Chapter 11.

Fourier Analysis 1

§11.1 Fourier Series

Definition (Periodic Functions)
A function f (x) is called a periodic function if ∃ p > 0
(called a period of f ) such that f (x + p) = f (x) for all x.
Ex. periodic : sin 2x, cos 3x, sin 2x + 3 cos 3x, · · ·
not periodic : x, x2, ex, cosh x, ln x, · · ·
If f (x) has period p, then f (x + np) = f (x) for all x and any
integer n = 1, 2, 3, · · · .
If f (x) and g(x) have period p, then af (x) + bg(x) has also
the period p for any constant a and b.

Theorem 1 (Orthogonality of Trigonometric System)

The trigonometric system
1, cos x, sin x, cos 2x, sin 2x, · · · , cos nx, sin nx, · · ·
is orthogonal on −π ≤ x ≤ π. That is,
∫ π ∫ π
cos nxdx = 0, sin nxdx = 0,
∫−ππ −π

cos nx cos mxdx = 0 (n ̸= m),

∫−ππ
sin nx sin mxdx = 0 (n ̸= m),
∫−ππ
sin nx cos mxdx = 0.
−π
∫ π ∫ π ∫ π
12dx = 2π, sin2 nxdx = π, cos2 nxdx = π
−π −π −π

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 2

[Fourier Series]
Suppose that f (x) is a function of period 2π.
Assume that it can be represented by a trigonometric series
(that is, the series converges and has the sum f (x).)
∑∞
( )
f (x) = a0 + an cos nx + bn sin nx .
n=1
What are a0, an, bn and how to find them?
Integrating on both sides from −π to π, by Theorem 1
∫ π
f (x)dx = 2πa0,
−π
and so ∫ π
1
a0 = f (x)dx.
2π −π
To determine an, multiplying on both sides by cos mx and in-
tegrating on both sides from −π to π, by Theorem 1
∫ π ∫ π
f (x) cos mxdx = a0 cos mxdx
−π −π
∞
∑ π( ∫
)
+ an cos nx + bn sin nx cos mxdx = πam,
n=1 −π

and so (m → n)
∫ π
1
an = f (x) cos nxdx.
π −π
Similarly, ∫ π
1
bn = f (x) sin nxdx.
π −π

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 3

Definition (Fourier series)

Fourier series of a function f (x) with period 2π
∞
∑ ( )
f (x) = a0 + an cos nx + bn sin nx .
n=1

Fourier coefficients (Euler formula)

∫ π
1
a0 = f (x)dx,
2π∫ −π
1 π
an = f (x) cos nxdx, n = 1, 2, · · · ,
π ∫−π
1 π
bn = f (x) sin nxdx, n = 1, 2, · · · .
π −π

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 4

Example 1
Find the Fourier series of the periodic function f (x),
{
−k if − π < x < 0
f (x) = , f (x + 2π) = f (x).
k if 0<x<π

Sol.
[∫ 0 ∫ π ]
1
a0 = −k dx + k dx = 0
2π −π
[∫ 0 0
∫ π ]
1
an = −k cos nx dx + k cos nx dx = 0
π −π
[∫ 0 ∫ 0π ]
1
bn = −k sin nxdx + k sin nx dx
π −π 0
2k 2k
= (1 − cos nπ) = (1 − (−1)n) ,
nπ nπ
Fourier series of f (x)
∑∞
2k 4k ( 1 )
(1 − (−1) ) sin nx =
n
sin x + sin 3x + · · · .
n=1
nπ π 3

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 5

Theorem 2
Let f (x) be periodic with period 2π and piecewise continuous
in [−π, π], and have a left-hand derivative and a right-hand
derivative at each point of that interval.
Then the Fourier series of f (x) is convergent and
the sum of the series
{
f (x ) ] if f is continuous at x = x0,
= 1[ 0
2 f (x0 + 0) + f (x0 − 0) if f is discontinuous at x = x0 .

Ex. {
x2 if x<1
f (x) =
x/2 if x≥1
left-hand limit: f (1 − 0) = 1
right-hand limit: f (1 + 0) = 1/2

In Example 1, (a) f is continuous at x = π2 .

π 4k 1 1 1
f ( ) = k = (1 − + − + · · · )
2 π 3 5 7
1 1 1 π
Thus 1 − + − + · · · = .
3 5 7 4

(b) f is discontinuous at x = 0.
sum of the series = 0 [ ] 1[ ]
1
= 2 f (0+0)+f (0−0) = 2 k +(−k) .

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 6

§11.2 Arbitrary Period. Even and Odd Functions.

Half-Range Expansion
1. From Period 2π to any period p = 2L
Let f (x) be a function of period 2L and
L
g(v) = f (x), x = v.
π
Then g(v) is a function of period 2π and have the Fourier series
∑∞
( )
g(v) = a0 + an cos nv + bn sin nv ,
n=1
∫ π
1
a0 = g(v)dv
2π∫ −π
1 π
an = g(v) cos nvdv,
π ∫−π
1 π
bn = g(v) sin nvdv.
π −π
Changing
π
v= x (−L ≤ x ≤ L ⇒ −π ≤ v ≤ π)
L
and g(v) = f (x), we obtain
∫ L ∫ L
1 π 1
a0 = f (x) dx = f (x)dx,
2π −L L 2L −L
∫ L ∫ L
1 nπx π 1 nπx
an = f (x) cos( ) dx = f (x) cos dx,
π −L L L L −L L
∫ L ∫ L
1 nπx π 1 nπx
bn = f (x) sin( ) dx = f (x) sin dx.
π −L L L L −L L

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 7

Definition
Fourier series f (x) of period p = 2L
∞
∑ ( nπx nπx )
f (x) = a0 + an cos + bn sin
n=1
L L
with the Fourier coefficients given by the Euler formulas
∫ L
1
a0 = f (x)dx,
2L −L
∫ L
1 nπx
an = f (x) cos dx,
L −L L
∫ L
1 nπx
bn = f (x) sin dx.
L −L L

Example 1(Periodic rectangular wave)

