Escolar Documentos
Profissional Documentos
Cultura Documentos
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
Bhimsen K. Shivamoggi
ISBN-13: 9780819432322
SPIE Vol. No.: PM178
Integral Transforms for Engineers
Integral Transforms
for Engineers
Larry C. Andrews
Bhimsen K. Shivamoggi
=-==
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
Larry C. Andrews
Bhimsen K. Shivamoggi
Orlando, Florida
March, 1999
vii
Preface
1
2 Introduction
(0.3)
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
* 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
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
where lv(x) is the Bessel function of the first kind (see Sec. 1.4), and
the Mellin transform
(0.9)
..
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
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
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
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
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
which reduces to
f"/2
f(x)f(y) = 2f(x + y) Jo cos2x-o sin2y-o dO (1.10)
f"/2
Example 1.2: Evaluate the integral Jo sin40 cos 50 dO.
rC 1) = r(1 + ~) = ~ ~ ~rG)
2
1.2 /The Gamma Function 11
1
~/2
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
lim
R-+oo
f fWd'
JcR =O
i
oo
--du=-- O< Re(z) < 1 (1.16)
o 1 +U SD7TZ '
(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.
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)
Show that
(b) (- !/ 2
) = ( ~~(~~~)!, n = 0,1,2, ...
(c) (a) =
k
f(a
k!f(a- k + 1)'
+ 1)
k = 0,1,2, ...
1.,.;-z
14.
Joo
2
a e ax-x
2
dx = 2 v 7Tea
2
Hint: 2ax - x2 = -(x - a)
2
+ a
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, ...
Show that
(a) B(x,y) = f(x)f(y)/f(x + y)
Hint: Let t = cos 28.
erf(z)
2
= y:;;: Jo e-
r 12
dt (l.I9)
erf(z) =
2 Jor ~o ----;;!t
y; <-Ir
00
2
n dt
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.
C(x) = f 2
COS(7Tt /2) dt (1.27)
erf(x)
-2 2 X
and
y = C(x)
0.5
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
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 2: O, b >O
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/";)]
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 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
22 Chap. 1/Special Functions
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
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
1.4/Bessel Functions 23
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.
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
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
24 Chap. 1/Special Functions
m + n is even.
For reference purposes, a short list of the basic identities of the Bessel
function follows.
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
1.4/Bessel Functions 25
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
26 Chap. 1/Special Functions
ln(x)
3 X
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)
n= -oo
show that
(a) cos(x sin O) = J0(x) + 2 2: 1 n(x)cos(2n0)
2
n=l
= ~ { 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
10.
I 2
x- J 2(x) dx = - 2x2 J,(x) - 3J,(x)
3
1 + 1x Jo(x) + 3li Jo(x) dx + C
3
11. Using the series representation (1.34), show that
d
(b) dz [z-vlv(Z)] = Z-vlv+!(Z)
1.5/Useful Engineering Functions 29
(a) 11/z(z) =
2
J
sinh z
1TZ
(b) L 112 (z) = J
2
1TZ
cosh z
h(t - a)
a t
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)
f(t)
a b t
da(t)
2s
a-s a
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.,
Recalling that
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
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
== f(a)g(a)
Since
we see that
* 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.
34 Chap. 1/Special Functions
EXERCISES 1.5
1. Show that
h(t) =i [1 + sgn(t)]
where the signum function is defined by
9. Show that
l>'(t)f(t) = f(0)8'(t) - f'(0)8(t)
!~ f~= t/Jit) dt = 1
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
(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*
and
1 JP . n1rt
bn = -
p
f(t)sm- dt,
-p p
n = 1,2,3, ... (2.4)
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
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)
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)
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
then
* 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+
2.2/Fourier Integral Representations 41
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,
where
and
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
-1
2 sin
cos sx dx = - -
7TS
s
and
B(s) = -1
7T
Jl -1
sin sx dx = O
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
* 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.
