|ξy
= − i sgn (ω) e e dξ − e e dξ
2π 0 0
|ω| ∞ ωix·ξ
∂ξ e−|ω|ξy
Z
= − i sgn (ω) e − e−ωix·ξ dξ
2π 0 −|ω|y
1 h ωix·ξ ∞
− e−ωix·ξ e−|ω|ξy 0


=i sgn (ω) e
2πy
Z ∞ i
ωix eωix·ξ + e−ωix·ξ e−|ω|ξy dξ


−
0
|ω|x ∞ ωix·ξ
Z
+ e−ωix·ξ e−|ω|ξy dξ


= e
2πy 0
- 88 - 4. Singular Integrals

|ω|x |ω| −|ωξ|y

 
x
Z
−ωix·ξ −|ωξ|y
= e e dξ = F e
2πy R y 2π
x x c1 y c1 x
= Py (x) = 2 2
= 2 ,
y yy +x y + x2
where c1 = Γ (1)/π = 1/π. That is,
1 x
Qy (x) = .
π y + x2
2

One can immediately verify that Q(x, y) = Qy (x) is a harmonic function in

the upper half-plane and the conjugate of the Poisson kernel Py (x) = P (x, y).
More precisely, they satisfy Cauchy-Riemann equations
1 2xy 1 x2 − y 2
∂x P = ∂y Q = − , ∂ y P = −∂x Q = .
π (y 2 + x2 )2 π (y 2 + x2 )2
In Theorem 4.9, we studied the limit of u(x, t) as y → 0 using the fact
that {Py } is an approximation of the identity. We would like to do the same for
v(x, y), but we immediately run into an obstacle: {Qy } is not an approximation
of the identity and, in fact, Qy is not integrable for any y > 0. Formally,
1
lim Qy (x) = ,
y→0 πx
this is not even locally integrable, so we cannot define its convolution with
smooth functions.
We define a tempered distribution called the principal value of 1/x, abbre-
viated p.v. 1/x, by
 
1 φ(x)
Z
p.v. , φ = lim dx, φ ∈ S .
x ε→0 |x|>ε x
To see that this expression defines a tempered distribution, we rewrite it as
φ(x) − φ(0)
  Z
1 φ(x)
Z
p.v. , φ = dx + dx,
x |x|<1 x |x|>1 x
this holds since the integral of 1/x on ε < |x| < 1 is zero. It is now immediate
that  
1
p.v. , φ 6 C(kφ0 k∞ + kxφk∞ ).
x
1
Proposition 4.13. In S 0 (R), we have lim Qy (x) = π
p.v. x1 .
y→0

Proof. For each ε > 0, the functions ψε (x) = x−1 χ|x|>ε are bounded and
define tempered distributions. It follows at once from the definition that in S 0 ,
1
lim ψε (x) = p.v. .
ε→0 x
Therefore, it will suffice to prove that in S 0
 
1
lim Qy − ψy = 0.
y→0 π
4.2. Poisson kernel and Hilbert transform - 89 -

Fix φ ∈ S , then by a change of variables, we have

xφ(x) φ(x)
Z Z
hπQy − ψy , φi = 2 2
dx − dx
R y +x |x|>y x
 
xφ(x) x 1
Z Z
= 2 2
dx + − φ(x)dx
|x|<y y + x |x|>y y 2 + x2 x
xφ(yx) φ(yx)
Z Z
= 2
dx − 2
dx.
|x|<1 1 + x |x|>1 x(1 + x )
If we take the limit as y → 0 and apply the dominated convergence theorem,
we get two integrals of odd functions on symmetric domains. Hence, the limit
equals 0. 
As a consequence of this proposition, we get that
1 f (x − t)
Z
lim Qy ∗ f (x) = lim dt,
y→0 π ε→0 |t|>ε t
and by the continuity of the Fourier transform on S 0 and by (4.22), we get
 
1 1
F p.v. (ξ) = −i sgn (ω) sgn (ξ).
π x
Given a function f ∈ S , we can define its Hilbert transform by any one of
the following equivalent expressions:
Hf = lim Qy ∗ f,
y→0
1 1
Hf = p.v. ∗ f,
π x
Hf =F −1 (−i sgn (ω) sgn (ξ)fˆ(ξ)).
The third expression also allows us to define the Hilbert transform of functions
in L2 (R), which satisfies, with the help of Theorem 1.26,
 1/2  1/2
|ω| |ω|
kHf k2 = kF (Hf )k2 = kfˆk2 = kf k2 , (4.23)
2π 2π
that is, H is an isometry on L2 (R). Moreover, H satisfies
H 2 f = H(Hf ) =F −1 ((−i sgn (ω) sgn (ξ))2 fˆ(ξ)) = −f, (4.24)
By Theorem 1.28, we have
Z Z
hHf, gi = Hf · gdx = F −1 (−i sgn (ω) sgn (ξ)fˆ(ξ)) · gdx
R ZR
= −i sgn (ω) sgn (ξ)fˆ(ξ) · ǧ(ξ)dξ
ZR
= f (x) · F [−i sgn (ω) sgn (ξ)ǧ(ξ)](x)dx
ZR
|ω|
= f (x) · F [−i sgn (ω) sgn (ξ) ĝ(−ξ)](x)dx
R 2π
- 90 - 4. Singular Integrals
Z
= f (x) · F −1 [i sgn (ω) sgn (η)ĝ(η)](x)dx
RZ

= − f · Hgdx = hf, −Hgi, (4.25)

R
namely, the dual/conjugate operator of H is H 0 = −H. Similarly, the adjoint
operator H ∗ of H is uniquely defined via the identity
Z Z
(f, Hg) = f · Hgdx = − Hf ḡdx = (−Hf, g) =: (H ∗ f, g),
R R
that is, H ∗ = −H.
Note that for given x ∈ R, Hf (x) is defined for all integrable functions f
on R that satisfy a Hölder condition near the point x, that is,
|f (x) − f (t)| 6 Cx |x − t|εx
for some Cx > 0 and εx > 0 whenever |t − x| < δx . Indeed, suppose that this
is the case, then
1 f (t) 1 f (t)
Z Z
lim Qy ∗ f (x) = lim dt + dt
y→0 π ε→0 ε<|x−t|<δx x − t π |x−t|>δx x − t
1 f (t) − f (x) 1 f (t)
Z Z
= lim dy + dt.
π ε→0 ε<|x−t|<δx x−t π |x−t|>δx x − t
Both integrals converge absolutely, and hence the limit of Qy ∗ f (x) exists
as ε → 0. Therefore, the Hilbert transform of a piecewise smooth integrable
function is well defined at all points of Hölder-Lipschitz continuity of the func-
tion. On the other hand, observe that Qy ∗ f is well defined for all f ∈ Lp ,
0
1 6 p < ∞, as it follows from the Hölder inequality, since Qy (x) is in Lp .
Ex. 4.14. Consider the characteristic function χ[a,b] of an interval [a, b]. It
is a simple calculation to show that
1 |x − a|
H(χ[a,b] )(x) = ln . (4.26)
π |x − b|
Let us verify this identity. By the definition, we have
1 χ[a,b] (x − y) 1 1
Z Z
H(χ[a,b] )(x) = lim dy = lim dy.
π ε→0 |y|>ε y π ε→0 |y|>ε
x−b6y6x−a
y
Thus, we only need to consider three cases: x − b > 0, x − a < 0 and
x − b < 0 < x − a. For the first two cases, we have
1 x−a 1 1 |x − a|
Z
H(χ[a,b] )(x) = dy = ln .
π x−b y π |x − b|
For the third case we get (without loss of generality, we can assume
ε < min(|x − a|, |x − b|))
Z −ε Z x−a 
1 1 1
H(χ[a,b] )(x) = lim dy + dy
π ε→0 x−b y ε y
4.2. Poisson kernel and Hilbert transform - 91 -

|x − a|
 
1 ε
= lim ln + ln
π ε→0 ε |x − b|
1 |x − a|
= ln ,
π |x − b|
where it is crucial to observe how the cancellation of the odd kernel 1/x
is manifested. Note that H(χ[a,b] )(x) blows up logarithmically for x near
the points a and b and decays like x−1 as x → ±∞. See the following
graph with a = 1 and b = 3:

The following is a graph of the function H(χ[−10,0]∪[1,2]∪[4,7] ):

It is obvious, for the dilation operator δε with ε > 0, by changes of variables

(εy → y), that
1 f (εx − εy)
Z
(Hδε )f (x) = lim dy
σ→0 π |y|>σ y
f (εx − y)
Z
= lim dy = (δε H)f (x),
σ→0 |y|>εσ y
so Hδε = δε H; and it is equally obvious that Hδε = −δε H, if ε < 0.
These simple considerations of dilation “invariance” and the obvious trans-
lation invariance in fact characterize the Hilbert transform.
Proposition 4.15 (Characterization of Hilbert transform). Suppose T is
a bounded linear operator on L2 (R) which satisfies the following properties:
(a) T commutes with translations;
(b) T commutes with positive dilations;
(c) T anticommutes with the reflection f (x) → f (−x).
Then, T is a constant multiple of the Hilbert transform.
- 92 - 4. Singular Integrals

Proof. Since T commutes with translations and maps L2 (R) to itself, ac-
cording to Theorem 1.62, there is a bounded function m(ξ) such that
Tcf (ξ) = m(ξ)fˆ(ξ). The assumptions (b) and (c) may be written as T δε f =
sgn (ε)δε T f for all f ∈ L2 (R). By part (iv) in Proposition 1.3, we have
F (T δε f )(ξ) =m(ξ)F (δε f )(ξ) = m(ξ)|ε|−1 fˆ(ξ/ε),
sgn (ε)F (δε T f )(ξ) = sgn (ε)|ε|−1 Tcf (ξ/ε) = sgn (ε)|ε|−1 m(ξ/ε)fˆ(ξ/ε),
which means m(εξ) = sgn (ε)m(ξ), if ε 6= 0. This shows that m(ξ) =
c sgn (ξ), and the proposition is proved. 
The next theorem shows that the Hilbert transform, now defined for func-
tions in S or L2 , can be extended to functions in Lp , 1 6 p < ∞.
Theorem 4.16. For f ∈ S (R), the following assertions are true:
(i) (Kolmogorov) H is of weak type (1, 1):
m C
({x ∈ R : |Hf (x)| > α}) 6 kf k1 .
α
(ii) (M. Riesz) H is of type (p, p), 1 < p < ∞:
kHf kp 6 Cp kf kp .

Proof. (i) Fix α > 0. From the Calderón-Zygmund decomposition of f at

height α (Theorem 3.20), there exist two functions g and b such that f = g + b
and
(1) kgk1 6 kf k1 and kgk∞ 6 2α.
(2) j bj , where each bj is supported in a dyadic interval Ij satisfying
P
b =
R
Ij j
m
b (x)dx = 0 and kbj k1 6 4α (Ij ). Furthermore, the intervals Ij and Ik
have disjoint interiors when j 6= k.
m
(3) j (Ij ) 6 α−1 kf k1 .
P

Let 2Ij be the interval with the same center as Ij and twice the length, and
m m
let Ω = ∪j Ij and Ω ∗ = ∪j 2Ij . Then (Ω ∗ ) 6 2 (Ω) 6 2α−1 kf k1 .
Since Hf = Hg + Hb, from parts (iv) and (vi) of Proposition 2.15, (4.23)
and (1), we have
(Hf )∗ (α) 6 (Hg)∗ (α/2) + (Hb)∗ (α/2)

m m
Z
−2
6(α/2) |Hg(x)|2 dx + (Ω ∗ ) + ({x ∈ / Ω ∗ : |Hb(x)| > α/2})
R
4
Z Z
2 −1 −1
6 2 |g(x)| dx + 2α kf k1 + 2α |Hb(x)|dx
α R R\Ω ∗
8 2 2
Z Z X
6 |g(x)|dx + kf k1 + |Hbj (x)|dx
α R α α R\Ω ∗ j
8 2 2X
Z
6 kf k1 + kf k1 + |Hbj (x)|dx.
α α α j R\2Ij
4.2. Poisson kernel and Hilbert transform - 93 -

For x ∈
/ 2Ij , we have
1 bj (y) 1 bj (y)
Z Z
Hbj (x) = p.v. dy = dy,
π Ij x − y π Ij x − y
m
since supp bj ⊂ Ij and |x − y| > (Ij )/2 for y ∈ Ij . Denote the center of Ij
by cj , then, since bj is mean zero, we have
1 bj (y)
Z Z Z
|Hbj (x)|dx = dy dx
R\2Ij R\2Ij π Ij x − y
 
1 1 1
Z Z
= bj (y) − dy dx
π R\2Ij Ij x − y x − cj
!
1 |y − cj |
Z Z
6 |bj (y)| dx dy
π Ij R\2Ij |x − y||x − cj |

m
!
1 (Ij )
Z Z
6 |bj (y)| 2
dx dy.
π Ij R\2Ij |x − cj |

m
The last inequality follows from the fact that |y − cj | < (Ij )/2 and |x − y| >
m
|x − cj |/2. Since |x − cj | > (Ij ), the inner integral equals
m m m
Z ∞
1 1
2 (Ij ) dr = 2 (Ij ) = 2.
m(Ij ) r2 (Ij )
Thus, by (2) and (3),
10 4 X 10 4 X
m
Z
(Hf )∗ (α) 6 kf k1 + |bj (y)|dy 6 kf k1 + 4α (Ij )
α απ j Ij α απ j
10 16 1 10 + 16/π
6 kf k1 + kf k1 = kf k1 .
α π α α
(ii) Since H is of weak type (1, 1) and of type (2, 2), by the Marcinkiewicz
interpolation theorem, we have the strong (p, p) inequality for 1 < p < 2. If
p > 2, we apply the dual estimate with the help of (4.25) and the result for
p0 < 2 (where 1/p + 1/p0 = 1):
kHf kp = sup |hHf, gi| = sup |hf, Hgi|
kgkp0 61 kgkp0 61

6kf kp sup kHgkp0 6 Cp0 kf kp .

kgkp0 61

This completes the proof. 

Remark 4.17. i) Recall from the proof of the Marcinkiewicz interpolation
theorem that
 the  coefficient
(1/2)1/2

1/p 10 + 16/π 1/2
2 + +2 , 1 < p < 2,



1 − 1/p 1/p − 1/2
Cp =  1/2

 1/p 0 (1/2) 1/2
2 (10 + 16/π)p + +2 , p > 2.


1/2 − 1/p
- 94 - 4. Singular Integrals

So the constant Cp tends to infinity as p tends to 1 or ∞. More precisely,

Cp = O(p) as p → ∞, and Cp = O((p − 1)−1 ) as p → 1.
ii) The strong (p, p) inequality is false if p = 1 or p = ∞, this can easily
be seen from the previous example Hχ[a,b] = π1 ln |x−a| |x−b|
which is neither
integrable nor bounded. See the following figure.

The integral

Hχ[1,2]

iii) By using the inequalities in Theorem 4.16, we can extend the

Hilbert transform to functions in Lp , 1 6 p < ∞. If f ∈ L1 and {fn } is a
sequence of functions in S that converges to f in L1 , then by the weak
(1, 1) inequality the sequence {Hfn } is a Cauchy sequence in measure:
for any ε > 0,
lim
m,n→∞
m({x ∈ R : |(Hf n − Hfm )(x)| > ε}) = 0.
Therefore, it converges in measure to a measurable function which we
define to be the Hilbert transform of f .
If f ∈ Lp , 1 < p < ∞, and {fn } is a sequence of functions in S that
converges to f in Lp , by the strong (p, p) inequality, {Hfn } is a Cauchy
sequence in Lp , so it converges to a function in Lp which we call the
Hilbert transform of f .
In either case, a subsequence of {Hfn }, depending on f , converges
pointwise almost everywhere to Hf as defined.

4.3 The Calderón-Zygmund theorem

From this section on, we are going to consider singular integrals whose ker-
nels have the same essential properties as the kernel of the Hilbert transform.
We can generalize Theorem 4.16 to get the following result.
Theorem 4.18 (Calderón-Zygmund Theorem). Let K be a tempered dis-
tribution in Rn which coincides with a locally integrable function on Rn \ {0}
and satisfies
|K(ξ)|
b 6 B, (4.27)
4.3. The Calderón-Zygmund theorem - 95 -
Z
|K(x − y) − K(x)|dx 6 B, y ∈ Rn . (4.28)
|x|>2|y|
Then we have the strong (p, p) estimate for 1 < p < ∞
kK ∗ f kp 6 Cp kf kp , (4.29)
and the weak (1, 1) estimate
C
(K ∗ f )∗ (α) 6
kf k1 . (4.30)
α
We will show that these inequalities are true for f ∈ S , but they can be
extended to arbitrary f ∈ Lp as we did for the Hilbert transform. Condition
(4.28) is usually referred to as the Hörmander condition; in practice it is often
deduced from another stronger condition called the gradient condition (i.e.,
(4.31) as below).
Proposition 4.19. The Hörmander condition (4.28) holds if for every x 6= 0
C
|∇K(x)| 6 n+1 . (4.31)
|x|
Proof. By the integral mean value theorem and (4.31), we have
Z Z Z 1
|K(x − y) − K(x)|dx 6 |∇K(x − θy)||y|dθdx
|x|>2|y| |x|>2|y| 0
1Z 1Z
C|y| C|y|
Z Z
6 n+1
dxdθ 6 n+1
dxdθ
0 |x|>2|y| |x − θy| 0 |x|>2|y| (|x|/2)
Z ∞
n+1 1 n+1 1
62 C|y|ωn−1 2
dr = 2 C|y|ωn−1 = 2n Cωn−1 .
2|y| r 2|y|
This completes the proof. 
Proof of Theorem 4.18. Since the proof is (essentially) a repetition of the proof
of Theorem 4.16, we will omit the details.
Let f ∈ S and T f = K ∗ f . From (4.27), it follows that
 n/2  n/2
|ω| |ω| b fˆk2
kT f k2 = kT f k2 =
c kK
2π 2π
 n/2  n/2
|ω| |ω| (4.32)
6 ˆ
kKk∞ kf k2 6 B
b kfˆk2
2π 2π
=Bkf k2 ,
by the Plancherel theorem (Theorem 1.26) and part (vi) in Proposition 1.3.
It will suffice to prove that T is of weak type (1, 1) since the strong (p, p)
inequality, 1 < p < 2, follows from the interpolation, and for p > 2 it follows
from the duality since the conjugate operator T 0 has kernel K 0 (x) = K(−x)
which also satisfies (4.27) and (4.28). In fact,
Z Z Z
hT f, ϕi = T f (x)ϕ(x)dx = K(x − y)f (y)dyϕ(x)dx
Rn Rn Rn
- 96 - 4. Singular Integrals
Z Z Z Z
= K(−(y − x))ϕ(x)dxf (y)dy = (K 0 ∗ ϕ)(y)f (y)dy
Rn Rn Rn Rn
=hf, T 0 ϕi.
To show that f is of weak type (1, 1), fix α > 0 and from the Calderón-
Zygmund decomposition of f at height α, then as in Theorem 4.16, we can
write f = g + b, where
(i) kgk1 6 kf k1 and kgk∞ 6 2n α.
R (ii) b = j bj , where each bj n+1
is supported in a dyadic cube Qj satisfying
P

Qj j
m
b (x)dx = 0 and kbj k1 6 2 α (Qj ). Furthermore, the cubes Qj and
Qk have disjoint interiors when j 6= k.
m
(iii) j (Qj ) 6 α−1 kf k1 .
P

The argument now proceeds as before, and the proof reduces to showing
that Z Z
|T bj (x)|dx 6 C |bj (x)|dx, (4.33)
Rn \Q∗j Qj
√
where Q∗j is the cube with the same center as Qj and whose sides are 2 n
times longer. Denote their common center by cj . Inequality (4.33) follows from
the Hörmander condition (4.28): since each bj has zero average, if x ∈ / Q∗j
Z Z
T bj (x) = K(x − y)bj (y)dy = [K(x − y) − K(x − cj )]bj (y)dy;
Qj Qj
hence,
Z Z Z !
|T bj (x)|dx 6 |K(x − y) − K(x − cj )|dx |bj (y)|dy.
Rn \Q∗j Qj Rn \Q∗j

However, by changing variables x − cj = x0 and y − cj = y 0 , and the fact

that |x − cj | > 2|y − cj | for all x ∈
/ Q∗j and y ∈ Qj as an obvious geometric
consideration shows, and (4.28), we get
Z Z
|K(x − y) − K(x − cj )|dx 6 |K(x0 − y 0 ) − K(x0 )|dx0 6 B.
Rn \Q∗j |x0 |>2|y 0 |
This completes the proof. 

4.4 Truncated integrals

There is still an element which may be considered unsatisfactory in our

formulation, and this is because of the following related points:
1) The L2 boundedness of the operator has been assumed via the hypothesis
that Kb ∈ L∞ and not obtained as a consequence of some condition on the
kernel K;
4.4. Truncated integrals - 97 -

2) An extraneous condition such as K ∈ L2 subsists in the hypothesis; and

for this reason our results do not directly treat the “principal-value” singular
integrals, those which exist because of the cancelation of positive and negative
values. However, from what we have done, it is now a relatively simple matter
to obtain a theorem which covers the cases of interest.
Definition 4.20. Suppose that K ∈ L1loc (Rn \ {0}) and satisfies the fol-
lowing conditions:
|K(x)| 6 B|x|−n , ∀x 6= 0,
(4.34)
Z
|K(x − y) − K(x)|dx 6 B, ∀y 6= 0,
|x|>2|y|
and Z
K(x)dx = 0, ∀0 < R1 < R2 < ∞. (4.35)
R1 <|x|<R2
Then K is called the Calderón-Zygmund kernel, where B is a constant
independent of x and y.
Theorem 4.21. Suppose that K is a Calderón-Zygmund kernel. For ε > 0
and f ∈ Lp (Rn ), 1 < p < ∞, let
Z
Tε f (x) = f (x − y)K(y)dy. (4.36)
|y|>ε
Then the following conclusions hold.
(i) We have
kTε f kp 6 Ap kf kp (4.37)
where Ap is independent of f and ε.
(ii) For any f ∈ Lp (Rn ), limε→0 Tε (f ) exists in the sense of Lp norm. That
is, there exists an operator T such that
Z
T f (x) = p.v. K(y)f (x − y)dy.
Rn
(iii) kT f kp 6 Ap kf kp for f ∈ Lp (Rn ).
Remark 4.22. 1) The linear operator T defined by (ii) of Theorem 4.21 is
called the Calderón-Zygmund singular integral operator. Tε is also called
the truncated operator of T .
2) The cancelation property alluded to is contained in condition
(4.35). This hypothesis, together with (4.34), allows us to prove the L2
boundedness and from this the Lp convergence of the truncated inte-
grals (4.37).
1
3) We should point out that the kernel K(x) = πx , x ∈ R1 , clearly sat-
isfies the hypotheses of Theorem 4.21. Therefore, we have the existence
of the Hilbert transform in the sense that if f ∈ Lp (R), 1 < p < ∞, then
- 98 - 4. Singular Integrals

1 f (x − y)
Z
lim dy
ε→0 π |y|>ε y
exists in the Lp norm and the resulting operator is bounded in Lp , as
has shown in Theorem 4.16.
For L2 boundedness, we have the following lemma.
Lemma 4.23. Suppose K satisfies the conditions (4.34) and (4.35) of the
above theorem with bound B. Let 
K(x), |x| > ε,
Kε (x) =
0, |x| < ε.
Then, we have the estimate
sup |K
b ε (ξ)| 6 CB, ε > 0, (4.38)
ξ
where C depends only on the dimension n.

Proof. First, we prove the inequality (4.38) for the special case ε = 1. Since
K̂1 (0) = 0, thus we can assume ξ 6= 0 and have
Z
K1 (ξ) = lim
b e−ωix·ξ K1 (x)dx
R→∞ |x|6R
Z Z
−ωix·ξ
= e K1 (x)dx + lim e−ωix·ξ K1 (x)dx
|x|<2π/(|ω||ξ|) R→∞ 2π/(|ω||ξ|)<|x|6R

=:I1 + I2 .
By the condition (4.35), K(x)dx = 0 which implies
R
Z1<|x|<2π/(|ω||ξ|)
K1 (x)dx = 0.
|x|<2π/(|ω||ξ|)
Thus, |x|<2π/(|ω||ξ|) e−ωix·ξ K1 (x)dx = |x|<2π/(|ω||ξ|) [e−ωix·ξ − 1]K1 (x)dx.
R R

Hence, from the fact |eiθ − 1| 6 |θ| (see Section 1.1) and the first condition in
(4.34), we get
Z Z
|I1 | 6 |ω||x||ξ||K1 (x)|dx 6 |ω|B|ξ| |x|−n+1 dx
|x|<2π/(|ω||ξ|) |x|<2π/(|ω||ξ|)
Z 2π/(|ω||ξ|)
=ωn−1 B|ω||ξ| dr = 2πωn−1 B.
0
To estimate I2 , choose z = z(ξ) such that e−ωiξ·z = −1. This choice can be
realized if z = πξ/(ω|ξ|2 ), with |z| = π/(|ω||ξ|). Since, by changing variables
x + z = y, we get
Z Z Z
−ωix·ξ −ωi(x+z)·ξ
e K1 (x)dx = − e K1 (x)dx = − e−ωiy·ξ K1 (y − z)dy
Rn n n
ZR R

=− e−ωix·ξ K1 (x − z)dx,
Rn
4.4. Truncated integrals - 99 -

which implies e−ωix·ξ K1 (x)dx = 1

e−ωix·ξ [K1 (x)−K1 (x−z)]dx, then
R R
Rn 2 Rn
we have
 Z Z 
I2 = lim − e−ωix·ξ K1 (x)dx
R→∞ |x|6R |x|62π/(|ω||ξ|)
1
Z Z
−ωix·ξ
= lim e [K1 (x) − K1 (x − z)]dx − e−ωix·ξ K1 (x)dx
2 R→∞ |x|6R |x|62π/(|ω||ξ|)
1
Z
= lim e−ωix·ξ [K1 (x) − K1 (x − z)]dx
2 R→∞ 2π/(|ω||ξ|)6|x|6R
1 1
Z Z
−ωix·ξ
− e K1 (x)dx − e−ωix·ξ K1 (x − z)dx.
2 |x|62π/(|ω||ξ|) 2 |x|62π/(|ω||ξ|)
The last two integrals are equal to, in view of the integration by parts,
1 1
Z Z
−ωix·ξ
− e K1 (x)dx − e−ωi(y+z)·ξ K1 (y)dy
2 |x|62π/(|ω||ξ|) 2 |y+z|62π/(|ω||ξ|)
1 1
Z Z
−ωix·ξ
=− e K1 (x)dx + e−ωix·ξ K1 (x)dx
2 |x|62π/(|ω||ξ|) 2 |x+z|62π/(|ω||ξ|)
1 1
Z Z
−ωix·ξ
=− e K1 (x)dx + e−ωix·ξ K1 (x)dx.
2 |x+z|>2π/(|ω||ξ|)
|x|62π/(|ω||ξ|) 2 |x|>2π/(|ω||ξ|)
|x+z|62π/(|ω||ξ|)

For the first integral, we have 2π/(|ω||ξ|) > |x| >

|x+z|−|z| > 2π/(|ω||ξ|)−π/(|ω||ξ|) = π/(|ω||ξ|),
and for the second one, 2π/(|ω||ξ|) < |x| 6 |x + −z

z| + |z| 6 3π/(|ω||ξ|). These two integrals are O

2π
|ω||ξ|

taken over a region contained in the spherical shell,

π/(|ω||ξ|) < |x| 6 3π/(|ω||ξ|) (see the figure), and
is bounded by 12 Bωn−1 ln 3 since |K1 (x)| 6 B|x|−n .
By |z| = π/(|ω||ξ|) and the condition (4.34), the
first integral of I2 is majorized by
1
Z
|K1 (x − z) − K1 (x)|dx
2 |x|>2π/(|ω||ξ|)
1 1
Z
= |K1 (x − z) − K1 (x)|dx 6 B.
2 |x|>2|z| 2
Thus, we have obtained
b 1 (ξ)| 6 2πωn−1 B + 1 B + 1 Bωn−1 ln 3 6 CB,
|K
2 2
where C depends only on n. We finish the proof for K1 .
To pass to the case of general Kε , we use a simple observation (dilation
argument) whose significance carries over to the whole theory presented in
this chapter.
Let δε be the dilation by the factor ε > 0, i.e., (δε f )(x) = f (εx). Thus if T
is a convolution operator
- 100 - 4. Singular Integrals
Z
T f (x) = ϕ ∗ f (x) = ϕ(x − y)f (y)dy,
Rn
then Z
δε−1 T δε f (x) = ϕ(ε−1 x − y)f (εy)dy
Rn Z

=ε−n ϕ(ε−1 (x − z))f (z)dz = ϕε ∗ f,

Rn
where ϕε (x) = ε ϕ(ε x). In our case, if T corresponds to the kernel K(x),
−n −1

then δε−1 T δε corresponds to the kernel ε−n K(ε−1 x). Notice that if K satisfies
the assumptions of our theorem, then ε−n K(ε−1 x) also satisfies these assump-
tions with the same bounds. (A similar remark holds for the assumptions of all
the theorems in this chapter.) Now, with our K given, let K 0 = εn K(εx). Then
K 0 satisfies the conditions of our lemma with the same bound B, and so if we
denote  0
0 K (x), |x| > 1,
K1 (x) =
0, |x| < 1,
then we know that |K b 10 (ξ)| 6 CB. The Fourier transform of ε−n K10 (ε−1 x)
b 0 (εξ) which is again bounded by CB; however ε−n K 0 (ε−1 x) = Kε (x),
is K 1 1
therefore the lemma is completely proved. 
We can now prove Theorem 4.21.
Proof of Theorem 4.21. Since K satisfies the conditions (4.34) and (4.35),
then Kε (x) satisfies the same conditions with bounds not greater than CB.
By Lemma 4.23 and Theorem 4.18, we have that the Lp boundedness of the
operators {Kε }ε>0 , are uniformly bounded.
Next, we prove that {Tε f1 }ε>0 is a Cauchy sequence in Lp provided f1 ∈
Cc (Rn ). In fact, we have
1
Z Z
Tε f1 (x) − Tη f1 (x) = K(y)f1 (x − y)dy − K(y)f1 (x − y)dy
|y|>ε |y|>η
Z
= sgn (η − ε) K(y)[f1 (x − y) − f1 (x)]dy,
min(ε,η)6|y|6max(ε,η)
because of the cancelation condition (4.35). For p ∈ (1, ∞), we get, by the
mean value theorem with some θ ∈ [0, 1], Minkowski’s inequality and (4.34),
that Z
kTε f1 − Tη f1 kp 6 |K(y)||∇f1 (x − θy)||y|dy
min(ε,η)6|y|6max(ε,η) p
Z
6 |K(y)|k∇f1 (x − θy)kp |y|dy
min(ε,η)6|y|6max(ε,η)
Z
6C |K(y)||y|dy
min(ε,η)6|y|6max(ε,η)
4.5. Singular integral operators commuted with dilations - 101 -
Z
6CB |y|−n+1 dy
min(ε,η)6|y|6max(ε,η)
Z max(ε,η)
=CBωn−1 dr
min(ε,η)

=CBωn−1 |η − ε|
which tends to 0 as ε, η → 0. Thus, we obtain Tε f1 converges in Lp as ε → 0
by the completeness of Lp .
Finally, an arbitrary f ∈ Lp can be written as f = f1 + f2 where f1 is
of the type described above and kf2 kp is small. We apply the basic inequality
(4.37) for f2 to get kTε f2 kp 6 Ckf2 kp , then we see that limε→0 Tε f exists in
Lp norm; that the limiting operator T also satisfies the inequality (4.37) is then
obvious. Thus, we complete the proof of the theorem. 

