Integral Transforms For Engineers

Integral Transforms for Engineers

by Larry C. Andrews and Bhimsen K. Shivamoggi
Integral Transforms
Integral transform methods provide effective ways to solve a variety of problems arising in the
engineering, optical, and physical sciences. This concise, easy-to-follow reference text
introduces the use of integral transforms, with a detailed discussion of the widely applicable
Laplace and Fourier transforms.
for Engineers
The text is suitable as a self-study for practicing engineers and applied mathematicians, as well
as a textbook for students in graduate-level courses in optics, engineering sciences, physics,
and mathematics. The math is straightforward. In most sections, applications relevant to
engineers and applied scientists are used in place of formal proofs. Numerous examples,
exercise sets, illustrations, and tables of transforms enhance the books usefulness as a
teaching tool and reference.
Contents: Special functions. Fourier integrals and Fourier transforms. Applications involving
Fourier transforms. The Laplace transformation. Applications involving Laplace transforms.
The Mellin transform. The Hankel transform. Finite transforms. Discrete transforms.
Bibliography. Appendix A: Review of complex variables. Appendix B: Table of Fourier
transforms. Appendix C: Table of Laplace transforms. Index.
Bhimsen K. Shivamoggi
Larry C. Andrews
Larry C. Andrews and Bhimsen K. Shivamoggi are professors of mathematics at the University
of Central Florida. Andrews is also a member of the Department of Electrical and Computer
Engineering and associate member of the Center for Research and Education in Optics and
Lasers (CREOL) and the Florida Space Institute. Shivamoggi is also a member of the
Department of Physics at U.C.F.
P.O. Box 10
Larry C. Andrews
Bellingham, WA 98227-0010
ISBN-10: 0819432326
ISBN-13: 9780819432322
SPIE Vol. No.: PM178
Integral Transforms
for Engineers
Larry C. Andrews
University of Central Florida
=-==
SPIE
===
=~
PTICAL ENGINEERING PRESS
A Publication of SPIE-The International Society for Optical Engineering
Bellingham, Washington USA
Library of Congress Cataloging-in-Publication Data
Andrews, Larry C.
Integral transforrns for engineers I Larry C. Andrews, Bhimsen K. Shivamoggi
p. cm.
Originally published: New York: Macrnillan, c1988.
Includes bibliographical references and index.
ISBN 0-8194-3232-6
1. Shivamoggi, Bhimsen K. II. Title.
QA432.A63 1999
515'.723-dc21 99-14143
CIP
Published by
SPIE-The Intemational Society for Optical Engineering

P.O. Box 10
Bellingham, Washington 98227-0010
Phone: 360/676-3290
Fax: 360/647-1445
E-mail: spie@spie.org
WWW: http://www.spie.org/
Copyright 1999 The Society of Photo-Optical lnstrumentation Engineers
(Originally published in 1988 by Macrnillan Publishing Company, New York.)
All rights reserved. No part of this publication may be reproduced or distributed

in any form or by any means without written perrnission of the publisher.
Printed in the United States of America.

Contents
Preface to the 1999 Printing vii
Preface lX
Introduction 1
1 SPECIAL FUNCTIONS 6
1.1 Introduction 6
1.2 The Gamma Function 7
1.3 The Error Function and Related Functions 16
1.4 Bessel Functions 21
1.5 U seful Engineering Functions 29
2 FOURIER INTEGRALS AND FOURIER TRANSFORMS 37
2.1 Introduction 37
2.2 Fourier Integral Representations 38
2.3 Proof of the Fourier Integral Theorem 47
2.4 Fourier Transform Pairs 49
2.5 Properties of the Fourier Transform 58
2.6 Transforms of More Complicated Functions 65
2.7 The Convolution Integrais of Fourier 78
2.8 Transforms Involving Generalized Functions 85
2.9 Hilbert Transforms 91
2.10 Additional Topics 97
3 APPLICATIONS INVOLVING FOURIER TRANSFORMS 102
3.1 Introduction 102
3.2 Boundary Value Problems 103
3.3 Heat Conduction in Solids 113
3.4 Mechanical Vibrations 125
3.5 Potential Theory 131
3.6 Hydrodynamics 141
3.7 Elasticity in Two Dimensions 151
3.8 Probability and Statistics 156
4 THE LAPLACE TRANSFORM 162
4.2 The Transforms of Some Typical Functions 164
v
vi Contents
4.3 Basic Operational Properties 170

4.4 Transforms of More Complicated Functions 182
4.5 The lnverse Laplace Transform 190
4.6 Complex Inversion Formula 200
4.7 Additional Topics 210
5 APPLICATIONS INVOLVING LAPLACE TRANSFORMS 218
5.2 Evaluating Integrais 218
5.3 Solutions of ODEs 221
5.4 Solutions of PDEs 229
5.5 Linear Integral Equations 238
6 THE MELLIN TRANSFORM 245
6.2 Evaluation of Mellin Transforms 246
6.3 Complex Variable Methods 254
6.4 Applications 262
6.5 Table of Mellin Transforms 273
7 THE HANKEL TRANSFORM 274
7.2 Evaluation of Hankel Transforms 276
7.3 Applications 285
7.4 Table of Hankel Transforms 290
8 FINITE TRANSFORMS 291
8.2 Finite Fourier Transforms 291
8.3 Sturm-Liouville Transforms 298
8.4 Finite Hankel Transform 303
9 DISCRETE TRANSFORMS 310
9.2 Discrete Fourier Transform 311
9.3 The Z Transform 321
9.4 Difference Equations 330
9.5 Table of Z Transforms 333
BffiLIOGRAPHY 335
APPENDIX A: REVIEW OF COMPLEX VARIABLES 337
APPENDIX B: TABLE OF FOURIER TRANSFORMS 340
APPENDIX C: TABLE OF LAPLACE TRANSFORMS 344
INDEX 349
Preface to the 1999 Printing
The use ofFourierintegrals in mathematics and physics applications dates back
to the pioneering work of Joseph Fourier (1768-1830). Since that time, the
notion of the integral transform has emerged as a related tool that owes much
of its success to the work of Oliver Heaviside (1850-1925), an English
electrical engineer who popularized the use of operational methods in
differential equations and electrical engineering. During the last decade or so
there have been significant generalizations of the idea of integral transforms
and many new uses of the transform method in engineering and physics
applications. Some of these new applications have prompted the development
of very specialized transforms, such as the wavelet transform, that have their
roots, however, deeply entrenched in the classical theory of Fourier. As a
result, knowledge of the properties and use of classical integral transforms,
such as the Fourier transform and Laplace transform, are just as important
today as they have been for the last century or so.
This text was written in 1988 as an introductory treatrnent of integral
transforms for practicing engineers and scientists, including the Fourier,
Laplace, Mellin, Hankel, finite, and discrete transforms. Like the fate of many
modem textbooks, the original publishing company changed hands and this
book went out of print after a few years. Nonetheless, a number of individuais
took the time to let us know they found the book useful as either a personal
reference text or as a classroom text, and also expressed their disappointment
in seeing it go out of print. We are therefore grateful to the SPIE PRESS for
agreeing to bring the book back into print. As authors, we have taken this
opportunity to correct several typographical errors that appeared in the first
printing, but would welcome hearing from anyone who finds additional
typographical errors that we did not catch or who cares to give any suggestions
for further improvements as well.
Larry C. Andrews
Orlando, Florida
March, 1999
vii
Preface
lN RECENT YEARS, INTEGRAL TRANSFORMS have become essential WOrking

tools of every engineer and applied scientist. The Laplace transform,
which undoubtedly is the most familiar example, is basic to the solution
of initial value problems. The Fourier transform, while being suited to
solving boundary-value problems, is basic to the frequency spectrum
analysis of time-varying waveforms. The purpose of this text is to introduce
the use of integral transforms in obtaining solutions to problems govemed
by ordinary and partial differential equations and certain types of integral
equations. Some other applications are also covered where appropriate.
The Laplace and Fourier transforms are by far the most widely used
of all integral transforms. For this reason they have been given a more
extensive treatment in this book than other integral transforms. However,
there are several other integral transforms that also have been used
successfully in the solution of certain boundary-value problems and in
other applications. Included in this category are Mellin, Hankel, finite,
and discrete transforms, which have also been given some discussion
here.
The text is directed primarily toward senior and beginning graduate
students in engineering sciences, physics, and mathematics who desire
a deeper knowledge of transform methods than can be obtained in in-
troductory courses in differential equations and other similar courses. It
can also be used as a self-study text for practicing engineers and applied
scientists who wish to leam more about the general theory and use of
integral transforms. We assume the reader has a basic knowledge of
ix
x Preface
differential equations and contour integration techniques from complex

variables. However, most of the material involving complex variables
occurs in separate sections so that much of the text can be accessible
to those with a minimum background in complex variable methods. As
an aid in this regard, wa have inclued a brief appendix relevant to our
use of the basic concepts and theory of complex variables in the text.
Also, because of the close association of special functions and integral
transforms, the first chapter is a short i~troduction to severa! of the
special functions that arise quite frequently in applications. This is con-
sidered an optional chapter for those with some acquaintance with these
functions, and thus it is possible to start the text with Chap. 2. Most
chapters are independent of one another so that various arrangements
of the material are possible.
Applications occur throughout the text and are drawn from the fields
of mechanical vibration, heat conduction, potential theory, mechanics
of solids and ftuids, probability and statistics, and severa! other areas.
A working knowledge in any of these areas is generally sufficient to work
the examples and exercises.
ln our treatment of integral transforms we have excised formal proofs
in several places, but then usually make an appropriate reference for the
more formal aspects of the theory. ln the applications we often make
the assumptions as to the commutability of certain limiting operations,
and the derivation of a particular solution sometimes may not be rigorous.
However, the approach adopted here is adequate in the usual applications
in engineering and applied sciences. We have included a large number
of worked examples and exercises to illustrate the versatility and adequacy
of this approach in applications to physical problems.
We wish to thank Jack Repcheck, Senior Editor of Scientific and
Technical Books department at Macmillan, for his assistance in getting
this text published in a timely manner. We also wish to express our
appreciation to the production staff of Macmillan for their fine efforts.
Finally, we wish to acknowledge Martin Otte who corrected several
errors during a final reading of the manuscript.
Introduction
The classical methods of solution of initial and boundary value problems

in physics and engineering scjences have their roots in Fourier's pioneering
work. An alternative approach through integral transform methods emerged
primarily through Heaviside's efforts on operational techniques. ln addition
to being of great theoretical interest to mathematicians, integral transform
methods have been found to provide easy and effective ways of solving
a variety of problems arising in engineering and physical science. The
use of an integral transform is somewhat analogous to that of logarithms.
That is, a problem involving multiplication or division can be reduced
to one involving the simpler processes of addition or subtraction by taking
logarithms. After the solution has been obtained in the logarithm domain,
the original solution can be recovered by finding an antilogarithm. ln the
sarne way, a problem involving derivatives can be reduced to a simpler
problem involving only multiplication by polynomials in the transform
variable by taking an integral transform, solving the problem in the
transform domain, and then finding an inverse transform. Integral trans-
forms arise in a natural way through the principie of linear superposition
in constructing integral representations of solutions of linear differential
equations.
1
2 Introduction
By an integral transform, we rnean a relation of the forrn*
J~= K(s,t)f(t) dt = F(s) (0.1)
such that a given functionf(t) is transforrned into another function F(s)

by rneans of an integral. The new function F(s) is said to be the transform
of f(t), and K(s,t) is called the kernel of the transforrnation. Both
K(s, t) andf(t) rnust satisfy certain conditions to ensure existence of the
integral anda unique transforrn function F(s). Also, generally speaking,
not more than one function f(t) should yield the sarne transforrn F(s).
When both of the lirnits of integration in the defining integral are finite,
we have what is called afinite transform.
Within the above guidelines there are a variety of kemels that rnay
be used to define particular integral transforrns for a wide class of functions
f(t). lf the kemel is defined by
0, t <o
K(s,t) = { e-r, 1 ;;::: 0 (0.2)
the resulting transforrn
(0.3)
is called the Laplace transform. When

1 .
K(s,t) = -=em (0.4)
y27T
we generate the Fourier transformt
1
;;;-::_
y27T -=
J"" eist f(t) dt = F(s) (0.5)
which, when tis restricted to the positive realline, leads to the Fourier
sine and Fourier cosine transforms
J~ L"" f(t)sin st dt = F(s) (0.6)
* We will always interpret integrais like (0.1) as the principal 1'{1/ue of the integral.
defined in general by PV r~f(x)dx =!i~ rRj(x)dx.
t Other definitions of K(s,t) for Fourier transforms involve the choices e'", e-'-'',
(I/2'1T)e'", among others.
lntroduction 3
and
J~ L"' f(t)cos st dt = F(s) (0.7)
The Laplace and Fourier transforms are by far the most prominent
in applications. Many other transforms have been developed, but most
have limited applicability. ln addition to the Laplace and Fourier transforms,
the next most useful transforms are perhaps the Hankel transform of
order v
L"" tlv(st)f(t) dt = F(s) (0.8)
where lv(x) is the Bessel function of the first kind (see Sec. 1.4), and
the Mellin transform
(0.9)
The Hankel transform arises naturally in solving boundary value problems

formulated in cylindrical coordinates while the Mellin transform is useful
in the solution of certain potential problems formulated in wedge-shaped
regions.
The integral transforms mentioned thus far are applicable to problems
involving either semnfinite or infinite domains. However, in applying
the method of integral transforms to problems fonnulated on finite domains
it is necessary to introduce finite intervals on the transform integral.
Transfonns of this nature are called finite integral transforms.
A basic problem in the use of integral transforms is to determine the
function f(t) when its transform F(s) is known. We refer to this as the
inverse problem. In many cases the solution of the inverse problem is
another integral transform relation of the type
fon(s,t)F(s) ds = f(t) (0.10)
where H(s, t) is another kernel and D is the domain of s. Such a result

is called an inversion formula for the particular transform. For example,
the inversion formula for the Fourier transform takes the form (see Sec.
2.4)
..
1 Joo e-'s'F(s)
. ds = f(t) (0.11)
V127T -oo
which is very much like the transform itself in Eq. (0.5). This means
that the problems of evaluating transfonns or inverse transforms are
4 lntroduction
essentially the sarne for Fourier transforms. This is not necessarily the
case for other transforms like the Laplace transform, however, where
the inversion formula is quite distinct from that of the transform integral.
Also, in the case of finite transforms, the inverse transform is in the
form of an infinite series.
The basic aim of the transform method is to transform a given problem
into one that is easier to solve. ln the case of an ordinary differential
equation with constant coefficients, the transformed problem is algebraic.
The effect of applying an integral transform to a partial differential equation
is to reduce it to a partial differential equation in one less variable. The
solution of the transformed problem in either case will be a function of
the transformed variable and any remaining independent variables. ln-
version of this solution produces the solution of the original problem.
The exponential Fourier transform does not incorporate any boundary
conditions in transforming the derivatives. Thus, it is best suited for
solving differential equations on infinite domains where the boundary
conditions usually only require bounded solutions. On the other hand,
the Fourier cosine and sine transforms are well suited for solving certain
problems on semiinfinite domains where the governing differential equation
involves only even-order derivatives. We will see that the Fourier transform
lends itself nicely to solving boundary-value problems associated with
the following partial differential equations:
(a) the heat equation:
V2u = a- 2u, - q(x,y,z,t) (0.12)
(b) the wave equation:
V2u = c- 2u,, - q(x,y,z,t) (0.13)
(c) the potential equation:
(0.14)
ln addition, it is useful in the solution of linear integral equations of the
form
f(x) = u(x) - J~=k(x, t)u(t) dt (0.15)
and certain ordinary differential equations. lnteresting applications of

these transform methods arise in hydrodynamics, heat conduction, potential
theory, and elasticity theory, among other areas. The Fourier transform
also lends itself to the theory of probability and statistics. For example,
it tums out that the moments of a random variable X are merely the
coefficients of (itl I k! in the Maclaurin series expansion of the characteristic
function C(t) of the random variable X, and this function is related to
lntroduction 5
the probability density function p(x) by the Fourier transform relation
C(t) = f~ooeitx p(x) dx (0.16)
While the Fourier transform is suited for boundary-value problems,

the Laplace transform is suited for initial-value problems. However, there
are other situations for which the Laplace transform can also be used,
such as in the evaluation of certain integrais and in the solution of certain
integral equations of convolution type like
J: U(T)k(t - T) dT == f(t), t >o (0.17)
ln addition to the transforms mentioned above, there are other less

well known transforms like the Hilbert transform and the Sturm-Liouville
transform, both of which are more limited in their usefulness than the
Fourier and Laplace transforms. Also, discrete transforms like the discrete
Fourier transform (which is the discrete analog of the Fourier transform)
and the Z transform (which is the discrete analog of the Laplace transform)
are becoming more prominent in various engineering applications where
it is either impossible or inconvenient to use more conventional transforms.
Much of our initial discussion will evolve around the problem of
calculating the transforms F(s) of given functions f(t), and also around
the related problem of finding inverse transforms of various functions
F(s). Our primary objective is to introduce methods to use the integral
transforms, rather than concerning ourselves too deeply with the general
theory itself. Therefore, we do not attempt to present the basic theorems
in their most general forms. However, the conditions put forth in the
theorems are generally broad enough to embrace most of the functions
that naturally arise in engineering and physical situations. Proofs of the
theorems are provided when feasible, but are sometimes based on heuristic
arguments instead of rigorous mathematical procedures. For example,
often we have the need in our proofs for interchanging certain limit
operations, like integration and summation, and in these situations we
normally operate under the assumption that such interchanges are
permissible.
1
Special Functions
1.1 lntroduction
Most of the functions encountered in introductory analysis belong to the
class of elementary functions. This class is composed of polynomials,
rational functions, transcendental functions (trigonometric, exponential,
logarithmic, and so on), and functions constructed by combining two or
more of these functions through addition, subtraction, multiplication,
division, or composition. Beyond these functions lies a class of special
functions which are important in a variety of engineering and physics
applications.
The use of integral transforms is heavily interlaced with special functions
like the gamma function, error function, Bessel functions, and so forth.
Also, functions such as the Heaviside unit function and the impulse
function, which are employed in a variety of engineering applications,
are briefiy discussed. Hence, a brief review of (or introduction to) some
of these special functions can be quite useful before discussing integral
transforms themselves. *
* For a more thorough treatment of special functions, see L. C. Andrews, Special Functions of
Mathematicsfor Engineers (SPIE Press, Bellingham, Wash.; Oxford University Press, Oxford, 1998)
6
1.2/The Gamma Function 7
1.2 The Gamma Function

One of the simplest but very important special functions is the gamma
function. Although it has less direct application than some of the other
special functions, knowledge of the properties of this function is a pre-
requisite for the study of Bessel functions and others which do have
direct .pplication.
Historically, the gamma function was discovered by L. Euler (1707-
1783) in 1729 who was concerned with the problem of interpolating
between the numbers
n! = L"" e- tn dt,
1
n == O, 1,2, ...
with nonintegral values of n. His studies eventually led him to the gamma
function relation
Re(z) >O* (1.1)
later termed the Eulerian integral of the second kind by A. M. Legendre

(1752-1833). Legendre is also responsible for the symbol r that is most
often used for the gamma function. The variable z in Eq. (1.1) may be
real or complex. Although the integral is improper, it has been shown
that it converges uniformly for all values of z for which Re(z) > O. The
function f(z) is bounded and differentiable and, in fact, an analytic function
throughout this domain.
By substituting z + 1 for z in Eq. (1.1) and performing an integration
by parts, we obtain
f(z + 1) = J: 1
e- f dt
= -e- 1 f ~~ + z L"" e- 1
tz-l dt
from which we deduce the very simple but important recurrence formula
f(z + 1) = zf(z) (1.2)
The value z= 1 in Eq. (1.1) greatly simplifies the integral and leads
to the result
f(l) = L"' e- 1
dt =1
* By Re(z), we mean the real part x of the complex variable z = x + iy. Similarly,
lm(z) refers to the imaginary part y. See also Appendix A.
8 Chap. 1/Special Functions
Hence, by repeated use of the recurrence formula (1.2), we see that

f(2) = I, f(3) = 1 2, f(4) = 1 2 3, while in general
f(n + I) = n!, n = 0,1,2,... (1.3)
Thus, we see that the gamma function is Euler' s extension of the factorial
function to nonintegral values of n. ln fact, the gamma function is an
extension of the factorial function to all complex numbers with a positive
real part. ln the next section we will extend this domain even further.
1.2.1 Analytic Continuation for Re(z) < O

An analytic continuation of f(z) to the left of the imaginary axis can be
accomplished through repeated use of the recurrence formula (1.2) ex-
pressed in the form
f(z) = f(z + 1)/z, z:fO (1.4)
The right-hand side of (1.4) is defined for all Re(z) > - 1, z :f O, and
thus this expression defines f(z) in this domain. Replacing z by z + 1
in (1.4) yields
f(z + l) = f(z + 2)/(z + 1), z :f -1
and when substituted into (1.4), we obtain
+ 2)/z(z + 1),
f(z) = f(z z :f O, -1 (1.5)
which now defines f(z) for ali Re(z) > - 2, z :f O, -1. Continuing this
process, we deduce that
f(z) = f(z + k) z :f 0,-1, ... ,-k + 1 (1.6)

z(z + l)(z + 2)(z + k - 1)'
where k is a positive integer.
Equation (1.6) can be used to define the gamma function at every z
with a negative real part except at negative integers and zero. The values
z = O, - 1, - 2, ... , are actually first -order poles of the function and thus
ln- n)l = oo, n = 0,1,2, ... (1.7)
The graph ofthe gamma function for z = x, a real variable, is sketched
in Fig. 1.1.
1.2.2 Additional Properties

One of the most common uses of the gamma function is in the evaluation
of certain integrais. That is, when it appears in applications it frequently
has the form suggested by Eq. (I. I) or some variation of it. For example,
if we set t = u 2 in (1.1), we find
1.2 /The Gamma Function 9
f(x)
\)
II
I
I
I
I
I
I
4
I I
I I
I I
I I
I
X
-41 -3 -21 3 4
I I
I I
I I
I I
I I
I
!
l
Figure 1.1 Graph of the gamma function
f(z) = 2 Loo e-" u2z-l du, Re(z) >O

2
(1.8)
whereas the substitution t = log(l I u) in ( 1.1) yields *
f(z) = Jori (log~)z-1 du, Re(z) >O (1.9)
Example 1.1: Evaluate the integral I

gamma function.
Solution: By making the substitution t = x3 , we have
* By log x, we mean the natural logarithm, often denoted by ln x.

and by comparison with Eq. (1.1), we deduce that
Loo e-x\4 dx = (l/3)f(5/3)
An interesting relation involving the product of two gamma functions

can be derived by using the representation (1.8). We first write
f(x)f(y) = 2 J: e-u2U2x-l du 2 L"" e-v v y-l dv

2 2
= 4 L"" L"" e-<u +u2lu2x-tv y-I du dv

2 2
where x and y are real variables. The change of variables

u = r cosO, v = r sinO
leads to
which reduces to
f"/2
f(x)f(y) = 2f(x + y) Jo cos2x-o sin2y-o dO (1.10)
Solving for the integral gives us the relation

f"/2 2x-o . 2y-IOdO- f(x)f(y)
x>O,y>O (1.11)
Jo cos sm - 2f(x+y),
f"/2
Example 1.2: Evaluate the integral Jo sin40 cos 50 dO.
Solution: Comparing this integral with that in Eq. (1.11), we see

that 2x - 1 = 5 and 2y - 1 = 4; thus, x = 3 and y = 5/2. ln terms
of gamma functions, we have
f"/2 . 4 5 f(3)f(5/2)
Jo sm Ocos O dO = 2f(1 1/ 2)
But, f(3) = 2 and
rC 1) = r(1 + ~) = ~ ~ ~rG)
2
1.2 /The Gamma Function 11
which follows from repeated applications of the recurrence formula

(1.2). Hence
r~12 4 5 2 2 2 s
Jo sin 6 cos 6 d6 = 9 7 5 = 3I5
Finally, we wish to derive an important relation connecting the gamma
function and the trigonometric functions. To doso, let us set x = 1 - z
and y = z in Eq. (1.10), which leads to
1
~/2
f(z)f(l - z) =2 tan2' - 16 dO (1.12)

0
where z may be real or complex. The change of variable u

converts this integral to the form
f(z)f(l - z) = 1=o Iuz-+ udu,

1
O< Re(z) < I (1.13)
where we (temporarily) restrict the real part of z as indicated in (1.13).

The evaluation of this last integral can now be accomplished through
contour integration in the complex plane.
Let us integrate the complex function
~--1
fW =I+~
which has a simple pole at ~ = - I and a branch point at ~ = O, around
the contour shown in Fig. 1.2 and then let p ~ O and R ~ ao. If we
write ~ = u along the upper boundary of the cut along the positive real
axis, then we must write ~ = ue 2~i along the lower boundary of this cut.
lm(~)
ReW
p
~-plane
Figure 1.2 Contour for evaluating the integral in (1.13)

The integral of f over the entire closed path leads to
tfWd' = LJWd' + LRf(u)du + LR!Wd' + J:f(ue 2";)du (1.14)

Based on Theorems A.3 and A.4 in Appendix A, we conclude that
lim
R-+oo
f fWd'
JcR =O
and since fW has a simple pole at ' = - 1, we find in the limit as

R--+= and p --+O that (1.14) reduces to
uz-1 fo (ue21Tiy-l
i
oo
- - du + du = 2m"Res{ -1}
ol+u oo1+u
or
z-1
i
oo
(1 _ e2,.;z) _u_ du = 2TriRes{ -1} (1.15)
o 1+u
Computing the residue, we obtain

Res{ -1} = lim ,z-l = ( -l)z-t
C-+-t
However, ( -l)z-t = e"i<z-tJ = -e";Z, and hence (1.15) becomes
from which we deduce

uz-1 7T
i
oo
--du=-- O< Re(z) < 1 (1.16)
o 1 +U SD7TZ '
Combining the results of (1.13) and (1.16), we obtain

f(z)f(l - z) = Tr/sinTrz (1.17)
which is valid for ali nonintegral values of z. By setting z = 1/2, we
get
f(l/2)f(l/2) = 7T
which leads to the special value
f(l/2) = y;. (1.18)
(11"/2
112
Example 1.3: Evaluate the integral Jo cot x dx.
Solution: Making use of (1.1 l) and (1.17), we get
f(l/4)f(3/4)
2f(l)
1 1T
2 sin(Tr/4)
and hence
There are many other identities involving the gamma function which
are too numerous to mention here. For reference purposes, a short list
of basic identities follows.
Basic Identities for f(z)
(Gl): f(z) = L=e-rtz-J dt, Re(z) >O
(G2): f(z) = 2L'"" e- t 2z-J dt,

12
Re(z) >O
(G3): f(z + l) = zf(z)

(G4): f(n + 1) = n!, n = 0,1,2, ...
(G5): f(l/2) = y;.
(G6): f(z) = ( f(z + k) , k = 1,2,3, ...

z z + 1)(z + 2) (z + k - 1)
(G7): y;. f(2z) = 22z-J f(z)f(z + !)
(2n)! -
(G8): f(n + 1/2) = 22n 1 y1r, n = 0,1,2, ...
n.
1T
(G9): f(z)f(l - z) = - . - , z nonintegral
SID 1TZ
(GlO): f(n + 1)- y21rnnne-n, n ~ao* (Stirling's formula)
* The symbol - means "behaves like" or "is asymptotic to."

EXERCISES 1.2
ln Probs. 1-6, give numerical values for the expressions (use the result
f(l/2) = y;
where necessary).
1. f(6) 2. f(3/2)
3. f(7 /2) 4. f( -1/2)
5. f( -9/2) 6. f(8/3)/f(2/3)
ln Probs. 7-9, verify the given identity.
7. f(a + n) = a(a + l)(a + 2) .. (a + n - l)f(a), n = 1,2,3, ...
8. f(n - a)/f(- a) = ( -1ta(a - 1)(a - 2) .. (a - n + 1), n =
1,2,3, ...
9. f(a)/f(a - n) = (a - l)(a - 2) .. (a - n), n = 1,2,3, ...
10. The binomial coefficient is defined by (a =I= 0)
(a)o = 1 (a)k = a(a - (ak! - k + 1) '

1) ..
k = 1,2,3, ...
Show that
(a) ) = k!(n n~ k)!' n = 0,1,2, ... , k = 0,1,2, ... ,n
(b) (- !/ 2
) = ( ~~(~~~)!, n = 0,1,2, ...
(c) (a) =
k
f(a
k!f(a- k + 1)'
+ 1)
k = 0,1,2, ...
(d) (~a)= (-1i(a + Z- 1

), k= 0,1,2, ...
ln Probs. 11-13, verify the given integral formula.
11. f(x) = p" L= e-pttx-I dt, X > O, p > O
12. f(x) = J~=exp(xt - e~dt, x >O
13. f(x) = (log bY L= tx-Ib- 1 dt, x >O, b > 1
Hint: Let u = t log b.

ln Probs. 14 and 15, use properties of the gamma function to obtain the
result.
1.,.;-z
14.
Joo
2
a e ax-x
2
dx = 2 v 7Tea
2
Hint: 2ax - x2 = -(x - a)
2
+ a
ln Probs. 16-19, use Eq. (1.11) to evaluate the integral

{.,.;z {.,.;z
16. Jo sin x dx
5
17. Jo sin xcos x dx
2
100 ~
1f/2
18.
10
~dx 19
o 1 + x4
Hint: Let x2 = tan 8.
20. Show that
.,.;z 2 1
i
o
sin n+ 8d(J =
i1t/2
o
cos 2n+i(Jd(J =
22n( ')2
(2n +
n.
I)!
, n = 0,1,2, ...
21. The beta function is defined by the integral
B(x,y) = Li tx-i(l - t)Y-i dt, X> O, Y >O
Show that
(a) B(x,y) = f(x)f(y)/f(x + y)
Hint: Let t = cos 28.
(b) B(x,y) = L= (1 :x~;x+y du

Hint: Let t = u/(1 + u).
22. Using the result of Prob. 21, evaluate the following:
(a) B(2,3) (b) B(l/2,1)
(c) B(2/3,1/3) (d) B(3/4,1/4)
23. By setting y = x in Eq. (1.11),
(a) show that
f(x)f(x) = zi-2x l.,.;z . 2x-l,~,. d..l.
2f(2x) o sm '+' '+'
Hint: Use the identity sin x cos x =! sin 2x.

(b) Evaluating the integral in (a), deduce the Legendre duplication
formula
y; f(2x) = z2x-i f(x)f(x + 1/2)
24. Using the result of Prob. 23, derive the relation

(2n)! -
f(n + I/2) = 22nn.I y'fl', n = O,I,2, ...
25. Verify the gamma function relation (n +v :f -I,-2,-3, ... )

f(2n + 2v + 1) = _I_ 22n+2 f(n + v + 1/2)
f(n +v+ I) y;
1.3 The Error Function and Related Functions

The error function is defined by the integral
erf(z)
2
= y:;;: Jo e-
r 12
dt (l.I9)
where the variable z may be real or complex. * This function is encountered

in probability theory, the theory of errors, the theory of heat conduction,
and various branches of mathematical physics.
By representing the exponential function in (l.I9) in terms of its power
series expansion, we have
erf(z) =
2 Jor ~o ----;;!t
y; <-Ir
00
2
n dt
from which we deduce (termwise integration of power series is permitted)

00
2 (-1tin+l
erf(z) = -= L (1.20)
VTI' n=O n!(2n + 1)
This series converges everywhere in the finite complex plane; therefore,

erf(z) is an entire function.
Examination of the series (1.20) reveals that the error function is an
odd function, i.e.,
erf(- z) = - erf(z) (1.21)
We also see that
erf(z) = erf(z) (1.22)
where z denotes the complex conjugate.
When z = O, it is clear that
erf(O) = O (1.23)
* The variable t can also be complex, although we generally assume it is real.

1.3/The Error Function and Related Functions 17
while for z ~ oo [jarg(z)l < 1rI 4], it follows from properties of the gamma
function that (in the limit)
erf(oo) = 2_ f'"' e- 12
dt = f(l~) = 1 (1.24)
v7T Jo v7T
The graph of erf(x), where x is real, is shown in Fig. 1.3.
1.3.1 Complementary Error Function

ln some applications it is useful to introduce the comp/ementary error
function
erfc(z) = v;2 f'"' e _,z dt

z (1.25)
Using properties of integrais, it follows that

2 f'"' _,z d
erfc(z) = v; Jo e t - v;2 Jore _,z d
t,

erfc(z) =1- erf(z). (1.26)
Hence, ali properties of erfc(z) are easily derived from those of erf(z).
1.3.2 Fresnel Integrais

Closely associated with the error function are the Fresnel integrais
C(x) = f 2
COS(7Tt /2) dt (1.27)
erf(x)
-2 2 X
Figure 1.3 The error function

and
S(x) = J: sin(1rf /2) dt (1.28)
These integrais come up in various branches of physics and engineering,

such as in diffraction theory and the theory of vibrations, among others.
From definition, we have the immediate results
C(O) = S(O) = O (1.29)
The derivatives of these functions are
2
C'(x) = COS(1TX /2), S'(x) = sin(1Tx2/2) (1.30)
and thus we deduce that both C(x) and S(x) are oscillatory. Namely,
C(x) has extrema at the points where r
= 2n + I (n = O,I,2, ... ), and
S(x) has extrema where x2 = 2n (n = I ,2,3, ... ). The largest maxima
occur first andare found to be C(l) = 0.77989 ... and S('\1'2) = 0.71397 ....
For x ~ oo, we can use the integral formulas (see Prob. I2 in Exer. 1.3)
f' cos fdt = {" sin fdt = ~ J~

to obtain the results
C(oo) = S(oo) = i (1.31)
The graphs of C(x) and S(x) for positive real x are shown in Fig. I.4.
y = C(x)
0.5
Figure 1.4 The Fresnel Integrais

1.3 /The Error Function and Related Functions 19
EXERCISES 1.3
1. Show that
(a) ra e- ~ 1 dt = y; erf(a)
(b)
rb
Ja e
-(2
dt =
v; [erf(b) -
2 erf(a)] =
v; [erfc(a) -
2 erfc(b)]
2. Show that
d 2 2
- erf(z) = -=e-z
dz v1T
3. Use integration by parts to obtain
1 -
J erf(z)dz = z erf(z) + y; e
2
z +C
where C is a constant of integration.

4. Evaluate
(a) erfc(O)
(b) erfc( oo)
5. Show that (p > 0)
= 1
f o
e-px erf(x) dx = - eP 214 erfc(p/2)
p
Hint: Replace erf(x) by its integral representation and interchange

the order of integration.
6. Show that (p :::: 0)
L= e-px-x2f4 dx = y; ep2 erfc(p)
Hint: Write x2 /4 + px = (x/2 + p) 2 - p 2 and make the change of

variable u = x/2 + p.
7. Using integration by parts,
(a) show that
= e -z2 1 i"" -
e -12
i
2
e- 1 dt =- - - dt
z 2z 2 z t2
(b) By repeated integration by parts, derive the asymptotic series

erfc(z)-
-z
v'e-
2
[
I + 2:= (- l t 1 X 3 X X (2n- I)]
7r Z n=l (2z 2
t
lzl - =, iarg(z)l < -rr/2
8. If X is a normal random variable, its probability density function is
( ) =-=-e 1 -(x-mP/2rr2
px
y2-rrCT
where m is the mean value of X and CT2 the variance. The probability
that X $: y is defined by
P(X $: y) = f= p(x) dx
(a) Show that
P(X $; y) = H1 + err(Y; 2:) J

(b) What is the probability P(X $; y) in the limit y - oo?
9. Considering the integral
a 2: O, b >O
as a function of the parameter a:

(a) Show that I satisfies the first-order linear differential equation
(DE)
di 2 -
da- 2ab I = -2y-rr.
(b) Evaluate I(O) directly from the integral.

(c) Solve the DE in (a) subject to the initial condition in (b) to deduce
that
7r
I(a) =b 2b2
ea erfc(ab).
10. Show that the Fresnel integrais satisfy

(a) C( -x) = -C(x) (b) S( -x) = -S(x)
11. Obtain the series representations

(a) C(x) = ~ ( -lt{-rr/2)2n x4n+l (b) S(x) =
= (_ I )n( /2)2n + I
n=O (2n)!(4n + 1) 2: 7r x4n+3
n=O (2n + 1)!(4n + 3)
1.4/Bessel Functions 21
12. Establish the integral formula

y;
i oo
o
e
- a2t2 d
t=-
2a
Then, writing a = (l - i)/y'2 and separating into real and imaginary
parts, deduce that
= ! J~
Jor= cos t = f"" sin r
2 dt dt
Jo 2 2
13. Using the definition of the error function (1.19), show that
(a) erf(Vfx) = (l + i) [C(xV2/7T) - iS(xV2/1T)]
(b) erf(V -ix) = (l - i) [C(xy'2/";) + iS(xV2/1T)]
(c) erf(Vfx) +erf(y' -ix) = 2 [C(xV2/7T) + S(xy'2/";)]
1.4 Bessel Functions

Bessel functions are closely associated with problems possessing circular
or cylindrical symmetry, such as the study of free vibrations of a circular
membrane and finding the temperature distribution in a circular cylinder.
These functions, of which there are several varieties, occur in so many
additional areas of application in engineering and physical science that
they are considered to be the most important functions beyond the el-
ementary ones studied in calculus.
Bessel functions of the first kind are defined by the series
= ( -l)k (z/2ik+v
fv(Z) = k~O k! f(k + v + 1) (1.32)
where the parameter v denotes the order of the given Bessel function.
When v = n (n = 0,1,2, ... ), Eq. (1.32) defines the Bessel function of
integer order
"" (- 1)k(z/2ik+n
Jn(Z) = t:o k! (k + n)! ' n = 0,1,2, ... (1.33)
the simplest representative of which is

= ~ (-lt(z/2)2k
J.o(Z) f:o (k!)2
(1.34)
The graphs of Jn(x), n = 0,1,2, are shown in Fig. 1.5, where xis real.
The parameter v in (1.32) may also take on negative values. For
example, when v = - n (n = O, l ,2, ... ), we get
Figure 1.5 Graph of ln(x), n = 0,1,2
oo ( -l)k(z/ 2)zk-n
l_iz) = L
k=O
k' (k _ )'
n
oo ( -1i(z/2)2k-n
= k~n k!(k- n)!
where we have used the fact that 1/(k - n)! = O (k = 0,1, ... , n - 1)
by virtue of Eq. (1.7). Finally, the change of index k = m + n yields
oo ( -l)m+n(z/ 2)zm+n
j_n(Z) =L m=O m.I(m + n.
)I

J -iz) = (- 1Y Jn(Z), n = 0,1,2, ... (1.35)
However, this last relation applies only to integral-order Bessel functions.
Rewriting (1.32) in the form
z)v oo ( -1)k(z/ 2)zk
Jv(Z) = ( 2 ~o k!f(k + V + 1)
it can be shown that the series on the right converges in the whole z-
plane. Therefore, the function (2/ztlv(Z) is ao entire function of z. However,
this does not necessarily imply that lv(z) is entire. If v< Oand nonintegral,
then clearly lv(z) has ao infinite discontinuity at z = O, and hence, cannot
represent an entire function. But, if v = n, n = 0,1,2, ... , then it can
be shown that Jv(x) is entire- a result that depends upon the relation
(1.35).
The Bessel functions are named in honor of F. W. Bessel (1784-
1846) who in 1824 carried out the first systematic study of the properties
of these functions and derived their goveming differential equation (see
Prob. 3 in Exer. 1.4). Nonetheless, Bessel functions were discovered
years earlier by Euler and others who were concerned with various
problems in mechanics, and the infinite series (1.34) was obtained by D.
Bernoulli in 1703- more than 120 years before Bessel's famous study
- in connection with bis investigation of the oscillatory behavior of a
hanging chain.
1.4.1 Basic Properties

The Bessel functions satisfy a large number of basic identities such as
~ [z"J.(z)] = z"lv-l(z) (1.36)
and
d
dz [z-vlv(Z)) = - Z-vlv+ 1(z) (1.37)
both of which follow from termwise differentiation of the series for z"J"(z)
and z-"lv(Z) (see Prob. 1 in Exer. 1.4). If we carry out the differentiation
in (1.36) and (1.37) and simplify the results, it follows that
(1.38)
and
J~(z)- ~lv(Z) = -lv+I(Z) (1.39)

z
The substitution of v = O in (1.39) leads to the special result
J(z) = -JI(Z) (1.40)
Direct integration of (1.36) and (1.37) give us the integral relations
J z"lv-l(z) dz = z"lv(Z) + C (1.41)
and
(1.42)
where C denotes a constant of integration. As a general rule, any integral
of the forro
m + n >O
where m and n are integers, can be evaluated with the use of (1.41) and
(1.42), coupled with standard integration techniques such as integration
by parts. When m + n is odd, the integral can be evaluated in closed

form, but will ultimately depend on the residual integral J J (z)dz when
0
m + n is even.
Examp/e 1.4: Reduce J z Jiz) dz to an integral involving only J (z).

2
0
So/ution: To use (1.42), we first write
Jz Jiz) dz = Jz [z- 12(z)] dz

2 3 1
and use integration by parts to get
Jlliz) dz = - z21t(Z) + 3 Jzlt(Z) dz

A second integration by parts on the last integral yields
Jz21 2(z) dz = - ilt(Z) - 3zJo(z) + 3 Jlo(z) dz

The last integral involving J0(z) cannot be evaluated in closed form,
and so our integration is complete.
For reference purposes, a short list of the basic identities of the Bessel
function follows.
Basic Jdentities for JJz)

,. ( _ k (z/Z)2k +"
(Jl): J.(z) = t:o k! f(k0 + 11 + 1)
(J2): 1 0(0) = 1; J.(O) = O, 11 >O

(13): J -iz) = ( -Itln(Z), n = O, 1,2, ...
(J4): ~ [z"J.(z)] = z"l.-t(Z)

(J5): ~ [z-"J.(z)] = -z-"lv+t(Z)
v
(J7):J~(z)- -Jv(Z) = -lv+t(Z)
z
(J8): lv-t(Z) - lv+t(Z) = 2J~(z)
2v
(J9): lv-t(Z) + J,+ t(Z) = - fv(Z)
z
(JlO): Jz"J.,_ 1(z) dz = z"J.,(z) + C

(Jll): f Z-vJ.,+ J(Z) dz = - Z-vJ.,(z) + C
1 f21r
(J12): Jo(Z) = 21T Jo eizcos8d(J
(z/2)"
(Jl3):J.,(z)-f(v + l)' v 1= -l,-2,-3, ... ,z-o
(114): J.,(z)- J~z cos[ z - (v + 1/2)~], lzl- ao, larg(z)l < 1T
Remark: ln certain applications it is important to recognize another

Bessel function Y.,(x), called a Besse/ function of the second kind and
order v. This function, defined by
_ J.,(x)cos V1T - J _.,(x)
Y.,(x ) - .
sm v1r
is a linear combination of J.,(x) and J _.,(x), and therefore satisfies the
sarne recurrence relations as J.,(x).
1.4.2 Modified Besse/ Functions

ln certain applications the Bessel function J .,(z) appears with a pure
imaginary argument. By setting z = iy in (1.32), we obtain
v (yj2)2k+v
J.,(ty) = l k~O k!f(k + V + 1) (1.43)
Except for the multiplicative factor i", the right-hand side of (1.43) defines
a real function, which is called the modified Bessel function of the first
kind and denoted by the symbol lv(Y ). Thus,
/.,(y) = ;-"J.,(iy) (1.44)
or by analytic continuation, we can generalize to complex arguments by
writing
(z/2)2k+v
/.,(z) = ~o k!f(k + v + 1)
(1.45)
Comparing this series representation with Eq. (1.-32) for Jv(z), it would
appear that /.,(z) and J.,(z) have many properties in common. lndeed, the
modified Bessel functions satisfy relations analogous to ali those for the
standard Bessel functions. ln particular, the modified Bessel functions
satisfy properties similar to those for J.,(z) given by (1.36)-(1.42) (see
the exercises). The major distinction in these functions perhaps is exhibited

by their graphs for real variable x. That is, the graph of lv(x) has an
oscillatory behavior like that of a sine or cosine except for decreasing
amplitude, while the graph of lv(x) shows no such oscillatory behavior
(see Figs. 1.5 and 1.6). This is comparable to the distinction between
circular functions and hyperbolic functions.
ln certain applications, particularly in probability theory, we find the
ln(x)
Figure 1.6 Graph of ln(x), n = 0,1,2
3 X
Figure 1.7 Graph of Kn(x), n = 0,1,2

appearance of another modified Bessel function, which we define by
Kv(Z) = !!_I -v(Z! - /,(z) (1.46)

2 sm v7r
This is called the modified Bessel function of the second kind and its
graph for certain integer values of v is given in Fig. 1. 7. Some properties
of this function are taken up in the exercises.
EXERCISES 1.4
1. Using the series representation (1.32), show that
(a) ~ [z"lv(z)] = z"lv- (z) 1
d
(b) dz [z-vlv(z)] = - z-vlv+ ,(z)
2. Based on (1.38) and (1.39), deduce that

(a) 2J~(z) = J,_,(z) - J,+t(z)
2v
(b) - lv(z) = lv-l(z) + J,+I(Z)
z
3. Verify that the series (1.32) is a solution of Bessel's differential
equation
iJ~(z) v2)1v(Z)
2
+ zJ;(z) + (z - =O
4. By comparing series, deduce that
(a) 1 11z(z) =
2
J
sin z
7T'Z
(b) J _,;z(z) = J 2
7T'Z
cos z
5. Using the Jacobi-Anger expansion
n= -oo
show that
(a) cos(x sin O) = J0(x) + 2 2: 1 n(x)cos(2n0)
2
n=l
(b) sin(x sinO) = 2 L1 n_ (x)sin[(2n -

2 1 1)0]
n=l
(c) cos x = 10(x) + 2 2: (-lt1 ix) 2

n=l
(d) sin x =2 2: (-1tlzn-l(x)

n=l
6. By integrating both sides of the result of Prob. 5(a), deduce that
J 0(x) = -1 i7T cos(x sinO) dO

7T o
7. By writing cos xt in an infinite series and using termwise integration,
deduce that
= ~ { cos xt dt
1
Jo(x)
7TJoyl1=7
ln Probs. 8-10, verify the given integral relation.
8. I 2
x J 0(x) dx = x 2J 1(x) + xJ0(x) - I J 0(x) dx + C
9. I x 3J 0(x) dx = (x3 - 4x)J1(x) + 2x 2J 0 (x) + C
10.
I 2
x- J 2(x) dx = - 2x2 J,(x) - 3J,(x)
3
1 + 1x Jo(x) + 3li Jo(x) dx + C
3
L"" e-x xJ (2x) dx = 1/2e

2
0
12. By expressing J0 (bx) in its series representation, use termwise in-

tegration to show that
L"" e-ax J (bx) dx = 1/'Va

0
2
+ b2 , a> O, b >O
13. By formally setting a = ic in the result of Prob. 12, deduce that

2 2
Jo(""cos(cx)J0(bx) dx = { 1/ybO, - c b>c

(a)
b <c.
(b) L"" sin(cx)J (bx) dx = { l/yb?_ cz,

0 bb: ~.
14. Integrate both sides of Prob. 13(a) with respect to c to deduce that
{"" sin x { 71'/2, Ol

(a) ~ [zvlv(Z)] = zvlv-!(z)
d
(b) dz [z-vlv(Z)] = Z-vlv+!(Z)
1.5/Useful Engineering Functions 29
16. Show that Ln(Z) = Iiz), n = 0,1,2, ...

17. Using the results of Prob. 15, show that
v
(a) /~(z) = lv-t(Z) - -lv(Z)
z
v
(b) /~(z) = lv+t(Z) + -lv(Z)
z
(c) l~(z) =i [lv-t(Z) + lv+ t(Z)]
2
(d) 1.-t(Z) - I v+ t(Z) =- lv(Z)
vz
18. By comparing infinite series, show that
(a) 11/z(z) =
2
J
sinh z
1TZ
(b) L 112 (z) = J
2
1TZ
cosh z
19. Show that
20. Show that

d
(a) dz[z"KJz)] = -z"K._ 1(z)
d
(b) dz [z-"K.(z)] = -z-"K.+ 1(z)
1.5 Useful Engineering Functions

ln the solution of various engineering and scientific problems it is helpful
to employ the use of special notation to identify functions that must be
prescribed in a piecewise fashion. There are a variety of functions for
which special notation has become standard, such as the step function,
rectangle function, signum function, ramp function, sine function, impulse
function, and so on. ln our brief introduction to such functions we
primarily will discuss the step function (and related functions) and the
impulse function. Because of their frequent occurrence in applications
involving the time domain, we will designate the independent variable
by t for discussion purposes only.
1.5.1 Heaviside Unit Function

Discontinuous functions occur quite naturally in circuit analysis problems
as well as in some problems involving mechanical systems. ln order to
deal effectively with functions having finite jump discontinuities, it is
30 Chap. 1/Speeial Functions
h(t - a)
a t
Figure 1.8 Heaviside unit function
helpful to introduce the unit step function, also widely known as the
Heaviside unit function in honor of its discoverer. * We denote this function
by the symbol h(t - a) and define it by (see Fig. 1.8)
h(t - a) = {~: :~ : (1.47)
This function has a jump discontinuity at t = a of unit magnitude. The

main utility of the Heaviside unit function is that it acts like a "switch"
to tum another function on or off at some time. For instance, the function
f(t) = h(t - l)cos 2Trt
is clearly zero for t < 1 and assumes the graph of the cosine function
for t > 1 as shown in Fig. 1.9.
A related function is the rectang/e function defined by
f(t) = {o1', a<t<b (1.48)

otherwise
Although it is customary to do so, we will not introduce any special
f(t)
Figure 1.9 Graph of f(t) = h(t - 1) cos21Tt
* See the footnote about Oliver Heaviside on p. 162.

a b t
Figure 1.10 Rectangle function
notation for this function since it can be easily expressed in terms of

the Heaviside unit function as (see Fig. 1.10)
f(t) = h(t - a) - h(t - b). (1.49)
The rectangle function (1.49) is useful in describing other functions
which are defined piecewise. For example, the function
fl(t), t< a
f(t) = { f2(t), a<tb
has the representation
f(t) = .fi(t)[l - h(t - a)] + fit)[h(t - a) - h(t - b)] + J;(t)h(t - b)
= .fi(t) + [j2(t) - f1(t)]h(t - a) + [j3(t) - f2(t)]h(t - b) (1.51)
1.5.2 Impulse Function

ln certain applications it is convenient to introduce the concept of an
impulse function which is the result of a sudden excitation administered
to a system, such as a sharp blow or voltage surge. Let us imagine that
the sudden excitation, which we will denote by da(t), has a nonzero value
ove r the short interval of time a - e < t < a + e, but is otherwise
zero. The total impulse (force times duration) imparted to the system is
thus defined by
fa+e
J
oo
I = -= da{t) dt = a-e da{t) dt (e> 0) (1.52)
The value of I is a measure of the strength of the sudden excitation.

ln order to provide a mathematical model of the function da(t), it is
convenient to think of it as having a constant value over the interval
a - e :5 t :5 a + e (see Fig. l.ll). Furthermore, we wish to choose
da(t)
2s
a-s a
Figure 1.11 Impulse function
this constant value in such a way that the total impulse given by (1.52)
is unity. Hence, we write
da(t) = (1/2s)[h(t - a + s) - h(t - a - s)] (1.53)
Now let us idealize the function da(t) by requiring it to act over shorter
and shorter intervals of time by allowing s - O. Although the interval
about t = a is shrinking to zero, we still want I = 1; i. e.,
lim I = lim J= da(t) dt = 1 (1.54)

e~o e---~>0 -oo
We can use the results of this limit process to define an "idealized"

unit impulse function, (t - a), which has the property of imparting a
unit impulse to the system at time t = a but being zero for ali other
values of t. The defining properties of this function are therefore
8(t- a) = O, t =f a (1.55)
J~oo (t - a) dt =1
By a similar kind of limit process, it is possible to define the integral
of a product of the unit impulse function and any continuous and bounded
function f; i.e.,
f~oo (t - a)f(t) dt = ~~ J~oo da(t)f(t) dt

1 fa+e
= lim - f(t) dt
e"'"'""+O 2B a-e
Recalling that
J: f(t) dt = f()(b - a), a<<b (1.56)

which is the mean value theorem of the integral calculus, we find that
= l
J-= ll(t - a)f(t) dt = ~~ 2e J(g) 2e
for some g in the interval a - e < g < a + e. Consequently, in the
limit we see that g ~ a, and deduce the sifting property
J~= ll(t - a)f(t) dt = f(a). (1.57)
Obviously the "function" 8(t - a), also known as the Dirac delta
function, * is not a function in the usual sense of the word. It has significance
only as part of an integrand. ln dealing with this function, therefore, it
is best to avoid the idea of assigning "functional values" and instead
refer to its integral property (1.57), even though it has no meaning as
an ordinary integral. Following more rigorous lines, the impulse function
can be defined as a limit of an infinite sequence of well-behaved functions
(see Probs. 11 and 12 in Exer. 1.5).
There are certain operational properties of the impulse function that
prove useful in practice. Our derivations of such properties, however,
will be based strictly upon formal manipulations of the symbols, i.e.,
they will not be rigorous. To begin we make the observation
J~= ll(t - a)f(t) dt = J~= ll(t)f(t + a) dt (1.58)
which follows from a simple shift in variable. Next, we write
I~= [f(t)ll(t - a)]g(t) dt == I~= 8(t - a)[f(t)g(t)] dt
== f(a)g(a)
Since
g(a) = J~= ll(t - a)g(t) dt
we see that
J~"" [f(t)(t - a)]g(t) dt = I~= [f(a)(t - a)]g(t) dt
from which we formally deduce

f(t)ll(t - a) = f(a)(t - a) (1.59)
* Named after Paul A. M. Dirac (1902-1984), who was awarded the Nobel prize (with
E. Schrdinger) in 1933 for his work in quantum mechanics.
Finally, we wish to develop a formal relationship between the impulse

function and the Heaviside unit function. To doso, let us use integration
by parts on the expression
J~oo h'(t - a)f(t) dt = h(t - a)f(t) [00 - J~oo h(t - a)f'(t) dt
= f(=) - L"' f'(t) dt

= f(a)
where we are assuming that f(t) is both continuous and bounded. By
comparison of this result with Eq. (1.57), we deduce that
J~oo h'(t - a)f(t) dt = J~oo i3(t - a)f(t) dt (1.60)
which suggests the formal relation

h'(t - a) = 8(t - a) (1.61)
Thus, in a purely formal sense we have extended the definition of derivative
to include discontinuous functions. That is, for t =/= a, the derivative of
h(t - a) is clearly zero, and at t = a the derivative is not defined in
the usual sense. We now say that the derivative of a function with a
finite jump discontinuity at some point will result in the presence of an
impulse function at that point. This generalized form of the derivative
can be very useful in practice in that it enables us to treat discontinuous
functions as if they were continuous functions; i.e., we do not have to
treat them piecewise (see Prob. 10 in Exer. 1.5).
EXERCISES 1.5
1. Show that
h(t) =i [1 + sgn(t)]
where the signum function is defined by
sgn(t) = { - 1,1 t<O

t >o
2. Show that
rb (t - to)f(t) dt = {f(to). a< t0 < b

Ja O, otherwise
3. Show that
oo 1
J -= i3(at)f(t) dt = jaj /(0)
4. Based on Eq. (1.59), deduce that

(a) t8(t) = O.
(b) 8(at) = 0/lai) 8(t).
(c) 8(t - a) = 8(a - t).
5. Show that if g(t) is a monotonic function for which g(a) = O, then
1
(a) 8[g(t)] = jg'(a)j 8(t - a).
(b) From (a), deduce that 8(at - b) = O/laj)8(t - b/a).
6. Show that the signum functon defined n Prob. 1 satisfies
d
dt sgn(t) = 28(t)
7. Using integration by parts, show formally that
(a) J~oo 8'(t)f(t) dt = -f'(O)
(b) J~ao 8<n>(t)f(t) dt = (-1t J<nl(O), n = 1,2,3, ...

8. If f(t) is a continuous differentiable function, show that it satisfies
the product rule
d
dt [8(t)f(t)J = 8(t)f'(t) + '(t)f(t)
9. Show that
l>'(t)f(t) = f(0)8'(t) - f'(0)8(t)
Hint: Use (1.59) and Prob. 8.

10. Let
n
g(t) = f(t) - L ak h(t - tk)

k=f
where f(t) is a piecewise continuous function having jump discon-

tinuities of magnitude a], a2' ... ' an at the points t], t2' ... ' tn.
Assuming that f'(t) is defined everywhere except at these
discontinuities,
(a) show that g(t) is everywhere continuous and that g'(t) = f'(t)
except for a finite number of points.
(b) Deduce that the generalized derivative of the piecewise differ-
entiable functionf(t) with finite jumps is the ordinary derivative,
where it exists, plus the sum of impulse functions at the dis-
continuities multiplied by the magnitude of the jumps.
11. Consider the sequence of rectangle functions defined by (n = 1,2,3, ... )
ltl < 1/n

ltl > 1/n
(a) Show that for each n the area enclosed by the rectangle is unity,
and deduce that
!~ f~= t/Jit) dt = 1
(b) More generally, if f is any function continuous at t = O and

everywhere bounded, show that
!~~ f~co t/Jif)j(t) dt = J(Q)

12. Any sequence of continuous and differentiable functions t/J 1(t), t/J 2(t),
... , t/Jn(t), ... satisfying the conditions of Prob. ll(a,b), is called a
delta sequence. Show that the following sequences are delta sequences:
(a) t/Jn(t) = n/-TT(l + n f),
2
n = 1,2,3, .. .
(b) t/Jn(t) = (n/v;)e -nzcz, n = 1,2,3, .. .
2
Fourier Integrais and
Fourier Transforms
2.1 Introduction
The concept of an infinite series dates back as far as the ancient Greeks
such as Archimedes (287-212 B.c.), who summed a geometric series in
order to compute the area under a parabolic are. ln the eighteenth century,
power series expansions for functions like eX, sin x, and arctan x were
first published by the Scottish mathematician C. Maclaurin (1698-1746),
and British mathematician B. Taylor (1685-1731) generalized this work
by providing power series expansions about some point other than
X= 0.
By the middle of the eighteenth century it became important to study
the possibility of representing a given function by infinite series other
than power series. D. Bernoulli (1700-1783) showed that the mathematical
conditions imposed by physical considerations in solving the vibrating-
string problem were formally satisfied by functions represented as infinite
series involving sinusoidal functions. ln the early 1800s, the French
physicist J. Fourier* carne across similar representations and announced
*Jean Baptiste Joseph Fourier (1768-1830) is known mainly for his work on the
representation of functions by trigonometric series in his studies on the theory of heat
conduction. His basic papers, presented to the Academy of Sciences in Paris in 1807 and
1811, were criticized by the referees for a Jack of rigor and consequently were not published
then. However, when publishing the classic Thorie analytique de la Chaleur in 1822, he
also incorporated his earlier work almost without change.
37
38 Chap. 2/Fourier Integrais and Fourier Transforms
in bis work on beat conduction tbat an "arbitrary function" could be

expanded in a series of sinusoidal functions. Some of Fourier's work
lacked rigor, but nevertbeless be provided tbe first real impetus to tbe
subject now bearing bis name. The Fourier integral was also first introduced
by Fourier as an attempt to generalize bis results from finite intervals
to infinite intervals. Tbe Fourier transform, wbile appearing in some
early writings of A. L. Caucby (1789-1857) and P. S. de Laplace (1749-
1827), also appears in tbe work of Fourier.
ln tbis cbapter we will discuss Fourier integral representations and
Fourier transforms, followed by a cbapter on applications involving tbe
Fourier transform.
2.2 Fourier Integral Representations

An important problem in matbematical analysis is tbe determination of
various representations of a given function f For example, a particular
representation may reveal information about tbe function tbat is not as
obvious by anotber representation. ln tbe calculus we are taugbt tbat
certain functions have power series representations of the form
(2.1)
wbere
n = 0,1,2, ...
Power series sucb as tbis are useful for numerical calculations in addition
to various otber uses. lf tbe function f is periodic witb period 2p, it may
bave a Fourier series representation*
f(x) = 2a ~ ( ancos -n1rx + bnsm-

1 + LJ . n1rx) (2.2)
0
n=l P P
wbere
1 JP n1rt
an = -
p -p
f(t)cos- dt,
p
n = 0,1,2, ... (2.3)
and
1 JP . n1rt
bn = -
p
f(t)sm- dt,
-p p
n = 1,2,3, ... (2.4)
* For a general discussion of Fourier series, see L. C. Andrews, Elementary Partia/

Differential Equations with Boundary Value Problems, Orlando: Academic Press, 1986.
2.2/Fourier Integral Representations 39
The theory of Fourier series shows that a periodic function satisfying

certain minimal requirements can be represented by the infinite sum of
sinusoidal functions given in (2.2). The formal* limit ofthis representation
as the period tends to infinity can be used to introduce the notion of a
Fourier integral representation. ln other words, while periodic functions
defined on the entire real axis have Fourier series representations, aperiodic
functions similarly defined have Fourier integral representations.
If f and f' are piecewise continuous functions on some interval [- p, p],
we say that f is piecewise smooth. If f has this property and is periodic
with period 2p, it has the Fourier series representation (2.2)-(2.4). To
formally obtain the Fourier integral representation of f from this series
as p ~ oo, we begin by substituting the integral formulas for a 0 , an,
and bn given by (2.3) and (2.4) into the Fourier series (2.2). This action
leads to
1
f(x) = -2 r
p -p
f(t)dt + ~
n=!
[.!.p r
-p
f(t)cos mrt cos mrx dt
p p
+ .!_ JP f(t)sin mrt sin mrx dt]

p -p p p
or
JP
f(x) = 2p
1
-pf(t)dt + p1 JP-pf(t) ~1 cos mr(tp-
= x)
dt (2.5)
where we have interchanged the order of summation and integration and

used the trigonometric identity
cos A cos B + sin A sin B == cos(A - B)
We now wish to examine what happens as we let p tend to infinity.
First, we must make the additional requirement that f is absolutely in-
tegrable, i.e.,
f~ if(t)ldt < 00 (2.6)
so that
lim -
1 Jp f(t)dt =O (2.7)
p-= 2p -p
For the remaining infinite sum in (2.5), it is convenient to let tls = rr/ p
and then consider the equivalent limit
* By "formal," we mean a procedure that is not mathematically rigorous.

1 f"'/!!>s
f(x) = lim -
lls-->0 1T
f(t)
-w/lls
2:
00
n= I
cos[nas(t - x)]as dt (2.8)
(Observe that as ~ O as p ~ oo.) When as is a small positive number,

the points nas are equally spaced along the s axis. ln such a case we
may expect the series in (2.8) to approximate the integral
L"" cos[s(t - x)]ds
in the limit as as ~ O. While this does not mean that the limit of the
series in (2.8) is defined to be the above, we may take, under appropriate
conditions on f 1 that (2.8) tends to the integral form
f(x) = ;
1 f""_""f(t) Jo("" cos[s(t - x)]ds dt (2.9)
Upon switching the order of integration, we get the equivalent form
f(x) = ;1 Jo("" f""-oo f(t)cos[s(t - x)]dt ds (2.10)
The purely formal procedure we just went through (since the passage
to the limit cannot be rigorously justified) has led us to an important
result known as Fourier's integral theorem.* We will state the theorem
here but not present its rather lengthy proof until Sec. 2.3.
Theorem 2.1 (Fourier Integral Theorem). If f and f' are piecewise con-
tinuous functions on every finite interval, and if
I~"" if(x)idx < 00
then
f(x) = ;1 Joroo I""_""f(t)cos[s(t - x)]dt ds
at points x where f is continuous. If x is a point of discontinuity of f,

the above integral converges to the average value ![f(x+) + f(x-)] of
the right-hand and left-hand limits. t
* For a rigorous discussion and a precise statement of the conditions under which
(2.10) holds, see E. C. Titchmarsh, Theory of Fourier Integrais, Oxford: Clarendon Press,
1937.
t Right-hand and left-hand limits are defined, respectively, by f(x+) = lim f(x + e)
e-O+
and f(x-) = lim f(x - e). At points of continuity it follows that f(x-) = f(x+) = f(x).
e-O+
The conditions listed in Theor. 2.1 are only sufficient conditions, not
necessary conditions. That is, there exist functions f that have valid
integral representations but which do not satisfy the conditions of this
theorem. Moreover, the conditions stated in Theor. 2.1 are not the most
general set of sufficient conditions that have been established over the
years. Nonetheless, these conditions are broad enough to cover most of
the functions commonly occurring in practice.
To emphasize the analogy between Fourier series and the Fourier
integral theorem, we rewrite (2.10) in the form
f(x)
1
= -7T' i"" f"" f(t)(cos stcos sx + sin stsin sx)dt ds
O -oo
or equivalently,
f(x) = L"" [A(s)cos sx + B(s)sin sx] ds (2.11)
where
A(s) = ;.1 f""-= f(t)cos st dt (2.12)
and
B(s) = ;.1 f""-= f(t)sin st dt (2.13)
ln this setting we refer to (2.11) as the Fourier integral representation

ofthe functionfwith coefficients defined by (2.12) and (2.13). The general
theory concerning such representations closely parallels that of Fourier
series.
Example 2.1: Find an integral representation of the form (2.11) for the
rectangle function f(x) = h(l - ixi), where h is the Heaviside unit
function (see Fig. 2.1).
f(x)
X
-1
Figure 2.1 Graph of f(x) = h(l - lxi)
42 Chap. 2/Fourier Integrais and Fourier Transfonns
Solution: The coefficients A(s) and B(s) are given by

1
A(s) = -
7T
J=
-=
1
h(l - lxi)cos sx dx = -
7T
J
1
-1
2 sin
cos sx dx = - -
7TS
s
and
B(s) = -1
7T
Jl -1
sin sx dx = O
Thus the Fourier integral representation becomes
f(x) = ;2 Jor= (sin

- - s) cos sx ds
8
Since x = O is a point of continuity of f in Exam. 2.1, we can use

the Fourier integral theorem to deduce that
f(O) = 1=~
7T
r= sins s ds
Jo
which leads to the interesting result*
r= sin s ds =! (2.14)
Jo s 2
Observe that at x = 1 there is a jump discontinuity in the function f
given above. At these points the Fourier integral converges to the average
value of the left-hand and right-hand limits. Hence, it follows that
1/2, X= -1
; L(-
2 = sin s
8
-) cos sx ds =
1
{1 2
i -l<x<l
X = 1
(2.15)
O, otherwise
Finally, it may be of interest to plot the "partial integral" of the
function f, defined by
S~(x) = ;J: (si;s) cossxds (2.16)
to see how it tends to f(x) as JL - oo. Recalling the identity

2 sin A cos B = sin(A + B) + sin(A - B)
we have
* This, of course, is a standard integral result that can also be derived by the use of
complex variable theory. It is an important result that we will refer to on several occasions.
Sx=-
( )
1 1~'- sin[s(l + x)] d
s+-
1 J"'
sin[s(l - x)] d
s
P- 7T o s 7T o s
=-
1 1"'( +x)
1
sin t
-dt+-
1 sin t
-dt
1"'(1-x)
7T o t 7T o t
= .!_{Si[JA,(l + x)] + SH~tO - x)]} (2.17)

7T
where Si(z) is the sine integral defined by
.( )
SlZ = iz -
o
sin t dt
t
(2.18)
Equation (2.17) is plotted in Fig. 2.2 for values JL = 4,16,128.
2.2.1 Cosine and Sine Integral Representations

lf the function f is an even function, i. e., if f(- x) = f(x), it follows from
properties of integrais that
A(s) = -7T1 J""

-oo
f(x)cos sx dx = -7T21""'
O
f(x)cos sx dx (2.19)
and
J""
B(s) = ;1 -=f(x)sin sx dx = O (2.20)
f(x) = L"" A(s)cos sx ds (2.21)
Figure 2.2 The partial integral of f(x) == h(l - !x!)