2.2/Fourier Integral Representations 43
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
.( )
SlZ = iz -
o
sin t dt
t
(2.18)
and
J""
B(s) = ;1 -=f(x)sin sx dx = O (2.20)
B(s) = -7T21""
o
f(x)sin sx dx (2.23)
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
A(s) =-
2l1r/Z cos x cos sx dx = 2 COS
(
7TS /2
)
2
1r0 1r 1 -s
and, therefore,
= -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
2.2/Fourier Integral Representations 45
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
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
46 Chap. 2/Fourier Integrais and Fourier Transforms
= {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
2 ioo s sin sx
3
-x
(b) e cos x = - 4 ds, x>O
7T o s + 4
<e>J<x> = (si:xr
2.3/Proof of the Fourier Integral Theorem 47
!~ 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
lim JP f(t)eit dt = O
-+OO -p
. 1 Joo sin t
hm - f(x + t) - - dt = f(x)
->OO 1T -oo f
. 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
:;;1rfoo
Jo -oo f(t)cos[s(t - x)]dt ds
1Joo
= :;; -oo f(t) Jor cos[s(t - x)]ds dt
1 r= foo
; Jo -=f(t)cos[s(t - x)] dt ds = f(x) (2.24)
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
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
F(s) = } Joo
~ ;;c_ eist f(t) dt (2.27)
V 27T -oo
and
!F(s)! :s ~ r-;--
} Joo lf(t)l dt (2.31)
y27T -oo
* Unless stated otherwise, we will generally assume that both t and sare real variables.
2.4/Fourier Transform Pairs 51
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)
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
-1 t<O
sgn(t) = { : (2.42)
1 t >o
ln this case, we find
1!-oo fo(t)
00
= .. ;;:c_
y21f'
~ os st dt + i .. }Joo fo(t)sin st dt
y21f' -oo
1
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)
~-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)
Example 2.4: Find the Fourier sine and cosine transforms of te- 01 ,
a> O.
and
L= te-a 1
COS st dt = (a2 - s 2)/(s 2 + a 2) 2
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
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
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.
56 Chap. 2/Fourier Integrais and Fourier Transforms
21
Example 2.6: Find the Fourier transform of e-a \ a > O.
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
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
2.4/Fourier Transform Pairs 57
g; { -sin mt
-;s } = J1T2 h(m- isj), m>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
(b) From (a), deduce that
cn:
(b) ~c{e
-ar .
sm at;s} =
J 2 2a - as
-
n s4 + 4a4
, a>O
* C(x) and S(x) are the Fresnel integrais (see Sec. 1.3.2).
2.5/Properties of the Fourier Transform 59
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
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
60 Chap. 2/Fourier Integrais and Fourier Transforms
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)
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)
2.5/Properties of the Fourier Transform 61
= 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)
from which we deduce
@'{f(t - a);s} = eias F(s)
Example 2.8: Find the Fourier inverse transform of 1/(i + ias + b),
b >o.
@'-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).
*Note that Theor. 2.3 implies that 3'- 1{F(s + a);t} = e;3'- 1{F(s);t}.
62 Chap. 2/Fourier Integrais and Fourier Transforms
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
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
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
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
= 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
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)
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)
(2.72)
= s-"f(a)
from which we deduce
l = y a-1 e -isydy=--e
o
f(a) -iwa/2
s"
(2.73)
~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
lim ( f(z)eiszdz = O
R-+= JcR
y
@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)
1 } - ( I e-ks)
g;{ 1(1z + JC) ;s = iV2TT 2JC - 2/C
J
1T i 1
= "2/C( -e -ks) , s> O
;' {
t(t2
1
+e)'
s} = J~2 _i_(l
e - e-kJI)sgn(s)
Solu1ion: By definition,
1
g;- 1{F(s);t} =---= J"" .
e-lls F(s) ds
V2TT -=
f!F-1{ i+ ias+
1 r}
b'
= Ja 2 ~ 4b exp[ -i(a + 2
ya + 4b)t]. t >O
Therefore,
;JF-1{ s 2
1 t}
+ ias + b'
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).
2.6/Transforms of More Complicated Functions 73
u i7T/2
-R o R X
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
or
74 Chap. 2/Fourier Integrais and Fourier Transforms
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
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
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
2is + 1
15. F(s) = -2 - l 16. F(s) = ~
s~ + 1
s + 1
76 Chap. 2/Fourier Integrais and Fourier Transforms
1- i
17. F(s) = (i + 4)2
18. By expanding J0(at) in a power series, use residue theory to deduce
that
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 -p p R X
2.6/Transforms of More Complicated Functions 77
sin;ts }
27. ~s { - 1
= ~ ~log 11-+-sI
t y2?T 1- s
(2.84)
* ln most cases of interest to us the function f(t) is real and thus f(t) = f(t).
