MA 201, Mathematics III, July-November 2022,

Fourier Integral and Fourier Transform

Lecture 18

Lecture 18 MA 201, PDE (2022) 1 / 23

Fourier Integral
We already know that Fourier series and orthogonal expansions can be used as
tools to solve boundary value problems over bounded regions such as intervals,
rectangles, disks and cylinders.

But we can understand that the modelling of certain physical phenomena will
give rise naturally to boundary or initial value problems over unbounded regions.

For example:
To describe the temperature distribution in a very long insulated wire, we can suppose
that the length of the wire is infinite, which gives rise to an initial value problem over
an infinite line.

To solve this type of problems, we can generalize the notion of Fourier series by
developing Fourier transform.

Lecture 18 MA 201, PDE (2022) 2 / 23

Fourier Integral
Recall that for a piece-wise C 1 and continuous function fp (t), its Fourier series in
the interval (−L, L) satisfies (fp is periodic with period 2L)
∞     
A0 X nπt nπt
fp (t) = + An cos + Bn sin , (1)
2 n=1
L L
where

L  
1 nπt
Z
An = fp (t) cos dt, n = 0, 1, 2, 3, . . . ,
L −L L
L  
1 nπt
Z
Bn = fp (t) sin dt, n = 1, 2, 3, . . . .
L −L L

We know
 
nπt 1
cos = [exp(inπt/L) + exp(−inπt/L)],
L 2
 
nπt 1
sin = [exp(inπt/L) − exp(−inπt/L)].
L 2i

Lecture 18 MA 201, PDE (2022) 3 / 23

Fourier Integral
Putting these expressions in (1),

∞   
A0 X 1
fp (t) = + An {exp(inπt/L) + exp(−inπt/L)}
2 n=1
2
 
1
+ Bn {exp(inπt/L) − exp(−inπt/L)} .
2i

From above, An and Bn can be absorbed in Cn as

1 1
Cn = (An − iBn ) , Cn = (An + iBn ) = C−n , n = 0, 1, 2, . . .
2 2

Subsequently, we obtain Fourier series for fp (t) in complex form as

∞
X
fp (t) = Cn einπt/L , (2)
n=−∞

where
L
1
Z
Cn = fp (τ )e−inπτ /L dτ, n = 0, ±1, ±2, . . . (3)
2L −L

Lecture 18 MA 201, PDE (2022) 4 / 23

Fourier Integral
Substituting (3) into (2):
∞  Z L 
X 1 π
fp (t) = fp (τ )e−inπτ /L dτ einπt/L . (4)
n=−∞
2π −L L

Define the frequency of the general term by

nπ
σn = , (5)
L

and the difference in frequencies between successive terms by

(n + 1)π nπ π
∆σ = σn+1 − σn = − = . (6)
L L L

Then (4) reduces to

∞  iσn t Z L 
X e
fp (t) = fp (τ )e−iσn τ dτ ∆σ
n=−∞
2π −L
∞  iσt Z L 
X e
= F (σn )∆σ, with F (σ) = fp (τ )e−iστ dτ . (7)
n=−∞
2π −L

Lecture 18 MA 201, PDE (2022) 5 / 23

Fourier Integral
If we plot F (σ) against σ, we can clearly see that the sum
∞
X π
F (σn )∆σ, ∆σ = (8)
L
n=1

is an approximation to the area under the curve y = F (σ).

Lecture 18 MA 201, PDE (2022) 6 / 23

Fourier Integral
Thus as, L → ∞, (7) can be written as

∞
X
lim fp (t) = f (t) = lim F (σn )∆σ
L→∞ L→∞
n=−∞
∞ ∞  ∞ 
1 iσt
Z Z Z
= F (σ)dσ = e f (τ )e−iστ dτ dσ, (9)
−∞ −∞ 2π −∞

which is known as the complex form of Fourier integral for the function f .

