Fourier Analysis

FIN

1
Contents

1 Fourier Series 3

2 Uniform Convergence of Fourier Series 5

3 Derivative and Integration of Fourier Series 5

2
 Among this note, every function is assumed to be bounded in its domain.

1 Fourier Series
Definition 1.1. We say a set of functions {ϕλ }λ∈Λ on (a, b) is orthonormal if for any i, j ∈ Λ,

⟨ϕi , ϕj ⟩ = δij .

Lemma 1.2 (Bessel’s Inequality). If {ϕ}n∈N is a orthonormal set, then for any f ∈ Cp (a, b), we have
∞
X 2 2
|⟨f, ϕn ⟩| ≤ ∥f ∥ .
n=1

Proof. Given N ∈ N, since

n 2
X
0≤ f− ⟨f, ϕn ⟩ ϕn
n=1

, we have
N
X N
X N
X
2 2 2 2
0 ≤ ∥f ∥ − 2 ⟨f, ϕn ⟩ + ⟨f, ϕi ⟩ ⟨f, ϕj ⟩ ⟨ϕi , ϕj ⟩ = ∥f ∥ − ⟨f, ϕn ⟩ .
n=1 i,j=1 n=1

Hence,
N
X 2 2
⟨f, ϕn ⟩ ≤ ∥f ∥ .
n=1

Since N is arbitrary, we are done.

2
Corollary 1.2.1 (Riemann-Lebesgue Lemma). If {ϕ}n∈N is a orthonormal set and ∥f ∥ < ∞, then

lim ⟨f, ϕn ⟩ = 0.
n→∞

Lemma 1.3. If G ∈ Cp (0, π), then

Z π  
2N + 1
lim G(x) sin x dx = 0.
N →∞ 0 2

Proof. Note that

Z π   Z π Z π
2N + 1 x x
G(x) sin x dx = G(x) cos sin(N x)dx + G(x) sin cos(N x)dx.
0 2 0 | {z } 2 0 | {z 2}
g(x) h(x)

nq q o
2 2
Note that g, h ∈ Cp (0, π) and π sin(N x), π cos(N x) : N ∈ N is an orthonormal set, by 1.2.1, we are
done.
Definition 1.4. The function DN (x) of the form
N
1 X
DN (x) = + cos(nx)
2 n=1

is called a Dirichlet kernel, where N ∈ N.

Here are some immediate consequences:
Z π
π
1. DN (x)dx = .
0 2
2. DN (x) is of period 2π.

3
  
1 2N + 1
3. DN (x) = sin x whenever sin(x/2) ̸= 0.
2 sin(x/2) 2
′
Lemma 1.5. If g ∈ Cp (0, π), and gR (0) exists, then
Z π
π
lim g(x)DN (x)dx = g(0+).
N →∞ 0 2

Proof.
π π π
g(x) − g(0+)
Z Z Z
2N + 1 π
g(x)DN (x)dx = sin( x)dx + g(0+)DN (x)dx = IN + g(0+).
0 2 sin(x/2) 2 2
|0 {z } 0
IN

Notice that
g(x) − g(0+) g(x) − g(0+) x/2 ′
= → gR (0)
2 sin(x/2) x sin(x/2)
as x → 0+ . Hence, I → 0 as N → ∞ by 1.3.

Theorem 1.6 (Pointwise Convergence of Fourier Series). Suppose f ∈ Cp (−π, π) with period 2π (periodic
extension). Then
a0 X f (x+) + f (x−)
+ an cos(nx) + bn sin(nx) → ,
2 2
n≥1

whenever fR′ (x), fL′ (x) exist, where

1 π
 Z
an = f (x) cos(nx)dx


π Z −π
1 π .
 bn = f (x) sin(nx)dx


π −π

Proof.
N Z π Z x+π Z x 
a0 X 1 1
+ an cos(nx) + bn sin(nx) = f (s)DN (s − x)ds = + f (s)DN (s − x)ds
2 n=1
π −π π x x−π

since both f and DN are of period 2π. Then since

1 x+π
Z Z π
y=s−x 1
f (s)DN (s − x)dx = f (x + y)DN (y)dy
π x π 0

and Z x Z π
1 y=x−s 1
f (s)DN (s − x)dx = f (x + y)DN (y)dy,
π x−π π 0

we have
N Z π
a0 X 1 1
+ an cos(nx) + bn sin(nx) = (f (x + y) + f (x − y))DN (y)dy → (f (x+) + f (x−))
2 n=1
2 0 2

if fR′ (x) and fL′ (x) exist.

If we write f ∼ (an , bn ), then an and bn are the Fourier coefficients of f .

4
2 Uniform Convergence of Fourier Series
Lemma 2.1. Suppose f ∈ C[−π, π] with f (−π) = f (π) and f ′ exists as long as f ′ ∈ Cp (−π, π), then
Xp
a2n + b2n
n≥1

converges, where f ∼ (an , bn ).

Proof. Suppose f ′ ∼ (An , Bn ), then by IBP, we know that An = nbn , Bn = −nan for all n ≥ 1 and A0 = 0.
Hence, v
N p N N r uN
X X 1 p X 1u X
a2n + b2n = A2n + Bn2 ≤ 2
t A2N + BN
2 .

n=1 n=1
n n=1
n n=1
P p
Then by 1.2, the RHS is bounded and therefore n≥1 a2n + b2n converges.
Theorem 2.2 (Uniform Convergence). Suppose f ∈ C[−π, π], f (−π) = f (π) and f ′ ∈ Cp (−π, π), then the
Fourier series of f converges uniformly to f on [−π, π].
PN
Proof. Suppose f ∼ (an , bn ) and SN = a20 + n=1 an cos(nx) + bn sin(nx) for all N ≥ 1. Then

N N
a0 X a0 X p
|SN | ≤ + |an | + |bn | ≤ + 2 a2n + b2n < ∞
2 n=1
2 n=1

by 2.1. Then by M-test, we are done.

3 Derivative and Integration of Fourier Series

Theorem 3.1. Suppose f ∈ C[−π, π], f (π) = f (−π) and f ′ ∈ Cp (−π, π). Then the Fourier series is
differentiate at each point x where f ′′ (x) exists and
X
f ′ (x) = n(−an sin(x) + bn cos(x)),
n≥1

where f ∼ (an , bn ).
Proof. Let g = f ′ and g ∼ (An , Bn ). Then

An = nbn , Bn = −nan , A0 = 0

for all n ≥ 1. Hence, X

g(x) = nbn cos(nx) − nan sin(nx) = f ′ (x)
n≥1
′ ′′
whenever g (x) = f (x) by 1.6.