80 Chap. 2/Fourier Integrais and Fourier Transforms
(g a f)(t) 1 f=
= ;;;-:_ g(u)f(t - u) du
v 27T -=
(gof)(t) = -./
y27T
J.-= g(t- v)f(v)dv
=
=. 11- f= f(v)g(t- v) dv
y27T -=
Qj;"-l{
Yft sin s . ;t }
s(l - IS)
and
1
g;- 1{ --.
I - IS
;t} = V27Te_, h(t)
J~2 e- 1 fr eu du,
-1
ltl < 1
O, t < -1
I-f=- - = (xz +
dx
az)(xz + bz)
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
I = 2 Loo j(f)g(f) df
= .!!._ r= e -(a+b)t dt
ab Jo
82 Chap. 2/Fourier Integrais and Fourier Transforms
or
Since both Fc(s) and Gc(s) are necessarily even functions, we can write
(2.89) as
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)
2.7/The Convolution Integrais of Fourier 83
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
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
to show that
("" 2
Jo J 0(ax)J0(bx) dx = 1rbK(ajb), O<a<b
fo
oo
x
smaxsm x -2 = -
2
b dx 7ra
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
(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
o;;; {
:-!/'c 12
lo(at) . } _
+ 1 ,s -
J~2 e -sIo(a ), a>O
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
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
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
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
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 ).
3'{sgn(t);s} =
J 2
-!..
7TS
Example 2.17: Find the Fourier transform ofthe Heaviside unit function
h(t).
* 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
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.
90 Chap. 2/Fourier Integrais and Fourier Transforms
7. If
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
integral)
(a nonintegral)
g;- 1{F(w);t} = -1
27T
J= ewt. F(w) dw = f(t)
-=
(2.I07)
and
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
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)
Thus,
f(t) = - -I
7T
i=
o
X(w)sin wt dw (2.Il3)
2.9/Hilbert Transjorms 93
fe(t) = -I
7T
1""o R(w)cos wt dw (2.118)
and
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
and
= J~oo e-;,, / 0
(t)sgn(t) dt (2.126)
or
1 Joo X(y)
R(w) =- - dy (2.129)
1T -oo w-y
Similarly, since
4 R( w)
= acosw- asinw X(w) = - wcosw- asinw
a>O
w +a
2 2 '
w2 + a2 '
96 Chap. 2/Fourier Integrais and Fourier Transforms
/(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
x(t) =!
7T
Joo
-oo (
x(r) dr
- T
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
(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)
I(t) = Re{ f 2
eit(s -s) ds}
f*(,y) = "'}
V 27T
f"" x f(x,y) dx
-oo
followed by
F(,Tj) 1 J""
= y;i; -= ei.,y f*(,y) dy
aY
8i'(2){iJxa/x,y);x~, Y~'YI } = -'Y/F(,'Y/) (2.147)
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~=
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
3.1 Introduction
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
=-
1lx+l e-iui du
2 x-1
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
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
where we define
k
y
Figure 3.1 Infinite beam on elastic foundation
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
propriate conditions)
f(t) 1 I""
= ~ r::>= . F(s) ds
e-rts (3.12)
V 2'7T -oo
J: J: jk(x,tW dt dx < oo
~
1
r::>=
foo u(t)k(x - t) dt = f(x), - =< x < = (3.15)
y2'7T -oo
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)
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
f(x) =
y27T
!--: J= -oo
lx - gl- 112 u() d
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
3.2/Boundary Value Problems 111
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.
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.
112 Chap. 3/Applications Involving Fourier Transforms
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
_ 2 e(I-)X
2- ' x<O
u(x) = 2
{ 2 - e-x, x2:::0
1 Joo tu(t)dt
u(x) = :; -oo [1 + (x - t) 2](1 + t 2)
"... 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 ... ,''
* For details, see J. Fourier, The Ana/ytical Theory of Heat, New York: Dover, 1955.
114 Chap. 3/Applications Involving Fourier Transforms
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.
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>
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)
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
u(x,t) 1 J""
= y; -
-=f(x + 2azy't)e-z 2
dz (3.36)
* 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.
3.3/Heat Conduction in Solids 117
Example 3.5: Solve the problem described by (3.26) when the initial
temperature distribution in the rod is given by
-oo<x<oo
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(x,t) = ~ ~ 2 foo
V 27T -oo
ii
O
g(x-, t-T)q(,T) dT d (3.41)
* To avoid confusion with partia! derivatives, we have chosen not to subscript the
transfonned functions by C to denote a cosine transform.
