Math 258 (Harmonic Analysis) Lecture Notes

Izak Oltman
last updated: August 29, 2020

These are lecture notes for math 258 (Harmonic analysis) at UC Berkeley instructed by
Professor Michael Christ during the Fall of 2020.

Contents
1 Introduction 2
1.1 Discrete Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

Index 7

1
 1. Introduction

Lecture 01 (8/27)

1 Introduction
Our story starts in 1822 when this guy was talking about heat on a metal rod. Let’s consider
(
the rod as a circle. Let’s paramatrize our circle x eix with x > 0, 2π . The heat equation
will be
¢̈
¨ut γuxx
¦
¨f ˆx uˆx, 0
¤̈
2
with u temp, t time. Fourier noticed that einx e γn t is a solution if the initial condition


is einx . He then made the big (controversial) leap that we can decompose initial conditions:

Q
ª

f ˆ x an einx
n ª

so that:

uˆx, t Qa e
n
n
 γn2 t inx
e

Fun fact: Fourier became the Governor of Egypt.

1.1 Discrete Fourier Transform

Let’s now talk about math. Consider the circle again, ˜z > C  Sz S 1 R~2πZ where x ( eix.
This is a group under both multiplication or addition.

 be the set of all homomorphisms ξ  G

If G is a finite abelian group, let G R~Z, then
 is a group.
G

Example 1.1. G ZN , then x ( e2πixn N (multiplicativity) with 0 B n @ N or x

~
(
nx~N mod Z (additive)

In general, finite abelian groups can be decomposed G  ZN1  ZN2   so that we get
homormorphisms

x ( M e2πix n j j ~Nj

and we get that the order of the group is easy to compute.

Let’s now define:

eξ ˆx e2πiξ ˆx

(I believe these are called characters) where ξ  G R~Z.

– 2–
 1. Introduction

 then
Proposition 1.1 (Orthogonality of Characters). If ξ and ξ are distinct elements of G, œ

`eξ , eξ e 0
œ

Where we can make the Hilbert space L2 ˆG with innerproduct `f, g e SGS1 P>
x G f ˆ x g ˆx 

Proof. Assume ξ, ξ > G

 with ϕˆx
œ
expˆ2πiξ ˆx and ψ ˆx expˆ2πiξ ˆx, then:
œ

`ϕ, ψ e Q ϕˆxψ̄ˆx Q ϕˆu  y  ψ ˆu  y 

x> G u

ŒQ ϕˆuψ̄ˆu‘ ϕˆyψ̄ˆy
u

with y arbitrary. If the inner product is not zero, we can divide by both sides, and get:

ϕˆy ψ̄ ˆy  1

for all y > G which is the same as ϕ ψ

The conclusion is that ™eξ  ξ > G ž is an orthonormal set in L2 ˆG. As a vector space,
ÂS SGS dimˆL2 ˆG. Therefore ™e  ξ > G
the dimension of this set is SG ž is an orthonormal
ξ
basis for L2 ˆG. Therefore, for all f > L2 ˆG,
2
Yf YL2 ˆG Q S `f, e e S ξ
2

ξ >G
Â

and
f Q `f, e e e ξ ξ
ξ >G
Â

Now let’s move to an infinite group. Let T R~2πZ which we can think of as the circle
(
Γ ˜Sz S 1 where x eix . We can also think of this as the interval π, π . Let Td d
1 T. L
Let f > L2 ˆTd . Define

fˆn ˆ2π 
 d
S f ˆx  e  in x
dx

with n > Zd . If we let en ˆx ein x , then fˆn ˆ2π d `f, en eL2 ˆTn  . Where:

`f, g e S Td
f ˆxg ˆxdx

Theorem 1.1 (Parseval). For d C 1, for all f > L2 ˆTd 

2
Yf YL2 ˆ2π 
d
Q SfˆnS 2

n>Zd

And

f Q fˆne in x

n>Zd

– 3–
 1. Introduction

What does this last equal sign mean? It means that

Yf ˆ   Q fˆne in x
YL2 ÐRÐÐ ª
0
S nSBR

Proof. We have Td , en ˆx eix n with n > Zd . Then ˜ˆ2π d~2 en  n > Zd  is orthonormala .

Thinking in finite abelian groups, x ( eix n Td Γ is a homomorphism.

The general theory of orthonormal sets is that for all f we know that:

Q S af, ˆ2π
n
 d~2
en f S2 B Yf Y2L2

we have equality if spanˆen  n > Zd  L2 .

To show this, we just use the Stone Weierstrauss theorem. If P spanˆen  n > Zd , this is
an algebra of continuous functions on Td . This is closed under conjugation. It contains 1 and
it separates points. Therefore by SWT we get that P is dense in C 0 ˆTd  with respect to Y YC 0 .

