Churchill, R. v. - Operational Mathematics

Operational Mathematics
Third Edition
Ruel V. Churchill
Professor Emeritusof Mathematics
Universityof Michigan
McGraw-HiII Book Company

New York St. Louis San Francisco Dsseldorf Johannesburg
Kuala Lumpur London Mexico Montreal New Delhi Panama
Rio de Janeiro Singapore Sydney Toronto
I
Operational Mathematics
Copyright O 1958, 1972 by MeGraw-Hill, lne. Al! rights reserved.

Copyright 1944 by MeGraw-Hill, Ine. Al! rights reserved.
Printed in the United States of Ameriea. No part of this publieation
may be reproduced, stored in a retrieval system, or transmitted, in
any form or by any means, electronie, mechanical, photocopying,
recording, or otherwise, without the prior written permission of the
publisher.
Library ofCongress Catalog Card Number 70-174611

07-010870-6
78910 KPKP 7832109
This book was set in Times Roman, and was printed and bound
by the Kingsport Press. The designer was Jo Jones; the
drawings were done by John Cordes, J. & R. Technical Services, Ine.
The editors were Howard S. Aksen and Madelaine Eiehberg. Matt
Martino supervised produetion.
Contents
Preface xi
Chapter 1. The Laplace TransformatioR 1

1. Introduction I
2. Definition of the Laplace Transformation 3
3. Sectionally Continuous Functions. Exponential Order 5
4. Transforms of Derivatives 7
5. Examples. The Gamma Function lO
6. The Inverse Transform 14
7. A Theorem on Substitution 16
8. The Use of Partial Fractions (Table 1) 17
9. The Solution of Simple Differential Equations 20
10. Generation of the Transformation 24
Chapter 2. Further Properties of the Transformation 27

11. Translation of F(t) 27
12. Step Functions 29
13. The Impulse Symbol c5(t - to) 33
14. Integrals Containing a Parameter 39
15. Improper Integrals 41
16. Convolution 43
17. Properties of Convolution 46
18. Differential and Integral Equations 50
19. Derivatives of Transforms 54
20. Series of Transforms 57
21. Differential Equations with Variable Coefficients 61
22. Integration of Transforms 65
v
vi CONTENTS
23. Periodic Functions 66

24. Partial ,Fractions 70
25. Repeated Linear Factors 73
26. Quadratic Factors 75
27. Tables of Operations and Transforms 78
Chapter 3. EIemeotary ApplicatioDS 85

28. Free Vibrations of a Mass on a Spring 85
29. Forced Vibrations without Damping 88
30. Resonance 91
31. Forced Vibrations with Damping 95
32. A Vibration Absorber 96
33. Electric Circuits 102
34. Evaluation of lntegrals 106
35. Exponential- and Cosine-integral Functions 110
36. Static Deftection of Beams 113
37. The Tautochrone 115
38. Servomechanisms 117
39. Mortality of Equipment 119
Chapter 4. Problems in Partial Differeotial EquatioDS 123

40. The Wave Equation 123
41. Displacements in a Long String 126
42. A Long String under lts Weight 130
43. The Long String lnitially Displaced 133
44. A Bar with a Prescribed Force on One End 135
45. Equations of Diffusion 143
46. Temperatures in a Semi-infinite Solid 145
47. Prescribed Surface Temperature 146
48. Temperatures in a Slab 151
49. A Bar with Variable End Temperature 153
50. A Cooling Fin or Evaporation Plate 153
51. Temperatures in a Composite Solid 155
52. Observations on the Method 159
Chapter 5. FunCtiODS01 a Complex Variable 162

53. Complex Numbers 162
54. Analytic Functions 163
55. Exponential and Trigonometric Functions 166
56. Contour lntegrals 168
57. Integral Theorems 170
58. Power Series 171
59. Singular Points and Residues 173
CONTENTS vii
60. Branches of Multiple-valued Functions 177

61. Analytic Continuation 179
62. Improper Cauchy Integrals 181
Chapter 6. The Inversion Integral 186

63. Analytic Transforms 186
64. Permanence of Forms 188
65. Order Properties of Transforms 189
66. The Inversion Integral 193
67. Conditions onf(s) 195
68. Conditions on F(t) 198
69. Uniqueness of Inverse Transforms 201
70. Derivatives of the Inversion Integral 202
71. Representation by Series of Residues 206
72. Residues at Poles 208
73. Validity ofthe Representation by Series 210
74. Alterations of the Inversion Integral 213
Chapter 7. Problems in Heat Conduction 219

75. Temperatures in a Bar with Ends at Fixed Temperatures 220
76. The Solution Established 222
77. The Series Form Established 224
78. Properties of the Temperature Function 226
79. Uniqueness ofthe Solution 228
80. Arbitrary End Temperatures 230
81. Special End Temperatures 232
82. Arbitrary Initial Temperatures 237
83. Temperatures in a Cylinder 240
84. Evaporation from a Thick Slab 245
85. Duhamel's Formula 247
Chapter 8. Problems in Mechanical Vibrations 253

86. A Bar with a Constant Force on One End 253
87. Another Form of the Solution 256
88. Resonance in the Bar with a Fixed End 258
89. Verification of Solutions 259
90. Free Vibrations of a String 264
91. Resonance in a Bar with a Mass Attached 266
92. Transverse Vibrations of Beams 268
93. Duhamel's Formula for Vibration Problems 270
Chapter 9. GeneraUzed Fourier Series 276

94. Self-adjoint Differential Equations 277
viii CONTENTS
95. Green's Functions 278

96. Construction of Green's Function 281
97. Orthogonal Sets of Functions 283
98. Eigenvalue Problems 288
99. A Representation Theorem 291
100. The Reduced Sturm-Liouville System 292
101. A Related Boundary Value Problem 293
102. The Transform y(x,s) 294
103. Existence of Eigenvalues 296
104. The Generalized Fourier Series 299
105. Steady Temperatures in a WalI 305
106. Verification ofthe Solution 308
107. Singular Eigenvalue Problems 309
Chapter 10. General Integral Transforms 317

108. Linear Integral Transformations 317
109. Kernel-product Convolution Properties 319
110. Example 320
111. Sturm-Liouville Transforms 325
112. Inverse Transforms 327
113. Further Properties 329
114. Transforms of Certain Functions 332
115. Example of Sturm- LiouvilIe Transformations 333
116. Singular Cases 336
117. A Problem in Steady Temperatures 341
118. Other Boundary Value Problems 343
Chapter 11. Finite Fourier Transforms 348

119. Finite Fourier Sine Transforms 348
120. Other Properties of S. 350
121. Finite Cosine Transforms 354
122. Tables of Finite Fourier Transforms 356
123. Joint Properties of C. and S. 357
124. Potential in a Slot 360
125. Successive Transformations 362
126. A Modified Sine 'Fransformation 368
127. Generalized Cosine Transforms 370
128. A Generalized Sine Transform 372
129. Finite Exponential Transforms E.{F} 376
130. Other Properties of E. 378
Chapter 12. Exponential Fourier Transforms 383

131. The TransformationE.{F} 383
CONTENTS ix
132. The Inverse Transformation 385

133. Other Properties of E. 388
134. The Convolution Integral for E. 391
135. Convolution Theorem 393
136. Tables of Transforms 396
137. Boundary Value Problems 397
Chapter 13. Fourier Transfonns 08 the HaJf Line 401

138. Fourier Sine Transformsf.(a) 401
139. Fourier Cosine Transformsfc(a) 404
140. Further Properties of S. and C. 405
141. Convolution Properties 406
142. Tables of Sine and Cosine Transforms 407
143. Steady Temperatures in a Quadrant 408
144. Deflections in an Elastic Plate 410
145. A Modified Fourier Transformation 1;. 414
146. Convolution for 1;. 416
147. Surface Heat Transfer 418
Chapter 14. Hankel Transfonns 420

148. lntroduction 420
149. Finite Hankel Transformations 421
150. lnversion of H.j 423
151. Modified Finite Transformations H.h 424
152. A Boundary Value Problem 426
153. Nonsingular Hankel Transformations 431
154. Hankel Transformations H." on the Half Line (x> O) 432
155. Further Properties of H." 434
156. Tables ofTransforms H..{F} 436
157. Axially Symmetric Heat Source 437
Chapter 15. Legendre and Other Integral Transfonns 441

158. The Legendre Transformation 1;,on the Interval (-1,1) 442
159. Further Properties of 1;, 444
160. Legendre Transforms on the lnterval (0,1) 446
161. Dirichlet Problems for the Sphere 448
162. Laguerre Transforms 452
163. Mellin Transforms 453
Bibliography 456
Appendixes
AppendixA Tables of LaplaceTransforms 458
x CONTENTS
Table A.l Operations 458

Table A.2 Laplace Transforms 459
Appendix B Tables of Finite Fourier Transforms 467
Table B.l Finite Sine Transforms 467
Table 8.2 Finite Cosine Transforms 469
Appendix C Table of Exponential F ourier Transforms 471
Appendix D Tables of Fourier Sine and Cosine Transforms 473
Table D.l Sine Transforms on the Half Line 473
Table D.2 Cosine Transforms on the Half Line 475
Index 477
Preface
This is an extensive revision of the second edition of "Operational

Mathematics" published in 1958. Chapters have been added on general
integral transforms, finite Fourier transforms, exponential Fourier trans-
forms, Fourier transforms on the half line, Hankel transforms, and on
Legendre and other integral transforms. The presentation of theory and
applications of the Laplace transformation has been revised. Tables of
several of the most useful transforms now appear in the Appendix or in
the text. Additional problems illustrate applications of the various integral
transformations.
The book is designed as a text and a reference on integral transforms
and their applications to problems in linear differential equations, to boundary
value problems in partial differential equations in particular. It presents the
operational properties of the linear integral transformations that are useful
in those applications. The selection of a transformation that is adapted to a
given problem, by observing the differential forms and boundary conditions
that appear in the problem, is emphasized in this edition. The Laplace
transformation receives special attention because of its many useful opera-
tional properties and the large class of problems to which it applies, in-
cluding applications outside the field of differential equations.
The applications to problems in physics and engineering are kept on a
fairly elementary level. They include problems in vibrations or displacements
in elastic bodies, in diffusion or heat conduction, and in static potentials.
No previous preparation in the subject of partial differential equations is
required of the reader.
This book is a companion volume to "Fourier Series and Boundary
Value Problems" and "Complex Variables and Applications." The three
books cover, respectively, these principal methods ofsolving linear boundary
xi
xii PREFACE
value problems in partial differential equations: the operational methods of

integral transforms, separation of variables and Fourier series, and con-
formal mapping. Generalized Fourier series and their applications are
presented in Chapter 9 here, with the aid of the theory of the Laplace trans-
formation. A summary of useful theory of functions of a complex variable
is given in Chapter 5. All three books are intended to present sound mathe-
matical analysis as well as applications. Conditions of validity of analytical
results are kept on a simple and practicallevel. More elegant conditions may
call for training in analysis beyond the level of advanced calculus.
The first four chapters are designed to serve as a text for a short course
in real Laplace transforms and their applications.
The impulse symbol or "delta function" is introduced in Seco13 in an
elementary and careful manner. No theory of distributions, or generalized
functions, is included in the book. A satisfactory presentation of the theory
would require considerable space and the introduction of concepts not
needed elsewhere in the texto Neither can the author justify a presentation of
the abstract theory of linear spaces in this book. An intuitive approach
from vectors to functions (Secs. 10 and 97) serves as a guide for writing inner
products of functions as integrals.
In preparing the three editions of this book the author has taken advan-
tage of improvements suggested by many students and teachers. He is
grateful to them for that assistance; also to authors referred to in the Bibliog-
raphy and footnotes, whose publications have influenced the selection of
material.
Ruel V. Churchill
1
The Laplace Transformation
1 INTRODUCTION
The operation of differentiating functions is a transformation from functions
F(t) to functions F'(t). If the operator is represented by the letter D, the
transformation can be written
D{F(t)} = F'(t).
The function F'(t) is the image, or the transform, of F(t) under the transforma-
tion; the function 3t2, for example, is the image of the function t3.
Another transformation of functions that is prominent in calculus is
that of integration,
I{F(t)} = I: F(t) dt.

The result of this operation is a functional f(x), the image of F(t) under the
transformation. A simpler transformation of functions is the operation of
multiplying all functions by the same constant, or by a specified function.
2 SECo 1] OPERATIONAL MATHEMATICS
In each of the above examples inverse images exist; that is, when the
image is given, a function F(t) exists which has that image.
A transformation T{F(t)} is linear if for every pair of functions Fl(t)
and F2(t) and for each pair of constants C 1 and C2, it satisfies the relation
(1)
Thus the transform of a linear combination of two functions is the same
linear combination of the transforms of those functions, if the transformation
is linear. Note the special casesof Eq. (1) when C2 = Oand when C 1 = C2 =
1. The examples cited above represent linear transformations.
The class of functions to which a given transformation applies must
generally be limited to some extent. The transformation D{F(t)} applies to
all differentiable functions, and the transformation 1{F(t)} to all integrable
functions.
Linear integral transformations of functions F(t) defined on a finite
or infinite interval a < t < b are particularly useful in solving problems in
linear differential equations. Let K(t,s) denote some prescribed function of
the variable t and a parameter s. A general linear integral transformation of
functions F(t) with respect to the kemel K(t,s) is represented by the equation
(2) T{F(t)} = rK(t,s)F(t) dt.

It represents a function f(s), the image,or transform,of the function F(t).
The class offunctions to which F(t) may belong and the range ofthe param-
eter s are to be prescribed in each case. In particular, they must be so
prescribed that the integral (2) exists.
We shall seethat with certain kemels K(t,s) the transformation (2),
when applied to prescribed linear differentialforms in F(t), change&those
forms into algebraicexpressionsinf(s) that involvecertain boundary values
ofthe function F(t). Consequently, classes of problems in ordinary differential
equations transform into algebraic problems in the image of the unknown
function. If an inversetransformation is possible,the solution of the original
problem may be determined. Boundary value problems in partial differenial
equations can be simplifiedin a similar way.
The operational mathematics presented in this book i:,the theory as
well as the application of such linear integral transformations that bears on
the treatment ofproblems in ordinary or partial differential equations. Later
on, we shall return to the general transformation (2)and to the question of
deciding upon the special cases that may apply to a given problem in differ-
ential equations. First we present the special case that is of greatest general
importance, the operational mathematics of the Laplace transformation.
Other prominent cases include the various Fourier transformations, to be
presented later.
THE LAPLACE TRANSFORMATION [SECo 2 3
When a = O and b = 00 and K(t,s) = e-sr, the transformation (2)

becomes the Laplace transformation. The direct application of this trans-
formation replaces the earlier symbolic procedure known as Heaviside's
operational calculus.1 The development of the transformation and the
accompanying operational calculus was begun before Heaviside's time;
Laplace (1749-1827) and Cauchy (1789-1857) were two of the earlier con-
tributors to the subject,2
In this chapter we present the basic operational property ofthe Laplace
transformation, the property that gives the image of differentiation of
functions as an algebraic operation on the transforms of those functions. In
the following chapters further properties of the transformation will be derived
and applied to problems in engineering, physics, and other subjects. Appli-
cations to boundary value problems in partial differential equations will be
emphasized.
Our study of the Laplace transformation leads to the theory of expand-
ing functions in series of the characteristic functions of Sturm-Liouville
systems. Such expansions form the basis of the method of solving boundary
value problems by separation of variables, a classical method of great
importance in partial differential equations.3 Furthermore, we can use that
theory to adapt the integral transformation (2) to certain types of linear
boundary value problems.
2 DEFINITION OF THE LAPLACE TRANSFORMATION
Ir a function F(t), defined for all positive values of the variable t, is multiplied
by e-sI and integrated with respect to t from zero to infinity, a new function
f(s) of the parameter s is obtained; that is,
{O e-SIF(t) dt = f(s).
As indicated in the preceding section, this operation on a function F(t) is
called the Laplaee transformation of F(t). It will be abbreviated here by the
symbol L{F}, or by L{F(t)}; thus
L{F} = foaJe-SIF(t) dt.

The new function f(s) is called the Laplaee transform, or the image,ofthe
objeet funetion F(t). Wherever it is convenient to do so, we shall denote the
I Oliver Heaviside, English electrical engineer, 1850-1925.
2 For historical accounts see J. L. B. Cooper, Heaviside and the Operational Calculus, Math.
Gazette, vol. 36, pp. 5-19, 1952, and the references given there.
3 That method is presented in the author's "Fourier Series and Boundary Value Problems,"
2d ed., 1963.
4 SECo 2] OPERATlONAL MATHEMATICS
object function by a capital letter and its transform by the same letter in
lowercase. But other notations that distinguish between functions and their
transforms are sometimes preferable; for example,
tfJ(s) = L{ f(t)} or .9(s) = L{y(t)}.
For the present, the variable s is assumed to be real. Later on, we shall
let it assume complex values. Limitations on the character of the function
F(t) and on the range of the variable s will be discussed soon.
Let us note the transforms of a few functions. First, if F(t) = 1 when
t > O, then
1 <X>
];
<X>
L{F} = f
o
e-sI dt = --e-st
s o
1
hence, when s > O, L{I} = -.
s
Ir F(t) = elctwhen t > O,where k is a constant, then
1 <X>
]
<X>
L{F} = fo ~e-st dt = -e-(s-k)1

k-s o
;
hence, when s > k, L{e"'}=~.s-k

With the aid of elementary methods of integration, the transforms of many
other functions can be written. For instance,
k s
L{ sin kt} = S2 + . '), L{cos kt} = S2 + ., ')
when s > O; but soon we shall have still simpler ways to obtain those trans-
forms.
It follows from elementary properties of integrals that the Laplace
transformation is linear in the sense defined by Eq. (1), Seco1. We can illus-
trate the use of this property by writing
L !e"I- !e-lct =!~ - !~.

{2 2 } 2s-k 2s+k'
when s> k and s> -k; that is,
L{sinhkt} = s - ~ (s > Ikl).

THE LAPLACE TRANSFORMATION [SEC.3 5
PROBLEMS
1. Use the linearity property and known transfonns to obtain these transfonnations,
where a, b, and e are constants:
(a) L{a + bt} = asS2

+b (s > O);
(b) L{a + bt + et2} = L{(a + bt) + et2}

as2+bs+2e
(s > O);
S3
a + bs
(s > O);
(e) L{asint + bcost} = S2 + 1
s
(s> lel);
(d) L{cosh et} = S2 - e2
(s> a and s> b).

(e) L{e'" - ebt} = (s _aa~sb- b)
2. Use trigonometric identities, such as 2 COS2 t = 1 + cos2t, 2 sinat sinbt =
cos (a - b)t - cos (a + b)t, and known transfonns to find f(s) when s > O, in case
F(t)is
(a) cos2 t; (b) sin2 t; (e) sin t sin 2t; (d) sin t cos t; (e) sin3 t =W- cos 2t) sin t.
s2+2 2 4s
Ans.(aL, ~2 . ,,; (b)s(S2+ 4); (e)(r + l)(r + 9);
1 6
(d)-r-- ; (e) " .. -.'
s +4
.
3. Show that the linearity property (1),Seco1, can be extended to linear combinations
of three or more functions when all the transfonns existo
4. If for all functions of some class and for every constant C a transfonnation T
satisfies the two conditions
T{F(t) + G(t)} = T{F(t)} +T{G(t)},

T{CF(t)} = CT{F(t)},
prove that the transformation is linear.
3 SECTIONALL y CONTINUOUS FUNCTIONS. EXPONENTIAL

ORDER
A function F(t) is seetionally continuous on a bounded interval a < t < b
if it is such that the interval can be divided into a finite number of subinter-
vals interior to each of which F is continuous and has finite limits as t
approaches either end point of the subinterval from the interior.
Thus the values of such a function may take at most a finite number of
finite jumps on the interval (a,b). The class of sectionally continuous functions
includes functions that are continuous on the closed interval a ~ t ~ b. The
integral of every function of this class, over the interval (a,b), exists; it is the
sum of the integrals of the continuous functions over the subintervals.
The unit step function
Sit) =O when O < t < k,
=1 when t > k,
is an example of a function that is sectionally continuous on the interval

O < t < T for every positive number T (Fig. 1). The Laplace transform of
this function is
oo . oo
1
f
oo
f o
Sk(t)e-SI dt =
k
e-sI dt = --e-sI;J
S k
thus whenever s > O,

-ks
e
L{Sit)} = s'
A function F(t) is of exponential order as t tends to infinity, provided
some constant IXexists suchthat the product
e-"'/IF(t)1
is bounded for all t greater than some finite number T. Thus F(t)1does not
grow more rapidly than M e"'las t -+ 00, where M is some constant. This is
also expressed by saying that F(t) is of the order of e"'/,or that F(t) is ((J(e"'/).
The function Sk(t) above, as well as the function tn, is of the order of
e"'las t -+ 00 for any positive IX;in fact, for the first function and, when
n = O,for the second, we may write IX= O. The function l is of exponential
order (IX~ 2); but the function el1is not of exponential order.
Ir a function F(t) is sectionally continuous on each bounded interval
O < t < T and if, for some constant a, F is ((J(e"'f),.then
the Laplace transform
of F exists whenever s > IX.This follows from a well-known comparison test
o (k,O) (T,O)
Fig. 1
. for the convergenceofimproper integrals (seeProbo14,Seco5). For in view

of the sectional continuity of F, and consequently of the product e-S'F(t),
that product is integrable over every bounded intervalO < t < T. AIso,
since F is C9(ea'),
a constant M exists such that for all positive t
le-S'F(t)1 < Me-(s-a),.
But the integral from O to 00 of M e- (s-a)' exists when 8 > OC;consequently,

not only the convergence but also the absolute convergence of the Laplace
integral .
is ensured when 8 > OC.

The above conditions for the existence of the transform of a function
are adequate for most of our needs; but they are 8ufficient rather than
necessary conditions. The function F may have an infinite discontinuity at
t = Ofor instance, that is, IF(t)1 --+00 as t --+O,provided that positive numbers
m, N, and T exist, where m < 1, such that IF(t~ < N/tm when O < t < T.
Then if F otherwise satisfies the above conditions, its transform still exists
because of the existence of the integral
{T e-S'F(t) dt.
For example, when F(t) = Ct, its transform can be written, after the
substitution of x for fit, in the form
'" 2 '"
(1)
fo
t-te-s, dt = -
JS o l e-x2 dx (8 > O).
The last integral has the value fi/2 (Prob. 10, Seco5); hence
(8 > O).
4 TRANSFORMS OF DERIVATIVES
By a formal integration by parts wehave
L{F'(t)} = 1'" e-S'F'(t)dt
= e-S'F(t)J~ + 8 {'" e-S'F(t)dt.

Let F(t) be of order of elZlas t approaches infinity. Then whenever s > oc,the
-
bracketed term becomes F(O),and it follows that
(1) L{ F'(t)} = sf(s) - F(O),
where f(s) = L{ F(t)}.
Therefore in our correspondence between functions differentiation of
the object function corresponds to the multiplication ofthe result function by
its variable s and the addition of the constant - F(O).Formula (1)thus gives
the fundamental operational property of the Laplace transformation, the
property that makes it possible to replace the operation of differentiation by
a simple algebraic operation on the transformo
As noted above, formula (1) was obtained only in a formal, ormanipula-
tive, manIler. It is not even correct when F(t) has discontinuities. The
following theorem will show to what extent we can rely on our formula.
Theorem 1 Let the function F(t) be continuous with a sectionally continuous

derivative F'(t), over every finite intervalO;;;;t ;;;;T. Also let F(t) be of
order of elZlas t -. oo. Then when s > oc,the transform of F'(t) exists, and
(2) L{F'(t)} = sL{F(t)} - F(O).
Since F(t) is continuous at t = O,the number F(O)here is the same as
F( +0), the limit of F(t) as t approaches zero through positive values.
To prove this theorem, we note first that
L{F'(t)} = lim
T oo o f T e-SIF'(t) dt,
ifthis limit exists. We write the integral here as the sum ofintegrals in each of
which the integrand is continuous. For any given T, let ti' t2," ., tn denote
those values of t between t = O and t = T for which F'(t) is discontinuous
(Fig. 2). Then
T ll l2 T
f o
e-SIF'(t)dt =
fo
e-stF'(t)dt +
f
I1
e-SIF'(t)dt +oo. +
fIn
e-SIF'(t)dt.
.F1t)
I
I
I
I
I
I
I I
01 (tl,O) (t2,OI (T,O)
Fig.2
After integrating each of these integrals by parts, we can write their sum as
tl t2 T T
e-stF(t)
]o
+ e-stF(t)
] tI
+ ... + e-stF(t)
]tn
+s
f
o
e-StF(t)dt.
Now F(t) is continuous so that F(t 1 - O) = F(t 1 + O),etc., and hence
(3) {T e-stF'(t)dt= -F(O) + e-STF(T)+ s{T e-stF(t)dt.

Since IF(t)1< Meat for large t for some constants O(and M, it follows
that
le-sTF(T)1 < Me-(s-a)T,
and since s > 0(,this product vanishes as T ~ oo. AIso the last integral in
Eq. (3) approachesL{F} as T ~ '00 because F is continuous and lP(eat).
Hence the limit as T ~ 00 of the right-hand member of Eq. (3) exists and
equals - F(O) + sf(s); therefore, the same is true of the left-hand member.
Thus Theorem 1 is pro ved.
If F is continuous when t ~ O except for a finite jump at to, where
to > O,the other conditions remaining as stated in the theorem, the above
proof is easily modified to show that our formula (2) must be replaced by the
formula
(4) L{ F'(t)} = sf(s) - F(O) - [F(to+ O)- F(to - O)]e-sto.
The quantity in brackets is the jump of F at to.
We use the symbol F' here and in the sequel to denote the function whose
value is the derivative of F wherever the derivative exists. In the case of our
step function Sk(t), for instance, S~(t) = O when O < t < k and when t > k,
but S~(k)has no value.
To obtain the transform of the derivative F" of the second order, we
apply Theorem 1 to the function F'. Let both F and F' be continuous when
t ~ O and lP(ea.t);also let F" be sectionally continuous on each bounded
interval. Then
L{F"(t)} = sL{F'(t)} - F'(O)
= s[sL{ F(t)} - F(O)]- F'(O).
Hence we have the transformation
(5) L{ F"(t)} = s2f(s) - sF(O) - F'(O).
When Theorem 1 is applied to F1n-1)(t)to write
L{Fln)(t)} = sL{Fln-l)(t)} - F1n-l)(0)
and again to write L{Fln-l)(t)} in terms of L{Fln-2)(t)}, and so on, the follow-
ing result is indicated.
Theorem 2 Let F and each of its derivatives of order up to n - 1 be con-

tinuous functions when t ~ O and &(eC%t); also let F(n)(t)be sectionally
continuous on each bounded intervalO < t < T. Then the transform of
the derivative F(n)(t)exists when s > ex,and it has the following algebraic
expressio~ in terms of the transform f(s) of F(t):
(6) L{F(n)(t)} = snf(s) - sn-1F(0) - sn-2F'(0)
- s"-3F"(0) - ... - F(n-l)(O) (s > ex).
Theorem 2 can be pro ved by using Theorem 1 and induction. Under
the conditions stated, suppose that formula (6) is valid when n is replaced by
some integer k where O < k < n. But Theorem 1 expresses L{ F(k+1)} in
terms of L{ F(k)}to which (6)now applies to show that formula (6)is true when
n is replaced by k + 1. But the formula is true when k = 1, according to
Theorem 1; it is therefore true when k = 2, hence when k = 3,. . . , n.
The Laplace transformation resol ves the differential form F'(t) in
terms of f(s), s, and the initial value F(O) when F satisfies the conditions
stated in Theorem 1. As a consequence ofTheorem 1, Theorem 2 shows how
that transformation resolves the iterates F"(t), F"'(t),... of that form in
terms of f(s), s, and initial values of F and its deriva tives.
5 EXAMPlES. THE GAMMA FUNCTION

In order to gain familiarity with the above fundamental operational property
of the transformation, let us first use it to obtain a few transforms.
Example 1 Find L{t}.

The functions F(t) = t and F'(t) = 1 are continuous, and F is
&(eC%t)
for any positive ex..Hence,
L{F'(t)} = sL{F(t)} - F(O) (s > O),

or L{l} = sL{t}.
Since L{ 1} = l/s, it follows that
1
L{t} = s2 (s > O).
Example 2 Find L{sin kt}.

The function F(t) = sin kt and its deriva tives are all continuous
and bounded, and therefore of exponential order, where ex= O. Hence
L{ F"(t)} = S2L{ F(t)} - sF(O) - F'(O) (s > O),
or -k2 L{sin kt} = s2L{sin kt} - k.
Solving for L{sin kt}, we see that
L{sinkt} = s + ~ (s > O).
Example 3 Find L{ tm} where m is any positive integer.

The function F(t) = tm satisfies all the conditions of Theorem 2
for any positive ex.Here
F(O) = F'(O)= .. . = F(m-l)(O)= O,
F(m)(t)= m!, F(m+l)(t) = O.
Applying formula (6) when n = m + 1, wefind that
L{F(m+l)(t)} = O = sm+lL{tm} - m!,
and therefore
(1) (s > O).
This formula can be generalized to the case in which the exponent

is not necessarily an integer. To obtain L{tk} where k> -1, we make
the substitution x = st in the Laplace integral, giving
(s > O).
The integral on the right represents the gamma function, or factorial

function, with the argument k + 1. Hence
(2) L{tk} = r(ks + 1) (k > -1, s > O).
Formula (1) is a special case of (2) when k is a positive integer (see

Probo 13).
Example 4 Find L{fh F(r) dT} when F is sectionally continuous and of

exponential order.
The function
(3) G(t)= {F(T) dT

is continuous (Sec. 14),and G(O)= O.AIso G'(t) =
F(t), except for those
values of t for which F(t) is discontinuous; thus G'(t) is sectionally
continuous on each finite interval. Ir the function G(t) is also l!J(e"t),
then according to Theorem 1,
L{ G'(t)} = sL{ G(t)} = L{ F(t)} (s> ex);
thus, if a > Oso that s > O,
(4) (s > a > O).

L{{ F(T)dT} = ~f(S)
To show that the integral (3) represents a function of exponential
order when F is sectionally continuous and of exponential order, we
first note that constants a and M exist such that jF(t)1< M eatwhenever
t ~ O, and if the number a is not positive, it can be replaced by a
positive number. Then
t t M
IG(t)1~ IF(T)I dT < M S ea'dT = -(eat - 1) (a > O),
SO o a
and therefore
(a > O).
This establishes the exponential order of the function (3).
PROBLEMS
1. State why each of the functions
- 1 when O< t < 2
(a) F(t) = te2t and (b) G(t) = t+ 1
{1 when t > 2,
is sectionally continuous on every interval.O < t < T and (!)(e"t)as t -+ 00, where a > 2
for F and a ~ Ofor G.
2. State why neither of the functions (a)(t - 1)-1 or (b)tan t is sectionally continuous
on the intervalO < t < 3.
3. Ir Sk(t)is the unit step function (Sec. 3),draw graphs of the following step functions
-
and find their transforms when s> O: (a) F(t) = 1 Sl(t); (b) G(t) = Sl(t) - S2(t).
AIso, note that both F and G vanish except on bounded intervals; thus their Laplace
integrals become definite integrals that exist for all s ( - 00 < s < 00). Show that
/(0) = 1 and g(O) = 1,and that
(a)les) = (1 - e-")/s (s:;' O);
-
(b) g(s)= (e-S e-2s)/s (s:;' O).
4. Obtain these transforms with the aid of Theorem 2:
(a) L{coskt} = S/(S2 + p) (s> O);
(b) L{sinh kt} = k/(S2 - k2) (s> IkJ).
5. Given the transform of e"', use Theorem 1 to show that
1
(s > k).
L{te'"} = (s - k)2
-
6. If G(t)= OwhenO;;; t ;;;k and G(t) = t k whent ;;;k, drawgraphs of Gand G'.
Giventhe transformof Sit), (a)apply Theorem 1to prove that g(s)= s-2e-ks (s> O);
(b) show that G(t) = f~Slr) d! and find g(s) from formula (4), Seco5.
7. Prove that the function F(t) = sin (eI2)is of exponential order (IX ;;;O)and that its
derivative F' is not of exponential order. Show that Theorem 1 ensures the existence
of the Laplacetransformof F' when s > O,in this case whereF' is not of exponential
order.
8. (a) Derive Eq. (4),Seco4. (b)Illustrate that equation by using it to find the transform
of the unit step function Sk(t).
9. (a) If a function F and its derivative F' are both (J)(e"')and continuous when t ;;;O
except possibly for finite jumps at a point to, and if F" is sectionally continuous on each
intervalO < t < T, apply Eq. (4), Seco4, to derive the formula
L{F"(t)} = s2f(s) - sF(O) - F'(O) - se-SIO[F(to+ O)- F(to - O)]
- e-srO[F'(to + O) - F'(to - O)] (s > IX,to > O).
(b) Show that the formula applies to the function

F(t) = sin t when O ;;; t ;;; 1t, F(t) = O when t ;;;1t
to give f(s) = (1 + e-~S)/(s2+ 1)when - 00 < s < oo.

10. Let J denote the second integral in Eq. (1), Seco3; then
J2 = fo'" e-x2 dx t'" e-y2 dy = fo'" fo'" e-(X2+y2)dx dy.

Evaluate the iterated integral here by using polar coordinates and show that J = .;:;r/2.
11. Prove that each linear combination AF(t) + BG(t) of two functions F and G of
exponential order is also of exponential order. -
12. Use properties of continuous functions to show that iftwo functions are sectionally
continuous on an interval (a,b) then (a) each linear combination of the two is also
sectionally continuous on that interval; (b)the product ofthe two functions is sectionally
continuous on the interval.
13. As noted in Seco5, the gammafunction is definedfor positivevaluesof r by the
formula
(r >0).
r(r) = t'" x,-Ie-x dx
(a) Integrate by parts to show that the function has the factorial property
f(r + 1) = rf(r).
(b) Show that f(1) = 1,and hencethat f(n + 1)= n! whenn = 1,2,....
(e) From the value of the integral J found in Probo 10, show that r(t) = .;:;r.
Then use the factorial property to find nt)
and formula (2)to show that L{Jt} = t.Jir/si.
14. Improper integrals Consideronly functions that are sectionally continuous on
each intervalO < t < T.
14 SEC.6] OPERATIONAL MATHEMATICS
(a) If O ;;;g(t) ;;;h(t) whenever t > O and if S~ h(t) dt exists, use the definition of
the improper integral
['" g(t) dt = lim [T g(t) dt

Jo T-",Jo
to prove the existence of that integral. [Note that the value of the last integral is non-
decreasing as T increases. It never exceeds 1 where 1 = S;> h(t) dt because S~ g(t) dt ;;;
f{;h(t) dt ;;;1, so it has a limit.] .
(b) Iff;> Ip(t)1dtexists,provethatf;> p(t)dtexistsby writingp(t) = (P(t)+ Ip(t)l]-
Ip(t)l. Note that O ;;;p(t) + Ip(t)1;;;2Ip(t)1 and apply the test in part (a) to those two
components of p(t).
(e) Prove this comparison test: If Iq(t)1;;;r(t) whenever t > O and if S;
r dt exists,
then the improper integral S'" q(t) dt is absolutely convergent, and the integral itself
exists. o
6 THE INVERSE TRANSFORM
Let the symbol L - 1 {f(s)} denote a function whose Laplace transform is f(s).
Thus if
L{ F(t)} = f(s),
then F(t) = L -l{f(S)}.
Using two of the transforms obtained in the foregoing sections, we can write,
for instance,
k
L -1 ~ = ekl, L -1 = sinkt:
{s - k } { S2 + k2 }
This correspondence between functions f(s) and F(t) is called the inverse
Laplace transformation, F(t) being the inverse transform of f(s).
In the strict sense of the concept of uniqueness of functions, the inverse
Laplace transform is not unique. The function F1(t) = ekt is an inverse
transform of 1/(s - k); but another, for instance, is the function (Fig. 3)
F2(t) = ekt when O < t < 2, or t > 2,
= 1 when t = 2.
For the transform of Fz{t) is
foa)e-srF2(t)dt = f: e-s1ekldt + ta) e-srl dt,

and this is the same as L{ ekl}. The function F2(t) could have been chosen
equally well as one that differs from Fl(t) at any finite set ofvalues oft, oreven
at such an infinite set ast = 1,2,3,....
THE LAPLACE TRANSFORMATION [SECo6 15
FZ(tI!
../
I
f(2,1)
I
I
I I
O
Fig.3
A theorem on uniquenessof the inversetransform will be proved later

(Sec. 69). To state it, we first define a c1ass of functions of exponential order.
Let the class tff denote the set of all functions F(t) defined on the half
line t > O,sectionally continuous on each bounded interval, and defined at
eachpoint to whereF is discontinuousas the mean value ofits limitsfrom the
right and left,
F(to) = t[F(to + O)+ F(to - O)] (to > O);
also, for each individual function F of the set, let constants M and IXexist
such that IF(t)1< M ea.twhen t > O. As we have seen, F then has a transform
f(s) defined on some half line s > IX.
The theorem states that no two functions of class tff can have the same
transforms. Thus if f(s) is the transform of some function F(t) of c1ass tff,
then F(t) is the unique inverse transform L -1 {J(s)} in tff.
For'example, the only function L -1{e-S/s} ofc1ass rf is the unit step
function S 1 (t) defined in Seco3 if S 1(1) is defined to be t. As another example,
-
the only function L - 1 {1/(s k)} of c1ass tffis ekt.
. It is well to note here that not everyfunction of s is a transformo The
kind of functions f(s) that are transforms of functions F(t) of broad c1asses
are limited, as we shall see (Chap. 6), by conditions of regularity that include
requirements that f be continuous on a half line s > IXand that f(s) -+ Oas
s -+ oo.
We have noted that if L{ F} and L{ G} exist, then
(1) L{ AF(t) + BG(t)} = Af(s) + Bg(s)

whenever A and B are constants. Let us restrict our functions of t to those of
c1ass tff, and functions of s to transforms of such functions. Then unique
inverse transforms exist, and the linearity property (1) can be written
(2) L -1{Af(s) + Bg(s)} = AF(t) + BG(t)

= AL -1{J(S)} + BL -1{g(S)};
that is, L - 1 is also a linear transformation of functions.

The most obvious way of finding the inverse transform of a given

function of s consists of reading the result from a table of transforms. A
fairIy extensive table is given in Appendix A. But we shall take up methods
of obtaining inverse transforms of certain combinations and modifications of
functions of s, as well as methods of resolving such functions into those
listed in the tables. With the aid of such procedures, we shall be able to make
much use of the transformation. In addition, there are explicit formulas for
L -l{f(S)}. The most useful of these formulas involves an integral in the
complex planeo To use this integral, we must let s be a complex variable and
we must be prepared to employ some theorems in the theory of functions of a
complex variable.
7 A THEOREM ON SUBSTITUTION
Let a function F(t) be such that its Laplace integral converges when
s > IX.Then, replacing the argument of the transform f(s) by s - a,wherea
is a constant, we have
when s - a > IX.Therefore
(1) f(s - a) = L{ e'"F(t)} (s > IX+ a).
Let us state this simple but important property as a theorem.
Theorem 3 The substitution of s a for the variable s in the transform

~
corresponds to the multiplication of the object function F(t) by the

function e"', as shown in formula (1).
To iIlustratethis property, let us recall that
mI
5"'; 1 = L{ tm} (m = 1,2,... ; s > O).
m!
Hence (s > a).
's - a)m+ 1 = L{ tme"t}
As another iIlustration,
s
L{coskt} = S2+ (s > O),
s+a
and therefore (s> -a).
L{e-a, cos kt} = (s + a)2 + k2
8 THE USE OF PARTIAL FRACTIONS (TABLE 1)

A few examples will show how the theory of partial fractions can be used in
finding inverse transforms of quotients of polynomials in s. In the next
chapter, a more systematic use of this procedure will be introduced.
Example 1 Find L -l{(S + 1)/(S2 + 2s)}.

The denominator of the function of s here is of higher degree
than the numerator and has factors that are linear and distinct.. There-
fore constants A and B can be found such that
s+l A B
-=-+-
s(s + 2) s s +2
for all valuesof s exceptOand - 2. Clearing fractions, we have
s + 1 = (A + B)s + 2A,
and this is an identity if A + B = 1and 2A = 1. Thus A = B = !, and
hence
s+l 11 1 1
+--
. - 2 s 2 s + 2'
Since we know the inverse transforms of the two functions on the right,
we have the result
L-
S + 1 =! ! - 2/
1
{ S2 }
+ 2s 2 + 2 e .
The procedure can be shortened for such a simple fraction by
writing
s + 1 = !(s + 2) + !s,
s+l 1 1 11
and hence +--
S(S+ 2) - 2 s 2 s + 2'
Example 2
Find L -lL(s : a)2}'
In view of"the repeatedlinear factor, we write

a2 A B e
=-+-+
s(s + a)2 s s+ a (s + a)2'
Clearing fractions and identifying coefficients of like powers of s as
before, or else by noting that
a2 = (s + a)2 - s(s + a) - as,
Table 1 A short table of transforms
F(t) f(s) a (s > a)
1 1 -1 O
s
- 1
2 e'" a
s-a
n!
3 t" (n = 1,2,...) O
5"+1
n!
4 t"e'" (n = 1,2,...) a
(5- a)"+ 1
k
5 sin kt O
S2 + k2
s
6 cos kt O
S2 + k2
k
7 sinh kt 1'"
S2 - P
5
8 coshkt Ikl
S2 - k2
k
9 e-al sinkt -a
(5 + a)2 + k2
. 5+a
10 e-al cos kt -a
(5 + a)2 + k2
11 Ji fi O
2.j?
1
12 O
Ji
13 r(k + 1) O
t1(k > -1) 7+1
14 r(k + 1) a
t1e"'(k>-1)
(5 - a)1+1
15 SJt) (Sec. 3) -e-u O

s
a-b
16 e'" - ebl(a > b) a
(s - a)(s - b)
b2 - a2
17 -sinat--sm
1 1 . bt O
a b (s2 + a2)(52 + b2)
18 cos at - cos bt
(b2 a2)s - O
(S2 + a2)(52 + b2)
THE LAPLACE TRANSFORMATION [SEC.8 19
we find that
a2 1 1 a
s(s + af = -;- s + a - (s + a)2.
Referring to Table 1, we can now write the result
a2
L-l = 1 - e-al - ate-al.
{ s(s + a)2}
Example 3
Since
s S (S2 + a2) - (S2 + b2)

(S2 + a2)(s2 + b2) = a2 - b2 (S2 + a2)(s2 + b2)
Example 4 Find F(t) if
f(s) =. Ss + 3
~. I '"
In view of the quadratic factor, we write

Ss + 3 A Bs + e
-~+
(s - 1)(s2+ 2s + 5) - s - 1 S2+ 2s + 5.
After clearing fractions and identifying coefficients of like powers of s,
wefindthat A = 1,B = -1, and e = 2; thus
1 s- 2
f(s) = -s - 1 -
1 s+l 3
=-- +-
s- 1 (s + 1)2+ 4 (s + 1)2+ 4.
Referring to Table 1, or to Theorem 3, we see that
F(t) = el - e-'(cos 2t - ! sin 2t).
9 THE SOLUTION OF SIMPLE DIFFERENTIALEQUATIONS

The application of the Laplace transformation to the solution of linear
ordinary differential equations with constant coefficients, or systems of such
equations, can now be made clear by means of examples. Such problems
can of course be solved also by methods studied in a first ~ourse in differential
equations. Later on, when we have developed further properties ofthe trans-
formation, we shall solve problems of this sort with greater efficiency. We
shall also be able to solve more difficult problems, especially in partial differ-
ential equations.
Example 1 Find the general solution of the differential equation

(1) Y"(t) + k2 Y(t) = O.
Let the value of the unknown function at t = O be denoted by
the constant A and the value of its first derivative at t = O by the
constant B; that is,
Y(O)= A, Y'(O)= B.
In view of the differential equation, we can write
L{ Y"(t)} + P L{ Y(t)} = O,
assuming that Yand Y" have transforms. If the unknown function Y
satisfiesthe conditions in Theorem 2, then
L{ Y"(t)} = s2y(S) - As - B,
where y(s) = L{ Y(t)}. Hence y(s) must satisfythe equation
s2y(S) - As - B + k2y(s)= O,
which is a simple algebraic equation. Its solution is clearly
s B k
y(s) = A s2 + k2 _
+ k s2 + - . ~
Now Y(t) = L -l{y(s)},and the inversetransformsofthefunctions

on the right in the last equation are known. Hence
(2)
Y(t) = A cos kt + ~ sin kt,
= A cos kt + B' sin kt,

where A and B' are arbitrary constants since the initial values Y(O)and
Y'(O)are not prescribed.
To verify our formal result given by formula (2), we need only to
find Y"(t) from that formula and substitute into Eq. (1) to see that the
differential equation is satisfied regardless of the values of A and B'.

Thus it is not necessary to justify the use of Theorem 2. However, our
function A cos kt + B' sin kt does satisfy all conditions in that theorem,
and the order of the steps taken above can be reversed to show in
another way t)lat our function satisfies the differential equation. These
remarks on verifying the solution apply also to the other examples and
problems that follow in this section.
Example 2 Find the solution of the differential equation

(3) Y"(t) - Y'(t) - 6Y(t) =2
satisfying the initial conditions
(4) Y(O) = 1, Y'(O) = O.
Applying the transformation to both members of the differential
equation, and letting y(s) denote the transform of Y(t), we obtain
formally the algebraic equation
2
s2y(S) - S - sy(s) + 1 - 6y(s) = -,
s
where we have used the initial conditions in writing the transforms of
Y"(t) and Y'(t). Hence
S2 - s + 2
(S2 - S - 6)y(s) = ,
s
or y(s) =
S2 - s+2 A
=-+-+-.
B e
s(s - 3)(s + 2) s s- 3 s+ 2
Evaluating the coefficients A, B, and e as in the preceding section, we
find that
11 8 1 4 1
y(s) = ---+--+--.
3s 15 s - 3 5 s + 2
Hence Y(t) = -j + /se3t + !e-2t.
It is easy to verify that this function Y satisfies the differential equation
(3) and both conditions (4).
Example 3 Find the functions Y(t) and Z(t) that satisfy the following system
of differential equations:
Y"(t) - Z"(t) + Z'(t) - Y(t) = et - 2,
2 Y"(t) - Z"(t) - 2Y'(t) + Z(t) = - t,
Y(O) = Y'(O)= Z(O) = Z'(O) = o.
22 SEC.9] OPERATIONAL MATHEMATICS
Let y(s)and z(s)denotethetransformsof Y(t)and Z(t),respectively.

Then in view of the differential equations and the initial conditions,
those transforms formally satisfy the following simultaneous algebraic
equations:
1
s2y(S) - S2Z(S) + sz(s) y(s) = s- 1 - -,
-
2
s _
1
2s2y(s)- S2Z(S)- 2sy(s)+ z(s) = -2"'
s
These equations can be written .
s-2
(s + l)y(s) - sz(s) =
s(s - 1)2'
1
2sy(s) - (s + l)z(s) =
S2(S- 1)"
Eliminating z(s), we find that
s2-2s-1
(S2- 2s - l)y(s) = s(s - W.
With the aid of partial fractions, we then find that
1 B A e
y(s) = s(S - 1)2 = S + s - 1 + (s - W
1 1 1
=---+
s s - .1 (s - 1)2'
Therefore Y(t) = 1 - e' + te'.
Likewise we find that
2s-1 1 1
z(s) = -'-'- -j 1)2 = -2"S + ;-- H?>
and therefore Z(t) = -t + te'.

Example4 Solvethe problem
Y"'(t) - 2Y"(t) + 5Y'(t) = O,
Y(O)= o, Y'(O)= 1,
Let e denote the unknown initial value Y"(O). Then
S3y(S) - s - e - 2s2y(s)+ 2 + 5sy(s)= O,

so that
C-2+s
y(s) = S(S2 - 2s + 5)
C- 2 1 s - 1 C+ 3 2
= S- [-:;- (s - 1)2 + ]
4 + 10 (s - 1)2 + 4'
C - 2 c + 3 C- 2
Thus Y(t) = --s- + e'( 10 sin2t - S- cos 2t) .
Since Y(n/8) = 1,it followsthat
C- 2 e"/8
1 = -+-(C + 3 - 2C + 4),
5 10j2
or that C = 7. Hence the solution is
Y(t) = 1+ e'(sin2t - cos 2t).
PROBLEMS
1. Use Theorem 3 to (a) obtain entry 9 from entry 5 in Table 1 and (b) show that
L -l{(S - a)-t} = e"'(1tt)-t, given L -l{S-t}.
2. Use partial fractions to find the inverse transforms of
a a2 a3
(a) s(s + a); (b) s(S2+ a2); (e) s(s + a)3'
Ans. (a) 1 - e-a,; (b) 1 - cos at; (e) 1 - (1 + at + ta2t2)e-a,.
3. Use partial fractions to obtain the inverse transforms shown in (a) entry 16 of
Table 1; (b) entry 17 of Table 1.
Solvethe followingproblemsand verifythat your solutionssatisfythe differential
equatios and any accompanyingboundary conditions.
4. Y"(t)- k2Y(t)= O(k # O). Ans. Y(t) = Ce'"+ C2e-kt.
5. Y"(t)- 2kY'(t) + k2Y(t) = O. Ans. Y(t) = e"'(C+ C2t).
6. Y"(t)+ k2Y(t) = a. Ans. Y(t) = C sinkt + C2coskt + a/k2.
7. Y"(t)+ 2Y'(t) + 2Y(t) = O,Y(O)= O,Y'(O)= 1. Ans. Y = e-' sin t.
8. Y"(t)+ 4Y(t)= sint, Y(O)= Y'(O)= O. Ans. Y(t)= sint - i sin2t. t
9. Y"(t)+ Y'(t) = t2 + 2t, Y(O)= 4, Y'(O)= -2. Ans. 3Y(t) = t3 + 6e-' + 6.
10. Y"'(t)+ Y'(t) = lOe2',Y(O)= Y'(O)= Y"(O)= O.
Ans. Y(t) = e2'- 5 - 2 sint + 4 cos t.
11. y(4/(t) = Y(t), Y(O) = Y'(O) = Y"(O) = O, Y"'(O) = 2.
Ans. Y(t) = sinh t - sin t.
12. X'(t) + Y'(t) + X(t) + Y(t) = 1, Y'(t) - 2X(t) - Y(t) = O,X(O)= O,Y(O)= 1.
. Ans. X(t) = e-r - 1, Y(t)= 2 - e-'.
13. Y"(t)+ 2Z'(t)+ Y(t) = O,Y'(t) - Z'(t) - 2Y(t) + 2Z(t) = 1 - 2t, Y(O)= Y'(O)=
Z(O)= O. Ans. Y(t) = 2(1 - e-' - te-'), Z(t) = Y(t) - t. .
14. Y"(t)+ Y(t) = t, Y'(O)= 1, Y(n) = O. Ans. Y(t) = t + n cos t.
15. y"(x) + y(x) = 1,y(O)= 1,y(tn) = O. Ans. y(x) = 1 - sinx.
16. y"(x)+ 2y'(x)+ y(x) = O,y(O)= O,y(l) = 1.
10 GENERATION OF THE TRANSFORMATION

We can show in a manipulative way why a linear integral transformation of
functions F(t) defined on the half line t ~ O,
(1)
T {F(t)} = L CX)K(t,s)F(t) dt = f(s),
must be essentially the Laplace transformation if it is to have the basic

operational property of that transformation, namely, that it replaces the
simple differentialform F'(t) by an algebraicform inf(s), s and the initial value
F(O)of the function F.
Let primes denote differentiation with respect to t. Then by a formal
integration by parts we can write
(2)
T {F'(t)} = SoCX)
K(t,s)F'(t) dt
= K(t'S)F(t)J~ - SoCX)K'(t,s)F(t)dt
= - K(O,s)F(O)- SoCX)
K'(t,s)F(t) dt,
provided that K(t,s)F(t) -> O as t -> oo. The kernel K must involve some
parameter s if the transform f is to correspond to a unique function F of
some large class; otherwise, f is merely the numerical value of the integral
(1) of the product K(t)F(t), a number that is the same for many functions F.
In order that the final integral in formula (2) will representf(s) except
for a factor A.(s),where A.is some function of s, we require that
(3) K'(t,s) = - A.(s)K(t,s).
Thus K(t,s) = ce-J.'. But it is convenient to write c = 1 and choose A.(s)as

the parameter, so we write A.(s)= s here. Then
(4) . K(t,s) = e-s"

T{F(t)}= LCX) e-stF(t)dt = L{F(t)},
THE LAPLACE TRANSFORMATlON [SECo 10 26
and formula (2) is the basic operational property of the Laplace transforma-
tion:
(5) T{F'(t)} = -F(O) + sT{F(t)}.

A corresponding procedure will be used later on for the generation of
integral transformations that will reduce other differential forms in functions
defined on some prescribed interval, in terms of the transform of the function
itself and prescribed boundary values of the function or its derivatives at the
ends of the interval.
Readers acquainted with the concept of a function F(t) (a ~ t ~ b) as
a generalized vector1 can see that a linear integral transformation
(6)
T{F(t)} = f K(t,s)F(t)dt = /(s)
is a generalization o/ linear vector trans/ormations. Our transformation (1) is
a special case of (6) in which a = Oand b is infinite.
A generalized vector, or function, F has an infinity of components
consisting of the values F(t) for each t; that is, the components are all the
ordinates of points on the graph of F(t) over the interval a ~ t ~ b. The
generalized scalar product, called the inner product, of two such vectors F
and G is the generalized sum of products of corresponding components,
namely
r F(t)G(t) dt.
Thus for each fixed s, the integral (6) is the inner product of the generalized
vectors K(t,s) and F(t).
We begin with ordinary vectors F in three-dimensional space having
rectangular cartesian components F1,F2,F3, which may be written F(l),
F(2), F(3). A general linear transformation of F into a new vector f can be
written in terms of components and constants Kij (i,j = 1,2,3). For the
first component we write
/1 = KllF1 + K12F2 + K13F3 = K1 . F
where K1 is a fixed vector with components Kll, K12, K13, and the dot
denotes the scalar product ofvectors. Similarly, K2 and K3 are fixed vectors
and
/2 = K2 F, .
/3 = K3 F. .
In space of n dimensions the general linear vector transformation is
determined by a familyof n fixed vectors K:
(7) ; = K . F (i = 1,2,..., n).
1 Churchill, R. V.:"Fourier Series and Boundary Value Problems," 2d ed., chap. 3, 1963.
The extension to generalized vectors, or functions, is now fairly natural.

A fixed function of the variable t and a parameter s, K(t,s), is a family of
generalized veetors, one veetor for eaeh fixed value of s of some set, say
e ~ s ~ d. The inner produet of K and F representsthe s component of
the transformed veetor f(s); that is,
(8)
f(s) = f K(t,s)F(t) dt.
This is the linear integral transformation (6).
Functions of certain elasses,as well as vectors, are elements of linear
veetor spaees. The abstract theory of linear spaees is independent of the
kind of elements involved. In that respeet the treatment of funetions as
veetors is not merely an intuitive approach or an analogy. But the abstraet
theory specifies properties of inner products without suggesting formulas
for them. The theory will not be needed here.
2
Further Properties of the
Transfor.mation
11 TRANSLATION OF F(t)
There are several further operational properties of the Laplace transforma-
tion that are important in the applications. Those properties whose deriva-
tions and applications do not necessarily involve the use of complex variables
will be taken up in this chapter.
We begin with an analog of Theorem 3 of the first chapter. According
to that theorem, the multiplication of the object function by an exponential
function corresponds to a linear substitution for s in the transformo Now
let us note the correspondence arising from the multiplication of the trans-
form by an exponential function.
Let F(t) ha ve a transform,
f(s) = {X) e-S'F(t) dt.
Then
27
28 SECo 11] OPERATIONAL MATHEMATlCS
where b is a constant, assumed to be positive. Substituting t + b = T, we

can write the last integral in the form
{X>e-StF(T- b)dT= s: O+ {X>e-StF(T - b) dT.

Thus if we define a function Fb(t)as follows,
(1) Fb(t) = O when O < t < b,
= F(t - b) when t > b,
we see that
f(s)e-bS = Lco e-S'Fb(T) d'C.
The following property is therefore established.
Theorem 1 If f(s) = L{F(t)}, thenfor any positive constant b,

(2) e-bSf(s) = L{Fb(t)},
where Fb(t)is the function defined by Eq. (1).
The function Fb(t)is illustratedin Fig. 4. Its graph is obtainedby
translating the graph of F(t) to the right through a distance of b units and
making Fb(t)identically zero between t = Oand t = b. We can refer to Fb(t)
as the translated function.
Our unit step function Sb(t) is the translation of the function So(t) = 1
(t > O). It serves as a familiar illustration of the above theorem, since its
transform is s-le-bs. This step function can be used to describe the transla-
tion of any function F(t) by writing. .
(t > O),
provided that F(t - b) is defined where t > O; that is, provided that F(t) has
numerical values for those values of t in the range t > - b. For example,
the function F(t) = sin kt is defined for all t, so in this case we can write
k -bs
(t > O,b ~ O).
Fb(t) = Sb(t)sin k(t - b) = L- 1 { S2 e+ k2 }
Rt)
o o
I
L
~
(b,O
Fig.4
FURTHER PROPERTIES OF THE TRANSFORMATION [SECo 12 29
On some occasions it is convenient to define F(t) as zero for all negative

values of t. When that is done, the graph of Fb(t) is simply a translation of
the graph of F(t), and
for all t.
Now consider the linear substitution esfor s in f(s), where e is a positive

constant. If the transform f(s) of a function F(t) exists whenever s > IX,then
whenever es > IX.

f(es) = {O e-CStP('r)d't
The substitution t = e't enables us to write that formula as
f(es) = ~foooe-~p(~) dt
in terms of the transform of F(t/e), to establish this theorem:
Theorem 2 [f L{ P(t)} = f(s) whenever s > IX,and if e is a positive eonstant,

then
(3)
We can write a = l/e to obtain an alternate form of (3):
(4) (a > O,s > alX).

L{F(at)} = ~f(~)
Given, for example, that L{ cos t} = S/(S2+ 1) when s > O,it follows
that
L {cos at} = -1 s/a

= 2
s
2 (s > O).
a (s/a)2+ 1 s +a
The effectof a generallinear substitution for s can beseenfrom formula
(3) and Theorem 3, Chap. 1, since
(5) (a > O).
12 STEP FUNCTIONS
When t ~ O,the bracket symbol [t] is used in mathematics to denote the
. . . , that does not exceed the number t. Thus [n] = 3
greatest integer, O, 1,2,
= [3]. The function [t]is therefore a step function of the type that is some-
times called a staircase function with unit rise and run :
[t] =O (O ;;;;
t < 1),
=1 (1 ;;;;
t < 2),
=2 (2 ;;;;
t < 3),.. ..
The staircase function with an arbitrary positive ron h, and with a unit rise
beginning at the origin t = O (Fig. 5), is then represented by the symbol

[1 + t/h] or 1 + [t/h]. The function c[l + t/h] has the rise e and the ron h.
To obtain the transform of the function shown in Fig. 5,
(1) (h > O),

Y(t) = [1 + i]
we may describe the function by means of a difference equation of the first

order together with an initial condition, as follows.
Y(t) = Y(t - h) + 1 (t ~ O),
=0 (t < O).
The function Y(t - h) is then the same as the translated function Yh(t) and,
in view of Theorem 1, the transform y(s) of the function Y(t) satisfies the
equation
- 1
y(s) = e h'y(S) + -
s (s > O).
Therefore the transform ofthe staircase function (1) is
=L l+ !. = !--.! >
(2) y(s)
{[ h] } s1 - j (s O).
Another useful step function is the unitfinite impulse function
1
(3)
(h, t - to) = h when to < t < to + h,
= O when t < to and when t > to + h,
10,1)
O Ih,OI
Fig.5
l(h,t-fo)
11
I '1
II I
I
'hl
1-:J
o (fo,O)
Fig.6
illustrated in Fig. 6. Let us introduce a variation of our unit step function

Sk(t), one that is defined to be zero when t is negative; namely,
(4) So(t - to) =O when t < to,
= 1 when t > to.
Then our function 1 can be written
(5)
It follows at once from the transform of So(t - to) that

1 -hs
(6) L{I(h, t - to)} = e-sto -, e (to ~ O,h > O,s> O).
It is clear from the definition (3) that

(7) lim I(h, t
h O
- to) = O when t :F to.
AIso, the area under the graph of the function 1 is unity for every positive h;
hence
(8)
h
lim
O f -et)
et) I(h, t - to) dt = 1.
The order of the positions of the limit and the integral here is important, for
in view of Eq. (7)
f et)
-et)
lim I(h, t
h O
- to) dt = O.
It is interesting to note the behavior of the transform of the finite unit
impulse function as h tends to zero. By evaluating the limit ofthe right-hand
member of Eq. (6), we find that
(9) limO L{I(h, t
h
- to)} = e-sto (to ~ O,h > O,S > O).
But note that
L{fim
h OI(h, t - to)} = L{O} = O.
Equation (9) presents the exponential function e -sto, or exp ( sto), not as the -
transform of a function, but as a limit of transforms of functions of the set
I(h, t - to) consisting of one function for each positive value ofthe parameter
h. Thus exp(- sto)approximatesthe transformof the impulsefunction
I(h, t - to) for small positive values of h, when to ~ O.
Example A partic1e of mass m, initially at rest at the origin X = O,is

subjected to a force in the form of a unit finite impulse I(h,t) beginning
at the instant to = O,in the direction of the X axis. 1ts displacements
X(h,t), which satisfy the conditions
(10) mX"(h,t) = I(h,t), X(h,O) = X'(h,O) = O (h > O),
where X' = dX/dt, could be found here by integration.
But if we use transforms to solve for X(h,t), we write
1 -e -hs
(11) ms2x(h,s)= L{l(h,t)}= . ,
assuming that X and X' are continuous functions of t of exponential

order. Thus
1 2 2 -hs
mx(h,s) = 2h ( S3 - S3e )
and therefore
(12)
that is
1 2
(13) when O ~ t ~ h,
X(h,t) = 2mh t
t2 - (t - h)2 1 1
= 2mh = ;;; ( t - 2.h) when t ~ h.
That function satisfies all conditions (10).

Now if h -+ O,the second of formulas (13) applies when t > Oto
give the limiting displacements X(t):
(14) X(t) lim X(h,t) = ~

= h-+O m
(t > O).
The displacements X(t) correspond to the idealized unit impulse given

to m at die instant t = O. As a verificationof the solution (14)of that
idealized case, we note that X"(t) = Owhenever t > O,and X( +0) = O.
Furthermore, the particle has the momentum mX'(t) = 1 whenever

t > O; thus if it is initially at rest, then its momentum jumps suddenly
from O to 1 at the instant t = O. The instantaneous impulse has the
same effect as an initial velocity l/m.
13 THE IMPULSESYMBOL 15(t- to)

The above example is a simple illustration of problems in differential equa-
tions involving instantaneous impulses. In order to reduce the number
of steps in formal solutions of such problems, we shall occasionally use the
unit impulse symbol c5(t- to) to a limited extent indicated below.
Let a function F(t), defined for all real t, be continuous on some interval
to ~ t ~ to + k. Then according to the basic law of the mean for integrals
to each h(O < h < k), there corresponds a number 0(0 < O < 1) such that
1 'O+h
fa>
- a>I(h, t - to)F(t) dt =h f
lo F(t) dt = F(to + Oh).
Consequently the impulse function 1 has the property
(1) lim
h"'O f a>
-a>
I(h, t - to)F(t) dt = F(to),
sometimes called the sifting property of selecting that value F(to) of F(t).
We abbreviate that property by writing
(2)
f:a> c5(t- to)F(t) dt = F(to),
where the entire left-hand member is a symbol that denotes the operation
represented by the left-hand member of Eq. (1). That symbol is selected so as
to suggest an integral of a function, because if c5(t- to) is replaced by
-
I(h, t to), the symboI does represent an integral whose value is approxi-
mately F(to) when h is a small positive number.
If to ~ O,then I(h, t - to) = Owhen t < Oand
(3) lim
h"'O of a> I(h, t - to)F(t) dt = F(to) (to ~ O,h > O)
according to Eq. (1); that is, in symbolic form
(4) (to ~ O).

foa> c5(t - to)F(t) dt = F(to)
When F(t) = 1, formula (2) becomes
(5)
. f~a> c5(t- to)dt = 1,
the symbolic form of Eq. (8), Seco 12. When F(t) = e-sI and to ~ O, the
symbolic integration formula (4) can be written
(6)
which is the symbolic fomi of Eq. (9), Seco12.

The letter f>used in our symbolic integrals has further significance in
differential equations. We usethe example treated in Seco12 as an illustration.
Let us write
(7) mX"(t) = f>(t); X(O) = X'(O) = O,
to signify that X is the limit as h -+ Oof the solution X(h,t) of problem (10)
there, the problem with f>(t)replaced by [(h,t). In the transformed problem
(11),Seco12,we may let h tend to zero and assumethat x(h,s)-+ x(s) where
x(s) = L{ X(t)} ; theresultis
(8)
But that result is obtained at onceby formally applying the operator L

to terms in problem (7)if L{ t>(t)}= 1,in accordance with formula(6),while
thesymbolL{ X"(t)} meansthe limit ash -+Oof the transformof X(h,t) so
that we may be justified in replacing it by S2X(S).Thus mx(s) = 1/s2,andour
result mX(t) = t followso
That symbolic method gives limits of solutions, as h -+ O, of other
problems in differentialequations with [(h, t - to) asa forcingfunction.We
shall use the method occasionally in the sequel. Results obtained by such
manipulations cleariy need to be verified if they are to be relied upon. More
often we sacrifice the brevity gained by using the f>symbol and useinstead the
function [(h, t - to), then find the limit of the solution as h -+ 00
The symbol t>(t - to), often called the Dirac delta function, is not a
function. In particular, it is not the limit of [(h, t - to) as h -+ O; that limit is
a function with value zero everywhere except at the one point t = to, a
function whoseintegral is zero (Sec.12). The symbolt>(t- to)may, however,
be replaced by the function [(h, t - to), when h is small, to give results that
approximate the effect of t>.
Generalizations of functions, called generalized functions or distribu-
tions, have been developed since 1950 which include the symbol f>(t to) and -
its generalized derivatives. There are now a variety of theories that give
sound developments of distributions. But even the introductory presenta-
tions of the theory of distributions seem too lengthy to present here.1
1 See, for instance; the artic1e entitled From Delta Functions to Distributions, by A. Erdlyi, in
"Modero Mathematics for the Engineer," Second Series, 1961, or A.-E. Danese, "Advanced
Calculus," vol. 2, pt. 6, 1965, or references listed in those books.
PROBLEMS
1 . With the aid of our theorems and known transforms find the inverse transforms
F(t) tabulated below and draw a graph of F(t).
I(s) F(t)
e-bs
(a) ----(b > O) 0(0 < t < b);t - b(t > b)
s
(b) fis-te-2'
e -,..
(e) S2 + 1 - S.(t) sin t
sint(O < t < n);O(t > n)
se -s/2
(e) S2 + n2
4
(f) (2s + 1
4
(g)--
2. From the Maclaurin series that represents (1 - X)-l when Ixl < 1, or from the
sum of an infinite geometric series, show that
1 1 1 e-los e-2h. 00 e-.Ios
-s 1 - e-Ios=-+-+-+"'=
s s s L- s
.=0
(h > O, s > O).
Apply the inverse transformation term by term to this infinite series, formally, to obtain
the result shown in Eq. (2), Seco12; namely,
-1 1 1 t
(h > O).
L {-;1 - e-los
} = 1 + h] [
3. Show that Eq. (2), Seco12,can be written in the form
L{[1 + ~]} = ~( 1 + coth ~).