We refer to (2.21) as a Fourier cosine integral representation. ln a similar

manner, if f is an odd function, i.e., f( -x) = -f(x), we obtain the
Fourier sine integral representation
f(x) = f" B(s)sin sx ds (2.22)
where A(s) = O and
B(s) = -7T21""
o
f(x)sin sx dx (2.23)
Finally, if f should be a function defined only on the interval O <

x< oo, we can represent it over this interval by either a Fourier cosine
integral ora Fourier sine integral, analogous to the ha/f-range expansions
of Fourier series. Consider the following example.
Example 2.2: Find a Fourier cosine and Fourier sine integral repre-
sentation of the function (see Fig. 2.3)
f(x) = {cos x, O< x < 7T/2
O, X> 7T/2
Solution: For a cosine integral representation, we compute
A(s) =-
2l1r/Z cos x cos sx dx = 2 COS
(
7TS /2
)
2
1r0 1r 1 -s
and, therefore,
f(x) =; L"" (c~s_:r~2 ) cos sx ds

ln the sarne manner
= -2 (s---;;-
- 2sin_ 1rsj2)
12
B(s) = -2 i1T
cos x sm sx dx _..;.__
1r0 1r s-1
X
2
Figure 2.3 Graph of f(x) = h(TT/2 - x) cos x, x > O
from which we deduce the sine integral representation
f(x) = ; Jo
2 ("" (S -s sin TrS/2) sm. sx ds
2 _
1
EXERCISES 2.2
1. By using the result of Eq. (2.14), show directly that
(a) ("" sin s cos s ds = :!!.
Jo s 4
(b) ("" sin ax dx = :!!. , a>0
Jo x 2
2. If
x<O
f(x) = { 0e :..x , x>O
(a) show that
f(x) = .!_ ("" cos sx + s sin sx ds
1r Jo i +1
(b) Verify directly that the above integral representation converges
to the value 1/2 at x = O.
3. Find an integral representation for
f(x) = {1O, - xZ, lxl <

lxi >
1
1
and deduce the value of the integral
I=
lo
= (sin X - XCOS X) cos -x
x3
1 dx
2
4. Use the result of Prob. 3 to deduce that
(a) L"" C- ;osxy dx =~
(b) ("" sin4x dx = .:!!.
Jo x2 4
ln Probs. 5-10, obtain the Fourier integral representation of the given
function.
~:
-l<x<O
5. f(x) = e-lxl, -ao <X< ao 6. f(x) = { - O<x<l
O, otherwise
1 f(x) = {sin x, lxl < 7T 8. f(x) = { si~'x, O<x<7T

O, lxl > 7T otherwise
= {cos x, O<x<7T
9 f(x) 10. f(x) =e-x\ -=<x< =
O, otherwise
11. Show that e-kx, k > O, has the half-range representations
(a) e-kx = 2k roo ~OS s:Z ds, X> 0
7T Jo s +
(b) e-kx =~roo S2SnSX ds, X> 0
7T Jo s + e
12. Show that e-xcos x has the half-range representations
(a) e-xcos x =; Loo (:: : !)cos sx ds, x >O
2 ioo s sin sx
3
-x
(b) e cos x = - 4 ds, x>O
7T o s + 4
13. Using the results of Prob. 11, show that

-x -2x 6 roo s sin sx
e - e = ; Jo (i + l)(i + 4) ds, x >O
ln Probs. 14-17, obtain the Fourier cosine and Fourier sine integral
representations of the given function.
o<X < p, K > o
14. f(x) ={ o.'K X >p
15. f(x) = { ~'- x, ~~~< 1

16. f(x) = {~:n x, ~~: < 7T
17. f(x) = {o,'

r O<x<1
x> 1
18. Determine whether or not the following functions are absolutely

integrable on the entire real axis.
(a) f(x) = {11 + xi,
O,
lxl < 1
lxl > 1 (b) J(x) = {!ir, lxl < 1
lxl > 1
(c) f(x) = e-lxl (d) f(x) = sin x
X
<e>J<x> = (si:xr
2.3/Proof of the Fourier Integral Theorem 47
2.3 Proof of the Fourier Integral Theorem

ln order to provide a more rigorous justification of the Fourier integral
theorem (Theor. 2.1) we start with the following central result.
Lemma 2.1 (Riemann-Lebesgue). If f is piecewise continuous and ab-

solutely integrable on the entire real axis, then
!~ J~=f(t)cos t dt = !~ J~=f(t)sin t dt =O
or equivalently,
Proof: We will present the proof only for the case when f is continuous
and has a bounded derivative f' on the real axis. A slight modification
of the proof is required for the case whenfhas some finite discontinuities.
Using integration by parts over the finite interval - p s t s p, we
get
P f(t)eit dt
J -p
= f(t~eit
l
lP -
-p
J- JP-p f'(t)eit dt
l
Clearly, f(t)eit is bounded on ali finite intervals for any . Thus, the
first term on the right-hand side vanishes in the limitas tends to infinity.
Also, because we assume that f' is bounded, it follows that
Irpf'(t)eiI dtl s rp if'(t)ldt <ao

for all . Hence, we conclude that
lim JP f(t)eit dt = O
-+OO -p
and if we formally allow the limits of integration to become infinite in

extent, i.e., p ~ ao, we get our intended result.
Remark: The assumption that f' is bounded in the proof of Lemma

2.1 is not required for the validity of the theorem. However, by adding
this assumption, our proof is much simpler than would otherwise be
required.
Lemma 2.2. If f is piecewise smooth and absolutely integrable on the

entire real axis, and if x is a point of continuity of f, then
. 1 Joo sin t
hm - f(x + t) - - dt = f(x)
->OO 1T -oo f
Proof: We first note that
. 1 JP -sin-t I" -1 JP -
d Im sin tt
d
Itm- t=
1T -p
-+00 t ->OO 1T -p t
_ 1 Joo -
--
sin-t dt
1T -oo [
=1
this Iast resuit following from (2.14). Hence, to prove the Iemma we wish
to show that
. 1 JP [f(x + t) - f(x)] sm
Itm- . t dt =0
-+OO 1T -p t
The function [f(x + t) - f(x)]/t is piecewise continuous for all
t =f. O, and at t = O, we find that
limf(x + t) - f(x) = f'(x)

t-+0 t
which exists since f is piecewise smooth. Thus, the conditions of Lemma
2.1 are satisfied by this function, and the above integral necessarily
vanishes, even in the limitas we formally allow p to become infinite .

2.3 .1 Convergence at a Point of Continuity
We are now prepared to prove our maio result, which is Theor. 2.1. We
will present the proof only for points of continuity of the function f,
leaving the proof for points of discontinuity to the exercises.
By changing the order of integration in the following iterated integral,
we obtain
:;;1rfoo
Jo -oo f(t)cos[s(t - x)]dt ds
1Joo
= :;; -oo f(t) Jor cos[s(t - x)]ds dt
= _!. Joo f(t) sin (t - x) dt

1T -oo f - X
= -1 Joo f(x + sin t

t) - - dt
1T -oo /
If we now a11ow ~ = and invoke Lemma 2.2, we obtain our desired
result
2.4/Fourier Transfonn Pairs 49
1 r= foo
; Jo -=f(t)cos[s(t - x)] dt ds = f(x) (2.24)
where x is a point of continuity of the function f.
EXERCISES 2.3
1. Prove that
. JP sin t 7T
(a) hm - - dt = -
-+00 o t 2
(b) lim Jo sin t dt = '!!.
-+00 -p t 2
2. Based on Lemma 2.1, show that if f and f' are piecewise continuous
on (O,p) and (- p,O), then
. 2 JP sin t
(a) hm- f(x + t) - - dt = f(x+)
-+00 7T o t
. 2
(b) hm-
Jo sin t
f(x + t) - - dt = f(x-)
-+00 7T -p t
3. Prove Theor. 2.1 for points of finite discontinuity of f, i.e., prove that
; L= f~=f(t)cos[s(t - x)]dt ds = ~[f(x+) + f(x-)]
2.4 Fourier Transform Pairs

Fourier' s integral theorem (Theor. 2.1) states that
f(x) = ;.1 Jor= J""

-=f(t)cos[s(t - x)] dt ds. (2.25)
Through the use of Euler' s formula, cos x = !( eix + e- ix), we can express
(2.25) in terms of complex exponential functions. That is,
f(x) = ;1 1""
0
Joo-= f(t)cos[s(t - x)]dt ds
= _1 r= Joo f(t) [eis(t-x) + e-is(t-x)]dt ds

27T Jo -oo
= -1
27T
J""-oo Joo-oo f(t)eis(t-x) dt ds
or
f(x) = - 1 Joo e -isx Joo eist f(t) dt ds (2.26)

27T -oo -oo
which is the exponential form of Fourier's integral theorem.

What we have established by the integral formula (2.26) is the pair
of transform formulas*
F(s) = } Joo
~ ;;c_ eist f(t) dt (2.27)
V 27T -oo
and
f(t) = v'1 Joo e-ist F(s) ds (2.28)

27T -oo
We define F(s) as the Fourier transform of f(t), also written as

F(s) = @i{f(t);s} (2.29)
and f(t) as the inverse Fourier transform of F(s), which may be written
as
f(t) = g;- 1{F(s);t} (2.30)
The location of the constant 1/27T in the definition of the transform pairs
is arbitrarily selected as long as (2.26) is satisfied. For reasons of symmetry
we have split the constant between the transform pairs, but in the literature
no universal agreement exists on the location of these constants. ln some
texts, the constant 1/27T is positioned in front of one of the transform
pairs with no constant in front of the other. There is also some variation
as to which integral represents the transform and which one represents
the inverse transform. ln practice, of course, these differences are of
little consequence but the user should be aware of them when consulting
different reference sources.
As an immediate consequence of (2.27), we observe that
!F(s)! :s ~ r-;--
} Joo lf(t)l dt (2.31)
y27T -oo
Hence, if f is absolutely integrable it follows that its transform function

F(s) is bounded. A similar argument applied to (2.28) shows that f(t)
is also bounded when F(s) is absolutely integrable. Furthermore, the
Riemann-Lebesgue Lemma (Lemma 2.1) shows that
lim F(s) =O (2.32)
11--=
* Unless stated otherwise, we will generally assume that both t and sare real variables.
2.4/Fourier Transform Pairs 51
Since the transform function F(s) associated with absolutely integrable

functions that are also piecewise smooth must satisfy this last relation,
it immediately rules out certain functions as possible transform functions.
For example, sines, cosines, and polynomials do not satisfy this relation.
Finally, it is a curious property that although the function f(t) may have
certain finite discontinuities, its transform F(s) can be shown to be a
continuous function. Because of this, the Fourier transform is sometimes
called a "smoothing process."
2.4.1 Fourier Cosine and Sine Transforms

ln Sec. 2.2.1 we found that when the function f is even, the Fourier
integral representation of f(x) reduces to
f(x) = E"' A(s)cos sx ds

=-
2
'1T' i"" cos sx i"" f(t)cos st dt ds
o o
(2.33)
Based on this relation we introduce the Fourier cosine transform
.<fc{f(t);s} = J~L'"'f(t)cosstdt = Fc(s), s>O (2.34)
and inverse cosine transform
.<f'c 1{Fc(s);t} = J~ L"" Fc(s)cos st ds = f(t), t >o (2.35)
These results are interesting in that they imply the equivalence of the
operators .<1'c and .<f'c 1 ln other words, the cosine transform and its
inverse are exactly the sarne in functional form.
Similarly, whenfis an odd function its Fourier integral representation
becomes
f(x) = -2
'1T' i"" sin sx i"" f(t)sin st dt ds
o o
(2.36)
which leads to the Fourier sine transform
.<1'5 {f(t);s} = J~ L"" f(t)sin st dt = F5(s), s>O (2.37)
and inverse sine transform

1
.<fi {F5 (s);t} = J~ L"" F5 (s)sin st ds = f(t), t >o (2.38)
Hence, we see that the Fourier sine transform and its inverse are also
exactly the sarne in functional form.
lf the function f is neither even nor odd, but defined only for t > O,
then it may have both a cosine transform and a sine transform. Moreover,
the even and odd extensions of f will then have exponential Fourier
transforms. To see the relations between these various transforms, let
us construct the even extension of f by setting
fe(t) = f(ltl), -oo<t<oo (2.39)
The Fourier transform of fe(t) leads to
~{.fe(t);s} 1 Joo
= --= fe(t)eist dt
Y21f' -oo O
= .. 11 Joo fe(t)cos st dt + i .. 11 Joo~ fe(t) st dt

y21f' -oo y21f' -oo
= J!; Loo f(t)cos st dt

~{fe(t);s} = ~c{f(t);s}, -oo<s<oo (2.40)
Based on (2.40), it is clear that the Fourier transform and cosine transform
of an even function give identical results. ln particular, their transforms
are even functions of s (see Prob. 20 in Exer. 2.4). The odd extension
of f is constructed by setting
fo(t) = f(ltl)sgn(t), -oo<t<oo (2.41)
where the signum function is defined by
-1 t<O
sgn(t) = { : (2.42)
1 t >o
ln this case, we find
~{fo(t);s} = -1- Joo fo(t)eist dt

\(i; -oo O
1!-oo fo(t)
00
= .. ;;:c_
y21f'
~ os st dt + i .. }Joo fo(t)sin st dt
y21f' -oo
1
= i J!; Loo f(t)sin st dt
Because the Fourier transform of an odd function is also an odd function

(see Prob. 20 in Exer. 2.4), we make the conclusion that the Fourier
transform and sine transform are related by

gf{fo(t);s} = igf5{f(t);jsj}sgn(s), -oo<s<oo (2.43)
The practical use of (2.40) is that if we want to evaluate the Fourier
transform of an even function, we can do so by simply calculating its
cosine transform. If the function we wish to transform is odd, we first
can find its sine transform and then use (2.43). These observations are
very useful when using tables to find transforms, since most ofthe known
transforms are either cosine or sine transforms. A short table of transforms
is presented in Appendix B.
2.4.2 Evaluating Transforms of Elementary Functions

As already pointed out, many elementary functions like sines, cosines,
polynomials, and in general any periodic function, do not have Fourier
transforms (at least in the usual sense) because they are not absolutely
integrable. A special class of elementary functions that do have Fourier
transforms and can be calculated by basic methods are those involving
exponential functions. Several such transforms are related to the integrais
I = r= e-arcos st dt == s
Jo 2
a
+ a2 '
a>O (2.44)
and
J =
ioo
o
e
-ar .
sm st dt =
s2 +a
S
2, a>O (2.45)
One way to verify these integral formulas is to use integration by parts

to obtain the relations
1
I=--~ ioo e-arsin st dt == -1 - -J
S
a a o a a
and
J = (s/a)I
Solving these last two equations simultaneously for I and J yields the
results given by (2.44) and (2.45).
Example 2.3: Find the Fourier transform of e-airi, a > O.
Solution: Because the function is even, we can use (2.40) to write

gf{e-airi;s} = gfc{e-ar;s}
= AL= e -arcos st dt
or, using (2.44), we see that

2 a
~{e-altl;s} =
J-
1rs
2 +
a
2' a>O
Because transform relations occur in pairs as given by Eqs. (2.27)

and (2.28), it follows that once we have established one transform or
integral relation, the other one is automatically known. For instance,
based on the result of Exam. 2.3, we have
~{e-altl;s} = 1 J""-= eiste-altl dt = J~1r s 2 a 2 a> O (2.46)

y21r +a
and thus it immediately follows that
~-I
{ s2
1
+ a2 '
}
1
1 J""
= __
vz:;;. -oo
e-ist ds
i + a2 a
J;
= _1 _ e-altl
2 '
a>O
(2.47)
Moreover, by interchanging the roles of t and s in (2.47), and taking the
complex conjugate of the resulting expression (which is real in this example
since we are dealing with even functions), we now deduce the additional
Fourier transform
~{-1-
2 2
s} = ! 0:_ e-aisi
t +a ' a..Vz '
a>O (2.48)
ln this fashion we see that the evaluation of a single Fourier transform

has the effect of giving us two transform relations from each integral.
It can be shown that both (2.44) and (2.45) are uniformly converging
integrais on any closed intervals for which a is positive and all closed
intervals involving s. Related integrais which can be formally derived by
differentiating or integrating (2.44) and (2.45) with respect to either pa-
rameter, a or s, can also be shown to converge uniformly. This means
that we can formally differentiate or integrate both sides of (2.44) and
(2.45) to produce new integral relations which can then be related to
other integral transforms. Consider the following examples.
Example 2.4: Find the Fourier sine and cosine transforms of te- 01 ,
a> O.
Solution: Formal differentiation of both sides of (2.45), first with

respect to a and then with respect tos, gives us, respectively,
- L"" te-ar sin st dt = - 2as/(i + a2f

and
L= te-a 1
COS st dt = (a2 - s 2)/(s 2 + a 2) 2
Thus, we deduce that
01:: {
.r s te
- at ;s} = J2 -7T (s2 2as ) ,
+a2 2
a >O
and
- 2 2
a>o
01:: {
3- c te
-at.,s} --
J -2 ( a2 - s2)2,
7T s +a
Example 2.5: Find the Fourier sine transform of (1/t)e-a 1 , a > O.
Solution: We begin by integrating both sides of (2.45) with respect

to the parameter a from a to oo, which leads to
l=o t
1
- e -atsin st dt = f=
a
s
s 2 + a2
da
7T Ia
= - - tan- -
2 s
= tan -1 -s
a
Thus, it follows that
[!li{!
s t
e-ats} =
'
J~7T tan- ~a' 1
a> O
If we allow a ~ o+ in the result of Exam. 2.5, we find
f!Ji 5{1/t;s} = J~ (2.49)
This result is only a formal result since neither 1/t nor V7T/2 satisfy the
conditions of the Fourier integral theorem. Nonetheless, it can be useful
to treat (2.49) as a limiting case of the transform relation given in Exam.
2.5. Using (2.43), we obtain the similar relation*
f!Ji{l/t;s} = iV7T/2 sgn(s) (2.50)
* Formal results like (2.49) and (2.50) are discussed in more detail in Sec. 2.8.
Finally, as a bonus we see that (2.50) provides a generalization of (2.14),

which is
oo sin st
o t i 1T'
- - dt = - sgn(s)
2
(2.51)
21
Example 2.6: Find the Fourier transform of e-a \ a > O.
Solution: From definition, we have
C:{e -aZtZ ;s}

.:r = . . 11 Joo eist-aZtZ dt
v21T' -oo
By writing
we find
~{e-a2f2;s} = _l_ e-sZ/4a2 J"" e-(at-is/za)Z dt
y:x; -=
1
=--=e
- sZf4aZ Joo e - xZ dx
av'27r -oo
= _1_ e-sZf4aZ f(l/2)

ayi2;-
where we have made the change of variable x == at - is/a and used
properties of the gamma function. Simplifying this last result leads
to
<E{ - aZtZ s} =--e
:J'e 1 - s2f4a2 a>O
, ay'2 ,
By setting a = 1/0 in the result of Exam. 2.6, we obtain the

interesting relation
3P{e_,zl2;s} = e-sz/2 (2.52)
which says that the function e- 1212 is self-reciprocal, i.e., it is its own
transform.
EXERCISES 2.4
1. Given the following functions, develop the even and odd extensions,
fe(t) and fo(t), respectively:
(a) f(t) = e-at (b) f(t) = e- 12sin t
(c) f(t) = (1 + t)e-a' (d) f(t) = e-' + sinh t
2. Determine the Fourier transform of the function f(t), given that
g;s{f(t);s} = J~ (2e-bs - 1), b >O
ln Probs. 3-10, determine the Fourier transform of each function.
3. f(t) = {~~ar, ~~~.a >O 4. f(t) = {~: ~t~~s~

s. f(t) = { 1.. ltl O
1
12. Use the results of Exam. 2.4 to determine
(a) ~{te-a/tl;s}, a >O
(b) ~{itie-alrl;s}, a >O
13. Given the triangle- function f(t) = (1 - ltl)h(l - ltj), show that
(a) ~{f(t);s} = 2j~ sinz(s/2)
1T. i
(b) From (a), deduce that
f~~ (si:xy dx = 1T
14. Lettingf(t) = 1/Yt in the sine and cosine forms of Fourier's integral
theorem given by (2.33) and (2.36), respectively, show that
(a) r~ cos t dt = r~ sin_t dt = J~
Jo Vi Jo Vt 2
IS. Use the result of Exam. 2.6 with a = (1 - i)/2 to derive
(a) ~{cos(t2 /2);s} = ~Z [cos(i /2) + sin(s 2/2)]

(b) @i{sin(r/2);s} = 0 [cos(ii) - sin(i/2)]
(c) @is{; cos(f/2);s} = J~ [C(s/y;) + S(sjy';)]*

16. Show that
2
-2 as4 + 2a4 ,
3
(a) @ic{e -arcos at;s} =
J- n s + 4a
3 2
a>O
cn:
(b) ~c{e
-ar .
sm at;s} =
J 2 2a - as
-
n s4 + 4a4
, a>O
17. From the results of Prob. 16, deduce that
g;-t t4 ~ e;t} = ~: e-kltl/v'2 [ cos(kt/\/2) + sin(kjtlfy'2)J. k >o
18. Evaluate the sine transforms

(a) @is{e-a1 ;s}, a >O
(b) g;s{e-a 1COS at;s}, a > O
(c) @is{e-a1sin at;s}, a > O
19. From the results of Prob. 18, evaluate
(a) g;S{r4: e;s} g;s{l: e;s}
3
(c) @is{ (t4 : k 4) 2 ;s} (d) g;s{ (14 : k4)2 ;s}

20. Prove the following properties of the Fourier transform for real func-
tions f(t):
(a) If f(t) is even, then F(s) is real and even.
(b) If f(t) is odd, then F(s) is imaginary and odd.
(c) If f(t) is neither even nor odd, then F(s) has ao even real part
and ao odd imaginary part.
2.5 Properties of the Fourier Transform

The calculation of integral transforms is often tedious and quite complex
in some instances. However, once we have derived the transforms of
some standard functions, we can deduce the transforms of many other
* C(x) and S(x) are the Fresnel integrais (see Sec. 1.3.2).
2.5/Properties of the Fourier Transform 59
functions in a simple way through the use of certain operational properties

associated with the transform. These operational properties are basically
consequences of the properties of integrais.
If f(t) satisfies the conditions of the Fourier integral theorem, its
Fourier transform F(s) is uniquely determined by the integral
F(s) = ~ ;;:;--:_
1
y2rr
J"'-oo
eisr f(t) dt (2.53)
Thus, there is only one transform function F(s) associated with each
functionf(t). However, if/(t) and g(t) are two functions that are identical
everywhere except at certain isolated points, then both f(t) and g(t) will
have the sarne transform, say F(s). This means that the inverse transform
of F(s) can be either f(t) or g(t). Of course, the distinction between
functions that differ only at isolated points is mostly of academic interest
and has little effect in practical applications. If we agree to define a
function at a point of finite discontinuity as the average of its left-hand
and right-hand limits, then f(t) is uniquely related to F(s) by the inverse
transform relation
-1 [/(t+) 1
+ J(r)] = ~ ;;:;--:_ J"' e-'. F(s) ds (2.54)
2 y2rr -"'
Another important property of the Fourier transform and inverse
Fourier transform is the linearity property.
Theorem 2.2 (Linearity property). If F(s) and G(s) are the Fourier trans-
forms, respectively, of f(t) and g(t), then for any constants C 1 and C2 ,
it follows that
3'{Ctf(t)+ C2g(t);s} = C 1F(s) + C2G(s)
g;- {CtF{s) + C2 G(s);t} = Ctf(t) + Czg(t)
1
Proof: Directly from the defining integral, we have
@i{Ctf(t) + Czg(t);s} 1 J""

= ~ ;;:;--:_ eist [Ctf(t) + C2g(t)] dt
V 2rr -oo
c J""
::= _ I _
yz; -oo
eist f(t) dt + c J"'
_2_
yz; -oo
eist g(t) dt
= CtF(s) + C2G(s)
A similar argument proves the inverse transform linearity property.
If a> O, then
8Ji{f(at);s} 1 J""
= ~ ;;c eit f(at) dt
y27T -oo
avz; Joo
= _1-
-oo
eiu(s/a) f(u) du
where we have set u = at. Thus, if F(s) is the Fourier transform off(t),
we have just shown that
8Ji{f(at);s} = (1/a)F(s/a), a>O (2.55a)
Similarly, if a < O it follows that
8Ji{f(at);s} = -(1/a)F(s/a), a<O (2.55b)
so that in general we have the scaling property
8Ji{f(at);s} = (1/iai)F(s/a), a f: O (2.56)
Example 2.7: Given that the Fourier transform of f(t) =

(sin by'l+7)jy1 + r is F(s) = yf;/2Jo(Vb2 - s 2 ), isi b, determine the Fourier transform of
(sin bya2 + r)jya 2 +r, b >o.
Solution: We first observe that

sin by7+"7 sin bjajy1 + (t/af
=
ya + r
2
iaiVI + (t/af
and then using (2.56), we see that
2
sin bya + r. } = {yr;;ji lo (yb2a2 - rii), !as!< bla!
8Ji{ .. !2 ..2 ,s
va + l O, !as!> b!a!
_{yr;;jio, lo(iajyb
-
2
- s2), !si b
2.5.1 Shift Properties

Multiplication of either f(t) or F(s) by a complex exponential causes a
shift in the transform variable upon completing the integration of the
transform or inverse transform. More precisely, we have the following
theorem.
Theorem 2.3 (Shifting property). If f(t) and F(s) are Fourier transform
pairs, then
(a) 8Ji{eia1(t);s} = F(s + a)
(b) 8f{f(t - a);s} = ia F(s)
Proof: From definition,
@'{eia1(t);s} = 1 JQO eist eiat f(t) dt

Y27T -oo
= _l_ JQO ei(s+a)l f(t) dt

y/2; -oo
= F(s +a)
ln the sarne fashion,
@'-I {/a'F(s);t} = 1
Y27T
J=-oo
e-its eias F(s) ds
1 JQO .
=-= e-<t-a>s F(s) ds
Y27T -oo
= f(t - a)
@'{f(t - a);s} = eias F(s)

Example 2.8: Find the Fourier inverse transform of 1/(i + ias + b),
b >o.
Solution: By completing the square, we have

1 1
i + ias + b - [s + (ia/2)f + [(az/4) + b]
Then, using Theor. 2.3, *
@'-t { 1
s 2 + ias + b'
r} = e-at/2@'-t { 1
s 2 + [(a 2 /4) + b]'
t}
=Jaz 2_; 4b exp[ -i(at +Vaz+ 4bltl>]
the last step of which follows from Eq. (2.47).
2.5.2 Transforms of Derivatives and Derivatives of Transforms

ln applications involving differential equations it is important to know
how the Fourier transform behaves on derivatives of a function. If f is
continuous everywhere and f' is piecewise smooth, and both f and f'
are absolutely integrable, then
*Note that Theor. 2.3 implies that 3'- 1{F(s + a);t} = e;3'- 1{F(s);t}.
1 Joo eist f'(t) dt

.?F{f'(t);s} = ---=
V27T -oo
= _1_f(tkst
vz:;
I-oo
00 - is
V27T -oo
I"" eist f(t) dt
where we have employed an integration by parts. Now if falso satisfies

lim f(t) =O
/t/->=
we then obtain
.?F{f'(t);s} = - isF(s) (2.57)
where F(s) is the Fourier transform of f(t). By repeated application of
(2.57), we can prove the following more general result.
Theorem 2.4 (Differentiation property). lf J, f', ... , fn -n are continuous

everywhere and absolutely integrable, fn> is piecewise smooth and ab-
solutely integrable, and
lim f(t) = lim f'(t) = ... = lim J<n-n (t) = O
/t/->= /t/->oo /t/->=
then
.?F{J<n>(t);s} = ( -ist F(s), n = 1,2,3, ...
where F(s) is the Fourier transform of f(t).
Remark: lf /(t) has a finite jump discontinuity at t = then f'(t) a,

contains an impulse at t = a (see Sec. 1.5.2). ln this case the Fourier
transform off'(t) must also contain the Fourier transform ofthe impulse
function. Such concepts, which involve the notion of generalized functions,
will be discussed in Sec. 2.8.
ln the case of the cosine and sine transforms, the above results are
somewhat different. For example, in the case of the cosine transform
we use integration by parts to obtain
.?Fc{f'(t);s} = ....jiJ; Loo f'(t)cos st dt
= -V2/7T /(0) + s....jii; L= f(t)sin st dt

.?Fc{f'(t);s} = sF5 (s) - y2/7T /(0) (2.58)
Similarly, it can be shown that

~s{f'(t);s} = -sFc(s) (2.59)
(see Prob. 9 in Exer. 2.5). For second derivatives, we are led to the
relations
~c{f"(t);s} = -iFc(s) - \/2/7T f'(O) (2.60)
and
(2.61)
the verification of which is left to the exercises (see Prob. 10 in Exer.
2.5). These last two formulas give us some indication of which transform
- cosine or sine - to use in a particular application. That is, in any
problem in which f(O) is known but f'(O) is not known, we should use
the Fourier sine transform of f"(t). ln the sarne way, if f'(O) is known
rather than f(O), the Fourier cosine transform should be used.
If the transform of f(t) is F(s), then the transform of tmf(t), m =
1,2,3, ... , can be found by repeated differentiation of F(s). To see this,
let us start with the Fourier integral
F(s) = 1 Joo
~ ;;c_ eist f(t) dt (2.62)
y27T -00
and formally differentiate both sides with respect to s. This action yields
F'(s) = 1 Joo eist [itf(t)] dt

~ ;;c_
y27T -oo
and thus we conclude that
~{if(t);s)} = -iF'(s) (2.63)
Of course, the validity of (2.63) requires that the transform of tf(t) exist.
Continued differentiation of (2.62) with respect to s leads to
p<m>(s) = _1_ Joo eist [(it)mf(t)] dt, m = 1,2,3, ... (2.64)

\12; -00
which we now formulate as a theorem.
Theorem 2.5. If f is absolutely integrable and piecewise smooth, and

if ~f(t) has a Fourier transform, then
~{~f(t);s} = (- i)m p<m>(s), m = 1,2,3, ...
where F(s) is the Fourier transform of f(t).
By combining Theors. 2.4 and 2.5, we arrive at the result

Jm
S'{t"'j<">(t);s} = (- i)m+n dsm [s"F(s)], m,n = 1,2,3,... (2.65)
Example 2.9: Find the Fourier transform of te1- 1212 .
Solution: Recalling Eq. (2.52), we have

${e-r2/2;s} = e-s2Jz
Then, using (2.63), it immediately follows that
${te-r2fz;s} = -i( -s)e-s2fz = is e-s2Jz
Finally, recalling Theor. 2.3, we deduce that
${ter-r2fz;s} = i(s - i)e-(s-iJl/Z
= (1 + is)eis e-(s2-J)/2
EXERCISES 2.5
1. lf f(t) is a real function with transform F(s), show that the complex
conjugate of F(s) satisfies
F(s) = S'{f(t);- s}
2. If f(t) is a complex function with transform F(s), show that the
complex conjugate of F(s) satisfies
F(s) = ${f(- t);s}
3. If F(s) is the Fourier transform of f(t), show that
Si{eibr/af(t/a);s} = aF(as + b), a> O
4. Show that
(a) Sic{f(at);s} = (1/a)Fc(s/a), a>O
(b) $ 5{f(at);s} = (1/a)Fs(s/a), a>O
5. Show that
(a) S'c{f(t)cos at;s} = i[Fc(s + a) + Fc(s - a)]
(b) S'5{f(t)cos at;s} = ![F5(s + a) + F5(s - a)]
6. Show that
(a) S'c{f(t)sin at;s} = UF5 (s + a) - F5 (s - a)]
(b) $ 5{f(t)sin at;s} = UFc(s - a) - Fc(s + a)]
7. Use the results of Probs. 5 and 6 to show that
2.6/Transforms of More Complicated Functions 65
- 2
-2 as4 + 2a4
3
Oi:
(a) :'JI c{e
- at
cos at;s}
J-
=
'1f' s + 4a
3 2
, a>O
m:
(b) :'Jic{e
-at .
smat;s}
J
= - 2 2a - as
4
'1f' s + 4a
4 , a>O
8. V se the results of Probs. 5 and 6 to evaluate

(a) ~s{e-a'cos at;s}, a >O
(b) ~s{e-a'sin at;s}, a >O
9. Verify that the Fourier sine transform satisfies the relation
~s{f'(t);s} = -sFc(s)
10. Derive the transform relations
(a) ~c{f"(t);s} = -s 2Fc(s) - y2/7T'f'(O)
(b) ~s{f"(t);s} = -s 2F5 (s) + y2/'Tf's/(O)
ln Probs. 11-l5, evaluate the Fourier transform of the given function
using known transforms and appropriate properties of the transform.
11. f(t) = (1 - t)e- 1' 1 12. f(t) = ebr-P
13. f(t) = t2e -t2!2 14. f(t) = e _,212 cos 2t
15. f(t) = t cos t2
ln Probs. 16-18, evaluate the Fourier inverse transform of the given
function using known transform relations and appropriate properties of
the transform.
e-2is
1
16. F(s) = 17. F(s) =
2
s + 4s + 7 s 2 + 4s + 7
18. F(s) = tan- 1(s/a) sgn(s)
Hint: Examine F'(s).
19. Show that, under appropriate assumptions on f and its derivatives,
~c{j<4 >(t);s} = iFc(s) + y2/'Tf' [sY'(O) - t<3>(0)]
20. Show that, under appropriate assumptions on f and its derivatives,
~s{f<4 >(t);s} = s4F 5 (s) - y2/'Tf' [sY(O) - sf"(O)]
2.6 Transforms of More Complicated Functions

When the functions involved in a Fourier transform or inverse transform
are of a more complicated nature, we usually must resort to techniques
other than the standard integration methods of calculus. Sometimes it is

useful to represent part of the integrand in a power series and perform
termwise integration on the resulting expression. * ln certain cases the
resulting integrated series can be summed to yield the transform we are
seeking. Other useful techniques are those involving the powerful methods
of complex variables.
To illustrate the power series method mentioned above, let us consider
the following example.
Example 2.10: Find the Fourier cosine transform of (a 2 - t 2)P- 112

x h(a - t), p > -1/2, a > O, where h(t) is the Heaviside unit
function.
Solution: From definition of the cosine transform, we have
3'c{(a2 - r)P- 112h(a - t);s} = J~ f (a 2 - t 2)P- 112 cos st dt
= J~ ~ -1)* ik
1r k=O
( (2k)!
ra(a2 - fy-1/2 t2k dt
Jo
where we have replaced the cosine function with its power series
representation. The substitution t = a sin (} in the above integral
leads to
J: (az - fy-1/2 tzk dt = azp+2k L1T/2 cos2P(J sin2k8 dO
= a2p+Zk f(p + !)f(k + !)

2 f(p + k + 1)
by use of Eq. (1.11) (see Chap. 1). Next, employing the duplication
formula of the gamma function [see (G7) in Sec. 1.2]
y:;f(2z) = 22z- 1 f(z)f(z + !)
we find that
2
@Pc{(az _ ry-112h(a _ t)s} = a Pf(p + !) ~ ( -1)kf(k + !)(asfk
' V21T k=O (2k)! f(k + p + 1)
= 2P_ 112 f(p + !)(a/s)P ~ ( -1}*(as/2)2k+p

k=O k!f(k + p + 1)
This last power series is recognized as the Bessel function JP(as) (see
Sec. 1.4), and thus we have our result
* The power series in such cases must converge everywhere.

~ c{(a 2 - r)P- 112h(a - t);s}

= 2p-l/ 2f(p + !)(a/sr Jp(as), p > -1/2, a> O
Because the Fourier cosine transform and inverse cosine transform

are identical operations, we can use the result of Exam. 2.10 to deduce
the additional cosine transform relation
- (az - szy-1/Zh(a - s)
~c{t P JP(at);s} = P_ 112aPr(p + !)
2
, p > -1/2, a> O (2.66)
Also, for p = O we obtain the special case

2 1
~{J0(at);s} = ~d/o(at);s} = { J;Va 2
- sz'
lsl <a
lsl >a (2.67)
O,
2.6.1 The Use of Residue Theory

The calculus of residues from complex variables is a powerful tool in
the calculation of many transform formulas. To deal with some of the
integrais that will arise, we will need to use the theorems from complex
variables provided in Appendix A.
To begin, we wish to derive the pair of transform formulas:
~c{ta-J;s} = J~7T f(a)

sa
cos(7Ta/2), s > O, O < a < 1 (2.68a)
ar; {a-I }
:1' st ;s = J2f(a) . (7Ta / 2) ,
-7T - sa sm s >O, O< a< 1 (2.68b)
Let us define the complex function f(z) = za- 1 e- sz and integrate it around
the closed contour shown in Fig. 2.4. From Cauchy's integral theorem
(Theor. A. I in Appendix A), it follows that
fcf(z)dz =O (2.69)
or
fCp
f(z)dz + JRp f(x)dx + JCR f(z)dz + JR(P f(iy)d(iy) = O (2.70)
Along the imaginary axis we have set z = iy, so that

J(iy) = ;a-lya-le-isy
If we now allow p ~ O and R ~ oo, we have from Theors. A.3 and
A.4 in Appendix A that
X
p R
Figure 2.4 Contour of integration
lim
p->0
J f(z)dz = O,
Cp
lim
R->=
J f(z)dz = O
CR
(2.71)
and hence (2. 70) reduces to
L'"" x"-le-sxdx + i" J~ y"-le-isydy = o

Now setting ;-" = e-hralz, we can rewrite this last expression as
(2.72)
ln order to evaluate the integral on the right, we make the change of

variable u = sx to get
L= x"-le-xdx = s-aL= e-u ua-1 du
= s-"f(a)
l = y a-1 e -isydy=--e
o
f(a) -iwa/2
s"
(2.73)
Finally, multiplying both sides of (2.73) by the constant factor y2/'TT'

and equating real and imaginary parts, we are led to the desired formulas
given by (2.68a) and (2.68b).
If we let a = 1/2 in (2.68a) and (2.68b), we get the special cases
@'c{1/yt;s} = 1/y~ (2.74a)
and
@'s{l/yt;s} = 1/VS (2.74b)
which shows that 1/Vt is self-reciprocal under cosine and sine

transformations.
The residue calculus is especially helpful in finding transforms of
rational functions. To illustrate the technique, let f(z) denote a complex
function with the following properties:
1. f(z) has a finite number of poles a~> a 2 , , an in the upper half-
plane.
2. f(z) is analytic along the real axis except at the .points b 1, b2 , ,
bm, which are simple potes.
3. if(z)eisz ~ O as z ~ oo, Im(z) > O.
Suppose we integrate the complex functonf(z)e;z, s >O, around the
contour shown in Fig. 2.5. By use of the residue theorem of complex
variables, we find that
~I r j(x)esxdX +
k=l JLk
i
k=l
r f(z)szdz + r
Jck JcR
f(z)eiszdz
n
= 27Ti L Res{f(z)eiz;ak}
k=l
where L 1 , L 2 , ,L;..+, are the straight line segments along the x axis
and C~o C2 , ,Cm are small semicircles with centers at the simple poles
b 1, b2 , , bm. ln the limitas R~ oo and the radii ofthe small semicircles
tend to zero, we obtain
f~= f(x)eisxdx - 7T ktl Res{f(z)eiz;bk} == 27Ti ktl Res{f(zksz;ak} (2.75)
where (see Theor. A.3 in Appendix A)
lim ( f(z)eiszdz = O
R-+= JcR
y

Thus, we have developed the general transform formula for rational

functions (changing x to t in the integral)
@i{j(t);s} = iy27T [
k=l
Res{f(z)i;ak} +! ~ Res{f(z)eiz;bk}]
2 k=l
(2.76)
for s >O.
lf f is either even or odd we can extend the result (2.76) to include
s < O by utilizing the relations (2.40) and (2.43). For more general f we
can set s = -a-< Oand integrate the functionf(z)e-;"z around a contour
similar to that in Fig. 2.5, but in the lower half-plane. Equivalently, we
can replace z by - z and integrate the function f(- z)e-iz around a
contour in the upper half-plane. The result of this latter approach is the
transform relation (see Prob. 9 in Exer. 2.6)
[~ Res{f( -z)e-i;ak}
@i{j(t);s} = iVbr
1
1 M
+2 "f; .
Res{f( -z)e_,.z;f3k}
1
]
(2.77)
where s < O. Here a 1 , a 2 , , aN are the poles of f(- z) in the upper

half-plane and /3~o /3 2 , , 13M are simple poles of f(- z) along the real
axis.
For calculating inverse Fourier transforms by this method, we simply
observe that the cases corresponding to positive and negative t are the
reverse of those for positive and negative s as a consequence of the fact
that the kemel of the inverse transform is the complex conjugate of the
kemel of the transform.
Example 2.11: Find the Fourier transform of 1/t(P + k 2), k > O.
Solution: The complex function

f(z) = 1/z(i + e)
has simple poles at z = O and z= ik. Hence, calculating residues
at z = O and z = ik, we find
Res{f(z)eiz;O} = lim zf(z)eiz = 1/e
z--+0
and
Res{f(z)eiz;ik} = lim (z - ik)f(z)eiz = -e -ks /2k 2
z--+ik
From (2. 76), we now have
1 } - ( I e-ks)
g;{ 1(1z + JC) ;s = iV2TT 2JC - 2/C
J
1T i 1
= "2/C( -e -ks) , s> O
Because f(t) = I/t(12 + JC) is an odd function, we can use (2.43) to

deduce that, for all s,
;' {
t(t2
1
+e)'
s} = J~2 _i_(l
e - e-kJI)sgn(s)
Example 2.12: Find the inverse Fourier transform of F(s) =

1/(s2 + ias + b), a > O, b > O.
Solu1ion: By definition,
1
g;- 1{F(s);t} =---= J"" .
e-lls F(s) ds
V2TT -=
= ~ 11 J" 00 eits F(-s) ds

y27T -=
where we have replaced s by - s. The complex function
1
F( -z) = zz - zaz
. +b
has simple poles at z = z 1 and z = z2 , where

i
Zt = 2(a + v'a2 + 4b)
i
Z2 = 2(a - Ya 2
+ 4b)
Clearly, z1 lies in the upper half-plane while z2 is in the lower half-
plane. Calculating the residue at z = z1 leads to*
Res{F(- z)eitz;zt} = eitz

.
I
2Z - la z=zl
and thus we have
* Recall that if z = a is a simple pole of f(z) = P(z)/Q(z), then
Res{f(z);a} = lim (z -Q~);(z) = P(a)/Q'(a).

z-tJ Z
f!F-1{ i+ ias+
1 r}
b'
= Ja 2 ~ 4b exp[ -i(a + 2
ya + 4b)t]. t >O
For t < O, we consider the complex function

1
F(z) = z2 + iaz + b
which has simple poles at z = z3 and z = z4 , where

i
z3 = -Z(a + ya 2 + 4b)
i .. I 2
Z4 = -Z(a- ya + 4b)
The pole at Z4 lies in the upper half-plane and
Therefore,
;JF-1{ s 2
1 t}
+ ias + b'
=Jaz ~ 4bexp[ -~(a- ya 2

+ 4b)t]. t<O
and by combining results, we deduce that for ali t, *
.'1'- Lz + i!s + b;t} =Jaz~ 4b exp[ -i(at + ya

1 2
+ 4bltl>]
Example 2.13: Find the Fourier cosine transform of (cosh at/cosh t)

and Fourier sine transform of (sinh at/cosh t), lal < I.
Solution: To start, let us integrate the complex function

e(a+is)z
f(z) = cosh z
around the rectangular contour shown in Fig. 2.6. The function f(z)
has simple poles at z = (n + l/2)7Ti, n = O, 1, 2, ... , but the
* Recall that we previously found this inverse transform relation througb use of the
shifting property (see Exam. 2.8).
-R + i1r ------+---.CJ. . .--, R + i1ri7T
u i7T/2
-R o R X
only pole enclosed by the rectangular contour C is i7T/2. Hence, from

the residue theorem (Appendix A) it follows that
fcf(z)dz = 27Ti Res{f(z);i7T/2} = 27Te<ia-shr/2

This can also be written as
J-R f(x)dx + fG f(z)dz + f f(z)dz + f f(z)dz = 27Te<ia-s)w/

2
R
~ ~
Along the line segment C2 we have z = R + iy, and thus

"' I
I Ii "'
eaR-sy ei(sR+ayl
fc2 f(z)dz
I =
o cosh(R + ty)
. dy
i
e-sy
:5 eaR dy
o jcosh(R + iy)j
lt can be shown that the last integral vanishes in the limit as R -
oo (see Prob. 20 in Exer. 2.6). The sarne in true of the integral along
the line segment C4 We have z = x + i7T along the segment C3 ,
which leads to
-R e<a+is)(x+iTr)
fc,
f(z)dz =
f
R cosh(x
R
+
. dx
l7T)
(a+is)x
= e(ia-s)Tr
I -R
_e_ _
cosh x
dx
Therefore, as R - oo we are left with

ao (a+is)x
(1 + e(ia-s)Tr)
J _e__ dx
-ao cosh x
= 27Te(ia-s)Tr/2
or
e<a+is)x - 2Tre(ia-s)Tr/2
J
oo
dx - (ia-s)Tr
-oo COS h X 1+e
2Tr
1T
= ------::-------=
cosh[ (ia ~ s)Tr]
By splitting this last expression into real and imaginary parts, we get
foo
Joo
-oo
cosh ax
COS h X
cos sx dx
1T
+ l
-oo
sinh ax .
h sm sx dx
COS X
a1r h s1r . a1r . h s1r

cos- cos
2
2- l. sm- 2
sm 2
a1r s1r . a1r . h s1r
T 1r cos
cosh 1r sm 2 sm 2
2
=------------------------+i-----------------
cos2-a1T cosh 2 STT + sin2 _a1r sinh 2 STT cos 2-a1r_ cosh 2s1r + sin2-a7r sinh 2~
2 2 2 2 2 2 2 2
and by comparing real and imaginary parts, and simplifying the algebra,
we deduce that
a1r S1T
cos-cosh-
oz: {cosh at. } _ ~ 1- 2 2
:f~Pc h ,s - V 21T ' lal < 1
cos t cos a1r + cosh S7T
a1r . hs1r
~v;;;-::
810 810
oz: {sinh
:JP s
at. } _
h ,s - 2~1T 2 2 lal < 1
cos t cos a1r + cosh s1r '
EXERCISES 2.6
l. Find the Fourier cosine transform of h(a - t), a > O and compare
this with Exam. 2.10 to deduce that
Jl/2(x) = J 2
1TX
sin x
2. Given thatf(t) = (t2 - Ith(l - ltl), n = 0,1,2, ... , show that

oz; ( -IYn! (2)n+l/2
:JP{f(t);s} = V2 ~ ln+ l/2(s)

;n
~{Pn(t)h(l - ltl);s} = Vs In+tds), n = 0,1,2, ...
where Pit) is the nth Legendre polynomial defined by
Pn(t) = 2"~! :;.[(12- l)"l n = 0,1,2, ...

4. Show that
~5{t(a 2 - f)P- 312h(a - t);s}
= 2p-Jf 2aPs 1 -Pf(p - 1/2)Jp(as), a> O, p > 1/2

5. Integrate both sides of Eq. (2.67) with respect to s to deduce the
Fourier sine transform of (1/t)J0(at).
6. Show that
CJi::
(a) 3-{ltl
-a
;s} = :;;
JZ f(l - a) . 1TQ'
lslt-a sm 2, O<a<1
z f(llslt-a
(b) ~{ltl-"sgn(t);s} = i
J
:;;
- a) 1Ta
cos 2' O<a<1
7. Given that r = ya 2
+i and tan 8 = s/a, show that
a> O, p >O
~s{e
- - J2
attp
1
;s} = -
1T r
f(p) sm
-p
. p(J, a> O, p >O
Hint: Consider the integral ofj(z) = zP-te-(s-iaJz around the contour

in Fig. 2.4.
8. Provide the details leading to Eq. (2.77).
ln Probs. 9-14, use residue theory to find the Fourier transform of the
given rational function.
1 2it
9. f(t) =f +9 10. f(t) = -2-1
t +
l l
11. j(t) = t4 + I 12. f(t) = -6 -1
t +
l P-t
13. f(t) = (12 + 4)2 14. j(t) = t(t2 + I)
ln Probs. 15-17, use residue theory to find the inverse Fourier transform
of the given rational function.
2is + 1
15. F(s) = -2 - l 16. F(s) = ~
s~ + 1
s + 1
1- i
17. F(s) = (i + 4)2
18. By expanding J0(at) in a power series, use residue theory to deduce
that
= t2k . d2k f= eist

Hint: Observe that f-=[
y--
+1
e,.., dt = (- 1i d 2k
s -=t
- 2- -
+1
dt
19. Following the suggestion in Prob. 18, derive the Fourier sine transform
of t(r + 1)- 1Jo(at).
20. Show that
(a) icosh(R + iy)i ;::: eR /4, R ~ oo.
(b) Use (a) to show that the integral along C2 in Exam. 2.12 satisfies
the inequality
ILJ<z)dz I 1
::5 47Te a-JJR, R~ oo
and thus deduce that lim
R-.-x-
r f(z)dz
Jc2 = o.
. . . sinh(z/2) . . cosh(z/2) ..
21. By mtegratmg the funct10ns . h e''z and . h e''" around the
sm z sm z
contour shown in the accompanying figure, deduce that
cn: {sinh(t/2). } _
3'c sinh t ,s -
J~2 sech 7TS
cn:
3's
{cosh(t/2).
h
} _
,s -
J~ tan h 7TS
SIO t 2
-R+ 27Ti~-----.._ ,___ _,...._ __, R + 27Ti
-R -p p R X
22. Given the transform function F(s) == i"14 jy'2s,

(a) show that
g;- {F(s);t} = O,
1
t < O
by integrating an appropriate complex function around a closed
contour in the upper half-plane.
(b) By using the contour shown in the accompanying figure, show
that
1 L= e-fr
g;- 1{F(s)t} = --= -= d~ t >o
' v'1r o v'~ '
and thus deduce that
g;-t{ ~ ;t} ~t h(t) ==
ln Probs. 23-30, establish the given transform relation.
23. f:F c{(l - t)h(l - t);s} == J~'TTS~(1 - cos s)
cn;; { sinh at . } _ _ 1_ sin a

24 3-c . h ,s - ~ ;;::;-::_
sm 1rt h
y 21r cos s + cos a
' lal < Tr
cn;; {cosh at } 1 sinh s
25
3-s sinh 1rt ;s == y'21r cosh s + cos a'
ial < Tr
e-altl } (y's2 + a2 + a)l/2

26. ~
{ --;s = , a>O
Vi 2?T(i + a 2)
sin;ts }
27. ~s { - 1
= ~ ~log 11-+-sI
t y2?T 1- s
28. ~c{r 1 (e-b' - e-a');s} = y'~?T log (;: : ::). a> b
29. ~s{r 1 (e-b' - e-a');s} = j~ tan-{ ~~ ~ :~]. a> b
30. ~ei~~a~; 1\s} = 1z lo(ay'b2 - s2)h(l - ls/bl)
2.7 The Convolution Integrais of Fourier

One ofthe most important operational properties ofthe Fourier transform
is the convolution theorem. To derive this important property, let us
begin by defining the convolution of two integrable functions f and g by
the expression
(fo g)(t) 1 Joo

=~~ f(u)g(t - u) du (2.78)
y2?T -oo
The Fourier transform applied to this convolution integral leads to
~{(fo g)(t);s} = -1 Joo eist Joo f(u)g(t - u) du dt

2?T -oo -oo
1 Joo foo-oo eis/ f(u)g(t
= 2?T -oo - u) du dt (2.79)
We can interpret (2.79) as ao iterated integral for which the order of

integration can be interchanged. Hence, by changing the order of integration
and making the change of variable t = u + x, we find
~{(fo g)(t);s} = -1 Joo foo eis(u+x>.J(u)g(x) dx du

-oo -oo
2?T
=~~1 Joo eisuf(u)du ~ ~1 Joo eisxg(x)dx

y2?T -oo y2?T -oo
and thus conclude that
~{(fo g)(t);s} = F(s)G(s) (2.80)
where F(s) and G(s) are the Fourier transforms, respectively, off(t) and
2.7/The Convolution Integrais of Fourier 79
g(t). By applying the inverse Fourier transform to both sides of (2.80),

we obtain the alternate form
g;- 1{F(s)G(s);t} = (fo g)(t) (2.81)
This shows that the inverse transform of a product of transform functions
can be found by convolving the inverse transforms of each product term.
Equation (2.81) is the Fourier convolution theorem, which sometimes
is expressed in the form
I~= e-isr F(s)G(s) ds = I~=f(u)g(t - u) du (2.82)
An interesting consequence of (2.82) follows by first setting t = O, which

yields
I~= F(s)G(s) ds = I~=f(u)g( -u) du (2.83)
For the special case where g(- u) = f(u), then*

G(s) = ff{g(u);s} = ff{f( -u);s} = F(s)
(see Prob. 2 in Exer. 2.5) and we obtain
I~,., F(s)F(s) ds = I~f(u)f(u) du

or
(2.84)
Equation (2.84) is called Parseval' s relation. ln physical applications

the quantity on the right-hand side represents the total energy in a
waveform, such as a sound wave or electrical signal. Thus, Parseval's
relation states that the total energy is given by the area under the IF<sW
curve. For this reason, the quantity IF(s)IZ is called the energy spectrum
or energy spectral density function of f(t). Engineers are undoubtedly
familiar with the Fourier series counterpart of Parseval's relation, which
has the physical interpretation that the power associated with a periodic
function equals the sum of the powers associated with its harmonic
components.
Returning now to the convolution integral (2.78), we observe that it
satisfies certain formal properties of ordinary products. For example, if
C is a constant, it is immediately clear that
fo (Cg) = (Cf) a g = C(fa g) (2.85)
* ln most cases of interest to us the function f(t) is real and thus f(t) = f(t).
Similarly, it follows that the distributive law holds, i.e.,

Jo(g + k) =Jog +Jok (2.86)
By definition,
(g a f)(t) 1 f=
= ;;;-:_ g(u)f(t - u) du
v 27T -=
and by making the change of variable v = t - u, we have
(gof)(t) = -./
y27T
J.-= g(t- v)f(v)dv
=
=. 11- f= f(v)g(t- v) dv
y27T -=
from which we deduce the commutative law

Jog=goJ (2.87)
Finally, the convolution integral (2. 78) also satisfies the associa tive law
Jo(gok) = (fog)ok (2.88)
but we leave the proof of this result to the exercises (see Prob. 8 in
Exer. 2.7).
Example 2.14: Use the convolution theorem to evaluate the inverse

Fourier transform
Qj;"-l{
Yft sin s . ;t }
s(l - IS)
Solution: Let us define

sin s 1
F(s) = - and G(s) = - - .
s I - IS
which have inverse Fourier transforms
g;- 1{ -sin- s ;t } = J7T2 h(l - Iti)

5
and
1
g;- 1{ --.
I - IS
;t} = V27Te_, h(t)
Thus, using (2.81), we have

~- 1 {F(s)G(s);t} = J~2 f"" -ao

e-<r-u)h(t - u)h(l - !ui) du
J~2 e- 1 fr eu du,
-1
ltl < 1
J~2 e-r fi eu du,

-1
t >1
O, t < -1
or
J~ (I _ e -<r+ ll), ltl < 1
~-~{ sins t } =
;'f
s(l - is)' J~ (e- 1
e- )e-
1
, t> 1
O, t < -1
Example 2.15: Evaluate the integral
I-f=- - = (xz +
dx
az)(xz + bz)
Solution: The given integral has the form
I = f~= F(x)G(x) dx
where F(x) = 1/(x2 + a 2) and G(x) = 1/(x2 + b2). Thus, by the use
of Eq. (2.83), we immediately can relate this integral to
J = f~oo J(t)g(- f) d/
where f(t) and g(t) are the inverse Fourier transforms of F(x) and
G(x), respectively. Recalling Exam. 2.3, we see that
f(t) =! J~ e-altl and g(t) =! J~ e-bltl

a 2 b 2
and using the fact that these are even functions, we deduce that
I = 2 Loo j(f)g(f) df
= .!!._ r= e -(a+b)t dt
ab Jo
or
2. 7.I Cosine and Sine Convo/ution Integrais

Other convolution integrais involving the cosine and sine transforms can
be derived similarly, but the resulting integrais are more complicated
than (2.82). For example, ifjand gare functions defined for t >O which
have cosine transforms Fc(s) and Gc(s), respectively, then from previous
results we know that
~{f(ltl);s} = Fc(s)
and
~{g(ltl);s} = Gc(s)
The substitution of these expressions into (2.82) yields
f~oo e-ist Fc(s)Gc(s) ds = f~oof(lul)g(lt- ui) du (2.89)
Since both Fc(s) and Gc(s) are necessarily even functions, we can write
(2.89) as
2 L"' cos (st)Fc(s)Gc(s) ds = Loo f(u)g(it - ui) du + J:oof(iui)g(it - ui) du

= fooo f(u)g(lt - ui) du + Loo f(u)g(t + u) du,
and thus we have derived the convolution integral (t > O)
roo l roo
Jo cos(st)Fc(s)Gc(s) ds = 2Jo f(u) [g(it - ui) + g(t + u)] du (2.90)
Without providing the details, we simply state that the following

convolution integrais can be derived in a likewise manner (see Exer.
2.7):
roo I roo
Jo cos(st)Fs(s)Gs(s) ds = 2Jo f(u) [g(u + t) + g(u - t)] du (2.91)
roo I roo
Jo sin(st)F5 (s)Gc(s) ds ; 2Jo f(u) [g(iu - ti) - g(u + t)] du (2.92)
roo l roo
Jo sin(st)Fc(s)Gs(s) ds = 2Jo f(u) [g(u + t) - g(u - t)] du (2.93)
EXERCISES 2.7
1. Verify the convolution integral (2.82) for
(a) f(t) = g(t) = h(l - ltl>
(b) f(t) = g(t) = e-r2f2
2. Use the convolution integral to find the inverse Fourier transform

of
1
F(s) = (1 - ls)2 = -
1 --is. -
1 --
is
3. Show that the convolution theorem can also be expressed in the
form
J~"" eist f(t)g(t) dt = J~"" F(u)G(s - u) du
and use this result to evaluate the Fourier transform of e-ltl sin t.
t
4. For a,b > O, show that
r
f oo
-oo (x 2 + a2)(x 2 + b2)

dx -
-
_'Tr_
a+b
S. Use the Fourier transform relation
@'{(a 2 - t2) -l/2h(a - lt!);s} = J~J0(as), a >O
to show that
("" 2
Jo J 0(ax)J0(bx) dx = 1rbK(ajb), O<a<b
where K(m) denotes the complete elliptic integral

f"/2
K(m) = Jo (1 - m2sin28) -l/2 dO
6. For O < a < b, show that
fo
oo
x

smaxsm x -2 = -
2
b dx 7ra
7. Verify Parseval's relation (2.84) for

(a) f(t) = e-alri, a> O
(b) f(t) = e-a 21
\ a> O
8. Show that the Fourier convolution (2. 78) satisfies the associative
property
Jo(gok) = (fog)ok
9. For real g(t) = f(t), show that (2.90) leads to the Parseval relation
L= iFc<sW ds =L"" if<uW du

10. Derive Eq. (2.91) by replacing G5(s) with its defining transform integral
and interchanging the order of integration.
11. Derive Eq. (2.92).
Hint: See Prob. 10.
12. Derive Eq. (2.93).
Hint: See Prob. 10.
13. For real g(t) = f(t), show that (2.91) leads to the Parseval relation
J: IFs<sW ds = L"' if<uW du

14. By substituting the functions
f(t) = r'", o< a < 1
2
g(t) = (1 - t Y- 112h(l - t), v> -1/2
into Eq. (2.90).
(a) show that
~ sin( a;)ro - a)f(v + !) Jo= s"-v- 11v(s) ds = f r"(l - t 2Y- 112 dt
(b) Using properties of the beta function (see Prob. 21 in Exer. 1.2),
evaluate the integral on the right in (a) and deduce that
r= 2a-v-lr( /2)
Joxa-v-llv(x)dx=f(v+ ::a/ ), O<a<l,v> -1/2
1 2
15. Using Eq. (2.90),
(a) show that
(E
3- c
{ ]
2
0(at) . }
1 ,s
_
-
J~ e -s ia " cosh
I 2
u d
2 u,
a >O
t+ 1T oya-u

2.8/Transforms Involving Generalized Functions 85
o;;; {
:-!/'c 12
lo(at) . } _
+ 1 ,s -
J~2 e -sIo(a ), a>O
where I 0(a) is the modified Bessel function of order zero.

Hint: Express cosh u in a power series and use properties of the
beta function (see Prob. 21 in Exer. 1.2).
2.8 Transforms Involving Generalized Functions

ln Sec. 2.4 we stated that the Fourier transform F(s) is a bounded function
provided the inverse transformf(t) is absolutely integrable, i. e., provided
f~oo IJ(t)l dt < 00 (2.94)
This is actually a sufficient condition of the Fourier integral theorem for

the existence of the transform of a given functionf(t). Sinusoidal functions,
the Heaviside unit function, polynomials, and so forth, do not satisfy
the condition (2.94) and, therefore, do not have a Fourier transform
in the usual sense. However, it is possible to extend our definition of
Fourier transform to include such functions if we are willing to consider
the notion of generalized functions such as the impulse function (t) and
its derivatives a<nl(t), n = 1,2,3' ....
The impulse function was first introduced in Sec. 1.5.2. lt is a useful
concept in a wide variety of physical problems involving the ideas of
line spectrum, impulsive forces, or point sources. Because the rules of
manipulation of impulse functions do not follow naturally from the methods
of classical analysis, such functions are often referred to as generalized
functions. The general theory of sue h functions has been put on a solid
mathematical basis under the title of the theory of distributions. * None-
theless, our treatment of impulse functions and their derivatives will
continue to be based upon formal manipulations as developed in Sec.
1.5.2.
2.8.1 Impulse Function

By using the "sifting property" of the impulse function
* The most notable pioneering work on generalized functions is contained in L. Schwartz,

Thorie des Distributions, Tomes I and 2, Paris: Hermann and Cie, 1950, 1951. See also
A. H. Zemanian, Distribution Theory and Transform Analysis, New York: McGraw-Hill,
1965.
J~= (t - a)f(t) dt = f(a) (2.95)
we immediately obtain the formal result
1
.:1f{(t);s} = --= J"" e'st. (t) dt = 1
--= (2.96)
v27r -= v27r
Hence, we sayjhe Fourier transform of the impulse function (t) is the
constant 1/V2Tr. On the other hand, we can use this transform relation
to deduce that
.:1'- 1{1;1} = .. 11 J"" .

e-m (1) ds - (s)
= V2Tr (2.97)
v2Tr -=
Of course, the integral in (2.97) has no meaning as an ordinary integral.
Its interpretation lies strictly in the fact that it is used to define the
generalized function (s).
To better understand how the concept of an impulse function might
arise in a practical situation, let us briefl.y discuss the case where f(t)
represents a function in the time domain. For example, if f(t) is a waveform
- an electrical signal like a voltage or current, or an acoustic wave or
optical wave, etc.- the Fourier transform off(t) describes the waveform
in the frequency domain. lt is customary in this setting to let s = w,
where w is the angular frequency variable. Although the waveform phys-
ically exists in the time domain, we can say that it consists of those
components in its frequency domain description called its spectrum.
Mathematically, we write*
1
F(w) =
v27r
J""-= ' f(t) dt = iF(w)iei<f>(w)
1
(2.98)
where iF(w)i is the spectrum amplitude function and cp(w) describes the
spectrum phase.
The sinusoidal functions f(t) = cos w0t, which represents one of the
simplest waveforms possible, does not satisfy the condition (2.94), and
hence, does not have a Fourier transform in the strict sense. Yet, from
a purely physical point of view we know this waveform has a single
frequency component, or fine spectrum, at w = w0 This apparent con-
tradiction can be explained by recognizing that the function cos w0t
cannot exist for all time - < t < = =
as we assume in the formal
definition of the Fourier transform. Like all waveforms, the cosine
waveform exists for only some finite interval of time, and as such will
* ln much of the engineering literature, the spectrum function is defined by a multiple

of the complex conjugate of (2.98).
2.8/Transforms lnvolving Generalized Functions 87
actually satisfy the condition of absolute integrability. For instance, let

us assume that f(t) = (cos w0t)h(T - jtj), where 2T is the interval of
time that the waveform is present. ln this case, the Fourier transform
of f(t) leads to
g;{f(t);w} = ~ 11- JT eiwt cos Wol dt

V 2'7T -T
= J~ LT cos wt cos w0t dt
or
F(w) = 1 [sin(w + w0 )T + sin(w - w0)T] (2.99)
y2'7T w + w0 w - w0
The graph of jF(w)j is shown in Fig. 2.7 for a fixed value of T. By allowing
T to become unbounded, it can be shown that the graph becomes more
and more narrow and peaked around w = w0 and w = - w0 Hence, in
a formal sense we may consider the Fourier transform of cos w0t for ali
time as the limit
~{
;}JP COS Wot'W
} .
= 1lm -=
1 [sin(w + w0)T sin(w - w0)T]
+ (2.100)
' r-.oo y2'7T W + Wo W - Wo
provided this limit is meaningful.

ln order to interpret the limit (2.100), let us recall that in the proof
of Theor. 2.1 (see Sec. 2.3) we established the limit relation
1
f(x) = :; 1Joo
00
0
-oo f(t) cos[s(t - x)] dt ds
(2.101)
= _!. Joo f(t) [ lim sin (t - x)] dt
1T -oo -'>00 f - X
Based on the sifting property (2.95), it becomes clear that the limit in
(2.101) leads to the formal definition
jF(w)j
w
-wo Wo
Figure 2. 7 Spectrum of sinusoidal pulse

1 . sin (t - x) "'( )
-tm1 =ut-x (2.102)
7T -->co t - X
With this interpretation of the limit, we see that (2.1 00) gives the expected
result*
$'{cos w0t;w} = YTT/2 [(w + w0) + (w - w0 )] (2.103)
Hence, the notion of impulse functions is consistent with our previous
claim that the spectrum of cos w0t should be a single line at frequency
w = w0 (and from symmetry of the transform, a line also at w = - w0 ).
Example 2.16: Find the Fourier transform of sgn(t).
Solution: Recalling the transform relation [see Eq. (2.50)]
$'H;s} = iJ~ sgn(s)

it therefore follows from inverse transform relations that
1 Jco e-m sgn(s) ds

-= = J21
-:-
Y27T -co 7T l(
Interchanging the roles of t and s, and taking the complex conjugate
of the result, leads to
J Jco
-=
Y27T -co
eist sgn(t) dt = -J~~
7T IS
and thus we deduce that
3'{sgn(t);s} =
J 2
-!..
7TS
Example 2.17: Find the Fourier transform ofthe Heaviside unit function
h(t).
Solution: We first note that

1
h(t) = 2[1 + sgn(t)]
Hence,
* We could also obtain (2.103) directly by writing cos CJJof =(e'..,+ e-'""'')/2 and using
the formal relation (2.97).
2.8/Transforms Involving Generalized Functions 89
~{h(t);s} = ~ [ ~{1;s} + ~{sgn(t);s}]

= J~ [ :s]
5(s) +
where we are using (2.97)* and the result of Exam. 2.16.

