Chapter 14

Integral Transforms
14.1 Goal
In this chapter, we introduce the idea of integral transforms and discuss some
of their applications. More specically, we will address the following:
1. What are integral transforms?
2. How many integral transforms are there?
3. What do they represent?
4. What are their applications?
5. What are their properties?
6. How do we compute them?
The last item will not be addressed except for a few examples to illustrate
how it can be done. We will simply use tables of known integral transforms.
We will look deeper into applications in the next chapters as we will focus on
one integral transform: the Fourier transform.
14.2 Integral Transforms: Denitions and Prop-
erties
We begin by giving a general idea of what integrals transforms are, and how
they are used. There is a whole family of integral transforms which includes
the Fourier transform, the Laplace transform, the Mellin transform, the Hankel
transform, ... In addition, as we will see, each transform goes with its inverse.
105
106 CHAPTER 14. INTEGRAL TRANSFORMS
Denition 105 An integral transform is an operator of the form:
F [)] = 1 (:) =
_
B
A
) (t) 1 (:, t) dt (14.1)
The function 1 is called the kernel of the transformation. It is the major
part of the transform. What makes the dierence between the various transforms
mentioned is mostly their kernel. It is also the interval of integration, [, 1] in
the general denition.
Remark 106 You will note that we used two notations: F [)] and 1 (:). As an
operator denoted F, an integral transform acts on a function. This is why we
used the notation F[)]. Some texts also use F[)] (:) because F[)] is a function
of a new variable we call : in the denition. For that reason, we also use the
notation 1 (:). In other words, we apply an integral transform to a function
) (t) and we obtain a new function 1 (:).
One of the features of integral transforms, one we will use in this chapters
and the next two, is that integral transforms eliminate partial derivatives with
respect to one of the variables. The new equation has one less variable. So, if
we start with a PDE having two variables and partial derivatives with respect
to both, then after applying an integral transform, we would have an ODE. For
example, if we apply an integral transform to the heat equation
n
t
= c
2
n
xx
to eliminate the time derivative, then we would be left with an ODE in r. If we
applied an integral transform to the equation
n
xx
+n
yy
+n
zz
= 0
to eliminate the derivative in r, we would be left with a PDE only involving j
and .. We could apply an integral transform again to eliminate the derivative
in j. We would be left with an ODE. Once the new equation is solved, we
need to express the solution in terms of the original variables and functions.
This is done using the inverse integral transform. This is why each transform
goes in pair with its inverse. A transform and its inverse form what is called
a transform pair. Both are needed when solving an equation. The following
ow chart illustrates better than words how integral transforms can be applied.
14.2. INTEGRAL TRANSFORMS: DEFINITIONS AND PROPERTIES 107
Figure 14.1: Using an Integral Transform to Solve a PDE
We nish this section by listing some common transforms and their inverse.
Transform Interval Transform Inverse Transform
Fourier Sine [0, 1) F
s
[)] = 1 (.) =
2

_
1
0
) (r) sin.rdr F
1
s
[1] = ) (r) =
_
1
0
1 (.) sin.rd.
Fourier Cosine [0, 1) F
c
[)] = 1 (.) =
2

_
1
0
) (r) cos .rdr F
1
c
[1] = ) (r) =
_
1
0
1 (.) cos .rd.
Fourier (1, 1) F [)] = 1 (.) =
1
p
2
_
1
1
) (r) c
i!x
dr F
1
[1] = ) (r) =
1
p
2
_
1
1
1 (.) c
i!x
d.
Finite Sine [0, 1] F
s
[)] = o
n
=
2
1
_
L
0
) (r) sin
:r
1
dr F
1
s
[o
n
] = ) (r) =
1

n=1
o
n
sin
:r
1
Finite Cosine [0, 1] F
c
[)] = C
n
=
2
1
_
L
0
) (r) cos
:r
1
dr F
1
c
[1] = ) (r) =
1

