4 Analytical methods (Part 1)

4.1 Fourier series/Fourier expansion

In this section, we are going to present one of the most powerful analytical methods –
Fourier series. Before doing that, let us first recall some complex notations.
p
Complex numbers. Let i = °1. For a complex number z = a + i b, its conjugate is
z̄ = a ° i b, and its magnitude is
p p
|z| = a2 + b2 = z z̄ .

Complex functions. f (x) is called a complex function, if we can write it in the form

f (x) = f1 (x) + i f2 (x)

where f1 (x) and f2 (x) are two real functions. The conjugate of f (x) is the complex
function f (x) = f1 (x) ° i f2 (x). We shall often use the relations

eix = cos x + i sin x , eix = e°ix = cos x ° i sin x .

Definition 4.1 (Orthogonal functions). Two complex functions f (x) and g(x) are
said to be orthogonal on the interval [a, b] if the following holds
Z b
(f, g) := f (x)g(x)dx = 0
a

where g(x) is the conjugate of g(x). For example, {eikx }1 k=1 is an orthogonal sequence
of functions on [°º, º] or [0, 2º], since
Z 2º
ikx ilx
(e , e ) = eikx eilx dx = 0 8 l 6= k
0

In fact, Z Z
2º 2º
1 Ø2º
Ø
eikx eilx dx = eikx e°ilx dx = ei(k°l)x Ø = 0 .
0 0 i(k ° l) 0

Similarly one can verify that the following three sequences

{cos kx}1
k=0 , {sin kx}1
k=0 , {cos kx, sin kx}1
k=0

are all orthogonal sequences of functions on [°º, º] or [0, 2º].

40
Definition 4.2 (Periodic functions). A function f (x) is called a periodic function
with period d if
f (x + d) = f (x) 8 x .

For example, eikx , cos kx and sin kx are all periodic functions with period 2º. But

cos 2kx, sin 2kx

are periodic functions with period 2º and also º.

Now, let us discuss the Fourier series. The idea of the Fourier series is to expand
a given function f (x) (maybe discontinuous) in terms of the cosine and sine functions.
We will consider the following two types of expansions:
1
X
f (x) = ck eikx = c0 + c°1 e°ix + c1 eix + c°2 e°2ix + c2 e2ix + · · · (4.1)
k=°1
f (x) = a0 + a1 cos x + b1 sin x + a2 cos 2x + b2 sin 2x + · · · (4.2)

Note that the right hand sides of (4.1) and (4.2) are all functions with period of 2º. So
function f (x) must be also a function with period 2º. Because of the periodicity, we can
consider any interval of length 2º for the Fourier expansions (4.1) and (4.2). We often
take [°º, º] or [0, 2º]. In our subsequent discussions, we will always use the interval
[°º, º].
Now suppose f (x) is a function with period 2º, i.e., f (x + 2º) = f (x) 8x . In this
case, the graph of f (x) in any interval of length 2º will be repeated in its neighboring
interval of length 2º.

Real Fourier series

We first discuss how to find the Fourier series (4.2).

We need to find all the coe±cients {ak } and {bk }. Recall that {cos kx, sin kx} are
orthogonal on [°º, º], namely for any k 6= l,
Z º
cos kx cos lxdx = 0, (4.3)
°º
Z º
cos kx sin lxdx = 0, (4.4)
°º
Z º
sin kx sin lxdx = 0 . (4.5)
°º

41
Find the coe±cient ak in (4.2). Multiply both sides of (4.2) by cos kx, integrate
then over [°º, º] and use the orthogonality (4.3)-(4.5). We have
Z º Z º
f (x) cos kxdx = ak cos kx cos kxdx.
°º °º

From this we obtain Z º

1
ak = f (x) cos kxdx (4.6)
º °º
since Z º Z º
2 1 + cos 2kx
cos kxdx = dx = º .
°º °º 2
Find the coe±cient bk in (4.2). Multiply both sides of (4.2) by sin kx, integrate then
over [°º, º] and use the orthogonality (4.3)-(4.5). We have
Z º Z º
f (x) sin kxdx = bk sin kx sin kxdx.
°º °º