4.5 Singular integral operators commuted with dilations

In this section, we shall consider those operators which not only commute
with translations but also with dilations. Among these we shall study the class
of singular integral operators, falling under the scope of Theorem 4.21.
If T corresponds to the kernel K(x), then as we have already pointed out,
δε−1 T δε corresponds to the kernel ε−n K(ε−1 x). So if δε−1 T δε = T we are back
to the requirement K(x) = ε−n K(ε−1 x), i.e., K(εx) = ε−n K(x), ε > 0; that
is K is homogeneous of degree −n. Put another way
Ω(x)
K(x) = , (4.39)
|x|n
with Ω homogeneous of degree 0, i.e., Ω(εx) = Ω(x), ε > 0. This condition
on Ω is equivalent with the fact that it is constant on rays emanating from the
origin; in particular, Ω is completely determined by its restriction to the unit
sphere S n−1 .
Let us try to reinterpret the conditions of Theorem 4.21 in terms of Ω.
1) By (4.34), Ω(x) must be bounded and consequently integrable on S n−1 ;
and another condition |x|>2|y| Ω(x−y) − Ω(x) dx 6 C which is not easily re-
R
|x−y|n |x|n
stated precisely in terms of Ω. However, what is evident is that it requires a
certain continuity of Ω. Here we shall content ourselves in treating the case
where Ω satisfies the following “Dini-type” condition suggested by (4.34):
Z 1
0 w(η)
if w(η) := sup |Ω(x) − Ω(x )|, then dη < ∞. (4.40)
|x−x0 |6η 0 η
|x|=|x0 |=1

Of course, any Ω which is of class C 1 , or even merely Lipschitz continuous,

satisfies the condition (4.40).
- 102 - 4. Singular Integrals

2) The cancelation condition (4.35) is then the same as the condition

Z
Ω(x)dσ(x) = 0 (4.41)
S n−1
where dσ(x) is the induced Euclidean measure on S n−1 . In fact, this equation
implies that
Z R2 Z
Ω(rx0 )
Z
K(x)dx = n
dσ(x0 )rn−1 dr
R1 <|x|<R2 R1 S n−1 r
 Z
R2
= ln Ω(x0 )dσ(x0 ).
R1 S n−1

Theorem 4.24. Let Ω ∈ L∞ (S n−1 ) be homogeneous of degree 0, and suppose

that Ω satisfies the smoothness property (4.40), and the cancelation property
(4.41) above. For 1 < p < ∞, and f ∈ Lp (Rn ), let
Ω(y)
Z
Tε f (x) = n
f (x − y)dy.
|y|>ε |y|
(a) Then there exists a bound Ap (independent of f and ε) such that
kTε f kp 6 Ap kf kp .
(b) limε→0 Tε f = T f exists in Lp norm, and
kT f kp 6 Ap kf kp .
(c) If f ∈ L2 (Rn ), then the Fourier transforms of f and T f are related by
Tcf (ξ) = m(ξ)fˆ(ξ), where m is a homogeneous function of degree 0. Explic-
itly,
 
πi
Z
m(ξ) = − sgn (ω) sgn (ξ · x) + ln(1/|ξ · x|) Ω(x)dσ(x), |ξ| = 1.
S n−1 2
(4.42)
Proof. The conclusions (a) and (b) are immediately consequences of Theorem
4.21, once we have shown that any K(x) of the form Ω(x)
|x|n
satisfies
Z
|K(x − y) − K(x)|dx 6 B, (4.43)
|x|>2|y|
if Ω is as in condition (4.40). Indeed,
Ω(x − y) − Ω(x)
 
1 1
K(x − y) − K(x) = + Ω(x) − .
|x − y|n |x − y|n |x|n
The second group of terms is bounded since Ω is bounded and
1 1 |x|n − |x − y|n
Z Z
n
− dx = dx
|x|>2|y| |x − y| |x|n |x|>2|y| |x − y|n |x|n
||x| − |x − y|| n−1 n−1−j
|x − y|j
P
j=0 |x|
Z
= dx
|x|>2|y| |x − y|n |x|n
4.5. Singular integral operators commuted with dilations - 103 -

Z n−1
X
6 |y| |x|−j−1 |x − y|j−n dx
|x|>2|y| j=0
Z n−1
X
6 |y| |x|−j−1 (|x|/2)j−n dx (since |x − y| > |x| − |y| > |x|/2)
|x|>2|y| j=0
Z n−1
X Z
−n−1
= |y| 2n−j
|x| dx = 2(2 − 1)|y| n
|x|−n−1 dx
|x|>2|y| j=0 |x|>2|y|

1
=2(2n − 1)|y|ωn−1 = (2n − 1)ωn−1 .
2|y|
To estimate the first group of terms, we notice that if |x| > 2|y|, the distance
|P Q| between the projections of x − y and x on the unit sphere as in the
following picture.

x P: x x
|x|
x−y
1 y Q: |x−y| 1
y
P P x−y
x−y
Q Q
θ

O O

|y|
Case 1: |x| ≥ |x − y|, sin θ ≤ |x| Case 2: |x| ≤ |x − y|, sin θ ≤ |y|
≤ |y|
|x−y| |x|

sin π−θ
By the sine theorem, we have sin θ
|P Q|
where |OP | = 1. Since |y| 6
= 2
|OP |q √
|x|/2, we have θ 6 π2 and so cos θ > 0. Thus, cos 2θ = 1+cos θ
2
> 1/ 2.
Then, we have
x−y x sin θ sin θ √ |y| |y|
− = |P Q| = π θ
= θ
6 2 62
|x − y| |x| sin( 2 − 2 ) cos 2 |x| |x|
|y|
since sin θ 6 |x| for both cases. So the integral corresponding to the first group
of terms is dominated by
Z ∞
|y| dx
 
dz dr
Z Z
n n n
2 w 2 n
=2 w(2/|z|) n = 2 ωn−1 w(2/r)
|x|>2|y| |x| |x| |z|>2 |z| 2 r
Z 1
w(η)dη
=2n ωn−1 <∞
0 η
in view of changes of variables x = |y|z and the Dini-type condition (4.40).
Now, we prove (c). Since T is a bounded linear operator on L2 which com-
mutes with translations, we know, by Theorem 1.62 and Proposition 1.3, that
T can be realized in terms of a multiplier m such that Tcf (ξ) = m(ξ)fˆ(ξ). For
such operators, the fact that they commute with dilations is equivalent with the
property that the multiplier is homogeneous of degree 0.
- 104 - 4. Singular Integrals

For our particular operators we have not only the existence of m but also
an explicit expression of the multiplier in terms of the kernel. This formula is
deduced as follows.
Since K(x) is not integrable, we first consider its truncated function. Let
0 < ε < η < ∞, and

 Ω(x)
, ε 6 |x| 6 η,
Kε,η (x) = |x|n
0, otherwise.


Clearly, Kε,η ∈ L1 (Rn ). If f ∈ L2 (Rn ) then K\ ˆ

ε,η ∗ f (ξ) = Kε,η (ξ)f (ξ).
d
We shall prove two facts about K d ε,η (ξ).
(i) supξ |K dε,η (ξ)| 6 A, with A independent of ε and η;
(ii) if ξ 6= 0, lim ε→0 K d ε,η (ξ) = m(ξ), see (4.42).
η→∞
For this purpose, it is convenient to introduce polar coordinates. Let x = rx0 ,
r = |x|, x0 = x/|x| ∈ S n−1 , and ξ = Rξ 0 , R = |ξ|, ξ 0 = ξ/|ξ| ∈ S n−1 . Then
we have
Ω(x)
Z Z
−ωix·ξ
Kε,η (ξ) =
d e Kε,η (x)dx = e−ωix·ξ dx
Rn ε6|x|6η |x|n
Z Z η 
0 −ωiRrx0 ·ξ 0 −n n−1
= Ω(x ) e r r dr dσ(x0 )
n−1
ZS Zε η 
0 −ωiRrx0 ·ξ 0 dr
= Ω(x ) e dσ(x0 ).
S n−1 ε r
Since Z
Ω(x0 )dσ(x0 ) = 0,
S n−1
we can introduce the factor cos(|ω|Rr) (which does not depend on x0 ) in the
integral defining K d ε,η (ξ). We shall also need the auxiliary integral
Z η
0 0 0 dr
Iε,η (ξ, x ) = [e−ωiRrx ·ξ − cos(|ω|Rr)] , R > 0.
ε r
Thus, it follows Z
K
d ε,η (ξ) = Iε,η (ξ, x0 )Ω(x0 )dσ(x0 ).
S n−1
Now, we first consider Iε,η (ξ, x0 ). For its imaginary part, we have, by chang-
ing variable ωRr(x0 · ξ 0 ) = t, that
Z η
0 sin ωRr(x0 · ξ 0 )
=Iε,η (ξ, x ) = − dr
ε r
Z |ω|Rη(x0 ·ξ0 )
0 0 sin t
= − sgn (ω) sgn (x · ξ ) dt,
|ω|Rε(x0 ·ξ 0 ) t
converges to
4.5. Singular integral operators commuted with dilations - 105 -
∞
sin t π
Z
0 0
− sgn (ω) sgn (x · ξ ) dt = − sgn (ω) sgn (x0 · ξ 0 ),
0 t 2
as ε → 0 and η → ∞.
For its real part, since cos r is an even function, we have
Z η
0 dr
<Iε,η (ξ, x ) = [cos(|ω|Rr|x0 · ξ 0 |) − cos(|ω|Rr)] .
ε r
If x · ξ = ±1, then <Iε,η (ξ, x ) = 0. Now we assume 0 < ε < 1 < η. For the
0 0 0

case x0 · ξ 0 6= ±1, we get the absolute value of its real part

Z 1
0 |ω| |ω| dr
|<Iε,η (ξ, x )| 6 −2 sin Rr(|x0 · ξ 0 | + 1) sin Rr(|x0 · ξ 0 | − 1)
ε 2 2 r
Z η Z η
dr dr
+ cos |ω|Rr|x0 · ξ 0 | − cos |ω|Rr
1 r 1 r
1
|ω|2 2
Z
6 R (1 − |x0 · ξ 0 |2 ) rdr
2 ε
Z |ω|Rη|ξ0 ·x0 | Z |ω|Rη
cos t cos t
+ dt − dt
|ω|R|ξ 0 ·x0 | t |ω|R t
|ω|2 2
6 R + I1 .
4
If η|ξ 0 · x0 | > 1, then we have
Z |ω|R Z |ω|Rη
cos t cos t
I1 = dt − dt
|ω|R|ξ 0 ·x0 | t |ω|Rη|ξ 0 ·x0 | t
Z |ω|R Z |ω|Rη
dt dt
6 +
|ω|R|ξ 0 ·x0 | t |ω|Rη|ξ 0 ·x0 | t

62 ln(1/|ξ 0 · x0 |).
If 0 < η|ξ 0 · x0 | 6 1, then
|ω|R/|ξ 0 ·x0 |
dt
Z
I1 6 6 2 ln(1/|ξ 0 · x0 |).
|ω|R|ξ 0 ·x0 | t
Thus,
|ω|2 2
0
|<Iε,η (ξ, x )| 6 R + 2 ln(1/|ξ 0 · x0 |),
4
and so the real part converges as ε → 0 and η → ∞. By the fundamental
theorem of calculus, we can write
Z η Z ηZ λ Z λZ η
cos(λr) − cos(µr)
dr = − sin(tr)dtdr = − sin(tr)drdt
ε r ε µ µ ε
Z λZ η Z λ
∂r cos(tr) cos(tη) − cos(tε)
= drdt = dt
µ ε t µ t
- 106 - 4. Singular Integrals

λη λ λη λ
cos(s) cos(tε) sin s sin s cos(tε)
Z Z λη
Z Z
= ds − dt = + ds − dt
µη s µ t s µη µη s2 µ t
λ
1
Z
→0 − dt = − ln(λ/µ) = ln(µ/λ), as η → ∞, ε → 0.
µ t
Take λ = |ω|R|x0 · ξ 0 |, and µ = |ω|R. So
Z ∞
0 0 0 dr
lim <(I ε,η (ξ, x )) = [cos |ω|Rr(x · ξ ) − cos |ω|Rr] = ln(1/|x0 · ξ 0 |).
ε→0
η→∞ 0 r
By the properties of Iε,η just proved, we have
2
|ω|
 
π
Z
|Kdε,η (ξ)| 6 + 2 0 0
R + 2 ln(1/|ξ · x |) |Ω(x0 )|dσ(x0 )
S n−1 2 4
π |ω|2 2
Z
6C( + R )ωn−1 + 2C ln(1/|ξ 0 · x0 |)dσ(x0 ).
2 4 S n−1
For n = 1, we have S = {−1, 1} and then S n−1 ln(1/|ξ 0 · x0 |)dσ(x0 ) =
0
R

2 ln 1 = 0. For n > 2, we can pick an orthogonal matrix A such that Ae1 = ξ 0 ,

and so by changes of variables and using the notation ȳ = (y2 , y3 , ..., yn ),
Z Z
0 0 0
ln(1/|ξ · x |)dσ(x ) = ln(1/|Ae1 · x0 |)dσ(x0 )
n−1 n−1
ZS S Z
−1 0
−1 0 0 A x =y
= ln(1/|e1 · A x |)dσ(x ) ==== ln(1/|e1 · y|)dσ(y)
S n−1 S n−1
Z 1
dy1
Z Z
= ln(1/|y1 |)dσ(y) = ln(1/|y1 |) √ dσ(ȳ) p
S n−1 −1 1−y12 S n−2 1 − y12
√ 2 Z 1 Z
z̄=ȳ/ 1−y1
====== ln(1/|y1 |) (1 − y12 )(n−3)/2 dσ(z̄)dy1
−1 S n−2
Z 1
=ωn−2 ln(1/|y1 |)(1 − y12 )(n−3)/2 dy1
−1
Z 1
=2ωn−2 ln(1/|y1 |)(1 − y12 )(n−3)/2 dy1
0
Z π/2
y1 =cos θ
====2ωn−2 ln(1/ cos θ)(sin θ)n−2 dθ = 2ωn−2 I2 .
0
For n > 3, we have, by integration by parts,
Z π/2 Z π/2
I2 6 ln(1/ cos θ) sin θdθ = sin θdθ = 1.
0 0
R π/2
For n = 2, we have, by the formula 0 ln(cos θ)dθ = − π2 ln 2 (see [GR,
4.225.3, p.531]),
Z π/2 Z π/2
π
I2 = ln(1/ cos θ)dθ = − ln(cos θ)dθ = ln 2.
0 0 2
Hence, S n−1 ln(1/|ξ · x |)dσ(x ) 6 C for any ξ ∈ S .
0 0 0 0 n−1
R
4.6. The maximal singular integral operator - 107 -

Thus, we have proved the uniform boundedness of K dε,η (ξ), i.e., (i). In view
of the limit of Iε,η (ξ, x ) as ε → 0, η → ∞ just proved, and the dominated
0

convergence theorem, we get

lim K
ε→0
d ε,η (ξ) = m(ξ),
η→∞

if ξ 6= 0, that is (ii).
By the Plancherel theorem, if f ∈ L2 (Rn ), Kε,η ∗ f converges in L2 norm
as ε → 0 and η → ∞, and the Fourier transform of this limit is ˆ
R m(ξ)f (ξ).
However, if we keep ε fixedRand let η → ∞, then clearly Kε,η (y)f (x −
y)dy converges everywhere to |y|>ε K(y)f (x − y)dy, which is Tε f .
Letting now ε → 0, we obtain the conclusion (c) and our theorem is com-
pletely proved. 
Remark 4.25. 1) In the theorem, the condition that Ω is mean zero on
S n−1 is necessary and cannot be neglected. Since in the estimate
Z 
Ω(y) Ω(y)
Z Z
n
f (x − y)dy = + f (x − y)dy,
Rn |y| |y|61 |y|>1 |y|n
the main difficulty lies in the first integral. For instance, if we assume
Ω(x) ≡ 1, f is a nonzero constant, then this integral is divergent.
2) From the formula of the symbol m(ξ), it is homogeneous of degree
0 in view of the mean zero property of Ω.
3) The proof of part (c) holds under very general conditions on Ω.
Write Ω = Ωe + Ωo where Ωe is the even part of Ω, Ωe (x) = Ωe (−x),
and Ωo (x) is the odd part, Ωo (−x) = −Ωo (x). Then, because of the uni-
form boundedness of the sine integral, i.e., =Iε,η (ξ, x0 ), we required only
|Ωo (x0 )|dσ(x0 ) < ∞, i.e., the integrability of the odd part. For the
R
S n−1
even part, the proof requires the uniform boundedness of
Z
|Ωe (x0 )| ln(1/|ξ 0 · x0 |)dσ(x0 ).
S n−1
This observation is suggestive of certain generalizations of Theorem
4.21, see [Ste70, §6.5, p.49–50].

4.6 The maximal singular integral operator

Theorem 4.24 guaranteed the existence of the singular integral transforma-

tion
Ω(y)
Z
lim f (x − y)dy (4.44)
ε→0 |y|>ε |y|n

in the sense of convergence in the Lp norm. The natural counterpart of this

result is that of convergence almost everywhere. For the questions involving
- 108 - 4. Singular Integrals

almost everywhere convergence, it is best to consider also the corresponding

maximal function.
Theorem 4.26. Suppose that Ω satisfies the conditions of the previous theo-
rem. For f ∈ Lp (Rn ), 1 6 p < ∞, consider
Ω(y)
Z
Tε f (x) = n
f (x − y)dy, ε > 0.
|y|>ε |y|
(The integral converges absolutely for every x.)
(a) limε→0 Tε f (x) exists for almost every x.
(b) Let T ∗ f (x) = supε>0 |Tε f (x)|. If f ∈ L1 (Rn ), then the mapping f →
T ∗ f is of weak type (1, 1).
(c) If 1 < p < ∞, then kT ∗ f kp 6 Ap kf kp .

Proof. The argument for the theorem presents itself in three stages.
The first one is the proof of inequality (c) which can be obtained as a rela-
tively easy consequence of the Lp norm existence of limε→0 Tε , already proved,
and certain general properties of “approximations to the identity”.
Let T f (x) = limε→0 Tε f (x), where the limit is taken in the Lp norm. Its
existence is guaranteed by Theorem 4.24. We shall prove this part by showing
the following Cotlar inequality
T ∗ f (x) 6 M (T f )(x) + CM f (x).
Let ϕ be a smooth non-negative function on Rn , which is supported in the
unit ball, has integral equal to one, and which is also radial and decreasing in
|x|. Consider
(
Ω(x)
, |x| > ε,
Kε (x) = |x|n
0, |x| < ε.
This leads us to another function Φ defined by
Φ = ϕ ∗ K − K1 , (4.45)
where ϕ ∗ K = limε→0 ϕ ∗ Kε = limε→0 |x−y|>ε K(x − y)ϕ(y)dy.
R

We shall need to prove that the smallest decreasing radial majorant of Φ is

integrable (so as to apply Theorem 4.10). In fact, if |x| < 1, then
Z Z
|Φ| =|ϕ ∗ K| = K(y)ϕ(x − y)dy = K(y)(ϕ(x − y) − ϕ(x))dy
Rn Rn
|ϕ(x − y) − ϕ(x)|
Z Z
6 |K(y)||ϕ(x − y) − ϕ(x)|dy 6 C dy 6 C,
Rn Rn |y|n
since (4.41) implies Rn K(y)dy = 0 and by the smoothness of ϕ.
R

If 1 6 |x| 6 2, then Φ = ϕ ∗ K − K is again bounded by the same reason

and K is bounded in this case.
Finally if |x| > 2,
4.6. The maximal singular integral operator - 109 -
Z Z
Φ(x) = K(x − y)ϕ(y)dy − K(x) = [K(x − y) − K(x)]ϕ(y)dy.
Rn |y|61
Similar to (4.43), we can get the bound for |y| 6 1
Z Z
|K(x − y) − K(x)|dx 6 |K(x − y) − K(x)|dx 6 C.
|x|>2 |x|>2|y|
Thus we obtain Z Z
|Φ(x)|dx 6C ϕ(y)dy 6 C.
|x|>2 |y|61
Therefore, we have proved that Φ ∈ L1 (R ) from three cases discussed above.
n

From (4.45), it follows, because the singular integral operator ϕ → ϕ ∗ K

commutes with dilations, that
ϕε ∗ K − Kε = Φε , with Φε (x) = ε−n Φ(x/ε). (4.46)
Now, we claim that for any f ∈ Lp (Rn ), 1 < p < ∞,
(ϕε ∗ K) ∗ f (x) = T f ∗ ϕε (x), (4.47)
where the identity holds for every x. In fact, we notice first that
(ϕε ∗ Kδ ) ∗ f (x) = Tδ f ∗ ϕε (x), for every δ > 0 (4.48)
because both sides of R(4.48)Rare equal for each x to the absolutely con-
vergent double integral z∈Rn |y|>δ K(y)f (z − y)ϕε (x − z)dydz. Moreover,
ϕε ∈ Lq (Rn ), with 1 < q < ∞ and 1/p + 1/q = 1, so ϕε ∗ Kδ → ϕε ∗ K in
Lq norm, and Tδ f → T f in Lp norm, as δ → 0, by Theorem 4.24. This proves
(4.47), and so by (4.46)
Tε f = Kε ∗ f = ϕε ∗ K ∗ f − Φε ∗ f = T f ∗ ϕε − f ∗ Φε .
Passing to the supremum over ε and applying Theorem 4.10, part (a), The-
orem 3.9 for maximal funtions and Theorem 4.24, we get
kT ∗ f kp 6k sup |T f ∗ ϕε |kp + k sup |f ∗ Φε |kp
ε>0 ε>0

6CkM (T f )kp + CkM f kp 6 CkT f kp + Ckf kp 6 Ckf kp .

Thus, we have proved (c).
The second and most difficult stage of the proof is the conclusion (b). Here
the argument proceeds in the main as in the proof of the weak type (1, 1) result
for singular integrals in Theorem 4.18. We review it with deliberate brevity so
as to avoid a repetition of details already examined.
For a given α > 0, we split f = g + b as in the proof of Theorem 4.18. We
also consider for each cube Qj its mate Q∗j , which has the same center cj but
√
whose side length is expanded 2 n times. The following geometric remarks
concerning these cubes are nearly obvious (The first one has given in the proof
of Theorem 4.18).
(i) If x ∈
/ Q∗j , then |x − cj | > 2|y − cj | for all y ∈ Qj , as an obvious
geometric consideration shows.
 (ii) Suppose x ∈ Rn \ Q∗j and assume that for some y ∈ Q j , |x − y| = ε. Then the
at x, of radius γn ε, contains Q j , i.e. B(x, r) ⊃ Q j , if r = γn ε.
(iii) Under the same hypotheses as (ii), we have that |x − y| > γn0 ε, for every y
Here γn and γn0 depend only on the dimension n, and not the particular cube Q
- 110 - 4. Singular Integrals

(ii) Suppose x ∈ Rn \ Q∗j and assume that

for some y ∈ Qj , |x − y| = ε. Then the closed B(x, r)
γn ε
ball centered at x, of radius γn ε, contains Qj ,
i.e., B(x, r) ⊃ Qj , if r = γn ε.
γn0 ε x
(iii) Under the same hypotheses as (ii), we Qj
y
have that |x − y| > γn0 ε, for every y ∈ Qj . Q∗j
ε

Here γn and γn0 depend only on the dimen- Rn \ ∪ j Q∗j

sion n, and not the particular cube Qj .
With these observations, and following the
development in the proof of Theorem 4.18, we Fig. 4.1 Observation for (ii) and (iii)
Observation for (ii) and (iii)
shall prove that if x ∈ R \ ∪j Qj ,
n ∗

XZ With these observations, and following the development in the proof of Th

sup |Tε b(x)| 6 prove
|K(x −that Rn \ ∪
y)if−x ∈K(x −j Qc∗jj, )||b(y)|dy
ε>0 Qj
j XZ (4.49)
1
Z sup |T ε b(x)| 6 |K(x − y) − K(x − c j )||b(y)|dy
ε>0 Qj
+ C sup |b(y)|dy, j
Z
r>0 m(B(x, r)) B(x,r) 1
+ C sup |b(y)|dy,
m(B(x, r))
with K(x) = Ω(x) |x|n
. r>0 B(x,r)

The addition of the maximal function

with K(x)to the
|x| . r.h.s of (4.49) is the main new
= Ω(x) n

The addition of the maximal function to the r.h.s of (6.23) is the main new ele
element of the proof.
To prove (6.23), fix x ∈ Rn \ ∪ j Q∗j , and ε > 0. Now the cubes Q j fall into thre
To prove (4.49), fix x ∈ Rn \ ∪j Q1)∗j for
, and
all yε∈ > |x −Now
Q j ,0. y| < ε;the cubes Qj fall into
three classes: 2) for all y ∈ Q j , |x − y| > ε;
1) for all y ∈ Qj , |x − y| < ε; 3) there is a y ∈ Q j , such that |x − y| = ε.
We now examine
2) for all y ∈ Qj , |x − y| > ε; XZ
3) there is a y ∈ Qj , such that |x − y| = ε. T ε b(x) =
Q
Kε (x − y)b(y)dy.
j j

We now examine
X ZCase 1). Kε (x − y) = 0 if |x − y| < ε, and so the integral over the cube Q j in (6
Tε b(x) = CaseK2).
ε (xKε−(x −y)b(y)dy.
y) = K(x − y), if |x − y| > ε, and(4.50)
therefore this integral over Q
j Qj Z Z
Case 1). Kε (x − y) = 0 if |x − y| < ε, and so theQ K(x − y)b(y)dy [K(x −Q
integral over the cube
= y) − K(x − c j )]b(y)dy.
j Q j j

in (4.50) is zero.
This term is majorized in absolute value by
Case 2). Kε (x − y) = K(x − y), if |x − y| > ε, and therefore this integral
over Qj equals
Z Z
K(x − y)b(y)dy = [K(x − y) − K(x − cj )]b(y)dy.
Qj Qj
This term is majorized in absolute value by
Z
|K(x − y) − K(x − cj )||b(y)|dy,
Qj
which expression appears in the r.h.s. of (4.49).
Case 3). We write simply
Z Z
Kε (x − y)b(y)dy 6 |Kε (x − y)||b(y)|dy
Qj Qj
4.6. The maximal singular integral operator - 111 -
Z
= |Kε (x − y)||b(y)|dy,
Qj ∩B(x,r)
by (ii), with r = γn ε. However, by (iii) and the fact that Ω is bounded, we have
Ω(x − y) C
|Kε (x − y)| = n
6 0 n.
|x − y| (γn ε)
Thus, in this case,
C
Z Z
Kε (x − y)b(y)dy 6 |b(y)|dy.
Qj m(B(x, r)) Qj ∩B(x,r)
If we add over all cubes Qj , we finally obtain, for r = γn ε,
XZ
|Tε b(x)| 6 |K(x − y) − K(x − cj )||b(y)|dy
j Qj

C
Z
+ |b(y)|dy.
m(B(x, r)) B(x,r)
Taking the supremum over ε gives (4.49).
This inequality can be written in the form
|T ∗ b(x)| 6 Σ + CM b(x), x ∈ F ∗,
and so
m({x ∈ R n
\ ∪j Q∗j : |T ∗ b(x)| > α/2})
6m({x ∈ R n
\ ∪j Q∗j : Σ > α/4}) + m({x ∈ R n
\ ∪j Q∗j : CM b(x) > α/4}).
R The first term in the r.h.s. is similar to (4.33), and we can get
Rn \∪j Q∗
j
n
m
Σ(x)dx 6 Ckbk1 which implies ({x ∈ R \ ∪j Q∗j : Σ > α/4}) 6
4C
α
kbk1 .
For the second one, by Theorem 3.9, i.e., the weak type estimate for the
maximal function M , we get m({x ∈ Rn \ ∪j Q∗j : CM b(x) > α/4}) 6
C
α
kbk1 .
The weak type (1, 1) property of T ∗ then follows as in the proof of the same
property for T , in Theorem 4.18 for more details.
The final stage of the proof, the passage from the inequalities of T ∗ to the ex-
istence of the limits almost everywhere, follows the familiar pattern described
in the proof of the Lebesgue differential theorem (i.e., Theorem3.13).
More precisely, for any f ∈ Lp (Rn ), 1 6 p < ∞, let

Λf (x) = lim sup Tε f (x) − lim inf Tε f (x) .

ε→0 ε→0

Clearly, Λf (x) 6 2T f (x). Now write f = f1 + f2 where f1 ∈ Cc1 , and

∗

kf2 kp 6 δ.
We have already proved in the proof of Theorem 4.21 that Tε f1 con-
verges uniformly as ε → 0, so Λf1 (x) ≡ 0. By (4.37), we have kΛf2 kp 6
2Ap kf2 kp 6 2Ap δ if 1 < p < ∞. This shows Λf2 = 0, almost everywhere,
- 112 - 4. Singular Integrals

thus by Λf (x) 6 Λf1 (x) + Λf2 (x), we have Λf = 0 almost everywhere. So

limε→0 Tε f exists almost everywhere if 1 < p < ∞.
In the case p = 1, we get similarly
m A
({x : Λf (x) > α}) 6 kf2 k1 6
α
Aδ
α
,
and so again Λf (x) = 0 almost everywhere, which implies that limε→0 Tε f (x)
exists almost everywhere. 

