M3/4P18: Fourier Analysis and Theory of Distributions

Dr. Claude Warnick

March 23, 2016

Abstract

Syllabus:
• Spaces of test functions and distributions,
• Fourier Transform (discrete and continuous),
• Bessel’s, Parseval’s Theorems,
• Laplace transform of a distribution,
• Solution of classical PDE’s via Fourier transform,
• Basic Sobolev Inequalities, Sobolev spaces.

i
 Contents

Introduction iv
A motivational example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv

1 Test functions 1
1.1 The space D(Ω) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 The space E (Ω) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 The space S . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Convolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4.1 Differentiating convolutions . . . . . . . . . . . . . . . . . . . . . . 10
1.4.2 Approximation of the identity . . . . . . . . . . . . . . . . . . . . . 13

2 Distributions 20
2.1 Functions as distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2 Derivatives of distributions . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3 Convergence of distributions . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4 Convolutions and the fundamental solution . . . . . . . . . . . . . . . . . 28
2.4.1 An example: Poisson’s equation . . . . . . . . . . . . . . . . . . . . 33
2.5 Distributions of compact support . . . . . . . . . . . . . . . . . . . . . . . 35
2.6 Tempered distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3 The Fourier Transform 41

3.1 The Fourier transform on L1 (Rn ) . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 The Fourier transform on L2 (Rn ) . . . . . . . . . . . . . . . . . . . . . . . 52
3.3 The Fourier transform on S 0 . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3.1 Convolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.4 The Fourier–Laplace transform on E 0 (Rn ) . . . . . . . . . . . . . . . . . . 65
3.5 Periodic distributions and Poisson’s summation formula . . . . . . . . . . 66

4 Sobolev Spaces and PDE (M) 80

4.1 Sobolev spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.1.1 The spaces W k,p (Ω) . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.1.2 The space H s (Rn ) . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

ii
 Contents iii

4.2 PDE Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4.2.1 Helmholtz equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.2.2 Spaces involving time . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.2.3 The heat equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.2.4 The wave equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

A Some background results 90

A.1 Differentiating functions of several variables . . . . . . . . . . . . . . . . . 90
A.2 Measure theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
A.2.1 Lp −spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
A.2.2 Convergence theorems . . . . . . . . . . . . . . . . . . . . . . . . . 93

B Bonus Material 94
B.1 Topological vector spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
B.1.1 Vector spaces and normed spaces . . . . . . . . . . . . . . . . . . . 94
B.1.2 Topological spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
B.1.3 Topological vector spaces . . . . . . . . . . . . . . . . . . . . . . . 105
B.2 Fréchet spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
B.2.1 Semi-norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
B.3 The test function spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
B.3.1 E (Ω) and DK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
B.3.2 D(Ω) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

Index 127

Acknowledgements

In preparing this course, I have benefitted from the notes of previous lecturers, in particular
Prof. Ari Laptev, Timothy Candy and Salvador Rodriguez-Lopez.
The textbooks that I have found most useful are:

• “Distributions, Theory and Applications”, Duistermaat and Kolk;

• “Introduction to the theory of distributions”, Friedlander and Joshi;

• “Fourier Analysis”, Körner;

• “Functional Analysis”, Rudin;

• “Topology”, Munkres.

Material marked with (M) is examinable only in the mastery question, while material
marked with (U) is not examinable.

Please pass any corrections to c.warnick@imperial.ac.uk.

 Introduction

Fourier analysis is a fundamental part of modern mathematics. Loosely speaking, Fourier

analysis is the study of general functions as a superposition of harmonic ‘waves’ of different
frequencies. It has important applications across mathematics, science and engineering,
including (but not limited to):

- Number theory
- PDE theory
- Probability
- Numerical analysis
- Quantum mechanics
- Fluids
- Radio transmission
- Acoustics
- Optics
- Signal processing

Much modern technology would be impossible without our understanding of Fourier

analysis. In this course we shall study Fourier analysis, but before we do this, we are
first going to need to set up some important analytical machinery, that of distribution
theory. Distributions are a generalisation of the notion of a function. Distribution theory
is an important topic in its own right, and we shall spend quite some time on it. One of
the main motivations for our study will be that distributions are the ‘right’ objects on
which to study Fourier analysis, in the same way that measurable functions are the ‘right’
objects on which to study integration.