From this we obtain Z º

1
bk = f (x) sin kxdx . (4.7)
º °º

Find the coe±cient a0 in (4.2). Multiply both sides of (4.2) by the constant 1, then
integrate over [°º, º] to obtain
Z º Z º
f (x)dx = a0 dx,
°º °º

therefore, Z º
1
a0 = f (x)dx , (4.8)
2º °º

that is, the first coe±cient a0 is the average of f (x) on [°º, º].
In summary, we can expand f (x) as follows:

f (x) = a0 + a1 cos x + b1 sin x + a2 cos 2x + b2 sin 2x + · · ·

where all the coe±cients are given by

Z º
1
a0 = f (x)dx ,
2º °º
Z
1 º
ak = f (x) cos kxdx ,
º °º
Z
1 º
bk = f (x) sin kxdx .
º °º

42
Complex Fourier series

Now we shall discuss how to find the complex Fourier series (4.1), namely,
1
X
f (x) = ck eikx . (4.9)
k=°1

We need to find all the coe±cients {ck }. By Definition 3.1, we know that {eikx } are
orthogonal on [°º, º], namely for any k 6= l,
Z º
ikx ilx
(e , e ) = eikx e°ilx dx = 0 8 k 6= l.
°º

Thus multiply both sides of (4.9) by e°ikx and use the orthogonality of {eikx }, we obtain
Z º Z º
°ikx
f (x)e dx = ck eikx e°ikx dx ,
°º °º

or Z º
1
ck = f (x)e°ikx dx . (4.10)
2º °º

That is, the Fourier series is

f (x) = c0 + c1 eix + c°1 e°ix + c2 e2ix + c°2 e°2ix + · · · (4.11)

with coe±cients ck defined by (4.10).

Remark 4.1. Note that in the Fourier series (4.11), the function f (x) can be a real
function. For a real function, one can choose the real Fourier expansion (4.2) or the
complex form (4.11).
Think about why we can choose the complex form (4.11) for a real function. Any
contradiction ?

4.2 Relation between the real and complex Fourier series

There are close relations between the real and complex Fourier series.

(a) The coe±cients ck in the complex form (4.11) can be derived from the coe±cients
ak and bk in the real form (4.2). In fact, we know

eikx = cos kx + i sin kx, e°ikx = cos kx ° i sin kx. (4.12)

43
 Multiply both sides of the second equation by f (x) and integrate over [°º, º]. We
obtain Z º Z º Z º
°ikx
f (x)e dx = f (x) cos kxdx ° i f (x) sin kxdx ,
°º °º °º

That implies
2ck = ak ° i bk . (4.13)
This can be written as
1 i
ck = ak ° bk .
2 2
Similarly, we can derive from the first equation of (4.12):
1 i
c°k = ak + bk .
2 2

(b) The real coe±cients ak and bk in (4.2) can be recovered from the complex coe±-
cients ck in (4.11). Using the formula
1 ° ikx ¢ 1 ° ikx ¢
cos kx = e + e°ikx , sin kx = e ° e°ikx .
2 2i
Therefore
Z
1 º
ak = f (x) cos kxdx = ck + c°k ,
º °º
Z
1 º 1
bk = f (x) sin kxdx = (c°k ° ck ).
º °º i

4.3 Examples of Fourier series

We now give some examples to illustrate the calculations of the Fourier series.

Example 4.1. Find the Fourier series of f (x) = cos2 x.

Solution. By definition, we have

Z º Z º
1 1 1 + cos 2x 1
a0 = f (x)dx = dx = ,
2º °º 2º °º 2 2
8
Z º Z º <0 , k 6= 2
1 1
ak = f (x) cos kxdx = (1 + cos 2x) cos kxdx =
º °º 2º °º :1 , k = 2 ,
2
Z º Z º
1 1
bk = f (x) sin kxdx = (1 + cos 2x) sin kxdx = 0 .
º °º 2º °º

44
Therefore the Fourier series of f (x) is
1 1
f (x) = + cos 2x .
2 2
This is a well-known formula. ]
• Try the Fourier expansions of the functions sin2 x, cos 2x, sin x + cos x, · · · .

Example 4.2. Find the Fourier series of f (x) = ±(x) on [°º, º]. This function is
called a delta function and it is one of the most important functions used in physics and
engineering. The delta function has the following properties
Z º
g(x)±(x)dx = g(0) 8 g 2 C [°º, º]
°º

and
±(x) = 0 for any x 6= 0.