Observe that, in Exam. 2.17, a superficial application of Eq. (2.57)
to (t) = h'(t) would have resulted in
~{5(t);s} = -is~{h(t);s}
leading to the incorrect transform relation

1
~{h(t);s} = - -.-
zsyl2;
That is, if sF(s) sG(s), it does not follow that F(s) = G(s), but
rather that
F(s) = G(s) + k8(s)
where k is a constant. This is a consequence of the property s5(s) =
O (recall Prob. 4a in Exer. 1.5).
EXERCISES 2.8
l. Show formally that
1 .
(a) ~{5(t - a);s} = yl2; e.a
(b) ~{eiat;s} = yl2; 5(s + a)
2. Find the Fourier transform of sin w0 t.
3. Show that for any real function of time, the amplitude spectrum
jF(w)j is necessarily an even function of w. Use this result to explain
why the Fourier transform of cos w0t leads to two impulse functions.
4. Using the property
~{J<n>(t);s} = (- istF(s), n = 1,2,3, ...
show formally that
= -is.
(a) ~{'(t);s}
(b) ~{(n)(t);s} = (-ist, n = 1,2,3, ...
*Observe that Bl{l;s} = Bl- 1{l;s} dueto the even property of f(t) = l.
S. Using the property

~{t"f(t);s} = (- itF(nJ(s), n = 1,2,3, ...
show formally that
(a) ~{t;s} = - iy27r S'(s)
(b) ~{t";s} = (- it y'2; s(n)(s), n = 1,2,3, ...
6. Based on the Fourier transform relation
~{ ~ ;s} = iJ~ sgn(s)

show that by writing
.!. = (- on-1
t"
dn-i
(n -l)!dtn-i
(!)
t
it can be formally deduced that

1 } . f" (is)n-i
= 1,2,3, ...
~{ t;s = 'y2 (n _ 1)! sgn(s), n
7. If
J~oc f(t) dt = F(O) f O

where F(s) is the Fourier transform of f(t), show that
~{focf(x) dx;s} = ~ F(s) + TrF(O)S(s)
Hint: Write f.,J(x)dx = J~oc h(t - x)f(x) dx and use the convolution
theorem.
8. Starting with the identity t8(t) = O,
(a) show formally that
'(t) = - ! S(t)
t
(b) Use the result in (a) to deduce that
s(n)(t) = (-1r (n!/t") (t), n = 1,2,3, ...
9. lf m and n are positive integers such that m < n, use Prob. 8 to
formally deduce that
tn-m s(n)(t) = (-on-m(n!/m!) a(m)(t)
2.9/Hilbert Transforms 91
10. Based on the result of Prob. 9, formally show that

(a) e-xt s<n)(t) = (n)k tn-k ()(k)(t)
(n) (
k=O
(b) f(t)S(n)(t) = -l)k f(n-k)(O)S(k)(t)

k k=O
ln Probs. 11-15, verify the given Fourier transform relation.
oz:{t nsgn( t),s} --

11. ;'lP J~ ( n!
. )n+l, n = O,I ,2,...
7T -ls
12 .9i'{tnh(t);s} = J~ ;n [s<n>(s) + (~!:n~!J. n == 0,1,2, ...
13 .9i'{C"h(t);s} = r(~a)isi"- 1 i"(l-a)sgn(s>, a > 1 (a nonintegral)

27T
14 .9i'{ltl-";s} = J~ f(l - a)isi"- 1 cos[!7T(l - a)], a > 1 (a non-
integral)
15. .9i'{ltl-"sgn(t);s} a > 1
(a nonintegral)
2.9 Hilbert Transforms

ln applications involving systems analysis to electrical networks, one
often finds the need to determine the frequency response function (i.e.,
the Fourier transform of a waveform) when only its real or imaginary
part is known. Basically the mathematical problem is to determine the
complex function
F(w) = R(w) + iX(w) (2.104)
from only knowledge of the real functions R(w) or X(w). If the function
f(t), which is the inverse Fourier transform of F(w), is a causal function
[i.e., f(t) = O for t < 0], then this mathematical problem can be solved
by means of Hilbert transforms, which fundamentally provide integral
relations between the real functions R(w) and X(w).
A related problem involving Hilbert transforms is to find a complex
representation of a real signal. For instance, if x(t) is a real signal and
x(t) is its Hilbert transform, then the analytic signal
u(t) = x(t) + ix(t) (2.105)
has a Fourier transform that leads to a one-sided frequency spectrum,

i.e., the spectrum vanishes for negative frequencies. Analytic signal rep-
resentations are particularly useful in the analysis of narrowband
waveforms.
For consistency of notation with that normally used in the discussion
of Hilbert transforms, let us redefine the Fourier transform of a given
function f(t) by
Bf{/(t);w} = f~"" e-iwt f(t) dt = F(w) (2.106)
The corresponding inversion formula then takes the form
g;- 1{F(w);t} = -1
27T
J= ewt. F(w) dw = f(t)
-=
(2.I07)
lf f(t) is a real function, it follows that
R(w) = f~""f(t)cos wt dt (2.I08)
and
X(w) =- J~= f(t)sin wt dt (2.I09)
where R(w) and X(w) are the real and imaginary parts of F(w) as given
in Eq. (2.104). We can immediately deduce from this that R(w) is an
even function and that X(w) is an odd function.
ln the special case when f(t) is an even function, we see that
R(w) = 2 L= f(t)cos wt dt (2.110)
and X(w) = O, and through the inverse cosine transform relation, it can
be shown that
f(t) = -I
7T
i""o R(w)cos wt dw (2.111)
Similarly, whenf(t) is an odd function we have R(w) = O and

X(w) =- 2 L= f(t)sin wt dt (2.112)
Thus,
f(t) = - -I
7T
i=
o
X(w)sin wt dw (2.Il3)
2.9/Hilbert Transjorms 93
We see, therefore, that iff(t) is an even (odd) function, it can be recovered

entirely from the real (imaginary) part of its Fourier transform F(w).
Regardless of whether a function is even or odd, it can always be
expressed as the sum of an even and an odd function. That is, by writing
f(t) = Uf(t) + !<- t)J + Uf(t) - J<- t)J
we have
f(t) = fe(t) + fo(t) (2.114)
where
fe(t) = ![f(t) + f(- t)J (2.115)
is an even function and
fo(t) = ![f(t) - f(- t)] (2.116)
is an odd function. From our discussion above, it now follows that R(w)
is the Fourier transform of fe(t) while X(w) is the Fourier transform of
fo(t). Hence, we have the transform pairs
R(w) = 2 L"" fe(t)cos wt dt (2.117)
fe(t) = -I
7T
1""o R(w)cos wt dw (2.118)
and
X(w) = -2 L"" / (t)sin wt dt

0 (2.119)
fo(t) = -7T1 1""

o
X(w)sin wt dw (2.120)
ln the case when f(t) is causal, it happens that f(- t) == Ofor t > O, and
therefore we deduce from (2.115) and (2.116) that
f(t) = 2/e(t) == 2/0 (1), t>O (2.121)
Under this condition, Eqs. (2.118) and (2.120) can be written as
f(t) = -7T21""o R(w)cos wt dw (2.122)
and
f(t) = -- 21""o X(w)sin wt dw

7T
(2.123)
Hence, a causal function f(t) can always be recovered from knowledge

of either R(w) or X(w), which suggests a possible relation between the
functions R(w) and X(w).
To obtain the relation between R(w) and X(w), we first note thatfe(t)
and fo(t) are related by
fe(t) = fo(t)sgn(t) (2.124)
/ 0 (1) = J.,(t)sgn(t) (2.125)
Hence, using (2.124) and properties of even and odd functions, we can
rewrite (2.117) as
R(w) = J~ e-;,, fe(t) dt
= J~oo e-;,, / 0
(t)sgn(t) dt (2.126)
Treating this last expression as the Fourier transform of a product, we

can use the convolution integral to evaluate it. Recalling that
.rf{fo(t);w} = iX(w) (2.127)
and
.cf{sgn(t);w} = 2/iw (2.128)
we find that (2.126) is equivalent to
1 2
R(w) = - 1T Joo iX(y) [.(
Y
)] dy
2 -oo l (IJ -
or
1 Joo X(y)
R(w) =- - dy (2.129)
1T -oo w-y
Similarly, since
iX(w) = J~oo e-;,, fo(t) dt
= J~oo e-;,, fe(t)sgn(t) dt
we likewise deduce that
X(w) = - .!_ Joo R(y) dy (2.130)

1T -oo (IJ - Y
Equations (2.129) and (2.130) form what we call a Hilbert transform pair.
They are also called Kramers-Kronig relations in electromagnetic theory
2.9/Hilbert Transforms 95
and are basically expressions of the causality condition, namely, the

effect must not precede in time the cause producing it.
What we have shown by (2.129) and (2.130) is that the real and
imaginary parts of the Fourier transform of a causal function satisfy
Hilbert transform relations. Conversely, it can also be shown (although
we do not give the proof here) that if the real and imaginary parts of
the Fourier transform F(w) satisfy Hilbert transform relations (2.129) and
(2.130), the inverse Fourier transform f(t) is a causal function.
Finally, there is an alternate form of the Hilbert transform pair that
is sometimes more convenient to use. We obtain this alternate form by
first recognizing that, since R(w) is even and X(w) is odd,
R( -w) = - .!_ J"" X(y) dy = R(w) (2.131)

7T -co (J) +Y
Summing (2.129) and (2.131) leads to
2R(w) = -1 foo X(y) (-1- - -1-) dy

7T -oo (J) - Y W + Y
R(w) =~roo y;(y) dy (2.132)

7T Jo w - y2
ln the same manner, we can show that
X(w) =- 2w roo R(y) dy (2.133)

7T Jo w2 - /
EXERCISES 2.9
ln Probs. 1-4, verify that the given functions satisfy the Hilbert transform
relations (2.129) and (2.130).
a
1. R(w) =- 2 ,
w +a2
2_ R(w) =1- cos w ~ w sin w, X(w) = sin w - w cos w

-w -w2
4 R( w)
= acosw- asinw X(w) = - wcosw- asinw
a>O
w +a
2 2 '
w2 + a2 '
S. Prove that the Hilbert transform of a constant is zero, i.e., show

that
J"'
-a>
~-o
w- y
6. Given that R(w) = - a/(w + a2), determine the causal functionf(t).
2
7. Given that X(w) = -w/(w2 + a2), determine the causal function

f(t).
8. Show that
(a) X(w) = .!.. J"' R(w) - R(y) dy
1T w - y
l"'
->
(b) X(w) = 2w R(w)2- R~y) dy
(c) X(w) =;Jof~ R'(~lo~~~

1T
+ yl
w - y
dy
Hint: Integrate by parts in the result of (b).

9. Let F(w) be a complex function of w that is analytic in the upper-
half w-plane and tends to zero as w ~ oo. [The condition that F(w)
has no poles in the upper-half w-plane is equivalent to the causality
condition that the inverse transform of F(w) vanishes for t < 0.]
Then show that (2.129) and (2.130) follow by applying Cauchy's
integral theorem to the integral
/(wl) = i
CW-
F(w)
(Wt- l6. ) dw, e~ o+
where w1 is real, and the contour C is shown in the following figure
as the sum of C 1 along the real axis and C2 at infinity.
lm(w)
Re(w)
w1
ie
-
2.10/Additional Topics 97
10. If x(t) is the Hilbert transform of x(t), defined by
x(t) =!
7T
Joo
-oo (
x(r) dr
- T
show that x(t) and .X(t) satisfy the orthogonality relation
J~= x(t)x(t) dt = o
11. Let x(t) be a real waveform which has no constant component. Show
that x(t) and its Hilbert transform .X(t) satisfy the Parseval relation
J~oo lx<tWdt = J~= lx(tWdt
2.10 Additional Topics

ln this final section on Fourier transforms we wish to touch briefty upon
some related topicso
2.10.1 Method of Stationary Phase

ln the use of Fourier transforms or inverse transforms we often have to
deal with integrais whose evaluation by standardized methods is either
difficult or impossible. ln such cases we may have to use a numerical
or approximatiol). method to obtain the desired results. Sometimes we
may only need to know the asymptotic behavior of the integral for large
values of a parameter o ln such instances Kelvin' s method of stationary
phase can be very useful.
To illustrate the technique, let us consider the integral
/(t) = f F(s)eirG<s> ds (2.134)
whose asymptotic behavior for t .- oo is that which we seek. lt is assumed

that a and b are real constants and that F(s) and G(s) are both twice
continuously differentiable real functions of s o Kelvin argued that the
integrand in (2.134), regarded as a function of s, oscillates with increasing
rapidity as t .- oo so that the contributions to I(t) of adjacent portions
of the integrand cancel one another except in the immediate vicinity of
the end points of the interval and in the vicinity of those points at which
G(s) is stationary, i.e., points where G'(s) = O. Moreover, in the first
approximation the contribution of stationary points is more significant
than that of the end points. Hence, if s0 is a point on the interval a <
S0 < b for which G'(s0 ) = Oand G'(s 0 ) =f O, then G(s) may be approximated
by the finite Taylor series
(2.135)
Similarly, we approximate F(s) by F(s0 ) , the first term in its Taylor
expansion. Under the assumption that the major contribution to the
integral occurs in the neighborhood of S0 , these approximations lead to
(2.136)
By formally extending the limits of integration over (- oo,oo) and using

the result (see Prob. 11 in Exer. 2.10)
yl2;"r + i1r/4
e-+ ia2u2 du =--e-
J oo
(2.137)
-= lal
we deduce that
l(t)- F(s 0 ) y2?T/tiG"(s0 )1 exp[itG(s0 ) i?T/4], f -H>O (2.138)
where the plus sign goes with G"(s 0 ) > O and the minus sign with G"(s 0 )
<o.
Example 2.18: Find the asymptotic behavior for t ~ oo of
l(t) = f cos[t(s2 - s)] ds
Solution: Here we first write
I(t) = Re{ f 2
eit(s -s) ds}
from which we identify F(s) = 1 and G(s) = s2 - s. Thus, calculating

the derivatives
G'(s) = 2s - 1, G"(s) =2
we find s0 = 1/2, and from (2.138) it immediately follows that
1T 1
l(t) - - cos -(t - 1T) t ~ 00
t 4 '
For a more thorough investigation of the method of stationary phase,
the reader is advised to consult A. Erdelyi, Asymptotic Expansions, New
York: Dover, 1956.
2.9.2 Multiple Fourier Transforms

The Fourier transform pairs (2.27) and (2.28) may be extended to functions
oftwo or more variables. For example, ifj(x,y) is a piecewise continuous
function such that
f~= f~oo lf(x,y)l dx dy < 00 (2.139)
we may consider the double Fourier transform pair
F(,Tj) = -1 Joo Joo ei<tx+.,yl f(x,y) dx dy (2.140)

21T -oo -oo
and
f(x,y) = _1 Joo foo e-i(X+Tjy) F(,Tj)dd'Yj (2.141)

21T -oo -oo
We can interpret (2.140) as an iterated Fourier transform applied se-
quentially to the variable x and then to y. That is, we define
f*(,y) = "'}
V 27T
f"" x f(x,y) dx
-oo
followed by
F(,Tj) 1 J""
= y;i; -= ei.,y f*(,y) dy
The inverse transform (2.141) follows in a similar manner. The double

Fourier transform and inverse double Fourier transformare also denoted,
respectively, by the notation
(2.142)
and
(2.143)
Operational properties of the Fourier transform developed in Sec. 2.5
carry over in a natural sort of way to the double Fourier transform
defined here. For example, the analogues of the shifting properties (Theor.
2.3) are
(2.144)
and
(2.145)
The double Fourier transform of derivatives leads to the results
8i'<2>{ :~(x,y);x~, Y~'Y/} = -iF(,Tj) (2.146)
aY
8i'(2){iJxa/x,y);x~, Y~'YI } = -'Y/F(,'Y/) (2.147)
and so on, and the convolution theorem assumes the form

= 17T Joo-= Joo

-=f(x-u, y-v)g(u,v)du dv (2.148)
2
The verification of these properties is left to the reader.
Generalizations of the double Fourier transform and its properties to
n-dimensional Fourier transforms is obvious. For example, in three vari-
ables we define
with a similar expression for the inverse transform. For reasons of the-
oretical tractability, one normally does not use Fourier transforms in
dimensions greater than two. Actually, even though physical systems ali
have three-dimensions, we are often able to ignore one or two dimensions
in our analysis by using some reasonable simplifying assumption, such
as symmetry, and so on. The classical example where simplification to
one or two dimensions is not possible, however, is the problem of diffraction
of X-rays by crystals, which must be analyzed in three dimensions.
EXERCISES 2.10
ln Probs. 1-3, verify the given asymptotic behavior of each integral.
I -
eits2 ds--1 J1T
-e;.,.j4 ,
1.
lo 2 t
t~=
3. Jor= cos (ts 2

- s) ds - 1 J1T ,2 21
4. Using properties of the Fresnel integrais (Sec. 1.3.2), show that
2
(a) fcos[t(s - s)]ds =
2
; [ J + cos~c( J2~) sin~s( J2~)]
(b) As t ~ oo, show that the result on the right-hand side in (a)
approaches that given in Exam. 2.18.
S. lf G'(s 0 ) = G"(s0 ) = ... = G<m-t)(s 0 ) = O, G(m)(S 0 ) =f O, show that
Eq. (2.138) is generalized to
b itG(s) f(1/ m ) [ m.I ]1/mexp [ztG(s
. ]
l1T
J F(s)e
a ds - m F(s 0 ) tjG<m>(so)l 0) 2m
6. Given the integral representation
Jix) = -1 l7T cos(x sin s - ns) ds, n = 0,1,2, ...

7T o
show that large-order Bessel functions have the asymptotic behavior
Hint: Use Prob. 5.

ln Probs. 7-10, use the method of stationary phase to find an asymptotic
expression for the given integral as t - oo, ln some cases it may be
necessary to use the result of Prob. 5.
8. f cos(ts4)tan s ds
9. f eil(s-sin s) ds 10.
(I
Jo sin[t(s + (/ -
sinh s) ]cos s ds
1
11. Show that

) -1-
(a
y:I;r
J""
-00
e ia2u2 du =1- -
lal
J2 [l"" yv
7T o
COS V d
---vz i"" -Vv- vd ]
o
sin V
-1- J"" e ia2u2 du=-e

1 iw/4
Vf; -oo lal
12. Show that
1
~(2){f(ax,by);x-g, y-11} = iabl F(/a, "1/b)
13. Show that

~<2J{f(x-a, y-b);x-, Y-"1} = ei<a~+ln,JF({,'r/)
14. Show that

~(2){/(x,y)cos wx;x-, Y-"1} = ![F(+w,'r/) + F(-w,'r/)]
15. Verify the convolution formula
~i2/<F(,'r/)G(,'r/);-x, "1-Y} = (f* g)(x,y)
where
<f* g)(x,y) J"" I'"'

1 _.., _.., f(u,v)g(x-u, y-v) du dv
;::: -211'
3
Applications Involving
Fourier Transforms
3.1 Introduction
The use of Fourier integrais and Fourier transforms in applications is

quite extensive. Therefore, we will make no attempt to consider all the
various applications involving these integrais, but rather briefty discuss
how they are used in severa! representative areas of application.
The basic aim of the transform method is to transform the given
problem into one that is easier to solve. ln the case of an ordinary
differential equation (ODE) with constant coefficients, the transformed
problem is algebraic. The effect of applying an integral transform to a
partial differential equation (PDE) is to exclude temporarily a chosen
independent variable and to leave for solution a PDE in one less variable.
The solution of the transformed problem in either case will be a function
of the transform variable s and any remaining independent variables.
Inverting this solution produces the solution of the original problem.
The exponential Fourier transform may be applied to derivatives of
all orders that may occur in the formulation of a given problem. However,
since it incorporates no boundary conditions in transforming these de-
rivatives, it is best suited for solving DEs on infinite domains where the
boundary conditions usually only require bounded solutions. On the other
hand, the Fourier cosine and Fourier sine transforms are well suited for
solving certain problems on semiinfinite domains where the governing
DE involves only even-order derivatives. That is, because the cosine
(sine) transform of an odd-order derivative involves the sine (cosine)
102
3.2/Boundary Value Problems 103
transform of the original function, these transforms are generally not

useful in solving DEs containing odd~ordered derivatives.
3.2 Boundary Value Problems

Fourier integrais and Fourier transforms re very useful in solving boundary
value problems on infinite domains. Such problems, however, are classified
as singular since they contain no finite boundaries. * ln this case we
normally prescribe boundary conditions of the form
y(x), y'(x) finite as lxl -+ oo (3.1)
Nonetheless, the use of Fourier transforms often forces us (at least
initially) to impose the more stringent requirements
y(x) -+O, y'(x) -+ Oas lxl -+ oo (3.2)
These stronger requirements are necessary to ensure that the Fourier
transforms of y'(x) and y"(x) exist. Even so, the formal solutions that
we generate by the transform method may not satisfy (3.2). ln such cases
we normally require y(x) to at least satisfy (3.1). Because of this, we
often tak:e the point of view in practice that the transform method produces
a "tentative solution" which must be independently verified that it is
indeed a proper solution of the problem.
Example 3.1: Use the Fourier transform to solve
y" - y = - h(l - lxl), - oo < x < oo
y(x) -+O, y'(x) -+O as lxl -+ oo
Solution: Let us introduce the Fourier transforms

[1i{y(x);s} = Y(s)
[1i{y"(x);s} = - s2 Y(s)
and
[1i{h(l - lxl);s} =- 1 J esx. dx = J--

I 2 sins
-
vfi; -1 1T s
Thus, by applying the Fourier transform to each term in the equation,
* For a discussion of singular boundary value problems, see Sec. 1.6.4 in L. C.

Andrews, Elementary Partia/ Differential Equations with Boundary Value Problems, Orlando:
Academic Press, 1986.
104 Chap. 3/Applications Involving Fourier Transforms
the DE is converted into the algebraic equation
- s2 Y(s)- Y(s) = - J2 -sin

7T
- s
s
with solution
Y(s) =
v;g sin s
s(s2 + 1)
Finally, the solution y(x) that we seek is simply the inverse Fourier
transform of Y(s), which we can obtain through use ofthe convolution
theorem. That is, we have
fffi- 1 { J~ si; s ;x} = h(l - lxl)

and

y(x) = fffi- 1{Y(s);x}
=-
}Joo e-iui h(l - lx- ul)du
2 -oo
=-
1lx+l e-iui du
2 x-1
The evaluation of this last integral leads to
sinh (l)e\ -oo<x< -1

y(x) = 1 - e- 1 coshx, -1~x~1
{
sinh (l)e-x, 1 <x<oo
(see Prob. 1 in Exer. 3.2).
This one example illustrates the basic technique used in solving

boundary value problems with constant coefficients by the transform
method. That is, the transform applied term-by-term to the DE leads to
an algebraic equation in the transform function Y(s). This algebraic equation
can be readily solved for Y(s), the inversion of which yields the desired
solution y(x). lt is perhaps the final step of inverting Y(s) that is generally
the most difficult part of the process. ln many cases, we simply represent
our solution as an integral whose evaluation is left to numerical procedures.
If the domain of interest in Ex. 3.1 were modified to the semiinifinite
region O < x < oo, then either the sine or cosine transform would be
3.2/Boundary Value Problema 105
used to solve the problem. The decision as to which of these two transforms
to use will depend upon the type of boundary condition specified at the
finite boundary x = O. That is, from Eqs. (2.60) and (2.61) in Sec. 2.5,
we recall
,q;c{y"(x);s} = -s2 Yc(s) - J~ y'(O) (3.3)
g;s{y"(x);s} = 2
-s Y5 (s) + J~ sy(O) (3.4)
and so if y(O) is specified we use the sine transform, whereas the cosine
transform is called for when y'(O) is specified. Problems involving these
transforms are taken up in the exercises.
Finally, we remark that the technique illustrated in Ex. 3.1 can readily
be generalized to boundary v:alue problems of the more general form
y"+ay'-by=f(x), -oo<x<oo (3.5)
y(x) --+ O, y'(x) --+ Oas lxl --+ ao
where a and b (b > O) are constants. Proceeding as before, the Fourier
transform applied to (3.5) leads to the algebraic problem
-(i + ias + b)Y(s) = F(s)
where F(s) is the Fourier transform of f(x). Solving for Y(s), we get
Y(s) =- F(s) (3.6)

i+ ias+ b
The inversion of (3.6) by use of the convolution theorem [see Eq. (2.81)
in Sec. 2.7] gives us the solution formula
y(x) = - f~""f(u)g(x - u) du (3.7)
where we define
g(X ) -- _1_ Oj;-1

~ ~ ~
{
2
1
b .,X } (3.8)
y21T s + zas +
Recalling Exam. 2.8 in Chap. 2, we have
g(x) == v'a/+ b exp[-

4
i<ax + \la 2
+ 4bjxj)J (3.9)
and our solution process is complete.

The use of Fourier integral representations to solve boundary value
problems is very similar to that of the Fourier transform. Let us illustrate
with the following example.
106 Chap. 3/Applications lnvolving Fourier Transforms
Example 3.2: Let us consider an "infinitely long" beam resting on an

elastic foundation as shown in Fig. 3.1. Such a problem might physically
correspond to a railroad track on a road bed. If f(x) represents a
prescribed distributed load on this beam, the classical beam theory
of Euler predicts that the vertical deftections y(x) will satisfy the
differential equation*
EJy<4>+ ky = f(x), -oo < x < oo
where E and I are physical constants and k is the foundation modulus
(i.e., a stiffness factor associated with the road bed). Solve for the
deftection y(x) when f(x) = Foh(l - lxl), where F0 is a constant.
Solution: We begin by assuming the solution y has the integral

representation
y(x) = L= [A(s)cos sx + B(s)sin sx] ds

where A(s) and B(s) are unknown functions. Differentiating this
expression with respect to x yields
y<4>(x) = L= s [A(s)cos sx + B(s)sin sx] ds

4
The functionf(x) = Foh(l - lxl) has the Fourier integral representation

(see Exam. 2.1 in Chap. 2)
f(x) = 2:o L= (si: s)cos sx ds

and by substituting these integral representations directly into the
goveming DE, we find
2
L= (Els 4
+ k)[A(s)cos sx + B(s)sin sx] dx = : 0
L"" (si: s) cos sx ds
Next, we equate like coefficients of cos sx and sin sx under the
integrais, which leads to the system of algebraic equations
(Eii + k)A(s) = 2F0 sin s

1T s
(Eis 4
+ k)B(s) = O
* See the discussion in Sec. 3.4.2 on vibrating beams.

3.2/Boundary Value Problema 107
k
y
Figure 3.1 Infinite beam on elastic foundation
Solving, we deduce that B(s) = O and
A(s) = 2F0 sin s

1r s(Els4 + k)
Therefore, our solution becomes
2Fo f= sin s
y(x) = -:;;:- Jo s(Els4 + k) cos sx ds
The final evaluation of this integral by using residue theory is left to

the exercises (see Prob. 8 in Exer. 3.2).
3.2.1 Linear Integral Equations

Another type of boundary-value problem includes equations of the form
f(x) = f k(x,t)u(t) dt (3.10)
where f(x) and k(x,t) are known functions and the function u(x) is to be
determined. Such equations are called linear integral equations of the
first kind. A familiar example of this kind of equation is the integral
transform
F(s) 1 J""
= ~ ;;:;: eist f(t) dt (3.11)
y2Tr -oo
where F(s) is known and we want f(t). The characteristic feature of an

integral equation is that the unknown function appears under the integral
- hence, the name integral equation.
Linear integral equations of the first kind are essentially integral trans-
forms where k(x,t) is the kemel. They are solved by finding an inverse
transform relation. For example, Eq. (3.11) has the solution (under ap-
108 Chap. 3/Applications Involving Fourier Transfonns
propriate conditions)
f(t) 1 I""
= ~ r::>= . F(s) ds
e-rts (3.12)
V 2'7T -oo
Another variety of integral equation has the more general form
f(x) = u(x) - J: k(x,t)u(t) dt (3.13)
called a linear integral equation of the second kind. Again it is assumed

that f(x) and k(x,t) are known functions and the unknown function is
u(x). If f(x) is identically zero on the interval a :s x ::s b, the integral
equation is called homogeneous, and nonhomogeneous, otherwise.
Integral equations like (3.13) withf(x) ::=O always possess the trivial
solution u = O. Nontrivial solutions of the homogeneous equation are
possible only for certain values of the parameter , called eigenvalues.
The corresponding nontrivial solutions are then called eigenfunctions.
The theory conceming such eigenvalue problems closely parallels the
corresponding theory involving ordinary differential equations. * When
the function f(x) is not identically zero, the resulting nonhomogeneous
equation usually has a solution only if is not one of the eigenvalues.
ln some instances the upper limit of integration in (3.13) is variable,
i. e.,
f(x) = u(x) - J: k(x,t)u(t) dt (3.14)
and we refer to such an equation as a Volterra equation. If the kemel

k(x,t) in either (3.13) or (3.14) satisfies the inequality
J: J: jk(x,tW dt dx < oo
then it is called a Hilbert-Schmidt kernel and (3.13) is further known as

a Fredholm integral equation.
Our approach here will not be to discuss the general theory or solution
techniques applicable to solving either (3.13) or (3.14). Rather, we will
single out a few examples of integral equations for which the transform
method is particularly well suited.
Perhaps the simplest integral equations to tackle by means of the
Fourier transform are those of convolution type,
~
1
r::>=
foo u(t)k(x - t) dt = f(x), - =< x < = (3.15)
y2'7T -oo
* See L. C. Andrews, Elementary Partial Differential Equations with Boundary V alue

Problems, Orlando: Academic Press, 1986. For a discussion of the theory of integral
equations, see W. L. Lovitt, Linear Integral Equations, New York: Dover, 1950.
in which the functions f(x) and k(x) are prescribed for all real values of
x andare assumed to possess Fourier transforms F(s) and K(s), respectively.
Taking the Fourier transform of each side of (3.15), we have formally
U(s)K(s) = F(s)
which may be written in the form
1
U(s) = F(s)K(s) = F(s)L(s) (3.16)
Now, if the inverse Fourier transform

f(x) = ~- 1
{L(s);x} (3.17)
exists, then by applying the convolution theorem we obtain the solution
of (3.15) in the form
u(x)
1
= r,;-:
foo f(t)f(x - t) dt (3.18)
y27T -oo
ln some cases it may tum out that L(s) = 1/K(s) does not have the
inverse transform given by (3 .17), but the related inverse transform
= ~- 1 {(-is:nK(s);x} = ~-
1
m(x) {M(s);x} (3.19)
does exist for some n (n = 1,2,3, ... ). Thus, by writing (3.16) in the form
U(s) = [(- is)nF(s)]M(s) (3.20)
we obtain the formal solution
u(x) 1 I""
= r,;-: J<n>(t)m(x - t) dt (3.21)
y27T -oo
Example 3.3: Solve
f(x) =
y27T
!--: J= -oo
lx - gl- 112 u() d
Solution: ln this case we apply the Fourier transform to each side

of the given integral equation to get
F(s) = lsl- 112 U(s)
and thus,
U(s) = lsl 112 F(s)
Because the function lsll/2 does not have an inverse Fourier trans-
form, we choose to rewrite U(s) in the equivalent form

U(s) = i sgn(s)isl- 112 ( -is)F(s)
By recalling the transform relations
8fr- 1{i sgn(s)isl- 112 ;x} = lxl- 112sgn(x)
and
8fr- 1{( -is)F(s);x} = f'(x)
we deduce by use of the convolution theorem that
V2rr Jr= lg- xl-

1 1
u(x) =- - IX lx - gl- 11'i'(g)dg - - - 112
f'(g)dg
V2rr - oo X
Ex. 3.3 involves an integral equation of the first kind. To illustrate

the integral transform method for integral equations of the second kind,
consider the following example.
Example 3.4: Solve the integral equation

1 foo
u(x) - - e-zlx - 11 u(t) dt = f(x), -oo<x<oo
2 -oo
Solution: Applying the Fourier transform to each side of the equation,

we find
U(s) - 22U(s) = F(s)

s +4
where U(s) and F(s) are the Fourier transforms, respectively, of u(x)
and f(x). This algebraic equation has solution
U(s) = F(s) + 2F(s)/(i + 2)
the inverse Fourier transform of which leads to
u(x) 1 Joo
= f(x) +----= e-v21x- 11 /(t) dt
y2 -oo
EXERCISES 3.2
1. Given the integral
x+l
I=
ix-1
e-lul du
show that
(a) I = r+
L-1
I eu du, - = <X< -1
o
(b) I =
lx-1
x+l
eu du +
ix+l
O
e-u du, -1sxs1
(c) I=
l
x-1
e-u du, 1 < x < =
(d) From (a), (b), and (c), deduce the value of I.
ln Probs. 2-6, use an appropriate integral transform to solve the given

problem.
2. y" - y = e-lxl, -=<X<=,
y(x) ~ O, y'(x) ~ O as Jxl ~ =
3. y" - y = e-x, O< x < =
y(O) = O, y(x) ~ O, y'(x) ~ O as x ~ =
4. y" - y = e-x, O< x < =
y'(O) = O, y(x) ~ O, y'(x) ~ O as x ~ =
S. y" - ey = -h(l - x), k >O, o< X< =
y'(O) = O, y(x) ~ O, y'(x) ~ O as x ~ =
6. y" - y = xe-x, O< x < =
y(O) = O, y(x) ~ =
O, y'(x) ~ O as x ~
Hint: $ 5{fe-';s} = V2f; 2s(3 - s )/(i + 2

1)3
7. Using Fourier transforms, show that
l 4' + K!y = f(x), -=<x<=
has the formal solution
y(x) 1 J""
= ~ r-c j(g)g(x - g) dg
y2'7T -00
where
g(x) = v;
2 3
lvT - -
K e-Kix! 2 [cos(Kx/\1'2) + sin(KJxJ/y2)]
8. Use residue theory to show that the integral solution in Exam. 3.2
can be reduced to
_ Fo [ -b(l+x)/vT b(l + x) + -b(l-x)/vT b(l-x)]
Y(x) - 2k e sm v' e sm y'2
2
where b4 = k/El.
ln Probs. 9-14, solve the given integral equation of the first kind.
oo . oo
7T e -x
9.
J-oo
u(t)erxr dt
Sn X
= -X- 10.
l0
u(t)cos xt dt =
2
12. _I_ Joo u()e -<x-g>2f4a2r dg = 1 e -x2f4a20 +r>

2av-:;i -= y'1 + t
oo 1
13. J-oo (
X-~
u()
1:)2
+a 2 d =
X
2
+
b2 , O < a < b
14. J~oo u()u(x - ) dg = e-x

2
15. Show that
u(x) = e-lxl + ioo ex-e u() d, 0<<1
has the solution
_ 2 e(I-)X
2- ' x<O
u(x) = 2
{ 2 - e-x, x2:::0
16. Given the homogeneous integral equation
1 Joo tu(t)dt
u(x) = :; -oo [1 + (x - t) 2](1 + t 2)
(a) show that
u(x) = -loo -,--

y27T sinxt
dt
o e - 1
(b) U sing the relation
1 2xoo 1
coth 7TX =- +-
7TX 7T L
n=IX
2
+n
2
show that (a) reduces to
u(x) - (7T2 coth 7TX

= y27T - 1)
2x
3.3/Heat Conduction in Solids 113
3.3 Heat Conduction in Solids

It is well known that if the temperature u in a solid body is not constant,
heat energy ftows in the direction of the gradient -V u with magnitude
k!Vu!. The quantity k is called the thermal conductivity of the material
and the above principie is called Fourier's law of heat conduction. This
law combined with the law of conservation of thermal energy, which
states that
"... the rate of heat entering a region plus that which is generated inside
the region equals the rate of heat leaving the region plus that which is
stored ... ,''
leads to the PDE*

V2u = a- 2u, - q(x,y,z,t) (3.22)
2
where a is another physical constant called the diffusivity. Eq. (3.22) is
commonly called the heat equation or diffusion equation. ln addition to
being the goveming equation in determining the temperature distribution
in homogeneous solids, it also occurs in problems involving electromagnetic
theory, diffusion processes, and the propagation of current in transmission
tines. Wben present, the term q is proportional to a heat source distributed
throughout the solid body.
The quantity V2 u in (3.22) is called the Laplacian and is a measure
of the difference between the value of u at a point and the average value
of u in a small neighborhood of the point. ln rectangular coordinates,
the Laplacian takes the form
(3.23)
Remark: We are adopting the standard notation of writing partial

derivatives as subscripted variables. Thus, we will write ux for aujax,
uxx for il 2u/ilx2 , uxy for a2ujayax, and so on.
The fundamental problem in the mathematical theory of heat conduction

is to solve Eq. (3.22) for the temperature u in a homogeneous solid when
the distribution of temperature throughout the solid is known at time t =
O and a certain boundary condition is prescribed at each exposed point
of the solid. There are three distinct kinds of boundary conditions that
might ordinarily be prescribed for such problems. The first is to specify
* For details, see J. Fourier, The Ana/ytical Theory of Heat, New York: Dover, 1955.
the temperature u along each finite surface of the solid (Dirichlet condition);
the second condition is to prescribe the flux of heat across the surface,
which is accomplished by specifying the normal derivative of u at the
surface (Neumann condition); and the third boundary condition is to
prescribe the rate at which heat is lost from the solid due to surface
radiation into the surrounding medium (Newton's law of cooling). This
last boundary condition is sometimes called Robin's condition.
When the unknown function u depends only upon the spatial variable
x and on time t, then (3.22) becomes
Uxx = a- 2U 1 - q(x,t) (3.24)
and when no heat source is present, (3.24) reduces further to
Uxx =a -2U 1 (3.25)
Both (3.24) and (3.25) are called one-dimensional heat equations. Among
other areas of application, these one-dimensional heat equations govern
the temperature distribution in a long rod or wire whose lateral surface
is impervious to heat (i. e., insulated). For modeling purposes we assume
in this case that the rod coincides with a portion of the x axis, is made
of homogeneous material, and has uniform cross section. Further, we
will assume that the temperature u(x,t) is the sarne at any pont in a
particular cross section of the rod, but may change from cross section
to cross section.
3.3.1 Heat Equation on an lnfinite Line

Let us first consider the flow of heat in the infinite medi um - oo < x <
ao when the initial temperature distribution f(x) is known and the region
is free of any heat sources. Physically, this problem might represent the
linear flow of heat in a very long slender rod whose lateral surface is
insulated. ln such cases, the solution will represent the temperature
distribution in the middle portion of the infinitely long rod, prior to the
time when such temperatures are greatly influenced by the actual boundary
conditions of the rod. The problem is mathematically characterized by
-oo<x<oo,t>O
B.C.: u(x, t) ~O, ux(x,t) ~ Oas lxl ~ oo (3.26)
I. C.: u(x,O) = f(x), - oo <x< oo
Remark: Since both initial conditions and boundary conditions are

prescribed in (3.26), we have designated the boundary conditions by
"B.C." and the initial conditon by "I. C." to clearly identify each type
of auxiliary condition. Also, the limiting boundary conditions are based
primarily upon physical considerations of heat flow, but in some cases

are prescribed merely to ensure that the relevant integral transforms
exist.
The infinite range on the spatial variable x suggests use of the exponential
Fourier transform rather than either the cosine or sine transform. However,
because the function we wish to transform depends upon more than one
independent variable, it is useful to adopt a special kind of notation to
desigate the variable being transformed. For example, the Fourier transform
of u(x,t) with respect to x is defined by
~{u(x,t);x~s}
1 J<X>
= ~ r:c_ eixu(x,t) dx = U(s,t) (3.27)
y27T -<X>
Similarly, we have that

~{ux(x,t);x~s} =- isU(s,t) (3.28)
~{u.u(x,t);x~s} = - iU(s,t) (3.29)
and
~{U1(X,t);x~s} = . . } J= eix U 1(x,t) dx

v 27T -00
1 afoo .
== ... 1 - e"x u(x ,t) dx
v 27T iJt -<X>
or
(3.30)
Using the above results together with the relation F(s) = @i{f(x);s},
we find that the Fourier transform applied to the problem described by
(3.26) leads to the transformed problem
UI + a2s2 U =O, t >o (3.31)
I. C. U(s,O) = F(s), -oo<s<oo
We recognize (3.31) as a first-order initial-value problem whose solution
is readily found to be
(3.32)
The solution of the original problem is now found by taking the inverse
Fourier transform of (3.32). The product form of (3.32) suggests use of
the convolution integral, which yields
u(x,t)
I J<X>
= ~ r:c_ f()g(x-,t) d (3.33)
y27T -00
116 Cbap. 3/Applications Involving Fourier Transforms
where
1
g(x,t) = @i-t{e-a2\s~x} = aVZt e-x2f4a2r (3.34)
Hence, we have obtained the formal solution
u(x,t) = ~ J"" f()e-(x-t;J2f4a2r d (3.35)

2ay'TTt -=
The integral (3.35) expresses the solution u(x,t) as a sum of the effects
of the initial temperature distribution f() at various points along the
infinite rod. The heat kemel
- -1- e -(x-t;)2f4a2t
2av:;i
physically represents the temperature distribution at point x and time t

dueto a concentrated "hot spot" (heat source) at point x = and time
t = O. That this heat kemel depends only on the difference x - is a
consequence of the fact that the medium is homogeneous, i. e., the coef-
ficients in the heat equation do not depend on x.
By making the change of variable z = ( - x)/2ay't, we can express
(3.35) in the equivalent form
u(x,t) 1 J""
= y; -
-=f(x + 2azy't)e-z 2
dz (3.36)
This form of the solution is particularly useful if the initial temperature

distribution is constant over various intervals of the rod. For example,
if f(x) = T0, - oo < x < oo, then (3.36) leads to the trivial result
u(x,t) = . . To;- Joo _

e z dz
2
= T0 (3.37)
v 1T -00
by using properties of the gamma function or error function. lt is interesting,

however, that (3.37) does not satisfy the limiting boundary condition
f.l(x,t) ~ Oas lxl ---+ oo
prescribed in (3.26). Moreover, neither the function f(x) = T0 nor the
solution u(x,t) = T0 have a Fourier transform in the usual sense. ln spite
of this, the solution (3.37) is valid!*
* Recall our discussion in Sec. 3.2. Also, since solutions given by (3.35) and (3.36)
can be derived independent of the Fourier transform method, their validity may extend
beyond that of the latter.
Example 3.5: Solve the problem described by (3.26) when the initial
temperature distribution in the rod is given by
-oo<x<oo
Solution: Using the transform relation

~{ -x2J4a2 } ~ ;-2 -a2s2
:!/'e ;s = v a e
we find that the solution (3.32) of the transformed problem becomes
U(s,t) = Vae-azszo + I)
Rather than use the convolution theorem, we can idvert this solution
directly to obtain
U(X,f) = 1 e -x2f4a2(1 + I)
Vl+t
Suppose now we consider the problem of heat ftow in an infinite rod
when a heat source is present. The problem is characterized by
Uxx = a- 2u1 - q(x,t), -oo <X< oo, t>0
B.C.: u(x,t) --+O, ux(x,t) --+O as lxl --+ oo, t >O (3.38)
I.C.: u(x,O) = O, -oo < x < oo
where q(x,t) is proportional to the heat source. For simplicity, we are
assuming the initial temperature is zero. By using the Fourier transforms
$i{u(x,t);x--+s} = U(s,t)
$i{q(x,t);x--+s} = Q(s,t)
we are led to the nonhomogeneous first-order initial-value problem
t>O (3.39)
I. C.: U(s,O) =O, -=<s<=
the solution of which is given by
U(s,t) = a2 L e-azszu- T)Q(s,T) dT (3.40)
Inverting (3.40) by use of the convolution theorem, we have
u(x,t) = ~ ~ 2 foo
V 27T -oo
ii
O
g(x-, t-T)q(,T) dT d (3.41)
where g(x,t) is defined by (3.34).

118 Cbap. 3/Applications lnvolving Fourier Transforms
3.3.2 Heat Equation on a Semiinfinite Line

Let us consider the problem of finding the temperature distribution near
the end of a long rod which is insulated. ln such a case we might model
the rod as if it were extended over the intervalO < x ::s If the initial=.
temperature distribution in the rod is f(x), the problem we wish to solve
is mathematically described by
Uxx =a -2U 1 , 0 <X<<=, t >0
B.C.: ux(O, t) = O, u(x,t) ~O, ux(x,t) ~ Oas x ~ = (3.42)
I. C.: u(x,O) = f(x), O<x<=
The fact that the interval is semiinfinite, together with the prescribed
boundary condition at x = O, suggests that the Fourier cosine transform
be used in this case. Hence, if we define*
.?fc{u(x,t);x~s} = U(s,t) (3.43)
it follows from properties of the cosine transform that
Bfduxx(x,t);x~s} = - s2 U(s,t) - V2/TT ux(O,t)
=- s2 U(s,t) (3.44)
Also, by setting F(s) = .?fc{f(x);s}, the transformed problem becomes
U, + a2s2U = O, t >O (3.45)
I. C.: U(s,O) = F(s), O<s<=
The solution of (3 .45) is
U(s,t) = F(s)e-a s 2 21
(3.46)
and by applying the inverse cosine transform, we get the formal solution
u(x,t) = V2/TT L"" e-azszr F(s)cos sx ds (3.47)
A more convenient form of the solution (3.47) can be developed by

first replacing F(s) with its integral representation
F(s) = yffj; i"" f()cos s d
* To avoid confusion with partia! derivatives, we have chosen not to subscript the
transfonned functions by C to denote a cosine transform.
which yields
00
2
u(x,t) = -
7T o o 1 1"" e-azszt f() cos scos sx d ds
1
=-
7T o o 1"" 1""
e-azszt f() {cos[s(x-)] + cos[s(x+)]} d ds (3.48)
Finally, interchanging the order of integration and using the result (see
Prob. 7 in Exer. 3.3)
f" e-bxzcos ex dx = iJ~ e-cz/4b, b>O (3.49)
the above expression becomes*
u(x,t) = ..1c
2a y1Tt o
1"" { [
f() exp - (x - 2
4a t
ef]
2
+ exp[ (x + 2 ) ]} d (3.50)
4a t
As our final example of one-dimensional heat conduction problems,
let us consider the problem where one end of a very long rod is exposed
to a time-varying heat reservoir. We will assume the initial temperature
distribution is zero, and thus our problem reads
0 <X< oo, t>0
B.C.: u(O, t) = f(t), u(x,t) ~O, ux(x,t) ~ Oas x ~ oo (3.51)
I. C.: u(x,O) = O, O< x < oo
This time we will use the Fourier sine transform, which leads to the
transforrned problem
Ut + a2s2 U = yl2f; a2sf(t), t>O (3.52)
I. C.: U(s ,0) = O, O< s < oo
where U(s,t) denotes the Fourier sine transform of u(x,t). The solution
of this initial-value problem is
U(s,t) = V2/7T a2s J:f(r)e-azsZ(t-T) dT (3.53)
* Equation (3.50) could also be derived directly from the convolution integral relation
given by Eq. (2.90) in Sec. 2.7.1.
and by using the transform relation [see Prob. 7(b) in Exer. 3.3)
we obtain from (3.53)
u(x,t) -
_ _ x_
r= (
f'
f(T)
) e -x2f4a2(t-T) dT
2ay?T o t - T 312
(3 .55)
Remark: The problem described by (3.51) can also be solved by

using the Laplace transform (see Sec. 5.4.1).
Example 3.6: Solve (3.51) for the special case whenf(t) = T1 (constant).
Solution: The solution for any f(t) is that given by Eq. (3.55). By
making the change of variable
z = x/2ay't-T
we find that (3.55) becomes
which for f(t) = T1 reduces to
u(x,t) = T1 V2r?T f""

x/2a\t
e -z2 dz
Recalling the definition of the complementary error function (see

Sec. 1.3.1)
2 {""
= y; Jx e-z
2
erfc(x) dz
we can express our solution as

u(x,t) = T1 erfc(x/2ayt)
The physical interpretation of the solution suggests that for any fixed
value of x, the temperature in the rod at that point will eventually
approach T1 ifwe wait long enough (t-oo). However, at any particular
instant of time t, the temperature u(x,t)- O as x - =, in agreement
with our prescribed boundary condition. Finally, we recognize that
the temperature u(x,t) is constant along any member of the family of
parabolas in the xt plane defined by
x/2aVt = constant
3.3/Heat Conduction in Solls 121
3.3.3 Heat Flow in an lnfinite Rectangle

Let us now consider the ftow of heat in a rectangular plate that is so
large we consider it mathematically unbounded in both the x and y
directions. If the initial temperature distribution is f(x,y) and the ftat
surfaces of the plate are insulated, the problem we wish to solve is
described by
Uxx + Uyy = a -2U 1, -oo<x<oo, -oo<y<oo,t>O
I.C.: u(x,y,O) = f(x,y), -oo < x < oo, -oo < y < oo (3.56)
We also assume that u and its normal derivatives vanish near the infinite
edges of the plate.
By introducing the double Fourier transforms (see Sec. 2.9.2)
@i(2){u(x,y,t);x~e. Y~'11} = U(,'7/,t) (3.57)
@i<z>{f(x,y);x~e. Y~'11} = F(,11) (3.58)
we find that (3.56) reduces to
U, + a 2(2 + 11 2)U = O, t>O (3.59)
I.C.: U(,'71,0) = F(,'1/)
The solution of this transformed problem is
U(,'7/,t) = F(,'1/)e-aZ(z + 71z) 1
(3.60)
and, therefore, we can write
u(x,y,t) = _1 I"" J""-= F(,'1/)e-a2(z+TJzlt e-i(fx+TJYl d d'11 (3.61)

2'7T -oo
This solution can also be expressed in the equivalent form (see Prob.
23 in Exer. 3.3)
u(x,y,t) = 21'7T J""-oo J""-oof(x',y')g(x-x', y-y', t) dx' dy' (3.62)
where
(3.63)
EXERCISES 3.3
ln Probs. 1-5, solve the heat conduction problem described by (3.26)
when the initial temperature distribution f(x) is prescribed as given.
1./(x) = {~~ ~~~~~ 2.f(x) ={~o. ~~~

3. f(x) = {To,
O,
O<x<e
otherwise
4. f(x)
0
e '
x<O
= { '-x
x>O
S. f(x) = e-lxl, -oo <x< oo Hint: See Prob. 6
6. Show that
7. Show that
(a) Loo e-bx2 COS ex dx = ~ ~ e-c2/ 4b, b >O
Hint: Expand cos ex in a Maclaurin series and use properties of the

gamma function.
(b) By differentiating the result of (a) with respect to c, deduce that
roo xe-b2x2f2 sin ex dx = ~3

Jo b
J! 2
e-c2f2b2
'
b >o
8. If f(x) = T0 (constant) and q(x,t) = S(x)S(t), where S is the delta
function, show that the problem described by (3.38) leads to
1 - 2j4a2t
u(x,t) = T0 + e x
2av;;i
9. A heat source of strength q(t)h(t), where h is the Heaviside unit
function, appears at the origin of a long rod at time t = O and moves
along the positive x axis with constant speed v. The problem is
characterized by
Uxx = u, - S(x- vt)q(t)h(t), -oo<x<oo,t>O
B.C.: u(x,t) ~O, ux(x,t) ~O as lxl ~ oo
I.C.: u(x,O) = O, -oo < x < oo
where S denotes the delta function. Show that
U(X,f) = ~ rt q(T) e-(x-VT)2j4(t-T) dT

2y7TJo~
10. (Duhamel's principie) If u(x,t;T) denotes the solution of
Uxx = a- 2u, u(x,O;T) = f(x,T), -oo <x< oo

show that
v(x,t) = J: u(x,t- T;T) dT

is a solution of
2
a vxx =V 1 - f(x,t), v(x,O) = O, -ao < x < ao
(Duhamel's principie basically points out the equivalence between
solving nonhomogeneous PDEs with a homogeneous initial condition
and solving homogeneous PDEs with a nonhomogeneous initial
condition.)
11. Show that a formal solution of
U:u = a -2Un O< X< ao, t >O
B.C.: u(O,t) = O, u(x,t) ~O as x ~ao
I.C.: u(x,O) = f(x), O< x <ao
(a) is given by
u(x,t) = J~ L"" e-a s 2 21

F(s)sin sx ds
(b) Show that the solution in (a) is equivalent to
2a~ L"" f() { exp[- (x 4: exp[-

2 2
u(x,t) = 2; ) ] - (x : 2; ) ] } d
4
Hint: See Prob. 7(a).

12. Solve Prob. 11 when
f(x) = {0,T xO>c

<x<c
0,
13. Solve the problem described by (3.42) when
f(x) = {0,T O <x<c

x >c
0,
14. Solve the problem described by (3.42) when f(x) = e-x, O < x
<ao.
15. Given the boundary-value problem
O< X< ao, t >O
B.C.: u(O,t) = O, u(x,t) ~ Oas x ~ ao
I. C.: O<x<ao
124 Cbap. 3/Applications Involving Fourier Transforms
show that
u(x,t) = T0 x(l + a 2t)- 312 exp[-x2/4(1 + a 2t)]
Hint: See Prob. 7(b).

16. If the boundary condition in (3.51) is f(t) = 1 1/y't, where 1 1 is
constant, show that
17. If the boundary condition in (3 .51) is
f(t) = {11 , O< t 0
B.C.: ux(O, t) = - f(t), u(x,t) ~O as x ~ oo
I. C.: u(x,O) =O, O<x<oo
show that
U (X,f )
a lt
= v-7 r 0~ yc--
f-T
f(-r)
e
-x2j4u2(t-T) d
T
19. Show that the solution in Prob. 18 can also be expressed n the form
u(x,t) = v'x_ r~- /(t-x~/4lh~)z-~ e-:O dz

7T J.r/1a\/
20. For the special case of f(t) = K (constant),

(a) show that the solution of Prob. 18 is
u(x,t) = K [ 2a J~ e-x 214 2

a r - x erfcCavr)]
(b) What is the temperature at the end x = O?

21. Find a solution in the form of Eq. (3.41) for
0 <X< oo, t > 0
3.4/Mechanical Vibrations 125
B.C.: u(O,t) = O, u(x,t) ~O as x ~ oo
I. C.: u(x,O) = O, O<x<oo
22. Find a solution in the form of Eq. (3.41) for

0 <X< oo, t>0
B.C.: ux(O,t) = O, u(x,t) ~ Oas x ~ oo
23. Show that

(a) g;(;lt {e-a2(2+...,2)r;~x, TJ~Y} = (1/2a2t) e-<x2+yZ)f4a2t
(b) Use the result of (a) to deduce the solution formula (3.62) of the
problem described by (3.56).
3.4 Mechanical Vibrations
The wave equation

(3.64a)
where c is a constant having the dimension of velocity, describes various
wave motions in nature and mechanical systems such as sound waves
emanating from a struck bell, surface waves propagating radially outward
when a pebble is dropped into a pool, and the deftections of a membrane
set in motion. The term q is proportional to an extemal "force" acting
on the system under investigation. A properly-posed problem involving
the wave equation consists of two initial conditions and one boundary
condition at each boundary point.
The one-dimensional wave equation
(3.64b)
is the goveming equation for such rudimentary problems as the transverse
oscillations of a tightly stretched string or the longitudinal vibrations of
a beam. The transverse vibrations of an elastic beam involve a fourth-
order PDE having similar characteristics (see Sec. 2.4.2).
3.4.1 Wave Equation on an Infinite Line

As our first illustrative example, we seek the transverse deftections of
an infinite string which is given an initial deftectionf(x) and initial velocity
g(x). Assuming there are no extemal forces acting on the string, the
problem is characterized by
Uxx = C-2 u,,, -oo<x<=,t>O
B.C.: u(x,t) ~ Oas lxl ~ ao
I. C.: u(x,O) = f(x), u,(x,O) = g(x), -ao< x <ao (3.65)
The infinite extent on x once again suggests the use of the Fourier
exponential transform. Hence, by introducing
.'Y'{u(x,t);x~s} = V(s,t) (3.66)
we get the transformed problem
Uu + c 2s2V = O, t>O (3.67)
I. C.: V(s,O) = F(s), V,(s,O) = G(s)
where F(s) = .'Y'{f(x);s} and G(s) = .'Y'{g(x);s}. By standard solution
techniques, we find
G(s) .
V(s,t) = F(s) cos cst +- sm cst (3.68)
cs
the inverse transform of which leads to the integral representation
u(x,t) =~ ~ f= e-isx [F(s) cos cst + G(s)sin cst] ds (3.69)

y27T -oo CS
Although (3.69) is a formal solution of (3.65), an interesting result

emerges if we choose to write the sine and cosine appearing in (3.69) in
terms of complex exponentials by use of the Euler formulas
1 . . 1 . .
cos X = Z(erx + e-u), sin x = ;<erx - e-'x)
2
Making the appropriate substitutions, we find
u(x,t) = _1_
2\(i;r
I=
-oo
[e-i<x-ct> + e-is<x+ct>] F(s) ds
+ _1_ f= [e-is(x-ct) - e-is(x+ct)] ~(s) ds

2y'z; -oo lCS (3.70)
The first integral in (3.70) is recognized as being equal to the expression
Uf(x - ct) + f(x + ct)]
The second integral can be similarly identified if we start with
g(z) =v'1271' foo -oo

.
e-,.zG(s)ds (3.71)
and integrate from x - ct to x + ct. Hence,
r+ct g(z)dz = 1 I"" [e-is(x-ct) - e-is(x+ct)] G.(s) ds

Jx-ct y'21r -oo IS
which implies that (3.70) is equivalent to

1[ ] 1 r+ct
u(x,t) = 2 f(x - ct) + f(x + ct) + c Jx-cr g(z)dz
2
(3.72)
Equation (3.72) is the well-known d'Alembert solution of the wave

equation on the infinite line. J. d'Alembert (1717-1783) introduced his
solution of the wave equation in 1747, six years before Daniel Bernoulli
(1700-1782) provided his series solution through separation of variables.
The method used by d' Alembert was to make a transformation of variables
to put the equation in what we now call canonical form. From this special
form he was able to solve the wave equation and finally produce the
solution (3.72).
3.4.2 Transverse Vibrations of an Elastic Beam

Consider a uniform elastic beam lying along the x axis in its equilibrium
position as shown in Fig. 3.2. If the beam is initially displaced from this
equilibrium position in the vertical plane it will cause the beam to vibrate
freely in the transverse direction. From the elementary theory of small
beam deftections it is known that the bending moment M at any cross
section of the beam satisfies the relation*
M = EI/R = Eluxx (3.73)
where E is the modulus of elasticity of the beam, I is the moment of
inertia of the cross section of the beam with respect to the neutral axis
passing through the centroid of each cross section, R is the radius of
curvature of the bent beam, and u is the deftection in the transverse
direction. Also, the balance of moments acting on an infinitesimal element
bounded by two adjacent cross sections of the beam requires that
yA
Mxx = - -uu
g
(3.74)
where 'Y is the specific weight of the beam, A is the area of the cross
section, and g is the acceleration of gravity. The term on the right-hand
side in (3.74) representing an inertia force actually behaves like a load
intensity along the entire length of the beam. Combining (3.73) and (3.74),
* See S. Timoshenko, Strength of Materiais, 3rd ed., New York: Van Nostrand, 1955.
- --- ....
.... -
\ :it:l displacement
X
Figure 3.2 Freely vibrating beam in transverse direction
we obtain the differential equation of motion

Uxxxx + a -2 u,, = o (3.75)
2
where a = ElgjyA.
For illustrative purposes, let us consider the case involving the free
vibrations of a uniform beam of infinite length where the motion is
produced by distorting the beam initially in the shape f(x) with zero
velocity. Assuming units such that a2 = 1, the problem is formulated
by
Uxxxx + u, = O, -oo<x<oo,t>O
B.C.: u(x,t) ~O, ux(x,t) ~ O as lxl ~ oo (3.76)
I. C.: u(x,O) = f(x), u,(x,O) = O, -oo < x < oo
Using the Fourier exponential transform, we convert (3.76) into the
initial value problem
U, + s4 U =O, t>O (3.77)
I. C.: U(s,O)) = F(s), U,(s,O) =O
with solution
U(s,t) = F(s)cos ts 2 (3.78)
By taking the inverse Fourier transform of (3.78), we find the motions
of the beam are described by
u(x,t) = -1- J"" .

e-zsx F(s)cos ti ds (3.79)
yl2; -oc
An alternate form of the solution can be obtained by first noting the

inverse transform relation (see Prob. 15 in Exer. 2.4)
g;;- 1 {cos(ti);s~} = ~1 r [cos(e/4t) + sin(e/4t)] (3.80)

2yt
and then using the convolution theorem on (2. 78) to get
u(x,t)
} Joo
= ~ ;;:;-::-. f(x-) [cos(e/4t) + sin(e/4t)] d (3.81)
2 V 211'1 -oo
Finally, making the change of variable 71 = x - we can also express

this last relation in the more compact form
u(x,t) = ~ lc Joo . [(x-71)

f(TJ) sm
2
l
+ -11'Jd71 (3.82)
2 V 11'1 -oo 4( 4
EXERCISES 3.4
ln Probs. l-4, solve the vibrating string problem described by (3.65)
when the initial conditions are prescribed by the given functions.
1. f(x) = e-lxl, g(x) = O, -oo <x < oo
2. f(x) = O, g(x) = e-lxl, -oo <x < oo
3. f(x) = {Ro,o' lxl < 1 g(x) = V0 , - oo < x < oo (F0 , V0 constants)

lxl > 1
4. f(x) = {R0,0 , O<x<a () {V0 , O<x<a
otherwise g x = O, otherwise
(F0 , V0 constants)
S. Solve Prob. l above by using the solution formula (3.69).
6. Show that
O <x <ao, t >O
B.C.: u(O, t) == O, u(x,t) ~O as x ~ oo
I. C.: u(x,O) == f(x), u,(x,O) = O, O < x < oo
has the solution

u(x,t) = Ufo(X + ct) + fo(X - ct)]
where fo denotes the odd extension of f over the entire axis.
7. Show that
0 <X< oo, t>0
B.C.: ux(O,t) = O, u(x,t) ~O as x ~ oo
I.C.: u(x,O) = e-x, u,(x,O) = O, O< x < oo

has the solution

e-' cosh x, x<t
( ' t)
U X = { e- r cosh t' x>t
8. Solve
-2
Uxx =C u,,, 0 <X< oo, t > 0
B.C.: u(O,t) = f(t), u(x,t) ~O as x ~ oo
I.C.: u(x,O) = O, u,(x,O) = O, O < x < oo
9. Solve
Uxx = Uu, 0 <X< oo, t > 0
B.C.: ux{O,t) = -1, u(x,t) ~O as x ~ oo
I.C.: u(x,O) = O, u,(x,O) = O, O < x < oo
10. Solve
Uxx =U 11 , O < X < oo, t > O
B.C.: ux{O,t) = A sin wt, u(x,t) ~O as x ~ oo
I.C.: u(x,O) = O, u,(x,O) = O, O < x < oo
11. Solve the problem described by (3.76) when
(a) f(x) = {F.o,o, O< x < oo

-oo<x<O
(b) f(x) = {F.o,o' lxl < 1

lxl >I
12. If f(x) = e-x 214 in the problem described by (3.76),
(a) show that , ,
I [e--/4(1+ir> e-x-/4(1-it>]
u(x,t) = - + ---:==-
2 y'I + it y'I - it
(b) By setting I + it = Reitt>, where
R = yf1+7, tan cf> =t
show that the solution in (a) becomes
u(x,t) = R-l/2e-r2(costt>t/4Rcos(x2~;cf>- ~)
13. Given the vibrating beam problem

U.rxxx + U 11 = O, 0 <X< oo, I> 0
B.C.: u(O,t)=I, U.u(O,t)=O, u(x,t)~Oasx~oo
I.C.: u(x,O) = O, u,(x,O) = O, O< x < oo

3.5/Potentilll Theory 131
use the Fourier sine transform to show that

(a) the solution of the transformed problem is
U(s,t) = 2J~ (1 - :os tsz)
(b) the solution of the original problem is
u(x,t) = 1- c(~)- s(~)

where C(x) and S(x) are the Fresnel integrais (see Sec. 1.3.2).
Hint: Recall Prob. 20 in Exer. 2.5 and Prob. 15 in Exer. 2.4.
14. Given the vibrating beam problem
Uxxxx + Uu = 0, 0 <X < oo, t>0
B.C.: u(O,t) = f(t), uxx<O,t) =O, u(x,t) ~O as x ~ oo
I.C.: u(x,O) = O, ut<x,O) =O, O< x < oo
show that
(a) the solution of the transformed problem is
U(s,t) = J~s {t(t- -r)sinTi dT

(b) the solution of the original problem is
u(x,t) 1 100 f(t

= ~c - x 2/2v 2) [cos(!v2) + sin(!v2)] dv
V 1T x/V'ft
3.5 Potential Theory

Perhaps the single most important PDE in mathematical physics is the
equation of Laplace, or potential equation. In two and three dimensions,
respectively, we have the rectangular coordinate representations
Uxx + Uyy = 0 (3.83a)
Uxx + Uyy + Uzz = 0 (3.83b)
whereas in general we write
(3.83c)
regardless of the coordinate system or number of dimensions.
Laplace's equation arises in steady-state heat conduction problems
involving homogeneous solids. * This sarne equation is satisfied by the
* The heat equation V2u = a- 2u, reduces to Laplace's equation when u, = O, i.e.,
when u is independent of time t.
gravitational potential in free space, the electrostatic potential in a uniform

dielectric, the magnetic potential in free space, the electric potential in
the steady ftow of currents in solid conductors, and the velocity potential
of inviscid, irrotational fluids. The mathematical formulation of ali potential
problems is essentially the sarne despite the physical differences of the
applications. Because of this, ali solutions of the potential equation are
coliectively called potential functions, and the study of the many properties
associated with these functions forms that branch of mathematics known
as potential theory. *
One of the most fundamental properties of the continuous solutions
of Laplace's equation is smoothness. This property is a consequence of
the fact that the equation describes "steady states." However, not ali so-
lutions are continuous. For example, the function u(x,y) = log(x2 + l)
satisfies (3.83a), but is discontinuous at (0,0). The continuous solutions
of Laplace' s equation that also have continuous second partial derivatives
in some domain R are commonly calied harmonic functions.
A properly-posed problem involving Laplace's equation consists of
finding a harmonic function in a region R subject to a single boundary
condition. The most common boundary conditions fali mainly into two
categories, giving us two primary types of boundary-value problems. If
R denotes a region in the plane and C its boundary curve, then one type
of problem is characterized by
V2u =O in R (3.84)
u =fone
which is called a Dirichlet problem or boundary-value problem of the
first kind. ln this problem we specify the value of u at each point of the
(finite) boundary. An example of a Dirichlet problem is to find the steady-
state temperature distribution in a region R given that the temperature
is known everywhere on the boundary of R. Another problem is char-
acterized by
V2u =O in R (3.85)
iJu/iJn = f on C
which is known as a Neumann problem or boundary-value problem of
the second kind. The derivative iJu/iJn is calied the normal derivative of
u and is positive in the direction of the outward normal to the boundary
curve C. t ln steady-state temperature problems, the normal derivative
specifies the heat ftow across the boundary of R. There is also a third
* For example, see O. D. Kellog, Foundations of Potential Theory, New York: Dover,
1953.
t Recall that u/n = Vu n, where n is the outward unit normal to C.
3.5/Potential Theory 133
boundary-value problem, called Robin's problem, in which the boundary

condition is a linear combination of u and its normal derivative. We will,
however, not give separate treatment of it.
3.5.1 Potential Problems in the Half-Plane

Let us consider the problem of finding a potential function u satisfying
the conditions
Uxx + Uyy = 0, -ex> < X < =, Y > 0
B.C.: u(x,O) = f(x), - =< x < = (3.86)

{ u(x,y) ~ Oas p ~ =
where p = (x 2 + /) 112 (see Fig. 3.3).