4. (a) Use the formal method indicated in Probo 2 to find F(t) when
1 2
I(s) = -sl+e' =- (s > O).
Draw the graph of F to see that F can be written in terms of the bracket symbol [t] in
the form
F(t) = 1 + (-1)1'1.
(b) Show that the function F in part (a) is described by the following difference
equation of the first order, when t > O:
F(t) + F(t - 1) = 2 where F(t) =O when t < O.
Transform that difference equation to verify the formula given in part (a) for j(s).
5. Apply Theorem 2 to the transformation found in Problem 4 to show that
L 1 + (- di] = !si +2..-

e M (h > O, s > O).
{ }
6. Find the function Y(t) that satisfies the following difference equation of the first
order and the accompanying initial condition.
Y(t) - eY(t - h) = F(t)
Y(t) = O when t < O,
where e and h are constants and h > O,and where F(t) = Owhen t < O. The expansion
of (1 - ce-hS)-l in powers of ce-hs is helpful here (compare Probo 2). Show that the
solution can be written
00
Y(t) I c"F(t -
= n=O nh);
but for each fixed value of t this series is a finite series because F(t - nh) = O when
t - nh < O. Thus an altemate form of the solution is
Y(t) = F(t) + cF(t - h) + c2F(t - 2h) + ... + cmF(t- mh),

wherem = O,1,2,... , whenmh < t < (m+ l)h. Verifythe solution.
7. Find the function Y(t)that satisfiesthe followingdifferenceequation of the second
order and the accompanyinginitial condition.
Y(t) - (a + b)Y(t - h) + abY(t - 2h) = F(t)
Y(t) = O when t < O,
where F(t) = Owhen t < O. The constants a and b are such that a # b, and the constant
h is positive. Note that the solution of the transformed problem can be simplified, with
the aid of partial fractions in the variable e-M, to
j(s) a b
y(s) = (
a - b 1 - ae-M- 1 - be-M' )
1 00
I (an+l - bn+l)F(t- nh).
Ans. Y(t) = -a - b n=O
00
8. Solve Probo 7 when b = a. Ans. Y(t) = I (n + l)a"F(t - nh).

n=O
9. Solve the difference-differential equation
Y'(t) - a Y(t - 1) = F(t),
where F(t) = b when t > Oand F(t) = Owhen t < O,under the ~ondition that Y(t) =O
when t ;;;; O.
a
Ans Y(t) =b t + -(t - 1)2 +... + -(t a" - n)"+l
. [ 2! (n+1)! J'
where n ;;;; t ;;;;n + 1 and n = O,1,2,.. . .
10. Initially a particle of mass m moving along the X axis is at the origin X = O with
a velocity Vo. The only external force that acts on that particle is an instantaneous
impulse Poat time to, in the X direction. Thus the displacement X(t) satisfies the follow-
ing symbolic equation and initial conditions:
mX"(t) = Poc5(t- to), X(O) = O,X'(O) = Vo.
Find the formal solution (writing L{X"} = S2X - vo)
X(t) m -
= vot + PO(t to)So(t - to) (t ~ O,to > O)
and show X(t) graphically. To verify the solution, show that X"(t) = O when t # to,
that X(O) = Oand X'(O) = Vo,and that the momentum mX'(t) undergoes a jump Po at
the instant to.
11. Replace the instantaneous impulse Poc5(t- to) in Probo 10 by the finite impulse
PoI(h,t - to) and then find the displacements X(h,t). Show that the limit of X(h,t) as
h Ois the function X(t) found before.
12. Problem 10 can be stated in terms of functions as folIows:
mX"(t)= Owhen t # to, X(O)= O, X'(O)= Vo

mX'(to + O)- mX'(to - O)= Po (to > O),
where X and X' are continuous except for the specified jump in X'. Use our formula
(Prob. 9, Seco5) for the transform of the second derivative in this case to obtain the
formula found before for X(t).
13. Show formalIy that the solution of the symbolic problem
X"(t) + k2 X(t) = Po c5(t - to), X(O) = X'(O) = O, (to ~ O)
where k and Po are constants, is (if we write L{X"} = S2Xhere)
14. Show that the formal solution of the symbolic problem (if we write L{ X'} = sx
here)
X'(t) = c5(t- to), X(O)= O (to ~ O)

-
is X(t) = So(t - to). In this sense, then, c5(t to) corresponds to the derivative of the
unit step function So(t - to) as we could anticipate from the symbolic formula
15. Let Al denote the change in /(h, t - to) when !J.t = -h. Draw the graph of the
function of t represented by the difference quotient
Al - /(h, t - to) - /(h, t - h - to)

T- h
If F and F' are continuous on an interval that includes the points to and to + 2h, show
that .
Al 1
f a)
-F(t)
-a) h
dt = --
h 'o f'OH F(-r: + h) - F(-r:)
h
d-r:
1 'O+h
= - 'o f F'(-r:+ 8h)d-r:

where O < 8 < 1, and hence that
-Al
lim
a)
f
, 0 -00 h
F(t) dt = - F'(to)'
A symbolic form of that formula is
f:", <5'(t - to)F(t)dt = -F'(to);

which is a symbolic integration by parts:
f:a) <5'(t- to)F(t)dt = - f:a) - <5(t to)F'(t)dt= -F'(to)'

In particular, if to ~ O,note that
L{<5'(t - to)} = - ddt e-sr] '='0 = se-sro.

16. LetFbecontinuous(t ~ O)exceptforajumpjoatto(to> O).ThenF(t)- joSo(t- to)
is continuouswhenevert ~ Oif F(to)is properlydefined. If F is exponentialorder and
F' is sectionallycontinuous,then
L{fc[F(t) - joSo(t - to)]} = {f(S) - jo e~sro] - F(O).

Show that this formula agrees with formula (4), Seco4, and note that its symbolic form
may be written (Prob. 14)
17. Since /(h,t) = l/h when 0< t < h and vanishes when t < O and when t > h,
describe the function /(h, - t) and show that
!~L: F(t)/(h, - t) dt = F(O)

when F is defined for all t and continuous over some intervalltl < a. Symbolically,
f'", F(t)b(-t) dt = F(O)= f:", F(t)b(t)dt;

thus b has the symbolic property
b( - t) = b(t).
14 INTEGRALS CONTAINING A PARAMETER
We now review some properties of integrals that will be useful.

Let x denote a para meter and f(x,t) a continuous function of the two
variables x and t together in a region a(x) ~ t ~ fJ(x),e ~ x ~ d, where a
and fJ are continuous functions. Then the integral
(l(X)
(1)
f a(x)
f(x,t) dt = g(x) (e ~ x ~ d)
represents a continuous function of x, which we have called g. Ifthe derivatives

a'(x) and fJ'(x) exist, and if of /ox as well as f is continuous, then the derivative
of the integral (1) is given by the generalized Leibnitz formula,
o
f
(l(X)
a(x) a
(2)
g'(x) = X f(x,t)dt + f[x,fJ(x)]fJ'(x) - f[x,a(x)]a'(x).
The foregoing two properties of the integral (1) are established in
advanced calculus from fairly elementary considerations. Let us indicate
here how the continuity requirements on the integrand can be relaxed in
establishing certain properties.
Let a, b, and e denote constants, and let the integrand of the integral
aX+b
(3)
f o f(x,t) dt = h(x)
(O ~ x ~ e)
be such that the region O~ t ~ ax + b, O~ x ~ e can be divided into

polygonal subregions by a finite number of straight lines (Fig. 7)
t = ax + b (i = 1,2,...,m)
and (j = 1,2,..., n),
where f(x,t) is continuous over each subregion up to its boundary; that is,
for each subregionf can be definedon the boundary in such a way that it is
continuous in x and t together over the closed subregion. We then say that f
is continuous by subregions.
e x Fig.7
01
When x 1 ~ X ~ X2 in Fig. 7, for instance, the integral can be written as

a sum ofintegrals ofthe type (1), such as
aX+b
(4) f a2X+ b2
f(x, t) dt
in which the integrand is continuous so that the integral represents a con-

tinuous function of x(x1 ~ x ~ X2). Similarly when x is between any two
adjacent x/s, h(x) is continuous. Consequently, the integral (3) represents
a seetionally eontinuous funetion h(x) on the interval (O,e)when f is continuous
by subregions.
As a special case, when the integrand is a sectionally continuous
function f(t), independent of x, the integral (3) represents a continuous
function of x on the closed interval [O,e].
The integral of the sectionally continuous function (3)
e e ax+b
(5)
fO h(x) dx = ff
O O f(x,t) dt dx
exists. It can be written as the sum of iterated integrals over each of the
subregions in which f(x,t) is continuous. When the order of integration is
interchanged in each subregion, the sum of the integrals represents the
iterated integral, with order interchanged, over the entire region.
Thus our condition of continuity by subregions permits the interchange
of order of integration with respect to t and x. In particular, when a = O so
that the limits are constants,
(6)
5: s: f(x,t) dt dx = s: 5: f(x,t) dx dt.
These conclusions can be reached by the same procedure for regions of more
general shapes. The argument used above applied, for instance, when the
lower limits ofthe integrals (5) and (6) are constants other than zero.
II

j
15 IMPROPER INTEGRALS
Corresponding properties can be established for convergent improper
integrals with the aid of a condition of uniform convergence with respect to
the parameter x. The concept of uniform convergence of integrals, which
applies only to improper integrals that involve a parameter, is illustrated in
the following example.
For each value of x(x ;; O)the particular Laplace integral
(1)
exists; it has the values

p(O) = O
(2)
p(x) = 1 when x > O.
It can be written as a definite integral plus a remainder,
p(x) = f: xe-xt dt + {X)xe-xt dt = 1 - e-xT + R(x,T),

whereR(x,T) = f~ xe-xt dt; then
R(O,T) = O
(3)
R(x,T) = e-xT when x > O.
Uniform convergence of the integral for all x in a prescribed interval

means not only that the integral exists for each x in the interval, but also that
the remainder can be made arbitrarily small in absolute value by taking T
sufficiently large, uniformly for all x in the interval. Thus when x ;; 1 and
is any small positive number, we see that
IR(x,t)1= e-xT ~ e-T <
when T> log (1M, a value that is independent of x, so that the integral
converges uniformly with respect to x when x ;; 1. In the same way we can
see that it converges uniformly when x ;; Xl whenever Xl > O.
But the convergence is not uniform in the range x ;; Oor x > O,because
the remainder (3) when x > O is small only if xT is large. Since x can be
arbitrarily small, the value of T that makes R(x, T) small must depend on x.
A consequence of this lack of uniform convergence in the range x ;; Ois the
discontinuity of the function p(x) at x = Othat is exhibited by Eq. (2).
The Weierstrass test for uniform convergence of the integral
(4) (O~ x ~ e)
{X' f(x,t) dt = q(x)
can be stated as follows. Letf(x,t) be continuous by subregions, in the sense

described in Seco14, over the region O ;;:;x ;;:;c, O ;;:;t ;;:;T, for each positive
constant T. Ir a sectionally continuous function M(t), independent of x,
exists such that
If(x,t)1 ;;:; M(t) (O ;;:;x ;;:;c, t > O)
and such that the integral f;;>M(t) dt exists, then the integral (4) is uniformly
convergent with respect to x(O ;;:;x ;;:;c). This test is established by first
noting that the integral (4) exists according to the comparison test (Probo 14,
Seco5),then that the absolute value of the remainder for the integral, for all x,
in the interval, do es not exceed the remainder f,;?M(t) dt in the integral of M.
The latter remainder is independent of x, and it tends to zero as T -. 000
The Weierstrass test sta tes useful sufficient, but not necessary, con-
ditions for uniform convergenceo Under those conditions the integral (4) is
also absolutely convergent.
Ir the integral (4) is uniforrnly convergent with respect to x and if its
integrand f(x,t) is continuous by subregions over the rectangle O ;;:;x ;;:;c,
O ;;:;t ;;:;T for each positive T, then the integral represents a sectionally
continuous function q(x); moreover
(5)
I: q(x) dx = I: {" f(x,t) dt dx = foooI: f(x,t) dx dt;
that is, the order of integration of the definite and improper integrals here can be
interchanged.
In particular, let f(x,t) be continuous in each rectangle O ;;:;x ;;:;c,
O ;;:; t ;;:; T, except possibly for finite jumps across a set of lines t = t(i =
1,2,. . o, n). Then q(x) is continuous ifthe integral (4) is uniforrnly convergent.
Suppose that Jf/Jxas well as f satisfies those continuity requirements and
that the integral f;;>(of/Jx)dt converges uniformly and the integral (4) exists.
Then the derivative of the latter integral exists and the order of differentiation
with respect to x and integration with respect to t can be interchanged:
d oo o
dxf f oo
(6) (O < x < c).
o f(x,t) dt = o oxf(x,t) dt
The continuity of q(x) can be seen by writing
(O ;;:; x ;;:;e).
q(x) = foTf(x,t) dt + R(x,T)
For each fixed T the integral on the right is a sum of integrals of the type
"
f"-1 f(x,t) dt
with continuous integrands; hence it is a continuous function of x so that the

change in its value is small when the change L1xin x is small. But IR(x,T)1and
IR(x + L1x,T)I are small for all x and x + L1xin the interval (O,e)as long as T
is taken sufficiently large, beca use of the uniform convergence of the integral
(4). Thus by taking Tlarge first, then lL1xlsmall, the continuity of q(x) follows;
that is, Iq(x + L1x) - q(x)1 is arbitrarily small when lL1xlis sufficiently small.
The proof of property (5) follows the plan commonly used in advanced
calculus, where f(x,t) is generally assumed to be continuous. The details are
left to Probo 11 at the end of Seco17.
Property (6)can be established in the following manner. Because ofthe
uniform convergence ofthe integral on the right in Eq. (6) and the continuity
properties of 8f/8x and f, for each r(0 < r < e) we can write
(7)
I: fooo : dt dx = fooo I:: dx dt
= fooof(r,t) dt - fooof(O,t)dt,
where we have used property (5) to invert the order of integration. Since
f:;' (8f/8x) dt is a continuous function of x, differentiationofthe first and last
membersof Eqs. (7)with respect to r leads to the equation
f_ ]
oo 8 d oo
o 8x
f(x,t)
x='
dt = - f
dr o f(r,t) dt
(O < r < e),
which is the same as Eq. (6) with x replaced by r.
16 CONVOLUTION
We now determine the operation on two functions of t that corresponds to
multiplying their transforms together. This convolution operation, which
gives the inverse transform ofthe product oftwo transforms directly in terms
of the original functions, is one of primary importance in operational
mathematics.
Let F(t) and G(t) denote any two functions that are sectionally con-
tinuous on each finite intervalO ;;;t ;;;T and (9(elXt),
and write
f(s) = L{F(t)}, g(s) = L{ G(t)} (s > a).
It will be convenient to define G(t) to be zero when t < O; but our final result
depends only on the values of F(t)and G(t)when t > O. According to Theorem
1, for each fixed -r('r;:;O),
e-Sfg(s)= L{G(t- -r)}= fooo e-stG(t- -r)dt,

where s > .CX.Hence
that is,
T
(1) f(s)g(s) = lim
ff
T-+oo o o
OO
F(T)e-srG(t - T) dt dT.
Since f(s) and g(s) exist when s > cx,the limit here exists.
The integrand ofthe inner integral in Eq. (1)is continuous by subregions
(Sec.14),over the rectangular region O ;;;;T ;;;;T, O ;;;;t ;;;;R, for each pair of
positive constants T and R. For if the jumps of F(T)and G(t) occur at
T = T~i = 1,2,..., m) and t = tJ{j = 1,2,..., n), then the jumps of the
integrand occur at the lines T = TIand t =T+ tj' In view of Eq. (5), Seco15,
the order of integration in Eq. (1) can be interchanged, provided that the
improper integral there is uniformlyconvergentwith respectto its parameter
T(O ;;;;T ;;;; T).
The uniform convergence is seen by noting that, owing to the exponen-
tial order of F and G, a constant N exists such that
(2)
where M(t) = N exp [-(s - cx)t]. The function M(t) satisfies the conditions
ofthe Weierstrass test (Sec.15); that is, M(t) is independent of Tand integrable
from zero to infinity.
Equation (1) can now be written in the form
(3) f(s)g(s) = lim

f
T-+oo o
ooe-sr
f o
T F(T)G(t - T)dTdt
= T-+oo
lim [11(T)+ lz(T)],
where
11(T) = foT e-sr IoTF(T)G(t - T)dTdt,
lz(T) = Irooe-sr I: F(T)G(t - T)dTdt.

In view of condition (2)
11z(T)1< N
fT ooe-(S-",)r
fo dTdt = s-cx
T NT e-(s-",)T;
therefore lim001z(T) = O.
T-+
FURTHER PROPERTIES OF THE TRANSFORMATION [SEC. 16 45
The region of integration for the integral 11(T) is the square O ~ 1: ~ T,

O~ t ~ T. But G(t - 1:)= Owhen 1:> t. Therefore
11(T) = f: {e-sI F(1:)G(t - 1:)d1: dt
and Eq. (3) reducesto
(4) f(s)g(s) = lim 11(T) =

T-ro f f
eo e-SI
o
l
o
F(1:)G(t - 1:)d1:dt.
The convolutionF * G of the functions F(t) and G(t) is defined as the

function
(5)
F(t) * G(t) = f~ F(1:)G(t - 1:)d1:,
so that Eq. (4) can be written
(6) f(s)g(s) = L{ F(t) * G(t)}.

We summarize our result as follows.
Theorem 3 1ff(s) and g(s) are the transforms oftwofunctions F(t) and G(t)
that are sectionally continuous on each interval O ~ t ~ T and of the
order of eal as t tends to infinity, then the transform of the convolution
F(t) * G(t) exists when s > IX; it is f(s)g(s). Thus the inverse transform of
the product f(s)g(s) is given by the formula
(7) L -l{J(S)g(s)} = F(t) * G(t).

The functions F(t) = t and G(t) = eal,for example, satisfy the conditions
of Theorem 3. Consequently
Partial fractions can also be used to obtain that resulto

When G(t) = F(t), we have the formula
(8) (f(s)JZ = L{F * F}.

As an example that will be useful in finding inverse transforms with the aid of
partial fractions, we note that
(9) L-l{(S2: P)2} = :2sinkt*sinkt
= :2 { sin k. sin k(t - .) d.
= 2~3(sin kt - kt cos kt).

The conditions stated in Theorem 3 are narrower than necessary for the
validity of formula (7). If F(t) = t-t for example, F is not sectionally con-
tinuous on an intervalO < t < T, but the Laplace integral of F(t) is absolutely
convergent and formula (7) is still valid if G(t) satisfies the conditions stated
in the theorem.
1
Thus L -1 - ~ *e' - 2t! r' e-<~
{ ~(s - 1)} - Fe - j Jo 2.fi.
If we make the substitution r = .J"i here, we find that
1 2e' ./l
(10)
L-1 { .;s(s-l) } = y1tr= f o e-r2dr=e'erf(.jh
where the error function erf (x), also callect the probability integral, is a
tabulated function defined by the equation
2 (X
(11) erf(x) = jJoe-r2dr.
It was shown in Probo 10 at the end of Seco5 that erf(oo) = 1.
Since the substitution of s + 1 for s in a transform corresponds to
multiplication of the object function by e-', it follows from Eq. (10) that
(12) L -1
~
{ s,s+l }
= erf(Ji).
17 PROPERTIES OF CONVOLUTION
By substituting the new variable of integration A = t - 1"in the convolution
integral (5), Seco 16, we find that the convolution operation is commutative;
that is,
(1)
F(t) * G(t)= G(t)* F(t) = {F(t - A)G(A)dA.
FURTHER PROPERTlES OF THE TRANSFORMATION [SECo 17 47
The operation is clearIy distributive with respeet to addition:

(2) F(t) * [G(t) + H(t)] = F(t) * G(t) + F(t) * H(t).
AIso, F * (kG) = k(F * G) if k is a eonstant.
Properties (1) and (2) are valid whenever the funetions are sectionalIy
eontinuous over an interval (O,T) that eontains the point t. For sueh fune-
tions the operation is also associative:
(3) F(t) * [G(t) * H(t)] = [F(t) * G(t)] * H(t).
To prove this we first write, in view of Eq. (1),
F*(G*H) = f~ F(-r:)
f~-t G(t - T - A.)H(A.)dA.dT
and observe that the iterated integral here represents an integration over tbe
triangular region shown in Fig. 8. When the order of integration is reversed,
the integral becomes
{ H(A.) {- A F(T)G(t - A.- T) dT dA.,
whieh represents H * (F * G) or (F * G) * H. Note that for eaeh fixed value

of t the integrand of the iterated integral is a funetion of T and A.that is
continuous by subregonsover tbe regon O ;;;;A. ;;;;t - T, O ;;;;T ;;;;t, because
of the sectional eontinuity of the funetions F, G, and H. Tbe interehange
of order of integration is thereforejustified (Sec. 14).
Tbe eonvolution of a seetionalIy eontinuous funetion F(t) and our
unit step funetion Sk(t)can be written
(4) when O ;;;;t ;;;;k,

F(t) * Sk(t) = f~ Sir:)F(t - T) dT = O
' '-k
= ik F(t - T)dT = i o F(A.)dA.
when t ;;;k.
Fig.8
o
Since the integral of a sectionally continuous function F(A.)is a continuous

function of its upper limit when the lower limit is a constant (Sec. 14) and
since the function (4) is continuous at t = k, it followsthat F * Sk is con-
tinuous on the intervalO ;;;;t ;;;;T ir F(t) is sectionally continuous there.
A proof, based on the continuity of F * Sk>that the convolution of any pair
of sectionally continuous functions is a continuous function, will be left to the
problems.
If two functions F(t) and G(t) are sectionally continuous on each
interval (O,T) and of the order of e'" as t tends to infinity, then for some
constant M 1
(t > O).
But ir ( is any positive number and ir M 2 represents the maximum value of

the function te-d when t ;:;O, then
where M = M 1 M 2' Consequently F * G is of exponential order:

(5) IF(t) * G(t)1< Me(a+<)1 ( > O,t > O).
Now, if three functions F(t), G(t), and H(t) satisfy the conditions of
Theorem 3, then the function G * H satisfies those conditions and its
transform is g(s)h(s). Therefore
f(s)g(s)h(s) = L{F(t) * [G(t) * H(t)]};
but since the convolution operation is associative, this equation can be
written in the form
(6) f(s)g(s)h(s) = L{F(t) * G(t) * H(t)}.
A similar formula can be written for the product of n transforms.

When G(t) = 1 in Theorem 3, we have a result noted earlier (Sec. 5):
Theorem 4 Division of the transform of a function by s corresponds to

integration of thefunction between the limits Oand t:
(7)
L -1Hf(S)} = I~ F(T)dT,
(8)
L -1 {s12f(S)} = {I: F(A.) dA. dT,
etc., for division by sn, provided F(t) is sectionally continuous and of the
order of eal(ex> O),wheres > ex.
FURTHER PROPERTIES OF THE TRANSFORMATION [SEC. 17 49
As examples, we note that
L - 1 L(s2 : k2)} = {sin k1:d7:= t(1 - cos kt),
L -1 {S2(S2 k+ k2)} = {s: sin kA.dA d7: = :2 (kt - sin kt).
PROBLEMS
1. Use transfonns to show that
tS
(a) 1 * 1 * 1 = tt2; (b) t * t * t = 5!;
(e) L 't2COS(t - 't)d't = 2t - 2sint.

2. Show that an alternate fonn of the transfonnation (8) Seco17,is
L -1 {~f(S)} = t * F(t) = t {F('t)d't - L 'tF('t) d't.

3. Showthat
1 1 1 1
)S-I=s-l+ JSl+s-l
( )
and in viewof fonnula (10),Seco16,that
L -1
{ --d-
v' s - 1}
= el+ ~ + e'erf(Jt).
v' 1[t
4. Generalize fonnula (10), Seco 16, by replacing s by sla2 to obtain transfonn 40,
Appendix A, Table A.2.
5. Generalize fonnula (12), Sec. 16, to obtain transfonn 44, Appendix A, Table A.2.
6. Find L -I{(S + 1)-1s-t} and generalize to obtain transfonn 41, Appendix A,
Table A.2.
7. Obtain the inverse transfonn
L-I{SS~\} =e'erf(Jt)-2~.
8. With the aid of transfonnation (9), Seco16, show that
L -1 {(s211 )3} = 3 (sin t - t cos t) - t2 sin t.

9. Let t = h and t = k be the points of discontinuity of a sectionallycontinuous
functionG(t)on an intervalO < t < T, and let jh andA denotethejumpsin thevalue
of G(t)at those points. With the aid of a graph of G(t),note that the function
(O~ t ~ T)
is continuous if its values at t = h and t = k are properly defined. Write F * G in

tenns of F * G, and of convolutions like the one in Eq. (4), Sec. 17, to show that if F(t)
is sectionally continuous, then F * G is continuous. AIso note how the result can be
generalized to pennit G(t) to have n jumps in the interval.
10. Show that the integral
00 x
u(x) = 1o I + x 2 t 2 dt
has the values u(x) = -n/2 when x < O, u(x) = n/2 when x > Oand u(O) = O. Examme
the remainder R(x, T) for this improper integral and show that the integral is not
unifonnly convergent with respect to x over any interval that contains the point x = O
in its interior or as an end point.
11. Prove property (5), Sec. 15, under the conditions stated there. This can be done
by first showing that for each positive number T
[ q dx = [s: f dt dx + [R(X, T) dx
and hencethat, correspondingto each positivenumber E,a number T.existssuch that
when T > T,
12. Extend property (5),Sec. 15,for the interchange of order of integration of definite
and improper integrals, to the following case:
I: Loo <jJ(x,t)dt dx = I: f <jJ(x,t) dx dt + Loo I: <p(x,t) dx dt
where b 6; Oand <jJis continuous by subregions over the region b < x < e, x < t < T,
whenever T> e, and f:><jJ dt is unifonnly convergent with respect to x(b < x < e).
Suggestions: Show the region graphically. Introduce a functionfto which fonnula
(5), Sec. 15, applies by writing
f(x,t) = <jJ(x,t) whenb < x < e and t > x,
f(x,t) = O wheneitherx < b or t < X.
18 DIFFERENTIALAND INTEGRAL EQUATIONS

General nonhomogeneous linear differential equations with constant
coefficients can be solved with the aid of the convolution property.
Example 1 Find the general solution of the differential equation

(1) Y"(t) + k2 Y(t) = F(t)
in terms of the constant k and the function F(t).
We assume for the present that F(t) is sectionally continuous
and of exponential order, that the derivative Y"(t) of the unknown
function satisfiesthose conditions, and that Y'(t) is continuous and of

exponential order. Then the equation in the transforms of Y(t) and
F(t) can be written
S2y(S) - sY(O) - Y'(O) + k2y(s) = I(s),
where Y(O) and Y'(O) are arbitrary constants. Hence
1 k s Y'(O) k
y(s) = k S2 + kd(s) + Y(O)S2 + k2 + k ? . .?,
and therefore
Y(t) = ~ (sin kt) * F(t) + Y(O)cos kt + Y~O)sin kt.
This general solution of Eq. (1) can be written
(2) Y(t) =~ { sin k(t ...:. T)F(T) dT + C 1 cos kt + C2 sin kt,

where CI and C2 are arbitrary constants.
The function CI cos kt + C2 sin kt is the general solution of the
homogeneous equation that arises when F(t) = O in Eq. (1). To verify
the solution (2), it is therefore sufficient to show that the first term
on the right is a solution ofEq. (1). If F(T)hasjumps at T = ti' t2"'" t"
when O < T < t, that term can be written
1 " 'l '
Z(t) =k fo sin k(t - T)F(T) dT + . .. + k f~ sin k(t - T)F(T)dT.
The Leibnitz formula (2), Seco 14, applies to each of the integrals in
this equation; thus
'. '
Z'(t) =
io
cos k(t - T)F(T)dT+ ... +
f~ cos k(t - T)F(T)dT + O.
By applying that formula again to the integrals here and collecting

the resulting integrals, we find that
Z"(t) = -k { sin k(t - T)F('r) dT + F(t)
or Z"(t) = -k2Z(t) + F(t). Therefore Z(t) satisfies Eq. (1), and the
function (2) is verified as the general solution of that equation for all
values of t.
The function Y(t) given by Eq. (2) is continuous, together with
Y'(t), when F(t) is sectionally continuous, in view of the above ex-
pressions for Z(t) and Z'(t). The condition of exponential order on
F(t) was not used in verifying the solution (2); the condition that F(t)
be sectionally continuous, on an interval containing all values of t to
be considered, was sufficient.
An equation in which the unknown function occurs inside an integral

is called an integral equation. In certain applied problems, to be illustrated
in the following chapter, the integral in the equation is a convolution integral.
Such ntegral equations of the convolution type transform into algebraic
equations.
Example 2 Solve the integral equation
Y(t) = at + {Y(-r) sin (t - -r)d-r.

We can write this equation in the form
Y(t) = at + Y(t) * sin t
and transform both members to get the algebraic equation
a 1
y(s) = S2 + y(s)S2 + l'
whose solution is
y(s) = a(s~ + s~}.
Therefore Y(t) = a(t + it3),
which can be verified directly as the solution of the above integral
equation.
The general integral equation of convolution type has the form
(3)
Y(t) = F(t) + { G(t - -r)Y(-r)d-r,
where the functions F(t) and G(t) are given and Y(t) is to be found. Since
the transformed equation is
y(s) = f(s) + g(s)y(s),
the transform of the unknown function is
f(s)
(4)
y(s) = 1 - g(s)"
Even if Eq. (3) is modified by replacing Y(t) by linear combinations

of Y(t) and its derivatives, the transform of the modified equation is an
FURTHER PROPERTIES OF THE TRANSFORMATlON [SECo 18 53
algebraic equation in y(s). For instance, the integrodifferential equation
(5) aY(t) + bY'(t) = F(t) + { G(t - T)Y(-r)dT,

where a and b are constants, givesrise to the transformed equation
(a + bs)y(s) - b Y(O) = f(s) + g(s)y(s),
which is easily solved for y(s).
The equation
(6) (O < b < 1),

F(t) = {(t - T)-bY'(T)dT
is known as Abel's integral equation. The unknown function could of course
be considered here as Y'(t) instead of Y(t); but no advantage is gained by
doing so. The solution of that equation is
(7) Y(t) = Y(O) + r(b)r(11 - b)tb-l * F(t),

valid when F(t) satisfies certain conditions of continuity. The formal deriva-
tiODof this solutioD is left to the problems.
PROBLEMS
Solve the following differential equations.
1. Y"(t) - k2 Y(t) = F(t), if Y(O)= Y'(O)= O(k # O).
Ans. 2kY(t) = e'''f~e-k'F(r)dt - e-k'f'o~rF(t)dt.
2. Y"(t) - 2k Y'(t) + k2 Y(t) = F(t).
Ans. e-kIY(t) = CI + C2t + f~ (t - t)e-krF(t)dt.
3. 2Y"(t) - 3Y'(t) - 2Y(t) = F(t), if Y(O)= Y'(O)= O. Verify that your function
satisfies the differential equation and the initiaI conditions.
4. Y"(t) + 4Y/(t) + 5Y(t) = F(t), if Y(O)= Y(7t/2)= O.
Ans. e2IY(t) = sin t f~/2 F(t)e2rcos t dt - COS t f~F(t)e2' sin 1:dt.
5. Y///(t)- Y'(t) = F(t). Verifyyour result.
6. Y///(t)- Y"(t)+ Y'(t) - Y(t) = F(t),if Y(O)= Y'(O)= Y"(O)= O.
Ans. 2Y(t)= f~ F(t- t)(e' - cost ~ sint) dt.
7. Show that the solution of the system of differentiaI equations
X/(t) - 2Y'(t) = F(t), X"(t) - Y"(t) + Y(t) = O,
under the conditions X(O) = X'(O) = Y(O)= Y'(O)= O,so that F(O)= O,is
X(t) = f~ F(t) dt - 2 {F(t) cos (t - t) dt,
Y(t)= - L F(t)cos(t - t) dt.

54 SEC, 19] OPERATIONAL MATHEMATICS
8. Solve the following system and verify your result:

X'(t) + Y(t) = F(t), Y'(t) + X(t) = 1, X(O)= 1, Y(O)= o.
9. Solvefor Y(t) and verifyyour solution:
{ Y(t) dt - Y'(t) = t,
Y(O)= 2.
Ans. Y(t) = 1 + cosht.
10. Derive the solution (7) of Abel's integral equation (6).
11 . Solve the integral equation
= ate-'.
Y(t) = a sin t - 2 { Y(t)cos (t - t)dt.
Ans. Y(t)
12. Find the solution of the integralequation
Y(t) = a sinbt + e { Y(t) sin b(t - t) dt

(a) when b2 > be; (b) when b = e.
Ans. (a) Y(t) = ab(b2 - be)-t sin (tJb2 - be); (b) Y(t) = abt.
13. Solve the nonlinear integral equation
Ans. Y(t) = 1.
2Y(t) + {Y(t)Y(t - t)dt =t+ 2.
14. Write Y"(t) = V(t). (a) Ir Y(O)= a and Y'(O)= b show that
Y'(t) = 1* V(t) + b, Y(t) = t* V(t) + bt + a;
(b) then convert the initial-value problem
Y"(t) + a(t)Y'(t) + P(t)Y(t) = F(t), Y(O) = Y'(O) = O,
in Y(t) into the following integral equation in V(t):
V(t) + {[cx(t) + P(t)(t - t)]V(t) dt = F(t).
19 DERIVATIVESOF TRANSFORMS
When the Laplace integral
(1)
f(s) = 1'" e-S'F(t)dt
is formally differentiated with 'respect to the parameter s by carrying out

the differentiation inside the integral sigo, the formula
f'(s) = 1'" e-st(-t)F(t)dt = L{ -tF(t)}

is obtained. Another formal manipulation with the integral (1) indicates

that f(s) ~ O as s ~ oo. We shall establish conditions under which those
formulas are valido
First, we note that if F(t) is ofthe order of (!'"as t ~ 00,then the function
tnF(t), where n = 0,1,2,. . ., is of exponential order. Let be a positive
number. Then constants N 1 and N 2 exist such that
(t > O),
where N 2 represents the maximum value of the function tne-" when t > O.
Thus the function tnF(t) is of the order of ellOI,where CXo= cx + .
Also let F(t) be sectionally continuous on each intervalO < t < T.
Then tnF(t) has that property. The absolute value of the integrand of the
Laplace integral of tnF(t)satisfiesthe condition
(2) (n = O, 1,2, . . . ),
where N denotes a constant. Consequently when s > cxo,
(3)
therefore L{~F(t)} ~ Oas s ~ oo. Moreover, if s ~ CXlwhere CXl> cxo,then

according to condition (2),
ItnF(t)e-SII< Ne-(Il,-IlO)1 = M(t), .
where this eXPQnential function M(t) is independent of s and integrable

from zero to infinity. It follows from the Weierstrass test (Sec. 15) that the
Laplace integral
(n = o, 1,2,. . . )
IX) ~F(t)e-SI dt
converges uniformly with respect to s when s ~ CXl> CXo.
Theorem 5 1f F(t) is sectionally continuous and of the order of (!"I,then

each ofthe Laplace integrals L{F(t)}, L{tF(t)}, L{t2 F(t)},.. ., is uniformly
convergent when s ~ CXIwhere CXI> cx;moreover
(4) lim f(s) = O and lim L{~F(t)} =O (n = 1,2,...).
S-'co
The function F(t)e-SI and its partial derivative of each order, with
respect to s, satisfy our conditions for the validity of formula (6), Seco 15.
Hence differentiation with respect to s can be performed in Eq. (1) inside
the integral sign, and the following theorem is established.
Theorem 6 Differentiation of the transforrn of a function corresponds to the

multiplication of the function by t : -
(5) (n = 1,2,...);
moreover, pnl(s) -. O as s -. oo. Those properties hold true whenever
F(t) is sectionally continuous and ofthe order of e"', if s > IXinformula (5).
Since a function is continuous wherever its derivative exists, it is true
that f(s) and each of its derivatives is continuous when s > IX.
To illustrate the last theorem, we can note that since
k
(s > O),
S2 + ,-2 = L{sin kt}
it follows that
-2ks
(S2 + '.2\2 = L{- t sinkt}.
Thus we have a formula that is useful in finding inverse transforms with the
aid of partial fractions:
. 2ks
(6) { k} (s > O).
L t S10 t = (S2 + k2)2
Transformation (6) can be obtained by another method that is some-
times useful, that of differentiating a Laplace integral with respect to a
parameter k, independent of s. When s > O,the Laplace integral
s
(7) L{cos kt} = f OO
o
e-s'cosktdt = ~s +
converges, and the integral
(8)
L{ :k cos kt} = fooo:k(e-st cos kt) dt
= - 100e-S't sin kt dt
converges uniformly with respect to k for all real k, according to the
Weierstrass test. AlI integrands here are continuous functions of k and t.
. Hence formula (6), Sec. 15,applies to show that the integral (8) represents
the derivative, with respect to k, of the integral (7); that is,
a S 2ks
o
-L{tS1Okt} = ak( ? .
)
.? = -,? . .?,?,
and this is the transformation (6).
FURTHER PROPERTIES OF THE TRANSFORMATlON [SEC. 20 57
20 SERIES OF TRANSFORMS
A useful method of finding inverse transforms L -1 {J(s)} is that of repre-
senting the functionfby an infinite series ofknown transforms, then applying
the operator L -1 to the terms of the series. For certain types of series in
powers of l/s, conditions for the validity of the procedure will now be
established.
Theorem 7 Let f(s) denote the sum of an infinite series of positive integral
powers of l/s which is absolutely convergent when s > IX,where IX~ O:
1 ao al
(1) f(s) =
ex>
n=O
an-
~+l
= -
S
+ - + ...
S2
(s > oc~ O).
Then the power series in t obtained by applying the operator L - 1 to

that series term by term converges to a function F(t) whose transform
is f(s), when s > IX:
ex> tn
(2)
F(t) = n=Oan,n. = L -l{J(S)} (t ~ O),
where O! = 1. A Iso, F is continuous when t ~ O, and of exponential

whenever OCl> oc.
order (!.J(e"lt)
Since series (1) is absolutely convergent when s ~ OClwhenever
OCl> a ~ O,it follows that lanl/sn+1 ~ lanl/a1+1 -+ O as n -+ oo. Therefore a
constant M, independent of s, exists such that lanl/~+1 < M; that is,
(3) lanl < M~+ 1 (n = O,1,2,.. . ; s ~ al > a ~ O).
Ir we write s = al here it follows that, when t ~ O,
tn tn (IXlt)n
(4)
I
-
ann! = la1-
n n! <
= Mal- n! .
I
The series of terms (altnn! converges to e"lt, so by the comparison test

series (2) is absolutely convergent, and its sum is of exponential order,
(t ~ O,al > a).
AIso, F is continuous because it is the sum of a convergent power series in t.

Let s be fixed (s > IX)and let T be a positive number. The series of
the ternis ane-Sttn/n! is uniformly convergent with respect to t over the
intervalO < t < T according to the Weierstrass test for infinite series of
functions, for in view of condition (3),
tn (sT)n
lanle-st, (s > a,O < t < T);
n. < Ms-.-n.
the numbers on the right are independent of t, and they are terms of a
convergent series. The series of the continuous functions ane-slr'fn! can
therefore be integrated term by term over the bounded interval (O,T)
(cf. Probo 15, Seco21); thus
T
a
I
T
o e-SIF(t) dt L~
00
= n=ono I o e-sIr' dt
= n=O
I: anrL{r'}
noL
- f T
ooe-str' dt
]
= I: - n. fT ooe-sIr' dt .
n=O( S'a: 1 a~
)
By subtracting the last series from series (1) which converges to f(s), we
find that
oo
a
(5) f(s) - Io
T
e-SIF(t)dt =
00
L ~ fT
n=on. e-sIr' dt.
We are to prove that the difference (5) vanishes as T -+ 00; that is,
to each positive number we are to exhibita correspondingnumber T.
such that
(6) /f(S) - f: e-SIF(t)dtl< wheneverT> T.o

It will follow that f(s) = L{F(t)}; then Theorem 7 will be provedo
The series in Eq. (5) is absolutely convergent because
f.
lanl 00 -st.,. d ::::;lanl L {r'} = lanl
(~
n., T e, - n.I
"
t ..n + 1
~
and series (1)is absolutely convergent. Hence if N = 1,2,. oo,
(8)
I
f(s) - I T
o
e-stF(t)dt
I
~ Ni:.l1a~1
n=O n. iooe-sIr' dt
T
+ RN(T)
where, in view of condition (7),
I: la~1
RN(T) = n=N n. T n=NS' i
ooe-str' dt ~ I: la:'l'
Since series (1) is absolutely convergent, it now follows that to each ( > O)
there corresponds a number N.. independent of T, such that
RN(T) < ! wheneverN > N..
Now consider the sum from n = O to N - 1 in condition (8). Let N
be some fixed integer greater than N. By successive integrations by parts
o
we find that
1
-
n! Tf co
e-stl"dt = -
e-ST
- + +
Tn
s [ n! (n - l)!s (n - 2)!S2
Tn-l
+ ... + - .
snJ
Tn-2 1
Since Iani < Ms"+l, it follows that, if sT > 1 and n ~ N - 1,
n! T f
lanl coe-Sltn dt < Me-ST
[
(ST)n
n!
+ (sT)n-l +... + 11
(n - 1)! J
~ Me-ST(ST)n ~ + (n - 1 1)! + ... + 1 < Me-sT(sT)NN.

[n! J
Thus the sum in condition (8) is less than MN2e-sT(sT)N, which vanishes
as T -+ co for that fixed N. Therefore, there is a number T. such that
MN2e-sT(sTt < f/2 when T> T.. and condition (6) follows to complete
the proof of the theorem.
Theorem 7 can be generalized to the following theorem:
Theorem 8 Let g(s) be represented by an absolutely convergent series when

s > a (a ~ O) of this type:
co 1
(9) (O< k ~ 1,s > a ~ O).
g(s) L
= n=O an-l sn+k
Then L{ G(t)} = g(s) when s > a, where the function G is continuous

when t > O, is l!1(ea,,)if al > a, and is represented by the convergent
series of inverse transforms:
co
a co tn+k-l
(10) (t > O).
G(t) = n~o L -1 { ;~: } = n~o an-l
A proof can be based on Theorem 7 by writing
a_l 1 ~ an-l a_l 1 ~ an
gs( ) =-+- 1.. -=-+- 1..-
~ ~n=l sn ~ ~n=Os"+l'
The last series here represents the transform of the function
CO tn
(11) F(t) = L an"
n=O n.
a function that is continuous (t ~ O), and l!1(ea,,)if al > a. AIso we know
that (Sec. 5)
(k > O,s > O).

Since k ;; 1, the function rc-1 is bounded when t ~ T> o. Then

..k-l
-l a-l..k-l tU
{ } (O< k ;; 1).
L g(s) = r(k) t + r(k) * F(t)
By writing the convolution here as an integral from t = O to t = plus an
integral from to t, we can modify the proof in Seco 17 to show that
rc-l * F(t), and hence L -1{g(S)}, is 19(e"")if (Xl> (X.
In view of Eq. (11) we can write
(12)
rc-1 * F(t) = E (t - t)k-l F(t) dt
'
L~
)
= n=On.
a
Io (t - tt-1tndt
) a
= L ...!!...tk-l* tn,
n=On!
where the term-by-term integration is valid because the power series for
F(t) converges uniformly with respect to t over the interval O < t < t
(Prob. 15, Seco21). By using known transforms, we find easily that
..k-l * tn =
tU
r(k)n! ("+k
r(n + 1 + k)
and therefore
rc-l
- )
r(k) * F(t) L an-
= n=O tn+k .-'
It now follows that L -l{g} is the function G represented by series (lO).

That convergent series is the product of rc-l, or t-(1-k), by a power series
in t. Hence G is continuous whenever t > O,and Theorem 8 is proved. "
Example To establish transformation 78, Appendix A, Table A.2, when

k = 1 there, we write the absolutely convergent representation
g(s) = s-te-l/s = n=O

~ (-l)n
n!
~s"+t (s > O).
According to Theorem 8 and the form 6, Appendix A, Table A.2, of

the elementary transformation L -l{s-n-t}, then, L{G} = g where
1
G(t) = - L) -(-l)n 2n+ l("+t
~ n=O n! (1)(3)(5)... (2n + Ir

The terms of that series reduce to (- 1f(2Jt)2" + 1j(2n + 1)! and

therefore
e-l/S 1
G(t)= L- 1 { sJS} = .fi sin(2Jt).
21 DlFFERENTIALEQUATIONS WITH VARIABLE COEFFICIENTS

We have seen that
d"
L{t"Y(t)} = (-1)" ds"L{Y(t)} = (-l)"y(")(s),
and therefore we can write the transform of the product of t" and any
derivative of Y(t) in terms of y(s); for instance,
d2
L{t2Y'(t)} = ds2[sy(s) - Y(O)] = sy"(s)+ 2y'(s),
d
L{tY"(t)} = - dS[S2y(S)- sY(O)- Y'(O)]
= -s2y'(S) - 2sy(s)+ Y(O).
A linear differential equation in Y(t) whose coefficients are polynomials in t
transforms into a linear differential equation in y(s) whose coefficients are
polynomials in s. In case the transformed equation is simpler than the
original, the transformation may enable us to find the solution of the
original equation.
If the coefficients are polynomials of the first degree, the transformed
equation is a linear equation of the first order, whose solution can be
written in terms ofan integral. To find the solution ofthe original equation,
however, the inverse transform of the solution of the new equation must be
obtained.
Example 1 Find the solution of the problem

Y"(t) + tY'(t) - Y(t) = O, Y(O) = O, Y'(O)= 1.
The transformed equation is
2 d
s y(s) - 1 - ds[SY(s)] - y(s) = O,
or
y'(s)+ (~ - s)y(S)= -~,
which is a linear equation of the first order. An integrating factor is
exp[f (~- s) dSJ = exp (2logs - !S2) = s2e-ts2,

so the equation can be written
Integrating, we have
l e
y(s)= -S2 + -ets2
S2 '
where e is a constant of integration. But e must vanish if y(s)isa

transform since y(s)must vanish as s tends to infinity. It followsthat
Y(t) = t,
and this is readily verified as the solution.
Example 2 Solve Bessel's equation with index zero,

t Y"(t) + Y'(t) + t Y(t) = O
under the conditions that Y(O) = 1 and Y(t) and its derivatives have
transforms.
The point t = Ois a singular point of this differentialequation
such that one of the solutions is a function that behaves like log t near
that singular point, and the Laplace transform of the derivative of
the function does not exist.
The transformed equation is
d 2 , d
ds y(s)- s - Y (O)]+ sy(s) - l - -y(s)
--[s ds = O,
or (S2 + I)y'(s) + sy(s) = O.
Separating variables, we have

dy s ds
y =- S2 + l'
and upon integrating and simplifying, we find that
e
(2)
y(s) = p-:::l'
where e is a constant of integration.
FURTHER PROPERTlES OF THE TRANSFORMATION [SEC. 21 63
Expanding the functi<;mfor y(s) by the binomial series,we have,

when s > 1,
=~ ~ -t = ~ I - ~~ - ~ -~ ~ - ...
y(s)
. S
I+
( S2 ) S [ 2 S2 2 2 2 !S4 ( ) J
- e - I) -
[ (- 1)
- -S 1 + n=1
ex> n(I)(3)...
2nn.s,
1 2n
(2n
. J - e n=O
ex> (-I)n(2n)!
L (2nn.1)2?H'"
s "
where O! = 1. The ratio test shows that series in positive powers
of l/s are absolutely convergent when s > 1. Theorem 7 therefore
applies to show that the operator L - 1 can be applied term by term
to that series to represent the continuous function whose transform is
y(s) as a convergent series in powers of t:
(3) L -1 { ()} e ~ (-I)n 2n

y S = n~O (2nn!)2t .
It is not difficultto verify that for all t the power series (3) is a
solution of Bessel's equation (1). Ir that function is to satisfy the
condition Y(O)= 1, then e = 1,and the solution(3)can be written
(4) Y(t) = J o(t)
where J o is Bessel' s function of the first kind with index zero:
ex>(-1)n t 2n t2 t4
(5) Jt= -...
o() n~o (n!)2 ( 2 ) - 1 22 +-~ ..~. .
We have shown above that, when s > 1,
1
(6)
L{Jo(t)}= J S2 + 1'
a result that is actually valid whenever s > O, because J o can be
represented by an integraP that shows that IJo(t)1~ 1.
The differential equation

(7) t2 Y"(t) + t Y'(t) + (t2 - n2) Y(t) = O
is Bessel's equation with index n. It can be verified that the function
ex> (-It t n+2k
(n = 0,1,2,...),
(8) Jn(t) = k=O kl(. n + k)I.( _2) L
known as Bessel's function of the first kind with index n, is a solution of that
equation. In the problems to follow we shall establish the transforms of
J1(t) and tnJn(t).
1 ChurchilI, R. V.. "Fourier Series and Boundary Value Problems," 2d ed.. p. 175, 1963.
PROBlEMS
1. Use the transformation (6) and Theorem 2 to find L{J o(at)} listed in Appendix A,
Table A.2 (transform 55).
2. If Y, Y', and Y" are of exponential order and Y, Y' are continuous (t ~ O)while
Y" is sectionaJly continuous, and if L{ Y(t)} = y(s), show that
L{t2Y"(t)} = s2y"(S)+ 4sy'(s) + 2y(s).
Solve the following differentiaJ equations for Y if Y and its derivatives are to
have transforms.
3. Y"(t)+ atY'(t) - 2aY(t) = 1, Y(O)= Y'(O)= O,a > O. Ans. Y(t) = t2/2.
4. tY"(t) + (t - l)Y'(t) + Y(t) = O,Y(O)= O. Ans. Y(t) = Ct2e-'.
5. tY"(t) + (2t + 3)Y'(t) + (t + 3)Y(t) = 3e-'. Ans. Y(t) = (C + t)e-'.
6. t2Y"(t) - 2Y(t) = 2t, Y(2) = 2. Ans. Y(t) = t2 - t.
7. When k is a constant, show that the equation
t2Z"(t) + 2tZ'(t) + kZ(t) = O
leads to the same differentiaJ equation in the transform z(s).
8. With the aid of transformation (6) show that
s
L{J(t)} = ~- 1.
S2 + l
Use formula (8) to show that J(t) = -J l(t), and hence that
L{J(t)} = p+t- s.
p+t
9. Use the transforms of Jo and J (Prob. 8) to show that
(a)
io -
' Jo(-r)Jo(t- -r)d-r= sint;
(b) JO(-r)Jl(t - "C)d"C= Jo(t) - costo

10. Expand the function S-l exp(-s-l) in powers of S-1 and apply Theorem 7
to verify that
L -1 He-l/.} = Jo(~,
and thus obtain the transformation 75, Appendix A, Table A.2.
Apply Theorem 8 to verify the following transformations:
1.
11. L-l{)se-l/.} = ~CoS(~.
(Cf. transformation 76, Appendix A, Table A.2.)
12. L -1 J e1/. =~ cosh (~.

%
{ Js } .fii
i
(Cf. transformation 77, Appendix A, Table A.2.)
-1 1 f'J .(t)
13 )
. L { (S2 + 1).+t} = (1)(3)(5)...(2n - 1) (n = 1,2,... .
(Cf.transformation57,AppendixA, TableA.2.)
1 1""t"k
14. L -1 { 1- 1} = t);:l r(nk) (k> O,s > 1).
15. Given that a power series L':=oA.-r. converges uniformly over an interval
O ~ -r ~ t to a sum F(-r),that is, the series converges to F(-r)there and
N-l
F(-r) L A.-r. +
= "=0 R~-r) (O ~ -r~ t)
where RN(-r) Ouniformly with respect to -r as N oo. Ir O~ e < 1, prove that

""
t-C* F(t) = A.ec * t", L
11=0
a result that was used to obtain Eq. (12), Seco20. Suggestion: Write
N-l
t-C * F(t) - L A.t-C * t. = t-C * RN(t)
n=O
and prove that the right-hand member vanishes as N 00, when t and e are kept
fixed.
22 INTEGRATIONOF TRANSFORMS
When a function F(t) is sectionally continuous and of the order of 't, then
its Laplace integral
f(x) = La>e-xtF(t) dt
is uniformly convergent with respect to x in every interval x;;; OCI,where
OCl> oc,according to Theorem 5. It follows from formula (5), Seco 15, that
when r > s > oc,
f f(x) dx = f La>e-xtF(t) dt dx = La>F(t) f e-xt dx dt.

If the function F is such that F(t)jt has a limit as t tends to zero, then the
latter function is also sectionally continuous and of exponential order.
Under those conditions the last equation can be written
.
f' f(x) dx = Jo F(t)e-st
s t
dt -
fa> fa>~-,t
Jo t
dt = g(s) - g(r),
where g(s) = L{F(t)jt}. But g(r) ~ Oas r ~ 00 (Theorem 5); hence
(1)
fs f(x) dx = Jofa>F(t)e-st
a>
t
dt (s > oc)
and we have established the following theorem:
Theorem 9 Division of the function F(t) by t corresponds to integration of

the tran~formf(s), in this manner:
(2)
Sufficient conditions for the validity of formula (2) are that F(t) be
sectionally continuous and of the order of eaJ,that s > exin formula (2),
and further that the limil of F(t)/t exists as t -. + O.
The function F(t) = sin kt, for example, satisfies the above conditions
when ex = O;in particular,t-l sinkt -. k as t -. O. Hencewhens > O,
sin kt k dx 1t S k
= arctan-.
""
(3) L{ - t } = fs x2 + k2 = -2 - arctan-
k s
Recalling how integration with respect to t corresponds to division
by s, we can now write the transform of the sine-integral function
'Sint
(4)
.
SI t = -dt.
o t i
This function is of some importance in applied mathematics. Its values are
tabulated in the more extensive mathematical tables. If k = l in Eq. (3),
it follows from Eq. (4) that
(5) L{Sit} = ~arccots

s = ~arctan~
s s (s > O).
As another illustration of Theorem 9 we note that

e-al_e-bl ""
1 1 x+a ""
L
{ t }
= f s
---
( x+a x+b )
dx=log-
x+bs ]
,
when s > -a and s> -b. Hence
e-al - e-bt S+ b
(6) L = log-.s + a
{ t }
When a = O and b = 1, we have the special case
(7) (s > O).
23 PERIODIC FUNCTIONS
Let a function F be periodic with period a over the half line t > O and
sectionally continuous over a period O < t < a. That periodicity can be
described by either of the two statements

F(t + a) = F(t) when t > O, or F(t) = F(t - a) when t > a.
The function is bounded over the half line and sectionally continuous over
each bounded subinterval, so it has a transformf(s) when s > O.
For convenience in examining the transform, we write
F(t) = O when t < O
and introduce a function Fo that is the same as F when O < t < a and zero
elsewhere; thus
Fo(t) = [1 - 80(t - a)]F(t),
Then F is described for all t by the difference equation

(1) F(t) - F(t - a) = Fo(t),
as we can see either directIy from graphs of F(t) and F(t - a) or else by
noting that when t > a, the equation becomes F(t) = F(t - a), and when
t < a, it becomes F(t) = Fo(t). Therefore
f(s) - e-asf(s) = fo(s) and f(s) = I fo(s)

- e-aso
Theorem 10 1f F(t) is periodic with period a when t > O, and sectionally

continuous, then
(2) (s > O).
Let us applythat formulato the function

M(c,t)= I when O < t < e,
= -1 when e < t < 2c,
M(c, t + 2c) = M(c,t).
This is sometimes called the square-wave function (Fig. 9).
Sin~
Mle,t)
(0,1) I 1
I I
I I
.I 1
o (2e,0) I
I I
I I
I L-
Fig.9
the transform of M(e,t) is

(1 - e-CS)2 1 - e-CS
s(1 - e-2C")= s(1 + e-c"}'
Hence
1 es
(3) (s > O).
L{M(e,t)} = stanh"2
The integral ofthe function M from Oto t is the function H(e,t) defined
as follows:
H(e,t) = t when O< t < e,

=2e-t when e < t < 2e,
H(e, t + 2e) = H(e,t).
This function, whose graph is the triangular wave shown in Fig. lO, has the
transform
1 es
(4) (s > O).
L{ H(e,t)} = S2tanh"2
Let G denote an antiperiodic function when t > O:

G(t + e)= -G(t) when t > O, or G(t) = -G(t - e) when t > e,
which is sectionally continuous over the interval O< t < c. Then
Hle,t)
(O,e)
Fig.10
jf
i
~
G(t + 2e)= - G(t + e) = G(t); that is, G is periodic with period 2e. M(e,t)
is an example of sueh a funetion, as is sin t when e = n, 3n, Sn, . .. ,
Ir Go(t) = G(t) when O < t < e and Go vanishes for aIl t outside that
interval, and if we write G(t) = Owhen t < O,we can see that G is deseribed
by the difference equation
(5) G(t) + G(t - e) = Go(t).
The transform g(s) therefore satisfies the equation
g(s) + e-esg(s) = go(s) = I: e-s'G(t) dt,

and we have established the foIlowing theorem:
Theorem 11 If G is seetionally eontinuous over an interval (O,e) and

G(t + e) = - G(t) when t > O, then
(6)
f~ e-s'G(t)dt
g(s) =- (s > O).
For the periodie function G1 with period 2e sueh that

G1(t) = G(t) when O < t < e, G1(t) = O when e < t < 2e,
where G is the above antiperiodic funetion, formula (2) beeomes
(7) L{ G1(t)} g e-s'G(t)dt- g(s) (s > O).

l - e-2es - I - e-es
For example, when G(t) = sin t and e = n, g(s) = (S2 + 1)-1 and the fune-
tion G1 shown in Fig. II has the transform
L {G (t)} = L sin t + sinti = I

(8) (s > O).
1 { 2 } (S2 + 1)(1 - e-'IS)
SimilarIy, the periodic funetion G2 sueh that
Git) = G(t) when O < t < e, G2(t + e) = G2(t)
where G(t + e) = - G(t),
o (271',0)
Fig.11
has the transform
therefore
(9) (s > O).
When G(t) = sin t, for example, then G2(t) = Isin ti if c = 1t and
(10) (s > O).

I L{lsin ti} = S2 ~ 1 coth ~
!
In case the antiperiodic function G has nonnegative values over the
interval (O,c),then G(t) ;;;Owhen c < t < 2c and the function G1 described
above is the half-wave rectificaton of G. It replaces the negative values of G
by zero. Also, in that case G2 is the full-wave rectification of G, and we can
write
G1(t) = HG(t) + IG(t)lJ, Git) = lG(t)1.
24 PARTIAL FRACTIONS
We shall now systematize the procedure of finding inverse transforms of

quotents of polynomals in s. Let p(s) and q(s) denote polynomials in the
variable s with no factor in common, and let the degree of p(s) be lower
than that of q(s). We shall see that the inverse transform of the function
f(s) = p(s)/q(s)exists and that it can be found when the elementary factors
of q(s) can be determined.
Consider first the case in which q(s) has a linear factor s - a, not
repeated. Let tP(s)denote the function that is left after removing that factor
from the denominator of f(s); that is,
(1) f(s) = p(s) = cP(s) .

q(s) s - a
Note that cP(s)may be a quotient of polynomials. According to the theory

. of partial fractions, a constant e exists such that
(2) cP(s) = ~ + h(s),

s-a s-a
where h(s) represents the sum of the partial fractions that correspond to the
other linear and quadratic factors of q(s), any of which may be repeated.
-
In order to determine the value of e, we multiply both members of

Eq. (2) by s - a, when s ::j: a, to obtain the equation
4>(s)= e + (s - a)h(s),
which is satisfied identicaIly for aIl values of s in a neighborhood of the
point s = a,exceptpossiblyat that point. But both membersofthe equation
are continuous functions of s at that point; thus their limits as s ~ a are
the same as their valueswhen s = a. Thereforee = 4>(a).The inverse
transform of the partial fraction corresponding to the factor s - a, or the
term in F(t) corresponding to that factor, is 4>(a)eat.
In view of Eq. (1), we can also write
S - a 1
lim 4>(s)= lim p(s)- = p(a)-,- '
s-a s-a [ q(S) ] q (a)
where we have evaluated the limit of (s - a)/q(s) as the limit of the quotient
of the derivatives of s - a and q(s), since q(a) = O and q'(a) ::j:O because
s = a is a simplezero of q(s). ConsequentIy 4>(a)= p(a)/q'(a).
When aIl the factors of q(s) are linear and not repeated and when
q(s) is written in the form
(3)
where all the constants an are distinct, we can write the inverse transform of
f(s) in fuIl. Let qn(s) denote the product of aIl the factors on the right of
Eq. (3) except the factor s - an, so that the function 4>(s)corresponding to
that factor is p(s)/qn(s). Then
(4) L- 1 P(S) = I
p(an) ~nt
{ q(s) } n= 1 qn(an) = I p(an) eant
n= 1 q'(an) .
The second sum here is sometimes caIled Heaviside's expansion.
The principal results are stated in the foIlowing theorem.
Theorem 12 lf f is the quotient p(s)/q(s) of two polynomials suchthat q(s)

has the higher degree and contains thefactor s - a which is not repeated,
then the term in F(t) corresponding to that factor can be written in either
of these two forms :
,1,. ""t p(a) at
(5) '
( )
'f' a t:: or q'(a{ ,
where 4>(s)is the quotient of p(s) divided by the product of all factors
of q(s) except s-a.
Theorem 12 is valid when the constant a is any complex number.
For complex arguments the exponential function is defined by the equation

(Sec. 55)
+iJl = (cos y + i siny) (x and y real).
Example Find F(t) when

152 - 4s
f(s) = (2s + l)(s2 + 1)"
We display the factors of the type s - a by writing
S2 - 15
f(s) = (s + t)(s - i)(s + i)'
Using the first of the two forms (5), or the first of the expansions (4),
we find that
( ) -.i -t/2 - l + 2i it -1 + 2i -it

F t - 4~e ("1"2
+ 1)2'e
1 + ( -1"2. + 1)( - 2")e
1
eit - e-it
= e-t/2 - 2
2i
= e-t/2 - 2 sin t.
PROBLEMS
1. Show that Theorem 9 applies and find the transfonn of: (a) (1 - cos at)/t;
(b) (1 - cosh at)/t; (e) (e' - cos t)/t.
Ans. (a) (b) See transfonns 105 and 106,Appendix A. Table A.2;
t
(e) log [(s2 + 1)/(s - 1)2](s > 1).
2. The condition that F(t)/t has a limit as t -+ +0 was used in proving Theorem 9.
Ir it is replaced by the condition that for some positive constants k and M, IF(t)1< Mr<
over an interval O < t < T, then g(s), the transfonn of F(t)/t, exists and vanishes as
s -+ oo. Thus fonnula (2) in the theorem is still valido Use this fact to establish trans-
fonn 36, Appendix A, Table A.2:
el" ti"
L ---=--- =~-~ (s > a and s > b),
{ zF }
by showingthat 11ft- i!"1/0 < M00n an interval (O,T).
3. (a) Use Theorem 11 to derive the transfonn (3), Seco23, of M(c,t).
(b) Apply fonnula (7), Sec. 23, to obtain the transfonn 68, Appendix A, Table A.2,
of the half-wave rectification of M(k,t).
j I (e) Use fonnula (9), Seco23, to verify that the transfonn of the full-wave rectifica-
tion of M(c,t) is l/s.
4. Sketch the graph of the periodic function F for which F(t) = t when -1 < t < 1
and F(t + 2) = F(t) for all t, and show that the transfonn of F is
l l
f(s)=--~ (s > O).
i S2 s smh s
I
,
5. Let F be the periodic function such that F(t) = t when 0< t < 1 and F(t + 1)= F(t),
and let G be this antiperiodic function: G(t) = t whenO< t O),
(a) f(s) = SI - s(1 - e-
I 2 I ~ (s > O),
(b) g(s) = SI - (SI+ s) I + e-s
f(s)
(e) gl(S) = I + e-s (s > O).
6. Use Theorem 12 to find the inverse transforms tabulated below, where a, b, and e
are distinct constants.
f(s) F(t)
I -e"'-e'"
(a) (s - a)(s - b) a-b
s ae'" - b"
(b) (s - a)(s a-b
-1 (b - c)e'" + (e - a)e'"+ (a - b)e<'
(e)
(a - b)(b - c)(c - a)
7. Use Theorem 12 to find the inverse transforms of the following functions.
(a) f(s) = . .
4s + I
. ... .. . Ans. F(t) = e'/z- e-'/z + e-' - l.
s s
(b) f(s) = SZ- kZ' (e) f(s) = SZ + kZ'
3sz
(el)f(s) =.. .. .
Use Theorem 12in solving the following differential equations.

8. Y"(t) - Y(t) = I + e3'. Ans. Y(t) = C1e'+ Cze-' - I + te3'.
9. Y"'(t) + Y"(t) - 4Y'(t) - 4Y(t) = F(t), if Y(O)= Y"(O)= Oand Y'(O)= 2.
Ans. Y(t) = sinh 2t + lzF(t) * (eZ'+ 3e-z,- 4e-').
10. y<4'(t)- 2Y"'(t) - Y"(t) + 2Y'(t) = 6F(t), if Y{t) and its first three derivatives
are zero when t = O.
25 REPEATED LINEAR FACTORS
We now consider partial fractions for the case in which the polynomial
q(s) contains a repeated linear factor (s - a)n+ 1. We write
(1) f(s) = p(s) - cp(s)

q(s)- (s - at+1'
where cp(S)is the quotient of polynomials obtained by removing the factor

-
(s af+ 1 from the denominator of the fraction p(s)/q(s). As before, the
degree of the polynomial p(s) is assumed to be lower than the degree of q(s).
Note that cp(s)and its derivatives are continuous functions at the point
s = a.
The representation of f(s) in partial fractions now has the form
cp(s) Ao AlAr An
(2) n+l =-+
(s-a ) s-a (s-a )2 +... + (s-a )'+1 + ... + (s-a )n+l +h(s),
where the numbers Ar are independent of s, and h(s) is the sum of the partial
fractions corresponding to the remaining factors of q(s). It follows from
Eq. (2) that
(3) cp(s)= Ao(s - a)n + ... + Ar(s - af-r + ... + An + (s - a)n+lh(s)
in a deleted neighborhood of the point s = a. When s -. a, we see that

An = cp(a).
To find the remaining coefficients A" we differentiate both members
of Eq. (3) with respect to s, n - r times, in order to isolate the number Ar.
When s -. a in the resulting equation, we find that
cp(n-r)(a) = (n - r)!Ar.
Equation (2) can now be written in the form
n cp(n-r)(a) l
(4)
f(s) L (n
= r=O - r.)r (s - ay+ 1 + h(s),
where O! = l and cp(O)(a) = cp(a).
Ir H(t) = L -1{h(s)}, it follows that the inverse transform off(s) is
n cp(n-r)(a)
(5) F(t) L
= r=O 'u uH,t'>t+H(t).
This equation can be simplified by recalling the formula for the derivative
of order n of the product of two functions u(s) and v(s), namely,
dn n ,
-(uv) = L n. u(n-r)(s)v(r)(s).
ds" r=o(n - r)!r!
When u = cp(s)and v = tT',then arv/as' = t'tT' and Eq. (5) reduces to the
form
(6) F(t) = ~!{:n[CP(S)e"]} s=a + H(t).