Solution. By definition, the Fourier coe±cients are

Z º
1 1
a0 = ±(x)dx = ,
2º °º 2º
Z
1 º 1 1
ak = ±(x) cos kxdx = cos 0 = ,
º °º º º
Z º
1
bk = ±(x) sin kxdx = 0 ,
º °º
therefore 1
1 1X
±(x) = + cos kx , x 2 [°º, º] (4.14)
2º º k=1
In the complex case, Z º
1 1
ck = f (x)e°ikx dx = ,
2º °º 2º
so we have 1
1 X ikx
±(x) = e , x 2 [°º, º] . (4.15)
2º k=°1
We have from (4.15) that
1
1 1 X ° ikx ¢
±(x) = + e + e°ikx
2º 2º k=1
1
1 1X
= + cos kx ,
2º º k=1

45
this is the same as (4.14).
Can the series (4.15) really reflect the behavior of ±(x) ?
Let us consider the partial sum of (4.15):
N
X N
ikx 1 °iN x X i(N +k)x
PN (x) = e = e e
k=°N
2º k=°N
2N
X
°iN x
=e eikx
k=0
1 ° ei(2N +1)x
= e°iN x
1 ° eix
1 1
ei(N + 2 )x ° e°i(N + 2 )x
=
eix/2 ° e°ix/2
sin(N + 12 )x
= .
sin 12 x
Then study the following questions

1. For each given N , show that

1
lim PN (x) = 2(N + ).
x!0 2
So PN (x) will tend to infinity at x = 0 when N goes larger and larger.

2. Plot the figure for PN (x) using Matlab; and calculate the integral
Z º
1
PN (x)f (x)dx
2º °º
approximately for N = 10, 20, 30, 40, 50, 100. Observe if PN (x) satisfies that
Z º
1
lim PN (x)f (x)dx = f (0).
N !1 2º °º

1
if so, P (x)
2º N
approximates ±(x).

Odd and even functions. A function f (x) is called an even function if it satisfies

f (°x) = f (x), 8x.

And it is called an odd function if it satisfies

f (°x) = °f (x), 8x.

It is easy to check the following properties:

46
 For any odd function f (x) on [°º, º], we have
Z º
f (x)dx = 0 .
°º

For any even function f (x) on [°º, º], we have

Z º Z º
f (x)dx = 2 f (x)dx .
°º 0

Example 4.3. Find the Fourier series of the odd function

f (x) = x , x 2 [°º, º] .

Solution. The Fourier coe±cients are

Z º
1
a0 = f (x)dx = 0 (why ?)
2º °º
Z
1 º
ak = f (x) cos kxdx = 0 (why ?)
º °º

Finally for the coe±cients bk , we have

Z Z
1 º 2 º
bk = f (x) sin kxdx = x sin kxdx
º °º º 0

By integration by parts, we obtain

Z º غ
2 2 Ø 2 cos kº
bk = cos kxdx ° x cos kxØ = ° ,
kº 0 ºk 0 k
that is,
2 2 2
b1 = 2, b2 = ° , b3 = , · · · , bk = (°1)k+1 ,
2 3 k
so the required Fourier series is
sin 2x sin 3x
x = b1 sin x + b2 sin 2x + · · · = 2(sin x ° + ° · · · ), °º < x < º .
2 3
Remark 4.2. Note that the Fourier series above does not converge at x = °º, º, as the
series is 0 at x = °º and º.

47
4.4 Sine series and cosine series
Every function f (x) can be written as a sum of an even and an odd function, i.e.,

f (x) = fe (x) + fo (x),

with
f (x) + f (°x) f (x) ° f (°x)
fe (x) = , fo (x) = .
2 2

A very important observation:

The Fourier series of an even function has only cosine terms, since
Z
1 º
bk = f (x) sin kxdx = 0 .
º °º

The Fourier series of an odd function has only sine terms, since
Z
1 º
ak = f (x) cos kxdx = 0 .
º °º

Example 4.4. The function f (x) = 1 is known on the half-period 0 < x < º. Find its
Fourier series when

(a) f (x) is extended to (°º, º) as an even function;

(b) f (x) is extended to (°º, º) as an odd function.