Find the Fourier series of the periodic function,
{
0 if 1 < |x| < 2
f (x) = , p = 2L = 4.
k if |x| < 1
1
∫1 nπx 1
∫1k
Sol. bn = 2 −1 k sin 2 dx = 0, a 0 = −1 kdx
4 = 2,
∫
1 1 nπx 2k nπ
an = k cos dx = sin .
2 −1 2 nπ 2
Fourier series of f (x)
k 2k ( π 1 3π 1 5π )
f (x) = + cos x − cos x + cos x − · · · .
2 π 2 3 2 5 2

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 8

Example 2(Periodic rectangular wave)

Find the Fourier series of the periodic function,
{
−k if − 2 < x < 0
f (x) = , p = 2L = 4.
k if 0<x<2
∫ 0 ∫
1 1 2
Sol. a0 = −kdx + kdx = 0,
4 ∫−2 4 0 ∫
1 0 nπx 1 2 nπx
an = −k cos dx + k cos dx = 0,
2 ∫ −2 2 2∫ 0 2
1 0 nπx 1 2 nπx
bn = −k sin dx + k sin dx
2 −2 2 2 0 2
2k
= (1 − cos nπ).
nπ
Fourier series of f (x)
4k ( π 1 3π 1 5π )
f (x) = sin x + sin x + cos x + · · · .
π 2 3 2 5 2

Fig. 263. Example 1

Fig. 264. Example 2

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 9

Example 3(Half-wave rectifier)

Find the Fourier series of the periodic function,
{
0 if − ωπ < t < 0
u(t) =
E sin ωt if 0 < t < ωπ
and u(t + 2π
ω ) = u(t).

Sol.
∫
ω π/ω E
a0 = E sin ωtdt = ,
2π 0 π
∫ π/ω
ω
an = E sin ωt cos nωtdt
π
{ 0
a1 = 0
= E − cos[(1+n)π]+1 − cos[(1−n)π]+1
a2 = · · · = 2π ( + )
{ 1+n 1−n
a1 = a3 = · · · = 0
= ,
a2 = a4 = · · · = − (n−1)(n+1)π
2E
∫
ω π/ω
bn = E sin ωt sin nωtdt
π
{ 0 E
b1 = 2
= ,
b2 = · · · = 0

Fourier series of the periodic function u(t)

E E
u(t) = + sin ωt
π 2
(
2E 1 1 )
− cos 2ωt + cos 4ωt + · · · .
π 1·3 3·5

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 10

2. Simplification: Even and Odd Functions

[Recall] Fourier series f (x) of period p = 2L
∑∞
( nπx nπx )
f (x) = a0 + an cos + bn sin
L L
n=1
∫ L
1
with coefficients a0 = f (x)dx,
2L −L
∫ L
1 nπx
an = f (x) cos dx,
L −L L
∫ L
1 nπx
bn = f (x) sin dx.
L −L L

If f is even, that is, f (−x) = f (x) for all x, its Fourier series
reduces to a Fourier cosine∞series
∑ nπx
f (x) = a0 + an cos
L
∫ L
n=1
1
with coefficients a0 = f (x)dx,
L ∫0
2 L nπx
an = f (x) cos dx
L 0 L
If f is odd, that is, f (−x) = f (x) for all x, its Fourier series
reduces to a Fourier sine series
∞
∑ nπx
f (x) = bn cos
L
∫ L
n=1
2 nπx
with coefficients an = f (x) sin dx
L 0 L

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 11

Summary (Even and Odd function of period 2π)

The Fourier series of an even function f (x) of period 2π is
∞
∑
f (x) = a0 + an cos nx (Fourier cosine series)
∫ πn=1 ∫ π
1 2
a0 = f (x)dx, an = f (x) cos nxdx.
π 0 π 0
The Fourier series of an odd function f (x) of period 2π is
∞
∑
f (x) = bn sin nx (Fourier sine series)
n=1
∫ π
2
bn = f (x) sin nxdx.
π 0

Theorem 1 (Sum and Scalar multiple)

(1) The Fourier coefficients of a sum f1 + f2 are the sums of
the corresponding Fourier coefficients of f1 and f2.
(2) The Fourier coefficients of cf are c times the corresponding
Fourier coefficients of f .
Example 5(Sawtooth Wave)
Find the Fourier series of the periodic function,
f (x) = x + π if −π < x < π and f (x + 2π) = f (x).
Sol. Let f (x) = f1(x)+f2(x),
{ where f1(x) = x and f2(x) = π.
a0 = π
Fourier coefficients of f2 =
an = bn = 0, n = 1, 2, · · ·
Fourier coefficients of f1

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 12

Since f1 is odd, a0 = an = 0, and

∫
2 π
bn = x sin nxdx
π [0 ∫ π ]
2 −x cos nx π 1
= [ ]0 + cos nxdx
π n n 0
2
= − cos nπ.
n
Fourier series of f
f = f1 + f2
∑∞
2
= π+ (− cos nπ) sin nx
n=1
n
1 1
= π + 2(sin x − sin 2x + sin 3x − · · · ).
2 3

Fig. 269. Partial sums S1 , S2 , , S3 , S2 0 in Example 5

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 13

3. Half-Range Expansions
Definition (Half Range Expansion of f (x), 0 ≤ x ≤ L)
[Even extension and periodic extension of period p = 2L]
f (x) extended as an even periodic function of period 2L
∑∞
nπx
f (x) = a0 + an cos
L
∫
1 L
n=1
∫
2 L
with a0 = L 0 f (x)dx, an = L 0 f (x) cos nπx L dx.
⇐ even periodic extension(Fourier cosine series)
[Odd extension and periodic extension of period p = 2L]
f (x) extended as an odd periodic function of period 2L
∑ ∞
nπx
f (x) = bn sin
n=1
L
∫
2 L
with bn = L 0 f (x) sin nπx
L dx.
⇐ odd periodic extension(Fourier sine series)

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 14

Example 6
Find the two half-range expansions of the function
{ 2k L
L x if 0 < x < 2 ,
f (x) = 2k
L (L − x) if L
2 < x < L.