Because we have specified u on the finite boundary, we recognize
that (3.86) is a Dirichlet problem. A physical situation leading to a problem
like this would be to find the steady-state temperature distribution in a
large rectangular plate, the ftat surfaces of which are insulated, when
the temperature is prescribed by f(x) along one edge of the plate and
tends to zero along each of the other edges.
ln deciding which integral transform to use, we note that x has an
infinite range of values while y has a semiinfinite range. On the variable
y we might use the Fourier sine transform but on x we use the Fourier
exponential transform. ln this case the Fourier exponential transform
leads to a somewhat easier problem to solve. We leave the solution by
means of the sine transform to the exercises (see Prob. 9 in Exer. 3.5).
Therefore, by introducing
~{u(x,y);x~s} = U(s,y) (3.87)
~{f(x);s} = F(s) (3.88)
we are led to the transformed problem
Uyy - s2 U = O, y >O (3.89)
B.C.: U(s,O) = F(s), U(s,y) ~O as y ~ =
Figure 3.3 Dirichlet problem for half-plane

Remark: Actually, to completely describe the limiting condition in

(3.89) we should also state u(x,y) ~O as lxl ~ ao.
The general solution of the differential equation in (3.89) is given by

V(s,y) == A(s)e-Y + B(s)eY
where A(s) and B(s) are arbitrary functions of s. A valid solution for all
values of s is readily found to be
U(s,y) == F(s)e-IIY, -ao< s <ao (3.90)
Thus, recalling that (see Exam. 2.3 in Chap. 2)
g;-l{e-IIY;s~x} == f!:_ y (3.91)

V;.x 2
+l
and applying the convolution theorem, we are led to the solution formula
Y
u(x,y) == -
Joo ( J(}1:) d+ Y >O (3.92)
1T' -oo X - Y C,
2 2,
which is the well-known Poisson integral formula for the half-plane.
Example 3.7: Solve the Dirichlet problem (3.86) when
f(x) == {To, lxl b
Solution: By substitutingf directly into the Poisson integral formula
(3.92), we obtain the solution
yT 0Jb
u(x,y) == - 1T' (t
dt
)2 + y2
-b - X
To [ tan-
== -:;;:
1 + b)
(X-Y- - tan- 1 -Y-(X - b)]
but using the trigonometric identity
tan(A _ B) == tan A - tan B

1 + tanA tanB
we can express our solution in the more convenient form
u(x,y) == -To tan

1T'
-I( 2by b )
x2 + y2
- 2
Curves in the upper half-plane for which the steady-state temperature
is constant are called isotherms. For our particular problem, these
curves are defined by the farnily of circular ares (C constant)

x 2 + l- Cy = b2
which have centers on the y-axis and endpoints on the x axis at x =
b (see Fig. 3.4).
The Neumann problem for the half-plane can be solved by a device

using the solution (3.92) of the Dirichlet problem. Let us suppose the
boundary condition in the original problem (3.86) is replaced by the
N eumann condition
B.C.: uy(x,O) = f(x), -oo <x< oo (3.93)
Physically, specifying the normal derivative on a boundary tells us the
heat ftux or fiow at this boundary. For example, an insulated boundary
has zero heat ftow.
To solve Laplace's equation subject to the boundary condition (3.93),
we define a function v(x,y) = uy(x,y) which has the properties
a
Vxx + Vyy = -(uxx + Uyy) = O
ay (3.94)
B.C.: v(x,O) = uy(x,O) = f(x)
Thus, the function v(x,y) is a solution of the Dirichlet problem described
by (3.94), and therefore assumes the form
v(x,y) = -
y J"" (
f() d
<:)2 + Y2 y>O (3.95)
7T -oo X - ~
U = T0
Figure 3.4 Family of isotherms
The solution function u(x ,y) that we are seeking can now be obtained
by a simple indefinite integration of (3.95), leading to
u(x,y) = Jv(x,y) dy
f""
l
:::: :;; -ooj(g)
J(x - yeidy + l dg
where we have interchanged the order of integration in this last step.

Finally, completing the above indefinite integration, we get
u(x,y) 1 J""
= Z1r -oc/(g) log[(x - g)2 + /] dg (3.96)
as the solution of the Neumann problem on the half-plane y > O. Of

course, because we performed an indefinite integration, an arbitrary con-
stant can be added to the solution (3.96). ln fact, it can be shown that
the solution to any Neumann problem is unique only to within an additive
constant.*
3.5.2 Potential Problems in the lnfinite Strip

Suppose we now consider the Dirichlet problem
-=<x<=,O<y<a (3.97)
B.C.: u(x,O) = f(x), u(x,a) = g(x), -ao< x < oo
Physically, this problem might correspond to finding the steady-state
temperature in an infinite slab whose faces are maintained at prescribed
temperatures (see Fig. 3.5).
Again we use the Fourier transform
@i{u(x,y);x-s} == U(s,y) (3.98)
u==f X
Figure 3.5 Infinite slab
* See L. C. Andrews, Elementary Partia/ Differential Equations with Boundary Value

Problems, Orlando: Academic Press, 1986.
which reduces (3.97) to

Uyy- s2U =O, O<y<a (3.99)
B.C.: U(s,O) = F(s), U(s,a) = G(s)
where F(s) and G(s) are Fourier transforms, respectively, of f(x) and
g(x). The solution of this boundary-value problem is easily shown to be
(see Prob. 13 in Exer. 3.5)*
U(s,y) = F(s) sinh.s(a - y) + G(s) s~nh sy (3.100)

smh sa smh sa
Recalling that Fourier transforms, inverse transforms and cosine transforms
are all the sarne for even functions, we deduce that
k(x y) = ~- 1 {sinh sy s--+x} = !J! sin(7Ty/a) (3 .10l)

' sinh sa' a 2 cosh(7Tx/a) + cos(7Ty/a)
the details ofwhich we leave to the reader. t Hence, using the convolution
theorem we can express the inverse transform of (3.100) in the form
u(x,y) = .Jz; [J~=f()k(x - , a - y)d + f~= g()k(x - , y)d]
(3.102)
3.5.3 Potential Problems in the Semiinfinite Strip

To illustrate the method of integral transforms for a semiinfinite strip,
we consider the physical problem of determining the steady-state tem-
perature distribution in a thick slab in which one of the infinite faces of
the slab is subjected to a prescribed temperature distribution f(x) and
the other two faces are insulated against the flow of heat. The mathematical
problem is described by (see Fig. 3.6)
Uxx + Uyy = O, O< x < oo, O< y < a
u(x,O) =!(x), uy(x,a) = O, O< x < oo (3.103)
B.C.: { ux(O,y)- O, O<y<a
Remark: The problem described by (3.103) is neither a Dirichlet nor

a Neumann problem since the boundary conditions are mixed.
* As a general rule, we use exponential functions as solutions of u" - k 2u = O when

the domain is infinite, but find it more convenient to use hyperbolic functions when the
domain is finite.
tA special case of the result (3.101) occurs in Prob. 24 in Exer. 2.6.
Figure 3.6 Semiinfinite slab
Because the independent variable x is defined on a semiinfinite domain

and the boundary condition ux(O,y) = Ois prescribed, we find the Fourier
cosine transform to be the appropriate tool in this case. Therefore, we
introduce
3'c{u(x,y);x~s} = U(s,y) (3.104)
3'duxix,y);x~s} =- iU(s,y) - V2/Trux{O,y)
=- s 2 U(s,y) (3.105)
and
3'df(x);s} = F(s) (3.106)
Using these results, we generate the transformed problem
Uyy - s 2U = O, O < y < a (3.107)
B.C.: U(s,O) = F(s), Uy(s,a) = O
with solution
U(s, ) = F(s) cosh(a - y)s (3.108)

y cosh as
lf we let (recall Exam. 13 in Sec. 2.6)
_ ar.- I{cosh(a - y)s. }

g ( x,y ) -ore h ,s~x
cos as
sin(TTy/2a) cosh(Trx/2a)
= cosh(Trx/a) - cos(Try/a) (3.109)
then through use of the convolution theorem, our solution becomes
u(x,y) 1
= ;;c
y21T o
1= f()[g(x + , y) + g(lx- l, y)] d (3.ll0)
EXERCISES 3.5
ln Prob. 1-4, use the prescribed function f to find a solution of the
Dirichlet problem (3.86).
1. f(x) = T0 , -= < x <= 2. f(x) = { ~: ~ ~ ~
3. f(x) = g: ~ ~ ~ 4 f(x) = { T0 , O < x ": 1
O, otherwtse
S. By assuming that (3.86) has a solution of the form
u(x,y) = L>O Y(s,y) [A(s)cos sx + B(s)sin sx] ds
(a) show that direct substitution of this expression into Laplace's

equation leads to the conclusion Y(s,y) = e-Y.
(b) By imposing the boundary condition in (3.86) upon the solution
given in (a), deduce that
A(s) = ;1 Joo-oo f(x)cos sx dx, B(s) = ;1 Joo-oo f(x)sin sx dx

ln Probs. 6 and 7, use the result of Prob. 5 and the prescribed function
f to find an integral representation of the solution of (3.86).
6. f(x) = {~.- xz, ~~~ ~! 7. f(x) = {~~X, ~~~

8. Use the Fourier sine transform to show that
Uxx + Uyy = O, O<x < =,y >O
u(O,y) =O, u(x,O) = f(x),
B.C.: { u(x,y)- O as (x 2 + l) - oo
(a) u(x,y)
y roo f(t) [(t -
= ; Jo
1
x)2 + y2 - (t
1
+ x)2 + y2
]
dt
(b) Show that whenf(x) = 1, the solution in (a) reduces to u(x,y) =

(2/'Tr) tan- 1(x/y).
9. Use the Fourier sine transform to solve the problem described by
(3.86).
10. Use the Fourier cosine transform to solve
Uxx + Uyy = 0, O< X< oo,y >O
140 Chap. 3/Apptications Involving Fourier Transfonns
ux(O,y) = O, u(x,O) = f(x)

B.C.: { u(x,y) """"O as (x 2 + y2) """" oo
and show that the solution has a form similar to that in Prob. 8.
11. Solve Prob. lO when
f(x) = {1,0, O<x<c
C< X< oo
12. Given the nonhomogeneous boundary-value problem

-x2
uxx + Uyy = -xe , -oo<x<oo,y>O
B.C.: u(x,O) = O, u(x,y) mite as y """" oo,
(a) show that the transformed problem via the Fourier transform has
the solution
U(s,y) = (1 - e-ls!y)~e-s2f4
2V2s
(b) Taking the inverse Fourier transform of (a), deduce that
u(x,y ) = -21Tl l""o (1 - e -sy) sin-xse -s2j4 ds

-
s
13. Verify that (3 .1 00) is the solution of the transformed problem (3 .99).
14. Find a solution similar to (3 .1 02) for the potential problem
Uxx + Uyy = 0, -oo<x<oo,y>O
B.C.: u(x,O) = f(x), uy(x,a) =O
15. Find a solution similar to (3.110) for the potential problem
Uxx + Uyy =O, o< X< oo, o< y <a
u(O,y) =O
B.C.: { u(x,O) = f(x), u(x,a) = g(x)
16. Use the double Fourier transform to solve the three-dimensional
Dirichlet problem
Uxx+ Uyy + Uzz =O, -oo<x<oo, -oo<y<oo,z>O
u(x,y,O) = f(x,y)
B.C.: { u(x,y,z) """"O as (x 2 + l + r) 112
"""" oo
and show that
u(x,y,z) = 2}
1T Joo
-= J""
-=f(-,TJ)g(x- ,y - TJ,Z) ds d71
where g(x ,y ,z) = z(x 2 + l + i)- 312

3.6/Hydrodynamics 141
3.6 Hydrodynamics
A ftuid ftow in three-dimensional space is called two-dimensional if the
velocity vector V is always parallel to a fixed plane (xy plane), and if
the velocity components parallel to this plane along with the pressure p
and ftuid density p are all constant along any normal to the plane. This
permits us to confine our attention to just a single plane which we
interpret as a cross section of the three-dimensional region under con-
sideration. Our discussion here will be limited to two-dimensional ftow
problems.
An ideal jluid is one in which the stress on an element of area is
wholly normal and independent of the orientation of the area. * ln contrast,
the stress on a small area is no longer normal to that area for a viscous
fluid in motion. lf the density p is constant, we say the ftow is incom-
pressible. Of course, the notions of an ideal ftuid or incompressible ftuid
are only idealizations that are valid when certain effects can be safely
neglected in the analysis of a real ftuid.
The velocity, pressure, and ftuid density are all interrelated through
a set of differential equations consisting of a continuity equation, equation
of motion, and an equation of state (such as the density equal to a
constant, etc.). The continuity equation is an expression ofthe conservation
of mass, which states that the ftow of mass into a region equals the ftow
of mass out of it, and assumes the form
~+V (pV) =O (3.111)
For an incompressible ftuid, this reduces to

VV =O (3.112)
which implies there are no sources nor sinks within the region of interest
(i.e., points at which fluid appears or disappears). The equation ofmotion
is a consequence of Newton's second law of motion. By equating the
rate of change of momentum of an element of ftuid to the total force
(i. e., the resultant F of the extemal forces and the net stresses acting
on the element), we obtaint
av
-+(V V)V = F- -Vp
1 1
+ vV2V +- vV(V V) (3.113)
ilt p 3
where v is the kinematic viscosity of the ftuid. lf the force F is conservative,
* An ideal ftuidis also called a perfect jluid or inviscid ftuid.

tFor details, see B. K. Shivamoggi, Theoretical Fluid Dynamics, Dordrecht: Martinus
Nijhoff, 1985.
there exists a scalar potential function A such that

F = - VA, (3.114)
and further, if the fluid is incompressible, then (3.113) reduces to
av-v X n + v(A + !vz + !!..) + vVxn =o (3.115)

at 2 p
where n = V x V is the vorticity and V is the magnitude of the velocity

vector. Upon taking the curl of Eq. (3.115), we get the two-dimensional
equation of motion
aw
- + u- + v-
aw aw = vV w
2
(3.116)
at iJx Jy
where
fi = (O,O,w), V = (u,v,O) (3.117)
and V2 = a2/Jr + J 2/Jl. Under these conditions, (3.112) now becomes
Ux + Vy = (3.118)
Equation (3.118) can be satisfied identically by introducing a scalar
function t/J, called the stream function, according to
U = - t/ly, V = t/Jx (3.119)
The stream function can be shown to be constant along any streamline,
and therefore the streamlines of a particular flow are the family of curves
t/1 = constant. Also, it follows that
(3.120)
so that in terms of the stream function, the equation of motion (3.116)
takes the form
~ vzt/1 + (Jt/1 ~ - t/1 ~) Vzt/1 = vV4t/J (3.121)

at Jy iJx iJx iJy
where V4t/J = V2V2t/J.

Finally, a flow is called steady if the velocity, pressure, and fluid
density are independent of time. If the velocity components are also
small and the viscosity large, ali terms on the left-hand side ofEq. (3.121),
which are due to inertia of the ftuid, may be neglected in a first ap-
proximation. Hence, in this case we find that (3.121) can be replaced by
the biharmonic equation
(3.122)
3.6/Hydrodyno.mics 143
which in two dimensions has the explicit form

t/Jxxxx + 2t/Jxxyy + t/lyyyy = O (3.123)
By solving (3.123) with appropriately prescribed boundary conditions,
the velocity components of the ftow can ultimately be determined through
the use of Eqs. (3.119).
3.6.1 lrrotational Flow of an Ideal Fluid

If the vorticity O is zero at every point within the region of interest, we
say the ftow is irrotational. This means that V x V = O, which in two
dimensions is described by
(3.124)
This relation combined with (3 .119) leads to Laplace' s equation
(3.125)
Clearly, solutions of Laplace's equation V t/J = O are also solutions of
2
the equation of motion given by (3.121).

The irrotational ftow of an ideal ftuid can also be described in terms
of a velocity potential function cf>. That is, the condition V x V = O
implies the existence of a potential function cf> such that V = - Vcf>, or
U = -cJ>x, V = -c/>y (3.126)
By combining (3.126) with (3.118), we find that the velocity potential cf>
is likewise a solution of Laplace' s equation
(3.127)
Thus, for irrotational ftows we have the choice of solving (3.125) for the
stream function t/J or solving (3.127) for the potential function cf>.
Example 3.8: Consider the irrotational ftow of ao ideal ftuid filling the
half-space y ~ O through the strip lxl :::: a (see Fig. 3.7). If the ftuid
is introduced normal to the region with prescribed velocity v = f(x),
find the resulting velocity components u(x,y) and v(x,y) within the
half-space.
Solution: Let us formulate and solve the problem in terms of the

velocity potential. ln this case the problem is characterized by
cf>xx + c/>yy =O, -oo<x<oo,y>O
B.C.: cf>y(x,O) = -f(x)h(a - lxl), -oo < x < oo
Figure 3.7 Fluid ftow into the upper half-plane
ln addition, it is customary to assume that the ftuid is at rest at large

distances from the plane y = O, i.e.,
B.C.: </Jx, </Jy ~O as y ~ oo
This problem is simply a Neumann problem for the half-plane,
and thus by use of (3.96) we immediately deduce that
<!J(x,y) = - 217T fa-af()log[(x-)2 + /] d

Recalling the relations (3 .126), it follows that the velocity components
are given by
u(x,y) =- <!Jx(x,y) = -7T

I fa ((X - 1:.))f()
+y d
2 2
-a X - ~
and
v(x,y) = - cf>y(x,y) = -7T

Y Ia (
J()
1:.)2 + y2 d
-a X - ~
The solution of Exam. 3.8 in terms of the stream function 1/J is left
to the exercises (see Prob. 4 in Exers. 3.6).
3.6.2 Surface Waves

We consider here the development of two-dimensional gravity waves on
a semiinfinite body of ftuid y :s Ofrom the action of an impulsive pressure
on the free surface y = O of the ftuid. We regard the ftuid as ideal and
initially at rest. lt then follows from the laws of hydrodynamics that the
motion will be irrotational throughout ali time, and hence we can describe
such motion in terms of a velocity potential q, satisfying Laplace's equation
cPxx + cPyy = O (3.128)
3.6/llydrodynalnics 145
Also, in this case the equation of motion (3.115) reduces to
v(A+ ~V2 + ~- <~>,) =o (3.129)
If F = -V A represents the force of gravity and y is the distance measured

from the free surface, then A = gy, where g is the acceleration of gravity.
Thus, the integrated form of (3.129) becomes*
p = -p(<f>, + gy + !V2) (3.130)
Let us introduce the function TJ(x,t) to denote the elevation of the
free surface relative to its equilibrium position y = O. For sufficiently
small displacements TJ, we may linearize the equation of motion (3.130)
by neglecting all terms of second order in amplitude. This assumption
permits us to omit the term !p V2 in (3.130) and to evaluate the potential
function y(x,O, t) = TJ (x,t)
1 (3.131)
which expresses the fact that a ftuid particle initially on the free surface
continues to remain on the free surface during the wave motion. The
dynamical boundary condition, corresponding to the requirement p = O
at the free surface, is
<f>,(x,O, t) - gTj(x,t) = 0 (3.132)
Combining (3.131) and (3.132), we obtain the single boundary condition
B.C.: </> 11(x,O,t) + g</>y(x,O,t) =O (3.133)
Finally, we complete the prescription of boundary conditions by invoking
the finiteness condition
B.C.: (3.134)
Because we have assumed the waves are generated by the action of
an impulsive pressure on the free surface y = O at time t = O, we find
from (3.130) that
I.C.: <f>(x,O,O) = (ljp) q(x) (3.135)
where q(x) is the impulsive pressure defined by
q(x) = J p(x,O, t) dt
* The constant of integration can be absorbed in the potential function f/1.

Finally, we assume the initial condition "' = O when t = O, which means

that [see (3.132)]
I.C.: l/>1(x,O,O) = O (3.136)
To determine the wave system produced by the above conditions,
we must solve Laplace's equation (3.128) subject to conditions (3.133)-
(3.136). lf <l>(s,y,t) and Q(s) are the Fourier transforms, respectively, of
tf>(x,y,t) and q(x), the transformed problem we need to solve is described
by
<l>yy - i<l> = O, < y < O, t > O
-ao
<1>11(s ,0, t) + g<l>y(s ,0, t) = O, t>O

B.C.: { <1>, <l>y- Oas y - -ao (3.137)
LC.: <l>(s,O,O) = (1/p)Q(s),

The solution of (3.137) is (see Prob. 5 in Exer. 3.6)
<l>(s,y,t) = (1/p)Q(s)cos(Vglslt)ell.v (3.138)
and consequently, by the inversion formula,
1
tf>(x,y,t) = - - J"" Q(s)cos(Vglslt)e 1' 1.v-isx ds (3.139)
pyfi; -oo
lf Q(s) is an even function of s, this last result becomes
lf>(x,y,t) = .. ~ [r= e.v Q(s)cos(sx - vgst) ds

py27T Jo
+ L"' e''' Q(s)cos(sx + V gst) ds] (3.140)
Generally speaking, the evaluation of integrais like (3.140) must be

accomplished by the use of numerical or approximation methods. However,
the asymptotic evaluation of(3.140) as t - ao can be achieved by Kelvin's
method of stationary phase (see Sec. 2.9.1 ). To recall, if s0 is a point
on the interval a < s0 F(s0) i7T]

F(s)e' s ds- V I
[
.. r-L exp itG(s0) - , t - = (3.141)
v 27T a t G"(so)l 4
where the plus sigo goes with G"(s 0 ) > O and the minus sigo with
G"(s 0) <O.
Now, in (3.140) there are two symmetrical groups of waves moving
in the two directions from the origin. If we consider only the right-
running waves, then we may drop the second of the two integrais. Writing
the first integral as
1 100 e YQ(s)cos(sx
5
- ygst) ds
p v'21T o
= _1_ Re{ roo eSYQ(s)eit(vKs-sx/t) ds} (3.142)

py1; Jo
we can identify G(s) = y'gs - sx/t. Then setting
G'(s) =
v'g-r - 2xv's = O (3.143)
2n(s
we see that s0 = gf/4x2 , and hence by applying the result of (3.141) to
the above integral, we deduce that (3.140) reduces to
cp(x,y,t) - 1p J7 (gt 2
~3 egt2yf4x2Q 4~ ) cos 4x - 1T)
4' (gt 2
t ~ ao (3.144)
3.6.3 Steady Slow Motion of a Viscous Fluid

Let us now consider the problem of a viscous, incompressible ftuid filling
the half-space y ;:::: O as discussed in Exam. 3.8 in Sec. 3.6.1. A viscous
ftuid cannot be irrotational so we must work directly with the stream
function 1/J in this case. If the ftow is steady and the velocity components
small, then the goveming equation is the biharmonic equation (3.124).
Assuming that all other conditions of the problem are the sarne as described
in Exam. 3.8, the present problem is characterized by
1/Jxxxx + 21/Jxzyy + 1/lyyyy = O, -ao< X< ao, Y >O (3.145)
B.C.: 1/Jy(x,O) = O, 1/Jix,O) = f(x)h(a - lxi), -ao< x <ao
Because the governing equation in (3.145) is fourth-order, it was necessary
to prescribe a boundary condition in addition to that in Exam. 3.8 to
maintain a well-posed problem. The above condition l{ly(x,O) = O merely
states that the horizontal velocity component u is zero along the x axis. *
As before, we assume the ftuid is at rest at large distances from the
plane y == O, i.e.,
B.C.: (3.146)
To solve (3.145), we will use the Fourier transform with respect to
x. Thus, we write
@'{1/J(x,y);x~s} == 'l'(s,y) (3.147)
* A moving viscous ftuid tends to adhere to the surface of an obstacle placed in its
path.
and the transformed problem takes the form of a fourth-order boundary-

value problem
'l'yyyy - 2s 2'1'yy + s4'1' = O, y >O
B.C.: 'l'y(s,O) = O, -is'l'(s,O) = F +(s) (3.148)
where
(3.149)
The characteristic polynomial of this differential equation is*

m4 - 2im2 + s4 = O
with m = s as double roots leading to the independent solutions
From physical considerations we must select only those solutions which

tend to zero as y ~ oo. Therefore, the bounded general solution of the
problem (3.148) is given by
'l'(s,y) = [A(s) + B(s)y]e-iIY (3.150)
where A(s) and B(s) are arbitrary functions. Application of the boundary
conditions in (3 .148) yields the relations
'l'y(s,O) =-+ B(s) = O
isiA(s)
is'l'(s,O) = - isA(s) = F +(s)
from which we deduce A(s) = iF +(s)/ s and B(s) = isiA(s). Hence,
'l'(s,y) = is- 1(1 + isiy)F +(s)e-I!Y (3.151)
ln order to invert (3.151), we will use the convolution theorem of
Fourier. ln this regard it is convenient to define
(3.152)
and then determine
8f- 1 {G(s,y);s~x} = 8f- 1
{is- 1 e-iIY;s~x} + iy81'- 1 {e-11Ysgn(s);s~x}
Based on Exam. 2.5 in Chap. 2, we have that

8f- 1 {is- 1 e-iiY;s~x} = 8f$ 1 {s- 1 e-Y;s~x}
= V2/7T tan- 1(x/y), y >O (3.153)
* We obtain the characteristic polynomial by assuming 'I' = emy.

and similarly,
~- 1 {ie- 1 1Ysgn(s);s~x} = ~s {e-Y;s~x}
1
(3.154)
= V2{; x/(x2 + l), y>O
Thus,
~- 1 {G(s,y);s~x} = y2/?r[tan- 1(x/y) + xyf(x 2 + l)l (3.155)
and through use of the convolution theorem we deduce the solution
1/J(x,y) =; J:af() [tan- (x;

1
g) + (x ~ ~/~ l] d (3.156)
The velocity components u(x,y) and v(x,y) can now be determined from
Eq. (3.119) (see Prob. 9 in Exers. 3.6).
EXERCISES 3.6
1. Show that the elimination of u and v between (3.119) and (3.126)
leads to the Cauchy-Riemann equations
1/Jx = -<{>y
and use these relations to calculate the stream function 1/J from the
expression for<{> given in Exam. 3.8.
2. Solve Exam. 3.8 for the special case where f(x) = V0 , constant.
3. For the special case of Exam. 3.8 where f(x) = (a 2 - x 2)- 112 ,
(a) show that
""( ) = ! J"" lo(a) ;~x-j~jy d/:.

o/ x,y 2 -oo ll e ~
(b) From (a), determine integral representations for the velocity com-
ponents and verify that u(x,O) = O when lxl < a.
4. ln terms of the stream function, the problem discussed in Exam. 3.8
is characterized by
1/Jxx + l/lyy =O, -=<X<=, Y >O
B.C.: 1/Jx{x,O) = f(x)h(a - lxi), -= < x < =.
Solve this problem for 1/J and show that it leads to the sarne expressions
for the velocity components u(x,y) and v(x,y) as derived in Exam.
3.8 from the potential function.
5. Verify that (3 .138) is the solution of the transformed problem given
by (3.137).
6. Solve the problem described by (3.128), (3.133)-(3.136) for the special

case q(x) = P0(x), where P0 is a constant, and show that
(a) cp(x,O,t) = (P0/27rp) L'"' [cos(sx - ygst) + cos(sx + ~t)] ds
(b) By making the substitutions
{ = \,lxjg (~ + gt/2x), w = yfgt2/4x
show that
dJ
cp(x,O,t) = -Po
--
Trpg dt
where
J = -zv'i!X rsin(w2 -
2
{ ) d{
(c) From (b), deduce that
cp(x,O, t) = Pou [cos(!Tru2)C(u) + sin(i?ru2)S(u)]

px
where C(u) and S(u) denote the Fresnel integrais (see Sec. 1.3.2)
and where u2 = gf /2Trx.
7. Determine the wave motion produced on the surface of a ftuid of
infinite depth y :::;; Oby an initial displacement 'TI = f(x) of the surface.
Use the method of stationary phase to evaluate this wave motion
asymptotically as t ~ =.
8. Determine the wave motion produced by an initial displacement 'TI
= f(x) on the surface of a ftuid of finite depth h. One now has for
the potential cp the following problem:
+ c{lyy = O,
cflxx - = <X < =, -h < Y < O
B.C.: {cflu(x,~t) + !!_cfly(x,O,t) = O
cpy(x, h, t) - O
I.C.: cp,(x,O,O) = f(x)
9. Use the solution (3.156) to determine the velocity components u(x,y)
and v(x,y) for the special case f(x) = V0 , constant.
10. Consider the steady, slow motion of a viscous, incompressible ftuid
filling the half-space y ~ O. ln terms of the stream function, this
boundary value problem is described by
1/Jxxxx + 21/Jxxyy + 1/lyyyy = O, -=<x<oo,y>O
B.C.: {1/ly(x,O) = g(x)h(a - lxl), 1/lx(x,O) =O
1/Jx,l/Jy ~o as y ~ <Xl
3.7/Elo.sticity in Two Dimensions 151
(a) Determine the solution for arbitrary g(x).

(b) Determine the velocity components u(x,y) and v(x,y) for the special
case when g(x) = V0 , constant.
3.7 Elasticity in Two Dimensions

Many of the great mathematicians since the time of Euler have attacked
the problem concerning the state of stress in various elastic bodies under
the action of given forces. ln this section we will briefty discuss some
two-dimensional problems for which the solution may be obtained by
Fourier transform methods.
The effect of body or surface forces on a two-dimensional body will
be to produce internal forces between various parts of the body. The
magnitudes of these internal forces are defined by the ratio of the force
to the area over which it acts, called the average stress. ln the limit as
the area shrinks to zero, we obtain the components of stress at a point
in the elastic medium. This stress is composed of two normal components,
CTxx and CTyy, and two shearing components, CTxy and CTyx for which CTxy =
CTyx * We adopt the convention that stresses are positive when a tension
is produced and negative when a compression occurs.
The DEs satisfied by the components of stress in an elastic medium
under the action of a force per unit mass having components <Fx,Fy)
may be obtained by applying Newton's second law of motion to a small
rectangular element of the medium. Writing the displacement vector as
u = (u,v), these equations of motion aret
2
iJCTXX O(Txy F au (3.157a)
--+-+p x=p-
dX ay af
CT CT o2V
~
ax + --1!
ay + pFY = Paf- (3.157b)
where p is the mass density ofthe elastic body. For equilibrium problems
the time derivatives on the right-hand sides can be set to zero. Also, in
the absence of body forces we have Fx = Fy = O, and in these cases
the equations of equilibrium take the simpler form
dCTxx + dCTxy =O (3.158a)
ax ay
(3.158b)
* ln this section the subscripted variables no longer represent partia! derivatives.

Instead, we will use the standard Leibniz notation for any partia! derivatives.
t For the derivation of these equations and a general discussion of two-dimensional
elasticity problems, see R. M. Little, Elasticity, Englewood Cliffs: Prentice-Hall, 1972.
The above equations of equilibrium together with appropriate boundary

conditions are still not sufficient for the determination of the stresses.
That is, the complete solution requires us to take into account a com-
patibility condition for the distribution of stress that arises from the
existence of a continuous displacement vector. This condition, specified
only in terms of the stresses, is given by the relation*
2
- 2 [CTxx - v(uxx + Uyy)]
2
+ --:2 [uyy - v(uxx
ou
+ CTyy)] = 2 ~ (3.159)
~ ~ k~
where v is the Poisson ratio of the material.

To solve this system of equations it is convenient to introduce a
scalar function x. called the Airy stress function, by setting
a2x
(1'
xy
= - -xy (3.160)
Using this relation, we find the equations of equilibrium (3.158) reduce

to
~
X
(u - a2x)
XX y 2
= o (3.161a)
a ( Uyy
y -
axx)
2
2 = 0 (3.161b)
Hence, it follows immediately that the equations of equilibrium (3.158)

are satisfied by
(3.162)
Lastly, the substitution of these expressions into the compatibility condition

(3.159) yields the biharmonic equation
4X + 2_il + 4X =O (3.163)
ox4 arai y4
Solving (3.163) for the Airy stress function, subject to appropriately
prescribed boundary conditions, leads to the stress components through
use of Eq. (3.162).
3.7.1 Elastic Equilibrium under Surface Forces

Let us consider the equilibrium of a solid body deformed by the application
of pressure to its bounding surfaces. We will assume the body forces
are zero throughout the body so that (3.163) is the goveming equation.
*For example, see S. P: Timoshenko and J. N. Goodier, Theory of Elasticity, New

York: McGraw-Hill, 1965.
3.7/Elasticity in Two Dimensions 153
The elastic body that we consider is bounded by a plane of infinite

extent defined by x = O. The x-axis is then normal to this plane and
taken positive in the direction into the medium. If the domain is compressed
by a surface pressure p, which varies along the surface, the problem is
mathematically described by
a4x a4x a4y
-+2--+-=0 O<x<oo, -oo<y<oo
iJx4 iJx2iJy2 al ' (3.164)
B.C.: o-xx(O,y) =- p(y), o-xy(O,y) = O, - oo < y < oo
Of course, ali stress components must satisfy the limiting condition
O"xx,O"yyO"xy- 0 as X - oo (3.165)
Formulated in terms of the Airy stress function by using (3.162), the
boundary conditions in (3.164) are equivalent to
a2x a2x
B.C.: - 2 (O,y) = - p(y), -(O,y) = O, -oo < y < oo (3.166)
iJy iJxiJy
By applying the Fourier transform
~{x(x,y);y-s} = G(x,s) (3.167)
to the governing equation in (3.164) and boundary conditions (3.166), we
obtain the transformed problem
d 4G 2d 2G 4
dx4 - 2s dx 2 + s G = O, x>O
(3.168)
2 . dG
B.C.: s G(O,s) = P(s), -ls dx(O,s) =O
where P(s) is the Fourier transform of p(y ). The bounded solutions of

this ODE are given by
G(x,s) = [A(s) + B(s)x]e-llx (3.169)
where A(s) and B(s) are unknown "constants." By application of the
boundary conditions in (3.168), we see that A(s) = P(s)/s 2 and B(s) =
P(s)/lsl; hence,
G(x,s) = -P(s) 11
(1 + jsjx)e- sx (3.170)
s2
x(x,y) = -1- J"" P(s)

- - 2 (1 + isix)e- 11 _.
ds
vz; -oo S
SX rsy (3.171)
The stress components that we desire are now calculated using (3.162),
and are given by
<Txx(x,y) = - ~ ~ J"" P(s)(l + isix)e-lslx-isy ds (3.172)

y27T -oo
<Tyy(x,y) = - ~; J"" P(s)(l - isix)e-lslx-isy ds (3.173)

V 27T -oo
<Txy(x,y) = - ~ J"" sP(s)e-llx-isy ds (3.174)

\[i; -oo
For the special cases where P(s) is either an even function or an odd
function, these equations take on a simpler form. Specifically, if P(s) is
an even function, then
<Txx<x,y) = -V2/7T J: P(s)(l + sx)e -sxcos sy ds (3.175)
<Tyy(x,y) = -V2/7T L= P(s)(l - sx)e-xcos sy ds (3.176)
<Txy(x,y) = -V2/7T X L= sP(s)e-xsin sy ds (3.177)
whereas when P(s) is an odd function, we find
<Txx(x,y) = -V2/7T L= P(s)(l + sx)e-xsin sy ds (3.178)
<Tyy(x,y) = - vzr;;. L= P(s)(l - sx)e -sxsin sy ds (3.179)
<Txy(x,y) = vzr;;. x L= sP(s)e-xcos sy ds (3.180)
By invoking the convolution theorems of Fourier, another representation

of these results is also possible (see Probs. 1 and 2 in Exer. 3.7).
EXERCISES 3.7
1. Use the convolution theorem to show that Eqs. (3.172)-(3.174) can
be expressed in the altemate form
2x3 Joo p(y _ 7])
<Txx = ---:;; -=(x2 + 112fd1]
z.xf= 1Jzp(y -7])

<Tyy = --;; -= (r
+ 11zf d1J
z.x2 [ 7]p(y - 7])
<Txy = ---:;; -= (xz + 7]2)2 d1]
3.7/Elasticy in Two Dimensions 155
2. Find a result similar to that in Prob. 1 for the special case when p(y)
is an even function.
3. Use the result of Prob. 1 to show that, when the prescribed pressure
is p(y) = polz(y), p 0 constant,
(a) the stress components ofthe problem described by (3.164) become
a- =- Po ( -1 + -rr() + -sm
1 . 2() )
n 2 2rr
- Po ( -1 + -() - -sm
1 . 2() )
2 rr 2rr
Po 2
a-xy = -cos
rr 8
where x = r cos () and y = r sin ()

(b) Find the maximum shear stress defined by
T = !(a-xx - a-yi + u;S 12

4. Repeat Prob. 3 for the case p(y) = p 0h(a - IYI), Po constant.
5. ln the problem given by (3.164) let the pressure be prescribed by
p(y) = Po (a 2 - y2)- 112 h(a - IYI), p0 constant

rr
Show that
(a) a-xx + O"yy = -\{i]; Po L Jo(as)e -sxcos sy ds
.
(b) O"yy - O"xx + 2ia-xy = j ;2 PoX Jo r~
sJ0(as)e-s<x+zyl ds
(c) At y = O, show that (a) and (b) reduce to

O"xx + O"yy = -v'2/rr Po(X2 + a2)-t/2
a-yy - U"xx + 2ia-xy = V2/rr PoXz(~ + az)-3/2

6. Find the stress components inside the semiinfinite medium y :::::: O such
that
-oo<x<oo,y>O
B.C.: a-yy(x,O) = O, O"xy(x,O) = q(x)

3.8 Probability and Statistics

Suppose that X is a random variable. The function P(x), called the dis-
tribution function, represents the probability that X < x, where x is a
real number. The distribution function has the following properties:
1. lim P(x) = O
X-+-oo
2. lim P(x) =1
3. P(x 1) s P(x2 ) when x 1 s x 2
If we think of X as a continuous variable, then there usually exists
a related function p(x) such that
P(x) = J:= p(u) du (3.181)
The function p(x) is called the probability density function (PDF) of the
random variable X. Once p(x) has been determined, various properties
of the random variable X can be calculated, such as the statistical moments
of X.
ln statistics, the moments m 1, m2 , , of the random variable X are
defined in terms of the expectation operator E. For example,
m 1 = E[X] = J~= xp(x)dx (3.182a)
(3.182b)
whereas in general,
mk = E[x*] = I~"" x*p(x)dx, k = 1,2,3, ... (3.183)
The first moment gives the average value of a random variable, and the
higher-order moments give additional information about the spread of
the distribution defining the random variable. The variance of the dis-
tribution is defined by
rl = f~"" (x - m1i p(x) dx = m2 - mi (3.184)
which follows by expanding the square and using relations (3.183).
3.8.1 Characteristic Functions

ln many situations it is convenient to introduce the notion of a characteristic
function C(t) from which the statistical moments of X also can be found.
3.8/Probability and Statistics 157
ln certain applications the characteristic function of a random variable

is easier to calculate than its PDF, and thus this function can be very
useful in such cases. We define this new function by the expectation
C(t) = E[eux] = I~= euxp(x)dx (3.185)
which we recognize as a Fourier transform relation given by

C(t) = Y2Tr @'{p(x);t} (3.186)
Hence, it follows that
C(O) = I~"" p(x) dx = 1 (3.186a)
C'(O) = i I~"" xp(x) dx = im 1 (3.186b)
C"(O) =- I~"" x p(x) dx = -

2
m2 (3.186c)
while in general, we deduce

mk = (-i)k c<k>(o), k = 1,2,3, ... (3.187)
This says that mk is the coefficient of (iti/ k! in the Maclaurin series
expansion of the characteristic function.
Finally, it also follows from properties of inverse Fourier tranforms
that if the characteristic function of a random variable is known, its
probability density function is related by*
p(x) =
2
I""
'7T -ao .
e -ltx C(t) dt (3.188)
Example 3.9: Find the characteristic function associated with the normal
or Gaussian density function
p(x) 1
= --e-<x-m)2f2rr2
-yi2;(T
where m denotes the mean (first moment) and if the variance.
Solution: From (3.186), we have
* For a more thorough discussion of random variables and characteristic functions,

see A. Papoulis, Probability, Random Variables, and Stochastic Processes, New York:
McGraw-Hill, 1965, or C. W. Helstrom, Probability and Stochastic Processes for Engineers,
New York: Macmillan, 1984.
C(t) = V27T .cf{p(x);t}

= (1/<T).cf{e-(x-m)2/2u2;t}
= eilm.cf{e-u !2;u~<Tt}
2
Recalling Exam. 2.6 in Chap. 2, we deduce

C(t) = eizm-rr2 zz;z
Our next example illustrates the way in which the characteristic function
is used to provide the desired PDF under a transformation of random
variables.
Example 3.10: Given that X and Y are independent normal random

variables with means zero and unit variances, find the PDF of the
random variable Z = XY.
Solution: To find pz(z), * we will first determine the characteristic

function for Z and then invert it according to (3.188).
Since Z = XY, the characteristic function of Z can be determined
by calculating
Cz(t) = E[eitz] = I~= I~= eitxy Px. y(x,y)dx dy
where Px,y{x,y) is the joint PDF of X and Y. Because X and Y are
assumed to be independent, it follows that the joint PDF is simply
the product of their individual PDFs. Hence, we have
Px,y(x,y) = px(x)py(y)
1
=--=e -x2/2
-1- e -y2j2
y27T y'2;
= _.!._e -(x2+y2)/2
27T
which leads to the double integral
Cz(t) = _1_
27T
I=
-oo
e-y2f2 I""
-oo
eitxy e-x2!z dx dy
Working with the innermost integral, we find
I~oo eitxy e-x2/2 dx = y27T .cf{e-x2/2;ty}

= y12; e r/2 -t
2
* When working with more than one PDF, we find it convenient to use the more
distinguishing notation Px(x) instead of just p(x).
U sing this result, the remaining integral yields
Cz(t) = _1_ J"" e-o + ,zlyz/z dy

vz:; -ao
y1 + t2
Substituting this expression for C2 (t) into (3.188), we obtain.
pz(Z) J""
= -1
27T -oo
e-itz Cz(t) dt
1 I"" e-itz
--
- 27T -oo y't+7 dt
= _!_ (""' cos zt dt
7Tj0 y't+7
where we have used the fact that C2 (t) is an even function. This last
integral is not an elementary integral nor does it lend itself to evaluation
by conventional means using basic complex variable theory. None-
theless, it is a well-known integral which leads to the final result*
1
pz(z) = - Ko(izi),
7T
where K0(x) is a modified Bessel function of the second kind and

order zero.
EXERCISES 3.8
1. A certain random variable X is known to have the characteristic
function given by C(t) = (1 - 2it) -t.
(a) Determine the first moment m 1 and variance a-2 of X.
(b) Find the PDF of X.
2. The uniform or rectangular distribution of a random variable X is
defined by
p(x) = {1/Za-, ix-mi s a-

o, ix-mi> a-
(a) Find the characteristic function of X.
(b) Show that the kth moment of X is given by
* For example, see I. S. Gradshteyn and I. M. Ryzhik, Table of Integrais, Series,

and Products, New York: Academic Press, 1980.
[k/2) - 2j
_ 1 "" cr mk-lj
mk- k. j~ (2j + l)!(k- 2j)!
where
= {~z_,l)/2,
keven
[k/2] kodd
3. The Laplace distribution of a random variable X is defined by
p(x) 1
= -e-lx-ml/u
2o-
(a) Show that the characteristic function of X is
C(t) = eimt/(1 + a-2t2)
(b) Use (3.188) to invert the characteristic function in (a) and show
that it leads to the Laplace PDF.
4. The negative exponential distribution of a random variable X is defined
by
1 -
p(x) = 2~ e x/lUZ h(x)
where h(x) is the Heaviside unit function.

(a) Find the characteristic function of X.
(b) Find the kth moment of X.
S. The Rayleigh distribution of a random variable R is defined by
(a) Find the characteristic function of R.

(b) Find the kth moment of R.
6. Given that Z = X + Y, where X and Y are independent random
variables,
(a) show that the characteristic functions of X, Y, and Z are related
by Cz(t) = Cx(t)Cy(t).
(b) If Z is a sum of N identically distributed but independent random
variables, i.e.,
show that
where Cx(t) denotes the characteristic function of any of the N

random variables X), x2, ... , XN.
7. Let the random variable Z be defined by the sum
z = X1Y1 + X2f2 + ... + XNYN
Assuming that each Xj and lj are independent normal random variables
with means zero and equal variances if-, use Prob. 6(b) to determine
Pz(Z)
(a) when N = 2.
(b) when N = 4.
(c) Compute the variance of Z for arbitrary N.
8. Objects illuminated by laser light exhibit what is called a speckle
pattem. The normalized intensity E of the speckle pattem is usually
govemed by the negative exponential PDF, PE(x) = e-xh(x). The
addition of N independent speckle pattems on an intensity basis leads
to a total intensity pattem described by
I = E) + E2 + ... + EN
Assuming each individual speckle pattem satisfies negative exponential
statistics, find the PDF plx) for the total intensity.
Hint: Use Prob. 6(b).
9. A square-law device consists of a squaring device followed by a
lowpass filter (see accompanying figure). lf the input X to the square-
law device is the sum of a deterministic signal and zero-mean Gaussian
noise, the output Y is known to have the PDF
py(y) = !e-W+Y>I2J0(AVY)h(y)
If the output Y is then integrated to produce the output Z, which is
assumed to be a sum of N independent samples of Y, what is the
PDF of Z?
Hint: Use Prob. 6(b).
X SQUARE LOWPASS y N
z
LAW - FILTER Z=LlJ
j=l
INTEGRATOR
4
The Laplace Transform
4.1 Introduction
Although the Fourier transform has been shown to be a useful tool in
a variety of applications, there are others for which it is not particularly
well suited. Generally these other applications involve initial value problems
for which the auxiliary data are prescribed at t = O. Also, many of the
functions which commonly arise in engineering and science applications
- like sinusoidal functions and polynomials - do not have Fourier
transforms in the usual sense without the introduction of generalized
functions. For these reasons, among others, it is useful to develop other
integral transforms.
There are numerous integral transforms that have been developed
over the years, many of which are highly specialized. The most versatile
of all integral transforms, including the Fourier transform, is the Laplace
transform. Laplace transforms date back to the French mathematician
Laplace who made use of the transform integral in his work on probability
theory in the 1780s. S. D. Poisson (1781-1840) also knew of the Laplace
transform integral in the 1820s and it occurred in Fourier's famous 1811
paper on heat conduction. Nonetheless, it was Oliver Heaviside* who
*Oliver Heaviside (1850-1925) was an English electrical engineer who set himself
apart from the established scientists of the day by studying alone primarily. He was a self-
made man, without academic credentials, but made significant contributions to the development
and application of electromagnetic theory. He called mathematicians "woodenheaded"
162
4.1/Introduction 163
popularized the use of the Laplace transform as a computational tool in

elementary differential equations and electrical engineering.
We can define the Laplace transform outright, but it is instructive to
formally derive it and its inversion formula directly from the Fourier
integral theorem. ln this way we will have a better perspective on the
relation between the Fourier and Laplace transforms. To begin, let us
suppose that f and its derivative f' are both piecewise continuous functions
for all t :=::: O. ln this case the Fourier transform of f(t) may not exist.
Regardless, it often turns out that we are interested in only the response
of a system due to the action of an agent f, which we generally assume
is zero for t < O. Such functions are referred to as causal functions in
the engineering literature and are usually best handled by the Laplace
transform.
While we don't require f to be absolutely integrable, let us assume
the related function
g(t) = e -cr f(t)h(t) (4.1)
does have this property, where c is a positive real constant and h(t) is
the Heaviside unit function. It follows therefore that f must satisfy the
condition
L"' e-ct jJ(t)j dt < oo
Because the function g satisfies the conditions of the Fourier integral

theorem (Theor. 2.1), we can write [see Eq. (2.26) in Sec. 2.4]
g(t) = _!_I"" e-ist I"" g(x)eisxdx ds

27T -oo -oo
or equivalently,
f(t)h(t) = ect I""

27T -=
e-ist r= f(x)e-(c-is)xdx ds
Jo (4.2)
lf we now introduce the change of variable p = c - is, then (4.2) takes

the form
f(t)h(t) = -1. ic+i=. i= eP1 e-Py(x)dx dp (4.3)

21T l c-= O
when his powerful new mathematical methods perplexed them, and he also alienated his
fellow electrical engineers who didn't know what to make ofhim. Because ofpoor relations
with his fellow scientists, much of his work did not receive the credit he deserved. Even
today his name is often not mentioned in connection with some of his most important
contributions to science, such as his pioneering work on discontinuous functions using
operational methods and his mathematical solution for the distortionless transmission line.
164 Chap. 4/The Laplace Transform
Thus, in a purely heuristic manner we have derived the pair of transform

formulas
F(p) = f' e-pr f(t) dt = .X{f(t);p} (4.4)
called the Laplace transform of f(t), and
f(t)h(t)
1
=- .
fc+i=. eP F(p) dp = .x- {F(p);t}
1 1
(4.5)
27Tl c-=
denoting the inverse Laplace transform. The path of integration in (4.5)
appears as illustrated in Fig. 4.1. It will be shown in Sec. 4.6 that this
integral converges if the real part of p exceeds some minimum value,
say c0 , such that all singularities of F(p) lie to the left of the line
Re(p) = c0 The inverse Laplace transform given in (4.5) then exists for
all c> c0
ln our subsequent discussion of Laplace transforms it is to be understood
thatf(t) is defined only for t ~ O. Hence, we will no longer write f(t)h(t)
in the inverse transform relation.
4.2 The Transforms of Some Typical Functions

The Laplace transform of many functions can be obtained through routine
formal integrations ofthe defining integral (4.4) by treating p as ifit were
Im(p)
c + joo
o Re(p)
C - joo
Figure 4.1 Integration path of inverse Laplace transform

4.2/The Transforms of Some Typical Functions 165
a real variable. For now we will proceed with such integrations and later
examine the conditions under which this formalism is valid.
Example 4.1: Find the Laplace transform of ear.
Solution: By definition,
2{eat;p} = L= e-pt eatdt

= L= e-(p-a)tdt
= e-(p-a)t t=
-(p - a) o
Thus, the integral diverges for Re(p) :::; a, while for Re(p) > a, we
get
2{ea1 ;p} = 1/(p - a), Re(p) >a
The restriction Re(p) > a that is required in Exam. 4.1 is typical in

the evaluation of Laplace transforms. That is, the transform integral (4.4)
will be meaningful only for complex values of p in some half-plane and
not for others, if indeed the integral converges at ali. Such restrictions
are important in developing the general theory of Laplace transforms
but are of little consequence in many of the applications of the transform.
That is, in practice we are generally satisfied if we know that the transform
exists for some values of p without knowing specifically for which values.
Notce that by allowing a --7 o+ in the result of Exam. 4.1, we get
the limiting case
2{1;p} = 1/ p, Re(p) >O (4.6)
Example 4.2: Find the Laplace transforms of cos at and sin at.
Solution: From definition we have
2{cos at;p} = L= e-pr cos at dt = p/(p2 + a 2), Re(p) >O
2{sin at;p} = L= e-pt sin at dt = a/(p 2

+ a 2), Re(p) >O
which follow directly from the results of Eqs. (2.44) and (2.45) in
Chap. 2. Also, we note the relations
2{cos at;p} = \/71'/2 .:1ic{e-P ;a} 1
2{sin at;p} = \/71'/2 .:1's{e-P ;a} 1

where f!fc and f!f s denote lhe Fourier cosine and Fourier sine transforms,
respectively.
Example 4.3: Find the Laplace transform of (, where x > -1.
Solution: From the defining integral,
where we have made the change ofvariable u = pt. Using properties

of the gamma function (see Sec. 1.2), we now deduce that
.!l'{tx;p} = f(x + 1)/ px+ 1, X > -1, Re(p) >O
We will consider some additional transforms in Sec. 4.4. For now,

let us examine conditions under which the Laplace transform exists.
4.2.1 Existence Theorem

Our evaluation of Laplace tra'.sforms thus far has been purely formal,
using elementary integration techniques of real variables. We have not
addressed the question as to which class of functions actually have
Laplace transforms. That is, like the Fourier transform, not ali functions
(even continuous functions) have a Laplace transform.
Suppose that f is a piecewise continuous function with the further
property that there exists a real number c0 such that
lim lf(t)le-c' = { 0 . . c> Co (4.7)

~->= no hm1t, c < c0
A function f satisfying this condition is said to be of exponential order

c0 , also written O(ec01 ). Equation (4.7) may or may not be satisfied if
c= c0
Bounded functions f(t), such as current and velocity, occurring in
the solution for time response of stable linear systems are of exponential
order zero. That is to say, the product lf(t)le -ct in such cases approaches
zero as t ~ oo for all c > O. Even some unbounded functions, like
electrical charge or mechanical displacement in systems displaying res-
onance, may have exponential order zero if they increase like (', n >
O. ln some unstable systems the response function may increase expo-
nentially like e01 Here we see that
lim e 01 e -c/ = o, c>a
/-->00
and thus deduce tbat e01 is of exponential order a. An example of a

function not of exponential order is f(t) = e1\ since

lim e12 e-cr ==
for any constant c > - =.
Remark: Functions that are identically zero for t :::: t 0 > O are said
to be of exponential order -=.
To establish that a given functionf(t) has a Laplace transform F(p),

we must show that the Laplace transform integral
F(p) = f' e-P1 f(t)dt
converges. This will be the case provided
IF(p)l :5 Loo ie-pt f(t)i dt = L"" e-ctif(t)i dt < =

where c = Re(p). Let f be piecewise continuous on t :::: O and of
O(eco1), and let c1 be a number such that c0 < c 1 < c. Because f(t) =
O(eco), it follows that for any given small positive constante, there exists
some t0 such that
if(t)ie-c' 1 <e when t >to
We now write
("" e-c/if(t)i dt = rio e-ctif(t)l dt + ("" e-c/lf(t)l dt

Jo Jo J~
where the first integral with finite limits exists because f is piecewise
continuous. Furthermore, the second integral satisfies
(""e -ct!f(t)l dt = (""e -(c-q)t!f(t)je -qt dt

Jro Jro
< e (""e -(c-q)l dt
Jro
But this last integral exists for c > c., and thus we have established
conditions under which the Laplace transform integral converges absolutely
in the half-plane Re(p) > c0 lt can also be shown that the Laplace
transform integral converges uniformly for Re(p) :::: c2 > c0 , where c2 is
any real number satisfying c0 < c2 :::; c.
ln summary, we have the following existence theorem.
Theorem 4.1 (Existence theorem). lf f is piecewise continuous on t :::: O

and is O(ec01), then f(t) has a Laplace transform F(p) in the half-plane
Re(p) > c0 Moreover, the Laplace transform integral converges both

absolutely and uniformly for Re(p) ;:::: c 2 > c0
Most functions met in practice satisfy the conditions of Theor. 4.1.

However, these conditions are sufficient rather than necessary to ensure
that a given function has a Laplace transform. For example, both r 1
and t- 112 have infinite discontinuities at t = O and are therefore not
piecewise continuous on t ;:::: O. Yet, while the integral
2{r 1;p} = L= e-pt r dt

does not exist, we have shown in Exam. 4.3 that*
2{rl/2.p}
'
=r=
Jo
e-ptrll2dt = VTT/p
Also, the Laplace transform may exist in certain instances whenfis not
of exponential order, although we will not pro vide any general discussion
of this case. t
4.2.2 Analytic Continuation

According to Theor. 4.1, every functionf(t) that is of exponential order
and belongs to the class of piecewise continuous functions has a Laplace
transform F(p). Due to properties of definite integrais, we also know
that F(p) is unique. Lastly, the transform function F(p) has the following
important property which we do not prove.:!:
Theorem 4.2. If f(t) is piecewise continuous on t ;:::: O and is O(ec'), its

Laplace transform F(p) is an analytic function of the complex variable
p in the half-plane Re(p) > c0
ln some cases the function F(p) may be analytic to the left of the
line Re(p) = c0 , although for our purposes it suffices to know that there
exists a half-plane where F(p) is indeed analytic.
Up to this point we have produced Laplace transforms
F(p) = Loo e-pt f(t) dt
* Since p is complex, one really needs also to specify the branch of the multivalued
function p' 1Z, which may be either yp or - VP in general.
t For a discussion of the convergence of the Laplace transform integral in the general
case, see W. R. LePage, Complex Variables and the Laplace Transform for Engineers,
New York: Dover, 1980.
:1: For a proofofTheor. 4.2, see R. V. Churchill, Operational Mathematics, New York:
McGraw-Hill, 1972.
by formal integration methods applicable to real integrais. ln other words,

we have treated the complex variable p as if it were a real variable x,
and then once F(x) was determined we obtained F(p) by simply replacing
x with p. It is reassuring to know that this formal procedure is actually
valid! What permits this type of formalism is the fact that F(p) is an
analytic function, and thus we can use analytic continuation in obtaining
F(p) from F(x), where xis real.
EXERCISES 4.2
ln Probs. 1-10, evaluate the Laplace transform of each function directly

from the defining integral.
1. f(t) =f 2. f(t) = e-ar - e-bt
3. f(t) = cosh at 4. f(t) = sinh at
5. f(t) = h(t - a), a>O 6. f(t) = te 21
7. f(t) = t sin kt 8. f(t) = cos 2kt
9. f(t) = sin at sin bt 10. f(t) =: ea
1
cosh kt
ln Probs. 11-20, determine which functions are of exponential order,
and for those which are, determine c0
11. f(t) = 1100 12. f(t) =r 112
13. f(t) = l 1
14. f(t) = sin 1
2
15. f(t) = (sin t)/t 16. f(t) = log 1

17. f(t) = r 312log t 18. f(l) = sin(e12)
19. f(t) = Vltan ti 20. f(t) = e' log t
ln Probs. 21-26, verify the Laplace transform relations.
21 .P{tl/2;p} = 2~ J~
22 .P{t5/2;p} = 81;3 J~
;p} = y; t!' erfc(p)
2
23 .P{e- 1214
24 .P{(t + a)- 112 ;p} = v;TP eap erfc(\fap), a >O
2 - -
25. 2{(! + a)- 312 ;p} = Va- 2y'1rp eap erfc(y'ap), a>O
26 .P{log t;p} = ~ [ f'(l) - log p]
Hint: First determine f'(z).

ln Probs. 27-30, the given functions are various types of pulses of unit
height and duration T. Determine their Laplace transforms.
= { o:
1 O<t<T
27. f(t) t>T 28. f(t) =
1, a<t<a + T, a > O
{ O, otherwise
29. f(t) = 30. f(t) =
2t/T, O< t < T/2 sin at, O<t<1r/a
2- 2t/T, T/2 < t < T { o, t > 1r/a
{
O, t> T
31. Prove that F(p) has no finite singularities (i.e., it is an entirefunction),
where F(p) is the Laplace transform associated with
(a) Prob. 27. (b) Prob. 28.
(c) Prob. 29. (d) Prob. 30.
32. Prove that if f(t) is piecewise continuous and f(t) is identically zero
for t > T, where Tis any positive number, then the transform function
F(p) is an entire function.
33. If f(t) is a polynomial, prove that its transform F(p) is a rational
function.
4.3 Basic Operational Properties

The use of the integral definition to compute Laplace transforms is a
tedious task and not required at all times. That is, once we have found
several transforms directly from the defining integral it may be possible
to find additional ones by using various properties of the integral transform
known as operational properties. Perhaps the most fundamental of these
is the linearity property, which we state first.
Theorem 4.3 (Linearity properiy). If F(p) and G(p) are the Laplace trans-
forms, respectively, of f(t) and g(t), then for any constants C, and C2o
.P{Cif(t) + C2g(t);p} = C,F(p) + C2G(p)
The proof of Theor. 4.3 is a simple consequence of the linearity
property of integrais and is left to the exercises (see Prob. 1 in Exer.
4.3).
4.3/Basic Operational Properties 171
For a > O, we find that
.T{f(at);p} = L"" e-pt f(at) dt
= -1
a o
i"" e-upfa f(u) du
the last step of which is the result of making the variable change u =
at. Thus, if F(p) denotes the Laplace transform off(t), the above relation
suggests that
1
.T{f(at);p} = - F(p/a), a>O (4.8)
a
which is called the scaling property. We illustrate this property in the
following example.
Example 4.4: Given that .T{e- 1214 ;p} = y; eP 2

erfc(p), find the Laplace
transform of e- 12
Solution: By writing e_,z = e-a2121\ we identify a = 2 in the scaling

property (4.8). Thus, it immediately follows that
2
.T{e- 1 ;p}
y;
= T eP erfc(p/2)
2j4
4.3.1 The Shifting Theorems

It sometimes happens that we need to calculate the transform of ea'l'(t),
where the transform of f(t) is known or is readily computed. Transforms
of this nature are easily handled because of the exponential function
occurring in the defining integral, and leads us to the first shift property.
Theorem 4.4 (Shifting property). If F(p) is the Laplace transform off(t),

then
.T{ea'l'(t);p} = F(p - a)
.:t{ealj(t);p} = L"" e-pt ealj(t) dt
= L"" e-<p-a)t f(t) dt

= F(p- a)
where F(p) is the Laplace transform of f(t).

Example 4.5: Find the Laplace transform of e- 21cos 3t.
Solution: Recalling Example 4.2, we have

2{cos 3t;p} = pf(p2 + 9)
and hence, through the shifting property it follows that
-21 p +2 p +2
5t{e cos 3t;p} = (p + 2)2 + 9 = p2 + 4p + 13
The Laplace transform has the effect of taking discontinuous functions

in the t-domain and "smoothing" them in the p-domain. Thus, one of
the most interesting and useful applications of the Laplace transform is
solving linear differential equations with discontinuous or impulsive forcing
functions, which are commonplace in circuit analysis problems and some
mechanical systems. The Heaviside unit function h(t - a) discussed in
Sec. 1.5 is used extensively in dealing with discontinuous and impulsive
functions. The property we need now involves the Heaviside unit function
and is called Heaviside's second-shift property, or the translation property.
Theorem 4.5 (Translation property). If F(p) is the Laplace transform of

f(t), then
5t{f(t - a)h(t - a);p} = e-apF(p), a> O
Proof: Here we see that
I:t{f(t - a)h(t - a);p} = E'" e-pt f(t - a)h(t - a) dt
= E"' e-pt f(t - a) dt
Introducing the new variable u = t - a, we get
5t{f(t - a)h(t - a)p} = ("" e-p(u+a) f(u) du

' Jo
= e-ap L"" e-pu f(u) du
= e-ap F(p)

The translation property (Theor. 4.5) can also be expressed as
5t{f(t)h(t - a);p} = e-ap 5t{f(t + a);p} (4.9)
which may be a more useful form in certain applications. Also, as a
consequence of the translation property, we have that
1
2{h(t - a);p} = -e -ap, a>O (4.10)
p
Example 4.6: Find the Laplace transform of the discontinuous function
f(t) = { ~: ~~~ < 2

Solution: We first express f(t) in terms of the Heaviside unit function,
i.e.,
f(t) = t 2 [h(t) - h(t - 2)] + 6h(t - 2)
= t2 + (6 - t 2)h(t - 2)
Then, using the linearity property and Eq. (4.9), we find
2{/(t);p} = 2{t2;p} + 2{(6 - t 2)h(t - 2);p}
= 2{t2;p} + e- 2 p 2{6 - (t + 2)2;p}
= 2{r;p} + e- 2p 2{2 - 4t - t2;p}
and thus deduce that
2{/(t);p} =2
- + e- 2 p (2- 4 --
-- 2)
p3 p p2 p3
4.3.2 Transforms of Derivatives and Integrais

The real merit of the Laplace transform is revealed by its effect on
derivatives. Here we will derive a relation between the Laplace transform
of the derivative of a function and the Laplace transform of the function
itself.
Suppose that f is a continuous function with a piecewise continuous
derivative f' on the interval t :2:: O. We further suppose that both f and
f' are of exponential order c0 Using integration by parts, we obtain
2{f'(t);p} = f" e-pt f'(t) dt
= e-pt f(t) [ + p L= e-pt f(t) dt

Becausefis O(ecof), it follows that e-P'l'(t) ~O as t ~ao, and consequently,
2{/'(t);p} = pF(p) - f(O) (4.11)
Remark: If f has a finite discontinuity at t = O, then we replace f(O)

in (4.11) withf(O+).
Similarly, if f and f' are continuous and f" is piecewise continuous

on t :::::: O, and if ali three functions are of exponential order c0 , we can
use (4 .11) to obtain
.:t{f"(t);p} = p.:l'{f'(t);p} - f'(O)
which simplifies to
.:l'{f"(t);p} = p 2F(p) - pf(O) - f'(O) (4.12)
By repeated application of (4.11) and (4.12), we arrive at the following
general result.
Theorem 4.6 (Differentiotion property). Jf f, f', ... ,J<n-n are all continuous
functions on t :::::: O, fn> is piecewise continuous on t :::::: O, and if all are
of exponential order c0 , then for n = 1,2,3, ... ,
.:l'{f(n)(t);p} = pnF(p) - pn-y(O) - pn-2f'(O) - ... - f(n-1)(0)
Although the real utility of Theor. 4.6 will not be observed until we
use it in the solution of differential equations, the following example
provides a novel way in which the differentiation property may be used.
Example 4.7: Find the Laplace transform of tn, n = 1,2,3, ... , by using
Theor. 4.6.
Solution: The functionf(t) = ~ and all its derivatives are continuous

functions of exponential order. Also, we see that
/(0) = f'(O) = ... = J<n- 0 (0) = O
f(n)(t) = n!
Substituting these results into Theor. 4.6 leads to
.:l'{f(n)(t);p} = n! .:l'{1 ;p} = pn .:t{f;p} - o - o - ... - o
and therefore
,p} -- n.lj p n+l , n -- 1,2 ,3 , ... *
(J){ln.
..L
where we are using the previous result .:l'{l ;p} = 1/p.
A different application of Theor. 4.6 involves the Laplace transform

of the integral of f.
* Notice that this result is simply a special case of that in Exam. 4.3.
Theorem 4. 7 (lntegration property). lf f is piecewise continuous on

t 2': O and is O(ecof), then
5t {J:f(u)du;p} = F(p)jp
Proof: Let us define
g(t) = J: f(u)du
which is a continuous function sincefis piecewise continuous; furthermore,
g(O) = O and g'(t) = f(t). Hence, g satisfies the conditions of Eq. (4.11),
which leads to
5t{f(t);p} = p 5t{J:f(u)du;p} -O
or
5t {J:f(u)du;p} = F(p)jp

Remark: Theorem 4. 7 is actually a special case of the convolution
iheorem presented in Sec. 4.5.
ln these last properties we have found that differentiation and integration

of a given function in the t-domain correspond roughly to multiplication
and division, respectively, in the p-domain. ln this fashion, the Laplace
transform has the effect of replacing the operations of calculus in the t-
domain with algebraic operations in the p-domain. It is primarily for this
reason that the Laplace transform is so useful in the solution of differential
and integral equations.
4.3.3 Derivatives and Integrais of the Transform

Sometimes we need to evaluate the transform of a function that is expressed
as t"f(t), where the transform of f(t) is easily obtained. To develop the
property we need in this regard, let us start with the transform relation
F(p) = L"" e-pt f(t) dt (4.13)
If f(t) is piecewise continuous on t 2': O and is O(ecof), then F(p) is an

analytic function in the half-plane Re(p) > c0 (Theor. 4.2). As a con-
sequence, F(p) has derivatives of all orders which we assume can be
formally obtained by differentiating (4.13) under the integral.* Thus, we

have
F'(p) =L'"' (-t)e-P'f(t)dt =-L"" e-P'[tf(t)]dt
~{tf(t);p} = - F'(p) (4.14)
Continued differentiation of (4.13) leads to
p<n>(p) =L"" (-tte-P'f(t) dt, n = 1,2,3, ...

and thus we have the following theorem.
Theorem 4.8. If f is piecewise continuous on t ;;::: O and is O(ec01 ), then

:l{tY(t);p} = (-It p<n>(p), n = 1,2,3, ...
Example 4.8: Evaluate the Laplace transform of te- 2' cos t.
Solution: Starting with the transform relation (see Exam. 4.2)

.P{cos t;p} = pf(p2 + I)
we use (4.14) to determine
2
d ( p ) p - 1
.P{t cos t;p} = - dp p2 + 1 = (pz + 1)2
Finally, we apply the shifting property (Theor. 4.4) to obtain

-21 (p + 2) 2 - 1 2
p + 4p + 3
.P{te cos t;p} = [(p + 2)2 + tf = (pz + 4p + 5i
By replacing p with u and integrating both sides of Eq. (4.13), we

find that
L"' F(u) du =L"" L"" e-u'f(t) dt du

= L"" (I: e-ut du) f(t) dt
= L"" e-pt [f(t)/tl dt
*Recai! from Theor. 4.1 that the integral in (4.13) is uniformly convergent in a half-
plane. Thus, differentiation under the integral sign is permitted as longas each new integral
produced in this fashion also converges uniformly.
where we have reversed the order of integration. Hence, we have derived

the following property.
Theorem 4.9. If f is piecewise continuous on t 2::: O and is O(ec01 ), and

f(t)/t has a Laplace transform, then
Y:{f(t)/t;p} = L'>O F(u) du
Example 4.9. Find the Laplace transform of the sine integral
Si(t) = {' sin u du

Jo u
Solution: Starting with the transform relation (see Exam. 4.2)

Y:{sin t;p} = 1/(p2 + 1)
and using Theor. 4.9, we obtain
y; {si~ t ;p} = f' d: u2 1 = tan -I i

Then, applying Theor. 4.7, we deduce that
fi {Si(t);p} = Y: { f' sin u du;p} = .!.. tan -I .!.. , Re(p) > 1

Jo u p p
Some additional properties of the Laplace transform involving integrais

are
1 j""
Y: { Jo(' f(u) }
-;;-du;p = p F(s) ds P (4.15)
fi {f oo
1
f(u)
---;;du;p
}
= p1 JofP F(s) ds (4.16)
and
("" f( ) } 1 f""
Y: { Jo : du;p = pJo F(s) ds (4.17)
the last of which is a simple consequence of the first two. The verification
of these properties is left to the exercises.
4.3.4 Periodic Functions

A function f is called periodic if there exists a constant T > O for which
f(t + T) = f(t) for ali t 2::: O. The smallest value of T for which the
property holds is called thefundamental period, or simply, the period.