3.3/Heat Conduction in Solids 119
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)
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
* Equation (3.50) could also be derived directly from the convolution integral relation
given by Eq. (2.90) in Sec. 2.7.1.
120 Chap. 3/Applications Involving Fourier Transforms
and by using the transform relation [see Prob. 7(b) in Exer. 3.3)
u(x,t) -
_ _ x_
r= (
f'
f(T)
) e -x2f4a2(t-T) dT
2ay?T o t - T 312
(3 .55)
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
This solution can also be expressed in the equivalent form (see Prob.
23 in Exer. 3.3)
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.
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
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
show that
(a) is given by
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)]
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
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) = _1_
2\(i;r
I=
-oo
[e-i<x-ct> + e-is<x+ct>] F(s) ds
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.
128 Chap. 3/Applications Involving Fourier Transforms
- --- ....
.... -
\ :it:l displacement
X
u(x,t)
} Joo
= ~ ;;:;-::-. f(x-) [cos(e/4t) + sin(e/4t)] d (3.81)
2 V 211'1 -oo
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
(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
u(x,t) = R-l/2e-r2(costt>t/4Rcos(x2~;cf>- ~)
show that
(a) the solution of the transformed problem is
* The heat equation V2u = a- 2u, reduces to Laplace's equation when u, = O, i.e.,
when u is independent of time t.
132 Chap. 3/Applications Involving Fourier Transforms
* 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
To [ tan-
== -:;;:
1 + b)
(X-Y- - tan- 1 -Y-(X - b)]
but using the trigonometric identity
v(x,y) = -
y J"" (
f() d
<:)2 + Y2 y>O (3.95)
7T -oo X - ~
U = T0
Figure 3.4 Family of isotherms
136 Chap. 3/Applications lnvolving Fourier Transforms
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
u(x,y) 1 J""
= Z1r -oc/(g) log[(x - g)2 + /] dg (3.96)
u==f X
(3.102)
u(x,y) 1
= ;;c
y21T o
1= f()[g(x + , y) + g(lx- l, y)] d (3.ll0)
3.5/Potential Theory 139
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
(a) u(x,y)
y roo f(t) [(t -
= ; Jo
1
x)2 + y2 - (t
1
+ x)2 + y2
]
dt
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
(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,z) = 2}
1T Joo
-= J""
-=f(-,TJ)g(x- ,y - TJ,Z) ds d71
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
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.
and
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).
q(x) = J p(x,O, t) dt
1 100 e YQ(s)cos(sx
5
- ygst) ds
p v'21T o
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)
* A moving viscous ftuid tends to adhere to the surface of an obstacle placed in its
path.
148 Chap. 3/Applications Involving Fourier Transforms
(3.149)
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
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
(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).
150 Chap. 3/Applications lnvolving Fourier Transforms
where
J = -zv'i!X rsin(w2 -
2
{ ) d{
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
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)
a ( Uyy
y -
axx)
2
2 = 0 (3.161b)
(3.162)
G(x,s) = -P(s) 11
(1 + jsjx)e- sx (3.170)
s2
from which we deduce
The stress components that we desire are now calculated using (3.162),
and are given by
154 Chap. 3/Applications Involving Fourier Transforms
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
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]
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
.
(b) O"yy - O"xx + 2ia-xy = j ;2 PoX Jo r~
sJ0(as)e-s<x+zyl ds
-oo<x<oo,y>O
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
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,
(3.182b)
whereas in general,
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
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.
Our next example illustrates the way in which the characteristic function
is used to provide the desired PDF under a transformation of random
variables.
Cz(t) = _1_
27T
I=
-oo
e-y2f2 I""
-oo
eitxy e-x2!z dx dy
* When working with more than one PDF, we find it convenient to use the more
distinguishing notation Px(x) instead of just p(x).
3.8/Probability and Statistics 159
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
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
[k/2) - 2j
_ 1 "" cr mk-lj
mk- k. j~ (2j + l)!(k- 2j)!
where
= {~z_,l)/2,
keven
[k/2] kodd
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)
show that
3.8/Probability and Statistics 161
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
or equivalently,
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
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.
Im(p)
c + joo
o Re(p)
C - joo
a real variable. For now we will proceed with such integrations and later
examine the conditions under which this formalism is valid.
Solution: By definition,
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
where f!fc and f!f s denote lhe Fourier cosine and Fourier sine transforms,
respectively.
Remark: Functions that are identically zero for t :::: t 0 > O are said
to be of exponential order -=.