Sol.
(a) even periodic expansion of f (x)
k 16k ( 1 2π 1 6π )
f (x) = − 2 2 cos x + 2 cos x + · · · .
2 π 2 L 6 L

(b) odd periodic expansion of f (x)

8k ( 1 πx 1 3πx 1 5πx )
f (x) = 2 2 sin − 2 sin + 2 sin + ··· .
π 1 L 3 L 5 L

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 15

§11.4 Approximation by Trigonometric Polynomials

[Minimum Square Error]
Let f (x) be a periodic function of period 2π and can be repre-
sented by a Fourier series
∑∞
( )
f (x) = a0 + an cos nx + bn sin nx .
n=1
Then the N th partial sum of the series
∑
N
( )
f (x) ≈ a0 + an cos nx + bn sin nx
n=1
is an approximation of the given f (x).
Let F (x) be a trigonometric polynomial of degree N
∑
N
( )
F (x) = A0 + An cos nx + Bn sin nx .
n=1
Consider ∫ π
E= (f (x) − F (x))2dx,
−π
called the square error of F relative to f on [−π, π].

Clearly, E ≥ 0.

Note: Here, we don’t choose E = max |f (x) − F (x)|.

x∈[−π,π]

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 16

N being fixed, we want to determine the coefficients An, Bn

such that
∫ πE is minimum. ∫ π ∫ π
E= |f (x)|2dx − 2 f (x)F (x)dx + |F (x)|2dx.
−π −π −π
Using the orthogonality of {sin nx, cos nx} and Euler’s formu-
las, we have
∫ π [ ]
∑
N
|F (x)| dx = π 2A0 +
2 2
(A2n + Bn2 ) , (why?)
−π n=1
and∫ ∫
π π
f (x)F (x)dx = A0 f (x)dx
−π −π
N [
∑ ∫ π ∫ π ]
+ An f (x) cos nxdx + Bn f (x) sin nxdx
−π −π
[
n=1
]
∑
N
= π 2A0a0 + (Anan + Bnbn) .
n=1
Thus
∫ [ ]
π ∑
N
E = |f (x)|2dx − 2π 2A0a0 + (Anan + Bnbn)
−π
[ ] n=1
∑
N
+π 2A20 + (A2n + Bn2 ) .
n=1
When A0 = a0, An = an, Bn = bn, its error, call it E ∗, is
∫ π [ ]
∑N
∗
E = |f (x)| dx − π 2a0 +
2 2
(a2n + b2n) ,
−π n=1

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 17

and so
{ }
∑
N
[ ]
E − E∗ = π 2(a0 − A0)2 + (an − An) + (bn − Bn)
2 2

n=1
≥ 0.
Thus E ≥ E ∗ and
E = E ∗ iff A0 = a0, · · · , BN = bN .
Theorem 1 (Minimum Square Error)
The total square error of a trigonometric polynomial F of de-
gree N relative to f on [−π, π] is minimum if and only if the
coefficients of F are the Fourier coefficients of f .
In this case,
∫ π [ ]
∑N
∗
0≤E = |f (x)| dx − π 2a0 +
2 2
(a2n + b2n) .
−π n=1

Hence
∑N ∫ π
1
2a20 + (a2n + b2n) ≤ |f (x)|2dx,
n=1
π −π
and
∞
∑ ∫ π
1
2
2a0 + (an + bn) ≤
2 2
|f (x)|2dx, Bessel’s inequality.
n=1
π −π

In fact,
∑∞ ∫ π
1
2a20 + (a2n + b2n) = |f (x)|2dx, Parseval’s identity.
n=1
π −π

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 18

Example 1(Sawtooth wave)

Compute the minimum square error E ∗ of F (x) with N =
10, 1000 relative to f (x) = x + π(−π < x < π) on [−π, π].
Sol. The Fourier series of f is
(−1)n+1
f = π + 2(sin x − sin 2x + · · · +
1
2 n sin nx + · · · ).
Consider
N +1
F (x) = π + 2(sin x − 21 sin 2x + · · · + (−1)N sin N x).
∫ π ∑
∗
( N
1)
Then E = (x + π) dx − π 2π + 4
2 2
2
.
−π n=1
n
and N = 20 → E ∗ = 0.6129, N = 1000 → E ∗ = 0.0126.
Notice that max |f (x) − F (x)| = π.
x∈[−π,π]

Note: In previous example, by the Parseval’s identity,

∑∞ ∫ π
2 4 1 2 8π 2
2π + 2
= (x + π) dx = .
n=1
n π −π 3
∑∞
1 π2
Thus we have 2
= .
n=1
n 6

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 19

§11.5 Sturm-Liouville Problems. Orthogonal functions

[Sturm-Liouville Problems]
1. Sturm-Liouville equation
[p(x)y ′]′ + [q(x) + λr(x)]y = 0
2. Sturm-Liouville boundary conditions
(a) k1y(a) + k2y ′(a) = 0
(b) l1y(b) + l2y ′(b) = 0
Assume that p, q, r, and p′ are continuous, and
r(x) > 0, a ≤ x ≤ b,
and k1, k2 are given constants, not both zero, and so are l1, l2,
not both zero.
3. Legendre’s equation
(1 − x2)y ′′ − 2xy ′ + n(n + 1)y = 0
=⇒ [(1 − x2)y ′]′ + λy = 0 (λ = n(n + 1))
4. Bessel’s equation
d2 y
2 + x̃ dx̃ + (x̃ − ν )y = 0
dy
x̃2 dx̃ 2 2

′′ ν2
=⇒ [xy ] + (− x − λx)y = 0 (x̃ = kx, λ = k 2)

[Eigenvlaues, Eigenfunctions]
The nonzero solutions of the Sturm-Liouville equation satisfying
the Sturm-Liouville boundary conditions is called eigenfunc-
tions and λ is called an eigenvalue.
Existence of eigenvalues : Eigenvalues of the a Sturm-
Liouville problem exist under some condition on p, q, r.
Reality of eigenvalues : If p, q, r, and p′ are real-valued
and continuous, r(x) > 0(or r(x) < 0) on a ≤ x ≤ b, then all
the eigenvalues of Sturm-Liouville problem are real.
Department of Mathematics V. Choi
 Chapter 11. Fourier Analysis 20

Example 1. Trigonometric functions as eigenfunctions

Find the eigenvalues and eigenfunctions of the Sturm-Liouville
problem
y ′′ + λy = 0, y(0) = 0, y(π) = 0.
This problem arises if an elastic string is stretched a little and
fixed at its ends x = 0 and x = π and then allowed to vibrate.