4.7 *Vector-valued analogues

It is interesting to point out that the results of this chapter, where our func-
tions were assumes to take real or complex values, can be extended to the case
of functions taking their values in a Hilbert space. We present this generaliza-
tion because it can be put to good use in several problems. An indication of
this usefulness is given in the Littlewood-Paley theory.
We begin by reviewing quickly certain aspects of integration theory in this
context.
Let H be a separable Hilbert space. Then a function f (x), from Rn to H
is measurable if the scalar valued functions (f (x), ϕ) are measurable, where
(·, ·) denotes the inner product of H , and ϕ denotes an arbitrary vector of H .
If f (x) is such a measurable function, then |f (x)| is also measurable (as a
function with non-negative values), where | · | denotes the norm of H .
Thus, Lp (Rn , H ) is defined as the equivalent classes of measurable
functions f (x) from Rn to H , with the property that the norm kf kp =
( Rn |f (x)|p dx)1/p is finite, when p < ∞; when p = ∞ there is a similar
R

definition, except kf k∞ = ess sup |f (x)|.

Next, let H1 and H2 be two separable Hilbert spaces, and let L(H1 , H2 )
denote the Banach space of bounded linear operators from H1 to H2 , with the
usual operator norm.
We say that a function f (x), from Rn to L(H1 , H2 ) is measurable if f (x)ϕ
is an H2 -valued measurable function for every ϕ ∈ H1 . In this case |f (x)| is
also measurable and we can define the space Lp (Rn , L(H1 , H2 )), as before;
here again | · | denotes the norm, this time in L(H1 , H2 ).
The usual facts about convolution hold in this setting. For example, R suppose
K(x) ∈ L (R , L(H1 , H2 )) and f (x) ∈ L (R , H1 ), then g(x) = Rn K(x−
q n p n

y)f (y)dy converges in the norm of H2 for almost every x, and

Z Z
|g(x)| 6 |K(x − y)f (y)|dy 6 |K(x − y)||f (y)|dy.
Rn Rn
Also kgkr 6 kKkq kf kp , if 1/r = 1/p + 1/q − 1, with 1 6 r 6 ∞.
4.7. *Vector-valued analogues - 113 -

Suppose that Rf (x) ∈ L1 (Rn , H ). Then we can define its Fourier trans-
form fˆ(ξ) = Rn e−ωix·ξ f (x)dx which is an element of L∞ (Rn , H ). If
f ∈ L1 (Rn , H ) ∩ L2 (Rn , H ), then fˆ(ξ) ∈ L2 (Rn , H ) with kfˆk2 =
 −n/2
|ω|
2π
kf k2 . The Fourier transform can then be extended by continuity to
a unitary mapping of the Hilbert space L2 (Rn , H ) to itself, up to a constant
multiplication.
These facts can be obtained easily from the scalar-valued case by introduc-
ing an arbitrary orthonormal basis in H .
Now suppose that H1 and H2 are two given Hilbert spaces. Assume that
f (x) takes values in H1 , and K(x) takes values in L(H1 , H2 ). Then
Z
T f (x) = K(y)f (x − y)dy,
Rn
whenever defined, takes values in H2 .
Theorem 4.27. The results in this chapter, in particular Theorem 4.18, Propo-
sition 4.19, Theorems 4.21, 4.24 and 4.26 are valid in the more general context
where f takes its value in H1 , K takes its values in L(H1 , H2 ) and T f and
Tε f take their value in H2 , and where throughout the absolute value | · | is
replaced by the appropriate norm in H1 , L(H1 , H2 ) or H2 respectively.
This theorem is not a corollary of the scalar-valued case treated in any obvi-
ous way. However, its proof consists of nothing but a identical repetition of the
arguments given for the scalar-valued case, if we take into account the remarks
made in the above paragraphs. So, we leave the proof to the interested reader.
Remark 4.28. 1) The final bounds obtained do not depend on the Hilbert
spaces H1 or H2 , but only on B, p, and n, as in the scalar-valued case.
2) Most of the argument goes through in the even greater generality
of Banach space-valued functions, appropriately defined. The Hilbert
space structure is used only in the L2 theory when applying the variant
of Plancherel’s formula.
The Hilbert space structure also enters in the following corollary.
Corollary 4.29. With the same assumptions as in Theorem 4.27, if in addition
kT f k2 = ckf k2 , c > 0, f ∈ L2 (Rn , H1 ),
then kf kp 6 A0p kT f kp , if f ∈ Lp (Rn , H1 ), if 1 < p < ∞.

Proof. We remark that the L2 (Rn , Hj ) are Hilbert spaces. In fact, let (·, ·)j de-
note the inner product of Hj , j = 1, 2, and let h·, ·ij denote the corresponding
inner product in L2 (Rn , Hj ); that is
Z
hf, gij = (f (x), g(x))j dx.
Rn
- 114 - 4. Singular Integrals

Now T is a bounded linear transformation from the Hilbert space

L (Rn , H1 ) to the Hilbert space L2 (Rn , H2 ), and so by the general the-
2

ory of inner products there exists a unique adjoint transformation T̃ , from

L2 (Rn , H2 ) to L2 (Rn , H1 ), which satisfies the characterizing property
hT f1 , f2 i2 = hf1 , T̃ f2 i1 , with fj ∈ L2 (Rn , Hj ).
But our assumption is equivalent with the identity (see the theory of Hilbert
spaces, e.g. [Din07, Chapter 6])
hT f, T gi2 = c2 hf, gi1 , for all f, g ∈ L2 (Rn , H1 ).
Thus using the definition of the adjoint, hT̃ T f, gi1 = c2 hf, gi1 , and so the
assumption can be restated as
T̃ T f = c2 f, f ∈ L2 (Rn , H1 ). (4.51)
T̃ is again an operator of the same kind as T but it takes function with values
in H2 to functions with values in H1 , and its kernel K̃(x) = K ∗ (−x), where
∗ denotes the adjoint of an element in L(H1 , H2 ).
This is obviousZon Zthe formal level since
hT f1 , f2 i2 = (K(x − y)f1 (y), f2 (x))2 dydx
n n
ZR ZR
= (f1 (y), K ∗ (−(y − x))f2 (x))1 dxdy = hf1 , T̃ f2 i1 .
Rn Rn
The rigorous justification of this identity is achieved by a simple limiting ar-
gument. We will not tire the reader with the routine details.
This being said we have only to add the remark that K ∗ (−x) satisfies the
same conditions as K(x), and so we have, for it, similar conclusions as for K
(with the same bounds). Thus by (4.51),
c2 kf kp = kT̃ T f kp 6 Ap kT f kp .
This proves the corollary with A0p = Ap /c2 . 
Remark 4.30. This corollary applies in particular to the singular in-
tegrals commuted with dilations, then the condition required is that
the multiplier m(ξ) have constant absolute value. This is the case, for
1
example, when T is the Hilbert transform, K(x) = πx , and m(ξ) =
−i sgn (ω) sgn (ξ).
L ECTURE N OTES ON c 2016 by Chengchun Hao
Introduction to Harmonic Analysis Email: hcc@amss.ac.cn

Chapter 5
Riesz Transforms and Spherical Harmonics

5.1 The Riesz transforms

We look for the operators in Rn which have the analogous structural charac-
terization as the Hilbert transform. We begin by making a few remarks about
the interaction of rotations with the n-dimensional Fourier transform. We shall
need the following elementary observation.
Let ρ denote any rotation about the origin in Rn . Denote also by ρ its induced
action on functions, ρ(f )(x) = f (ρx). Then
Z Z
−ωix·ξ −1
(F ρ)f (ξ) = e f (ρx)dx = e−ωiρ y·ξ f (y)dy
n Rn
ZR
= e−ωiy·ρξ f (y)dy = F f (ρξ) = ρF f (ξ),
Rn
that is,
F ρ = ρF .
Let `(x) = (`1 (x), `2 (x), ..., `n (x)) be an n-tuple of functions defined on
R . For any rotation ρ about the origin, write ρ = (ρjk ) for its matrix realiza-
n

tion. Suppose that ` transforms like a vector. Symbolically this can be written
as
`(ρx) = ρ(`(x)),
or more explicitly
X
`j (ρx) = ρjk `k (x), for every rotation ρ. (5.1)
k

Lemma 5.1. Suppose ` is homogeneous of degree 0, i.e., `(εx) = `(x), for

x
ε > 0. If ` transforms according to (5.1) then `(x) = c |x| for some constant c;
that is
xj
`j (x) = c . (5.2)
|x|

115
- 116 - 5. Riesz Transforms and Spherical Harmonics

Proof. It suffices to consider x ∈ S n−1 due to the homogeneousness of degree

0 for `. Now, let e1 , e2 , ..., en denote the usual unit vectors along the axes. Set
c = `1 (e1 ). We can see that `j (e1 ) = 0, if j 6= 1.
In fact, we take a rotation arbitrarily such that e1 fixed under the acting
of ρ, i.e., ρe1 = e1 . Thus, we also have e1 = ρ−1 ρe1 = ρ−1 e1 = ρ> e1 .
From ρe1 = ρ> e1 = e1 , we get ρ11 = 1 and ρ1k = ρj1 = 0 for k 6= 1
   −1  
10 10 1 0
and j 6= 1. So ρ = . Because = and ρ−1 = ρ> ,
0A 0A 0 A−1
we obtain A−1 = A> and det A = 1, i.e., A is a rotation in Rn−1 . On the
other hand, by (5.1), we get `j (e1 ) = nk=2 ρjk `k (e1 ) for j = 2, ..., n. That
P

is, the n − 1 dimensional vector (`2 (e1 ), `3 (e1 ), · · · , `n (e1 )) is left fixed by all
the rotations on this n − 1 dimensional vector space. Thus, we have to take
`2 (e1 ) = `3 (e1 ) = · · · = `n (e1 ) = 0.
Inserting again in (5.1) gives `j (ρe1 ) = ρj1 `1 (e1 ) = cρj1 . If we take a
rotation such that ρe1 = x, then we have ρj1 = xj , so `j (x) = cxj , (|x| = 1),
which proves the lemma. 
We now define the n Riesz transforms. For f ∈ Lp (Rn ), 1 6 p < ∞, we set
yj
Z
Rj f (x) = lim cn n+1
f (x − y)dy, j = 1, ..., n, (5.3)
ε→0 |y|>ε |y|
Γ ((n+1)/2) π (n+1)/2
with cn = π (n+1)/2
where 1/cn = Γ ((n+1)/2)
is half the surface area of the
Ω (x)
unit sphere S of R . Thus, Rj is defined by the kernel Kj (x) = |x|
n n+1 j
n , and
xj
Ωj (x) = cn |x| .
Next, we derive the multipliers which correspond to the Riesz transforms,
and which in fact justify their definition. Denote
Ω(x) = (Ω1 (x), Ω2 (x), ..., Ωn (x)), and m(ξ) = (m1 (ξ), m2 (ξ), ..., mn (ξ)).
Let us recall the formula (4.42), i.e.,
Z
m(ξ) = Φ(ξ · x)Ω(x)dσ(x), |ξ| = 1, (5.4)
S n−1
with Φ(t) = − πi
2
sgn (ω) sgn (t) + ln |1/t|. For any rotation ρ, since Ω com-
mutes with any rotations, i.e., Ω(ρx) = ρ(Ω(x)), we have, by changes of
variables, Z Z
ρ(m(ξ)) = Φ(ξ · x)ρ(Ω(x))dσ(x) = Φ(ξ · x)Ω(ρx)dσ(x)
n−1 n−1
ZS ZS
= Φ(ξ · ρ−1 y)Ω(y)dσ(y) = Φ(ρξ · y)Ω(y)dσ(y)
S n−1 S n−1
=m(ρξ).
Thus, m commutes with rotations and so m satisfies (5.1). However, the mj
ξ
are each homogeneous of degree 0, so Lemma 5.1 shows that mj (ξ) = c |ξ|j ,
with
5.1. The Riesz transforms - 117 -
Z
c =m1 (e1 ) = Φ(e1 · x)Ω1 (x)dσ(x)
S n−1

πi
Z
= [− sgn (ω) sgn (x1 ) + ln |1/x1 |]cn x1 dσ(x)
S n−1 2
πi
Z
= − sgn (ω) cn |x1 |dσ(x) (the 2nd is 0 since it is odd w.r.t. x1 )
2 S n−1
πi Γ ((n + 1)/2) 2π (n−1)/2
= − sgn (ω) = − sgn (ω)i.
2 π (n+1)/2 Γ ((n + 1)/2)
Here we have used the fact S n−1 |x1 |dσ(x) = 2π (n−1)/2 /Γ ((n + 1)/2). There-
R

fore, we obtain
ξj ˆ
R j f (ξ) = − sgn (ω)i f (ξ), j = 1, ..., n. (5.5)
d
|ξ|
This identity and Plancherel’s theorem also imply the following “unitary”
character of the Riesz transforms
Xn
kRj f k22 = kf k22 .
j=1
By m(ρξ) = ρ(m(ξ)) proved above, we have mj (ρξ) = k ρjk mk (ξ) for
P

any rotation ρ and then mj (ρξ)fˆ(ξ) = k ρjk mk (ξ)fˆ(ξ). Taking the inverse
P

Fourier transform, it follows

F −1 mj (ρξ)fˆ(ξ) =F −1 ρjk mk (ξ)fˆ(ξ)
X

ρjk F −1 mk (ξ)fˆ(ξ) =
X X
= ρjk Rk f.
k k
But by changes of variables, we have
F −1 mj (ρξ)fˆ(ξ)
 n Z
|ω|
= eωix·ξ mj (ρξ)fˆ(ξ)dξ
2π n
 n ZR
|ω|
= eωiρx·η mj (η)fˆ(ρ−1 η)dη
2π Rn

=(F (mj (ξ)fˆ(ρ−1 ξ)))(ρx) = ρF −1 (mj (ξ)fˆ(ρ−1 ξ))(x)

−1

=ρRj ρ−1 f,
since the Fourier transform commutes with rotations. Therefore, it reaches
X
ρRj ρ−1 f = ρjk Rk f, (5.6)
k
which is the statement that under rotations in Rn , the Riesz operators transform
in the same manner as the components of a vector.
We have the following characterization of Riesz transforms.
- 118 - 5. Riesz Transforms and Spherical Harmonics

Proposition 5.2. Let T = (T1 , T2 , ..., Tn ) be an n-tuple of bounded linear

transforms on L2 (Rn ). Suppose
(a) Each Tj commutes with translations of Rn ;
(b) Each Tj commutes with dilations of Rn ;
(c) For every rotation ρ = (ρjk ) of Rn , ρTj ρ−1 f = k ρjk Tk f .
P

Then the Tj is a constant multiple of the Riesz transforms, i.e., there exists a
constant c such that Tj = cRj , j = 1, ..., n.

Proof. All the elements of the proof have already been discussed. We bring
them together.
(i) Since the Tj is bounded linear on L2 (Rn ) and commutes with transla-
tions, by Theorem 1.62 they can be each realized by bounded multipliers mj ,
i.e., F (Tj f ) = mj fˆ.
(ii) Since the Tj commutes with dilations, i.e., Tj δε f = δε Tj f , in view of
Proposition 1.3, we see that F Tj δε f = mj (ξ)F δε f = mj (ξ)ε−n δε−1 fˆ(ξ) =
mj (ξ)ε−n fˆ(ξ/ε) and F δε Tj f = ε−n δε−1 F Tj f = ε−n δε−1 (mj fˆ) =
ε−n mj (ξ/ε)fˆ(ξ/ε), which imply mj (ξ) = mj (ξ/ε) or equivalently mj (εξ) =
mj (ξ), ε > 0; that is, each mj is homogeneous of degree 0.
(iii) Finally, assumption (c) has a consequence by taking the Fourier trans-
form, i.e., the relation (5.1), and so by Lemma 5.1, we can obtain the desired
conclusion. 
One of the important applications of the Riesz transforms is that they can
be used to mediate between various combinations of partial derivatives of a
function.
2
Proposition 5.3. Suppose f ∈ Cc2 (Rn ). Let ∆f = nj=1 ∂∂xf2 . Then we have
P
j
the a priori bound
∂ 2f
6 Ap k∆f kp , 1 < p < ∞. (5.7)
∂xj ∂xk p

Proof. Since F (∂xj f )(ξ) = ωiξj F f (ξ), we have

 2 
∂ f
F (ξ) = − ω 2 ξj ξk F f (ξ)
∂xj ∂xk
  
iξj iξk
= − − sgn (ω) − sgn (ω) (−ω 2 |ξ|2 )F f (ξ)
|ξ| |ξ|
= − F Rj Rk ∆f.
2
Thus, ∂x∂j ∂x
f
k
= −Rj Rk ∆f . By the Lp boundedness of the Riesz transforms,
we have the desired result. 

Proposition 5.4. Suppose f ∈ Cc1 (R2 ). Then we have the a priori bound
5.1. The Riesz transforms - 119 -

∂f ∂f ∂f ∂f
+ 6 Ap +i , 1 < p < ∞.
∂x1 p ∂x2 p ∂x1 ∂x2 p

Proof. The proof is similar to the previous one. Indeed, we have

iξj iξj ξ12 + ξ22
F ∂xj f =ωiξj F f (ξ) = ω |ξ|F f (ξ) = ω F f (ξ)
|ξ| |ξ| |ξ|
iξj (ξ1 − iξ2 )(ξ1 + iξ2 )
=ω F f (ξ)
|ξ| |ξ|
− sgn (ω)iξj − sgn (ω)i(ξ1 − iξ2 )
=− F (∂x1 f + i∂x2 f )
|ξ| |ξ|
= − F Rj (R1 − iR2 )(∂x1 f + i∂x2 f ).
That is, ∂xj f = −Rj (R1 − iR2 )(∂x1 f + i∂x2 f ). Also by the Lp boundedness
of the Riesz transforms, we can obtain the result. 
We shall now tie together the Riesz transforms and the theory of harmonic
functions, more particularly Poisson integrals. Since we are interested here
mainly in the formal aspects we shall restrict ourselves to the L2 case. For Lp
case, one can see the further results in [Ste70, §4.3 and §4.4, p.78].
Theorem 5.5. Let f and f1 , ..., fn all belong to L2 (Rn ), and let their respective
Poisson integrals be u0 (x, y) = Py ∗ f , u1 (x, y) = Py ∗ f1 , ..., un (x, y) =
Py ∗ fn . Then a necessary and sufficient condition of
fj = Rj (f ), j = 1, ..., n, (5.8)
is that the following generalized Cauchy-Riemann equations hold:
 n
 X ∂uj

 = 0,
∂xj

j=0 (5.9)
 ∂uj ∂uk
= , j 6= k, with x0 = y.


∂xk ∂xj


Remark 5.6. At least locally, the system (5.9) is equivalent with the exis-
∂g
tence of a harmonic function g of the n + 1 variables, such that uj = ∂x j
,
j = 0, 1, 2, ..., n.
Proof. Suppose fj = Rj f , then fbj (ξ) = − sgn (ω) |ξ|j fˆ(ξ), and so by (4.15)
iξ

 n Z
|ω| iξj
uj (x, y) = − sgn (ω) fˆ(ξ) eωiξ·x e−|ωξ|y dξ, j = 1, ..., n,
2π Rn |ξ|
and  n Z
|ω|
u0 (x, y) = fˆ(ξ)eωiξ·x e−|ωξ|y dξ.
2π Rn

The equation (5.9) can then be immediately verified by differentiation under

the integral sign, which is justified by the rapid convergence of the integrals in
question.
- 120 - 5. Riesz Transforms and Spherical Harmonics
 n R
|ω|
Conversely, let uj (x, y) = 2π Rn j
fb (ξ)eωiξ·x e−|ωξ|y dξ, j = 0, 1, ..., n
∂u ∂u
with f0 = f . Then the fact that ∂u
∂xj
0
= ∂yj , j = 1, ..., n, and Fourier
= ∂x0j
inversion theorem, show that
ωiξj fb0 (ξ)e−|ωξ|y = −|ωξ|fbj (ξ)e−|ωξ|y ,
iξ
therefore fbj (ξ) = − sgn (ω) j fb0 (ξ), and so
|ξ|
fj = Rj f0 = Rj f, j = 1, ..., n.


5.2 Spherical harmonics and higher Riesz transforms

We return to the consideration of special transforms of the form

Ω(y)
Z
T f (x) = lim f (x − y)dy, (5.10)
ε→0 |y|>ε |y|n

where Ω is homogeneous of degree 0 and its integral over S n−1 vanishes.

We have already considered the example, i.e., the case of Riesz transforms,
y
Ωj (y) = c |y|j , j = 1, ..., n. For n = 1, Ω(y) = c sgn y, and this is the only
possible case, i.e., the Hilbert transform. To study the matter further for n > 1,
we recall the expression
Z
m(ξ) = Λ(y · ξ)Ω(y)dσ(y), |ξ| = 1
S n−1
where m is the multiplier arising from the transform (5.10).
We have already remarked that the mapping Ω → m commutes with ro-
tations. We shall therefore consider the functions on the sphere S n−1 (more
particularly the space L2 (S n−1 )) from the point of view of its decomposition
under the action of rotations. As is well known, this decomposition is in terms
of the spherical harmonics, and it is with a brief review of their properties that
we begin.
We fix our attention, as always, on Rn , and we shall consider polynomials
in Rn which are also harmonic.
Pn α
Definition 5.7. Denote α = (α1 , ..., αn ), |α| = j=1 αj and x =
α1 αn
x1 · · · xn . Let Pk denote the linear space of all homogeneous poly-
nomials of degree k, i.e.,
n X o
Pk := P (x) = aα xα : |α| = k .
Each such polynomial corresponds its dual object, the differential operator
P (∂x ) = aα ∂xα , where ∂xα = ∂xα11 · · · ∂xαnn . On Pk , we define a positive inner
P
0
product hP, Qi = P (∂x )Q̄. Note that two distinct monomials xα and xα in
5.2. Spherical harmonics and higher Riesz transforms - 121 -

Pk are orthogonal w.r.t. it, since there exists at least one i such that αi > αi0 ,
α0
then ∂xαii xi i = 0. hP, P i = |aα |2 α! where α! = (α1 !) · · · (αn !).
P

Definition 5.8. We define Hk to be the linear space of homogeneous

polynomials of degree k which are harmonic: the solid spherical harmon-
ics of degree k. That is,
Hk := {P (x) ∈ Pk : ∆P (x) = 0} .
It will be convenient to restrict these polynomials to S n−1 , and there to de-
fine the standard inner product,
Z
(P, Q) = P (x)Q(x)dσ(x).
S n−1
For a function f on S n−1
, we define the spherical Laplacean∆S by
∆S f (x) = ∆f (x/|x|),
where f (x/|x|) is the degree zero homogeneous extension of the function f to
Rn \ {0}, and ∆ is the Laplacian of the Euclidean space.1
Proposition 5.9. We have the following properties.
(1) The finite dimensional spaces {Hk }∞k=0 are mutually orthogonal.
(2) Every homogeneous polynomial P ∈ Pk can be written in the form
P = P1 + |x|2 P2 , where P1 ∈ Hk and P2 ∈ Pk−2 .
(3) Let Hk denote the linear space of restrictions of Hk to the unit sphere.2
The elements of Hk are the surface spherical harmonics of degree k, i.e.,
Hk = {P (x) ∈ Hk : |x| = 1} .
P∞
Then L (S ) = k=0 Hk . Here the L2 space is taken w.r.t. usual measure,
2 n−1

and the infinite direct sum is taken in the sense of Hilbert space theory. That
is, if f ∈ L2 (S n−1 ), then f has the development
X∞
f (x) = Yk (x), Yk ∈ Hk , (5.11)
k=0
where the convergence is in the L (S n−1 ) norm, and
2
Z XZ
2
|f (x)| dσ(x) = |Yk (x)|2 dσ(x).
S n−1 k S n−1

(4) If Yk (x) ∈ Hk , then ∆S Yk (x) = −k(k + n − 2)Yk (x).

1
This is implied by the well-known formula for the Euclidean Laplacian in spherical polar coordi-
nates:  
∂ ∂f
∆f = r1−n rn−1 + r−2 ∆S f.
∂r ∂r

2
Sometimes, in order to emphasize the distribution between Hk and Hk , the members of Hk are
referred to as the surface spherical harmonics.
- 122 - 5. Riesz Transforms and Spherical Harmonics

(5) Suppose f has the development (5.11). Then f (after correction on a

set of measure zero, if necessary) is indefinitely differentiable on S n−1 (i.e.,
f ∈ C ∞ (S n−1 )) if and only if
Z
|Yk (x)|2 dσ(x) = O(k −N ), as k → ∞, for each fixed N. (5.12)
S n−1

Proof. (1) If P ∈ Pk , i.e., P (x) = aα xα with |α| = k, then

P
n n n
αj −1
X X X X X
α1 αn
xj ∂xj P = xj aα α j x 1 · · · x j · · · xn = αj aα xα = kP.
j=1 j=1 j=1

On S , it follows kP =
n−1
where ∂P
∂ν
denotes differentiation w.r.t. the out-
∂
∂ν
ward normal vector. Thus, for P ∈ Hk , and Q ∈ Hj , then by Green’s theorem
 
∂P ∂ Q̄
Z Z
(k − j) P Q̄dσ(x) = Q̄ −P dσ(x)
S n−1 S n−1 ∂ν ∂ν
Z
= [Q̄∆P − P ∆Q̄]dx = 0,
|x|61
where ∆ is the Laplacean on R . n

(2) Indeed, let |x|2 Pk−2 be the subspace of Pk of all polynomials of the
form |x|2 P2 where P2 ∈ Pk−2 . Then its orthogonal complement w.r.t. h·, ·i
is exactly Hk . In fact, P1 is in this orthogonal complement if and only if
h|x|2 P2 , P1 i = 0 for all P2 . But h|x|2 P2 , P1 i = (P2 (∂x )∆)P1 = hP2 , ∆P1 i,
so ∆P1 = 0 and thus Pk = Hk ⊕ |x|2 Pk−2 , which proves the conclusion. In
addition, we have for P ∈ Pk
|x| P0 (x), k even,
 k
2
P (x) = Pk (x) + |x| Pk−2 (x) + · · · +
|x|k−1 P1 (x), k odd,
where Pj ∈ Hj by noticing that Pj = Hj for j = 0, 1.
(3) In fact, by the further result in (2), if |x| = 1, then we have
P0 (x), k even,

P (x) = Pk (x) + Pk−2 (x) + · · · . +
P1 (x), k odd,
with Pj ∈ Hj . That is, the restriction of any polynomial on the unit sphere is a
finite linear combination of spherical harmonics. Since the restriction of poly-
nomials is dense in L2 (S n−1 ) in the norm (see [SW71, Corollary 2.3, p.141])
by the Weierstrass approximation theorem,3 the conclusion is then established.
(4) In fact, for |x| = 1, we have
∆S Yk (x) =∆(|x|−k Yk (x)) = |x|−k ∆Yk + ∆(|x|−k )Yk + 2∇(|x|−k ) · ∇Yk
=(k 2 + (2 − n)k)|x|−k−2 Yk − 2k 2 |x|−k−2 Yk
= − k(k + n − 2)|x|k−2 Yk = −k(k + n − 2)Yk ,
Pn
since j=1 xj ∂xj Yk = kYk for Yk ∈ Pk .

3
If g is continuous on S n−1 , we can approximate it uniformly by polynomials restricted to S n−1 .
5.2. Spherical harmonics and higher Riesz transforms - 123 -
P∞
(5) To prove this, we writeR (5.11) as f (x) = k=0 ak Yk (x), where the
0

Yk0 are normalized such that S n−1 |Yk0 (x)|2 dσ(x) = 1. Our assertion is then
equivalent with ak = O(k −N/2 ), as k → ∞. If f is of class C 2 , then an
application of Green’s theorem shows that
Z Z
0
∆S f Yk dσ = f ∆S Yk0 dσ.
S n−1 S n−1
Thus, if f ∈ C ∞ , then by (4)
Z Z Z ∞
X
r r 0 r
0
∆S f Yk dσ = f ∆S Yk dσ = [−k(k + n − 2)] aj Yj0 Yk0 dσ
S n−1 S n−1 S n−1 j=0
Z
=[−k(k + n − 2)]r ak |Yk0 |2 dσ = ak [−k(k + n − 2)]r .
S n−1
So ak = O(k −2r ) for every r and therefore (5.12) holds.
To prove the converse, from (5.12), we have for any r ∈ N
∞
X ∞
X
k∆rS f k22 =(∆rS f, ∆rS f ) = ( ∆rS Yj (x), ∆rS Yk (x))
j=0 k=0
∞
X ∞
X
=( [−j(j + n − 2)]r Yj (x), [−k(k + n − 2)]r Yk (x))
j=0 k=0
∞
X
= [−k(k + n − 2)]2r (Yk (x), Yk (x))
k=0
∞
X
= [−k(k + n − 2)]2r O(k −N ) 6 C,
k=0
if we take N large enough. Thus, f ∈ C ∞ (S n−1 ). 