Solution. By definition of even and odd functions, we have

(a) f (x) is an even function,

Z º Z
1 1 º
a0 = f (x)dx = 1dx = 1 ,
2º °º º 0
Z Z
1 º 2 º
ak = f (x) cos kxdx = cos kxdx = 0 ,
º °º º 0
Z
1 º
bk = f (x) sin kxdx = 0 ,
º °º
therefore the Fourier series of f (x) is

f (x) = 1, °º < x < º.

This recovers the original constant function.

48
(b) f (x) is an odd function,
Z º
1
a0 = f (x)dx = 0 ,
2º °º
Z
1 º
ak = f (x) cos kxdx = 0 ,
º °º
Z Z
1 º 2 º
bk = f (x) sin kxdx = sin kxdx
º °º º 0
8
2 ° ¢ <0, k is even
k
=° (°1) ° 1 =
kº : 4 , k is odd ,
kº

so the Fourier series of f (x) is

4 n sin x sin 3x sin 5x o
f (x) = + + + ··· , °º < x < º.
º 1 3 5
This is very diÆerent from the original constant function.

4.5 Some properties of Fourier series

Introduce
Z º n
X
© ™2
E(A0 , A1 , · · · , An , B0 , B1 , · · · , Bn ) = f (x) ° (Ak cos kx + Bk sin kx) dx .
°º k=0

Then we claim that

The best trigonometric approximation of f (x) on [°º, º] in the mean-

square sense is its Fouries series, i.e.,

E(a0 , a1 , · · · , an , b0 , b1 , · · · , bn ) = min E(A0 , A1 , · · · , An , B0 , B1 , · · · , Bn )

8 Ak ,Bk 2R1

where {ak } and {bk } are the Fourier coe±cients of f (x).

To see this, let us assume {Ak , Bk }nK=0 is a minimizer of E, then

Z º Xn
@E © ™
=2 f (x) ° (Ak cos kx + Bk sin kx) cos kxdx
@Ak °º K=0
Z º
© ™
=2 f (x) ° Ak cos kx cos kxdx
Z°º
º
=2 f (x) cos kxdx ° 2ºAk = 0 ,
°º

49
therefore Z º
1
Ak = f (x) cos kxdx = ak ,
º °º
6 0. Similarly we have
for k =
Z º
1
A0 = f (x)dx = a0 ,
2º °º

and Z º
1
Bk = f (x) sin kxdx = bk , 8k .
º °º
This indicates that the minimizer {Ak , Bk }nk=0 is the Fourier coe±cients of f (x). ]
• Think about why we can claim what we get is the minimizer, not the maximizer.
Think about the diÆerence between E({Ai }, {Bi }) and E({ai }, {bi }).

Our second claim is:

Let Fn (x) be the truncated Fourier series

Fn (x) = a0 + a1 cos x + b1 sin x + · · · + an cos nx + bn sin nx ,

then we have Z º Z º
Fn2 (x)dx ∑ f 2 (x)dx .
°º °º

First we have by using the orthogonality that

Z º
° ¢
f (x) ° Fn (x) Fn (x)dx
°º
Z n
X Z n
º ° º ©X ™2
= f (x) ak cos kx + bk sin kx)dx ° aK cos kx + bk sin kx dx
°º k=0 °º K=0
= 0,

thus
Z º Z º
2
° ¢2
f (x)dx = f (x) ° Fn (x) + Fn (x) dx
°º
Z°º
º ° Z º Z º
¢2 ° ¢
= f (x) ° Fn (x) dx + 2 f (x) ° Fn (x) Fn (x)dx + Fn2 (x)dx
Z°º
º ° Z º°º °º
¢2
= f (x) ° Fn (x) dx + Fn2 (x)dx
Z°º
º
°º

∏ Fn2 (x)dx .
°º

50
]

• Think about the interesting question. If we define a sequence {Æn } by

Z º
Æn = Fn2 (x)dx,
°º

then the sequence {Æn }1

n=0 must be monotonely increasing.

51
4.6 Solution of the Laplace’s equation
In this section, we are going to apply the Fourier series to solve an important diÆerential
equation, i.e., the Laplace’s equation:
@ 2u @ 2u
+ = 0, (x, y) 2 ≠ (4.16)
@x2 @y 2
with the boundary condition

u(x, y) = u0 (x, y), (x, y) 2 @≠ (4.17)

where ≠ is the unit circle, i.e.,

≠ = {(x, y); x2 + y 2 < 1}.