This result can be stated as follows.
;,
Theorem 13 1f f(s) is the quotient p(s)/q(s) of two polynomials such that

q(s) has the higher degree and contains the factor (s - a)n+ 1, then the
term in F(t) corresponding to that factor is c1Iia,t),where
1 an
(7)
c1In(s,t)= n! a~[4>(s)e"]
and 4>(s)is the function indicated by Eq. (1).
The term in F(t) corresponding to a factor (s - a)2in q(s),for instance, is
(8) c1I1(a,t) = [4>'(a) + 4>(a)t]e"t,
and the term corresponding to a factor (s - a)3 is

(9) c1I2(a,t) = H4>"(a) + 24>'(a)t + 4>(a)t2]eat.
As an example, if
1
f(s) = (s - l)(s - 2)2'
then the term in F(t) correspondingto the factor s - 1 is . To correspond
with the factor (s - 2)2, we have 4>(s)= (s - 1)- 1 and 4>'(s)= - (s - 1)- 2
so that 4>(2)= 1 and 4>'(2)= - 1. In view of formula (8) the term in F(t)
is (-1 + t)e2t. Hence
F(t) = + (t - l)e2t.
Since the number a may be imaginary and since a factorization of
every polynomial into linear factors, real or imaginary, exists, Theorems
12and 13providea systematicway offinding inversetransformsof quotients
of polynomials in all cases where the factors of the denominator can be
determined. Ir, however,imaginaryfactors are present,the results are given
in terms of imaginary functions. The reduction of the results to real forms
is sometimes tedious. To obtain the real form of the inverse transform
directly, and to observe the character of that function in general, we may
proceed as follows.
26 aUADRATIC FACTORS
In this section we assume that the polynomials p(s) and q(s) have real
coefficients.The imaginaryzeros of q(s) then occur in pairs, each pair
consisting of some complex number a + ib and its conjugate. The corre-
sponding linear factors of q are
(1) ql = s - a - ib, q2 = s - a + ib (b ::/=O),
where a and b are real numbers. The product of those factors is the real
quadratic factor (s - a)2 + b2.
,~
Let q have the factors (1), not repeated,and let <Pland <Plbe the
quotients of polynomials obtained by removing the factors ql and ql in
turo from the fraction p(s)/q(s); thus
(2) f(s) = p(s) =

q(s)
<Pl(S)
S- a -
ib
= s -<Pl(S)
a + ib
.
According to Theorem 12 the sum of terms in F(t) corresponding to those
two linear factors is
(3)
This is the component of F corresponding to the quadratic factor
(s - a)l + b1. We shall represent it in terms of real-valued functions.
The rational function <Pl(S)is continuous at the point s = a + ib.
In a neighborhood of that point excluding the point itself
(4) <Pl(S)= (s - a - ib)f(s).
Similar1y, <Plis continuous at the point a - ib and
(5) <P2(S) = (s - a + ib)f(s)
throughout a neighborhood of that point with that point deleted. From
elementary properties of complex conjugates we can see that the conjugate
of f(s), a rational function with real coefficients, is f(S). Consequently
<Pis) = (s - a + ib)f(s) = (s - a - ib)f(s) = <Pl(S).
In view of the continuity of <Pland <Plit follows that

(6)
Since e-ibt = cOS bt - i sin bt, it is the conjugate of eibt. Thus formula
(3) can be written
G(t) = eQt[<pl(a
+ ib)eibt+ <Pl(a+ ib)eib'J.
The factor in brackets is twice the real part of the first term, 2 Re[<P
1(a+ ib)eib'J.
So if we let r 1 and 81 represent polar coordinates of the point representing
the complex number <Pl(a+ ib),which is never zero,
(7) <Pl(a+ ib)= rleiOI = rl(cos81 + isin(1),
we can write the component G in the real form
(8)
G(t) = 2eQtr1
cos (bt + (1) (b =FO,rl > O).
Ir a = O, the component G(t) is the simple periodic function
2rl cos (bt + 8), and if a < O, it is a damped periodic function. Those
components, corresponding to cases a ~ O,represent stable oscillationsin
the theory of control of mechanical and electrical systems where t denotes

time. Stable oscillations are bounded as t -+ oo. When a> O, the com-
ponent G(t) represents an unstable oscillation.
To illustrate the use of formula (8), we find F(t) when
2s - 2 2s - 2 <Pl(S)
f(s) = (s + l)(s2 + 2s + 5) = (s + I)[(s =

2s - 2
where <Pl(S)= (s + I)(s + 1 + 2i)"
Th A..
( 1 2)
' -4 + 4i 1 - i .J2 -i,,/4
en '1'1 - + I = 2i(4i) = ~ = le
and G(t)= .J2e-t cos (2t - nJ4)= e-t(cos 2t + sin 2t). After adding the
term corresponding to the linear factor s + 1,we find that
F(t) = e-t(cos 2t + sin 2t - 1).
In case q contains the square of the quadratic factor (s - a)2 + b2,
we write
(9) f(s) = p(s) = <Pl(S) = <P2(S)

q(s) (s - a - ib)2 (s - a + ib)2
and apply formula (8), Seco25, to get the corresponding component of F(t):
G(t) = e"t[<p'1(a+ ib) + t<Pl(a + ib)]eibt
+ eat[<p(a - ib) + t<P2(a - ib)]e-ibt.
But <P2(S)= (s - a + ib)2f(s), and as before we find that <P2(S)

= <Pl(S),also
that <p(s)= <P'1(S),
so that
(lO) G(t) = 2e"tRe [<p(a+ ib)eibt+ t<P1(a+ ib)eibt].
We use absolute values and arguments of the complex numbers
<P1(a + ib) and <p'1(a+ ib) displayed by the polar forms
(11)
where rl #=O because <P1(a+ ib) #=O. Then formula (10) becomes
(12) G(t) = 2e"t[Plcos(bt + ifJl)+ rltcos(bt + 81)] (rl > O).
Qur results can be summarized as follows.
Theorem 14 Whenf(s) = p(s)Jq(s)where p(s) and q(s) are polynomials with

real coefficients and q has the higher degree and contains a nonrepeated
-
real quadratic factor (s a)2 + b2, where b #=O, the component of
F(t) corresponding to thatfactor is thefunction G(t) given by formula (8).
If q contains the square of that quadratic factor, the component of F(t)

is given by formula (12).
We note that the component (12) represents a stable oscillation only
if the real part of the imaginary zeros a :t ib of q is negative: a < O. If a = O,
the component contains the term 2r1 t cos (bt + ()1) representing an unstable
oscilIation.
When the quadratic factor appears to a degree n + 1(n = 2,3, . . . ),we
can see from the procedure we used when n = 1 and from formula (5), Seco25,
that the component in F will contain a term of the type Cea'tncos (bt + ()1),
where C is a constant.
When p and q are not polynomials in s, partial fractions in s cannot
be used. Such cases will be treated in Chap. 6 with the aid of residues and
contour integrals.
27 TABLES OF OPERATIONS AND TRANSFORMS

Appendix A, Table A.1, contains a list of operations on F(t) with corre-
sponding operations on f(s). That table of operations summarizes several
of our resu1ts on the theory of the Laplace transformation.
The table of Laplace transforms (Appendix A, Table A.2) gives a
fairly extensive list of transforms of particular functions. Derivations of a
number of them have been presented above. References to some more
extensive tables can be found in the Bibliography.
Transforms 82 to 84, Appendix A, Table A.2, of functions that are
prominent in problems of diffusion and conduction of heat, will now be
derived. A more direct derivation will be made later on in Chap. 6 with the
aid of contour integral s and functions of a complex variable; but we wish
to use those transforms earlier, in Chap. 4.
First we perform formal manipulations that will suggest the inverse
transforms of the two functions
(1) y(s) = ~ e-k./S (k ;?; O,s > O),

Js
(2) z(s) = e-k,(S (k > O,s > O).
We see that (.Jiy)' = -kz/(2.Ji) and 2z' = -ky, so that y and z satisfythis
system of differentialequations with coefficientsthat are linear in s:
(3) 15y'(s) + y(s) + kz(s) = 2z'(s)+ ky(s) = O.
The corresponding system Jor the inverse transforms Y(t) and Z(t) is
therefore
2(- t Y)' + Y + kZ = O, - 2tZ + k Y = O.

Thus -2tY' - Y + kZ = Oand

k2 1 . k
(4) 2 Y'(t) = ( 2t2 - t ) Y(t), Z(t) = 2t Y(t).
The solution of system (4) is

C k2 Ck k2
Y(t) = -fi exp ( - 4t ) ' Z(t) = 2.jt3 exp ( - 4t ) .

But when k = O, then y(s) = 1/j""i, and we know that Y(t) is then 11Ft,
so if C is independent of k, it follows that C = 1/.j;., and our formal results
can be written
(5) L ~exp - k2 = ~e-kJS (k ~ O),

{ Ft ( 4t)} .Js
(6) L ~exp - k2 = e-kJS (k > O).

{ 2.jit3 ( 4t )}
Let us now prove that transformations (5) and (6) are correct when
k = 1 and s > O. The function
(7) Y1(t)= ~exp -~ when t > O,Y(O) = O,
Ft ( 4t)
is continuous when t ~ O,and bounded. Hence when s > O,
eo eo 2
1 dt 1 dt
.j;.Yl(S) f ( 4t) -fi = e-../S o exp - (JSi - 20 ) J -fi'
= o e-S1exp- f [
We substitute1: for 2-fi to write
(8) fie../SY1(S) = Leo exp [- (f1: - ~rJ d1:,

and make a further substitution tJ;r = 1/).to see that
2
2 eo .JS
= .fi f
1 d)'
(9) o exp
.j;.e../SY1(s) [- (2).
- ) J ).2'
By first adding corresponding members of equations (8) and (9), then
substitutingx fort0). -
1/)" we can write
2 .fieo 2
.fi
.
2.j;.e-Isy f[ (
(s) = .fi o exp - 2).
1
1
- J (2 +
)
1
).2 ) d)'
~
eo
2
=v r: fs -eo
exp (_X2) dx = 2 -.
S
.--
Therefore
Yl(S) = L{J;r exp( - ~t)} = j-se-,s (S> O),

and when k> O,transformation (5) follows by writing Yl(k2s) and applying
Theorem 2. Formula (5) was established earlier when k = O.
Ir k > O, the funetion Z, defined as follows,
k k2
Z(t) = -exp -- when t > O,Z(O) = O,
'2~ ( 4t)
is continuous and bounded when t ~ O; also Z = tkt - 1 Y where
Y(t) = ~exp - k2 and (S > O).
.jid ( 4t )
Theorem 9 therefore applies to give the transformation
k dx
f
a:>
(s > O),
z(s) ="2 s e-k./X ~ = e-k./S
which is transformation (6).
Finally, in view of transformation (6), we note that
L -l l -k-.s = - k
-e
{s } 2.;:;c o I
' e -k2/( 4f) . -i..- d. = - 2
fi f a:>
e -J.2 dJI.
k/(2../i)
1
Therefore
(10) (k ~ O,s > O),
where erf(x) is the error function defined in Seco 16. Equation (10) can be
written
(11) (k ~ O,s > O),
where the complementary error function erfe (x) is defined as

a:>
2
(12)
erfc (x) = 1- erf(x) = fi fx e-J.2dA.
o,
PROBLEMS
1. Find the inversetransformstabulated below:
f(s) F(t)
s+a a -b -b' a- e
a - b -c'
(a) (s + b)(s + e)2 (b- e)2e + [ b - et- (b- e)2e
J
s+3
(b) (2s + l)(s2 + 2s + 2)
2b4
(e) (S2+ b2)
s2-b2
t cos bt
(d) (S2 + b2)2
2Ss2
(St - 8)e-' + 6 sin 2t + 8cos 2t
(e) (s + 1)2(S2+ 4)
2. Without finding F(t), determine whether the oscillations of that function are stable
or unstable when its transform is
(a) f(s) = - .152_+.1 Ans. Stable.
s
(b) f(s) = .- . -~~. Ans. Unstable.
3. Solve the following differential equations:

(a) Y"(t) - 2Y'(t) + Y(t) = 1. Ans. Y(t) = (CI + C2t)e'+ 1.
(b) Y"(t)+ Y(t) = 2 sint, if Y(O)= Oand Y'(O)= -1. Ans. Y(t) = - t cos t.
.
(e) 4Y"'(t) + 4Y"(t) + Y'(t) = F(t).

(d)y(4)(t) + 2Y"(t) + Y(t) = O, if Y(O)= O, Y'(O) = 1, Y"(O)= 2, and
Y"'(O)",; -3. Ans. Y(t) = t(sin t + cos t).
4. In Seco26, ifwe write <PI(a+ ib) = (XI+ iPI, show that formula (8) for the com-
ponent G can be written
G(t) = 2e"'((XIcos bt - PI sin bt).
5. With the aid of transformation (S), Sec. 27, find L -1{s-texp(-kJS)}listed as

transformation 8S in Appendix A, Table A.2.
.
6. Solve for Y(t):
-
(a) (t t2)Y"(t)+ 2Y'(t) + 2Y(t) = 6t, Y(O) = Y(2) = O.
Ans. Y(t) = t2 - 8t-l(t - 1)380(t - 1).
. (b) Y(t) + 2 J~ Y(x) cos (t - x) dx = ge2'.
Ans. Y(t) = Se2'+ 4e-' - 6te-'.
7. Draw the graph of the ramp function Y such.that
Y(t) =t- Y(t - 1) and Y(t) =O when t ;;;! O,
and show that Y(t) = t - n when 2n;;;!t ;;;!2n + 1, Y(t) = n + 1 when 2n + I ;;;!
2n + 2, n = O, 1,2, .. . , and that

l
(s > O).
L{Y(t)} = s2(1+ e-')
8. When F, G, and H are sectionallycontinuous, show that
F(t) * G(t) * H(t) = E F(t - x) J: G(x - y)H(y) dy dx.

9. When F is sectionally continuous and G and its derivative G' are continuous over
an interval O;;:;t ;;:;T, show that, at each point t(0 < t < T) where F is continuous,
d
dt[F(t) * G(t)] = F(t) * G'(t) + G(O)F(t).
10. Under the following set of conditions the limit of the integral
(x > e),
~(x) = focof(x,t) dt
as x - 00, is the same as the integral ofthe limito (a) Letf(x,t) be continuous over each
rectangle e ;;:; x ;;:; e, O ;;:; t ;;:; T, except possibly for finite jumps across a finite number
=
oflinest ti in each rectangle,and (b)let the above integralconvergeuniformlywith
respect to x when x> c. Also, (e) letfhave a limit
F(t) = x~co
lim f(x,t)
uniformly with respect to t(t ,p ti) on each interval O < t < T; that is, for each positive E
there is a number N. independent of t such that
IF(t) - f(x,t)1 < E whenever x> N. (O < t < T),
and (d) let f~ F(t) dt exist. Then prove that
lim q(x) =
X-CIO lo
co F(t) dt.
Note: The difference of the last integral and q(x) can be written
I: [F(t) - f(x,t)] dt + f; F(t)dt - f; f(x,t)dt = tsq.

Take T large first, independent of x, then x large, to make Itsclsmall.
11. Use a remainder to prove that the integral
x
q(x) = l CO
~dt
o t + x
. is not uniformly convergent with respect to x on the half line x > l (cf. Probo 10).
Show that lim q(x) = rr/2 while the integral of the limit of the integrand, as x
is zero. x~co
00, -
12. If F(t) ~ G(t) when t > O,then f(s) ~ g(s)for every real value of s for which the
transformsf and g existo Prove that property, and iIIustrate it with some particular
functions. Note that the case F(t) = 2 sin t, G(t) = sin 2t shows that the converse of
the property is not always valido
FURTHER PROPERTIES OF THE TRANSFORMATlON [SEC. 27 83
13: Inilial-value theorem Ir F is continuous over the half line t ~ O except possibly
for a finite number of finitejumps, while F' is sectionally continuous over each bounded
interval there, and if F and F' are of exponential order, then the limit of the product
sf(s), as s - 00, exists and equals the initial value of F:
s~oo
lim sf(s) = F(O).
Use a formula for L{ F'} and Theorem 5 to prove that theorem.1

14. Illustrate the initial-value theorem (Prob. 13) with these particular functions:
(a) F(t) = A; (b) F(t) = So(t - to); (e) F(t) = t; (d) F(t) = cos kt; (e) F(t) = 11"[1 - So(t - to)].
15. Irf(s) = p(s)/q(s)where p and q are polynomials whose terms of highest degree
are a.s' and b.s"+1, respectively, use the initial-value theorem (Prob. 13) to prove that
F(O)= a.lb..
16. Let F and F' be continuouswhileF" is sectionallycontinuousovereach interval
O~ t ~ T, and let all three functions be of exponential order. Then if s2f(s) has a
limit as s - 00, prove that
lim s2f(s) = F'(O) and F(O)= o.
s~oo
17. Illustrate the modified initial-value theorem presented in Probo 16 in case F(t)
is (a) sin kt; (b) at + bt2; (e) 1 - cosh t.
18. Ajinal-valueIheorem Let both F and F' be bounded over the half line t ~ Oand
continuous there except possibly for a finite number of finite jumps; also let 1F'(t)1
be integrable from t = O to t = oo. Then
lim sf(s)
3-+0
= t-Q'J
lim F(t).
Prove that theorem. In the formula for L{F'} note that our conditions ensure the
continuity of that Laplace integral with respect to s when s ~ O (Sec. 15).
19. Illustrate the final value theorem (Prob. 18) in case (a) F(t) = A + Be-';
(b) F(t) = So(t - to); (e) F(t) = t - (t - I)So(t - 1).
20. Generalizedconvolulion For a function F(t;r) of two variables, defined over the
quadrant t > O,-r > O,letf(s,-r) denote its transform with respect to t. Using the same
parameter s, let](s,s) denote the iterated transform:
J(s,s) = Loo e-s,!(s,-r) d-r = Loo e-S<Loo e-stF(t,-r) dt d-r. .

Indicate formally why J(s,s) is the transform of the generalized convolution F * (t) of
F(t.-r) defined below:
/ J(s,s) = L{F * (t)} where

F * (t) = E F(t - -r,-r)d-r.
It is convenient to write F(t,-r)= Owhen either t < Oor -r < o. (See Probo 21 for condi-
tions ofvalidity ofthe generalized-convolution property.) When F(t,-r)= F(t)G(t),note
that the property reduces to the convolution property (6), Seco16.
21. Prove that the generalized-convolution property written in Probo 20 is valid
when s> IXunder these conditions on F: IF(t,-r)1< M exp[IX(t+ -r)]for all positive
I Similar theorems with conditions on I(s) rather than F(I) will be found in Chap. 6.
84 SECo 27] OPERATlONAL MATHEMATlCS
t and " and F is continuous by subregions (Seco14) over each rectangle, O ~ t ~ T,

O ~ . ~ R (cf. Seco 16).
22. When F(t,.) = 80(t - .) in Probo 20, show that J(s,s) = ts-2 when s > O, also
that F * (t) = tt, and thus verify the generalized-convolution p~rty in this caseo
23. When F(t,.) = J o(2Jh), we found (Probo 10,Seco21) thitf(s,~ = S-l exp (- ./s)o
Apply the generalized convolution property (Probo 20) to derive the integration formula
(t ~ 0)0
24. When. is fixed and positive, let Z(t,.) denote the solution of the differential
equation with constant coefficients (independent of t and -r)
m dn
Z(t,-r) + J1 andt"Z(t,.)= F(-r)

such that Z and dnZ/dtn(n = 1,2, ,m - 1)all vanishwhen t = O. Write the corre-
o o o
sponding equation for the iterated transform z(s,s) of Z(t,.) (Probo 20) and compare
it with the equation for the transform y(s) of Y(t), where
m
Y(t) + L any,n)(t) = F(t),
n= 1
Y(O) = Y'(O)= o o o = y'm- 1)(0)= O,
to show that y(S) = SZ(S,S).Thus deduce the following formula for the solution of the
problem in Y in terms of the solution of the simpler problem in Z:
Y(t) = io
'
-Z(t - " .)d..
oot
25. Use the formula in Probo 24 to find the solution of the problem
Y"(t) - Y(t) = F(t), Y(O)= Y'(O)= O,
in the form Y(t) = J~F(r)sinh(t - Note that - F(-r)is a particular solution of
-r)d-ro
the differential equation in Z(t,-r)o
3
Elementary Applications
The properties of the Laplaee transformation that we have derived up to

this point enable us to solve many problems in engineering and physies
involving ordinary linear and partial differential equations. In this ehapter
we shall solve a number of problems in elastie vibrations involving ordinary
differential equations. They are problems in whieh our method is very
eonvenient, although not essential. We shall also treat some simple appliea-
tions of integral equations.
The next ehapter eontains applieations that involve partial differential
equations. The solution of problems of this type is a primary objeetive of
this book. In later ehapters we shall extend our treatment of sueh problems.
28. FREE VIBRATIONS OF A MASS ON A SPRING

Let a body of mass m attaehed to the end of a eoil spring (Fig. 12)be given an
initial displaeement and an initial veloeity and allowed to vibrate. The other
end of the spring is assumed to be kept fixed, and the spring is assumed to
85
86 SECo 28] OPERATiONAL MATHEMATICS
..x Fig.12
obey Hooke's law, SOthat the force exerted by the free end is proportional
to the displacement of that end. The factor k of proportionality is called
the spring constant. We also assume that the mass of the spring can be
neglected in comparison with the mass m and that no frictional forces or
external forces act on m.
Let X denote the displacement of m from the position of equilibrium;
that is, let the origin O denote the position of m when the spring is not
deformed. Then according to Newton's second law of motion,
d2X
(1) (k > O).
m dt2 = kX -
Let Xo denote the initial displacement and Vothe initial velocity, so that the
function X(t) satisfies the conditions
(2) X(O) = Xo, X'(O) = Vo.
We can determine the function X(t) by applying the Laplace trans-

formation to both members of Eq. (1) and using the conditions (2). Thus if
x(s) denotes the transform o~X(t), it follows that
m[s2x(s) - sXo - vo] = -kx(s),
5 1
and therefore
X(5) = Xo 2 . n. L' + Vo 52 + (k/m)"
Hence
(3) X(t) = Xocos wot + ~ sin wot

Wo
= jxo2 + (~:r sin (wot + ex),
xowo. WoXo
where Wo = A
m
-, tan ex= -, Vo smex = y,Wo 2Xo 2 + Vo2'
. The motion described by formula (3) is a simple vibration with angular

frequency wo, called the natural frequency of this system, and phase angle ex,
and with the amplitude [xo2 + (VO/wo)2]t.
If a viscous damping force proportional to the velocity also acts upon
the mass m, as indicated by the presence of a dashpot c in Fig. 13,the equation
of motion becomes
(4) mX"(t) = -kX(t) - cX'(t),
.
ELEMENTARY APPLICATIONS [SEC. 28 87
Fig.13
where the eoefficient of damping e is a po sitive constant. Let the mas s start
from the origin with initial velocity Vo:
X(O) = O, X'(O) = Vo.
The equation in the transform x(s) becomes
ms2x(s) - mvo = - kx(s) - esx(s)
or, if we write 2b = e/m, and again write illo2 = k/m,
(5) x(s) =? -.Vo ? = . n1

Vo
1 - l'
Ir b2 < illo2, that is, if the coefficient of damping is smalI enough that
e2 < 4km,
then the formula for the displacement is
(6)
In the case of critical damping, that is, when illo = b, or

e2 = 4km,
it folIows from Eq. (5) that
(7)
We can see from this formula that the mass m moves in the direction of Vo
until the time t = l/b, then reversesits direction and approaches O as t tends
to infinity.
When e2 > 4km, a similar motion of the mass takes place. The dis-
cussion of this case and the case of other initial conditions is left to the
problems.
The mathematical problem treated in this section can be interpreted
also as a problem in electric circuits. This welI-known analogy between
problems in vibrations of mechanical systems and electric-circuit theory will
be observed for other problems in this chapter. NaturalIy, the notation and
terminology differ in the two types of problems.
In the electric circuit shown in Fig. 14,let Q be the charge accumulated
in the capacitor e at time t, and 1 the current in the circuit, so that
(8) I(t) = Q'(t).
./
Fig.14
This equation requires that I(t) be positive when Q(t) is increasing. Ir the
positive sense of ftow I(t) of positive charges is taken in the clockwise
direction in Fig. 14, then Q(t) measures the charge on the upper plate of the
capacitor.
The circuit has a resistance R and a coil of inductance L. Since the
sum of the three voltage drops in the circuit is zero in this case of no impressed
voltage, then
1
(9)
LI'(t) + CQ(t) + RI(t) = O.
The system of first-order differential equations (8) and (9) can be used
directly to determine I(t) and Q(t) in terms of initial values of those functions.
By eliminating I(t) from the two equations, however, we see that
1
(10) LQ"(t) + RQ'(t) + CQ(t) = O.
Except for the notation used, this equation is the same as Eq. (4). When the
resistance is negligible, R = O,the equation reduces to our Eq. (1).
The initial conditions in the electrical problem can be made the same
as those in the mechanical problem. For example, if the capacitor has an
initial charge Qo and if the initial current is 10, then
Q(O) = Qo, Q'(O) = 10,
which are the same as the initial conditions (2) in our first mechanical
problem.
29. FORCED VIBRATIONS WITHOUT DAMPING

Let an external force F(t) act upon the mass in the mechanical system of the
last section, assuming there is no damping (Fig. 15). The displacement X(t)
of the mass m then satisfies the differential equation
(1) mX"(t) = - kX(t) + F(t).
Ir the initial conditions are
(2) X(O) = Xo, X'(O)= Vo,
~
m
Ftt)
Fig.15 O~x
the equation in the transform x(s) becomes

m[s2x(s)- sXo - vo] = -kx(s) + f(s),
where f(s) is the transform of the force function F(t). Let Wo again denote
the natural frequency of the system,
wo=A.
Then we can write
xos + Vo 1 1
(3) x(s) = s2 + Wo2 + -m f(s)---z- ..
s + Wo
Hence the displacement for any F(t) can be written, with the aid of the
convolution, as
~ sin wot + ~
(4) X(t) = Xocos wot +
Wo mwo o f
' sin wo(t - -r)F(r) d-r,
a result that satisfies conditions (1) and (2) above.

But the motion of the mass under particular external forces F(t) is
more interesting than the general formula (4). In these special cases it is
often easier to refer to the transform (3) than to formula (4).
When F(t) is a constant F o, as in the case when the X axis is vertical
and the force of gravity acts on m, Eq. (1) can be written
mX"(t)= -k[ X(t) - ~J.

If Y = X - Fo/k,this becomes mY" = -kY; so the motion is the same as
freevibrations if displacementsare measuredfrom a new origin at a distance
of Fo/kunitsfromO. Note that Fo/k is the staticdisplacementXo of the
free end of the spring due to the applied forceFo, sinceFo = kX o.
In case
(5) F(t) = Fo when O < t < to,
=0 when t > to,
then
f(s) = F{~ - exp(;tos1
Ir Xo = Vo = O,it follows from Eq. (3) that
1 - exp(-toS)
(6) x(s) = FO
[
m S(S2 + (02) S(S2 + (02) ].
Now
and ifwe write tf(t) = sin2 !wot when t > O,

=0 when t < O,
it follows from Eq. (6) that
2Fo
(7) X(t) =k [tf(t) - tf(t - to)].
The graph of this function can be drawn easily by composition of ordinates.
When to is approximately !n/wo, the graph is the full-drawn curve in Fig. 16.
When to = 2n/wo, it follows from Fig. 16 that the mass m performs
one oscillation and then remains at the origin (Fig. 17).
The step function (5) is proportional to the unit finite impulse function
(3), Seco12, with impulse starting at t = O.The area under the graph of our
function is Foto. Ir this is kept fixed, Foto = Mo, while to -+ O and Fo -+ 00,
the force F(t) becomes the impulse represented formally by the equation
(8) F(t) = M ol5(t),
where l5(t)is the unit impulse symbol (Sec. 13). The constant M o represents
the increase in momentum of the mass m at the instant t = O.
X(t)
Fig.16
1
..
ELEMENTA.RY APPLICATIONS [SECo 30 91
Xt
lO, 2FO)
"
o
Fig.17
The formal transform of M o <5(t)was found to be M o. This suggests

that, in view of Eq. (3), when Xo = Vo= O,
Mo 1
x(s) = - 2 "
m s + Wo
and hence that
Mo.
(9) X (t ) = -SInWoto
mwo
It will be left as an exercise to show that formula (9) follows rigorously

as a limiting case of formula (7). But the formal solution can be verified as
well by noting that, according to formula (9),the momentum of the mass m,
M(t) = mX'(t) = Mo cos wot,
satisfies the condition M( +0) = Mo. Ir the mass started from rest, then its
momentum jumped to the value M o at the instant t = 00
30. RESONANCE
Let the external force in the problem of the last section be

F(t) = Fo sin wt,
where Fo and w are positive constants. Then according to Eq. (3), Seco29,
w
(1)
and, if w ::f:wo,
l Fow .
(2) X(t) = Xocos wot + -Wo
[Vo +
m (w
2
- Wo
2
)J
SIn wot
- Fo
20
m(w - Wo2)SInwto
That is, the motion is the superposition of two simple harmonic motions,
one with frequency Wo and known as the natural component of vibration,
and the other with frequency w which is called the forced component of the
vibration. Note that the natural vibrations are not present in case
Fow
Xo = O, Vo= 2 2 '
m(Wo - w )
However, if the frequency of the periodic force F is the same as the

natural frequency of the oscillator, w = wo, then
(3)
The presence of the repeated quadratic factor in the denominator here shows
that X(t) will contain an unstable component (Sec. 26) having the form ofthe
product of t by a cosine function. In fact,
l Fo . Fo
(4) X(t) = Xo cos wot + ---"2vowo
+ _ SIRwot - - t cos wot.
Wo ( 2m) 2 mwo
In view of the last term here, the amplitude of the oscillations of m increases
indefinitely.
In this case the force F(t) is said to be in resonance with the system.
We note in particular that if Xo = O and Vo = - Fo/(2mwo) the resonance
type of motion reduces to
Fo
X(t) = .-,.2mwo
- t cos wot,
shown in Fig. 18.
--
-- --
X(t)
"'"
"'".....-
.....---
o (.:. ,o/ ' t

---u
--- -- ---
Fig.18
ELEMENTARY APPLlCATIONS [SEC. 30 93
PROBLEMS
1. A good mechanical oscillator for demonstrating vibrations can be constructed by

selecting a length of coil spring such that when a weight of! lb H = mg,g = 32ft/sec2)
hangs at rest from the free end, the spring is extended by 0.2 fl. Show that the spring
t
constant then has the value k = Ib/ft, and that the natural frequency of the oscillator
is Wo = 4JiO = 12.65rad/sec, or about 2 cycles/sec,and hence the period T of
oscillation, T = 2n/wo, is approximately ! seco
2. Let the supported upper end of the spring in Probo 1 be moved vertically with
displacement Z(t), where Z(O)= O. Let X(t) denote the vertical displacement of the
mass m from its position on the lower end of the spring at time t = Owhen the spring
has its natural length. AIso let Z and X be positive for displacements downward.
If damping forces are neglected, note why X(t) then satisfies the equation
mX"(t)= - k[X(t) - Z(t)]+ mg.
Let Y(t) be the displacement of the mass m from its static position when supported by
the spring while Z(t) = o. Show that
mY"(t) = -k[Y(t) - Z(t)] = -kY(t) + kZ(t),
and hence that Y also represents the displacements in a horizontal oscillator with an
external force kZ(t) applied to the mass.
3. The upper end of the spring in the oscillator described in Probs. l and 2 is suddenly
lowered l in. at the instant t = O and kept in that position during twice the natural
period 2T = 4n/wo sec (about l sec), then suddenly returned to its initial position,
so that
Z(t) = -h[1 - So(t - 2T) (t > O)
Ifthe mass m (ofweight t lb) initially hangs at rest, show that it then makes two oscilla-
tions 2 in. downward from and back to its initial position, then remains there after
the instant t = 2T,since
1 . 2 nt Y(t) = O when t ~ 2T.
whenO;; t;; 2T,
Y(t) = "6sm T
4. In Probo 2 if Z(t) = vot and Y(O) = Y'(O)= O,show that
Y(t) = vot - ~ sin wot

Wo
and desocibe the motion of the upper end of the spring and the mass.
5. Let the upper end of the spring in the oscillator described in Probs. l and 2 be
moved vertically by means of a rotating eccentric or cam with the periodic displacement
Z = l2 sin wot ft, with the natural frequency Wo = 4JiO rad/sec of that oscillator.
If Y(O)= Y'(O)= O,show that
l. Wo
Y(t ) = 24 smwot - 24 tcoswot.
Note that the vibration of the mass is unstable under that periodic 2-in. oscillation of
the support, and that within only four cycles of the oscillations (t = 8n/wo < 2 sec),
displacements of the mass mount to approximately l ft.
6. If the coefficient of damping for the oscillator shown in Fig. 13 is so large that
e2 > 4km and if X(O) = Oand X'(O) = vo,(a) show that
X(t) = :
V. et
smh (at) exp ( - 2m') wherea =
p-4km
2m .
(b) Find X'(t) and show that the mass m moves in the direction of vo until the
instant
11 e+2am
t = 2a og e- 2am'
when it turns and approaches the origino
7. Let the mass m in the damped oscillator shown in Fig. I3 be released from rest
with an initial displacement X(O) = xo, and write 2b = e/m. (a) If e2 < 4km and
WI = (W02 - b2)t, show that
. b 1t
wh eresm(X=-, O;;(X< "2;
Wo
(b) if e2 = 4km show that
X(t) = xoe-bt(l + -bt),
and hence that the mass never moves across the origino
8. When the force on the undamped oscillator shown in Fig. 15is the periodic function
F(t) = Fo sin (wt + 8),where 8 is a constant, show that resonance occurs when w = Wo.
9. When t = O. let the current in the circuit shown in Fig. 14 be zero and let the
capacitor have a positive charge Qo. If R > 2jL!C, derive the formula
Rt . (Xt t
2Qo 4L
l(t)= --exp
(XC (
--
2L)
smh-
2L (
2
(X= R --
C)[ ]
for the current. Show that l(t) ;;O and 1(00) = O.
10. The current 1, and the charge Q on the capacitor in the circuit shown in Fig. 19,
are functions of t that satisfy the conditions
L-dI + -Q + RI = E 1(0)= O,
dt C o' Q = f~ I(T)dT,
when Q and 1 are initial1y zero, where t is the time after cIosing the switch K, and the
electromotive force Eo is constant. If b = R(2L)-1 and Wl2 = (LC)-l - b2 > O,
derive the formula
1 = EOe-bl sin wlt.
WIL
Fig.19
31. FORCED VIBRATIONS WITH DAMPING
When an external force F(t) acts parallel to the X axis on the mass in the
damped oscillator shown in Fig. 13,the equation of motion becomes
(1) mX"(t) = - kX(t) - eX'(t) + F(t).
Ir X(O)= X'(O) = O,the solution of the transformed equation is
1 1
(2) x(s) = -m f(s)
Again, consider the periodic force

F(t) = Fo sin wt
and the case e2 < 4km. Then b < wo, and Eq. (2) becomes
Fo w
(3) (W12 = W02 - b2).
x(s) = -;;(S2 + w2)[(S + b)2 + Wl2]
It follows from Theorem 14, Seco26, that X(t) consists of terms of the types
A cos (wt + e) and Be-bt cos (wI + el), where A, B, e, and el are constants.
ConsequentIy the component of oscillation with frequency WI is nearIy
damped out after a sufficiently long time, and the periodic forced component
of vibration
(4) Xp(t) = A cos (wt + e) (A > O)
remains. Its amplitude A is, according to Seco 26, 214>1(iw)1 where
4>1(S) = (s - iw)x(s). Thus
(5) A = IFol 1 .
m I(iw+ b)2 + wl21
The square of the absolute value in the denominator can be written as
IW02 - W2 + 2ibwl2,which is (W2- W02)2 + 4b2w2,and by expanding and
completing the square, we can write that polynomial in w as (W2 + 2b2 -
Wo 2)2 + 4b2(wo 2 - b2). Thus
IFol
(6) A= - ~ u~ ~-~ ~_..
The amplitude A of the periodic vibration depends on the frequency w

of the force Fo sin wt, and it varies directIy with the amplitude IFol of that
force. When Fo is fixed, the value of w for which A is greatest is the resonance
frequency wr of our damped oscillator under that periodic force.
Let the coefficient of damping be small enough that e2 < 2km. Then
2b2 - W02 < O, and we can see from formula (6) that A is greatest when
002 = 0002 - 2b2. Thus the resonance frequency in that case is
(7) 00, = Jooo~ - 2b2 (2b = ;,e2 < 2km),

a frequencyonly slightly less than 000when e is small. When 00= 00" the
amplitude A of the undamped component of vibration has the value
(8) A = ~= IFol
, 2bmool eOOl'
We can seethat the expression(3) for X(5)cannot contain a repeated
factor of the type (52 + (2)2 in its denominator unless e = Oso that b = O.
Hence no value of the frequency w of our periodic force will induce a com-
ponent of X(t) of type t cos (wt + 9) as it did for the undamped oscillator.
The amplitude A of the forced component of vibration is still given by
formula (5) when the force Fosin wt is replaced by a force Fosin (wt + IX)
(Prob. 3, Seco32).
In case e2 ~ 2km, it tums out that the amplitude A of the forced
vibration tends toward an upper bound IFol/kas 00tends to zero. When w
is small and t is large enough that wt is near an odd multiple of n/2, the
magnitude of the force Fo sin wt is then near IFol. Under static conditions
the force IFol would displace the end of the spring a distance IFol/k. The
formula for X(t) when e2 = 2km, found in Probo 4, Sec. 32, shows that when
w is small the mass may oscillate slowlywith an amplitude near IFol/k.
32. A VIBRATION ABSORBER

We have seen that for the simple damped oscillator with an excitingforce
Fo sin wt the forced component of vibration A cos(wt + O) remains un-
damped. Let another spring and mass be connected in series with the
original mass (Fig. 20), where that second oscillator is undamped. We shall
see that if the spring constant and mass of the auxiliary oscillator are chosen
so that the natural frequency of that oscillator coincides with the fixed
frequency w of the exciting force, then the forced vibrations of the first
mass will be eliminated.
F!tl
mI
m
A k kl
I
OL ..x
Fig.20
ELEMENTARY APPLlCATlONS [SECo 32 97
Let X and X 1 denote the displacements of the masses m and mi'

respectively, from the positions they have when both springs have their
naturallengths. If the exciting force is
F(t) = Fo sin rot,

and if m and mi are initially at rest at their respective origins, then the
displacements X(t) and X i(t) satisfy the following system of differential
equations:
d2X . dX
m dt2 = -kX + k1(X1- X) - c-; + Foslnrot
d2X1
mld(2 = -k1(X1 - X),
X(O) = X'(O) = X 1(0) = X'I(O) = o.
The transforms x(s) and XI(s) of X(t) and X l(t) therefore satisfy the simul-
taneous algebraic equations
Foro
(ms2 + cs + k + k1)x(s) - k1xi(S) = -y- 'l'
S +ro
kix(s) - (m1s2 + k1)Xi(S) = o.

Eliminating x1(s), we find that
mls2 + ki
(1) '
x(s) = Foro(s+rops
2 2) ( )
where
(2)
In view ofthe presence ofthe quadratic factor S2 + ro2in the denomina-

tor of the right-hand member of Eq. (1), it follows that X(t) will contain a
term of the type
(3) ecos (rot + O)
unless kdml = ro2,in which case the numerator ofthe above fractioh cancels
with the.factor in the denominator, leaving
1
(4)
x(s) = FOromlp(s)'
The inverse transform X(t) of the function (4) represents a damped
oscillation of the mass m if each of the four roots, real or imaginary, of the
equation p(S)= Ohas a negative real part (Secs.24 to 26). Now the poly-
nomial p(s) is the determinant of this system of homogeneous equations in
x and Xl:
(ms2+ es + k + kl)x - klxl = O,
(5)
-klx + (mls2 + kl)XI = O.
When the determinant vanishes, that system is satisfied by numbers x and
Xl' both differentfrom zero.
Let s be any one of the four roots of the equation p(s) = O and let
z,z1 denote a pair of nonvanishing roots of the system(5)that corresponds
to that value of s, where z and z1may be either real or imaginary numbers.
We now write X = z and Xl = Zl in Eqs. (5), then multiply the members of
those equations by the complex conjugates Zand Zl' respectively,and add
to obtain the equation
(6) (mzz + mlzlzl)s2 + ezzs + B = O,
where B = kzz + kl(zz - ZZI- ZIZ + ZIZ)
= klzI2 + kllz - zl12 > O.
Ifwe write A = mlzI2 + mllzl12,then Eq. (6)becomes
(7) As2 + elzl2s + B = O,
where the coefficients A, clzl2, and B are positive numbers that depend on
the value of s. However, the number s is given in terms of those positive
coefficients by the quadratic formula
1
(8) s = 2A (- clzl2 :f: .Je21z14 - 4AB).
Whether the radical here is real or imaginary, it follows from formula (8)
that the real part of s is negative. This completes the proof that the oscillation
X(t) is damped.
Thus the forced component of the vibration of the main mass m is
eliminated by the system mI' kl, if the natural frequency of that system
coincides with the frequency of the exciting force:
(9) [k; = OJ.

V~
Since all components of the vibration of m are then damped, that mass
approaches a fixed position as t increases.
This is the principIe of the Frahm vibration absorber, which has been
used in such practical appliances as electric hairclippers.l Note that, in
1 Den Hartog, J. P., "Mechanical Vibrations," 4th ed., pp. 87 fr., 1956.
ELEMENTARY APPLlCATIONS [SECo 32 99
view of Eq. (9), the absorber is designed for a fixed frequency ro of the exciting
force Fo sin rot.
By solving the above equations for XI (s),we can see that the mass mi
has an undamped component of vibration of the type (3).
PROBlEMS
1. If the initialconditionsX(O)= X'(O) = Oused in Seco31 for the damped oscillator
with exciting force F(t) are replaced by the conditions X(O)= XO,X'(O)= vo, show
that the additional terms in the formula for X(t) represent damped oscillations.
2. When F(t) = FoD(t)and X(O)= X'(O) = Ofor the damped oscillator considered in
Seco31, where e2 < 4km, show formally that
Fo b.. e 2 k 2
X( t) = -e- s!nrolt = -,rol =- - b ,
mrol (b 2m m )
and note that the jump in the momentum of the mass m at the instant t = Ois Fo.
3. Show that formula (6), Seco31, for the amplitude A of the forced component of
vibration for the damped~scillator is valid (a) when the exciting force F(t) is Focos (J)f;
(b) when F(t) = Fo sin (rot + IX).
4. If e2 = 2km for the damped oscillator considered in Seco31, and if F(t) = Fosin wt
and X(O)= X'(O) = O,show that
X(t) = ~[COS 92cos (rot+ 9) - e-b. cos 91 cos (bt + (2)]

where 2b = e/m and 91 and 92 are these arguments of complex numbers:
91 = arg [ - 2bro + (ro2 - 2b2)],
Hence when ro is small and rot is large, show that X(t) is approximately (Fo/k)sin rol.
5. In Seco31, if e2 = 4km, F(t) = Fosinrot, and X(O)= X'(O)= O,show that X(t)
has the form A cos (rot + IX)+ (B + Ct)e-bt, and when ro is small that IAI is near its
upper bound IFol/k.
6. Let the oscillator describedin Probs. 1 and 2, Seco30,for whichk = Ib/ft and t
t
mg = lb, be subject to a damping force - eY'(t) where e = f4Ib/ft/sec. (a) If the
upper end of the spring is fixed so that Z(t) = O,and if Y(O)= Yoand Y'(O)= O,show
that after about nine cydes of oscillation of the mass the amplitude of the displacements
Y(t) is reduced to approximately o.llyo!. (b) When the upper end is oscillated so that
Z(t) = Zosin rot, show that resonance occurs if ro = ror where ror = )159.5 rad/sec,
a frequency slightly less than roo(cf. Seco31). (e) When ro = rorin part (b),show that the
amplitude A ofthe undamped forced compone~t ofvibration exceeds 121zol.
7. Let the two masses in the system treated in Seco32 have arbitrary initial displace-
ments and velocities. When condition (9) is satisfied by the eIements of the absorber,
show that the vibration of the main mass again contains no undamped component.
8. (a) If the exciting force Fosin rot in the system of Seco32 is replaced by the force
Fosin (rot + IX),where IXis any constant, show that when condition (9) is satisfied the
vibration of m is again entirely damped.
(b) If the exciting force is replaced by Al sin rolt + A2 sin ro2t,where the A's and
ro's are constants and rol f= ro2' show that the values of mi and kl cannot be adjusted
so that alI undamped vibrations of m are absorbed.
9. Let the force F(t) be removed from the mass m in Fig. 20 and let the end A of the
spring k be moved horizontalIy so that its distance from the walI is F(t)/k. Show that
the differential equations of motion of m and mi are then the same as in the original
problem.
10. A system of two masses and two springs in series, with no damping, is shown in
Fig. 21. The weights of the two masses are mlg = 8lb and m~ = 2lb (g = 32 ft/sec2).
The spring constants have the values kl = 24lb/ft and k2 = 8lb/ft. If a periodic force
F = Focos 2nct acts on m2' find the frequencies c (cycles per second) of the force for
which resonance occurs in the vibration of mi' Assume zero initial displacements and
velocities. Also show tltat resonance occurs in the vibration of m2for those same values
of e. Ans. e = 4n-I, 4 J3 n-l cycles/sec.
Fig.21
11. For the system shown in Fig. 21, if mi = m2 = ls, kl = ?s, k2 = ?s, and if both
masses start from rest in their equilibrium positions, find the formula for the displace-
ment X l(t) of mi when the force F(t) on m2 is arbitrary.
Ans. X l(t) = (6 sin t - J6 sin t J6) * F(t).
12. InitialIy the two masses shown in Fig. 22 are at rest and the spring has its natural
length. ThenaconstantforceF = Fo(t > O)actsonml' Ifthesystemisfreefromdamp-
ing, find the formula for the displacement of m2 and note that the displacement consists
of a simple harmonic motion superimposed upon a uniformly accelerated motion.
Also show that ifFo > O,the spring is always compressed by an amount proportional to
1 - cosrot,wherero2= k/ml + k/m2'
Ans. 2w2(ml + m2)X 2(t) = Fo(ro2t2 - 2 + 2 cos rot).
I Fig.22
13. If the force F = O in the system shown in Fig. 22 and if the masses are initialIy
released at rest from positions Xl = al and X 2 = a2' show that the frequency of
vibration of the masses has the value ro = (k/ml + k/m2)t. When a2 = -alml/m2'
I show that XI = al cos rot and X 2 = a2 cos rot and describe the vibration.
I 14. Let the force in the system shown in Fig. 22 be the periodic force F = Fosin rot,
where ro2 = k/m2' and let a viscous dampingforce -eX(t) act on mi' Find the steady-
state vibration of m2 and note that it does not depend on c. Ans. - Fok- 1sin rot.
! 15. In Fig. 23 the end E of the first spring has a periodic displacement Y = A sin rot.
i The two springs are identical. Find the value of ro for which this damped system is in
resonance, when e2 < 4km. Ans. 2m2ro2= 4km - e2.
I
1
[
ELEMENTARY APPLlCATlONS [SEC. 32 101
Fig.23
E
~
I
I
~y
k
I
~X
m
16: In Fig. 24 the two masses are unit masses (m=:=1)and the two springs are identical.
Initially, the springs have their naturallength, the first mass has velocity vo, and the
second mass is at rest. Find the undamped component of vibration of those masses.
Ans. !(vo/.jk) sin t .jk.
Fig.24
17. For the undamped system of two equal masses and three identical springs shown
in Fig. 25, the initial conditions are XI(O) = al' X2(O)= a2, X'I(O)= X(O)= O.
(a) If lall * la21,show that the components of vibration of each mass have frequencies
-fl0 and J3k/m. (b) If a2 = al' show that both masses vibrate in unison with
frequency -fl0. (e) If a2 = - al' showthat the massesvibrate in oppositedirections
with frequency J3k/m.
f I
Fig.25 L-.. Xl L-. X2

18. A dampedabsorber Figure 26 shows an absorber with elements mi, kl, and e in
which the damping is located in the absorber itself. When a force Fosin wt acts on
the main mass m, the undamped vibrations of m cannot be completely absorbed here,
but the coefficient of damping e can be adjusted to give an optimum range of the ampli-
tudes ofthose vibrations. Let mand mI be initially at rest in their positions of equilibrium.
Derive the formula
x(s) =~ mls2 + es + k
S2 + W2(mls2 + es + k)(ms2 + es + k + k) - (es + k)2
for the transform of the displacement X(t) of m, then show that the amplitude A of the
forced component A cos (wt + O)of X(t) is given by the formula
A2 (mw2 - k)2 + e2w2
F? = [(mw2 - k)(mw2 - k) - mkw2]2 + e2w2[(m+ m)w2 - kp.
(A graphical examination of A2 as a function of W2 would indicate a value of e for
whichthe rangeof valuesof A2willbe as smallas possiblefor all w.)
See Den Hartog, op. cit., pp. 93 ff.,ror a detailed discussion.
Fig.26
33. ELECTRIC CIRCUITS

In Seco28 we used Kirchhoff's voltage law: for each closed circuit in an
electrical network the impressed voltage equals the sum of the voltage drops
across the elements in that circuit. His law 01 currents states that at each
junction point of branches of a network the total current into the point
equals the total current away from the point. Let us now present additional
applications of these laws for instantaneous behavior of voltage and current,
in setting up differential equations for networks.
An electrical analog of the Frahm vibration absorber discussed in
Sec. 32 is indicated in Fig. 27, where the impressed voltage is
V = Va sin rot.
According to the law of currents, applied at the junction P,
(1)
We apply the voltage law to the circuit on the left to see that
l l
(2)
V = LI'(t) + RI(t) + CQ(t) + CI QI(t),
where Q'(t) = I(t) and Q'l(t) = II(t); then to the circuit on the right to see that
(3)
~l QI(t) = LII;(t).
Ll
Fig.27
.1
Let Q2(t) denote the difference petween quantities of charge on the

two capacitors, so that
(4) Q~(t) = 1(t) - 11(t) = 12(t).
In terms of Q and Q2 Eqs. (2) and (3) take the form
LQ"(t) = - ~Q(t) + ~ [Q2(t) - Q(t)] - RQ'(t) + Vosin wt,

(5) 1 1
LIQ2(t) =- CI [Qz(t) - Q(t)].
Except for differences in notation the system (5) of differential equations
in Q(t)and Q2(t)is precisely the system written in Seco32 for the displacements
X(t) and Xl (t) of the masses m and mi in the absorber (Fig. 20). Since all the
electrical coefficients here, R, L, etc., are positive constants, the signs of the
coefficientsin Eqs. (5) match those in the equations for X and X l' The
initial conditions for the network match those chosen for the absorber, and
the analogy is complete if
(6) Q(O) = Ql(O) = 1(0) = 12(0) = O.
Under conditions (6) the transforms of 1(t)and 12(t)are sq(s) and SQ2(S),
which correspond to sx(s) and SXl(S) for the absorber. According to the
results found in Seco 32 therefore, the current 1(t) has an undamped com-
ponent of type B cos (wt + IX)unless
1
(7) - = W2.
LICI
But when the elements ofthe LI C 1circuit satisfy condition (7)for a prescribed
frequency w of the impressed voltage V,the current 1(t)contains only damped
components; thus 1(t) ~ O as t ~ oo. The currents 11(t) and 12(t) in the
LI C1circuit, however, do have undamped components oftype B cos (wt + IX).
Note that Eqs. (1) to (3) with conditions (6) can be solved directly for
the currents either by replacing Q(t) by f~ 1(.) d. or q(s) by i(s)/s, etc. Thus
the transformations of Eqs. (2) and (3) gives the equations
(LS + R + ~s)i(S) + cljl(S) = v(s),

(8)
LISi(s) - (LIS + Clls)il(S) = O.

Upon eliminating il(s), we find that
(9) i(s) = v(s)y(s) if y(s) = s(LIs2 + CI-I)

p(s) ,
where P(s)is the polynomial introduced in Seco32:

p(s) = (L1S2 + Cl-l)(Ls2 + Rs + C-l + Cl -1) - Cl -2.
The function y(s) depends only upon the characteristics of the network,
not upon the impressed voltage. Since the formal transform of the unit
impulse symbol b(t) is unity, it follows from Eq. (9) that y(s) = i(s) when
V(t) = b(t) formally; that is, Y(t) represents the current 1 produced by the
voltage b(t). When the network is considered as a system with input V(t)
and output I(t), then, in the language of systems analysis, y(s) is the transfer
function for the system; in view of Eq. (9) the output corresponding to any
input V(t) is given by the formula
(lO) I(t) = V(t) * Y(t).
PROBLEMS
1. If 11(O)= 12(0)= Oin the network shown in Fig. 28, while the capacitors have the
same initial charge Qo, show that
wherero = (CL)-t.
e
11 R 12
Fig.28
2. Ifl1(0) = 12(0) = Oin thenetwork shown in Fig. 28 and ifQI(O) = alandQ2(0) = a2,
find the undamped component 1 (t) of the current 11(t), the steady-state current through
the first inductance coil. Ans. 1 = -!(al + a2)wsin rot,ro = (CL)-t.
3. Show that the network indicated in Fig. 28 represents an electrical analog of the
mechanical system shown in Fig. 24 where QI and Q2 correspond to the displacements
of the two masses.
4. In Fig. 29 the currents 11(t) and 12(t)and the charge Q(t) are initially zero.
(a) Show that the current 11(t) is unstable under the impressed voltage V =
Vosin (rot + ex)if ro2 = (L1q-l + (L2Q-I.
(b) Show that the system indicated in Fig. 22 is a mechanical analog to the
network here with X I(t) and X 2(t)corresponding to Ql(t) and Q2(t),where Q'l(t) = 11(t),
Q(t) = 12(t), and Q(t) = Ql(t) - Q2(t).
Fig.29
5. A prescribed current l(t) is supplied to the network shown in Fig. 30. The initial
values of 1I(t)and Q(t)are zero. Use the following notation: V(t) is the resulting potential
drop from the junction A to the junction B, m02 = (LC)-I, 2b = (RC)-I, and
ml2 = m02 - b2, given that b2 < m/.
(a) Show that the transfer function from input l(t) to output 1(t) is
y( s)
- m o2
- S2 + 2bs + mo2'
thus thatI (t) = l(t) * Y(t), where Y(t) = m/ml -Ie-br sin mi t.
(b) Find V(t) when l(t) = lo, a constant. Ans. V(t) = 10(Cm)-e-br sin mlt.
Fig.30
6. In a simple LC circuit with inductance coil and capacitor in series, let the impressed
voltage be a periodic function, with period T, of the type
N
V(t) = ao + L (a. cos nmt + b. sin nmt),
n=1
where m = 2n/T. Show that the current l(t) in the circuit is unstable ifthe frequency m
of V(t) has any one ofthe values
l 1
m- - (m = 1,2,..., N),
-myU
providedthat amand bmare not both zero.
7. In the network shown in Fig. 31 all currents and chargesare zero at the instant
t = Owhen the switch K is closed. If the impressed voltage is given by the equation
V = Vosin mt, where m2 = 2(LC)-I, find the steady-state value 12(t) of the current
12(t)and note that it is independent of the value of the resistance R.
Ans. 12(t) =- VoCmcos mt.
Fig.31
8. In the system shown in Fig. 25 apply an exciting force to the first mass and viscous
damping to the second mass. Show that the resulting system is a mechanical analog
ofthe network shown in Fig. 31.
9. Initially the currents and the charge on the capacitor are zero in the network shown
in Fig. 32. The voltage drop from Al to B1 caused by the mutual inductance of the
two coils is MI;(t), and the drop from A2 to B2 caused by that mutual inductance is
Ml(t),whereM2 < L1L2. Theimpressedvoltageisgivenbytheequation V = Vocoscot.
When the values of C and L2 are adjusted so that L2C = co-2, show that the current
Il(t) in this idealized resistance-free circuit has a simple periodic variation at all times
with a frequency greater than co.
Fig.32
10. Show that the network in Fig. 33 represents an analog of the damped absorber
shownin Fig. 26 if initial currents and charges are zero, where charges Q and Q - Q
correspond to displacements X and Xl and the voltage V corresponds to the force
Fo sin cot.
Q
~c 1
Ql
R 11
Fig.33
34. EVALUATION OF INTEGRALS

Laplace transforms can be used to evaluate some types of integrals containing
a parameter. The manipulative procedure is often direct and simple, but, as
the examples and exercises here will indicate, proofs of the reliability of
those formal results can be challenging. In our proofs we use this special
case of the theorem on uniqueness stated in Seco6: A function f(s) cannot
have more than one inverse transform F(t) that is continuous over the half
line t ~ Oand of exponential order.
Example 1 Evaluate the integral
(1)
ELEMENTARY APPLICATIONS [SECo 34 107
Here F(t) is the Fourier cosine transform (Chap. 10) of (X2 + a2)-1.
Note that F is bounded: F(t)1;;;; f~ (X2 + a2)-1 dx = n/12al; also,
-
F( t) = F(t).
Let us first proceed formalIy to transform with respect to t and
interchange the order of integration with respect to x and t:
eo dx eo s dx
(2) I(s) = fo L{costx}
x +a
2 2 = fo
eo
s 1 1
= - . f
o ( x2 + a2 - x2 + S2 )
d
x
n
= 2a s +
1
a
if a > Oand s > O. Thus the integral has the value
n
(3) F(t) = - e-al (a > O,t E:;;O).
2a
TOsee that the formal step (2)is sound, we first note that, whenever
T and s are positive constants,
T T
eo cos tx eo 1
(4)
o fe-sI
ox+a
2f 2 dx dt =
ox+a
2 2
o f
e-sI cos tx dt dx, f
because the absolute value of the entire integrand does not exceed
(X2 + a2) - 1, a function independent of t whose integral from x = O
to x = 00 exists. The integral F(t)e-st therefore converges uniformly
with respect to t when t E:;;O by the Weierstrass test and (Sec. 15) the
interchange of order of integration with respect to x and t made in
step (4)is valido Also, F is continuous (t E:;;O)and bounded so, according
to the theorem on uniqueness, if L{ F} is given by formula (2), then F
must be the exponential function (3).
The final integral in Eq. (4) can be evaluated by integration by
parts, and the equation can be written
(5)
fOTe-SIF(t)dt = Leog(x,T)dx,
where T) - e-SI(xsin Tx - s cos Tx) + s

g(x, - (X2 + a2)(x2 + S2) .
This continuous function of x and T has a limit g(x,oo),
g(x,oo) = T-+eo
lim g(x, T) = S(X2 + a2)-I(x2 + S2)-I,
uniformly with respect to x when x E:;;Osince

Ig(x, T) - g(x,oo)1 < e -sT 2 2
X
2
+s
2
<
= M e -sT ,
(x +a )( x +s )
where M is the maximum value of the quotient of polynomials in X.
To eachpositive number f there corresponds a number T.,independent

of x, such that Me-sT < f when T> T.. Moreover, the second integral
in Eq. (5) converges uniformly with respect to T, according to the
Weierstrass test. The limit of the integral (5) as T ~ 00 is therefore
fO' g(x,oo)dx (Prob. 10, Seco 27), since the latter integral exists; thus
Eq. (2) follows from Eq. (4).
Example 2 Evaluate the integral
(6) F(t) = (a) sin tx dx.

Jo x
The formal procedure is simple. When t > Oand s > O
. dx dx n
(7) f(s) = i o
a)
L{smtx}-
x
= ia)
oX+s
2 2 =_
2
;
s
hence F(t) = n/2. In view of Eq. (6), F(t) is an odd function: F( - t) =
- F(t), and F(O)= O. Therefore
n n
(8) F(O) = O.
F(t) = 2" (t > O); F(t) = -2" (t < O);
The Weierstrass test cannot be used here to justify step (7).
In fact, if our formal result (8) is correct, then F is discontinuous at the
origin t = O,and the integral (6) cannot converge uniforrnlyover an
interval that includes the origino
Now the function S(r) = sin rlr, r =1=
O, S(O) = 1, that is,
r2 r4
(9) S(r)= 1 - -3! + -5! - . . . (-oo<r<oo)
is continuous, and IS(r)1~ 1, for all real r. Note that the sum of the
alternating sees (9) is positive and not greater than unity when
Irl ~ 1, and if Irl > 1, then Isinrlrl < 1. To see that the integral (6)
exists, we note that its integrand tS(tx) is continuous in t and x and,
if to, xo, and XI are positive constants, an integration by parts shows
that
= cos txo - cos tx 1 -

f Xo
XI sin tx dx
x txo tXI f
Xo
XI cos:x dx
tx
and this has a limit as XI ~ oo. Thus F(t) exists because
sin tx cos tx
F( t) =
XO
fo -
x
dx+--cos txo
txo f XI)
a)
tx
y- dx (t > O),
and because F(O)= Oand F( - t) = - F(t).

ELEMENTARY APPLlCATIONS [SEC. 34 109
When t > O,the substitution r = tx showsthat our integral has

a constant value, F(t) = SO'r-l sin r dr. Hence F is bounded, con-
tinuous when t > O, and F( + O) exists. Ir we establish formula (7) for
/(s), then F must be the step function (8).
The remainder for integral (6) can be written
sin tx cos tX cos tx
R(t,X) =
f x
co
-dx
x
= -
tX
-
f
co
x tx
r dx
and we can seethat, when t ~ to > O,for eachpositive number there

is a number X independent of t such that
(
l 1 2
IR(t X )I ::;;-
::;;- < when X> X(. +-
, - tX tX - toX
Thus integral (6) converges uniformly with respect to t
(t ~ to > O),and wecanwrite, with the aid of anelementaryintegration,
T co 1
T
f lO
e-SIF(t) dt =
f f
o X
-
lo
e-sI sin tx dt dx
= Lco h(T,x) dx - Lco h(to,x) dx

where h is this continuous function:
e-sI sin tx
h(t,x) = - x 2 +s 2 st-
tx
+ cos tx (5 > O).
( )
The integral SO' h(to,x) dx converges uniformly with respect to
to(O < 1) according to the Weierstrass test, so it is continuous in
;;;;; to
to at the point to = O; thus
T co
dx
(lO)
f o
e-SIF(t) dt =
fo
h(T,x) dx + f
co
ox+s
-y- (5 > O),
Let N(s) denote the maximum value of(sT + l)e-sT. Then

N(s)
(11) Ih(T,x);;;;;I 2 2 and Ih(T,x) I =< sT z-+ 1 e -sT .
x + s s
The uniform convergence of the integral SO' h(T,x) dx and the fact that
h(T,x) -. Oas T -. 00, uniform1y with respect to x, foIlow from con-
ditions (11). Thus as T -. 00, that integral vanishes, and Eq. (10) leads
to step (7), so the bounded continuous function F has the transform
tn/s, and therefore F(t) = tn when t > O. This completes the proof.
. IncidentaIly, we have shown that Si(00) = tn, for when t > O,we
found that F(t) = !1t, and Eq. (6) can be written
co sin r . 1t
(12) F(t)= o -;:-dr=Sl(oo)="2' f

35. EXPONENTIAL- AND COSINE-INTEGRAL FUNCTIONS

Convenient forms of the exponential-integral function are
oo -y oo -Ix
(1) E1(t) =
f I
=-dy =
y f 1
~dx
X
(t > O).
Formally, the transform of the second integral is

oo dx 1 oo 1 1 1 x 00
(2) =-
s) s f 1
- - - dx = -log - (s > O).
J,1 x(x + (X X + s) s x+s ] 1
The proof that expression (2) represents L{El(t)} is left to the problems.
The basic form of transform 100, Appendix A, Table A.2, is, therefore,
1
(3) L{E1(t)} = -log(s + 1) (s > O).
s
In Seco 22 we found that the transform of the sine-integral function,
Si t, is S-1 arccot s. Now let us wrte the transform of the cosine-integral
function
(4) Ci t =- oocos r dr = - (00cos tx dx (t > O).
f.I r J1 x
An integration by parts, first over a bounded interval (l,X), shows that
. sin t 1 oo sin tx
(5) Clt=---
t t f
1
-
X2
dx (t > O);
thus Ci t is continuous when t > O,and Ci (00) = O.

The procedure used in Example 2, Sec. 34, can be applied to the final
integral in formula (4) to prove that
oo dx 1
(6)
L{Ci t} = - s J,1 xx(2 +s 2) =- _2s log(s2 + 1) (s > O).
Next let us derivethe transform (No. 98,AppendixA, Table A.2)of the

function cos t Si t - sin t Ci t which can be written
cost
,-dr
sinr
+ smt
. oocos r
-dr
1o r f.I r
oo sin r oo dr
= cost fo -drr, -
f (sinrcost - cosrsint)-
r
when t > O. By replacing Si ((0) here by tn (Sec. 34) and substituting tx
for r - t, we can write the last expression in the form
1t oo sin (r - t) 1t oo sin tx
(7) -cos t -
2 f,r dr = -cos t -
2 f -dx.
ox+l
When we transfonn that final fonn (7), we find that

. . ' log s
(8) L{ cos t S1t - Sin t C1 t } = z-
s + 1 (s > O).
In like manner we can derive the transformation (No. 99, Appendix A,

Table A.2)
(9) L{ cos t Ci t + sin t Si t} = - S2

s log
+ S1 (s > O).
Finally, let us evaluate the Laplace integral

oo -tx
(10)
g(x) = fo t:+ 1dt (x > O).
Its transfonn with respect to x is found to be

n s Alogs
g(s) = 2 S2 + 1 - S2 + 1 (s > O).
In view of formula (8) we have the integration fonnula

(11) g(x) = (tn - Si x) cos x + Ci x sin x,
which is transformation 120, Appendix A, table A.2:
(s > O).
~ I} = (~n - Sis) cos s + Ci s sins
.
(12) LL2
PROBLEMS
Establish the integration formulas in Probs. l to 3, where t ~ O.
1. l '" sin tx
o x,(x2:1- I)dx
1t
= 2(1 -
-/
e ),
the Fourier sine transform of X-1(X2 + 1)-1 (Chap. 13).
'" sin2 tX '" 1 - eos2tx
2. f -r-
-'" x
dx =
o xl,x d = 1tt.
- exp ( - tx2) d
3.
lo
'" l
x
2 X
-- y'1tt.
c:
Obtain formally the integration formulas in Probs. 4 to 7.

'" .
4.
f-'" x sm tx
X2 + a2 dx = 1te-aI
(a > O, t > O).
5. (x ~ O).
1)fi, d,= 1te' erfe fi
f.o.'"(-r +e-X<
(ef. transformation 111, Appendix A, Table A.2.)
eo sin tx eo n 1-
6. fo fi dx f
= 2 o sin ty2 dy = ( 2t)
(t > O).
[Note that y4 + S2 = (y2 + S)2 - 2si = (y2 - y.j2s + s)(y2+ y.j2s + s).]
eo 1 /1r
7. fo exp(-tx2)cos2txdx =2 V-e-' (t > O).
8. WhenT > to> Oands > O,provethat

-s, eoe-'X eo exp [-(x + s)to] - exp [-(x + s)T] d .
T
f'o e f
-
X
dx d t =
f x(x + s)
x,
then prove that the limit of that integral as to -> Oand T -> 00 is the integral (2),Seco35,
which represents L{E(t)}.
9. Use the transformation (3), Seco35, to show that, when a > O,
logs - loga .
L{e"'E(at)} = s-a (s> a; cf. transformatlOn 97, Appendix A, Table A.2).
Obtain formally the transforms given in Probs. 10 to 12, if a > O and b> O
(see Probo 9).
eo -sr
I
10.
fo t+a ~dt =L - = easE(as)
{ t+a }
(cf.transformation118, AppendixA,Table A.2.)
11. L{(t /a~t b+b)} = ebSEI(bs)- easE(as) (a * b).
12. Ltt: a)2} = 1 - aseasE(as)(cf.transformation119,AppendixA,Table A.2.)

13. Prove that the remainderfor the integral Ci t = - SI x- COS
tx dx satisfiesthe
condition IR(t,X)1< 2J(toX)when t ~ to-> O and hence that the integral converges
uniformlywhent ~ to. Then completethe proof of transformation(6),Seco35.
When a > Oand t > O,obtain formallythe Fourier transformationsof(x + a)-
stated in Probs. 14and 15.
eo sin tx n . ..
14. fo x + adx = (2 - SI at) cos at + Ci at Sin ato.
eo cos tx n . . .
d S C
1.
5
fo x + a x = ( 2- 1at ) Sin at - at cos ato
16. Complete the derivation of formula (11), Seco35.

17. Obtain formally the transformation (9), Seco35.
18. The beta function is defined as follows:
B(x,y) = f1 r"-(1 - r)Y- dr (x > O,Y > O).

, Jo
Since L{r-I} = r(x)s-X, note that
L{tX-I * tY-I} = r(x)r(y)s-X-y = r(x

r(x)r(Y)L{tX+Y-l}.
+ y)
Write the convolution here in tenns of the integral representing B(x,y) to show that the
beta function can be written in tenns of gamma functions (Sec. 5) as
r(x)r(y)
(x > O, Y > O).
B(x,y) = r(x + y)
36. STATIC DEFLECTION OF BEAMS
Let Y(x) denote the static transverse displacement of a point at distance x

from one end of a uniform beam, caused by a load distributed in any manner
along the beam (Fig. 34). Under certain idealizing assumptions, primarily
that displacements Y(x) and slopes Y'(x) are small, it is shown in mechanics
that the internal bending moment M(x) exerted by any span of the beam
upon an adjacent span, through their common cross section Ax, is propor-
tional to the curvature of the beam at position x. In fact
(1) M(x) = ElY"(x),
where E is Young's modulus, the modulus of elasticity in tension and com-
pression for the material, and 1 is the moment of inertia of the area Ax with
respect to the neutral axis of the cross section, the line in Ax about which Ax
turns when the beam bends. Note that Y"(x) is approximately the curvature
since Y'(x) is small.
Ir F(x) denotes the internal shearing force at A.x>it can be seen that
F(x) dx = dM(x); that is (Prob. 14, Seco39), F(x) = M'(x),or
(2) F(x) = ElY"'(x).
Let W(x) represent the transverse load per unit length along the beam.
Then W(x) dx = dF(x), and it follows from Eq. (2) that
(3) y(4)(X) = aW(x) (a-1 = El).
r l
I Wo I
I I.
0.----------------------
I
.."
., x-e x-2e
y
Fig.34
The shear F(x) is a continuous function except at points where con-

centrated loads or supports acto WhenW(x) is sectionally continuous then,
in view of Eqs. (2) and (3), Y"'(x) is continuous and y(4)(X) is sectionally
continuous. These are just the continuity conditions that are implied by
our formula for the transform, with respect to x, of y(4)(X).
Although the simple differential Eq. (3) can be solved by successive
integrations, the requirement that the solution Y(x) and its derivatives of
the first three orders be continuous at all points often involves tedious labor
when W(x) is sectionally continuous. The simplicity of Eq. (3) enables us to
use the Laplace transformation even though that transformation is not
adapted to the two-point boundary conditions which will accompany that
equation. Certain Fourier transformations specified by the boundary con-
ditions and the differential form d4Y/dx4 are properly adapted to such
problems (Chap. ll).
Let us determine the displacements Y(x) in a beam whose end x = Ois
built into a rigid support while the end x = 2e is free, or unsupported
(Fig. 34). The load W(x) per unit length is zero over the span O < x < e
and a constant Woover the span e < x < 2e. The total load Wo is woe.
Equation (3) can now be written
(4) y(4)(X) = aWoSe(x) (O< x < 2e),
where Se(x) is our unit step function. The end conditions are
(5) Y(O) = Y'(O) = O, Y"(2e) = Y"'(2e) = O,
since no bending or shear acts on the end x = 2e..
Let y(s) denote the Laplace transform of Y(x) when Y(x) satisfies
Eq. (4) on the semi-infinite range x > O, together with the first pair of
boundary conditions (5). In view of the continuity conditions to be satisfied
by Y(x) and its derivatives, we may expect that
S4y(S) - S3(0) - S2(0) - sY"(O) - Y"'(O)= aw)e-es.

s
We have used a convenient extension, awoS.(x) when x > 2e, ofthe function
aW(x) in the right-hand member of Eq. (4) where O < x < 2e. The par-
ticular extension used is immaterial because we first seek a function Y(x)
that satisfiesEq. (4)when O < x < 2e and conditions Y(O)= Y'(O)= O,and
contains two arbitrary constants A = Y"(O)and B = Y"'(O).Since
A B 1 -es
y (S) = 3"
S
+ "4
S
+ awose
S
,
a function that satisfies those requirements is

(6) Y(x) = tAx2 + iBx3 + l4awO(x- e)4Se(x).
When X> c, it follows from Eq. (6) that
Y"(x) = A + Bx + tawo(x - C)2, Y"'(x) = B + awo(x - c).