Familiar examples of periodic functions are cos t and sin t which have
period 2n, but there are many others whose definition is not so easily
given. Periodic functions appear in a wide variety of engineering applications
where the Laplace transform may be used. Such applications rely heavily
on the following theorem conceming periodic functions.
Theorem 4.10. Let f be piecewise continuous on t =:: O and of O(ec01 ).

lf f is also periodic with period r, then
.P{f(t);p} = [l/(1 - e-pT)) r e-pt f(t) dt
r
Proof: Let us write the Laplace transform as
.P{f(t);p} = e-pt f(t) dt + J; e-pt f(t) dt
By making the change of variable t = u + T in the last integral, we

obtain
.P{f(t);p} = r e-pt f(t) dt + Loo e-p(u+T) f(u + T) du
=r e-P'f(t)dt + e-pT Loo e-puf(u)du
=r e-pt f(t) dt + e-pT .P{f(t);p}
Solving for .P{f(t);p} yields

.:l'{f(t);p} = [1/(1 - e-pT)] LT e-pr f(t) dt.

Example 4.10: Find the Laplace transform of the half-wave rectified
sinusoidal (see Fig. 4.2)
f(t) = {sin t, O<t<n
O, 17' < t < 217'
where f(t + 2n) = f(t).
f(t)
o 17' t
Figure 4.2 Graph of a half wave rectified sinusoidal
Solution: Since T = 21r, we compute

z,. i"' 1 + e-,.p
io
e-P1 f(t) dt =
o
e-p1 sin t dt = --,:---
2
p +1
and therefore
EXERCISES 4.3
1. Prove the linearity property (Theor. 4.3)
:J!{C1/(t) + Czg(t);p} = C 1F(p) + CzG(p)
2. If F(p) denotes the Laplace transform of f(t), show that
:J!{e-bt/a f(t/a);p} = aF(ap + b), a> O
3. Given that :J!{(sin t)/t;p} = tan- (1/p), find the Laplace transform
1
of (sin at)/t where a > O.

4. Use Probs. 2 and 3 to find the Laplace transform of e31 (sin 5t)/t.
ln Probs. 5-20, evaluate the Laplace transform of the given function
using known transforms and operational properties.
5. f(t) = 3t4e51 6. f(t) == 2e - 21 cos 23t
7. f(t) = e41 cosh 5t 8. f(t) == e- 1(3 cos 6t - 5 sin 6t)
~:: < 1 ~: ~ < 3

2
9. f(t) = {;: 10. f(t) = {: !_1,
ll. f(t) = {sin t,

O,
O<t<1r
t > 1T
12. f(t) = {C?S t,
sm t,
0<t<1T
t > 1T
!, O<t<l
13. f(t) = l, l<t<2
{ 3 - t, 2<t<3
O, t>3
15. f(t) = L e-"214 du 16. f(t) = L e

4
" cosh 5u du
17. f(t) = t2 sin kt 18. f(t) = 5te 31 sin2t

19. f(t) = (sinh t)/t 20. f(t) = (e-a1 - e-b1)/t
21. If .,<t'{f(t);p} = (1/p)e- 11 P, find the Laplace transform of e- 21/(31).

22. Given that 2{sin yt;p} = (l/2p) V1r/ p e- 114 P,

(a) use Eq. (4.11) to determine the transform of 0/Vt) cos y[.
(b) use Theor. 4.9 to determine the transform of (1/t) sin y[.
23. Given that 2{r 112 ;p} = v;TP, show that
ln Probs. 24-30, establish the given transform relation.
24
(/){ cos at - cos bt . } =!1 p
2
+ a2
2 og 2
oL
t
,p
p + bz
. tan- 1p 1
25. 2{tSI(t);p} = - -2 - - ( 2
p p p + 1)
26. 2 { e;;p} = J~ ea 1 24
P erfc C~), a> O
Hint: Use Exam. 4.4.

27. 2 { e -aVr.;p} = -1 - -a J'T-e
r a2f4 P erfc ( -a-) , a>O
p 2p p 2yp
{' f(u)
28. 2 { Jo --;;- du;p
}
=p P
1 foo F(s) ds
29. 2{J,oo f~) du;p} =; J: F(s) ds
foof() } 1( 00
30. 2:{Jo : du;p = pJo F(s) ds

oo J(u)
-du =
ioo F(s) ds
io u o
32. Use Prob. 29 to evaluate the Laplace transform of
(a) the cosine integral defined by
CI'( t ) = fi -
cos-ud U, t>O
00 u
(b) the exponential integral defined by
E (t) = foo e-u du, t >o.

1
J, u
33. The Laguerre polynomials are defined by
Show that
;t'{Ln(t};p} = ~ (p ; 1) n
34. If f(t) = (1/t)g'(t), show that
F(p) = Loo sG(s) ds

l(b) = Loo e-pu2 COS bu du, b ~O, p >O
as a function of the parameter b,

(a) show that I satisfies the first-order linear DE
di - .!?_] =o
db 2p
(b) Evaluate /(0) directly from the integral.
(c) Solve the DE in (a) subject to the initial condition in (b) to deduce
the result
/(b) = ! j~ e-b2f4p
2 p
36. Use the result of Prob. 35 to deduce the Laplace transform of
(a) (1/yt) cos avi.
(b) Differentiate the transform relation in part (a) with respect to a
and deduce the Laplace transform of sin ayt.
(c) By integrating the transform relation in part (a) with respect to
a from O to 1, deduce the Laplace transform of (1/t) sin y/.
ln Probs. 37-40, use Theor. 4.10 to find the Laplace transform of the
given periodic function.
37. f(t) = lsin ti
38. J'"'(t)
= { - 11 ', oc << 1t << c2c f(t + 2c) = Jr.(t)
39. f(t) = { '

t o< t <c f(2t + 2c) = f(t)
2c- t, c< t < 2c
40. f(t) = {631,' o< t < 2

2<t<4
f(t + 4) = f(t)
4.4 Transforms of More Complicated Functions

When the function whose transform is needed leads to a nonelementary
integral, we must usually resort to advanced integration techniques or
various tricks and manipulations to accomplish our task. Let us illustrate
with some examples.
Example 4.11: Find the Laplace transform of erf(t), which is the error
function discussed in Sec. 1.3.
Solution: From the defining integral of the Laplace transform and

that of the error function, we have
.'l'{erf(t);p} = L= e-pi erf(t) dt

2 r
= Jor= e-pi y; Jo e-xz dx dt
Recharacterizing the region of integration O s: x s: t, O s: t < oo, by

x s: t < oo, O s: x < oo, we can interchange the order of integration
(see Fig. 4.3) to get
2 r= r=
.'l'{erf(t);p} = y; Jo e-xz Jx e-pi dt dx
= _2_ r= e-(xZ+px) dx
pyi;Jo
= _2_ eP2j4 r= e -(x+p/2)2 dx
pyi; Jo
t
Figure 4.3 Region of integration
4.4/Transjorms of More Complicated Functions 183
where we have written

x2 + px = (x + p/2) 2 - lI 4
Finally, making the change of variable u =x + p/2 leads to
.:t'{erf(t);p} = .2C ePz/4 f"" e -uz du

py7T Jp/2
and we deduce that
.:t'{erf(t);p} = (1/ p)eP214 erfc(p/2), Re(p) >O
Several Laplace transform relations involve the integral formula
y; -2ab
io
=
e -aZxZ-bZfxZ dx=-e
2a '
a> O, b ~O, (4.18)
the derivation of which is sirrlar to that of Prob. 35 in Exer. 4.3.
Example 4.12: Find the Laplace transform of Ofvli)e-k 1', k ~ O.
Solution: From definition,
.:e{~~ e-kf';p} = L"" e-pt .:(/-k/t dt
where we have set t = x2 in the last step. Referring now to Eq.

(4.18), we see that
.:e{~~ e-kl';p} = J~ e-2vTP, k 2: O, Re(p) >O
By formally differentiating both sides of the result of Exam. 4.12 with

respect to the parameter k, we obtain the new transform relation
.:eL~ e-kl';p} = J~ e- vTP, 2

k >O, Re(p) >O. (4.19)
Another transform whose calculation involves the integral formula

(4.18) is given in the next example.
Example 4.13: Find the Laplace transform of erfc(ljyt).
Solution: Here we see that

Changing the order of integration (see Fig. 4.4) leads to
2 loo e-x2
.P{erfc(l/yt);p} = ~r
v 1T o
i"" l/x2
e-pt dt dx
= p~ Loo e-x2-pfx2 dx
Based on Eq. (4.18), it is clear that

.P{erfc(l/yt);p} = (1/ p)e- 2Yii, Re(p) >O
If we use the scaling property (4.8), we can readily generalize the

result of Exam. 4.13 to
.P{erfc(a/yt);p} = (ljp)e- 2aYii, Re(p)>O (4.20)
When the function f(t) that we wish to transform is continuous and
has a power series expansion about t = O that converges for ali values
t;;::: O, it may be useful to first expressf(t) in this power series representation
and then take the Laplace transform of the series termwise. ln this case
the transformed series is a Laurent series of the function F(p), and in
many cases of interest it is possible to sum this transformed series and
obtain an explicit representation of F(p). *
1
X= Vt
* See also Watson's lemma (Theor. 4.16) in Sec. 4.7.

4.4/Transjorms of More Compcated Functions 185
Example 4.14: Find the Laplace transform of sin y't.
Solution: Using the well-known series for sin x, we have

. = (- 1t(yt)2n+t r
= (- 1 1n+I/2
Slfi Vt = n~O (2n + 1)! = n~O (2n + 1)!
Therefore, the Laplace transform of this expression leads to
oL
(L){ r.. } - ~ (
sm yt,p - n~o(2n + l)!oL t
-1r
(L){ n+l/2. }
,p
"" (-1r f(n + 3/2)

= ~o (2n + 1)! pn+3/2
where we are recalling Exam. 4.3. By using property (G8) in Sec.

1.2, we can write
(2n + 2)! - (2n + 1)! -
f(n + 3/2) = 22n+2(n + 1)! v'rr = 22n+ln! v'rr
and thus we find that
-
5t{sin y't;p} = -
1 J1i- L --1t, ( -1 )n
= (
2p p n=O n. 4p
We recognize this last series as that of an exponential function, and
hence
-
5t{sin y't;p} = 1P J'TT e- 114P, P Re(p) >O
2
4.4.1 Transforms Jnvolving Bessel Functions

Bessel functions arise in a variety of applications involving the Laplace
transform and therefore it is important to know certain transform relations
of these functions.
Example 4.15: Find the Laplace transform of tv 12Jv(2y't).
Solution: Here we find it convenient to first express the given function

in a power series; thus,
lv/2] (2y't) = lv/2 L _..:.(-____.1)n:.___ [n+v/2
v n=O n!f(n + V + 1)
= i (-tr tn+v
n=O n!f(n + V + 1)
186 Chap. 4/Tbe Laplace Transform
By taking the Laplace transform of this series, we get
(f){ v/2 ( ~ r.) } _ ~ ( -l)n 5f{ n+v. }

aL t lv 2 V t ,p - :=-o n!f(n + v + 1) t ,p
"" <-W
= ~O n!pn+P+ I
1 (-l)"(l)n
= pv+t~0 ~
oo
p
Re(p) >O
The scaling property, Eq. (4.8), allows us to generalize the result of

Exam. 4.15 to
5f{tv12Jv(2yat);p} = (av12 /pv+ 1)e-a1p, a>O,Re(p)>O (4.21)
The special case v = O then yields
5f{Jo(2'0);p} = (1/ p)e-afp, a> O, Re(p) >O (4.22)
Example 4.16: Find the Laplace transform of tvJ,(t), v> -1/2.
Solution: Again relying on the series definition of the Bessel function,

we have
and thus
5f{tvJ (1) } ~ ( - l)n CO{ 2n + 2v }
v ;p = :=-o n!f(n + V + 1)22n+v aL t ;p
00
1r f(2n + 2v + 1)
n~O n!f(n
(-
= + V + 1)22n+vp2n+Zv+l
By using the duplication formula of the gamma function (see (G7) in

Sec. 1.2], it can be shown that
f(2n + 2v + 1)
f(n + v + 1) = v;1 2 2n+2v
f(n + v + 1/2)
(see Prob. 25 in Exer. 1.2). Hence, our transform relation reduces
to
5f{t"J,(t);p} = 2v ~ ( -trnn + v + I/2) (_!_)"
y; p2v+l n=O n! p2
4.4/Transforms of More Complicoted Functions 187
This last series can be summed by using properties of the binomial

series
(1 + x)-a = ~
n=O
(-a) n
x" =~
n=O
(-1t;(n +a) xn,
n.f(a)
lxl < 1
where we are using the fact that

-a)= (-l)na(a + 1) (a+ n- 1) = (-1Yf(n +a)
( n n! n!f(a)
Therefore, we see that our Laplace transform series is a binomial
series for which
or
v 2v f(v + 1/2)
;t'{t Jv(t);p} = C 2 + 1/2' v> -1/2, Re(p) > 1
v1r(p
4
+ 1Y
When v = O in the result of Exam. 4.16, we obtain the special case
5l'{J0(t);p} = 1/VJT+l, Re(p) > 1 (4.23)
Also, by using Eq. (4.11) together with the relation J(t) == -J1(t), we
have
2{J(t);p} == P 2{Jo(t);p} - Jo(O)
or
-5t{J,(t);p} = <PIVP2 + o - t
5t{J,(t);p} = (Vp 2 + 1 - p)/VJT+l, Re(p) > 1 (4.24)
EXERCISES 4.4
ln Probs. 1-8, verify the Laplace transform relation involving error
functions.
1. 5t{erf(t/2a);p} = (lj p)e 02 2
P erfc(ap), a >O
2. 2{erf(I/yt);p} = (1/ p)(l - e- 2v'P)
3. 2{erf(yt);p} = 1/ PYP+l
4. 5t{ea'erf(Vaf);p} = J~ (-
P p- a
1
-), a> O
5 .X{ea'erfc(yrat);p} = ypp(V P1 + Vaa) , a> O
4r. 3p + 8
6 .X{t erf(2 v t);p} = p 2 (p + )312
4
21 1
7. .X{1jy( - aea erfc(ayt);p} =V , a> O
p+a
8. .x{eb<bt+alerfc(byt + _a_);p} = --==-e--a=v_p_

2yt VP<VP + b)
ln Probs. 9-15, use infinite series to derive the given Laplace transform
relation.
9. .X{(sin2t)/t;p} = (1/4)log(l + 4/ p 2)
10 . .X{(l/V1Tt) cos 2Vat;p} = Jp e-afp, a> O
11 . .X{ -sin at
-;p } = tan- 1 a/p, a>O
1
12. .X{erfc(l/yt);p} = (1/ p)e- 2-vp
a>O
v> -1/2
ln Probs. 16-22, verify the Laplace transform relation involving Bessel

functions.
16. !t{e-a'J0 (bt);p} = 1/Vp 2 + 2ap + a 2 + b2 , b >O
17 .X{tJ0(at);p} = pj(p 2 + a 2) 312, a> O
18 . .X{te-'J0(t);p} = (p - l)/(p2 - 2p + 2) 3/ 2
19 .;l{tJI(t);p} = 1/(p2 + 1)3/2
(2n)! __
20 .X{fJn(at);p} = -2n'(p2 + a2) n 1/2, a> O, n = 1,2,3, ...
n.
22. 2{Ju(t);p} = <Vll+l - p)" ;\(il+t

23. Show that
(a) .5t{J0(t)sin t;p} = 4
1
sinG tan -I ~)
V"PVl+4
1
(b) .5t{J0(t) cos t;p} = 4
_ _ cosG tan -I ; )
2
VPVP + 4
24. Evaluate 2{ 1- /o(t) ;p}

25. If F(p) = .P{f(t);p}, show that
.P{ foo f f(x) dx;p} = F(log p)

Jo r(x + 1) p
ln Probs. 26-28, use Laplace transforms to verify the Bessel function
relation.
26. f" Jo(2Vxf)cos x dx = sin t
27. L"" J (2y/xt)sin x dx = cos t

0
28. L"' J (2Vxt)J (x) dx = J (t)

0 0 0
29. Use the result of Prob. 31 in Exer. 4.3 to deduce the value of
L"" J (x) dx.

0
30. If Ln(t), n = 0,1,2, ... , are the Laguerre polynomials of Prob. 33 in

Exer. 4.3, use Laplace transforms to show that
~ Ln~t) = e Jo(2y/t)
n=O n.
31. Show that
2{1 - C(~) - S(V2/7rt);p} = (1/ p)e-v'P cos VP
where C(x) and S(x) are the Fresnel integrais (see Sec. 1.3.2).
32. Using the sifting property of the impulse function (see Sec. 1.5.2),
show that
(a) .P{cS(t);p} = 1
(b) 2{8(t - a);p} = e-ap, a> O
4.5 The Inverse Laplace Transform

Thus far we have concerned ourselves only with the problem of finding
the transform function F(p), given the function f(t). However, the use
of Laplace transforms is effective in applications only if we can also
solve the inverse problem of finding f(t), given the function F(p). ln
symbols, we write the inversion formula
f(t) = ,;e- 1{F(p);t} (4.25)
First of ali, there is the question of whether a given function F(p) of
a particular class of functions actually represents the Laplace transform
of some function f(t). ln this regard we have the following theorem.
Theorem 4.11. If f is piecewise continuous on t ;::: O and is O(ecof), and

if F(p) is the Laplace transform of f(t), then
lim F(p) =O
IPI->oo
The proof of Theor. 4.11 follows that of Theor. 4.1 and is left to the
exericises (see Prob. 37 in Exer. 4.5). The real significance of Theor.
4.11 is that if F(p) is any function for which lim F(p) =/= O, then it does
IPI-+oo
not represent the Laplace transform of any piecewise continuous function
of exponential order. This condition roles out many functions as possible
Laplace transforms, such as polynomials in p, t!', cosp, and so forth.
Given a particular function f(t), we know that its transform F(p) is
uniquely determined as a consequence of the properties of integrais.
Moreover, F(p) is an analytic function even iff(t) has certain discontinuities.
The situation for the inverse transform, however, is not the sarne. For
instance, if f(t) and g(t) are two functions that are identical except for
a finite number of points, they will have the sarne transform, say F(p),
since their integrais are identical. Therefore, we can claim that either
f(t) or g(t) is the inverse transform of F(p). That is to say, the inverse
transform of a given function F(p) is uniquely determined only up to an
additive null function. * This result is known as Lerche's theorem. Null
functions are normally of little consequence in applications and so the
difficulty of finding unique inverse Laplace transforms is mostly of academic
interest. If we can find a continuous function f(t) that is the inverse
transform of F(p), that is the one we use.
When constructing inverse Laplace transforms, we find in many routine
problems that the desired inverse transforms of F(p) can be obtained
* A null function n(t) is one for which J: n(u) du = O for all t.

4.5/The Inverse Laplace Transform 191
directly from existing tables of transforms (see Appendix C). For instance,
in Exams. 4.1 and 4.2 we derived the transform relations
2{e0 ';p} = 1/(p - a)
and
cP{cos at;p} = pj(p2 + a 2)
and hence, we immediately have the inverse transform relations
;e-t {-1-;t} =
p-a
eat
and
;e-t {
pz
P
+ az
;t} = cos at
Also, many of the operational properties used in finding the transform

itself likewise can be used in constructing the inverse transform. For
example, the Linearity property (Theor. 4.3) and shifting property (Theor.
4.4) become, respectively,
;e- 1{C 1F(p) + C2G(p);t} = Ctf(t) + Czg(t) (4.26)
and
::e- 1{F(p - a);t} = ear 5f- 1{F(p);t} (4.27)
Example 4.17: Find ;e- 1{(p - 5)/(p2 + 6p + 13); t}
Solution: Completing the square in the denominator, we get

p - 5 - p - 5 - (p + 3) - 8
2
p + 6p + 13 - (p + 3) + 4 - (p
2
+ 3)2 + 4
Then, using (4.26) and (4.27), we obtain
:;e-1 { p - 5 } :;e-1 { (p + 3) - 8 }
p2 + 6p + 13 ;t = (p + 3)2 + 4 ;t
= e-Jr;e-1 { p - 8
p2 + 4'
t}
= e- 3t [ 5f- I { pz : 4
;t} - 4 ;t'- I { pz :
4
;t}]
which gives the result
;e-t { 2
P-
p + 6p
5
+ 13
;t} = e- 3'(cos 2t- 4 sin 2t)
4.5 .1 Partia[ Fractions

ln many cases of practical importance we wish to find the inverse transform
of a rational function, i.e., a functian having the form
F(p) = R(p)/Q(p)
where R(p) and Q(p) are polynomials in p. The inverse transform in
such cases can often be found quite easily be representing F(p) in terms
of its partia[ fractions. The partial fraction representation is the sarne
as that found in the calculus, for example, as a means of integrating
certain rational functions. Here it is assumed that R(p) and Q(p) have
no common factors and that the degree of R(p) is lower than that of
Q(p). Let us illustrate the technique with some examples.
Example 4.18: Find Il- 1{2/(p + 1) (p

2
+ 1); t}
Solution: Using partial fraction expansions, we write
2 =~+Bp+C
(p + 1)(p2 + 1) p + 1 p2 + 1
and clearing fractions yields
2 = A(p2 + 1) + (Bp + C)(p + 1)
Setting p = - 1, we find A = 1, and equating like coefficients of p 2
and p 0 gives the equations
O= A+ B,
2 =A+ C,
from which we deduce B = -1 and C = 1. Thus, we find
If:-1 {
. (p + 2 + r}- {-1-r}
l)(p2 +
1) , -
It-1
p 1,
- 5f- I { p2 : 1 ;t} + If- I { p2 ~ 1 ;t}

= e -t - cos t + sin t
Solution: Let us write

4.5/The lnverse Laplace Transform 193
and by clearing fractions, we have

p + 1 = Ap(p + 2) 3 + B(p + 2) 3
+ Cp 2(p + 2)
2 + Dp 2(p + 2) + Ep 2
Solving for the constants yields C = -A = 1/16, B = 1/8,
D = O, andE = -1/4. Hence,
{~;r}+ ~x- 1 {~2 ;r}

1 1 1
x- {p2rp ++ 2)3;r} = -/6 x-
1 1 }
+ 16 X
-1 {
P + 2 ;t - 41 X -1 {
(p
1
+ 2)3 ;t
}
= _ _!_ + ! 1 + I.e-21 _ ! 12 e-21

16 8 16 8
ln Sec. 4.6 we will develop more systematic means of finding the

inverse Laplace transform of rational functions. Specifically, we will use
methods of complex variables to derive what are known as Heaviside's
expansion theorems. These theorems provide an approach to determining
the unknown constants in partial fraction expansions that is more so-
phisticated than clearing fractions and matching like terms, and so on,
as we have done here.
Finally, we wish to present one more example involving partial fractions
that also makes use of a previously derived property of the Laplace
transform.
Example 4.20: Find x- 1{log (1 + 1/p2); t}
Solution: Letting F(p) = log (1 + 1/p2), we first observe that
F'(p) =- 2 = -2(!- _P_)

p(p2 + 1) + p p2 1
where we have used a partial fraction expansion. Now recalling the
property [see Eq. (4.14)]
X{if(t);p} = - F'(p)
which can also be written as
x- 1{F'(p);t} = - tf(t)
we see that in our particular case
x- 1
{F'(p);t} = -20 - cos r)= -tf(t)
Hence, we deduce that
f(t) = 2(1 - cos t)/t

which is the inverse Laplace transform of F(p) = log(l + 1/p 2).
4.5.2 Series Method

ln constructing Laplace transforms, we previously found it useful in
certain situations to express the functionf(t) in terms of its power series
and take the transform termwise. This sarne technique also proves fruitful
in constructing inverse Laplace transforms. That is, if F(p) has a convergent
Laurent series expansion of the form
a0 a1 a2 a3
F(p) = -
p
+ p2 + p3 + p4 + (4.28)
then under suitable conditions we can invert (4.28) to obtain*

12 t3
f(t) = a0 + a1t + a2 21 + a3 3! + ... (4.29)
ln some cases the resulting series (4.29) may be summable to a known

function.
Example 4.21: Find the inverse Laplace transform of (1 j p )e -I/P.
Solution: Using infinite series, we have

00 00
1 -1 1 <-1t <-1r
Pe IP = P~o n! Pn = ~o n! Pn+ I
lnverting the series term by term yields
00
::t-1
{
_1 e-lfp;t } = 2: __
1t ::t-1 { --;t
(- 1 }
p n=O n! pn+l
00
( -1t ln
2:
= n=O (n!)
2
and thus we see that
::t-
1
{~ e-l/p;t} = lo(2Vt)
4.5.3 Convolution Theorem

1t is often expedient to resolve a Laplace transform into the product of
two transforms, i.e., F(p)G(p), when the inverse transform of both F(p)
* See also Watson's lemma in Sec. 4.7 (Theor. 4.16).

and G(p) are known. This sarne situation arises when one of them represents
an arbitrary function - typically an input to some physical system. The
inversion of the product F(p )G(p) can then be obtained by application
of what we call the convolution theorem.
ln order to derive the convolution theorem, let us begin by writing
the product of the transforms of f(t) and g(t) as the iterated integral
F(p)G(p) = L"" e-px f(x) dx L"" e-pu g(u) du

= L"" L"" e-p(x+u) f(x)g(u) dx du
The change of variable x = t - u then leads to
F(p)G(p) = L"" L"" e-pr f(t - u)g(u) dt du (4.30)
We can interpret the integrais in (4.30) as an iterated integral over the

region u :s t < oo, O :s u < oo, as shown in Fig. 4.5. If we change the
order of integration, we find that the region of integration is characterized
by O :s u :s t, O :s t < oo, and thus
F(p)G(p) = L"" Le-pr f(t - u)g(u) du dt
= L"" e -pr Lf(t - u)g(u) du dt
We now define the Laplace convolution integral of f(t) and g(t), i.e.,
(f * g)(t) = L f(t - u)g(u) du (4.31)

ln terms of the convolution integral, we have shown that

F(p)G(p) = ..P{ lf* g)(t);p} (4.32)
and by formally taking the inverse Laplace transform of(4.32), we deduce
that
.x- 1{F(p)G(p);t} = <f*g)(t) = Lt(t- u)g(u)du (4.33)
which is our intended result.

The integral in (4.33) arises in many applications where it represents
a superposition of effects of magnitude g(u), occurring at time t = u,
for which f(t - u) is the influence function or response of a system to
a unit impulse delivered at time t = u.
ln summary, we have the following theorem.
Theorem 4.12 (Convolution theorem). lf f(t) and g(t) are piecewise con-
tinuous functions on t ~ O andare O(eco'), and if F(p) and G(p) are the
Laplace transforms, respectively, of f(t) and g(t), then the inverse Laplace
transform of the product F(p)G(p) is given by the formula
.,P- 1{F(p)G(p);t} = <f* g)(t)
Once again we point out that the conditions stated in Theor. 4.12 are
more stringent than required for the validity of (4.33). That is, there exist
functions satisfying (4.33) that are not piecewise continuous on t ~ O.
Some examples illustrating this point are included in the exercises (see
also Exam. 4.23).
Example 4.22: Find .x- 1 {1/p2(p 2 + e);t}.
Solution: Let us select F(p) = 1/p 2 and G(p) = 1/(p2 + k 2), whose
inverse transforms are
and
.x-l {l/(p2 + e);t} = g(t) = (1/k) sin kt
Thus, using the convolution theorem (4.33), we write
.x-l {1/ p 2(p 2 + e);t} = <f* g)(t) = L (t - u)(l/k) sin ku du
which leads to the result

.x- 1 {1/p2(p2 + k 2);t} = (l/k 3) (kt- sinkt)
The convolution integral (4.31) satisfies certain properties which often
prove useful in practice. For example, by making the change of variable

v = t - u in (4.31), we find
U* g)(t) =- f f(v)g(t - v)dv = J: g(t - v)f(v)dv
from which we deduce the commutative relation

(f* g)(t)= (g * f)(t) (4.34)
We also write this as simply f * g = g * f. From definition, it is clear
that
f* (Cg) = (Cf) * g = CU* g) (4.35)
where C is a constant, and
f*(g + k) =f*g +f*k (4.36)
which is the distributive relation. The convolution integral (4.31) also
satisfies the associative relation
(4.37)
but this is a little more difficult to prove.
Our final example in this section illustrates how the convolution theorem
can be used in a less direct manner to derive a certain result.
Example 4.23: Prove that

lo(t) = ~ {I cos tx dx
7T Jo yl - x2
Solution: Recalling Eq. (4.23), we have
5f {Jo(t);p} = ~ r:::r-71
yp + 1
U sing the fact that
:;e-t {1/~;t} = (1/y7Tt)e-at
which follows from the shift property (Theor. 4.4).applied to the result
in Exam. 4.3, we obtain by way of the convolution theorem,
1
Jo(t) = 5f -1 { c-:--; .. I
1 . ;t }
yp+z yp-z
=.!.. t u-1/2 e-iu (t- u)-1/2 ei(l- u) du

7T Jo
=-
1i/ ei(t - 2u)
du
7T 0 yu(t - u)
By letting u = tv, this becomes

1
Jo(t) = -
ii eit(l-2v)
dv
7T 0 y'v(l - v)
and then by introducing x = 1 - 2v, we have
Jo(t) = -
II itx
1
~dx
7T -I V 1 -X
4
=- I y'l""=7 I
1 1
1 cos tx dx+-i sin tx dx
7T -t 7T -I y't - x2
Finally, by using properties of even and odd functions, we conclude
that the integral involving sin tx vanishes and the remaining integral
leads to
Jo(t) = 3_ t cos tx dx
7T Jo v'1 -r
EXERCISES 4.5
ln Probs. 1-10, determine the inverse Laplace transform using the table
in Appendix C and various operational properties.
1 3p + 7
1. F(p) = 2p + 3 2. F(p) =
p
2
+5
2p 1
3. F(p) = (p - 3)s 4. F(p) = p2 - 6 p + 10
p 5p- 2
5. F(p) = 2
p -
6
p + 13 6. F(p) = 3p 2 + 4 p + 8
2p + 3 p2
7. F(p) = 4 4
2
p + p +
5 8. F(p) = (p + 2)4
e-sp
10. F(p) = (p + l)e-"P
9. F(p) = (p - 3t p2 + p + 1
ln Probs. 11-20, evaluate the inverse Laplace transform by the method
of partial fractions.
1 1
11. F(p) = p(p + 1) 12. F(p) = (p - l)(p + 2)(p + 4)
p2 3p- 2
13. F(p) = (p + Z)3 14. F(p) = p3(p2 + 4)
4.5/The lnverse Laplace Transform 199
p+l I
15. F(p) = (pz - 4p)(p + 5)z 16. F(p) = - 4- -
p - 1
p 4p 2 - 16
17. F(p) = (pz + az)(pz + bz) 18. F(p) = p3(p + 2)z
19
i- 3
' F(p) = (p + 2)(p - 3)(l + 2p + 5)
3p 2 - 6p + 7
20. F(p) = (pz - 2p + 5)z
21. Given that ::r 1{F(p);t} = f(t), show for constants, a, b, and k that
(a) 5/!- 1 {F(kp);t} = (1/k)f(t/k), k >O
(b) 5f- 1{F(ap + b);t} = (l/a)e-bt/a f(t/a), a> O
22. Given that f(t) = t sin t is the inverse Laplace transform of F(p)
= 2p/(p2 + 1)2 , use Theor. 4.7 to evaluate
2 -1 {l/(pz + l)z;t}
ln Probs. 23-28, use infinite series to find the inverse Laplace transform
of the given function.
23. F(p) = log(l + 1/p) 24. F(p) = log(l + 1/ p 2)
p+1 p-a
25. F(p) = log-- 26. F(p) = 1og--b
p- 1 p-
21. F(p) = 1/pyp + 4 28. F(p) = (tjyp)e- 11 P
ln Probs. 29-36, use the convolution theorem to find the inverse Laplace
transform of the given function.
29. 1/ PY p + 4 30. 2/(p + l)(p 2 + l)
31. pf(pz + az)z 32. 1/ p 2(p + 1)
2
33. 1/yp(p - 1) 34. 1/(p - l)pyp

2 2
35. 1/(p + 1) (p + 4) 36. 1/(yp - I)
37. Prove Theor. 4.1 1.
38. Show that
(a) I * 1 * 1 == t /2
2
(b) t * t * t = 1 /5!
5
m-1 n (m - l)!n!tm+"
(c)t *I= ( )I ;m,n=1,2,3, ...
m + n.
39. Starting wth f(t) = Lux- (t - 1 ur-l du' x,y > O,

(a) use the convoluton theorem to show that
F(p) = f(x)f(y)/ px+\
(b) From the result in (a), establish the formula
J 1
0
lf- 1 (I - u)Y- 1 du = f(x)f(y)/f(x + y), x >O, y >O
40. Show that

(a) J: J 0 (t - u)J0 (u) du = sin t
(b) J: cos(t - u)sin u du = (1/2)t sin t
(c) J: sin(t - u)Jo(u) du = (l/2)tJ (t) 1
(d) I: J1(t - u)J0 (U) du = J (t)

0 - COS t
41. Show that
-'1T 1,v
I
o
ezau
u(t - u)
du = eat lo(at)
42. Show that
!t-t {Pv'} 1 ;t} = J'1Tt + e, erf(yt)

43. Use the result of Prob. 42 to evaluate :E- 1 {1/(1 + yp);t}
44. Prove that
4.6 Complex Inversion Formula

The techniques introduced in Sec. 4.5 for evaluating inverse Laplace
transforms are adequate for a wide variety of routine applications involving
the Laplace transform. Mostly they make use of tabulated results coupled
with operational properties ofthe transform. More complicated problems,
however, may lead to transforms that are not in tables, and in such cases
we may need more powerful direct methods for constructing the necessary
inverse Laplace transforms. Some of the methods of complex variables
can be particularly useful in deriving more direct inversion formulas.
ln Sec. 4.1 we gave an heuristic argument that led to the integral
formula
} fc+ico
f(t) = -. ept F(p) dp (4.38)
27TZ c-ico
where f(t) and F(p) are Laplace transform pairs. ln order to provide a
4.6/Complex lnversion Formula 201
more rigorous derivation of this formula, let us assume that both f(t)
andf'(t) are continuous functions on t ~O and thatf(t) is O(ec1). Then,
based on Theors. 4.1 and 4.2, we know that the Laplace integral
F(p) = Loo e-pu f(u) du (4.39)
converges absolutely and uniformly in the half-plane Re(p) ~ c2 > C0

and that F(p) is analytic in the half-plane Re(p) > C0 Substituting the
integral representation (4.39) for F(p) in the following integral, we obtain
f:i~ eP 1
F(p) dp = f:i~ eP Loo e-pu f(u) du dp
1
(4.40)
Our goal at this point is to show that by allowing ~ oo, we can derive
Eq. (4.38) from Eq. (4.40). Because ofthe requirements we have imposed
on f(t), we can interpret (4.40) as an iterated integral and interchange
the order of integration. Doing so yields
c+iA ept F(p) dp = loo f(u) fc+iA eP<t-u) dp du

f c-iA O c-iA
= 2i loo f(u)ec<r-ul sin (t - u) du

o t-u
= 2iect I"" f(t + x)e-c<r+xl sin X dx
-t X
where in the last step we have made the change of variable u = t + x.

If we now invoke Lemma 2.2, it follows that
1 fc+iA
lim - . e~'1 F(p) dp = ec1 f(t)e-ct
A->oo 27TZ c- iA
= f(t), t >o (4.41)

which is the sarne as (4.38).
Let us summarize this result in the form of an inversion theorem.
Theorem 4.13 (lnversion theorem). If F(p) is an analytic function of the

complex variable p in the half-plane Re(p) > c, and further, if F(p) is
O(p- "), * where k is real and k > 1, then the inversion integral
1 fc+ioo
f(t) = 27Ti c-ioo ePI F(p) dp
converges to the real functionf(t), which is independent of c and whose
* By saying F(p) is O(p-k), we mean there exists a positive real constant M such that
lpkF(p! < M whenever !PI is sufficiently large.
Laplace transform is F(p) for Re(p) > c. Also, the function f(t) is
O(ec"') and is continuous everywhere, and f(t) = O when t :5 O.
The conditions stated in Theor. 4.13 are quite severe in that they
exclude, for example, the simple function 1/p, which is O(p-~ where
k = 1. Also, these conditions are not satisfied by transforms of functions
that are discontinuous or for which f(O) =/= O. Nonetheless, by using a
Fourier integral theorem and stating conditions on the functionf(t) rather
than on F(p), these conditions can be relaxed so that the inversion
integral formula is valid in nearly all practical cases of interest to us. *
We do note that the conditions stated in Theor. 4.13 ensure the existence
of the inverse Laplace transform of F(p), and moreover, ensure that the
inverse Laplace transform f(t) is that function for which 5t'{f(t);p} =
F(p).
ln order to better understand the conditions stated in Theor. 4.13,
2
let us consider the function F(p) = ' , which is analytic everywhere in
2
the complex plane. Therefore, integration of ''' along the imaginary
axis p = iy yields
~ fi= '''2 dp = _1 I"" eiry e- y2 dy

2m -i= 21T -=
The integral on the right can be interpreted as a Fourier transform of

2
the function e-y , and based on the result of Exam. 2.6 in Chap. 2, we
deduce that
1
-.
fi= eP 1'
2
dp = -1- Bf{e-Y2;t}
2m -i= y-2;
1 -t /4 2
=--=e (4.42)
2y'7T
However, the Laplace transform of this last function is (see Prob. 23 in
Exer. 4.2)
.;e{ 2~ e-' 214

;p} = ~ '
2
erfc(p) =/= F(p) (4.43)
What we are illustrating here is that the inversion integral of F(p) may
exist without representing the function f(t) whose Laplace transform is
F(p). The reason it happens in this particular case is because the function
2
F(p) = ' is not O(p-~ for any positive k. Thus we have violated a
condition of Theor. 4.13.
* See, for instance, R. V. Churchill, Operational Mathematics, New York: McGraw-

Hill, 1972, Chap. 6.
4.6/Complex Inversion Formula 203
The complex inversion integral for the inverse Laplace transform can
often be evaluated quite readily through use of the theory of residues.
Let us suppose that F(p) is an analytic function in the complex p-plane,
except for a finite number of isolated singularities a~> a 2 , , aN. By
integrating the function
eP' F(p)
around the closed contour shown in Fig. 4.6, where c and R are selected
such that no singularity ak lies to the right of Re(p) = c and the radius
R of the circular are CR is large enough to enclose all singularities of
F(p), we have the result
,! eP'F(p )dp = JB eP'F(p )dp +

t A
J~ eP'F(p )dp
N
= 27Ti 2: Res{eP'F(p);ak} (4.44)

k=l
The integral along CR can be split into three parts, i.e.,
f cR
eP'F(p )dp ={
JnJ
eP' F(p )dp +J
JKL
eP' F(p )dp
+ fu eP' F(p)dp (4.45)
Based upon Theorem A.6 in Appendix A, it follows that
lm(p)
J
K Re(p)
Re(p) = c
L
lim
R-oo JJKL
r eP1 F(p )dp = O (4.46)
However, sinee the ares BJ and LA do not lie in the seeond and/ or third
quadrants, the vanishing of the remaining two integrais in (4.45) in the
limit must be treated separately. Under the assumption of Theor. 4.13,
F(p) is of order O(p- k) and therefore it follows that there exists a eonstant
M sueh that
IF(p)l <M/R\ M >O, k > 1 on CR
We now eonsider
II1I = I L 1
epr F(p)dp I
::s r
JBJ
iepriiF(p )iidPi
::s (M/Rk) JBJ iepriidPi

Along the are BJ we set p = Re;8 , ()0 ::s () ::s Tr/2, so that
idpi = RdO
and therefore
/ I ::s __
M l.,.;z eRr cos e dO
1I Rk-1 Bo
::5 ~ rtf>o eRt sin q, d"'

Rk-1 Jo 'I'
where c/J = Tr/2 - () and c/Jo = sin- 1 (e/R). Using the obvious inequality
sin cfJ ::s sin c/Jo = c/R
we deduee that
M rtf>o
liii ::5 Rk-1 Jo ect dcfJ
::5 ~ ect sin- I (e/R)

Rk-1
As R ~ =, we ean use the approximation

sin- 1(c/R) =e/R, R>>c
to find
lim
R.....~J>oo
liii ::5 lim (Me/R")
R--+oo
ect =o (4.47)
4.6/Complex lnversion Formula 205
Hence, the integral along the are BJ vanishes in the limit R~ oo, Similar
arguments can be used to show that the integral in (4.45) along LA also
vanishes as R ~ oo. Based on the above results, we deduce that
lim { epr F(p) dp = O (4.48)

R->oo JcR
and Eq. (4.44) reduces to
N
f(t) = ;e- {F(p);t} =

1
L Res{eP F(p);ak}
1
(4.49)
k=l
lf F(p) = R(p)/Q(p), e.g., when both R(p) and Q(p) are polynomials,
then Eq. (4.49) can be expressed in more explicit forms. For instance,
if Q(p) has simple zeros at p = a~> a 2 , ... , aN, then (4.49) becomes
f(t) = ;e-I {R(p)

Q(p)'
t} = k=li Q'(ak)
R(ak) eakr (4.50)
This last result is due essentially to Heaviside and is widely known as

the Heaviside expansion theorem. Other similar expansion theorems in-
volving higher-order poles can also be readily obtained.
Example 4.24: Find the inverse Laplace transform of pj(p 2 + a 2).
Solution: Although the inverse transform of this function is familiar

to us, we wish to calculate it again using (4.50). Here we see that
Q(p) = p 2 + a 2 has simple zeros at p = ia. Also, R(p)/Q'(p) =
1/2 and hence,
1 1
;t' _ 1 { p } = Res { peP ;w + Res peP ; - za

2 ;t
} { }
2 2 2 2 2
p+a p+a p+a
= !(eiat + e -iat)
= cos at
We can formally extend the result (4.49) to the case in which F(p)
has infinitely many isolated singularities by allowing N to become infinite.
ln this case, we obtain
f(t) = ;e-I {F(p);t} = L Res{eP F(p);ak}

1
(4.51)
k=l
Example 4.25: Find the inverse Laplace transform of

coshxyp
F(p) = .. I ' o< X < 1
p cosh v p
206 Cbap. 4/Tbe Laplace Transform
Solution: At first it might appear that p = O is a branch point of

F(p) because of the presence of yp. That this is not the case can
be seen by writing
coshxyp 1 + (xyp) 2/2!
+ (xyp)4/4! + ...
F(p) = .= _ __:_~~-__;_=-=-__;_--
P cosh yp p[l + (ypi/2! + (y p) 4/4! + ... ]
1 + x 2p/2! + x4p 2/4! + .. .
= p(l + p/2! + p 2/4! + ... )
from which it is clear that p = O is not a branch point but a simple
pole of F(p). ln addition, there are infinitely many other simple poles
given by the roots of the transcendental equation
cosh YP =O
The solutions of this equation are given by
YP = (n - 1/2)7ri, n = 1,2,3, ...
or
p = an = - (n - 1/2)277"2 , n = 1,2,3, ...
Calculating the residues at p = O and p = an yields
Res{eP' F(p);O} = coshxyp I =1

cosh YP p=o
and
Res{eP' F(p);an} = eP' cosh x\fp

I
cosh YP + 2YP smh YP
1
p=a.
= 4( -Ir cos[(n - 1/2)7rx] e-rn-t/2)2.rlt

7r(2n - 1)
!!_ 1 { cosh xyp ;t}

pcosh yp
4
=1+- L (2n-l)n
oo
1T' n=l
(
- 1)
cos[(n - 1/2)7rx]e-<n-t/2)2.r2t
4.6.1 Multivalued Functions

The method we have presented for evaluating inverse Laplace transforms
when F(p) has isolated singularities can be modified to include the case
when F(p) has branch points. We do this by deforming the contour of
Fig. 4.6 to exclude the branch point(s) and then take the limit as the
radius of the small circle around the branch point(s) tends to zero. Let
us illustrate the procedure with an example.
Example 4.26: Find the inverse Laplace transform of F(p) == e-aYP,

a> O.
Solution: The function F(p) has a branch point at the origin p

O. Hence we integrate the function
epr F(p)
around the contour shown in Fig. 4. 7. Along C2 we write
where x is real and positive, and on C3 we write
Therefore, from the residue calculus, we find
JAept-aVp dp + Jc,r ept-aVp dp _ JRrp e-xt-iaVx dx

B
+ r ept-avP dp _ JR e-xt+iaVx dx + r ept-avP dp == 0

J~ p J~
lt can be shown that the integrais along C~> CP, and C4 all tend
to zero as p ~O and R ~ oo. Hence, it follows that
2-l{e-aY>;t} == ~ r= e-xt+iaVx dx- ~ r= e-xt-iaVx dx

27Tl Jo 27Tl Jo
lm(p)
Figure 4. 7 Contour of integration

Ii""
=-
7T o
-
e -xt sin(ay'x) dx
The change of variable x = u 2 further leads to

.P- 1{e-aYP;t} = -2
7T
i""o ue-"21 sin au du
= V2/7T 3's{ue-"21 ;a}
From Exam. 2.6 in Chap. 2 we know that
.'1P{e-u2t;a} = (2!)-1/2 e-a2f4t
and thus
3P{ue-uzt;a} = -i! [ (2t)-l/2 e-a2/4t]
= ia(2t)-3/2 e-a2f4t
However, for a > O the Fourier and Fourier sine transforms are
related by
3'{ue-u2t;a} = i 3's{ue-uzr;a}
and so we finally deduce the intended result*
.P-t {e-aYP;t} = a(47Tf)-t/2 e-a2f4t, a> O
EXERCISES 4.6
ln Probs. 1-12, use the complex inversion formula to evaluate the inverse
Laplace transform of the given function having isolated singularities.
1 p
1. F(p) = (p + l)(p _ 3)2 2. F(p) = (p + 1)3(p _ ti
p2 p2
=~
+4
3. F(p) = (p2 + 4i 4. F(p)
p
S. F(p)
1
= p 2 cosh p 6. F(p) = co~h xp ' O< x < 1
psmhp
sinhxy'p sinhxy'p
7. F(p) = . r:, = r:
8. F(p) y' ,
smh vP 4
v p cosh p
4
O<x<l O<x<I
* See aJso Prob. 22(a) for another derivation of the Fourier sine transform
.'9is{ue-\a}.
1 10. F(p) = sinhxp , O<x< 1

9. F(p) = p(' + 1) p 2 coshp
1 1 1
11. F(p) = 2p - . h 12. F(p) = c
psm p p cosh vp
13. Show that
2:-1 {/o(ryp)
Jo(byp)'
t} = ~ ~ knlo(knr)
b n= 1 J1(knb)
e-k~r
where the kn are solutions of J0(knb) = O, n = 1,2,3, ...

14. U sing the result of Exam. 4.26,
(a) determine the inverse Laplace transform of (1/p)e-av'P by way
of the convolution theorem.
(b) Differentiate the result in (a) with respect to a to deduce that
2:-1 { J]/-avP;t} = J1T/-a 2

/4t, a> o
ln Probs. 15-20, use the complex inversion formula to evaluate the
inverse Laplace transform of the given function with branch points.
1 1
15. F(p) = r::--: 16. F(p) = .. 1
vp-a pyp+a2
1
17. F(p) =c 2
18. F(p) =c
v p(p- a) VP +a
19. F(p) = log(l + 1/ p) 20. F(p) = log(l + 1I p 2)

I(t) = L~ e-bu 2
cos tu du, t::::: O, b >O
as a function of the variable t,

(a) express cos tu in a Maclaurin series and perform termwise in-
tegration to find a series representation for /(t).
(b) Sum the series in (a) to deduce the result
I(t) = ! J~ e-r2f4b
2 b
22. Using the result of Prob. 21 and
(a) differentiating with respect to t, show that
("" ue- 12u212 sinaudu = ~JTT e- 02121 t>O

Jo t 2t '
ZlO Chap. 4/The Laplace Transform
(b) integrating with respect to t, show that
L"" (1ju)e- 1 2
u sin au du = (Tr/2) etf(a/2Vt)
23. Using contour integration around the branch point p = O, show that
(a) :;e-t{_!_ e-avP;t} = 1 - _!_ (""! e-xt sin ay; dx, a> O
p 'TTJo x
(b) Use the result of Prob. 22(b) to arrive at
:;e-t{~ e -avP;t} = etfc(a/2yt), a>O
24. From the defining integral of the Laplace transform, show that
(a) ::e{--v:;i +
1
- e'etf(Vt);p} = YP
p-1
(b) Apply the complex inversion formula to ypj(p - I) and use
the result in (a) to deduce that
+ ;p } = Vr;p -
::t{ 1 yt -
TrePerfc(y p)
1
(c) Finally, establish the relation
::e L~ 1 ;p} = lP -::e{yr(t 1

+ 1) ;p}
and thus conclude that
::e{Vtt(t1+ 1) ;p} = TrePetfc(yp)

Hint: Let t = u2 in the transform integral.
4. 7 Additional Topics
ln earlier sections we have found that the transform function F(p) has
certain useful properties, most of which can be established by relying
on the Laplace transform integral for the definition of F(p). For example,
under appropriate conditions on the inverse transformf(t), we have found
that F(p) is analytic (Theor. 4.2) and that F(p) ~O as IPI ~ = (Theor.
4.11). ln addition, certain operational properties of this function were
developed in Sec. 4.3. Here we will discuss briefty further asymptotic
properties of F(p) and extend the definition of the Laplace transform to
functions defined on the entire realline rather than only on the positive
realline.
4.7.1. Asymptotic Properties of F(p)

The behavior of a function f(t) in the neighborhood of the origin is, in
some sense, reflected in the behavior of its Laplace transform F(p) as
IPI ----+ oo. For instance, if f(t) is a polynomial of degree k, i.e., if
f(t) = ao + a1t + ai2 + ... + ai<, k = 1,2,3, ...
then its Laplace transform is
a0 a1 2a2 k!ak
F(p) = -
p
+ 2p + -p3 + + ~
p
From these equations, we immediately deduce that
lim pF(p) = ao = /{0) (4.52)
!PI-+=
Not only is this particular property true of all polynomials, but applies
to a relatively large class of functions.
Theorem 4.14 (lnitial Value Theorem). Iff(t) is piecewise continuous on

t ~ O and is O(ec01 ), then its Laplace transform F(p) satisfies
lim pF(p) = limf(t) = /(0)
IPI-+= ,_.o
Proof: We will present the proof only for the stronger case where
f(t) is continuous on t ~ O.
By Eq. (4.11) in Sec. 4.3,
.P{f'(t);p} = L"" e-pt f'(t) dt = pF(p) - /(0)
Iff'(t) is piecewise continuous and of exponential order, then its Laplace

transform satisfies (Theor. 4.11)
lim f"" e -p'f'(t) dt = O

IPI-+= Jo
lim [pF(p) - /(0)] =O
IPI-+oo
or
lim pF(p)
IPI-+oo
= limf(t) = /(0)
t->0
A related result is given in the next theorem.
Theorem 4.15 (Final Value Theorem). If f(t) is piecewise continuous on

t ;;::= O and is O(ecot), and if the integral
L"" f'(t) dt
exists, then the transform F(p) satisfies
lim pF(p) = lim f(t) = f(oo)
p-+0 t-+a:J
Proof: As in the proof of Theor. 4.14, we will present the proof here
only for the case when f(t) is continuous on t ;;::= O.
Again we start with the relation
.T{f'(t);p} = L"" e-pt f'(t) dt = pF(p) - /(0)
Noting that
lim
p .....
r=
oJo
e-pt f'(t) dt = r= f'(t) dt
Jo
= lim
t-+oo Jor' f'(u) du
= lim f(t) - /(0)
/-+00
we deduce that
limf(t) - /(0) = lim pF(p) - /(0)
(-+00 p-+0
or, upon simplification,

lim pF(p)
p-+0
= f(oo)
Example 4.27: Find the Laplace transform of the cosine integral Ci(t),
defined by
Ci(t) = J' cosu u du,

00
t >O
Solution: Although this transform can be found directly from Theors.

4.7 and 4.9, we wish to illustrate another technique using the final
value theorem (Theor. 4.15).
lfwe setf(t) = Ci(t), thenf'(t) = (cos t)/t and thus
tf'(t) = cos t
Taking the Laplace transform, using Theorem 4.8 on the left-hand

side, we obtain
d p
dp [pF(p) - f(O)] = P2 + 1
and upon integration, this leads to
pF(p) - f(O) = I p2: 1 dp
or
pF(p) = (1/2) log(l + p 2) +C
where f(O) has been absorbed in the constant of integration to give
us the constant C. By the final value theorem,
lim pF(p) = limf(t) = O
p-+0 t-+oo
Thus, we deduce that C = O and consequently,

!t{Ci(t);p} = F(p) = (1/2p) log(l + p 2)
A result related to Theors. 4.14 and 4.15 is Watson's lemma below,

which defines an asymptotic expansion of F(p) <I PI ~ =) for a certain
class of functions f(t).
Theorem 4.16 (Watson's lemma). If f(t) is O(ec01 ) and if, in some neigh-
borhood of t = O, the function f(t) has the Maclaurin series expansion
00
f(t) = L a~ tn, ltl <R

n=O n.
then the transform function F(p) has the asymptotic series
00
F(p)- L ~:I'
n=O P
The result of Watson's lemma follows formally by applying the Laplace
transform termwise to the Maclaurin series expansion of f(t). * Let us
illustrate the use of this theorem in the following example.
Example 4.28: Find an asymptotic expansion for the complementary

error function erfc(x).
*For a rigorous proof of Watson's lemma, see I. N. Sneddon, The Use of Integral
Transforms, New York: McGraw-Hill, 1972, pp. 188-190.
214 Cbap. 4/The Laplace Transform
Solution: ln Exam. 4.11 in Sec. 4.4, we have shown that

2
.:t'{erf(t);p} = (1/ p)' / 4 erfc(p/2)
Now the error function has the Maclaurin series expansion [see Eq.
(1.20) in Sec. 1.3]
2
erf(t) = - 2:oo _:__;.____
(-1tt2n+l
v; n=O n!(2n + 1)
so based upon Watson's lemma, we deduce that
.!. e-.../)2;4 erf<c(p 12) _ ~ ~ ( -1t(2n)!
. . 1- LJ 2n+2
p y'Tr n=O 1
n.p
__1 ~ ( -1t f(n + 1/2)
1rp n=O (p/2)2n+i
where we have used the duplication formula of the gamma function.
Finally, setting x = p /2, we get the desired asymptotic expansion
erfc(x)- e-x2 ~ ( -l)n r~n+~ 1/2)' lxl--' 00
1r n=O X
4.7.2 Two-Sided Laplace Transform

lf the function f(t) is defined over the entire real axis, we may consider
the integral
2 +{f(t);p} = f~"" e-pt f(t) dt = F +(p) (4.53)
known as the two-sided Laplace transform or bilateral Laplace transform.

For pore imaginary p, say p = is, we see that (4.53) becomes
F +(is) = J~ e-ist f(t) dt (4.54)
which is a multiple of our definition for the Fourier transform of f(t).