Theorem 5.10 (Hecke’s identity). It holds

 −n/2
− |ω| |x| 2 |ω| |ω| 2
F (Pk (x)e 2 )= (−i sgn (ω))k Pk (ξ)e− 2 |ξ| , ∀Pk ∈ Hk (Rn ).
2π
(5.13)
Proof. That is to prove
 −n/2
|ω|
Z
−ωix·ξ− |ω| |x| 2 |ω| 2
Pk (x)e 2 dx = (−i sgn (ω))k Pk (ξ)e− 2 |ξ| .
Rn 2π
(5.14)
Applying the differential operator Pk (∂ξ ) to both sides of the identity (cf.
Theorem 1.10)
 −n/2
|ω|
Z
−ωix·ξ− |ω| |x|2 |ω| 2
e 2 dx = e− 2 |ξ| ,
Rn 2π
we obtain
- 124 - 5. Riesz Transforms and Spherical Harmonics
−n/2
|ω|
Z 
−ωix·ξ− |ω| |x|2 |ω| 2
(−ωi) k
Pk (x)e dx = 2 Q(ξ)e− 2 |ξ| .
Rn 2π
Since Pk (x) is polynomial, it is obvious analytic continuation Pk (z) to all of
Cn . Thus, by a change of variable
 n/2 Z
|ω| |ω| 2 |ω| 2
Q(ξ) =(−ωi) k
Pk (x)e−ωix·ξ− 2 |x| + 2 |ξ| dx
2π Rn
 n/2 Z
|ω| |ω| 2
=(−ωi)k Pk (x)e− 2 (x+i sgn (ω)ξ) dx
2π Rn
 n/2 Z
|ω| |ω| 2
=(−ωi)k Pk (y − i sgn (ω)ξ)e− 2 |y| dy.
2π Rn
So,
 n/2 Z
|ω| |ω| 2
Q(i sgn (ω)ξ) =(−ωi) k
Pk (y + ξ)e− 2 |y| dy
2π Rn
 n/2 Z ∞
|ω|
Z
n−1 − |ω| 2
=(−ωi) k
r e 2 r
Pk (ξ + ry 0 )dσ(y 0 )dr.
2π 0 S n−1
Since Pk is harmonic, it satisfies the mean value property, i.e., Theorem 4.5,
thus Z Z
Pk (ξ + ry 0 )dσ(y 0 ) = ωn−1 Pk (ξ) = Pk (ξ) dσ(y 0 ).
S n−1 S n−1
Hence
n/2 ∞
|ω|
 Z Z
n−1 − |ω| r2
Q(i sgn (ω)ξ) =(−ωi) k
Pk (ξ) r e 2 dσ(y 0 )dr
2π 0 S n−1
n/2
|ω|
 Z
|ω| 2
=(−ωi)kPk (ξ) e− 2 |x| dx = (−ωi)k Pk (ξ).
2π Rn
Thus, Q(ξ) = (−ωi) Pk (−i sgn (ω)ξ) = (−ωi)k (−i sgn (ω))k Pk (ξ), which
k

proves the theorem. 

The theorem implies the following generalization of itself, whose interest
is that it links the various components of the decomposition of L2 (Rn ), for
different n.
If f is a radial function, we write f = f (r), where r = |x|.
Corollary 5.11. Let Pk (x) ∈ Hk (Rn ). Suppose that f is radial and
Pk (x)f (r) ∈ L2 (Rn ). Then the Fourier transform of Pk (x)f (r) is also of the
form Pk (x)g(r), with g a radial function. Moreover, the induced transform
f → g, Tn,k f = g, depends essentially only on n + 2k. More precisely, we
have Bochner’s relation
 k
|ω|
Tn,k = (−i sgn (ω))k Tn+2k,0 . (5.15)
2π
5.2. Spherical harmonics and higher Riesz transforms - 125 -

Proof. Consider the Hilbert space of radial functions

 Z ∞ 
2 2 2k+n−1
R = f (r) : kf k = |f (r)| r dr < ∞ ,
0
with the indicated norm. Fix now Pk (x), and assume that Pk is normalized,
i.e., Z
|Pk (x)|2 dσ(x) = 1.
S n−1
Our goal is to show that
k
|ω|

(Tn,k f )(r) = (−i sgn (ω))k (Tn+2k,0 f )(r), (5.16)
2π
for each f ∈ R.
|ω| 2
First, if f (r) = e− 2 r , then (5.16) is an immediate consequence of Theo-
rem 5.10, i.e.,
 −n/2
− |ω| r2 |ω| |ω| 2
(Tn,k e 2 )(R) = (−i sgn (ω))k e− 2 R
2π
 k
|ω| |ω| 2
= (−i sgn (ω))k (Tn+2k,0 e− 2 r )(R),
2π
 k
− |ω|
which implies Tn,k f = |ω| 2π
(−i sgn (ω))k
Tn+2k,0 f for f = e 2
r2
.
|ω| 2
Next, we consider e− 2 εr for a fixed ε > 0. By the homogeneity of Pk and
the interplay of dilations with the Fourier transform (cf. Proposition 1.3), i.e.,
F δε = ε−n δε−1 F , and Hecke’s identity, we get
|ω| |ω|
ε|x|2 ε|x|2
F (Pk (x)e− 2 ) = ε−k/2 F (Pk (ε1/2 x)e− 2 )
− |ω| |x|2
=ε−k/2−n/2 δε−1/2 F (Pk (x)e ) 2

 −n/2
|ω| |ω| 2
=ε−k/2−n/2 (−i sgn (ω))k δε−1/2 (Pk (ξ)e− 2 |ξ| )
2π
 −n/2
|ω| |ω| 2
= (−i sgn (ω))k ε−k/2−n/2 Pk (ε−1/2 ξ)e− 2 |ξ| /ε
2π
 −n/2
|ω| |ω| 2
= (−i sgn (ω))k ε−k−n/2 Pk (ξ)e− 2 |ξ| /ε .
2π
|ω|
 −n/2 |ω| 2
This shows that Tn,k e− 2 εr = |ω|
2
2π
(−i sgn (ω))k ε−k−n/2 e− 2 r /ε , and
so
 −k−n/2
− |ω| εr2 |ω| |ω| 2
Tn+2k,0 e 2 = (−i sgn (ω))0 ε−0−(n+2k)/2 e− 2 r /ε
2π
 −k−n/2
|ω| |ω| 2
= ε−k−n/2 e− 2 r /ε .
2π
- 126 - 5. Riesz Transforms and Spherical Harmonics

|ω|
 k |ω|
Thus, Tn,k e− 2 εr = |ω|
2 2
2π
(−i sgn (ω))k Tn+2k,0 e− 2 εr for ε > 0.
To finish the proof, it suffices to see that the linear combination of
− |ω| 2
{e 2 εr }0<ε<∞ is dense in R. Suppose the contrary, then there exists a (al-
|ω| 2
most everywhere) non-zero g ∈ R, such that g is orthogonal to every e− 2 εr
in the sense of R, i.e.,
Z ∞
|ω| 2
e− 2 εr g(r)r2k+n−1 dr = 0, (5.17)
0
Rs 2
for all ε > 0. Let ψ(s) = 0 e−r g(r)rn+2k−1 dr for s > 0. Then, putting
ε = 2(m + 1)/|ω|, where m is a positive integer, and by integration by parts,
we have Z ∞ Z ∞
2 2
0= e−mr ψ 0 (r)dr = 2m e−mr ψ(r)rdr.
0 0
2
By the change of variable z = e−r , this equality is equivalent to
Z 1 p
0= z m−1 ψ( ln 1/z)dz, m = 1, 2, ....
0
Since the polynomials are uniformly dense in the space of continuous func-
tions on the closed interval [0, 1], this can only be the case when ψ( ln 1/z) =
p
2
0 for all z in [0, 1]. Thus, ψ 0 (r) = e−r g(r)rn+2k−1 = 0 for almost every
r ∈ (0, ∞), contradicting the hypothesis that g(r) is not equal to 0 almost
everywhere.
 k
Since the operators Tn,k and |ω| 2π
(−i sgn (ω))k Tn+2k,0 are bounded and
agree on the dense subspace, they must be equal. Thus, we have shown the
desired result. 
We come now to what has been our main goal in our discussion of spherical
harmonics.
Theorem 5.12. Let Pk (x) ∈ Hk , k > 1. Then the multiplier corresponding
Pk (x)
to the transform (5.10) with the kernel |x| k+n is

Pk (ξ) Γ (k/2)
γk , with γk = π n/2 (−i sgn (ω))k .
|ξ|k Γ (k/2 + n/2)
Remark 5.13. 1) If k > 1, then Pk (x) is orthogonal to the constants on the
sphere, and so its mean value over any sphere centered at the origin is
zero.
2) The statement of the theorem can be interpreted as
 
Pk (x) Pk (ξ)
F k+n
= γk . (5.18)
|x| |ξ|k
3) As such it will be derived from the following closely related fact,
 
Pk (x) Pk (ξ)
F k+n−α
= γk,α k+α , (5.19)
|x| |ξ|
5.2. Spherical harmonics and higher Riesz transforms - 127 -
 −α
|ω| Γ (k/2+α/2)
where γk,α = π n/2
2
(−i sgn (ω))k Γ (k/2+n/2−α/2) .
Lemma 5.14. The identity (5.19) holds in the sense that
Pk (x) Pk (ξ)
Z Z
k+n−α
ϕ̂(x)dx = γk,α k+α
ϕ(ξ)dξ, ∀ϕ ∈ S . (5.20)
Rn |x| Rn |ξ|
It is valid for all non-negative integer k and for 0 < α < n.
Remark 5.15. For the complex number α with <α ∈ (0, n), the lemma
and (5.19) are also valid, see [SW71, Theorem 4.1, p.160-163].
Proof. From the proof of Corollary 5.11, we have already known that
 −n/2
− |ω| 2 |ω| |ω| 2
F (Pk (x)e 2 ε|x|
)= (−i sgn (ω))k ε−k−n/2 Pk (ξ)e− 2 |ξ| /ε ,
2π
so we have by the multiplication formula,
Z Z
− |ω| 2 |ω| 2
Pk (x)e 2 ε|x|
ϕ̂(x)dx = F (Pk (x)e− 2 ε|x| )(ξ)ϕ(ξ)dξ
Rn Rn
−n/2
|ω|
 Z
|ω|
k −k−n/2 |ξ|2 /ε
= (−i sgn (ω)) ε Pk (ξ)e− 2 ϕ(ξ)dξ,
2π Rn
for ε > 0.
We now integrate both sides of the above w.r.t. ε, after having multiplied the
equation by a suitable power of ε, (εβ−1 , β = (k + n − α)/2, to be precise).
ThatZ is
∞ Z
|ω|
ε|x|2
εβ−1 Pk (x)e− 2 ϕ̂(x)dxdε
0 Rn
−n/2 ∞
|ω|
 Z Z
|ω|
β−1 −k−n/2 |ξ|2 /ε
= (−i sgn (ω)) k
ε ε Pk (ξ)e− 2 ϕ(ξ)dξdε.
2π 0 Rn
(5.21)
By changing the order of the double integral and a change of variable, we get
Z Z ∞
|ω| 2
l.h.s. of (5.21) = Pk (x)ϕ̂(x) εβ−1 e− 2 ε|x| dεdx
Rn 0
−β Z ∞
|ω| 2
Z 
t=|ω|ε|x|2 /2
==== Pk (x)ϕ̂(x) |x| tβ−1 e−t dtdx
Rn 2 0
 −β
|ω|
Z
= Γ (β) Pk (x)ϕ̂(x)|x|−2β dx.
2 Rn
Similarly,
 −n/2
|ω|
Z
r.h.s. of (5.21) = (−i sgn (ω)) k
Pk (ξ)ϕ(ξ)
2π Rn
Z ∞
|ω| 2
ε−(k/2+α/2+1) e− 2 |ξ| /ε dεdξ
0
- 128 - 5. Riesz Transforms and Spherical Harmonics
−n/2 −(k+α)/2
t= |ω| |ω| |ω| 2
 Z 
2
|ξ|2 /ε k
==== (−i sgn (ω)) Pk (ξ)ϕ(ξ) |ξ|
2π Rn 2
Z ∞
tk/2+α/2−1 e−t dtdξ
0
−n/2 −(k+α)/2
|ω| |ω|
 
k
= (−i sgn (ω)) Γ (k/2 + α/2)
2π 2
Z
Pk (ξ)ϕ(ξ)|ξ|−(k+α) dξ.
Rn
Thus, we get
 −(k+n−α)/2
|ω|
Z
Γ ((k + n − α)/2) Pk (x)ϕ̂(x)|x|−(k+n−α) dx
2 R n
 −n/2  −(k+α)/2
|ω| |ω|
= (−i sgn (ω))k Γ (k/2 + α/2)
2π 2
Z
· Pk (ξ)ϕ(ξ)|ξ|−(k+α) dξ
Rn
which leads to (5.20).
Observe that when 0 < α < n and ϕ ∈ S , then double integrals in the
above converge absolutely. Thus the formal argument just given establishes
the lemma. 
Proof of Theorem 5.12. By the assumption that k > 1, we have that the inte-
gral of Pk over any sphere centered at the origin is zero. Thus for ϕ ∈ S , we
get
Pk (x) Pk (x)
Z Z
k+n−α
ϕ̂(x)dx = k+n−α
[ϕ̂(x) − ϕ̂(0)]dx
Rn |x| |x|61 |x|
Pk (x)
Z
+ k+n−α
ϕ̂(x)dx.
|x|>1 |x|
Pk (x)
Obviously, the second term tends to |x|>1 |x| k+n ϕ̂(x)dx as α → 0 by the
R

dominated convergence theorem. As in the proof of part (c) of Theorem 4.26,

Pk (x)
|x|k+n
[ϕ̂(x) − ϕ̂(0)] is locally integrable, thus we have, by the dominated con-
vergence theorem, the limit of the first term in the r.h.s. of the above
Pk (x) Pk (x)
Z Z
lim [ϕ̂(x) − ϕ̂(0)]dx = [ϕ̂(x) − ϕ̂(0)]dx
α→0 |x|61 |x|k+n−α |x|61 |x|
k+n

Pk (x) Pk (x)
Z Z
= k+n
ϕ̂(x)dx = lim ϕ̂(x)dx.
|x|61 |x| ε→0 ε6|x|61 |x|k+n

Thus, we obtain
Pk (x) Pk (x)
Z Z
lim k+n−α
ϕ̂(x)dx = lim ϕ̂(x)dx. (5.22)
α→0+ Rn |x| ε→0 |x|>ε |x|k+n

Similarly,
5.3. Equivalence between two classes of transforms - 129 -

Pk (ξ) Pk (ξ)
Z Z
lim k+α
ϕ(ξ)dξ = lim ϕ(ξ)dξ.
α→0+ Rn |ξ| ε→0 |ξ|>ε |ξ|k

Thus, by Lemma 5.11, we complete the proof with γk = limα→0 γk,α . 

Pk (y)
For fixed k > 1, the linear space of operators in (5.10), where Ω(y) = |y|k
and Pk ∈ Hk , form a natural generalization of the Riesz transforms; the latter
arise in the special case k = 1. Those for k > 1, we call the higher Riesz
transforms, with k as the degree of the higher Riesz transforms, they can also
be characterized by their invariance properties (see [Ste70, §4.8, p.79]).

5.3 Equivalence between two classes of transforms

We now consider two classes of transforms, defined on L2 (Rn ). The first

class consists of all transforms of the
Z form
Ω(y)
T f = c · f + lim f (x − y)dy, (5.23)
ε→0 |y|>ε |y|n

where c is a constant,
R Ω ∈ C (S ) is a homogeneous function of degree
∞ n−1

0, and the integral S n−1 Ω(x)dσ(x) = 0. The second class is given by those
transforms T for which
F (T f )(ξ) = m(ξ)fˆ(ξ) (5.24)
where the multiplier m ∈ C ∞ (S n−1 ) is homogeneous of degree 0.
Theorem 5.16. The two classes of transforms, defined by (5.23) and (5.24)
respectively, are identical.

Proof. First, support that T is of the form (5.23). Then by Theorem 4.24, T is
of the form (5.24) with m homogeneous of degree 0 and
Z  
πi
m(ξ) = c + − sgn (ω) sgn (ξ · x) + ln(1/|ξ · x|) Ω(x)dσ(x), |ξ| = 1.
S n−1 2
(5.25)
Now, we need to show m ∈ C ∞ (S n−1 ). Write the spherical harmonic de-
velopments
X∞ ∞
X XN N
X
Ω(x) = Yk (x), m(x) = Ỹk (x), ΩN (x) = Yk (x), mN (x) = Ỹk (x),
k=1 k=0 k=1 k=0
(5.26)
where Yk , Ỹk ∈ Hk in view
R of part (3) in Proposition 5.9. k starts from 1 in the
development of Ω, since S n−1 Ω(x)dx = 0 implies that Ω(x) is orthogonal to
constants, and H0 contains only constants.
Then, by Theorem 5.12, if Ω = ΩN , then m(x) = mN (x), with
Ỹk (x) = γk Yk (x), k > 1.
- 130 - 5. Riesz Transforms and Spherical Harmonics
h i
But mM (x) − mN (x) = S n−1 − πi 1
R
2
sgn (ω) sgn (y · x) + ln |y·x|
[ΩM (y) −
ΩN (y)]dσ(y). Moreover by Hölder’s inequality,
sup |mM (x) − mN (x)|
x∈S n−1
!1/2
2
πi
Z
6 sup − sgn (ω) sgn (y · x) + ln(1/|y · x|) dσ(y)
x S n−1 2
Z 1/2
2
× |ΩM (y) − ΩN (y)| dσ(y) → 0, (5.27)
S n−1
as M , N → ∞, since4 for n = 1, S 0 = {−1, 1},
2
πi π2
Z
− sgn (ω) sgn (y · x) + ln(1/|y · x|) dσ(y) = ,
S0 2 2
and for n > 2, we can pick a orthogonal matrix A satisfying Ae1 = x and
det A = 1 for |x| = 1, and then by a change of variable,
2
πi
Z
sup − sgn (ω) sgn (y · x) + ln(1/|y · x|) dσ(y)
x n−1 2
ZS  2 
π 2
= sup + (ln(1/|y · x|)) dσ(y)
x S n−1 4
π2
Z
= ωn−1 + sup (ln |y · Ae1 |)2 dσ(y)
4 x n−1
2 ZS
π
= ωn−1 + sup (ln |A−1 y · e1 |)2 dσ(y)
4 x S n−1
2
z=A y π
Z
−1
==== ωn−1 + (ln |z1 |)2 dσ(z) < ∞.
4 S n−1
Here, we have used the boundedness of the integral in the r.h.s., i.e., (with the
notation z̄ = (z2 , ..., zn ), cf. [Gra04, p.A-20,p.267])
Z 1
dz1
Z Z
2 2
(ln |z1 |) dσ(z) = (ln |z1 |) √ dσ(z̄) p
S n−1 −1 1−z12 S n−2 1 − z12
√ Z 1 Z
y=z̄/ 1−z12
2
====== (ln |z1 |) (1 − z12 )(n−3)/2 dσ(y)dz1
−1 S n−2
Z 1
=ωn−2 (ln |z1 |)2 (1 − z12 )(n−3)/2 dz1
Z−1π
z1 =cos θ
====ω n−2 (ln | cos θ|)2 (sin θ)n−2 dθ = ωn−2 I1 .
0
If n > 3, then, by integration by parts,
Z π Z π Z π
2
I1 6 (ln | cos θ|) sin θdθ = −2 ln | cos θ| sin θdθ = 2 sin θdθ = 4.
0 0 0

4
There the argument is similar with some part of the proof of Theorem 4.24.
5.3. Equivalence between two classes of transforms - 131 -
R π/2
If n = 2, then, by the formula 0 (ln(cos θ))2 dθ = π2 [(ln 2)2 + π 2 /12], cf.
[GR, 4.225.8, p.531], we get
Z π Z π/2
2
I1 = (ln | cos θ|) dθ = 2 (ln(cos θ))2 dθ = π[(ln 2)2 + π 2 /12].
0 0
Thus, (5.27) shows that
∞
X
m(x) = c + γk Yk (x).
k=1
Since Ω ∈ C ∞ , we have, in view of part (5) of Proposition 5.9, that
Z
|Yk (x)|2 dσ(x) = O(k −N )
S n−1
as k → ∞ for every fixed N . However, by the explicit form of γk , we see that
γk ∼ k −n/2 , so m(x) is also indefinitely differentiable on the unit sphere, i.e.,
m ∈ C ∞ (S n−1 ).
Conversely, suppose m(x) ∈ C ∞ (S n−1 ) and let its spherical harmonic de-
velopment be as in (5.26). Set c = Ỹ0 , and Yk (x) = γ1k Ỹk (x). Then Ω(x),
given by (5.26), has mean value zero in the sphere, and is again indefinitely
differentiable there. But as we have just seen the multiplier corresponding to
this transform is m; so the theorem is proved. 
As an application of this theorem and a final illustration of the singular
integral transforms we shall give the generalization of the estimates for partial
derivatives given in 5.1.
Let P (x) ∈ Pk (Rn ). We shall say that P is elliptic if P (x) vanishes only
at the origin. For any polynomial P , we consider also its corresponding differ-
ential polynomial. Thus, if P (x) = aα xα , we write P ( ∂x ∂
) as
∂ α
P P
) = aα ( ∂x
in the previous definition.
Corollary 5.17. Suppose P is a homogeneous elliptic polynomial of degree k.
∂ α
Let ( ∂x ) be any differential monomial of degree k. Assume f ∈ Cck , then we
have the a priori estimate
 α  
∂ ∂
f 6 Ap P f , 1 < p < ∞. (5.28)
∂x p ∂x p

∂ α
Proof. From the Fourier transform of ∂x f and P ∂x∂
f,



     
∂ ∂
Z
F P f (ξ) = e−ωix·ξ P f (x)dx = (ωi)k P (ξ)fˆ(ξ),
∂x Rn ∂x
and  α 
∂
F f (ξ) = (ωi)k ξ α fˆ(ξ),
∂x
we have the following relation
- 132 - 5. Riesz Transforms and Spherical Harmonics
 α     
∂ α ∂
P (ξ)F f (ξ) = ξ F P f (ξ).
∂x ∂x
α
Since P (ξ) is non-vanishing except at the origin, Pξ(ξ) is homogenous of degree
0 and is indefinitely differentiable on the unit sphere. Thus
 α    
∂ ∂
f =T P f ,
∂x ∂x
where T is one of the transforms of the type given by (5.24). By Theorem 5.16,
T is also given by (5.23) and hence by the result of Theorem 4.24, we get the
estimate (5.28). 
L ECTURE N OTES ON c 2016 by Chengchun Hao
Introduction to Harmonic Analysis Email: hcc@amss.ac.cn

Chapter 6
The Littlewood-Paley g-function and Multipliers

In harmonic analysis, Littlewood-Paley theory is a term used to describe a

theoretical framework used to extend certain results about L2 functions to Lp
functions for 1 < p < ∞. It is typically used as a substitute for orthogonality
arguments which only apply to Lp functions when p = 2. One implementa-
tion involves studying a function by decomposing it in terms of functions with
localized frequencies, and using the Littlewood-Paley g-function to compare
it with its Poisson integral. The 1-variable case was originated by J. E. Little-
wood and R. Paley (1931, 1937, 1938) and developed further by Zygmund and
Marcinkiewicz in the 1930s using complex function theory (Zygmund 2002
[1935], chapters XIV, XV). E. M. Stein later extended the theory to higher
dimensions using real variable techniques.

6.1 The Littlewood-Paley g-function

The g-function is a nonlinear operator which allows one to give a useful

characterization of the Lp norm of a function on Rn in terms of the behavior of
its Poisson integral. This characterization will be used not only in this chapter,
but also in the succeeding chapter dealing with function spaces.
Let f ∈ Lp (Rn ) and write u(x, y) for its Poisson integral
 n Z
|ω|
Z
ωiξ·x −|ωξ|y ˆ
u(x, y) = e e f (ξ)dξ = Py (t)f (x − t)dt
2π Rn Rn
as defined in (4.15) and (4.17). Let ∆ denote the Laplace operator in Rn+1 + ,
∂2
Pn ∂ 2
that is ∆ = ∂y2 + j=1 ∂x2 ; ∇ is the corresponding gradient, |∇u(x, y)|2 =
j

| ∂y | + |∇x u(x, y)| , where |∇x u(x, y)|2 = nj=1 | ∂x

∂u 2 2 ∂u 2
|.
P
j

Definition 6.1. With the above notations, we define the Littlewood-Paley

g-function g(f )(x), by

133
- 134 - 6. The Littlewood-Paley g-function and Multipliers
Z ∞ 1/2
2
g(f )(x) = |∇u(x, y)| ydy . (6.1)
0
We can also define two partial g-functions, one dealing with the y differ-
entiation and the other with the x differentiations,
!1/2
Z ∞ 2 Z ∞ 1/2
∂u 2
g1 (f )(x) = (x, y) ydy , gx (f )(x) = |∇x u(x, y)| ydy .
0 ∂y 0
(6.2)
2
Obviously, g = g12 + gx2 .
The basic result for g is the following.
Theorem 6.2. Suppose f ∈ Lp (Rn ), 1 < p < ∞. Then g(f )(x) ∈ Lp (Rn ),
and
A0p kf kp 6 kg(f )kp 6 Ap kf kp . (6.3)

Proof. Step 1: We first consider the simple case p = 2. For f ∈ L2 (Rn ), we

have Z Z ∞ Z ∞ Z
2 2
kg(f )k2 = |∇u(x, y)| ydydx = y |∇u(x, y)|2 dxdy.
Rn 0 0 Rn
In view of the identity
n Z
|ω|

u(x, y) = eωiξ·x e−|ωξ|y fˆ(ξ)dξ,
2π Rn
we have n Z
|ω|

∂u
= −|ωξ|fˆ(ξ)eωiξ·x e−|ωξ|y dξ,
∂y 2π Rn
and  n Z
∂u |ω|
= ωiξj fˆ(ξ)eωiξ·x e−|ωξ|y dξ.
∂xj 2π Rn
Thus, by Plancherel’s formula,
Z " 2 n 2
#
∂u ∂u
Z X
|∇u(x, y)|2 dx = + dx
Rn Rn ∂y j=1
∂x j

2 n 2
∂u X ∂u
= +
∂y L2x j=1
∂xj L2x
" n
#
(−|ωξ|fˆ(ξ)e (ωiξj fˆ(ξ)e
X
−1 −|ωξ|y −1 −|ωξ|y
= kF )k22 + kF )k22
j=1
n " n
#
|ω|

k − |ωξ|fˆ(ξ)e−|ωξ|y k22 + kωiξj fˆ(ξ)e−|ωξ|y k22
X
=
2π j=1
6.1. The Littlewood-Paley g-function - 135 -
n
|ω|

=2 ω 2 k|ξ|fˆ(ξ)e−|ωξ|y k22
2π
 n
|ω|
Z
= 2 ω 2 |ξ|2 |fˆ(ξ)|2 e−2|ωξ|y dξ,
Rn 2π
and so Z ∞ Z  n
|ω|
2
kg(f )k2 = y 2 ω 2 |ξ|2 |fˆ(ξ)|2 e−2|ωξ|y dξdy
0 Rn 2π
 n Z ∞
|ω|
Z
2 ˆ
= 2 2
ω |ξ| |f (ξ)| 2
ye−2|ωξ|y dydξ
n 2π 0
ZR  n  n
|ω| 2 ˆ 1 1 |ω|
= 2 2
ω |ξ| |f (ξ)| 2
2 |ξ|2
dξ = kfˆk22
R n 2π 4ω 2 2π
1
= kf ||22 .
2
Hence,
kg(f )k2 = 2−1/2 kf k2 . (6.4)
We have also obtained kg1 (f )k2 = kgx (f )k2 = 1
2
kf k2 .
Step 2: We consider the case p 6= 2 and prove kg(f )kp 6 Ap kf kp . We
define the Hilbert spaces H1 and H2 which are to be consider now. H1 is the
one-dimensional Hilbert space of complex numbers. To define H2 , we define
first H20 as the L2 space on (0, ∞) with measure ydy, i.e.,
 Z ∞ 
0 2 2
H2 = f : |f | = |f (y)| ydy < ∞ .
0
Let H2 be the direct sum of n + 1 copies of H20 ; so the elements of H2 can
be represented as (n + 1) component vectors whose entries belong to H20 .
Since H1 is the same as the complex numbers, then L(H1 , H2 ) is of course
identifiable with H2 . Now let ε > 0, and keep it temporarily fixed.
Define  
∂Py+ε (x) ∂Py+ε (x) ∂Py+ε (x)
Kε (x) = , ,··· , .
∂y ∂x1 ∂xn
Notice that for each fixed x, Kε (x) ∈ H2 . This is the same as saying that
Z ∞ 2 Z ∞ 2
∂Py+ε (x) ∂Py+ε (x)
ydy < ∞ and ydy < ∞, for j = 1, ..., n.
0 ∂y 0 ∂xj
In fact, since Py (x) = (|x|2 +yc2n)y(n+1)/2 , we have that both ∂P∂y
y
and ∂Py
∂xj
are
bounded by (|x|2 +y2 )(n+1)/2 . So the norm in H2 of Kε (x),
A
Z ∞
2 2 ydy
|Kε (x)| 6A (n + 1)
(|x| + (y + ε)2 )n+1
2
Z0 ∞
dy
6A2 (n + 1) 6 Cε ,
0 (y + ε)2n+1
and in another way
- 136 - 6. The Littlewood-Paley g-function and Multipliers

∞
ydy A2 (n + 1)
Z
2 2
|Kε (x)| 6 A (n+1) = (|x|2 +ε2 )−n 6 C|x|−2n .
ε (|x|2 + y 2 )n+1 2n
Thus,
|Kε (x)| ∈ L1loc (Rn ). (6.5)
Similarly,
2 Z ∞ Z ∞
∂Kε (x) ydy ydy
6C 2 2 n+2
6C 2 2 n+2
6 C|x|−2n−2 .
∂xj 0 (|x| + y ) ε (|x| + y )
Therefore, Kε satisfies the gradient condition, i.e.,
∂Kε (x)
6 C|x|−(n+1) , (6.6)
∂xj
with C independent of ε.
Now we consider the operator Tε defined by
Z
Tε f (x) = Kε (t)f (x − t)dt.
Rn
The function f is complex-valued (take their value in H1 ), but Tε f (x) takes
its value in H2 . Observe that
Z ∞  21 Z ∞  12
2 2
|Tε f (x)| = |∇u(x, y + ε)| ydy 6 |∇u(x, y)| ydy 6 g(f )(x).
0 ε
(6.7)
Hence, kTε f (x)k2 6 2−1/2 kf k2 , if f ∈ L2 (Rn ), by (6.4). Therefore,
|K̂ε (x)| 6 2−1/2 . (6.8)
Because of (6.5), (6.6) and (6.8), by Theorem 4.27 (cf. Theorem 4.18), we get
kTε f kp 6 Ap kf kp , 1 < p < ∞ with Ap independent of ε. By (6.7), for each
x, |Tε f (x)| increases to g(f )(x), as ε → 0, so we obtain finally
kg(f )kp 6 Ap kf kp , 1 < p < ∞. (6.9)
Step 3: To derive the converse inequalities,
A0p kf kp 6 kg(f )kp , 1 < p < ∞. (6.10)
In the first step, we have shown that kg1 (f )k2 = 21 kf k2 for f ∈ L2 (Rn ).
Let u1 , u2 are the Poisson integrals of Rf1 , fR2 ∈ L2 , respectively. ThenR we have
∞
kg1 (f1 + f2 )k22 = 41 kf1 + f2 k22 , i.e., Rn 0 | ∂(u1∂y+u2 ) |2 ydydx = 41 Rn |f1 +
f2 |2 dx. It leads to the identity
Z Z ∞
∂u1 ∂u2
Z
4 (x, y) (x, y)ydydx = f1 (x)f2 (x)dx.
Rn 0 ∂y ∂y Rn
This identity, in turn, leads to the inequality, by Hölder’s inequality and the
definition of g1 ,
1
Z Z
f1 (x)f2 (x)dx 6 g1 (f1 )(x)g1 (f2 )(x)dx.
4 Rn Rn
6.1. The Littlewood-Paley g-function - 137 -
0
Suppose now in addition that f1 ∈ Lp (Rn ) and f2 ∈ Lp (Rn ) with kf2 kp0 6
1 and 1/p + 1/p0 = 1. Then by Hölder inequality and the result (6.9).
Z
f1 (x)f2 (x)dx 6 4kg1 (f1 )kp kg1 (f2 )kp0 6 4Ap0 kg1 (f1 )kp . (6.11)
Rn
Now we take the supremum in (6.11) as f2 ranges over all function in
0
L ∩ Lp , with kf2 kp0 6 1. Then, we obtain the desired result (6.10), with
2