A motivational example

In order to give some motivation for why we introduce distributions, let us look at a
simple problem in differential equations. We seek a function ψ ∈ C 2 (R; C) which is 2π

iv
 A motivational example v

periodic1 , ψ(θ + 2π) = ψ(θ) and which satisfies the equation:

d2 ψ
− +ψ =f (1)
dθ2
where f (θ) = f (θ + 2π) is a given function with the same period. Let us suppose that
there exist numbers fn , ψn ∈ C such that
∞
X
f (θ) = fn einθ (2)
n=−∞
X∞
ψ(θ) = ψn einθ (3)
n=−∞

Periodic sums of this type are known as Fourier series. Clearly the sum on the right hand
side has the correct periodicity. It is not a priori obvious that f , ψ can be represented by
such a sum. Assuming however that (2) holds, we can calculate the coefficients fn . To do
this, we note that
( h i2π
ei(n−m)θ)
Z 2π
−imθ inθ i(n−m) 0 = 0 n 6= m
e e dθ =
0 2π n=m
So that
∞
!
Z 2π Z 2π
1 −imθ 1 X
f (θ)e dθ = fn e inθ
e−imθ dθ
2π 0 2π 0 n=−∞
∞  Z 2π 
X 1 −imθ inθ
= fn e e dθ
n=−∞
2π 0

= fm .

In passing from the second line to the third, we have interchanged two limits, which
should ring alarm bells. Let’s suspend our disbelief for the time being and plough on
regardless.
We can insert (2) and (3) into (1), to find that if the equation is satisfied we have:
d2 ψ
0=− +ψ−f
dθ2
∞ ∞ ∞
!
d2 X
inθ
X X
=− 2 ψn e + ψn einθ − fn einθ
dθ n=−∞ n=−∞ n=−∞
∞  2

X d
= −ψn 2 einθ + ψn einθ − fn einθ
n=−∞
dθ
∞
X  2
(n + 1)ψn − fn einθ .

=
n=−∞

1
alternatively, one can think of ψ and f as mapping S 1 → C
vi Introduction

Again, we’ve interchanged limiting operations, but again we’ll ignore this for the time
being. Our previous argument which shows us how to extract the numbers fn from the
function f (θ) suggests that the sum can only vanish if each term separately vanishes, so
we expect that ψ will solve the equation if:
fn
ψn = 2 .
n +1
We hope then that a solution to (1) is given by:
∞ ∞ Z 2π
1 X einθ

X fn inθ −inα
ψ(θ) = e = f (α)e dα
n=−∞
1 + n2 2π n=−∞ 1 + n2 0

making another interchange of limits, we have:

∞
Z 2π !
1 X ein(θ−α)
ψ(θ) = f (α) dα
0 2π n=−∞ 1 + n2
Z 2π
= f (α)G(θ − α)dα,
0
where we introduce a new function in the last line:
∞
1 X einθ
G(θ) := . (4)
2π n=−∞ 1 + n2
By the Weierstrass M -test, the sum defining G converges uniformly to a 2π-periodic
continuous function.In fact, it’s possible to perform this sum explicitly, and we find:
cosh(π − θ)
G(θ) = , 0 ≤ θ < 2π. (5)
2 sinh π
We can plot this function, and it is shown in Figure 1.

Exercise 1.1. Consider the meromorphic function

1 cos [(π − θ)z]
F (z) =
1 + z2 2 sin πz
a) Show that the poles of F (z) occur at z = ±i and z = n, for n ∈ Z, with:
cosh(π − θ) cos(nθ)
Res[F, ±i] = − , Res[F, n] = .
4 sinh π 2π(1 + n2 )
(N )
b) For N ∈ N, let Γi be the curves
 
(N ) 1
Γ1 (t) = N + (1 + it) , −1 ≤ t ≤ 1
2
 
(N ) 1
Γ2 (t) = N + (−t + i) , −1 ≤ t ≤ 1
2
 
(N ) 1
Γ3 (t) = N + (−1 − it) , −1 ≤ t ≤ 1
2
 
(N ) 1
Γ4 (t) = N + (t − i) , −1 ≤ t ≤ 1
2
 A motivational example vii

GHΘL

1
2
tanh Π

Θ
0 Π 2Π 3Π 4Π

Figure 1 The function G(θ)

and let Γ(N ) be the closed contour which results from following Γ1 , Γ2 , Γ3
and Γ4 in turn. Sketch Γ(N ) , together with the locations of the poles of F in
the complex plane.

c) Show that if 0 ≤ θ ≤ 2π, then:

Z
C
F (z)dz ≤ ,
Γ(N ) N

where C is a fixed constant, independent of N .

d) By applying Cauchy’s residue theorem, show that for 0 ≤ θ ≤ 2π:

N
cosh(π − θ) X einθ C
− 2
≤ ,
2 sinh π 1+n N
n=−N

and conclude that

∞
cosh(π − θ) X einθ
= ,
2 sinh π n=−∞
1 + n2

with the sum converging uniformly in θ.

It is straightforward to verify that G has the following properties:

i) It is 2π-periodic: G(θ + 2π) = G(θ).

ii) It is everywhere continuous

iii) Restricted to the open interval (0, 2π), G is smooth and satisfies:

−G00 (θ) + G(θ) = 0.

viii Introduction

iv) The first derivative G0 (θ) has finite jump discontinuities at θ ∈ 2πZ, and2 :
G0 (0+ ) − G0 (0− ) = G0 (0+ ) − G0 (2π− ) = −1.

