Fourier Analysis

Ripan Das
 Contents

Chapter 1. Lecture 1 5
1. Introduction 5

Chapter 2. Lecture 2 7

Chapter 3. Lecture 3 10
1. Approximate identity 12
2. Fourier transformation 13

Chapter 4. Lecture 4 15

Chapter 5. Lecture 5 16

Chapter 6. Lecture 6 19

Chapter 7. Lecture 7 20

Chapter 8. Lecture 8 21

Chapter 9. Lecture 9 22

3
 CHAPTER 1

Lecture 1

1. Introduction

We will restrict ourselves to the space R, Rn or S 1 , Tn . Note that they all are topological group.
Therefore we may do Fourier analysis in locally compact topological group which is not very easy.
Also we need some sort of measure on this groups to define Lp spaces. Thus we will start developing
theorems for Euclidean spaces and some compact subspaces of this spaces. When we use complex
analysis, we mainly restrict our attention on R. We will use Lebesge measure on Rn . Since S 1 is
compact and therefore finite measure space all Lp spaces are comparable.

Proposition 1.1. Show that Lp (S 1 ) ⊆ Lq (S 1 ) if p > q.

Proposition 1.2. Show that ℓp (S 1 ) ⊆ ℓq (S 1 ) if p ≤ q.

1
But this results are not true for Lp (R) or Lp (Rn ). For example take the function f (x) = ∈
|x|1/q
Lp (R) \ Lq (R) for p < q. However we have the following results.

Proposition 1.3 (Minkowski’s integral inequality). Suppose (Ω1 , F1 , µ1 ) and (Ω2 , F2 , µ2 ) be

two sigma finite measure spaces and consider the space (Ω1 ×Ω2 , F1 ⊗F2 , µ1 ⊗µ2 ). Let f : Ω1 ×Ω2 →
R be a non-negative measurable function and 1 ≤ p < ∞. Then
Z Z p 1/p Z Z 1/p
p
f (x, y)dµ2 (y) dµ1 (x) ≤ f (x, y) dµ1 (x) dµ2 (y).
Ω1 Ω2 Ω2 Ω1

Proof.

Proposition 1.4. (1) Let p < r < q be any and f ∈ Lp (λn ) ∩ Lq (λn ). Then f ∈ Lr (λn ).
(2) Let 1 ≤ p, q ≤ ∞ and p < r < q. Then f ∈ Lr (λn ) ⇒ f + f1 + f2 where f1 ∈ Lp (λn ) and
f2 ∈ Lq (λn ).
(3) Let (Ω, F, µ) be a measure space. For 1 < p < ∞, (Lp (µ))∗ = Lq (µ) where 1/p + 1/q = 1.
The dual of L1 (µ) is L∞ (µ) but L1 (µ) ⊊ (L∞ (µ))∗ .

Proof. (1)
(2)
□
1 1 n
Let f (x) = , x ∈ R for α > 0 or f (x) = , x ∈ R and α > 0 are radially symmetric
|x|α ∥x∥α
function such that they are not member of Lp (λn ) for any p ≥ 1.
Z Z Z Z
1 n 1 n−1 n−1
α dλ (x) = α
r drdω = vol(S ) rn−1−α dr.
R n ∥x∥ S n−1 R r R
5
6 1. LECTURE 1

Thus showing it for R is enough. But such functions can be inside appropriate Lp (λn ) + Lq (λn ) by
writing f as
f = f · 1{x∈R:∥x∥>1} + f · 1{x∈R:∥x∥≤1} .
Now we define three important operation on function spaces. Since Lebesgue measure is invari-
ant under translation and rotation, we expect the same for functions in Lp (λn ). For any f ∈ Lp (λn )
we define translation of f by a vector x ∈ Rn as
(τx f )(y) := f (y − x) (y ∈ Rn ).
Similarly, define dilation of a function by δ > 0 as
1
fδ (x) := n f (x/δ) (x ∈ Rn ).
δ
The reason for defining dilation like this is if f ∈ L1 (λn ) then
∥f ∥1 = ∥fδ ∥1 .
We also define the rotation of a function. Let f ∈ Lp (λn ) and A ∈ SO(n) we define
(Af )(x) := f (Ax) (x ∈ Rn ).
 CHAPTER 2

Lecture 2

Definition 0.1. A function f : Rn → C is called radial if f (Ax) = f (x) for all A ∈ SO(n).
That is f (x) = f (y) whenever ∥x∥ = ∥y∥ .