A0p = 1/4Ap0 , but where f is restricted to be in L2 ∩ Lp . The passage

to the general case is provided by an easy limiting argument. Let fm be a
sequence of functions in L2 ∩ Lp , which converges in Lp norm to f . No-
tice that |g(fm )(x) − g(fn )(x)| = k∇um kL2 (0,∞;ydy) − k∇un kL2 (0,∞;ydy) 6
k∇um − ∇un kL2 (0,∞;ydy) = g(fm − fn )(x) by the triangle inequality. Thus,
{g(fm )} is a Cauchy sequence in Lp and so converges to g(f ) in Lp , and we
obtain the inequality (6.10) for f as a result of the corresponding inequalities
for fm . 
We have incidentally also proved the following, which we state as a corol-
lary.
Corollary 6.3. Suppose f ∈ L2 (Rn ), and g1 (f ) ∈ Lp (Rn ), 1 < p < ∞. Then
f ∈ Lp (Rn ), and A0p kf kp 6 kg1 (f )kp .
Remark 6.4. There are some very simple variants of the above that
should be pointed out:
(i) The results hold also with gx (f ) instead of g(f ). The direct inequal-
ity kgx (f )kp 6 Ap kf kp is of course a consequence of the one for g. The
converse inequality is then proved in the same way as that for g1 .
(ii) For any integer k > 1, define
Z ∞ k 2
!1/2
∂ u
gk (f )(x) = k
(x, y) y 2k−1 dy .
0 ∂y
Then the Lp inequalities hold for gk as well. both (i) and (ii) are stated
more systematically in [Ste70, Chapter IV, §7.2, p.112-113].
(iii) For later purpose, it will be useful to note that for each x,
gk (f )(x) > Ak g1 (f )(x) where the bound Ak depends only on k.
It is easily verified from the Poisson integral formula that if f ∈
L (Rn ), 1 6 p 6 ∞, then
p

∂ k u(x, y)
→ 0 for each x, as y → ∞.
∂y k
Thus,
Z ∞ k+1
∂ k u(x, y) ∂ u(x, s) k ds
k
=− s k.
∂y y ∂sk+1 s
By Schwarz’s inequality, therefore,
- 138 - 6. The Littlewood-Paley g-function and Multipliers

2 2
! Z
∞ ∞
∂ k u(x, y) ∂ k+1 u(x, s) 2k
Z 
−2k
6 s ds s ds .
∂y k y ∂sk+1 y

Hence, by Hardy’s inequality (2.17) (on p.51, with q = r = 1 there), we

have
Z ∞ k 2
2 ∂ u
(gk (f )(x)) = (x, y) y 2k−1 dy
0 ∂y k
Z ∞ Z ∞ k+1 2
! Z
∞ 
∂ u 2k −2k
6 k+1
(x, s) s ds s ds y 2k−1 dy
0 y ∂s y
Z ∞ Z ∞ k+1 2
!
1 ∂ u
= k+1
(x, s) s2k ds dy
2k − 1 0 y ∂s
Z ∞ k+1 2
1 ∂ u
6 k+1
(x, s) s2k+1 ds
2k − 1 0 ∂s
Z ∞ k+1 2
1 ∂ u
= (x, s) s2(k+1)−1 ds
2k − 1 0 ∂sk+1
1
= (gk+1 (f )(x))2 .
2k − 1
Thus, the assertion is proved by the induction on k.
The proof that was given for the Lp inequalities for the g-function did not, in
any essential way, depend on the theory of harmonic functions, despite the fact
that this function was defined in terms of the Poisson integral. In effect, all that
was really used is the fact that the Poisson kernels are suitable approximations
to the identity.
There is, however, another approach, which can be carried out without re-
course to the theory of singular integrals, but which leans heavily on character-
istic properties of harmonic functions. We present it here (more precisely, we
present that part which deals with 1 < p 6 2, for the inequality (6.9)), because
its ideas can be adapted to other situations where the methods of Chapter 4 are
not applicable. Everything will be based on the following three observations.
Lemma 6.5. Suppose u is harmonic and strictly positive. Then
∆up = p(p − 1)up−2 |∇u|2 . (6.12)

Proof. The proof is straightforward. Indeed,

∂xj up = pup−1 ∂xj u, ∂x2j up = p(p − 1)up−2 (∂xj u)2 + pup−1 ∂x2j u,
which implies by summation
∆up = p(p − 1)up−2 |∇u|2 + pup−1 ∆u = p(p − 1)up−2 |∇u|2 ,
since ∆u = 0. 
6.1. The Littlewood-Paley g-function - 139 -

n+1
Lemma 6.6. Suppose F (x, y) ∈ C(R+ ) ∩ C 2 (Rn+1
+ ), and suitably small at
infinity. Then
Z Z
y∆F (x, y)dxdy = F (x, 0)dx. (6.13)
Rn+1
+ Rn

Proof. We use Green’s theorem

 
∂v ∂u
Z Z
(u∆v − v∆u)dxdy = u −v dσ
D ∂D ∂N ∂N
where D = Br ∩ Rn+1 + , with Br the ball of radius r in R
n+1
centered at the
origin, N is the outward normal vector. We take v = F , and u = y. Then, we
will obtain our result (6.13) if
Z Z
y∆F (x, y)dxdy → y∆F (x, y)dxdy,
D Rn+1
+

and  
∂F ∂y
Z
y −F dσ → 0,
∂D0 ∂N ∂N
as r → ∞. Here ∂D0 is the spherical part of the boundary of D. This will
certainly be the case, if for example ∆F > 0, and |F | 6 O((|x| + y)−n−ε ) and
|∇F | = O((|x| + y)−n−1−ε ), as |x| + y → ∞, for some ε > 0. 

Lemma 6.7. If u(x, y) is the Poisson integral of f , then

sup |u(x, y)| 6 M f (x). (6.14)
y>0

Proof. This is the same as the part (a) of Theorem 4.9. It can be proved with a
similar argument as in the proof of part (a) for Theorem 4.10. 
Now we use these lemmas to give another proof for the inequality
kg(f )kp 6 Ap kf kp , 1 < p 6 2.
Another proof of kg(f )kp 6 Ap kf kp , 1 < p 6 2. Suppose first 0 6 f ∈
D(Rn ) (and at least f 6= 0Ron a nonzero measurable set). Then the Poisson
integral u of f , u(x, y) = Rn Py (t)f (x − t)dt > 0, since Py > 0 for any
x ∈ Rn and y > 0; and the majorizations u(x, y) = O((|x| + y)−n ) and
|∇u| = O((|x| + y)−n−1 ), as |x| + y → ∞ are valid. We have, by Lemma 6.5,
Lemma 6.7 and the hypothesis 1 < p 6 2,
Z ∞ Z ∞
2 2 1
(g(f )(x)) = y|∇u(x, y)| dy = yu2−p ∆up dy
0 p(p − 1) 0
2−p Z ∞
[M f (x)]
6 y∆up dy.
p(p − 1) 0
We can write this as
g(f )(x) 6 Cp (M f (x))(2−p)/2 (I(x))1/2 , (6.15)
R∞
where I(x) = 0
y∆u dy. However, by Lemma 6.6,
p
- 140 - 6. The Littlewood-Paley g-function and Multipliers
Z Z Z
p
I(x)dx = y∆u dydx = up (x, 0)dx = kf kpp . (6.16)
Rn Rn+1
+ Rn
This immediately gives the desired result for p = 2.
Next, suppose 1 < p < 2. By (6.15), Hölder’s inequality, Theorem 3.9 and
(6.16), we have, for 0 6 f ∈ D(Rn ),
Z Z
p p
(g(f )(x)) dx 6 Cp (M f (x))p(2−p)/2 (I(x))p/2 dx
Rn Rn
Z 1/r0 Z 1/r
0
6Cpp p
(M f (x)) dx I(x)dx 6 Cp0 kf kp/r p/r
p kf kp = Cp0 kf kpp ,
Rn Rn
where r = 2/p ∈ (1, 2) and 1/r + 1/r0 = 1, then r0 = 2/(2 − p).
Thus, kg(f )kp 6 Ap kf kp , 1 < p 6 2, whenever 0 6 f ∈ D(Rn ).
For general f ∈ Lp (Rn ) (which we assume for simplicity to be real-valued),
write f = f + − f − as its decomposition into positive and negative part; then
we need only approximate in norm f + and f − , each by a sequences of positive
functions in D(Rn ). We omit the routine details that are needed to complete
the proof. 
Unfortunately, the elegant argument just given is not valid for p > 2. There
is, however, a more intricate variant of the same idea which does work for the
case p > 2, but we do not intend to reproduce it here.
We shall, however, use the ideas above to obtain a significant generalization
of the inequality for the g-functions.
Definition 6.8. Define the positive function
Z ∞Z  λn
∗ 2 y
(gλ (f )(x)) = |∇u(x − t, y)|2 y 1−n dtdy. (6.17)
0 Rn |t| + y
Before going any further, we shall make a few comments that will help to
clarify the meaning of the complicated expression (6.17).
First, gλ∗ (f )(x) will turn out to be a pointwise majorant of g(f )(x). To un-
derstand this situation better we have to introduce still another quantity, which
is roughly midway between g and gλ∗ . It is defined as follows.
Definition 6.9. Let Γ be a fixed proper cone in Rn+1 + with vertex at the
origin and which contains (0, 1) in its interior. The exact form of Γ will
not really matter, but for the sake of definiteness let us choose for Γ the
up circular cone:
Γ = (t, y) ∈ Rn+1

+ : |t| < y, y > 0 .
For any x ∈ Rn , let Γ (x) be the cone Γ translated such that its vertex
is at x. Now define the positive Luzin’s S-function S(f )(x) by
Z Z
2 2 1−n
[S(f )(x)] = |∇u(t, y)| y dydt = |∇u(x − t, y)|2 y 1−n dydt.
Γ (x) Γ
(6.18)
 y
6.1. The Littlewood-Paley g-function - 141 -
Γ Γ(x)
We assert, as we shall momentarily prove,
that (0, 1)
π
4
Proposition 6.10. O x t
g(f )(x) 6 CS(f )(x) 6 Cλ gλ∗ (f )(x). (6.19) Fig. 6.1 Γ and Γ (x) for n = 1
Figure 1: Γ and Γ(x) for n = 1
What interpretation can we put on the inequalities relating these three quan-
tities? A hint is afforded by considering three corresponding approaches to the
boundary for harmonic functions.
(a) With u(x, y) the Poisson integral of f (x), the simplest approach to the
boundary point x ∈ Rn is obtained by letting y → 0, (with x fixed). This
is the perpendicular approach, and for it the appropriate limit exists almost
everywhere, as we already know.
(b) Wider scope is obtained by allowing the variable point (t, y) to approach
(x, 0) through any cone Γ (x), (where vertex is x). This is the non-tangential
approach which will be so important for us later. As the reader may have al-
ready realized, the relation of the S-function to the g-function is in some sense
analogous to the relation between the non-tangential and the perpendicular ap-
proaches; we should add that the S-function is of decisive significance in its
own right, but we shall not pursue that matter now.
(c) Finally, the widest scope is obtained by allowing the variable point (t, y)
to approach (x, 0) in an arbitrary manner, i.e., the unrestricted approach. The
function gλ∗ has the analogous role: it takes into account the unrestricted ap-
proach for Poisson integrals.
Notice that gλ∗ (x) depends on λ. For each x, the smaller λ the greater gλ∗ (x),
and this behavior is such that that Lp boundedness of gλ∗ depends critically
on the correct relation between p and λ. This last point is probably the main
interest in gλ∗ , and is what makes its study more difficult than g or S.
After these various heuristic and imprecise indications, let us return to firm
ground. The only thing for us to prove here is the assertion (6.19).
Proof of Proposition 6.10. The inequality S(f )(x) 6 Cλ gλ∗ (f )(x) is obvious,
since the integral (6.17) majorizes that part of the integral taken only over Γ ,
and
 λn
y 1
> λn
|t| + y 2
since |t| < y there. The non-trivial part of the assertion is:
g(f )(x) 6 CS(f )(x).
y
It suffices to prove this inequality for x = 0. Let By Γ
us denote by By the ball in Rn+1+ centered at (0, y) (0, y)

and tangent to the boundary of the cone Γ ; the ra-

dius of By is then proportional to y. Now the partial O t

Γ and1:BΓyand By
Fig. 6.2 Figure
- 142 - 6. The Littlewood-Paley g-function and Multipliers

derivatives ∂u
∂y
and ∂x
∂u
k
are, like u, harmonic func-
tions. Thus, by the mean value theorem of harmonic
functions (i.e., Theorem 4.5 by noticing (0, y) is the
center of By ),
∂u(0, y) 1 ∂u(x, s)
Z
= dxds
∂y m(By ) By ∂s
where m(By ) is the n + 1 dimensional measure of By , i.e., m(By ) = cy n+1
for an appropriate constant c. By Schwarz’s inequality
2 2
∂u(0, y) 1 ∂u(x, s)
Z Z
6 dxds dxds
∂y (m(By ))2 By ∂s By
2
1 ∂u(x, s)
Z
= dxds.
m(By ) By ∂s
If we integrate this inequality, we obtain
!
Z ∞ 2 Z ∞ 2
∂u(0, y) ∂u(x, s)
Z
y dy 6 c−1 y −n dxds dy.
0 ∂y 0 By ∂s
However, (x, s) ∈ By clearly implies that c1 s 6 y 6 c2 s, for two positive
constants c1 and c2 . Thus, apart from a multiplicative factor by changing the
order of the double integrals, the last integral is majorized by
Z Z c2 s  2 2
∂u(x, s) ∂u(x, s) 1−n
Z
−n 0
y dy dxds 6 c s dxds.
Γ c1 s ∂s Γ ∂s
This is another way of saying that,
Z ∞ 2 2
∂u(0, y) ∂u(x, y)
Z
00
y dy 6 c y 1−n dxdy.
0 ∂y Γ ∂y
The same is true for the derivatives ∂xj , j = 1, ..., n, and adding the corre-
∂u

sponding estimates proves our assertion. 

We are now in a position to state the Lp estimates concerning gλ∗ .
Theorem 6.11. Let λ > 1 be a parameter. Suppose f ∈ Lp (Rn ). Then
(a) For every x ∈ Rn , g(f )(x) 6 Cλ gλ∗ (f )(x).
(b) If 1 < p < ∞, and p > 2/λ, then
kgλ∗ (f )kp 6 Ap,λ kf kp . (6.20)

Proof. The part (a) has already been proved in Proposition 6.10. Now, we prove
(b).
For the case p > 2, only the assumption λ > 1 is relevant since 2/λ < 2 6
p.
Let ψ denote a positive function on Rn , we claim that
Z Z
∗ 2
(gλ (f )(x)) ψ(x)dx 6 Aλ (g(f )(x))2 (M ψ)(x)dx. (6.21)
Rn Rn
6.1. The Littlewood-Paley g-function - 143 -

The l.h.s. of (6.21) equals

Z ∞Z Z 
2 ψ(x) λn −n
y|∇u(t, y)| y y dx dtdy,
0 t∈Rn x∈Rn (|t − x| + y)λn
so to prove (6.21), we must show that
ψ(x)
Z
sup λn
y λn y −n dx 6 Aλ M ψ(t). (6.22)
y>0 x∈R n (|t − x| + y)
However, we know by Theorem 4.10, that
sup(ψ ∗ ϕε )(t) 6 AM ψ(t)
ε>0
for appropriate ϕ, with ϕε (x) = ε−n ϕ(x/ε). Here, we have in fact ϕ(x) =
(1 + |x|)−λn , ε = y, and so with λ > 1 the hypotheses of that theorem are
satisfied. This proves (6.22) and thus also (6.21).
The case p = 2 follows immediately from (6.21) by inserting in this in-
equality the function ψ = 1 (or by the definitions of gλ∗ (f ) and g(f ) directly),
and using the L2 result for g.
Suppose now p > 2; let us set 1/q + 2/p = 1, and take the supremum of the
l.h.s. of (6.21) over all ψ > 0, such that ψ ∈ Lq (Rn ) and kψkq 6 1. Then, it
gives kgλ∗ (f )k2p ; Hölder’s inequality yields an estimate for the right side:
Aλ kg(f )k2p kM ψkq .
However, by the inequalities for the g-function, kg(f )kp 6 A0p kf kp ; and by
the theorem of the maximal function kM ψkq 6 Aq kψkq 6 A00q , since q > 1, if
p < ∞. If we substitute these in the above, we get the result:
kgλ∗ (f )kp 6 Ap,λ kf kp , 2 6 p < ∞, λ > 1.
The inequalities for p < 2 will be proved by an adaptation of the reasoning
used for g. Lemmas 6.5 and 6.6 will be equally applicable in the present situa-
tion, but we need more general version of Lemma 6.7, in order to majorize the
unrestricted approach to the boundary of a Poisson integral.
It is at this stage where results which depend critically on the Lp class first
make their appearance. Matters will depend on a variant of the maximal func-
tion which we define as follows. Let µ > 1, and write Mµ f (x) for
 1/µ
1
Z
µ
Mµ f (x) = sup |f (y)| dy . (6.23)
r>0 m(B(x, r)) B(x,r)

Then M1 f (x) = M f (x), and Mµ f (x) = ((M |f |µ )(x))1/µ . From the theorem
of the maximal function, it immediately follows that, for p > µ,
1/µ 1/µ
kMµ f kp = k((M |f |µ )(x))1/µ kp = k((M |f |µ )(x))kp/µ 6 k|f |µ kp/µ = kf kp .
(6.24)
This inequality fails for p 6 µ, as in the special case µ = 1.
The substitute for Lemma 6.7 is as follows.
- 144 - 6. The Littlewood-Paley g-function and Multipliers

Lemma 6.12. Let f ∈ Lp (Rn ), p > µ > 1; if u(x, y) is the Poisson integral
of f , then
n
|t|

|u(x − t, y)| 6 A 1 + M f (x), (6.25)
y
and more generally
n/µ
|t|

|u(x − t, y)| 6 Aµ 1 + Mµ f (x). (6.26)
y
We shall now complete the proof of the inequality (6.20) for the case 1 <
p < 2, with the restriction p > 2/λ.
Let us observe that we can always find a µ ∈ [1, p) such that if we set
λ0 = λ − 2−pµ
, then one still has λ0 > 1. In fact, if µ = p, then λ − 2−p
µ
>1
since λ > 2/p; this inequality can then be maintained by a small variation of
µ. With this choice of µ, we have by Lemma 6.12
 n/µ
y
|u(x − t, y)| 6 Aµ Mµ f (x). (6.27)
y + |t|
We now proceed the argument with which we treated the function g.
(gλ∗ (f )(x))2
 λn
1 y
Z
1−n
= y u2−p (x − t, y)|∆up (x − t, y)|dtdy
p(p − 1) Rn+1 +
y + |t|
1 2−p ∗
6 A2−p
µ (Mµ f (x)) I (x), (6.28)
p(p − 1)
where
 λ0 n
y
Z
∗ 1−n
I (x) = y ∆up (x − t, y)dtdy.
R+n+1 y + |t|
It is clear that
 λ0 n
y
Z Z Z
∗ 1−n
I (x)dx = y ∆up (t, y)dxdtdy
R n n+1
R+ Rxn y + |t − x|
Z
=Cλ0 y∆up (t, y)dtdy.
Rn+1
+

The last step follows from the fact that if λ0 > 1

Z   λ0 n Z   λ0 n
−n y −n y
y dx =y dx
Rn y + |t − x| Rn y + |x|
Z  λ0 n
x=yz 1
== dz
Rn 1 + |z|
=Cλ0 < ∞.
So, by Lemma 6.6
6.2. Fourier multipliers on Lp - 145 -
Z Z
∗
I (x)dx = Cλ0 up (t, 0)dt = Cλ0 kf kpp . (6.29)
Rn Rn
Therefore, by (6.28), Hölder’s inequality, (6.24) and (6.29),
1/2
kgλ∗ (f )kp 6 CkMµ f (x)1−p/2 (I ∗ (x))1/2 kp 6 CkMµ f k1−p/2
p kI ∗ k1 6 Ckf kp .
That is the desired result. 
Finally, we prove Lemma 6.12.
Proof of Lemma 6.12. One notices that (6.25) is unchanged by the dilation
(x, t, y) → (δx, δt, δy), it is then clear that it suffices to prove (6.25) with
y = 1.
Setting y = 1 in the Poisson kernel, we have P1 (x) = cn (1 + |x|2 )−(n+1)/2 ,
and u(x − t, 1) = f (x) ∗ P1 (x − t), for R each t. Theorem 4.10 shows that
|u(x − t, 1)| 6 At M f (x), where At = Qt (x)dx, and Qt (x) is the smallest
decreasing radial majorant of P1 (x − t), i.e.,
1
Qt (x) = cn sup 0 2 (n+1)/2
.
|x0 |>|x| (1 + |x − t| )

For Qt (x), we have the easy estimates, Qt (x) 6 cn for |x| 6 2t and Qt (x) 6
A0 (1 + |x|2 )−(n+1)/2 , for |x| > 2|t|, from which it is obvious that At 6 A(1 +
|t|)n and hence (6.25) is Rproved.
Since u(x − t, y) = Rn Py (s)f (x − t − s)ds, and Rn Py (s)ds = 1, by
R

Hölder inequality, we have

0
u(x − t, y) 6kPy1/µ f kµ kPy1/µ kµ0
Z 1/µ
µ
6 Py (s)|f (x − t − s)| ds = U 1/µ (x − t, y),
Rn
where U is the Poisson integral of |f |µ . Apply (6.25) to U , it gives
|u(x − t, y)| 6A1/µ (1 + |t|/y)n/µ (M (|f |µ )(x))1/µ
=Aµ (1 + |t|/y)n/µ Mµ f (x),
and the Lemma is established. 

6.2 Fourier multipliers on Lp

In this section, we introduce briefly the Fourier multipliers on Lp , and we

prove two (or three) main multiplier theorems.
In the study of PDEs, we often investigate the estimates of semigroups. For
example, we consider the linear heat equation
ut − ∆u = 0, u(0) = u0 .
−1 −t|ωξ|2
It is clear that u = F e F u0 =: H(t)u0 is the solution of the above
heat equation. The natural question is: Is H(t) a bounded semigroup from Lp
- 146 - 6. The Littlewood-Paley g-function and Multipliers

to Lp ? In other word, is the following inequality true?

2
kF −1 e−t|ωξ| F u0 kp . ku0 kp , for 1 6 p 6 ∞.
Of course, we have known that this estimate is true. From this example, we
can give a general concept.
Definition 6.13. Let ρ ∈ S 0 . ρ is called a Fourier multiplier on Lp if the
convolution (F −1 ρ) ∗ f ∈ Lp for all f ∈ S , and if
kρkMp = sup k(F −1 ρ) ∗ f kp
kf kp =1

is finite. The linear space of all such ρ is denoted by Mp .

Since S is dense in Lp (1 6 p < ∞), the mapping from S to Lp : f →
(F −1 ρ) ∗ f can be extended to a mapping from Lp to Lp with the same norm.
We write (F −1 ρ) ∗ f also for the values of the extended mapping.
For p = ∞ (as well as for p = 2) we can characterize Mp . Considering the
map:
f → (F −1 ρ) ∗ f for f ∈ S ,
we have
ρ ∈ M∞ ⇔ |F −1 ρ ∗ f (0)| 6 Ckf k∞ , f ∈ S. (6.30)
Indeed, if ρ ∈ M∞ , we have
kF −1 ρ ∗ f k∞
|F −1 ρ ∗ f (0)| 6 kf k∞ 6 Ckf k∞ .
kf k∞
On the other hand, if |F −1 ρ ∗ f (0)| 6 Ckf k∞ , we can get
kF −1 ρ ∗ f k∞ = sup |F −1 ρ ∗ f (x)| = sup |[(F −1 ρ) ∗ (f (x + ·))](0)|
x∈Rn x∈Rn

6Ckf (x + ·)k∞ = Ckf k∞ ,

which yields kρkM∞ 6 C, i.e., ρ ∈ M∞ .
But (6.30) also means that F −1 ρ is a bounded measure on Rn . Thus M∞ is
equal to the space of all Fourier transforms of bounded measures. Moreover,
kρkM∞ is equal to the total mass of F −1 ρ. In view of the inequality above and
the Hahn-Banach theorem, we may extend the mapping f → F −1 ρ ∗ f from
S to L∞ to a mapping from L∞ to L∞ without increasing its norm. We also
write the extended mapping as f → F −1 ρ ∗ f for f ∈ L∞ .
Theorem 6.14. Let 1 6 p 6 ∞ and 1/p + 1/p0 = 1, then we have
Mp = Mp0 (equal norms). (6.31)
Moreover,
M1 = ρ ∈ S 0 : F −1 ρ is a bounded measure


(6.32)
Z
−1
kρkM1 =total mass of F ρ = |F −1 ρ(x)|dx
Rn
and
6.2. Fourier multipliers on Lp - 147 -

M2 = L∞ (equal norm). (6.33)

For the norms (1 6 p0 , p1 6 ∞)
kρkMp 6 kρk1−θ θ
Mp kρkMp1 , ∀ρ ∈ Mp0 ∩ Mp1 (6.34)
0

if 1/p = (1 − θ)/p0 + θ/p1 (0 6 θ 6 1). In particular, the norm k · kM p

decreases with p in the interval 1 6 p 6 2, and
M1 ⊂ Mp ⊂ Mq ⊂ M2 , (1 6 p 6 q 6 2). (6.35)
0
Proof. Let f ∈ Lp , g ∈ Lp and ρ ∈ Mp . Then, we have
kρkMp0 = sup k(F −1 ρ) ∗ gkp0 = sup |h(F −1 ρ) ∗ g(x), f (−x)i|
kgkp0 =1 kf kp =kgkp0 =1
−1
= sup |(F ρ) ∗ g ∗ f (0)| = sup |(F −1 ρ) ∗ f ∗ g(0)|
kf kp =kgkp0 =1 kf kp =kgkp0 =1
Z
= sup | ((F −1 ρ) ∗ f )(y)g(−y)dy|
kf kp =kgkp0 =1 Rn

= sup k(F −1 ρ) ∗ f kp = kρkMp .

kf kp =1

The assertion (6.32) has already been established because of M1 = M∞ .

The Plancherel theorem immediately gives (6.33). In fact,
 n/2
|ω|
−1
kρkM2 = sup k(F ρ) ∗ f k2 = sup kρfˆk2 6 kρk∞ .
kf k2 =1 kf k2 =1 2π
On the other hand, for any ε > 0, we can choose a non-zero measurable set E
such that |ρ(ξ)| > kρk∞ − ε for ξ ∈ E. Then choose a function f ∈ L2 such
that supp F f ⊂ E, we can obtain kρkM2 > kρk∞ − ε.
Invoking the Riesz-Thorin theorem, (6.34) follows, since the mapping f →
(F −1 ρ) ∗ f maps Lp0 → Lp0 with norm kρkMp0 and Lp1 → Lp1 with norm
kρkMp1 .
Since 1/q = (1 − θ)/p + θ/p0 for some θ and p 6 q 6 2 6 p0 , by using
(6.34) with p0 = p, p1 = p0 , we see that
kρkMq 6 kρkMp ,
from which (6.35) follows. 

Proposition 6.15. Let 1 6 p 6 ∞. Then Mp is a Banach algebra under

pointwise multiplication.

Proof. It is clear that k · kMp is a norm. Note also that Mp is complete. Indeed,
let {ρk } is a Cauchy sequence in Mp . So does it in L∞ because of Mp ⊂ L∞ .
Thus, it is convergent in L∞ and we denote the limit by ρ. From L∞ ⊂ S 0 ,
we have F −1 ρk F f → F −1 ρF f for any f ∈ S in sense of the strong
topology on S 0 . On the other hand, {F −1 ρk F f } is also a Cauchy sequence
in Lp ⊂ S 0 , and converges to a function g ∈ Lp . By the uniqueness of limit
- 148 - 6. The Littlewood-Paley g-function and Multipliers

in S 0 , we know that g = F −1 ρF f . Thus, kρk − ρkMp → 0 as k → ∞.