Sol. In Sturm-Liouville Problem,

p = 1, q = 0, r = 1, a = 0, b = π, k1 = l1 = 1, k2 = l2 = 0.
[Case λ = −ν 2 < 0]
A general solution : y = c1eνx + c2e−νx
From the conditions, a solution for IVP : y = 0
[Case λ = 0]
A general solution : y = c1 + c2x
From the conditions, a solution for IVP : y = 0
[Case λ = ν 2 > 0]
A general solution : y = A cos νx + B sin νx
From the conditions, y(0) = A = 0, y(π) = B sin νπ = 0
ν = ±1, ±2, · · · , λ = ν 2 = 1, 4, · · · .
Taking B = 1,
y = sin νx, ν = 1, 2, · · · .
eigenvalues of the problem : λ = ν 2 = 1, 4, · · · ,
eigenvectors of the problem : y(x) = sin νx, ν = 1, 2, · · · .

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 21

[Orthogonal Functions]
Definition (Orthogonality)
Functions y1(x), y2(x), · · · on [a, b] are called orthogonal on
[a, b] w.r.t. the weight function r(x) > 0 if
∫ b
r(x)ym(x)yn(x)dx = 0 (m ̸= n).
a
The norm ∥ym∥ of ym is defined by
√
∫ b
∥ym∥ = r(x)|ym(x)|2dx.
a

The functions y1, y2, · · · are called orthonormal on [a, b] if

they are orthogonal on [a, b] and all have norm 1. That is,
∫ b {
1 if m = n
r(x)ym(x)yn(x)dx = δmn = .
a 0 if m ̸
= n

Note: If r(x) = 1, the functions are called orthogonal in-

stead of orthogonal with respect to r(x) = 1. Then the
orthonormality and its norm are defined by
√
∫ b ∫ b
ym(x)yn(x)dx = 0, ∥ym∥ = |ym(x)|2dx.
a a

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 22

Example 2. Orthogonal Functions

Show that
ym(x) = sin mx, m = 1, 2, · · ·
are orthogonal on [−π, π].
Sol. For m ̸= n,
∫ π ∫ π
ym(x)yn(x)dx = sin mx sin nxdx
−π ∫ π
−π
1
= [cos(m − n)x − cos(m + n)x]dx
2 −π
= 0.
Note that
∫ π ∫ π
1
2
ym (x)dx = [1 − cos 2mx]dx = π
−π 2 −π
sin x sin 2x sin 3x
and so √ , √ , √ , · · · are orthonormal on [−π, π].
π π π

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 23

Theorem 1.(Orthogonality of Eigenfunctions)

Suppose that p, q, r, and p′ in the Sturm-Liouville problem
[p(x)y ′]′ + [q(x) + λr(x)]y = 0
(a) k1y(a) + k2y ′(a) = 0, (b) l1y(b) + l2y ′(b) = 0
are real-valued and continuous, and r(x) > 0 on a ≤ x ≤ b.
Let ym(x) and yn(x) be eigenfunctions of the Sturm-Liouville
problem corresponding to different eigenvalues λm and λn, re-
spectively.
Then ym, yn are orthogonal on [a, b] w.r.t. the weight r, that
is, ∫ b
r(x)ym(x)yn(x)dx = 0 (m ̸= n).
a
If p(a) = 0, then the condition (a) can be dropped.
If p(b) = 0, then the condition (b) can be dropped.
If p(a) = p(b), then the condition (a),(b) can be replaced by
y(a) = y(b), y ′(a) = y ′(b).
(periodic boundary conditions)

Proof. By assumption,

′ ′
(pym ) + (q + λmr)ym = 0,
(pyn′ )′ + (q + λnr)yn = 0.

Multiplying the 1st equation by yn, the 2nd equation by −ym,

and adding,

(λm − λn)rymyn = ym(pyn′ )′ − yn(pym

′ ′
) = [p(ymyn′ − ym
′
yn)]′.
Department of Mathematics V. Choi
 Chapter 11. Fourier Analysis 24

Integrating over x from a to b, we have

∫ b
(λm − λn) rymyndx = [p(ymyn′ − ym ′
yn)]ba
a
= p(b)[ym(b)yn′ (b) − ym
′
(b)yn(b)]
−p(a)[ym(a)yn′ (a) − ym ′
(a)yn(a)]
≡ I.
Recall that λm ̸= λn. If we show I = 0, then the proof is
completed.

Case 1. p(a) = p(b) = 0 =⇒ I = 0.

Case 2. p(a) ̸= 0, p(b) = 0. From the condition (a),

k1yn(a) + k2yn′ (a) = 0, ′
k1ym(a) + k2ym (a) = 0.
Let k2 ̸= 0. Multiplying the 1st equation by ym(a), 2nd equation
by −yn(a), and adding,
k2[ym(a)yn′ (a) − ym
′
(a)yn(a)] = 0 =⇒ I = 0.
If k2 = 0, then k1 ̸= 0 by assumption, ym(a) = yn(a) = 0 and
I = 0.

Case 3. p(a) = 0, p(b) ̸= 0. Similar to Case 2.

Case 4. p(a) ̸= 0, p(b) ̸= 0. Similar to Case 2, 3.