Since ≠ is a circle, it is easier to use the polar coordinates:

x = r cos µ , y = r sin µ .

Under the transformation, we have

u(x, y) = u(r cos µ, r sin µ) = w(r, µ) .

Further, the domain ≠ and the equation (4.16) are transformed into

! = {(r, µ); 0 ∑ r < 1, ° º ∑ µ < º}

and
1 @ ° @u ¢ 1 @ 2 u
r + 2 2 =0. (4.18)
r @r @r r @µ
The boundary condition (4.17) changes into

w(1, µ) = u0 (cos µ, sin µ) . (4.19)

We are now going to find the solutions of (4.18). First, we can easily check that the
following functions
1, r cos µ, r sin µ, r2 cos 2µ, r2 sin 2µ, · · · (4.20)
are all solutions of (4.18). For example, we take w(r, µ) = rk cos kµ for k ∏ 2, then

wr = krk°1 cos kµ ,
1 @
(rwr ) = k 2 rk°2 cos kµ ,
r @r

52
while
1
2
wµµ = °k 2 rk°2 cos kµ ,
r
k
therefore w(r, µ) = r cos kµ is a solution to the equation (4.18). Note that (4.18) is
a linear equation, so any combination of two solutions w1 (r, µ) and w2 (r, µ) is still a
solution (why ?). Thus the following combination of the above special solutions is a
genernal solution:

w(r, µ) = a0 + a1 r cos µ + b1 r sin µ + · · · + ak rk cos kµ + bk rk sin kµ + · · · , (4.21)

where ak and bk are arbitrary constants.

But we have to determine the coe±cients ak and bk . This can be done by using the
boundary condition (4.19). For this, we let r = 1 in (4.21) and obtain

w(1, µ) = a0 + a1 cos µ + b1 sin µ + · · · + ak cos kµ + bk sin kµ + · · · .

We know that w(1, µ) = u0 (cos µ, sin µ), so the coe±cients ak and bk are nothing else but
the Fourier coe±cients of u0 , i.e.,
Z º
1
a0 = u0 (cos µ, sin µ)dµ , (4.22)
2º °º
Z
1 º
ak = u0 (cos µ, sin µ) cos kµdµ , k = 1, 2, · · · (4.23)
º °º
Z
1 º
bk = u0 (cos µ, sin µ) sin kµdµ , k = 1, 2, · · · . (4.24)
º °º

This indicates that w(r, µ) in (4.21) is the desired solution of the boundary value problem
(4.18) with the coe±cients ak and bk given by (4.22)-(4.24).

Example 4.5. Find the solution of the following Laplace equation:

( ° @u ¢ 1 @ 2 u
1 @
r @r
r @r + r2 @µ2 = 0, 0 ∑ r < 1, °º ∑ µ < º
u(1, µ) = µ, °º ∑ µ < º .

and ( ° @u ¢ 1 @ 2 u
1 @
r @r
r @r + r2 @µ2 = 0, 0 ∑ r < 1, °º ∑ µ < º
u(1, µ) = ±(µ), °º ∑ µ < º .

53
4.7 Orthogonal functions
In this section, we introduce some further knowledge on orthogonal functions.
For a given positive function w(x) on [a, b], we define an inner product
Z b
(f, g)! = !(x) f (x) g(x)dx
a

for any two real functions f (x) and g(x) on [a, b]. And !(x) will be called a weight
function. We will often use the following norm:
nZ b o 12
kf k! = !(x) f 2 (x) dx .
a

Definition 4.3 (Weighted orthogonal functions). Let f (x) and g(x) be two real
functions on [a, b]. f (x) is said to be orthogonal to g(x) with respect to the inner product
(·, ·)! if (f, g)! = 0.
A sequence of functions {fk }1 k=0 is said to be orthonormal with respect to the inner
product (·, ·)! if the following holds:

(fm , fn )! = 0 8 m 6= n

and each fk is unitary, i.e.,

kfk k! = 1 .

• Check if function cos x is orthogonal to g(x) = sin x with respect to the inner product
(·, ·)! for !(x) = 1, x, x2 .
• Verify that any sequence of orthogonal functions {gk }1
k=1 on the interval [a, b] are
linearly independent.