n=1
C
n
cos
:r
1
Laplace [0, 1) L[)] = 1 (:) =
_
1
0
) (r) c
sx
dr L
1
[1] = ) (r) =
1
2i
_
c+i1
ci1
1 (:) c
sx
d:
Hankel [0, 1) H [)] = 1
n
() =
_
1
0
rJ
n
(r) ) (r) dr H
1
[1
n
] = ) (r) =
_
1
0
J
n
(r) 1
n
() d
Mellin [0, 1) ' [)] = 1 (:) =
_
1
0
r
s1
) (r) dr '
1
[1] = ) (r) =
1
2i
_
c+i1
ci1
r
s
) (:) d:
Remark 107 For the inverse Laplace and Mellin transforms, c is a real num-
ber.
Remark 108 For the Hankel transform, J
n
is the Bessel function of the rst
kind, or order :
1
2
.
We are not going to study all these transforms in detail. In this chapter, we
will only study the sine and cosine transforms. In the next two, we will study
the Fourier transform.
108 CHAPTER 14. INTEGRAL TRANSFORMS
14.3 The Spectrum of a Function
Recall, when we rst introduced the separation of variables method, when we
applied the initial condition, we had to express c(r) as
c(r) =
1
sinr +
2
sin2r +
3
sin3r +...
=
1

n=1

n
sin:r
We found that this was possible with

n
= 2
_
1
0
c(r) sin:rdr
This was called the Fourier sine expansion of c(r). When we are computing
the coecients
n
, we are nding the contribution sin:r makes to c(r). For
example, if
c(r) = 2 sinr + 3 sin5r
then only sinr and sin5r play a role in c(r), for every other :, sin: does
not play a role. So, in a way, nding these coecients
n
tells us for each
natural number : if sin:r plays a role in c(r) and how much of a role.
Sometimes, instead of writing the Fourier sine expansion of a function, we
need the Fourier cosine expansion. Other times, as we will see in the next
chapters, we express a function in terms of sine and cosine functions. In other
words, given a function ) (r), we write
) (r) =
1

n=0
[
n
sin:r +1
n
cos :r] (14.2)
We will discuss this in greater detail in the next chapters. This is called nding
the Fourier series expansion of ) (r). When we nd the coecients
n
and 1
n
,
we nd by how much of sin:r and cos :r ) (r) is made up .
Denition 109 In the expansion given in equation 14.2, the quantity
_

2
n
+1
2
n
is called the spectrum of the function ) (r). It measures the amount of ) (r) with
frequency :.
In the case of Fourier series, the spectrum is discrete. It is only dened
for integer values. We will see in later chapters that if a function is peri-
odic, it can be expanded as a Fourier series and hence has a discrete spectrum.
For functions which are not periodic, we must use integral transforms. The
equivalent of Fourier series for non-periodic functions is the Fourier transform
1 (.) =
1
p
2
_
1
1
) (r) c
i!x
dr. This is, most of the time, a complex valued
function in other words, 1 (n) = C (n) + io (n). In this case, we dene the
spectrum as
_
C (n)
2
+o (n)
2
. It measures the amount of ) (r) with frequency
n. The only dierence is that this time, it is a continuous function.
14.4. SOME PROPERTIES OF THE SINE AND COSINE TRANSFORMS109
In later chapters, we will use the spectrum of a function in applications. We
now turn to the sine and cosine transforms, look at some of their properties and
applications.
14.4 Some Properties of the Sine and Cosine
Transforms
We begin with general properties the Fourier sine and cosine transforms have.
We will then give the transform of some standard functions. These properties
are very important. They will be used to transform a PDE problem into an
easier problem.
Proposition 110 All the integral transforms and their inverses in the table
above are linear operators.
Proof. We prove the result for the Fourier sine transform. The proof for
the remaining transforms is left as an exercise. If ) and q are two functions
and c is a real number, we need to prove that F
s
[) +q] = F
s
[)] + F
s
[q] and
F
s
[c)] = cF
s
[)].
F
s
[) +q] =
2

_
1
0
() (r) +q (r)) cos .rdr
=
2

_
1
0
[) (r) cos .r +q (r) cos .r] dr
=
2

_
1
0
) (r) cos .rdr +
2

_
1
0
q (r) cos .rdr
= F
s
[)] +F
s
[q]
And
F
s
[c)] =
2

_
1
0
(c) (r)) cos .rdr
= c
2

_
1
0
) (r) cos .rdr
= cF
s
[)]
Proposition 111 (Transforms of derivatives) Let ) be a function such that
) (r) !0 and )
0
(r) !0 as r !1. The following are true:
1. F
s
[)
0
] = .F
c
[)]
2. F
s
[)
00
] =
2