For such functions Z Z Z

f (x)λn (x) = f (r, ω)rn−1 drdω.
Rn S n−1 R
For example, volume of ball of radius r is given by
Z Z rZ
n
dλ (x) = rn−1 drdω.
B(0,r) 0 S n−1

Now we compute the volume of S n using the Gaussian. For time being assume that
Z Z n
2 2
e−∥x∥ dλn (x) = e−x dλ(x) .
Rn R
Anyway, it is easy to see just by applying Fubini’s theorem n-times. Thus we have
Z Z ! Z 
−∥x∥2 n −r2 n−1
e dλ (x) = e r dr · dω
Rn [0,∞) S n−1

i.e., Z n
−x2
Z e dλ(x)
n−1 R
Vol(S )= dω = Z .
2
S n−1 e−r rn−1 dr
[0,∞)

Definition 0.2. A function on Rn is called homogeneous of degree α ∈ R if f (rx) = r( x) for

all x ∈ Rn and for all r ∈ R∗ .

Proposition 0.3. If f is a homogeneous function on Rn then f ∈

/ Lp (λn ) for any 1 ≤ p ≤ ∞.

Proof.

Observation 0.4. An example of a radial homogeneous function is f (x) = ∥x∥α . Suppose g is

a radial homogeneous function of degree α on Rn then
g(x) = g(rω) = rα g(ω) = c ∥x∥α .
Hence all radial and homogeneous functions are of the form f (x) = c ∥x∥α for some c ∈ R where α
is degree of homogeneity.

Basic facts about Lp −spaces.

(1) For any measure space (Ω, F, µ), the space Lp (µ) is Banach space for 1 ≤ p ≤ ∞ and
L2 (µ) is a Hilbert space.
7
8 2. LECTURE 2

(2) If Ω = Rn then Lp (λn )’s are separable for 1 ≤ p < ∞.

(3) The space
Cc∞ (Rn ) := {f ∈ Rn → R : supp(f ) is compact}
is a dense subspace of Lp (λn ) for all 1 ≤ p < ∞.

Definition 0.5 (Convolution). For any two function f, g ∈ L1 (λn ), we define convolution of
f and g as Z
(f ∗ g)(x) := f (x − y)g(y)dλn (y) (x ∈ Rn )
Rn
whenever the integral exists.

Note that f ∗ g is the same as g ∗ f by change of variable.

Proposition 0.6. For f, g ∈ L1 (λn ) we have

∥f ∗ g∥1 ≤ ∥f ∥1 ∥g∥1

i.e., L1 (λn ) with respect to convolution is a Banach algebra.

Z
Proof. By definition ∥f ∗ g∥1 = |(f ∗ g)(x)|dλn (x). Thus,
Rn
Z Z Z
|(f ∗ g)(x)|dλn (x) = f (x − y)g(y)dλn (y) dλn (x)
Rn Rn Rn
Z Z
≤ |f (x − y)||g(y)|dλn (y)dλn (x)
Rn Rn
Z Z 
≤ |g(y)| |f (x − y)|dλ (x) dλn (y)
n
Rn Rn
≤ ∥f ∥1 ∥g∥1 (as Lebesgue measure is translation invariant).

Proposition 0.7. For f ∈ L1 (λn ) and g ∈ Lp (λn ) we have f ∗ g ∈ Lp (λn ) i.e.,

∥f ∗ g∥p ≤ ∥f ∥1 ∥g∥p .

Thus Lp (λn ) is a L1 (λn )−module.

Proof.
Z 1/p
p n
∥f ∗ g∥p = |(f ∗ g)(x)| dλ (x)
Rn
Z Z p 1/p
n n
= f (x − y)g(y)dλ (y) dλ (x)
Rn Rn
Z Z 1/p
p p n
≤ |f (x − y)| |g(y)| dλ (x) dλn (y)
Rn Rn
≤

Definition 0.8. Let (Ω, F, µ) be a measure space and f : Ω → C be a measurable function.