Thus, in a formal sense, there is a relation between the two-sided Laplace
transform and the Fourier transform.
If the integral in (4.53) converges, it will usually do so only for
restricted values of p in a vertical strip of the complex p-plane. To better
understand this, let us rewrite the integral in two parts as
J~oo e-P'f(t)dt = J:oo e-P'f(t)dt +L"" e-ptf(t)dt

=L"" e+P'f(-t)dt +L"" e-P'f(t)dt
We see, therefore, that the two-sided Laplace transform is simply a sum

of two ordinary Laplace transforms, i. e.,
2 + {f(t);p} = 2{!(- t);- p} + 2{f(t);p} (4.55)
The second transform in (4.55) exists in the plane Re(p) > c~> where c 1
is the abscissa of convergence. Similarly, the first transform will exist
in the plane Re(- p) > - c2 , or Re(p) < c2 , where - c2 is the abscissa
of convergence in this case. The two-sided Laplace transform of f(t)
exists only if these two half-planes overlap. That is, it will exist only in
the strip c 1 < Re(p) < c2 lf c2 < c~> the two-sided Laplace transform
does not exist, and if c2 = c 1 , the strip contracts to the vertical line
Re(p) = c 1 Finally, if c2 = c 1 = O, the two-sided Laplace transform
is then actually a Fourier transform [see Eq. (4.54)].
12
Example 4.29: Find the two-sided Laplace transform of e-
Solution: Here f(t) = f(- t) = e- 12 , for which c 1 = -ao and c2 =

ao. Thus the two-sided Laplace transform exists for ali values of
Re(p).
By definition,
({)
oL + { e -t2 ;p} = I""
-oo
e -pt e -t2 dt
The inversion theorem for the two-sided Laplace transform has the
sarne form as for the ordinary Laplace transform, except that for the
two-sided transform the interval of convergence must also be established
in order to uniquely establish the inverse transform. For example, consider
the two functions
0 t<O
f(t) = { :_21 -t
e -e , t>O
and
g(t) = {e=:;
e '
t <o
t >o
whose two-sided Laplace transforms are, respectively,
Re(p) > -1
and
- 2 < Re(p) < - 1
Except for the strip of convergence specified in each case, these are
identical transforms. This means that to use the inversion formula to
find the inverse transforms of F + (p) and G + (p), we must set up contours
over different vertical strips in the complex p-plane for each case. Spe-
cifically, these particular inversion formulas are
1 fc+ioo
f(t) = - 7Tl. . eP
1
F +(p) dp, c> -1
2 c-100
and
1 fc+ioo
g(t) = - 7Tl. . 1
eP G+(P) dp, -2<c< -1
2 c-100
For a more detailed discussion of the two-sided Laplace transform,

the reader is advised to consult W. R. LePage, Complex Variables and
the Laplace Transform for Engineers, New York: Dover, 1980, or
D. V. Widder, The Laplace Transform, Princeton: Princeton University
Press, 1941.
EXERCISES 4. 7
1. Verify the initial value theorem for the following functions:
(a) f(t) = 5 + 4 cos 2t
(b) f(t) = (3t - 2)3
(c) f(t) = erf(t)
2. Verify the final value theorem for the following functions:
(a) f(t) = 3 + e- 21(cos t + sin t)
(b) f(t) = 1 + e_,z
(c) f(t) = erfc(l/y't)
ln Probs. 3 and 4, use the technique of Exam. 4.27 to find the Laplace
3. Si(t) = r sinu u du,
Jo
t >O 4. Ei(t) = (oo e-u du,
J, u
t >O
5. Use the final value theorem to evaluate the following limit:

lim t"12 J,(2yt)
(-.oo
6. If f(t) and f'(t) are continuous functions on t ~ O, f"(t) is piecewise

continuous on t ;::: O, and ali three functions are of exponential order

c0 , show that f(O) = O and
lim p 2F(p) = f'(O)

IPI-=
7. Given the function
F(p) = ("" e-px dx

Jo 1 + x2
(a) show that its inverse Laplace transform leads to the Maclaurin
expansion
f(t) = L (-l)ntZn, Iti < 1

n=O
(b) Use Watson's lemma to deduce the asymptotic expansion
F( ) - ~ ( -1r<2n)! IPI ~ oo
P f:o p2n+l '
ln Probs. 8-10, use the technique of Prob. 7 to derive the given asymptotic
formula.
~ -lt n!
l
oo e-px (
8. - -dx-L.J n+l
o 1+ X n=O p
00
-lt
9. j<
(""
0
e-px COS X dx- L 2n+i
(
n=O P
1(""
10. f(a) Jo e -px xa-1 (1 + x)c-a-1 dx
_ _!__ ~ ( -lt(a>n(l + a - c)n
LJ
p a n=O n.9 p n '
where (a)n = f(a + n)/f(a), n = 0,1,2, ....

ln Probs. 11-14, find the two-sided Laplace transform of the given function
and state the strip of convergence.
e"1 t<O 12 .ftt> = e -ltl
11. .f(/) = { a/
e . t>O
a<h
sin t
13. .f(t) = t- 14. f(t) =e-a't'
2
15. By integrating the function F + (p) = eP 214a along the imaginary axis
of the complex p-plane, find the inverse transform of this two-sided
Laplace transform.
5
Applications Involving
Laplace Transforms
5.1 lntroduction
Like the Fourier transform, the Laplace transform is used in a variety
ofapplications. Perhaps the most common usage ofthe Laplace transform
is in the solution of initial value problems. However, there are other
situations for which the properties of the Laplace transform are also
very useful, such as in the evaluation of certain integrais and in the
solution of certain integral equations. ln this chapter we will briefly
discuss applications of the Laplace transform in ali of the above named
areas.
5.2 Evaluating Integrais

An interesting application of Laplace transforms involves the evaluation
of certain integrais, particularly those containing a free parameter. ln
some cases we simply recognize the integral as a special case of a Laplace
transform for a particular value of the transform variable p. Other integrais
may be solved by first taking the Laplace transform of the integrand
with respect to a free parameter (not the variable of integration). The
resulting integral is hopefully easier to evaluate than the original, and
by applying the inverse Laplace transform we obtain our desired result.
This latter procedure is direct and often simple, but it requires the in-
terchange of two limit operations, so some caution should be exercised
in its usage.
218
5.2/Evaluating Integrais 219
Example 5.1: Evaluate the integrais
I= r= sin t dt
Jo t
J = fo
= _ sin t d
e 1-- t
t
Solution: Both of these integrais are special cases of known Laplace

transform integrais. For example, consider the Laplace transform
relation (recall Exam. 4.9 in Chap. 4)
w{sin t ;p } =
.z, - -
t o
1=
sin-tdt = tan _ 1 -1
e -pr -
t p
By setting p = O and p = 1 in this result, we obtain, respectively,
I= Ii{ si~ t ;p=O} = tan- 1
==~
J = Ii{ si~ 1
;p = 1} = tan -I 1 = ~
r= cos tx dx t> o
Jo x2 + 1
Solution: Let us define the integral by f(t) and take the Laplace
transform with respect to t. This action leads to
F(p) = L"" (xz + l)~xz + pz) dx

p
= pz - 1 Jo
r=(xz + 1 -1
xz + pz dx
1 )
= p2 ~ 1(~ -2:)
7T/2
=p+ 1'
and thus by taking the inverse Laplace transform, we obtain*
If- 1{F(p);t} = f(t) = (7T/2)e- 1
, t >O
* We might also recognize the integralf(t) as a multiple ofthe Fourier cosine transform
of the function l/(x 2 + 1).
220 Chap. 5/Applications Involving Laplace Transforms
EXERCISES 5.2
ln Probs. 1-10, use known Laplace transforms or transform properties
to evaluate the given integral.
1. L=te- 21cos t dt
oo -31 -61
3. Joo sinh t dt
-ao (
4.
i
o
e - e
t
dt
S. J: J0 (t) dt 6. LootJo(t) dt
7. r=cos 6t - cos 4tdt 8. L=e- 1erf(y/t) dt

Jo t
10. L=xe-x2erfc(x) dx
ln Probs. 11-16, use the technique illustrated in Exam. 5.2 to evaluate

the given integral.
i=Xsin txdx
11. L=e-lxz dx, t >o 12. O X +
2 1 ' t>O
13. Looexp(-x 2 rjx2 )dx, 14 ioosin txdX,

-
t >o
t> o O X
15. LOOS0Xdx, t>O 16. J~ooCOS tx 2

dx, t >o
ln Probs. 17-20, introduce a parameter t somewhere in the integrand

and then use the method ofExam. 5.2 to verify the given integral relation.
17. J"" -oo

2
sin x dx = J:!
2
18. r=
Jo
X COS x 3 dx = 1T
0
3 3f(l/3)
19 r=sin X dx = 1T
O< a< 1
. Jo xa 2f(a)sin 0!7T/2'
20 r=cos X dx = 7T O<a<1
Jo xa 2f(a)cos 0!7T /2'
5.3/Solution of ODEs 221
5.3 So/ution of ODEs

ln solving linear ordinary differential equations (ODEs) by the Laplace
transform method, we first convert the equation in the unknown function
y(t) into an equation in Y(p), and then if possible, solve for Y(p). The
inversion of Y(p) will then give the solution y(t) of the original ODE.
This technique clearly has advantages only if the equation for Y(p) is
easier to solve than the equation for y(t), and if Y(p) is also invertable.
ln the case of an equation with constant coefficients, the transformed
equation for Y(p) tums out to be an algebraic one, and the Laplace
transform method is therefore a powerful tool for solving this class of
ODEs.
Because the appearance of the forms y(O), y'(O), y"(O), and so on, in
the transforms of the derivatives of y(t), the Laplace transform method
is best suited to initial value problems, i.e., those where the auxiliary
conditions are all imposed at t == O. Furthermore, the solution arises in
the Laplace transform method with the initial conditions automatically
built into it, unlike the conventional approach where one constructs the
solution by adding a particular integral to the complementary function
and then imposes the auxiliary conditions on it.
Let us illustrate the Laplace transform technique on some typical
initial value problems.
Example 5.3: Solve the initial value problem

y(O) = 2, y'(O) = 6
Solution: By introducing the Laplace transforms
X{y(t);p} == Y(p)
J?{y'(t);p} == p Y(p) - y(O)
== pY(p) - 2
2
X{y"(t);p} == p Y(p) - py(O) - y'(O)
= p 2 Y(p) - 2p - 6
and
2{t2e3';p} = 2!/(p - 3) 3
we find that the given initial value problem is transformed into the
algebraic equation
[p 2 Y(p) - 2p - 6] - 6[pY(p) - 2] + 9Y(p) = 2/(p - 3) 3
or
(p 2 - 6p + 9)Y(p) = 2(p - 3) + 2/(p - W
222 Chap. 5/Applications lnvolving Laplace Transforms
Solving for Y(p), we obtain

2 2
Y(p) =P - 3 + (p - 3)5
and by taking the inverse Laplace transform, we have
y(t) = .~r'{Y(p);t} = :e-'{p ~ 3

;t} + ..ce-'Lp ~ 3>5;1}
y(t) = 2e 31 + (1/12)t4e31
y" + 4y = f(t), y(O) = 1, y'(O) = O
where
f(t) = {41,4, :S;t:S;l
t> 1
Solution: By first writing the forcing function f(t) in terms of the

Heaviside unit function, we have
f(t) = 4t[l - h(t - 1)] + 4h(t - 1)
= 4t - 4(t - l)h(t - 1)
the Laplace transform of which is (recall Theor. 4.5)
4 4
..ct'{f(t);p} = -p2 - -p2 e-P
ln this case the transformed initial value problem leads to

4 4
[p 2 Y(p) - p] + 4Y(p) = 2 - 2 e-p
p p
with solution
p 4 4 -p
Y(p) = l + 4 + pz(pz + 4) - pz(pz + 4) e
Using a partial fraction expansion on the last two terms on the right-
hand side of the above expression, we obtain
Y(p) = _P_
p2 +4
+ _.!.. - _1_ - (_.!.. -
p2 p2 + 4 p2 p2 + 4
_1_) e-p
the inversipn of which yields

y(t) = cos 2t + t - ! sin 2t - [(t - 1) - ! sin 2(t - 1)]h(t - 1)
The above examples illustrate the basic procedure used in the method
of Laplace transforms. And although these problems can be solved by
other techniques, the Laplace transform method offers the advantage of
solving the problem directly without first producing the general solution
of the DE. Moreover, in the case of Exam. 5.4 we found the solution
without splitting the problem into two problems, one over each interval
where f(t) is defined, as required by more conventional methods. Forcing
functions of this nature, as well as discontinuous or impulsive ones, are
commonplace in circuit analysis problems and in certain problems involving
mechanical vibrations.
While in general the Laplace transform method works best on constant-
coefficient equations, there are some variable-coefficient equations which
also lend themselves to the transform method. Consider the following
examples.

ty" + y' + ty = o, y(O) = 1, y'(O) =O
Solution: Upon taking the Laplace transform of the given equation,

we find
d 2 d
- dp[p Y(p) - p] + [pY(p) - 1] - dp Y(p) =O
which reduces to
dY
(p2 + 1) dp + py = o
Here we see that the transformed problem is another DE rather than
an algebraic equation. However, since the largest power of t occurring
in the given equation is unity, the transformed equation is a first-
order DE whereas the original DE was second order. The general
solution of this first-order linear DE is readily found to be
Y(p) == Ajy' p 2 +1
where A is an arbitrary constant. The inverse Laplace transform of
this result gives us
y(t) = AJ0 (t)
where J0 (t) is the Bessel function of arder zero. The initial conditions
require the choice A = 1.

y" + ty' + y = o, y(O) = 1, y'(O) =O
224 Chap. S/ Applications Involving Laplace Transforms
So/ution: The transformed problem becomes

2 d
[p Y(p) - p] - dp[pY(p) - 1] + Y(p) =O
or
dY
- - pY = -1
dp
The solution of this first-order linear DE is
Y(p) = ep2/2(A - J e-p2f2 dp)
= eP212 [A - y27T erf(pjy'2)]

where A is an arbitrary constant and erf(x) is the error function. To
determine the value of A, we use the initial value theorem (Theor.
4.14) which requires that
y(O) == 1 == lim p Y(p)
p .... oo
Hence, we find that A = yf2;,- and therefore

Y(p) = y21r eP 212 erfc(p/y2)
the inversion of which yields (recall Prob. 23 in Exer. 4.2)
y(t) = e-r212
Notice that in solving variable-coefficient DEs, the solution obtained

by the Laplace transform method involved arbitrary constants that had
to be resolved by use of the initial conditions. This is quite distinct from
solutions of constant-coefficient DEs obtained through the transform
method.
5.3.1 Impulse Response Function

To better understand the transform method and its relation to standard
solution techniques, let us apply the Laplace transform to the general
initial value problem
y" + ay' + by = f(t), y(O) = ko, y'(O) = kt (5.1)
where a and b are known constants. The coefficient of y" has been set
to unity for mathematical convenience. If we introduce the Laplace
transforms .P{y(t);p} = Y(p) and .P{f(t);p} = F(p), then (5.1) reduces
to the algebraic equation
[p 2 Y(p) - pko - kt1 + a[pY(p) - ko] + bY(p) == F(p)
or
(p2 + ap + b)Y(p) = (p + a)k0 + k1 + F(p) (5.2)
Solving for Y(p), we have
(p + a)k0 + k 1 F(p)
Y(p ) = + -=--;:__-
2 (5.3)
p2 + ap + b p + ap + b
and by taking the inverse Laplace transform, we obtain
y(t) = 2-lr~2++ a~~o: :

1
;t} + 2-l{ p2 :;:; + b ;t } (5.4)
~ ~
YH(t) yp(t)
Here it is interesting to observe that the solution (5.4) has naturally
split into two parts- the function YH(t), which is a solution of the initial
value problem
y" + ay' + by = O, y(O) = k0 , y'(O) = k 1 (5.5)
and yp(t), which satisfies
y" + ay' + by = f(t), y(O) = O, y'(O) = O (5.6)
We can physically interpret the function YH(t) as the response of the
system described by (5.1) entirely due to the initial conditions in the
absence of an external disturbance f(t). On the other hand, the function
yp(t) represents the response of the sarne system which is at rest until
time t = O, at which time it is subject to the external input f(t). By
separating the solution in this fashion we can use the Laplace transform
as an effective tool for analyzing the basic characteristics of a system
in response to each of the input parameters.
ln network analysis as well as other areas of application the analyst
is often interested only in the system response to an external stimulus
when the system is "at rest." That is the part of the solution above that
we have designated by yp(t). If we assume that YH(t) = O, then we can
represent the response of a system to the input f(t) by
y(t) = 2- 1{ F(p)
p 2 + ap + b
;t}
= J:g(t - u)f(u) du (5.7)
where we have used the convolution theorem and defined
g(t) =2 -1{ p 2 1
+ ap +
b ;t } (5.8)
226 Cbap. S/ Applications Involving Laplace Transforms
Equation (5.7) represents the response of any linear system, characterized

by y" + ay' + b, to the general input f(t). The function g(t) is called
the response function of the system. Expressed as g(t - u), it is also
called the one-sided Green 's function in much of the literature. *
Example 5.7: Construct the one-sided Green's function for the system
described by y" - 2y' + 5y = f(t).
Solution: By use of Eq. (5.8), we first construct the response function
g(t) = ..P-l{p2 _;p + 5;t} = ..P-J{(p- 1\2 + 4;t}

which yields
g(t) = !e 1
sin 2t
Hence, the one-sided Green's function is simply
g(t - u) = !er-u sin 2(t - u)
lf the forcing function to a system is the impulse function 8(t), then

Eq. (5.7) yields
y(t) = J:g(t - u)8(u) du = g(t) (5.9)
Hence, we see that the response function g(t) is actually the response
of the system to a unit impulse. For this reason, it is often called the
impulse response function of the system. According to (5.7), all other
solutions for general forcing functions f(t) are simply superpositions of
f(u) with the "fundamental solution" g(t - u).
Example 5.8: Use the impulse response function to finda general solution
of
y" - 2y' + 5y = f(t), y(O) = Yo, y'(O) = Vo
Solution: From Exam. 5.7, we know that the impulse response func-
tion is
g(t) =! e 1
sin 2t
Hence, based on Eq. (5.4) the general solution is given by
* See Chap. 2 in L. C. Andrews, Elementary Partia/ Differentia/ Equations with

Boundary Value Prob/ems, Orlando: Academic Press, 1986.
y(t) = :;e-'{YoP2 + (vo- 2Yo) ;t} +! f'l-"sin 2(t- u)f(u) du

p - 2p + 5 2 Jo
= :;e-'{Yo(P - 1)
(p -
+ (vo - Yo)
1l + 4
r} + ! Jo
' 2
f'e 1 -"sin 2(t - u)f(u) du
so that upon taking the inverse Laplace transform, we arrive at the

solution
y(t) = e1[y0cos 2t + !(vo - Yo)sin 2t]
+ !J:e -"sin2(t- 1
u)f(u)du
The above expression represents a solution to the problem for any

set of initial conditions and input function f(t).
EXERCISES 5.3
ln Probs. 1-15, use the Laplace transform to solve the given initial value
problem.
1. y" - y = e cos 1
t, y(O) = O, y'(O) = O
2. y" + 2y' + y == 3te- 1
, y(O) = 4, y'(O) == 2
3. y" - 4y' + 4y = t, y(O) = 1, y'(O) = O
4. y" - 3y' + 2y == 4e21 , y(O) = -3, y'(O) =5
S. y" + 2y' + 5y = e-r sin t, y(O) = O, y'(O) = 1
6. y"' - 3y" + 3y' - y = t2e 1
, y(O) = 1, y'(O) = O, y"(O) == -2
7. 2y"' + 3y" - 3y' - 2y = e- 1
, y(O) = O, y'(O) = O, y"(O) = 1
8. y"' - y" + 4y' - 4y = t, y(O) == O, y'(O) = O, y"(O) = 1
9. y" + 4y = f(t) == { ~~s 41 ~ ~ ~ :s 'IT y(O) == O, y'(O) = 1

10. y" + 4y = sin t - h(t - 2'1T)sin(t - 2'1T), y(O) = O, y'(O) = O
11. y" + ty' - y == O, y(O) = O, y'(O) =1
12. y" + aty' - 2ay = 1, y(O) = O, y'(O) = 1
13. ty" + (2t + 3)y' + (t + 3)y = 1
3e- , y(O) = O
14. ty" + (t - l)y' + y = O, y(O) = O

15. fy" - 2y = 2t, y(O) =2
ln Probs. 16-21, use the Laplace transform to construct the impulse
response function and the one-sided Green' s function for the given dif-
ferential operator M, where D = d/dt.
16. M =(D-a? 17. M = (D - a)(D - b),
a i= b
18. M = D2 + 5 19. M = D2 + 4D + 7
20. M = 4D2 - 8D +5 21. M = D2 - D - 2
22. The small motions y(t) of an undamped spring-mass system are gov-
emed by the initial value problem
my" + ky = f(t), y(O) = Yo. y'(O) = Vo
where m is the mass, k is the spring constant, andf(t) is an extemal
(driving) force. Show that the impulse response function of this system
is
g(t) = (1/w0}sin w0t,
23. Using the impulse response function given in Prob. 22, find the
response of the spring-mass system given that
(a} f(t) = P (constant)
(b} f(t) = P cos wt, w i= w0
(c) f(t) = P cos w0t
24. When resstve forces are taken into account for the spring-mass
system in Prob. 22, the motions are called damped. ln such cases
the goveming DE is modified to
my" + cy' + ky = f(t)
where c is a positive constant. Determine the impulse response function
for each of the following cases of damping:
(a) underdamped (c2 < 4mk)
(b) critically damped (c2 = 4mk)
(c) overdamped (c2 > 4mk}
25. Determine the impulse response function for each of the following
differential operators (D = d/ dt):
(a) M = Dn, n = 2,3,4, ...
(b) M = D 2(D2 - 1)
(c) M = D 4 - 1
(d) M = D3 - 6D2 + llD - 6
5.4/Solutions of PDEs 229
5.4 Solutions of PDEs

The Laplace transform is especially well-suited for solving initial-boundary
value problems for which some auxiliary conditions are prescribed at
t = O. Such problems arise naturally in the solution of the heat equation
and the wave equation, the independent variable t generally being in-
terpreted as the time variable. The Laplace transform, however, is not
generally appropriate in the solution of potential problems. For example,
the Laplace transform of uxx(x,y), x > O, leads to
;t{uxx(x,y);x ~ p} = p 2U(p,y) - pu(O,y) - u;r(O,y) (5.10)
To use this expression would require knowledge of both u and ux at the
boundary x = O. Yet, prescribing both u and ux on the boundary would
generally lead to an ill-posed problem (i.e., not usually solvable). ln heat
conduction problems this is equivalent to prescribing both the temperature
and heat flux at the boundary, which may not be compatible. Even if
either u or ux is left undetermined untillater in the problem, the problem
would almost surely become unwieldy at some point.
Since we have previously discussed physical situations leading to the
heat equation and wave equation (see Chap. 3), here we will simply
illustrate the solution technique of the Laplace transform.
5.4.1 Heat Conduction

Let us start by considering a very long homogeneous rod, one end of
which is exposed to a time-varying heat reservoir. If we assume the
initial temperature distribution is ooc along the rod, we have the math-
ematical problem described by*
O< X< ao, t >O
B.C.: u(O, t) = f(t), u(x,t) - O as x - oo (5.11)
By applying the Laplace transform to the PDE and boundary conditions,
we have
Uxx- {p/a2) U =O, O< X< oo
B.C.: U{O,p) = F(p), U(x,p)- O as x - oo (5.12)

where U(x,p) = ;t{u(x,t);t-p} and F(p) = 5t'{f(t);p}. The general solution
* The problem described by (5.11) can also be solved by applying the Fourier sine
transform to the variable x, but the Laplace transform is an easier tool to use in this case.
230 Chap. S/Applications Involving Laplace Transforms
of this second-order linear DE is

U(x,p) = A(p)exvp/a + B(p)e -xvp/a (5.13)
where A(p) and B(p) are arbitrary functions of p. However, to satisfy
the condition U(x,p) ~ O as x ~ ao, we must choose A(p) = O. The
remaining boundary condition demands that B(p) = F(p), and thus
U(x,p) = F(p)e-xvp/a (5.14)
From Exam. 4.26 in Chap. 4, we have that
.x-l{e-xvP/a.p~t} = X e-x2f4a2t (5.15)

' 2a\ITTt 312
and therefore, through use of the convolution theorem, we arrive at the
result
u(x,t)
X 1t
.. r o (t-T)3/2
= 2avTT
j(T)
An alternate form of the solut!Qgj_5.16) can be derived by making

the change of variable z = x/2a\lt - T, which leads to
ln particular, when the temperature at the end is given by f(t) = T0

(constant), we find that (5.17) reduces to
u(x,t) = T0 erfc(x/2a\lt) (5.18)
where erfc(x) is the complementary error function (see Sec. 1.3.1).
Suppose we now consider a homogeneous rod of unit length where
the initial temperature of the rod is zero. lt is assumed that the end x =
O is maintained at zero temperature while the end at x = 1 is kept at
constant temperature T0 The problem is characterized by
Uxx = a -2U1, O<x<1,t>O
B.C.: u(O,t) = O, u(l,t) = To (5.19)
I. C.: u(x,O) = O, O<x<1
Application of the Laplace transform to (5 .19) leads to the transformed
problem
Uxx- (p/a 2) U = O, O< x < 1
B.C.: U(O,p) =O, U(l,p) = T0 /p (5.20)
the general solution of which is*

U(x,p) == A(p)cosh(xy'pla) + B(p)sinh(xy'pla) (5.21)
Imposing the boundary conditions requires that A(p) == O and B(p) ==
ToiP sinh(Ypla); hence,
sinh(xy'p I a)
U(x p) - T.0 (5 22)
' - p sinh(Y pia)
ln order to invert (5.22), we use the method of residues from Sec.

4.6. The function U(x,p) has a simple pole at p == O and simple poles
at p == - n2,l-a2 , n = I ,2,3, .... Therefore, we find that
eP sinh(xyPia)
1
}
.;e- {U(x,p);p~t} == T0 [ Res {
1
. YP
p smh( pia)
;p == O
~
1
{eP sinh(xy'pla) ;p == 2 2 2} ]
+ LJ Res -n1ra
n= 1 p sinh(y'PI a)
2 ~ (- It . _n2,-2a21] t
u(x,t) == To [ X+- L.. - - sm(mrx)e (5.23)
1T n=I n
5.4.2 Mechanical Vibrations

Let u(x,t) denote the transverse displacement of a semiinfinite stretched
string, one end of which is fixed far out on the x axis and the other end
looped around the point x == O. The string is presumed to be at rest
initially with the looped end later moved in some prescribed manner
perpendicular to the x axis, i.e., u(O,t) == f(t). Because the string is
initially at rest, it follows that f(O) == O. If no extemal forces are acting
on the string, the above conditions are characterized by the boundary
value problem
0 <X< oo, t>0
B.C.: u(O, t) == f(t), u(x,t) ~Oas x ~ oo (5.24)
I. C.: u(x,O) ==O, ulx,O) == O, O < x < oo
* As a general rule, we use hyperbolic functions in the general solution when the
domain is finite and exponential functions [see (5.13)] when the domain is infinite.
t The standard method of solving this problem is by use of separation of variables,
e.g., see L. C. Andrews, Elementary Partia/ Differential Equations with Boundary V alue
Problems, Orlando: Academic Press, 1986.
232 Chap. 5I Applications Involving Laplace Transforms
The Laplace transform applied to (5.24) yields

Uxx - (p/c) 2 U = O, O< X< 00 (5.25)
B.C.: U(O, p) = F(p), U(x,p) ~O as x-? oo
the solution of which is
U(x,p) = F(p)e-xp/c (5.26)
The translation property of the Laplace transform (Theor. 4.5) enables
us to invert (5.26) directly, from which we get
t -.5 xf c (5.27)
u(x,t) = f(t - x/c)h(t - x/c) = {J(t _ x/c),
t >x/c
The interpretation of this solution is that a point on the string x units
from the origin remains at rest until time t = x/c, and then it executes
the sarne motion as the loop at the point x = O. ln other words, the
displacement which is imposed on the end x = O propagates down the
string with velocity c.
Suppose the semiinfinite string is now fixed at x = O and subject to
the extemal force f(x,t) = - fo3(t - xfv), which is a concentrated load
moving with speed v according to x = vt. If we assume the string is
initially at rest, the resulting motion of the string is govemed by
c2uxx = U11 + / 03(! - x/v), O <X < oo, t > O
B.C.: u(O,t) = O, u(x,t) ~ Oas x ~ oo (5.28)
I. C.: u(x,O) = O, u,(x,O) = O, O < x < oo
The transformed problem of (5.28) reads

Uxx - (p/c) 2 U = (fo/c2) e-xpfv, O< X< oo
(5.29)
B.C.: U(O,p)=O, U(x,p)~Oasx-oo
This is a nonhomogeneous ODE whose solution is readily found to be
fov2 (e -xpjv - e -xpfc) v ..J. c

2 2)2 T
U(x,p) = (
c - v P (5.30)
{ _fox e-xpfc v =c
2cp '
and by use of the translation property, the inversion of this expression
yields
v =I= c
v =c
(5.31)
For our final example involving mechanical vibrations, let us consider

a semiinfinite beam which is initially at rest along the x axis and then
at time t = O given a transverse displacement b at the end x = O. The
subsequent displacements are solutions of
Uxxxx + a- 2U = 0,
11 0 <X< oo, t > 0
B.C.: {"((O,t)) = b , uxx(O,t) = O (5.32)
u x,t ~ 0 asx~ oo
I.C.: u(x,O) = O, u1(x,O) = O, O< x < oo
By application of the Laplace transform, we obtain the transformed

problem
Uxxxx + (p/a) 2 U == O, O< X< oo
U(O,p) = b/p, UxiO,p) =O

B.C.: { U(x,p)~ Oasx~ oo (5.33)
The general solution of this fourth-order ODE is

U(x,p) = e-xvPti [A(p)cos(xyp/2a) + B(p)sin(x~)]
+ exvPti [C(p)cos(xyp/2a) + D(p)sin(xypf2a)] (5.34)
To satisfy the boundedness condition in (5.33), we must set C(p) =
D(p) = O. The remaining boundary conditions in (5.33) lead to A(p) =
bfp and B(p) = O; thus, our solution reduces to
U(x,p) = (b/p)e-xvPti cos(xyp/2a) (5.35)
Now let us use Euler's formula cos z = !(e;z + e-iz) so that we may
rewrite (5.35) as*
(5.36)
Then, using the inverse transform relation (see Prob. 14 in Exer. 4.6)
2- 1{(1/p)e-avP;t} = erfc(a/2Yt) (5.37)
we deduce that
u(x,t) = ~ [erre(~ J:/) + erfc (~ J~t) J

=b [1- 2 (x J-i)
1 erf 2 at
1 erf 2 -;;! - 2 (x Jl)J (5.38)
* Note that 1 ::t i = y'Zi.

234 Chap. S/ Applications Involving Laplace Transfonns
Finally, recalling Prob. 13 in Exer. 1.3, we obtain the solution in terms

of the Fresnel integrais, i.e.,
u(x,t) = b [1 - C(xjy2mrt) - S (x/y2mrt)] (5.39)
EXERCISES 5.4
1. lf the boundary condition in (5.11) is
u(O,t) =f(t) = {~: O<t+b

t?:. b
(a) show that the subsequent temperature distribution is
T1erfc(x/2ayt), O< t < b
u(x,t) = { T1[erf(x/2ayt - b) - erf(x/2ayt)], t?:. b
(b) Show that u(x,t) given in (a) is continuous at t = b.

2. Solve the problem described by (5.11) when u(O, t) = f(t)
Tt/Vt.
3. Given the heat conduction problem
Uxx = a -2U 1, 0 <X< oo, t>0
B.C.: ux(O, t) = - f(t), u(x,t) ~O as x ~ oo

show that its solution can be expressed as
u(x,t) ac_
= ~y7T l'
Oyt-r
f(r) exp [
~ r:--: -r Jdr
(
4a 2 t - r
)
4. Show that the solution in Prob. 3 can be expressed in the altemate

form
u(x,t) = ..V~-J=
7T x/2a
vf(t
t
- x2/4a 2z2)z- 2 e-z dz
2
5. For the special case f(t) = K (constant), show that the solution of
Prob. 3 is
u(x,t) = K [2ayt/7Te-x 2
/
4 2
ar - x erfc(x/2ayt)]
Hint: Use Prob. 4 and integration by parts.

6. A semiinfinite condricting solid has initial temperature T0 Radiation
into a medium x < Oat temperature zero is assumed to be such that
the ftux at the face x = O is proportional to the difference in tem-
peratures of the face x = O and the medium x < O. Given that the
mathematical formulation of the problem is
0 <X< oo, I> 0
B.C.: ux(O, t) = ku(O, t), u(x,t)-+ Oas x - oo
I. C.: u(x,O) = To, O<x<oo

show that the temperatures inside the conducting solid are given by
u(x,t)
= 2k10 i"" e_,,z (z cos xz + kzk sinxz) dz
1T o z z2 +
ln Probs. 7-10, use the Laplace transform to solve the given heat conduction
boundary-value problem.
7. Uxx = a -2U1 , O<x<l,t>O
B.C.: u(O,t) = T0 , Ux(l,t) = O
I. C.: u(x,O) =O
8. Uxx = a -2U1, O<x< l,t>O
B.C.: u(O,t) = O, u(l,t) = O
I. C.: u(x,O) = T0
9. 0.25 Uxx =U 1 - 1, 0 <X< 10, t > 0
B.C.: ux(O,t) =O, u(lO,t) = 20
I. C.: u(x,O) = 50
10. Uxx = U1 - 2x, O<x<l,t>O
B.C.: u(O,t) =O, u(l,t) = O
I.C.: u(x,O) = x(l - x)
Uxx = Uro O<x<I,t>O
B.C.: u(O,t) = O, u(l,t) = T0
I.C.: u(x,O) = T0 , O<x<l
(a) show that the solution of the transformed problem can be expressed
in the form
T0 {
U(x,p) = - 1 - e-vp
[ 1 - e
V
-Z(l-x)Vp]}
p 1 - e- 2 P
(b) By expanding (1 - e- 2v'P)- 1 in a series of ascending powers of

e- 2V'P, show that
U(x,p) =; [1- e-xv'P + e-<2 -x)vp- e-<2+x)vp + .. ]
(c) Inverting the series in (b) termwise, deduce that
u(x,t) = T0 {erf(x/2\/'t) + erfc[(2- x)/2\/'t]
- erfc[(2+x)/2yt] + }
12. A heat source of strength q(t)h(t), where h is the Heaviside unit
function, appears at the origin of a long rod at time t = Oand moves
along the positive x axis with constant speed v. The problem is
characterized by
Uxx = u, - (x - vt)q(t)h(t), -oo<x<oo,t>O
B.C.: u(x,t)- Oas lxl - oo
I.C.: u(x,O) = O, -oo <x< oo
Using the Laplace transform, show that
u(x,t) = -1- ft q(T)(t - T)- 112 e-<x- dT

zv; o 1JT)ZJ4(t- T)
13. Solve the heat conduction problem

Uxx = a - 2 u1 - (x)(t), - oo <x< oo, t>O
B.C.: u(x,t) -O as lxl - oo
I.C.: u(x,O) = T0
14. Using the Laplace transform, finda bounded solution of the exterior
temperature distribution problem for a sphere described by
u,, + (2/r)u, = u,, 1 <r< oo, t >O
B.C.: u(l,t) = T1, u(r,t) - O as r - oo
I.C.: u(r,O) = T0
15. The temperature distribution u(r,t) in a thin circular plate, which is
initially at 0C, has its faces insulated and its boundary held at
temperature T1, is govemed by the boundary value problem
u, + (1/r)u, = u,, O< r< 1, t >O
B.C.: u(l,t) = T1
I.C.: u(r,O) = O
Use the Laplace transform to find a bounded solution.
Hint: Recall Prob. 13 in Exer. 4.6.
16. Given the boundary value problem

0 <X< oo, t >0
B.C.: u_.(O,t) = f(t), u(x,t) ~O as x ~ oo
I.C.: u(x,O) = O, u,(x,O) =O
show that u(x,t) = g(t - x), where g(z) = O, z > O, and
g(z) =- Lf(T) dT, z 2: O

17. Consider the motions of a string fastened at the origin but whose
far end is looped around a frictionless peg that exerts no vertical
force on the loop. The string is initially supported at rest along the
x axis and is released at time t = O, moving downward under the
action of gravity. Determine the subsequent displacements given that
the problem is characterized by
O< x < oo, t >O (g constant)
B.C.: u(O,t) =O, u(x,t) ~O as x ~ oo
I. C.: u(x,O) = O, u,(x,O) =O
18. Show that the boundary-value problem
Uxx = Um 0 <X< oo, t >0
B.C.: u..(O,t) =O, u(x,t) ~O as x ~ oo
I. C.: u(x,O) = e-x, ur(x,O) =O
has the solution
e- 1 coshx, x<t
u(x t) = { e-"cosht,
' x>t.
ln Probs. 19-25, use the Laplace transform to solve the given boundary-
value problem.
-2
19. Uxx = C U 11 , 0 <X< oo, t >0
B.C.: u(O,t) = O, ux(x,t) ~O as x ~ oo
I. C.: u(x,O) = O, u,(x,O) = Vo
20. -2
Uxx = C Um 0 <X< oo, t >0
B.C.: u(O,t) =O, u(x,t) ~ O as x ~ oo
I. C.: u(x,O) =A, u,(x,O) =O

21. uxx = c- 2U 11 -A sin 1TX, O<x<l,t>O
B.C.: u(O,t) =O, u(l,t) =O
I. C.: u(x,O) =O, u,(x,O) =O
22. -2
Uxx = C Um O<x<1,t>O
B.C.: u(O,t) = O, u(l,t) = 1
I. C.: u(x,O) =O, u,(x,O) = O
-2
23. Uxx = C Um O<x< 1,t>O
B.C.: Ux(O,t) =O, u(l,t) = 1
I. C.: u(x,O) = O, ub,O) =O
-2
24. Uxx =C Um O<x<l,t>O
B.C.: u(O,t) = O, ux(l,t) =Ar
I. C.: u(x,O) =O, u,(x,O) = O

-2
25. Uxx =C Um O<x<1,t>O
B.C.: u(O,t) = O, ux(l,t) = O
I. C.: u(x,O) =O, u,(x,O) = x
5.5 Linear Integral Equations

Integral equatons of the form
J:u(r)k(t - r) dr = f(t), t> o (5.40)
are known as V o/terra equations of convolution type (see also the discussion
in Sec. 3.2.1). The Laplace transform provides a useful technique for
the solution of such equations in which f(t) and k(t - r) are known
functions and u(t) is to be determined.
Using the result of Eq. (4.32) in Sec. 4.5.3 to take the Laplace transform
of (5.40), we obtain
U(p )K(p) = F(p) (5.41)
from which it follows that
U(p) = F(p)/K(p) (5.42)
Inverting (5.42) leads to the solution
u(t) = oT- 1 {:~~;t} (5.43)
If the function 1/K(p) has an inverse Laplace transform, say
2-
1
{ K:p) ;t} = g(t) (5.44)
5.5/Linear Integral Equations 239
then we can use the convolution theorem (Theorem 4.12) to express the
solution (5.43) as
u(t) = L f(T)g(t - T) dT, t >o (5.45)
ln some cases it may happen that the inverse transform (5.44) does
not exist, but 1/pK(p) has an inverse transform. Thus, (5.42) becomes
U(p) = pF(p) L(p) (5.46)
where L(p) = 1/pK(p). Now if f(t) is a differentiable function such that
f(O) = O, then
.:f- 1{pF(p);t} = f'(t) (5.47)
and the inversion of (5.46) yields
u(t) = L f'(T)e(t - T) dT, t >o (5.48)
where e(t) = .:f- 1{L(p);t}.
J: U(T)J (t -
0 T) dT = Sn t, t >0
where J0(t) is the Bessel function.
Solution: Recalling that

.:t'{J0(t);p} = 1/Vp2 + 1
we find that the Laplace transform applied to the integral equation
leads to
U(p)jy'p 2 + 1 = 1/(p2 + 1)
The solution of this transformed problem is

U(p) = 1/y'p2 + 1
and therefore we deduce that
u(t) = J 0(t)
By substituting u(t) = J 0(t) into the original integral equation, we
get the interesting integral formula
L lo(T)J0(t - T) dT = sin t
u(t) - J: e 1
-T u('r) dr = f(t), t >O
Solution: This is an integral equation of the second kind (see Sec.

3.2.1). By taking the Laplace transform of each term in the equation,
we find
U(p)
U(p)- - - = F(p)
p- 1
which leads to
U(p) = (p - l)F(p) = F(p) + F(p)

p-2 p-2
Now taking the inverse transform of this last expression, we get the
formal solution
u(t) = f(t) + L e2<t--T> /(r) dr
5.5.1 The Tautochrone Problem

The great Norwegian mathematician Niels Abel (1802-1829) studied a
particular integral equation of the Volterra type which has severa! important
applications. The most famous application is the tautochrone problem,
which is to determine a curve passing through the origin for which the
time required for a particle to slide down the curve is independent of
the starting point. The particle is allowed to slide freely from rest under
the action of gravity and the reaction of the curve on which it is constrained
to move (see Fig. 5.1). It is this curve that is called the tautochrone.
If we assume the particle is initially at rest at P(x,y), then its kinetic
energy is zero and its potential energy is mgy, where m is the mass of
P(x,y)
Figure 5.1 Tautochrone

the particle and g is the gravitational constant. At some intermediate

point Q(, TJ), in accordance with the conservation of total energy, we
can equate the gain in kinetic energy to the loss of potential energy,
which leads to
imv 2 = mg(y- TJ) (5.49)
where v is the instantaneous speed of the particle. Solving for v, we
obtain
v = ds/dt = y2g(y - TJ)
where s denotes the arclength along the curve. The time of travei from
P to Q is thus given by the expression
1 J.Q ds
t =- (5.50)
\(ii p yy:::-;,
From the calculus we have the arclength relation
ds = -y1 + (d/dTJ) 2 d11 = -u(TJ)dTJ (5.51)
where the negative sigo reftects the fact that 11 is a decreasing variable
from P to Q. Substituting this last expression into (5.50) yields
t = __1_ f11 U(TJ) dTJ

\(ii y yy:::-;,
and therefore the total time of descent T from P to the origin is given
by the Volterra integral equation
T = 1 (Y U(TJ) dTJ (5.52)
y2g Jo yy:::-;,
where T is a fixed constant.
To solve (5.52) for the unknown function u(y), we apply the Laplace
transform to the variable y, which gives us
T\12gfp = VTT/p U(p)
or
U(p) = Ty2g/TTp (5.53)
The inverse Laplace transform of this expression is
TJ2g
u(y) =- - (5.54)
1T y
Rewriting Eq. (5.51) in terms of x and y, and substituting (5.54) into the
square of the resulting expression leads to
242 Chap. 5/ Applications Involving Laplace Transforms
~
2 2
1 + (dx) == 2gT ==
dy TTzY y
where a = 2gT2/TT2 Separating variables yields

a=y
dx ==
J -y-dy
and by substituting y == a sin2l0, this last expression simplifies to

dx == a cos 2l0 dO = (a/2)(1 + cos O)d() (5.55)
Observing that x = O when y = O, the solution of (5.55) leads to the
set of parametric equations for the tautochrone
x = (a/2)(0 + sinO), y == (a/2)(1 - cos O) (5.56)
which represent a curve called a cycloid. This sarne curve is generated
by a point P on a circle of radius a/2 as the circle rolls along the lower
side of the line y = a.
Abel's integral equation (5.52) is a special case of the more general
integral equation
U(T)
f
t
O ( t - T)
a dT = J(t), t >o (5.57)
where f(t) is given and a is a constant such that O < a < 1. If we formally
apply the Laplace transform to (5.57), we get
f(1 - a)pa- 1U(p) = F(p)
or
F(p) pF(p)
= f(l (5.58)
U(p) - a)pa ro - a)pa
Assuming that f(t) is differentiable and that f(O) = O, then

2- 1{pF(p);t} = f'(t) (5.59)
Also, we have that
~ a)p";t} == -r(-1-~-a-~-)f-(a-)
1
2 -l{ro
TTasin
==--(a- 1
(5.60)
1T
and thus we deduce that
( ) = sin TTa
U(
1T
1 1
(
O t-T
f'(T)
)l-a
d
T, t >o (5.61)
EXERCISES 5.5
ln Probs. 1-10, we finda continuous solution ofthe given integral equation.
1. r u(r)
Jo y!l--;
dr = ,fi 2. r vu(r)
Jo t - r
dr = 1 + t + 3t 2
3. J:u(r)u(t - r) dr = 4 sin 2t 4. u(t) = t + J:(t - r)u(r) dr
S. u(t) = 4t 6. u(t) + J:e-r u(t - r) dr = 1

- 3 J:u(r)sin(t - r) dr
7. u(t) = a sin t - 2 J: u(r) cos (t - r) dr
8. u(t) = t + l J: (t - du(r) dr
9.
,i(
u(r)
O t- r
) 113 dr = t(l + t)
10. u(t) = l sin 2t + J: u(r)u(t - r) dr
ln Probs. 11-15, solve the given integrodifferential equation.
11. J:u(r)cos(t - r) dr = u'(t), u(O) =1
12. J:u(r)cos 2(t - r) dr = 1 - u'(t), u(O) =3
13. {u'(r)u(t- r) dr= 24t3 , u(O) =O
14. J:u"(r)u'(t - r) dr = u'(t) - u(t), u(O) = O, u'(O) = O
15. f(t) = L(t- r)-" u'(r) dr, O<a<1
16. Show that

u(t) + J:u(r)sin w(t - r) = f(t)
*
u(t) = f(t) - J:f(r)sin O(t - r) dr, 0 2 = w(l + w)

244 Chap. S/ Applications Involving Laplace Transforms
17. Show that
f(t) =L l0(2~)u(T) dT, f(O) =O
u(t) = f'(0)/0(2yt) + J: 10(2~)/(T) dT

18. Solve the integral equation
f>n<t- T)U(T) dT = Jn+lt)
(a) for the case n = O.

(b) for the cases n = 1,2,3, ...
19. Show that
U(T)
u(t) + l vt-
t
o
;-:---_
~
T
dT = f(t), u(O) = /(0) = O
has a formal solution of the form
u(t) = Lr(t - T)g(T) dT
and identify the function g(t).

ln Probs. 20-22, find formal solutions of the given integral equations.
20. f(t) = i (
r
o t2
u(s)
- s2
)"ds, O< a < 1
21. f(t) = 2 J,
e su(s)
~ds
22. f(t) =t
),
("" ~
s- t
ds
6
The Mellin Transform
6.1 Introduction
Generally speaking, unlike the Fourier and Laplace transforms, we find
that the Mellin transform is not very useful in a direct manner. It is
quite effective, however, in the derivation of certain properties of integrais,
in summing series, and in statistics. ln this sense, we generally think of
the Mellin transform as a sort of indirect tool in applications.
As in the case of the Laplace transform, the Mellin transform and
its inversion formula can be formally derived from the Fourier integral
theorem
g(u) = -1 Joo e-tu

. Joo g(t)e't'. dt de (6.1)
211" -~ -~
Let us begin by introducing the change of variables x = e', y = eu, and

s = c + ;e, where c is a fixed constant. Then, after some algebraic
manipulation, we find
c+ioo
g(logy)y-c = ~i
2 f
c-oo
y- L""g(logx)x-cx'- 1 dxds (6.2)
If we now define
f(x) = g(log x )x -c (6.3)
245
246 Chap. 6/The Mellin Transform
then (6.2) leads to the pair of transform formulas
F(s) = L""xs-l f(x) dx (6.4)
and
c+ioo
f(x) = 27Ti
1 Jx -s F(s) ds (6.5)
c-ioo
We define (6.4) as the Meflin transform of f(x), and (6.5) is the related
inversion formula. We also use the notations
F(s) = .M{f(x);s} (6.6)
and
f(x) = .,u- 1{F(s);x} (6.7)
respectively, for the Mellin transform and its inverse.
H the integrais
both converge for real a and b such that a< b, then the Mellin transform
of f(x) converges uniformly to F(s) in any finite region interior to the
infinite vertical strip a O, and if F(s) is defined
by (6.4), then the inversion formula (6.5) is valid for c > k.
6.2 Evaluation of Mellin Transforms

ln this section we will calculate the Mellin transform of several functions
and develop certain operational properties. ln many cases we find that
we can relate the desired Mellin transform to known results involving
either the Fourier or Laplace transform.
Example 6.1: Find the Mellin transform of e-x.
Solution: From definition,
.M{e-x;s} = L""x-le-x dx
6.2/Evaluation of Mellin Transforms 247
but this is precisely the definition of the gamma function. Hence, we

immediately deduce that
Re(s) >O
Observe that, in Exam. 6.1, we could also relate the Mellin transform
of e-x to the Laplace transform of xx-1, i.e.,
(6.8)
Example 6.2: Find the Mellin transforms of cos x and sin x.
Solution: Starting with cos x, we have
At{cos x;s} = f"

x'- 1cos x dx
but the integral on the right is recognized as a multiple of a special

case of the cosine transform
f:Fc{x'- 1 ;} = J~f'x-l cos x dx
= J~ f(s) cos (11rs) O< Re(s) <I

7T ' 2 '
[recall Eq. (2.68a) in Sec. 2.6.1]. Therefore we deduce that ( = 1)

At{cos x;s} = f(s)cos(!7Ts ), O < Re(s) < 1
Similarly, the result
At{sin x;s} = f(s )sin(!7Ts ), O< Re(s) < 1
follows directly from the sine transform given by Eq. (2.68b) in Sec.
2.6.1.
Example 6.3 Find the Mellin transform of 1/(1 + x).
Solution: From the defining integral,

1 }
At { - - ;s =
l~-x'-1
-dx
1+x o1+x
By imposing the restriction O < Re(s) < 1, the integral is exactly
that given by Eq. (1.13) in Sec. 1.2.2. Hence, it follows that
AtL ~ x ;s} = f(s)f(l - s), O< Re(s) < 1

248 Chap. 6/The Mellin Transfonn
Also, by using properties of the gamma function, we can write this

transform in the more convenient form
1 s}
At { - - = -Tr- O< Re(s) < 1
1+x ' sin 1rs '
6.2.1 Operational Properties

The Mellin transform enjoys certain operational properties analogous to
those of the Fourier and Laplace transforms. For example, if C, and C2
are any constants, then we have the linearity property
At{C,f(x) + C26'(x);s} = C 1F(s) + C2G(s) (6.9)
where F(s) and G(s) are the Mellin transforms, respectively, off(x) and
g(x). This property is a simple consequence of the linearity property of
integrais.
By making the simple change of variable x = t/a in the following
integral
At{f(ax);s} = Looxs-J f(ax) dx
=~[r-'
a o
J(t) dt
we deduce the scaling property

At{f(ax);s} = (1/as)F(s), a>O (6.10)
where F(s) is the Mellin transform of f(x). Also, we find that
Al{xaf(x);s} = Loor+a- f(x) dx

which leads to the translation property
Al{xaf(x);s} = F(s + a) (6.11)
Similarly, it can be shown that
At{f(xa);s} = (1/a)F(s/a), a>O (6.12)
and
At{(l/x)f(l/x);s} = F(l - s) (6.13)
the proofs of which are left to the exercises (see Probs. 16 and 17 in
Exer. 6.2).
lff(x) is continuous on x 2:: O and has a Mellin transform F(s), then
6.2/Evallllllion of Mellin Transforms 249
..ttl{f'(x);s} = L~xs-l f'(x) dx
= x- 1f(x)~~- (s- 1) L~x.- 2 f(x)dx

Now, if it happens that u 1 and u 2 exist such that
lim x'- 1f(x) = O, <T 1 < Re(s) < u 2 (6.14a)
x-+0
limr- 1/(x) = O, <Tt < Re(s) < <Tz (6.14b)
then the Mellin transform of f'(x) leads to

..ttl{f'(x);s} = -(s - 1)F(s - 1), <Tt < Re(s) < u 2 (6.15)
provided F(s - 1) exists in the stated vertical strip of the s plane. A
second application of (6.15) yields
..tU{f"(x);s} = -(s- l)..ttl{f'(x);s- 1}
or
..tU{f"(x);s} = (s - 1)(s - 2)F(s - 2) (6.16)
Continued application of (6.15) eventually leads to the general result
f(s)
.M{j<n>(x);s} = ( -1r F(s - n), n = 1,2,3, ... (6.17)
r (s - n)
provided
limx-k-tf<k>(x) = O, k = 1,2, , n-1 (6.18)
x-+0
The convolution theorem for the Mellin transform is given by
.M- 1{F(s)G(s);x} = i
=
o
d
f(x/u)g(u) __!!_
u
(6.19)
the derivation of which we leave to the exercises (see Prob. 25 in Exer.

6.2). Additional properties of the Mellin transform are also provided in
the exercises.
Example 6.4: Find the Mellin transform of l/(1 + ax)m, m > O.
Solution: Let us first find the Mellin transform of f(x) =

1I (l + x )m. From the defining integral
.M{f(x);s} = L~(l ~-~)m dx

but recalling the definition of the beta function (see Prob. 21 in Exer.
1.2)
B(x,y) = Lao(I :x~;x+y dt, x>O,y >O
we see immediately that

f(s)f(m - s)
Al{f(x);s} = B(s, m - s) = f(m)
Now using the scaling property (6.10), we have

1 } f(s)f(m - s)
Al { (1 + axr;s = aSf(m) ' o< Re(s) < m
Examp/e 6.5: Find the Mellin transform of x-vlv(ax), a> O, v> -1/2.
Solution: ln this case the Mellin transform
Al{x-vlv(ax);s} = Laoxs-v-Ilv(ax) dx
can be evaluated most easily in an indirect manner using known

results from the Fourier cosine transform. First we recall the cosine
transforms
and
~c{xs-I ;} = y2/7T f(s)cos(!'TTs)g-s
[see Eqs. (2.66) and (2.68a) in Sec. 2.6]. Then, using the cosine
transform relation
Laof(x)g(x) dx = LaoFc()Gc() d
[see Eq. (2.90) in Sec. 2.7], we obtain
(ao s-v-IJ ( ) d = J~ f(s) COS(!7TS) ra cs( 2 _ ;;2)v-l/2 di:

Jo X v ax X 7T 2v-l/2a"f(v + 1/2) Jo ~ a ~ ~
However, if O < Re(s) < l, then

1
ra-s(a2- gzy-1/2 d = !a2v-s f r<s+l)/2(1 - t)v-1/2 dt
Jo 2 Jo
l 2v-s r(! - !s)f(v + !)
=-a
2 f(v - is + 1) '
6.2/Evalua.tion of Mellin Transforms 251
from which it follows that
io
oo
- -1
xs v J(ax)dx =a
v
v-sf(s)f(!- ls)cos(i7Ts)
v1Trf(v -is+ 1)
The integral on the left is the Mellin transform we seek. By use of
properties of the gamma function, we can write
1 1
f(s )f(z - zs )cos(z1TS)
1
= f(7Tf(S)
2 + zS
1 = ~C
1v 1T 2s-t f(lzS )
)
and hence we deduce that

-v av-s2s-v-l f(lzS ) a> O, v> -1/2
Al{x J.,(ax)~s} = f( ) ,
1
v- zS +1
EXERCISES 6.2
ln Probs. 1-15, evaluate the Mellin transform of the given function.
When possible, use known integral results from previous chapters.
1. f(x) = h(a - x), a>O 2. f(x) = e-hx, b>O
3. f(x) = xae-bX, a,b >O 4. f(x) = e-b'x\ b >O
S. f(x) = log(b/x)h(x- 1), b >O 6. f(x) = x" log(b/x)h(x- 1),
a,b > O
7. f(x) = x"/(1 + xt, 8. f(x) == x"(l - x)h- 1h(l - x),
a,b >O a>O
[y?+t- x]"
9 f( ) = h(t - x) a>O 10. f(x) = .. 12"":-: ,
x (x- 1)", vx2 + 1
a>O
Hint: Let x = b/W+J..
I
11. f(x) = - --a, a> O 12. f(x) = 1/(l + x")m,
1 +x a,m >O
13. f(x) = lv(x), V > - l/2 14. f(x) = Jv(V-;;), v> -l/2
15. f(x) = Jix2), v> -l/2
Hint: ln Probs. 14 and 15, use the result of Prob. 13 and Eq. (6.12).
ln Probs. 16-28, verify the given operational property of the Mellin
transform.
16. Al{f(x");s} = (l/a) F(s/a), 17. Al{(l/x)f(l/x);s} = F(l - s)
a> O
18. Al{(log x)f(x);s} = F'(s) 19. Al{(x d/dx)f(x);s} = -s F(s)

20. Al{(x d/dxtf(x);s} = (-stF(s), n = 1,2,3, ...
21. Al{xnj<n>(x);s} = ( -1tf(~(:) n) F(s), n = 1,2,3, ...
22. Al{rf"(x) + xf'(x);s} = s2F(s)
23 .,u{J:f(u) du;s} = -(1/s) F(s + 1)
24. Al{Loof(u) du;s} = (1/s) F(s + 1)
25 .,u-'{F(s)G(s);x} = Loof(x/u)g(u) du/u
1 Jc+ioo
26. Al{f(x)g(x);s} = 1T'i c-ioo F(p)G(s- p) dp
2
27 .,U{Loou"'.f(u)g(xu) du;s} = F(l +m - s)G(s)
28 .,U{Looumf(u)g(x/u) du;s} = F(l + m + s)G(s)
ln Probs. 29-33, verify the given Mellin transform relation.
29 .,u-{r(s)F(l - s);x} = ~{f(t);x}

1
30.4C {cos(!1T's)f(s)F(l - s);x} = y:;;j2 ~c{f(t);x}
31. .tr'{sin(!1rs)f(s)F(l - s);x} = v:;;;2 ~slf(t);x}
32 .At{e-xcos.Pcos(x sin <f>);s} = f(s )cos(s<f>), -1r /2 < <f> < 'TT'/2, Re(s) > O
33. At{e-xcos.Psin(x sin <f>);s} = f(s)sin(s<f>), -'TT'/2 < <f> < 'TT'/2, Re(s) > -1
34. Starting with the Dirichlet series
00
g(s) = "' an
LJ--:;
n=l n
and the integral representation

6.2/Evaluation of MeUin Transforms 253
(a) show, by summing over all positive integers n, that

g(s )f(s) = Looxs-l f(x) dx
where
n=l
(b) From (a), deduce the inverse transform relation

1 fc+ioo
f(x) = - . . x-s g(s)f(s) ds
21'fl c-100
35. Show that
At{ex ~ 1
;s} = '(s )f(s)
where '(s) is the Riemann zeta function defined by
I 00
'(s) = 2: -..
Re(s) > 1
n=ln
Hint: See Prob. 34.

36. Using Prob. 35, show that
(a) At{ ex ~ 1 ;s} - At{ ~ ~ 1 ;s} = zl-s '(s)f(s)
(b) At{ ex ~ 1 ;s} = (1 - 2(-sms)
37. Starting with the Mellin transform relation
At { -1 -+1 x's } 1'(

=- -
sin 1rs
use Prob. 23 to deduce that
At{log(l + x);s} = 1rj(s sin 7TS)
38. Starting with the Mellin transform relation
1
At{-- 2
1+x '
s} = --7T...,.--
2 sin(!7Ts)
use Prob. 24 to deduce that
At{ (~- tan-\);s} = 2s co:(!7Ts)

6.3 Complex Variable Methods

Up to this point we have evaluated Mellin transforms primarily by relating
the transform integral to known results involving other integral transforms.
A more direct evaluation of the Mellin transform may sometimes be
accomplished by complex variable methods, particularly when the function
f(x) is a rational function. Also, since the inversion formula ofthe Mellin
transform is formulated in the complex plane, it likewise lends itself to
contour integration techniques similar to those used in evaluating inverse
Laplace transforms. ln this section we will briefty illustrate the method
of residues as it applies to both the Mellin transform and its inversion
formula.
6.3.1 Me/lin Transforms

Suppose that f(z) is a rational function having no poles on the positive
real axis. Suppose further that there exists real constants <T 1 and O"z such
that
lim z"f(z) = O, O"t < Re(s) < O"z
z-->0 (6.20)
limt'f(z) = O, O"t < Re(s) < O"z
lzl--->=
If f(z) has N poles at z = at> az, , aN, then we begin by integrating
the function z- t_{(z) around the contour shown in Fig. 6.1, which encloses
all poles of f(z). Based on the Residue Theorem (see Appendix A)
J. z-t f(z) dz = 2?Ti Res{z-t_r(z);ak}

Jc k=t
or
rRXs-t_r(x)dx + r Zs-t_r(z)dz + rp(Xez,.;y-t_r(x)dX + r z-t_r(z)dz

JP JcR JR Jcp
N
= 2?TiL Res{z-t_{(z);ak}
k=l (6.21)
Because of the limit relations (6.20), it can be shown that the integral
around the large circle of radius R vanishes in the limitas R --+ =, as
does the integral around the small circle of radius p as p --+ O. Hence,
in the limit as R --+ = and p --+ O, we obtain
(6.22)
6.3/Complex Variable Methods 255
However,
and thus (6.22) becomes
(o""'x-l f(x) dx = - ~e-i"'s ~ Res{z-t_{(z);ak} (6.23)

J< sm 'Tf'S k=l
Eq. (6.23) can be easily modified to include the case wheref(z) has poles
on the positive real axis (see Prob. 8 in Exer. 6.3).
Example 6.6: Find the Mellin transform of 1/(1 + x2).
Solution: We first note that conditions (6.20) hold if O < Re(s)

< 2. Also, the poles of f(z) = 1/(1 + i) are simple poles at z =
i, and the residues at these poles are
s-1 } s-1 1
Res _z__ .i = _z- = - - ei"'s/2
{ l+l' 2i 2
and
where we have observed that i= e;"'12 and -i = e3;"'12 From (6.23),

256 Chap. 6/The MeUin Transform
it now follows that

s-1 -i1rs 1
l
oo
_x__ dx = ~ . -(ei'"s/2 + e3i'ITs/2)
o 1 + x2 sin 7TS 2
= __;!!.--. !(e-i'lrs/2 + ei'ITs/2)
sm 7TS 2
7T cos (7Ts/2)
= sin 7TS
One further simplification leads to our final result
..L~ x2;s} = 2sin~7Ts/2)' O<Re(s)<2
Observe also that we could obtain this result from Exam. 6.3 and
the operational property given by Eq. (6.12).
6.3.2 Inverse Mellin Transforms

ln Exam. 6.1 we obtained the Mellin transform
Re(s) >O (6.24)
To formally recover the function e-x from the transform function f(s),
we consider the inversion formula
c+ioo
..- {f(s);x} = 2~; Jx- f(s) ds

1
c-ioo
(6.25)
Ali of the poles of f(s) occur in the left-half plane, viz., at s = -n

(n = 0,1,2, ... ). Thus, since f(s) is analytic in the right-halfplane, we
can evaluate (6.25) by considering the sarne kind of contour used in
evaluating inverse Laplace transforms [see Fig. 6.2(a)]. Dueto the asymp-
totic behavior of f(s) for large Is I, it can be shown that the integral along
the large circular are tends to zero for ali x as the radius of the circular
are tends to infinity. Based on the calculus of residues, we then conclude
that
..- {f(s);x} = 2:Res{x-r(s);-n}

1
n=O
(6.26)
Now, by writing [recall property (G9) in Sec. 1.2.2]

7T
f(s) = . (6.27)
f(l - s )sm 7TS
lm(s) lm(s)
c + i
c + i
Re(s) Re(s)
----+------+----~--
c - i
c - i
(a) (b)
Figure 6.2 Contours of integration
we find
Res{x-f(s); -n} = f(l

x-
)
I =
( n'.-l)nx"
- SCOS1TS s=-n
and therefore it follows from (6.26) that
.J.l-l{f(s);x} = L"" ( -l)nxn = e-x (6.28)

n=O n!
ln general, the result we obtain by enclosing the poles of F(s) in the
left-half plane by a contour as shown in Fig. 6.2(a) will be restiicted to
the interval O < x < x 0 where x 0 is determined such that the integral
along the circular are vanishes in the limit as the radius tends to infinity.
If the number of poles along the negative real axis is infinite, this will
lead to an ascending series in x like that in (6.28). ln some cases the
function F(s) may also have poles along the positive real axis which we
then enclose by a contour as shown in Fig. 6.2(b). This time the result
obtained will be valid for x > x 0 , and will be a descending series in x.
Example 6.7: Use the method of residues to find the inverse Mellin
transform of
F( ) = f(s)f(m - s) a>O,O<Re(s)<m
s ar(m) '
Solution: The function F(s) has two sets of poles, viz., at s = -n

(n = 0,1,2, ... ) and at s = m + n (n = 0,1,2, ... ). Thus, to find the
inverse Mellin transform we must apply the residue theory to both

contours shown in Fig. 6.2.
By first choosing a contour as shown in Fig. 6.2(a), we have that
.M- 1{F(s);x} = :LRes{x-F(s);-n}

n=O
for some restricted values of x to be determined. Using Eq. (6.27),

the residue at s = - n is found to be
R { -Fi( )
es x s ;- n
}
= ax-r(m - s)
r(m )'f(l - S ) COS 7TS
I
s= -n
_ ( -ltf(m + n)( )n
- n!f(m) ax
However, using the identity (see Prob. lO in Exer. 1.2)
(-1rr(m +n) = (- 1r(m + n- 1) = (-m)

n!f(m) n n
we get the ascending binomial series
1
_,u- {F(s);x} =~o( -nm) (axt
It is known that this series converges for Jaxl < I,* and hence we
have
_,u- 1{F(s);x} = 1/(1 + ax)m, O<x< I/a
Choosing a second contour as shown in Fig. 6.2(b), the residue
calculus applied to the remaining poles of F(s) on the positive real
axis leads to
_,u- 1{F(s);x} = - LRes{x-F(s);m+n}
n=O
where the negative sign in front of the summation is due to integrating

in the negative direction along the closed contour in Fig. 6.2(b).
Recalling Eq. (1.6) in Sec. 1.2, we find that
Res{x -sf() } . (s-m-n)f(s)f(m-s)

s ;m + n = 11m -'-----"-.-'---'---'------'-
s->m+n (ax) f(m)
= _ Iim f(s )f(m - s + n + l)
s->m+n(axY(m- s)(m- s + 1) (m- s + n- 1)
*These are the sarne values of x for which the integral along the circular are in Fig.
6.2(a) vanishes in the limit as the radius tends to infinity.
6.3/Complex Varillble Methods 259
(-ltf(m+n)
=
Hence, we obtain the descending binomial series
.,u- 1{F(s);x} = ~~ (-nm) (ax)-n

(ax) n=O
which converges for iaxi > 1, and therefore

.,u- 1{F(s);x} = 1/(1 + ax)m, x > 1/a
Combining results, we finally deduce that
_,u-l{f(s)f(m- s)x} = 1 O<x<=

af(m) ' (1 + ax)m'
in agreement with Exam. 6.4.
6.3.3 Transforms in Polar Coordinates

ln the application of Mellin transforms in polar coordinates (r, O) it is
frequently necessary to consider inverse Mellin transforms of functions
sueh as F(s )cos slJ and F(s )sin slJ, where F(s) is the Mellin transform
of a real function f() (e.g., see Sec. 6.4.3). The technique illustrated
below for evaluating such inverse transform is based on a paper by W. J.
Harrington, * who pro vides a more rigorous treatment and justification
of the formal result.
Suppose that F(s) is the Mellin transform of a real function f().
Then, we have formally
.,U{f(reie);r ~s} = L=r-1 f(reie) dr
= e -is8L=s-1 f() d
where we have set = rei8 Hence, we deduce that
Al{f(rei8);r~s} = e-is8 F(s)
*W. J. Harrington, "A property ofMellin transforms," SIAM Review, 9, No. 3, 542-
547, July (1967).
which leads to
.,.u- 1{F(s)cos s(};s ~r} = Re[f(rei11)] (6.29a)
.,.u- 1{F(s)sins8;s ~r}= -Im[f(rei6)] (6.29b)
Example 6.8: Find the inverse Mellin transforms of

1T cos s(} d 1T sin s(}
. an .
sm S1T sm S1T
Solution: We first recall Exam. 6.3, which gives us
.,.u- I{ 1T
sin S1T ;x
}
= 1 +1 x
Therefore, from (6.29a) it follows that
.,.u-t{'TT cos s(}s ~r} = Re [

sin S1T '
1 ]
1 + rei8
= r
1 + cos (}
1 + 2r cos (} + r 2
and similarly, from (6.29b),
{1Tsins(} } rsin(}
11 _ 1
.;n ---s ~r = ---------=
sin S1T ' 1 + 2r cos (} + r
2
EXERCISES 6.3
ln Probs. 1-6, use residue calculus to evaluate the Mellin transform of
the given function.
1
l. f(x) = (x + a)(x +h)' a,b >O
1 1
2. f(x) = (1 + x) 3 3. f(x) = - - 3
1 +x
l X
4. f(x) =- - 4 S. f(x) = (1 + rf
1 +x
1
6. f(x) = (x2 + a2)(xz + b2) ' a#- b, a,b >O
7. Show that Eq. (6.23) can be equivalently expressed in the form

N
JU{f(x);s} = _..;!-. LRes{(e-i1Tz)'- 1f(z);aJ
sm 1TS k=t
8. If f(z) is a rational function satisfying conditions (6.20), and if f(z)
also has poles b~> b2 , ,bM on the positive real axis, show that
N
.M{f(x);s} = __;!!-- LRes{(e-i,.zY-1(z);ak}

sm rrs k=t
N
-rr cot rrs L Res{f(z);bk}

k=l
ln Probs. 9 and 10, use the result of Prob. 8 to find the Mellin transform
1 1
9. f(x) = - - 10. f(x) = -
1- X 4 -x2
11. Show that
(a) .M{erfc(x);s} = r(s ; );v;
1
s
(b) Using residue methods, show that
- 1{
r(s
-2-
+ 1) } 2 "" (-1tx2n+l
.M y; 1TS
;x =1- r= L
V1Tn=On.
'(2n +1
) = erfc(x)
12. Show that

(a) .M{E1(x);s} = f(s)/s
where E 1(x) is the exponential integral defined by
E (x)
1
= {""e-t dt
Jx t
(b) Using residue methods, find an ascending series for the inverse
Mellin transform of F(s) = f(s)/s.
13. Show that
(a) .M{Ci(x);s} =- f(s) cos(rrs/2)
s
where Ci(x) is the cosine integral defined by
- -t dt
.( ) = - J,""cos
C IX
t X
(b) Using residue methods, find an ascending series for the inverse
Mellin transform of F(s) = - f(s)cos(rrs/2)/s.
14. Use residues to show that
(a) .M-1{ 1T x} = ~ - ~ ( -l)nx2n+l O<x<1

2s COS(rrs/2) ' 2 n=O 2n + 1 '
_ 1{ 7r } = ( -IY
(b) .At 2s cos(Trs/2) ;x = ~o (2n + l)x 2n+l'
x>I
(c) From (a) and (b), deduce that
.At- 1{ 7r
2s COS(Trs/2)
;x} = ~2 - tan- x ' 1
x>O
ln Probs. 15-26, use residue methods to determine the inverse Mellin

= r~:.
2
15. F(s) ) 16. F(s) = -.-
sm 1TS
7r
7r
17 ' F(s) = 2 sin(Trs/2) 18. F(s) = f(s)cos(7Ts/2)
2- 1r(s/2)
19. F(s) = f(s)sin(Trs/2) 20. F(s) = f(l - s/2)
21. F(s) = c~s scp, 22. F(s)

sin scp
= -.-, - - <
7r
<P
7T
< -
smsa smsa 2 2
7r 7r
-2 < <P < 2
23. F(s) = 7r c_os sO 24. F(s) = 7T ~in sO
s sm sTr s sm S7T
Hint: See Prob. 37 in Exer.
6.2.
25. F(s) = cos s(} 26. F(s) = sin s()

COS S7r COS S1T
6.4 Applications
Although the Mellin transform is not nearly as versatile in applications
as are the Fourier and Laplace transforms, there are some areas of
application where it can be a useful tool. ln particular, it is useful in the
summation of certain series, in finding the distribution function for products
of random variables, and in solving for the potential function in a wedge-
shaped region. Our discussion of these applications, however, will be
intentionally brief since any deeper treatment would require mathematical
knowledge beyond the stated prerequisites.
6.4/Applications 263
6.4.1 Summation of Series

The discussion in this section is based on a paper by G. G. MacFarlane,
who considered more general series than those presented here. *
Given that f(x) and F(s) are Mellin transform pairs, they are related
by the inversion formula
c+ioo
f(x) = 1
7Ti J -
2 c-ioo
x s F(s) ds
Replacing x by nx, where n = 1,2,3, ... , this inversion formula becomes

c+iC/C)
f(nx)
2 Jn-sx-sF(s) ds,
= ~i
c-ioo
n = 1,2,3, ... (6.30)
Now let us sum both sides of (6.30) over ali positive integer values of
n, interchanging the order of summation and integration on the right-
hand side. This action leads to
c+ioo
00
l;((nx) = 1 J
7Ti . x-s~(s)F(s) ds (6.31)
2 c-too
where ~(s) is the Riemann zeta function defined by

1 00
~(s) = 2: -.,, Re(s) > 1 (6.32)

n=ln
Thus we have the result
Lf(nx) = .M -I{~(s )F(s) ;x} (6.33)

n=l
The importance of Eq. (6.33) is that we can replace the series on the
left, which in some cases may converge very slowly, by the inverse
Mellin transform on the right. This inverse transform leads to another
infinite series, but in certain cases it may be summed exactly. ln other
cases this new series may converge faster than the original series, which
is important for computational purposes. Proficiency in this technique
of summing series requires knowledge of some properties of the zeta
function, most of which were developed in 1859 by G. Riemann (1826-
*G. G. MacFarlane, "The application of Mellin transforms to the summation of slowly

convergent series," Phil. Mag., (vii), 40, 188 (1949).
1866). Some of the relations involving the zeta function that prove useful
in our work here include the following: *
,(0) = -1/2 (6.34)
,(1) = 00 (6.35)
,(2) = 7T
2
/6 (6.36)
=
,(4) 7T
4
/90, (6.37)
'( -2n) = O, n = 1,2,3, ... (6.38)
r<O) == -Hog 27T (6.39)
7T''(l - s) == i-r(s),(s)cos('ITs/2) (6.40)
Example 6.9: Sum the series 2: (cos an) / n 2

n~t
Solution: The given series defines the function
f(x) == cos ax
2
X
whose Mellin transform is
Al{f(x);s} = L"'x.- 3 cos ax dx

= Al{cos ax; s- 2}
= - (1/a'- 2)f(s - 2)COS(7TS/2)
ln deriving this transform we have used the result of Exam. 6.2 and
the scaling property (6.10). Substituting this result into Eq. (6.33)
yields
~ cos an ==
LJ-- -
_{
-j{ 1
1_ ,(s)f(s- 2)COS(7Ts/2);x= 1}
2 8 2
n~l n a
U sing property (6.40) of the zeta function, we can express this inverse
Mellin transform in the more convenient form
~ cos an = _ a
n=t n
2
2
2
Al_ 1{(27T)' '(1 - s)f(s- 2) ;x =
a f(s)
t}
2
= _ a .M-t{(27T)'
2 a
'(1 - s)
(s - l)(s - 2) '
x== t}
*For a detailed discussion ofthe zeta function, see E. T. Whittaker and G. N. Watson,
A Course of Modern Analysis, Cambridge: Cambridge University Press, 1965.
Thus, we have three simple poles at s = O, 1, 2, the residues of

which are
Res{O} = -1/2, Res{I} = 1r/a, Res{2} = -7r
2
/3a 2
The details of evaluating these residues are left to the exercises (see
Probs. 1 and 2 in Exer. 6.4). Combining results, we finally obtain
~ cos an = _ a _! + ~ _
2 2 2 2
( 7r ) =a _ 7ra + 7r
2
n= 1 n2 2 2 a 3a 4 2 6
Observe that in Example 6.9 the limit a - o+ leads to the special
case
ao 1 2
L-=~ (6.41)
n= I n2 6
6.4.2 Products of Random Variables

Unlike the distribution of sums of independent random variables, the
distribution of products of independent random variables has received
relatively little attention in the literature. This is particularly true of
products involving more than two variables. It has been shown that the
probability density function (PDF) for such products can be obtained
through use of the Mellin transform. * The general theory conceming
these problems is too complex for our purposes, but we can illustrate
the utility of the Mellin transform by examining certain products involving
only two random variables.
Let us define Z = XY, where X and Y are independent, positive,
random variables. From probability theory it is known that the PDF of
Z is related to the joint PDF of X and Y by
pz(Z) =~ JLpx.y(x,y) dy dx (6.42)
where Dz is the region of the xy plane such that xy :s z. (ln this section
the variable z is considered to be real.) Since X and Y are independent
random variables, we can express their joint density function as a product
of their respective PDFs, and thus (6.42) becomes
d rz/xr=
Pz(Z) == dz Jo Jo Px(x)py(y) dy dx
*See M. D. Springer and W. E. Thompson, "The distribution of products of beta,

gamma, and Gaussian random variables," SIAM J. Appl. Math. 18, No. 4, 721-737, June
1970.
i"
which simplifies to
dx
Pz(z) = o
Px(x)py(z/x)-
X
(6.43)
We recognize (6.43) as the convolution integral of the Mellin transform

[see Eq. (6.19)], and, therefore, we can express (6.43) in the alternate
forrn
(6.44)
where F 1(s) and F 2(s) are the Mellin transforrns, respectively, of Px(x)
and py(y).
Although we will not provide the details, we can generalize (6.44) to
the case involving products of N random variables. That is, if
N
z = nxj (6.45)
j=l
where X~> X 2 , ... , XN are independent, positive, random variables, then

it can be shown that
Pz(Z) = .l- 1ll]Fis);z} (6.46)
where Fj(s) is the Mellin transform of the PDF for Xi, j = 1,2, ... ,N.
Further generalizations to products involving both positive and negative
random variables have been established, as well as generalizations to
certain quotients of randorn variables.
Example 6.10: Find the PDF of the product Z = XY, where X and Y
are independent Cauchy random variables, each having density function
p(x) = 2/7T(l + x2), x >O
Solution: Using (6.43), we find irnmediately (upon simplification)

4 ("" xdx
Pz(z) = 'TTz J0 (xz + l)(xz + l)
which we can interpret as a Mellin transforrn of the rational function
f(x) = 1/(x2 + l)(x 2 + z2)
for the special value s = 2. Using the rnethod of residues, it can be
shown that (see Prob. 11 in Exer. 6.4)
'TT(Zs-2 - 1)
.M{f(x);s} = 2(z2 - l)sin(TTs/2)
Therefore, the result we need is

. 2(zs-2 - 1)
Pz(z) = hm 7T(z2 -
.~2 1)sm
. ('Tf'S / 2)
which, through an application of Leibnitz's rule, leads to

4log z
Pz(z) = 7T2(z2 _ 1)' z>O
We could equally well use Eq. (6.44) to find the PDF by first
noting that
Al{px(x);s} = Al{py(y);s} = 1/sin(7Ts/2)
However, the details of this approach are left to the exercises (see
Prob. 12 in Exer. 6.4).
Finally, observe that, because Px(x) and py(y) are even functions,
we could extend our result to products involving Cauchy random
variables that extend over -ao< x <ao, -ao< y < oo. The PDF
for Z = XY in this case assumes the form
log z2
Pz(Z) = 7Tz(l _ l), -ao<z<oo
the verification of which we leave to the reader.
6.4.3 Distribution of Potential in a Wedge

Let us consider the boundary value problem of determining a potential
function in an infinite wedge-shaped region in the xy plane (see Fig. 6.3).
ln polar coordinates the problem is mathematically formulated by
r 2u,, + ru, + u88 =O, O<r<oo, -a<O<a
B.C.: u(r,-a) = f(r), u(r,a) = g(r)
{ u(r,O) ~O as r~ (6.47)
ao, 101 <a
Figure 6.3 Infinite wedge

Among other areas of application, this problem describes the steady-

state temperature distribution in an infinite wedge, given the temperature
distribution along the boundaries of the wedge.
Using the differentiation property of the Mellin transform (see Prob.
22 in Exer. 6.2)
(6.48)
we find that under the Mellin transform the potential problem (6.47) is
transformed to
uflfl + s2U = o, -a<()< a (6.49)
B.C.: U(s,-a) = F(s), U(s,a) = G(s)
The general solution of the transformed DE is
U(s, O) = A(s )cos s() + B(s )sin s()
and by imposing the boundary conditions in (6.49), we obtain
U(s,()) = F(s) sin ~(a - ()) + G(s) sin ~(a+ O) (6.50)

sm 2as sm2as
The inversion of this solution by the convolution theorem, given by Eq.