Now we are going to demonstrate that

Any sequence of linearly independent functions {¡k }1 k=0 defined on

1
[a, b] can generate a sequence of functions {qk }k=0 which are orthonor-
mal with respect to the inner product (·, ·)! .

Gram-Schmidt orthogonalization is one of such orthogonalizing techniques. Below

we introduce the Gram-Schmidt orthogonalization.

54
Gram-Schmidt orthogonalization process.
Given a sequence {¡k }1k=0 of linearly independent functions defined on [a, b], we are
going to construct a sequence of orthonormal functions {qk }1
k=0 as follows:

0) Set
qe0 (x) = ¡0 (x) .
Normalize qe0 (x):
qe0 (x)
q0 (x) = ;
keq0 k !
1) Set
qe1 (x) = ¡1 (x) ° Æ10 q0 (x) ,
choose Æ10 such that
Z b
(e
q1 , q0 )! = !(x) q̃1 (x)q0 (x)dx = 0,
a

that gives, Z b
Æ10 = (¡1 , q0 )! = !(x) ¡1 (x)q0 (x)dx.
a
Normalize qe1 (x):
qe1 (x)
q1 (x) = .
keq1 k !
k) Suppose q0 , q1 , · · · , qk are constructed such that

(qi , qj )! = 0 8i 6= j and kqi k! = 1 .

We then construct qk+1 by

n o
qek+1 (x) = ¡k+1 (x) ° Æk+1,0 q0 (x) + · · · + Æk+1,k qk (x)

with
Æk+1,i = (¡k+1 , qi )! , i = 0, 1, · · · , k.
Normalize qk+1 :
qek+1 (x)
qk+1 (x) = .
keqk+1 k!

Then the sequence {qk }1

k=0 constructed above is an orthonormal sequence, i.e.,

(qi , qj )! = 0 8 i 6= j ; kqi k! = 1 .

55
Example 4.6. Given the sequence of polynomials

1, x, x2 , · · · , xk , · · · ,

on the interval [°1, 1], use the Gram-Schmidt orthogonalization process to construct an
orthonormal sequence of polynomials, and write down the first three constructed polyno-
mials explicitly.

Solution (exercise). The three polynomials are

p p ≥
1 3 45 1¥
P0 (x) = p , P1 (x) = p x , P2 (x) = p x2 ° .
2 2 8 3

Example 4.7. Check if the Chebyshev polynomials

T0 (x) = 1, T1 (x) = x, T2 (x) = 2x2 ° 1, T3 (x) = 4x3 ° 3x

p
are orthogonal on [°1, 1] with respect to the weight function w(x) = 1/ 1 ° x2 .

Solution. Use the transformation x = cosµ.

Example 4.8. Based on a given sequence of functions {¡i (x)}1 k=0 , which is orthogo-
nal with respect to the inner product (·, ·)! , use the Gram-Schmidt orthogonalization to
construct an orthonormal sequence of functions with respect to (·, ·)! . (exercise)

Example 4.9. Expand a given function f (x) on [a, b] in terms of a given orthogonal
sequence of functions {¡k (x)}1
k=1 with respect to the inner product (·, ·)! .

Solution. Let
f (x) = Æ1 ¡1 (x) + Æ2 ¡2 (x) + Æ3 ¡3 (x) + · · · .
Think about how to find the coe±cients {Æk }. ]

56
4.8 Fourier transform
Fourier transforms play a very important role in mathematics, physics and engineering.

4.8.1 Definition and examples

Recall that
eikx = cos(kx) + i sin(kx).
From this expression, we can easily see that the magnitude of k determines the intensity
of the oscillation of function exp(ikx), and k measures the frequencies of the oscillation.
To better understand the relation between the magnitude of k and the oscillation of
exp(ikx), , one may plot and compare the figures of sin ºx, sin 4ºx and sin 8ºx.

Definition 4.4. For a given function f (x) defined on (°1, 1), the Fourier transform
of f is a function fb depending on frequency:
Z 1
b
f (k) = f (x)e°ikx dx , °1 < k < 1 . (4.25)
°1

The inverse Fourier transform of fb(k) recovers the original function f (x):
Z 1
1
f (x) = fb(k)eikx dk , °1 < x < 1 . (4.26)
2º °1

Example 4.10. Find Fourier transform of the delta function f (x) = ±(x).