The last two of conditions (5) are then satisfied if
(7) B = -awoc,
and the displacements in the beam are given by the equation
(8) Y(x) = aWo(ic2x2 - !CX3 + f.rtx - C)4S.(X)] (O;;;; X ;;;; 2c).
This function satisfies Eqs. (4) and (5) and the continuity conditions.
Since Y"'(O)= B = -awoc, it follows from Eq. (2) that the shear at
-
x = O has the value - woc,or F(O)= Wo as we should expect, since the
magnitude of the vertical force exerted by the support must be the same as
the totalload on the beam. Actually, the displacements Y(x) for this can-
tilever beam can be found by first noting that the shear - Y"'(x)/a at each
section is the totalload on the span to the right of that section.
The bending moment at the end x = Ois Y"(O)/aor Ala:
M(O)= ~cWo (Wo= cWo)'

If the end x = 2c were pin-supported (hinged or simply supported) so
that it can rotate freely about a fixedaxis, then the conditions at that end
become
Y(2c) = Y"(2c)= O.
37. THE TAUTOCHRONE

We shall now discuss a problem in mechanics that leads to a simple integral
equation of the convolution type.
The problem is that of determining a curve through the origin in a
vertical xy plane such that the time required for a particle to slide down the
curve to the origin is independent of the starting position. The particle
slidesfreelyfrom rest under the action of its weight and the reaction of the
curve on which it is constrained to move. The required curve is called the
tautochrone.
Let a denote the length of arc of the curve, measured from the origin O,
and let (x,y) be the starting point and (~,tI)any intermediate point (Fig. 35).
Equating the gain in kinetic energy to the loss of potential energy, we have
I da 2
2m( dt ) = mg(y ti), -
where m is the mass of the particle and t is time. Thus
da = - J2gJY='"iidt,
o %
Fig.35
and upon separating variables and integrating from r = y to r = 0,we have

"=y da
TJ2i= S,,=0.j y - r,
where T is the fixed time of descent. Now
a = H(r),
where the function H(r)depends upon the curve, and therefore
(1)
T J2i = I (y - r)-tH'(r)dr.
This is an integral equation of convolution type in the unknown function
H'(y). We may write it in the form
T fii = y-t * H'(y).
Let h(s)be the Laplace transform of H(y) with respect to the variable y.
Since H(O) = O,it followsformallyfrom Eq. (1)that
T fii~ = Sh(S)L{h} = Sh(S)~.

Thatis,
sh(s)= Tf! ~;
hence
(2) H'(y) = ~fii

1t
~.
VY
We can see that this function does satisfy our integral equation (1) by sub-
stituting it into that equation and performing the integration.
dX
Since H'(y) = da =
dy ) 1+
( dy )
2,
the differential equation of the curve in terms of the variables x and y is,
according to Eq. (2),
dx
1+ 2 = 2gT2 =~,
( dy ) n2y y
where a = 2gT2jn2. Separating variables here, we have

~
dx = V---ydY,
and the necessary integration can be performed easily by substituting
y = a sin2 !O, for we then find that
1 a
dx = a COS2 20 dO = 2(1 + cos O) dO.
Noting that x = Owhen y = O,we see that the parametric equations of the
tautochrone are therefore
a
(3) x = ~(O + sinO), y = 2(1 - cos O).
These equations represent the cycloid generated by a point P on a

circle of radius !a as the circle rolls along the lower side of the line y = a.
The parameter Ois the angle through which the radius drawn to the point P
has tumed, where the initial position of P is at the origino Qur tautochrone
is of course just one arch of this cycloid. Since a = 2gT2jn2, the diameter
of the generating circle is determined by the time T of descent.
The above problem can be generalized in various ways so as to lead to
other interesting questions; in fact, it was a generalization of the problem
of the tautochrone that led the great Norwegian mathematician Niels Abel
(1802-1829) to introduce the subject of integral equations.1
Ir the time T of descent is a function F(y), for example, our integral
equation (1) becomes
(4)
j2gF(Y) = I: (y - 17)-tH'(17)d17.
38. SERVOMECHANISMS
Simple integral equations as well as differential equations arise in the theory
of automatic control. As a special case we consider servomechanisms that
force the angle of tum 0o(t) of a rotating shaft to follow closely the angle of
tum 0(t) of a pointer or indicator, where t denotes time. The shaft and
1 See Bcher, M., "Integral Equations," p. 6, 1909.
material rigidity attached to it have a total moment of inertia 1 that is much

greater than that of the pointer. Mechanisms of that general type are used,
for instance, in directing antiaircraft guns and in aircraft control.
Let (t)
be the angle of deviation between shaft and pointer or the
difference between output angle and input angle:
(1) (t)= 00(t) - 0(t).
An auxiliary system or servomechanism can be designed to measure (t)
and feed back to the shaft a component of torque that is proportional to
the deviation <l>(t).The servomechanism may contain its own source of
power, motors, generators, and other electrical equipment. In order to
provide damping in the system, let the servo also supply a component of
torque proportional to the rate of deviation '(t).Then, since the product of
1 by the angular acceleration of the shaft equals the torque applied to the
shaft,
(2) -
10~(t) = k<l>(t) - e<l>'(t),
where k and e are positive constants.
Ir the shaft is initially at rest and its angle of tum is measured from the
initial position, then 00(0) = 0~(0) = O. AIso, in view of Eq. (1), <1>(0) =
. - 0(0) and in terms of transforms Eq. (2) becomes
1s280(s) = -(k + es)ljJ(s) - e0(0),
where we have assumed that 00(t), 0~(t), and (t), and therefore 0(t), are
continuous when t ;E; O. Since80(s)= ljJ(s) + 8{s),it followsthat
(3) ljJ(s) = - 1s2(J(s) + e0(0).

1S2 + es + k
Under an input 0(t) = At, it followsfrom Eq. (3)that
A. e k e2
(t) = - ~ e-bt smwt ( b = 21'W2 =1- 412) .
Ir k > c2/(4I), the angle of deviation (t) has a damped oscillation with
initial value zero. Since Isin wt/(wt)1 < 1 and (be)-l is the maximum value
of te-bt, that oscillation is small at all times when b is large, because
I(t)1< IAlte-bt ~ I~.
Ir in addition to the two components of torque shown in Eq. (2), a

component proportional to the accumulated angle of deviation is produced
by the servo, then
(4) 10~(t) = - k<l>(t)

- c<l>'(t)
- b {(-r)d-r,
where b is a positive constant. When E>0(0)= E>~(O) = Otherefore,
1S2[ <jJ(s) + 8;(s)]= - (k+ es + ~) <jJ(s) - eE>;(O).

In terms of positive numbers B, C, and K, where
lC = e, lK2 = k, lB3 = b,
the last equation can be written
(5) <jJ(s) -- - s3
s38;(s) + CsE>;(O)
+ Cs2 + K2S + B3'
In the special case E>;(t)= At, K2 = BC, Eq. (5) becomes
As
(6) <jJ(s) =-
and since the )?olynomials in the numerator and denominator here are of
degrees one and three in s, respectively, it folIows from our earlier observa-
tions (see the problems, Seco27) that c1I(O)= O; that is, the initial value ofthe
angle of deviation is zero. When C > B, all values of s that make the poIy-
nomial in the denominator vanish ha ve negative real parts, and con-
sequently (Sec. 26) c1I(t)contains only damped components. But when
C < B, c1I(t)has an unstable oscillation.
39. MORTALlTYOF EQUIPMENT

Let the function F(t) denote the number of pieces of equipment on hand at
time t, where the number is large enough that we can consider it as a con-
tinuous variable instead of a variable that takes on only integral values.
The equipment wears out in time, or is lost from service for other reasons,
so that, out of no pieces of new equipment introduced at time t = O,the
number N(t) in service at time t is given by the fonnula
(1) N(t) = noH(t),
where H(t) is a function that determines the surviving equipment after
t units of time. Note that H(O) = 1,necessarily.
If R(r) is the total number of replacements up to time 't',then R'('t')d't'
is the number of replacements during the time interval from t = to 't'
t = 't' + d't'and the number of survivals at any future time t, out of these
replacements, is
R'('t')H(t - 't')d't'.
The total amount of equipment in service at time t is the sum of these
survivals from the replacements during every time interval d't' between
T = O and T = t, increased of course by the survivals from the new equipment

on hand at time t = O. Therefore
(2)
F(t) = F(O)H(t) + { R'(T)H(t - T) dT.
We have assumed here that the equipment F(O)on hand at time t = Ois all
new; then R(O) = O.
Ir the amount F(t) that must be in service at each instant is known and
if the survival factor H(t) is known, then Eq. (2) is an integral equation of
convolution type in R'(t). Its solution gives the formula by which replace-
ments must be made.
The equation is an integral equation in the survival factor H(t) when
F(t) and R(t) are known.
In either case, the transformed equation is
(3) . f(s) = F(O)h(s) + s r(s)h(s).
Then
(4) ()
- f(s) - F(O)h(s)
rs - s h(s) ,
and R(t) is the inversetransform of that function.
Suppose the mortality is exponential in character so that
H(t) = e-kt
and that the amount of equipment on hand is to be a constant,
F(t) = b.
Then h(s)= l/(s + k) and f(s) = bIs,and it followsfrom Eq. (4)that
1
r(s) = bkz.
s
Therefore replacements must be made at such arate that the total equipment
replaced up to time t is
R(t) = bkt,
a result that is easily verified as the solution of Eq. (2). Thus replacements
must be made at the rate of bk pieces per unit time.
PROBLEMS
1. In the example solved in Seco 36 let the end x = 2c of the beam be pin-supported,
rather than free, with no other changes in conditions. Find the vertical force exerted
by the beam on the pin, and the vertical force and bending moment exerted on the
supportatx = O. Ans. *wo;HWo;ncWo'
2. Both ends x = Oand x = 2e of a beam are pin-supported. Find the vertical force
on each support when a transverse load Wo is distributed uniformly over the span
e < x < 2e. Ans. Wo/4; 3Wo/4.
3. Solve Probo 2 if both ends are built in rather than pin-supported.
Ans. f,Wo;tiWo'
4. The end x = O of a beam is built in and the end x = 2e is pin-supported. The
load per unit length is bx on the span O < x < e, and b(2e - x) on the span e < x < 2e.
Find the vertical force on each support in terms of the total load Wo on the beain.
Ans. ~~Wo;gwo.
5. In addition to a distributed load W(x) a single concentrated load W acts between
the ends of a beam, at position x = b. Then the shear F(x) is continuous except for a
jump W at x = b. Apply formula (4), Seco4, to the function Y"'(x) to show that the
transform of Eq. (3),Seco36, is
S4y(S) - s3y(0) - s2y'(0) - sY"(O) - Y"'(O) - aWe-bs= aw(s).
.Compare this equation in y(s) with the one obtained formally by replacing W(x) in
Eq. (3), Seco36, by W(x) + Wt5(x - b) and transforming as if Y"'(x) were continuous,
where t5(x) is the unit impulse symbol.
6. In Seco37, let the time T of descent be proportional to.JY, T.j2g = 2BJY,
whereB > 1. Showthat the curveof descentis the linex = yJB2=l.
7. Ire = k = 21 in Eq. (2), Seco38, find the output angle 80(t) corresponding to the
constant input angle 8{t) = l(t > O), and compare them graphically. Assume that
80(0) = 80(0) = O. AIso, note that the value of the output lags behind that of the
input until t = 3n/4. Ans. 80(t) = 1 - J2e-t sin(t + n/4).
8. For the servomechanism corresponding to Eqs. (4) and (5), Seco38, consider this
special case: 8{t) = At, B3 = CK2. Ir a represents the argument of the complex
number K + Ci, derive the formula
CI>(t)= 2 ~ rf2[Ce-Ct - JC2 + K2sin(Kt + a)]
for the angle of deviation, and note the undamped component of Cl>(t).
9. Let the servomechanism discussed in Seco 38 supply only a corrective torque
proportional to the deviation angle CI>(t)while the shaft itself is subject to a damping
torque proportional to its angular velocity. Ir 80(0) = 80(0) = O,show that
k 8Js),
80(s) = rs2 + es + 'o
where e, k, and 1 are positive constants. When 8;(t) = A, showthat Bo(t)approaches

A as t increases.
10. When H(t) = e-Ict,where H(t) is the survival factor in Seco39 and k is a positive
constant, (a) find the replacement function R(t) corresponding to an arbitrary amount
F(t) of equipment on hand; (b) find R(t) when F(t) = At + B.
Ans. (b) R(t) = (A + Bk)t + !Akt2.
-
11. When H(t) in Seco39 is the step function l Sk(t) so that every piece of new
equipment survives for k units of time, (a) show that the number of replacements R(t)
up to time t required to maintain F(t) pieces in serviee at that time is

R(t) = F(t) - F( +0) + F(t - k) + F(t - 2k) + F(t - 3k) + ...
if we define F(t) to be zero when t < O. (b) When F(t) = A(t > O),show that R(t) =
A[t/k], where [t] is the bracket symbol, and draw the graph of R(t).
12. A particle mass m moves on a vertical X axis under two forees: the force of gravity
and a resistance proportional to the velocity. If the axis is taken positive downward,
the equation of motion is
mX"(t) = mg - kX'(t).
Show that its solution, under the conditions X(O) = O,X'(O)= vo, is
where b = k/m, and discuss the motion.

13. Each one of a set of radioactive elements El, E2' E3' and E4 disintegrates into the
succeeding one at arate proportional to the number of atoms present, exeept for the
end product E4 which is a stable elemento If N(t), where i = 1, 2, 3, 4, denotes the
number of atoms of element E present at time t and if the distinct positive constants
CI, C2' and C3are the respective coefficients of decay of the first three elements, then
N~(t)= -cINI(t), N(t) = -c2N2(t) + cINI(t),
N3(t)= - c3N3(t)+ c2N2(t), N~(t)= c3N3(t).
When onlyM atoms of El are presentinitially,derivethe formula
N4(t) =1- c2c3e-Qt - clc3e-c2t - clc2e-c3t

M (C2- CI)(C3- CI) (CI- C2)(C3- C2) (CI- C3)(C2- C3r
14. In Seco36 the shear F(x) is the internal force in the direction of the Yaxis exerted
at a cross section Ax of the beam upon the span to the right of that section, and M(x)
is the bending moment exerted on that span there. If F(O)= Fo, M(O)= Mo, and
W(x) is the load per unit length, apply equilibrium conditions to a span extending x
units to the right ofthe section Ao, at x = O,to show that
Fo + I: W(~) d~ - F(x) = O,
Mo + Fox + J: (x - ~)W(~)d~ - M(x) = O.

Thus derive these relations between shear, load, and bending moment:
F'(x) = W(x), M'(x) = F(x).
4
Problems in Partial Differential
Equations
40 THE WAVE EQUATION

Several functions in physics and engineering satisfy the partial differential
equation
(1)
known as the wave equation in two independent variables x and t. We shall

use literal subscripts to indicate partial derivatives; then Eq. (1) can be
written
Brief derivations of this equation for some elementary physical functions will
now be given. The derivations are helpful in writing modifications of the
equation and in setting up boundary conditions for specific problems.
123
y
...-..-
::::-.I
-H I
I
I
Y I
I
I
" I
o
Fig.36
First, let Y(x,t) denote the displacement at time t away from the x axis
of a point (x, Y) of a string in the x Y plane under tension P (Fig. 36). The
string is assumed to be flexible enough that all bending moments transmitted
between its elements can be neglected; thus each element pulls tangentially
on an adjacent element with a force of magnitude P. Suppose further than
conditions are such that the magnitude H of the x component of the tension
remains essentially constant at all times for all points (x, Y); in particular all
displacements Y(x,t) are assumed small compared with the length of the
string. Finally let the slope angle IXremain small in order that each small
element of length of the string may be approximated by the length of its
projection Llx on the x axis. All these idealizing assumptions are satisfied,
for instance, by strings of musical instruments under ordinary conditions of
operation.
The vertical component of tension is the vertical force Vexerted by the
part ofthe string to the left ofpoint (x, Y)on the part to the right ofthe point.
It is proportional to the slope of the string, - VI H = tan IX(Fig. 36); that is,
(2) V(x,t) = - HY,Jx,t).

This is the basic formula for deriving the equation of motion.
Now consider an element oflength ofthe string whose projection on the
x axis is Llx. If p denotes the mass per unit length, the mass of the element is
approximately p Llx. If no external forces act on the string, the application of
Newton's second law to the element gives, in view of formula (2),
(3) pLlxY,,(x,t) = -Hy"(x,t) + HY,,(x + Llx,t)
approximately, when Llx is small; that is,
y.II(x,t) = H y,,(x + Llx,

A
t) - y"(x,t)
.
P l.lX
When we let Llx approach zero, this becomes Eq. (1), where
H
(4) a2 =-.
p
PROBLEMS IN PARTlAL DIFFERENTIAL EQUATIONS [SECo 40 125
If in addition to the internal force a force F(x,t) per unit of massacts in

the Y direction along the string, then the additional term p L\x F(x,t) appears
on the right in Eq. (3). The resulting modification ofEq. (1) is
(5) Y,r(x,t) = a2Yxx(x,t) + F(x,t).
When the Yaxis points vertically upward and the weight of the string is to be
taken into account, for instance, F(x,t) = - g, where g is the acceleration of
gravity.
As another physical example consider the longitudinal displacements
in a cylindrical or prismatic elastic bar. The values of the variable x are
marked on the bar so as to designate the cross section that is x units from
one end when the bar is neither stretched nor compressed (Fig. 37). For each
value of x the longitudinal displacement Y(x,t) is measured from a fixed
origin outside the bar, an origin in the plane occupied by the cross section at
x when the bar is unstrained and in some position of reference. Thus, if the
bar is moved lengthwise as a rigid body, Y(x,t) is a constant at each time t.
Since Y(x + L\x, t) is the displacement of the cross section at x + L\x,
an element of the bar whose naturallength is L\x is stretched by the amount
Y(x + L\x, t) - Y(x,t) at time t. According to Hooke's law the force exerted
by the bar upon the left-hand end ofthe element to produc~ that extension is
- AE Y(x + L\x,t) - Y(x,t).
whereA is the area ofthe crosssectionand E is Young's modulus of elasticity

of the material, the modulus in tension and compression. When L\x tends to
zero, it follows that the internal force from left to right at the cross section is
(6) F(x,t) = -AE~(x,t).
This basic formula corresponds to Eq. (2) for the string.

Let p denote the mass of the material per unit volume. When we apply
Newton's second law to an element ofthe bar,
(7) pA L\xY,ix,t) = -AEYx(x,t) + AEYx(x + L\x, t),
O
I
"
j "
Fig.37
o G)---%---t~ ~
'-y
Fig.38
we find as before that Y(x,t) satisfiesEq. (1),where

E
(8) a2 =-.
p
When the elastic bar is replaced by a column of air, Eq. (1) has further
applications in acoustics.1
Again, the function Y(x,t) may represent the angle of turn of a cross
section x units from one end of an elastic cylindrical shaft under torsion
(Fig. 38). Let 1 denote the moment of inertia of the cross section with respect
to its axis, Es the modulus of elasticity of the material in shear, and p the
density of the material. Then, by steps analogous to those used above, we
find that the internal torque -r acting through a cross section at position x is
(9) -r(x,t) = - lEs Yix,t)
and that Eq. (1) is satisfied by the angle Y(x,t), where
(10) a2 =-.Es
p
Finally, it should be remarked that Eq. (1) is a special case of the
telegraph equation
(11) Yxx(x,t) = KLY,,(x,t) + (RK + SL)Y,(x,t) + RSY(x,t),
where Y(x,t) represents either the electric potential or the current at time t at
a point x units from one end of a transmission line or cable.2 Here the
elements of resistance, inductance, etc., are distributed along the cable or
through the circuit, in contrast to the lumped elements in elementary circuits
(Sec. 33) that lead to ordinary differential equations in the currents. The
cable has resistance R, electrostatic capacity K, leakage conductance S, and
self-inductance L, all per unit length. When R and S are so small that their
effect can be neglected, Eq. (11) reduces to Eq. (1), where a2 = (KL)-l.
41 DISPLACEMENTS IN A LONG STRING

Let Y(x,t) represent the transverse displacements of the points of a semi-
infinite stretched string, a string having one end fixed so far out on the x axis
1 Lord Rayleigh. "Theory of Sound," vols. I and 2, Dover, 1945.
2A derivation of Eq. (1I) is outlined in Probo 6, Seco93.
PROBLEMS IN PARTlAL DIFFERENTIAL EQUATIONS [SECo 41 127
that the end may be considered infinitely far from the origin, and having its
other end looped around the Y axis. The loop, initially at the origin, is
moved in some prescribed manner along the Yaxis (Fig. 39) so that Y = F(t)
when x = O and t ~ O, where F(t) is a prescribed continuous function and
F(O) = O. Ir the string is initiallyat rest on the x axis, let us find the formula
for Y(x,t).
The above conditions on Y(x,t) can be written
(1) Yrr(x,t)= a2 Yxx(x,t) (x > O,t > O),
(2) Y(x,O) = Yr(x,O)= O (x > O),
(3) Y(O,t)= F(t), lim Y(x,t) = O
x->oo
(t ~ O),
where a2 = H/p (Sec.40). The equation ofmotion (1) implies that no external
forces act along the string.
A problem composed of such conditions is called a boundary value
problem in partial differential equations. We shall use a formal procedure to
solve the problem and then show how our result can be verified as a solution.
Ir y(x,s) is the Laplace transform of Y(x,t), then, in view of the initial
conditions (2), L{ Yrr} = s2y. AIso,
if the function e-srY(x,t) satisfies conditions under which the indicated

interchange of order of integration with respect to t and differentiation with
respect to x is valid (Sec. 15). When both members of the partial differential
equation (1) are transformed and conditions (2) are used, we therefore obtain
the equation s2y = a2yxx in the transform of our unknown function.
From conditions (3) we find that y(O,s)= I(s), where I(s) is the trans-
form of F(t), and limx->00y(x,s) = O, provided that the order of integrating
with respect to t and taking the limit as x --+oc>can be interchanged. The
Ji1t)
o
Fig.39
128 SECo 41] . OPERATlONAL MATHEMATICS
transformed boundary value problem in y(x,s) is therefore

d2y S2
(4) -dX2 - -y =O (x > O),
. a2
(5) y(O,s) = f(s), lim y(x,s) = O.
x-oc
Here we have used the symbol for ordinary rather than partial differentiation
because s is only a parameter in the problem ; no differentiation with respect
to s is involved.
A convenient form of the general solution of Eq. (4) is
y(x,s) = Cle-sx/a + C2e'x/a,
where CI and C2 may be functions of s. This solution could of course be
obtained by transforming the members of Eq. (4) with respect to x. We
consider s as positive, since Laplace transforms generally exist for all s
greater than some fixed number. Then C2 = O,if y(x,s) is to approach zero
as x tends to infinity. The first of conditions (5) is also satisfied if C 1 = f(s),
and the solution of the transformed problem is
(6) y(x,s) = e-(x/a)s(s).
The translation property (Sec. 11) enables us to write the inverse trans-
form of y(x,s) at once:
(7) when t > ~

Y(x,t) = F( t - ~) = a'
=0 when t ~~.a
Since F(t) is continuous and F(O) = O,we can see that the function Y(x,t)
described by formula (7) is continuous when x ~ O and t ~ O, including
points on the line x = at in the xt planeo The function clearly satisfies the
boundary conditions (2) and (3).
Any function of t - x/a is easily seen to be a solution of the wave
equation (1) when its derivative of the second order exists. Let F(t) satisfy
these additional conditions: F'(t) and F"(t) are continuous when t ~ O
except possibly for finite jumps at t = ti (i = 1, 2, . . .). The function Y(x,t)
described by formula (7) then satisfies Eq. (1) except possibly at points (x,t)
on the lines t - x/a = ti and t - x/a = O in the quadrant x > O,t > O. In
this sense, then, the function (7) is verified as a solution of our boundary
value problem regardless of the validity of formal steps taken to obtain that
function. In case F"(t) is continuous whenever t ~ O, and F(O)= F'(O)=
F"(O)= Oso that Y(x,t) and its partial derivatives of first and second order
are also continuous on the line t = x/a, then the function (7)satisfiesEq. (1)
PROBLEMS IN PARTlAL DIFFERENTIAL EQUATIONS [SEC. 41 129
with no exceptions when x > O and t > O and represents a solution of the
boundary value problem in the ordinary sense. Note that conditions of
exponential order are not involved in the verification of our solution.
According to formula (7), a point of the string x units from the origin
remains at rest until the time t = x/a. Starting at that time, it executesthe
same motion as the loop at the left-hand end. The retarding time x/a is the
time taken by a disturbance to travel the distance x with velocity a. Since
a = (H/p)t, note that it has the physical dimensions ofvelocity. The vertical
force from left to right at a point (Sec. 40) is -HYix,t); thus the vertical
force exerted on the loop to make it move in the prescribed manner, as
found from formula (7), is
H
V(O,t)= -F'(t)
a = JPijF'(t).
Finally, let us observe instantaneous positions ofthe string correspond-

ing to the loop movement
(8) F(t) = sin nt when O ~ t ~ 1,
=0 when t ~ 1;
thus the loop is lifted to the position Y = 1 and returned to Y = O, where it
remains after time t = 1. In this case F(t - x/a) = Owhen t - x/a ~ 1,that
is, when x ~ a(t - 1).AIsoF(t - x/a) = sin n(t - x/a) when O ~ t - x/a ~
1,that is, when a(t - 1) ~ x ~ atoThenfromformula(7)wefindthat
(9) Y(x,t) = O when x ~ a(t - 1)or when x ~ at,
= sin n(t - x/a) when a(t - 1) ~ x ~ ato
Thus the string coincides with the x axis except on an interval of length a,
if t > 1, where it forms one arch of a sine curve ending at x = ato As t
increases, the arch moves to the right with velocity a (Fig. 40).
It is interesting to note that for each fixed x in formula (9) the function
Yr(x,t)has a jump n at t = x/a and at t = 1 + x/a and hence s2y(X,S)is not
the transform of Yrt(x,t). On the other hand it can be shown that d2y/dx2 is
not the transform of y"'ix,t), because of the discontinuities of Yx(x,t); but the
transform of Yrt - a2 Yxx is zero.
(t>l)
"
Th"'b
O alt-l)-O
Fig.4O
The function Y(X,t) above can also represent the longitudinal dis-
placements in a semi-infinite elastic bar (Sec. 40), initially at rest and un-
strained. The distant end of the bar is held fixed and the end x = O is
displaced in a prescribed manner, Y(O,t) = F(t). When F(t) is given by
formula (8), the solution (9) or Fig. 40 shows that at time t the section of the
bar fram x = a(t - 1) to x = at is strained while the remainder is unstrained
and at res.
42 A LONG STRING UNDER ITS WEIGHT

Let the semi-infinite string be stretched along the positive half of a hori-
zontal x axis with its end x = O fastened at the origin and with its distant
end looped around a vertical support that exerts no vertical force on the
loop (Fig. 41). In view of formula (2), Seco40, for the vertical force at points
of the stretched string, Yx(x,t) vanishes at the distant end. The string is
initially supported at rest along the x axis. At the instant t = O the support
is removed and the string moves downward under the action of gravity. Let
us find the displacements Y(x,t).
As noted in Seco40, Eq. (5), the equation of motion is
(1) y,/(x,t) = a2 Yxx(x,t) - g (x > O,t > O).
The boundary conditions are

(2) Y(x,O) = Y,(x,O) = O (x ~ O),
(3) Y(O,t)= O, lim Yx(x,t) = O (t ~ O).
x'" 00
The transformed problem is found formally to be
(4)
(5) y(O,s) = O, lim y'(x,s) = O,

x'" 00
~
,
I
I
I % I
O I I
I
{
lat.-'lgt2) I
Fig.41
PROBLEMS IN PARTIAL DIFFERENTIAL EQUATIONS [SEC. 42 131
where the primes denote derivatives with respect to x. Since the constant
- g/S3is a particular solution of Eq. (4), the general solution ofthat equation
is
When conditions (5) are applied, we find that
(6) y(x,s) = _gL13 - :3e-<X/O)S).

The displacements Y(x,t) can be written in the form
g 2
(7) when x ~ at,
Y(X,t) = - 2a2(2axt - x )
when x ~ at.
The details of this step and the verification of the solution (7) are left to the
problems. An instantaneous position of the string is shown in Fig. 41. We
note that up to time t the segment of the string to the right of the point x = at
has moved like a freely falling body.
PROBLEMS
Solve these boundary value problems and verify your results:

1. YAx,t) + x Y;(x,t)= O; Y(x,O) = O, Y(O,t)= t.
Ans. Y = Oif t ;;;;!X2, y = t - !X2 if t ~ !X2.
2. Yx(x,t)+ 2x Y;(x,t)= 2x; Y(x,O) = Y(O,t)= 1.
Ans. Y = 1 + t if t ;;;;x2, y = 1 + x2 if t ~ x2.
3. xYx(x,t) + Y;(x,t)+ Y(x,t) = xF(t); Y(x,O) = Y(O,t)= O.
4. Wxx(x,t) + W,Ax,t) - 2 W,,(x,t) = O(x > O,t > O);

W(x,O) = W,(x,O)= limx_ooW(x,t) = O, W(O,t)= F(t).
Ans. W(x,t) = F(t - 2x), where F(r) = O if r < o.
5. ~Ax,t) - 2W,x(x,t) + W,,(x,t) = O(O < x < 1, t > O);
W(x,O) = Jv,(x,O)= W(O,t)= O; W(l,t) = F(t) (t > O).
Ans. W(x,t) = xF(x + t - 1),where F('r)= O if T < o.
6. In the problem on the semi-infinite string falling under its own weight (Sec. 42),
(a) give the details in deriving formula (7) for displacements Y(x,t) and verify the result
fully; (b) use formula (7) to verify that the vertical force at time t exerted on the string
by the support at the origin equals the weight gpat of the curved segment of the string.
7. In Seco42let the weight ofthe string be replaced by a general vertical force of F(t)
units per unit mass of string, so that another special case of Eq. (5), Seco40, is involved.
132. SECo 42] OPERATIONAL MATHEMATICS
Show that the displacements are
Y(x,t) = G(t) - G( t - ~)
where G(t) = J~J~ F(-r) d-rdr if t ~ O, G(t) = O if t ;;;O.
8. The end x = Oof a semi-infinite stretched string is looped around the Yaxis which
exerts no vertical force on the loop, and the distant end is fixed on the x axis. The string
is displaced into the position Y = e - X(x ~ O) and released from rest in that position
at the instant t = O. If no external forces act on the string, set up the boundary value
problem for the displacements Y(x,t) and find its solution in the form Y = e-a, cosh x
if x ;;;ato Y = e-X cosh at ifx ~ ato Suggestion: Find A(s) so that A(s)e-X is a particular
solution of the nonhomogeneous equation a2y" - s2y = -se-X in y(x,s).
9. The force per unit area on the end x = O of a uniform semi-infinite elastic bar
(Fig. 42) is F(t): - EYx(O,t)= F(t). Ifthe infinite end is fixed and the initial displacement
and velocity of each cross section are zero, set up the boundary value problem for
longitudinal displacements Y(x,t) and derive the solution
Y(x,t) = iG( t - ~)
where G(r) = J~ F(-r) d-r if r ~ O, G(r) = O if r ;;;O; thus Y(x,t) = O when t ;;;x/a. Note
that the displacement of the end x = Ois Y(O,t)= (a/E)G(t).
w - f3
-
~,
-r+-r--t ~~
~
iIfi
~~:::~ O
Fig.42
1
10. When the pressure F(t) in Probo 9 is a finite impulse: F(t) = EFo when t < to.
F(t) = Owhen t > to, show that Y = Owhen t ;;;x/a, y = (at - x)Fo when x/a ;;;t ;;;
to + x/a, and Y = Foato when t ~ to + x/a. Study this function Y(x,t) graphically.
11. Under the instantaneous impulse ofpressure F(t) = 1<5(t)in Probo 9 show formally
that
al
if x < at, Y(x,t) = O if x > ato
Y(x,t) = E
Note that in this hypothetical case a part of the bar is displaced into a region already
occupied by another parto
12. Let the pressure F in Probo 9 be constant, F(t) = Fo.
(a) Show formally that Ey(x,s) = aFos-2 exp( -sx/a) and hence
Y(x,t) = FoE-l(at - x)So(at- x).

Verify that result as a solution of the boundary value problem.
PROBLEMS IN PARTIAL DIFFERENTIAL EQUATIONS [SECo 43 133
(b) Draw graphs of Y(X,t) here as a function of x with t fixed, and as a function
of t with x fixed. Note the jumps in Yxand Y;where x = at and the removable discon-
tinuities in the functions Yxxand Y;,there.
13. For the function Y(x,t) and its transform y(x,s) found in Probo 12, show that
L{ Y;,}is zero, not S2y as assumed in the formal solution; also that L{ YxJ is zero, not
d2y/dx2. Note, however,that L{Y;, - a2yxx}= O= s2y- a2 d2y/dx2.
14. If a rigid mass m is attached to the end x = O of the semi-infinite bar in Probo 9
(Fig. 42), when the bar has unit cross-sectional area, and the force F(t) is a constant Fo,
show why the boundary condition at that end becomes
mY;,(O,t)= EYx(O,t)+ Fo.
Write b = E/am and derive the formula
x
Y(x,t) = -aFo xlI
E [ t - -a - -b + -e-
b
b(' Xla
- o t- - .
a)
Js (
15. An unstrained semi-infinite elastic bar is moving lengthwise with velocity - Vo
when, at the instant t = O, the end x = O is suddenly brought to rest, the other end
remaining free (Fig. 43). Set up and solve the boundary value problem for the longi-
tudinal displacements Y(x,t). AIso show that the force per unit area exerted by the
support at x = O upon the end of the bar is Evo/a.
Ans. Y = -vot when t ~ x/a, Y = -vox/a when t ~ x/a.
- Vo
.
,I
~ x
~
I
Ag.~
16. Let the end x = O of the bar in Probo 15 meet an elastic support at the instant
t = O,such that the pressureexertedby that support upon the end is proportional to
the displacement of that end: - EYAO,t) = - BY(O,t),or
:fx(O,t)= b Y(O,t) (b = B/E).
Show that the velocity ofthat end is -voe-ab' since abY(O,t) = -vo(1 - e-ab').
43 THE LONG STRING INITIALLY DISPLACED
Let the ends of a semi-infinite string stretched along the positive x axis be
kept fixed, and let the string be given some prescribed displacement Y = 1I(x)
initially a~d released from that position with initial velocity zero. Here
11(0) = 11(00) = O. Then th~ boundary value problem in the transverse
displacements Y(x,t) is
(x > O,t > O),.
(1) Y(x,O) = cI>(X), Y,(x,O) = O,
Y(O,t)= O, lim Y(X,t) = O.

x-+00
The problem in the transform y(x,s) is therefore

(2) s2y(X,S) - scI>(x) = a2yxx(x,s) (x > O),
(3) y(O,s)= O, lim y(x,s) = O,

x-+ 00
where yxx(x,s) = d2y/dx2. We shall solve the ordinary differential equation

(2) by using the Laplace transformation with respect to x. Let u(z,s) denote
that transform of y(x,s); that is,
u(z,s) = foooe-ZXy(x,s)dx.
Since y(O,s)= O,when we transform both members ofEq. (2), we obtain

the equation
where q>(z)is the transform of cI>(x).Let the unknown function of s,yx(O,s)be

denoted by C. Then the solution of the last equation can be written
C s 1
u(z,s) = Z2 - (s2/a2) - a2q>(Z) Z2 - (s2/a2)'
and performing the inverse transformation with respect to z, with the aid of
the convolution, we ha ve
(4) y(x,s) = aC sinh sx

s a
- ~ I
x cI>(~)sinh ~ (x
a o a
- ~)d~.
In view ofthe condition requiring y(x,s) to vanish as x tends to infinity, it

is necessary that the coefficient of x/a on the right of Eq. (4) should vanish
as x becomes infinite. Writing the hyperbolic sines in terms of exponential
functions, we find that coefficient to be
aC
2s
- ~
2ao I
x cI>(~)e-(s~/a) d~.
Since the limit of this function is to be zero as x -+ 00, we have
(5)
Substituting this into Eq. (4), we can write the result in the form
2ay(x,s) = LX' cI>(~)e-[s(~-xllla

d~ - {XJW(~)e-[s(x+~lIIa
d~
+ J: W(~)e-[S(x-~)]/a d~.
The integralshere can be reduced to Laplace integrals. In the first one

we substitute. = (~- x)/a, in the second. = (~+ x)/a, and in the third
. = (x - ~)/ato get
(6) 2y(x,s) = IJ:>W(x + a.)e-St d.

x'a
i
""
- f W(-x + a.)e-St d. + W(x - a.)e-St d..

x/a O
In order to combine the last two integrals, let W(x) represent the odd extension
of the function W(x):
(7) tPl(X) = W(x) ifx ~ O,
= -W( -x) ifx ;;;;O;
thus Wl( -x) = -Wl(X) for all x. Then Eq. (6) can be written
2y(x,s) = L{W(x + at)} + L{W(x - at)}
and therefore
(8) Y(x,t) = t[W(x + at) + Wl(X - at)].
It is easily seen that a twice-differentiable function of the variable
x + at, or of the variable x - at, as well as each linear combination of such
functions, satisfies the linear homogeneous wave equation Y" = a2yxx. Qur
continuous function (8)therefore satisfiesthe wave equation for all x and t
where W"(x + at) and Wl(x - at) existo It satisfies the condition Y,(x,O)= O
when x > Owherever W'(x)exists, and it satisfies the remaining conditions of
problem (1) because W(x) = W(x) when x > O, W1(-at) = -W(at) and
W(oo,t) = O. .
Instantaneous positions of the string can be sketched by adding
ordinates as suggested by formula (8). When t is fixed, the graph of !(x-
. at), for instance, is obtained by translating the graph of the function !w(x),
defined for all real x, to the right through the distance al.
44 A BAR WITH A PRESCRIBED FORCE ON ONE END

Let the end x = O of an elastic bar of length e be kept fixed, and let F(t)
denote a prescribed force per unit area acting parallel to the bar at the end
x = e (Fig. 44). Ir the bar is initially unstrained and at rest, the boundary
--+
o % %
Fig.44
value problem in the longitudinal displacements Y(x,t) is the following,

Y,,(x,t) = a2Yxx(x,t) (O< x < e, t > O),
Y(x,O) = Y,(x,O)= O,
Y(O,t)= O, EYx(e,t) = F(t),
The transform of Y(x,t) therefore satisfies .the conditions

s2y(X,S) = a2yxx(x,s),
.y(O,s)= O, EYx(e,s)= f(s),
and the solution of this transformed problem is readily found to be
a sinh (sx/a)
(1)
y(x,s) = Ef(s) s cosh (se/a)"
CONSTANT FORCE
Consider first the case
(2) F(t) = Fo.

Then
aFo sinh (sx/a)
(3)
y(x,s) = E S2 cosh (se/a)
aFo 1 se
and, when x = e, y( e ,S) = - E s"2tanh -.a
In Seco23 we found that S-2 tanh (bs/2) is the transform of the triangular
wave function H(b,t) of period 2b (Fig. 10). Therefore the displacement of
the end x = e is
aFo 2e
(4) Y(e,t) = EH ( ~,t ) ;
that is, the end moves byjerks as indicated in Fig. 45.
PROBLEMS IN PARTIAL DIFFERENTIAL EQUATlONS [SECo 44 137
~~
(0.2C;0)
Q (4cti ,01
~
Fig.45
To find the displacement of an arbitrary point using formula (3) we

may write, when s > O,
sinh (sx/a) e-[(c-x)sJ/a- e-[(c+x)sJ/a
cosh (se/a) = 1+ e-(2cs/a)
00
= [e-[(C-X)sJ/a- e-[(c+X)sJ/a] L (_1)ne-(2ncs/a),

n=O
00
since (1 + z)-1 =L
o
(- 1tzn when O < z < 1. Therefore
aFo 00 1 (2n+ l)e - X
y(x,s)=E~(-lt { s2exp -s[ a ]

- :2 exp [- s(2n + ~)e + x]}.
Formally applying the inverse transformation to the terms of the infinite
series,and using braces to denote this translation of the linear function -r:
(5) {-r - k} = (-r- k)So(-r- k),
we obtain the formula
F.
(6) Y(x,t)= ;({at - e + x} - {at - e - x} - {at - 3e + x}
+ {at - 3c - x} + {at - Se+ x} - {at - Se- x} - .. .).
For any fixed t the series (6) is finite since each ofthe braces is to be replaced
by zero when the quantity inside is negative. The number of nonvanishing
terms in the series increases as t increases.
The function (6) can be verified directly as the solution of our problem.
Its graph, for a fixed value of x, is shown in Fig. 46.
.Y(x.t)
3c-x
ti
k ti
Fig.46
A GENERAL FORCING FUNCTION
When the pressure on the end x = e is F(t), let us write
G(t) = {F(r) dr if t ~ O, G(t) = O if t ;;;O.

Then L{G} = J(s)/s, and formula (1) can be written
y(x,s) =!!:. (s)sinh (sx/a) o

E g cosh (se/a)
Thus if we write qn(x) = [(2n + l)e - xJ/a then, as in the above case,
a co
y(x,s) = E n~o (-lt{exp [-sqn(X)] - exp [-sqn( -x)]}g(s)
and formally,
a co
(7) Y(x,t) = - I
E n=O
(-lt{G[t - qn(x)J- G[t- qn(-x)]}o
AN IMPULSE OF PRESSURE
In the hypothetical case where

(8) F(t) = 1 ()(t),
J(s) = 1 in formula (1) and
(9) (x,s) = al sinh(sx/a) o
y E s cosh (se/a)
The displacements Y(x,t) can be represented by a seriesof the step functions
So{t - [(2n + l)e:!: xJ/a}. Those displacementsare not continuous.
To see the behavior of the end x = e, we note that
al 1 se
y(e,s)= -E -s tanh-.a
The displacement of that end is therefore represented by the square-wave

function (Seco23, Figo9),
al 2e
(10) Y(e,t) = E M ( a,t ) o
Thus the end jumps suddenly back and forth between two fixed posi-
tions. It is possible to demonstrate a close approximation to that behavior by
substituting for the bar a loosely wound coil spring. Ir, when the spring is
hanging from one end, the free lower end is given a sharp tap, the lower end
tends to move as indicated.
PROBLEMS
1. A string is stretched between two fixed points (0,0)and (c,O)oIf it is displaced into
the curve Y = b sin (nxjc) and released from rest in that position at time t = O,set up
and solve the boundary value problem for the displacements Y(x,t). Verify the result
fully and describe the motion of the string. Ans. Y(x,t) = b cos (natjc) sin (nxjc).
2. If the initial displacement of the string in Probo 1 is changed to
Y(x,O) = b sin nnx (O~ x ~ e),
e
where n is any integer. derive the formula
nnat o nnx
Y(x,t) = bcos-sm-.
e e
Note that the sum of two or more of these functions with different values of n and b
is a solution of the equation of motion that satisfies aIl the boundary conditions if the
initial displacement Y(x,O) corresponding to such a superposition of solutions is a
certain linear combination of the functions sin (nnxjc).
3. As a special case of the initiaIly displaced semi-infinite string considered in Seco43,
where Y(O,t)= Y(co,t) = O, let the end segment O~ x ~ 1 be plucked so that the
initial displacement has this form:
Y(x,O) = sin nx if O~ x ~ 1, Y(x,O)= O if x ~ 1,
where again Y,(x,O)= O when x > O. Show the displacement Y(x,t) graphically as a
function of x when t is fixed and at > 1, and note that the displacement is a wave
formed by letting one cycle of the curve Y = t sin nx, -1 < x < 1, move to the right
with velocity a. (Boys have used such plucking of a kite string to dislodge parachutes
hooked onto the string and blown toward the kite.)
4. A long string stretched along the entire x axis is released at rest from a prescribed
initial displacement. Thus
( - co < x < 00, t > O),
Y(x,O) = (x), Y,(x,O)= O (-00 < x < 00),
x-lim
- 00Y(x,t) = O, lim Y(x,t) = O
x-oo (t ~ O).
Assuming "(x)
exists for all x, derive the formula
Y(x,t) = t[(x+ at) + (x at)] -
and verify that solution.
5. A semi-infinite string stretched along the positive x axis with ends fixed,
Y(O,t)= limx_ooY(x,t) = O,is given an initial velocity Y,(x,O)= g(x), where g is section-
ally continuous on each bounded interval, limx_oog(x) = O,and g is integrable from
zerotoinfinity. (a)Ifg istheoddextensionofg,sothatg(r) = g(r)andg(-r) = -g(r)
if r > O,show that the transverse displacements can be written
Y(x,t) = 2
a
_f
1 x+ar
x-al
_1
g(r)dr = 2 [G(x+ at) - G(x - at)],
a
where G(r) = f~gl(P)dp [cf.Eqs. (2),(3),and (6),Seco43]. Verifythat resulto(b)Show

Y(x,t) graphically as a function ofx when at > 1in case g is the step function 1 - So(x - 1).
6. In Seco44, when the force per unit area at the end x = e of the bar (Fig. 44) is
constant, F(t) = Fo, and Y(x,O) = Y;(x,O)= Y(O,t)= O, show that the force per unit
area exerted by the bar on the support at the end x = Ois the function shown in Fig. 47.
Note that the force becomes twice the applied. force during regularly spaced intervals
of time.
EYz(O,t)
n r--
I
I
2Fo II I
I
o 3c
a
Fig.47
7. Let the constant force Fo in Probo 6 be replaced by a finite impulse of duration 4e/a:
. 4e . 4e
F(t) = Fo IfO < t<-, a F(t) = O If t>-.
a
(a) Show that the force per unit area exerted by the bar on the support at the end
x = Ois that shown in Fig. 47 up to the time t = 3e/a after which time the force is zero.
(b) Show that the end x = e moves with constant velocity Vo = aFo/E during
the time intervalO < t < 2e/a, then with velocity - Voduring an equal time interval
and that Y(e,t) = Owhen t ~ 4e/a.
8. A constant longitudinal force Fo per unit area is applied at the end x = e of an
elastic bar (Fig. 48) whose end x = Ois free and which is initially at rest and unstrained.
Set up the boundary value problem for the longitudinal displacements Y(x,t) and show
that Ey(x,s) = aFos- 2 cosh (sx/a)/sinh (se/a).
(a) Derive the formula
Y;(O,t)= Vo
.;0
f So[t - (2n + 1)~J a where Vo
2aFo.
= E'
hence note that the free end moves with velocity Vo from time t = e/a until t = 3e/a,
then with velocity 2vo until t = Se/a,etc.
- -
(b) In terms ofthe function {T k} = (r k)So(T k), show that -
F. 00
Y(x,t) = ; .~o [{at + x - e(2n+ 1)}+ {at - x - e(2n+ 1)}].
o e
Fig.48
9. In Probo 8, let the force Fo be replaced by the force
F(t) = Fo cos rot,
where ro = naJ(2e). Show that [formula (10), Seco23]
nEY(e,t) = 2eFoIsinrotl.
10. The end x = O of a bar or heavy coil spring (Fig. 49) is free. The end x = e is
displaced in a prescribed manner, Y(c,t) = G(t). If the bar is initially unstrained and
at rest, find the transform y(x,s) of longitudinal displacements.
J 01
Fig.49
(a) When G(t) = Fot, show that the displacement Y(O,t) of the free end is repre-
sented by the function obtained by integrating the function EYx(O;r)in Fig. 47 from Oto
t and show the displacement graphically.
(b) If G(t) = bt when t ;;4eJa and G(t) = 4beJa when t ;;4eJa, show that the
free end moves with a uniform velocity 2b to a new position and remains there, as shown
in Fig. 50.
Y(O,t)
Fig.50
11. An unstrained elastic bar is moving lengthwise with velocity Vo when at the
instant t = O its end x = e meets and adheres to a rigid support (Fig. 51).
(a) If A is the area of the cross section of the bar and M denotes the square-wave
function (Fig. 9), show that the force on the support can be written
AEvo 2e
-AEy"(e,t) = ---;-M ( -;i',t) (a2= ~).
- Vo
[
O 1-.
Fig.51
(b) In terms ofthe function {t - k} = (t - k)So(t - k), show that

1 - X
e e+ X 3e - X 3e + X
VoY(x,t) = t - {t - a- } - { t - a- } + {t - a- } + {t - a- } -. .. ,
and hence that Y(x,2e/a)= Oand Y,(x,2e/a)= -Vo for all X (O< X < e. Thus if the
end X = e is free to leave the support, the bar wiII move after time t = 2e/a as a rigid
unstrained body with velocity - Vo .
12. A steel bar 10 in. long is moving lengthwise with velocity Vowhen one end strikes a
rigid support squarely (Fig. 51). Find the length of time of contact of the end of the bar
with the support and note that the time is independent of vo' For steel take E to be
30 x 106Ib/in.2, and mass per unit volume p such that gp = 0.28Ib/in.3, where
g = 384 in./sec2. (See Probo 11.) Ans. 0.0001 seco
13. An unstrained cylindrical shaft is rotating with angular velocity Jwhen it ends
x = :t e are suddenlycIamped(Fig.52). Derivethe formula
8(x s) = ~ 1
, S2 [- cosh (sx/a)
cosh (se/a)]
for the transform of the angular displacements 0(x,t) of the cross sections, where
a2 = EJ p (Sec. 40).
(a) Ir H denotes the triangular-wave function (Fig. 10), show that the displace-
ment of the middle cross section is
0(0,t) = Jt ift ~ ~
- a'
Je 2e e . e
0(0,t) = -a - JH-,t - -a ft f;-.a
( a )
(b) Describea stretched-stringanalogfor 0(x,t) here,as seenfromthe boundary

valueproblem.
- "
( -e
. r I,
O
+ ~-
e
Fig.52
14. Let M denote the square-wave function (Fig. 9). Show that the torque acting on
the support at the end x = e of the shaft in Probo 13 is a-1 E.IJM(2e/a, t).
15. The end x = Oof a cylindrical shaft is kept fixed. The end x = e is rotated through
an angle 80 and, when all parts have come to rest, that end is released; thus 0(x,0) =
8ox/e (Fig. 53). Show that the angular displacement of the free end at each instant can
be written
a 2e
0(e,t) = 80 - 80cH ( -;' t) ,
where H is the triangular-wave function (Fig. 10).

of-
Fig.53
~ e
45 EQUATIONS OF DIFFUSION
Let U(x,y,z,t) denote the temperature at a point (x,y,z) within a solid at time t.
Let S be aplane or curved surface passed through that point and let the
coordinate n represent directed distance along the line normal to S at the
point. Then the flux of heat, the quantity of heat per unit area per unit time
transferred across S at the point by conduction, is given by the formula
d
(1)
<II(x,y,z,t)= - K dn U(x,y,z,t),
where the coefficient K is the thermal conductivity of the material. The

negative sign that appears with this directional derivative of U gives the
sense of flow, where positive flux signifies flow in the positive direction of the
normal.
Formula (1) is an empiricallaw or basic postulate for the conduction
of heat in solids. It is likewise a basic postulate for the flux of a substance
undergoing simple diffusion into a porous solid if U represents the concentra-
tion of the diffusing substance and K the coefficient 01 diffusion. Ir the
temperature or concentration is a function U(x,t) of x and t only, then the
flux in the x direction is
(2) <II(x,t)= -KUix,t).
When the temperature function has the simple form U(x,t), consider
a portion of the solid in the shape of a prism parallel to the x axis (Fig. 54).
Since there is no temperature variation with respect to distance normal to
the lateral surface of the prism, no heat is conducted across that surface.
According to formula (2), an element of the prism extending from position
Fig.54
x to x + ~X receives heat by conduction through its two bases at the rate

-AKUx(x,t) + AKUx(x + ~x,t),
where A is the area of the cross section of the prism. This is the quantity
of heat gained by the element per unit time if no heat is generated or lost
inside the element. Another expression for that rate can be written in terms
of the coefficient e of heat eapacity per unit volume, the quantity of heat
required to raise the temperature of a unit volume of the material by one
degree. Then when ~x is small, eA ~x U(x,t) is a measure ofthe instantaneous
heat content of the element, and
(3) eA ~x U,(x,t) = A[KU x(x + ~x, t) - KU Ax,t)].
Ir K depends on x, then the first factor K in Eq. (3)is actually K(x + ~x, t)
and the second, K(x,t). When we divide by ~x and let ~x tend to zero, we
obtain the partial differential equation
(4) eU,(x,t) = [K(x,t)U Ax,t)]x'
When K is either constant or independent of x, the equation becomes
(5) U,(x,t) = kU xx(x,t),
where k = K/e and k is called the thermal diffusivity of the material. Equa-
tion (5) is the simple form of the heat equation or the equation of diffusion.
When U(x,t) represents the concentration of a diffusing substance in
weight per unit volume, Eq. (3) follows as before if e = 1 and K is the coeffi-
cient of diffusion. Thus, when K is constant, U(x,t) satisfies Eq. (5) where
k= K.
The heat equation or equation of diffusion for U(x,y,z,t)
(6) U, = k(Uxx + Uyy + Uzz),
when coefficients are constant, can be derived in a similar manner by con-
sidering a three-dimensional element of volume ~x ~x ~Z.1
Modifications of the heat equation are easily written. For example,
let U(x,t) represent temperatures in a slender wire along the x axis. Let heat
be generated at a time rate R(x,t) per unit volume of the wire and let heat
loss from the lateral surface take place at arate proportional to the difference
between the temperature of the wire and the temperature Uo of the surround-
ings. The modification of Eq. (3) then shows that
1
(7) U,(x,t) = kUxix,t) + -R(x,t)
e - h[U(x,t) - Uo],
where h is a positive eoefficient of surfaee heat transfer.

1 See chap. 1 of the author's "Fourier Series and Boundary Value Problems," 2d ed., listed in
the Bibliography, for a full derivation of the heat equation.
In the problems to follow we assume that the coefficients K, k, e, and h

are constants. AIso, we often consider semi-infinite solids, where results are
relatively simple. Such bodies are not necessarily large in any absolute sense.
In the diffusion of hardening materials into steel, for instance, test bars of
length 1 in. or even less may be represented quite accurately as semi-infinite
solids.
46 TEMPERATURES
IN A SEMI-INFINITE
SOLIO
Let us now derive a formula for the temperatures U(x,t) in a semi-infinite
solid x ~ O,initially at temperature zero, when a constant flux of heat is
maintained at the boundary x = O (Fig. 55). In this idealized case of a thick -
slab of material, we shall substitute for the thermal condition at the right-hand
boundary the condition that U tends to zero as x tends to infinity. The
boundary value problem is then
(1) Ulx,t) = kUxx(x,t) (x > O,t > O),
(2) U(x,+O) = O (x > O),
(3) -KU,,(O,t) = <Po, lim U(x,t) = O (t -> O).
Let u(x,s) be the transform, with respect to t, ofthe temperature function

U(x,t). Transforming the members of Eqs. (1) and (3), we have the following
problem in ordinary differential equations which u(x,s) must satisfy:
su(x,s) = ku",,(x,s) (x > O),
- Ku,,(O,s)= <Po,
s
lim u(x,s) = O.
x-oo
The solution of that problem is
u(x,s) = <PO{e-M/S/k.
Ksys
Accordingto formula (5),Seco27, we can write
L-1 {fie - M/S/k}

=~ exp( - ::J
-'
4>0-!
-10 x
-1
Fig.55
and in view of the factor l/s in our formula for u(x,s) it follows that
4>
, X2 d.
U(x,t) = ~
K ~ -
1t1o
exp ---
( 4h ) .Ji
4>ox
f oo
= Kv :1t x/(2./ktA2exp (- A )dA,
1 2
where the second integral is obtained from the first by the substitution
A = x/(2.)kT). Upon integrating the last integral by parts, we find that
2
U(x,t) = 4>~ r2Jkt

Kv 1tL
exp (
- X
4k t)
- 2x
f oo exp (- A(!)dA .
x/(2./kt ]
We can therefore write our formula in terms of the complementary error
function (Sec. 27) in the form
-~
x erfc ~ -
(4)
[
U(x,t) = 4>0 2 {kiexp
K V -; ( 4kt) ( 2Jkt)J
.
We can show that the function (4) satisfies all our conditions (1), (2), and (3).
We observe that
1
K 1t
U(O,t)= 24>1f fi.
Thus the temperature of the face of the solid must vary as Jt in order that
the fluxof heat through the faceshall be constant.
47 PRESCRIBED SURFACE TEMPERATURE

Let the temperature of the face of a semi-infinite solid x ;;;;Obe a prescribed
function F(t) of time. If the initial temperature is zero, the temperature
function U(x,t) is the solution of the boundary value problem
U,(x,t) = kUxx(x,t) (x> O,t > O),
U(x,O) = O (x > O),
U(O,t)= F(t), lim U(x,t) = O
x"" 00
(t > O).
The transform u(x,s) of U(x,t), therefore, satisfies the conditions

su(x,s) = kuxx(x,s) (x > O),
u(O,s) = f(s), lim u(x,s) = O,
x"" 00
where f(s) is the transform of F(t). It follows that

(1) u(x,s) = f(s)e-xJ;ik.
Let us first study the flux of heat through the face of the solid,
~(t) = -KU",(O,t).
The transform of this function, - Ku",(O,s),according to formula (1), is
K K 1
(2) cp(s)= y'k.fif(S) = y'ksf(s) .fi'
Since s-t = L{(nt)-t} and sf(s) = L{F'(t)} + F(+0), assuming that F is a
continuous function,then
cp(s)= ~ [F(+0) + L{F'(t)}~ ]

y'k .fi .fi
and with the aid of the convolution it follows that
(3) ~(t) =~
pjt [
F(+0) +
fo~
r F'(t - 't) d't
]
,
when F is continuous. Ir F( + O)# O,the flux is infinite initially, of the order

of t-t as t ~ O. In particular if F(t) = Fo,a constant, then
~(t) = KFo(nkt)-t.
The total amount of heat that has been absorbed by the solid through
a unit area of the face at time t is
Q(t) = {~('t) d't.

1 K 1
Hence
q(s) = -cp(s) = r;: /:f(s),
s ykys
and therefore
(4)
For example, when

(5) F(t) = Fo if t < to and F(t) = O if t > to,
it follows from formula (4) that
Q(t) = 2KFo r: when t ;;;to,

Pyt
= 2KFo when t ~ to.
p(jt-~)
148 SECo 47] OPERATIONAL MATHEMATIC!:
o f Fig.56
This function is shown in Fig. 56. Its greatest value is

(t;
2KFoVnk'
which is assumed at the instant t = to.
Returning to formula (1)for u(x,s) and noting that, according to formuJ
(6), Seco27,
e-x./Sik = L exp -~ x ,
{ 2v1nkt3
( 4kt)}
we can write,with the aid of the convolution,
U(x,t) = 2#
X
io
t F(t -
Ti
T) X2
exp ( - 4h ) dT.
Substitutinga new variable of integration, we have for the general tempera-
ture formula
oo 2
2
fi f
(7) U(x,t) = - F t x
txl../ki ( - 4k,F ) e-;.2 dA..
When the temperature of the surface is constant,

(8) F(t) = Fo,
the temperature within the solid is therefore
(9)
U(x,t) =.Fo erfc
I (2~).
Since this is a function of Fo and x/Jki only, it follows that the rapidity o
heating is proportional to k; for if k is increased and t decreased so that k
is unchanged, the temperature at any given distance x from the face is thc
same. It is also interesting to note that for a fixed k two points Xl and x2 \Vi
have equal temperatures at times ti and t2 provided x/-.; = X2/.jt;, thaJ
is, if
(10)
This is sometimes called the law of times in the conduction of heat in semi-
infinite solids.
Ir Fo is positive, formula (9) assigns a positive value to U(x,t) for each
x and t beca use the complementary error function has only positive values.
At each interior point of the solid U(x,t) is small when t is sufficiently small,
but the temperature does not retain its initial value U(x,O)= O during an
interval of time after t = O. That instantaneous though highly attenuated
transfer of the thermal disturbance at the surface can be noted in other
cases. In Seco46, for example, the flux
-KUjx,t) = 4>0erfc (~fit)

is clearly different from zero immediately after the thermal disturbance took
place at the surface x = O. The instantaneous transfer is a consequence of
the basic postulate on the flux: <1>= - K dU/dn (Sec. 45).
PROBLEMS
.
1 The initial temperature of a semi-infinite solid x ~ Ois zero. The surface tempera-
ture is this step function:
U(O,t)= Fo ifO < t < to, U(O,t)= O if t > to.
Obtain the temperature distribution formula
x
U(x,t) = Foerfc" r,:;
~kt
XX
= Fo erf - erf-
[ 2Jk(t - to) 2.jki J
2. A thickslab ofiron withthermal diffusivityk = 0.15 cgs (centimeter-gram-second)
unit is initially at OC throughout. Its surface is suddenly heated to a temperature of
500C and maintained at that temperature for 5 min, after which the surface is kept
chilled to OC. (See Probo 1.) Find the temperature to the nearest degree at a depth of
10 cm below the surface (a) at the end of 5 min; (b) at the end of 10 mino
Ans. 146C; (b) 82C.
3. Solve Probo 2 if the slab is made of firebrick with diffusivity k = 0.007 cgs unit.
Ans. (a)OC; (b) Oc.
4. The surface of a thick slab of concrete for which k = 0.005 cgs unit, initially at OC,
undergoes temperature changes described in Probo 2. Show that at each instant the tem-
perature at any depth x in the concrete slab is the same as the temperature at the depth
X2 = J30x in the iron slab. Generalizethis result for materialswith diffusivitiesk
and k2 and any common time interval to of heating the surfaces of the slabs.
5. At time t = O,the brakes of an automobile are applied, bringing the automobile to a
stop at time to. Assuming that the rate of generating heat at the surface of the brake
bands varies linearly with the time, then

Ux(O,t)= A(t - to)
where A is a positive constant and x is the distance from the face of the bando Ir to is not
large, the band can be assumed to be a semi-infinte solid x ~ O. Ir the initial temperature
of the band is taken as zero show that the temperature at the face is
2A fk r;,
U(O,t)= 3 V nV t(3to - 2t)
Hence show that this teniperature is greatest at the instant t = tto and that ths maximum
temperatureis j2U(O,to). .
6. Theinitialtemperatureofa semi-infinitesolidx ~ Ois zero. The inwardfluxofheat
through the facex = Ois a prescribed function 4I(t)oftime and the distant face is kept at
temperaturezero. Derivethe temperatureformula
(x > O).
Also show that the temperature of the face can be written

r
1 (k
U(O,t)= K.,Jn i
O 4I(1:")(t - 1:")-t d1:".
7. An impulse of heat of quantity Qo per unt area is introduced through the face
x = O of a semi-infinite solid x ~ Oat the instant t = Oand the face is nsulated when
t > O,so that -KUx(O,t) = Qo c5(t).(That instantaneous source ofheat may be realized
approximately by burning a layer of highly combustible fuel at the face.) Ir U(x,O)= O
when x > O,show formally that the subsequent temperature is
X2
U(x,t) = ~QK /ff
-1ftexp --4kt (t > O).
( )
(See Probo 8 for a verification.) Observe that U(x,t) > Owhenever t > O.
8. The function I/I(x,t)= t-t exp [-x2/(4kt)] and its derivative I/Ix are known as
fundamental svlutivns vf the heat equativn I/Ir= kl/lxx'
(a) Verify that 1/1satisfies that equation when t > O,that 1/1 iO,t) = Oand I/I(oo,t) =
O when t > O,and that I/I(x,+O)= O when x > O; then complete a verification of the
formal solution of Probo 7 by showing that the total heat content of the solid per unit
area offace, f~ cU(x,t) dx, relative to the initial heat content, is Qo.
(b) Show that the functions 1/1and I/Ixare unbounded in a neghborhood of the
point x = t = O when x > O and t > O. Note that functions AI/I(x,t), where A is an
. arbitrary constant, could be added to the solution (4),Seco46, to producea familyof
solutions of the problem in that section if unbounded solutions were permitted.
9. For a solid whose temperature is a function U(x,t) show that the total quantity
of heat transmitted across a unit area of a plane x = Xo can be written, formally,
PROBLEMS IN PARTlAL DIFFERENTIAL EOUATIONS [SECo 48 151
For the solid in Seco47 show that the formal result is
Q(xo) = -/CKs-o
limJif(s) (K = ck).
For the solid in Probo 7, show that Q(xo) = Qo.
10. The initial temperature distribution of a solid x ~ Ois prescribed:
U(x,O) = g(x) (x > O).
If U(O,t) = O when t > O and UAoo,t) = O, show formally that, when t > O,
1 OJ
(r - X)2 (r + X)2
U(x,t)= 2;;ki fo g(r){exp [ -~ [

J - exp -"4kt J } dr.
11. If the initial temperatureis uniformin Probo10,g(x) = Uo' showthat
U(x,t)= Uoerf (~).

12. Show that the sum of the temperature function found in Probo 10 and the function
(7), Seco47, represents the temperature in the solid x ~ Owith initial temperature g(x)
and face temperature F(t).
48 TEMPERATURES IN A SLAB
The initial temperature of a slab of homogeneous material bounded by the
planes. x = Oand x = 1is Uo. Let us find a formula for the temperatures-in
this solid after the face x = O is insulated and the temperature of the face
x = 1is reduced to zero (Fig. 57).
The temperature function U(x,t) satisfies the following conditions
Ut(x,t) = kUxAx,t) (O< x < 1,t > O),

U(x,O) = Uo (O < x < 1),
U AO,t) = O, U(1,t)= O (t > O).
The transform therefore satisfies the conditions
(1) su(x,s) - Uo = kuxx(x,s),
(2) u(1,s)= O.
U(x,O)-UQ u-o
o x
Fig.57
The solution of the ordinary differential equation (1) that satisfies the
first of the conditions (2) is
u(x,s) = USo
+ e cosh xA,
where e is determined by the second of conditions (2). Thus
(3) u(x,s) = uo
[! - ! coshx# ].
s s cosh1#
Let us write
q=f,
and note that 1

coshIq = e-lq(eXq+ e-Xq)1 + e 21q
coshxq
<XJ
= [e-(l-X)q + e-(I+x)q] Lo (-I)"e-2",q

<XJ
= L (-I)"{exp
"=0
[-q(ml - x)] + exp [ -q(ml + x)]},
where m = 2n + 1. We have seen(Sec.27) that
(ex ~ O).
Therefore it follows formally from Eq. (3) that

<XJ
(2n + 1)1- X
(4)
U(x,t) = Uo - L
Uo "=0 (-1)" { [
erfc Jki
2 kt ]
(2n + 1)1+ x
+ erlC
~
[ 2ykt
r,::
]} .
We shall not take up the verification of this formula sincea complete
discussionwould be lengthy. However, it is not difficult to show with the
aid of the ratio test that the series converges uniformly with respect to x
and t and that the series can be differentiated term by termo Since the value of
the complementary error function here decreases rapidly as n increases, the
convergence of the seriesis rapid, especiallywhen t is small. Moreover the
error function is one that is tabulated so that the series is well adapted to
computation.
Fig.58
49 A BAR WITH VARIABLE END TEMPERATURE

Let us determine a formula for the temperature V(x,t) in a bar with its lateral
surface insulated against the flow of heat when the initial temperature is
zero and one end is kept at temperature zero while the temperature of the
other end is a prescribed function of t.
If we take the unit of length as the length of the bar (Fig. 58) and select
the unit of time such that (l/k) av/at' becomes av/at, that is, so that t = kt'
where t' is the original and t the new variable, our boundary value problem
can be written as follows:
Vt(x,t) = V xJx,t) (O< x < 1, t > O),
V(x,O) = O (O< x < 1),
V(O,t) = O, V(l,t) = F(t) (t > O).
The solution of the transformed problem is found to be
u(x,s)
= f( s) sinhx-fi1: .
sinhv s
Proceeding as in the last section, and using the convolution, we find
that
'"
2 '" (m - X)2
(1) V(x,t) = V~L
7tn=O {f F t-
t(m-x)/./l [
4A2 J e-A2dA
'"
(m + X)2
- ft(m +x)/./l F[t - 4A2 J e-A2 dA} (m = 2n + 1).
When the temperature of the facex = 1 is constant,
(2) F(t) = Fo,
our formula can be written
'" 2n + 1 + X 2n + 1 - x
(3)
V(x,t) = Fo t [erf( 2jt ) - erf( 2.Ii )J .
Details are left to the problems.
50 A COOLlNG FIN OR EVAPORATION PLATE
A thin semi-infinite plate occupies the space x ~ O,O~ Z ~ Zo, - 00 < y < oo.
Heat transfer takes place at the faces Z = O and Z = Zo,into a medium at
temperature zero,accordingto the linear lawofsurfaceheat transfer(Sec.45);
u-o
t t
S-::'::==::Y
+ + :c
U{x.O)-O
Fig.59
but the thickness Zoofthe plate is small enough that variations oftemperature
with z can be neglected. Ir the initial temperature is zero and the end x = O
is kept at temperature Fo (Fig. 59), the boundary value problem for the
temperature function becomes
Ut(x,t) = kU xAx,t) - hU(x,t) -t ~i. (x > O, t > O),
(1)
U(x,O)= O, U(O,t)= Fo, lim U(x,t) = O.
Here U(x,t) may also represent the concentration of moisture in the

plate when the plate is initially dry and evaporation takes place into a dry
medium, if the concentration at the end x = Ois a constant Fo. In that case
k = K, the coefficient of diffusion (Sec. 45), and the positive constant h is a
coefficient of evaporation.
In terms of transforms, our problem (1) becomes
kUxAx,s) - (s + h)u(x,s)= O (x > O),
Fo
u(O,s)= -,s lim u(x,s) = O.
x-""
The solution of this problem is
Fo r;+h
(2) u(x,s) = - exp ( - Xv --;- ) .
Knowing the inverse transform of e - xJiik, we can write, with the aid
of our property on substitution of s + h for s (Sec. 7),
xe-ht X2
L -1 {exp
(-xPi)} = 2~ nkt3 exp ( - 4kt) .
It follows from formula (2) that
t
(3)
U(x,t) = ~
2~
F. x
""
o ie-ht exp ---
(
X2
4kt ) tt
dt
2F. hx2
~
= '\! n Jtx../ki exp ( -A2 - 4k12A ) dA,
where the second integral is obtained by the substitution A = tx/~.