(6.19), leads to the formal result
,.m [ {"" gm-1 J(g) dg
u(r, ()) = 2a cos(m(J) Jo g2m + 2rngmsin(m()) + ,zm
("" gm-1 g(g) d ]
+ Jo em - 2r"'msin(m8) + ,zm (6.51)
where we are using the inverse Mellin transform relation (see Prob. 22
in Exer. 6.3)
.!l-l{ sin scfJ s~r} = _!_. rmsin(mc/J) (6 _52 )

sin 2as ' 2a 1 + 2rmcos(mc/J) + r 2m
where m = 7T /2a.
EXERCISES 6.4
F(s) = (2/as) '(s)f(s - 2)cos(7Ts/2)
show that Res{F(s );O} = -1/2.
F(s) = (27T)s
a
,(1 -
s)
(s - l)(s - 2)
show that
27T 7T
(a) Res{F(s);I} =- -'(0) =-
a a
47T2 7T2
(b)Res{F(s);2}=- 2 '(-1)= - -2
a 3a
3. Show that
00
"" sm an _ 7T - a
L.J O< a< 27T
n=l n - 2 ,
4. Show that
~ _,2 I .. ~-~ -1TZk2
L.Je = --+ y7TL.Je
n=l 2 k=O
5. Show that
(a) ~ (-I~"- I = (I - 21 -ms)
n=l n
(b) ~o (2n ~ IY = (I - 2 -.),(s)
6. From the results of Prob. 5, deduce that

(a) L (-1)"-Y(nx) = AC 1{(1 - 2'-ms)F(s);x}
n=l
n=O
ln Probs. 7-10, use Eq. (6.33) or the results of Prob. 5 to sum the given
series.
oo ( -l)n-1 00
7. L 2 cos an S. L sm an
n=l n n=l 2n - 1
oo (- l)n-1 ~ J 1[(2n + 1)a]
9. L sin an 10. L.J
2n + I
n=l n n=O
11. Show that

1 } 7T(Zs-2 - I)
.l
{ (x2 + l)(x2 + z2) 's = ---:=---'----'---
2(l - 1)sin(7Ts/2)
12. By using Eq. (6.44),

(a) show that the PDF of Exam. 6.10 can also be expressed as
Pz(Z) = Ai -1{ sin2(7Ts/2)

1 ;z
}
(b) U sing the method of residues, show that the inverse Mellin trans-
form in (a) leads to the result of Exam. 6.10.
13. Given that X and Y both have the PDF p(x) = (a + l)xah(l - x),
X> O,
(a) show that Z = XY has the PDF
(a + 1)2 1
Z h(l - z)log ~
0
Pz(Z) =
2
z> 0
(b) If Z is a product of N independent random variables, all having
the above density function, show that
(a + l)N-1 ( 1)N-1
Pz(Z) = (N _ 1)! zah(l - z) log ~ , z>O
14. The PDF of W = 1/f is known to satisfy the Mellin transform

relation
Al{pw(w);s} = Al{py(y);-s+2}
Use this re1ation to show that the PDF of Z = X j Y, where X and
Y both have the PDF p(x) = (a + l)xah(l - x), x > O, is given
by
[(a + 1)/2] Z 0
' O:::;z:::;1
Pz(Z) = { [(a + 1)/2] Z-a-2, z > 1.
15. Recall from Exam. 3.10 in Sec. 3.8 that the product of two independent,
zero-mean, Gaussian random variables with equal variances has the
PDF
Pz(Z) = (1/7T)Ko(lzl)
where K 0(x) is a modified Bessel function of the second kind. Use
this result to deduce that the Bessel function K0 (z) has the integral
representation
16. Use the result of Prob. 15 to deduce that the PDF of Z = XY, where
X and Y have the sarne PDF p(x) = e-xh(x), is given by
pz(z) = (2/7T)Ko(Vz /2)h(z)
17. Use the result of Prob. 14 to find the PDF of Z = X/Y, where X
and Y are positive independent random variables, each with PDF
p(x) == V2/TT e-x 212 h(x)
18. For the special case of (6.47) where the boundary conditions are
u(r, -a) == u(r,a) == f(r)
(a) show that the solution of the transformed problem is
U(s, O) == F(s) cos s()

cos sa
(b) Following the technique in Sec. 6.3.3, show that
At _ 1 { cos s() .s ~
cos sa'
r} = r"'
a
12
(1 + r"')cos (m() /2)
1 + 2r"'cos (m8) + r 2m'
where m == TT /a.
(c) Using the result in (b), find a formal solution of the potential
problem by using the convolution theorem.
19. Finda formal solution of the problem described by (6.47) when the
boundary conditions are prescribed by
u(r,a) == u(r,-a) == {~: O<r<a

r>a
20. Find the steady-state temperature distribution inside the infinite wedge-
shaped region O < r< oo, O < () < a, if the boundary () = O is held
at temperature zero while the other boundary is maintained at
T. O< r< a
u(r,a) = { 0,0 ' r>a
21. Shaw that the integral equation
u(x) == f(x) + Jor= u()g(x/) T

d

1
u(x) == - .
fc+i= X-sF(s) ds
2m c-i= 1 - G(s)
22. lf g(x) == /(1 + x) in Prob. 21,
(a) show that
u(x) = f(x) + ~ J.c+i= (~TT)x-sF(s) ds

2m c-i= SID S'TT - 'TT
272 Chap. 6/The MeUin Transform
(b) lf 7T = sin om, where O < a < 1/2, show that
Jtl
-I
o
SID a7T
o
o
o
,X-
} _ t an a7T ( X -a - XI
2
+a)
{
SID S7T - SID a7T 7T 1- X
(c) From (a) and (b), deduce the formal solution
u(x) =f(x) + x (tan7Ta7T) L= f(~) [(x/~)-;2-~ ~(x/~)a] d~

23. Show that the integral equation
f(x) = L= g(x~)u(~) d~
1 fc+i= F(l - S) -s d
( )
UX =- X S
27Ti c-i= G(l - S)
Hint: See Prob. 26 in Exer. 6.20
24. Use the result of Prob. 23 to solve the Laplace integral equation
-1- = i= e-xl; u(~) d~, x>O

1 +X O
25. Show that the integral equation
u(x) = f(x) + L= g(x)u() d

1 fc+i= [F(s) + G(s)F(l - s)] -s d
( )
UX =- X S
27Ti c-i= 1 - G(s)G(l - s)
26. Use the result of Probo 25 to solve the integral equation
u(x) = f(x) + y Jo
1 r= e-xl; u(~) d~, x >O
where
log(l + x)
O<x<1
f(x) ~ { lo~/+ x) _ !, x>1
X X
6.5/Table of Mellin Transforms 273
6.5 Table of Mellin Transforms

The following is a short table of Mellin transforms and their inverses.
Table 6.1 Mellin Transforms
No.
I ic+i=
f(x) =---: x- F(s) ds
27Tt '-i=
F(s) = r x'- 1 f(x) dx
7T
I +X sin 1rs
2 7T cot 1rs
I - X
I f(s)f(m- s)
3 a >O.m >O
(I + ax)m' a'f(m)
f(s/a)f(m - s/a)
4 a >O,m >O
a f(m)
5 f(s)/a'
6 f(s/2)/2a"
f(s)
7 cos ax, a> O - . COS(7Ts/2)
a'
f(s) .
8 sin ax, a > O -sm(7rs/2)
a'
a"-'2-- 1f(s/2)
9 x-"lv(ax), a> O, v> -1/2
f(11 + 1 - s/2)
s +
2'-lf ( -2- JJ)
lO J"(ax), a >O, v > -1/2
s
a'r -v - -+ I)
( 2
7
The Hankel Transform
7.1 Introduction
Hankel transforms arise naturally in solving boundary-value problems
formulated in cylindrical coordinates. They also occur in other applications
such as determining the oscillations of a heavy chain suspended from
one end, first treated by D. Bernoulli. This latter problem is of some
special historical significance since it was in this analysis of Bernoulli in
1703 that the Bessel function of order zero appeared for the first time.
To formally derive the Hankel transform and its inversion formula,
we start with the two-dimensional Fourier transform pair (see Sec. 2.9.2)
F(,"fJ) = _!_f= J= ei<x~+.v">f(x,y)dxdy (7. 1)

2'7T -oo -oo
and
f(x,y) = _!_ J=
2'7T -oo
J=
-oo
e-i<x~+.v"> F(,"'J) d d"fJ (7.2)
ln applications involving these transform pairs, such as in optics, for

instance, the problem often exhibits circular symmetry for natural reasons.
When this happens, it may be expected that a simplification will result
since one radial variable will suffice in place of two independent rectan-
gular coordinates. Thus, let us assume that f(x,y) == f(r), where r =
(x 2 + l), 112 and transform (7 .1) and (7 .2) into polar coordinate repre-
sentations. ln terms of the polar coordinates
274
7.1/lntroduction 275
x =r cosO, y = r sinO
(7.3)
= p cos ~. Ti= p sin ~
we find that (7.1) becomes
= _1 f""
Ff "')
,p,"' 2'11" o Jofz" eiprcos<O-l rf(r) dO dr (7.4)
where we have recognized that x + Y'Y/ = pr cos (O - ~). Recalling

property (J12) in Sec. 1.4.1, we have
-1 12" e'Pr... cos(8-t/J) dO = lo(pr) (7.5)

2rr o
where J 0(x) is the Besselfunction of order zero. This result suggests that
F(p,~) is a function of p alone, i.e., F(p,~) = F(p), and in this case,
(7 .4) leads to
F(p) = {"' rf(r)J0(pr) dr (7.6)
called the Hankel transform of order zero. Clearly, the substitutions (7.3)
into Eq. (7.2) will produce the similar result
f(r) = L"" pF(p)J0(pr) dp (7.7)
which represents the inversionformula for the Hankel transform of order

zero.
When circular symmetry does not necessarily prevail, we define the
general Hankel transform of order v by
Yfv{f(r);p} = L"' rf(r)lv(pr) dr = F(p), v> -1/2* (7.8)
the corresponding inversion formula of which takes the forro
Yt'; 1 {F(p);r} = L"" pF(p)lv(pr) dp = f(r), v> -1/2 (7.9)
A heuristic argument for the special case v = n/2 - 1 can be presented

to derive (7.8) and (7.9), which parallels the above development for
n = 2, by considering the multiple Fourier transform of order n applied
to a radially symmetric function. t
*The restriction v > -1/2 may be extended to a larger interval in certain transform
results.
tSee I. N. Sneddon, The Use of Integral Transforms, New York: McGraw-Hill, 1972,
pp. 79-83.
276 Chap. 7/The Hankel Transform
The basic requirement for the existence of the Hankel transform (7 .8)
is that the function Vrf(r) be piecewise continuous and absolutely integrable
on the positive realline. The proof of the Hankel inversion formula (7.9)
is similar to, but more complicated than, the corresponding proof of the
Fourier inversion theorem. This is due primarily to the fact that the
Hankel inversion formula relies on a good understanding of the properties
of Bessel functions, which are more complicated than those of the cor-
responding kemels of the Fourier transform. The interested reader can
consult Sneddon for the proof. *
7.2 Evaluation of Hankel Transforms

As in the case of the Mellin transform, we find that certain Hankel
transforms are directly related to known transforms previously calculated.
Also, many Hankel transforms and basic properties can be developed
for general values of v, although most applications involve only the cases
v = O and v = 1. ln this section we will determine the Hankel transform
of certain functions and develop some of the fundamental operational
properties of the transform. When possible, we will keep the value of v
arbitrary.
Example 7.1: Find the Hankel transform of rvh(a - r), a > O.
Solution: From definition, we have
;;[Jrvh(a - r);p} = J: r v+ 1v(pr) dr

1
= v\2lap Xv+IJv(X) dx
p o
the last step resulting from a change of variable. Now using Eq. (1.42)
in Sec. 1.4.1, we deduce that
a>O,v> -1/2
Example 7.2: Find the Hankel transform of rv-te-a', a >O.
Solution: Here we see that the Hankel transform can be expressed

as a Laplace transform, i.e.,
*1. N. Sneddon, The Use of Integral Transforms, New York: McGraw-Hill, 1972,
pp. 301-309.
7.2/Evaluation of Hankel Transforms 277
~v{r"-le-a';p} = L"" r"e-a'lv(pr) dr

= ::t {r"lv(pr);r ~ a}
Thus, recalling Exam. 4.16 in Sec. 4.4.1 and the scaling property of
the Laplace transform, we immediately have the result
'jJf {r"-le-ar, } = (2p)'T(v + 1/2) a> O, v> -1/2

v ,p .,./(2 2)v+l/2'
y1T p +a
The special case v = O in Exam. 7.2 leads to the results
ou; {
<11-o 1 -ar ;p } = . . 1
-e , a> O (7.10)
1 2
r vP +a2
~o{;;p} = ~' p>O (7.11)
Also, since multiplication by r corresponds to differentiation in the Laplace

transform domain, we obtain
or
a>O (7.12)
2
Example 7.3: Find the Hankel transform of order zero of e-a r2.
Solution: From the defining integral, we have
'Je0{e -aZrZ;p} = L"' re -a2' 210(pr) dr
By expressing the Bessel function as an infinite series and interchanging

the order of integration and summation, we find*
ou; { -a2r2, } = ~ ( -l)n(p/2)2n f"" ,Zn+l -a2r2 d
<11- o e ,p LJ ( ') 2 e r
n=O n. O
*Aiso, note that Jl'0{e-"2' 2 ;p} = L:t{J0(pyt);t-+ a}, which can be evaluated through
use of Exam. 4.15 in Sec. 4.4.
the last step of which follows a change of variables. This last integral
is simply n!, and thus
'Je { -az~. } = _1 ~
2 2
( -1t(p /4a t
Oe ,p 2a2 n~O n!
or
ou; { -a2~. } - _1_ -p2/4a2 a >o
dLo e ,p - 2a2 e '
Observe that the special case a2 = 1/2 in Exam. 7.3 leads to

(7.13)
which shows that e-~;z is self-reciprocal. It is interesting that this sarne
function is self-reciprocal under the Fourier transform [see Eq. (2.52) in
Sec. 2.4].
7 .2.1 Operational Properties

Because the Hankel transform and inverse Hankel transform are exactly
the sarne in functional form, it follows that each operational property of
the transform is likewise a property of the inverse transform. For example,
as a consequence of the linearity property of integrais, we deduce that
'Je~{Ctf(r) + Czg(r);p} = C 1F(p) + C2G(p) (7.14a)
1
'Je; {C1F(p) + C2G(p);r} = C1f(r) + Czg(r) (7.14b)
where C 1 and C2 are arbitrary constants and F(p) and G(p) are the Hankel
transforms, respectively, of f(r) and g(r).
If we make the simple change of variable x = ar in the defining
integral
'Je~{f(ar);p} = Loo rf(ar)J~(pr) dr, a>O
we obtain
'Je~{f(ar);p} = 2a1 100

o
xf(x)Jv(pxfa) dx
from which we deduce the scaling property

'Je~{f(ar);p} = (l/a2)F(p/a), a>O (7.15)
where F(p) is the Hankel transform of f(r).
It is not possible to derive a simple shift formula for the Hankel
transform because the addition formula for the Bessel function, even for
integer values of v, takes a very complicated form. That is, in contrast

with the simple addition formula
for the exponential function, which is the basis of the shift formulas for
the Fourier and Laplace transforms, we have the Neumann-Lommel
addition formula
m=-cx:>
for the Bessel functions. Although it is possible to derive a shift formula

for the Hankel transform based on this addition formula, it becomes
unwieldly and not very useful.
The lack of a simple addition formula for the Bessel functions also
precludes the existence of a simple convolution theorem for the Hankel
transform. However, a simple relation of Parseval type can be easily
derived for this transform. If F(p) and G(p) are Hankel transforms of
f(r) and g(r), respectively, then
f' pF(p)G(p) dp = Loo pF(p) Loo rg(r)Jv(pr) dr dp
= Loo rg(r) Loo pF(p)Jv(pr) dp dr

where we have interchanged the order of integration. 1t now follows that
ioo pF(p)G(p) dp = f" rf(r)g(r) dr (7.16)
O < a < b, v > - 1/2
Solution: The given integral has the form
I = L"" xF(x)G(x) dx
where
F(x) = (1/x)Jv+l(ax)
and
G(x) = (1/x)Jv+l(bx)
We recognize from Ex. 7.1 that F(x) and G(x) are the Hankel transforms,
respectively, of
and
g(r) = V/bv+t)h(b- r), b >O
Hence, based on the Parseval relation (7.16), we have
I=- 1- r2v+l drfa

(abt+ 1 o
which leads to
r=1 1
Jo ~lv+ 1 (ax)Jv+l(bx)dx = 2(v + 1)
(a)v+l
b O< a< b, v> -1/2
The Bessel functions satisfy the three-term recurrence formula [see

Property (J9) in Sec. 1.4.1]
lv-t(Z)- (2v/z)Jv(Z) + lv+t(Z} =O (7.17)
Based on this relation, we obtain
Yt'v{;f(r};p} =;v L= rf(r) [lv+l(pr) + lv-l(pr}] dr
or
Yt'v{;f(r);p} =;v [ Yt'v+t{f(r);p} + Yt'v-t{j{r);p}].v> -1/2 (7.18)

ln solving boundary-value problems by means of the Hankel transform
it is necessary to develop formulas connecting the Hankel transform of
functions to the Hankel transform of their derivatives. Most important
among these formulas are the special cases
Yt't{f'(r};p} = - pYt'o{f(r};p} (7.19)
and
Yt'o{; :)rf(r)];p} = pYt't{f(r);p} (7.20)
To derive (7.19), we start with an integration by parts to get
Yt't{f'(r);p} = L= rf'(r)J1(pr) dr
= rf(r)J1(pr}
= r= d
Jo f(r) dr [rJ 1(pr)] dr (7.21)
I 0
-
Recalling the asymptotic relations [see (J13) and (J14) in Sec. 1.4.1]
(7 .22a)
z -7 00 (7.22b)
we see that the first term on the right in (7 .21) vanishes provided
lim y;f(r) = O
lim rf(r) = o.
The first limit condition is satisfied by any function whose Hankel transform
exists, and the second condition imposes on f(r) a certain behavior near
r = O. ln addition, we note that [see Eq. (1.36) in Sec. 1.4.1]
d
dr[rJ 1(pr)] = prJo(pr)
so that (7.21) reduces to
;tt'I{f'(r);p} =-p f" rf(r)Jo(pr) dr
which in turn leads to (7.19). Equation (7.20) follows in a similar fashion

but is left to the reader to verify (see Prob. 25 in Exer. 7.2).
Both (7 .19) and (7 .20) have generalizations to Hankel transforms of
order v. For instance, Eq. (7.19) is a specialization of either
;t{ v{f'(r);p} = -:V[ (V + l);t{v-l{f(r);p} - (V -l);t{v+ I{f(r);p}J (7 .23)
or
;tevV- 1 :,[r1 -vf(r)];p} = -p;tt'v-I{f(r);p} (7.24)
while (7 .20) is a specialization of
;tev{r:I :rv+I_t(r)];p} = p,rt'v+I{f(r);p} (7.25)
The verification of these results is also left to the exercises (see Probs.
26-28 in Exer. 7.2).
Hankel transforms provide significant simplifications in solving partial
differential equations that lead to Bessel's equation
ry"(r) + ry'(r) + (r 2 - v2 )y(r) =o (7.26)
To understand why this is so, let us first define the function
d
g0(r) = y"(r) + -r1 y'(r) = -1r -d
r
[ry'(r)] (7.27)
282 Cbap. 7/Tbe Hankel Transfonn
and apply the Hankel transform of order zero. This action leads to
:1eo{go(r);p} = p:1et{y'(r);p}
= -p2:1eo{y(r);p} (7 .28)
where we have used (7.20) and (7.19), respectively. More generally, if
we define
1d v2
g ..(r) = ; dr[ry'(r)] - r2 y(r)
= ?: ![,zv+t :r(~y(r)) J
1 (7.29)
and apply the Hankel transform of order v, we obtain the similar result
(7 .30)
by use of (7.25) and (7.24), respectively. Hence, we see that the Hankel
transforms of g0(r) or gv(r), which involve derivatives of y(r), are related
directly to Hankel transforms of y(r).
EXERCISES 7.2
ln Probs. 1-15, verify the given Hankel transform relation. When possible,
use known integral transforms results from previous chapters.
1 ou; { v -ar. } = 2a(2pt f(v + 3/2) a>O

dL v re ,p c 2 2 3/2 '
y7T(p +a y+
2. ou; {
dL v -
1e -ttr ;p} = . . I 2
1 2
( P
... I 2 2
)v '
a >O
r v p +a a+ vP +a
Hint: Differentiate both sides of Prob. 3 with respect to a.
S. :1eo{(a2 - r 2)11-- 1h(a- r);p} = 21-L-If(~-t)(~rJ,..(ap), a> o
6. :1evV(a2 - r 2)~'-"- 1 h(a - r);p} = 2~'-"- 1

f<~-t - v)a~'p"-~'J,..(ap),
a> O, IL >v~ O
7. :reo{ ~ J 1(ar);p} = ~ h(a - p), a> O
v-1' . _ pv(a2 _ p2)"'-v-l

8. ;r{v{r Jiar),p} - 2~'- v-lf(JL - v)a~'-h(a - p), a> O, JL >v;:::: O
Hint: See Prob. 6.
9. ;reo{ Va 2
h(a -
_
r)i p } = Jmrp J112(ap), a> O
2
Hint: See Prob. 8.
10. ;rev{
rvh(a - r)
v'a2 - r2 ;p
}
=
JTT2/ v+ 1/2
lv+ 1/z(ap), a >o
Hint: See Prob. 8.
1 . }
11. ;rt'0{ -smar;p = ~h(a
1 2
- p)
2
, a> O
r va - p
1 . } ah(p - a)
12. ;re 1{ - sm ar;p
r
= V
p p2 -a2
, a >O
13. :re~{~e-ar2;p} = ~(1- e-p /4a), 2

a >O
14. ;rt' 1{ ~ cos (br2);p} = ~ [ 1 - cos(:;)]. b >O
15. ;rt'1{ ~ sin (br2);p} = ~ sin(:;), b >O
16. Integrate both sides of the Hankel transform relation in Prob. 7 with
respect to a from O to b and deduce that
:reo{?[l- J (br)];p} = h(b- p)log(~),

0 b>O
1
f
b
Hint: J 1(ar) da = - [1 - J0(br)]
o r
17. Show that

(a) ;rt'v{r'- 1 ;p} = .!l{lv(pr);r~(s + 1)}
(b) Evaluate the Mellin transform in (a) and thus show that
_1 2T(s/2 + v/2 + 1/2)
;rt'v{r' ;p} = ps+lf(v/2- s/2 + 1/2), -1-v<s<1+v
284 Chap. 7/The Hankel Transfonn
18. Starting with the integral formula
1= cos(ax)J (bx) dx = I/yb

0
2
- a2 , O<a 1
19. Show that
Yfo{~ e-a2~14 ;p} = ~e-P212a[0 (~~), a>O
20. Show that
~exp(p ~ b )/o(~:).
2 2
Yfo{e-arl lo(br);p} = a> O, b >O
21. Use the integral representation
1 = 1_ r= e-(rZ + a2)x X-1/2 dX

v'rz + a2 VTr Jo
to deduce that
Yfo{.yr / +a2;p} =.!.P e-ap, a> o
22. Use the result of Eq. (7 .10) and other appropriate properties to show
that
2az - pz
(a) .rt'o{re-a';p} = ( 2 z)s;z a>O
p +a
(b) .rt'o{e-a';p} = (pz +a a2)3/2 a>O

-ar a + vy'pz + az ( P )"
.rt',{e ;p} = (pz + az)3/2 a+ ypz + az ' a> O
24. Show that r"e-~12 is a self-reciprocal function with respect to the

Hankel transform of order v.
25. Verify Eq. (7.20).
26. Verify Eq. (7.23).
27. Verify Eq. (7.24).
28. Verify Eq. (7.25).
7.3 Applications
One of the principal uses of the Hankel transform is in the solution of
initial boundary-value problems involving cylindrical coordinates. We
will briefty illustrate some of the classical examples of such problems
where the Hankel transform is an effective tool.
7.3.1 Potential Problems

Let us first consider the axisymmetric Dirichlet problem for a half-space
which is mathematically characterized by
u,, + (1/r) u, + Uzz = O, O< r< oo, z >O
u(r,O) = f(r), O < r < oo
B.C.: (7.31)
{ u(r,z) ~ O as y?- + z2 ~ oo, z >o
If we apply the Hankel transform of order zero to the variable r in (7.31),

we obtain the transformed problem
Uzz - p U
2
= O, z >O
B.C.: U(p,O) = F(p) (7.32)

{ U(p,z) ~O as z ~ oo
where
:1t'0{u(r,z);r ~ p} = U(p,z) (7.33)
Jeo{flr);p} = F(p) (7.34)
and we are using the result of Eq. (7.28). Clearly, the solution of (7.32)
is
U(p,z) = F(p)e-pz (7.35)
and inverting this result by means of the Hankel inversion formula, we
have
or
u(r,z) = L"" pF(p)e-pzJ (pr) dp

0 (7.36)
Example 7.5: Solve the problem described by (7.31) for the special case
where
u(r,O) = f(r) = Yr I+a

2 2
, a>O
286 Chap. 7/The Haokel Transform
Solution: Based on Eq. (7.10), we see that

Jl'o{ftr);p} = F(p) = (1/p) e-ap
Hence, substituting this result into (7 .36) leads to the Laplace transform
integral
u(r,z) = f' e-<z + a)p J0(pr) dp
= 2{J0(pr);p ~ (z + a)}
from which we conclude
1
u(r,z) = ---r:;;====;;
2 2
v'r + (z + a)
Next, we consider the problem of finding a function u(r,z) which is
harmonic in the half-space z > O, and which satisfies mixed boundary
conditions (i.e., conditions where u is prescribed over part ofthe domain
and the normal derivative of u prescribed over the remaining part of the
domain). A specific example of this kind of problem is described by
u, + (1/r)u, + Uzz = O, O< r< oo, z >O
u(r,O) = u0 , O< r< 1
B.C.: uz(r,O) = O, 1 <r< oo (7.37)
{ u(r,z) ~O as Vr 2
+r~ oo, z>O
where u0 is a constant. Such a problem might describe the electrostatic

potential of an electric field due to a uniformly charged ftat circular disc
of unit radius.
Because of the mixed boundary conditions, it is preferable to write
down the general solution of the transformed problem and invert it before
imposing the boundary conditions. Thus, taking the Hankel transform
of order zero with respect to r, we obtain the transformed equation
Uzz- p2 U =O, z >O, (7.38)
where U(p,z) is the Hankel transform of u(r,z). The bounded solution
of this equation is
U(p,z) = A(p}e-pz (7.39)
where A(p) is an arbitrary function of p.
The inversion of (7 .39) leads to the integral equation
u(r,z) = {"' pA(p)e-pz J 0(pr) dp (7.40)
for the determination of the function A(p). Now imposing on this formal
7.3 I Applications 287
solution the mixed boundary conditions in (7.37), we get the pair of

equations
L"" pA(p)Jo(pr) dp = u 0, O< r< 1 (7.41a)
L"" p A(p)J (pr) dp = O,

2
0 1 < r< oo (7.41b)
Equations of this variety are known as dual integral equations, but the
general theory concerning them goes beyond the intended scope of this
text. * However, to solve this particular pair of equations we simply start
with the observations (see Probs. 13 and 14 in Exer. 1.4)
("" sin p 7f'
Jo J 0(pr)pdp = 2' O<r<1 (7.42a)
L"" J (pr) sin p dp = O,

0 1 <r<oo (7.42b)
and hence, by comparison with Eqs. (7.41a,b), we deduce that

2u0
A( p) =- smp (7.43)
7rp 2
Thus, the solution of (7 .37) is finally given by
'
u(r,z) = -2u0 i"" e-pzJ (pr) -sin-p dp
0 (7.44)
7f' o p
7.3.2 Vibration Problems

Consider the transverse displacement u(r ,t) of a large thin membrane
which is deformed symmetrically under the action of an externa! normally
applied pressure p(r,t). lf Tis the tension in the membrane, c is a physical
constant proportional to yT and having dimension of velocity, f(r) and
g(r) denote the initial displacement and velocity, respectively, of the
membrane, then the subsequent displacements of the membrane are de-
scribed by solutions of the nonhomogeneous boundary-value problem
u,, + (1/r)u, = c- 2u11 - (1/T)p(r,t), O< r< oo, t >O
B.C.: u(r,t) finite as r~ oo (7.45)
I.C.: u(r,O) = f(r), u1(r,O) = g(r), O< r< oo
*For a discussion of such equations, see I. N. Sneddon, Mixed Boundary-Value Problems

in Potential Theory, Amsterdam: North Holland, 1966.
Taking the Hankel transform of order zero with respect to r, we are

led to the initial value problem
c
Uu + c2p2U = pTP(p, t), t >O
(7.46)
I. C.: U(p,t) = F(p), U,(p,t) = G(p)
Using standard solution techniques, we find that (7.46) has the solution
U(p,t) = F(p)cos cpt + _!__ G(p)sin cpt + cT(' P(p,T)sin[cp(t - T)]dT

cp p Jo
(7.47)
The subsequent inversion of (7 .47) then gives us the formal result for
the displacements
u(r,t) = {"' pU(p, t)J0(pr) dp (7.48)
EXERCISES 7.3
1. Finda formal solution of the heat-conduction problem
u,, + -1 u, = a -2u,, O< r<=, t >O

r
B.C.: u(r,t) finite as r~=
I.C.: u(r,O) = f(r)
2. Solve the problem described by (7.31) for the special case where
u(r,O) = f(r) = 1/(r2 + a 2 ) 312 , a> O
3. Show that the boundary-value problem
u,, + (1/r) u, + Uzz = -f(r), O< r<=, z >O
B.C.: u(r,O), =O
{ u(r,z) ---+O as y'r + l---+ =, z>O
(a) has the formal solution
u(r,z)
roo F(p)
= Jo
(1 - pe
-pz) J (pr) dp
0
(b) ln the special case

f(r) = a/(r2 + a 2) 312 , a >O
show that (a) reduces to
( ) 1[1
u r z =- og
' 2 w(r,O)
w(r,O)-
+1
1- 1og w(r,z)-
w(r,z) + 1
1]
where
w(r ,z) = J+ (z
r
2
a)
2
+1
4. Show that the steady-state temperature distribution problem

Urr + (l/r) U, + Uzz = O, O< r< oo, z >O
- Kuz(r,O) = (Q/ a )h(a 2
- r), a> O
B.C.:
{ u(r,z) ~O as yr + z 2 2
~ oo, z >o
S. Show that the potential problem for

v(r,z) = Qjy,Z + i + u(r,z)
where u(r ,z) satisfies
u,, + (1/r)u, + Uzz =O, O< r<=, -a<z<a
B.C.: u(r,a) = -Q/~
v (r ,z) -
_
~
Q
1_2 + z2
_ Q cosh zp -ap 1 ( ) d
h e o pr p
l""
vr o cos ap
6. Let u(r,z) denote the steady state temperature in a slab bounded by
O < r < oo, O < z < 1. If the face z = O is kept at u = O and the
face z = 1 is insulated except that heat is supplied through a circular
region, such that
uz(r,l) = h(c - r)
find the subsequent temperature distribution in the slab.
7. The small transverse oscillations of a heavy chain of uniforrn line
density t:r, suspended vertically from one end, are determined by the
equation
a2y a ay
>0
t:r af = g~:r ax (x a) + p(x,t), 0 <X< oo, t
where the origin of the x coordinate is taken at the position of equilibrium

of the lower end and the x axis is taken along the equilibrium position
of the chain, pointing upward. The function p(x,t) is the intensity of
the extemal transverse force. If the initial conditions are given by
iJy
I. C.: y(x,O) =O, -(xO)=O
iJt '
show that for subsequent times
y(x,t) = ~o{~ L P(p,T)sin[p(t - T)] dT;p ~ 2j~}

where
Note: The constant g is simply the gravitational constant.
7.4 Table of Hankel Transforms

The following is a short table of Hankel transforms of order zero and
their inverses.
Table 7.1 Table of Hankel Transforms
f(r) = r pF(p)Jo(pr) dp F(p) = r rf(r)l0(pr) dr
h(a - r), a<O

,_, ,
2 r lsl >I 2'f[(s + 1)/2]
p'+ 'r[(I - s)/2]
3 1/r 1/p
4 (1/r)e-"', a<O l/yp2 + a2
5 e -o' , a>O
-u2,2
6 e , a>O
7 (1/r)sin ar, a>O h(a - p)/ya2 - p2
8 (l/r 2)sin ar, a>O

8
Finite Transforms
8.1 Introduction
The integral transforms considered thus far are applicable to problems
involving either semiinfinite or infinite domains. However, in applying
the method of integral transforms to problems formulated on finite domains
it is necessary to introduce finite intervals on the transform integral. We
then find that it is possible to derive their inverses from the theory of
Fourier series. Transforms ofthis nature are called.finite transforms and
sometimes afford a more convenient method of solution than the classical
methods which often require much ingenuity in assuming at the outset
a correct solution form.
ln this chapter we will introduce finite Fourier transforms and the
finite Hankel transform, the latter being a special case of the more general
Sturm-Liouville transform.
8.2 Finite Fourier Transforms

Let us begin by considering the simplest cases of finite transforms, which
are known as finite sine or cosine transforms. The general theory of
these transforms is based on the theory of Fourier series, with which
we assume the reader is familiar.
According to the theory of Fourier series, a functionfwhich is piecewise
continuous and has a piecewise continuous derivative f' on the interval
291
292 Cbap. 8/Finite Transforms
O :::; x :::; 1T has the Fourier sine series
f(x) = 2: bn sin nx, O<x<1r (8.1)

n=J
where
bn = -2 f1T f(x)sin nx dx, n = 1,2,3, ... (8.2)

1T o
This series converges pointwise to f(x) at points where f is continuous
and to the average value llf(x+) + f(x- )] at other points. A similar
statement holds true for the Fourier cosine series
1 <X>
f(x) = 2a0 + n~J an cos nx, O<x<1r (8.3)
where
2 f1T
an = - f(x)cos nx dx, n = 0,1,2, ... (8.4)
1T o
8.2.1 Finite Sine Transform

We define the .finite sine transform of f(x) by
S.,.{f(x);n} = Fs(n) = r /(x)sin nx dx, n = 1,2,3, ... (8.5)
which is distinct from previous transforms in that the transform function

F.(n) is actually a sequence of numbers rather than a continuous function.
By comparing this expression with (8.2), we see that
n = 1,2,3, ... (8.6)
and thus deduce from Eq. (8.1) that
s; I {F.(n);x} = f(x)
n=J
F.(n)sin nx,
= ~
1T
i
O<x<1r (8.7)
ln this case Eq. (8.7) represents the inversion formula for the inverse
finite sine transform, which is generally attributed to Doetsch. *
The finite sine transform is appropriate for solving differential equations
containing only even-order derivatives. (Recall the discussion in Sec.
(3.1.) Fortunately, many of the DEs of interest fali into this category.
*See G. Doetsch, "Integration von differentialgleichungen vermittels der endlichen

Fourier-transfonnation," Math. Ann., 112, p. 52 (1935).
8.2/Finite Fourier Transforms 293
There are numerous examples of functions f(x) whose finite sine

transform can be found through elementary integration techniques applied
to (8.5). Some of these are considered in the exercises. Our interest here
is mostly concemed with the operational properties of this transform.
Iffandf' are continuous on O:::; x:::; 1T' andf" is piecewise continuous
on the sarne interval, then using integration by parts, we find
S"{f"(x);n} = r f"(x)sin nx dx
= f'(x)sin nx~~- n L" f'(x)cos nx dx

The first term on the right is identically zero and a further integration
by parts leads to
S1T{f"(x);n} = -nf(x)cos nx I~ -r n
2
f(x)sin nx dx
or, upon simplifying,

S"{f"(x);n} = - n 2Fs(n) + n[f(O) - ( -l)".f(7r)] (8.8)
Hence, we see that the finite sine transform of f"(x) depends upon the
transform of f(x) and upon the values of f(x) at the boundary points x =
O and x = 'TT'.
Although we have defined the finite sine transform for the particular
intervalO:::; x:::; 'TT', it is easy to generalize to other intervallengths. For
instance, the generalization of (8.5) and (8.7) to an interval of length p,
rather than 'TT', merely involves a scale transformation with x replaced
by 'TT'X/p. This leads to the more general sine transform
n'TT'X
SP{f(x);n} = F.(n) = l
o
P
f(x)sin- dx,
p
n = I ,2,3,... (8.9)
with corresponding inversion formula

_ 2 ~ . n'TT'x
sp I {F.(n);x} = f(x) =-L.. Fs(n)sm - , o< X < p (8.1 O)
Pn=l P
ln the sarne manner, we find that (8.8) becomes
nz'TT'z n'TT'
Sp{f"(x);n} 2 Fs(n) +- [f(O) - ( -l)".f(p)]
= -- (8.11)
p p
Still other generalizations are possible, but we will not discuss them. *
*See Chap. II in R. V. Churchill, Operational Mathematics, 3rd ed., New York:

McGraw-Hill, 1972.
294 Chap. 8/Finite Transforms
8.2.2 Finite Cosine Transform

ln a similar fashion as we did for the finite sine transform, we define
the finite cosine transform of f(x) by
C11'{f(x);n} = Fc(n) = J: f(x)cos nx dx, n = 0,1,2,... (8.12)
the inversion formula of which is

C; 1{Fc(n);x} = f(x)
O<x<1r (8.13)
The operational property concerning the second derivative of f(x) takes

the form
CTI'{f"(x);n} =- n2Fc(n) + ( -1tf'(7T) - f'(O) (8.14)
8.2).
The generalizations of these results to an interval of length p are
P n1rx
Cp{f(x);n} = Fc(n) =
lo
f(x)cos- dx,
p
n = 0,1,2,... (8.15)
c;t {Fin);x} = f(x)

~ n1rx
= -1 Fc(O) + -2 LJ Fc(n)cos - , O<x<p (8.16)
P P n=i P
and
= - n ~ Fc (n) + ( -l)Y'(p)
2
CP{f"(x);n} - f'(O) (8.17)

p
8.2.3 Applications
The steady-state temperature distribution u(x,y) in a long square bar with
one face held at constant temperature T0 and the other faces held at zero
temperature is governed by the boundary-value problem
Uxx + Uyy = 0, O<x<1r,O<y<1T
u(O,y) =O, U(1T,y) = (8.18)
B.C.: { u(x,O) =O, u(X,1T) = To
If we. apply the finite sine transform with respect to the variable x, we
arrive at
Uyy - n2 U = O, O< y < 1T
0, n even (8.19)
B.C.: U(n,O) = O, U(n,'TT) = { 2To/n, n odd
where*
U(n,y) = r u(x,y)sin nx dx (8.20)
The solution of (8.19) is readily found to be

= 2T0 sinh ny
U( n,y) . h ' n = 1,3,5, ... (8.21)
n sm n'TT
and U(n,y) = O, n = 2,4,6, .... The inversion of this result leads to
4T0 ~ sinh ny .
u(x,y) = - L. . h sm nx (8.22)
1T n=l n sm n'TT
(odd)
which is our solution of (8.18)

For a second example on the use of finite transforms, let us consider
the problem of heat conduction in a solid bounded by the parallel planes
x = O and x = 1. If the faces of the solid are thermally insulated and
the initial temperature in the solid is f(x), the subsequent temperatures
u(x, t) are solutions of
Uxx =a -2 U 1, O<x<1,t>O
B.C.: ux(O,t) =O, ux(1,t) = O,t>O (8.23)
I. C.: u(x,O) = f(x), O<x<1
2
where a is the thermal diffusivity of the material of the solid.
Because the boundary conditions in (8.23) involve the derivative of
u, we must use the finite cosine transform this time. Also, the length of
the x-interval is unity instead of 1T, so we use the general form of the
transform given by (8.15) with p = 1. ln this case the transformed
problem is
U, + a2n2 1T2U = O, t > O (8.24)
I. C.: U(n,O) = F(n)
where
U(n,t) = L1
u(x,t)cos n'TTX dx (8.25)
*For simplicity of notation, we will drop the s subscript on the transform function.
and
F(n) = f f(x)cos mrx dx (8.26)
The solution of (8.24) is

T~(
u n,t) = c~(n) e -u2n2x,2t (8.27)
the inversion of which y1elds
u(x,t) = F(O) + 2 L 2 2
F(n) e- " n x21 cos mrx (8.28)
n=l
However, using (8.26) we can express this solution in the more convenient
forrn
u(x,t) = Lf(~)d~ + ~1 [f f(~)cos mr~ d~]

2
2 2 21
e-" n x cos mrx (8.29)
EXERCISES 8.2
ln Probs. 1-10, evaluate the finite sine transforrn and finite cosine transforrn
1. f(x) = 1, 0 <X < 1T' 2. f(x) = X, 0 <X < 1T'
3. f(x) = (x/2)(1T' - x), 4. f(x) = eX, O< x < 4
O<x<1r
O<x<1 6. f(x) = x(p - x), O< x < p
7. f(x) = sin kx, 0 <X< 1T', k =f 1,2,3, .. .
8. f(x) = cos kx, 0 <X< 1T', k =f 1,2,3, .. .
9. f(x) = x(l - x 2), 10. f(x) = (x/6)(x2 - 3x + 2),
O<x<l O<x<1
11. If j(1r - x) = f(x), show that F.(n) = O when n is even.
12. Verify the identities:
(a) s'lr {f(x)cos kx;n} = UF.(n + k) + F.(n - k)]
(b) C1T {f(x)cos kx;n} = i[Fc(n - k) + Fc(n + k)]
(c) S1T {f(x)sin kx;n} = i[Fc(n - k) - Fc(n + k)]
(d) C1T {f(x)sin kx;n} = UF.(n + k) - F.(n - k)]
13. Show that
(a) s1T {f('TT' - x);n} = ( -l)n-lps(n)
(b) C1T {f(1r - x);n} = ( -ltFc(n)
14. Show that

(a) C" {f"(x);n} = - n 2Fc(n) + ( -1tf'(7r) - f'(O)
2 2
(b) CP {f"(x);n} =-n~ Fc(n) + ( -1tf'(p) - f'(O)
p
15. Show that
(a) S" {f'(x);n} = - n C"{f(x);n}
(b) C" {f'(x);n} = n S"{f(x);n} - /(0) + ( -l)j(1r)
ln Probs. 16-20, use the finite sine transform or finite cosine transform
to solve the given boundary-value problem
16. Uxx = a- 2u1 , O<x< 1,1>0
B.C.: u(O,t) = O, u(l,t) = O
I. C.: u(x,O) = 3 sin 1TX - 5 sin 47Tx
17. O<x < 10, t >O
B.C.: u(O,t) = 10, u(lO,t) = 30
I. C.: u(x,O) =O
18. 0 <X< p, l > 0
B.C.: u..(O, t) = O, ux(p,t) = O
I. C.: u(x,O) = T0 sin2(7Tx/p)
19. U 11 = c 2Uxx- 2ku,, 0 <X< 1, t > 0 (0 < k < 7TC)
B.C.: u(O,t) = O, u(l,t) = O
I. C.: u(x,O) = sin 1TX, u,(x,O) = O
20. Uxx + Uyy = O, O< X < 1T, O < Y < 1
ux(O,y) = O, ux(1r,y) = O
B.C.: {
u(x,O) = T0 cos x, u(x,l) = T0 cos2x
21. Find a formal solution of the boundary-value problem
U1 = Uxx + g(x,l), O <X < 1T
B.C.: u(O,t) = O, u(1r,t) = O
I. C.: u(x,O) = f(x)
22. A uniform string of length p is stretched tightly between two fixed
points at x = O and x = p. It is displaced a small distance e at a
point x = b, O < b < p, and released from rest at time t = O.
Starting with the equation of motion
Uxx = C- 2U 11
show that subsequent displacements are described by

2ep2 ~ 1 . mrb . mrx mrct
u(x,t) = 7T2b( p -
b) Li 2 sm -sm -cos - -
n=l n p p p
8.3 Sturm-Liouville Transforms

A Sturm-Liouville problem is a boundary-value problem of the general
form
d
dx [p(x)y'] + [q(x) + r(x)]y = O, a<x O and r(x) > O in a <
x < b, and p'(x), q(x), and r(x) are
ali continuous functions in the interval a :5 x :5 b. Here h 1 and h2 are
given constants. (Other types of boundary conditions are also possible. *)
It is known that, under appropriate conditions, a collection of nontrivial
solutions {<f>ix)}, n = 1,2,3, ... , exists corresponding to a set of values
{n}, n = 1,2,3, ... , for the parameter . We refer to the set {n} as the
eigenvalues of the problem and the corresponding nontrivial solutions
{</>n(x)} as the eigenfunctions.
Sturm-Liouville problems provide a direct method of determining a
linear integral transformation that is appropriate for a given linear boundary-
value problem, with the finite Fourier sine and cosine transforms as
special cases. ln such situations the integral transform assumes the general
form
F(n) = J: r(x)<f>n(x)f(x) dx, n = 1,2,3, ... (8.31)
where the function r(x) is called a weighting function. We call F(n) the
Sturm-Liouville (S-L) transform of the function f(x), and K(x,n) =
r(x)<f>n(x) is the kernel of the transform. t
For example, the finite Fourier sine transform (Sec. 8.2) is a S-L
transform associated with the Sturm-Liouville problem
y" + y =O, O<x<1r (8.32)
y(O) = O, y(7T) =o
*For a general discussion of Sturm-Liouville problems, see Chap. I in L. C. Andrews,

Elementary Partia[ Differential Equations with Boundary Value Problems, Orlando: Academic
Press, 1986.
tSometimes the kemel of the transform is defined as simply the eigenfunction <(J"(x).
8.3/Sturm-LiouviUe Transforms 299
Here we identify p(x) = 1, q(x) = O, and r(x) = 1. The eigenvalues

and eigenfunctions of (32) are
cf>n(x) = sin nx, n = 1,2,3, ... (8.33)
Similarly, the finite Fourier cosine transform is closely associated with
the Sturm-Liouville problem
y" + y =O, O<x<1r
(8.34)
y'(O) = O, y'(1T) =o
which has eigenvalues and eigenfunctions given by
cf>ix) = cos nx, n = 0,1,2, ... (8.35)
The most important property of the eigenfunctions of a Sturm-Liouville
problem is the orthogonality relation
J: r(x)cf>m(x)cf>n(X) dx = O, m f= n (8.36)
which is useful in the development of ao inversion formula for the S-L

transform (8.31). To obtain this inversion formula, let us assume that
the functionf(x) in (8.31) can be represented by the generalized Fourier
series
f(x) = L Cnc/>n(X), a<xm(x), m = 1,2,3, ... , and integrate the result
over the inverval a ::::; x ::::; b, we get
b ~ fb
fa r(x)cf>m(x)f(x) dx = ~I Cn a r(x)cf>m(X)cpix) dx
where we have interchanged the order of integration and summation.
Based on the orthogonality property (8.36), it is clear that all terms of
the above series vanish except for the one cortesponding to n = m.
Also, based on (8.31) we recognize the left-hand side as F(m), and thus
the expression reduces to
F(m) = Cm I: r(x)[cf>m(.~)f dx, m = 1,2,3, ...

Solving now for Cm (and changing the dummy index m to n), we find
that
n = 1,2,3, ... (8.38)
where
n = 1,2,3, ... (8.39)
Because the integrand in (8.39) is positive, it is clear that this integral

never vanishes. Finally, the substitution of (8.38) into (8.37) yields the
inversion formula
f(x) = 2: F(n) ll<f>ix)ll- 2 <f>ix), a<x<b (8.40)

n=l
lt can be shown that when f, f', and f" are all continuous functions over
the closed interval a ::s x ::s b, the series (8.40) converges to f(x) over
the open interval a < x < b (and possibly at the endpoints).
8.3.1 Generalized Finite Fourier Transforms

ln addition to the finite Fourier transforms, the next most useful transforms
are those in the category of generalized finite Fourier transforms. These
transforms are associated with Sturm-Liouville problems usually having
the sarne differential equation as in (8.32) and (8.34), but with different
boundary conditions. For example, let us consider the Sturm-Liouville
problem
y" + y = 0, 0<X O
1t can be shown that the only nontrivial solutions of (8.41) correspond
to the case > O. By setting = k 2 > O, we find that the solution of
the DE satisfying y(O) = O is
y =C sin kx
where C is an arbitrary constant. The second boundary condition in
(8.41) leads to
h sin kb + k cos kb = O (8.42)
Hence, denoting the nth solution of this transcendental equation by kn, *
the eigenvalues and eigenfunctions of (8.41) are represented by
n-- k2.
n' <f>ix) = sin knX, n = 1,2,3, ... (8.43)
ln this case we can define the generalized finite Fourier transform
T{f(x);n} = F(n) = J: f(x)sin knx dx, n = 1,2,3,... (8.44)
*The actual values k., n = 1,2,3, ... , must be determined by a numerical procedure.
8.3/Sturm-Liouville Transforms 301
To obtain the corresponding inversion formula for (8.44), we first

calculate
= !(b _ sin k11 b cos k"b)

2 kn
but since sin knb = -(kn/h)cos knb [see Eq. (8.42)], we have
111/>ix)ll2 1( 1 b) ,
= 2 b + h cos2kn n = 1,2,3, ... (8.45)
lt now follows from (8.40) that the inversion formula is

r 1
{F(n);x} = f(x)
~( F(n) ) .
= 2n~t b + (1/h)cos2knb sm knx, 0 <X< b (8.46)
ln solving differential equations with this transform, it is important

to know the transforms of derivatives of a given functionf(x). ln particular,
it can be shown that (see Prob. 5 in Exer. 8.3)
T{f"(x);n} =- k~ F(n) + [hf(b) + f'(b)]sin knb + knf(O) (8.47)
where F(n) is the transform of f(x).
Other generalized finite Fourier transforms are introduced in the ex-
ercises (see Probs. 6-8 in Exer. 8.3).
8.3.2 Applications
To illustrate the use of the generalized finite Fourier transform (8.44) in
solving partial differential equations, let us consider the heat conduction
problem
O<x<l,t>O
B.C.: u(O,t) =O, u(l,t) + ux(I,t) = O,t>O (8.48)
I. C.: u(x,O) = u0 , O<x<l
where u0 is a known constant.
lntroducing the transform
T{u(x,t);x ~ n} = U(n,t) (8.49)
it follows that
T{uxx(x,t);x ~ n} = - k~ U(n,t) (8.50)

T{u,(x,t);x ~ n} = U,(n,t) (8.51)
and
T{uo;x ~ n} = u 0 fsin knx dx = ~:(I - cos kn) (8.52)
Hence, the transformed problem is

U, + a 2k~ U = O, t > O (8.53)
I. C.: U(n,O) = (uo/ kn)(l - cos kn)
with solution
kn)e-a k~t
2
U(n,t) = (uo/kn)(l - COS (8.54)
Using (8.46) to invert (8.54), we finally deduce that
_ ~ (1 - COS kn)sin knX -a2k2t
U(X,t) - 2Uo L.J 2 e " (8.55)
n= 1 ki1 + COS kn)
EXERCISES 8.3
ln Probs. 1-4, determine the generalized finite Fourier transform defined
by (8.44) of the given function.
1. f(x) = x 2. f(x) = 1 - x
3. f(x)= h(a - x), 4. f(x) = ~
O<a<b
5. Show that
T{f"(x);n} = -k~ F(n) + [hf(b) + f'(b)]sin knb + knf(O)
where F(n) is the transform of f(x) and each kn satisfies (8.42).
6. Develop a generalized finite Fourier transform and inversion formula
associated with the Sturm-Liouville problem
y" + y =O, O<x<b
y'(O) =O, hy(b) + y'(b) =o
y" + y =O, O<x<1
hy(O) - y'(O) = O, y'(l) =o
8.4/Finite Hankel Transform 303

d
dx (eV) + exy == O, O<x<l
y(O) == O, y(l) =o
9. Use the generalized finite Fourier transform (8.44) to solve the heat
conduction problem (u 1 and u2 are known constants)
O<x<l,t>O
B.C.: u(O,t) = U~o u(l,t) + ux{l,t) = u2
I.C.: u(x,O) = U1
10. Use the generalized finite Fourier transform defined in Prob. 6 to

solve the heat conduction problem (u0 is a known constant)
O<x<l,t>O
B.C.: ux(O,t) =O, 2u(l,t) + ux{l,t) = u0
I.C.: u(x,O) =O
8.4 Finite Hankel Transform

A particular Sturm-Liouville transform that has proven useful in the
solution of certain boundary-value problems formulated in cylindrical
coordinates is the finite Hankel transform. The Sturm-Liouville problem
that motivates the introduction of this transform is
d ,?
dx (xy')- ;-Y + xy = 0, o< X< b, Jy(O)I < =, y(b) == o (8.56)
where the DE is Bessel' s equation. The eigenvalues and eigenfunctions

of (8.56) are given, respectively, by
n = 1,2,3, ... (8.57)
where k~o k 2 , k3 , , km are chosen to satisfy
n = 1,2,3, ... (8.58)
Recognizing from (8.56) that the weighting function is r(x) = x, we
define the finite Hankel transform of order v of the function f(x) by
Hv{f(x);n} = F(n) = J: xf(x)Jv(knx) dx, n = 1,2,3,... (8.59)

Based on the known relationship*
IIJ,(k,x)ll
2
= J: x[J.,(k,x)f dx
=l b2 [J.,+,(k,b)f (8.60)
it follows that the inversion formula for the finite Hankel transform
assumes the form
_, { 2 ~ F(n) J., (k,x)
H v F(n);x} = f(x) = b2 ;!:, [J.,+,(k,b)f , 0< X < b (8.61)
It is known that if f and f' are piecewise continuous functions on the

interval O ::5 x ::5 b, the Bessel series (8.61) converges pointwise to f(x)
at points x where f is continuous and to Uf(x+) + f(x-)] at points x
where f has finite discontinuities.
8.4.1 Basic Properties

There are several properties of the finite Hankel transform that are mere
consequences of the properties of Bessel functions. First, let us consider
the finite Hankel transform off'(x). Using integration by parts, we find
H.,{f'(x);n} = J: xf'(x)J.,(k,x) dx
b rb d
= xf(x)J.,(knx) - Jo f(x) dx [xJ.,(knx)] dx (8.62)
l0
The first term on the right in (8.62) for v ?: O is zero at both endpoints
[recall Eq. (8.58)], and by using the identity (see Prob. 8 in Exer. 8.4)
! [xJ,(kx)] = (v :V 1)kxJ,_,(Ia:) - (v;:, 1)w,+ 1(kx)
we see that (8.62) reduces to
H.,{f'(x);n} = (v :V 1)k,H,_ {f(x);n}

1
- (v;:, 1)k,Hv+l{f(x);n}, v> o (8.63)
Clearly, the case v = O has to be handled separately, but it leads to

(see Prob. 9 in Exer. 8.4)
*See Chap. 6 in L. C. Andrews, Special Functions of Mathematics for Engi-

neers (SPIE Press, Bellingham, Wash.; Oxford University Press, Oxford, 1998)
H 0{f'(x);n} = - Ho{;f(x);n} + knHl{f(x);n} (8.64)
ln a similar manner we can show that (see Probs. 10 and 11 in Exer.

8.4)
H v{ ~f(x);n} = ~; [ Hv-l {/(x);n} + H v+ 1 {f(x);n}], v > O (8.65)
and
Hv {xv-y'(x);n} = -knHv-l {xv-lf(x);n}, V> O (8.66)
Also, by defining the function
1 1d
g(x) = f"(x) + - f'(x) = - -d [xf'(x)] (8.67)
X X X
we find that two integrations by parts leads to

rb d
H 0{g(x);n} = Jo dx [xf'(x)]J0(knx) dx
= xf'(x)J0(knx) 1:- knxf(x)J(knx) 1:

b k
+
l
O
xf(x) [k~J(knx) + ..!!.
X
J(knx)] dx (8.68)
The first term on the right above vanishes at both endpoints provided
that f'(x) is bounded at x = O. Also, using the identity JMx) = -J1(x)
and the fact that (see Prob. 13 in Exer. 8.4)
k~ J (knx) + knX J (knx) = - k~Jo(knX) (8.69)
the remaining expressions on the right in (8.68) simplify to
Ho{g(x);n} = knbf(b)Jl(knb) - k~ J: xf(x)Jo(knx) dx

which we can write as
H 0{/"(x) + (1/x)f'(x);n} = knbf(b)J1(knb) - k~F(n) (8.70)
More generally, it can likewise be shown that (see Prob. 12 in Exer.
8.4)
Hv{f"(x) + (1/x)f'(x) - (v2/r)/(x);n}
= knbf(b)Jv+l(knb) - k~F(n) (8.71)
Other versions of the finite Hankel transform are also possible, but
we will not discuss them. The interested reader may consult Chap. 8 in
I. N. Sneddon, The Use of Integral Transforms, New York: McGraw-
Hill, 1972.
8.4.2 Applications
Consider the heating of a long circular cylinder of radius b, whose initial
temperature throughout is zero. If the temperature on the lateral surface
is described by the function f(t), the subsequent temperature u(r,t)
throughout the cylinder is govemed by
O< r< b, t >O
B.C.: u(b,t) = f(t), t>O (8.72)
I. C.: u(r,O) =O, O<rO (8.73)
I.C.: U(n,O) = O
where
U(n,t) = J: ru(r,t)J (knr) dr

0 (8.74)
and we have used (8.70). The solution of (8.73) is readily found to be
U(n,t) = a2knbJt(knb) L f('r)e-a

2
kMt - T) dT (8.75)
the inversion of which yields the formal solution
u(r,t) = :
2 2
~1 It(~b) [L f(T)e-a k~(t-
2
T) dT ]J0 (knr) (8.76)
EXERCISES 8.4
ln Probs. 1-5, find the finite Hankel transform of order zero of the given
function.
1. f(x) = 1, 0 <X< 1 2. f(x) = {~: O<x<1
1<x<2
3. f(x) = 1 - r, 0 <X< 1 O<x<1
S. f(x) = log x, O<x<1
6. Show that (0 < x < b)

HJxv;n} = (bv+l/kn)Jv+l(knb)
7. Show that (0 < x < 1)
v+ 1 }
H v{X ;n = 2(v k2+ 1) lv+ 1(kn)
n
8. Verify the identity
d [Xlv(kx)]
dx = (v + 1) kxlv-l(kx) - (v - 1) kxlv+ 1(kx),
~ ~ v>O
ln Probs. 9-12, verify the given property ofthe finite Hankel transform.
9. Ho{f'(x);n} + H0{(1/x)f(x);n} = knH1{f(x);n}

10. Hv{(l/x)f(x);n} = (kn/2v)[Hv-llf(x);n} + Hv+ 1{f(x);n}], v > O
11. Hv{Xv-I_r'(x);n} = -knHv-l{xv-I_r(x);n}, v >O
12. Hvlf"(x) + (1/x)f'(x) - (v 2/x2)/(x);n} = knbf(b)lv+l(knb) - k~F(n)
13. Given that y = J0(x) is a solution of Bessel's equation
y" + (1/x)y' + y = O
show that J 0(kr) satisfies
J'(kr) + (1/kr)J(kr) + J 0(kr) = O
14. The displacements of a thin circular membrane of unit radius are
approximately governed by the boundary-value problem
O<r<1,t>O
B.C.: u(l,t) =O
1,0<r<1/2
I.C.: u(r,O) = { O, 112 <r< 1 u,(r,O) = O
Show that the displacements are given by

~ Jl(kn/2)
u(r,t) = n-7:1 kn[JI(kn)12 lo(knr)cos knct
where J 0(kn) = O, n = 1,2,3, ...
u,, + (1/r)u, = c- 2u11 , O<r<l,t>O
B.C.: u(l,t) =O
I.C.: u(r,O) = O, u,(r,O) = 1
show that
2 ~ sin(k,ct)
u(r,t) =- LJ k 2 J (k) 10(k,r)
C n=l n I n
where J0(k,) = O, n = 1,2,3, ....