Solution.
Z 1 Z 1
fb(k) = °ikx
f (x)e dx = ±(x)e°ikx dx = 1, for all frequencies k .
°1 °1

So the Fourier transform of the delta function is a constant function.

Example 4.11. Find the Fourier transform of the function:

8
<1 , |x| ∑ a
f (x) = square pulse = .
:0 , |x| > a

Solution. Z Z
1 a
2 sin ka
fb(k) = f (x)e °ikx
dx = e°ikx dx = .
°1 °a k
• Think about whether this function fb(k) makes sense at k = 0.

57
Example 4.12. For a > 0, find the Fourier transform of the function:
8
< e°ax , x∏0
f (x) =
:°eax , x<0.

Solution. By definition, we have

Z 1 Z 1 Z 0
b
f (k) = f (x)e °ikx
dx = °ax°ikx
e dx + °eax°ikx dx
°1 0 °1
1 1 °2ik
= ° = 2 .
a + ik a ° ik a + k2

• Justify the above process yourself.

Example 4.13. Find the Fourier transform of

8
< 1, x>0
f (x) = sign function =
: ° 1, x<0.

Solution. We have
Z 1 Z 1 Z 0
fb(k) = °ikx
f (x)e dx = e °ikx
dx + °e°ikx dx.
°1 0 °1
Ø1
But what is e°ikx Ø0 ? It is di±cult to know.
To solve this problem, we consider the function
8
< e°ax , x>0
fa (x) = ,
: ° eax , x<0

it is easy to see that

lim fa (x) = f (x) ,
a!0+

then we can compute as follows:

Z 1 Z 1
b
f (k) = f (x)e °ikx
dx = lim+ fa (x)e°ikx dx
°1 °1 a!0
Z 1
°2ik
= lim+ fa (x)e°ikx dx = lim+ 2
a!0 °1 a!0 a + k 2
°2i 2
= = .
k ik

58
Example 4.14. Find the Fourier transformation of the constant function

f (x) = 1, 8 x 2 (°1, 1) .

Solution. We have
Z 1 Z 1 Z 0
fb(k) = e °ikx
dx =
dx + °ikx
e°ikx dx
e
°1 0 °1
Z Z 0
© 1 °ax °ikx ™
= lim+ e e dx + eax e°ikx dx
a!0 0 °1
8
© 1 1 ™ <0 k 6= 0
= lim+ + =
a!0 a + ik a ° ik :? k=0

What is fb(0) ? Note that fb(k) looks like a delta function. Let fb(k) = Ʊ(k), then by
the inverse Fourier transform we have
Z 1
1 Æ
1 = f (x) = fb(k)eikx dk = ,
2º °1 2º
so
Æ = 2º
or
fb(k) = 2º±(k), °1 < k < 1 .

4.8.2 Two identities for Fourier transforms

(1) For a function f (x) on (°1, 1) and its Fourier transform fb(k), we have
Z 1 Z 1
2º 2
|f (x)| dx = |fb(k)|2 dk . (4.27)
°1 °1

(2) The inner product of any functions f and g satisfies

Z 1 Z 1
2º f (x)ḡ(x)dx = fb(k)g(k)dk
b̄
°1 °1

where ḡ(x) is the conjugate of g(x).

59
Example 4.15. Check the relation (4.27) for the following function
8
<e°ax , x>0
f (x) =
:0 , x<0.

Solution. We have
Z 1 Z 1
º
2º 2
|f (x)| dx = 2º e°2ax dx = ,
°1 0 a

while Z 1 Z 1
fb(k) = f (x)e °ikx
dx = e°ax e°ikx dx
°1 0
1 Ø1 1
=° e°ax°ikx Ø0 = ,
a + ik a + ik
therefore Z Z 1 Z 1
1 Ø Ø dk dk º
Øfb(k)Ø2 dx = = = ,
|a + ik|2 2 2
°1 °1 °1 a + k a
that verifies (4.27). Here we have used the transformation k = a cos µ/ sin µ. ]

4.8.3 Important properties of Fourier transform

This subsection discusses some more properties of Fourier transforms.

(1) One can directly verify from definition that for any complex number Æ,

c (k) = Æfb(k).
Æf

(2) One can directly verify from definition that

f[
+ g(k) = fb(k) + gb(k).