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
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
* 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.
4.2/The Transforms of Some Typical Functions 169
EXERCISES 4.2
13. f(t) = l 1
14. f(t) = sin 1
2
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
170 Chap. 4/The Laplace Transform
= { 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.
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
= -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.
1
2{h(t - a);p} = -e -ap, a>O (4.10)
p
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)
where F(p) is the Laplace transform of f(t).
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.
* Notice that this result is simply a special case of that in Exam. 4.3.
4.3/Basic Operational Properties 175
5t {J:f(u)du;p} = F(p)jp
where F(p) is the Laplace transform of f(t).
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.
*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.
4.3/Basic Operational Properties 177
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.
r
Proof: Let us write the Laplace transform as
f(t)
o 17' t
Figure 4.2 Graph of a half wave rectified sinusoidal
4.3/Basic Operational Properties 179
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
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
foof() } 1( 00
CI'( t ) = fi -
cos-ud U, t>O
00 u
(b) the exponential integral defined by
Show that
;t'{Ln(t};p} = ~ (p ; 1) n
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)
Example 4.11: Find the Laplace transform of erf(t), which is the error
function discussed in Sec. 1.3.
= _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
y; -2ab
io
=
e -aZxZ-bZfxZ dx=-e
2a '
a> O, b ~O, (4.18)
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
1
X= Vt
oL
(L){ r.. } - ~ (
sm yt,p - n~o(2n + l)!oL t
-1r
(L){ n+l/2. }
,p
-
5t{sin y't;p} = 1P J'TT e- 114P, P Re(p) >O
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
"" <-W
= ~O n!pn+P+ I
1 (-l)"(l)n
= pv+t~0 ~
oo
p
from which we deduce
Re(p) >O
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
(1 + x)-a = ~
n=O
(-a) n
x" =~
n=O
(-1t;(n +a) xn,
n.f(a)
lxl < 1
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
from which we deduce
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
188 Chap. 4/The Laplace Transform
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
11 . .X{ -sin at
-;p } = tan- 1 a/p, a>O
1
a>O
v> -1/2
29. Use the result of Prob. 31 in Exer. 4.3 to deduce the value of
~ 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
190 Chap. 4/The Laplace Transform
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
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
:;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)
192 Chap. 4/The Laplace Transform
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,
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
::t-
1
{~ e-l/p;t} = lo(2Vt)
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
We now define the Laplace convolution integral of f(t) and g(t), i.e.,
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).
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
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
=-
1i/ ei(t - 2u)
du
7T 0 yu(t - u)
198 Chap. 4/The Laplace Transform
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
(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.
J 1
0
lf- 1 (I - u)Y- 1 du = f(x)f(y)/f(x + y), x >O, y >O
-'1T 1,v
I
o
ezau
u(t - u)
du = eat lo(at)
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: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
* 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.
202 Chap. 4/The Laplace Transform
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
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.
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
f cR
eP'F(p )dp ={
JnJ
eP' F(p )dp +J
JKL
eP' F(p )dp
lm(p)
J
K Re(p)
Re(p) = c
L
Figure 4.6 Contour of integration
204 Chap. 4/The Laplace Transform
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
idpi = RdO
and therefore
/ I ::s __
M l.,.;z eRr cos e dO
1I Rk-1 Bo
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
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
1T' n=l
(
- 1)
cos[(n - 1/2)7rx]e-<n-t/2)2.r2t
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.
lm(p)
Ii""
=-
7T o
-
e -xt sin(ay'x) dx
= 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}.
4.6/Complex Inversion Formula 209
2:-1 {/o(ryp)
Jo(byp)'
t} = ~ ~ knlo(knr)
b n= 1 J1(knb)
e-k~r
1
17. F(p) =c 2
18. F(p) =c
v p(p- a) VP +a
I(t) = L~ e-bu 2
cos tu du, t::::: O, b >O
I(t) = ! J~ e-r2f4b
2 b
22. Using the result of Prob. 21 and
(a) differentiating 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
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
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
4.7/Additional Topics 211
functions defined on the entire realline rather than only on the positive
realline.
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,
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
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
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
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(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.
*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
1r n=O X
12
Example 4.29: Find the two-sided Laplace transform of e-
({)
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,
216 Chap. 4/The Laplace Transform
Re(p) > -1
and
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
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
transform of the given function.