16. Find a formal solution of
u, + (1/r)u, = a- 2u,, O< r< b, t >O
B.C.: u(b,t) =O
I.C.: u(r,O) = f(r)
17. The axisymmetric motion of a viscous ftuid in concentric circles
about the axis of rotation of an infinitely long cylinder is govemed
by the boundary-value problem
u, + (1/r)u, - (1/r2)u = (1/1))U1, O< r< b, t >O
B.C.: u(b,t) = u0
I.C.: u(r,O) =O
where 11 is the coefficient of kinematic viscosity of the ftuid.
(a) By taking the finite Hankel transform of order one, obtain the
transformed problem
U, + 71k~U = -71u0bk,J0(k,b), t >O
I.C.: U(n,O) = O
(b) Solve the problem in (a) and show that its inversion leads to
u(r,t) = - 2uo
b
i
nsl
J,(k,r) (1 -
k,Jo(k,b)
e-'IJki~, J,(k,b) = o
(c) Using the identity
r= -2 ~ J,(k,r) O<r<b
n=l k,Jo(k,b)'
show that the long-time solution in (b) simplifies to u(r,t) = u0 r/b,

t ~ oo. This shows that eventually all of the ftuid will rotate as a
rigid body.
18. The equation goveming the horizontal deftection y(x,t) of a heavy
chain of uniform line density cr, fixed at x = 1, and acted on by ao
extemal transverse force of intensity p(x,t) is
a2y a ay
cr--:1 = gcr-(x-) + p(x,t), O<x< 1,t>O
ar ax ax
(a) If the chain is initially at rest under gravity, show that the horizontal
deflection at any subsequent time is given by
4 ~ lo(kn yx) r . ~r
y(x,t) = ~ r L.J [J (k )) 2 Jl q(t - T)Sm(!kn V gT) dT
CTVgn=l I n O
where
1r
q(t) =2Jo p(x,t)J0(kn yx) dx
Hint: Under the substitution x = z\ show that the goveming

equation takes the form
a2y
cr t2
1
= 4g cr az2
(a 2y
+ z1 ay)
az + p(z ,t)
2
(b) If the chain is released from rest in the position y = e(l - x),
O < x < 1, and swings freely, show that the horizontal deflections
at subsequent times are described by
~ lo(kn yx) ,A r
y(x,t) = Se f:t k!Jt(kn) cos(!~~. v g t)
9
Discrete Transforms
9.1 lntroduction
ln many engineering applications the function (signal) under consideration
is a continuous function of time that needs to be processed by a digital
computer. The only way this can be accomplished is to sample the
continuous function at discrete intervals of time. The sampled signal x*(t)
is then processed as an approximation to the true signal x(t).
The relation between a continuous function x(t) and its sample values
x(kT), k = O, 1, 2, ... , where T is a fixed interval of time, is one of
prime importance in digital processing techniques. If the Fourier transform
of x(t) is nonzero only over a finite range of the transform variable, it
turns out that the continuous function x(t) can always be recovered
(theoretically) from knowledge of only its sample values x(kT), provided
that the sampling rate is fast enough. This remarkable result is known
as the sampling theorem and plays a central role in digital processing
techniques. Functions whose transform is zero everywhere except for a
finite interval are known as band-limited waveforms in signal analysis.
Such functions do not actually exist in the real world, but theoretical
considerations of band-limited waveforms is fundamental to the digital
field. If the function under consideration is closely approximated by a
band-limited waveform, then the sampled version of the function gives
a reasonably accurate description of that function, provided the sample
values are taken at a rate that is at least twice the highest frequency
that is significant in the continuous waveform. This restriction is important
310
9.2/Discrete Fourier Transform 311
in that it eliminates the problem of "aliasing" (high frequency components

impersonating low frequencies).
9.2 Discrete Fourier Transform

Suppose we are able to sample the continuous function x(t) at the discrete
times t = kT, k = 0,1,2, ... (see Fig. 9.1).* The sampled function
x*(t) then consists solely of the sample values {x(kT)}, obtained through
our sampling procedure. If we idealize the situation by assuming the
sampling is done instantaneously, it is convenient to represent the sampled
function by
x*(t) = 2: x(t)S(t - kT) = L x(kT)S(t - kT) (9.1)

k= -oo k= -oo
where S is the impulse function (see Sec. 1.5.2). The sampled function
x*(t) is really a traio of impulses in this sense, but is treated as if it were
a continuous function of t through use of the properties of the impulse
function.
ln reality, we cannot obtain an infinite number of samples as suggested
by Eq. (9.1). Hence, in practice we must settle for N samples over a
total time duration NT, and in this case, Eq. (9.1) is replaced by the
finite sum
N-1
x*(t) = L x(kT)S(t - kT) (9.2)

k=O
9.2.1 Discrete Fourier Transform Pairs

ln engineering applications involving time and frequency domain analysis
it is customary to define Fourier transforms pairs by the expressions
X(w) = f~"" x(t)e-iwt dt (9.3)
and
x(t) = -1 f"" X(w)eiwt dw (9.4)

27T -00
*We are using the term "continuous" here merely to distinguish the function x(t) from
its sampled version x*(t). Our only real assumption regarding x(t) is that it has a Fourier
transform.
312 Chap. 9/Discrete Transforms
-31 -21 -1 o 121 31 41 t

Figure 9.1 Sampled function
or, in terms of frequency f = w/2TT,

X(f) = J~"" x(t)e-Z1rift dt (9.5)
and
(9.6)
Equations (9.5) and (9.6) as transform pairs have a certain appeal because
the constant 1/2Tr has been absorbed. ln our discussion to follow, we
will henceforth convert to these definitions of Fourier transforms in order
to be consistent with the majority of the literature on discrete transforms.
Let us begin by taking the Fourier transform of the sampled function
x*(t) given by (9.2); thus, we obtain
X*(f) = ~~ f~"" x(k1)(t 2

- k1)e- -rrift dt
N-1
= 2: x(k1)e- 2-rrifkT (9.7)
k=O
At this point X*(f) is a continuous function of frequency f, and represents

our approximation to the true frequency function X(f) associated with
the continuous function x(t). If one is interested in a few frequency
samples of X*(f) taken at various values of f, then (9.7) is convenient
to use. Owing to the periodic nature of the complex exponential, however,
we find that X*(f) is a periodic function with period 1/1, the reciprocai
of the sampling interval. Hence we can obtain frequency information-
only up to this value and then the function repeats itself. If we desire
to calculate X*(f) at N points, it is usually best to choose these N points
evenly spaced over one period. Thus, we pick the discrete frequencies
f = j/ NT, j = O, 1, ... ,N -1, which cover one period. At this point it is

convenient to introduce the simplified notation
x(k) = x(kT), X(j) = X*(j / NT)
so that (9. 7) becomes
N-1
X(j) = 2: x(k)e-Z7Tiik/N, j = 0,1, ... ,N-1 (9.8)
k=O
which is called a discrete Fourier transform (DFT).

ln order to derive the discrete inverse Fourier transform, we first
make the observation that (m = 0,1, ... ,N-1)
2:
N-1
e-27Ti(k-m)j/N = { O, k :f m (9.9)
j=O N, k =m
the verification of which is left to the reader (see Prob. 6 in Exer. 9.2).
Now multiplying (9.8) by e27Timi/N and summing from j = O to N -1, we
find
N-1 N-1 N-1
L X(j)e27Timi/N = L x(k) L e-27Ti(k-m>i/N
j=O k=O j=O
= Nx(m) (9.10)
where we have made use of (9.9). Changing the free index m to k in
(9.10), we have derived the inverse transform relation
N-1
x(k) = ..!_ L X(j)e 27Tiik/N, k = 0,1, ... ,N-1 (9.11)
N j=O
Equations (9.8) and (9.11) constitute what are called DFT pairs. ln
some areas of the literature, however, the factor 1/N that appears in
(9.11) is found in (9.8) instead. Thus, once again, we caution the reader
to carefully check the definitions when using properties of these transforms
found in other reference sources.
Example 9.1: Find the DFf of the four-point sequence {x(k)} = {1,1,0,0},
and then find the inverse DFT of the result.
Solution: The DFT in this case is given by

3
X(j) = 2: x(k)e-i7Tjk/Z, j = 0,1,2,3,
k=O
which leads to
3
X(O) = L x(k) = 2
k=O
3
X(l) = L x(k)e-i"'k 12 = 1 - i
k=O
3
X(2) = L x(k)e-i"'k = O
k=O
3
X(3) L x(k)e-3iTrk/2 = 1 + i
k=O
Using these four values to calculate the inverse DFT, we have
3
x(k) =!L X(j)eiTrjk/2, k = 0,1,2,3
4 k=O
from which we recover the original sample values
x(O) = 1, x(l) = 1, x(2) = O, x(3) = O
As already pointed out, one of the major distinctions in the DFT as
compared with the continuous Fourier transform is that both {x(k)} and
{X(j)} form periodic sequences with period N; that is,
x(k + N) = x(k), ali k (9.12)
and
X(j + N) = X(j), allj (9.13)
the verification of which we leave to the exercises (see Prob. 8 in Exer.
9.2). To geometrically display this periodicity property, the sample values
are often represented as equally spaced marks around a circle as depicted
in Fig. 9.2.
Other than the periodicity property, the operational properties of the
DFT correlate very closely with the corresponding operational properties
Figure 9.2 Sample values arranged on a circle

of the continuous Fourier transform. For example, the DFT has the shift
properties
N-1
2: [x(k)e-21Tiik/N] e-21rimk!N = X(j + m) (9..14)
k=O
and
N-1
2: x(k + m)e -21Tijk/N = X(j)e21Tiim/N (9.15)
k=O
The convolution theorem takes the form

N-1 .. N-1
-1 2: X(j)Y(j)e 21T'Jk/N = (x * y)(k) = 2: x(m)y(k - m) (9.16)
N j=O m=O
and Parseval' s relation is given by

N-1 N-1
~i~ lXU)iZ = ~o lx(k)!Z (9.17)
The proofs of these properties, and some additional ones, are left to the
exercises.
Example 92: Compute the convolution ofthe two four-point sequences

{x(k)} = {1,2,3,4} and {y(k)} = {5,6,7,8}
Solution: The convolution is defined by
N-1
(x * y)(k) = 2: x(m)y(k - m)
m=O
To evaluate this convolution in a simple fashion, we display the two

sequences around two concentric circles as shown in Fig. 9.3. The
8 5
7 5 8 6
6 7
k= o k=1
Figure 9.3 Convolution
value of the convolution at k = O is then

(x * y)(O) = (1)(5) + (2)(8) + (3)(7) + (4)(6) = 66,
while at k = 1 the outer circle is rotated counterclockwise by one
position so that
(x * y)(l) = (1)(6) + (2)(5) + (3)(8) + (4)(7) = 68
Continuing in this fashion, we obtain the remaining values
(x * y)(2) = (1)(7) + (2)(6) + (3)(5) + (4)(8) = 66
and
(x * y)(3) = (1)(8) + (2)(7) + (3)(6) + (4)(5) = 60
9.2.2 Fast Fourier Transform

The DFT and its inverse have been defined, respectively, by
N-l
X(J) = L x(K)W1K, J = 0,1, ... ,N -1 (9.18)

K=O
and
N-l
x(K) = _!_ 2: X(J)W- 1
\ K = O,l, ... ,N-1 (9.19)
N J=O
where
(9.20)
Because of subscripting, etc., in the material to follow, it is more convenient
in this setting to use capital ndices J and K rather than lower case as
before. To actually perform the indicated operations in a direct computation
ofthe DFT would necessitate N operations for each sample output, where
an operation is defined as one complex multiplication and addition. The
complete DFf of a signal of length N would then require N 2 operations.
For applications involving large N, the required computational time can
often be prohibitive, even on a high-speed computer. For real-time analysis
the computational time may be prohibitive even for moderate size N.
ln 1965, Cooley and Tukey published an algorithm that, under certain
conditions, significantly reduces the number of computations required
to compute a DFf. * This algorithm, which has become known as the
fast Fourier transform (FFf), is one of the most important contributions
*J. W. Cooley and J. W. Tukey, "An algorithm for the machine calculation ofcomplex
Fourier series," Math. Comp., 19, pp. 297-301, April 1965.
in this century to the field of numerical computations. Whereas N 2 op-

erations are required for computing a DFT, the FFT only requires N
log 2N operations, when N is a power of 2. Today there are several
variations of this algorithm, but they are ali based on the sarne principie.
ln order to derive the original form of the algorithm let us assume
that N = 2m, where m is a positive integer. Using base 2, the ndices
J and K can be represented by
J = 2m-llm-l + ... + 211 +lo= (Jm-IJilo) (9.21)
and
K = 2m- 1Km-l + ... + 2K 1 + Ko = (Km-IKIKo) (9.22)
where each li and Ki is either zero or one. For example, the number
121 in base 2 is represented by
121 = 26 1 + 25 1 + 24 1 + 231 + 22 0 + 20 + 1 = (1111001)
Thus if we introduce the notations
and
X(J) = XUm-l ... lo)
then Eq. (9.18) takes the form
X(Jm-I"Jo) = 2: ... 2: x(Km-IKo)WJ(2m-lKm-l+...+Ko) (9.23)
Ko Km-l
Certain simplifications can now take place by making use of the following
theorem and its obvious generalizations.
Theorem 9.1. For N = 2m, the function w = e-ZTTi/N satisfies the identity
Proof: Consider
WJ2m-lKm-l = exp[ -27Ti(2m-llm-l ~~- + lo)zm-!Km-1]
= exp[- 1Ti(2m- 1lm-l + ... + lo)Km-11
From properties of the complex exponential, it follows that
exp(- 7Ti2m- 1lm-1Km-l) =1
exp(- 1Tizm-zlm-zKm-1) = 1
and thus we deduce our result

wnm-IKm-1 = exp(- 7TiJoKm-1) = WJo2m-IKm-l

The result of Theorem 9.1 is readily generalized to
although we will not present the proof. Hence, starting with the innermost
summation in (9.23), we obtain successively
AI(JoKm-2Ko) = 2 x(Km-IKo)WJo2m-IKm-l
Km-1
.,. J 1K m-3 K o) --
A 2( Jo " A I(JoK m-2 K O) w<2JI +Jo)2M-2Km-2
L.J
Km-2
Km-p
.,. J ) _ ""A
A m (Jo (.,. J K )W(2m-1Jm-I+ ... +Jo)Ko
m-1 - L.J m-1 Jo m-2 O
Ko
(9.25)
The last result Am(J0 Jm_ 1) is our desired output, but in bit-reversed
order; that is,
(9.26)
This bit-reversal is an inherent property of the algorithm. To obtain the
output in proper order, the input could first be scrambled. Either way,
the scrambling of input or output data is simply part of the entire FFT
process.
Example 9.3: Use a four-point FFT to compute the Fourier transform

of
{x(K)} = {1,2,3,4}
Solution: We first note that the process consists of three steps of
calculations:
I
AI(JoK0) = L x(K 1K 0 )W 21oK1

K1=0
I
A2(Joll) = L AI(JoKo)W2J1Ko+JoKo
Ko=O
X(Jilo) = A2(JoJ1)
where W = e -i'IT/2 = i.
Step 1:
I
A1(00) = L x(K10) = x(OO) + x(lO) = 1 + 3 = 4

K1=0
I
A1(0l) = L x(K11) = x(01) + x(11) = 2 + 4 = 6

K1=0
I
A 1(10) = L x(K 10)W 2K' = x(OO)- x(lO) = 1- 3 = -2

K1=0
I
A1(11) = L x(K11)W 2
K' = x(Ol)- x(ll) = 2- 4 = -2
K1=0
Step 2:
I
Az(OO) = L A1(0K0) = A1(00) + A1(01) = 4 + 6 = 10

Ko=O
I
Az(Ol) = L AI(0Ko)W 2Ko = A1(00) - A 1(01) = 4 - 6 = -2

Ko=O
I
Az(IO) = L AIOKo)WKo = A1(10) + A1(ll)i = -2 + 2i

Ko=O
I
Az(ll) = L A10Ko)W 3Ko = A1(10) - A1(11)i = -2- 2i

Ko=O
Step 3:
X(O) = X(OO) = A2(00) = 10
X(l) = X(01) = A2(10) = -2 + 2i
X(2) = X(lO) = A 2(01) = -2
X(3) = X(ll) = A2(11) = -2- 2i
Hence, the desired transform is
{X(J)} = {10, -2 + 2i, -2, -2- 2i}
EXERCISES 9.2
ln Probs. 1-4, find the DFf of the given four-point sequence and then
find the inverse DFf of the result.
1. {1,2,3,4} 2. {0,1,0,-1}
3. {1,2/3,1/3,0} 4. {1 ,2, 1,2}
320 Chap. 9/Discrete Transfonns
5. Calculate the convolution of the sequences

(a) from Probs. 1 and 2.
(b) from Probs. 3 and 4.
6. Prove that
N-1
~ e-21Ti(k-m)j/N = { O, k =f m
j=O N, k =m
Hint: Sum the series as a geometric series.

7. Prove the linearity properties
N-1
(a) ~ [x(k) + y(k)]e-Z1rijk/N = X(j) + Y(j), j = 0,1, ... ,N-1
k=O
N-1
1
(b)- ~ [X(j) + Y(j)]e 21Tijk/N = x(k) + y(k), k = 0,1, ... ,N -1
N j=O
8. Prove the periodicity properties

(a) x(k + N) = x(k), all k
(b) X(j + N) = X(j), allj
9. Prove the shift properties
N-1
(a) ~ [x(k)e-21TijkfN]e-21Timk/N = X(j + m)
k=O
N-1
(b) ~ x(k + m)e-21Tijk/N = X(j)e21Tijm/N
k=O
10. Prove the convolution theorem

N-1 N-1
(a) _!_ ~ X(j) Y(j)e 21Tijk/N = ~ x(m)y(k - m)
N j=O m=O
(b) From (a), deduceParseval's relation

N-1 N-1
~ j~ IXU)I 2
= ~o lx<kW
11. Verify Parseval's relation for the sequences given in Exam. 9.1.
12. A sequence {x(k)} is said to be even if x(N - k) = x(- k) = x(k).
We say the sequence is odd if x(N - k) = x(- k) = - x(k). Prove
the following properties concerning even and odd sequences involving
DFT pairs:
(a) {X(j)} is even if and only if {x(k)} is even.
9.3/The Z Transform 321
(b) {X(j)} is odd if and only if {x(k)} is odd.

(c) {X(j)} is real and even if and only if {x(k)} is real and even.
(d) {X(j)} is real and odd if and only if {x(k)} is pure imaginary and
odd.
(e) {X(j)} is pure imaginary and even if and only if {x(k)} is pure
imaginary and even.
(0 {X(j)} is pure imaginary and odd if and only if {x(k)} is real and
odd.
ln Probs. 13-16, use the FFT to evaluate the Fourier transform of the
given four-point sequence.
13. {1,1,0,0} 14. {0,1 ,0, -1}
15. {1,2/3,1/3,0} 16. {1 ,2, 1,2}
9.3 The Z Transform

Communication systems using pulse modulation techniques, where severa!
different messages may be interlaced on a time-sharing basis, rely on
sample values of the signals taken at regular spaced intervals of time.
ln this case the sample values constitute the full available information
about the signals. Another area of application based on sample values
includes control systems in which feedback is applied on a basis of sample
values of some quantity which is to be controlled. The Z transform is
utilized heavily in these areas of application, although it has also proven
useful in other applications as well.
The Z transform is an operation that converts a discrete signal into
a complex frequency domain representation. ln this regard it is the discrete
analog of the Laplace transform, and thus will have many properties in
common with the Laplace transform.
Since the Laplace transform is ordinarily associated with causal func-
tions, i.e., functions which are identically zero for t <O, we will consider
only that class of functions in the present discussion. Let x(t) be a
continuous function on t ~ O and of exponential order c0 The related
sampled function x*(t) then has the representation
x*(t) = L x(nT)8(t - nT) (9.27)

n=O
where T denotes the time interval between samples. ln developing prop-

erties of the Z transform, we normally assume that the sampled function
consists of an infinite number of samples as indicated in Eq. (9.27). Of
course, in practice there may be only a finite number of these sample
values that are actually available for processing. The Laplace transform
of the above sampled function formally leads to
;t'{x*(t);p} = ~ L~ e-pt x(nT)8(t - nT) dt

n=O
= L x(nT)e-pnT (9.28)
n=O
It is now notationally convenient to make the substitution

z = epT (9.29)
which transforms the axis of convergence Re(p) of the Laplace
CT
transform into a circle in the z plane. Hence, in the z plane we have the
new function
X(z) = ;t'{x*(t);p} = X*(p)
and (9.28) becomes
X(z) = 2: x(nT)z-n, lzl > errT (9.30)

n=O
We say that X(z) is the Z transform of x(nT), but which we will also
call the Z transform of the continuous function x(t). Using the principie
of analytic continuation, we can deduce that X(z) is an analytic function
of z outside the circle lzl = errr. The series on the right in (9.30) is the
Laurent series expansion of X(z) about the origin. Because the spacing
T between samples has no effect on developing properties and on the
use ofthe Z transform, it is conventional to set T = 1. Also, corresponding
to the notation ;t'{f(t);p} = F(p) that is used in the Laplace transform,
we introduce the similar notation
Z{x(t);z} = X(z) = 2: x(n)z-n (9.31)

n=O
in which we have set T = 1.
Remark: Although our discussion of Laplace transforms is usually

based on the possibility of discontinuities in the time function, we normally
assume in the use of the Z transform that the function x(t) is continuous.
However, if x(t) has discontinuities between the sampling points, the
sampled function x*(t) will be insensitive to these points. Also, if a
discontinuity in x(t) occurs at kT, we can simply replace x(kT) by x(kT+).
9.3.1 Evaluating Z Trans/orms

Ifthe function to be transformed consists of only its sample values {x(n)},
then we can immediately write down its Z transform. For instance, if
{x(n)} = {2,1,0,5}, then
X(z) = 2 + z- 1 + 5z- 3
When we start with a continuous function x(t), the Z transform normally
leads to an infinite series that in many cases can be summed exactly.
Let us consider some examples.
Example 9.4: Given that x(t) = 1, find Z{l;z}.
Solution: We first note that the sample values of x(t) are given by
x(n) = 1, n = 0,1,2, ...
Hence,
00 1
Z{l;z} = L z-n = 1
n=O - Z
-1' lzl > 1
where we have summed the series as a geometric series. We also
write this result as
Z{1 ;z} = z/(z - 1), lzl > 1
Example 9.5: Find Z{d;z}, a =I= O.
Solution: Here x(n) = an, n = 0,1,2, ... , so that
Z{a 1 ;Z} = L anz-n = L (a/zt

n=O n=O
or
Z{a 1 ;z} = z/(z - a), lzl >a
Example 9.6: Find Z{t;z}.
Solution: Since x(n) = n, n = 0,1,2, ... , we get
Z{t;z} = L nz-n
n=O
=z 2: nz-(n+ I)
n=O
d 00
""
-z-L.JZ -n
dz n=O
from which we deduce (using the result of Example 9.4)

Z{t;z} = z/(z - 1)
2
, lzl > 1
Example 9.7: Find Z{l/f(t + l);z}

Solution: Here x(n) = 1/f(n + 1) = 1/n!, and thus
oo -n
Z{1/f(t + l);z} = 2: ~
n=On.
= e 11z, all z
Other Z transforms can be derived in a similar manner. A table of

Z transforms involving elementary functions can be found at the end of
this chapter.
Based upon the definition of the Z transform, it is clear that whenever
two functions x(t) and y(t) have the sarne values at t = n, they will have
the sarne Z transform. Hence the Z transform is not one-to-one. For
example, if y(t) = x(t) + sin 1rt, then x(t) and y(t) have the sarne Z
transform. There is, however, a one-to-one relationship between the
samples {x(n)} and the transform function X(z). Only when the conditions
of the sampling theorem are satisfied will we be able to uniquely determine
the time function x(t) from its sample values {x(n)}.
9.3.2 Properties of the Z Transform

Many of the operational properties of the Z transform are simple discrete
analogs of the properties of the Laplace transform. For example, directly
from definition we have the linearity property
Z{Ctx(t) + C2 y(t);z} = CtX(z) + C2Y(z) (9.32)
where C1 and C2 are arbitrary constants.
The first shift theorem takes the form
Z{x(t + l);z} = zX(z) - zx(O) (9.33)
To derive this property, we note that
Z{x(t + l);z} = 2: x(n + l)z-n

n=O
= z 2: x(n + l)z-<n+ll
n=O
=z 2: x(n)z-n
n=t
the last step of which follows the change of index n ~ (n - 1). By

adding the n = O term to the sum and subtracting it back out, we obtain
(9.33). The second shift property is given by
Z{x(t - a)h(t - a);z} = z-a X(z) (9.34)
9.3).
If Z{x(t);z} = X(z), then
Z{tx(t);z} = L nx(n)z-n
n=O
d
L
00
- z- x(n)z-n
dz n=O
Z{tx(t);z} = - z X'(z) (9.35)
Also, it is easily shown that (see Prob. 12 in Exer. 9.3)
Z{e-a1x(t);z} = X(eaz) (9.36)
Additional properties are taken up in the exercises.
9.3.3 Inverse Z Transforms

There are several ways in which inverse Z transforms can be evaluated.
First, if it is known that
00
X(z) = L CnZ-n (9.37)

n=O
then we immediately deduce*

x(n) = z-t {X(z);n} = Cn' n = 0,1,2, ... (9.38)
ln some cases we try to represent X(z) in terms of functions whose
inverse transforms are known. Partial fraction techniques are particularly
useful in such cases.
Example 9.8: Find the inverse Z transform of

z+3
X(z) = z- 2
*Because the uniqueness of the Z transform and its inverse extends only to the sample
values {x(n)} and not to x(t), we will consider {x(n)} as the inverse Z transform of X(z)
rather than x( t).
Solution: By writing
z 3
X(z) = -- + --
z-2 z-2
it follows from Exam. 9.5 and the second shift property (9.34) that
x(n) = z- 1{X(z);n}
= r + 3. 2n-lh(n - 1)
1, n=O
== { 5r- 1, n = 1,2,3, ...
Example 9.9: Find the inverse Z transform of

3
X(Z) = z - 9l + 5z - 1
4z3 - sl + 5z - 1
Solution: Since the numerator is of the sarne degree as the denom-

inator, we first divide and then apply a partial fraction expansion to
obtain
2
X(z) = !+ -7z + 15z/4 - 3/4
4 4z3 - sl + 5z - 1
1 4 9/4 5/16
= 4 - z - 1 + z - 1/2 + (z - 1/2)2
Therefore, with the aid of entries 13, 15, and 5 in the table and the
second shift property, we deduce that
1
x(n) = 4[h(n)- h(n-1)] - 4h(n-1) + 4 9(1)n-l
2 h(n-1)
2
+ 5 (n-1) (1)n-
2 h(n-1)
16
= ~h(n) + [ 03 : 5
n)Gr- ~]h(n-1). n=0,1,2, ...
An altemate way of finding the inverse transform in Exam. 9.9 is to

use long division to obtain
z3 - 9l + 5z - 1 1 7 _1 41 _2
4z3 - 8z2 + 5z - 1 = 4- 4z - 16z + ...
from which we conclude x(O) = 1/4, x(l) = -7/4, x(2) = -41/16, ....
This method, of course, provides no general formula for x(n) but it does
eliminate the need to factor the polynomial in the denominator.
The convolution theorem of the Z transform is (see Prob. 21 in Exer.

9.3)
z- {X(z)Y(z);n} = (X*Y)(n)
1
(9.39)
n
where
(X*Y)(n) = 2: x(k)y(n - k) (9.40)
k=O
Example 9.10: Use the convolution theorem to find the inverse Z trans-
form of the product
(z - 2)(z - 3)
Solution: By setting
X(z) = - z- , Y(z) = -z-

z- 2 z- 3
we can immediately obtain the individual inverse transform relations
z- {X(z);n} = x(n) = 2n
1
z- {Y(z);n} = y(n) = 3n
1
Thus, from the convolution theorem (9.39), we have

n
k=O
= 3n (~)k
k=O 3
We recognize this last finite series as a geometric series, from which
we deduce the result
n = 0,1,2, ...
The inverse Z transform can be evaluated in many cases by using a

complex inversion formula similar to that used in evaluating inverse
Laplace transforms. To derive this formula, we retum to Eq. (9.37)
written as
1
X(z) = Co + CJZ- + c 2z- 2 + + CnZ-n + ...
and then multiply both sides by zn-l to get
X(z)zn-1 = CoZn-1 + CJZn-2 + ... + CnZ-1 + ...
We then integrate both sides of this expression around a closed contour
lzl = R such that X(z) is analytic on and outside the closed contour. On
the right ali terms will vanish except the one involving z-I, and we are
left with the inversion formula
Cn = f
2~; izi=R X(z)zn-J dz, n = 0,1,2,... (9.41)
Thus, if X(z)~- 1 has poles at z = ako k = 1,2, ... ,N, within the circle
lzl = R, we deduce that
N
x(n) = cn = L Res{X(z)zn- 1;ak}, n = 0,1,2, ... (9.42)

k=l
Example 9.11: U sing (9 .42), obtain the inverse Z transform of
X(z) =z+ 3
z- 2
Solution: We first note that

X(z)zn-1 = (z + 3)zn-J
z- 2
has simple poles at z = O and z = 2 when n = O, and only the
simple pole z = 2 when n :::: 1. Thus, as in Exam. 9.8, we find
x(O) = Res{O} + Res{2} = -23 + 25 = 1

x(n) = Res{2} = 5 zn- 1
, n = 1,2,3, ...
EXERCISES 9.3
ln Probs. 1-10, evaluate the Z transform of the given function.
2
2. t
3. cos bt 4. sin bt
S. h(t - 1) 6. h(t) - h(t - 1)
7. h(t - 1) - h(t - 2) s. oI t)h(t - 1)
9. h(t - k), k = 1,2,3, ... 10. a 1- 1 h(t - 1)
11. Show that
Z{x(t - a)h(t - a);z} = z-ax(z), a >O
12. Show that

Z{e-a1X(t);z} = X(eaz)
13. Use Eq. (9.35) to find the Z transform of tecr.
14. Use the result of Problem 12 to find the Z transform of ec1cos bt.
ln Probs. 15-20, find the inverse Z transform of the given function
without the use of residues. If necessary, use the table of Z transforms
on p. 334.
z l+1
15. -2--1 16. -2--1
z + z -
2
17. z 18. z
(z + l)(z + 3) (z + l)(z + 3)
l +1
19. (z - 1)3 20. (zz + l)(z - 2)
21. Verify the convolution theorem

n
z- 1{X(z)Y(z);n} = L x(k)y(n - k)
k=O
where X(z) and Y(z) are, respectively, the Z transforms of x(n) and
y(n).
22. Verify the initial and final value theorems
(a) lim X(z) = x(O)
lzl-->oc
(b) lim (z - l)X(z) = lim x(t) = x(oo)
z-->oo
ln Probs. 23-30, use Eq. (9.42) to evaluate the inverse Z transform of

the given function.
23 _z_ l+l
24. -2--1
"l+l z -
25.-----
l 26. z
(z + l)(z + 3) (z + l)(z + 3)
28
(i + 1)(z - 2)
zsinb
30
i - 2z cos b + 1
9.4 Difference Equations

Difference equations arise in a variety of applications. ln particular, they
are closely related to differential equations and their theory basically
parallels that of differential equations.
To get some idea of how differences and derivatives are related, let
us start by considering the definition of derivative,
'( ) . y(t + T) - y(t)

y t = 1T->0
1m T
Rather than passing to the limit, suppose we now think of T as fixed,

i.e., T = 1. We then define
y(t) = y(t + 1) - y(t) (9.43)
which is called thefirst-order difference. Using (9.43) as a definition, we
can formally define a second-order difference by
2y(t) = [y(t)]
= [y(t + 1) - y(t)]
= [y(t + 2) - y(t + 1)] - [y(t + 1) - y(t)]
or
2
y(t) = y(t + 2) - 2y(t + 1) + y(t) (9.44)
Continuing in this fashion, we can construct 3y(t), 4y(t), and so forth.

Although we can interpret t as a continuous variable in (9.43) and
(9.44), it is generally regarded as a discrete variable n in most applications.
Also, for notational convenience it is customary to adopt the notation
y(n) = Yn, y(n + 1) = Yn+ 1,. .. . Such notation is suggestive of the
recurrence formulas that occur in many applications, e.g., the power
series method of solving ordinary differential equations. Adopting this
notation, we will henceforth define Z transforms by
Z{yn;z} = L YnZ-n (9.45)

n=O
ln solving difference equations by the method of Z transforms, we

will find the results of the following theorem helpful. [Note that theorem
9.2 is simply special cases of (9.33).]
Theorem 9.2. lf Z{yn;z} = Y(z), then

(a) Z{Yn+I;z} = z[Y(z) - Yol
(b) Z{Yn+z;z} = z2[Y(z) - Yol - ZY1
9.4/Difference Equations 331
n=O
n---i>n-1
n=l
= Z [~ YnZ-n - Yo]
n=O
= z[ Y(z) - Yo]
Similarly,
n=2
from which (b) follows.

To compare one of the subtle distinctions between differential equations
and difference equations, let us start with the first-order linear differential
equation
y' - y =O, y(O) = 3 (9.46)
whose solution is
y(t) = 3e 1 (9.47)
Replacing y' in (9.46) with the difference (9.43), we get the corresponding
difference equation
Ay- Yn =O, Yo =3 (9.48)
or
Yn+l - 2yn =O, Yo = 3 (9.49)
Applying the Z transform to (9.49) leads to
z[Y(z) - 3] - 2Y(z) = O
332 Chap. 9 /Discrete Transforms
from which we obtain*

Y(z) = 3z/(z - 2)
The inversion of this function leads to the solution
Yn = 3. zn, n =O, 1,2, ... (9.50)
One of the things that this simple example is illustrating is that the
"natural base" for exponential functions in difference calculus is 2 rather
than e as in ordinary calculus. There are several other correspondences
of this nature that become evident by pursuing how the di:fference operator
ll works on various types offunctions. However, here we will not develop
such correspondences since our primary interest is in solving difference
equations. t
Example 9.12: Use the Z transform to solve the difference equation

Yn+2 + 3yn+l + 2yn =O; Yo = 1, Yt = 2
Solution: Application of the Z transform yields

l[Y(z) - 1] - 2z + 3z[Y(z) - 1] + 2Y(z) = O
or
(z
2
+ 3z + 2)Y(z) = z2 + 5z
from which we find
l + 5z
Y(z) = (z + l)(z + 2)
4z 3z
z+ 1 z+2
Inverting this last result leads to the desired solution
Yn = 4(-lt- 3(-2t
= <- 1r[4 - 3 zn1. n = 0,1,2, ...
EXERCISES 9.4
ln Probs. 1-10, use the Z transform to solve the given di:fference equation.
1. Yn+2 - 5yn+l + 6yn = O; Yo = O, Yt =
2. Yn+2 + 2Yn+l - 3yn = O; Yo = O, Yt = 1
on the other hand, the Z transform applied directly to (9.47) yields Y(z) = 3z/(z - e).
tFor more discussion of finite differences, see K. S. Miller, An lntroduction to the
Calculus of Finite Differences and Difference Equations, New York: Dover, 1960.
9.4/Difference Equations 333
3. 3yn+2 - 5Yn+l + 2yn = O; Yo = 1, Y1 = O

4. Yn+z - 4Yn+l + 4yn = O; Yo = 1, Y1 = 4
S. Yn+2 - 2Yn+l + 2yn = O; Yo = O, Y1 = 1
6. Yn+2 - 5yn+l + 6yn = 4n; Yo = O, Y1 = 1
1. Yn+2 - 1Yn+l + lOyn = 16n; Yo = 6, Y1 = 2
8. Yn+2 - 5yn+l + 6yn = 2n + 1; Yo = O, Y1 = 1
9. Yn+2 + 4Yn+l - 5yn = 24n - 8; Yo = 3, Y1 = -5
Yo = O, Y1 = O
11. Use the Z transform to solve the variable-coefficient difference equation
(n + l)yn+ 1 - Yn = O; Yo = 1
12. Show that
3y(t) = y(t + 3) - 3y(t + 2) + 3y(t + 1) - y(t)
13. Prove that
2
Z{Yn+3 ;z} = l[Y(z) - Yo] - Z Y1 - ZY2
ln Probs. 14 and 15, use the result ofProb. 13 to solve the given difference
equation.
14, Yn+3 - 2Yn+2 - Yn+l + 2yn = O; Yo = O, Y1 = 1, Y2 = 1
15. Yn+3 - 2Yn+2 - Yn+l + 2yn = n + r;
2
Yo = O, Y1 = 1, Y2 = 1
16. The sequence ofnumbers 0,1,1,2,3,5,8,13,21, ... , where each number
after the first is the sum of the two preceeding numbers, is called a
Fibonacci sequence. Finda formula for Yn such that y0 = O, y 1 = 1,
Y2 = l, Y3 = 2, ....
9.5 Table of Z Transforms

The following is a short table of Z transforms for reference purposes.
Table 9.1. Z Transforms
Z{x(t);z} = 2: x(n)z-n = X(z)

n=O
No. x(t), t ~o x(n), n = 0,1, ... X(z)
z
h(n)
z- I
z
2 n
(z- V
z(z + I)
3
(z - 1)3
z
4 a', a> O a"
z- a
5 ta', a > O na"
z(z - cos b)
6 cos bt cos bn
z2 - 2z cos b + I
z sinb
7 sin bt sin bn
z2 - 2zcosb +I
z(z - cosh b)
8 cosh bt cosh bn
z2 - 2zcoshb + 1
z sinh b
9 sinh bt sinh bn
z2 - 2z cosh b + 1
z(z - e'cos b)
10 e'"'cos bt
ze'sin b
11 e''sin bt
12 h(t - 1) h(n - 1)
z- I
13 h(t) - h(t - I) h(n) - h(n - I)
14 h(t - k), k = 1,2, ... h(n - k)

t 1
(z- I)
1
15
z- a
16
1
-h(t- 1)
1
-h(n-1) log--
z
t n z- 1
17
f(t + 1) n!
Bibliography
Listed below are some of the standard references on integral transforms.

Bach of these references supplies numerous additional references, including
many of the related research papers.
R. N. Bracewell, The Fourier Transform and Its Applications, New York:
McGraw-Hill, 1978.
E. O. Brigham, The Fast Fourier Transform, New Jersey: Prentice-Hall,
1974.
H. S. Carslaw and J. C. Jaeger, Operational Methods in Applied Math-
ematics, Oxford: Oxford University Press, 1941.
R. V. Churchill, Operational Mathematics, New York: McGraw-Hill,
1972.
B. Davies, Integral Transforms and Their Applications, New York:
Springer-Verlag, 1985.
D. F. Elliott and K. Rao, Fast Transforms: Algorithms, Analyses, Ap-
plications, New York: Academic Press, 1982.
W. R. LePage, Complex Variables and the Laplace Transform for En-
gineers, New York: Dover, 1980.
J. W. Miles, Integral Transforms in Applied Mathematics, London: Cam-
bridge University Press, 1973.
A. Papoulis, The Fourier Integral and Its Applications, New York:
McGraw-Hill, 1963.
I. N. Sneddon, Fourier Transforms, New York: McGraw-Hill, 1951.
335
336 Bibliography
I. N. Sneddon, The Use of Integral Transforms, New York: McGraw-

Hill, 1972.
C. J. Tranter, Integral Transforms in Mathematical Physics, London:
Methuen, New York: John Wiley & Sons, 1951.
E. C. Titchmarsh, lntroduction to the Theory of Fourier Integrais, 2nd
ed., Oxford: Oxford University Press, 1948.
H. J. Weaver, Applications of Discrete and Continuous Fourier Analysis,
New York: John Wiley & Sons, 1983.
Appendix A
Review of
Complex Variables
ln this appendix we present a very brief review of some of the most

basic concepts and theorems from complex variables that have direct
bearing on material in this text. There is no attempt for completeness,
and proofs of the theorems will not be provided. The reader who desires
a more thorough coverage of these topics can consult any introductory
text on complex variables.
A complex variable is one that can be represented by
z =X + iy (A.l)
where x and y are real variables and i = y'-=1. The variable xis called
the real part of z, denoted by x = Re(z), and y is called the imaginary
part of z, also written as y = lm(z).
If f is a function depending on the complex variable z, we say that
f is a complex function. Such functions can always be represented in
the form
f(z) = u(x, y) + iv(x, y) (A.2)
where both u and v are real functions.
If the derivative f' (z) exists at all points z of a region D in the complex
plane, then f(z) is said to be analytic in D. Necessary and sufficient
conditions for f(z) to be an analytic function are the Cauchy-Riemann
equations
au av au av
-=-
ax ay' ay
= ax
(A.3)
337
338 Appendix A
where these partial derivatives are assurned to be continuous. A point

where f(z) is not analytic is called a singular point.
Integration of a cornplex function is actually line integration along
sorne rectifiable curve (or path, or contour), in the cornplex plane. The
rnost fundamental integral theorern concerning analytic functions bears
the narne of Cauchy.
Theorem A.l (Cauchy's integral theorem). Ifj(z) is analytic on and inside

a closed path C in the cornplex plane, then
Tcf(z)dz =O
If there is sorne neighborhood of a singular point z = a of a function

j(z) throughout which f(z) is analytic, except at the point itself, then
z = a is called an isolated singularity. Every functionf(z) has a residue
at each of its isolated singular points, the value of which rnay be zero.
ln general, the value of the residue is the value of the integral
1
- . i, f(z) dz
27Tl Yc
around any closed contour containing the isolated singularity. Multiple-
valued functions have singularities that are called branch points. For
example, the function f(z) = z 112 has a branch point at z = O.
One kind of isolated singular point is called a pole. If f(z) is not
finite at sorne point z = a, but the product (z - a)mf(z) is analytic for
sorne integer m, we say that f(z) has a pole of order m at z = a. A
pole of order one is also called a simple pole. The residue for a function
having a pole of order m at z = a is given by
1 dm-1
Res{f(z); a} = ( - 1)' lirn d m-I [(z - a)mf(z)] (A.4)
m . z-->a z
If f(z) = p(z)/q(z) has a sirnple pole at z = a, then also
p(a)
Res{f(z); a} = q'(a) (A.5)
The evaluation of an integral around a closed contour containing a finite

nurnber of poles of a given function f(z) relies on the following residue
theorem.
Theorem A.2 (Residue theorem). Iff(z) is a single-valued function which

is analytic on and inside a closed path C, except for finitely-many isolated
Appendix A 339
singular points a 1 , a 2 , , aN inside C, then
Tcf(z)dz = 27Ti i-, Res{f(z); ak}
The evaluation of real definite integrais can often be accompiished

by using the residue theorem together with a suitabie function f(z) and
a suitabie contour C, the choice of which in some cases may require a
certain amount of ingenuity. ln the evaiuation of such integrais it is
frequently necessary to reiy on some of the following theorems.
Theorem A.3. lf, on a circular are C with radius R and center at z=

O, we have zf(z) ~O uniformly as R~ oo, then
lim
R-.oo
r
JcR
f(z)dz =o
Theorem A.4. lf, on a circular are CP with radius p and center at z =

a, we have (z - a)f(z) ~O uniformiy as p ~O, then
Iim r
p--.o Jc.
f(z)dz = o.
Theorem A.S. If, on a circular are CP with radius p and center at

z = a, and intercepting at an angle a at z = a, f(z) has a simple pole at
z =a, then
lim r f(z) dz = ai Res{f(z); a}
p-->O Jc"
where a > O if integration is counterclockwise and a < O otherwise.
Theorem A.6. lf, on a circular are CR with radius R and center at

z = O, we have f(z) ~ O uniformly as R ~ oo, then
(a) lim
R-->oo
r
JcR
eim'f(z) dz =o (m >O)
provided that CR lies in the first and/or second quadrants.
(b) lim
R-->oo
r
JcR
em'f(z) dz =O (m > O)
provided that CR lies in the second and/or third quadrants.

AppendixB
Table of
Fourier Transforms
Listed below are short tables of Fourier transforms, cosine transforms,

and sine transforms.
Table 8.1. Fourier Transforms
No. f(t) = yl2;

I
21T
r.
-=
e-'"F(s)ds F(s) = -yl2;
I r.-=
e'''f(t )dt
y21r (s)
2 -I A i sgn(s)
3
tz + az a>O J;i
2a
I -11
--e
4
z + az)z ' a>O - J'IT2 2a
is e-11
5
f- a 2
t(f + a 2 )'
a>O ;J~ (2e-ll - l)sgn(s)
6 iul yl2; (s + a)
7 e-11, a>O ~;a

sz + az
8 te-11, a>O A (s2

2ais
+ az)z
340
Appendix B 341
Table 8.1. (continued)
No. f(t) = .r .
y'2;
21!' -oo
e-"'F(s)ds F(s) =
r
y'2;
21!' -00
t!"f(t)dt
9 ltle-lrl, a>O
ji 2
a -s
2
1T (s2 + a2)2
-u2t2 J -s2/4a2
10 e ' a>O Via e
II cos(t2/2) ~ [cos(s2/2) + sin(s2/2)]
12 sin(t2/2) ~ [cos(s2/2) - sin(s 2/2)]
2a 3
13 e-ltllv'i[cos(at/VZ> + sin(alti/VZ)],
y; s4 + a4
a>O
_ , sin t I
14 e11 - y'2; arctan(2/s 2)
t 211'
15 sgn(t) ji!'ITS
16 t sgn(t) -/,! 1T s2
17 h(l - ltl)
ji sins
1T s
I ( sin s/2r
18 (I - ltl)h(l - ltl)
\/211' s/2
19 h(t) A [a(s) + ~J
I
20 8(t - a) --e'"'
y'2;
21 Jtj-a, O<a<l
/!;. f(l - a) . 'l!'a
isl'-a StnT
22 ltl-a sgn(t ), O<a<l 1

)2 :;;
f(l - a) 'll'a
JsJ' a COST
i"
23 P.(t )h(l - Iti) v~Jn+I/2(S)
342 Appendix B
Table 8.2. Cosine Transforms
No. f(t) =A r Fc(s)cos st ds Fc<s> = J;r f(t)cosstdt
Y2Tr 8(s)
l l
2
Vi Vs
3
P + a2' a>O J"!!
2a
e-
4 e -at , a>O A a a2
s2 +
5 te-a', a>O
A a2-s2
(s2 + a2)2
-a212 l _,2,..2
6 e ' a>O aVi e
7 e- cos at, a> O j"'J. as2 + 2a 3
Tr s 4
+ 4a 4
3 2
e- sin at, A2a - as
8 a>O
s4 + 4a 4
9 cos(f/2) 0 [cos(i/2) + sin(s2/2)]
lO sin(f/2) 0 [cos(s2/2) - sin(s 2/2)]
t-1' O<p<l /2f(p)

ll ; 7COS(7rp/2)
(a 2 - f)P- 112h(a - t), 2- 112f(p + l/2)(a/sYJ.(as),

12
a>O p > -l/2
Table 8.3. Sine Transforms
No. /(t) =A r Fs(s)sin st ds Fs(s) = J!;. r /(t)sin st dt
J~ sgn(s)
l
-
l l
2
Vi Vs
J~ e-
t
3 a>O
f + a2'
Appendir B 343
Table 8.3. (continued)
No. f(t) ==A r Fs(s)sin st ds Fs(s) = Ar f(t)sin st dt
1 -Se -as
--
4 + a2)2, a>O
(t2 yZ; a
5 a>O -(1 - e-as)

fz1
t(t2 + a 2)' a2
6 e -ai , a>O JI s
:;;: s2 + a2
7 te-a', a>O JI7T

2as
(s2 + a 2f
te -a2t2' _s_ e-s2;<a2

8 a>O
2yZa 3
9
1 -ar
-e
t
J! arctan(s I a)
lO e-ai cos at, a>O JI s

:;;: s + 4a4
4
3
ll e-ar sin at, a>O JI 4

2
2a s
:;;:s +4a 4
12 tp-1/2 , O<p<1 fz 7 f(p) .

S10(7Tp/2)
2p-3/2aPsl-pf(p - l/2)Jp(as ),
13 t(a 2 - fY- 312 h(a - t), a>O
p > 1/2
Appendix C
Table of
Laplace Transforms
Listed below is a short table of Laplace transforms and their inverses.
Table C.l. Laplace Transforms
No. F(p) = r e -p'f(t)dt f(t)

1 f+i>O
=~ . eP'F(p)dp
7Tl c-=
-1
p
1
2
pi
r-1
3 -p"1 (n = 1,2,3, ... ) (n- 1)!
1 1
4
v'P \(;i
1
5 PJ/2
2\/t/rr
1 ,.,-I
6 x>O
p;' f(x)
1
7 --
p-a
e"'
1
8 te"'
(p- a)2
344
Appendix C 345
Table C.l. (continued)
No. F(p) =r e-P'f(t)dt f(t) =-.

27T
f+'~
I
1 c-i~
eP'F(p)dp
9
(p - a)"
(n =1, 2, 3, ...) 1n-!eat
(n - 1)!
I 1x-leat
10 x>O
(p- aY' f(x)
eat - ebt
11 a~ b
(p - a )(p - b) ' a-b
p ae"' - beh'
12 a~ b
(p- a)(p- b)' a-b
1
13 - sin at
pz + az a
p
14 cos at
Pz + az
15 Pz _ az -1 sm
. h at
a
p
16 pz _ az cosh at
17
(p - a)z + bz !b e"'sin bt
p-a
18 e"'cos bt
(p - af + b2
1
19
p(pz + az) 20 - cos at)
a
20
pz(pz + az) ~(at - sin at)
a
21 1 .
a (sm at - at cos at
)
(pz + a2)z 2 3
p t
22 a sin at
(pz + az)2 2
p2
23 1a (sm
. at + at cos at)
(pz + az)z 2
p 1
24 az ~ bz ;;---z<cos at - cos bt)
(pz + az)(pz + bz) -a
25
P4 + 4a4 4~ 3 (sin at cosh at - cos at sinh at)
346 Appendix C
No. F(p) = r e-P'f(t)dt /(t) = -'1TI.

2
l Lc+i~
c-IQO
. eP'F(p)dp
26 l . . h
a SID at SID at
2 2
27
2~3 (sinh at - sin at)
I
28 a (cosh at - cos at)
2 2
29 10(at)
30 a> O, v> O
31
yP(p- a)
l
32
33 8(1 - a)
l
34 -e-op
p , a>O h(t - a)
35 (t - a)h(t - a)
I -
36 a>O r= cos(2y'at)
y7Tt
_I_ -a/p I -
37 a>O . c= sin(2y'at)
PVPe ' yTTa
.!_ -a/p
f)(v-ll/2 _
38 a> O, v> O
p "e , (~ J,_,(2y'at)
39 _a_ e-a2f4t
2v:;f
40
1 -
41 - e-v'p a~ O erfc(a/2yi()
p ,
42 sin at
arctan(a/p)
Appendix C 347
No. F(p) ==r e-'f(t)dt 1

f(t) = - .
f+l=.
27TI c-=
e'"F(p)dp
1
43 - arctan(a/p) S(at)
p
2 2 _I_ e-2t4a2
44 erfc(as ),
e" a>O
a v;,
1 2 2
45 - e" P erfc(as ), a2=0 erf(t/2a)
p
46 eP erfc( v'ap), a>O

Va
1TVt(t + a)
47 VP1 e"P erfc(VaP ), a2=0

y:;,(t +a)
48 erf(a/yp) _!_ sin(2aVt)

1TI
49 VP1 e"1Perfc(v'aJP), a2=0 I e-2,,i;;,

v'1Tt
Index
Abel, N., 240 asymptotic formula for, 281

Abel's integral equation, 242 differential equation for, 27, 281, 303,
Absolutely integrable, 39, 85 307
Addition formula, graph of, 22, 26
Bessel functions, 279 Jacobi-Anger expansion. 27
Airy stress function, 152-154 modified, 25-29, 159
Analytic function, 337 of the first kind, 21-25
Andrews, L. C., 6, 38, 103, 108, 136, 226. of the second kind, 25
231' 298, 304 properties of, 24-25
Applications, series representation of. 21
discrete transforms, 330-333 Bessel equation, 27, 281, 303, 307
double Fourier transform, 121 eigenfunctions of, 303
finite transforms, 294-296, 301, 302, 306 eigenvalues of, 303
Fourier transforms, 102-161 Beta function, 15
Hankel transform, 285-290 Bibliography, 335-336
Laplace transform, 218-244 Biharmonic equation, 142. 147-149, 152
Mellin transform, 266-272 elasticity, 152-154
Archimedes, 37 viscous fluid, 147-149
Asymptotic formula, Bilateral Laplace transform, 214-216
Bessel functions, 281 Binomial coefficient, 14
complementary error function, 19-20, Boundary value problems, 103-112, 131-
214 138
mixed, 137-138
Beam theory, 106 of the first kind, 132
Bernoulli, D., 23, 37, 127, 274 of the second kind, 132
Bessel, F. W., 22 singular, 103
Bessell functions, 21-29, 60, 66-67, 74- Branch point, 338
76. 78, 84-85, 101' 149, 155, 159,
161, 185-189, 197-198, 200, 209, 220, Cauchy, A. L., 38
223, 239, 244, 250-251. 269-270, 274- Cauchy integral theorem, 338
290 Cauchy-Riemann equations, 149, 337
addition formula for, 279 Causal function, 91, 163
349
350 Index
Characteristic function, 4-5, 156-161 applications, 330-333

Churchill, R. V., 168, 202, 293 fast Fourier, 316-319
Complementary error function, 17, 19-20, Fourier, 3ll-319
120, 122, 124, 169, 171' 180, 183-184, inverse, 313, 328
187-188, 210, 214, 216, 220, 224, 230, properties of, 315-316, 324-325
233-234, 236, 261 sampling continuous functions, 3ll
asymptotic series for, 19-20, 214 table of, 334
Complex variables, 337-339 z, 321-334
analytic function, 337 Doetsch, G., 292
branch point, 338 Double Fourier transform, 98-101, 121,
Cauchy's integral theorem, 338 274-275
Cauchy-Riemann equations, 337 Dual integral equations, 287
poJes, 338 Duhamel's principie, 122-123
residue theorem, 338-339 Duplication formula, 13, 15, 66
Convolution formula for
cosine transform, 82, 138
discrete Fourier transform, 315 Eigenfunctions, 108, 298-299
Eigenvalues, 108, 298-299
double Fourier transform, 99-100
Fourier transform, 78-85, tt5, ll7, 129, Elasticity, 151-155
Airy stress function, 152
137, 149
Laplace transform, 194-198, 225, 230 compatibility condition, 152
Mellin transform, 249, 266 equations of motion, 151
equilibrium, 152-154
sine transform, 82
Z transform, 327 stress components. 151, 153-154
Cooley, J. W., 316 Elliptic integral, 83
Cosine integral, 180, 261 Energy spectrum, 79
Cosine integral representation, 43-46 Entire function, 16
Cosine transform, 51-53, 62-63, 67-68, Erdelyi, A., 98
ll8-ll9, 247, 250 Error function, 16-21, 124, 182-184, 187-
inverse, 51 188, 200, 210, 214, 216, 220, 224,
233-234, 236
D'Aiembert, J. 127 complementary. See complementary er-
ror function
D'Aiembert solution, 127
graph of, 17
Delta function. See Impulse function
properties of, 16-17
Delta sequence, 36
Derivative properties of Euler formulas, 49, 126, 233
cosine transform, 62-63 Euler, L., 7, 23, 106, 151
double Fourier transform, 99 Exponential integral, 180, 216, 261
finite Fourier transforms, 293-294 Exponential order, 166
finite Hankel transform, 304-305
Fourier transform, 61-64 Fast Fourier transforms, 316-319
generalized finite Fourier transform, 301 Fibonacci sequence, 333
Hankel transform, 280-281 Final value theorem,
Laplace transform, 173-177 Laplace transform, 212
Mellin transform, 249 Z transform, 329
Difference, 330 Finite transforms, 291-309
Difference equations, 330-333 applications, 294-296, 306
Diffusion equation. See Heat equation cosine, 294
Dirac delta function. See Impulse function Fourier, 291-298
Dirac, P. A. M., 33 generalized Fourier, 300-301
Dirichlet condition, 114, 132 inverse, 292, 294, 300, 302, 304
Dirichlet problem, 132 Hankel, 303-309
axisymmetric, 285-286 sine, 292-293
half-plane, 133-135 Sturm-Liouville, 298-309
infinite strip, 136-137 Fluid ftow. See Hydrodynamics
Dirichlet series, 252 Fourier integral representations, 38-49,
Discrete transforms, 310-334 106-107
Index 351
cosine, 43-46 axisymmetric, 288

sine, 43-46 finite interval, 230-231, 295-296, 301-
Fourier integral theorem, 40 302
exponential form of, 50 infinite line, 114-117
proof of, 47-49 infinite rectangle, 121
Riemann-Lebesgue lemma, 47 semiinfinite tine, 118-120, 229-230
Fourier, J., 37, 113 Heaviside expansion theorem, 205
Fourier series, 38, 299 Heaviside, 0., 162
cosine, 292 Heaviside unit function, 29-31, 41, 66,
generalized, 299 88-89, 104, 172-173, 222, 232, 236,
sine, 292 276, 328
Fourier's Iaw, 113 Helstrom, C. W., 157
Fourier transform, 2, 49-161, 208 Hilbert-Schmidt kernel, 108
applications, 102-161 Hilbert transform, 91-97
convolution integral, 78-85, 99-100, Hydrodynamics, 141-151
115, 117, 129, 137-138, 149 continuity equation, 141
cosine, 51-53 equation of motion, 141
definition of, 2, 50 ideal fluid, 141
discrete, 311-334 incompressible flow, 141
fast, 316-319 irrotational flow, 143-144
inverse, 50, 59 sources and sinks, 141
method of stationary phase, 97-98 steady flow, 142, 147-149
multi pie, 98-101, 121, 274-275 stream function, 142
properties of, 58-65 surface waves, 144-147
residue methods, 67-74 two-dimensional flow, 141
sine, 5!-53 viscous fluid, 141
table of, 340-343 vorticity, 141
Fredholm integral equation, 108
Fresnel integrais, 17-21, 58, 131, 189, 234 Impulse function, 31-36, 85-91, 122, 150,
graph of, 18 189, 226, 232, 236, 311, 321-322
series representation of, 20 definition of, 32
sifting property of, 33
Gamma function, 7-16, 66, 166, 170, 247, Impulse response function, 224-227
250-251, 256-259, 264 Influence function, 196
argument negative, 8 Initial value problems, 221-228
graph of, 9 Initial value theorem,
poJes of, 8 Laplace transform, 211
properties of, 13 Z transform, 329
Gaussian random variable, 20, 157 Integral equations, 4-5, 107-112, 238-244
Generalized finite Fourier transform, 300- convolution type, 108, 238-244
301 dual, 287
Generalized functions, 85-91 Fredholm, 108
See also Impulse function Hilbert-Schmidt kernel, 108
Goodier, J. N., 152 of the first kind, 107, 238-239
Gradshteyn, I. S., 159 of the second kind, 108, 240
Green's function, 226 Volterra type, 108, 238-244
Inversion formula, 3
Hankel transform, 3, 274-290 cosine transform, 51
applications, 285-290 discrete Fourier transform, 313
definition of, 3, 275 double Fourier transform, 98
finite, 303-309 finite cosine transform, 294
inverse, 275 finite Hankel transform, 304
properties of, 278-282 finite sine transform, 292
table of, 290 Fourier transform, 50, 59
Harmonic function, 132 generalized finite Fourier transform, 301
Harrington, W. J., 259 Hankel transform, 275
Heat equation, 4, I 13-125, 229-231 Laplace transform, 164-201
352 Index
Melln transform, 246 Multiple Fourier transform. See Double

sine transform, 51 Fourier transform
Z transform, 328
Neumann condition, 114, 132
Jacobi-Anger expansion, 27 Neumann problem, 132, 135-136, 144
half-plane, 135-136, 144
Kellogg, O. D., 132 Normal random variable, 20, 157
Kernel, 2 Null function, 190
Hilbert-Schmidt, 108
Kramers-Krnig relations, 94 One-sided Green's function, 226
Operational properties of
Laguerre polynomials, 181, 189
discrete Fourier transform, 315-316,
Laplacian, 113
324-325
Laplace, P. S., 38, 162
Fourier transform, 58-65
Laplace's equation. See Potential equation Hankel transform, 278-282
Laplace transform, 2, 162-244
Laplace transform, 170-181, 211-213
applications, 218-244
Mellin transform, 248-251
convolution integral, 194-198, 225, 230 Z transform, 324-325
definition of, 2, 164 Orthogonality relation for
~xistence theorem for, 167
Hilbert transform, 97
mverse, 164, 190-200 Sturm-Liouville problem, 299
of perodic functions, 177-179
properties of, 170-181, 211-214
table of, 344-347 Papoulis, A., 157
two-sided, 214-216 Parseval's relation for
Legendre, A. M., 7 cosine transform, 84
Legendre polynomials, 75 discrete Fourer transform, 315
Leibnitz's rule, 267 Fourier transform, 79
LePage, W. R., 168, 216 Hankel transform, 279
Lerche's theorem, 190 Hilbert transform, 97
Linear integral equations. See Integral sine transform, 84
equations Partia! fractions, 192-194, 325-326
Linearty property of Periodic functions, 177-179
discrete Fourer transform, 320 Periodic sequences, 314
Fourer transform, 59 Piecewise smooth, 39
Hankel transform, 278 Poisson integral formula, 134
Laplace transform, 170, 191 Poisson, S. D., 162
Mellin transform, 248 Potes, 338
Z transform, 324 Potential equation, 131-138, 143-144
Little, R. M., 151 axisymmetric, 285-286
Lovitt, W. L., 108 half-plane, 133-136, 143-144
infinite strip, 136-137
MacFarlane, G. G., 263 rectangle, 294-295
Maclaurn, C., 37 semiinfinte strip, 137-138
Mellin transform, 3, 245-273 Potential function, 132, 143
applications, 262-272 Principal value, 2
convolution integral, 249, 266 Probability and statistics, 4-5, 156-161,
definition of, 3, 246 265-267
inverse, 246, 256-259, 263 characteristic functions, 4-5, 156-161
polar coordinates, 259-260 density functions, 156, 159-161, 265-
properties of, 248-251 267
residue methods, 254-273 distribution function, 156
table of, 273 statistical moments, 156
Miller, K. S., 332
Mixed boundary value problem, 286 Rectangle function, 30
Modified Bessel functions, 25-29, 159 Residue methods,
graph of, 26 Fourier transform, 67-74
See also Bessel functions Laplace transform, 200-210
Index 353
Mellin transform, 254-262 signum function, 34

Z transform, 327-328 sine integral, 43, 177, 216
Residue theorem, 338-339 Spectrum, 86
Riemann, G., 263 Springer, M. D., 265
Riemann-Lebesgue lemma. 47 Stationary phase method, 97-101, 146-147
Riemann zeta function. See Zeta function Statistical moments, 156
Robin's condition, 114 Stream function, 142
Ryzhik, I. M., 159 Sturm-Liouville problem, 298, 300, 303
Sturm-Liouville transform, 298-309
Sampled function, 311-312, 321 applications, 301-302, 306
Scaling property of finite Hankel, 303-309
Fourier transform. 60 generalized finite Fourier, 300-301
Hankel transform, 278 inverse, 300, 304
Laplace transform, 171
Mellin transform, 248 Tautochrone problem, 240-242
Schwartz, L., 85 Taylor, B., 37
Self-reciprocal function, 56, 278, 284 Thompson, W. E., 265
Shift properties of Timoshenko, S., 127, 152
discrete Fourier transform, 315 Titchmarsh, E. C., 40
double Fourier transform, 99 Translation property of
Fourier transform, 60-61 Laplace transform, 172, 232
Laplace transform, 171-173, 191 Mellin transform, 248
Mellin transform, 248 Tukey, J. W., 316
Z transform, 324-325 Two-sided Laplace transform, 214-216
Shivamoggi, B. K., 141
Signum function, 34, 52. 88, 90-91, 110, Unit step function. See Heaviside unit
148-149 function
Sine integral, 43, 177, 216
Sine integral representation, 43-46 Vibrating
Sine transform, 51-53, 63, 119-120, 208 beam, 127-129, 233-234
inverse, 51 membrane, 287-288
Singularity, string, 125-127, 231-232
branch point, 338 Volterra integral equation, 108, 238, 241
isolated, 338
poJe, 338 Watson, G. N., 264
Sneddon, I. N., 213, 275, 276, 287, 306 Watson's lemma, 213-214
Special functions, 6-36 Wave equation, 4, 125-131, 231-234
beta function, 15 infinite line, 125-127
Bessel functions, 21-29 semiinfinite line, 231-232
cosine integral, 180, 261 Whittaker, E. T., 264
elliptic integral, 83 Widder, D. V., 216
error functions, 16-21
Fresnel integrais, 17-21, 58, 131, 189, Zemanian, A. H., 85
234 Zeta function, 253, 263-264, 268-269
gamma function, 7-16 properties of, 264
Heaviside unit function, 29-31 Z transform, 321-334
impulse function, 31-36 applications, 330-334
Laguerre polynomials, 181 inverse, 325-328
Legendre polynomials, 175 properties of, 324-325, 329
rectangle function, 30 table of, 334
by Larry C. Andrews and Bhimsen K. Shivamoggi
Integral Transforms
Integral transform methods provide effective ways to solve a variety of problems arising in the
engineering, optical, and physical sciences. This concise, easy-to-follow reference text
introduces the use of integral transforms, with a detailed discussion of the widely applicable
Laplace and Fourier transforms.
for Engineers
The text is suitable as a self-study for practicing engineers and applied mathematicians, as well
as a textbook for students in graduate-level courses in optics, engineering sciences, physics,
and mathematics. The math is straightforward. In most sections, applications relevant to
engineers and applied scientists are used in place of formal proofs. Numerous examples,
exercise sets, illustrations, and tables of transforms enhance the books usefulness as a
teaching tool and reference.
Contents: Special functions. Fourier integrals and Fourier transforms. Applications involving
Fourier transforms. The Laplace transformation. Applications involving Laplace transforms.
The Mellin transform. The Hankel transform. Finite transforms. Discrete transforms.
Bibliography. Appendix A: Review of complex variables. Appendix B: Table of Fourier
transforms. Appendix C: Table of Laplace transforms. Index.
Larry C. Andrews
Larry C. Andrews and Bhimsen K. Shivamoggi are professors of mathematics at the University
of Central Florida. Andrews is also a member of the Department of Electrical and Computer
Engineering and associate member of the Center for Research and Education in Optics and
Lasers (CREOL) and the Florida Space Institute. Shivamoggi is also a member of the
Department of Physics at U.C.F.
P.O. Box 10
Larry C. Andrews
Bellingham, WA 98227-0010
ISBN-10: 0819432326
ISBN-13: 9780819432322
SPIE Vol. No.: PM178

Integral Transforms For Engineers

Diunggah oleh

Informasi Dokumen

Judul Asli

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

Integral Transforms For Engineers

Diunggah oleh

Hak Cipta:

Format Tersedia

Integral Transforms for Engineers

Integral Transforms for Engineers

University of Central Florida

SPIE-The Intemational Society for Optical Engineering

Copyright 1999 The Society of Photo-Optical lnstrumentation Engineers

(Originally published in 1988 by Macrnillan Publishing Company, New York.)

All rights reserved. No part of this publication may be reproduced or distributed

Printed in the United States of America.

4.3 Basic Operational Properties 170

lN RECENT YEARS, INTEGRAL TRANSFORMS have become essential WOrking

differential equations and contour integration techniques from complex

The classical methods of solution of initial and boundary value problems

By an integral transform, we rnean a relation of the forrn*

J~= K(s,t)f(t) dt = F(s) (0.1)

such that a given functionf(t) is transforrned into another function F(s)

the resulting transforrn

is called the Laplace transform. When

J~ L"" f(t)sin st dt = F(s) (0.6)

J~ L"' f(t)cos st dt = F(s) (0.7)

L"" tlv(st)f(t) dt = F(s) (0.8)

The Hankel transform arises naturally in solving boundary value problems

fon(s,t)F(s) ds = f(t) (0.10)

where H(s, t) is another kernel and D is the domain of s. Such a result

f(x) = u(x) - J~=k(x, t)u(t) dt (0.15)

and certain ordinary differential equations. lnteresting applications of

the probability density function p(x) by the Fourier transform relation

C(t) = f~ooeitx p(x) dx (0.16)

While the Fourier transform is suited for boundary-value problems,

J: U(T)k(t - T) dT == f(t), t >o (0.17)

ln addition to the transforms mentioned above, there are other less

1.2 The Gamma Function

Re(z) >O* (1.1)

later termed the Eulerian integral of the second kind by A. M. Legendre

Hence, by repeated use of the recurrence formula (1.2), we see that

1.2.1 Analytic Continuation for Re(z) < O

f(z) = f(z + k) z :f 0,-1, ... ,-k + 1 (1.6)

1.2.2 Additional Properties

Figure 1.1 Graph of the gamma function

f(z) = 2 Loo e-" u2z-l du, Re(z) >O

whereas the substitution t = log(l I u) in ( 1.1) yields *

f(z) = Jori (log~)z-1 du, Re(z) >O (1.9)

Example 1.1: Evaluate the integral I

Solution: By making the substitution t = x3 , we have

* By log x, we mean the natural logarithm, often denoted by ln x.

and by comparison with Eq. (1.1), we deduce that

Loo e-x\4 dx = (l/3)f(5/3)

An interesting relation involving the product of two gamma functions

f(x)f(y) = 2 J: e-u2U2x-l du 2 L"" e-v v y-l dv

= 4 L"" L"" e-<u +u2lu2x-tv y-I du dv

where x and y are real variables. The change of variables

Solving for the integral gives us the relation

Solution: Comparing this integral with that in Eq. (1.11), we see

which follows from repeated applications of the recurrence formula

f(z)f(l - z) =2 tan2' - 16 dO (1.12)

where z may be real or complex. The change of variable u

f(z)f(l - z) = 1=o Iuz-+ udu,

where we (temporarily) restrict the real part of z as indicated in (1.13).

Figure 1.2 Contour for evaluating the integral in (1.13)

The integral of f over the entire closed path leads to

tfWd' = LJWd' + LRf(u)du + LR!Wd' + J:f(ue 2";)du (1.14)

and since fW has a simple pole at ' = - 1, we find in the limit as

Computing the residue, we obtain

However, ( -l)z-t = e"i<z-tJ = -e";Z, and hence (1.15) becomes