.) (0) .
2
F
s
[)]
3. F
c
[)
0
] =
2

) (0) +.F
s
[)]
110 CHAPTER 14. INTEGRAL TRANSFORMS
4. F
c
[)
00
] =
2

)
0
(0) .
2
F
c
[)]
Proof. We prove part 1. The remaining parts are left as exercises. For these
proofs, we will use integration by parts. Also, recall that to compute an integral
of the form
_
1
a
) (r) dr, we must compute lim
u!1
_
u
a
) (r) dr. With this in mind,
we have
F
s
[)
0
] =
2

_
1
0
()
0
(r)) sin.rdr
=
2

_
lim
u!1
_
u
0
()
0
(r)) sin.rdr
_
=
2

_
lim
u!1
_
) (r) sin.rj
u
0
.
_
u
0
) (r) cos .rdr
__
(integration by parts)
=
2

_
lim
u!1
) (n) sin.n ) (0) sin0 .
_
1
0
) (r) cos .rdr
_
= .
2

_
1
0
) (r) cos .rdr
= .F
c
[)]
For this computation we also used the fact that lim
u!1
) (n) sin.n = 0 since by
assumption lim
u!1
) (n) = 0 and we know that sin.n is bounded.
Proposition 112 (Derivatives of transforms) Let ) be a function. such
that ) (r) !0 and )
0
(r) !0 as r !1 The following are true:
1. F
c
[r) (r)] =
d
dn
F
s
[) (r)]
2. F
s
[r) (r)] =
d
dn
F
c
[) (r)]
Proof. We start from the expression on the right and we use Leibniz rule, which
we reviewed several chapters earlier (see theorem 34). We have
d
dn
F
s
[) (r)] =
d
dn
_
2

_
1
0
) (r) sin.rdr
_
=
2

_
1
0
d
dn
() (r) sin.r) dr (by Leibniz rule)
=
2

_
1
0
r) (r) cos .rdr
= F
c
[r) (r)]
We now list the sine and cosine transforms of some standard functions. Be-
fore we do so, we introduce some new functions which play a role in the tables
we are about to give.
14.4. SOME PROPERTIES OF THE SINE AND COSINE TRANSFORMS111
Denition 113 We dene four important functions.
1. The Heaviside function, denoted H (r a), is dened to be
H (r a) =
_
0 if r < a
1 if r a
2. The Reected Heaviside function, denoted H (a r), is dened to be
H (a r) =
_
0 if r a
1 if r a
3. The error function, also called the Gauss error function, denoted
erf (r), is dened to be
erf (r) =
2
p

_
x
0
c
t
2
dt
4. The complementary error function, denoted erf c (r) is dened to be
erf c (r) = 1 erf (r)
=
2
p

_
1
x
c
t
2
dt
The graphs of erf (r) and erf c (r) are shown in gure 14.4.
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
x
y
erf (r) and erf c (r) (dashed line)
112 CHAPTER 14. INTEGRAL TRANSFORMS
Proposition 114 The following table gives the sine transform pair for some
standard functions.
) (r) =
_
1
0
1 (.) sin.rd. 1 (.) =
2

_
1
0
) (r) sin.rdr
0 0
) (ar)
1
a
1
_
.
a
_
c
ax
2.
(a
2
+.
2
)
r

1
2
_
2
.
_1
2
H (a r)
2
.
[1 cos .a]
r
1
1
r
a
2
+r
2
c
a!
r
r
4
+ 4
1
2
c
!
sin.
tan
1
a
r
1 c
a!
.
r
2
) (r)
2

1
00
(.)
erf c
_
r
2
p
a
_
2

1 c
a!
2
.
Proposition 115 The following table gives the cosine transform pair for
some standard functions.
) (r) =
_
1
0
1 (.) sin.rd. 1 (.) =
2

_
1
0
) (r) sin.rdr
0 0
) (ar)
1
a
1
_
.
a
_
c
ax
2a
(a
2
+.
2
)
r

1
2
_
2
.
_1
2
H (a r)
2
.
sina.
c
ax
2 1
p
a
c

.
2
4a
sinar
r
H (a .)
a
r
2
+a
2
c
a!
r
2
) (r)
2

1
00
(.)
Finding the Fourier transform of a function is a matter of computing the
corresponding integral, which can be hard. We illustrate this with an example.
14.5. THE SINE TRANSFORM: AN APPLICATION 113
Example 116 Find the Fourier cosine transform of ) (r) = c
ax
, (a 0, r 0).
Note these conditions mean that ) (r) !0 as r !1.
F
c
[)] =
2