Therefore, Mp is a Banach space.
Let ρ1 ∈ Mp and ρ2 ∈ Mp . For any f ∈ S , we have
k(F −1 ρ1 ρ2 ) ∗ f kp =k(F −1 ρ1 ) ∗ (F −1 ρ2 ) ∗ f kp 6 kρ1 kMp k(F −1 ρ2 ) ∗ f kp
6kρ1 kMp kρ2 kMp kf kp ,
which implies ρ1 ρ2 ∈ Mp and
kρ1 ρ2 kMp 6 kρ1 kMp kρ2 kMp .
Thus, Mp is a Banach algebra. 
In order to clarify the next theorem we write Mp = Mp (R ) for Fourier
n

multipliers which are functions on Rn . The next theorem says that Mp (Rn ) is
isometrically invariant under affine transforms of Rn .
Theorem 6.16. Let a : Rn → Rm be a surjective affine transform1 with
n > m, and ρ ∈ Mp (Rm ). Then
kρ(a(·))kMp (Rn ) = kρkMp (Rm ) .
If m = n, the mapping a∗ is bijective. In particular, we have
kρ(c·)kMp (Rn ) =kρ(·)kMp (Rn ) , ∀c 6= 0, (6.36)
kρ(hx, ·i)kMp (Rn ) =kρ(·)kMp (R) , ∀x 6= 0, (6.37)
Pn
where hx, ξi = i=1 xi ξi .
Proof. It suffices to consider the case that a : Rn → Rm is a linear transform.
Make the coordinate transform
ηi = ai (ξ), 1 6 i 6 m; ηj = ξj , m + 1 6 j 6 n, (6.38)
which can be written as η = A ξ or ξ = Aη where det A 6= 0. Let A be the
−1 >

transposed matrix of A. It is easy to see, for any f ∈ S (Rn ), that

 n Z
|ω|
−1
F ρ(a(ξ))F f (x) = eωixξ ρ(a(ξ))fˆ(ξ)dξ
2π n
 R n Z
|ω|
=| det A| eωix·Aη ρ(η1 , · · · , ηm )fˆ(Aη)dη
2π n
 n ZR
|ω| >
=| det A| eωiA x·η ρ(η1 , · · · , ηm )fˆ(Aη)dη
2π Rn

=| det A|(F ρ(η1 , · · · , ηm )fˆ(Aη))(A> x)

−1

= F −1 ρ(η1 , · · · , ηm ) F f ((A> )−1 ·) (η) (A> x).




It follows from ρ ∈ Mp (Rm ) that for any f ∈ S (Rn )

kF −1 ρ(a(ξ))F f kp
=| det A|−1/p kF −1 ρ(η1 , · · · , ηm ) F f ((A> )−1 ·) (η)kp

1
An affine transform of Rn is a map F : Rn → Rn of the form F (p) = Ap + q for all p ∈ Rn ,
where A is a linear transform of Rn and q ∈ Rn .
6.2. Fourier multipliers on Lp - 149 -

=| det A|−1/p Fη−1 ∗ kf ((A> )−1 ·)kLp (Rn−m )

1 ,··· ,η m
ρ(η 1 , · · · , ηm ) Lp (Rm )

6kρkMp (Rm ) kf kp .
Thus, we have
kρ(a(·))kMp (Rn ) 6 kρkMp (Rm ) . (6.39)
Taking f ((A> )−1 ·) = f1 (x1 , · · · , xm )f2 (xm+1 , · · · , xn ), one can conclude
that the inverse inequality (6.39) also holds. 
Now we give a simple but very useful theorem for Fourier multipliers.
Theorem 6.17 (Bernstein multiplier theorem). Assume that k > n/2 is an
integer, and that ∂xαj ρ ∈ L2 (Rn ), j = 1, · · · , n and 0 6 α 6 k. Then we have
ρ ∈ Mp (Rn ), 1 6 p 6 ∞, and
n
!n/2k
1−n/2k
X
kρkMp . kρk2 k∂xkj ρk2 .
j=1
Pn
Proof. Let t > 0 and J(x) = j=1 |xj |k . By the Cauchy-Schwartz inequality
and the Plancherel theorem, we obtain
Z Z n
X
−1 −1 −1 n/2−k
|F ρ(x)|dx = J(x) J(x)|F ρ(x)|dx . t k∂xkj ρk2 .
|x|>t |x|>t j=1
Similarly, we have
Z
|F −1 ρ(x)|dx . tn/2 kρk2 .
|x|6t
Choosing t such that kρk2 = t−k nj=1 k∂xkj ρk2 , we infer, with the help of
P

Theorem 6.14, that

Z n
!n/2k
1−n/2k
X
kρkMp 6kρkM1 = |F −1 ρ(x)|dx . kρk2 k∂xkj ρk2 .
Rn j=1
This completes the proof. 
The first application of the theory of the functions g and will be in the
gλ∗
study of multipliers. Our main tool when proving theorems for the Sobolev
and Besov spaces, defined in the following chapters, is the following theorem.
Note that 1 < p < ∞ here in contrast to the case in Theorem 6.17. We give
the theorem as follows.
Theorem 6.18 (Mihlin multiplier theorem). Suppose that ρ(ξ) ∈ C k (Rn \
{0}) where k > n/2 is an integer. Assume also that for every differential
α
∂
monomial ∂ξ , α = (α1 , α2 , ..., αn ), with |α| = α1 + α2 + ... + αn , we have
Mihlin’s condition
 α
∂
ρ(ξ) 6 A|ξ|−|α| , whenever |α| 6 k. (6.40)
∂ξ
- 150 - 6. The Littlewood-Paley g-function and Multipliers

Then ρ ∈ Mp , 1 < p < ∞, and

kρkMp 6 Cp,n A.
The proof of the theorem leads to a generalization of its statement which we
formulate as a corollary.
Corollary 6.19 (Hörmander multiplier theorem). The assumption (6.40)
can be replaced by the weaker assumptions, i.e., Hörmander’s condition
|ρ(ξ)| 6A,
 α 2
∂ (6.41)
Z
2|α|−n
sup R ρ(ξ) dξ 6A, |α| 6 k.
0<R<∞ R6|ξ|62R ∂ξ
The theorem and its corollary will be consequences of the following lemma.
Its statement illuminates at the same time the nature of the multiplier trans-
forms considered here, and the role played by the g-functions and their vari-
ants.
Lemma 6.20. Under the assumptions of Theorem 6.18 or Corollary 6.19, let
us set for f ∈ L2 (Rn )
F (x) = Tρ f (x) = (F −1 (ρ(ξ)) ∗ f )(x).
Then
g1 (F )(x) 6 Aλ gλ∗ (f )(x), where λ = 2k/n. (6.42)
Thus in view of the lemma, the g-functions and their variants are the char-
acterizing expressions which deal at once with all the multipliers considered.
On the other hand, the fact that the relation (6.42) is pointwise shows that to a
large extent the mapping Tρ is “semi-local”.
Proof of Theorem 6.18 and Corollary 6.19. The conclusion is deduced from
the lemma as follows. Our assumption on k is such that λ = 2k/n > 1. Thus,
Theorem 6.11 shows us that
kgλ∗ (f )(x)kp 6 Aλ,p kf kp , 2 6 p < ∞, if f ∈ L2 ∩ Lp .
However, by Corollary 6.3, A0p kF kp 6 kg1 (F )(x)kp , therefore by Lemma
6.20,
kTρ f kp = kF kp 6 Aλ kgλ∗ (f )(x)kp 6 Ap kf kp , if 2 6 p < ∞ and f ∈ L2 ∩Lp .
That is, ρ ∈ Mp , 2 6 p < ∞. By duality, i.e., (6.31) of Theorem 6.14, we have
also ρ ∈ Mp , 1 < p 6 2, which gives the assertion of the theorem. 
Now we shall prove Lemma 6.20.
Proof of Lemma 6.20. Let u(x, y) denote the Poisson integral of f , and U (x, y)
the Poisson integral of F . Then withˆdenoting the Fourier transform w.r.t. the
x variable, we have
û(ξ, y) = e−|ωξ|y fˆ(ξ), and U
b (ξ, y) = e−|ωξ|y Fb(ξ) = e−|ωξ|y ρ(ξ)fˆ(ξ).
6.2. Fourier multipliers on Lp - 151 -
 n R
|ω|
Define M (x, y) = 2π Rn
eωix·ξ e−|ωξ|y ρ(ξ)dξ. Then clearly M
c(ξ, y) =
e−|ωξ|y ρ(ξ), and so
U
b (ξ, y1 + y2 ) = M
c(ξ, y1 )û(ξ, y2 ), y = y1 + y2 , y1 , y2 > 0.
This can be written as Z
U (x, y1 + y2 ) = M (t, y1 )u(x − t, y2 )dt.
Rn
We differentiate this relation k times w.r.t. y1 and once w.r.t. y2 , and set y1 =
y2 = y/2. This gives us the identity
Z
(k+1)
U (x, y) = M (k) (t, y/2)u(1) (x − t, y/2)dt. (6.43)
Rn
Here the superscripts denote the differentiation w.r.t. y.
Next, we translates the assumptions (6.40) (or (6.41)) on ρ in terms of
M (x, y). The result is
|M (k) (t, y)| 6A0 y −n−k , (6.44)
Z
|t|2k |M (k) (t, y)|2 dt 6A0 y −n . (6.45)
Rn
In fact, by the definition of M and the condition |ρ(ξ)| 6 A, it follows that
 n
|ω|
Z
(k)
|M (x, y)| 6 |ω| k
|ξ|k e−|ωξ|y ρ(ξ)dξ
2π n
 nR Z ∞
|ω|
6Aωn−1 |ω| k
rk e−|ω|ry rn−1 dr = A0 y −n−k ,
2π 0
which is (6.44).
To prove (6.45), let us show more particularly that
Z
|xα M (k) (x, y)|2 dx 6 A0 y −n ,
Rn
where |α| = k.
By Plancherel’s theorem
 n/2  α
|ω| ∂
α (k)
kx M (x, y)k2 = |ω|k
(|ξ|k ρ(ξ)e−|ωξ|y ) . (6.46)
2π ∂ξ 2
So we need to evaluate, by using Leibniz’ rule,
 α  β  γ
∂ −|ωξ|y
X ∂ ∂
k
(|ξ| ρ(ξ)e )= Cβ,γ k
(|ξ| ρ(ξ)) e−|ωξ|y .
∂ξ β+γ=α
∂ξ ∂ξ
(6.47)
Case I: (6.40) =⇒ (6.45). By the hypothesis (6.40) and Leibniz’ rule again,
we have
 β
∂
(|ξ|k ρ(ξ)) 6 A0 |ξ|k−|β| , with |β| 6 k.
∂ξ
Thus,
- 152 - 6. The Littlewood-Paley g-function and Multipliers
α
∂
(|ξ|k ρ(ξ)e−|ωξ|y )
∂ξ
X X
6C |ξ|k−|β| y |γ| e−|ωξ|y 6 C |ξ|r y r e−|ωξ|y .
|β|+|γ|=k 06r6k
Since for r Z> 0 Z ∞
2r −2|ωξ|y
y 2r
|ξ| e dξ =Cy 2r
R2r e−2|ω|Ry Rn−1 dR
Rn
Z0 ∞
=Cy −n z 2r e−2|ω|z z n−1 dz 6 Cy −n ,
0
we get
kxα M (k) (x, y)k22 6 A0 y −n , |α| = k,
which proves the assertion (6.45).
Case II: (6.41) =⇒ (6.45). From (6.46) and (6.47), we have, by Leibniz’
rule again and (6.41),
kxα M (k) (x, y)k2
 1/2
 β 0 2  β 00 2
∂ ∂
X Z
6C  |ξ|k ρ(ξ) e−2|ωξ|y y 2|γ| dξ 
Rn ∂ξ ∂ξ
|β 0 |+|β 00 |+|γ|=k
 1/2
 β 00 2
X
|γ| 
XZ 0 ∂
6C y |ξ|2(k−|β |) ρ(ξ) e−2|ωξ|y dξ 
2j y6|ξ|62j+1 y ∂ξ
|β 0 |+|β 00 |+|γ|=kj∈Z
0 j y2
X X
6C (2j+1 y)k−|β | y |γ| e−|ω|2
|β 0 |+|β 00 |+|γ|=k j∈Z
 1/2
 β 00 2
∂
Z
00 |+n/2
· (2j y)−|β (2j y)2|β 00 |−n ρ(ξ) dξ 
j
2 y6|ξ|62j+1 y ∂ξ
j 2 j 2
XX X X
6CA (2j y)|γ|+n/2 y |γ| e−|ω|2 y = CAy −n/2 (2j y 2 )r e−|ω|2 y
|γ|6k j∈Z 06r6k j∈Z

6Cy −n/2 ,
which yields (6.45).
Now, we return to the identity (6.43), and for each y divide the range of
integration into two parts, |t| 6 y/2 and |t| > y/2. In the first range, use the
estimate (6.44) on M (k) and in the second range, use the estimate (6.45). This
together with Schwarz’ inequality gives immediately
Z
(k+1) 2 −n−2k
|U (x, y)| 6Cy |u(1) (x − t, y/2)|2 dt
|t|6y/2

|u(1) (x − t, y/2)|2 dt
Z
+ Cy −n
|t|>y/2 |t|2k
6.3. The partial sums operators - 153 -

=:I1 (y) + I2 (y).

Now
Z ∞ 2 Z
X ∞
2 (k+1) 2 2k+1
(gk+1 (F )(x)) = |U (x, y)| y dy 6 Ij (y)y 2k+1 dy.
0 j=1 0

However, by a change of variable y/2 → y,

Z ∞ Z ∞Z
I1 (y)y 2k+1
dy 6C |u(1) (x − t, y/2)|2 y −n+1 dtdy
0 |t|6y/2
Z0
6C |∇u(x − t, y)|2 y −n+1 dtdy = C(S(f )(x))2
Γ
6Cλ (gλ∗ (f )(x))2 .
Similarly, with nλ = 2k,
Z ∞ Z ∞ Z
I2 (y)y 2k+1
dy 6C y −n+2k+1 |t|−2k |∇u(x − t, y)|2 dtdy
0 0 |t|>y
∗
6C(gλ (f )(x))2 .
This shows that However by Remark 6.4 (iii) of g-
gk+1 (F )(x) 6 Cλ gλ∗ (f )(x).
functions after Corollary 6.3, we know that g1 (F )(x) 6 Ck gk+1 (F )(x). Thus,
the proof of the lemma is concluded. 

6.3 The partial sums operators

We shall now develop the second main tool in the Littlewood-Paley theory,
(the first being the usage of the functions g and g ∗ ).
Let ρ denote an arbitrary rectangle in Rn . By rectangle we shall mean, in
the rest of this chapter, a possibly infinite rectangle with sides parallel to the
axes, i.e., the Cartesian product of n intervals.
Definition 6.21. For each rectangle ρ denote by Sρ the partial sum opera-
tor, that is the multiplier operator with m = χρ = characteristic function
of the rectangle ρ. So
F (Sρ (f )) = χρ fˆ, f ∈ L2 (Rn ) ∩ Lp (Rn ). (6.48)
For this operator, we immediately have the following theorem.
Theorem 6.22.
kSρ (f )kp 6 Ap kf kp , f ∈ L2 ∩ Lp ,
if 1 < p < ∞. The constant Ap does not depend on the rectangle ρ.
However, we shall need a more extended version of the theorem which
arises when we replace complex-valued functions by functions taking their
value in a Hilbert space.
- 154 - 6. The Littlewood-Paley g-function and Multipliers

Let H be the sequence Hilbert space,

X
H = {(cj )∞
j=1 : ( |cj |2 )1/2 = |c| < ∞}.
j
Then we can represent a function f ∈ Lp (Rn , H ), as sequences
f (x) = (f1 (x), · · · , fj (x), · · · ),
where each fj is complex-valued and |f (x)| = ( ∞ 2 1/2
. Let < be
P
j=1 |fj (x)| )
∞
a sequence of rectangle, < = {ρj }j=1 . Then we can define the operator S< ,
mapping L2 (Rn , H ) to itself, by the rule
S< (f ) = (Sρ1 (f1 ), · · · , Sρj (fj ), · · · ), where f = (f1 , · · · , fj , · · · ). (6.49)
We first give a lemma, which will be used in the proof of the theorem or its
generalization. Recall the Hilbert transform f → H(f ), which corresponds to
the multiplier −i sgn (ω) sgn (ξ) in one dimension.
Lemma 6.23. Let f (x) = (f1 (x), · · · , fj (x), · · · ) ∈ L2 (Rn , H ) ∩
Lp (Rn , H ). Denote Hf
e (x) = (Hf1 (x), · · · , Hfj (x), · · · ). Then
kHf
e kp 6 Ap kf kp , 1 < p < ∞,
where Ap is the same constant as in the scalar case, i.e., when H is one-
dimensional.

Proof. We use the vector-valued version of the Hilbert transform, as is de-

scribed more generally in Sec. 4.7. Let the Hilbert spaces H1 and H2 be both
identical with H . Take in R, K(x) = I · 1/πx, where I is the identity map-
ping on H . Then the kernel K(x) satisfies all the assumptions of Theorem
4.27 and Theorem 4.24.ZMoreover,
lim K(y)f (x − y)dy = Hf
e (x),
ε→0 |y|>ε
and so our lemma is proved. 
The generalization of Theorem 6.22 is then as follows.
Theorem 6.24. Let f ∈ L2 (Rn , H ) ∩ Lp (Rn , H ). Then
kS< (f )kp 6 Ap kf kp , 1 < p < ∞, (6.50)
where Ap does not depend on the family < of rectangles.

Proof. The theorem will be proved in four steps, the first two of which already
contain the essence of the matter.
Step 1: n = 1, and the rectangles ρ1 , ρ2 , · · · , ρj , · · · are the semi-infinite
intervals (−∞, 0).
It is clear that S(−∞,0) f = F −1 χ(−∞,0) F f = F −1 1− sgn 2
(ξ)
F f , so
I − i sgn (ω)H
S(−∞,0) = , (6.51)
2
6.3. The partial sums operators - 155 -

where I is the identity, and S(−∞,0) is the partial sum operator corresponding
to the interval (−∞, 0).
Now if all the rectangles are the intervals (−∞, 0), then by (6.51),
I − i sgn (ω)He
S< =
2
and so by Lemma 6.23, we have the desired result.
Step 2: n = 1, and the rectangles are the intervals (−∞, a1 ), (−∞, a2 ), · · · ,
(−∞, aj ), · · · .
Notice that F (f (x)e−ωix·a ) = fˆ(ξ + a), therefore
F (H(e−ωix·a f (x))) = −i sgn (ω) sgn (ξ)fˆ(ξ + a),
and hence F (eωix·a H(e−ωix·a f (x))) = −i sgn (ω) sgn (ξ − a)fˆ(ξ). From this,
we see that
fj − i sgn (ω)eωix·aj H(e−ωix·aj fj )
(S(−∞,aj ) fj )(x) = . (6.52)
2
If we now write symbolically e−ωix·a f for
(e−ωix·a1 f1 , · · · , e−ωix·aj fj , · · · )
with f = (f1 , · · · , fj , · · · ), then (6.52) may be written as
e −ωix·a f )
f − i sgn (ω)eωix·a H(e
S< f = , (6.53)
2
and so the result again follows in this case by Lemma 6.23.
Step 3: General n, but the rectangles ρj are the half-spaces x1 < aj , i.e.,
ρj = {x : x1 < aj }.
(1)
Let S(−∞,aj ) denote the operator defined on L2 (Rn ), which acts only on the
x1 variable, by the action given by S(−∞,aj ) . We claim that
(1)
Sρj = S(−∞,aj ) . (6.54)
This identity is obvious for L2 functions of the product form
f 0 (x1 )f 00 (x2 , · · · , xn ),
since their linear span is dense in L2 , the identity (6.54) is established.
We now use the Lp inequality, which is the result of the previous step for
each fixed x2 , x3 , · · · , xn . We raise this inequality to the pth power and integrate
w.r.t. x2 , · · · , xn . This gives the desired result for the present case. Notice that
the result holds as well if the half-space {x : x1 < aj }∞ j=1 , is replaced by the
half-space {x : x1 > aj }∞ j=1 , or if the role of the x 1 axis is taken by the x2
axis, etc.
Step 4: Observe that every general finite rectangle of the type considered is
the intersection of 2n half-spaces, each half-space having its boundary hyper-
plane perpendicular to one of the axes of Rn . Thus a 2n-fold application of the
result of the third step proves the theorem, where the family < is made up of
finite rectangles. Since the bounds obtained do not depend on the family <, we
- 156 - 6. The Littlewood-Paley g-function and Multipliers

can pass to the general case where < contains possibly infinite rectangles by
an obvious limiting argument. 
We state here the continuous analogue of Theorem 6.24. Let (Γ, dγ) be a σ-
finite measure space,2 and consider the Hilbert space H of square integrable
functions on Γ , i.e., H = L2 (Γ, dγ). The elements
f ∈ Lp (Rn , H )
are the complex-valued functions
R R f (x, γ) =2 fγ (x) on Rn ×Γ , which are jointly
measuable, and for which ( Rn ( Γ |f (x, γ)| dγ)p/2 dx)1/p = kf kp < ∞, if p <
∞. Let < = {ργ }γ∈Γ , and suppose that the mapping γ → ργ is a measurable
function from Γ to rectangles; that is, the numerical-valued functions which
assign to each γ the components of the vertices of ργ are all measurable.
Suppose f ∈ L2 (Rn , H ). Then we define F = S< f by the rule
F (x, γ) = Sργ (fγ )(x), (fγ (x) = f (x, γ)).
Theorem 6.25.
kS< f kp 6 Ap kf kp , 1 < p < ∞, (6.55)
2 n p n
for f ∈ L (R , H ) ∩ L (R , H ), where the bound Ap does not depend on the
measure space (Γ, dγ), or on the function γ → ργ .

Proof. The proof of this theorem is an exact repetition of the argument given
for Theorem 6.24. The reader may also obtain it from Theorem 6.24 by a
limiting argument. 

6.4 The dyadic decomposition

We shall now consider a decomposition of Rn into rectangles.

First, in the case of R, we decompose it as the union of the “disjoint” in-
tervals (i.e., whose interiors are disjoint) [2k , 2k+1 ], −∞ < k < ∞, and
[−2k+1 , −2k ], −∞ < k < ∞. This double collection of intervals, one col-
lection for the positive half-line, the other for the negative half-line, will be the
dyadic decomposition of R.3
Having obtained this decomposition of R, we take the corresponding prod-
uct decomposition for Rn . Thus we write Rn as the union of “disjoint” rectan-
gles, which rectangles are products of the intervals which occur for the dyadic
decomposition of each of the axes. This is the dyadic decomposition of Rn .

2
If µ is measure on a ring R, a set E is said to have σ-finite measure if there exists a sequence {En }
of sets in R such that E ⊂ ∪∞ n=1 En , and µ(En ) < ∞, n = 1, 2, · · · . If the measure of every set
E in R is σ-finite, the measure µ is called σ-finite on R.
3
Strictly speaking, the origin is left out; but for the sake of simplicity of terminology, we still refer to
it as the decomposition of R.
6.4. The dyadic decomposition - 157 -

The family of resulting rectangles will be

denoted by ∆. We recall the partial sum op-
erator Sρ , defined in (6.48) for each rectan-
gle. Now in an obvious sense, (e.g. L2 con-
vergence)
X 1
Sρ = Identity.
ρ∈∆

Also in the L case, the different blocks,

2

Sρ f , ρ ∈ ∆, behave as if they were indepen-

dent; they are of course mutually orthogonal.
Fig. 6.3 The dyadic decomposition
To put the matter precisely: The L2 norm of f Figure 1: The dyadic decomposition
can be given exactly in terms of the L2 norms of Sρ f , i.e.,
X
kSρ f k22 = kf k22 , (6.56)
ρ∈∆
(and this is true for any decomposition of Rn ). For the general Lp case not as
much can be hoped for, but the following important theorem can nevertheless
be established.
Theorem 6.26 (Littlewood-Paley square function theorem). Suppose f ∈
Lp (Rn ), 1 < p < ∞. Then
X
k( |Sρ f (x)|2 )1/2 kp ∼ kf kp .
ρ∈∆

The Rademacher functions provide a very

useful device in the study of Lp norms in terms r1 (t)
of quadratic expressions.
These functions, r0 (t), r1 (t), · · · , rm (t), · · ·
1 1 t
are defined on the interval (0, 1) as follows: 2

r0 (t)

1, 0 6 t 6 1/2,
r0 (t) =
−1, 1/2 < t < 1, Fig. 6.4 r0 (t) and r1 (t)
r0 is extended outside the unit interval by pe- Figure 1: r0 (t) and r1 (t)
riodicity, i.e., r0 (t + 1) = r0 (t). In general, rm (t) = r0 (2m t). The sequences
of Rademacher functions are orthonormal (and in fact mutually independent)
over [0, 1]. In fact, for m < k, the integral
Z 1 Z 1 Z 2m
m k −m
rm (t)rk (t)dt = r0 (2 t)r0 (2 t)dt = 2 r0 (s)r0 (2k−m s)ds
0 0 0
Z 1 Z 1/2 Z 1
= r0 (s)r0 (2k−m s)ds = r0 (2k−m s)ds − r0 (2k−m s)ds
0 0 1/2
"Z k−m−1 Z k−m #
2 2
=2m−k r0 (t)dt − r0 (t)dt
0 2k−m−1
- 158 - 6. The Littlewood-Paley g-function and Multipliers
Z 1 Z 1 
−1
=2 r0 (t)dt − r0 (t)dt = 0,
0 0
R1
so, they are orthogonal. It is clear that they are normal since 0 (rm (t))2 dt = 1.
For our purposes, their importance arises from the following fact.
Suppose ∞
P∞
2
and set m=0 am rm (t). Then for every
P
m=0 |am | < ∞ F (t) =
1 < p < ∞, F (t) ∈ L [0, 1] and
p

X∞
Ap kF kp 6 kF k2 = ( |am |2 )1/2 6 Bp kF kp , (6.57)
m=0
for two positive constants Ap and Bp .
Thus, for functions which can be expanded in terms of the Rademacher
functions, all the Lp norms, 1 < p < ∞, are comparable.
We shall also need the n-dimensional form of (6.57). We consider the
unit cube Q ⊂ Rn , Q = {t = (t1 , t2 , · · · , tn ) : 0 6 tj 6 1}. Let
m be an n-tuple of non-negative integers m = (m1 , m2 , · · · , mn ). Define
rm (t) = rm1 (t1 )rm2 (t2 ) · · · rmn (tn ). Write F (t) = am rm (t). With
P
Z 1/p
p
kF kp = |F (t)| dt ,
Q
we also have (6.57), whenever |am |2 < ∞. That is
P

Lemma 6.27. Suppose |am |2 < ∞. Then it holds

P

∞
!1/2
X
kF kp ∼ kF k2 = |am |2 , 1 < p < ∞. (6.58)
m=0

Proof. We split the proof into four steps.

Step 1: Let µ, a0 , a1 , · · · , aN , be real numbers. Then because the
Rademacher functions are mutually independent variables, we have, in view
of their definition,
Z 1 Z 1 Z 2m Z 1
µam rm (t) µam r0 (2m t) −m µam r0 (s)
e dt = e dt = 2 e ds = eµam r0 (s) ds
0 0 0 0
−1 µam −µam
=2 (e +e ) = cosh µam .
and for m < k
Z 1 Z 1
µam rm (t) µak rk (t) m k
e e dt = eµam r0 (2 t) eµak r0 (2 t) dt
0 0
Z 2m Z 1
−m µam r0 (s) µak r0 (2k−m s) k−m
=2 e e ds = eµam r0 (s) eµak r0 (2 s) ds
0 0
Z 1/2 Z 1
k−m k−m
= eµam eµak r0 (2 s) ds + e−µam eµak r0 (2 s) ds
0 1/2
"Z k−m−1 Z k−m #
2 2
=2m−k eµam eµak r0 (t) dt + e−µam eµak r0 (t) dt
0 2k−m−1
6.4. The dyadic decomposition - 159 -
Z 1 Z 1 Z 1
−1 µam −µam µak r0 (t) µam rm (t)
=2 (e +e ) e dt = e dt eµak rk (t) dt.
0 0 0
Thus, by induction, we can verify
Z 1 P N Z 1
µ N
Y
a r (t)
e m=0 m m
dt = eµam rm (t) dt.
0 m=0 0
2
If we now make use of this simple inequality cosh x 6 ex (since cosh x =
P∞ x2k P∞ x2k x2
k=0 (2k)! 6 k=0 k! = e for |x| < ∞ by Taylor expansion), we obtain
Z 1 N N
Y Y 2 2 2
PN 2
eµF (t) dt = cosh µam 6 eµ am = eµ m=0 am ,
0 m=0 m=0
PN
with F (t) = m=0 am rm (t).
Step 2: Let us make the normalizing assumption that Nn=0 am = 1. Then,
2
P

since e µ|F | µF
6e +e −µF
, we have
Z 1
2
eµ|F (t)| dt 6 2eµ .
0
Recall the distribution function F∗ (α) = m{t ∈ [0, 1] : |F (t)| > α}. If we
take µ = α/2 in the above inequality, we have
Z Z
2 α2 α2 α2
− α2 α
F∗ (α) = dt 6 e e 2 |F (t)| dt 6 e− 2 2e 22 6 2e− 4 .
|F (t)|>α |F (t)|>α
R∞ 2
From Theorem 2.16, the above and the formula xb e−ax dx = Γ ((b +
√ 0
1)/2)/2 ab+1 , it follows immediately that
 Z ∞ 1/p
p−1
kF kp = p α F∗ (α)dα
0
 Z ∞ 1/p
2
p−1 − α4
6 2p α e dα = 2(pΓ (p/2))1/p ,
0
for 1 6 p < ∞, and so in general
∞
!1/2
X
kF kp 6 Ap |am |2 , 1 6 p < ∞. (6.59)
m=0
Step 3: We shall now extend the last inequality to several variables. The case
of two variables is entirely of the inductive procedure used in the proof of the
general case.
We can also limit ourselves to the situation when p > 2, since for the
case p < 2 the desired inequality is a simple consequence of Hölder’s in-
equality. (Indeed, for p < 2 and some q > 2, we have kF kLp (0,1) 6
kF kLq (0,1) k1kLqp/(q−p) (0,1) 6 kF kLq (0,1) by Hölder’s inequality.)
We have
- 160 - 6. The Littlewood-Paley g-function and Multipliers