Case 5. p(a) = p(b). By the condition y(a) = y(b), y ′(a) =

y ′(b),
I = 0.
Department of Mathematics V. Choi
 Chapter 11. Fourier Analysis 25

Example 4. Orthogonality of the Legendre Polynomial

Legendre’s equation (1 − x2)y ′′ − 2xy ′ + n(n + 1)y = 0 may
be written
[(1 − x2)y ′]′ + λy = 0 (λ = n(n + 1)),
which means that Legendre’s equation is a Sturm-Liouville equa-
tion with p = 1 − x2, q = 0 and r = 1.
We know that for n = 0, 1, · · · , hence λ = 0, 1 · 2, 2 · 3, · · · ,
the Legendre polynomials Pn(x) are solution of the problem.
Hence these are eigenfunctions, and they are orthogonal on
[−1, 1], that is,∫
1
Pm(x)Pn(x)dx = 0 (m ̸= n)
−1

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 26

§11.6 Orthogonal Series. Generalized Fourier Series

Notation For a given r(x) > 0 on [a, b],
∫ b
(ym, yn) = r(x)ym(x)yn(x)dx,
a √
√ ∫ b
∥y∥ = (y, y) = r(x)y 2(x)dx.
a
Recall that {yn} is an orthonormal set w.r.t. r if
{
0, m ̸= n
(ym, yn) = δmn = .
1, m = n
[Orthogonal series]
Let y0, y1, y2, · · · be orthogonal w.r.t. weight r(x)(> 0) on
[a, b]. Let f (x) be a function that can be represented by a
convergent series
∞
∑
f (x) = amym(x)
m=0
= a0y0(x) + a1y1(x) + a2y2(x) + · · · .
This is called an orthogonal series, orthogonal expansion
or generalized Fourier series. Then we have
∞
∑
(f, yn) = am(ym, yn) = an(yn, yn) = an∥yn∥2.
m=0
Thus
(f, ym)
am = (Fourier constants)
∥ym∥ 2

and
∞
∑ (f, ym)
f (x) = ym (generalized Fourier series).
m=0
∥ym ∥2
Department of Mathematics V. Choi
 Chapter 11. Fourier Analysis 27

Example Fourier series

In the periodic Sturm-Liouville problem
y ′′ + λy = 0, y(π) = y(−π), y ′(π) = y ′(−π),
the eigenfunctions {1, cos kt, sin kt}k≥1 are orthogonal and a
function f (x) can be represented by a convergent series(called
Fourier series)
Sol. Note that p(x) = 1, q(x) = 0, r(x) = 1 > 0 on [−π, π]
and the eigenfunctions(solutions) are
1, cos x, sin x, cos 2x, sin 2x, · · · .
For m, n = 1, 2, · · · ,
(cos mx, sin nx) = 0,
(sin mx, sin nx) = πδmn, (cos mx, cos nx) = πδmn,
(cos mx, 1) = 0, (sin mx, 1) = 0, (1, 1) = 2π.
An eigenfunction expansion is Fourier series
∞
∑
f (x) = a0 + (am cos mx + bm sin mx)
m=1

where (Euler formulas)

π∫
1
a0 = f (x)dx,
2π −π
∫ π
1
am = f (x) cos mxdx,
π −π
∫
1 π
bm = f (x) sin mxdx.
π −π

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 28

Example 1. Fourier-Legendre series

In the Legendre’s equations,
[(1 − x2)y ′]′ + λy = 0 (λ = n(n + 1))
the Legendre polynomials {Pm(x)}m≥0 are orthogonal and a
function f (x) can be represented by a convergent series(called
Fourier-Legendre series)
Sol. Note that p = 1 − x2, q = 0, r = 1. The Legendre poly-
nomials Pm(x), m = 0, 1, 2, · · · are solutions(eigenfunctions) of
the equation and orthogonal.
An eigenfunction expansion is Fourier-Legendre series
∞
∑
f (x) = amPm(x)
m=0
= a0P0 + a1P1(x) + a2P2(x) + · · ·
where 1∫
1
am = f (x)Pm(x)dx
∥Pm∥ −1
∫ 1
2m + 1
= f (x)Pm(x)dx.
2 −1

Ex. sin πx = 0.95493P1(x) − 1.15824P3(x) + · · · .

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 29

Note: Bessel’s inequality & Parseval equality

Let y0, y1, y2, · · · be an orthonormal set on [a, b]. The or-
thogonal expansion of f (x) w.r.t the weight function r(x) and
its N th partial sum SN (x) are
∑∞ ∑N
f (x) = amym(x), SN (x) = amym(x),
∫ m=0
b
m=0

where am = r(x)f (x)ym(x)dx.

a
∫ b
0 ≤ r(x)|f (x) − SN (x)|2dx
∫a b ∫ b ∫ b
= r|f |2dx − 2 rf SN dx + r|SN |2dx
a a a
∫ b ∑
N ∫ b ∫ b ∑
N
= r|f |2dx − 2 am rf ymdx + r| amym|2dx
a m=0 a a m=0
∫ b ∑
N
= r|f |2dx − |am|2 (by orthogonality of ym)
a m=0

Hence we have Bessel’ inequality

∑N ∫ b
|am|2 ≤ ∥f ∥2 = r(x)|f (x)|2dx for N = 1, 2, · · · .
m=0 a
If y0, y1, y2, · · · on [a, b] are a complete orthonormal set in
a set of functions S, then every f in S satisfies the so-called
Parseval equality
∑∞ ∫ b
|am|2 = ∥f ∥2 = r(x)|f (x)|2dx.
m=0 a

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 30

§11.7 Fourier Integral

Example 1 Rectangular Wave
Consider the nonperiodic function
{
1 if − 1 < x < 1
f (x) = .
0 if otherwise
Let 
 0 if − L < x < −1
fL(x) = 1 if − 1 < x < 1

0 if 1<x<L
be the periodic function of period 2L(L > 1). Then
f (x) = lim fL(x).
L→∞

Investigate the Fourier coefficients of fL.

bn = 0,
∫ 1
1 1
a0 = dx = ,
2L −1 L
∫
1 1 nπx 2 sin(nπ/L)
an = cos dx = .
L −1 L L nπ/L
This sequence of Fourier coefficients is called the amplitude
spectrum of fL because |an| is the maximum amplitude of the
wave an cos(nπx/L).
For increasing L these amplitudes become more and more dense
on the positive wn-axis, where wn = nπ/L, and will eventually
be everywhere dense on the positive wn-axis(and will decrease
to zero).
Department of Mathematics V. Choi
 Chapter 11. Fourier Analysis 31

Fig. 280. Waveforms and amplitude spectra in Example 1

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 32

[From Fourier Series to Fourier Integral]

Let
f (x) = lim fL(x),
L→∞
∫∞
where fL(x) = fL(x + 2L) and −∞ |f (x)|dx < ∞.