Now C 0 ` L2 as a dense subspace (which is a fact proved in measure theory).

Let’s now work in Rd for d C 1. Let ξ > Rd then x x ξ > R is a homomorphism Rd ( R.
(
Instead if we take x eix ξ this is a homomorphism Rd Γ. So we define:

fÂˆξ  S Rd
f ˆx  e  ix ξ
dx `f, eξ eL2 ˆRd 

where eξ ˆx eix ξ .

There are a couple issues

1. eξ > L2 ˆRd 
2. f ˆxe  ix ξ need not be in L1 if f > L2
We need to be less ambitious,
Definition 1.1. fÂˆξ  RRd f ˆx e  ix ξ dx `f, eξ eL2 ˆRd  for all f > L1 ˆRd 

This isn’t too bad, note that L1 9 L2 ˆRd  is dense in L2 ˆRd .

What can we say about these fourier coefficients, if f > L1 then f is bounded and con-
tinuous. To see this note:

fˆξ  u  fÂˆξ  S f ˆ x e  ix ξ
ˆe
 ix u
 1dx

if we let u go to zero. The integrand on the right goes to zero pointwise, then we apply the
dominated convergence theorem (RHS is dominated by 2Sf S.
a
calculus exercise

– 4–
 1. Introduction

Proposition 1.2. For all f > L1 9 L2 ˆRd  then f > L2 ˆRd  and YfÂY2L2 ˆ2π d Yf Y2L2

This means that we need

VS f ˆ x e  ix ξ
dxV P S Sf ˆxe
 ix ξ
Sdx

for most ξ (not rigorous). So we require a lot of cancellation in the integral on the LHS.

This proposition is the first part of Plancharel’s theorem. Thinking of  L1 9 L2 L2 , ,

then this extends uniquely to F  L2 ˆRd  L2 ˆRd  which is linear, bounded and YF f Y2L2
ˆ2π d Yf Y2L2 and for all f > L2

d Â
f ˆ2π  ˆf ˇ


where

fˇˆy  S f ˆξ eiy ξ dξ

for f > L1 9 L2 and this check also extends to a thing. And we get fˇˆy  fˆy . And finally
we have:

Yf  ˆ2π 
 d
S S ξ S BR
fÂˆξ eix ξ dξ YL2 ÐRÐÐ ª
0

Proof. Let’s prove Yf Y2L2 ˆ2π d YfÂY2L2 for all f > L2 9 L1 .

Will prove this first for a dense subspace of L2 which is contained in L1 .

Claim: (proved later) There exists a subspace V ` L2 ˆRd  such that V is dense in
L2 , V is contained in L1 , for every f > V there exists C @ ª such that for all ξ > Rd ,
SfÂˆξ S B Cf ˆ1  Sξ S d and f is continuous and has compact support.


Given f  Rd C supported where SxS B R @ ª (R Rˆf ). For large t A 0, let

ft ˆx f ˆtx. If t is sufficiently large, then the support is contained in ˆπ, π d . Therefore
we can think of ft  Td C.

If we compute the Fourier coefficients, we get:

ˆ2π 
 d
S Td
f t ˆx  e  in x
dx fÂt ˆn

(that hat is the one for Td ) on the other hand the LHS is just:

t  d
ˆ2π 
 d
S
Rd
f ˆ y e i n~ty
 ˆ
dy t  d
ˆ2π 
 dÂ
f ˆn~t

(that hat is the one for Rd )

– 5–
 1. Introduction

We have by a change of variables that:

2 d 2
Yft YL2 ˆTd  t Yf YL2 ˆRd 


while by Plancheral on the Torus:

2
Yft YL2 ˆTd  cd Q SfˆnS t
2
cd tœ  2d
Q Sfˆt  1
nS2
n>Zd n

(first hat is Td , the second is Rd ) Therefore:

Yf YL2 ˆRd 
2
cd tœ  d
Q Sfˆt
n
 1
nS2

If we take Rd and tile it of cubes of sidelength t  1 with lattices t 1 n, n > Zd . Then the RHS


is just:

cdœ
S Rd
Sgt S
2

with gt some function such that gt ˆx fˆt 1 n. So letting t


ª we would like the Riemann
sum to converge to the integral.

Claim:

S Rd
2 t
Sgt S ÐÐ SRª

d
SfÂS
2

this follows from the dominated convergence theorem.

Recall we assumed that SfÂˆξ S B Cf2 ˆ1  Sξ S  2d > L1 ˆRd . This serves as a dominator. We
know that gt ˆξ  ÐÐ
tª
fÂˆξ  for all ξ. And we also have that

Sgt ˆξ S Sfˆt
 1
nS B C ˆ1  St 1 nS   d
B C ˆ1  Sξ S
œ  d
®
unif

– 6–
 Index

Index
lecture 01 (8/27), 2

– 7–