The formula (3)can be ehanged to a more usefulform with the aid of

the integration formula (see Probo 14 below)
(4)
~ 100 exp ( -AZ - ~~) dA = eZaerfe (r +~) + e-Za erfe (r - ~).
This formula can be verified by noting that its two members have the same
derivative with respeet to the parameter r and that both members vanish as
r tends to infinity. With the aid offormula (4) the solution (3) of our problem
becomes
(5)
U(x,t) = ~[ebXerfc
(2~ + Jht) + e-bx erfe (2~- Jht)1
whereb =.Jhik. Note that U(x,t) > Ofor eaeh x whenevert > O,if Fo > O.
In obtaining Eq. (5) we have found a desirable form of the inverse
transform of the funetion (2). The verifieation of our formal solution of
problem (1)is left to the problems.
51 TEMPERATURES IN A COMPOSITE SOLlD

Let us write a formula for temperatures U(x,t) in a solid x ~ O,eomposed
of a layer O < x < a of material initially at uniform temperature A in perfeet
thermal contact with a semi-infinite solid x > a of another material initially
at temperature zero, when the face x = O is kept insulated (Fig. 60).
If the thermal eonduetivity and diffusivity are K1 and k1, respeetively,
in the first part and Kz and kz in the second part, the boundary value problem
is the following one:
(1) Ur(x,t) = k1 Uxx(x,t) (O< x < a, t > O),
(2) Ur(x,t) = kz Uxx(x,t) (x > a, t > O),
(3) U(x,O)= A(O< x < a), U(x,O)= O (x > a),
(4) UAO,t)= O, lim U(x,t) = O
x-oo
(t > O),
U(%,O)-A
o a %
Kl.kl
Fig.60
(5) U(a - O,t) = U(a + O,t) (t > O),

(6) Kl UAa - O,t) = K2UAa + O,t) (t> O).
Condition (5) states that the temperature at the interface x = a is the
same after t = Owhen the point approaches the interface from either direc-
tion. Condition (6)states that the flux of heat out of the first part through the
interface must be equal to the flux into the second part, at each instant.
The problem in the transform of U(x,t) is then
(7) xu(x,s) - A = k1uxx(x,s) (O < x < a),
(8) su(x,s) = k2ux::.(x,s) (x > a),
(9) uiO,s) = O, lim u(x,s) = O,
x-+00
u(a - O,s) = u(a + O,s),
(10)
K1ux(a - O,s) = K2ux(a + O,s).
The solution of Eq. (7) that satisfies the first of conditions (9) is
fS A
u(x,s)= Cl coshx-vI; + -; (O~ x < a),
and the solution of (8) that satisfies the second of conditions (9) is
(x > a).
u(x,s) = C2exp (-xJf)
By applying the conditions (10) to these functions, the values of Cl and C2
are easily found. The formulas for u(x,s) can then be written
- A 1 - A e-ala-x) + e-ala+X
(O< X < a),
(11) u(x,s)- -; ( 1 -~. 1 - Ae-2aa )
A(l + A)e-aplx-a)- e-aI2a+px-pa)
(12) u(x,s) = (x > a),
where
r; (k;
u = vk;' Jl = -VI<;'
and therefore UJl= .;;k; and IAI< 1.
Equation (11) can be written in the form
u(x,s) =
A
!s - 1-
2
A f
"=0 s
A"(e-alma-x) + e-alma+x (O< X < a),
where m = 2n + 1, and (12) can be written

.
u(x,s) = 1 + A
A 2
f
n=O
An (e-a(2na+/lx-/la)
s
- e-a(2na+2a+/lx-/la)) (x > a).
It therefore follows that

1 - A00 (2n + l)a - x
(13)
U(x,t)= A - A-y- ~ An{erfe [ 2.jk;i ]
+ erfe (2n + l)a + x (O< x < a),
[ ~
2yll.lt ]}
1 + A 00 2na + Jl(x- a)
(14)
U(x,t)= A-y- ~ An{ erfe [ 2.jk;i ]
- erfe (2n + 2)a + Jl(x - a) (x > a).

[ ~
2yll.lt ]}
In sueh problems if U(x,t) represents the eoneentration of a substance
diffusing in a eomposite porous medium, the eoneentrations in adjaeent
layers of different media may maintain some ratio a (a # 1), the ratio of
equilibrium concentrations in the two media (see Probo 13 below). InterfaCe
eondition (5) would then be replaeed by the eondition
(15) U(a + O,t) = aU(a - O,t).
PROBLEMS
1. Completethe derivationofformula (1),Seco49.
2. Whenx = 1,showthat thegeneraltermofthe seriesinformula(3),Seco49,becomes
and hence that U(I,t) = Fo.

3. In Seco50, verify the solution (5) ofproblem (1).
4. In the problem (1) of Seco50, show that the flux into the plate at the end x = Ois
given by the formula
(O,t)
= KFo(~e-ht + AerfJht).
5. In the problem (1) of Seco50, make the substitution
V(x,t) = ehIU(x,t)
and show that the resulting problem in V(x,t) is a special case of one solved in Seco47.
158 SECo 51] OPERATIONAl MATHEMATICS
6. The faces x = Oand x = 1 of a slab of material

for whieh k = 1 are kept at tempera-
tures U =O U = 1, respeetively, until the temperature distribution beeomes
and
U = x. After time t = Oboth faces are held at temperature U = O. Derive the tempera-
ture formula
oo
2n+1-x 2n+l+x
(t > O).
U(x,t)= x - J;o ( erfe 2.1t erfe 2../i )
7. In Prob. 6, let the face x = 1 be insulated when t > Owhile U = Oas before at the
face x = O. Find the flux of heat outward through the face x = O.
00 2n + l
[
Ans. K 1 - 2 "~o(-1)" erfe 2../i J.
8. Heat is generated in a long bar x $;;Oat arate R(t) units per unit volume [Seco45,
Eq. (7)],and the lateral surface of the bar is insulated. Ir the initial temperature is zero
and the end x = Ois kept at that temperature while the infinite end is insulated (Fig. 61),
derive the temperature formula
2 oo X2
U(x,t) = Q(t) - r= Q( t
V 1t
l -
4U2 ) exp(-''2)dl,
!:x;./fi
l t
where Q(t) = -
e o i
R(t) dt.
u-0-c~;;;;:::;;;;~~:::r:;:~!;::;:::)
O UIz,O-O -
-1/4-
--.,.7,
Fig.61
9. In Probo 8, let the rate of generation of heat be a eonstant B and show that
B
i
U(x,t) = e oerf 2../fT
t ~dt.
10. The lateral surface of a bar of unit length is insulated while its ends x = O and
x = 1are kept at temperaturezero. Heat is generatedthroughoutthe bar at a eonstant
rate of B units per unit volume. Ir the initial temperature is B(x - x2)/(2K), show that
the bar retains that initial temperaturedistributionwhent > O.
11. Units of time and distance are ehosen so that k = 1 and a given bar has unit
length. The lateral surface and the end x = O are insulated. The initial temperature of
the bar is zero, and the end x = 1 is kept at that temperature. Ir heat is generated
throughout the bar at arate of R(t) units per unit volume, derive the temperature
formula
cU(x,t)= E R(t)dt - I: R(t - t)E(x,t) dt,

00
2n + 1 + X 2n + 1 - X
where
[
E(x,t)= ~ (-1)" erfe( 2..!t ) + erfe( 2,!t )J .
PROBlEMS IN PARTIAl DIFFERENTIAl ENQUATIONS [SECo 52 159
12. Let the semi-infinite evaporation plate (Sec. 50, Fig. 59) have a uniform initial
concentration of moisture U(x,O) = uo, and let the end x = O be kept dry, U(O,t) = O,
while the infinite end is impervious to moisture. Derive, and verify, this formula for
the concentration :
-h, f x
U(x,t) = uoe er rv;.
2yKt
13. The entire surfaceof a long porous cylinder is impervious to moisture. The
cylinder consistsof two semi-infiniteparts, the first ( - 00 < x < O)of material with a
coeffieientof diffusion of moisture K and the second (O < x < 00) with a eoefficient K 2
(Fig. 62). Initially, the concentration of moisture is zero in the part x < O and Uo in
the part x > O.Let IXbe the ratio of equilibrium concentrations in the two parts so that,
as indicated in Seco51, the concentration U(x,t) satisfies the eonditions
U( +O,t) = IXU(-O,t),
at the interface x = O. Derive the formula
Uo -x
U(x,t) = - rr;;-; whenx < O,
IX + p erfe ( 2yKt )
X
= - Uo + p erf" rr;;-;) whenx > O,
IX + p(
IX
.yK2t
where p = .JK/K2' Show that the ratio ofthe limiting coneentrations, as t - 00, in
two parts is 0(.
L---G:r~~
~UI#IJ)-O K~::::::~
~UI#IJ)-"
O
Fig.62
1
14. Derive the integration formula (4), Seco50. Suggestion:Write
z(x) = ~ro exp [- (l- I)]dl;

then the integral (4), Seco 50, is e - 2Dz(a).Show that z'(x) = 4e4xerfe(r + xlr) and
z(0)= 2erfer, thenuseintegrationbypartsto findz(x). .
52 OBSERVATIONS ON THE METHOD
All the problems treated in this chapter involve partial differential equations
and boundary conditions that are linear,that is, offirst degree in the unknown
function and its derivatives. The limitation of our operational method to
the treatment of such linear boundary value problems is a natural one, since
we have presented no formula giving the Laplace transform of the product
/
of two functions in terms of the transforms of the individual functions. It is

known that the transform of the product of two arbitrary functions can be
expressed by a convolution integral of the two transforms, where the integra-
tion is one in the complex plane of s. But it is safe to say that no advantage
can be anticipated in replacing nonlinear differential forms by complex
nonlinear integral forms.
We have solvedproblems with constant coefficients. Ir the coefficients
are functions of t, the variable with respect to which the transformation is
made, the transformed problem is not likely to be simpler than the original
one. For even when the coefficients are polynomials in t, the transformed
problem involves derivatives with respect to s in place of the derivatives
with respect to t present in the original one. Ir the coefficients are not func-
tions of t, the transformed problem will be simpler.
We have made the transformation with respect to time t in all our
problems in partial differential equations. Ir the physical problem involved
the first derivative U" the initial value U(x,O) was prescribed; if it involved
Y,,, then Y(x,O) and Y,(x,O)were both prescribed. Consequently, when we
applied the formula for the transformation of those derivatives, u(x,s) or
y(x,s) was the only unknown function arising. But suppose the Laplace
transformation with respect to x had been applied in the temperature prob-
lems. Then if L{ U(x,t)} = u(z,t),
L{ Uxx(x,t)} = Z2U(Z,t)- zU(O,t) - U)O,t),

and not both of the functions U(O,t) and U)O,t) could be prescribed, since
both the temperature and the flux of heat at the surface x = O cannot be
prescribed. Thus one of those unknown functions must be determined with
the aid of other boundary conditions, and the procedure becomesunwieldy;
the Laplace transformation is not the integral transformation that is adapted
to the boundary conditions with respect to x here.
The operational method of solving partial differeQ#ialequations is of
coursenot limitedto equations of the second order. We shall soon take up
a more powerful method of obtaining inverse transforms, and then we can
attack problems whose solutions depend upon more involved inverse trans-
formations than those in this chapter.
As we have illustrated here and in the preceding chapter, the operational
method is well adapted to the solution of many problems in differential
equations in which some of the given functions or their derivativesare dis-
. continuous. That is one of the remarkable features of the method.
In case the boundary value problem involves more than two indepen-
dent variables, say x, y, and t, a Laplace transformation with respect to t
stillleaves us with a partial differential equation with independentvariables
x and y. The new problem may be adapted to solution by one ofthe Fourier
transformations or another integral transformation with respect to x or y,
depending on the differential forms and boundary conditions involved there.

That procedure of using successive transformations will be illustrated in
Chap. 11.
As the problems in the. present chapter have indicated, transformation
methods help to show whether the number and type of boundary conditions
accompanying a given partial differential equation are adequate to determine
a definitesolution of the boundary value problem. A physicalinterpretation
of the problem servesas another guide for setting boundary conditions that
lead to just one solution.
A complete treatment or rigorous solution of a boundary value prob-
lem, however, consists first of proving that the formal.solution does satisfy
the differential equation and all boundary conditions and continuity require-
ments, then showing that the function is the only one that satisfies all those
conditions. That procedure of establishing a unique solution. is apt to be
lengthy. We shall illustrate it in later chapters.
5
Functions of a Complex Variable
We give now a synopsis of theorems and definitions from the basic theory of
functions of a complex variable, material needed to develop futher the
theory ofthe Laplace transformation. For complete treatments ofthe topics,
including proofs of statements and theorems, the reader may refer to books
on the subject.l Derivations of extentions of the Cauchy integral formulas
are given in the final section of this chapter.
53 COMPLEX NUMBERS
We shall not review the algebra of complex numbers except to remind the
reader of properties of absolute values and conjugates. We write
z = x + iy = r(cos 9 + i sin 9),
1 See, for instance, the author's "Complex Variables and Applications," 2d OO.,1960, and
references listed in the Bibliography in that book.
162
FUNCTIONS OF A COMPLEX VARIABLE [SECo 54 163
x = Re z, y = 1m z, r = Izl and e ;;., arg z.
Ir Zl and Z2 are complex numbers, then
if Z2 "" O.
The triangle inequalities are
IZ1 :t z21 ;; Iz11 + Iz21,
Neither the statement z 1 > z 2 nor z 1 < z2 has meaning unless z 1 and z 2 are
both real numbers.
Ir m and n are integersand z "" O,then
me . me
znfn = r"'fn ( ros --;-+ i SID --;-) , Izmfnl= Izlmfn.
The conjugate operation on z, Z = x - iy, has these distributive

properties:
Zl + Z2 = Zl + Z2, ZlZ2 = ZlZ2,
AIso zz = IzZ..
A neighborhood of a point Zoin the complex plane is a two-dimensional
circular domain, or disk, consisting of all points z such that Iz - zol < (,
where ( is some positive number.
54 ANAL YTIC FUNCTIONS
Let w denote the values of a single-valued function j of the complex variable z,

defined on some set of points in the z planeo We write
w = j(z) = u(x,y) + iv(x,y)
where u and v are real-valued functions of two real variables. Ir w = Z2 for
all z, for example, then
j(z) = (x + iy)2 = X2 - y2 + 2xyi;
thus u(x,y) = x2 - y2 and v(x,y) = 2xy. Ir w = Re z then u(x,y) = x and
v(x,y) = O.
Ir w is a single-valued function z! such that
z! = .j;(cos~ + iSin~) (r> O, -1t < e < 1t),

then it is convenient to write the components u and v as functions of r and e:
u(r,e) = ~ cos!O, v(r,e)= ~ sin!O
(r > O,-1t < e < 1t).
Here u and v are single-valued in the domain consisting of all points of the
complex plane except those for which y = O and x ~ O.
A function f has a limit Woat a point Zo,
lim f(z) = WO,

%-+%0
if and only if for each positive number lO there is some number ~ such that
If(z) - wol < lO whenever Iz - zol < ~ and z :1=Zo.
Thus the value of f is arbitrarily close to Wo throughout some deleted

neighborhood of Zo, a two-dimensional neighborhood with Zo excluded.
The definition requires f to be defined throughout some neighborhood of
the point Zo, except possibly at the point itself.
Ifwe write Re Wo= uo, 1mWo= vo, and Zo= Xo + iyo, then it tums
out that the condition
lim f(z)
%-%0
= %-%0
lim (u(x,y) + iv(x,y)]= Uo+ ivo
is satisfied if and only if the two-dimensional limits of the real-valued

functions u and v at the point (xo,Yo) are Uoand vo, respectively. Since f is
continuous at Zo only if its limit Wois the same as its value f(zo), then it is
continuous there if and only if both u and v are continuous at the point
(xo,Yo).
The derivative of f at a point z is defined as
f'(z) = 6:lim0 f(z + ~z)

~z
- f(z)
Jo
provided the limit exists, that is, provided that for each positive number lO
there is a number ~ such that
I f(z + ~; - f(z) - f'(z) <

I lO
whenever I~zl < ~(~z :1=O).
The two-dimensional limit here with respect to the variable ~z (Fig. 63)
requires that, iff'(z) exists, the partial derivatives offirst order of the functions
u and v exist at the point (x,y) and satisfy the Cauchy-Riemann conditions
au av au av
-=-, -=--,
ax ay ay ax
at the point. AIso, when f'(z) exists, we find that
f'(z) = ux(x,y)+ ivx(x,y)= Vy(x,y)- iuy(x,y).
FUNCTlONS OF A COMPLEX VARIABLE [SECo 54 165
Fig.63 o
Suppose that U and V are real-valued functions of x and y whose

partial derivatives of first-order are continuous at a point (x,y).! Then in
order for the function f = u + iv to have a derivative f'(z) at the point
z = x + iy, it is necessary and sufficient that u and v satisfy the Cauchy-
Riemann conditions at the point.
For example, the components u and v of the function Z2satisfy those
conditions everywhere; hence
d o o
dz (Z2) = oX (X2 - y2) + i ox (2xy) = 2(x + iy) = 2z.
In contrast, the components of the function Izl2 = X2 + y2, whose partial
derivatives are continuous everywhere, satisfy the Cauchy-Riemami
conditions at the origin only, where 2x = Oand 2y = O. Thus the derivative
of Izl2 exists at z = Oonly, where its value is zero.
A function f is analytic at a point Zoif its derivative f'(z) exists at every
point in some neighborhood of zo. Analyticity implies continuity, but not
conversely. The functions Re z, z, and Iz12,for instance, are everywhere
continuous, but nowhere analytic because their derivatives do not exist
throughout any neighborhood.
Formulas for derivatives of sums, products, quotients, and powers of
functions and of composite functions g[f(z)J correspond to those in calculus.
Consequently the sum, product, or quotient of two analytic functions is
analytic except, in the case of the quotient, at those points where the de-
nominator vanishes. An analytic function g(w) of an analytic function
w = f(z) is analytic for values of z in a domain (a connected open region)
where w is analytic and where the values of w are interior to a domain where
g(w) is analytic.
The function l/z is an example of a quotient of two functions that are
everywhere analytic. The function is analytic at all points except the origino
Its derivative is -1/z2 if z =1=O.
An entire function is one that is analytic for all finite Z. Every poly-
nomial in z is an entire function.
1The continuity of those partial derivatives ensures the continuity of the functions u and v
themselves, at the point.
55 EXPONENTIALAND TRIGONOMETRIC FUNCTIONS

The exponential function, written exp z or eZ,can be defined by the equation
(1) exp z = eX(cos y + i sin y) (z = x + iy),
where y is the radian measure of the argument of the trigonometric functions.
Its components u and v have partial derivatives which are continuous and
satisfy the Cauchy-Riemann conditions everywhere. Therefore exp z is an
entire function. Its derivative is the function itself:
d
(2)
dz exp z = uX<x,y)+ ivX<x,y)= exp z.
We note that lexp zl = ~ > O for all z; thus is never zero. When
z = x, the function expz becomesthe real-valued function ~.
The polar representation of z can now be written reie:
I z = r(cos lJ + i sin lJ)~ r exp (ilJ).
The hyperbolic cosine and sine are these linear combinations of the
entire functions and e-Z:
(3)
hence they are entire. Their components are shown by the formulas
cosh z = cosh x cos y + i sinh x sin y,
sinh z = sinh x cos y + i cosh x sin y.
Each of the remaining hyperbolic functions of z,
l
tanh z = sinh z
cosh z' coth z = tanh z'
l l
sechz = cosh z' csch z = sinh z'
is analytic except at those points where the denominator on the right-hand
side vanishes.
The circular functions of z can be defined as follows:
. eiz .
(4) smz = - e-1Z tanz=sinz
2 2i cosz'
the remaining three being the reciprocals of these. The functions cos z and
sin z are entire. In view of these definitions, it follows that
cos z = cosh iz = cos x cosh y - i sin x sinh y,
sinz = - i sinh iz = sinx cosh y + i cosh x sinh y.
AIso, sin z and cos z are periodic with period 2n, while sinh z and cosh z are
periodic with period 2ni.
The forms of the trigonometric identities are the same as those for
functions with real arguments; for instance,
sin2 z + COS2Z = 1, cos 2z = COS2Z - sin2z,

sin(zl + Z2) = sin Zl cos Z2 + COSZl sin Z2'
The same is true for the relations between the six hyperbolic functions.
Furthermore, the formulas for the derivatives of all those functions retain
the same form when the variable is real. Thus for all z in a region where w(z)
is analytic, the composite functions sin w(z) and cos w(z) are analytic functions
of z and
d
-sm
. w = cosw-
dw d
-cosw
. dw
= -sm w-.
dz dz' dz dz
The absolutevalues of sin z and cos z are not bounded as Iyl -. oo. It is left to
the problems to show that
(5) IsinZl2 = sin2 x + sinh2 y, Icos Zl2 = COS2x + sinh2 y,
(6) Isinhzj2 = sinh2x + sin2 y, Icosh Zl2 = sinh2 x + COS2y.
From those formulas we can see that the zeros of sin z and cos z are just the
real zeros of those functions, while the zeros of sinh z and cosh z are pure
imaginary numbers.
PROBLEMS
1. (a) Interpret the triangle inequalities (Sec.53) in tenns oflengths of sidesoftriangles.
(b) Extend the first triangle inequality to apply to n complex numbers as follows:
IZ1+ Zz + ... + z.1 ;;; Iz11 + Izzl + ... + Iz.l.
2. Ir Izzl =FIz31,prove that
3. Ir n = 2, 3, . . . , show that zll. has n distinct values.

4. Ir a. and b. are real numbers, prove that the function
f(z) = alx + bly + e + i(a2x + bzy) (z = x + iy)
is an entire function if and only if the coefficients are such that f(z) = (al - ibl)z + c.
5. Provethat the functionf(z) = XZ + iyZis nowhereanaIytic.
6. State whythe functionssin() and exp(cosz)are entire.
7. Show that (a) e-Z = lje"; (b) exp (Zl + Z2) = (exp zl)(exp Z2);(e) lexp (-z2)1 is not
bounded for all z.
8. In Seco55, prove (a) identities (5); (b) identities (6).
9. Prove that (a) sin z = O if and only if z = :tmr; (b) sinh z = O if and only if z =
:tmri, wheren = O,1,2,....
I
I 10. Prove that tanh z is analytic except at the points z = :t(n - !)ni, where n = 1,
2,... .
11. Show that (a) tan (z + n) = tan z; (b) tanh (z + ni) = tanh Z.
12. Establish the inequalities sinh Ixl ~ Isinh zl ~ cosh X.
13. Use the definition of f'(z) when dz = dx, then when dz = idy, to show that
f' = Ux+ ivx= Vy- iuywhen f'(z) exists. Hence deduce that the components u and v
necessarily satisfy the Cauchy-Riemann conditions at points wheref'(z) exists.
I 14. In polar coordinates, where z = reiO,r > O, verify that the Cauchy-Riemann
conditions take the form
I ou 1 ov,
or = -;: 00'
I
and that f'(z) = e-iO[u,(r,O)+ iv,(r,9).
56 CONTOUR INTEGRALS
Consider first a complex-valued function G of a real variable t, with com-
ponents V and V,
G(t) = V(t) + iV(t).
We call it sectionallycontinuous over a bounded interval if both the real-

valued functions V and V have that property. Its derivativeis V'(t) + iV'(t)
wherever V' and V' exist. Its integral
(1)
f G(t) dt = f V(t) dt + i f V(t) dt
has the properties of integrals of real-value functions as to integrals of sums,

multiplication by a constant, reversal of limits of integration, and integra-
tion by parts. Also, when G is sectionally continuous on the interval
a < t < b, it is absolutely integrable over the interval and
(2)
If G(t)dtl ~ f'G(t)'dt.
The limit b ofthe integral (1)can be replaced by <X)ifthe corresponding
integrals of V and Vexist. The comparison test for convergence of integrals
(Prob. 14, Seco5) is easily extended to the improper integral of G. Thus if
G is sectionally continuous over each bounded subinterval of the half line

t > a and if IG(t)1~ M(t) where the real-valued function M is integrable
from ato 00, then G is integrable from a to oo. In particular, G is integrable
there if IGIis integrable,and a proof1of inequality (2)can be extendedto the
improper integral to write
(3)
Ir" G(t) dtl ~ LX) IG(t)1dt.
Now let e denote either an are of a smooth curve
(4) x = q,(t), y = Ijt(t),
where q" Ijt, q,', and Ijt' are continuous and the derivatives do not vanish
simultaneously for any of the values of t, or a curve formed by joining a
finite number of such ares endwise. Then e is called a contour. If e is closed
and do es not intersect itself, it is called a closed contour; boundaries of
triangles, rectangles, and circles are examples.
Ir the parametric equations (4) represent a contour e over which
a ~ t ~ b and if j(z) is sectionally continuous over e, that is, if the function
j[q,(t) + iljt(t)], a < t < b, is sectionally continuous, then the contour
integral of j over e can be defined in terms of an integral of type (1), namely
(5)
fe j(z) dz = lb f[ q,(t) + iljt(t)][q,'(t) + iljt'(t)]dt.
The integrand is a sectionally continuous function of t. Inequality (2)

enables us to write
(6)
lfe j(z) dzl ~ fe If(z)lldzl~ ML,
if If(z)1~ M for all z on e, L is the lengthof e, and Idzlis the elementof
are length of e :Idzl= Iq,'(t) + W(t)1 dt = ds.
Using the components u(x,y) and v(x,y) ofj(z), we can write the contour
integral (5) in terms of real-line integrals over e as follows:
(7)
fe j(z) dz = fe (udx - v dy) + i fe (u dy + v dx),
a result that is arrived at formally by writingf = u+ iv and dz = dx + i dy,

then expanding the productj(z) dz.
I Churchill, R. V., "Complex Variables and Applications," 2d ed., p. 96, 1960.
57 INTEGRALTHEOREMS
We now state basic theorems on contour integrals of analytic functions.
Cauchy integral theorem lf afunctionfis analytic at all points interior to

and on a closed contour C, then
(1) fe j(z) dz = O.
A simply connected domain D is an open-connected region such that
every closed contour within it endoses only points of D. The interior of a
closed contour and the entire z plane are examples, but the exterior of a
closed contour is not simply connected, nor is the domain between two
concentric cirdes.
The Cauchy integral theorem permits C to be any closed contour
within a simply connected domain throughout which f is analytic. The
theorem is easily extended so as to permit C to be replaced by the boundary
B of certain multiply connected regions, as follows.
Let Co denote a closed contour, and CI, C2, ..., Cn a finite number
of closed contours within Co with no interior points in common (Fig. 64).
Let R be the closed multiply connected region consisting of all points on
and within Co except for points interior to Cj(j = 1,2,..., n). Let B denote
the boundary of R, consisting of Co and all Cj described in such a direction
that the regio n R lies on the left of the boundary. Then iff is analytic at all
points of R, it is true that
(2)
fB f(z) dz = O.
The value offat each point interior to a closed contour C is determined
by the values on C by the following theorem.
Cauchy's integral formula Let f be analytic everywhere within and on a

closed contour C. Then if Zois a point interior to C,
(3)
1
f(zo) = ---: -
if(z)
2m e z - Zo
dz,
where the integration is counterclockwise around C.
y
o Fig.64
Here, too, e can be replaced by the boundary B of a multiply connected

region under the conditions prescribed for formula (2).
Derivatives of the integral (3) with respect to Zo exist and are given by
the integral of the derivative of the integrando In particular,
"
f (Zo) --
(4) 1
-
f f(z)
ni e (z - zo)
3 dz.
Consequently the derivative f' is analytic wherever f is analytic; thus the

derivative <n) of any order is analytic there.
Ir Zo and z are points of a simply connected domain D over which f is
analytic, then a contour integral between the points,
(5)
f z
ZO
f(~) d~ = F(z),
has a value F(z) independent of the contour e as long as e les entirely

within D. Moreover that single-valuedfunction F is analytic over D. It is
an antiderivative,or indefiniteintegral, off:
(6) F'(z) = f(z) for all z in D.
Any two antiderivatives off differ by a constant. In terms ofan antiderivative
F of the analytic functionf, we can write
ZI
fZO
f(z) dz = F(Zl) - F(zo)
over any contour interior to D from point Zo to point z1.
58 POWER SERIES
Let f be analytic over the interior of a circle Iz - zol = ro. Then at each
interior point z it is true that
(1)
that is, Taylor's series converges whenever Iz -

zol < ro and its sum is f(z).
That theorem is a consequence of Cauchy's integral formula.
The region of convergence of any power series
00
(2) L an(z -
n=O
zo)n,
where an are complex constants, is always a disk about the point Zo,
-
Iz zol < ro, unless the series converges only when z = Zoor for all z. Let
f(z) denote the sum of the series. The series can be differentiated term by
172 SEC, 58] OPERATIONAL MATHEMATICS
term to give a power series that converges over that disk to j'(z). Thus
series (2) represents an analytic function over that disk; in fact, it is Taylor's
series representation, about the point Zo, of its sum f(z). Thus
l
an = ,f(nJ(zo) (n = 1,2,. ..).
n.
The power series (2) converges absolutely and uniformly with respect to z
over each interior disk Iz - zol ~ rl, where rl < ro.
Since is an entire function, its Maclaurin series (zo = O),
ex> l
(3) e% = l + _zn (lzl < (0),
n= 1 n!
represents that function for all z. Consequently,the function g(z) = exp (Z2)
has the power series representation
(lzl < (0),
which must be Maclaurin's series for g, so it follows that g(2nJ(0)= (2n)!jn!

and g(2n-1)(0) = O.
With the aid of Taylor's series we find that the zeros of an analytic
function are isolated. More precisely, if a function f is analytic at a point Zo
and not identically zero over some neighborhood of that point, then there
is a neighborhood of Zosuch that f(z) :F Oat any of its points except possibly
Zo itself.
A generalization of Taylor's series also follows from Cauchy's integral
-
formula. If f is analytic over an an,nular region R, r1 < Iz zol < ro, it is
represented there by Laurent' s series:
ex>
(4) f(z) =
n=
L- co
Aiz - zot
where, if e is any closed contour described counterclockwisearound the

annulus R (Fig. 65), .
A l
(5)
n = 2ni c(zJ -f(z)zot+1 d
Z
(n = O,:t 1, :t2,.. .).
We note in particular that
A_1 = 21 . _1tlJe f(z)dz.

Taylor's and Laurent's series can be integrated term by term within
their domains of convergence. It follows that a representation of f by a
Fig.65 o x
convergent series of form (4) is necessarily the Laurent series, and hence
the values of the integrals (5) can be seen from such a representation.
For example, from Maclaurin's series (3)for eZwe can write this Laurent
series representation:
(6) exp ~= 1+ ~Z + ~2! ~Z2 + ~3! ~Z3 + ... (lzl > O).
( Z)
Here A_1 = 1, A_2 = 1, and Zo= O. According to formula (5), if e is a
closed contour described counterclockwise around the origin, then
fe exp (~) dz = 2ni, fe z exp (~) dz = ni.

The region of convergence of a series of type (4) is always an annulus
with center at Zo. The series represents f as a sum of two functions: the sum
of the series of nonnegative powers of z - Zoand the sum of the series of
negative powers. The first function is analytic when Iz - zol < ro and the
second one when Iz - zol > r1.
Two power series of type (2) in which Zo is the same for both can be
added, multiplied, or divided to produce series of that type which converge
to the corresponding combinations of the sums of the series, under natural
restrictions on the regions. Those operations apply to some types of Laurent
series too,ofcourse,as wecansee by replacingz - Zoin series (2)by(z - ZO)-1
or by multiplying the series by a negative power of z - Zo. But in the general
case the product of two Laurent series is a double series.1
59 SINGULAR POINTS AND RESIDUES
Ir a function is analytic at some point in every neighborhood of a point Zo,

but not at Zo itself, then Zo is called a singular point of the function. Ir the
function is analytic at all points except Zo, in some neighborhood of Zo, then
1 For properties of double series see, for example, E. Hille, "Analytic Function Theory," vol. 1,
pp. 114 ff., 1959.
Zo is an isolated singular point. The function (Z2 + 1)-1, for example, is

analytic everywhere except for the two isolated singular points Z = :t i.
About an isolated singular point Zo a function always has a Laurent
series representation:
A-I A_2
(1) f( Z) = -Z - Zo
+
(z - Zo)
2 + . .. + Ao + Al (Z - Zo) +...
(O< Iz - zol < ro),
where ro is the radius ofthe neighborhood in whichfis ana1ytic except at Zo.

The coefficients An are given by formula (5), Seco58, where the inner radius
r 1 of the annu1us, around which C is described, is now zero. In particular,
(2) A_1= _21. f f(z)dz,

1t)c
where C can be any dosed con tour, described counterdockwise, containing

Zoin its interior and such that, except for Zo,f is analytic within and on C.
The complex number A - 1 ,the coefficientof(z - zo)- 1 in the expansion
(1),is called the residueoff at the isolated singular point zo; 21tiA- 1 is the
value of the integral of f in the positive direction around a contour that
endoses no other singular points but Zo. The expansion (6), Seco 58, for
instance, shows that the residue of the function exp (I/z) at the point z = O
has the value A-I = 1.
Ir C is a cIosed contour within and on which f is analytic except for a
finite number of singular points Z1, Z2, .. . , Zm,interior to the region bounded
by C, the residue theorem states .that
(3)
fe f(z) dz = 21ti(P1 + P2 + . . . + pm>,
where Pn denotes the residue off at Znand where the integration is in the
positive direction around C. Note that the singular points Znare necessarily
isolated because of their finite number.
In the representation (1) the series of negative powers of (z - zo) is
called the principal part of f(z) about the isolated singular point Zo. The
point Zois an essential singular point offifthe principal part has an infinite
number ofnonvanishingterms. It is a poleofordermirA - m =1= Oand A-n = O
when n > m. It is called a simple pole when m = 1; thus, if Zo is a simple pole,
a number ro exists such that
(4) (A_1 =1=

O,0< Iz - zol < ro).
The function exp (I/z) has an essential singular point at the origin, in
view of its representation (6), Seco58. The function
cos z l 00 (- I)n l l Z2
-=-
z2
L:
Z2n=O (2n)!
-Z2n=___+-+...
Z2 2! 4!
(lzl > O)
has a pole of order 2 at z = O,where the residue of the function is zero.

If a function is not analytic at Zobut can be made so by merely assigning
a suitable value to the function at that point, then Zois a removable singular
point of the function. Thus the function
sin z Z2 Z4
-=1--+--... (lzl > O)
z 3! 5!
is analytic when z :Fo. Ir we define its value to be unity when z = O,the
function is represented by the above convergent power series for all z and
is therefore entire. Consequently z = O is a removable singular point of
Z-1 s.inz.
Whenfhas a pole of order m at Zo, let us write
(5) (O< Iz - zol < ro).
From the Laurent expansion (1) offabout Zoit follows that Zois a removable
singular point of 4>,that 4>is analytic at Zoif we make the definition
(6)
Then <pis represented by Taylor's series when Iz - zol < ro, and it can be
seen from Eq. (5)that
(7)
This is a useful formula for the residue off at a pole. For a simple pole
(m = 1) it becomes
(8) A -1 = 4>(zo) = lim (z -
%~%o
zo)f(z).
-
We also note that If(z)1 OCJas z -
Zo whenever Zo is a pole.
Conversely, for a givenj, suppose that a positive integer m exists such
that the function 4>,defined by Eq. (5), is analytic at Zowhen 4>(zo)is suitably
defined and that this value 4>(zo):F O. Then Taylor's series for 4>shows that
f has a pole of order m at Zoand that the residue offthere is given by formula
(7), or by (8) if m = l. For example, if
e-Z e-Z - l
f(z) = Z2+ n2 = z + ni z - ni
then corresponding to the singular point z = ni,

e-Z
4>(z)= ~
z+m
andm = I,sothatniisasimplepole. Here4>(ni)
==e-1tij(2ni);thustheresidue
of j at this pole is ij(2n).
Let functions p and q be analytic at Zo, where p(zo) # O. Then the
function
(9) j(z) = p(z) = p(z)

q(z) q(zo) + q'(zo)(z - zo) + . . .
has a simple pole at Zo if and only if q(zo) = O and q'(zo) # O. The residue
of j at. the simple pole is given by the formula
p(zo) .
(10) A-l = -,-(
q Zo)
Ir q(zo) = q'(zo) = O and q"(zo) # O, then Zo is a pole of j of order 2, and
conversely. SimilarIy for poles of higher order. Since formulas for residues
in terms of p and q are awkward when m > 1, we rely on formula (7) or the
Laurent expansion instead.
PROBLEMS
1 . When e is the boundary of a square with opposite vertices at the points z = Oand
z = 1 + i, evaluate the integral f e Z2dz directly in terms of realline integrals to show that
its value is zero.
2. When e is the circle Izl = 2 described counterclockwise, use Cauchy's integral
theorem or Cauchy's integral formula to prove that
(a) r~
Jcz + 9
dz = O;
(b) fe sec~dz = O;
(e)
i_Z2 + 4
cZ- 1 dz = IOni; (d)

i_ sinhz
cZ + 1tI
2
.dz = n.
3. Establish these expansions in the regions indicated:

1 ao
(a) - = I (-l).z"
1 + z .=0
(Izl < 1);
1 ao
(b) ~=
1 z - I
.=0
Z2. (lzl < 1);
ao Z2.-
(e) sinhz = I
n=1
(lzl< (0);
Z ao 1
(lzl> 1).
(d) z - 1 = .~o?
4. Showthat the functiontan z is analyticexceptfor simplepoles at z = :t(n - t)1t

and that its integral around the square bounded by the lines x = :t 2, Y = :t 2 has the
value -41ti.
5. Show that the integral of tanh z around the circle Izl = 21t has the value 81ti.
6. Showthat the functionssin(I/z) and cos(l/z) are analyticexceptat z = Oand that
z = Ois an essential singular point. Find the residues there. Ans. 1; O.
7. Find the residues of these functions at their singular points:
1 . 1
(a) zz +
- l.l' (b) sinz' (d) Z3cosh?;
Ans. (a) 2; (b) :t 1; (e) -sin Zo; (d) t; (e) O.

8. If e is the circle Izl = 8 described counterclockwise, show that
_21 '
i eZl
-:--
1tI e Sinh z
dz = 1- 2 cos 1tt + 2 cos 21tt.
9. If e is the circleIzl = 4 described counterclockwise, show that
10. Show that the singular point z = Oofthe function

1
f(z) = sin(1tlz)
. is not isolated.
11. With the aid of the Cauchy-Riemann conditions prove that the components u and
v of an analytic function f are harmonic functions, that is, they are continuous with
continuous partial derivatives up to the second order and satisfy Laplace's equation
60 BRANCHES OF MULTIPLE-VALUED FUNCTIONS
When z :f=O,the function
(1) zt = ~(cos~ + iSin~)

has one value corresponding to a particular choice of Oand a second value
when that argument is increased by 2n. Those two values of zt, which differ
only in algebraic sign, are the only possible values for a given z; thus the
function (1) is double-valued.
A branch of a multiple-valued function f is a single-valued function
that is analytic in some region and whose value at each point there coincides
with one ofthe values oflat the point. The function
(2) 11(Z) = ~(CoS~ + iSin~) (-n < O< n, r > O),

for example, is a branch of the double-valued function (1). The definition
of the derivative can be used to show that 11 is analytic everywhere exce~t
at the origin and points on the negative real axis. Since 11(z) tends to i'-;'
when O-. n and to - i~ when O-. -n, the function has no limit as
z -. -r (r> O). Thusl1 cannot be defined on the negative real axis so as to
make the function continuous there, and the ray O = n must be excluded
from the region of analyticity (Fig. 66).
The negative real axis is called a branch cut of 11' a boundary that is
needed to define the branch in the greatest possible region. Note that each
point of the branch cut is a singular point of 11 that is not isolated. The
origin is a branch point, a point associated with the multiple-valued function
in this way: each branch of the function has a branch cut running from that
point. Since the function (1) is double-valued unless the range of Ois limited .
to the value 2n or less, its branches must have cuts running out from the
ongm.
A second branch of the function (1) with the same cut as 11 is
(3)
liz) = ~(cos ~ + i sin~) (n < O< 3n, r > O).
Here 12(Z)= -11 (z). The number of branches is unlimited. A branch with
the positive real axis as its branch cut is
(4) (O< O < 2n, r > O);

liz) = ~(cos~ + iSin~)
one that has the positive y axis as a cut is the function (1) wi;:hthe restriction
n/2 < O < 5n/2, r > O; etc.
A branch of the n-valued function zl/n (n = 2,3, . . .) with a ray O = 00
as the branch ~ut is described by the conditions
(5) zl/n = ~(cos ~ + i sin~) (00 < O < 00 + 2n, r > O).
o x
Fig.66
FUNCTIONS OF A COMPlEX VARIABLE [SECo 61 179
For this function it can be shown that

d 1 ZI/.
_(ZI/')
dz = -n -.z
The logarithmic function log z, the inverse of the exponential function,

is infinitely multiple-valued. It can be written
(6) log z = log (reiO) = Log r + iO,
where Log r denotes the real natural logarithm of the positive number r.
The branch known as the principal value of log z is the function
(7) Log z = Log r + i(} ( -n< () < n, r > O),
whose branch cut is the negative real axis. Its derivative is l/z.
When e is a complex number, we define zCthus:
(8) ZC= exp (clog z) (z :/=O).
Unless e is an integer, the function ZCis multiple-valued. Ir e = m/n where
m and n are integers and n :/=O, :t 1, ZCbecomes the n-valued function zm/..
Branches of ZCcan be written by using branches of log z in definition (8).
The inverse trigonometric and hyperbolic functions, which can be
written in terms of logarithms, are further examples of multiple-valued
functions.
61 ANAL YTIC CONTINUATION
Ir a function is single-valued and analytic throughout a region, it is uniquely

determined throughout the regio n by its vales over an are, or over a sub-
region, within the given region. That theorem can be proved with the aid
of Taylor's series.
Letfl denote a given function analytic in a region R l' and let R be a
greater region containing R1. A functionf, analytic throughout R and equal
to fl whenever z is in R 1, may exist. Ir so, there is only one such function in
view ofthe foregoing theorem. The functionfis called the analytic continua-
tion of fl into the larger region R.
As an example, let fl be definedby this power series:
00
fl(Z) I z'
= .=0 (lzl< 1).
Then f is analytic in the region Izl < 1,the region of convergence ofthe series;
but the function is undefined for all other values of z because the series
diverges whenever Izl ;;;;;1. The series is the Mac1aurin series representing
the function (1 - Z)-1 in the region; thus f(z) = (1 - Z)-1 when Izl < 1.
This second representation of f1 discloses the analytic continuation

1
f(z) = 1- z (z ~ 1)
off1 into the entire z plane, excluding the point z = 1,becausefis everywhere
analytic except at that point and f(z) = f1 (z) when Izl < 1.
As another example, consider the Laplace transform
f1(Z) = {O e-zl dt = 1'" e-XIcos yt dt - i {O e-J<Isin yt dt.

The Laplace integral here exists only if x > O,where its value is l/z. Thus
f1(Z) = l/z (x > O),and the function is analytic in that right half planeo But
since l/z is ana1ytic everywhere except at the origin, the ana1ytic continuation
of f1 out of the half plane is the function
1
f(z) =z (z~ O).
We also note that ifthe Laplace integral were known to represent an analytic
function in the half plane and to have the value l/x when z = x there, then
the integral represents l/z there because the latter function is analytic and
has the values l/x on the x axis.
Finally, consider the branch (2), Seco 60, of the function zt. In the
half plane y > O it is given by the conditions
(O < 8 < n, r > O).

f1(Z) = fi(cos~ + iSin~)
It is not analytic on the negative real axis. The function
[(z) = fi(cos ~ + i sin~) (O < 8 < 2n, r > O)
is ana1ytic except on the positive real axis, and its values coincide with those
off1 in the half plane y > O. Thus f is the analytic continuation of f1 down-
ward across the negative real axis.
The following principie of rejlection is easily established with the aid
of the Cauchy- Riemann conditions. Let a function w = f(z) be analytic in
. some region R that includes a segment of the x axis and is symmetricwith
respect to that axis. Hf(x) is real whenever x is a point on that segment, then
(1) f(z) = w
wheneverz is in R. Conversely,if conditionl!lis satisfied,thenf(x) is real.
Condition (1) can be written f(z) = f(z), or f(z) = f(z).
As examples, the entire functions Z2, eZ, and sin z are real when z is
real, and the complex conjugate of each function is the same as the function
of Z. But the entire functions iz and Z2 + i are not real when z = x; their
conjugates are different from iz and Z2 + i.
62 IMPROPER CAUCHY INTEGRALS

In the following chapter we shall use extensions of the Cauchy integral
formula and Cauchy's integral theorem in which the closed contour C is
replaced by a straight line x = y, the boundary of a half plane x ~ y over
which the functionf is analytic. The following proof of those extensions will
serve to illustrate a prominent type of application of contour integration.
First we introduce the notion of order of magnitude of a function of z
for large Izl. Order conditions serve our purpose more fully than the concept
of analyticity at the infinite point of the complex planeo
A function f is of the order of i', written (!J(i'), as Izl -+ ex:>in a specified
part ofthe z plane, if positive numbers M and ro exist such that Iz-kf(z)1 < M
there whenever Izl > ro; that is, when Izlis sufficiently large,
Theorem 1 Let f be analytic over a right half plane Re z ~ y and (!J(Z-k)

there as Izl -+ 00, where k > O. Then ifRe Zo > y, f(zo) is given by the
improper Cauchy integralformula
l Y+;oo f(z) l oo f(z)
=
2m fY-;OOz - 21tf-00 Z - Zo
(1) f(zo) ; -dz = -- -dy,
Zo
where the integration is along the line x = y; thus z = y + iy.

To prove the theorem, let CR denote the arc x ~ y of a circle Izl = R
where R > Iyl and R > IZol,so that the point Zo is interior to the region
bounded by the arc and the line x = y (Fig.67). Ifwe write p = JR2 - y2,
-y+i/3
R x
Fig.67 -y-i/3
then according to Cauchy's integral formula,
(2) f(zo) = ~ [f
2m CRz
f(z)
- Zo
dz -
j Y+ifJ f(z)
y-ifJ Z - Zo
dZ
],
where C R is described in the positive sense.
Whenpoint z is on CR,then Iz - zol~ R - IZol and, since If(z)1<
Mlzl- k when Izl = R and R is sufficiently large, it follows that
f(z) M 1
- ro.
The integrands of the integrals in formula (2) are continuous. The length of
CR is less than 2nR; therefore (Sec. 56)
I fCRz
f(z)
- Zo z
d
I
2nRM
< Rk(R- IZol)- Rk(l - IZolR
-1)
- 2nM
and the last member clearly tends to zero as R -+ 00,since k > O. It also
tends to zero as p -+ 00, because R2 = p2 + y2. Sincef(zo) is independent
of p, we see from formula (2) that
1 . Y+ifJ f(z)
(3) f(zo) = ---; bm
j
2m fJ~oo y-ifJ Z - Zo
- dz;
1 . f(y + iy) d
+ Jp f(y
.- iy) d
].
fJ
= -- l 1m . y y
2n fJ~ 00 [J o Y + IY - Zo o Y - IY - Zo
The integrands of the last two integrals are of the order of 1/1+ 1 as
y -+ 00,and sincek + 1 > 1, the improper integrals, as p -+ 00, exist and
(4) f(zo) = -~
2n [i oo f(Y. + iy) dy +
o y + IY - Zo I o f(Y. + iy) dY
-00 y + IY - Zo ] ,
This is the extension (1) of Cauchy's integral formula.

Formula (1) states that the integral of the function
cP(z)= f(z)
z - Zo
along the line x = y is the residuef(zo) of cPat its singular point Zo. The proof
was based upon the analyticity of cP(z)(z - zo) throughout the half plane and
the fact that cPis (f}(l/zk+ 1) there. But Taylor's series.,representation of the
analytic functionf about the point Zo shows that Zo is a removable singular
point of cPif and only iff(zo) = O. Thus if cP is also analytic at zo, its integral
along the line x = y is zero, the value f(zo), where f(z) = (z - zo)cP(z), and
we have established the following extension of the Cauchy integral theorem.
Theorem 2 Let cjJbe analytic over a half plane x ~ y and (9(Z-k) as Izl -+ 00
there, where k > l. Then
Y+iOJ OJ
(5)
f y-iOJ cjJ(z)dz=i f -OJcjJ(Y+ iy)dy=O.
The proof of formula (5) applies as well when cjJis analytic and (9(Z-k)
in the left half plane x ~ y, if k > 1. Also, the residue theorem (Sec. 59) can
be extended in the same manner to the case in which cjJhas a finite number
of singular points interior to either half planeo For the left half plane that
extension of the residue theorem can be stated as follows.
Theorem 3 Let cjJ be analytic when x ~ y except for n singular points

z 1 , Z 2, . . . , zn interior to that half plane, where cjJhas residues P 1 , P2 , . . . ,
Pn. Then if for some real numbers M, k, and ro, where k > I and ro
exceeds the greatest of the numbers Iz11,Iz21,. .. ,Iznl, cjJsatisfies the
order condition IcjJ(z)1< Mlzl-k when Izl > ro in that half plane, it is true
that
Y+iOJ
(6)
f Y-IOJ
. cjJ(z)
dz = 2ni(pl + P2 + . . . + Pn).
As examples, consider first the function f(z) = e-z/z whichsatisfies

the conditions in Theorem 1 when y = 1, for it is analytic over the halfplane
x ~ I and (9(I/z) there because
I e-1
when x ~ 1.
If(z)1 = Gje-x ~ -g-
Thus formula (1) applies to f, giving the integration formula
l+iOJ e-Z dz = --e2ni -zo
f 1- iOJz(z - zo) Zo
when Re Zo > l.
The function cjJ(z) = z-2 satisfies the conditions in Theorem 2 when

y = 1; therefore
(7)
f
1+iOJdZ
l-iOJ
2 =
Z
"
I
f OJ
-OJ (1
dy
. 2 = O.
+ IY)
When t is real and t ~ Oand y = 1, the conditions in Theorem 3 are
satisfiedby the function eZl/(Z2+ 1)with simple poles :!:i; therefore
1 1 + OJ eZt eit e - it .
2ni f 1- OJZ2 + 1
(8) - -dz=---=smt (t ~ O).
2i 2i
As we shall see in the following chapter, the left-hand member of formula (8)
is an integral representation of the inverse Laplace transform of 1/(s2+ 1).
PROBLEMS
1. Use Maclaurin's series for cos Z to prove that the function cos.j; is single-valued
and entirewhendefinedas l at z = O.
2. Point out why the function
gl(Z) = exp(-~cos~)[ cos(~sin~) - isin (~sin~)J

where r > O and O< () < 2n, is a branch (Sec. 60) of the double-valuedfunction
exp(-.j;) that is analyticeverywhereexceptat pointson the realaxiswherex ~ O.
3. The branchg1ofexp(-.j;) describedin Probo2 is not analyticalongthe positive
real axis. Showwhy its analyticcontinuation from the first quadrant into the fourth
quadrant is the branch
g2(Z) = exp(-~cos~)[ cos(~sin~) - isin (~sin~)J

wherer > Oand -!n ~ ()~!n.
4. (a)If)/ > O,provethat the branch g2ofexp( -.j;) describedin Probo3 is (!)(z-~as
IzI--. ex)in the half plane x ~ )/ for each fixedpositivenumber k. [Note that Ig2(z)1
<
exp(-..fifi) there.] (b)ApplyTheorem2, Seco62,to showthat
f Y+lOO z"g2(Z)
y-loo
dz = O (n = 0,1,2,...;)/ > O).
5. Show that the complex-valued Laplace transform of sin t,
(Re z > O),

fl(z) = L"" e-Z'sintdt
is analytic over the halfplane x > Oand that its analytic continuation into the rest ofthe
z plane,excludingthe points :t i, is the functionfez) = 1/(Z2+ 1). .
6. Use Theorem 1, Seco62, to prove that, if k is real and positive,
f-100~-_?:!.
lOO
Z2- k2- k
(k > O).
Verifythat resultby integratingwith respectto y.

7. In the half planex ~ l showthat
coshx -<
Itanhzl -< -:-- - l
h- s10 x-tan h l'
and that the function Z-1 tanh z satisfies the conditions in Theorem 1, so that
l tanh z
-- f l +ioo
dz=-tanh Zo
ifRe Zo > 1.
2ni 1-100z(z- zo) Zo
8. ApplyTheorem2 to prove that
I +100e"
2dz = O ift ~ O.
S1-100 Z
FUNCTlONS OF A COMPlEX VARIABLE [SECo 62 185
9. Show that the integral (8), Seco62, vanishes if t ~,O.

10. Use Theorems 2 and 3 to prove that, if'Y> O,
1 !sin t - tt cos t ift ~ O,
;
f
Y+iOO
2m y-ioo(z
Z
e"
+ 1)
z dz
-
- {O ift ~ O.
Thus when t ;;;O,the left-hand member represents L -1{(SZ + l)-Z}.
11. Prove Theorem 1, Seco62, by using open rectangular contours CfJwith sides
along the lines x = p, y = ::!:P (Fig. 68) in place of the circular arcs C R .
.
+iP lP+iP
o 'Y
+-. c
{3-i{3
Fig.68
12. Use the residue theorem (Sec. 59) and give details of the proof of Theorem 3,
Seco62.
6
The Inversion Integral
We shall now extend our theory ofthe Laplace transformation by letting the
letter s in the transform f(s) represent a complex variable. As before F(t)
represents a real-valued function of the positive real variable t; but the
transformf(s) can assume complex values. We shall see that properties ofthe
transformation airead y obtained by assuming that s is real carry over to the
case in which s is complex.
63 ANALYTIC TRANSFORMS
When s is a complex variable,
s = x + iy,
the Laplace transform of a real-valuedfunction F(t),
(1)
186
THE INVERSION INTEGRAL [SECo 63 187
can be written in the form
(2) f(s) = u(x,y) + iv(x,y),

where the components u and v are real integrals:
u(x,y) = {O e-XI cos ytF(t) dt,

(3)
v(x,y) = - {O e-xI sin ytF(t) dt.

Since the integrands of the integrals (3) are the real and imaginary
coefficientsof the integrand e-SIF(t)of the Laplace integral (1),we note that
le-XI cos ytF(t)1 ;;:;;
le-SIF(t)l,
(4)
1- e-xI sin ytF(t)1 ;;:;;
le-SIF(t)1.
These inequalities will be useful in discussing the convergence of the Laplace
integral.
Under broad conditions on F its transform f(s) is analytic over a right
half plane and bounded there. Let F be sectionally continuous over each
intervalO < t < T and of exponential order as t
and a exist such that IF(t)1< Mea.1when t ~
- oo. Then constants M
O. Consequently whenever
x ~ xo, where Xo > a,
(5) le-SIF(t)1= e-X/IF(t)1< M exp [-(xo - a)t].
The integrands of the real-valued integrals (3)are continuous functions of x, y,

and t except for possible finitejumps of F(t). Their absolute values are less
than M exp [ -(xo - a)t] when x ~ Xo, in view of inequalities (4) and (5).
According to the Weierstrass test, then, the integrals (3) converge uniformly
with respect to x and y on the half plane x ~ Xo. Thus they represent con-
tinuous functions u(x,y)and v(x,y) there (Sec.15).
When F is (!)(~/), we noted earlier (Sec. 19) that tF(t) is (!)(expXlt)
whenever Xl > a. It follows that
(6) (x > a),

uix,y) = {D e - -
XI( t) cos ytF(t) dt
because the integral here converges uniformly with respect to x and y when
x ~ xo(xo > a); also Uxis continuous there. From the second of Eqs. (3)
we see that the integral (6) also represents v.(x,y); hence u and v satisfy the
first Cauchy-Riemann condition over the half plane x > a. In the same
manner we find that the second condition Uy = - Vxis satisfied there. Those
conditions along with the continuity of the first-order partial deriva tives of u
and v establish the analyticity of f over the half plane x > a (Sec. 54).
Now l' = Ux+ ivx wherever f is analytic; thus when X> rx
I'(S) = - foO:>
te-xt(COS yt - i sin yt)F(t) dt = L{ - tF(t)}.
Since tF(t) is also sectionalIy continuous and of exponential order, our
result applies as welIto that function; thus
(x> rx),
and so on for <nJ(s).

With the aid of condition (5) we find that f is bounded over the half
plane x ~ Xosince (Sec. 56)
o:> o:> M
l/(s)1 ~ e-xtIF(t)1 dt < M e-(xo-cr)t dt = -.
f.o f.o Xo - rx
SimilarIy, we find that II'(s)1 < M/(xo - rx)2,etc., so the derivatives of f are
bounded over the half planeo
FinalIy we note that f(s) is real when s = x because F(t) is real, and it
folIows from the principie of reflection (Sec. 61), or directIy from Eqs. (2) and
(3), that f(s) = f(s). Our results can be stated as folIows.
Theorem 1 Let a real-valued function F be sectionally continuous over each

intervalO < t < T and (9(ec1t)
as t -+ oo. Then its transform
(s = x + iy)
f(s) = {O:>e-stF(t) dt = L{F}
is an analytic function of s over the half plane x > rx. Its Laplace
integral is absolutely and unlformly convergent over each half plane
x ~ xo(xo > rx),and f and its derivatives are bounded there. Also,
(7) <nJ(s)= L{( -ttF(t)} (n = 1,2,...; x > rx),
(8) f(s) = f(s) (x> rx).
Formula (7) was derived in Seco19 when s is real.
The conditions in Theorem l can be relaxed. For instance F may
become infinite at t = to in such a way that (t - to)kF(t) remains bounded as
t -+ to, where k < 1. Then the conclusions in the theorem are still valid. As
.an example, the transform of t-t is analytic in the half plane x > O, and
formula (7) applies to it.
64 PERMANENCEOF FORMS
We have seen that the Laplace integral of F(t) represents a functionf(s) that
is analytic in a half plane x > ex,where s = x + iy. Ir the integration is
performed when s = X, a real function cp(x)is obtained that is identical with

J(s) to the right of point s = exalong the real axis; that is, cp(x)= J(x) when
x > ex.Ir cp(s)is an analytic function in the half plane, then it must be identical
with J(s), since two different analytic functions cannot be identical along a
line (Sec. 61).
It follows that transforms can be found by carrying out the integration
as if s were a real variable. That the function J(s) so found is analytic when
Re s > excan be seen in the particular cases; but it is true in general because
the integration formulas are the same whether the parameter in the integral
is complex or real. The transform of t2, for instance, was found to be 2s- 3
when s is real. Now t2 is (9(e<Xt)for any positive ex,and 2s-3 is analytic when
Re s > O.Therefore, L{ t2} = 2s - 3 for all complex s in the half plan e x > O.
All our transforms of particular functions, tabulated in Appendix A,
Table A.2, are valid when s is complex. We seldom need the value of ex
which determines the half plane in which s lies; the existence of the number ex
usually suffices.
The operational properties of the transformation developed in the first
two chapters and tabulated in Appendix A, Table A.t, are likewise valid
when s is a complex variable in some half plane x > ex. For the sake of
simplicity, we may make an exception of operation 10, Appendix A, Table
A.l:
LHF(t)} = 1(%)J(J.)dJ.,
and agree that s is real here.
The permanence of the forms of those properties is again a consequence
ofthe fact that the steps used in their derivations are independent ofthe real
or complex character of the parameter s. However, the derivations could be
rewritten when s = x + iy by taking the corresponding steps with the real
integrals (3), Seco63, that represent u(x,y) and v(x,y).
65 ORDER PROPERTIES OF TRANSFORMS

When Isl -+-00 in a half plane of convergence of the Laplace integral, the
behavior of the transformJ(s) is governed by regularity properties of the
object function F.
Theorem Z Let a jncton F as we/l as ts dervatve F' be sectonally con-

tinuous over each bounded intervalO < t < T and let F itself be (9(elXt).
Then in any half plane Re s ~ Xo where Xo > ex,its transJormJ(s) tends
to zero as Isl -+- 00:
(1) lim J(s) = O (Res ~ Xo > ex).
Isl-+(%)
For points s =x + iy in the half plane x ;;;Xo we are to prove that for
each positive number E there is a number '. such that
(2)
/f(s)1 < E wheneverIsl> '. and x ;;;Xo.
The Laplace integral of F converges uniformly with respect to s in the
half plane x ;;;Xo (Theorem 1), so the remainder for that improper integral
can be made uniformly small in absolute value for all s there. That is, to the
given number E there corresponds a number T. independent of s such that
(3) ILooe-SIF(t)dt! ;;; I f:' e-SIF(t)dtl + IR(s,T.)1

and IR(s,T.)I < tE for all s in the half planeo Now we shall determine '. such
that the first term on the right of condition (3) is less than tE when Isl > '..
The interval (O,T.) consists of a finite number of subintervals interior to
each of which F and F' are continuous and have limits from the interior at
the end points. Consider a subinterval (O,TI) as a sample. Over it we can
apply integration by parts to the complex-valued integral (Sec. 56); thus
TI
-s fo e-SIF(t)dt
TI (TI
= e-S'F(t)J o - Jo e-S'F'(t)dt.
The absolute value ofthe right-hand member does not exceed the function
(TI
IF(O)I + exp (-xT.)IF(TI)1 + Jo e-xtlF'(t)1dt

and, since x ;;;Xowhere Xo may be negative, and F' is bounded, IF'(t)1< NI,
this function is less than MI, where
MI = IF(O)I+ exp (lxolT.) [IF(T)I+ N T].
Similarly for the remaining subintervals of (O,T.). Thus a number M
exists such that
(4) when x ;;;Xo.

IsllLT e-S'F(t)dt/ < M
Then from condition (3)weconc1udethat
iflsl > 2M
E '
that is, condition (2) is established by selecting r. as 2M/E.
Weaker-order properties onf(s), such as the boundedness of xf(x + iy),
can be derived by assuming only that F itself is sectionally continuous and
of exponential order (Probs. 7 and 8). Stronger ones will now be noted in
case F or its derivatives are continuous and of exponential order.
Theorem 3 If F is continuous and F' and F" are sectionally continuous over
each intervalO ;;;t ;;;T while both F and F' are (9(e"'t),then
(5) lim sf(s) = F(O) when x ~ Xo > oc;

1.1 00
in particular,f is &(1/s) in that half plane:
(6) Isllf(s)1 < M (x ~ Xo > oc).
Under the conditions stated we know that
(7) L{F'(t)} = sf(s) - F(O).
Since F' satisfies the conditions on F in Theorems 1 and 2, then L{ F'} is

bounded over the ha1f p1ane x ~ Xl and tends to zero as Isl -> 00 there.
Therefore sf(s) is bounded and, since it equa1s L{F'} + F(O), its limit as
Isl -> 00 exists and is given by formula (5). .
As illustrations the functions 1,.cos t and t satisfy the conditions in
Theorem 3 whenever oc> O. Their transforms l/s, S/(S2+ 1), and 1/s2 are
(9(l/s) and satisfy condition (5). In fact the 1ast of those transforms is (9(1/s2),
a conc1usion that follows from properties of the function t with the aid of
the following extension of Theorein 3.
Theorem 4 If F and F' are continuous and F" and F'" are sectionally con-
tinuous over each intervalO ;;;t ;;;T while F, F', and F"are (9(e"'t),then
s2f(s) - sF(O)is bounded over any half plane x ~ Xo where Xo > oc;in
fact
(8) lim [s2f(s)
1.1 00
- sF(O)] = F'(O) (x ~ Xo > oc),
and the additional condition F(O) = O is then adequate and necessary

for f(s) to be (9(1/S2)in the half planeo
Since our formula for L{ F"} is valid here, then
L{F"(t)} + F'(O)= s2f(s) - sF(O)

In viewofTheorems 1 and 2 the function on the left is bounded and has the
limit F'(O)as Isl-> 00, so the same is true for the function s2f(s) - sF(O).
The second component -sF(O) of that function is bounded if and only if
F(O) = O, in which case s2f(s) is bounded over the half planeo
Extensions ofTheorem 4 are evident. For instance,f(s) is (9(1/s3)over
the half plane if F, F', and F" are continuous and the next two derivatives
are sectionally continuous, while F and its first three derivatives are (9(e"'t)
and F(O)= F'(O)= O;alsos3f(s)-> F"(O)as Isl -> 00 in the half planeo
PROBlEMS
1 . State why none of the following functions can be Laplace transforms of real-valued
sectionally continuous functions of exponential order (Theorem 1).
1 1
(a) s; (b) ssins
+ l'. (ti) 7;
S
(e)~.
S+I
2. Use Theorem 2 to show that none of the following functions can be Laplace
transforms of a function F of exponential order such that F and F' are sectionally
continuous.
s
(a) 1; (b) s - 2;
3. Determine an order property of the transform f(s) from the character of the
function F in each of the following cases, and verify by writing f(s). (a) F(t) = sin t;
(b) F(t) = cosh t; (e) F(t) = t sin t; (ti) F(t) = (t - 1)80(t - 1).
l l 1
Ans. (a) (!J2'
s x ~ Xo > O; (b) (!J-,x
s ~ Xo > 1; (e) (!J3'X
s ~ Xo > O.
4. Ir a function F is sectionally continuous over a bounded interval O < t < T and

F(t) = O whenever t > T, state why F(t) is (!J(e''')for every real IX,and why its transform
f(s) is an entire function of the complex variable s.
5. The step function l - 80(t - 1) satisfies the conditions on F in Probo 4. Show
directly that its transform can be written
f(s) = l - s e-s whenever s # O,and feO) = 1;
then represent f as a power series in s convergent for all s to verify that f is entire.
6. Ir a function F is sectionally continuous over some interval O < t < T but con-
tinuous when t ~ T, and ifF' is sectionally continuous over every such interval and both
F and F' are (!J(e"'),use the formulafor L{F'} to prove that the transformf(s) is (!J(I/s)
over the half plane Re s ~ Xo when Xo > IX.
7. Ir F is sectionally continuous over each interval O < t < T and (!J(e"'), prove that its
transform f(s) satisfies the condition that xf(x + iy) is bounded over the half plane
x ~ Xowhen Xo > IX;consequently
lim f(x + iy)
x-a>
= o.
8. According to the Riemann-Lebesgue theorem in the theory of convergence ofFourier
series,l if a function G is sectionally continuous over an interval a < t < b, then
lim
y co f G(t)cos yt dt = lim f
b
a y oo a
b G(t)sin yt dt = O,
a result that is plausible if graphs of the integrands for large values of y are considered.
, See the author's "Fourier Series and Boundary Value Problems," 2d OO.,p. 87, 1963.
(a) If G is sectionallycontinuous over each intervalO < t < T, and absolutely

integrable from Oto 00, extend the above result to improper integrals by showing that
lim
1,1->o i > e-i)'tG(t)dt = O.
(b) Write G(t)= e-X'F(t)where F is sectionallycontinuous over each interval
(O,T) and l!I(eat)to prove that the transform f(s) of F satisfies this condition when x is
fixed :
lim
1,1->
f(x + iy) =O (x > a).
66 THE INVERSION INTEGRAL
According to our extension ofCauchy's integral formula (Sec. 62), a function

f(s) that is analytic and (!J(S-k) over a half plane Re s ~ y, where k > O,can
be represented in terms of its values on the boundary of the half plane by a
line integral:
1 Y+iOOf(z)
(1) f(s) = ---; -
I
dz
2m y-ioo s - z
(Res> y).
Supposef(s) = L{F(t)}. Ifwe formally apply the inverse Laplace transforma-

tion to the functions of s in formula (1) and interchange the order of the
operation L - 1 and the integration, we find that
1 Y+iOO .
I f
eyl oo
(2) F(t) = ---; e'Zf(z)dz = - e'Y'f(y + iy) dy.
2m y-ioo 21t -<X> .
This is an integral transformation of f(s) that may represent in a direct

manner the inverse Laplace transform of that function.
The improper integral(2)existsifand onlyifthe integrals of eiy'f(y + iy)
from - 00to Oand from Oto 00 both existoFor some transformsf(s) those
improper integral s fail to exist, at least for certain values of t, while the
Cauchyprincipalvalueof the integral (2),
f
eyl (l.
(3) _
2 lim e'Y'f(y+ iy) dy,
1t (1 00 -(1
does existo This is illustrated in Probo 7, Seco70. Ir the improper integral (2)
does exist, its principal value exists and is equal to the improper integral.
The principal value (3) is called the complexinversionintegral for the
Laplace transformation. We use the symbol Li - 1 for that linear integral
transformation of f(s), a symbol that is intended to suggest an integration
as well as an inverse Laplace transformation; thus
1 Y+i(l
(4) Li - 1 {J(s)} = ---: lim
1 i(l e'Zf(z)dz.
2m (1 <X>Y-
y
-y+i(3
z=-y+iy
o x
-
-y i(3
Fig.69
We assume that f(5) = f(s), a condition that is satisfiedby transforms

of real-valued functions F(t) (Sec. 63). Then although the inversion integral
is an integral in the complex plane along the line x = y (Fig. 69), we can write
it as a real improper integral. Since z = y + iy, where y is fixed, the integral
in Eq. (4) can be written as .
ieyt[f~f/ Y'f(y + iy)dy + I: eiy'f(y + iY)dY].