A function which has its period going to infinity is called a non-periodic

function.

Note that Fourier integral is a valid representation of the non-periodic function

f (t), a function whose period goes to infinity provided that

(a) in every finite interval, f (t) is defined and is piecewise continuous,

R∞
(b) the improper integral −∞ |f (t)|dt exists.

Lecture 18 MA 201, PDE (2022) 7 / 23

Fourier Integral
Now proceed to obtain the trigonometric representation of the Fourier integral:
Equation (9) can, after writing e−iσ(τ −t) = cos σ(τ − t) − i sin σ(τ − t), be expressed as

Z ∞ Z ∞ 
1
f (t) = f (τ ) e−iσ(τ −t) dσ dτ
2π −∞
Z ∞  Z−∞
∞ 
1
= f (τ ) cos σ(τ − t)dσ dτ.
2π −∞ −∞

(Since sin σ(τ − t) is an odd function of σ, the part containing sine term will vanish.)

Also note that cos σ(τ − t) is an even function of σ, therefore

Z ∞ Z ∞ 
1
f (t) = f (τ ) e−iσ(τ −t) dσ dτ
2π −∞ −∞
1 ∞
Z Z ∞ 
= f (τ ) cos σ(τ − t)dσ dτ. (10)
π −∞ 0

Now, the integral has a representation with cosine terms only (with respect to σ).

Lecture 18 MA 201, PDE (2022) 8 / 23

Fourier Integral
Fourier cosine and sine integrals: Expanding cos σ(τ − t) in (10),

∞ Z 0 ∞ 
1
Z Z
f (t) = f (τ ) cos στ cos σt dτ +
f (τ ) cos στ cos σt dτ dσ
π 0 −∞ 0
Z Z 0
1 ∞
Z ∞ 
+ f (τ ) sin στ sin σt dτ + f (τ ) sin στ sin σt dτ dσ. (11)
π 0 −∞ 0

Changing τ to −τ in the first and third integrals:

1 ∞
Z Z ∞ Z ∞ 
f (t) = f (−τ ) cos στ cos σt dτ + f (τ ) cos στ cos σt dτ dσ
π 0
Z Z0 ∞ 0
1 ∞
Z ∞ 
+ f (−τ )(− sin στ ) sin σt dτ + f (τ ) sin στ sin σt dτ dσ.
π 0 0 0
(12)

Remark: Above integral representation does not demand that the function f
should be defined in (−∞, ∞).
Therefore, we can now think of defining Fourier integral of a function defined in (0, ∞)
only.

Lecture 18 MA 201, PDE (2022) 9 / 23

Fourier Integral

For a function f defined in (0, ∞), we can use the concept of even or odd
functions (or even or odd function extension) to define Fourier integral and
subsequently, Fourier transform.

For an even function f or its even extension, Fourier integral of f is defined as

(the first and second integrals become the same while the third and fourth
integrals cancel each other in (12))
∞Z ∞
2
Z
f (t) = f (τ ) cos στ cos σt dτ dσ. (13)
π 0 0

Equation (13) is called the Fourier cosine integral of f .

Lecture 18 MA 201, PDE (2022) 10 / 23

Fourier Integral
For an odd function f or its odd extension, Fourier integral of f is defined as
(the first and second integrals cancel each other while the third and fourth
integrals become the same in (12))
∞Z ∞
2
Z
f (t) = f (τ ) sin στ sin σt dτ dσ. (14)
π 0 0

Equation (14) is called the Fourier sine integral of f .

With these three definitions of Fourier integrals ((9), (13) and (14)),
we are in a position to define Fourier transforms including Fourier sine and cosine
transforms.

Lecture 18 MA 201, PDE (2022) 11 / 23

Complex Fourier Transform
Recalling equation (9):
∞  ∞ 
1 iσt
Z Z
−iστ
f (t) = e f (τ )e dτ dσ,
−∞ 2π −∞

we can now define the complex Fourier transform pair.