For α ≥ 0, we define
df (α) = µ({ω ∈ Ω : |f (ω)| > α}).
 2. LECTURE 2 9
Z
Proposition 0.9. Let f ∈ Lp (λn ) then ∥f ∥pp = pxp−1 df (x)dλ(x).
[0,∞)

Proof. Observe
Z Z Z
p−1 p−1
px df (x)dλ(x) = px dλn (y)dλ(x)
[0,∞) [0,∞) {y∈Rn :|f (y)|>x}
Z Z |f (y)|
= pxp−1 dλ(x)dλn (y)
Rn 0
p
= ∥f ∥p .
□

Definition 0.10 (Weak Lp spaces). For a measurable function f : Ω → C, we define

∥f ∥pp,∞ := sup αp df (α).
α>0

Proposition 0.11. For f ∈ Lp (µ), we have ∥f ∥pp,∞ ≤ ∥f ∥pp .

Proof. By definition,
∥f ∥pp,∞ = sup αp df (α)
α>0
Z
p
= sup α dµ(ω)
α>0 {ω∈Ω:|f (ω)|>α}
Z
≤ sup |f (ω)|p dµ(ω)
α>0 {ω∈Ω:|f (ω)|>α}
Z
≤ |f (ω)|p dµ(ω)
Ω
= ∥f ∥pp .
□
 CHAPTER 3

Lecture 3

Proposition 0.1. For f ∈ L1 (λn ) and g ∈ Lp (λn ); 1 ≤ p ≤ ∞ and δ > 0, we have

(f ∗ g)δ = fδ ∗ gδ .
1
Proof. Recall that (f ∗ g)δ (x) = n (f ∗ g)(x/δ.) Thus
δ
Z
1 1
(f ∗ g)(x/δ) = n f (x/δ − y)g(y)dλn (y)
δn δ Rn
   
x−u
Z
1 u y
= 2n f g dλn (u) (put u = )
δ Rn δ δ δ
Z
= fδ (x − u)gδ (u)dλn (u)
Rn
= (fδ ∗ gδ )(x).
□

Proposition 0.2. For a ∈ Rn and f ∈ L1 (λn ), g ∈ Lp (λn ); 1 ≤ p < ∞ we have

τa (f ∗ g) = (τa f ) ∗ g = f ∗ (τa g).
In other words, τa commutes with convolution.

Proof. □

If f ∈ Cc∞ (R) and g ∈ L1 (R) then

f (x + h − y) − f (x − y)
Z Z
d
(f ∗ g)(x) = lim g(y)dλ(y) = f ′ (x − y)g(y)dλ(y).
dx h→0 R h R
To push the limit inside integration, use Lebesgue’s dominated convergence theorem. In this
scenario f ∗ g ∈ Cc∞ (R). An analogous proof works in Rn − the directional derivative exists (as
above) and are Cc∞ . In particular
Z
∇(f ∗ g)(x) = (∇f )(x − y)g(y)dλn (y).
Rn

In the light of Proposition (2) 0.7, for any f ∈ L1 (λn ) we can define a linear map Tf : Lp (λn ) →
Lp (λn ) by Tf (g) = f ∗ g.
We can also define the convolution of measure that is

Definition 0.3. If µ is a probability measure on Rn and f ∈ Lp (µ) then define

Z
(f ∗ µ)(x) := f (x − y)dµ(y).
Rn
10
 3. LECTURE 3 11

Definition 0.4. For any two measure µ1 , µ2 on Rn we define

Z Z 
(1) (µ1 ∗ µ2 )(A) := dµ1 (x − y) dµ2 (y).
A Rn

Example 0.5. Let δ0 be the Dirac δ measure at 0. Then (f ∗ δ0 )(x) = f (x).

For f, g ∈ L1 (λn ) we use the notation

Z
⟨f, g⟩ := f (x)g(x)dλn (x).
Rn

Note that the integral on the right-hand side may not exist.

Observation 0.6. For f, g ∈ L1 (λn ), the integral

Z
f (x)g(x − a)dλn (x)
Rn

exists for almost every a ∈ Rn because the expression is equal to (g # ∗ f )(a) where g # (x) = g(−x)
for x ∈ Rn .

Let f, g, h ∈ L1 (λn ) then

Z Z
(2) ⟨f ∗ g, h⟩ = f (y)g(x − y)h(x)dλn (x)dλn (y)
Rn Rn
Z Z 
(3) = f (y) g(x − y)h(x)dλ (x) dλn (y)
n
Rn Rn
#
(4) = ⟨f, g ∗ h⟩.