(3) The Fourier transform of df

dx
is ik fb(k), i.e.,

b
df
(k) = ik fb(k) .
dx
To see this, we use Z 1
1
f (x) = fb(k)eikx dk
2º °1

60
 to obtain Z 1
df 1
(x) = ik fb(k)eikx dk.
dx 2º °1

Comparing with definition

Z 1 b
df 1 df
(x) = (k)eikx dk
dx 2º °1 dx

gives
b
df
(k) = ik fb(k).
dx
Rx fb(k)
(4) The transform of F (x) = a
f (x)dx is ik
+ C ±(k), i.e.,

fb(k)
Fb(k) = + C ±(k)
ik
To see this, we use
dF (x)
= f (x),
dx
or
d° ¢
F (x) + C = f (x) 8C 2 R1 .
dx
Taking the transform on both sides,
° ¢
ik Fb(k) + 2C º±(k) = fb(k) ,

this is ,
fb(k)
Fb(k) = + C±(k) 8 C 2 R1 .
ik

(5) The Fourier transform of F (x) = f (x ° d) is e°ikd fb(k).

In fact, we have Z 1
Fb(k) = F (x)e°ikx dx
Z°1
1
= f (x ° d)e°ikx dx
Z°1
1
= f (y)e°ik(y+d) dy
°1

=e °ikd
fb(k) .

61
 (6) The transform of g(x) = eixd f (x) is fb(k ° d).
By definition, we have
Z 1 Z 1
°ikx
gb(k) = g(x)e dx = f (x)e(id°ik)x dx
Z°1
1
°1
0
= f (x)e°ik x dx = fb(k 0 ) = fb(k ° d) .
°1

(7) The convolution of G and h is the function

Z 1
u(x) = G(x ° y)h(y)dy ,
°1

we often write
u(x) = (G § h)(x)
or ≥ ¥ Z 1
u(x) = G § h (x) = G(x ° y)h(y)dy .
°1

We now show that

u b b
b(k) = G(k) h(k).
In fact, we have
Z 1 Z 1Z 1
°ikx
b(k) =
u u(x)e dx = G(x ° y)h(y)e°ikx dydx
Z°1
1 Z 1
°1 °1
Z 1 Z 1
°ikx
= G(x ° y)h(y)e dxdy = h(y) G(x ° y)e°ikx dxdy
Z°1
1
°1
Z 1 °1 °1
0
= h(y)e°iky dy G(x0 )e°ikx dx0 (let x ° y = x0 )
°1 °1
b b
=G(k) h(k) , °1 < k < 1 .

4.8.4 Application of Fourier transform for diÆerential equations

Fourier transforms can be applied to solve diÆerent types of diÆerential equations. Here
we consider one example.
Consider the diÆerential equation

d2 u
° + a2 u = h(x) , °1 < x < 1 . (4.28)
dx2

62
In order to solve the equation, we apply the Fourier transform
Z 1
b(k) =
u u(x)e°ikx dx
°1
to each term of the equation to obtain
°(ik)2 u b(k) = b
b(k) + a2 u h(k) .
This gives
b
h(k)
b(k) =
u . (4.29)
a + k2
2

Let G(x) be a function such that

b 1
G(k) = ,
a2 + k2
then we know
bh(k)
b(k) =
u b b
= G(k)h(k).
a + k2
2

By the convolution property, the solution u(x) can be given by

Z 1
u(x) = (G § h)(x) = G(x ° y)h(y)dy . (4.30)
°1

To find G(x), we consider function f (x) = e°a|x| . By definition, we have

Z 1
b
f (k) = f (x)e°ikx dx
°1
Z 1 Z 0
°(a+ik)x
= e dx + e(a°ik)x dx
0 °1
1 Øx=1 1 Øx=0
Ø Ø
= ° e°(a+ik)x Ø + e(a°ik)x Ø
a + ik x=0 a ° ik x=°1
1 1 2a
= + = 2 .
a + ik a ° ik a + k2
this shows
d1 1
f (k) = 2 ,
2a a + k2
so we have
1
G(x) = e°a|x| .
2a
Now we get from (4.30) that
Z 1
1
u(x) = e°a|x°y| h(y)dy, °1 < x < 1 . (4.31)
2a °1
]
• Check if the function u(x) in (4.31) is indeed a solution to the diÆerental equation
(4.28).

63