N
X N
X N
X
F (t1 , t2 ) = am1 m2 rm1 (t1 )rm2 (t2 ) = Fm1 (t2 )rm1 (t1 ).
m1 =0 m2 =0 m1 =0
By(6.59), it follows
Z 1 !p/2
X
|F (t1 , t2 )|p dt1 6 App |Fm1 (t2 )|2 .
0 m1
Integrating this w.r.t. t2 , and using Minkowski’s inequlaity with p/2 > 1,
we have
Z 1 X !p/2 p/2 !p/2
X X
|Fm1 (t2 )|2 dt2 = |Fm1 (t2 )|2 6 k|Fm1 (t2 )|2 kp/2
0 m1 m1 m1
p/2
!p/2
X
= kFm1 (t2 )k2p .
m1
However, Fm1 (t2 ) = am1 m2 rm2 (t2 ), and therefore the case already
P
m2
proved shows that
X
kFm1 (t2 )k2p 6 A2p a2m1 m2 .
m2
Inserting this in the above gives
Z 1Z 1 !p/2
XX
|F (t1 , t2 )|p dt1 dt2 6 App a2m1 m2 ,
0 0 m1 m2
which leads to the desired inequality
kF kp 6 Ap kF k2 , 2 6 p < ∞.
Step 4: The converse inequality
kF k2 6 Bp kF kp , p>1
is a simple consequence of the direct inequality.
In fact, for any p > 1, (here we may assume p < 2) by Hölder inequality
1/2
kF k2 6 kF kp1/2 kF kp0 .
We already know that kF kp0 6 A0p0 kF k2 , p0 > 2. We therefore get
kF k2 6 (A0p0 )2 kF kp ,
which is the required converse inequality. 
Now, let us return to the proof of the Littlewood-Paley square function the-
orem.
Proof of Theorem 6.26. It will be presented in five steps.
Step 1: We show here that it suffices to prove the inequality
X 1/2
2
|Sρ f (x)| 6 Ap kf kp , 1 < p < ∞, (6.60)
ρ∈∆ p
6.4. The dyadic decomposition - 161 -
0
for f ∈ L2 (Rn ) ∩ Lp (Rn ). To see this sufficiency, let g ∈ L2 (Rn ) ∩ Lp (Rn ),
and consider the identity
XZ Z
Sρ f Sρ gdx = f ḡdx
ρ∈∆ Rn Rn

which follows from (6.56) by polarization. By Schwarz’s inequality and then

Hölder’s inequality
Z Z X ! 21 ! 21
X
f ḡdx 6 |Sρ f |2 |Sρ g|2 dx
Rn Rn ρ ρ
! 21 ! 21
X X
6 |Sρ f |2 |Sρ g|2 .
ρ ρ
p p0
Taking the supremum over all such g with the additional restriction that
kgkp0 6 1, gives kf kp for the l.h.s. of the above inequality. The r.h.s. is ma-
jorized by
X 1/2
Ap0 |Sρ f |2 ,
p
since we assume (6.60) for all p. Thus, we have also
!1/2
X
Bp kf kp 6 |Sρ f |2 . (6.61)
ρ
p
To dispose of the additional assumption that f ∈ L , for f ∈ Lp take fj ∈ 2

L2 ∩ Lp such that kfj − f kp → 0; use the inequality (6.60) and (6.61) for fj
and fj − fj 0 ; after a simple limiting argument, we get (6.60) and (6.61) for f
as well.
Step 2: Here we shall prove the inequality (6.60) for n = 1.
We shall need first to introduce a little more notations. We let ∆1 be
the family of dyadic intervals in R, we can enumerate them as I0 , I1 ,
· · · , Im , · · · (the order is here immaterial). For each I ∈ ∆1 , we con-
sider the partial sum operator SI , and a modification of it that we now
define. Let ϕ ∈ C 1 be a fixed function with the following properties:

1, 1 6 ξ 6 2, ϕ(ξ)
ϕ(ξ) = 1
0, ξ 6 1/2, or ξ > 4.
Suppose I is any dyadic interval, and
1 2 3 4 ξ
assume that it is of the form [2k , 2k+1 ].
Define S̃I by Fig. 6.5 ϕ(ξ)
Figure 1: ϕ(ξ)
F (S̃I f )(ξ) = ϕ(2 ξ)fˆ(ξ) = ϕI (ξ)fˆ(ξ).
−k
(6.62)
- 162 - 6. The Littlewood-Paley g-function and Multipliers

That is, S̃I , like SI , is a multiplier transform where the multiplier is equal to
one on the interval I; but unlike SI , the multiplier of S̃I is smooth.
A similar definition is made for S̃I when I = [−2k+1 , −2k ]. We observe that
SI S̃I = SI , (6.63)
since SI has as multiplier the characteristic function of I.
Now for each t ∈ [0, 1], consider the multiplier transform
∞
X
T̃t = rm (t)S̃Im .
m=0

That is, for each t, T̃t is the multiplier transform whose multiplier is mt (ξ),
with
X∞
mt (ξ) = rm (t)ϕIm (ξ). (6.64)
m=0
By the definition of ϕIm , it is clear that for any ξ at most three terms in the
sum (6.64) can be non-zero. Moreover, we also see easily that
dmt B
|mt (ξ)| 6 B, (ξ) 6 , (6.65)
dξ |ξ|
where B is independent of t. Thus, by the Mihlin multiplier theorem (Theorem
6.18)
kT̃t f kp 6 Ap kf kp , for f ∈ L2 ∩ Lp , (6.66)
and with Ap independent of t. From this, it follows obviously that
Z 1 1/p
p
kT̃t f kp dt 6 Ap kf kp .
0
However, by Lemma 6.27 about the Rademacher functions,
Z 1 Z 1Z X p
p
kT̃t f kp dt = rm (t)(S̃Im f )(x) dxdt
0 0 R1
Z !p/2
X
>A0p |S̃Im f (x)|2 dx.
R1 m
Thus, we have
!1/2
X
|S̃Im (f )|2 6 Bp kf kp . (6.67)
m
p
Now using (6.63), applying the general theorem about partial sums,
Theorem 6.24, with < = ∆1 here and (6.67), we get, for F =
(S̃I0 f, S̃I1 f, · · · , S̃Im f, · · · ),
!1/2 !1/2
X X
|SIm (f )|2 = |SIm S̃Im (f )|2 = kS∆1 F kp
m m
p p
6.4. The dyadic decomposition - 163 -
!1/2
X
6Ap kF kp = Ap |S̃Im (f )|2 6 Ap Bp kf kp = Cp kf kp , (6.68)
m
p
which is the one-dimensional case of the inequality (6.60), and this is what we
had set out to prove.
Step 3: We are still in the one-dimensional case, and we write Tt for the
operator
X
Tt = rm (t)SIm .
m
Our claim is that
kTt f kLpt,x 6 Ap kf kp , 1 < p < ∞, (6.69)
with Ap independent of t, and f ∈ L ∩ L . 2 p
PN
Write TtN = m=0 rm (t)SIm , and it suffices to show that (6.69) holds,
with Tt in place of Tt (and Ap independent of N and t). Since each SIm is a
N

bounded operator on L2 and Lp , we have that TtN f ∈ L2 ∩ Lp and so we can

apply (6.61) to it, which has already been proved in the case n = 1. So
N
!1/2
X
Bp kTtN f kLpt,x 6 |SIm f |2 6 Cp kf kp ,
m=0
p
by using (6.68). Letting N → ∞, we get (6.69).
(1)
Step 4: We now turn to the n-dimensional case and define Tt1 , as the op-
erator Tt1 acting only on the x1 variable. Then, by the inequality (6.69), we
get
Z 1Z Z
(1) p p
|Tt1 f (x1 , x2 , · · · , xn )| dx1 dt1 6 Ap |f (x1 , · · · , xn )|p dx1 ,
0 R1 R1
(6.70)
for almost every fixed x2 , x3 , · · · , xn , since x1 → f (x1 , x2 , · · · , xn ) ∈
L2 (R1 ) ∩ Lp (R1 ) for almost every fixed x2 , · · · , xn , if f ∈ L2 (Rn ) ∩ Lp (Rn ).
If we integrate (6.70) w.r.t. x2 , · · · , xn , we obtain
(1)
kTt1 f kLpt ,x 6 Ap kf kp , f ∈ L2 ∩ Lp , (6.71)
1
with Ap independent of t1 . The same inequality of course holds with x1 re-
placed by x2 , or x3 , etc.
Step 5: We first describe the additional notation we shall need. With ∆ rep-
resenting the collection of dyadic rectangles in Rn , we write any ρ ∈ ∆, as ρ =
Im1 × Im2 × · · · × Imn where I0 , I1 , · · · , Im , · · · represents the arbitrary enu-
meration of the dyadic intervals used above. Thus if m = (m1 , m2 , · · · , mn ),
with each mj > 0, we write ρm = Im1 × Im2 × · · · × Imn .
(1)
We now apply the operator Tt1 for the x1 variable, and successively its
analogues for x2 , x3 , etc. The result is
- 164 - 6. The Littlewood-Paley g-function and Multipliers

kTt f kLpt,x 6 Anp kf kp . (6.72)

Here
X
Tt = rm (t)Sρm
ρm ∈∆
with rm (t) = rm1 (t1 ) · · · rmn (tn ) as described in the previous. The inequality
holds uniformly for each (t1 , t2 , · · · , tn ) in the unit cube Q.
We raise this inequality to the pth power and integrate it w.r.t. t, making use
of the properties of the Rademacher functions, i.e., Lemma 6.27. We then get,
as in the analogous proof of (6.67), that
!1/2
X
|Sρm f |2 6 Ap kf kp ,
ρm ∈∆
p
if f ∈ L2 (Rn ) ∩ Lp (Rn ). This together with the first step concludes the proof
of Theorem 6.26. 

6.5 The Marcinkiewicz multiplier theorem

We now present another multiplier theorem which is one of the most im-
portant results of the whole theory. For the sake of clarity, we state first the
one-dimensional case.
Theorem 6.28. Let m be a bounded function on R1 , which is of bounded
variation on every finite interval not containing the origin. Suppose
(a) |m(ξ)|
R 6 B, −∞ < ξ < ∞,
(b) I |m(ξ)|dξ 6 B, for every dyadic interval I.
Then m ∈ Mp , 1 < p < ∞; and more precisely, if f ∈ L2 ∩ Lp ,
kTm f kp 6 Ap kf kp ,
where Ap depends only on B and p.
To present general theorem, we consider R as divided into its two half-lines,
R as divided into its four quadrants, and generally Rn as divided into its 2n
2

“octants”. Thus, the first octants in Rn will be the open “rectangle” of those
ξ all of whose coordinates are strictly positive. We shall assume that m(ξ) is
defined on each such octant and is there continuous together with its partial
derivatives up to and including order n. Thus m may be left undefined on the
set of points where one or more coordinate variables vanishes.
For every k 6 n, we regard Rk embedded in Rn in the following obvious
way: Rk is the subspace of all points of the form (ξ1 , ξ2 , · · · , ξk , 0, · · · , 0).
Theorem 6.29 (Marcinkiewicz’ multiplier theorem). Let m be a bounded
function on Rn that is C n in all 2n “octant”. Suppose also
6.5. The Marcinkiewicz multiplier theorem - 165 -

(a) |m(ξ)| 6 B,
(b) for each 0 < k 6 n,
∂km
Z
sup dξ1 · · · dξk 6 B
ξk+1 ,··· ,ξn ρ ∂ξ1 ∂ξ2 · · · ∂ξk
as ρ ranges over dyadic rectangles of Rk . (If k = n, the “sup” sign is omitted.)
(c) The condition analogous to (b) is valid for every one of the n! permuta-
tions of the variables ξ1 , ξ2 , · · · , ξn .
Then m ∈ Mp , 1 < p < ∞; and more precisely, if f ∈ L2 ∩ Lp , kTm f kp 6
Ap kf kp , where Ap depends only on B, p and n.

Proof. It will be best to prove Theorem 6.29 in the case n = 2. This case is
already completely typical of the general situation, and in doing only it we can
avoid some notational complications.
Let f ∈ L2 (R2 ) ∩ Lp (R2 ), and write F = Tm f , that is F (F (x)) =
m(ξ)fˆ(ξ).
Let ∆ denote the dyadic rectangles, and for each ρ ∈ ∆, write fρ = Sρ f ,
Fρ = Sρ F , thus Fρ = Tm fρ .
In view of Theorem 6.26, it suffices to show that
X 1/2 X 1/2
2 2
|Fρ | 6 Cp |fρ | . (6.73)
p p
ρ∈∆ ρ∈∆
The rectangles in ∆ come in four sets, those in the first, the second, the third,
and fourth quadrants, respectively. In estimating the l.h.s. of (6.73), consider
the rectangles of each quadrant separately, and assume from now on that our
rectangles belong to the first quadrant.
We will express Fρ in terms of an integral involving fρ and the partial sum
operators. That this is possible is the essential idea of the proof.
Fix ρ and assume ρ = {(ξ1 , ξ2 ) : 2k 6 ξ1 6 2k+1 , 2l 6 ξ2 6 2l+1 }. Then,
for (ξ1 , ξ2 ) ∈ ρ, it is easy to verify the identity
Z ξ2Z ξ1 2 Z ξ1
∂ m(t1 , t2 ) ∂
m(ξ1 , ξ2 ) = dt1 dt2 + m(t1 , 2l )dt1
2 l 2 k ∂t 1 ∂t 2 2k ∂t 1
Z ξ2
∂
+ m(2k , t2 )dt2 + m(2k , 2l ).
2l ∂t 2
Now let St denote the multiplier transform corresponding to the rectangle
(1)
{(ξ1 , ξ2 ) : 2k+1 > ξ1 > t1 , 2l+1 > ξ2 > t2 }. Similarly, let St1 denote the
(2)
multiplier corresponding to the interval 2k+1 > ξ1 > t1 , similarly for St2 .
(1) (2)
Thus in fact, St = St1 · St2 . Multiplying both sides of the above equation by
the function χρ fˆ and taking inverse Fourier transforms yields, by changing the
order of integrals in view of Fubini’s theorem and the fact that Sρ Tm f = Fρ ,
(1) (1) (2) (2)
and St1 Sρ = St1 , St2 Sρ = St2 , St Sρ = St , we have
- 166 - 6. The Littlewood-Paley g-function and Multipliers

Fρ =Tm Sρ f = F −1 mχρ fˆ
 n Z h Z ξ2 Z ξ1 ∂ 2 m(t , t )
|ω| 1 2
i
= e ωix·ξ
dt1 dt2 χρ (ξ)fˆ(ξ) dξ
2π R2 2l 2k ∂t1 ∂t2
 n Z Z ξ1
|ω| h ∂ i
+ eωix·ξ m(t1 , 2l )dt1 χρ (ξ)fˆ(ξ) dξ
2π R2 2k ∂t1
 n Z Z ξ2
|ω| h ∂ i
+ eωix·ξ m(2k , t2 )dt2 χρ (ξ)fˆ(ξ) dξ
2π R2 2l ∂t2
−1 k l
+ F m(2 , 2 )χρ (ξ)f (ξ) ˆ
 n Z Z 2l+1 Z 2k+1 2
|ω| ωix·ξ ∂ m(t1 , t2 )
= e χ[2k ,ξ1 ] (t1 )χ[2l ,ξ2 ] (t2 )dt1 dt2
2π R2 2l 2k ∂t1 ∂t2
· χρ (ξ)fˆ(ξ)dξ
 n Z Z 2k+1
|ω| ∂
+ e ωix·ξ
m(t1 , 2l )χ[2k ,ξ1 ] (t1 )dt1 χρ (ξ)fˆ(ξ)dξ
2π R 2 2k ∂t 1
 n Z Z 2l+1
|ω| ∂
+ eωix·ξ m(2k , t2 )χ[2l ,ξ2 ] (t2 )dt2 χρ (ξ)fˆ(ξ)dξ
2π R 2 2l ∂t 2
k l
+ m(2 , 2 )fρ
 n Z 2l+1 Z 2k+1
|ω|
Z
= eωix·ξ χ[t1 ,2k+1 ] (ξ1 )χ[t2 ,2l+1 ] (ξ2 )χρ (ξ)fˆ(ξ)dξ
2π 2l 2k R2
2
∂ m(t1 , t2 )
· dt1 dt2
∂t1 ∂t2
 n Z 2k+1 Z
|ω| ∂
+ eωix·ξ χ[t1 ,2k+1 ] (ξ1 )χρ (ξ)fˆ(ξ)dξ m(t1 , 2l )dt1
2π 2k R2 ∂t 1
 n Z 2l+1 Z
|ω| ∂
+ eωix·ξ χ[t2 ,2l+1 ] (ξ2 )χρ (ξ)fˆ(ξ)dξ m(2k , t2 )dt2
2π 2l R 2 ∂t 2
k l
+ m(2 , 2 )fρ
Z 2k+1
∂ 2 m(t1 , t2 ) ∂
Z
(1)
= St fρ dt1 dt2 + St1 fρ m(t1 , 2l )dt1
ρ ∂t 1 ∂t 2 2 k ∂t 1
Z 2l+1
(2) ∂
+ St2 fρ m(2k , t2 )dt2 + m(2k , 2l )fρ .
2 l ∂t 2
We apply the Cauchy-Schwarz inequality in the first three terms of the above
w.r.t. the measures |∂t1 ∂t2 m(t1 , t2 )|dt1 dt2 , |∂t1 m(t1 , 2l )|dt1 , |∂t2 m(2k , t2 )|dt2 ,
respectively, and we use the assumptions of the theorem to deduce
2
Z
2 ∂ 2m  Z ∂ 2 m 
|Fρ | . |St fρ | dt1 dt2 dt1 dt2
ρ ∂t1 ∂t2 ρ ∂t1 ∂t2
6.5. The Marcinkiewicz multiplier theorem - 167 -

2k+1  Z 2k+1 ∂
Z
(1) ∂ 
+ |St1 fρ |2 l
m(t1 , 2 ) dt1 m(t1 , 2l ) dt1
2k ∂t1 2k ∂t1
 Z 2l+1 ∂  Z 2l+1
∂ 
(2) 2 k k
+ |St2 fρ | m(2 , t2 ) dt2 m(2 , t2 ) dt2
2l ∂t2 2l ∂t2
+ |m(2k , 2l )|2 |fρ |2
∂ 2m ∂m(t1 , 2l )
Z Z
0 2 (1)
6B |St fρ | dt1 dt2 + |St1 fρ |2 dt1
ρ ∂t1 ∂t2 I1 ∂t1
k

2 ∂m(2 , t2 )
Z
(2) 2
+ |St2 fρ | dt2 + |fρ |
I2 ∂t2
==1ρ + =2ρ + =3ρ + =4ρ , with ρ = I1 × I2 .
To estimate k( ρ |Fρ |2 )1/2 kp , we estimate separately the contributions of each
P

of the four terms on the r.h.s. of the above inequality by the use of Theorem
6.25. To apply that theorem in the case of =1ρ we take for Γ the first quadrant,
2 m(t ,t )
and dγ = | ∂ ∂t 1 2
1 ∂t2
|dt1 dt2 , the functions γ → ργ are constant on the dyadic
rectangles. Since for every rectangle,
∂ 2 m(t1 , t2 )
Z Z
dγ = dt1 dt2 6 B,
ρ ρ ∂t1 ∂t2
then
!1/2 !1/2
X X
1 2
|=ρ | 6 Cp |fρ | .
ρ ρ
p p
Similarly, for =2ρ , =3ρ and =4ρ , which concludes the proof. 
L ECTURE N OTES ON c 2016 by Chengchun Hao
Introduction to Harmonic Analysis Email: hcc@amss.ac.cn

Chapter 7
Sobolev Spaces

7.1 Riesz potentials and fractional integrals

Let f be a sufficiently smooth function which is small at infinity, then the

Fourier transform of its Laplacean ∆f is
F (−∆f )(ξ) = ω 2 |ξ|2 fˆ(ξ). (7.1)
From this, we replace the exponent 2 in |ξ|2 by a general exponent s, and
thus to define (at least formally) the fractional power of the Laplacean by
(−∆)s/2 f = F −1 ((|ω||ξ|)s fˆ(ξ)). (7.2)
Of special significance will be the negative powers s in the range −n < s <
0. In general, with a slight change of notation, we can define
Definition 7.1. Let s > 0. The Riesz potential of order s is the operator
Is = (−∆)−s/2 . (7.3)
For 0 < s < n, Is is actually given in the form
1
Z
Is f (x) = |x − y|−n+s f (y)dy, (7.4)
γ(s) Rn
with
π n/2 2s Γ (s/2)
γ(s) = .
Γ ((n − s)/2)
The formal manipulations have a precise meaning.
Lemma 7.2. Let 0 < s < n.
(a) The Fourier transform of the function |x|−n+s is the function
γ(s)(|ω||ξ|)−s , in the sense that
Z Z
−n+s
|x| ϕ(x)dx = γ(s)(|ω||ξ|)−s ϕ̂(ξ)dξ, (7.5)
Rn Rn
whenever ϕ ∈ S .
(b) The identity F (Is f ) = (|ω||ξ|)−s fˆ(ξ) holds in the sense that

169
- 170 - 7. Sobolev Spaces
Z Z
Is f (x)g(x)dx = fˆ(ξ)(|ω||ξ|)−s ĝ(ξ)dξ
Rn Rn
whenever f, g ∈ S .

Proof. Part (a) is merely a restatement of Lemma 5.14 since γ(s) = |ω|s γ0,s .
Part (b) follows immediately from part (a) by writing
1
Z Z
Is f (x) = f (x − y)|y|−n+s
dy = (|ω||ξ|)−s f \
(x − ·)dξ
γ(s) Rn Rn
Z Z
−s ˆ
= (|ω||ξ|) f (ξ)e ωiξ·x
dξ = (|ω||ξ|)−s fˆ(ξ)e−ωiξ·x dξ,
Rn Rn
so Z Z Z
Is f (x)g(x)dx = (|ω||ξ|)−s fˆ(ξ)e−ωiξ·x dξg(x)dx
Rn ZR
n Rn

= (|ω||ξ|)−s fˆ(ξ)ĝ(ξ)dξ.
Rn
This completes the proof. 
Now, we state two further identities which can be obtained from Lemma 7.2
and which reflect essential properties of the potentials Is .
Is (It f ) = Is+t f, f ∈ S , s, t > 0, s + t < n. (7.6)
∆(Is f ) = Is (∆f ) = −Is−2 f, f ∈ S , n > 3, 2 6 s 6 n. (7.7)
The deduction of these two identities have no real difficulties, and these are
best left to the interested reader to work out.
A simple consequence of (7.6) is the n-dimensional variant of the beta func-
tion,1
γ(s)γ(t) −n+(s+t)
Z
|x − y|−n+s |y|−n+t dy = |x| (7.8)
Rn γ(s + t)
with s, t > 0 and s + t < n. Indeed, for any ϕ ∈ S , we have, by the definition
of Riesz potentials and (7.6), that
ZZ
|x − y|−n+s |y|−n+t dyϕ(z − x)dx
n n
Z R ×R Z
= |y|−n+t |x − y|−n+s ϕ(z − y − (x − y))dxdy
n Rn
ZR
= |y|−n+t γ(s)Is ϕ(z − y)dy = γ(s)γ(t)It (Is ϕ)(z) = γ(s)γ(t)Is+t ϕ(z)
Rn

γ(s)γ(t)
Z
= |x|−n+(s+t) ϕ(z − x)dx.
γ(s + t) Rn
By the arbitrariness of ϕ, we have the desired result.

1
The beta function, also called the Euler integral of the first kind, is a special function defined
by B(x, y) = 01 tx−1 (1 − t)y−1 dt for <x > 0 and <y > 0. It has the relation with Γ -function:
R

B(x, y) = Γ (x)Γ (y)/Γ (x + y).

7.1. Riesz potentials and fractional integrals - 171 -

We have considered the Riesz potentials formally and the operation for
Schwartz functions. But since the Riesz potentials are integral operators, it
is natural to inquire about their actions on the spaces Lp (Rn ).
For this reason, we formulate the following problem. Given s ∈ (0, n), for
what pairs p and q, is the operator f → Is f bounded from Lp (Rn ) to Lq (Rn )?
That is, when do we have the inequality
kIs f kq 6 Akf kp ? (7.9)
There is a simple necessary condition, which is merely a reflection of the
homogeneity of the kernel (γ(s))−1 |y|−n+s . In fact, we have
Proposition 7.3. If the inequality (7.9) holds for all f ∈ S and a finite
constant A, then 1/q = 1/p − s/n.

Proof. Let us consider the dilation operator δε , defined by δε f (x) = f (εx) for
ε > 0. Then clearly, for ε > 0
1
Z
(δε−1 Is δε f )(x) = |ε−1 x − y|−n+s f (εy)dy
γ(s) Rn
z=εy −n 1
Z
==ε |ε−1 (x − z)|−n+s f (z)dz
γ(s) Rn
−s
=ε Is f (x). (7.10)
Also
kδε f kp = ε−n/p kf kp , kδε−1 Is f kq = εn/q kIs f kq . (7.11)
Thus, by (7.9)
kIs f kq =εs kδε−1 Is δε f kq = εs+n/q kIs δε f kq
6Aεs+n/q kδε f kp = Aεs+n/q−n/p kf kp .
If kIs f kq 6= 0, then the above inequality implies
1/q = 1/p − s/n. (7.12)
If f 6= 0 is non-negative, then Is f > 0 everywhere and hence kIs f kq > 0, and
we can conclude the desired relations. 
Next, we observe that the inequality must fail at the endpoints p = 1 (then
q = n/(n − s)) and q = ∞ (then p = n/s).
Let us consider the case p = 1. It is not hard to see that the presumed
inequality
kIs f kn/(n−s) 6 Akf k1 , (7.13)
Rcannot hold. In fact, we can choose a nice positive −n
function ϕ ∈ L1 with
ϕ = 1 and a compact support. Then, with ϕε (x) = ε ϕ(x/ε), we have that
as ε → 0+ ,
Is (ϕε )(x) → (γ(s))−1 |x|−n+s .
- 172 - 7. Sobolev Spaces

If kIs ϕε kn/(n−s) 6 Akϕε k1 = A were valid uniformly as ε, then Fatou’s

lemma2 will imply that
Z
|x|−n dx < ∞,
Rn
and this is a contradiction.
The second atypical case occurs when q = ∞. Again the inequality of the
type (7.9) cannot hold, and one immediate reason is that this case is dual to the
case p = 1 just considered. The failure at q = ∞ may also be seen directly as
follows. Let f (x) = |x|−s (ln 1/|x|)−(1+ε)s/n , for |x| 6 1/2, and f (x) = 0, for
n/s
|x| > 1/2, where ε is positive but small. Then f ∈ Ln/s (Rn ), since kf kn/s =
|x|−n (ln 1/|x|)−1−ε dx < ∞. However, Is f is essentially unbounded
R
|x|61/2
near the origin since
1
Z
Is f (0) = |x|−n (ln 1/|x|)−(1+ε)s/n dx = ∞,
γ(s) |x|61/2
as long as (1 + ε)s/n 6 1.
After these observations, we can formulate the following Hardy-Littlewood-
Sobolev theorem of fractional integration. The result was first considered in
one dimension on the circle by Hardy and Littlewood. The n-dimensional re-
sult was considered by Sobolev.
Theorem 7.4 (Hardy-Littlewood-Sobolev theorem of fractional integra-
tions). Let 0 < s < n, 1 6 p < q < ∞, 1/q = 1/p − s/n.
(a) If f ∈ Lp (Rn ), then the integral (7.4), defining Is f , converges absolutely
for almost every x.
(b) If, in addition, p > 1, then kIs f kq 6 Ap,q kf kp .
(c) If f ∈ L1 (Rn ), then m{x : |Is f (x)| > α} 6 (Aα−1 kf k1 )q , for all α >
0. That is, the mapping f → Is f is of weak type (1, q), with 1/q = 1 − s/n.

Proof. We first prove parts (a) and (b). Let us write

Z Z
−n+s
γ(s)Is f (x) = |x − y| f (y)dy + |x − y|−n+s f (y)dy
B(x,δ) Rn \B(x,δ)

=:Lδ (x) + Hδ (x).

Divide the ball B(x, δ) into the shells Ej := B(x, 2−j δ) \ B(x, 2−(j+1) δ),
j = 0, 1, 2, ..., thus
X∞ Z ∞ Z
X
−n+s
|Lδ (x)| 6 |x − y| f (y)dy 6 |x − y|−n+s |f (y)|dy
j=0 Ej j=0 Ej

2
Fatou’s lamma: If {fk } is a sequence of nonnegative measurable functions, then
Z Z
lim inf fk dµ 6 lim inf fk dµ.
k→∞ k→∞
7.1. Riesz potentials and fractional integrals - 173 -

∞ Z
X
6 (2−(j+1) δ)−n+s |f (y)|dy
j=0 Ej
X∞ Z
6 (2−(j+1) δ)−n+s |f (y)|dy
j=0 B(x,2−j δ)
∞
(2−(j+1) δ)−n+s m(B(x, 2−j δ))
X Z
= |f (y)|dy
j=0
m(B(x, 2−j δ)) B(x,2−j δ)
∞
(2−(j+1) δ)−n+s Vn (2−j δ)n
X Z
= |f (y)|dy
j=0
m(B(x, 2−j δ)) B(x,2−j δ)
∞
X Vn δ s 2n
s n−s
6Vn δ 2 2−sj M f (x) = M f (x).
j=0
2s − 1
Now, we derive an estimate for Hδ (x). By Hölder’s inequality and the con-
dition 1/p > s/n (i.e., q < ∞), we obtain
Z 1/p0
(−n+s)p0
|Hδ (x)| 6kf kp |x − y| dy
Rn \B(x,δ)
Z Z ∞ 1/p0
(−n+s)p0 n−1
=kf kp r r drdσ
S n−1 δ
Z ∞ 1/p0
1/p0 (−n+s)p0 +n−1
=ωn−1 kf kp r dr
δ
 1/p0
ωn−1 0
= 0
δ n/p −(n−s) kf kp = C(n, s, p)δ s−n/p kf kp .
(n − s)p − n
By the above two inequalities, we have
|γ(s)Is f (x)| 6 C(n, s)δ s M f (x) + C(n, s, p)δ s−n/p kf kp =: F (δ).
Choose δ = C(n, s, p)[kf kp /M f ]p/n , such that the two terms of the r.h.s. of
the above are equal, i.e., the minimizer of F (δ), to get
|γ(s)Is f (x)| 6 C(M f )1−ps/n kf kps/n
p .
Therefore, by part (i) of Theorem 3.9 for maximal functions, i.e., M f is
finite almost everywhere if f ∈ Lp (1 6 p 6 ∞), it follows that |Is f (x)| is
finite almost everywhere, which proves part (a) of the theorem.
By part (iii) of Theorem 3.9, we know kM f kp 6 Ap kf kp (1 < p 6 ∞),
thus
kIs f kq 6 CkM f k1−ps/n
p kf kps/n
p = Ckf kp .
This gives the proof of part (b).
Finally, we prove (c). Since we also have |Hδ (x)| 6 kf k1 δ −n+s , taking
α = kf k1 δ −n+s , i.e., δ = (kf k1 /α)1/(n−s) , by part (ii) of Theorem 3.9, we get
m{x : |Is f (x)| > 2(γ(s))−1 α}
- 174 - 7. Sobolev Spaces

6m{x : |Lδ (x)| > α} + m{x : |Hδ (x)| > α}

6m{x : |Cδ s M f (x)| > α} + 0
C
6 −s kf k1 = C[kf k1 /α]n/(n−s) = C[kf k1 /α]q .
δ α
This completes the proof of part (c). 