Fourier series of fL :
∑∞
( ) nπ
fL(x) = a0 + an cos wnx + bn sin wnx (wn = )
L
∫ n=1
L
1
= fL(v)dv
2L −L
∞ ∫ L
1 ∑[
+ cos wnx fL(v) cos wnvdv
L n=1 −L
∫ L
]
+ sin wnx fL(v) sin wnvdv .
−L
Setting △wn = wn+1 − wn = Lπ ,
∫ L
1
fL(x) = fL(v)dv
2L −L
∑∞ ∫ L
1 [
+ (cos wnx)△w fL(v) cos wnvdv
π n=1 −L
∫ L
]
+ (sin wnx)△w fL(v) sin wn vdv .
−L
Let L → ∞ (△w → 0).
Then the Fourier series of fL converges to the Fourier integral
of f .
Department of Mathematics V. Choi
 Chapter 11. Fourier Analysis 33

Definition (Fourier integral)

Fourier integral of f (x)
∫ ∞
f (x) = [A(w) cos wx + B(w) sin wx]dw
0
where
∫ ∞ ∫ ∞
1 1
A(w) = f (v) cos wvdv, B(w) = f (v) sin wvdv.
π −∞ π −∞

Theorem 1
If f (x) is piecewise continuous in every finite interval and has
a left-hand
∫ ∞ derivative and right-hand derivative at every point,
and −∞ |f (x)|dx < ∞, then f (x) can be represented by a
Fourier integral of f (x) and
Fourier integral
{
f (x0) if f is continuous at x = x0,
=
2 [f (x0 +) + f (x0 −)] if f is discontinuous at x = x0 .
1

If f (x) is an even
∫ ∞ function, then B(w) = 0,
f (x) = A(w) cos wxdw, Fourier cosine integral
0 ∫
2 ∞
where A(w) = f (v) cos wvdv.
π 0
If f (x) is an odd∫ function, then A(w) = 0,
∞
f (x) = B(w) sin wxdw, Fourier sine integral
0∫
2 ∞
where B(w) = f (v) sin wvdv.
π 0

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 34

Example 2 Single pulse

Find the Fourier integral representation of the function
{
1 if |x| < 1
f (x) = .
0 if |x| > 1

Sol. Since f (x) is even,

∫ 1 B(w) = 0 and
2 2 sin w
A(w) = cos wvdv = .
π 0 ∫ ∞πw
2 cos wx sin w
Fourier integral of f (x) : f (x) = dw.
π 0 w
Note: 1. Dirichlet’s discontinuous factor
π
∫ ∞  2 if 0 ≤ x < 1
cos wx sin w
dw = π4 if x = 1 .
0 w 
0 if x > 1
∫ ∞
sin w π
2. In particular(x = 0), dw = .
0 w 2
3. The integral
∫
2 a cos wx sin w
dw
π 0 w
approximates f (x), and shows oscillations near the points of dis-
continuity of f (x) which is known as the Gibbs phenomenon.

Fig. 283. The approximation integral for a = 8, 16, and 32, illustrating the development of the
Gibbs phenomenon

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 35

Example 3 (Laplace integral)

Derive the Fourier cosine and Fourier sine integrals of f (x) =
e−kx where x > 0 and k > 0.
Sol. Note that the Laplace transform of cos wt is
∫ ∞
s
L(cos wt) = e−st cos wtdt = 2 2
.
0 s +w
(a) By integration by parts or the Laplace transform,
∫ ∞
2
A(w) = e−kv cos wvdv
π 0
2 k
= 2 2
,
πk + w
Fourier cosine integral of f :
∫ ∞
2k cos wx
f (x) = e−kx = 2 2
dw (x > 0, k > 0).
π 0 k +w
∫ ∞
cos wx π −kx
(*) 2 2
dw = e (x > 0, k > 0).
0 k +w 2k
(b) By integration by parts or the Laplace transform,
∫ ∞
2
B(w) = e−kv sin wvdv
π 0
2 w
= ,
π k2 + w2
Fourier sine integral of f :
∫ ∞
2 w sin wx
f (x) = e−kx = 2 2
dw (x > 0, k > 0).
π 0 k +w
∫ ∞
w sin wx π −kx
(*) 2 + w2
dw = e (x > 0, k > 0).
0 k 2
Department of Mathematics V. Choi
 Chapter 11. Fourier Analysis 36

§11.9 Fourier Transform. Discrete and Fast Fourier

Transforms
[Complex Form of the Fourier Integral]
Fourier integral of f (x):
∫ ∞
f (x) = [A(w) cos wx + B(w) sin wx] dw
0
where
∫
1 ∞
A(w) = f (v) cos wv dv,
π ∫−∞
1 ∞
B(w) = f (v) sin wv dv.
π −∞
Then
∫ ∞∫ ∞
1
f (x) = f (v)[cos wx cos wv + sin wx sin wv]dvdw
π ∫0
∞[∫ ∞
−∞
1 ]
= f (v) cos(wx − wv)dv dw.
π 0 −∞
Notice that the integral in brackets is an even function of w,
and (odd function)
∫ ∞ ∫ ∞
1 [ ]
f (v) sin(wx − wv)dv dw = 0.
2π −∞ −∞
Using the Euler formula eix = cos x + i sin x, we have
f (v)[cos(w(x − v)) + i sin(w(x − v))] = f (v)eiw(x−v)
and ∫ ∞ ∫ ∞
1
f (x) = f (v)eiw(x−v)dvdw.
2π −∞ −∞