When the variable of integration in the first integral here is replaced by - y,
the expressioninside the brackets becomes .
I: [e-iy'f(y - iy) + eiY'f(y + iy)Jdy.

The integrand of this last integral is 2 Re eiy'f(y + iy), because the complex
conjugate of eiy'f(y + iy) is e-iy'f(y - iy). We write
f(x + iy) = u(x,y) + iv(x,y);
then
Re [eiY'f(y + iy)J = u(y,y) cos yt - v(y,y) sin yt
and therefore
oo
eyt
(5) Li -1 {J(s)} =-
; o f [u(y,y)cos yt - v(y,y)sinytJdy.
When f(s) satisfies certain conditions, we shall see that the value of
the real integral (5) is independent of y for all values of y greater than a
prescribed number. Even for simple functions f, however, the integration
involved in the real form (5) is generally difficult. To evaluate the inversion
integral, we shalI use the complex form (4) in conjunction with the theory of
residues or processes of changing the path of integration.
In the following sections conditions on either J or F will be established

under which the inversion integral represents the inverse transform F(t) of
J(s).
Other formulas for the inverse transformation are known; 1 but the
inversion integral (4) has greater general utility than the others.
67 CONDITIONS ON I(s)
The following theorem gives conditions on a function J that are sufficient
to ensure that L- 1 {J(s)} exists and represents a function F whose transform
is J(s). Useful properties of F(t) also follow from the conditions on J(s),
properties that give the initial value F(O), the continuity and exponential
order of F.
Theorem 5 Let J be any Junction oJ the complex variable s that is analytic

and (9(S-k) for all s(s = x + iy) over a half plane x ~ ex,wherek > 1;
alsolet J(x) bereal whenx ~ ex.ThenJorall real t theinversionintegral
oJJ(s) along any Unex = y, where y ~ ex,convergesto a real-valued
Junction F that is independentoJy,
(1) (Id < ex),
whose Laplace transJorm is the givenJunction J(s):
(2) L{F(t)} = J(s) (Re s > ex).
Furthermore F(t) is (9(e"t),it is continuous ( - ex)< t < ex), and
(3) F(t) =O when t ~ O.

The integrand etzJ(z) of the inversion integral, where z = y + iy and
y ~ ex,is everywhere continuous in y and t sinceJis analytic when Re z ~ ex.
Thus the integral along each bounded segment Iyl ~ Yo of the line x = y
exists as a continuous function of t. According to the order condition on J,
positive constants M and Yo exist such that
(4) when Iyl > Yo.
Since k > 1, the improper integral s of Mlyl-k from Yo to ex)and from - ex)
to - Yo exist, so the inversion integral exists as an improper integral, and
it is equal to its principal value.
1 See books by Doetsch and by Widder listed in the Bibliography.
AIso, since f(x) is real, then, according to the reflection principIe for
analytic functions,f(z) = f(z), and the inversion integral takes its real form
00
eyt
(5) L-l{J(S)} = - [u(y,y)cos ty - v(y,y) sin ty] dy.
1t 1o
Now the components u(y,y) and v(y,y) of f(y + iy), as well as f itself, satisfy
condition (4). Thus the absolute value of the integrand of integral (5) is less
than 2My-k when y > Yo, a function that is integrable from Yo to 00 and
independent of t. Consequently, integral (5) converges uniforrnly with
respect to t, and it follows that L-1{J} is a continuous real-valued function
F(t) for all rt.al t.
To show that F(t) is independent of y when y ~ el, we use a second
path ofintegration x = y',where y' > y. Sinceetzf(z)is analytic when x ~ y,
the integral of that function around the boundary of the rectangle ABCD
in Fig. 70 vanishes, according to Cauchy's integral theorem. On sides BC
and AD, Izl ~ fJand, since If(z)1< Mlzl-\
M N(t)
letz' (z)1 < etx- <-
J fJk = fJk
where N(t) is the greater ofthe two numbers Mety and Mety'. The length of
-
those sides is y' y; hence the absolute value of the integral of etzf(z) along
either side BC or AD does not exceed N(t)(y' - y)fJ-k,a quantity that vanishes
as fJ -+ oo.
The vanishing of the contour integral around the boundary can be
expressed in the form
fAB
etzf(z) dz +
iCD
etzf(z)dz =
f AD
etzf(z) dz r etzf(z) dz.
- JBC
o 'Y x
A DI'Y'-iP
I I Fig.70
The left-hand member tends to zero as P-+ 00 because the member on the
right does so. But we have seen that the limit as P-+ 00 of each of the
integrals on the left exists when k > 1. It follows that
Y+iP Y'+IP
p
lim f
oo y-IP
e'Zf(z)dz = lim
p f
oo y'-IP
e'Zf(z)dz,
so the function F defined by Eq. (1) is independent of y.

Let us write y = IXin Eq. (1). Then
oo
l l
e-CZ/IF(t)1= -
2n I
f oo
- 00
.
eUYf(1X+ iy) dy ~ -
I 2n - 00f If(1X+ iy)1dy.
The last integral exists because f is lP(y-k) as Iy/ -+ 00 where k > 1, and f
is continuous. Its value is independent of t. Therefore F(t) is lP(eCZ/)
as t -+ oo.
To evaluate F when t ~ O,it is convenient to select a positive number
for y; this can be done since y can be any number in the range y ~ IX.Then
over the half plane x ~ y, le'zl = e'" ~ l when t ~ Oand
le'Zf(z)1< Mlzl-k where k > 1.
It follows from our extension of Cauchy's integral theorem (Theorem 2,
Sec.62)that
Y+loo
f Y-JC()
. e'Zf(z) dz = O (t ~ O);
that is, F(t) = Owhen t ~ O. In particular, F(O)= O.

Finally, the transform of F(t) must exist when Re s > IXbecause F is
continuous and lP(eCZ/), so
T oo
l
L{F(t)} = 2 lim
nT oo
_ f o
e-sI
f -00
e'Zf(z) dy dt (Re s > IX),
when z = IX+ iy. Since the convergence of the inversion integral is uniform
with respect to t, the order of integration can be interchanged:
oo T
l
(6) - L{F}=-lim
2nT
f(z)
00 f f
e-(S-z)ldtdy
-00
oo
o
= .2- lim
2n T 00 f- [ 00
f(z) - f(z) e-(S-Z)T
s - z s- z ]
dy.
The first term inside the brackets here is independent of T and, in

view of the extended Cauchy integral formula (Theorem 1, Seco 62), its
integral exists and has the value 2n.f(s):
~
(7)
2n f -
oo
00 s
f(rx + iy!
- rx- ,y
dy = f(s) (Res> rx).
The limit as T -. 00 of the

integral of the second term inside the brackets
in Eq. (6) is zero. For if we write s = a + ib, where a > <x,and note that
Is - zl ~ a - <Xand le-(s-z)TI = e-(a-a)T, it follows that
I
f-",s-z
'" f(z) e-(s-Z)T dy
l ~ e-(~a)T
a (Xf'"-'" If(z)1 dy.
The member on the right vanishes as T -. 00 becausefis absolutely integrable
along the line x = (x. Thus
(8) L{ F(t)} = L{L-1[f]} = f(s) (Re s > (X).
The proof of Theorem 5 is now complete.
68 CONDITIONS ON F{t)
The foregoing conditions under which the inversion integral formula is
valid are severe. They are not satisfied, for example, by the function l/s,
which is @(l/sk) where k = 1, nor by transforms of functions F that are
discontinuous or for which F(O):f: O. By using a Fourier integral theorem
and qualifying the function F instead of f, we can relax the conditions so
that the inversion integral formula can be seen to apply in nearly all cases
of interest to uso In fact we shall see that our formula is only a modified
form of the Fourier integral formula.
Let G be a function defined for all real t, sectionally continuous over
each bounded interval ofthe taxis, and absolutely integrable from - 00 to oo.
We agree to define G(t) at each point to where G is discontinuous as the mean
value of its limits from the right and left at to:
(1) G(to) = ![G(to + O)+ G(to - O)].
Also let the derivative G' be sectionally continuous over each bounded
interval; this condition actually implies that G itself is sectionally continuous
(Prob. 10, Seco70).
Then G(t) is represented in terms of trigonometric functions of t by
the F ourier integral formula 1
1 "' '"
(2) G(t) = -
1t o f f
G('r)cos y(t - .) d. dy
-00
(-oo<t<oo)
1 p oo. .
= - lim [G(.)e,y(t-f)+ G(.)e-,y(t-f)]d. dy.
f. f -'"
21tP-+00o
Since the integral of IG(.)I from - 00 to 00 exists, the inner integralof each
term within the brackets converges uniformly with respect to y to a con-
1 See, for instance, the author's "Fourier Series and Boundary Value Problems," 2d ed., pp. 83,
115 Ir., 1963.
tinuous function of y whose integral from y = O to y = p exists. Thus the

iterated integral in the second line of formula (2) can be written as the sum
p eiyt
f.o f co
-co
G(r)e-iy<d1:dy +
f o
-P
eiYI
f co
-co
G(1:)e-iYt d1: dy.
Hence the Fourier integral formula has the exponential form
(3) G(t) = ~ lim

21tp-co f f
p
-P
eiYI co G(1:)e-iyt d1:dy
-co
( - 00 < t < 00).
Note that the existence of the principal value of the improper integral with
respect to y from - 00 to 00 is ensured here, but not the convergence of the
improper integral itself.
Now let F denote a function defined when t ;;;O and (9(eat)as t -+ 00,
such that F', and therefore F itself, is sectionally continuous over each interval
O < t < T. If we write
(4) when t > O, G(t)= O when t < O,

where y > IX,then G satisfies the sufficient conditions stated above for the
representation (3), so
(5) G(t)= ~ lim

f p eiYI
f.ocoe-(Y+iy)tF(1:) d1:dy (Itl < 00).
21tP-co -P
The inner integral represents f(y + iy), where f(s) is the Laplace transform
of F(t). Ir z = y + iy, it follows that, for all real t,
I Y+iP
(6) eY1G(t) = : lim e'Zf(z) dz = Li -1{J(S)}.
2m P-co Jy-ip
When t < O, G(t) = O,and therefore Li -1{J(S)} = O.
When t > O,eY1G(t)= F(t), and Eq. (6) represents F(t) as the inversion
integral Li-1 {J(s)}. But sincethe function G representedby formula (5)has
the mean value (1) at eachpoint to of discontinuity, the inversion integral
converges to F(to) at a point to where F has a jump, provided we agree that
(7) F(to) = t[F(to + O)+ F(to - O)] (to > O).
When t = O,the value of the inversion integral is ![F(+ O)+ O].
The existence of the inversion integral (5)in the form of the principal
value of the improper integral along the line x = y was established here,
but the improper integral itself may not exist under the conditions stated.
The following theorem is now established.
Theorem 6 1f f(s) is the Laplace transform of a function F of exponential

order (9(eat)whose derivative F' is sectionally continuous over each
intervalO < t < T, then the inversion integral of f along any line
Re s = y, where y > oc,exists and represents F(t):
(8) (t > O).
At a point to(to > O) where F is discontinuous,the inversion integral
has the mean value (7) of F(t); when t = O, it has the value tF( + O),
and whent < O,it has the valuezero.
The conditions here on F can be relaxed because the conditions on G
for the validity of the Fourier integral formula (2) can be modified in various
ways.l For instance, on each bounded interval let the function G have at
most a finite number of infinite discontinuities but still be absolutely inte-
grable over the entire taxis, where each interval can be subdivided into a
finite number of open intervals such that G is monotonic (nondecreasing or
nonincreasing) over each one. Then the Fourier integral expression (2) has
the value t[G(t + O)+ G(t - O)] at each point t where G(t + O)and G(t - O)
exist.
As a consequence of those m9difiedconditions, for instance,
(y > O,t > O),
because the function G(t) = e-ytct when t> O, G(t)= O when t < O, is
represented by its Fourier integral formula when t > O and y > O, and
because L{t-i} = (n/s)i when Re s> O.
Even though its conditions are on F, the function usually sought,
rather thanf, Theorem 6 serves as a useful supplement to Theorem 5. The
following example and corollary illustrate this.
Example Let,j'; denote the branch J;eiOI2, r > O, 101< n/2, where
s = reiO.Show that the inversionintegral formula applies,when y > O,
to the function
f(s) =
sys+l
f .
On a half plane Re s ~ oc> O, the branch ,j'; is analytic and
s,j'; + 1 never vanishes sin<::eRe (s + l/,j';) > O. Thus f is analytic
there but only (9(1/s), which fails to satisfy the order condition in
Theorem 5. By carrying out one step of a division, however, we can
write f as a sum of two components,
1 1
(9) f(s) = - - -
S S2,j'; +s
1 See, for instance, H. S. Carslaw, "Fourier's Series and Integrals," chap. 10, 1930, and E. C.
Titchmarsh, "Theory of Fourier Integrals," p. 13, 1937.
"
such that Theorem 5 applies to the second, which is (9(S-5/2)and analytic

over the half plane, and real when s is real, while Theorem 6 applies
to the first to give Li-l{l/s} = 1 when t> O. Thus Li-l{f} = F(t)
where
~
F(t) = Li-l
{s }
+ Li-l
;/
{ S2 s + s}
= 1 + Fl(t) (t > O),
and F1 is continuous (t ~ O); also F1(0) = O and F1 is (9(e<Xt)

if IX> O.
ConsequentIy F is continuous when t > Oand (9(e<Xt), and F( + O)= 1.
At the end of this chapter we present methods for simplifying such
representations as Li-l{f} of our function F.
The example suggests a corollary to Theorems 5 and 6.
Corollary lf f(s) can be written in the form

(10) (Res ~ IX)
wherefl satisjies the conditions in Theorem 5 andf2 is the transform of a

function F2 satisfying the conditions in Theorem 6, then Li -1{f} is the
inverse transform F(t) of f(s). Moreover F is (9(e<xt), F is sectionally
continuous on each bounded interval, and F( + O)= F2(+ O).
69 UNIQUENESS OF INVERSE TRANSFORMS

In order to make a simple analysis of uniqueness of the inverse Laplace
transformation, we consider the class S of real-valued functions on the half
Une t > O, defined in Seco6. The class consists of all functions F that are
sectionally continuous over each bounded interval, where F(t) is dejined as the
mean value of its Umitsfrom right and left at each point of discontinuity, and
such that each function F is (9(et)for some IX.Then F has a transform f(s)
when Re s > IX.Our theorem on uniqueness of inverse transforms can be
stated as follows.
Theorem 7 Amongallfunctions of class S no two distinctfunctions canhave

the same Laplace transform over a right half plane Re s > ex,or for all
real s on a half Une s > IX.
Since transforms of functions of class S are analytic functions of s
over some right half plane, any two transforms are the same over a half
plane if they are identical for all real s on a half line s > IX(Sec. 64).
In particular, the theorem shows that no two functions of exponential
order that are continuous when t ~ Ocan have identical transforms for all
real s on a half line s > a.
To prove the theorem, first let S' denote the subclass of all functions F
in class S for which F' is also sectionally continuous over each bounded
interval on the half line t > O. According to Theorem 6 if F belongs to e',

it is represented by the inversion integral of its transform f That is, when
F(t) is (9(e"'),then F(t) = Li -1 {J(s)} when t > Oregardless of the value ofy
in the inversion integral as long as y > IX.In case a function F in e' has the
vanishing transform f(s) = O for all s in some right half plane, it follows
that F(t) = Ois the only function of class e' whose transform vanishes over a
right half planeo
Next let G and H be two functions of c1ass $, so that no conditions are
imposed on their derivatives. Let both be (9(e"/)and suppose that they have
identical transforms g(s):
(1) L{ G(t)} = L{ H(t)} = g(s) (Re s > IX).
Now the function
(t ~ O)
(2) F(t) = f~[G(r) - H(!)] d!
is continuous and (9(eP/)whenever {3> IX(Sec. 17). Moreover, on each

interval where G and H are continuous,
(3) F'(t) = G(t) - H(t);
thus F' is sectionally continuous over each bounded interval, and F is
therefore in c1ass e'.
We can choose {3as a positive number. Then from condition (1) and
the representation (2) of F as a special convolution, it follows that
1
(4) L{F(t)} = -[g(s)
s - g(s)] = O (Re s > {3),
and since F is in e', we conc1ude that F(t) = O when t> O. Therefore

F'(t) = O(t> O);that is, wherever G and H are continuous,
(5) G(t) - H(t) = O.
At each point where G or H has a jump, it followsthat the one-sidedlimits
of G(t) - H(t) vanish; thus
!(G(t + O)+ G(t - O)] = t[H(t + O)+ H(t - O)] (t > O).
Consequently G(t) = H(t) whenever t > O,and Theorem 7 is proved.1
70 DERIVATIVES OF THE INVERSION INTEGRAL

When the solution of a boundary value problem is found in the form of an
inversion integral Li - 1{J}, it is often possible to verify the solution by noting
1 For theorems on uniqueness that permit L -1{f} to have certain int'inite discontinuities, see
G. Doetsch, "Handuch der Laplace-Transformation," vol. 1, pp. 72 ff., 1950.
properties of the function f The two theorems in this section are useful for
that purpose. Their proofs foIlow from Theorem 5 and properties of uni-
formly convergent integrals (Sec.15).
When the inversion integral in Theorem 5 is differentiated with respect
to t under the integral sign, we obtain L -1 {sf(s)} :
_1 a I oo
f f
oo
(1) -;-[etzf(z)]dy = 2
_ etzzf(z) dy,
21/: - 00ut 1/: - 00
where z = y + iy. If sf(s) asweIl asf(s) satisfiesthe conditions imposed on

fin that theorem, the integral (1) converges uniformly with respect to t and
represents F'(t), the derivative with respect to t of L -1 {f(s)}. Then F'
satisfies the conditions stated for the function F in the theorem. The addi-
tional condition needed here is that sf(s) is @(S-k)where k > 1; that is, that
f(s) is @(S-k-l).
If we replace f(s) above by sf(s), we see that F"(t) is represented by
L -1{S2f(s)}, and so on, giving the foIlowing theorem.
Theorem 8 Let f be a function of s that is analytie and @(s-k-m) over a half

plane x 6 ex,where s = x + iy, k > 1, and m is a positive integer. Also
let f(x) be real (x 6 ex). Then the inversion integral off along any Une
x = y, where y 6 ex,converges to the inverse transform off, a real-valued
function F,
(2) L-1{f(S)} = F(t) (t 6 O),
and the derivatives of F are given by the formula
(3) (n = 1,2,... , m);
furthermore, F and each of its derivatives (3) are continuous when t 6 O
and @(e"'t),
and their initial values are zero:
(4) F(O) = F'(O)= .. . = F(m)(o) = O.
Consider again, as an example, the inverse transform F1 of the second

function on the right in Eq. (9), Seco68;
F1(t) = L-I
{ ;/+ }.
S2 s s
Its transform is @(s-5/2),so m = 1, and it foIlows from Theorem 8 that both

F1 and F~ are continuous when t 6 Oand both are @(e"'t)when ex> O,and
that
F1(O) = F~(O)= O. .
It wilI be recaIled that formula (3) cannot be valid unless either F or f
is subjected to rather severe restrictions. For according to our basic property
of transfonning derivatives,
p<n)(t)= L-1{s"f(s) - s"-1F(O) - ... - F(n-1)(0)}.
If F(n)(t)is to be the inverse transform of s"f(s), then it is necessary that
F(O),. .. , F(n-1)(0) all vanish.
Of course the function f in the above theorem may involve variables
independent of s provided the statements in the theorem are understood to
apply for fixed values of those parameters. Conceming differentiation and
continuity with respect to such a parameter p, the following theorem can be
stated from properties of unifonnly convergent integrals.
Theorem 9 Letfbe a continuousfunction oftwo variables p and s (s = x + iy)

whenx ~ ex,analytie with respect to s over that halfplanefor eaehfixed p,
and such that If(p,s)1 < Mlsl-k there where k > 1 and the constants
M, k, and IXare independent of p. Also let f(p,x) be real when x ~ IX.
Even if p has an unbounded range of values, let f(p, IX+ iy) be bounded
for all p and y. Then the inverse transform off with respect to s, which
can be written
(y ~ IX, t ~ O),
is a continuous real-valued function of its two variables p and t(t ~ O)
for whicha constantN independentof p exists suehthat IF(p,t)1< NzJ;
in particular, F is a bounded funetion of p for eaehfixed t. 1f the partial
derivative J;,(p,t) also satisfies the conditions imposed here on f, then -
(5) :p F(p,t) = Li - 1{:pf(P,S)},

and IFp(p,t)1< N(/'t.
The theorem can be applied to J;,(p,s)to obtain corresponding results
for Fpp(P,t), and so on for derivatives of higher order. Applications of
Theorems 8 and 9 are made in the folIowing chapters.
PROBLEMS
1. Apply Theorem5 to the functions(a)f(s) = (S2 - 1)-1 and (b)g(s)= s-2e-s to
show that their inverse transforms F(t) and G(t) are represented by inversion integrals,
also to find order and continuity properties of F and G and to show that F(O)= G(O)= O.
Verify those properties by writing F and G as known inverse transforms.
2. ApplyTheorem8 to the functionf(s) = S-4 to deduce that the inverse transform
F(t) and its derivatives offirst and second order at least are continuous when t ;; O,and
19(e"")if IX> O, and that F(O) = F'(O) = F"(O) = O. Verify those properties by recaIling
that F(t) = t3/6.
3. Apply Theorem 6 to the transform of the function e-' and simplify the inversion
integral, when }' = O,to establish the integration formula
lo
eo cos ty + y sin ty d
2
Y + 1
y=ne
-, (t > O).
4. When f(s) in Theorem 5 is the transform (s + 1)-2 of te-', simplify the inversion
integral, when )' = O,to pro ve the integration formula
lo
eo O - y2)cos ty + 2y sin ty d = nte-t
(y2 + 1)2 Y
(t ;;;;O).
5. Let ~ denote the branch j;eiBI2, -n < IJ< n, r > O,where s = reiO,an analytic
function over each half plane Re s ;;;;IX> O. If f(s) = exp (-~), prove that If(s)1 <
exp(-.;i) overthe halfplane and thatf(s) is l!I(s-m)there for every positive integer m.
Show that Theorem 8 applies to establish f(s) as the transform of a function F(t) with
continuous derivatives of all order (t ;;;;O)such that
F(O) = O and F(n)(o)= O (n = 1,2,.. .).
[Compare the transformation (6), Seco27.]
6. Use power series to prove that the function g, where
g(z)= 1 - ze-S if z i' O, g(O) = 1,
is entire and hence bounded over a neighborhood of the origin in the z plane. Then
apply our corollary, Seco68, to the function
f(s) = !e-lls
s = !s - sO - S2e-l/')
to prove that L -1{J} is a continuous function F(t) when t > O,l!I(e.')if IX> O,and that
F(+O) = 1.
7. State why the inversion integral formula applies when}' = 1to the known transform
f(s) = s-le-s for all t; but show that when t = 1 the Cauchy principal value Li-l{J}
must be used because the improper integral itself fails to exist for that value of t. AIso
verify that, when t = 1, Li-l{J(S)} = t.
8. The entire functiong(s) = exp (S2)is not l!I(S-k)over a right half plane for any
positive k. But prove that Li-l{g}, integrated along the imaginary axis, is the function
eo
l
F(t) = -n f
o
exp (- y2) cos ty dy.
A known integration formula [Eq. (7), Seco74] shows that F(t) = tn-t exp (- t2/4), and
thus (transformation 109,Appendix A, Table A.2)
L{F}= L{2fiexp (-~)}= texp(s2)erfcs i' g(s). .

Thus the inversion integral may exist without representing the inverse transformo
9. Show why the function G, where

when t > O, G(O)= t, G(t)= O when t < O,
satisfies all conditions for its representation by the exponential form of the Fourier
integral formula. When t = O,however, show that the Cauchy principal value (3),Seco68,
must be used in that representation, for the improper integral itself,with respect to y, does
not existo
10. Prove that F(t) itself is sectionally continuous over an interval if its derivative
F'(t) is sectionally continuous there. Note that F' is continuous over each of a finite
number of subintervals, such as a ~ t ~ b, when properly defined at the end points, so
F is continuous when a < t < b and, when a < to < b, then
r' F'('r:) dt = F(t) - F(to) (a < t < b).

J,o
Deduce that F(a + O)and F(b - O)exist.
71 REPRESENTATIONBY SERIES OF RESIDUES
Throughout this and the following two sections, f(s) denotes a function
that is analytic for all finite values of the complex variable s except for a
set of isolated singular points
confined to some left half plane Re s < y. We assume also thatf(s) satisfies
conditions under which its inversion integral along the line x = y converges
to the inverse transform F(t), say the conditions in either Theorem 5, Seco67,
or Theorem 6, Seco68. Then F(t) can be represented formally by a series,
finite or infinite, depending on the number of singular points, and we shall
establish practical conditions under which the representation is valido
Since etz is an entire function of z, the singular points z = Snoff(z) are
the singular points of the integrand etzf(z) of the inversion integral. Let
Pn(t) denote the residue of the integrand, for any fixed t, at the isolated
singular point Sn:
(1) Pn(t) = residue of etzf(z) at z = Sn'
According to the residue theorem, the integral of etzf(z) around a path

encIosing the points SI' S2, . . . , SN has the value
21ti[Pl(t) + P2(t) + . . . + PN(t)].
Let the path be made up of the line segment joining the points y - iPN,
Y + iPN,and some contour CNbeginning at the second and ending at the
52
O
Fig.71
first of these two points and Iying in the half plane x ~ y (Fig. 71). Then
N
l
(2) _2l .f
nI
Y+PN
y-PN
etzf(z)dz + _
2nI '
fCN
etzf(z) dz = L Pn(t).
n=1
As PN- 00, the value of the first integral here tends to L-1{J(S)},
sincethe inversionintegralis the limitofthe correspondingintegralinvolving
--
P,as P 00 in any manner. Let the numbers PN(N = 1,2,...) be selected
so that PN 00, and let the curves CNtogether with the line x = y endose
the points S1, S2, ... , SN, if the number of singular points is infinite. If the
number is finite, let all of them be endosed when N is greater than some
fixed number. Then if j(z) and CNsatisfy additional conditions under which
_1 . lim
(3)
f
2nI N-oo CNetzf(z) dz = O,
it follows, by letting N - 00 in Eq. (2), that 00
L-1{J(S)} = L Pn(t),
n= 1
or by a finite series in case the number of singular points is finite. The series
on the right is necessarilyconvergent because the limit, as N 00, of the
left-hand member of Eq. (2) exists.
-
Since the inversion integral represents F(t) by hypothesis, the inverse
transform off (s)is represented as the series of residues of etzf(z):
00
(4) F(t) L1 Pn(t).
= n=
It is not essential that the line x = y and the curve CNendose exactIy
N of the poles, of course. F or example, if two poles are induded in the ring
between CN and CN+1, the residues at these two poles are simply grouped as
a single term in the series.
It is sometimes convenient to use the series representation (4) directly
and formally, without regard to conditions under which the inversion
integral represents F(t) and the integral over CN tends to zero. The function
F so obtained may be such that its transform can be shown to be f(s), or
such that it satisfies all conditions in a problem whose solution was sought
by the transformation method. As we shall see in the problems at the end
of this chapter, however, there are simple cases in which such a formal
procedure leads to incorrect results.
72 RESIDUES AT POLES
At singular points Sn' which are poles of the functionf, useful formulas for
the residues Pn(t) of e'Zf(z) can be found.
Let Sn be a pole of f of order m. Over a neighborhood of that pole
with the point Snitself excIuded (O< Iz - snl < cJ, f is represented by a
Laurent series
A_1 A_2 A-m ~ {

.
(1) f( ) ,;y
Z = -Z - Sn + (Z - Sn)2 + ... + (Z - Snr + j=f.J O AJ,z - S ,
where A-m =FO and the A's depend on Sn' Thus (z - Sn)'''f(z)is represented
there by the power series
IX)
(2) A_1(z - snr-1 + A_2(z - snr-2 + ... + A-m +

j=O
AJ{z - sny+m.
Taylor's series expansion ofthe entire function e'z about the point Snis
~-1
(3) [
e'z. = exp(snt) 1 + t(z - Sn) + ... + (m - 1)!(Z - Snr-1 + ... . ]
The coefficientof (z - sn)-l in the Laurent expansion of e'zf(z) over the
domain O< Iz - snl < cn, the residue Pn(t) of that function at Sn' is the
coefficientof (z - Snr- 1 in the product of the power series(2)and (3). Thus
we find that
~-1
~
Pn(t) [
= exp(snt) A-1 + tA_2 +... + (m - 1)!A-m . ]
Formula (4) has a useful feature. It presents the residue of e''i(z) at a
pole Snof order m as the inverse Laplace transform, with respect to z, of the
principal part of the Laurent expansion (1) of the function f
When sn is a simple pole (m = 1) of f
(5) Pn(t) = exp (snt)A":: 1 = exp (snt) lim [(z - sn)f(z)].
z s,.
Note that A_1 is the residue offitself at sn. In particular, if
(6) fez) = p(z)

q(z)
where p and q are analytic at Snand q(sn) = Owhile q'(sn) :i=O and p(sn) :i=O,
then
P(Sn)
(7) Pn(t) = ~ exp (snt).
q (Sn)
Ir all singular points off are simple poles andf has the fractional form (6),
then formula (4),Seco71,can be written
(8) F(t) = L - 1
P S)
{
qS )} = f P,
n= 1q Sn)
Sn)
exp (snt).
In caseP and q are polynomials, the number of poles offis finite, and formula
(8) becomes Heaviside's expansion (4), Seco24.
Assume now that f(z) = f(z) except at the singular points off, all of
which are isolated, a condition that is satisfied if f(z) is real over the real
axis except at singular points (Sec. 61). Let SI = a + ib, where b :i=O, be a
pole off of order m. Then over a domain O < Iz - sIl < c1,
(9) f (z) = - A_l + ... + (Z A-m m + ~

~ AJ,z
{
- \i
Sl1 (A-m :i= O).
Z - SI - SI) j=O
The operation of taking complex conjugates is distributive to the terms

of a convergent series; that is, if an infinite series ~zn converges to a sum S,
then the series ~zn converges to S, as we can see by considering remainders.
It follows from representation (9) and the condition f(z) = f(z) that
(10)
when 0< Iz - sIl < Cl, that is, when 0< Iz - sIl < Cl. By writing S for
z here, we see that f(s) has a Laurent series expansion in the domain
O < Is - sIl < Cl with the A's as coefficients, and, since A-m :i=O, SI is also
a pole off of order m. According to formula (4) then, the sum of the residues
of etzf(z) at the poles S1 = a + ib and S1is P1(t) + P1(t), where
m-l
...
-
Pl(t) - e
(a+ib)t
[A_l + tA_2 + + (m -
t
I)!A-m ] .
Therefore the component of F(t) corresponding to the pair of poles a :t ib
(b :i=O) of f(s) is the real-valued function
(11) P1(t)+ P1(t)= 2~tRe {eibt[ A_l + tA_2 +... + (~:-11)!A_mJ}.

Let us write the residue A - 1 off at a pole SI =a+ ib in its polar form as
(12) A_1 = r1 exp (iOl).
Ir s1is a simple pole, formula (11) then gives the component of F corresponding
to the poles a :t ib as
(13) Pl(t) + Pl(t) = 2e"trl cos (bt + (1) (b :F O,r > O),
a result that includes the component G(t) obtained in Seco26, wherefwas a
quotient of polynomials.
In case the singular points off consist only of the simple poles O, :t inro
(n = 1,2,.. .), wherefhas residue ro at z = Oand residues r" exp (i0,,)at inro,
formula (4), Seco71, becomes
00
(14) F(t) L r"cos(nrot

= ro + 2 ,,=1 + O,,);
then F is a periodic function of t with period 21C/ro.

Ir SI = a + ib is a second-orderpole off, we write
(15) (Pl > O).
Then if b :F O,according to formulas (11) and (12),

(16) Pl(t) + Pl(t) = 2e"t[rlcos(bt + (1) + p1tcos(bt + "'1)]'
Whena = O,the term 2p 1t cos (bt + '" ) is one of resonance type representing
an unstablecomponent of F(t), since Pl > O. .
Similarly if ib is a pole of order m off(z) then F(t) contains an unstable
component of the type Ctm- 1 cos (bt + O)where C :FO. In fact we can see
from the above formulas that F(t) contains an unstable component iff(z) has a
real or imaginary pole a + ib of any order such that a > O, or a pole z = O
or z = ib of order m = 2, 3, . . .. This is a generalization of results found in
Seco26 where partial fractions could be used.
73 VALlDITYOF THE REPRESENTATIONBY SERIES
We now establish sufficient conditions on f under which the integral s of

etzf(z) over certain useful contours CN tend to zero as N -+ 00 as assumed in
Sec.71. Since the integral s must exist, the contours should not pass through
singular points off, nor should they become arbitrarily close to such points
as N increases if the limit of the integrals is to exist. The ease of finding an
order of magnitude of If(z)1 when z is on CN depends on the particular
function f and the type of paths chosen for CN'
We treat three types of paths CN that are especialIy useful. In each
case we assume thatf(z) is (!)(Z-k)on CN where k > O; that is, constants M
and k, independent of N, exist such that

M
(1) for points z on CN (k > O),
If(z)1 < Izlk
and we shall prove that, since PN-+ 00 as N -+ 00,
(2) lim
f et%f(z)dz =O (t > O).
N-<e CN
If k> 1, statement (2) holds also when t = O.

RECTANGULAR PATHS
Order properties of many functions on horizontal and vertical lines are

easily determined. Let CN be an open rectangie with sides along lines
y = :tPN and x = - PN as indicated in Fig. 72, where PN -+ 00 as N -+ oo.
Then iffsatisfies the order condition (1) on CN, the integrand ofthe inversion
integral satisfies the condition
M
let%f(z)1< -x when z is on CN.
Izlk
The integrals over the sides y = :t PN are then such that

M M - e-t/lN
I
Y y ety
(3) et%f(z)dx < -
II-/IN' 1 pNk -/IN Xdx = -t PNk
when t > O. Thus they tend to zero as N -+ oo.
The length ofthe third side of CN is 2PN,and the absolute value ofthe
-
integrand et%f(z)is less than M exp ( PNt)PN-k there, so when t > O, the
integrals over those sides vanish as N -+ oo.
Thus condition (2), where t > O, is established. Moreover, if k > 1,
then ScNf(z) dz -+ O as N -+ 00, so that condition (2) holds when t = O.
y
'PN+iPN "(+i(3N
CN
-(3N "(
o %
(3N-i(3N "(-i(3N
Fig.72
CIRCULAR ARCS
Let CN be ares Izl = RN, x ~ y. Then RN2= {3N2+ y2 and RN-+ 00 as

N -+ oo. Ir y > O, the length of eaeh of the two ares of CN in the strip
O ~ x ~ y tends to y as N -+ oo. Sineefsatisfies eondition (1),then le'Zf(z)1<
Me'YRN-k on those ares, and the integrals over them tend to zero as N -+ oo.
Regardless of the sigo of y, the integral over the are CN in the region
x ~ O satisfies the eondition
M 3"/2
(4)
IfCN e'Zf(z)dz < ---
1 RN "/2 f exp (tRNeos O)RNdO
2M ,,/2
= --- exp (- tRN sin </J)RN
d</J.
RN 1o
The graph of sin </Jshows that sin </J> 2</J/nwhen O < </J< n/2. Therefore
the final member of eondition (4) is less than
2M "/2 A,. nM 1 - exp (-tRN)
f
</J
--- exp - 2tR N- Nd'1' = - k

RN o ( n)R t RN
when t > O,and this quantity vanishes as N -+ oo.
Therefore eondition (2) is satisfied when CN are those circular ares.
Moreover, if k> 1, eondition (2) follows when t = Obeeause
(5) [ f(z) dz < Mk2nRN -+ O as N -+ oo.

I JCN I RN
PARABOllC ARCS
Transforms involving fi arise in solutions of heat equations. It is often

desirable to seleetexpanding paths CNon whieh is bounded away from ~
zero. But Re ~ .fi
= eos 0/2 which vanishes on the negative real axis
(O = n). We may ehoose paths on whieh 1m ~
never vanishes by taking CN
as ares of eurves .fi
sin 0/2 = ~, where aN -+ 00 as N -+ oo.
Then CNis a parabolie are
2aN
(6) (x ~ y)
r = 1- eos O
with vertex at x = -aN,Y = Oand foeus at the origin, resemblingthe are
shownin Fig.71. Its equation can be written in the form
(7) y2 = 4aN(x+ aN) [x ~ y, {3N2= 4aN(y+ aN)]
from whieh we find that
dy 2 = x + 2aN.
1+
( dx) x + aN
Since Izl ~ aN and f satisfy condition (1) on CN,
(8)
I JCN
r eIZf(z)dz I
< 2~
aN f -aN e'xJX X ++ 2aN
y aN dx.
Let N be large enough that aN > 21yl. Then when t > O,
-taN + 2aN ta N -taN dx

(9)
f -aN
Ix
e
jWf;
+ X aN
dx < exp --
( 2 )~f 2 aN
-aN
-
Jx + aN
and when we evaluate the last integral, we find that the right-hand member
of condition (9) vanishes as N -+ oo. AIso,
1 y + 2aN y JIx
aN f-taN f
(10) --- e d d
y IX! + 2aN x< aNkJaN/2 -taN e x
xx+aN
which tends to zero as N -+ 00 if t > O, so condition (2) is satisfied on the
parabolic ares. Again we find that the right-hand member of condition (8)
vanishes as N -+ 00 when t = Oin case k > l. Our results can be summarized
as follows.
Theorem 10 Let f be analytic everywhere except for isolated singular points

sn(n = 1,2,...) in a half planex < y and such that its inversionintegral
along the line x = y represents the inverse Laplace transform F(t) of
f(s). On contours CN(N = 1,2,...) in the left hand plane x ;;;y, let f
satisfy the order condition If(z)1 < Mlzl-k where M and k are positive
constants independent of N, and CN is either (a) the rectangular path
shown in Fig. 72, (b) circular arcs Izl = (PN2+ y2)t, x ;;;y, or (c) the
parabolic arcs (7), where PN -+ 00 as N -+ oo. Then whenever t > O,the
series of residues Pn(t)of eIZf(z) at sn converges to F(t),
00
(11) F(t) = L Pn(t)
n= 1
(t > O),
if its terms corresponding to points sn within the ring between successive

paths CNand CN+1are grouped as a single terms of the series.
Also, if k > 1 and Li - 1{J} represents iF( + O)when t = O,then
00
(12) iF( + O)= L Pn(O).
n= 1
74 ALTERATIONSOF THE INVERSION INTEGRAL

When the singular points of f(s) are not all isolated, it is often possible to
reduce the inversion integral to a desirable form by introducing a new path of
integration. Weillustrate the procedure by findingan inversetransformthat

was obtained by another method in Seco27.
Let us find F(t) when
1
(1) f(s) = - exp (-st),
s
where s = reioand st is the single-valued function
(2) (r > 0, -n < (}< n).

st = ";efi = ~(cos~ + iSin~)
Since the function (2) is analytic everywhere except on the branch cut
(} = n, r ~ 0, f is analytic in the same region. In the half plane Re s ~ y,
wherey is a positive constant, -!n < (}< !n and cos!lJ > l/./i; therefore
l,skf(s)1= ,-1<-1exp(-";cos~) < ,-I<-le-.;]2.

This last function of r is continuous when r ~ y and vanishes as r -+ 00;
hence it is bounded, and therefore f is 19(s-~ in the half plane for every
value of the constant k, and in particular if k > 1. According to Theorem 5,
the inversion integral along the line x = y therefore represents the inverse
transform F(t),
1 Y+i{l dz
(3) F(t) = --: lim
f
2m {I-+oo y-I{I
(P exp ( - zt)-
Z
(t ~ O).
The sum of the integral in formula (3) and the integral along the path
ACDD'C' A', consisting of the circular ares and line segments shown in
Fig. 73, is zero, since the integrand isanalytic except on the nonpositive real
axis. Thus if IAc denotes the integral of e'=j(z) over the arc AC, etc., we can
write
(4)
Let R and ro denote the radii of the large and small cicular arcs; thus
R2 = y2 +p2 so that P -+ 00 when R -+ oo. Along the arc AC, z = Rel,
dz = iReiOdO, and zt = JRei0/2. Hence the integrand of the integral is a
continuous function of (}whenever f. ~ 0, wheref. is the anglebetweenDC
or D'C' and the negative real axis. For any fixed R, the limit ofthe integrals
IAC and IA'c', as f.-+ 0, therefore exists. Likewise for any fixed positive ro
the limits of the integrals over the other parts of the path exist. Since formula
(4)is true for everypositive f. and the integral on the left is independent of f.,
it follows that we can let each of the integrals on the right have their limiting
valuesas f. -+ 0, and consider hereafterthe path in Fig. 74.
y
'Y + ifJ
c
x
c'
'Y-ifJ
Fig.73 Fig.74
Write J AC = lim o JAC, and so on, for the integrals over the other
arcs and lines. We shall now let ro approach zero and R tend to infinity, so
that the left-hand member ofEq. (4),which is incidentally independent of ro,
becomes the inversion integral or F(t), except for a factor - 2ni.
When z is on the circle r = ro, Z = roei9,and the integral over that
circle can be written .
JDD, = i L-" exp(troei9 - efi9) dO,

and the integrand is a continuous function of Oand ro when r o ;;O. Therefore
-"
lim JDD.= i
ro-+O fn dO = -2ni.
On the line CD, z = rei("-f) and zt = ';ei("-fl/2; thus as ~ O,

Z ~ -r and zt ~ ifi On the limiting position of D'C', however,z = -r
and zt = - i';. Therefore
. dr
JCD + J D'C' =
f R
ro
e-tre-''; - +
r Iro
R . dr
e-tre,f;-
r
' R -tr sin ,;d r,
= 2
fro e -
r
. .
and hm (JCD + JD,c') = 21 f. R e _trsin';
-dr
ro O o r
./R sin JL
= 4i f. exp ( - tJL2)- dJL.
o JL
Consequently we can write, in view offormula (4),

1 . 2 oo 2 sin JL
(5) F(t) = : hm (JAC + JC'A') + 1 - - f. exp(-tJL )-dJL.
2m R-+oo n o JL
216 SECo 74] OPERATlONAl MATHEMATICS
Now on ares AB and A' B', Re (tz - zt) ~ ty; thus the integrand of the
integrals J AB and J B'A' satisfies the condition
Iz-l exp (tz - zt)1~ R-le'Y.

Since the length of each are tends to y as R -+ 00, those integrals tend to zero
as R -+ oo.
-
Finally, on ares BC and B'C', Re (tz zt) ~ tR cos so that e
" "/2
IJBe!~
I ,,/2
exp (tR cos e) de =
f.o exp (- tR sin cp)dcp,
and JC'B' satisfies the same condition. The value of the last integrand does
not exceed exp ( - 2tRCP/n),as pointed out under Circular Ares, Seco73, and
an elementary integration then shows that J BCand JC'B'tend to zero as
R -+ 00, if t > O.
The limit in formula (5) therefore vanishes, and
2 '" sinp.
(6) F(t) = l - - exp (-tp.2)-dp. (t > O).
f.
n o P.
When we integrate both members of the known 1 formula
'" l O(2
(7) exp (- tp.2)COSocp.dp. = - A- exp -- (t > O),
f.o 2 t ( 4t )
with respect to oc,from zero to 1, we find that

l
2 '" sin p. 1 OC2
- f. exp ( - tp.2)- dp.= e f. exp ( -4 ) doc
n o P. V nt o t
2 l/(2 jij
= fi f.o exp (_).2) d)..
Thus we can write our result in the form
l
(8) 1 - erfc - .
F(t) = 1 - erf ( 2.fi } - ( 2-fi }
I See, for instance, R. V. Churchill, "Complex Variables and Applications," 2d ed., p. 171,
exercise 7, 1960. Formula (7) also follows from the result in Probo 7, Seco35.
PROBLEMS
1 . Instead of simply using partial fractions, show that Theorems 5 and 10 apply to
1(s)when
2s + 1 1
(a) f(s) = S(S2 + 1)' (b) f(s) = S2(S+ 1)'
and find F(t) as a sum of residues. Ans. (a) F(t) = 1 - cos t + 2 sin t.
2. Use identities (6),Seco55, to prove that the function tanh z is bounded over each of
the half planes x ~ y and x ~ -y when y> O, and on the lines y = INn(N = 1,
2, . . .) uniformly with respect to N. Thus show that the function
f(s) = tanh
S2
s
satisfies all conditions in Theorems 5 and 10 such that F(t) is represented by a series of
residues as follows:
F(t) = 1 - .! f
n2 n =1 (2n
1
.. ,
co (2n
s
- I)nt (t ~ O).
Note that, according to Seco23, F is the triangular-wave function H(2,t) shown graphi-
cally in Fig. lO.
3. In Probo 4, Seco24, we found that the function
1 1
f(s) = S2 - s sinh s
is the transform of the periodic function F for which
F(t) =t when -1 < t < 1,F(t + 2) = F(t).
Show that s = Ois a removablesingular point of f, that f satisfiesall conditionsin
Theorems6 and 10,and thus that F has the sine seriesrepresentation
2 00 (-1)"+1
F(t) = 1t
- n= 1 n sin nnt.
4. Use' Maclaurin's series for coshz to show that the function g(s)= cosh ;,
s #- O,g(O)= 1, is an entire function of s for any branch of ;. Then show that the
function
1
f(s) =-
s cosh .jS
is analytic except for simple poles s = O,s = -(n - t)2n2and, with the aid of one of
identities (6), Seco5}. that f satisfies all conditions in Theorems 5 and 10 when CN are
ares of parabolas v
r sin 8/2 = Nn. Thus find L -1{J} in the form
= 1 + -4 '--
~ (-I)n (2n - 1)2n2
F(t) -
nn=12n-1
exp
[ 4 ]
t (t > O).
5. Write F(t) fonnally as a series of residues when
f(s) = 4s2 + n2s .

2 tanh
2 . nt nt '" 1 (2n - l)nt

Ans. n F(t) = ntstn2' + oos2' - n~2n(n- l)ooS 2 .
6. Whenf(s) = s-2e-s, show that F(t) is representedby the inversionintegral of f,
where y > O,but that the residueof e/%
f(z) at the lone singularpoint off represents
F(t)only whent ~ 1. This exampleand those in Probs. 7 and 8 belowiUustratehow
- the fonnal procedureof writing the sum of residuesPn(t)may fail to givethe inverse
transfonn. Note that in this case lel%f(z)1-+ OCJas x -+ - OCJif t < 1.
7. StatewhyL,-l{s-le-'} = 80(t- 1)ifY> O,but showthat the sumof residues
Pn(t),oonsisting ofjust one tenn, fails to represent 80(t - 1)when t < I.
8. Apply the real fonn of the inversion integral, fonnula (5), Seco66, to the entire
function
f(s) = l - s e-S [s "p 0,f(0) = 1],
to find F(t) directly. Here we can write y = Oand use the integration fonnula (8), Seco34.
Note that the residue procedure can represent F(t) only where F(t) = Obecause e'"f(z)
is entire.
9. From properties ofthe transfonn ofthe function erfc (tt-i) noted in Seco74, prove
that the function and each ofits derivatives tends to zero as t -+ +0.
10. When,JS = ~i912( -n < (J< n, r > O),use the procedure in Seco74 to prove
that
L;l -
1
{ ,JS }
=-
2
n0 i '"
o
exp(-12) d1 =-
1
jitt
(t > O)
11 . Generalize Theorem 10 for rectangular paths by letting the sides of CNlie on lines
x = -CPNand y = IPN and usingonlya oonditionof boundednessIf(z)1 O,where e and B are positive oonstants independent of N.
12. Prove Theorem 10 when CN are semicircles Iz - yl = PN'x ~ y, and If(z)1<
Mlzl-k on CN. -
7
Problems in Heat Conduction
We shall now illustrate the use of the theory just developed in solving further
boundary value problems in the conduction of heat in solids. We present
examples of problems that cannot be fully treated with the more elementary
theory used in Chap. 4.1
The formal solution of the problem in the next section is followed by
a complete mathematical treatment of that problem. The purpose is to
illustrate a means of rigorously establishing the solutions of such problems.
Since the procedure is lengthy, the reader is advised to use it sparingly in
his workon the sets of problems that follow. A clear understanding of the
formal method of solution is of primary importance.
1AItemative methods for solving a few of the problems in this and the following chapter are
presented in chap. 7 of the author's "Fourier Series and Boundary Value Problems," 2d ed.,
1963.
219
75 TEMPERATURES IN A BAR WITH ENDS AT

FIXED TEMPERATURES
Let U(x,t) denote the temperature at any point in a bar (Fig. 75) with insulated
lateral surface and with its ends x = O and x = 1 kept at temperatures zero
and Fa, respectively, when the initial temperature is zero throughout.
In Seco49 we obtained a formula for U(x,t) in the form of a series of
error functions, a series that converges rapidly when t is small. We shall
now obtain another series representation of this temperature function.
This series will converge rapidly for large t. Let us proceed formally to the
solution here, leaving the fuIl justification of our result to the foIlowing
sections.
We have taken the unit of length as the length of the bar, and we
observed earlier that, by a proper choice of the unit of time, we can write
k = 1 in the heat equation, where k is the diffusivity. The boundary value
problem in U(x,t) is then
Ur(x,t) = Uxx(x,t) (O< x < 1, t > O),

U(x,+O) = O (O < x < 1),
U(+O,t)= O, U(l - O,t) = Fa (t > O),
where Fa is a constant.
The problem in the transform of U(x,t) is
su(x,s) = uxx(x,s) (O< x < 1),
u( +O,s) = O, u(1 - O,s) = Fa.

s
Since this problem in ordinary differential equations has a solution that is

continuous at x = O and x = 1, u( +O,s) = u(O,s) and u(l - O,s) = u(l,s).
The solution is
(1) u(x,s) = Fasinh xjS

s sinhjS ,
where the symbol fi denotes some branch ofthe double-valued function sto
As long as fi
represents the same branch, regardless of which one, in both
,I
L
i
i
{:::::::::~:~::::}o
O x-l . Fig.75
PROBLEMS IN HEAT CONDUCTION [SECo 75 221
numerator and denominator, the quotient of hyperbolic sines can be written
(2) sinh x-fi - x-fi + (x-fi)3/3! + . .. - x + (x3s/3!)+ ...

sinh -fi - -fi+ (-fi)3/3! + ... - 1+ s/3!+ ... '
except at those points where sinh -fi = O,namely,
(3) s = O, (n = 1,2,.. o).
The final member of Eqs. (2) is the quotient of two power series in s
that converge for every s. It follows that u(x,s) is an analytic function of s
except for isolated singular points (3)0 It also follows from Eqso (1) and (2)
that
(4) lim su(x,s) = Fox;

s-o
therefore when O, the function u(x,s) has a simple pole at s = O with
x =1-'
residue Fox. We note that u is real-valuedwhen s is real.
Now let the branch cut of .jS be taken as the positive real axis, so
that -fi is analytic on the negative real axis. Then u(x,s)has the fractional
form p(x,s)/q(s),where the functions
p(x,s) = Fosinh
s x-fi, q(s)= sinh-fi
are analytic at s = - n2n2, and
Hence s = - n2n2 are simple poles, where the residues of e"u(x,s) are
22 .'
P (x,-n n ) _"2,,2, F.01 SIn nnx " ' 2 2
(5) ' e
2 2 = 2 2 ( - 1) 2nm exp ( - n n t) o
q ( -n n ) -n n
At least formally then (Sec. 71) the inverse transform of u(x,s) is the sum of
the residues (4) and (5) of e"u(x,s) at the poles (3); that is,
2 00 (-1)"
(6)
[
U(x,t) = Fo x + - L-
n "= 1 n ]
exp ( - n2n2t) sin nnx .
This formal solution can be verified by showing that the function

defined by formula (6) satisfies all the conditions of our boundary value
problem (see Probs. 5 and 6, Seco81). But we shall now see that the theory
in the precedingchapter enables us to make the verificationin another way
that has some advantages.
76 THE SOLUTION ESTABLlSHED
We have seen that the function
sinh x.jS
u(x,s) = F.o 1:
s sinhy s
is analytic with respect to s in any half plane Re s ~ y where y > O. To
examine its order in that half plane, let us write
thenRe.jS = .fi cos (()/2)> ~ ~ .JYi2. Thus

sinhx.jS 1: 1 - exp ( - 2X.jS)
I
.
smh.jS I
= I
exp [(x - l)y s]
1 - exp(- 2y's) ] [ I
1 + exp(- 2x.jYji) r.:;

) exp[-(1 - x)y r/2]~ M exp[-(1 - x) 02],
< 1 - exp -
(2Ji72y/2 y'l'"-
where M -
= 2/[1 exp (- 2.JYi2)]. Thus when XI < 1, constants Mm'
independent of x over the interval O~ x ~ Xl> exist such that for each
- integer m = 2, 3,. . .
~lu(x,s)1< FoM~-lexp[-(1 - XI).fi1i] < Mm (Res ~ y);
that is, u satisfiesthe order condition
(1) (m = 2,3, .. ., Re s ~ y, O~ x ~ XI)'
According to Theorem 8, Seco70, the inversion integral of u along the

line Re s = y, therefore, represents the inverse transform U(x,t) of u(x,s),
- (2) U(x,t) = L- l {u(x,s)} (t ~ O,O ~ x < 1),
and U is a real-valued continuous function of t that satisfies the required
initial condition
(3) U(x, + O) = U(x,O) = O (O~ x < 1)
and also the condition
(4) Ut(x,t) = L - l {su(x,s)} (O ~ x < 1).
Now in view of Theorem 9, Seco70, U is a continuous function of x
and t together when t ~ Oand O ~ x ~ XI; thus
(5) U( + O,t) = U(O,t) = L - l {u(O,s)} = O
since U(O,S)= O. Likewise, the deriva tives

cosh x.jS uxx(x,s) = su(x,s),
u..,(x,s) = Fo !:. !:'
V s smh V s
are l!'i(s-m)in the half plane, for each integer m, uniformly with respect to x
when O ;;;;x ;;;;XI < l. This is evident by comparing those functions with
u itself. Therefore
ux..,(x,t) = L - l {ux..,(x,s)} = L - l {su(x,s)} (O< X < 1),

and it follows from Eq. (4) that U satisfies the heat equation
(6) U,(x,t) = Uxx(x,t) (O< x < 1).
We have now shown that our function (2) satisfiesall conditions of
our boundary value problem exceptthe end condition
(7) U(I - O,t) = Fo (t > O).
It is evident that U satisfies the condition
U(I,t) = L-I{U(I,s)} = L-I{~O} = Fo (t> O); I
but this does not assure us that U(x,t) -. Fo as x -. 1, which is the condition ~
the temperature function should satisfy.
By writing hyperbolic sines in terms of exponential functions and
carrying out one step of a division, we find that I
sinh x.Js !: exp[-(2 - x).Js] - exp(-x.Js)

(8) = exp[ - (1 - x)V s] + .
sinh.Js - exp (.Js) - exp(-.Js) I
It is the first term on the right that has a weak order with respect to s when I
x = 1. Now we can write ~
(9) u(x,s) = Foexp

s [ -(1 - x).jS] + g(x,s)
where
(10)
g(x,s) = ~oexp [-(1 + x).jS]exp[1 -2(1
- exp-(-2.jS)
x).jS] - 1.
On the half plane Re s ~ y we can see that

IF.I 2
(O ;;;;x ;;;; 1)
Ig(x,s)1< ~exp(-J12)1
r - exp(- .fiY
2y)
so that g is l!'i(s-m)there for each integer m, uniformly with respect to
x (O;;;X ;;;1). Consequently, Li l{g} represents a function G(x,t) that is

continuous in x and t and, since g(l,s) = O,
G(I - O,t) = G(l,t) = L-l{g(1,S)} = O (t ;;;O).
With the aid of the inversion integral evaluated in Seco 74, we find
! that
(O ;;;x ;;; 1, t > O).

(11) L-l{~O exp [-(1 - X)j;]} = Foerfc(12~)
It now follows from Eq. (9) that
1- X
(O ;;;x ;;; 1, t > O),
. (12) U(x,t) = Foerfc ( 2.Ji ) + G(x,t)
and U(l - O,t) = Fo + G(I,t) = Fo when t > O.
- Our function (2) thus satisfies the end condition (7), and it is therefore
completely established as a solution of our boundary value problem.
Moreover, we have shown that the transform of our temperature function
is the function u(x,s) from which we obtained U(x,t). Some interesting
properties of U will follow from the order properties of U.
We still have to prove that the series obtained in the last section
represents our solution (2).
-
77 THE SERIES FORM ESTABLlSHED

We have seen that the function
. sinh xj;
u(x,s) = F.o
s sinh j;
is analytic except for the poles s = Oand s = - n2n2 and that its inversion
integral converges to a function U(x,t) that is a solution of our boundary
value problem. The series representation of U(x,t), given in Seco75, is valid
- provided the integral
1Cn e'%u(x,z)dz,
taken along a curve Cn of a family of curves (n = 1,2,...) between the poles,
tends to zero as n tends to infinity (Sec. 71).
To select curves Cn so that lu(x,z)1has a suitable order property when-
ever z is on Cn, we first note that
Isinh ;;12 = sinh2 (-fi cos~) + sin2(-fi sin~).

Hence if fi sin!8 = an,where

(1) (n = 1,2, .. .),
then
(2)
Isinhj;12 = sinh2(fi cos~) + 1 ~ 1.
Let us therefore take Cn as the arc of the parabola

an2
(3) r- -- 2an2
- sin2(}/2- 1 - cos()
that lesto the leftof the line~ = y, wherey is any positiveconstantand
z = ~+ ir(Fig. 76). Then when z is on Cn
(4) -zu(x,z) 2 = sinh2(xfi cos (}/2)+ sin2(xfi sin (}/2)<1

I
Fa 1
sinh2(fi cos (}/2)+ 1 = ,

because O~ x ~ 1 and sinh2y increases when Iyl increases. Thus u(x,z)
is O(z- 1)on Cnand, according to Theorem 10,Seco73, the seriesof residues
of eZ1u(x,z)converges to the inversion integral for all positive values of t.
The series (6),Seco75, therefore does represent a solution of our problem;
that is, our solution can be written
2
[
00 (-I)n
(5) U(x,t)= Fo x + - L -
;n= 1 n ]
exp (- n2'nh) sin n;x (t > O).
-a~
Fg.76
We have established the sum of series (5) as a solution of our boundary

value problem by identifying it with our solution L- 1{u}. Owing to the
uniforrn convergence of such series when t ~ to > O (Prob. 5, Seco 81)
however, we find easily that series (5) represents a function that does satisfy
the end conditions U(I - O,t) = Fo, U(+0, t) = O and the heat equation.
Hence a simpler verification consists of those observations coupled with
our brief proof that LI{U} satisfies the initial condition (3), Seco76, after
establishing the series representation of the inversion integral.
We found in Seco 76 that L-I{U} = O when t = O and O~ x < 1.
Certain order conditions satisfied by u on Cn are actualIy adequate to show
that the series representation (5) of L -1 {u} is valid when t = O and con-
sequently that
2 ex> (- 1)" .
(6) L1 -smn1rx
x= -- 1tn= n
(O ~ x < 1).
For if is a smalI positive angle and z is on either part of any of the parabolic
paths Cn on which 181 < 1t - , it can be seen from the Eq. (4) that for each
fixed x (x < 1)a number M. independent ofn exists such that lz1'u(x,z)1< M..
for any constant k. This order property, together with the inequality (4)
satisfied on the remaining part of Cn, is sufficient to show that the representa-
tion (5) is valid when t = Oand x < 1. The proof, made by first selecting
smalI and then n large, is left to the problems. It establishes the Fourier
sine series representation (6) of the function x on the interval O ~ x < 1.

We shalI use a similar procedure in Chap. 9 to establish a generalization of
Fourier series representations of arbitrary functions.
78 PROPERTIES OF THE TEMPERATURE FUNCTION

It was shown in Seco 76 that for any integer m the transforrn u(x,s) of our
temperature function U(x,t) is (i)(s-m)in a right half plane of the variable s,
uniforrnly for all x in any interval O ~ x ~ Xl where Xl < 1. The derivatives
of u with respect to x also satisfy that order property. As a consequence,
our temperature function possesses the folIowing properties, according to
Theorems 5, 8, and 9 of the preceding chapter.
The function U(x,t) is a continuous function of both x and t when
t ~ Oand O ~ x < 1, and each of its derivatives with respect to x or t has
this continuity property.
At any interior point of the bar, the temperature begins to change
very slowly at the time t = O,since U(x,O)= Oand
(1) U,(x,O) = Utt(x,O)= Uttt(x,O)= ... = O (O~ x < 1).

Of course, U(x,t) is not identicalIy zero because u(x,s) is no1. In fact, from
the representation (3), Seco 49, of U(x,t) as a series of error functions, we can
see that
l
(2) whenever t > Oand O < x < l.
F:o U(x,t) > O
The flux of heat through any section x = xo,

cI>(xo,t)= -KUixo,t),
where K is thermal conductivity, has the transform
,/., - - K coshxo';; if O ~ Xo < 1.

'1'(xo,s) - Fo r:. r:
v s smhv s
From the order of cjJwe can see that the flux also changes very slowly from
its initial value cI>(xo,O)
= O,since
(3) cI>,(xo,O)= cI>rr(xo,O)
= ... = O when O ~ Xo < 1.
In Seco 76 we found that
l - X
U(x,t) = Foerfc ( 2ft ) + G(x,t) (O~ x ~ 1,t > O),

where G and its derivativesvanish as t -+ O. Therefore,the flux through the
right-hand faceof the bar can be written
(4)
and
(5) lim
,-o [
cI>(I,t) + F~ J
v 1I:t
= o.
That is, the flux at that face is (!)(I/.Ji) as t -+ O; it becomes infinite as

t -+ Olike - FoKIFt.
The total quantity of heat that has passed through a unit area of the
face x = l up to time t can be written
in view offormula (4). Thus

Since Gx(l,t) is continuous when t ~ O,

(6) lim Q(1,t) = O,
t-O
a condition that would not be satisfied if there were an instantaneous source
of heat over the surface x = 1 at t = O. Such a source is an idealization of an
actual situation in which a large quantity of heat is generated over a surface
in a very short time interval, by combustion, for instance.
79 UNIQUENESS OF THE SOLUTION

Qur treatment of the boundary value problem is not strictly complete until
we have shown that our solution is the only one possible. The physical
problem of the temperatures in a bar with prescribed initial temperature
and prescribed thermal conditions at the boundary must have just one solu-
tion. If we have completely stated the problem as one in mathematics, that
problem must also have a unique solution.
The conditions we have imposed on U(x,t), namely,
(1) U,(x,t) = U xX<x,t) (O< x < 1, t > O),
(2) U(x,+O) = O (O< x < 1),
(3) U(+O,t)= O, U(1 - O,t) = Fo (t > O),
are not sufficient to ensure just one solution. They do not exclude the
possibility of instantaneous sources of heat at the ends of the bar at t = O.
The equation of conduction (1) is the statement that heat distributes itself
interior to the bar after"the time t = O,by conduction. In the derivation of
that equation, it is assumed that the functions U, U" Ux, and Uxx are con-
tinuous with respect to the two variables x and t, interior to the solid after
conduction begins. We shall therefore require our solution to have those
properties of continuity. Physically, the presence of heat sources interior
to the bar when t > Ois then prohibited.
Let the required temperature function satisfy the conditions (1), (2),
and (3) and the following continuity and order conditions.
l. U is a continuous function of x and t when t > Oand O ;;;; x ;;;;1,
and U, Ut, and U x are continuous when t ~ Oand O ;;;;x < l.
2. Constants exand M exist such that IU(x,t)1 < M eat over the strip
t> O, O < x < 1; also for each Xl (O < Xl < 1), a constant Ml exists such
that the functions Ut and Ux are less in absolute value than M 1eat over the
strip t > O,O < x < Xl'
We could write ex= Ohere, simply requiring the functions to be bounded

over the regions specified; but it is easier to show that our functions are
PROBLEMS IN HEAT CONDUCTlON [SECo 79 229
&(e'tt)than to pro ve they are bounded for all t. Note that V xx satisfies the
eonditions imposed on VI because V, = V xx'
We have seen that the funetion
(4) V(x,t) = L - 1 Fosi~hxfi

{ ssmhfi }
satisfies conditions (1), (2). and (3) and that it is eontinuous, together with
V, and Vx, when t ~ Oand O ~ x < 1. We saw also that
l - X
V(x,t) = Foerfe ( 2Jt ) + G(x,t) (O~ x ~ 1,t > O),

\vhere G is continuous and &(e')when t ~ O and O~ x ~ 1, if IX> O,
aecording to the eharaeter of g(x,s). The error funetion here is eontinuous
in x and t when t > Oand O~ x ~ l. Hence the funetion (4) satisfiesour
eontinuity requirements 1.
The first of eonditions 2 is satisfied beeause that error funetion is
bounded over the strip t.> O,O < x < 1, and G is l!J(e')there. The order
condition there on V, and V x when O < x < Xl is satisfied according to the
representation (4).
Qur funetion (4) therefore satisfies eonditio~s (1), (2), (3), and the
eontinuity and order eonditions. If a seeond funetion V does so, then the
funetion
(5) W(x,t) = V(x,t) - V(x,t)
also satisfies those eontinuity and order eonditions, and also these homo-
geneous eonditions:
(6) W;(x,t) = JtYxx(x,t) (O< x < 1, t > O),
(7) W(x,O) = O (O< x < 1),
(8) W(O,t) = O, W(I,t) = O (t > O).
In writing eonditions (8), we have used the faet that W is eontinuous when
t > Oand O ~ x ~ 1. We shall prove that this problem in W has only the
trivial solution W(x,t) == O in the class of funetions that satisfy the con-
tinuity and order eonditions above.
Sinee W satisfies those eonditions, its transform w(x,s) exists when
Re s> IXas well as the transforms w'(x,s) and w"(x,s) of JtYxand Wxx, where
the primes denote derivatives with respeet to x. Moreover, those transforms
are eontinuous funetions of x and s when O ~ x < l. In view of condition (7)
the transform of W; is sw. It follows from conditions (6) and (8) that for
everys in the right half plane Re s > IX
(9) sw(x,s) = w"(x,s) (O< x < 1),

(10) w(O,s) = O, w(l,s)= O.