3. Si(t) = r sinu u du,
Jo
t >O 4. Ei(t) = (oo e-u du,
J, u
t >O
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 '
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.
I= r= sin t dt
Jo t
J = fo
= _ sin t d
e 1-- t
t
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 = ~
Example 5.2: Evaluate the integral
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
= 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
10. L=xe-x2erfc(x) dx
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
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 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.
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.
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- 1{ F(p)
p 2 + ap + b
;t}
= J:g(t - u)f(u) du (5.7)
g(t) =2 -1{ p 2 1
+ ap +
b ;t } (5.8)
226 Cbap. S/ Applications Involving Laplace Transforms
Example 5.7: Construct the one-sided Green's function for the system
described by y" - 2y' + 5y = f(t).
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
= :;e-'{Yo(P - 1)
(p -
+ (vo - Yo)
1l + 4
r} + ! Jo
' 2
f'e 1 -"sin 2(t - u)f(u) du
+ !J:e -"sin2(t- 1
u)f(u)du
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
* 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
u(x,t)
X 1t
.. r o (t-T)3/2
= 2avTT
j(T)
~
1
{eP sinh(xy'pla) ;p == 2 2 2} ]
+ LJ Res -n1ra
n= 1 p sinh(y'PI a)
from which we deduce
2 ~ (- It . _n2,-2a21] t
u(x,t) == To [ X+- L.. - - sm(mrx)e (5.23)
1T n=I n
* 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
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
v =I= c
v =c
(5.31)
5.4/Solutions of PDEs 233
(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
EXERCISES 5.4
u(x,t) ac_
= ~y7T l'
Oyt-r
f(r) exp [
~ r:--: -r Jdr
(
4a 2 t - r
)
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)]
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
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)
11. Given the boundary-value problem
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
- 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
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.
5.4/Solutions of PDEs 237
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
20. -2
Uxx = C Um 0 <X< oo, t >0
B.C.: u(O,t) =O, u(x,t) ~ O as x ~ oo
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
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
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
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
J: U(T)J (t -
0 T) dT = Sn t, t >0
L lo(T)J0(t - T) dT = sin t
240 Chap. S/Applications Involving Laplace Transforms
u(t) - J: e 1
-T u('r) dr = f(t), t >O
P(x,y)
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
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
~ a)p";t} == -r(-1-~-a-~-)f-(a-)
1
2 -l{ro
TTasin
==--(a- 1
(5.60)
1T
( ) = sin TTa
U(
1T
1 1
(
O t-T
f'(T)
)l-a
d
T, t >o (5.61)
5.5/Linear Integral Equations 243
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
8. u(t) = t + l J: (t - du(r) dr
9.
,i(
u(r)
O t- r
) 113 dr = t(l + t)
*
has the formal solution
o
;-:---_
~
T
dT = f(t), u(O) = /(0) = O
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(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
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 < u < b, where u = Re (s). ln such cases the
transform F(s) is analytic in this vertical strip. Also, if xk -y(x) is absolutely
integrable on the positive real axis for some k > O, and if F(s) is defined
by (6.4), then the inversion formula (6.5) is valid for c > k.
.M{e-x;s} = L""x-le-x dx
6.2/Evaluation of Mellin Transforms 247
1 s}
At { - - = -Tr- O< Re(s) < 1
1+x ' sin 1rs '
=~[r-'
a o
J(t) dt
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
provided
limx-k-tf<k>(x) = O, k = 1,2, , n-1 (6.18)
x-+0
.M- 1{F(s)G(s);x} = i
=
o
d
f(x/u)g(u) __!!_
u
(6.19)
but recalling the definition of the beta function (see Prob. 21 in Exer.
1.2)
Examp/e 6.5: Find the Mellin transform of x-vlv(ax), a> O, v> -1/2.
Al{x-vlv(ax);s} = Laoxs-v-Ilv(ax) dx
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
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 )
)
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
252 Chap. 6/The Mellin Transform
1 Jc+ioo
26. Al{f(x)g(x);s} = 1T'i c-ioo F(p)G(s- p) dp
2
g(s) = "' an
LJ--:;
n=l n
n=l
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
(6.22)
6.3/Complex Variable Methods 255
However,
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).
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
Observe also that we could obtain this result from Exam. 6.3 and
the operational property given by Eq. (6.12).
c-ioo
(6.25)
n=O
(6.26)
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
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) '
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
*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)
=
= 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).