_
1
0
c
ax
cos .rdr
From a table on integrals,
_
c
ax
cos .rdr =
c
ax
a
2
+.
2
[. sin.r a cos .r]
=
a
a
2
+.
2
c
ax
_
.
a
sin.r cos .r
_
Thus
F
c
[)] =
2

a
a
2
+.
2
c
ax
_
.
a
sin.r cos .r
_

1
0
=
2

a
a
2
+.
2
14.5 The Sine Transform: An Application
We show how the Fourier sine transform can be used to solve a PDE. Consider
the following IBVP
_
_
_
PDE n
t
= c
2
n
xx
0 < r < 1 0 < t < 1
BC n(0, t) = 0 < t < 1
IC n(r, 0) = 0 0 r < 1
We illustrate the technique used in three steps. First, we transform the problem
by eliminating the r variable. Next, we will solve the easier problem. Finally,
we will apply the inverse transform to get the solution in terms of the original
problem.
Step 1: Apply the Fourier sine transform with respect to r Our equa-
tion can be rewritten
n
t
c
2
n
xx
= 0
We apply F
s
, the Fourier transform with respect to r to each side to get
F
s
_
n
t
c
2
n
xx

= F
s
[0]
Using linearity of F
s
and the fact that F
s
[0] = 0, we get
F
s
[n
t
] c
2
F
s
[n
xx
] = 0
which can be written as
F
s
[n
t
] = c
2
F
s
[n
xx
]
114 CHAPTER 14. INTEGRAL TRANSFORMS
Now, we evaluate the Fourier sine transform of each quantity separately.
Remember that we are taking the sine transform with respect to r. So,
we have a dr integral.
F
s
[n
t
] =
2

_
1
0
n
t
(r, t) sin.rdr
=
0
0t
_
2

_
1
0
n(r, t) sin.rdr
_
(Leibniz rule)
=
d
dt
F
s
[n]
=
d
dt
l (t)
Note that while n was a function of r and t, its transform F
s
[n] is a
function of . and t. The new variable . will be treated as a parameter.
Hence our notation F
s
[n] = l (t). For the second transform, we get using
proposition 111
F
s
[n
xx
] =
2

.n(0, t) .
2
F
s
[n]
=
2.

.
2
l (t)
Replacing in the PDE, we get
d
dt
l (t) = c
2
_
2.

.
2
l (t)
_
d
dt
l (t) +.
2
c
2
l (t) =
2.c
2

We need to nd the IC for this new equation, by applying the same trans-
form to the original IC. So, we get
F
s
[n(r, 0)] = 0
l (0) = 0
So, our ODE problem is
_
_
_
d
dt
l (t) +.
2
c
2
l (t) =
2.c
2

0 < t < 1
l (0) = 0
Step 2: Solve the ODE This is an ODE we have already solved using the
integrating factor technique (multiply each side by c
!
2

2
t
). Its solution is
l (t) =
2
.
_
1 c
!
2

2
t
_
14.6. PROBLEMS 115
Step 3: Applying the inverse transform to the solution F
s
changed the
variable r into .. F
1
s
will do the reverse. In this process, t is just a con-
stant. Since l (t) = F
s
[n], it follows that
n(r, t) = F
1
s
[l (t)]
To nd n(r, t) we can either compute the integral corresponding to F
1
s
or
use tables. From the tables, we see that F
1
s
_
2

1 c
a!
2
.
_
= erf c
_
r
2
p
a
_
.
Our function is almost that. We have l (t) =
2
.
_
1 c
!
2

2
t
_
=

1 c
a
2
!
2
t
.
. Thus, using linearity of F
1
s
, we have
F
1
s
[l (t)] = F
1
s
_

1 c
a
2
!
2
t
.
_
= F
1
s
_
2

1 c
a
2
!
2
t
.
_
= erf c
_
r
2
c
p
t
_
14.6 Problems
1. Finish proving proposition 110.
2. Finish proving proposition 111.
3. Finish proving proposition 112.
4. Solve the ODE:
_
_
_
d
dt
l (t) +.
2
c
2
l (t) =
2.c
2

0 < t < 1
l (0) = 0
5. Solve using the sine or cosine transform
_
_
_
PDE n
t
= c
2
n
xx
0 < r < 1 0 < t < 1
BC n
x
(0, t) = 0 < t < 1
IC n(r, 0) = H (1 r) 0 r < 1