Motivated by this, we define the convolution of measures as

⟨µ1 ∗ µ2 , f ⟩ := ⟨µ1 , µ#
2 ∗ f ⟩.

Note that
Z
(5) f (x)d(µ1 ∗ µ2 )(x) = ⟨µ1 ∗ µ2 , f ⟩ = ⟨µ1 , µ# ∗ f ⟩
Rn
Z Z 
(6) = f (x + y)dµ2 (y) dµ1 (x)
Rn Rn
Z
(7) = f (x + y)d(µ1 ⊗ µ2 )(x, y)

Taking f = 1A , we get our equivalent form of (1). Also for + : Rn × Rn → Rn , (2) becomes
Z Z
f d(µ1 ∗ µ2 ) = (f ◦ +)(x, y)d(µ1 ⊗ µ2 )(x, y)
Rn

i.e., µ1 ∗ µ2 is push forward of the product measure µ1 ⊗ µ2 under +.

12 3. LECTURE 3

1. Approximate identity

Recall that L1 (λn ) is a Banach algebra under ∗. But we claim that L1 (λn ) has no identity
element under ∗. Suppose not, then there is some 0 ̸= g ∈ L1 (λn ) such that f ∗ g = f for all
f ∈ L1 (Rn ). Then for all δ > 0 and for all f ∈ L1 (λn ) we have

fδ ∗ gδ = fδ ⇒ f ∗ gδ = f.

Thus g = gδ in L1 (λn ) i.e., g is homogeneous (a.e.). Hence g ∈

/ L1 (λn ).

Proposition 1.1. There is no g ∈ L1 (λn ) such that f ∗ g = f for all f ∈ Lp (λn ); 1 ≤ p < ∞.
L1
Proof. If so, for all ψ ∈ Cc∞ , g ∗ ψ = ψ. Let f ∈ L1 (λn ) be any consider a sequence ψn −→ f.
Then
∥ψn ∗ g − f ∗ g∥1 ≤ ∥ψb − f ∥1 ∥g∥1 .
As n → ∞, the right-hand side also goes to 0. Thus we get f ∗ g = f for all f ∈ L1 (λn ) which
contradicts the previous observation. □

Definition 1.2 (Approximate identity). Let {kt } ⊆ L1 (λn ) be a sequence of net of functions
in L1 (λn ) with the property that
Z
(1) kt (x)λn (x) = 1 for all t,
Rn
Z
(2) |kt (x)|dλn (x) ≤ c for all t,
Rn

(3) given any ϵ > 0, the integral

Z
|kt (x)|dλn (x) → 0
{x∈Rn :∥x∥>ϵ}

as t → ∞.
Such a sequence of net is called approximate identity.

Construction of approximate identity. Take a positive function k ∈ L1 (λn ) such that

Z
f (x)dλn (x) = 1.
Rn

1
Define kt (x) = k(x/t). Then conditions (1) and (2) in the definition 2.2 is satisfied. Note that
tn
Z Z
n
kt (x)dλ (x) = k(x)dλn (x) → 0
{x∈Rn :∥x∥>ϵ} {x∈Rn :∥x∥>ϵ/t}

as t → 0 by monotone convergence theorem.

2
For example we may take k(x) = e−π∥x∥ . Thus
1 2
kt (x) = e−π∥x∥ /t
.
tn/2
 2. FOURIER TRANSFORMATION 13

Lemma 1.3. For a ∈ Rn and f ∈ Lp (λn ), we have

∥τa f − f ∥→ 0

as a → 0.

Proof. Let f ∈ Cc∞ (Rn ) be any function.

Proposition 1.4. Let f ∈ Lp (λn ) and {kt } be an approximate identity. Then f ∗ kt → f in

Lp (λn ).

Proof.

Proposition 1.5. Let f ∈ Lp (λn ) and g ∈ Lq (λn ) where 1/p + 1/q = 1. Then f ∗ g is bounded
and continuous.

Proof.

Proposition 1.6. Let f ∈. Then (f ∗ kt )(x) → f (x) as t → ∞ uniformly.

Proof.

Remark 1.7. Proposition 2.4 is not true for L∞ (λn ). Let f ∈ L∞ (λn ) be any. If the statement
L∞
is true then f ∗ kt −−→ f . But by Proposition 2.5 f ∗ kt becomes continuous and bounded. Thus f
becomes continuous. Since we have started with any L∞ −function, we arrived at a contradiction.