The continuity of w with respect to X actually extends to the point
X = 1. To prove it, we write
'O GO
W(X,S)= i O e-S'W(x,t) dt + i
.
'0 e-S'W(x,t) dt
(to > O),
where the second integral is a continuous function of X when O ;;; x ;;; 1

that vanishes when x = 1. Since IW(x,t)1 < M eal and Re s > 0(, then
le-S'W(x,t)1 < M throughout the strip t > O, O < x < 1. Thus we can
take to small enough to make the absolute value of the first integral arbit-
rarily small independent of x when O < x < 1. Then for that to the absolute
value of the second integral is arbitrarily small when x is sufficiently close
-
to 1. Hence w(x,s) Owhen x - 1; that is, w(l - O,s) = O= w(1,s).
We have now shown that for each s in a right half p1ane,wis a continuous
function of x (O;;;x ;;;1)that satisfies the linear ordinary differential equation
(9) with constant coefficients, also that w' is continuous when O ;;; x < 1,
and w(O,s)= O. In the theory of linear differential equations it is shown that,
when the value w'(O,s)of the derivative at x = Ois also prescribed, such an
equation has one and only one solution satisfying those continuity con-
ditions. In our case that solution is
w'(O,s) . r:. )
w(x,s) = -fi smh Xv s (O ;;;x ;;; 1 .
But w(l,s) = O,and sincesinh -fi = O,only whens = -n2n2 (n = 0,1,2...)

and not for all s in a right halfplane, then w'(O,s)= O; hence
(11) w(x,s)= O.
Since w(x,s) is the transform of a continuous function W of exponential
order and Jv,is continuous, ourTheorem 7, Seco69, on uniqueness ofinverse
transforms applies to show that W(x,O) = O; that is,
(12) V(x,t) = U(x,t) (O;;;x < 1).
The proof that the problem consisting of conditions (1), (2), (3), and
the continuity and order conditions has just one solution is now complete.
80 ARBITRARY END TEMPERATURES

. Let the temperature of the end x = 1 of the bar be a prescribed function
F(t) (Fig. 58). The temperature function U then satisfiesthe heat equation
U, = Un, the initial condition U(x,+O)= O,and the end conditions
U( +O,t) = O, U(l - O,t)= F(t) (t > O).
As noted in Seco49. the solution of the transformed problem is
(1) ( )
- f( ) sinhx.fi
uX,s - s r:.
sinh v s
Let V(x,t) denote the temperature function found in the preceding

sections when F(t) = 1. Then
(2) u(x,s) = sf(s)v(x,s)
since v(x,s) = sinh x.fi.

ssinh.fi
Now sv(x,s)is the transform of V,(x,t)when O~ x < 1. In view of the con-
volution property, it follows from Eq. (2) that
(3)
U(x,t) = {F(t - T)V,(x;r)dT.
It was shown that V(x,t) is represented by a series:

2 <X>(-1)"
V(x,t) = x + -11:,,=L1- n
exp ( - n211h)sin mr:x.
The series obtained by differentiating this series term by term with respect
to t does not converge when t = O; but it was shown that the function
V,(x,t)is continuous when t ~ Oand O~ x < 1and that
V,(x,O) = O (O~ x < 1).
The differentiated series simply fails to represent V,(x,t)at t = O.
To arrive at another form of the temperature function U, we assume
F continuous, F' sectionallycontinuous, and F of exponential order. Then
L{ F'(t)} = sf(s) - F( + O),
and u(x,s) = F( +O)v(x,s) + L{F'(t)}v(x,s).
Consequently we have the formula
(4)
U(x,t) = F( +O)V(x,t) + { F'(t - T)V(X,T) dT.
The two formulas (3) and (4) give the temperature U(x,t) in terms of
the temperature V(x,t) corresponding to a fixed surface temperature. They
are two forms of Duhamel's formula (Sec. 85). The above series for V(x,t)
can be substituted into formula (4), and it can be shown that the temperature
function can be written
(5) U(x,t) = xF(t) + 2F(+ O) ~

t..,-(- 1)"
e-n2"2I Sin. mrx
n n= 1 n
'
200(-1)"
L -sinn
+ -nn=l mrx i o
F'(t - "t)e-n2,,2fd"t.
81 SPECIAL END TEMPERATURES

When the end temperature F(t) is a specific function, a convenient formula
for U(x,t) may be found directIy from the transform u(x,s).
For example, let
(1) F(t) = At
in the problem of the last section, where A is a constant. Then
sinh xfi
u(X,s) = A 1:'
S2sinh v s
a functionwith a poie ofthe second order at s = O.Wenotedin Seco75that
sinhxfi - x + (x3s13!)+ (xSs215!)+ ...
sinh fi - 1 + (sI3!) + (s215!)+ ...
By carrying out the indicated division here, the first two terms are found
to be x + X(X2 - 1)s3!; hence u(x,s) has the following representation in a
neighborhood of s = O:
X X(X2 1) 00
u(x,s)=
. A ["2
s + 3'-.S L an(x)~J .
+ n=O
The residue of eZlu(x,z)at z = O is therefore (Sec. 72)
x(x2 - 1)
[
A xt + 3! J.
The residue of IU(X,Z)at the simplepole z = -n2n2 is
' sinh x'-; 2A( _1)n-l . 2 2
z z cosh
2A.-;.-; zJ z=-n2,,2= n3n3 Sin (mrx) exp (-n n t).
ConsequentIythe formula for the temperatures can be written
X3 - x 2 00 (_1)n-l
(2) U(x,t) = A -
[ 6
+ xt + "3
n
L
n=l n
3 exp ( - n2n2t) sinmrx .
J
That function can be verified completely as a solution of the boundary

value problem by just the same procedure that was used in Secs.76 and 77.
But the procedure can be simplified in this case in view of the fact that
u(x,s) is (9(S-2) in a right halfplane and on the parabolas Cn, uniformly with
respectto x when O ;;;x ;;; 1. ConsequentIyU is a continuousfunctionof
its two variables for all x and t (O ;;;x ;;; 1, t ~ O); also the series representa-
tion (2) is valid at t = O. Since U(x,O) = O,it follows from Eq. (2) that
12 ex>(_I)n-l
x - X3 = "3
1t n=1 n
3 sin n1tx (O ;;; x ;;; 1).
which is the Fourier sine series expansion of the function x - X3 on the

intervalO ;;; x ;;; 1.
PROBLEMS
1. A bar with its lateral surface insulated is initialIy at unifonn temperature A

(Figo 77)0 Its ends x= O and x = 1are then kept at constant temperatures B and C,
respectivelyo Show fonnalIy that its temperature distribution can be written
U(x,t ) = B +-xC - B + -2 ~ A - B + (-l)n(c - A) o n7tX n27t2kt

L.., sm-exp --,:- o
1 7tn=1 n 1 ( 1 )
Fig.77
2. A slab of iron with diffusivity k = 0.15 cgs unit, 20 cm thick, is initially at OC

throughout,and one faceis kept at that temperatureoIts other faceis suddenlyheated
to 500C and kept at that temperatureo Use the fonnula in Probo 1 to compute the
temperature of the midsection of the slab after 5 min, to the nearest degree, and compare
the result with a corresponding temperature found in Probo2, Seco47, for a semi-infinite
solid of the same material. Ans. 145C.
3. For fixed values of A, B, C, and k, note that the temperature U(x,t) in Probo 1 is
detennined by the two numbers xii and tl120For two slabs with those same fixed values
and widths 11and 12,show why the temperature ofa section x = XI in the first at time ti
is the same as the temperature of the corresponding section x = x2(x2112 = xdl) of
the second at time t2 if
Thus the slowness of heating varies as the square of the widtho

4. When k = 1= 1 for the bar in Probo 1, let Uo(x,t) denote the temperatures in case
C = Oand W(x,t) the temperatures when C is replaced by F(t), so that W(l,t) = F(t),
,.
234 SEC. 81] OPERATIONAL MATHEMATICS
W(O,t) = B, and W(X,O)= A. Ir V(X,t) is the function (5), Seco80, so that V, = V xx'
V(x,O) = V(O,t) = Oand V(I,t) = F(t), show by superposition that
W(x,t) = V o(x,t) + V(x,t).
5. When t ~ to > O and O ~ x ~ 1, prove that the series (6), Seco75, and the series
obtained by differentiating that series once or twice with respect to x or once with
respect to t are unifonnly convergent with respect to x and t. The series then represent
continuous functions of x and t, and tennwise differentiation is valid, since the terms
themselves are continuous functions. As a consequence, show that the function V(x,t)
defined by the series satisfies the heat equation V, = Vxx when O < x < l and t > O,
and end conditions V( +O,t) = O, V(I - O,t) = Fo, when t > O.
6. According to Abel's test for uniform convergence, a series 1::",. An T,,(t) converges
unifonnly with respect to t if the series 1::'=1 An converges, and if functions T,,(t) are
bounded unifonnly with respect to t and n and such that T,,+(t) ~ T,,(t). Given that
the Fourier series representation (6), Seco77, is valid, use Abel's test when O ~ t ~ to
to show that the function defined by Eq. (6), Seco75, is continuous with respect to t
when t ~ Oand that it satisfiesthe initial condition V(x,+O) = Ofor eachfixedx such
that O < x < l. This, together with the result in Probo 5, again establishes the solution
(6), Seco 75.
7. Give details of the proof outlined in Seco 77 of the Fourier sine series representation
(6) ofthe function x(O ~ x < 1). Suggestion: Show first that when z is on the parabolic
arc Cn and 181< 1t - (, then
ZU(X,Z)
~ cosh(x-fi cos8/2) < 2 exp[-(1 - x>-fi sin (/2].
I Fo I
- cosh(-fi cos8/2) l + exp(- 2-fi sin(/2)
8. As in Seco80 let V(x,t) denote temperatures in the slab with faces x = O and x = l
when V(x,O) = V(O,t) = O, V(l,t) = 1, and k = 1, and write V(x,t) = O when t < O.
Let V(x,t) be the temperatures under the same conditions except that the face x = l is
kept at temperature A from t = O ~o t = to and thereafter at temperature zero; thus
V(I,t) = A - ASo(t - to) (t > O).
(a) Show that

V(x,t) = A[V(x,t) - V(x, t - to)].
(b) Note that the total quantity of heat conductedacrossa unit areaof a section
x = xo, Q = - K So Vx(xo,t)dt, is the value of L{ Vx(xo,t)} when s = O if that trans-
fonn exists when s = O. Thus for the above slab with temperatures V show fonnalIy
that
Q = -AKto.
9. Obtainthesolutionof the temperatureproblem
V,(x,t) = V xx(x,t) (O< x < 1,t > O),
V(x,O) = 1, V(O,t) = V(I,t) = O,
I For a proofsee the author's "Fourier Series and Boundary Value Problems," 2d ed., chap. 10,
1963, which also treats uniqueness of solutions.
~
~
PROBLEMS IN HEAT CONDUCT/ON [SECo 81 235
in the fonn
~
U(x,t ) = -4 L..- sin(2n - 1)1tx
1
2 2
exp [ - (2n - 1) 1t t] .
1t n= 1 2n -
10. Derive the fonnula
- ~ ~ (_l)n-I
I -
[- (2n - 1)21t2t
U( ) (2n - 1)1tx
x,t - 1t /;;'1 2n - I cos 21 exp 412 J
for the temperatures in a wall with its face x = O insulated and its face x = 1 kept at
temperature U = 1, if the initial temperature is zero (Fig. 78) and if k = l.
UI%,OI-O U-l
o %
Fig.78
11. Establish the fonnula in Probo 10 as a solution of the boundary value problem.
12. Find U in Probo 10 as a series of error functions.
Ans. U = I - U o(x,t), where U o is the function (4), Seco48, when Uo = k = l.
13. Obtain the solution of the problem in Seco48 in the fonn
) - 4uo ~ (-Irl (2n- l)1tx

U(
x,t - 1t nf 2n - I cos 21 exp [- (2n - 1)21t2kt
412 J.
14. Let the temperatureof the facex = 1of the wall in Probo10 be F(t),where F is
continuous,F' is sectionallycontinuous,and F(O)= O. Derive the temperaturefonnula
4 00 (-l)n-I (2n - l)1tx
L1 2n
U(x,t) = F(t) - -1t n= . cos ..., Gn(x,t),
where
' (2n - 1)21t2r
Gix,t) = Jo F'(t - r) exp [ - 412 J dr.
15. The face x = O of a slab is insulated. The temperature of the face x = 1t is
U = t(t ~ O). Ir the slab is initially at temperature zero throughout and k = 1.
(a) derive the fonnula
1t2 - X2
U(x,t) = t - ---z-
- .1tn=1
16 f
(-I)"cOS (n - tlx exp -(2n - 1)2~
(2n-l) [ 4J (O~ x ~ 1t,t ~ O).
(b) Verify this solution of the problem.

. ..
16. Solve the boundary value problem

cU,(x,t) = KU xx(x,t) + R(t), Ux(O,t)= U(I,t) = U(x,O)= O,
for the temperatures in an internally heated bar (cf. Probo 11, Seco51) in the form
U(x,t)= L R(t - -r)G(x;t)d-r,

4 00 (_1)n+ 1 (2n - I)nx (2n - 1)2n2kt
L 2n-
G(x,t)= -nc n=1 I cos 2 exp [ - 4 J
where k = K/c. Note that G(x,t) is formally the temperature distribution when
R(t) = b(t).
17. Ir heat is introducedthrough the face x = 1 of a slab at a uniform rate A per
unit area while the face x = O is kept at the initial temperature zero of the slab, the
temperature function U satisfies the conditions
U,(x,t) = kU xx(x,t) (O< x < 1,t > O),
U(x,O)= U(O,t)= O,
Writem = n - t and derive the solution
A 2 00 (-I)n..
U(x,t)= _
[
K x + 2n n= L mr 1
s10 (mnx) exp (-m2n2kt)
J.
18. At the face x = Oof a wall, heat transfer takes place into a medium at temperature
zero according to the linear law of surface heat transfer, so that
Ux(O,t)= hU(O,t) (h > O).
Units are chosen so that k = 1,and the wall has unit thickness. Ir the other conditions
are those indicated in Fig. 79, set up the boundary value problem for temperatures
U(x,t) and show that
( )-
uX,s - I h sinhx-fi + -fi coshx-fi .
s h sinh -fi + JS cosh -fi
, ,
U(.x.O)-O
U(1,t)-l
0'-
-10 %
-
----- Fig.79
Obtain formally the solution
U(x,t) -- hx + I
h+ I
- 4 L-
~ 2sinan(1-. 2x) exp ( - an2t),
n=1 an - s10 an
where IXI, 1X2, . . . are the positive roots of the equation

IX
tan IX= - Ii'
Show how those roots can be approximated graphicalIy when the value of h is known,
and note that IX.is only slightly greater than (n - t)71:when n is large. To show that
s = Oand s = -IX/ are the only singular points of u(x,s),write p(z) = h sinh z + z cosh z
and z = A + ip.and prove that p(j';) = O only if s = O or s is real and negative by
showing that
2Ip(z)1 ~ Ie\/(h + A)2+ p.2- e-AJ(h - Af + p.21> O

if A > O or if A < O. Hence A = O and z = ip. when p(z) = o.
82 ARBITRARY INITIALTEMPERATURES
Let the initial temperature distribution in a bar or slab be a prescribed func-
tion g(x) of the distance from one face. When the lateral surface of the bar is
insulated and the ends are kept at temperature zero (Fig. 80), units can be
selected so that the boundary value problem for the temperature in the bar
. becomes
U,(x,t) = Uxx(x,t) (O< x < 1, t > O),
U(x,+O) = g(x) (O< x < 1),
U(+O,t) = U(l - O,t) = O (t > O).
The problem in the transform of U(x,t),

(1) u"(x,s) - su(x,s) = - g(x) (O < x < 1),
(2) u(O,s) = u(l,s) = O,
can be solved by any one of several methods, including a Laplace trans-

formation with respect to x, or the process ofusing a Green's function to be
described in Chap. 9. Regardless of the method used, the solution can be
written in terms of Green's function R for the problem, in the form (see
Probs.9 to 11, Seco83)
(3) u(x,s)= - f R(x,~,s)g(~)d~,
O' U(x,O)-g(x) O'
Fg.80 o
l.
238 SECo 82J OPERATIONAL MATHEMATICS
where R is this symmetric funetion of x and ~:
(4) R(x,~,s)= - sinh [(1 - x)JS] sinh~JS (O~ ~ ~ x),

JS sinhJS
- - sinh[(1- ~)JS] sinhxJS (x ~ ~ ~ 1).
- JS sinhJS
By writing Maclaurin's series in powers of JS
for the hyperbolie sines
and properIy defining R(x,~,O), we see that, regardless of whieh branch of
st is represented by ./S, R is an analytie funetion of s exeeptfor the singular
points
(5) (n = 1,2,.. .).
Ir g is seetionally eontinuous, the order of integration with respect to ~ in
Eq. (3)and differentiationwith respect to s can be interchanged,and hence
u(x,s) is analytic except for the singular points (5). By taking the branch
cut ofJS along the positive real axis and selecting q(s) as sinh JS
in formula
(7), Seco 72, we find that the points (5) are simple poles of u(x,s), where the
residues of e'lu(x,s) are
%
exp ( - n2n2t)
- 1
"leos nn [f.o g(~)sin nn(1 - x) sin nn~d~
+ f g(~) sin nnx sin nn(1 - ~) d~ 1

Since sin nn(1 - x} = - cos nn sin nnx, the residues can be written
2 exp ( - n2n2t) sin nnx f gR) sin nn~ d~.
Formally, therefore, the solution of our problem is

~ l
(6)
U(x,t) = 2 n~1 exp (- n2n2t) sin nnx f. o g(~) sin nn~ d~.
When t = O, the series here is the Fourier sine series for the function g(x),
on the intervalO < x < 1. In fact, the boundary value problem in U(x,t)
here is especially well adapted to the classical method of solution by using
separation of variables and Fourier series, a method we shall discuss in
Chap. 9.
VERIFICATION
Solution (6) can be established by the procedure used in Secs. 76 and 77.
Since some variations are needed, we outline the method here. Over any
half plane Res ~ y > O we find that IR(x,~s)1< ME(x,~,r)f0-' where

E(x,~,r)= exp (-Ix - ~1j;12), r = Isl, and M is a constant. Since g is
H g
bounded and E d~ and E d~ are 19(r-t), it follows from Eq. (3) that u is
19(r-l). This is not sufficient to show that the inversion integral applies to u.
For the sake of brevity we assume g" continuous and g'" sectionally
continuous (O~ x ~ 1). We integrate both integrals by parts in the expres-
slOn
u(x,s)= - I: R(x,~,s)gR) d~ - f R(x,~,s)g(~) d~,
and introduce an integral P of R with respect to ~,where
(7) P(x,~,s)= - sinh [(1 - x)j;] cosh~j; (O~ ~ < x),

ssinhj;
- cosh[(1- ~)j;] sinhxj; . (x < ~ ~ 1),
- ssinhj;
and thus isolate the term In u of weak order by writing
l
g(x)
(8) U =-
s
+ g(O)P(x,O,s)
- g(1)P(x,l,s)+ fo Pg'(~)d~.
The integral in Eq. (8) is 19(r-3/2)in the half plane Re s ~ 1'. From Seco
76 we find that the inversion integrals of P(x,O,s)and P(x,l,s) exist and satisfy
certain boundary conditions. Since Li -1{g(x)fs} = g(x), we can conclude
that Li - 1{u} represents a function U(x,t) that satisfies the conditions
U(x, +0) = g(x) and U(+0, t) = U(l - O,t) = O.
From Eq. (8) we can show that
u' = g(OW,,(x,O,s)- g(l)P..,(x,l,s) + fg'(~)Px(X,~,S) d~

and, by an integration by parts here, that u'(x,s) is 19(r-3/2)when O < x < 1.
That u"(x,s)also has this same order followsfrom the fact that u" = su - g
and integration by parts twice in Eq. (8). Thus Uxx = Li - 1{u"} and, since
-
s(u gjs) is 19(r- 3/2), then
(O< x < 1). .

Li -1 {s( U- ~)} = [U(x,t) - g(x)], = U,(x,t) = Uxix,t)
The function
(9) U(x,t) = Li - 1 {u(x,s)} (y > O)
is therefore established as a solution of our problem.
From Eqs.J.3) and (4) it can be shown that u(x,s) is O(r-t) when s is on
the parabolas V
r sin (J/2 = (n - t)n. Therefore the series (6) represents the
solution (9) when t > O.
In addition to this, we can see from Eq. (8) that the inversion integral
of the function u(x,s) - g(x)/s converges to zero when t = O and that it is
represented by its series of residues when t = O, provided O< x < 1. It
follows that
00 l
(lO)
f
2 n~l sin nnx o g(~)sin nn~d~ - g(x) =O
(O< x < 1),
which is the Fourier series expansion mentioned above.

The results hold true if g or its derivatives are sectionally continuous
instead of continuous. The proof is longer, since it involves the writing of
each integral as the sum of integral s over intervals on which the functions
are continuous.
83 TEMPERATURES IN A CYLlNDER
Let us derive formally the temperature function U(r,t) for a solid circular
cylinder of infinite length whose initial temperature is zero and whose
surface is kept at unit temperature (Fig. 81). Units of time and length can
be chosen so that the boundary value problem becomes
I
U,(r,t) = Urr(r,t)+ -Ur(r,t) (O~ r < 1, t > O),
r
(1) U(r,O) = O (O~ r < 1)

U(I,t) = 1 (t > O),
where the cylindrical coordinate r is the distance from the axis ofthe cylinder.
I
I
I
I
U(r,OI-OI U-l
I
I
I
~~--T----
,," r I 1 .......
Fig.81
....
The function U is to be continuous interior to the cylinder, at r = O in

particular. We can assume that U is bounded over the domain O < r < 1,
t > O. Then the formal problem in the transform u(r,s) is
d2u l du
(2) su =- +-- (O ;;; r < 1,Re s > O),
dr2 r dr
1
(3) u(r,s) continuous (O ;;;r ;;; 1, Re s > O).
u(l,s) = s'
Let J1.denote a complex constant. The substitutionst = J1.Zand

y(z) = Y(J1.z)in Bessel's equation (7), Seco21, leads to the form
(4) z2y"(Z) + zy'(z) + (J1.2z2- n2)y(z) = O (n = O, 1,2,. . .).
The solution of Bessel's equation (4), which is continuous at z = O as well

as elsewhere, is y = CJ n(J1.z),where C is any constant. Equation (2) is a
special case of that equation in which n = O, J1.2 = -s, and z = r, so its
solution that satisfies our continuity requirement is
(5) u(r,s) = CJo(irj;)
where (Sec. 21) for each branch of j;

2k
00 1 r
(6)
Jo(irj;) = k~o(k!)2 ( 2) ~.
Since this power series in s converges for all complex s, it represents an entire
function of s for each fixed value of r.
The condition u(l,s) = l/s determines C, so that
(7) u(r,s) = Jo(irj;).

sJo(ij;)
Now the zeros ofBessel's function Jo(z) are all real and simple; they consist
of an infinite sequence of positive numbers (Xn(n= 1,2,...) together with
their negatives - (Xn,such that (Xn-. 00 as n -. 00.1 Their values are tabulated ;
in particular, to four significant figures.
(Xl = 2.405, (X2 = 5.520, (X3 = 8.654, (X4 = 11.79.
Thus JO((Xn) = O and J((Xn) =FO. In view of formula (7), the

singular points of u(r,s)in the plane of the complex variable s consist of the
simple poles
(8) s = O, s = -(X/ (n = 1,2,.. .),
1For such properties of Bessel functions see, for instance, chap. 8 of the author's "Fourier
Series and Boundary Value Problems," 2d 00., 1963. The function Jo(iz) is aIso written as 10(z).
since s = -IX/ whenever ifi = :f:IX",

for a branch of fi that is analytic
over the negative real axis so that J - IX"2 is defined, say the branch
fi = .jGiei812where O < (J < 21t.
The residue of u(r,s)e"1
at s = Ois 1,and at s = - IX"2 it is
e"1 Jo(irfi) 2 Jo(-IX"r) 2
-
s dJo(ifi)/ds ]$=_,,"2
=-
-IX" J~(-IX,,)
exp ( - IXt)
" ,
where J~(z)= dJo(z)/dz= -J1(z) = J1(-z). FormalIy, then,
CO J (IXr)
(9)
U(r,t)= l - 2 ,,=1L (
IX"~ 1 IX", exp(-IX,,2t).
To write this formula in terms of standard units of length and time,

centimeters and seconds, for example, let P denote the radial distance and
T the time in such units. Ir the radiusof the cylinderis Poand the thermal
diffusivity of the material is k, then to transform the heat equation in problem
(1) into Uf = k(Upp + Up/p) (O~ P < Po),we put
P kT
r = -, t = -Z,
Po Po
where r and t are the variables used in formula (9). AIso write V(P,T) = A U(r,t)
so that the constant surface temperature is arbitrary:
V(Po,T) = A.
Our temperature formula then takes the form
(10) V(P,T) = A
[ - ,,=f
I 2 JO(IX"p/PO)exp
1 IX"J 1(IX,,) (
- IX/~T
Po }J
.
PROBLEMS
1 . The initial temperature of a slab is U(x,O)= Ax. Ir the faces x = O and x = I
are kept at temperature zero, derive the temperature formula
2AI 00 (-1)"-1 n21t2kt . n1tx
U(x,t)= - L
1t ,,=1 n
exp
(
- y-
I )
Sln-.
I
2. Derive the following formula for the temperature function in Probo 1:
~ (2" + 1)1+ X (2n + 1)/- x

L- { en
U(x,t)= Ax - Al ,,=0 [ 2, /ktkt ] - erf[ ~Ktr:. ]} .
'"
3. Ir the slab in Probo 1 is 20 cm thick and is made of iron for which k = 0.15 cgs unit,
and if the initial temperature varies uniformly through the slab from O to 100C,
find to the nearest degree the temperature at the center after the faces have been kept at
OC(a) for 1min, (b) for 100mino Ans. (a) 48C; (b) OC.
4. The facesx = Oand x = Iof a slab are insulated (Fig. 82)and the initial temperature
distribution is prescribed: U(x,O)= g(x). Find the transform of the temperature
function U(x,t) in the form
'
l
i
u(x,s) = - k og@R(x,~,s) d~,
where R(x,~,s)= - coshx# cosh(1- ~>#

#sinh/#
when x ~ ~ and R(x,~,s) = R(~,x,s). Then obtain the formula
where
2
i'
an= 1 o g(x)cosdx
mr:x
(n = O, 1, 2, . . .).
U(x.O)-g(x)
o %
1-
Fig.82
5. In Probo4, if g(x) = A when O < x < 1/2and g(x) = Owhen 1/2 < x < 1,show that
-
U(x,t ) -- A 2A ~ (_l)n- (2n - 1)1tx (2n - 1)21t2kt
2 +-'- 2
1tn=n- 1 cos I exp [ - 12 J
.
6. The facex = Oof a slab is kept at temperature zero while the face x = 1 is insulated.
If k = 1 and the initial temperature is U(x,O)= x, derive the temperature formula
-~ -
U(x,t)
00 (-1)"+
- 1t2n~ (2n -
. (2n - l)1tx
1)2SIn 2 exp [- (2n 1)21t2t
4 J.
7. The initial temperature of a cylinder of infinite length is zero. If the surface r = 1
is kept at temperature A from t = Oto t = to and at temperature zero thereafter, and if
k = 1, derive the following formula for the temperatures in the cylinder:
W(r,t) = A[U(r,t) - U(r, t - to),
where U(r,t) is the function defined by formula (9), Seco83, when t 6; Oand U(r,t) = O
when t < O.
8. The flux of heat into a cylinder of infinite length through its surface r = l is a
constant, so that U,(I,t) = Ao If k = l and the initial temperature is zero, derive the
temperature formula
where p, P2' oare the positive roots ofthe equation J (fJ) = O,given that the equation
o o
has only real roots and that J'(Pn)"# 00 Note that according to Bessel's equation
-J;;(x) = Jo(x) - J(x)/x, and since -J;;(x) = J~(x) it follows that J~(Pn)= JO(Pn)o
9. Use symbolic transforms, as in Seco13,to indicate formally that the solution of the
problem
R(O,~,s)= R(I,~,s) = O,
where O < ~ < 1, is Green's function (4), Seco820

10. Multiply the members of all equations in Probo 9 by - g@ and integrate from
~ = Oto ~ = l. Formally interchange the operations of differentiation with respect to
x and integration with respectto ~, as might be justifiedif (x - ~ in Probo 9 were
replacedby l(h, x - ~),to indicatethat the function
u(x,s)= - f R(x,~,s)gR)d~
is the solution of the problem (Seco82)

d2u
~-su=
dx -g(x) ' u(O,s) = u(l,s) = 00
Write R in the form (4), Seco82, and verify the solutiono

11. (a) Note that Green's function R defined by Eqso (4), Seco82, is a continuous
function of x (O~ x ~ 1) when O < ~ < 1, but show that its derivative R" has a unit
jump at x = ~:
(O< ~ < 1)0
(b) Integrate the members of the equation

d2R
-dX2 - sR = (x - ~) (O < ~< 1)
with respectto x from ~ -

( to ~ + (and let ( tend to zero to indicate formally that
R" has a unit jump with respect to x at x = ~o
12. Show that the derivative R,,(x,~,s) of Green's function R in Probo 4 has a unit
jump with respect to x at the point x = ~(O< ~ < 1);also showthat R itselfis a con-
tinuous functionof x (O~ x ~ 1)that satisfiestheseconditions:
d2R
k- - sR = O (x "# 0,
dX2
84 EVAPORATION
FROM A THICK SLAB
We now illustrate a type of problem that is especiallywell adapted to treat-
ment by the Laplace transformation in contrast with other methods.
Let C(x;r)denote the concentration of moisture at time 't" in a porous
semi-infinite solid (x ~ O)with a uniform initial concentration Co (Fig. 83).
Evaporation of moisture takes place at the face x = O into a dry adjacent
medium according to the linear law tbat the outward flux is E[C(O,'t")- O],
where the positive constant E is a coefficient of evaporation. A boundedness
condition on C will take the place of a condition of diffusionat a right-hand
boundary. The condition that C(x,'t")is 19(eat)uniformly with respect to x,
whenever ex> O,will be sufficient. Then if K is the coefficient of diffusion
of moisture in the solid, C(x,'t")satisfies the conditions
Ct = KCxx, (x > O,'t"> O); C(x,O) = Co, (x > O);

(1)
KCx(O,'t") = EC(O,'t"), lC(x,'t")1< M~t.
In terms of new variables t, V, and a positive constant h, where

E
(2) t = K't", V(x,t) = C(x,'t")
Co ' h= K'
the problem in the relative concentration V(x,t) becomes
VI = Vxx, (x> O,t > O); V(x,O) = 1, (x > O);

(3)
Vx(O,t)= hV(O,t), IV(x,t)1< Ne{J1
for some constant N and each positive constant p. The problem in the
transform of V becomes,formally,
su(x,s) - 1 = u"(x,s) (x > O,Re s ~ y > P),

(4)
N
u'(O,s) = hu(O,s),
lu(x,s)1< y-p'
---
O
--10 C(X, O) = Co
--
x
Fig.83
Ir.j';=jr ei912where -11: < (J < 11:and s = rei9,the solution of problem (4)
can be written as
I h
(5)
u(x,s) = -s - s(h + .j';s) exp(- x.j';).
SinceRe.j'; < Owhen e =F11:,
then h + .j'; =FOand u is an analytic
function of s everywhere except on the nonpositive real axis. The ray (J = 11:
is the branch cut of.j'; and of u. The inversion integral on the line Re s = y
applies to both terms on the right in Eq. (5) since the second term is (Il(S-3/2),
uniformly with respect to x. According to our theorems in Chap. 6, it
represents a function
(6) V(x,t) = I - Li - 1 (t > O,x ~ O)

{ s(h +h .j';s) exp (- x.j'; )}
that is (Il(eY')uniformly in x where y can be any positive number. Moreover
that function is continuous in x and t, and V(x, + O)= l. When x ~ Xo > O,
the function inside the braces is (Il(s-m)when Re s ~ y, for each integer m,
and it follows readily that V, = V xx when t > O and x > O. Thus the
function (6) satisfies all the conditions (3) except possibly the condition
V x(O,t) = hV(O,t).
The procedure in Seco74 of altering the path of integration of the
inversion integral can be applied to the last term in Eq. (6) with no essential
change. In particular we find that the integral around the small circle
Isl = ro in Figs. 73 and 74,
1 9=" -x.s
-
211:i
f e
9= -" s(h + .j';)
lf'ds
tends to 1 as ro -. O. The integrIs over the rays s = re-i" and s = rei"

constitute the rest of the value of V(x,t) - 1, and the result is this more
usefulrepresentation of the function (6):
(7) V( x,t ) =~ a) hsinx~ + ~cosx~ e-" dr

J
11:o
2
r (h +r )
(t > o, x ~ O).
Since Iy-l sin yl O,we can see that the integrand in
formula (7)satisfiesthe condition .
xh sinx~/(x~) + cos x~ e-" < xoh + 1 e-rt

I
~(h2 + r) I
h2~
when O ;;; x < xo. Thus the integral (7) converges uniformly with respect
to x(O;;; x < xo)when t > Oand
h e-rt dr
f
a)
V( + O,t)= V(O,t)= -
11: o
h- 2 +ryr
r.: (t > O).
The integral of the derivative with respect to x is uniformly convergent in x

when t > Oand x ~ O,so that
h2
Ux( + O,t) = Ux(O,t)= -
1t o f co e-rt
-h2
+ryr
dr
1: (t > O).
Thus Ux(O,t)= hU(O,t) when t > O,so the function U with representations
(6) and (7) satisfies all conditions in boundary value problem (3).
That function has the transform (5); thus
l h l h
u(Os) =- - =- -
, s s(h + JS) s - h2 JS(s - h2)
and it follows from Probo 4, Seco17,or transform 40, Appendix A, Table A.2,
that
2 fco
(8) U(O,t)= exp (h2t) erfc (h.jt) = ~Jo exp(-y2 - 2h.jty)dy ~ l.
Using formulas (7) and (8) and the substitutions (2), we can write the
concentration C in the slab and at the face as
(9) C(X,-r=
)
2hCo
-
1t f
o
co h sin xy + y cos xy
2 2
y(y + h )
2
exp ( - K -ry ) dy,
(10) C(O,-r)= Coexp (h2K-r)erfc(h~).

Another representation of C(x,-r),in terms of error functions, can be written
from our formula for u(x,s) (Prob. 1, Seco85).
85 DUHAMEL'S FORMULA
In Seco80 we obtained a formula for the temperatures in a bar with variable
end temperature, in terms ofthe temperature function when the end tempera-
ture is constant. We now obtain a more general formula that simplifies
heat conduction problems in the same way.
Let U(x,y,z,t) be the temperatures in any solid, filling a regio n R, that
is initialIy at temperature zero throughout. Let the temperature at every
point of some part S of the boundary be a prescribed function F(t) of time,
and let the remainder S' of the boundary be kept at temperature zero
(Fig. 84).
Ir A(U) represents the linear differential form
l a au a au a au
(1) A(U)=- - K- +- K- +- K- ,
e[ ax( ax ) ay( ay ) az( az )]
S'
Fig.84
the equation of conduction can be written

(2) Vt = A(V) [(x,y,z) in R, t > O].
We assume that the thermal coefficients K and e are either constants or
independent of t. The boundary value problem in V consists of Eq. (2) and
the conditions
(3) V(x,y,z,O) = O interior to R,
(4) V=O on S'
= F(t) onS.
The transform u(x,y,z,s) then satisfies the conditions
(5) su = A(u) inR,
(6) u=O on S'
= f(s) onS.
Let V(x,y,z,t) be the temperature function V when F(t) = 1; that is,
V satisfies the heat equation (2), the initial condition (3), and the surface
conditions
(7) V=O onS'
= 1 on S.
Then the transform v(x,y,z,s) satisfies the differential equation (5) and the
conditions
(8) v=O onS'
-1 onS.
s
Since s is a parameter in the linear homogeneous differential equation
(5), the product of sf(s) by the solution v is also a solution. But according to
conditions (8),
sf(s)v = O onS'
= f(s) onS;
thus the function sf(s)v also satisfies all the conditions (5) and (6). Ir the
problem consisting of conditions (5) and (6) is to have but one solution, then
sf(s)v must be the same as the solution u:
(9) u(x,y,z,s) = sf(s)v(x,y,z,s).
Since sv is the transform of v" it follows from Eq. (9) with the aid of
the convolution that
(lO) U(x,y,z,t) = I: F(t - T)1;(X,y,z,T) dT

o '
f
= ot oF(t - T)V(X,y,Z,T) dT.
These are two cases of Duhamel's formula representing temperatures U
corresponding to variable surface temperatures in terms of temperatures V
in the simpler problem where surface temperatures are constant. A more
general case is given in Probo 14.
Given that Vis the solution of the simpler problem, the solution (10)
ofthe problem in U can be verified directly under reasonable conditions on F.
PROBlEMS
1. From formula(5),Seco84, for u(x,s)and transform 86,AppendixA, Table A.2,
obtainthisformulafor the concentrationofmoisturein the slabx ;;; withevaporation
at the surface:
C(X,f)
= c{ erf(~) + exp (hx + h2 Kf) erfc (~ + h~) J.
2. Identify the two expressions in formula (8), Seco85, for the concentration U(O,t)
at the face of the semi-infinite solid and thus show that < U(O,t)~ 1 when t ;;;O.
3. A layer of finite thickness extending from the face of the slab in Seco84 is initially
dry. The initial concentration of the rest of the semi-infinite slab is uniformo Choose
units so that the relative concentration U(x,t) satisfies the conditions
U, = U""',X> O,t > 0, U(x,O)= So(x - 1), u ,.(O,t) = hU(O,t),
and a boundedness condition; also U and Ux are to be continuous when t > 0, at

x = 1 in particular. Showformallythat
2su(x,s) = exp [-(1 - x)../S] - exp[-(1 + x)j;]h - Jf:. (O~ x ~ 1),

h + yS
and use the tablesof transformsto get theseparticular results:
U(O,t) = exp (h + h2t) erfc (zfi+ hJt),

U(l,t)= ~erf(jr) + exp(2h+ h2t)erfc
(jr + hfi).
U(%.O)-O
A'-
-
Fig.85
4. The face x = of a semi-infinite solid (Fig. 85) is exposed to a medium at constant

temperature A. Heat is transferred from that medium to the face of the solid according
to the law that the flux of heat is E[A -
U(O,t)],where U(x,t) is the temperature in the
solido Thus the boundary condition at the face becomes
U..(O,t)= h[U(O,t)- A],
where h = E/K. Ir the initial temperature is zero, derive the followingformulawith
the aid of the tablesin AppendixA:
U(x,t) = A[ erfc (~) - exp(hx+ h2kt)erfc(hJkt + ~) J

5. Let V(x,t) be the temperaturefunctionfor the solid in Probo4 when A = 1. Let
W(x,t) be the temperature of the solid when the constant A is replaced by a function
II>(t),
so that the medium to which the face is exposed has a variable temperature. Derive
the formula
W(x,t) = f~ II>(t- t)V,(x,t)dt.

6. The temperature of the faceof a semi-infinitesolid x ~ varies in the following
manner:
U(O,t)= A sin rot.
Taking the initial temperature as zero, for convenience, show that, when t is large, the
temperature U(x,t) is approximately V(x,t), where
V(x,t) = A sin (rot - x~) exp(-x~),

a simple periodic function of time. Note that the closed contour corresponding to the
one shownin Fig.73willendose two simplepoless = :!:iwofu(x,s)in this case. (Also,
see Probo 7.)
7. Use the following elementary method to obtain the formula in Probo 6 for the
undamped component V of the temperature function U. Assume V periodic in t with
frequency ro, for each x, and write V(x,t) = 1m W(x,t) where W(x,t) = F(x)ei"", F is
complex-valued, and Jv. = kWn, W(co,t) = 0, W(O,t)= Aei..,. Then V, = kVn,
V(co,t) = 0, and V(O,t)= A sin rot. Thus write the problem in ordinary differential
equations in the function F and show that F(x) = A exp[-(1 + i)xJW/J2k].
8. The diffusivity k of the earth 's soil in a certain locality is 0.005 cgs unit. The tempera-
ture of the surface of the soil has an annual variation from -
8 to 22C. Assumingthe
variation is approximately sinusoidal (Prob. 6), show that the freezing temperature
will penetrate to a depth of approximately 170cm (considerably less, because of the
latent heat of freezing).
9. In Probo 8 find the approximate depth at which the variation of temperature with
time is 6 months out of phase with the variation of the surface temperature. Show that
the amplitude of the variation at that depth is less than 1C.
Ans. x = 705 cm = 23 n, approximately.
10. Let the functions V(x,t) and W(y,t) satisfy the heat equations V. = kJ.~x and
~ = kjy, respectively. Prove by direct substitution that the product of those tem-
perature functions,
U(x,y,t) = V(x,t)W(y,t),
satisfies the heat equation
u, = k(Uxx + Uyy).
Ir in addition V(O,t) = V(a,t) = O and W(O,t) = W(b,t) = O, and if V(x,O) = f(x) and
W(y,O) = g(y), then show that U(x,y,t) represents the temperatures in a rectangular
plate (Fig. 86) with insulated faces, if the edges are at temperature zero and the initial
temperature is
U(x,y,O) = f(x)g(y).
U-o
(a,b)
U-o U(x,y,O)- fTx)g(y) U-o
Fig.86 o U-o x.
11. Use the product of solutions (Prob. 10) to obtain the following formula for the
temperaturesin an infiniteprism with a squarecrosssection,if the initial temperature
is A and the surfacetemperatureis zero, if the unit of length is the side of the square,
and k = 1:
V(x,y,t) = A U(x,t)U(y,t),
where U is the temperature function found in Probo 9 of Seco81.
12. With the aid of Probo 10, derive the formula
4 x 00 sin(2n - I)lty
U(x,y,t) = -7t erf ( 2y r.t.) n=L 1 2n - l exp[ - (2n - l )2lt2t]
for the temperatures in the semi-infiniteslab x ;;;; O,O;;;; Y ;;;;1 with its boundary at
temperature zero, if U(x,y,O)= 1 (Fig. 87) and k = 1.
252 SECo85] OPERATIONALMATHEMATICS
YI U-O
U-O
"'''1
O
U(x,y,Oj
U-O
-1
\ " Fig.87
13. Generalize the method of Probo 10 to the case of three dimensions, and give an
illustration of its use in finding temperatures in a cube.
14. Use the generaJized convolution property found in Probo 20, Seco27, to make a
formal derivation ofthe following more general Duhamel formula. Let A(U) denote the
differential form (1), Seco85, P the point (x,y,z), n the distance normal to the boundary
surface S or S', and let a(P), b(P),H, F, and G represent prescribed functions. The tem-
perature function U(P,t) in the region R is to satisfy the conditions
U, = A(U) + H(P,t), U(P,O)= G(P) (P in R);
dU
(P on S or S').
a(P)U+ b(P)dn = F(P,t)
When the functions H(P,t) and F(P,t) are replaced by H(P,t') and F(P,t') respectively,
where t' is any fixed vaJue of t, V(P,t,t') denotes the solution of the problem. Write the
problem in the transform v(P,s,t')and transform its members with respect to t', using the
same parameter s, to obtain a problem in the iterated transform D(P,s,s)and thus derive
this generaJization of formula (10):
a f'
U(P,t) = ai Jo V(P, t - T, T) dT.
Note that if a(P) = F(P,t) = Owhen P is on S', then the surface S' is insulated. Thus the
generalized formula could be applied to Probo 14, Seco81.
t
A
8
Problems in Mechanical Vibrations
This chapter contains further illustrations of the uses of those properties

ofthe Laplace transformation that involve complex variables. The problems
taken as illustrations deal with vibrations and resonance in continuous
mechanical systems-systems in which the mass and elastic characteristics
are distributed over the system. Consequently these problems are boundary
value problems in partial differential equations, of the type treated in
Chap. 4. It is the intention here to present fairly simple physical problems
in their mathematical form, although the mathematical problems may have
more important physical interpretations. Some electrical analogs of
mechanical vibrations, involving transmission lines, are included among the
exercises.
86 A BAR WITH A CONSTANT FORCE ON ONE END

In Seco 44 we derived a formula for the longitudinal displacements in an
elastic bar in the form of a prism, when one end of the bar is fixed and a
253
Fg.88
constant force Fo per unit area acts parallel to the bar on the other end
(Fig. 88). Let all parts of the bar be initially at rest and unstrained. The
displacements Y(x,t) then satisfy the conditions in the boundary value
problem
Y,,(x,t) = a2 Yxx(x,t) (O< x < c, t > O),
(1) Y(x,O) = Y,(x,O) = O,
Y(O,t)= O, Ey"(c,t)= Fo,

where a2 = E/p, E is Young's modulus of elasticity, and p is the mass
per unit volume of the material.
Let us obtain another formula for Y(x,t) here. From problem (1)we
found this transform of Y(x,t):
aFo sinh (sx/a)
(2) y (x,s ) = - ,
E S2cosh (sc/a)
an analytic function of s everywhere except at the origin and at the zeros
:tsn of cosh (sc/a), where
(2n - l)na.
(3) (n = 1,2,...).
Sn= 2c 1
When s ::/=O and s ::/=:tsn, y is continuous in x and s, real-valued when s
is real, and y(O,s)= O.
Now the quotient of hyperbolic functions in formula (2),
Q(x,s) = sinh (sx/a)

cosh (sc/a)
fails to enhance the order property of y over a right or left half plane of the
variable s. For if s = A + ip., then (Sec. 55)
(4) IQ( )12 = sinh2 (Ax/a) + sin2 (p.x/a)

x,s sinh2 (Ac/a) + COS2(p.c/a)
sinh2 (Ax/a) 1
~ sinh2 (Ac/a) + sinh2 (J..c/a) (J..::/=O).
Thus Q is bounded over either half plane 1..1.1;;;y if Y > O, since IQI2~
1 + 1/sinh2 (ye/a). But Q does not always vanish there as Isl-+ OCJbecause
PROBLEMS IN MECHANICAL VIBRATIONS [SECo 86 255
IQI2~ sinh2(Ax/a)/cosh2 (AC/a), and this quotient remains fixed when A is

constant and Jl --+ :t oo.
Therefore y(x,s) is (9(1/s2) over half planes A ~ Y and A ~ -y, when
y > O, uniformly with respect to x, and L -l{y} is a continuous function
Y(x,t) with transform y(x,s). AIso, Y(x,O)= Y(O,t) = O. But the order
properties of the derivatives Yx and Yxxare not adequate for the representa-
tion of derivatives of Y by corresponding inversion integrals.
We can see from Eq. (4) that Q is bounded on horizontal lines
Jl = :t mea/cbetween the singular points :t Snsince
IQI2 ::; cosh2(Ax/a)::; 1

- cosh2 (Ac/a) -
for all points s on those lines. Thus y is (9(1/s2)on those lines, and L-l{y}
is represented by the infinite series of residues of esty(x,s) (Theorem 10,
Seco73).
We can write
a function that is analytic at the point s = O. In viewof Eq. (2) then the
origin is a simple pole of (f'y(x,s) when x =1= O, and the residue there is
(aFo/E)(x/a), or Fox/E.
The residue of y itself at the simple pole Sncan be written
(5) -Foa2 sinh (snx/a) =-4Foc (-lt . (2n- l)nx

sIn
Ecs; sinh (snc/a) n2E (2n - 1)2 2c'
It is real-valued. Thus according to formula (13), Seco72, in which (}l = O

and a = O,the sum of the residues of (fty(x,s) at the pair of poles :tsn, or
Snand sn, is
(6)
- 8Foc (-l)n . (2n - l)nx (2n - l)nat
Pn + Pn = n2E (2n - 1)2 SIn 2c COS 2c .
Finally, then, the inversion integral L-1{y(X,S)}is a function Y(x,t)

represented by this infinite series:
(7) Y - Fo 8c ~ (_1)n . (2n - l)nx (2n - 1)1tat

- E [ x + n2 n~l (2n - 1)2SIn 2c COS 2c ].
E;very term in that convergent series is periodic in t with period
4c fE
(8) To = -a = 4cyE:
so the function Y itself has that periodicity: Y(x, t + To) = Y(x,t). Note,
however, that the series is not twice differentiablewith respect to x or t.

Neither the series nor the inversion integral representation can be used
directly to verify that our function Y satisfies the wave equation. We shall
sum the series (7)in order to represent Y in a form that enables us to complete
the verification of our solution of problem (1).
87 ANOTHER FORM OF THE SOLUTION

In establishing the inversion integral representation of our function Y, we
noted that Y(x,O) = O. Since the function is also represented by series (7),
Seco86, it follows that
(1) x = -- 8e ~
i.J ( - 1)n
sm (2n -
o l)nx
(O ~ x ~ e).
n2 n=1 (2n - 1)2 2e
Each term of series (1) is an odd function of x, as is x itself, so the series
represents the function x over the interval - e ~ x ~ eo When x is replaced
by x + 2e, each term merely changes sign so the sum of the series is an
antiperiodic function for all x, with period 2e; consequendy it is periodic
with period 4e. Thus for all real x the series represents the triangular-wave
function H shown in Fig. 89, described by the two conditions
(2) H(x) = x (-e ~ x ~ e), H(x + 2e) = -H(x) ( - co < x < co)o
'The function H also has these properties for all x:
(3) H(x) = -H( -x) = H(x + 4e) = H(2e - x).
Now the sum of the series in Eq. (1) can be written
(4) 8e
2i.J
~ (-I)n-1
2sm
. o (2n - l)nx
= H(x ) (- co < x < co).
.n n=1(2n - 1) 2e
The series is, in fact, the Fourier series representing the odd periodic function
H for all Xo
-
We write m = (2n l)nf(2e) and note that
2 sin mx cos mat = sinm(x + at) + sinm(x - at).
H[x)
Fig.89
Then our series representation (7), Seco86, of Y(x,t) can be written

F. 4c a) (- 1)n-l
y= ; {x - 71:2
n~1 (2n - 1)2[sin m(x + at) + sin m(x - at)]}
and, in view of formula (4), Y has this representation:
(5) Y(x,t) = Fo[x

E - !H(x + at) - !H(x - at)].
We can use formula (5) to complete the verification of the solution of
our boundary value problem (1), Seco86. We first note that H(x) has deriv-
atives everywhere except at the points x = :!:(2n - l)c (n = 1,2,.. .).
Since any twice-differentiablefunction of x + at or x - at satisfies the
linear homogeneous-wave equation Y;,= a2y"x,and x itself satisfiesit, our
function (5) satisfies that equation except when x and t are such that
x :!: at = :!:(2n - 1)c, for which values the partial derivatives of Y do not
exist.
To see that the condition Ey"(c,t) = Fo is satisfied by our function (5),
we note that
(6) EYx(c,t) = Fo[1 - fH'(c + at) - !H'(c - at)].
But since H(x) = H(2c - x), then H'(x) = -H'(2c - x). Therefore
H'(e + at) = - H'(2e - e - at) = - H'(e - at) and, exceptfor the values of
t such that e :!: at = :!:(2n - l)e, Eq. (6) reduces to EYx(e,t) = Fo.
The condition Y;(x,O)= O is easily verified from formula (5). This
completes the verificationof our solution.
Formula (5)is well adapted to graphical descriptions of the variation
of y with either x or t. The graph of H(x + at) for a fixed value of t, for
example, is obtained by translating the graph of H(x) to the left through a
distance ato
The displacement at the end x = e can be written
F.
(7)
Y(e,t) = ; [e - !H(at + e) + fH(at - e)]
Fo
= -[e
E + H(at - e)]
since H(at + e) = - H(at + e - 2e). Y(e,t) is shown in Fig. 45.
The reader can show that the force per unit area on the fixedsupport,
Ey"(O,t),assumes the values 2Foand zero periodically.
Incidentally,we have established this transformation here:
-1 2a sinh (sx/a)
(8) L = 2x - H(x + at) - H(x - at).
{ s2 cosh (se/ a) }
88 RESONANCEL'N THE BAR WITH A FIXED END
Let the end x = O of the bar again be fixed while a simple periodic force
F(t) = A sin rot (ro > O)
per unit area acts lengthwise at the end x = e. Ifthe bar is initially unstrained
and at rest, only the end conditionat x = e in problem (1), Seco86, need be
changed to read
(1) EYx(e,t) = A sin rot.
The transform of the displacement now becomes

Aaro sinh (sx/a)
(2) Y(x,s) =
7>, ~
2
. n'
2
S cosh (se/a)
.
Here y is an analytic function of s except at the origin s = O, a
removable singular point, and the points s = :t iro and s = :t Snwhere
(2n - I)na.
s = I (n = 1,2,...).
n 2e
If iro is not equal to any one of the numbers Sn' that is, if
ro#--
(n - !)na
(n = 1,2,...),
e
the singular points :t iro and :t Snare all simple poles unlessthe value of x
is such that some are removable, in which case they serve as simple poles
with residue zero.
r:
According to formula (13), Seco72, the component of Y corresponding
to the pair of poles :t iro of y is a-term of the type b cos(rot + O); here the
real numbers b and Odepend on x. Similar components correspond to a
pair of simple poles :t Sn' Thus Yis represented formally by a series of
this type:
00
(3) Y(x,t) = bo(x)cos[rot + Oo(x)] + ,,=1

L bn(x)cos[ro"t + On(x)],
where ro" = (n - !)na/e. We shall find the b's and O's in Seco89 and establish
formula (3) as a solution of our boundary value problem in Y. The series
shows that each section of the bar moves as a superposition of two periodic
motions, one with frequency ro and the other with frequency rol = na/(2e),
or period TI = 4e/a.
But if the frequency ro of the external force coincides with one of the
frequencies ro", say ro = ro, for some positive integer r, then the function
se
q(s) = (S2 + ro,2)cosh-
a
PROBLEMS IN MECHANICAL VIBRATIONS [SEC.89 259
in the denominator of expression (2) for y(x,s) is such that q(iw,) = q'(icD,)= O
while q"(iw,) =FO. Consequently the points s = :t iw, are poles'of y f the
second order, and corresponding to that pair of poles, according to formula
(16), Seco72, Y(x,t) contains an unstable component of the type
(4) tb,(x)cos[w,t + I/I,(x)], where b,.(x) =Fo.
The remaining component of Y is periodic with frequency Wl>the common
frequency of all its periodic terms.
That unstable oscillation of sections of the bar is called resonanee.
The periodic external force is in resonance with the bar when its frequency
w coincides with any one of the resonance frequencies
(5) w, = (r - e!)1la (r = 1,2,...).
The frequenciesw, depend on the physical properties of the bar and the
mannerin whichthe bar is supported. If the end x = O of the bar is free
or elastically supported, for instance, we find that the set of resonance
frequenciesis differentfrom the set (5).
To produce resonance, the external force need not be restricted to the
simple form A sinw,t. For any prescribed force F(t) at the end x = e, the
transform of Y(x,t) is
a sinh (sx/a)
(6)
y(x,s) = "Ef(s) s cosh (se/a)'
Consequently if F(t) contains a term A, sin w,t or B, cos w,t, where w, is

a number of the set (5), then y(x,s) will contain a term with the product
(S2 + w/) cosh (se/a) in the denominator, and Y(x,t) will have an unstable
component of type (4). In fact, if F is any periodic function with a frequency
w" then resonance will occur, as can be seen from the form in Seco 23 of
the transform of a periodic function.
89 VERIFICATIONOF SOLUTIONS
When the formal solution of a boundary value problem can be written in
terms of a finite number of simple functions, that form is usually best for
verifying the solution, as we illustrated in Seco87. But to establish a solution
represented only by an infinite series or by an improper integral, the
procedure based on properties of its transform, used in Chap. 7, may be
useful. Let us illustrate a combination of the two methods for the problem
in the preceding section.
The following transform of displacements Y(x,t) in a bar with its end
x = Ofixed and with pressure A sin wt applied at the end x = e was found
formally:
B sinh (sx/a)
(1)
y(X,S)= S2+ ro2s cosh (se/a)
The quotient of hyperbolic functions here was shown (Sec. 86) to be bounded
uniformly in x over half planes Re s ~ ')1and Re s ;;;- ')1 when ')1 > O;
cosh (sx/a)/cosh (se/a) is also bounded there. Hence y is l!J(S-3)and Yx is
l!J(s- 2) over those half planes, uniformly in x, and L- 1 {y} along a line
Re s = ')1represents a continuous function Y(x,t) such that Y(x,O) = Y(O,t) = O;
also Y has continuous derivatives of first order,
Y,(x,t)= L- 1{sy(x,s)}, y':(x,t)= L- 1 {yx(x,s)},
such that Y,(x,O)= Oand Ey':(e,t) = A sin rot. Hence Y satisfies all boundary
conditions in the problem, but the order of y is not adequate to show directly
that Y/t = a2yxx.
To prove that Y does satisfy the wave equation, we write
1 1 ro2
S2 + ro2 = S2 - S2(S2 + ro2)'
(2) u(x s) = -
B sinh (sx/a)
v(x s)
- Bro2sinh (sx/a)
= S3(S2 + ro2) cosh (se/a).
, S S2 cosh (se/a)' ,
Then y(x,s) = u(x,s) + v(x,s)
where v is l!J(s-S)and Vxxis (9(S-3). It follows readily that L-l{V} satisfies
the wave equation.
The function L- 1{u} can be written as a linear combination of
known solutions of the wave equation since (Sec. 87)
2au(x,s)= Bs-1 L{2x - H(x + at) - H(x - at)}
where H is the triangle-wavefunction shown in Fig. 89. Thus L-l{U} is
the function
B '
U(x,t) = 2a fo [2x - H(x + a-r)- H(x - a-r)]d-r
x+al
= _B 2xt - -l
2a [ af
H@ d~ .
x-al ]
In terms of the periodic function
(-<X) < r < (0)

(3) G(r) = I: H(~)d~
we can write U in the form
B
(4)
U(x,t) = 2a2[2axt - G(x + at) + G(x - at)],
fram which we see that V" = a2V"",oTherefore our function

-1 -1 B sinh (sx/a)
(5)
Y(X,t) = L {y(X,S)} = L { s(s2 + ro2) cos h (se/)a }
satisfies the wave equation and all the boundary conditionso
We shall use Theorem 10, Seco73, to prave that the series of residues
of e"y(x,s) converges to the inversion integral for each frequency ro. Then
we shall have a series representation of our solution (5).
Qur function y is analytic in s except for the poles
(6) :tiro and wherem =- (n - !)n

(n = 1,2,.. ')0
n e
As noted in Seco86, the quotient sinh (sx/a)/cosh (se/a) is bounded on the
horizontal lines 1ms = :t ann/eo Thus y(x,s) is (!)(s- 3) on all such lines for
which ann/e > ro. Since y has that order over the left half plane Re s < -y,
the infinite series of residues of esty(x,s) converges to L-l{y} when t ~ O
and O ~ x ~ eo
Let us write that series explicitly when ro is not one of the resonance
frequencies amn. Then the poles of y are all simple, and formula (13),
Seco72, gives the sum of the residues of esty at each of its pairs of conjugate
poles. The residue of y itself at the poles s = iro and s = iamn are easily
found to be
B sin (rox/a) B( -I)n sin mnX
(7) -1
o
an-I
d o
2ro2 cos (roe/a) emn(amn2 - ro2)'

respectively. The sum of the residues of e"y at the pair of poles :t iro is
therefore
B sin (rox/a) n
cos rot - - .
ro2cos (roe/a) ( 2) '
at the pair :t iamn the sum is
2B( - 1)" sin mnx
cos amnt - _
n
o
2 2
emnamn
( - ro ) ( 2)
Thus we have this representation of our solution (5):
(8) Y(x,t) = Bsin (rox/a) sin rot

ro2 cos (roe/a)
2n-1
( mn=~n ) o
When ro coincides with one of the resonance frequencies am" then y

has a conjugate pair of poles of order two, and the series for Y contains a
resonance term of type (4), Seco88.
PROBLEMS
1. The end x = O of an elastic bar is free while a constant longitudinal force Fo per .
unit area acts longitudinally at the end x = c (Fig. 48). The bar is initially at rest and
unstrained. One formula for the longitudinal displacements Y(x,t) in the bar was
found in Probo 8, Seco44. Derive this altemate formula:
Y = -Fo 2
x + 3a t - c -
22 2 12c2
r '-- ---y-cos-cos-
~ (_I)n mrx mrat
.
6Ec[3 n n= n c c ]
2. Reducethe solutionof Probo1 to the form
F.
Y(x,t) = 12~c[6x2+ 6a2t2- 2C2 + Q(x + at) + Q(x - at)]
where Q is this periodic function:
(-c<x<c), Q(x + 2c) = Q(x) (-00 < x < 00).
Use that form to verify the solution of the boundary value problem.
3. An unstrained elastic bar or heavy coiled spring is clamped along its length c
so as to prevent longitudinal displacements, then hung from its end x = O (Fig. 90).
At the instant t = O,the clamp is releasedand the bar or coil vibrateslongitudinally
because of its own weigh. Thus
Y,,(x,t) = a2Yxx(x,t)+ g (O< x < c, t > O),
where g is the acceleration of gravity. Complete the boundary value problem, write
m = n - 1. and derive the formula
g 4C2 <X>1 . mnx mnat
Y(x,t) = -
2a2 [
x(2c - x) - - "-sm-cos-
n3 nf'm3 c c ]
.
Fig.90
4. Reducethe solution of Probo3 to the form
Y(x,t) = 4:2 [2P(x)- P(x + at) - P(x - at)],

where the function P is described by the conditions
P(x) = x(2c - x) (O~ x ~ 2c), P(x + 2c) = -P(x) (-00 < x < 00).
Use that form to verify the solution of the problem.
5. The ends x = :t e of an unstrained shaft spinning about its axis with angular
velocity w are brought to rest at the instant t = O (Fig. 52, Probo 13, Seco44). Derive
this formula for the angular displacements of its cross sections:
O( ) -- 8cw ~ (_1)n-1 (2n - 1)1tx.
SIn (2n - 1)1tat
.
'--
1t a "=1 (2n - l )
O' x,t 2 2 cos
2e 2e
6. Reducetheformulafore in Probo5 to the form
w
e (x,t) = 2a [H(x + at) - H(x - at)],
where H is the triangular-wave function shown in Fig. 89 and represented by series (4),
Seco87. Use this simple representation of e to verify the solution of the boundary
value problem in Probo 5.
7. The end x = O of an elastic bar is fixed while a force proportional to t2 acts
10ngitudinalIy at the end x = l so that Yx(l,t) = At2, where Y(x,t) denotes longitudinal
=
displacements. The unit of time is such that a l. If the bar is initially at rest and
unstrained, derive the formula
Y(x,t) =A [ X(t2 - 1)+ _

3
x3
-"4
4
~
1t n=1 m
00 (-1)"
sin m1txcos mnt
]
where m = n-t. Verifythe solutionfullywith the aid of the inversionintegral
representation of Y. (Also, seeProbs.8 and 9, below.)
8. Reduce the solution of Probo 7 to the form
X3 l l
where
Y(x,t) = A [ X(t2 - 1) + 3" + '2R(x + t) + '2R(x- ]
t)
X3
R(x) = x - 3"
(-1 ~ x ~ 1), R(x + 2) = -R(x) (-00 < x < (0).
Show R graphically. Use this simple representation of Y to verify the solution of

Probo7.
9. Showthat the function Y found in Probo 7 and its derivatives of first and second
order are everywhere continuous in x and t and (9(e'")uniformly with respect to
x (O~ x ~ 1). Prove that no other function of that type can satisfy alI conditions in
the boundary value problem. (Cf. Seco79; but here the proof of uniqueness is simpler
becauseof the favorablecontinuityconditions.)
10. The end x = O of an elastic bar is fixed while the end x =e is displaced longi-
tudinalIy in the manner Y(c,t) = A sin wt. Show that the resonance frequencies are
w = w" where
w,--- r1ta (r = 1, 2, . .. ).
e
Note that the lowest frequency W1 is twice as great as the lowest in the set (5), Seco88.
11 .
If the end x = O of the bar in Probo 10 is free, rather than fixed, show that the
resonance frequencies are the same as the set (5), Seco88.
12. A simple periodic longitudinal force acts uniformly on all elements of an elastic
bar with its end x = O fixed and its end x = e free, so that the displacements Y(x,t)
satisfy the conditions
1';,(x,t) = a2Y"",(x,t) + B sin (Ot,
Y(O,t) = Y,,(c,t) = Y(x,O) = 1';(x,O)= O.
Show that resonance occurs when (O= (o" where
(0= (r - t)na
, e (r = 1,2,...).
13. A simple periodic transverse force acts uniformly on all elements of a stretched
string of length e with fixed ends, so that the transverse displacements Y(x,t) satisfy
the equation 1';, = a2y= + B sin (Ot. Let the string be initially at rest in its position
of equilibrium. Show that resonance occurs only when (O= (o,where
(O, = (2r -e 1)1ta (r= 1,2,...).
14. The supports of the ends of a stretched string vibrate in a transverse direction
so that
Y(O,t)= A sin at, Y(c,t) = B sin pt.
The string is initially at rest on the x axis. In general, show that resonance takes place
if either a or p has one of the values n1ta/c. But if p = IX,show that the resonance
frequencies are 2n1ta/c when B = -A, and (2n - l)na/c when B = A, where
n = 1,2, Hint: When p = a and B = -A, showthat
y( ) -
x,s -
~
S2 + a2
sinh [s(c - 2x)/(2a)]
sinh [sc/(2a)] .
90 FREE VIBRATIONS OF A STRING
A string stretched between the origin and a point (e,O)is given a prescribed
initial displacement Y(x,O) = g(x) and released from rest in that position.
To find its transverse displacements Y(x,t), we solve the problem
Y,,(x,t) = a2Yx.x<x,t) (O < x < e, t > O; a2 = ~),

Y(x,O) = g(x), Y,(x,O) = Y(O,t) = Y(e,t) = O.
The function g must naturally be continuous (O~ x ~ e), and vanish

when x = Oand when x = e. Let g' be sectionallycontinuous.
The solution of the transformed problem in y(x,s),
a2y" - s2y = -sg(x), y(O,s)= y(e,s)= O,
can be written (cf. Seco82)
p(x,S)
y(x,s) = a sinh (sela
when s ,,:0, where

. (e - x)s
x
" . h -~sa.""
p (x,s ) = SInh
a f.o g."( ) SIn a
. xs c
" . (e - ~)sa"
+ SInh -
a f
x
( ) SInh
g."
a
.".
When s = 0, the solution is y(x,O)= o.

Now p is an entire function of s, and the singular points of y are the
simple poles s = :t Sn' where
iml:a
s =- (n = 1,2,...).
n e
Ir we write
2 c . ml:~
(1) bn = - f. g(~)SIn-a~,
e o e
then p(x,sn)=tbne cos mI:sin (ml:xfe),and the residue ofy at Snis tbn sin (ml:xfe).
Thus the sum ofthe residues of esty(x,s) at the pair of poles :tsn is (Sec. 72)
.
bnSIn-COS-.
nnx nnat
e e
The formal solution of the boundary value problem is then
(2) "
en
Y(x ,t) -- f..J bn sln -cos-
. nnx nnat
n= 1 e e '
where the coefficients bn are given by formula (1).
Since Y(x,O)= g(x), formula (2)implies that
en nnx
L bn sin- e = g(x)
n= 1
(O ~ x ~ e).
Each term of this series is an odd periodic function of x with period 2e.
Consequently, for all x the series should represent the odd periodic extension
G of g defined by the conditions
G(x) = g(x) when ~ x ~ e,
(3)
G(-x) = -G(x), G(x + 2e) = G(x) for all x.
~ ,.
Then G is everywhere continuous, and

CO mtx
(4) L bnsin-
n= 1 e
= G(X) ( - 00 < x < (0).
Since the terms in series (2) can be written as

1 . n7t(x + at) b . n7t(x- at)
2[b n Sin e + n Sin 2 ]'
our formal solution has the simple representation
(5) Y(x,t) = t[G(x + at) + G(x - at)].
The function (5) is easily verified as a solution of our boundary value
problem. 118form is a convenient one to use in studying the motion of the
string.
91 RESONANCE IN A BAR WITH A MASS ATTACHED

The end x = of an elastic bar is fixed. To the end x = e a rigid mass M
is attached, and a periodic 10ngitudinal force B sin rot ac18 on that mass
(Fig. 91). The elastic displacements within the mass itself are assumed
negligible, and the bar is assumed to be too heavy to be considered as a
coiled spring with negligible mass. We shall find the frequencies ro for
which resonance takes place.
Let A denote the area of the cross section of the bar, m the mass of
the bar, and Y(x,t) the longitudinal displacements in the bar. The force
exerted by the bar on the attached mass Mis then -AEYic,t), and so the
end conditions are
(1) Y(O,t)= 0, MY,t(c,t) = -AEy"(c,t) + B sin wt.
Ir the bar is initially at rest and unstrained, the remaining conditions in
the boundary value problem are
(2) Y,t(x,t) = a2 y"ix,t)

(O < x < e, t > 0, a2 = ~) ,
(3) Y(x,O) = Y,(x,O) = O.
0-
L!BSintl1tFig.91
PROBlEMS IN MECHANICAl VIBRATIONS [SECo 91 267
This is also the problem of the torsional vibrations in a shaft with

one end fixed and with a flywheel, on which a periodic torque acts, attached
to the other end. Note that one end condition here involves a derivative
of the second order. An interpretation of the problem in terms of torsional
vibrations in a propeller shaft is given in the problems at the end of the
chapter.
The problem in the transform y is
a2y"(x,s) - S2y(X,s)= O,
(4)
y(O,s) = O,
+ w..
AEy'(c,s) + Ms2y(c,s) = s2Bw
Since m = Acp = AEc/a2, we can write the solution of problem (4) in

terms of a function q, where
(5) q(z) = Mz sinh z + m cosh z,
in this form:
(6) y( )- Bcw sinh (sx/a)
x,s - as(S2 + w2)q(sc/a)'
Thus y is analytic in s except at the points s = :1::iw and at the zeros
of q(sc/a). But if we write z = ..1. + iJ1.,we find from Eq. (5) that
2Iq(z)1 = I(Mz + m)eZ - (Mz - m)e-ZI
~ IIMz + mieA - IMz - mle-AI

= 1[(M..1. + m)2 + M2J1.2]t - [(M..1.- m)2 + M2J1.2]te-AI,
from which we can see that Iq(z)1 > O if ..1.> O or if ..1.< O. Therefore q(z)
can vanish only if ..1.= O,and thus z = iJ1.,where J1.is a root of the equation
- M J1.sin J1. + m cos J1. = O; that is,
m 1
(7) tanJ1. = --o
MJ1.
The graphs of the functions tan J1.and m/(MJ1.)show that Eq. (7) has an
infinite sequence of roots :l::J1.n
where O< J1.1< n/2 and J1.n
- (n - l)n ~ O
as n ~ ex)(Fig.92). Ir w = aJ1.n/c,
where n is some positive integer,then the
Fig.92
two points s = :t. iafJ.n/care poles of y of the second order and Y contains
a term of resonance type bn(x)t cos [wnt + 8n(x)] with the resonance frequency
(8) Wn = a:n (tan fJ.n= ::fJ.n' n = 1,2,...) .