These properties are sometimes summarised in the formula:

∞
X
− G00 (θ) + G(θ) “ = ” δ(θ − 2πn). (6)
n=−∞

Here δ(θ) is the so called “Dirac delta function” whose defining properties are often given
as: δ(θ) = 0 when θ 6= 0, and:
Z 
δ(θ)dθ = 1,  > 0.
−

With this property understood, we can recover the jump condition on the derivative of G
by integrating (6). Obviously such an object doesn’t make sense within our usual notions
of a function. Assuming that we can insert our series representation (4) for G into (6)
and commute the derivative and the sum, we arrive at:
∞ ∞
1 X inθ X
e “=” δ(θ − 2πn). (7)
2π n=−∞ n=−∞

Showing that we can give a meaning to this expression will occupy quite a substantial
part of the course.
Now, let us return to our candidate solution to (1):
Z 2π
ψ(θ) = f (α)G(θ − α)dα,
0

And let us put to once side our discussion involving Fourier series and just take G as
being defined by (5) together with the assumption that it is periodic. Clearly, since G is
2π-periodic, so is ψ. Suppose θ ∈ [0, 2π]. Then we can write
Z θ Z 2π
ψ(θ) = f (α)G(θ − α)dα + f (α)G(θ − α)dα
0 θ
Z θ Z 2π
= f (α)G(θ − α)dα + f (α)G(θ − α + 2π)dα
0 θ
Z θ Z 2π
cosh(π − θ + α) cosh(−π − θ + α)
= f (α) dα + f (α) dα
0 2 sinh π θ 2 sinh π
Exercise 1.2. Suppose that f ∈ C 0 (R) is a continuous function with period
2π, i.e. f (θ) = f (θ + 2π). For θ ∈ [0, 2π), define:
Z θ Z 2π
cosh(π − θ + α) cosh(−π − θ + α)
ψ(θ) := f (α) dα + f (α) dα
0 2 sinh π θ 2 sinh π
and extend ψ to a function on R by periodicity: ψ(θ) = ψ(θ + 2π).
2
The notation f (x+ ) means lim&0 f (x + ). Similarly, f (x− ) = lim&0 f (x − ).
 A motivational example ix

a) Show that ψ ∈ C 0 (R).

b) By directly differentiating the formula, show that:

θ 2π
sinh(π − θ + α) sinh(−π − θ + α)
Z Z
ψ 0 (θ) = − f (α) dα − f (α) dα
0 2 sinh π θ 2 sinh π

and show that ψ ∈ C 1 (R).

c) Differentiating again, show that

ψ 00 (θ) = −f (θ) + ψ(θ)

Conclude that ψ ∈ C 2 (R) is a solution to

−ψ 00 (θ) + ψ(θ) = f (θ), θ ∈ R.

Exercise 1.3. Let φ ∈ C 2 (R) be 2π-periodic, i.e. φ(θ) = φ(θ +2π), and suppose
that φ satisfies:
−φ00 (θ) + φ(θ) = 0, θ ∈ R.

a) Show that if φ attains a maximum at θ0 ∈ R, then φ(θ0 ) ≤ 0.

b) Show that if φ attains a minimum at θ0 ∈ R, then φ(θ0 ) ≥ 0.

c) Show that φ ≡ 0.

d) Conclude that there is at most one ψ ∈ C 2 (R) satisfying ψ(θ) = ψ(θ + 2π)
and solving:
−ψ 00 (θ) + ψ(θ) = f (θ), θ ∈ R,
where f ∈ C 0 (R) is a given 2π-periodic function.

The conclusion of these two exercises is the following result

Lemma 0.1. Suppose that f ∈ C 0 (R) is a 2π-periodic function, and define:

Z 2π
ψ(θ) = f (α)G(θ − α)dα.
0

Where G : R → R is the 2π-periodic function given by:

cosh(π − θ)
G(θ) = for 0 ≤ θ < 2π.
2 sinh π

Then ψ ∈ C 2 (R) is the unique 2π-periodic function satisfying:

−ψ 00 + ψ = f.
x Introduction

Some remarks are in order at this stage. In the derivation of our solution ψ, we
performed several manipulations which are not justified without further assumptions. In
particular, in order to justify the interchange of various limiting operations, we would
expect that we require the Fourier series of f to converge at least uniformly. However,
there are C 0 functions with period 2π whose Fourier series do not converge pointwise
everywhere, let alone uniformly [see the Functional Analysis course]. On the other hand,
we have verified that the result that we arrived at is indeed correct. Part of the aim
of this course will be to provide a framework in which we can rigorously carry out the
manipulations in this example.
We also introduced the “Dirac delta function”. Making sense of this and the formula
(6) in which it appears will require us to consider a class of objects which generalise our
familiar idea of smooth functions.