2. Fourier transformation

Definition 2.1. Let f ∈ L1 (λn ) and ξ ∈ Rn . Then define

Z
ˆ
f (ξ) := f (x)e−2πi⟨ξ,x⟩ dλn (x).
Rn

Remark 2.2. (1) We can’t define Fourier transform as an absolutely convergent integral
unless f ∈ L (λn ).
1

(2) If f ∈ L1 (λ) then for any complex number ξ = σ + it we can define fˆ(ξ) as
Z
|fˆ(ξ)| ≤ |f (x)|e2πtx dλ(x).
R

But this integral exists for all ξ ∈ C if f ∈ Cc∞ (R). By Fubini and Morera’s theorem fˆ is
entire. In particular if f ∈ Cc∞ (Rn ) then Fourier transformation of f can be extended to
Cn as a holomorphic function. However we must continue to use ⟨z, x⟩ = z1 x1 + · · · + zn xn
where z = (z1 , . . . , zn ) ∈ Cn and x = (x1 , . . . , xn ) ∈ Rn .
(3) Suppose f has rapid deacy say f (x) = O(e−α|x| ). Then,
Z
ˆ
f (σ + it) = f (x)e2πtx e−2πiσx dλ(x)
R

is well defined for all |t| < α. Thus, the Fourier transformation admits an extension to a
2
strip in C. Again for g(x) = eπx , ĝ(ξ) admits an extension to C i.e., ĝ(z) is entire. By
2
identity theorem ĝ(z) = e−πz is forced.
14 3. LECTURE 3

Definition 2.3. Let µ be a finite measure on Rn . Then we define Fourier transformation of µ

as Z
µ̂(ξ) := e−2πi⟨ξ,x⟩ dµ(x) (ξ ∈ Rn ).
Rn

Properties 2.4. Let f ∈ L1 (λn ). Let a, ξ ∈ Rn , δ > 0 and A ∈ SO(n) be arbitrary. Then
[
(1) (τ −2πiξa fˆ(ξ).
a f )(ξ) = e
(2) fˆδ (ξ) = fˆ(δξ).

Proof.
2 2
Example 2.5. On R, we know that if g(x) = e−πx then ĝ(ξ) = e−πξ . Thus,
1\ √ 2
√ e−πx2 /δ = gˆδ (x) = ĝ( δx) = e−πδx .
δ
Remark 2.6. Fourier transformation is defined above works for L1 −functions only. We aim
to extend it.
 CHAPTER 4

Lecture 4

Properties 0.1. Continuation of Properties (3) 2.4.

c (ξ) = fˆ(Aξ).
(3) Af
(4) If f is radial then fˆ is also radial. In particular for f : R → C, f is even implies fˆ is even.
(5) fˆ ≤ ∥f ∥ .
1
∞
(6) fˆ is uniformly continuous.
(7) (Riemann-Lebesgue Lemma) For f ∈ L1 (λn ), fˆ ∈ C0 (Rn ).
(8) For f, g ∈ L1 (λn ) we have
f[∗ g = fˆ · ĝ.

Proof.
2
Example 0.2. Let g(x) = e−π∥x∥ for x ∈ Rn . For the Gaussian, dilation by δ will mean dilation
√
by δ but we still denote it by gδ . Thus
1 2
gδ (x) = n/2 e1π/δ∥x∥ .
δ

15
 CHAPTER 5

Lecture 5

Take the indicator function of the interval [−a, a] for a > 0. Then the Fourier transform of
e2πiξa − e−2πiξa
1[−a,a] at ξ is 1̂[−a,a] (ξ) = . Note that this function is differentiable.
2πiξ
Question 0.1. Under what conditions on f does the Fourier transformation of f become an
infinitely many differentiable function?
Z
d ˆ d
1
Let f ∈ L (λ). Then f (ξ) = f (x)e−2πiξx dλ(x). That is
dξ dξ R
" #
fˆ(x + h) − fˆ(x) e−2πi(ξ+h)x − e−2πiξx
Z
lim = lim f (x) dλ(x)
h→0 h h→0 R h
e−2πihx − 1
Z
= lim f (x)e−2πiξx dλ(x).
h→0 R h

Thus if f (x) · 2πix ∈ L1 (λ), then we can conclude. Suppose xk f (x) ∈ L1 (λ) for all k = 1, . . . , n.
Then for any polynomial function p(x) = a0 + · · · + an xn , we have

dn
 
d
p := a0 + · · · + an n .
dξ dξ

Therefore
 
d
p fˆ(ξ)
dξ
 
d
makes sense and p fˆ(ξ) = qd
· f (ξ) for some polynomial function q.
dξ

Proposition 0.2. Let f ∈ L1 (λ) and f ′ ∈ L1 (λ). Also let |f (x)| → 0 as |x| → ∞. Then

fˆ′ (ξ) = 2πiξ fˆ(ξ).