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 37

Definition
Complex Fourier Integral
∫ ∞∫ ∞
1
f (x) = f (v)eiw(x−v)dvdw.
2π −∞ −∞

Note: ∫ ∞ ∫ ∞
1 [ 1 −iwv
] iwx
f (x) = √ √ f (v)e dv e dw.
2π −∞ 2π −∞
Definition
Fourier transform of f (x)
∫ ∞
1
F(f ) = fˆ(w) = √ f (x)e−iwxdx.
2π −∞
inverse Fourier transform of fˆ(w)
∫ ∞
1
f (x) = F −1(fˆ) = √ fˆ(w)eiwxdw.
2π −∞

Example 1
{
1 if − 1 < x < 1
Find the Fourier transform of f (x) = .
0 if otherwise
Sol.
1∫
1
fˆ(w) = √ e−iwxdx
2π −1
[ −iwx ]1
1 e
= √
2π −iw −1
√
2 sin w
= .
π w
Department of Mathematics V. Choi
 Chapter 11. Fourier Analysis 38

[Physical interpretation : Spectrum]

Note that
∫ ∞
1
f (x) = √ fˆ(w)eiwxdw (1).
2π −∞
The representation (1) of f (x) becomes clear if we think of
it as a superposition of sinusoidal oscillations of all possible
frequencies, called a spectral representation.
In (1), the spectral density fˆ(w) measures the intensity of
f (x) in the frequency interval [w, w + w△w].
∫ ∞
|fˆ(w)|2dw : total energy of the physical system
∫−∞b
|fˆ(w)|2dw : contribution of the frequencies w in [a, b] to
a
the total energy

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 39

(Ex.) A mechanical system giving a single frequency

harmonic oscillator(mass on spring) my ′′ + ky = 0.
Multiplication by y ′ gives
my ′′y ′ + kyy ′ = 0.
Setting v = y ′ and by integration,
1 2 1 2
mv + ky = E0 = const.
2 2
1 2
Here, 2 mv : kinetic energy,
1 2
2 ky : potential energy,
E0: total energy of the system.
A general solution is
y = a1 cos w0t + b1 sin w0t = c1eiw0t + c−1e−iw0t
where w02 = mk and c1 = a1−ib
2
1
, c −1 = a1 +ib1
2 = c¯1.
Substituting y and v = iw0(c1eiw0t − c¯1e−iw0t),
1 2 1 2
E0 = mv + ky
2 2
1 [ −iw0 t 2 −iw0 t 2
]
2
= m(−w0 )(c1e iw0 t
− c¯1e ) + k(c1eiw0 t
+ c¯1e )
2
= 2mw02c1c¯1 = 2k|c1|2.
Hence the energy is proportional to the square of the amplitude
|c1|.

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 40

Theorem 2 (Linearity)
The Fourier transform is a linear operation: if F(f ) and F(g)
exist, then for any constants a and b, F(af + bg) exists, and
F(af + bg) = aF(f ) + bF(g).

Proof It is true because integration is a linear operation.

Theorem 3 (Fourier transform of the Derivative of f (x))

Let f (x) be continuous and f (x) → 0 as |x| → ∞. Let f ′(x)
be absolutely integrable on the x-axis. Then F(f ′) = iwF(f ).
Proof Integrating by parts, we obtain
∫ ∞
1
F(f ′) = √ f ′(x)e−iwxdx
2π −∞ ∫ ∞
1 [ ]
= √ f (x)e−iwx|∞
−∞ − (−iw) f (x)e−iwxdx .
2π −∞
Since f (x) → 0 as |x| → ∞, F(f ′) = iwF(f ).

Note: Two successive application of the formula gives

F(f ′′) = iwF(f ′) = −w2F(f ).
Example 3
w 2
−ax2 √1 e− 4a ,
Using F(e )= 2a
a > 0, find the Fourier transform
of xe−x .
2

Sol.
−x2 1 −x2 ′ 1 −x2 iw − w2
F(xe ) = F(− (e ) ) = − iwF(e ) = − √ e 4 .
2 2 2 2

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 41

Definition
convolution of f and g
∫ ∞
(f ∗ g)(x) = f (p)g(x − p) dp
−∞

Note: f ∗ g = g ∗ f.
Theorem 4 (Convolution Theorem)
Suppose that f (x) and g(x) are piecewise continuous, bounded,
and absolutely integrable on the x−axis. Then
√
F(f ∗ g) = 2πF(f )F(g).
Proof
∫ ∞
1
F(f ∗ g)(w) = √ (f ∗ g)(x)e−iwxdx
2π ∫−∞ ∫
∞ ∞
1
= √ f (p)g(x − p)e−iwxdpdx
2π ∫−∞ ∫−∞
∞ ∞
1
= √ f (p)g(x − p)e−iwxdxdp.
2π −∞ −∞
Take x − p = q, then
∫ ∞∫ ∞
1
F(f ∗ g)(w) = √ f (p)g(q)e−iw(p+q)dqdp
∫ 2π∞
−∞ −∞

= f (p)F(g)e−iwpdp
√−∞
= 2πF(f )F(g).
∫ ∞
Note: (f ∗ g)(x) = fˆ(w)ĝ(w)eiwxdw
−∞

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 42

[Discrete Fourier Transform(DFT)]

Let f (x) be a periodic function of period 2π.
Assume that N measurements of f (x) are taken over [0, 2π] at
regularly spaced points
2πk
xk = , k = 0, 1, · · · , N − 1.
N
We say that f (x) is being sampled at these points.
We want to determine a complex trigonometric polynomial
∑
N −1
q(x) = cneinx
n=0
that interpolates f (x) at xk , that is,
∑
N −1
fk ≡ f (xk ) = q(xk ) = cneinxk , k = 0, 1, · · · , N − 1.
n=0
We must determine c0, c1, · · · , cN −1.
Multiplying e−imxk and summing over k from 0 to N − 1,
∑
N −1 ∑
N −1 N
∑ −1
fk e−imxk = cnei(n−m)xk
k=0 k=0 n=0
∑
N −1 ∑
N −1
i(n−m) 2πk
= cn e N .
n=0 k=0
Note that 
∑
N −1 
N [ if n = m
2πk 2πi(n−m) ]N
ei(n−m) N = 1− e N

 2πi(n−m) = 0 if n ̸= m.
k=0
1−e N

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 43

∑
N −1
Thus fk e−imxk = N cm, and so (m → n)
k=0
N −1
1∑
cn = fk e−inxk , fk = f (xk ).
N
k=0

Definition Discrete Fourier Transform

Discrete Fourier transform of the signal f = [f0, f1, · · · , fN −1]T
f̂ = [fˆ0, fˆ1, · · · , fˆN −1]T ,
where
∑
N −1
fˆn = N cn = fk e−inxk , fk = f (xk ).
k=0

This is the frequency spectrum of the signal.