7.2 Bessel potentials

While the behavior of the kernel (γ(s))−1 |x|−n+s as |x| → 0 is well suited
for their smoothing properties, their decay as |x| → ∞ gets worse as s in-
creases.
We can slightly adjust the Riesz potentials such that we maintain their es-
sential behavior near zero but achieve exponential decay at infinity. The sim-
plest way to achieve this is by replacing the “nonnegative” operator −∆ by
the “strictly positive” operator I − ∆, where I = identity. Here the terms non-
negative and strictly positive, as one may have surmised, refer to the Fourier
transforms of these expressions.
Definition 7.5. Let s > 0. The Bessel potential of order s is the operator
Js = (I − ∆)−s/2
whose action on functions is given by
Js f = F −1 G
cs F f = Gs ∗ f,
where
Gs (x) = F −1 ((1 + ω 2 |ξ|2 )−s/2 )(x).
Now we give some properties of Gs (x) and show why this adjustment yields
exponential decay for Gs at infinity.
Proposition 7.6. Let s > 0.
R∞ |x|2 s−n
(a) Gs (x) = (4π)n/21Γ (s/2) 0 e−t e− 4t t 2 dtt .
(b) Gs (x) > 0, ∀x ∈ Rn ; and Gs (x) ∈ L1 (Rn ), precisely, Rn Gs (x)dx =
R

1.
(c) There exist two constants 0 < C(s, n), c(s, n) < ∞ such that
Gs (x) 6 C(s, n)e−|x|/2 , when |x| > 2,
and such that
1 Gs (x)
6 6 c(s, n), when |x| 6 2,
c(s, n) Hs (x)
where Hs is a function that satisfies
7.2. Bessel potentials - 175 -
 s−n
 |x| + 1 + O(|x|s−n+2 ), 0 < s < n,
2
Hs (x) = ln |x| + 1 + O(|x|2 ), s = n,
1 + O(|x|s−n ), s > n,


as |x| → 0.
0
(d) Gs (x) ∈ Lp (Rn ) for any 1 6 p 6 ∞ and s > n/p.

Proof. (a) For A, s > 0, we have the Γ -function identity

Z ∞
−s/2 1 dt
A = e−tA ts/2 ,
Γ (s/2) 0 t
which we use to obtain Z ∞
2 −s/2 1 2 dt
2
(1 + ω |ξ| ) = e−t e−t|ωξ| ts/2 .
Γ (s/2) 0 t
Note that the above integral converges at both ends (as |ξ| → 0, or ∞). Now
take the inverse Fourier transform in ξ and use Theorem 1.10 to obtain
Z ∞
1 −1 2 dt
Gs (x) = Fξ e−t e−t|ωξ| ts/2
Γ (s/2) t
Z ∞ 0
1  2
 dt
= e−t Fξ−1 e−t|ωξ| ts/2
Γ (s/2) 0 t
Z ∞
1 |x| 2
s−n dt
= n/2
e−t e− 4t t 2 .
(4π) Γ (s/2) 0 t
(b) We have easily3 Rn Gs (x)dx = F Gs (0) = 1. Thus, Gs ∈ L1 (Rn ).
R
2 |x|2
(c) First, we suppose |x| > 2. Then t + |x| 4t
> t + 1
t
and also t + 4t
> |x|.
This implies that
|x|2 t 1 |x|
−t − 6− − − ,
4t 2 2t 2
from which it follows that when |x| > 2
Z ∞
1 1 s−n dt
− 2t − 2t − |x| − |x|
Gs (x) 6 e e t 2 e 2 6 C(s, n)e 2 ,
(4π)n/2 Γ (s/2) 0 t
2|s−n|/2 Γ (|s−n|/2)
where C(s, n) = (4π)n/2 Γ (s/2)
for s 6= n, and C(s, n) = 4
(4π)n/2 Γ (s/2)
for
s = n since

2
3
Or use (a) to show it. From part (a), we know Gs (x) > 0. Since Rn e−π|x| /t dx = tn/2 , by
R

Fubini’s theorem, we have

Z ∞
1 s−n dt
Z Z
|x|2
Gs (x)dx = n/2
e−t e− 4t t 2 dx
Rn Rn (4π) Γ (s/2) 0 t
Z ∞
1 s−n dt
Z
|x|2
= n/2
e−t e− 4t dxt 2
(4π) Γ (s/2) 0 Rn t
Z ∞
1 −t n/2
s−n dt
= e (4πt) t 2
(4π)n/2 Γ (s/2) 0 t
Z ∞
1 −t 2s −1
= e t dt = 1.
Γ (s/2) 0
- 176 - 7. Sobolev Spaces

∞ 1 ∞ ∞
dt dt dy
Z Z Z Z
− 2t − 2t
1 1
− 2t − 2t
e e 6 e + e dt = e−y + 2e−1/2
0 t 0 t 1 1/2 y
Z ∞
62 e−y dy + 2 6 4.
1/2
Next, suppose that |x| 6 2. Write Gs (x) = G1s (x) + G2s (x) + G3s (x), where
Z |x|2
1 1 −t − |x|
2
s−n dt
Gs (x) = n/2
e e 4t t 2 ,
(4π) Γ (s/2) 0 t
Z 4
2 1 −t − |x|
2
s−n dt
Gs (x) = e e 4t t 2 ,
(4π)n/2 Γ (s/2) |x|2 t
Z ∞
3 1 −t − |x|
2
s−n dt
Gs (x) = n/2
e e 4t t 2 .
(4π) Γ (s/2) 4 t
2
Since t|x|2 6 16 in G1s , we have e−t|x| = 1 + O(t|x|2 ) as |x| → 0; thus after
changing variables, we can write
Z 1
1 2 1 s−n dt
1
Gs (x) =|x| s−n
n/2
e−t|x| e− 4t t 2
(4π) Γ (s/2) 0 t
Z 1 Z 1
1 1 s−n dt
− 4t O(|x|s−n+2 ) 1 s−n
=|x| s−n
n/2
e t 2 + n/2
e− 4t t 2 dt
(4π) Γ (s/2) 0 t (4π) Γ (s/2) 0
s−n Z ∞
2 n−s−2
|x| dy 2 n−s−4
O(|x|s−n+2 ) ∞ −y s−n dy
Z
−y s−n
= e y 2 + e y 2 2
(4π)n/2 Γ (s/2) 1/4 y (4π)n/2 Γ (s/2) 1/4 y
=c1s,n |x|s−n + O(|x|s−n+2 ), as |x| → 0.
|x|2 |x|2
Since 0 6 4t 6 41 and 0 6 t 6 4 in G2s , we have e−17/4 6 e−t− 4t 6 1,
thus as |x| → 0, we obtain
 s−n s−n+1
|x|
Z 4  n−s
− 2 n−s , s < n,
dt 
2
G2s (x) ∼ t(s−n)/2 = 2 ln |x| , s = n,
|x| 2 t 2
 s−n+1
s−n
, s > n.
|x|2
Finally, we have e−1/4 6 e− 4t 6 1 in G3s , which yields that G3s (x) is
bounded above and below by fixed positive constants. Combining the estimates
for Gjs (x), we obtain the desired conclusion.
(d) For p = 1 and so p0 = ∞, by part (c), we have kGs (x)k∞ 6 C for
s > n.
Next, we assume that 1 < p 6 ∞ and so 1 6 p0 < ∞. Again by part (c),
0 0
we have, for |x| > 2, that Gps 6 Ce−p |x|/2 , and then the integration over this
range |x| > 2 is clearly finite.
0
On the range |x| 6 2, it is clear that |x|62 Gps (x)dx 6 C for s > n. For the
R
0
case s = n and n 6= 1, we also have |x|62 Gps (x)dx 6 C by noticing that
R
7.2. Bessel potentials - 177 -
 q Z 2 q
2 2
Z
ln dx = C ln rn−1 dr 6 C
|x|62 |x| 0 r
for
R any q 2> q0 since lim
ε
R 2 r→0 r ln(2/r) =R 0. For the case s = n = 1, we have
1
|x|62
(ln |x| ) dx = 2 0 (ln 2/r) dr = 4 0 (ln 1/r)q dr = 4Γ (q + 1) for q > 0
q
R1
by the formula 0 (ln 1/x)p−1 dx = Γ (p) for <p > 0. For the final case s < n,
R2 0
we have 0 r(s−n)p rn−1 dr 6 C if (s − n)p0 + n > 0, i.e., s > n/p.
Thus, we obtain kGs (x)kp0 6 C for any 1 6 p 6 ∞ and s > n/p, which
implies the desired result. 
We also have a result analogues to that of Riesz potentials for the operator
Js .
Theorem 7.7. (a) For all 0 < s < ∞, the operator Js maps Lr (Rn ) into itself
with norm 1 for all 1 6 r 6 ∞.
(b) Let 0 < s < n and 1 < p < q < ∞ satisfy 1/q = 1/p − s/n. Then there
exists a constant Cn,s,p < ∞ such that for all f ∈ Lp (Rn ), we have
kJs f kq 6 Cn,s,p kf kp .
(c) If f ∈ L1 (Rn ), then m{x : |Js f (x)| > α} 6 (Cn,s α−1 kf k1 )q , for all
α > 0. That is, the mapping f → Js f is of weak type (1, q), with 1/q =
1 − s/n.
Proof. By Young’s inequality, we have kJs f kr = kGs ∗ f kr 6 kGs k1 kf kr =
kf kr . This proves the result (a).
In the special case 0 < s < n, we have, from the above proposition, that the
kernel Gs of Js satisfies
 −n+s
|x| , |x| 6 2,
Gs (x) ∼ −|x|/2
e , |x| > 2.
Then, we can write
Z Z 
−n+s −|y|/2
Js f (x) 6Cn,s |f (x − y)||y| dy + |f (x − y)|e dy
|y|62 |y|>2
 Z 
−|y|/2
6Cn,s Is (|f |)(x) + |f (x − y)|e dy .
Rn
We now use that the function e ∈ Lr for all 1 6 r 6 ∞, Young’s
−|y|/2

inequality and Theorem 7.4 to complete the proofs of (b) and (c). 
The affinity between the two potentials is given precisely in the following
lemma.
Lemma 7.8. Let s > 0.
(i) There exists a finite measure µs on Rn such that its Fourier transform µbs
is given by
|ωξ|s
µbs (ξ) = .
(1 + |ωξ|2 )s/2
- 178 - 7. Sobolev Spaces

(ii) There exist a pair of finite measures νs and λs on Rn such that

(1 + |ωξ|2 )s/2 = νbs (ξ) + |ωξ|s λbs (ξ).
Remark 7.9. 1) The first part states in effect that the following formal
quotient operator is bounded on every Lp (Rn ), 1 6 p 6 ∞,
(−∆)s/2
, s > 0. (7.14)
(I − ∆)s/2
2) The second part states also to what extent the same thing is true of
the operator inverse to (7.14).
Proof. To prove (i), we use the Taylor expansion
X∞
s/2
(1 − t) = 1 + Am,s tm , |t| < 1, (7.15)
m=1
s s
( −1)···( 2s −m+1) (− 2s )(1− 2s )·(m− 2s −1)
where Am,s = (−1)m Cs/2m
= (−1)m 2 2 m! P
= m!
. All
the Am,s are of same sign for m > 2 + 1, so |Am,s | < ∞, since (1 − t)s/2
s

remains bounded as t → 1, if s > 0. Let t = (1 + |ωξ| ) . Then

2 −1
s/2 ∞
|ωξ|2
 X
=1+ Am,s (1 + |ωξ|2 )−m . (7.16)
1 + |ωξ|2 m=1
However, G2m (x) > 0 andR Rn G2m (x)e−ωix·ξ dx = (1 + |ωξ|2 )−m .
R

We noticed already that G2m (x)dx = 1 and so kG2m k1 = 1.

Thus from the convergence of |Am,s |, it follows that if µs is defined by
P
∞
!
X
µ s = δ0 + Am,s G2m (x) dx (7.17)
m=1
with δ0 the Dirac measure at the origin, then µs represents a finite measure.
Moreover, by (7.16),
|ωξ|s
µbs (ξ) = . (7.18)
(1 + |ωξ|2 )s/2
For (ii), we now invoke the n-dimensional version of Wiener’s theorem,
to wit: If Φ1 ∈ L1 (Rn ) and Φ c1 (ξ) + 1 is nowhere zero, then there exists a
Φ2 ∈ L (R ) such that (Φ
1 n c1 (ξ) + 1)−1 = Φ c2 (ξ) + 1.
For our purposes, we then write
X∞
Φ1 (x) = Am,s G2m (x) + Gs (x).
m=1
Then, by (7.18), we see that
|ωξ|s + 1
Φ
c1 (ξ) + 1 = ,
(1 + |ωξ|2 )s/2
which vanishes nowhere. Thus, for an appropriate Φ2 ∈ L1 , by Wiener’s theo-
rem, we have
7.3. Sobolev spaces - 179 -

(1 + |ωξ|2 )s/2 = (1 + |ωξ|s )[Φ

c2 (ξ) + 1],
and so we obtain the desired conclusion with νs = λs = δ0 + Φ2 (x)dx. 

7.3 Sobolev spaces

We start by weakening the notation of partial derivatives by the theory of

distributions. The appropriate definition is stated in terms of the space D(Rn ).
Let ∂ α be a differential monomial, whose total order is |α|. Suppose we
are given two locally integrable functions on Rn , f and g. Then we say that
∂ α f = g (in the weak sense), if
Z Z
α |α|
f (x)∂ ϕ(x)dx = (−1) g(x)ϕ(x)dx, ∀ϕ ∈ D. (7.19)
Rn Rn
Integration by parts shows us that this is indeed the relation that we would
expect if f had continuous partial derivatives up to order |α|, and ∂ α f = g had
the usual meaning.
Of course, it is not true that every locally integrable function has partial
n
derivatives in this sense: consider, for example, f (x) = ci/|x| . However, when
the partial derivatives exist, they are determined almost everywhere by the
defining relation (7.19).
In this section, we study a quantitative way of measuring smoothness
of functions. Sobolev spaces serve exactly this purpose. They measure the
smoothness of a given function in terms of the integrability of its derivatives.
We begin with the classical definition of Sobolev spaces.
Definition 7.10. Let k be a nonnegative integer and let 1 6 p 6 ∞. The
Sobolev space W k,p (Rn ) is defined as the space of functions f in Lp (Rn ) all
of whose distributional derivatives ∂ α f are also in Lp (Rn ) for all multi-
indices α that satisfies |α| 6 k. This space is normed by the expression
X
kf kW k,p = k∂ α f kp , (7.20)
|α|6k
(0,...,0)
where ∂ f = f.
The index k indicates the “degree” of smoothness of a given function in
W k,p . As k increases, the functions become smoother. Equivalently, these
spaces form a decreasing sequence
Lp ⊃ W 1,p ⊃ W 2,p ⊃ · · ·
meaning that each W k+1,p (Rn ) is a subspace of W k,p (Rn ) in view of the
Sobolev norms.
We next observe that the space W k,p (Rn ) is complete. Indeed, if {fm } is a
Cauchy sequence in W k,p , then for each α, {∂ α fm } is a Cauchy sequence in
- 180 - 7. Sobolev Spaces

Lp , |α| 6 k. By the completeness of Lp , there exist functions f (α) such that

f (α) = limm ∂ α fm in Lp , then clearly
Z Z Z
|α| α α
(−1) fm ∂ ϕdx = ∂ fm ϕdx → f (α) ϕdx,
Rn Rn Rn
for each ϕ ∈ D. Since the first expression converges to
Z
|α|
(−1) f ∂ α ϕdx,
Rn
it follows that the distributional derivative ∂ α f is f (α) . This implies that fj →
f in W k,p (Rn ) and proves the completeness of this space.
First, we generalize Riesz and Bessel potentials to any s ∈ R by
I s f =F −1 |ωξ|s F f, f ∈ S 0 (Rn ), 0 ∈ / supp fˆ,
J s f =F −1 (1 + |ωξ|2 )s/2 F f, f ∈ S 0 (Rn ).
It is clear that I −s = Is and J −s = Js for s > 0 are exactly Riesz and Bessel
potentials, respectively. we also note that J s · J t = J s+t for any s, t ∈ R from
the definition.
Next, we shall extend the spaces W k,p (Rn ) to the case where the number k
is real.
Definition 7.11. Let s ∈ R and 1 6 p 6 ∞. We write
kf kḢps = kI s f kp , kf kHps = kJ s f kp .
Then, the homogeneous Sobolev space Ḣps (Rn ) is defined by
n o
Ḣps (Rn ) = f ∈ S 0 (Rn ) : fˆ ∈ L1loc (Rn ), and kf kḢps < ∞ , (7.21)
The nonhomogeneous Sobolev space Hps (Rn ) is defined by
Hps (Rn ) = f ∈ S 0 (Rn ) : kf kHps < ∞ .

(7.22)
If p = 2, we denote Ḣ2s (Rn ) by Ḣ s (Rn ) and H2s (Rn ) by H s (Rn ) for sim-
plicity.
It is clear that the space Hps (Rn ) is a normed linear space with the above
norm. Moreover, it is complete and therefore Banach space. To prove the com-
pleteness, let {fm } be a Cauchy sequence in Hps . Then, by the completeness
of Lp , there exists a g ∈ Lp such that
kfm − J −s gkHps = kJ s fm − gkp → 0, as m → ∞.
Clearly, J g ∈ S and thus
−s 0
is complete.
Hps
We give some elementary results about Sobolev spaces.
Theorem 7.12. Let s ∈ R and 1 6 p 6 ∞, then we have
(a) S is dense in Hps , 1 6 p < ∞.
(b) Hps+ε ⊂ Hps , ∀ε > 0.
(c) Hps ⊂ L∞ , ∀s > n/p.
7.3. Sobolev spaces - 181 -

(d) Suppose 1 < p < ∞ and s > 1. Then f ∈ Hps (Rn ) if and only if
∂f
f ∈ Hps−1 (Rn ) and for each j, ∂xj
∈ Hps−1 (Rn ). Moreover, the two norms are
equivalent:
n
X ∂f
kf kHps ∼ kf kHps−1 + .
j=1
∂xj Hps−1
(e) Hpk (Rn ) = W k,p (Rn ), 1 < p < ∞, ∀k ∈ N.

Proof. (a) Take f ∈ Hps , i.e., Js f ∈ Lp . Since S is dense in Lp (1 6 p < ∞),

there exists a g ∈ S such that
kf − J −s gkHps = kJ s f − gkp
is smaller than any given positive number. Since J −s g ∈ S , therefore S is
dense in Hps .
(b) Suppose that f ∈ Hps+ε . By part (a) in Theorem 7.7, we see that Jε maps
Lp into Lp with norm 1 for ε > 0. Form this, we get the result since
kf kHps = kJ s f kp = kJ −ε J s+ε f kp = kJε J s+ε f kp 6 kJ s+ε f kp = kf kHps+ε .
(c) By Young’s inequality, the definition of the kernel Gs (x) and part (d) of
Proposition 7.6, we get for s > 0
kf k∞ =kF −1 (1 + |ωξ|2 )−s/2 (1 + |ωξ|2 )s/2 F f k∞
=kF −1 (1 + |ωξ|2 )−s/2 ∗ J s f k∞
6kF −1 (1 + |ωξ|2 )−s/2 kp0 kJ s f kp
=kGs (x)kp0 kf kHps 6 Ckf kHps .
(d) From the Mihlin multiplier theorem, we can get (ωξj )(1 + |ωξ|2 )−1/2 ∈
Mp for 1 < p < ∞ (or use part (i) of Lemma 7.8 and properties of Riesz
transforms), and thus
∂f
=kF −1 (1 + |ωξ|2 )(s−1)/2 (ωiξj )F f kp
∂xj Hps−1
=kF −1 (1 + |ωξ|2 )−1/2 (ωξj )(1 + |ωξ|2 )s/2 F f kp
=kF −1 (1 + |ωξ|2 )−1/2 (ωξj ) ∗ J s f kp 6 CkJ s f kp = Ckf kHps .
Combining with kf kHps−1 6 kf kHps , we get
n
X ∂f
kf kHps−1 + 6 Ckf kHps .
j=1
∂xj Hps−1
Now, we prove the converse inequality. We use the Mihlin multiplier the-
orem once more and an auxiliary function χ on R, infinitely differentiable,
non-negative and with χ(x) = 1 for |x| > 2 and χ(x) = 0 for |x| < 1. We
obtain
- 182 - 7. Sobolev Spaces

n
X
(1 + |ωξ| )2 1/2
(1 + χ(ξj )|ξj |)−1 ∈ Mp , χ(ξj )|ξj |ξj−1 ∈ Mp , 1 < p < ∞.
j=1
Thus,
kf kHps =kJ s f kp = kF −1 (1 + |ωξ|2 )1/2 F J s−1 f kp
Xn
−1
6CkF (1 + χ(ξj )|ξj |)F J s−1 f kp
j=1
n
X ∂f
6Ckf kHps−1 + C kF −1 χ(ξj )|ξj |ξj−1 F J s−1 kp
j=1
∂xj
n
X ∂f
6Ckf kHps−1 + .
j=1
∂xj Hps−1

Thus, we have obtained the desired result.

(e) It is obvious that W 0,p = Hp0 = Lp for k = 0. However, from part (d), if
∂f
k > 1, then f ∈ Hpk if and only if f and ∂x j
∈ Hpk−1 , j = 1, ..., n. Thus, we
can extends the identity of W k,p = Hpk from k = 0 to k = 1, 2, .... 
We continue with the Sobolev embedding theorem.
Theorem 7.13 (Sobolev embedding theorem). Let 1 < p 6 p1 < ∞ and
s, s1 ∈ R. Assume that s − np = s1 − pn1 . Then the following conclusions hold
Hps ⊂ Hps11 , Ḣps ⊂ Ḣps11 .

Proof. It is trivial for the case p = p1 since we also have s = s1 in this case.
Now, we assume that p < p1 . Since p11 = p1 − s−sn
1
, by part (b) of Theorem 7.7,
we get
kf kHps11 = kJ s1 f kp1 = kJ s1 −s J s f kp1 = kJs−s1 J s f kp1 6 CkJ s f kp = Ckf kHps .
Similarly, we can show the homogeneous case. Therefore, we complete the
proof. 

Theorem 7.14. Let s, σ ∈ R and 1 6 p 6 ∞. Then J σ is an isomorphism

between Hps and Hps−σ .

Proof. It is clear from the definition. 

Corollary 7.15. Let s ∈ R and 1 6 p < ∞. Then

(Hps )0 = Hp−s
0 .

0
Proof. It follows from the above theorem and the fact that (Lp )0 = Lp , if
1 6 p < ∞. 
Finally, we give the connection between the homogeneous and the nonho-
mogeneous spaces.
7.3. Sobolev spaces - 183 -

/ supp fˆ. Then

Theorem 7.16. Suppose that f ∈ S 0 (Rn ) and 0 ∈
f ∈ Ḣps ⇔ f ∈ Hps , ∀s ∈ R, 1 6 p 6 ∞.
Moreover, for 1 6 p 6 ∞, we have
Hps =Lp ∩ Ḣps , ∀s > 0,
Hps =Lp + Ḣps , ∀s < 0,
Hp0 =Lp = Ḣp0 .
References - 185 -

References

Abe11. Helmut Abels. Short lecture notes: Interpolation theory and function spaces. May 2011.
Bec75. William Beckner. Inequalities in Fourier analysis on Rn . Proc. Nat. Acad. Sci. U.S.A., 72:638–
641, 1975.
BL76. Jöran Bergh and Jörgen Löfström. Interpolation spaces. An introduction. Springer-Verlag, Berlin,
1976. Grundlehren der Mathematischen Wissenschaften, No. 223.
Din07. Guanggui Ding. New talk of functional analysis. Science press, Beijing, 2007.
Duo01. Javier Duoandikoetxea. Fourier analysis, volume 29 of Graduate Studies in Mathematics. Amer-
ican Mathematical Society, Providence, RI, 2001. Translated and revised from the 1995 Spanish
original by David Cruz-Uribe.
Eva98. Lawrence C. Evans. Partial differential equations, volume 19 of Graduate Studies in Mathematics.
American Mathematical Society, Providence, RI, 1998.
GR. I. S. Gradshteyn and I. M. Ryzhik. Table of integrals, series, and products. Seventh edition.
Gra04. Loukas Grafakos. Classical and modern Fourier analysis. Pearson Education, Inc., Upper Saddle
River, NJ, 2004.
HLP88. G. H. Hardy, J. E. Littlewood, and G. Pólya. Inequalities. Cambridge Mathematical Library.
Cambridge University Press, Cambridge, 1988. Reprint of the 1952 edition.
HW64. Richard A. Hunt and Guido Weiss. The Marcinkiewicz interpolation theorem. Proc. Amer. Math.
Soc., 15:996–998, 1964.
Kri02. Erik Kristiansson. Decreasing rearrangement and Lorentz L(p, q) spaces. Master’s thesis, Luleå
University of Technology, 2002. http://bioinformatics.zool.gu.se/~erikkr/
MasterThesis.pdf.
Mia04. Changxing Miao. Harmonic analysis and its application in PDEs, volume 89 of Fundamental
series in mordon Mathematics. Science Press, Beijing, 2nd edition, 2004.
Rud87. Walter Rudin. Real and complex analysis. McGraw-Hill Book Co., New York, third edition, 1987.
SS03. Elias M. Stein and Rami Shakarchi. Complex analysis. Princeton Lectures in Analysis, II. Prince-
ton University Press, Princeton, NJ, 2003.
Ste70. Elias M. Stein. Singular integrals and differentiability properties of functions. Princeton Mathe-
matical Series, No. 30. Princeton University Press, Princeton, N.J., 1970.
Ste93. Elias M. Stein. Harmonic analysis: real-variable methods, orthogonality, and oscillatory inte-
grals, volume 43 of Princeton Mathematical Series. Princeton University Press, Princeton, NJ,
1993. With the assistance of Timothy S. Murphy, Monographs in Harmonic Analysis, III.
SW71. Elias M. Stein and Guido Weiss. Introduction to Fourier analysis on Euclidean spaces. Princeton
University Press, Princeton, N.J., 1971. Princeton Mathematical Series, No. 32.
WHHG11. Baoxiang Wang, Zhaohui Huo, Chengchun Hao, and Zihua Guo. Harmonic analysis method for
nonlinear evolution equations, volume I. World Scientific Publishing Co. Pte. Ltd., 2011.
Index

Vn : the volume of the unit ball in Rn , 44 entire function, 33

Hk , 121
Lp p
∗ : weak L spaces, 44 Fatou’s lemma, 172
C: complex number field, 2 Fourier inversion theorem, 12
C0 (Rn ), 1 Fourier transform, 2
Γ -function, 44
Hk , 121 Gagliardo-Nirenberg-Sobolev inequality, 68
N0 = N ∪ {0}, 17 Gauss summability, 6
Pk , 120 Gauss-Weierstrass integral, 10
R: real number field, 2 Gauss-Weierstrass kernel, 9
ωn−1 = 2π n/2 /Γ (n/2): surface area of the unit gradient condition, 95
sphere S n−1 , 10 Green theorem, 139
∗
gλ -function, 140
Hörmander condition, 95
Abel method of summability, 6 Hadamard three lines theorem, 35
analytic, 33 Hardy inequality, 51
Hardy-Littlewood maximal function, 61
Bessel potential, 174 Hardy-Littlewood maximal operator, 61
beta function: n-dimensional variant, 170 Hardy-Littlewood-Paley theorem on Rn , 55
Bochner’s relation, 124 Hardy-Littlewood-Sobolev theorem of
fractional integrations, 172
Calderón-Zygmund decomposition harmonic conjugate, 86
for functions, 72 harmonic function, 77, 82
of Rn , 70 Hausdorff-Young inequality, 39
Calderón-Zygmund kernel, 97 heat equation, 14
Calderón-Zygmund singular integral operator, Hecke’s identity, 123
97 Heine-Borel theorem, 64
Calderón-Zygmund Theorem, 94 Heisenberg uncertainty principle, 20
Cauchy-Riemann equations, 86 higher Riesz transforms, 129
generalized ..., 119 Hilbert transform, 89
convolution, 1 Hilbert transform
Cotlar inequality, 108 Characterization, 91
holomorphic, 33
decreasing rearrangement, 45 homogeneous Sobolev spaces Ḣps (Rn ), 180
dilation, 2
dilation argument, 99 Jensen’s inequality, 51
Dini-type condition, 101
distribution function, 41 Laplace equation, 77
dyadic decomposition of Rn , 156 Lebesgue differentiation theorem, 66
Littlewood-Paley g-function, 133
elliptic homogeneous polynomial of degree k, Littlewood-Paley square function theorem, 157
131 locally integrable, 61
187
- 188 - Index

Lorentz space, 49 Rademacher functions, 157

Luzin’s S-function, 140 Riemann-Lebesgue lemma, 4
Riesz potentials, 169
Marcinkiewicz interpolation theorem, 52 Riesz transform, 116
maximal function, 61 Riesz-Thorin interpolation theorem, 36
maximal function theorem, 63
maximum modulus principle, 34 Schwartz space, 17
maximum principle, 34 Sobolev embedding theorem, 182
Mean-value formula for harmonic functions, 81 Sobolev space W k,p (Rn ), 179
Minkowski integral inequality, 2 solid spherical harmonics of degree k, 121
multiplication formula, 9 Stein interpolation theorem, 40
multiplier theorem
Bernstein’s multiplier theorem, 149 Tchebychev inequality, 67
Hörmander’s multiplier theorem, 150 tempered distribution, 21
Marcinkiewicz’ multiplier theorem, 164 The equivalent norm of Lp , 43
Mihlin’s multiplier theorem, 149 translation, 2
translation invariant, 26
truncated operator, 97
nonhomogeneous Sobolev spaces Hps (Rn ), 180
unitary operator, 15
partial g-functions, 134
partial sum operator, 153 Vitali covering lemma, 57
Phragmen-Lindelöf theorem, 34
Plancherel theorem, 15 Weierstrass kernel, 9
Poisson equation, 77 Weighted inequality for Hardy-Littlewood
Poisson integral, 10, 82, 83 maximal function, 73
Poisson kernel, 9, 83 Whitney covering lemma, 58
principal value of 1/x, 88 Wiener’s theorem, 178

quasi-linear mapping, 51 Young’s inequality for convolutions, 39