260 Chap. 6/The Mellin Transform
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)
.,.u- I{ 1T
sin S1T ;x
}
= 1 +1 x
Therefore, from (6.29a) it follows that
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
also has poles b~> b2 , ,bM on the positive real axis, show that
N
ln Probs. 9 and 10, use the result of Prob. 8 to find the Mellin transform
of the given function.
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)
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
_ 1{ 7r } = ( -IY
(b) .At 2s cos(Trs/2) ;x = ~o (2n + l)x 2n+l'
x>I
.At- 1{ 7r
2s COS(Trs/2)
;x} = ~2 - tan- x ' 1
x>O
= 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)
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
f(x) = 1
7Ti J -
2 c-ioo
x s F(s) ds
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
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-
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)
f(x) == cos ax
2
X
~ 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.
6.4/Applications 265
~ 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
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
i"
which simplifies to
dx
Pz(z) = o
Px(x)py(z/x)-
X
(6.43)
z = nxj (6.45)
j=l
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
. 2(zs-2 - 1)
Pz(z) = hm 7T(z2 -
.~2 1)sm
. ('Tf'S / 2)
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
where we are using the inverse Mellin transform relation (see Prob. 22
in Exer. 6.3)
EXERCISES 6.4
1. Given the function
F(s) = (2/as) '(s)f(s - 2)cos(7Ts/2)
show that Res{F(s );O} = -1/2.
6.4/Applications 269
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
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
(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
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)
6.4/Applications 271
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)
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
Jtl
-I
o
SID a7T
o
o
o
,X-
} _ t an a7T ( X -a - XI
2
+a)
{
SID S7T - SID a7T 7T 1- X
f(x) = L= g(x~)u(~) d~
has the formal solution
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
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
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)
and
f(x,y) = _!_ J=
2'7T -oo
J=
-oo
e-i<x~+.v"> F(,"'J) d d"fJ (7.2)
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-<I>l rf(r) dO dr (7.4)
called the Hankel transform of order zero. Clearly, the substitutions (7.3)
into Eq. (7.2) will produce the similar result
*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. *
= 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
*1. N. Sneddon, The Use of Integral Transforms, New York: McGraw-Hill, 1972,
pp. 301-309.
7.2/Evaluation of Hankel Transforms 277
ou; {
<11-o 1 -ar ;p } = . . 1
-e , a> O (7.10)
1 2
r vP +a2
or
a>O (7.12)
2
Example 7.3: Find the Hankel transform of order zero of e-a r2.
*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.
278 Chap. 7/The Hankel Transform
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 '
we obtain
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:>
I = L"" xF(x)G(x) dx
where
F(x) = (1/x)Jv+l(ax)
and
G(x) = (1/x)Jv+l(bx)
280 Chap. 7/The Hankel Transform
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
or
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
-
7.2/Evaluation of Hankel Transforms 281
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
or
;tevV- 1 :,[r1 -vf(r)];p} = -p;tt'v-I{f(r);p} (7.24)
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.
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
9. ;reo{ Va 2
h(a -
_
r)i p } = Jmrp J112(ap), a> O
2
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
16. Integrate both sides of the Hankel transform relation in Prob. 7 with
respect to a from O to b and deduce that
1
f
b
Hint: J 1(ar) da = - [1 - J0(br)]
o r
~exp(p ~ b )/o(~:).
2 2
Yfo{e-arl lo(br);p} = a> O, b >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
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.
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
Example 7.5: Solve the problem described by (7.31) for the special case
where
= 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
for the determination of the function A(p). Now imposing on this formal
7.3 I Applications 287
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)
EXERCISES 7.3
u(r,z)
roo F(p)
= Jo
(1 - pe
-pz) J (pr) dp
0
( ) 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
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
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 '
3 1/r 1/p
5 e -o' , a>O
-u2,2
6 e , a>O
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.
where
2 f1T
an = - f(x)cos nx dx, n = 0,1,2, ... (8.4)
1T o
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.
S"{f"(x);n} = r f"(x)sin nx dx
S1T{f"(x);n} = -nf(x)cos nx I~ -r n
2
f(x)sin nx dx
= - n ~ Fc (n) + ( -l)Y'(p)
2
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
8.2/Finite Fourier Transforms 295
0, n even (8.19)
B.C.: U(n,O) = O, U(n,'TT) = { 2To/n, n odd
where*
U(n,t) = L1
*For simplicity of notation, we will drop the s subscript on the transform function.