In vector notation,
f̂ = FN f ,
where FN = [enk ], N × N Fourier matrix,
2πink 2πi
enk = e−inxk = e− N = wnk , w = wN = e− N .
Thatis, 
ˆ  0  
f0 w w · ·
0
w 0
f0
 ˆ  w0 w1 · · wN −1   f1 
 f1     
   0 2(N −1)  
 ·  = w w · · w 2
  ·  .
    · 
 ·  · · · · ·
fˆN −1 w0 wN −1 · · w(N −1)(N −1) fN −1

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 44

Example 4 DFT, Sample of N = 4 values

Find the discrete Fourier transform of the signal f = [0, 1, 4, 9]T .

Sol. N = 4, w = e−2πi/N = e−πi/2 = −i.

 0 0 0 0  
w w w w 0
w0 w1 w2 w3 1
f̂ = F4f =  0 2 4 6
 
w w w w  4
w0 w3 w6 w9 9
    
1 1 1 1 0 14
1 −i −1 i  1 −4 + 8i
=    
1 −1 1 −1 4 =  −6  .


1 i −1 −i 9 −4 − 8i

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 45

Theorem
A N × N Fourier matrix FN and its complex conjugate FN =
[wnk ] satisfy
FN FN = FN FN = N I,
where I is the N × N unit matrix, hence FN−1 = 1
N FN .

Proof Let GN = FN FN = [gjk ], where

gjk = (wj wk )0 + (wj wk )1 + · · · + (wj wk )N −1
= W 0 + W 1 + · · · + W N −1
where W = wj wk = [e2πi/N ]j [e−2πi/N ]k .

In case j = k, since W = 1, gjj = N .

In case j ̸= k, since W N = (wj wk )N = [e2πi]j [e2πi]k = 1,

1 − WN
gjk = = 0.
1−W
Hence GN = N I.

Similarly, FN FN = N I. The theorem is proved.

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 46

[Fast Fourier Transform(FFT)]

Let N = 2M .
2πi 4πi 2πi
2
wN 2
= w2M = (e− N )2 = e− 2M = e− M = wM .
The signal f = [f0, f1, · · · , fN −1]T is split into two vectors fev
and fod,
fev = [f0, f2, · · · , fN −2]T , fod = [f1, f3, · · · , fN −1]T .
Determine the DFTs
f̂ev = [fˆ0, fˆ2, · · · , fˆN −2]T = FM fev,
f̂od = [fˆ1, fˆ3, · · · , fˆN −1]T = FM fod.
We obtain the components of the DFT of the given vector f by
the formulas
fˆn = fˆev,n + wN
n ˆ
fod,n, n = 0, 1, · · · , M − 1,
fˆn+M = fˆev,n − wn fˆod,n, n = 0, 1, · · · , M − 1.
N

Proof The components of the DFT satisfy

∑
N −1 ∑
M −1 ∑
M −1
(2k+1)n
fˆn = kn
wN fk = 2kn
wN f2k + wN f2k+1.
k=0 k=0 k=0
2
Using wN = wM ,
∑
M −1 ∑
M −1
fˆn = kn
wM n
fev,k + wN kn
wM fod,k .
k=0 k=0
− 2πiM − 2πiM
Notice that M
wN =e =e
N = e−πi = −1. Thus
2M

fˆn = fˆev,n + wN
n ˆ
fod,n, n = 0, 1, · · · , M − 1,
fˆn+M = fˆev,n − wn fˆod,n, n = 0, 1, · · · , M − 1.
N

Department of Mathematics V. Choi

 Chapter 11. Fourier Analysis 47

Example 5 FFT, Sample of N = 4 values

Using FFT, find the discrete Fourier transform of the signal
f = [0, 1, 4, 9]T .
Sol. N = 4 and M = 2.
Thus
w = wN = e−2πi/N = e−πi/2 = −i
and w = wM = e−2πi/M = e−πi = −1.
( ) ( )( ) ( )
fˆ 1 1 f0 f0 + f2
f̂ev = ˆ0 = F2fev = =
f2 1 −1 f2 f0 − f2
( ) ( )( ) ( )
ˆ
f 1 1 f1 f1 + f3
f̂od = ˆ1 = F2fod = =
f3 1 −1 f3 f1 − f3
and so
fˆ0 = fˆev,0 + wN
0 ˆ
fod,0
= (f0 + f2) + (f1 + f3) = f0 + f1 + f2 + f3,
fˆ1 = fˆev,1 + wN
1 ˆ
fod,1
= (f0 − f2) − i(f1 − f3) = f0 − if1 − f2 + if3,
fˆ2 = fˆev,0 − wN
0 ˆ
fod,0
= (f0 + f2) − (f1 + f3) = f0 − f1 + f2 − f3,
fˆ3 = fˆev,1 − wN
1 ˆ
fod,1
= (f0 − f2) + i(f1 − f3) = f0 + if1 − f2 − if3.
When f0 = 0, f1 = 1, f2 = 4, f3 = 9,
fˆ0 = 14, fˆ1 = −4 + 8i, fˆ2 = −6, fˆ3 = −4 − 8i.

(Compare Ex.4.)
Department of Mathematics V. Choi