Proof. We use integration by parts.

Z
′
ˆ
f (ξ) = f ′ (x)e−2πiξx dλ(x)
R
Z
−2πiξx
=e f (x)]∞
−∞ + 2πiξ f (x)e−2πiξx dλ(x)
R
= 2πiξ fˆ(ξ).

□
16
 5. LECTURE 5 17

\

d

Similarly we can show that p f (ξ) = q(ξ)fˆ(ξ) for some polynomial function q. To recall
dξ
this, let f, g ∈ Cc∞ (R) then
Z
df df
⟨ (x), g(x)⟩ = (x)g(x)dλ(x)
dx R dx
 ∗
d
= ⟨f (x), g(x) ⟩
dx
 ∗
d d
but =− .
dx dx
2
Proposition 0.3. Let g(x) = e−πx for x ∈ R. Then
2
ĝ(ξ) = e−πξ (ξ ∈ R).
d
Proof. Note that xn g(x) ∈ L1 (λ) thus we can talk about ĝ(ξ).
dξ
Z
d ˆ d
f (ξ) = g(x) (e−2πiξx )dλ(x)
dξ R dξ
Z
2
= i e−πx (−2πx)e−2πiξx dλ(x)
R Z 
−2πiξx −πx2 ∞ −2πiξx πx2
= i [e e ]−∞ − e (−2πiξ)e dλ(x)
R
= −2πξĝ(ξ).
Observe that ĝ(ξ) satisfies a differential equation
d
ĝ(ξ) = −2πξĝ(ξ).
dξ
2
Solving this differential equation gives us ĝ(ξ) = ce−2πξ . Putting ξ = 0 on both sides of the
2
equation shows c = 1. Hence ĝ(ξ) = e−πξ . This proof does not use complex analysis at all. □
We are now going to define a new class of functions that is invariant under the action of the
Fourier transformation.

Definition 0.4 (Schwartz class function). A function f ∈ C ∞ (R) is said to be in S (R) if

sup(1 + |x|)m |f (n) (x)| < ∞
x∈R

for all m, n ∈ N0 = N ∪ {0}.

Check that S (R) is a topological vector space where the topology is inherited as subspace
topology on S (R). For f ∈ S (R) we denote
pm,n (f ) := sup(1 + |x|)m |f (n) (x)|.
x∈R

Then for all m, n ∈ N0 , pm,n are semi norm which separates functions of S (R). Hence S(R) is a
locally convex metriziable topological vector space where
∞
X 1 pn (f − g)
d(f, g) =
2n 1 + pn (f − g)
n=1
18 5. LECTURE 5

where pn ’s are semi-norms defined above. Thus any sequence fn → f in S (R) if and only if
pn (fm − f ) → 0 as m → ∞ for all n ∈ N.

Proposition 0.5. (1) If f ∈ S (R) then fˆ ∈ S (R).

(2) Cc∞ (R) ⊊ S (R).
(3) S (R) is a proper dense subspace of Lp (λ) for all 1 ≤ p < ∞.

Proof.
 CHAPTER 6

Lecture 6

Let f ∈ S (R) then by definition of S (R) we have

c
|f (x)| ≤ .
(1 + |x|n )
As |x| → ∞, |f (x)| → 0 and f ∈ Lp (λ) for all 1 ≤ p < ∞. For any polynomial function p we
have p(x) · f ∈ S (R) then f ′ ∈ S (R). Also under pointwise multiplication and convolution S (R)
becomes an algebra.

Proposition 0.1. For f, g ∈ S (R) we have

⟨f, ĝ⟩ = ⟨fˆ, g⟩.

Proof.

19
CHAPTER 7

Lecture 7

20
CHAPTER 8

Lecture 8

21
CHAPTER 9

Lecture 9

22