296 Chap. 8/Finite Transforms
and
F(n) = f f(x)cos mrx dx (8.26)
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
EXERCISES 8.2
ln Probs. 1-10, evaluate the finite sine transforrn and finite cosine transforrn
of the given function.
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)
8.2/Finite Fourier Transforms 297
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
J: r(x)cf>m(x)cf>n(X) dx = O, m f= n (8.36)
where
n = 1,2,3, ... (8.39)
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).
*The actual values k., n = 1,2,3, ... , must be determined by a numerical procedure.
8.3/Sturm-Liouville Transforms 301
111/>ix)ll2 1( 1 b) ,
= 2 b + h cos2kn n = 1,2,3, ... (8.45)
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
302 Chap. 8/Finite Transforms
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
7. Develop a generalized finite Fourier transform and inversion formula
associated with the Sturm-Liouville problem
y" + y =O, O<x<1
hy(O) - y'(O) = O, y'(l) =o
8.4/Finite Hankel Transform 303
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
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)
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)
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
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)
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<r<b
Taking the finite Hankel transform of order zero, we obtain
U, + a2k;U = a2knbf(t)Jt(knb), t >O (8.73)
I.C.: U(n,O) = O
where
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
8.4/Finite Hankel Transform 307
v+ 1 }
H v{X ;n = 2(v k2+ 1) lv+ 1(kn)
n
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.
show that
2 ~ sin(k,ct)
u(r,t) =- LJ k 2 J (k) 10(k,r)
C n=l n I n
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)'
(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
(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
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
and
*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
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
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.
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
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 proofs of these properties, and some additional ones, are left to the
exercises.
(x * y)(k) = 2: x(m)y(k - m)
m=O
8 5
7 5 8 6
6 7
k= o k=1
Figure 9.3 Convolution
316 Chap. 9/Discrete Transforms
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.
9.2/Discrete Fourier Transform 317
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
318 Chap. 9/Discrete Transforms
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.
Step 1:
I
A1(11) = L x(K11)W 2
K' = x(Ol)- x(ll) = 2- 4 = -2
K1=0
Step 2:
I
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
values that are actually available for processing. The Laplace transform
of the above sampled function formally leads to
= L x(nT)e-pnT (9.28)
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
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.
Z{t;z} = L nz-n
n=O
=z 2: nz-(n+ I)
n=O
d 00
""
-z-L.JZ -n
dz n=O
324 Chap. 9/Discrete Transforms
= z 2: x(n + l)z-<n+ll
n=O
=z 2: x(n)z-n
n=t
9.3/The Z Transform 325
Z{tx(t);z} = L nx(n)z-n
n=O
d
L
00
- z- x(n)z-n
dz n=O
from which we deduce
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.
*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).
326 Chap. 9/Discrete Transforms
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, ...
= ~h(n) + [ 03 : 5
n)Gr- ~]h(n-1). n=0,1,2, ...
Example 9.10: Use the convolution theorem to find the inverse Z trans-
form of the product
(z - 2)(z - 3)
Solution: By setting
z- {Y(z);n} = y(n) = 3n
1
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, ...
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(z) =z+ 3
z- 2
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
9.3/The Z Transform 329
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
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
330 Chap. 9/Discrete Transforms
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)
n=O
n---i>n-1
n=l
= Z [~ YnZ-n - Yo]
n=O
= z[ Y(z) - Yo]
Similarly,
n=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
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
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)
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
Tcf(z)dz =O
lim
R-.oo
r
JcR
f(z)dz =o
Iim r
p--.o Jc.
f(z)dz = o.
(a) lim
R-->oo
r
JcR
eim'f(z) dz =o (m >O)
(b) lim
R-->oo
r
JcR
em'f(z) dz =O (m > O)
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)
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
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
i"
23 P.(t )h(l - Iti) v~Jn+I/2(S)
342 Appendix B
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
Tr s 4
+ 4a 4
3 2
e- sin at, A2a - as
8 a>O
s4 + 4a 4
J~ sgn(s)
l
-
l l
2
Vi Vs
J~ e-
t
3 a>O
f + a2'
Appendir B 343
1 -Se -as
--
4 + a2)2, a>O
(t2 yZ; a
6 e -ai , a>O JI s
:;;: s2 + a2
9
1 -ar
-e
t
J! arctan(s I a)
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
-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
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
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
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
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
Bhimsen K. Shivamoggi
ISBN-13: 9780819432322
SPIE Vol. No.: PM178