The Fourier transform of a function f ∈ L1 (R) is defined by

∞
1
Z
F{f (t)} = g(σ) = √ e−iστ f (τ ) dτ. (15)
2π −∞

The inverse Fourier transform is defined as

∞
1
Z
f (t) = F −1 {g(σ)} = √ g(σ)eiσt dσ. (16)
2π −∞

Here g(σ) is in the frequency domain and f (t) in the time domain.
Lecture 18 MA 201, PDE (2022) 12 / 23
Fourier cosine transform

Recalling equation (13):

∞Z ∞
2
Z
f (t) = f (τ ) cos στ cos σt dτ dσ,
π 0 0

we can now define the Fourier cosine transform of f as

r
∞
2
Z
Fc {f (t)} = gc (σ) = f (τ ) cos στ dτ. (17)
π 0

The inverse Fourier cosine transform is defined as

r Z
2 ∞
f (t) = Fc−1 {gc (σ)} = gc (σ) cos σt dσ. (18)
π 0

Lecture 18 MA 201, PDE (2022) 13 / 23

Fourier sine transform

Recalling equation (14):

∞Z ∞
2
Z
f (t) = f (τ ) sin στ sin σt dτ dσ,
π 0 0

we can now define the Fourier sine transform of f as

r Z ∞
2
Fs {f (t)} = gs (σ) = f (τ ) sin στ dτ. (19)
π 0

The inverse Fourier sine transform is defined as

r Z
2 ∞
f (t) = Fs−1 {gs (σ)} = gs (σ) sin σt dσ. (20)
π 0

It is to be noted that
any integral transform is always accompanied by its inverse transform.

Lecture 18 MA 201, PDE (2022) 14 / 23

Some Properties of Fourier transform
Linearity Property:
If F{f1 (t)} = g1 (σ), F{f2 (t)} = g2 (σ), then

F{c1 f1 (t) ± c2 f2 (t)} = c1 F{f1 (t)} ± c2 F{f2 (t)} = c1 g1 (σ) ± c2 g2 (σ),

where c1 and c2 are constants.

Theorem I:
If F{f (t)} = g(σ), then
F{f (t − a)} = e−iσa g(σ).

Proof:
By definition,
Z ∞
1
F{f (t − a)} = √ e−iστ f (τ − a)dτ
2π −∞
Z ∞
1
= √ e−iσ(ξ+a) f (ξ)dξ, by taking τ − a = ξ
2π −∞
= e−iσa g(σ).

This is known as the shifting property.

Lecture 18 MA 201, PDE (2022) 15 / 23

Some Properties of Fourier transform
Theorem II:
If F{f (t)} = g(σ), then
1
F{f (at)} = g(σ/a).
a

Proof:
By definition,
Z ∞
1
F{f (at)} = √ e−iστ f (aτ )dτ
2π −∞
Z ∞
1
= √ e−i(σ/a)ξ f (ξ)dξ/a, by taking aτ = ξ
2π −∞
1
= g(σ/a).
a

This is known as the scaling property.

Lecture 18 MA 201, PDE (2022) 16 / 23

Some Properties of Fourier transform
Theorem III:
If F{f (t)} = g(σ), then
F{eiat f (t)} = g(σ − a).

Proof:
By definition,
Z ∞
1
F{eiat f (t)} =√ e−iστ eiaτ f (τ )dτ
2π −∞
Z ∞
1
= √ e−i(σ−a)τ f (τ )dτ
2π −∞
= g(σ − a).

This is known as the translation property.

Lecture 18 MA 201, PDE (2022) 17 / 23

Some Properties of Fourier transform
Theorem IV:
If F{f (t)} = g(σ) with f (t) as continuously differentiable and f (t) → 0 as |t| → ∞,
then
F{f ′ (t)} = iσg(σ).

Proof:
By definition,
Z ∞
1
F{f ′ (t)} =√ e−iστ f ′ (τ )dτ
2π −∞
 Z ∞ 
1
= √ [f (τ )e−iστ ]∞−∞ + iσ e−iστ
f (τ )dτ
2π −∞
= iσg(σ).

Lecture 18 MA 201, PDE (2022) 18 / 23

Some Properties of Fourier transform

Theorem V (Extension of Theorem IV):

If f (t) is n-times continuously differentiable and f (k) (t) → 0 as |t| → ∞ for
k = 1, 2, . . . , (n − 1), then the Fourier transform of the n-th derivative of f (t) is given
by
F{f (n) (t)} = (iσ)n F{f (t)} = (iσ)n g(σ).

Theorem VI (Particular form of Theorem V - important from the point

of view of solving differential equations):
′
If f (t) is twice continuously differentiable; f (t) → 0 as |t| → ∞ and f (t) → 0 as
|t| → ∞, then the Fourier transform of the second derivative of f (t) is given by
′′
F{f (t)} = (iσ)2 F{f (t)} = (−σ 2 )g(σ).


Lecture 18 MA 201, PDE (2022) 19 / 23
Some examples of Fourier integrals
Example: Find the Fourier integral representation of the following non-periodic
function
(
eat , t < 0,
f (t) = a > 0.
e−at , t > 0,

Solution :The Fourier transform of f (t)

F{f (t)} = g(σ)

Z ∞
1
= √ f (τ )e−iστ dτ
2π −∞
Z 0 Z ∞
1
= √ e(a−iσ)τ dτ + e−(a+iσ)τ dτ
2π −∞ 0
1 2a
= √ .
2π a2 + σ 2

Taking the inverse

Z ∞
1
f (t) = F −1 {g(σ)} = √ g(σ)eiσt dσ
2π −∞
Z ∞
2a cos σt
= dσ.
π 0 a2 + σ 2

Lecture 18 MA 201, PDE (2022) 20 / 23

Convolution

A convolution is an integral that expresses the amount of overlap of one function f2 as

it is shifted over another function f1 .

It, therefore, “blends” one function with another.

In other words, the output which produces a third function can be viewed as a
modified version of one of the original functions.

Definition:
The convolution of two functions f1 (t) and f2 (t), −∞ < t < ∞, is defined as
Z ∞ Z ∞
1 1
f1 (t)∗f2 (t) = √ f1 (τ )f2 (t−τ ) dτ = √ f1 (t−ξ)f2 (ξ) dξ = f2 (t)∗f1 (t),
2π −∞ 2π −∞
provided the integral exists for each t.

This is known as the convolution integral.

Lecture 18 MA 201, PDE (2022) 21 / 23

Convolution: Some algebraic properties

Property I (Commutative):

f1 ∗ f2 = f2 ∗ f1

Property II (Associative):

f1 ∗ (f2 ∗ f3 ) = (f1 ∗ f2 ) ∗ f3

Property III (Distributive):

(αf1 + βf2 ) ∗ f3 = α(f1 ∗ f3 ) + β(f2 ∗ f3 )

Lecture 18 MA 201, PDE (2022) 22 / 23

Convolution Theorem

If f1 (t) and f2 (t) are Fourier transformable, i.e., we have F {f1 (t)} = g1 (σ) and
F {f2 (t)} = g2 (σ), then

F{f1 (t) ∗ f2 (t)} = g1 (σ)g2 (σ).

In other words,
the Fourier transform of the convolution of two functions is equal to the product of the
individual Fourier transforms of those functions.

We can also write

∞
1
Z
f1 (t) ∗ f2 (t) = F −1 {g1 (σ)g2 (σ)} = √ f1 (τ )f2 (t − τ ) dτ.
2π −∞

That is,
the inverse Fourier transform of product of Fourier transforms of two functions is equal
to the convolution of these two functions.
Lecture 18 MA 201, PDE (2022) 23 / 23