The representation of Y(x,t) as a series of residues can be written in
the usual way. Except for a term of resonance type in case W = Wn, each
term of the series is periodic in t, but the terms have no common periodo
92 TRANSVERSE VIBRATIONS OF BEAMS
Let Y(x,t) denote the transverse displacement of a point at distance x from

one end of a bar or beam, at time t. As in the case of static displacements
discussed in Seco36, the instantaneous bending moment transmitted through
a cross section is EIYxx(x,t), where 1 is the moment of inertia of the cross
section with respect to its neutral axis. The shearing force at a cross section
is Ela3Y/ax3. When the cross section is uniform and no external force
acts along the beam, the displacement function Y(x,t) satisfies the equation
a2y a4y
(1)
at2 + a2 ax4 = O
under certain idealizing assumptions, where Ap is the mass of the beam

per unit length.
Let the end x = Obe hinged, so that no bending moment is transmitted
across the section at x = O and the displacement is zero there. Let the end
x = e be hinged on a support which moves parallel to the Yaxis in a simple
harmonic manner (Fig. 93). Ir the beam is initially at rest along the x axis,
the boundary conditions that accompany Eq. (1) are then
Y(x,O) = Y,(x,O)= O,
Y(O,t) = Y",X<O,t)
= O,
Y(c,t) = B sin wt, Yxx(c,t) = O.
Let us find the frequencies w at which resonance will occur.
11
lB sin c"t
I
1-
(e,O)
Fig.93
PROBLEMS IN MECHANICAL VIBRATlONS [SECo 92 269
The problem in the transform y(x,s) is

d4y
(2) a2 - + s2 y = O
dX4 '
Bw
y(O,s) = YxiO,s) = Yxie,s) = O,
y(e,s) = S2 + w"
It wilI be convenient to write
. 2
(3) S = aq .
Then S2 = - a2q4, and the general solution of Eq. (2) can be written
(4) y(x,s) = CI sin qx + C2cosqx + C3sinh qx + C4coshqx,

where the C's can be functions of the parameter s. When those constants
are determined so that the boundary conditions on y(x,s) are satisfied, the
solution (4) becomes
(5) y (x,s )
= 2
Bw 2 sin qx sinh..qe + sinh qx sin qe .
s +w 2 sm qe smh qe
The Maclaurin series representations of the sine and hyperbolic sine
functions show that the final fraction in Eq. (5) is an analytic function of q2,
and therefore of s, except at points where the denominator vanishes. The
point s = Ois a removable singular point. The denominator vanishes when
qe = :i:n1t and qe = :i:in1t; that is, whenever q2e2 = :i:n21t2. In view of
Eqs. (3)and (5)then, the singular points of y(x,s) are
in21t2a
(6) s = :i:iw, s = :i:---y-
e
(n = 1,2,...).
Ir the value of w is distinct from all the numbers
(7) (n = 1,2,...),
the derivative of the entire denominator in the expression (5) for y(x,s)
does not vanish at any of the points (6),nor does the numerator exceptfor
particular values of x. Those points are therefore simple poles of y(x,s).
In this case the formula for the displacements has the form
co
Y(x,t) = bo(x) cos [wt + Oo(x)]+ L bn(x) cos [wnt +

n=1
On(x)].
Ir the valueof w coincides with one of the numbers Wngiven by formula

(7), then y has poles ofthe second order at the points s = :i: iwn, and a term
of resonance type appears in Y(x,t). Hence the resonance frequencies are
the frequencies (7).
In case the hinge at x = e is kept fixed and a simple harmonic bending

moment acts on that end of the beam, the conditions at x = e have the form
Y(e,t) = O, Yxx(e,t)= B sin rot.
The reader can show that the expression for y(x,s) then has the same
denominator as it does in Eq. (5). The resonance frequencies are again those
given by formula (7). Other cases are included in the problems at the end
of the chapter.
93 DUHAMEL'S FORMULA FOR VIBRATION PROBLEMS
As in the case of problems in heat conduction (Sec. 85), the convolution

property of the transform displays a relation between the solutions of
problems in vibrations with variable boundary conditions and corre-
sponding problems with fixed conditions. Consider for example the trans-
verse displacements Y(x,t) in a string. If both damping and an elastic support
are present (Fig. 94), the equation of motion has the form
(1) Y;lx,t) = a2 ~ix,t) - b Y;(x,t) - h Y(x,t).
To permit the end x = O to be elasticallysupported or kept fixed or
to slide freely along the Yaxis, we can write the condition
(2) A,l(Y) =O at x = O,
where A,l(Y) = h1Y - k1~. If a prescribed force F(t) acts on the end x = e,
a fairly general boundary condition is
(3) A,2(Y) = F(t) at x = e,
(4) Y(x,O) = :Y;(x,O)= O,

the transform y(x,s) satisfies the conditions
(5) (S2 + bs + h)y(x,s) = a2Yxix,s),
(6) A,[y(O,s)]= O, A,2[y(e,s)]= f(s).
y .F(t
o (e,O) x
Fig.94
Let Z(X,t) represent the displacement Y(x,t) in the special case m

which F(t) = l.
Then z(x,s) satisfies Eq. (5) and the boundary conditions
1
A.[z(O,s)] = 0, A.2[Z(C,S)] = -.
s
It follows that the product sf(s)z(x,s) satisfies the conditions (5) and (6),
and therefore
(7) y(x,s) = sf(s)z(x,s).
In view of the convolution property then,

o t
(8)
I
Y(x,t) = ot o F(t - t)Z(x;r)dt,
or
(9)
Y(x,t)= E F(t - t)Zt(X,t) dt.
These are two forms ofDuhame1's formula for the resolution ofthe problem
in Y(x,t) with a variable end condition into one with a fixed end condition.
The derivation of these formulas can be extended easily to other
problems. In the case of transverse vibrations of bars, for instance, the
derivative of the fourth order with respect to x replaces Yxx(x,t),and addi-
tional boundary conditions are involved. In the case of transverse displace-
ments in a membrane, the Laplacian of Y replaces Yxx in our problem.
In these cases, the steps in the derivation of Duhamel's formula are the
same as in the case treated above.
PROBLEMS
1. An elastic bar with its end x = O kept fixed is initially stretched so that its longi-
tudinal displacements are Y(x,O) = Ax (O~ x ~ e), then released from rest in that
position at the instant t = Owith its end x = e kept free. Derive these two formulas
for its displacements:
Y(x,t) 8Ac ~ (-Ir

-- -=r 1 . (2n - I)nx COS (2n - I)nat
'-
n n= 1 (2n - 1)2 Sin 2e 2e
= tA[H(x + at) + H(x - at),

where H is the triangular-wave function (2), Seco87 (Fig. 89). Verify the solution.
2. The end x = O of an elastic bar is elastically supported (Fig. 95) S9 that the
longitudinal force exerted on that end is proportional to the longitudinal displacement:
y"(O,t) = h Y(O,t) (h > O).
Fig.95
When a longitudinal periodic force F(t) = A sin rot acts at the end x = e, show that
resonance takes place if ro has any one of the values aa../e (n = 1,2,...), where a.. is
a positive root of the equation tan a. = he/a.. Show those roots graphically (cf. Fig. 92).
3. A string stretched from the origin to a point on the x axis is initially straight with
a uniform velocity in the direction of the Y axis, as if a moving frame supporting the
end points is brought to rest at the instant t = O. There is a smalldampingforceper
unit length proportional to the velocity. Select units so that the transverse displace-
ments satisfy the conditions
Y,,(x,t)= Yxx(x,t)- 2bY,(x,t) (O< x < n, t > O,O< b < 1),
Y(O,t)= Y(n,t) = Y(x,O)= O, Y,(x,O)= Vo.
1[p is a branch of the function (S2 + 2bs)t, show that
L{Y} = y(x s) = Vocosh (np/2) - cosh (x - n/2)p

, p2 cosh (np/2)
and obtain formally this formula for displacements:
Y(x,t) = 4voe-h'
n .=1
fsin (2n - l)x sin mt
(2n-l)m
4. A string is stretched from the origin to the point (e,O)and given an initial veiocity
Y,(x,O)= g(x) but no initial displacement. Derive this formula for its displacements:
2
Y(x,t) = -
00 I . nnx . nnat
L -sm-sm-e
na.=ln e i
o
e . nn~
g(~)sm-d~.
e
5. In Probo 4 let G(x) be the odd periodic extension, with period 2e, of the velocity
function g(x) and show formaIly that
2 Y,(x,t) = G(x + at) + G(x - at)
x+ar
I
and Y(x,t) = -2
f
a x-at
G(~)d~.
Verify that integral form of Yas a solution of the problem.

6. Let I(x,t) denote the current and V(x,t) the voltage at distance x from one end of
a transmission line, at time t. FoIlowing the notation used in the telegraph equation (11),
Seco40, and indicated in Fig. 96, let R, L, S, and K represent resistance, inductance,
leakage conductance, and capacitance to ground, respectiveIy, per unit length of line.
Then the functions I(x,t) and V(x,t) satisfy the system of partial differential equations
(a) -Ji"=RI+LI,, -Ix=SV+KYr,
called the transmission Uneequations. By differentiating the members of equations (a)
and eliminating one of the functions and its derivatives, show that either 1 or V satisfies
" ---- x-e

Source R L load
1 Fig.96
1
the telegraph equation. Note that equations (a) are better adapted to initial and end
conditions that involve values of both I and V, however, than is the telegraph equation
(see Probs. 7 to 9).
7. Supposethat R = S = O in the transmission line shown in Fig. 96, and that the
current and voltage are zero initiaUy. If one end is kept grounded, V(O,t)= O, and
the current at the end x = e is proportional to t, find the transform v(x,s) of V(x,t)
using the transmission line equations (a), Probo 6. Thus show that V(x,t) is an eIectrical
analog of the displacements Y(x,t) in the bar in Seco86, where a2corresponds to (KL)- 1.
8. In the transmission line shown in Fig. 96, R = S = O and I(x,O)= V(x,O)= O.
If V(O,t)= A sin rot and the end x = e is grounded, find the values of ro for which
the voltage and current become unstable, using the transmission line equations (a),
Prob.6. Ans. ro = mrc-l(KL)-t (n = 1,2,...).
9. Solve Prob. 8 when the condition that the end x = e is grounded is replaced by the
condition that the circuit is open at that end.
Ans. ro = (n - t)1tc-l(KL)-t (n = 1,2,...).
10. One end of a beam is built into a rigid support (Fig. 97); thus if Y(x,t) denotes
transverse displacements,
Y(O,t)= YJO,t) = o.
y;- .
o x
Fig.97
The pin-support of the other end x = e vibrates, so that
Y(c,t) = B sin rot, Yxx(c,t)= O.

Initially the beam is at rest along the x axis. If a2 is the coefficient in Eq. (l), Seco92,
show that resonance occurs in the vibration of this beam if ro = aa.//C2,where ex.is
any positive root of the equation tan ex= tanh ex. Show how the roots ex. can be
approximated graphically.
11 .One end of a bar of length e is free. The other end is built into a support that
undergoes a transverse vibration Y = B sin rot (Fig. 98). The bar is initially at rest
with no displacements. Show that resonance takes place in the transverse vibrations
of the bar when ro = aa..2/C2,where ex.is any positive root of the equation cos ex=
-sechex and a2 is the coefficient in Eq. (1), Sec. 92. Also show how the roots ex.can
be approximated graphically.
y
YlO,tl-B sin (lit Fig.98
12. Both ends of a beam of length e are built into rigid supports. The beam is initially
at rest with no displacements. A simple periodic force per unit length acts along the
entire span, in a direction perpendicular to the beam, so that the transverse displace-
ments Y(x,t) satisfy the equation
Show that resonance occurs when w = 4aa.//C2,where ex.is any positive root of the
equation tan ex= - tanh ex.
13. A membrane, stretched across a fixed circular frame r = e, is initiallyat rest in

its position of equilibrium. Ir a simple periodic force per unit area in a direction
perpendicular to the membrane acts at all points, the transverse displacements Z(r,t)
satisfy an equation of the type
Ztt = b2( Z" + ~Zr) + A sinwt.

Show that the resonance frequencies are w = bcx./e,where ex.is any positive root of
the equation J o(ex)= O.
14. The end x = Oof the propeller shaft for a ship is connected rigidly to a massive
flywheel that turns with unifonn angular velocity w. The end x = e supports a propeller
whose moment of inertia with respect to the axis of the shaft is l. In addition to the
steady-state torque transmitted between shaft and propeller, the action in the water
induces a perturbation in the nature of a torque proportional to sin 4wt exerted on
the propeller. Let 0(x,t) denote the corresponding perturbation in the angular dis-
placements of the circular cross sections of the shaft. Thus 0 (x,t) = Owhen the shaft
operatesunder steady-stateconditionswithangularvelocityw. Ir 1p is the polar
moment of inertia of the cross section of the shaft, of radius ro (1p = tro 4), and if E.
is the modulus of elasticity in shear for the material in the shaft, show that
10tt(e,t) = - E.J p0 ",(e,t) + B sin 4wt,
and that the boundary value problem in o (x,t) is an analog of the problem in Seco91.
where a2 now represents EJ p and p is mass of the material per unit volume. Show that
torsional resonance takes place when w = aA..I(4e),where A..is any positiveroot of
the equation tan A. = k/A.and where k = epl p/l.
15. The propeller shaft in Probo 14 is 150 ft long with a diameter of 10in., and made
of steel. The moment of inertia 1 of the propeller equals that of a steel disk 4 in. thick
and 4 ft in diameter. The steel weighs 0.28 lb/in.3 (gp = 0.28),and Es = 12 X 106 lb/in. 2
Show that the lowest ftywheel speed at which torsional resonance takes place is approx-
imately 135 revolutions per minute.
9
Generalized Fourier Series
We shall use theory developed in Chap. 6, on the inversion integral for the
Laplace transforrnation, to assist us in establishing some basic theory of
eigenvalue problems in differential equations. The theory shows that the
solutions of certain types of homogeneous boundary value problems in
ordinary differential equations form orthogonal sets of functions, and that an
arbitrary function can be represented by a series of functions of such a set, a
generalized Fourier series.
A classical method of solving some types of linear boundary value
problems in partial differential equations begins with a separation ofvariables,
then requires a representation of a prescribed function by a generalized
Fourier series. The orthogonal functions in the series are the solutions of an
eigenvalue problem that arises from the separation of variables. The method
will be illustrated at the end of this chapter.
In the following chapter we shall see how such eigenvalue problems
are encountered in a process of designing an integral transformation to fit
276
...
GENERALIZED FOURIER SERIES [SECo 94 277
the requirements of a given boundary value problem. Then the inverse

transformation is represented by the corresponding generalized Fourier
series.
94 SELF-ADJOINT DIFFERENTIAL EQUATIONS
We begin with some concepts and an existence theorem from the theory
of linear ordinary differential equations.
The adjoint D!y of the linear differential form
(1)
of the second order is this related form :
(2) D!J = (IX2(X)y(X)]" - (IXl(X)y(X)]' + IXO(X)y(X).
It is useful in constructing forms that represent derivatives of other forms.

With the aid of a second function v(x), for instance, a form that is an exact
derivative can be written
(3)
vD2y - + IXIVY].
yD!v = d~(rJ.2VY'- (rJ.2V)'Y
The Lagrange identity (3) is easily verified.

We find that the form
(4) ~2Y = (r(x)y'(x)]' - Q(x)y(x)
is self-adjoint, ~!y = ~y. For it, Lagrange's identity becomes
(5)
V~2Y - y~2V = :x(Vy' - v'y)r].
When both members of the differential equation
(6) y"(x) + Pl(X)Y'(x) + Po(x)y(x) = f(x)
are multiplied by an integrating factor
(7)
r(x) = exp [f Pl(X) dx J.
the equation assumes its self-adjoint form
(8) (r(x)y'(x)]' - Q(x)y(x) = F(x),
where Q(x) = - r(x)Po(x) and F(x) = r(x)f(x).
Adjoints of differentialforms of order higher than 2 are defined in a
manner that corresponds to the definition (2) (Prob. 3, Seco97). But linear
differential equations of higher order do not always have self-adjoint forms.
..
An existence theerem for solutions ofthe homogeneous form ofEq. (6),

in which the coefficient Po(x) may involve a parameter A in a lin~ar manner,
will now be cited. The theorem can be proved by a method of successive
approximations.l
THEOREM
Let A, B, and e be continuous functions of x over some closed interval 1 and

let Xo be a fixed point on 1. Let A denote a complex-valued parameter and
Yo and Yl two prescribed constants independent of A. Then the initial-value
problem
y" + A(x)y' + [B(x) + AC(X)]Y= O,
(9)
y(xo,A.)= Yo, y'(xo,A.)= Yl
in the complex-valued function y(X,A) has one and only one solution y =
u(X,A.)such that u and u' are continuous functions of x and A.for all A and all x
on the interval 1 and, for each x on 1, u and u' are entire functions of the
complex parameter A.
In Seco96 we shall see that further restrictions are needed to ensure a
unique solution of a two-point boundary value problem in linear ordinary
differential equations. This fact is illustrated by the simple problem
(10) y" + AY = O, y(O,A) = O, y(l,A.) = O,
which clearly has a solution Y(X,A.)==O. But when the parameter A has any
one ofthe values n2n2 (n = 1,2,.. .),problem (10)has the additional solutions
(11) y = k sin nnx,
where the constant k is arbitrary. AIso, note that the problem
(12) y"(x) + n2y(x) = O, y(O)= O, y(l) = 1,
has no solution, since the function k sin nx that satisfies the first two condi-
tions fails to satisfy the condition y(1) = 1 for any value of the constant k.
95 GREEN'S FUNCTIONS
On an interval a ~ x ~ b let r, Q, and F denote prescribed continuous func-

tions of x such that r' is also continuous and r(x) > O. In the boundary
value problem
(1) [r(x)y'(x))' - Q(x)y(x) = F(x), y(a) = y(b) = O,
1 See, for instanee, E. L. Inee, "Ordinary Di!ferential Equations," pp. 72 !f., 1927, also, p. 123 for
more on adjoints, or E. C. Titehmarsh, "Eigenfunetion Expansions," p. 6, 1946.
GENERALlZED FOURIER SERIES [SECo 95 279
. the unknown function y can be interpreted as the static transverse displace-

ments in a stretched string attached to an elastic foundation. The elastic
support may have variable properties so that it contributes a transverse
force - Q(x)y(x) per unit length as well as a variable tension r(x) in the string.
An external transverse force - F(x), per unit length, acts along the string.
(Cf. Secs. 40 and 93.) The ends are fixed at the points (a,O)and (b,O). We make
the interpretation to assist us in writing the solution of problem (1), at first
formally, in terms of solutions of simpler problems.
Let F(x) be replaced formally by the unit impulse symbol <5(x- ~),
where a < ~ < b. This corresponds to the application of a negative unit
transverse force at the point x = ~ (Fig. 99). When the string has that
concentrated load, we write G(x,~) for the displacements y(x). Since the
vertical force from left to right is - r(x)y'(x), we note that the value of - ry'
must jump by the amount -1 at x = ~; that is,
1
(2)
GA~ + o, ~) - Gi ~ - O,~) = r(~r
The displacement function G(x,~) itself should be continuous in the

square region a ;;;x ;;;b, a ;;; ~ ;;;b, and its derivative Gx(x,~) should be
continuous there except for the jump prescribed by Eq. (2) at the diagonal
x =~. Since <5(x- ~) is zero except when x = ~,
(3) (x # ~);
therefore, in view ofthe conditions imposed on r and Q, the function Gxix,~)

is continuous except at the diagonal x = ~. AIso, G(x,~) satisfies the bound-
ary conditions
(4)
A function that satisfies conditions (2) to (4) and the stated continuity
conditions is called Green's function for problem (1). Specifically, and for a
problem with general linear homogeneous boundary conditions, we describe
the function as follows.
y-G(x,f
o (a,O) (b,O) x
Fig.99
J
I
DEFINITION
Green's function G(X,~)for this boundary value problem with self-adjoint

I differential equation:
(5) (r(x)y'(x)]' - Q(x)y(x) = F(x),
I
(6) Aly(a) + A2y'(a) = O, Bly(b) + B2Y'(b) = O,
is a solution of the homogeneous differential equation (3) when a < x < ~
and when ~ < x < b that satisfies the following conditions. It is continuous
in x and ~ together over the square a ;;;x ;;;b, a ;;;~ ;;;b (Fig. 100). Its
derivativeGxiscontinuousovertheclosedtriangleT{a;;; x;;; b,a;;; ~;;; x)
when properly defined on the diagonal x = ~,and continuous over the closed
triangle T2when again properly defined on the diagonal, but Gx has the jump
l/r(~) from TI to T2 at x = ~ as specifiedby condition (2). AIso, when ~
is fixed (a < ~ < b), G satisfies boundary conditions (6):
I (7) AIG(a,~) + A2Gx(a,~) = BIG(b,~) + B2Gx(b,~)= O.
Of course, one of the A's and one of the B's here must not vanish. A
convenient way to specify this is to write
(8) Al = cosa, A2 = sina, BI = cosp, B2 = sinp,
where a and pare constants. Note that Gxx has the continuity properties of
Gx, in view of Eq. (3).
We now indicate how the solution of problem (1) may be written as a
weighted superposition of solutions G of the problem in which the load F(x)
is replaced by concentrated loads 15(x - ~).
The impulse symbol has the formal properties 15(- x) = 15(x)and
F(x) = f F(~)15(~ - x) d~ = f F(~)15(x
that is, F(x) is a superposition of impulses. If all members of the symbolic

- ~) d~;
equations
(r(x)Gx(x,~)]x - Q(x)G(x,~) = 15(x - ~),
G(a,~) = G(b,~) = O,
~
(O,b)
10,a)
o (b,O) x Fig.100
are multiplied by F(~) and integrated with respect to ~ from ~= a to ~ = b,

the result is the three statements in problem (1) when we write
(9)
y(x) = f F(~)G(x,~)d~
and identify (ry')' with the symbol S: F(~)(r(x)Gx(x,~)]xd~.

In order to verify the formal solution (9), we first note that y(a) =
y(b) = Obecause G(x,~)satisfies the boundary conditions (4). When we write
y(x) = r F@G(x,~) d~ + f F(~)G(x,~) d~,
the integrand of each integral and its first two partial derivatives with respect
to x are continuous functions of (x,~) in the triangular regions 1;. and T2,
respectively, in Fig. 100. Differentiation by the Leibnitz formula (Sec. 14) is
therefore valido We find that
since G is continuous, and that
(10)
(ry')' = f F(~)[r(x)Gx(x,mx d~ + F(x)r(x)J(x),
where J(x) = G..,(x,x - O)- Gx(x,x + O). Thus J(x) is the jump in the
value of Gx as the point (x,~) moves across the diagonal from triangle TI into
triangle ~ (Fig. 100). According to Eq. (2), J(x) = l/r(x). In view of Eq. (3),
Eq. (10) can now be written
(ry')' = Q(x) f F(~)G(x,~)d~ + F(x) = Qy + F;

that is, the function (9) also satisfies the differential equation in problem (1).
The solution (9) of problem (1) is now established if Green's function G
exists. In fact the solution is valid if F is a sectionally continuous function.
For modifications of the procedure when the boundary conditions are
type (6), see Probs. 8 and 9, Seco97.
96 CONSTRUCTION OF GREEN'S FUNCTION

A function f(x) is of class C' on an interval a ~ x ~ b if both f and f' are
continuous over the interval.
Let r be of class C' over that interval, where r(x) > O, and let Q be
continuous there. Then, according to the existence theorem stated in Seco94,
the initial-value problem

(1) [r(x)u'(x)J' - Q(X)U(X)= 0, u(a) = 0, u'(a) = 1,
has a unique solution u(x) of class C' on that interval. Let v(x) be the unique
solution, of that class, of the initial-value problem
(2) [r(x)v'(x)J' - Q(x)v(x) = 0, v(b) = 0, v'(b) = ~.
Then on the square region a ~ x ~ b, a ~ ~ ~ b, the function

A(~)u(x) (x ~ ~)
(3)
G(x,~) = { B(~)v(x) (x ~ ~),
where A and B are continuous functions, satisfies the self-adjoint differential
equation (3), Seco95, and the boundary conditions
G(a,~)= 0, G(b,~) = O.
It is continuous there, on the diagonal x = ~ in particular, and its derivative
Gxhas the jump prescribed by Eq. (2), Seco95, at x = ~ if A.and B satisfy
the simultaneousequations
1
(4) B( ~)v(~) - A( ~)u(~) = 0,
B(~)v'(~) - A(~)u'(~) = r(~)'
According to Lagrange's identity (5), Seco 94, the determinant

- v(~)u'(~)+ u(~)v'(~)of system (4) satisfies the condition
d
(5) --[(vu' - uv')r] = -[(ru'Y - Qu]v + [(rv'Y - Qv]u.
d~
The right-hand member vanishes since u and v satisfy the differential equa-
tionsinproblems(1)and(2). Hence(vu' - uv')risconstantand,sincev(b)
=
and v'(b) = 1,
-[v(~)u'R) - u(~)v'(~)]r(~) = u(b)r(b).
Then in case u(b) =F0, we find that the solution of system (4) is
u(~) v(~)
B(~) = u(b)r(b)' AR) = u(b)r(b)'
and formula (3) for Green 's function becomes
(6) G(x,~) = u(x)v(~) when x ~ ~,

u(b)r(b)
= u(~)v(x) when x ~ ~.
u(b)r(b)
Note that G is symmetric in its two arguments:

(7) G(x,~) = G(~,x),
a consequence of the self-adjoint character of the differential equation.
In case u(b) = O then, since u(a)= O and u'(a)= 1, the function
w = u(x) is not identically zero, and it is a solution of class e of the homo-
geneous problem
(8) (rw')' - Qw = O,
w(a)= w(b)= O.
Since v satisfies the same differential equation and v(b) = O,it followsthat
v(x) = ku(x) when u(b) = O,where ku'(b) = 1. The following theorem is now
established.
Theorem 1 1J the homogeneous problem (8) has no solution oJ class e on the

interval a ~ x ~ b other than w(x) ==O, then Green's Junction G(x,~)
exists Jor the boundary value problem
(9) (r(x)y'(x)' - Q(x)y(x) = F(x), y(a) = y(b) = O,
where r is oJ class e, Q is continuous, F is sectionally continuous, and
r(x) > O, on the interval a ~ x ~ b. Green's Junction is expressed in
terms oJ solutions oJ the initial-value problems (1) and (2) by Jormula (6);
it is symmetric in x and~. The solution, oJ class e, oJ problem (9) is
(10) y(x) = f F(~)G(x,~) d~.

Although the initial-value problems (1) and (2) are simpler than the
boundary value problem (9), they can be solved explicitly only for limited
types of coefficients r and Q.
Under the conditions as()umed in Theorem 1, suppose that there is.
another solution z(x) of problem (9), in addition to y(x). Then the function
w = y - z satisfiesthe homogeneousproblem (8)whichhas only the solution
w(x) ==O. Thus z = y, and we ha ve the following theorem on uniqueness.
Theorem 2 Under the conditions stated in Theorem 1, the solution (10) oJ

problem (9) is the only one oJ class C.
Examples and additional material on Green's functions are given in
the problems following Seco97.1
97 ORTHOGONAL SETS OF FUNCTIONS

Consider now real-valued functions of x that are continuous over an interval
a ~ x ~ b. A function J may be called a generalized vector, with one
component J(x) corresponding to each value of x on the interval (Sec. 10).
1 For further theory on Green's functions for ordinary and partial differential equations see,
for instance, R. Courant and D. Hilbert, "Methods of Mathematical Physics," vol. 1, 1953.
A generalized scalar product, the inner product (J,g), of two functions

f and g is defined as the sum of products of corresponding components:
(1)
(J,g) = f f(x)g(x) dx = (g,f).
The norm of a function J, the generalized concept of the length of a vector, is
(fJ)t; that is,
(2)
IIfII = {f If(xW dx y.
Two functions f and g are orthogonal on the interval if (J,g) = O. This
signifies nothing about actual perpendicularity, but instead that the product
of the functions changes sign in such a way that
(3)
f f(x)g(x) dx = O.
Corresponding to units of length that differ from one coordinate axis
to another in vector analysis, we introduce a nonnegative weight function p
and define the inner product with respect to p, of two functions, as
(4) (J,g) = f p(x)f(x)g(x) dx = (gj) [P(x)~ O].

The norm 11 fII is then (fJ)t; and f and g are orthogonal if (J,g) = O.
Let a set of functions Yn(n = 1,2,...) be orthogonal on the interval,
with weight function p. It is convenient to normalize the set by dividing each
function by its norm. Thus the set
(5) (a ~ x ~ b, n = 1,2,...)
is orthogonal with unit norms, an orthonormal seto It corresponds to a set of

mutually perpendicular unit reference vectors. A prescribed function f may
possibly be represented as a generalized linear combination, an infinite
series, of those orthonormal functions,
00
(6) f(x) = : cncPn(x) (a < x < b).
n=1
Ir so, and if it is true that (cPmJ)= ::= 1 cn(cPm,cPJ,we can determine the
constants cn. F or (cPm,cPn)
= Owhen n "" m, and 11 cPm11= 1. Thus (J,cPm)= cm;
that is
(7) (n = 1,2,.. .).

Series (6) with coefficients (7) is the generalized Fourier series for f
corresponding to the orthonormal set of functions </1n.The coefficients cn
are called the Fourier constants for the function f In case the orthogonal sets
are generated by certain eigenvalue problems known as Sturm-Liouville
problems, we shall establish useful conditions on f under which its genera-
lized Fourier series converges to f(x).
The set of functions Yn(x) = sin nx is a well-known example of ortho-
gonal sets. The set is orthogonal with unit weight function on the interval
(O,n)and IIYnl12 = n/2 for each n; thus the orthonormal set consists of the
functions
(8)
</1n(x) = A sin nx (O< x < n, n = 1,2,.. .).
F or this set series (6) becomes the Fourier sine series:

co 2 co
"
f(x) =
A-nn=l Cnsin nx = -
nn=l
5.o
f(~) sin n~d~sin nx,
which converges to f(x) when O < x < n at each point where f is continuous,
if f' is sectionally continuous on the interval.l
PROBLEMS
1. Verify Lagrange's identities (3) and (5), Seco94.
2. Use Lagrange's identity to show that v(x) is an integrating factor for the differential
equation
if v is not identically zero and satisfies the homogeneous equation D~v = 0, where
D! is the adjoint of D2. Thus show that D2y = f reduces to this equation of first order:
a].vy' + [al v - (a2v)']y= f v(x)f(x) dx:
3. The adjoint of the linear differentialform

D"y = ao(x)y(x) + al(x)y'(x) + a2(x)y"(x) + ... + an(x)y(n)(x)
of order n, is the form
D:y = aoy - (aly)' + (a2Y)"- ... + (-l)n(any)(n).
When n ==3, establish the Lagrange identity
1 See the author's "Fourier Series and Boundary Value Problems," 2d ed., 1963; also, ortho-
gonal sets are treated somewhat more fully in Chap. 3 of that book.
where B3 is this bilinear form in v and y:

B3 = CXIVY + [CX2Vy' - (CX2V)'Y] + [CX3VY" - (CX3V)'Y'+ (CX3V)"Y].
4. Show that this special case of D.y in Probo 3,

D4y = cxo(x)y(x)+ y(4)(X),
is self-adjoint, and that it satisfies the Lagrange identity
d
vD4y - yD4v = dx(vy'" - v'y" + v"y' - v"'y).
5. Find Green's function for the elementary problem

y"(x) = F(x), y(O) = O, y(1)= O,
and solve for y in the form
y(x) = (x - 1) I: ~F(~)d~ + x f (~ - I)F(~)d~.

Ans. G(x,~) = x(~ - 1) when x ~ ~, G(x,~) = G(~,x).
6. Obtain Green's function (4), Seco82, for the problem
u"(x,s) - su(x,s) = - g(x), u(O,s) = u(l,s) = O (s > O).
7. Let F be a sectionally continuous function, with only one jump at a point Xo, for
the sake of simplicity, on the interval a < x < b. Carry out the verification of the
solution (Sec. 95)
of the boundary value problem

y(x) = f F(~)G(x,~)d~
[r(x)y'(x)]' - Q(x)y(x) = F(x), y(a) = y(b) = O (x #<xo),

where G is Green's function for the problem. Note that the solution y(x) is of class C'
if y'(xo) is defined as f: F(~)G x<.x'o'~)d~.
8. Let the constants cxandP be real and let w = u(x) and w = v(x) be the solutions of
the homogeneous self-adjoint equation (rw')' - Qw = Othat satisfy these initial condi-
tions:
u(a) = sin cx, u'(a) = -cos CX,
v(b)= sinp, v'(b)= -cosp.
ConstructGreen's functionfor this problem(Sec.95):
[r(x)y'(x)]'- Q(x)y(x)= F(x) (a < x < b),
y(a) cos cx+ y'(a) sin cx= O, y(b) cos P+ y'(b) sin P = O,
in case e #<O, where
e = r(a)[v(a) cos cx+ v'(a) sin cx]
= -r(b)[u(b)cosP + u'(b)sinP],
in the form
G(x,~) = u(x)v(~)
e
when x ;; ~
= u(~)v(x) when ~ ;; x.
e
9. When F is continuous(a ;; x ;;b), verify the solution
y(x) = f F(~)G(x,~)
d~
of the boundary value problem in y for which Green's function G was constructed in
Probo 8.
10. When the constant k is neither zero nor an integer, use Green's function (Probs. 8
and 9) to solve this problem for y:
y"(x) + ey(x) = F(x), y'(0)= O, y'(7t)= o.
Ans. y(x)k sin k7t = I: F(~) cos k~ d~ cos k(7t - x) + f F@ COS k(7t - ~)d~ cos kx.
11. Write the differential equation in the problem

(x + l)y"(x) + y'(x) = F(x), y(O)= O, y'(I) = O,
in self-adjoint form and use Green's function (Probs. 8 and 9) to solve the problem in
the form
y(x) =- f: F(~)log(~ + l)d~ - log(x + 1) f F(~)d~.
12. Interpret the symmetry property G(x,xo) = G(xo,x) of Green's function in terms
of displacementsofthe stringconsideredin Seco95 and concentratedloads,at pointsx
and Xo.
13. Supposea function w of c1assC' is not identicalIyzero and satisfiesthe homo-
geneous problem
[r(x)w'(x)]' - Q(x)w(x) = O, w(a) = w(b) = O.
Then if the problem

[r(x)y'(x)]' - Q(x)y(x) = F(x), y(a) = y(b) = O,
has a solutionof cIass C', use Lagrange's identity to show that F must be orthogonal to
w on the interval (a,b):
f F(x)w(x) dx = O.
IIIustrate that result using the problem
y"(x) + y(x) = - 3 sin 2x, y(O) = y(7t) = O,
whichhas the familyof solutionsy = e sin x + sin 2x.

14. Show that the set of functions

1 mrx
.fi' A-cos-
e e (n = 1,2,...)
is orthonormal on the interval (O,e)with unit weight function, and that the generalized
Fourier series for a functionJ, with respect to the set, is the Fourier eosine series
l 2 '" mrx mr.;
-
e o [ f(.;)d.; + -e cos- e [o f(.;)cos-d.;e
n= 1
(O < x < e).
98 EIGENVALUE PROBLEMS
The homogeneoustwo-point boundary value problem
[r(x)Y'(X,A)]'- [q(x) + Ap(X)]Y(X,A)= O (a < x < b)
(1)
A1y(a,A) + A2Y'(a,A) = O, B1y(b,A) + B2Y'(b,A)= O,
involving a parameter A in the manner indicated, is called a Sturm-Liouville
problem or system.l The constants Al, A 2, B l' and B2 are real and independ-
ent of A. The functions p, q, r, and r' are real-valued and continuous, and
p(x) > Oand r(x) > O,over the entire interval a ~ x ~ b.
Problem (1) has the trivial solution y = O,for everyvalue of A. But for
certain values Anof A,called eigenvalues or characteristic numbers,the problem
has a non trivial solution
Y(X,An) = Yn(x),
an eigenfunction or characteristic function. Note that y = Ym(x)and y = y.(x)

are solutions of different problems obtained by writing Amand An,respectively,
for A in problem (1). Also note that if Yn(x)is an eigenfunction, then CYn(x)is
also one, where C is any constant other than zero.
The Sturm-Liouville problem is a particular type of eigenvalue prob-
lem. Such problems arise in the process of solving boundary value problems
in partial differential equations by the classical method of separation of
variables, a method we shall illustrate in this chapter.2
Since eigenfunctions satisfy the Sturm-Liouville differential equation,
their deriva tives of second order are to exist over the open interval (a,b). We
assume the functions are of class C' (Sec. 96) on the interval a ~ x ~ b,and
show later on that eigenfunctions of that class do exist. AIso, we write the
Sturm-Liouville equation in terms of our self-adjoint operator ~2 in the
form
~2Y - APY = O, where ~2Y = (ry')' - qy.
1 The first extensive development of the theory of such systems was published by 1. C. F. Sturm
and 1. Liouville in the first three volumes of Journal de mathmatique, 1836-1838.
2 See also the author's "Fourier Series and Boundary Value Problems," 2d ed., 1963.
GENERAlIZED FOURIER SERIES [SECo 98 289
Let y and z denote any two eigenfunctions of problem (1) corresponding

to eigenvalues A and J1.,respectively. Then both functions satisfy the first
boundary condition,
(2) Aly(a) + AzY'(a)= O,
Since Al and A2 are not both zero, the determinant ofthis pair of simultaneous
linear equations in Al and A2 must vanish; that is, W(a) = O,where W is the
Wronskian of the functions y and z:
y(x) z(x)
(3) W(x) = = y(x)z'(x) - y'(x)z(x),
I y'(x) z'(x) I
and similarly for the second boundary condition. Thus the functions y and z
must be such that
(4) W(a) = O, W(b) = O.
AIso, they satisfy the Sturm-Liouville equations
~2Y = Ap(X)y(x), ~2Z = J1.p(x)z(x),
and from these and Lagrange's identity (5), Seco94, it follows that
(5) (J1.- A)PYZ= y~2Z - Z~2Y

d d
= d)Yz' - y'z)r) = d)W(x)r(x)).
In view of conditions (4), therefore,
(6)
(J1.- A) f pyz dx = W(b)r(b) ...,.W(a)r(a) = O.
In case A.=f.J1.,it follows that y and z are orthogonal on the interval

(a,b), with weight function p:
(7) f p(x)y(x)z(x) dx = O
(A =f. J1.).
If the eigenfunctions y and z correspond to the same eigenvalue, then

J1.= A in Eq. (5), and so (Wr)' = O and Wr is constant; but the constant is
zero because W(a) = O. Since r(x) > O,it follows that the Wronskian of y and
z vanishes :
(8) W(x) = y(x)z'(x) - y'(x)z(x) == O (a ~ x ~ b).
In that case we now prove y and z linearly dependent.
Consider the linear combination of y and z,
(9) X(x) = [A2z(a) - Al~'(a))y(x) - [A2y(a) - Ay'(a))z(x),
which also satisfies the homogeneous equation f!)2X = A.x. The reader can
X(a) = Oand X'(a) = O because
verify that it satisfies the initial conditions
W(a) = O. Therefore X(x) == O, the unique solution of that equation under
those conditions. Moreover, the coefficient of y(x) in formula (9) cannot
vanish, in view of the second of boundary conditions (2) and because z(a)
and z'(a) cannot both vanish, for the determinant of the pair of equations
A2z(a) - Alz'(a) = O,' A1z(a)+ A2z'(a)= O
in z(a) and z'(a) has the value A22 + Al 2. That value is positive for our real
numbers Al and A2.
Likewise, the coefficient of z(x) in formula (9) is a nonzero constant.
SinceX(x) == O,it followsthat there is a constant k such that
(10) z(x) = ky(x) (k*O,a~x~b).
Finally, we show that all eigenvalues are real and have corresponding
real-valued eigenfunctions. For if y(x) is an eigenfunction corresponding to
an eigenvalue ..1.,then by writing complex conjugates of all terms of the
Sturm-Liouville problem (1), we can find that
f!)2Y= y, Aly(a) + A2Y'(a) = Bji(b) + B2y'(b) = O.
Thus y is an eigenfunction corresponding to and, according to Eq. (6), it
follows that
f
( - ..1.) p(x)y(x)y(x) dx = O.
But y is not identically zero, and yy = IYl2~ O,while p(x) > O,so the integral
here has a positive value; therefore, - A.= O. Thus = ..1.,
so A.is real.
Now since y and y are eigenfunctionscorresponding to the same real
eigenvalue ..1.,then y and y are linearly dependent; that is, for some constant
k, y = ky. In case k = -1, then y = - y and iy = iy, so iy is a real-valued
eigenfunction. If k * -1, then (1 + k)y is the eigenfunction y + y, or 2 Re y,
which is real-valued. That is, each eigenfunction can be made real-valued by
multiplying it by an appropriate complex constant.
We have now established the following properties of characteristic
numbers and functions.
Theorem 3 Each eigenvalue ..1.n of the Sturm-Liouville problem (1) is real.

The eigenfunction Yn corresponding to ..1.nis unique up to an arbitrary
constant factor k other than zero, and k can be chosen so that the eigen-
function is real-valued. Eigenfunctions Ym and Yn corresponding to
different eigenvalues ..1.m
and A.nhave the orthogonality property
f p(x)Ym(x)Yn(x)
dx = O (A.m* ..1.n)'
99 A REPRESENTATIONTHEOREM
In the sections to follow, the existence of an infinite sequerice o eigenvalues
Anofthe Sturm-Liouville system will be established. We then 'show that the
generalized Fourier series of orthonormal eigenfunctions, for each function f
of a broad class, does converge to f(x) on the interval (a,b).
Let Yndenote real-valued eigenfunctions of the system
(1) [r(x)y'(x)]' - [q(x) + Ap(x)]y(x) = O (a < x < b),
A1y(a)+ A2J'(a)= O, B1y(b)+ B2J'(b)= O,
in case eigenfunctionsexist. Then the normalized eigenfunctions
(2)
are orthonormal on the interval (a,b) with weight function p:

b
Oifmi=n
(3) fa p(x)~m(x)~n(x)dx = { 1 if m = n (m,n = 1,2, o o 0)0
The generalized Fourier series of those functions, for a function f on the

interval,is
f
00
(4) L en~n(x)
n= 1 where en = p(x)f(x)~n(x) dxo
Our theorems on representing the inversion integral by a series of
residues are useful in establishing the following representation theorem for
the series (4)0
Theorem 4 Let p, q, r, r', and (pr)" be eontinuous real-valued funetions of x

sueh that p(x) > O and r(x) > O, when a ~ x ~ b, and let the real-
valued eonstants Al, A2, B l' and B2 be independent of A. Then the
Sturm-Liouville problem (1) has an irifinite sequenee of eigenvalues
An(n= 1,2,.. .), all real, and at most afinite number of them are positive.
The eorresponding real-valued orthonormal eigenfunetions ~n are of
class C' on the interval a ~ x ~ b; also ~n(x) and, when Ani= O,
An-t~~(x) are bounded with respeet to both x and An Eaeh seetionally
eontinuous funetion f with seetionally eontinuous derivatives l' and f" is
represented by itsgeneralized Fourier series(4)at all points x (a < x < b)
where f is eontinuouso1
The proof of the theorem is lengthyo We carry it out fully for the case
A2 = B2 = O and indicate in the problems some variations needed to
1 The theorem is true without the condition onf", but in our proofit is convenient to assumef"
sectionaIly continuous.
prove the general case. The proof exhibits interesting devices for finding
properties of solutions of differential equations when the equations are too
general to solve.
100 THE REDUCED STURM-LlOUVILLE SYSTEM

Under the conditions stated in Theorem 4 on p, q, and r, the Sturm-Liouville
problem can be reduced to a simpler form by substituting certain new
variables t and z for x and y. In the Sturm-Liouville differential equation we
first write y(x) = v(x)X(x) and x = x(t) or t = t(x), where v and t(x) are yet
to be determined, then write X(x) = z(t). Ir dots denote deriva tives with re-
spect to t, then the equation (ry')' = (q + Ap)y takes the form
.. 2rvl + r'v . X , PX2
(1) z+ x - -; z. -- + 1\.- z,
( rv x) (Q r )
where Q(t) is independent of A.
Equation (1) can now be simplified by determining t(x) or x(t) so that
px2jr = e2, where e is a constant, then v(x) so that the coefficientof i
vanishes. Details are left to the problems. The results are
x
P(~)
(2) t = .!.
e f a [ r(~)J
t d~
,
z(t) = (pr)*y(x),
and, to reduce the interval a < x < b to O < t < 1, we write

b
P(~)
(3) e=
f a [r(~)J
t d~.
Then the Sturm-Liouville equation takes the reduced form

(4) z(t) - [Q(t) + Jl]z(t) = O (O< t < 1),
where Jl = e2 A and z is a function of t and Jl.
Under the substitutions (2) the boundary conditions in the Sturm-
Liouville system retain their linear homogeneous forms. Thus
(5) alz(O) + a2i(0) = O, b1z(l) + b2i(1) = O,
where the a's and b's are real-valued constants related to the A's and B's.
In particular, if A2 = B2 = O so that y(a) = y(b) = O, then z(O)= z(l) = O,
and therefore a2 = b2 = O.
The reduced Sturm-Liouville problem consisting of the simpler
Sturm-Liouville equation (4) and conditions (5) transforms back into the
original problem in y under the substitutions (2). Orthogonality of eigen-
functions zit) on the interval (0,1) with unit weight function corresponds to
orthogonality of Yn(x) on (a,b) with weight function p. The reduced system
can be used to prove Theorem 4.
101 A RELATED BOUNDARY VALUE PROBLEM
To shorten the analysis, we now consider the case where a2 = b2 = Oin the
boundary conditions of our reduced Sturm-Liouville problem. Except
for changes in notation, that problem then reads
(1) X"(x) - [A + q(x)]X(x) = O, X(O) = X(l) = O.
We introduce, to serve as a guide to the analysis, a boundary value
problem in partial differential equations whose solution is associated with
problem (l). Let Y(x,t) denote transverse displacements in a string stretched
between points (0,0)and (1,0)when the string is attached along its span to an
elastic medium that may be nonuniform. If the string starts from r:est with a
prescribed initial displacement Y = f(x), and if the unit of time is properly
chosen, then
Y,r(x,t)= y':x(x,t) - q(x) Y(x,t) (O < x < 1, t > O),
(2)
Y(O,t)= Y(l,t) = Y,(x,O)= O, Y(x,O) = f(x).
The differential equation and both two-point boundary conditions art
homogeneous. Hence the method of separation of variables may apply to
problem (2). First, to find all functions not identically zero of type X(x)T(t)
that satisfy al! homogeneous conditions, we can write
XT" = X"T - qXT, X(O)= X(l) = T(O) = O,
T"(t) X"(x ) q(x )X (x )
Consequently - = =A
T(t) X(x) ,
where the parameter A is independent of x and t, because T"jTis independent
of x and equal to a function that is independent of t. The function X therefore
satisfies all conditions in problem (1). Thus A must be an eigenvalue Anof
that Sturm-Liouville problem and, except for an arbitrary constant factor,
X(x) must be the corresponding orthonormal eigenfunction 4>n(x).The
function T(t) then satisfies the system T" = AnT, T(O) = O. The functions
X.T can therefore be written
(3) (n = 1, 2, . . .),
where the constallts Cnare arbitrary.

A sum of functions (3) also satisfies all homogeneous conditions in
problem (2). (Note that the sum itself is not a product of a function of x alone
by a function of t alone.) Formally, the series
<X>
(4) Y(x,t) = L
n=l
cn4>n(x)
cos r';=
satisfies all conditions in the problem, including the condition Y(x,O) = f(x),
if the numbers Cnare the Fourier constants of f(x) corresponding to the

orthonormal functions 4Jn(x),so that
. IX)
(5) f(x) = n=L1 cn4Jn(x) (O < x < 1),
where
(6) (n = 1,2,.. .).
Thus the Sturm-Liouville expansion (5) is needed to complete the formal

solution of problem (2) by the method of separation of variables.
Ir we now apply the Laplace transformation with respect to t to the
boundary value problem (2) in Y(x,t) and establish order properties of the
transform y(x,s), we may be able to show that L-l{y} = f(x) when t = O,
or that
when t = O.
L-l{y(X,s) - f~X)} = O
Then wecan anticipate that the sum of residuesofy may bejust the series(5),
and that carefulanalysiswill show that the seriesconvergesto f(x).
102 THE TRANSFORM y(x,s)

Formally, the Laplace transform of the function Y(x,t) in problem (2),
Seco101, satisfies the conditions
y"(X,S) - [S2 + q(x)y(x,s)
= -sf(x),
(1)
y(O,s) = y(1,s) = O.
To construct Green's function for problem (1) here, we introduce, as
in Seco96, the unique solutions u(x,s) and v(x,s) of these initial-value prob-
lems:
(2) U" - (S2 + q)u = O, u(O,s) = O, u'(O,s) = 1,
(3) v" - (S2 + q)v= O, v(l,s) = O, v'(l,s) = 1.
Note that u and v are actually functions of x and S2,hence even functions of s;
they are entire functions of S2 (Sec. 94) and therefore of S. Then for all s
such that u(l,s) :F O,Green's function can be written as
u(x,s)v(~,s)
(x ~ ~)
u(l,s)
(4)
G(x,~,s) = u(~,s)v(x,s)
{ (x ~ ~).
u(l,s)
Thus ifjis sectionally continuous, the solution of problem (1) is

l h(x,s)
y(x,s) = -s fo j(~)G(x,~,s)d~ = -s u(l,s)'
(5)
where
(6)
h(x,s) = v(x,s) f: f(~)u(~,s) d~ + u(x,s) f j(~)v(~,s) d~.
For each fixed x, h is an entire function of s. This follows from the Cauchy-
Riemann conditions, because the products uv are entire functions of s.
Consequently y is an analytic function of s except at the zeros of the
entire function u(l,s), which are necessarily isolated. But when u(1,s) = O,
we see from problem (2) that u(x,s) is a solution of our Sturm-Liouville
problem (1), Seco101, in which A.= S2. Ir A.nis the corresponding eigenvalue
and A denotes one of the two square roots of that real number, then
s = :t Sn where Sn =..::,. Thus if ljJnis the normalized eigenfunction,
u(x, :t sn) = kljJn(x)for some nonzero constant k. Since u'(O,sn)= 1, it follows
that ljJ~(O)=1:O and
(7)
But v(l,:tsn) = O so when u(l,:tsn) = O, then v(x,:tsn) = Ku(x,:tsn), where

K =1:Oand, since v'(l,:tsn) = 1,
ljJn(x)
(8)
v(x,:tsn) = ljJ~(1)'
The function ljJn'not identically zero, satisfies the conditions
(9)
From Eq. (6) to (8) we find that
(10)
ljJ~(1)h(x,:tsn) = :~~~; f j(~)ljJn(~) d~
ljJn(x)
= Cn ljJ~(O) ,
where Cnis the Fourier constant for our functionj Also, note that
(11) ljJ~(O) f u(x,:tsn)ljJix)dx = f [ljJn(x)]2dx = 1.

Eliminating q between Eqs. (2) and (9), we find that
so, in view of the boundary conditions on u and <Pn,
(12)
(S2 - s/) f u(x,s)<Pn(x)dx = - <p~(I)u(l,s).
Consequently formula (5) can be written

<p~(I)h(x,s) s
y(x,s) = 1 s 2 -s .
I
~
o u(x,s)<Pn(x)dx n
It follows that I snare simple poles of y. If Sn=1=

O,the residue of y at I Snis,
in view of formulas (10)and (11),
and the sum of those equal residues is cn<Pn(x).Similarly, if Sn= O,we find
that the residue ofy at snis cn<Pn(x).Hence the sum ofresidues ofthe function
y at all its poles s = ISn is formally the generalized Fourier series forf:
00
L cn<Pn(x).
n=1
We have yet to prove that an infinite sequence of eigenvalues exists

and that the series of residues converges to f(x).
103 EXISTENCE OF EIGENVALUES
Again, let u(x,s) be the unique solution of the initial-value problem

(1) u" - (S2 + q)u = O, u(O,s)= O, u'(O,s)= 1.
Then the solution of this problem in z(x,s):
(2) z" - S2Z = q(x)u(x,s), z(O,S) = O, z'(O,S)= 1
can be written in the form
(3) z(x,S) = ! sinh sx + !

s s o I
x q(~)u(~,s)sinh s(x - ~)d~
when s =1= o, as is easily verified or derived (cf. Probo 17, Seco 104). The
solution (3) is unique because the difference w - z of two solutions w and z
of problem (2) must satisfy a completely homogeneous initial-value problem,
-
so that w z ==O. But problem (1) shows that z = u(x,s)is the solution of
problem (2). Therefore u satisfies the integral equation
(4)
su(x,s) = sinh sx + f: q(~)u(~,s) sinh s(x - ~) d~,
where s is any complex number.
Likewise, for all s the solution v ofthe initial value problem (3), Seco102,
satisfies the integral equation
(5)
sv(x,s) = sinh s(x - 1) + f q(~)u(~,s) sinh s(~ - x) d~.
We write p = Isl and (J = IRe si. Then
when O~ x ~ 1;
thatis,e-uXlsinh sxl ~ 1ande-uxlcosh sxl ~ 1whenO ~ x ~ 1. Forafixeds

let M(s) denote the maximum value of the continuous function plu(x,s)le-UX
on the interval O ~ x ~ 1. The function assumes its maximum value at
some point x = k on the interval. When we write x = k in formula (4) and
introduce the factor e-uk there, we find that
M(s) ~ 1 + f: Iq(~)u(~,s)sinh s(k - ~)Ie-ukd~.
When the integrand here is written in the form
we can see that
M(s) ~ 1 + qo M(s), where

p
Therefore
(6) whenever p > 2qo.
A similar argument applies to v, using Eq. (5); thus
(7) plu(x,s)le-UX< 2, plv(x,s)le-U(1-X) < 2 whenever p > 2qo.
From Eq. (4) we now find that

(8) su(x,s) = sinh sx + s-leUXR1(x,s),
(9) u'(x,s) = cosh sx + s-leUXR2(x,sy,
where R 1 and R2 are continuous functions whose absolute values are less
than 2qo when p > 2qo. In particular,
1
(10) su(l,s)e-a = e-a sinh S + -R1(l,s),
S
where if
We found that u(l,s) cannot vanish unless S has real or pure imaginary
values :t.JI,;. If S is real and not zero, S = :tu, then e-a sinh S approaches
:ti as o' -+ 00, and if follows from Eq. (10) that a number')l exists such that
u(l,s) =FO when o' ~ ')l. All real zeros of the entire function u(1,s),therefore,
le on a bounded interval of the real axis; hence they are finite in number.
When S is apure imaginary number, o' = Oand s = ir,so
ru(l,ir)= sin '1 - .!.R1(1,ir) ('1 real).

'1
Note that ru(l,ir)is an odd function ofr since u(x,s) is an even function of s.
AIso u(x,ir)is real-valued because itis the solution ofthe initial-value problem
(1) in which the coefficients are all real, so that satisfies the same problem
and hence = u. Thus the function r-lR1(1,ir)is real-valuedand odd. It
is continuous and vanishes as '1 -+ <X)so its graph intersects that of sin '1at
an infinite sequence of points '1 = :trj (j = 1,2,.. .),where rjis arbitrarily
close to an integral multiple of 11:when rjis sufficiently large.
An infinite sequence of eigenvalues
(11)
(A'n~ O if n ~ m, An < O if n > m),
or simple poles :tsn = :t.JI,; of y(x,s),therefore exists. When n > m, then
sn = irn'where 'Inis real and, according to Eqs. (8) and (9),
(12)
(13)
With the aid of the last two equations we find that the norm 11rnu(x,sn) 11
tends to 1/J2 as n -+ 00, and that a constant K, independent of n, exists

such that the normalized eigenfunctions l/Jnsatisfy conditions
(14) 14>n<x)1< K,
Since u is of class C' in x, then 4>nis also of that class. Qur principal results
are stated as follows.
Theorem 5 The reduced Sturm-Liouville problem (l), Seco 101, has an
than afinite
infinite number
set (ll) ofthem are A.n.
of eigenvalues positive. The value of
All eigenvalues - A.nand
arjeal, is arbitrarily
no more
close to an integral multiple of 1t when n is sufficiently large. The corre-
sponding orthonormal eigenfunctions <Pnare of class C', and <Pn(x}and
<p~(x}satisfy the boundednessconditions (14).
104 THE GENERALlZED FOURIER SERIES
To prove that the series of residues of y converges to f(x}, we first find order
properties of Green's function G and our function y.
Againwewriteu = IResi and p = Isl. When x ~ ~,
) - u(x,s}vR,s} - [se-"Xu(x,s}][se-"(l-~)v(~,s)]
G(x,~,s- u(l,s) - s2e-"u(1,s)e"(~-X) .
In view of conditions (7) and Eq. (10), Seco103, then
4
(1)
.h
IG(x,~,s)1 < pe1 -" sms + s -l R l'S
(1 )1
This condition is valid also when x ~ ~, owing to the sYlOlOetryof Green's

function. The denominator vanishes only at points lying along the axis of
ilOaginaries or at points on a segment u < y of the real axis, since the zeros
:t Snof u(l,s) are so distributed. In case all the zeros are pure ilOaginary
nulObers, we let y denote any positive number.
Throughout either of the two half planes u ~ y, the denominator in
condition (1) is bounded away frolO zero, since
e-"Isinh si ~ !<1 - e-2y) > O
and s-lR1(1,s) -t Oas p -t oo. Therefore G is (I)(p-l) over those halfplanes.
For points s = :tu + il] on lines 1]= :t(N - t)1t (N = 1,2,.. .),
e-"Isinh si = e-" cosh u = t(1 + e-2") >. t.
It follows frolO condition (1) that G is (I)(p- 1) on those lines as p -t oo.
Hence for all points s in a right half plan e or on the open rectangular paths
CN used in Theorem 10, Seco 73, in which PN = (N - t)1t, a constant M
independent of x, ~, and N exists such that
M
(2) IG(x,~,s)1< - when Re s ~ y or when s is on CN.
p
It follows that our function
(3) y(x,s) = -s ff(~)G(x'~'S)d~

is at least bounded over the halfplane and on CN.
Since for a fixed x, Green's function satisfies the homogeneous differen-

tial equation of which u(~,s) and vR,s) are solutions, we can make the
substitution
(~ :F x)
into Eq. (3) to write

1 l 1 l
y(x,s) + -
s o f
f(~)G~~(x,~,s)d~ = -
s o f
f(~)q(~)G(x,~,s) d~.
The right-hand member here is &(s- 2) on the half plane Re s ~ 'Yand on the
paths CN, in view of condition (2). The left-hand member of the equation
therefore satisfies the same conditions, and so its inversion integral along
the line Re s = 'Y has the value zero when t = O; also (Sec. 73) its integral
over CN vanishes as N -. oo. That is, if CN denotes the closed rectangle
with the side along Re s = 'Yadded to CN, then
-. oo.
(4) ICN [ y(x,s) + ~ I: f(~)G~~(x,~,s)
d~ ds -. O ] as N
Now let us assume 1", and therefore f' and J, sectionally continuous.
It will suffice to let f have just one jump of an amount jo at a point XO
(O< Xo < 1). Let x be a fixed point of the interval at whichf is continuous,
say O < x < Xo. The jump in G~ at the point ~ = x is 1, as we can see from
the expression for G in terms of u and v. Then
l x xo l
fo fG~~ d~ = fo fG~~ d~+ f

x
fG~~ d~ + J~ fG~~d~
= fG~lo + fG~I~o + fGd;o - I: f'G~ d~
= - f(x) - f(O)G~(x,O,s)- joG~(x,xo,s)
+ f(l)G~(x,l,s) - I: f'G~ d~,
if we write f(O) for f( + O) and f(l) for f(1 - O). Note that G~x,O,s)is
u'(O,s)v(x,s)/u(l,s)and u'(O,s)= 1, etc. Thus in terms of the function
(5) (x s) - f(O) v(x,s) + . u(x,s)v'(xo,s) - f(l) u(x,s)

g , - su(l,s) lo su(l,s) su(l,s)
the bracketed integrand in Eq. (4) is the sum
(6)
The final member of the sum (6) is (!J(s-Z)on CNas we can see by
integrating the sectionally continuous function l' G~by parts and recalling
that G itself and therefore such terms as jG(x,x,s) and gf"G d~ are (!J(S-l).
Thus the integral around CN of that member tends to zero as N -+ oo.
We find, corresponding to Eqs. (8) and (9), Seco103, that
(7) sv(x,s)= -sinh s(l - x) + S-lea(1-X)R3(x,s),
(8) v'(x,s) = coshs(l - x) + s-lea(1-X)R4(x,s),
where R3 and R4 are bounded when Isl > 2qo and O ~ x ~ 1.
We write the coefficient of jo in the sum (5) in the forms
1
~ 1 [sinh sx + s-leaXR1(x,s)][cosh s(l
s u( ,s) - xo) + S-lea(1-xo)R4(xo,s)]
= e-a sinh sx cosh s(l - xo)

se-a sinh s + R1(1,s) + gl(X,XO's),
where it is readily verified that gl is (!J(s-z) on CN. Since l/(a + b) can be

written as l/a -
(b/a)/(a + b), that coefficient is
sinh sx cosh
. s(l - xo) + R1(1,s)sinh
;--. sx cosh s(l - xo) + g 1(X,Xo,S)
s smh s s smh s se-a smh S + R1(1,s )
= sinh s(l + x - xo) - sinh s(l - x - xo) + gz(x,xo,s) + gl(X,XO's),

2ssinhs
where gz is (!J(s-Z)on CN. Note that 0< 1 + x - Xo < 1 and 11 - x - xol

< 1. Thus the two terms of weak order here are of the type
tf(e,s)= si~h es
s smh s where - 1 < e < 1.
Actually, tf is the transform of a square-wave function which vanishes over
an interval containing the origin t = O(Prob. 9, Seco104).
An examination ofthe integral of 1tf1 around CN shows that JCNtfds -+ O
as N -+ 00 (Prob. 11 below). Similarly, the integrals around CN of the
coefficients of f(O) andf(l) in the sum (5) vanish as N -+ 00; the reduction
in terms of the function tfis simplerfor those coefficients.Thus JCNgds -+ O
as N -+ oo.
Consequently, condition (4) reduces to the equation
(9)
.
11m
i[
N oo CN
-
f(X)
y(x,s) - -
S]
ds = O;
that is, the sum of the residues of the integrand at the poles inside CN con-
verges to zero as N -+ 000 But the sum of residues of y were found in Seco102
to be 'I:.cnl/Jn(x);therefore
f
00
(lO) L cnl/Jn{x) -
n= 1
f{x) = O,
where Cn = f{x)l/Jn{x)dx.
There are no essential changes in the proof of formula (lO) when
Xo < x < 1. Thus :rheorem 4 is established for the reduced form of the
Sturm-Liouville problem with boundary conditions X{O) = X(1) = O. A
similar procedure, but more involved, can be followed to prove the theorem
when the more general boundary conditions of type (5), Seco 100, are usedo
Theorem 4 as stated in Seco 99 then follows by transforming the
reduced problem back to the original Sturm-Liouville problem with the
substitutions (2), Seco 100.
PROBLEMS
1 . Prove that the eigenvalues A..of the Sturm-Liouville problem
y"(x) - (A.+ h)y(x) = O, O< x < e; y(0) = y(c) = O,
where h is real and constant, consist of the numbers -h- n2n2/el (n = 1,2,0 o0)0 Note
that not more than a finite number of them are positive, and that - h is not an eigenvalueo
Show that the normalized eigenfunctions 4J. are .j2jC sin (nnx/c) and that according
to Theorem 4, iff" is sectionally continuous (O< x < e), then f(x) is represented at
each point wherefis continuous by its Fourier sine series on the interval (O,e):
2 QO nnx nn~
f(x) =- Lsin -
e .= 1 e l c
o
f(~) sin - e d~ (O< x < e).
2. Writef(x) = 1 (O < x < e) in Probo 1 to show that
in .=f 12n
~sin
- 1
(2n -e l)xx = 1 (O< x < c)o
3. Find all A..and 4J.(x)for the Sturm-Liouville problem

y"(x) - A.y(x)= O, y'(0) = y'(e) = O,
and obtain the Fourier cosine series expansion of f(x),
1 2 nnx
f(x) = -e lc
o
f(~) d~ +
QO
-e.=L1cos - e l
c
o
nn~
f(~) cos -
e
d~ (O< x < e),
as a specialcaseofTheorem 4. Note that when A. = Othe general solution ofthe Sturm-

Liouville equation is y = Ax + B, where A and B are arbitrary constantso
4. Find all characteristic numbers and functions of the systems
(a) y"(x) - A.y(x)= O, y(0)= y'(c)= O,
(b) y"(x) - A.y(x)= O, y(0) = O, y(l) + hy'(1) = O,
where h is a positive constant. Note the orthogonality of the characteristic functions

as given by Theorem 3.
Anso A. = -a/, y.(x) = sina.x, where in (a), a. = (n - !)7tc-l, and in (b),
a. > and tan a. = -ha. o
5. Use the substitutions (2), Seco100, to reduce the problem
(X4y')' - (A - 2x2)y = 0, y(1)= 0, y'(2)= 0,
to the form
z"(t) - JlZ{t) = 0, z(O)= 0, z(1) - z'(1) = O.
Then use them to derive the following expressions for the eigenvalues and eigenfunctions
of the problem in y:
Al = 0, A.= -4a.2 (n = 2,3,.. o)
where a. are the positive roots of the equation tan a = a;

(n = 2, 3, . o.).
6. For the fourth-order eigenvalue problem

d4y
dX4 - Ay = 0, y(0) = y'(0) = y(c) = y'(e) = 0,
prove that two eigenfunctions y and z and their corresponding eigenvalues A and .L
must satisfy the condition
J: y(x)z(x) dx
(A - .L) = 0,
and hence that ifA '# .L,y is orthogonal to z on the interval (O,e);also that all eigenvalues
must be real.
7. In Seco 100, complete the derivation of form (1) of the Sturm-Liouville equation
and of the substitutions (2) that produce the reduced form (4).
8. Derive formulas (7) and (8), Seco104, for sv(x,s) and v'(x,s).
9. Let'l' be this periodic square-wave function:
'I'(c,t)= if < t < 1 - e or if 1 + c < t < 2,
'I'(c,t)= 1 if 1 - e < t < 1 + c,
'I'(c,t + 2)= 'I'(t) whent > 0,
-
where ;;;e < 1; also '1'( c,t)= - 'I'(e,t). Show '1' graphicallyas a function of t.
Use Theorem 10, Seco23, to prove that its transform is the function t/J(e,s)used in
Seco104:
sinh es
L{'I'(e,t)} = t/J(c,s)= s sinh s (-1 < e < 1, Re s > O).
10. Apply Theorem 6, Seco68, to the square-wave-function 'I'(e,t) in Probo 9 to prove

that L-I{t/J} = Owhent = O.
11 . If we write s =v+ i", the sides of the cIosed rectangle CN used in Seco104 lie
on the lines v = y(y > O), v = - PN, and" = IPN, where PN = (N - t)7t, N = 1,2,....
For the function
sinh es
(-1 < e < 1)
tf(e,s)= s sinh s
establish this property, used in Seco104:
lim
N-C() JCN
f tf(c,s)ds = O,
with theaid ofthe result found in Prob.lO.
12. If the function f is continuous (O~ x ~ 1) and satisfies the boundary conditions
in the reduced Sturm-Liouville problem (1), Seco 101, so that f(O) = f(l) = O,and if!"
is sectionally continuous, prove that condition (9), Seco104, is satisfied uniformly in x;
hence that
<X>
f(x) L
= n=l c.4>.(x) (O~ x ~ 1)
and this generalized Fourier series for f converges uniformly.

13. Write the reduced system (1), Seco101, in the form
X"(x) - q(x)X(x) = AX(X), X(O)= X(l) = O,
and show that, if zero is not an eigenvalue of the system, each eigenfunction X. is a
solution of the homogeneous integral equation
X.(x) = A. fX.(~)G(x,~,O)d~
where A.is the eigenvalue corresponding to X. and G is Green's function (4), Seco102.
Notation used in Probs. 14 to 18: Let N.[X] and N[X]and their derivatives
with respect to the real-valued parameters exand Pdenote these boundary values on the
interval O~ x ~ 1:
N.[X] = X(O)cosex+ X'(O)sin ex, N[X]= X(l)cosP + X'(l)sinp,

N~[X] = -X(O) sin a + X'(O)cosex, Np[X] = -X(l)sinp + X'(1)cosP.
Also, u(x,s) and v(x,s) denote the solutions of these initial-value problems, so arranged
that N.[u] = N[v] = O:
(a) u" - (S2 + q)u = O, u(O,s) = sin ex, u'(O,s)= -cos ex,
(b) v" - (S2 + q)v = O, v(1,s) = sin p, v'(l,s) = -cosp,
and y(x,s) is the solution of the problem
(e) y" - (S2 + q)y = -sf(x),
The following problems display several steps in proving Theorem 4 for the Sturm-
Liouville problem
(d) X"(x) - [A+ q(x)]X(x) = O,
14. Use Green 's function (Probs. 8 and 9, Seco97) to write the solution of the boundary
value problem (c) above in the form
Y(x,s) = N~U][V(X'S) f: f(~)u(~,s)d~ + u(x,s) f f(~)v(~,s)d~ 1

15. Let A. and l. be eigenvalues and normalized eigenfunctions of Probo (d) above
andwrites. = .J A..Notewhy:ts. arezerosofNp[u] and singularpointsofthe function
y (Prob. 14). Recall that (Prob. 8, Seco97) Np[u] = -N Av] and that (Sec. 98) eigen-
functions corresponding to the same eigenvalue are linearly dependent; then show that,
corresponding to Eqs. (7) and (8), Seco102,
4>.(X) 4>.(x)
u(x,:ts.) = - N~[4>.], v(x,:ts.) = - Np[4>.r
16. Verify that (cf. Seco 102 and Probo 15)
N p[u]NM.] =- (S2 - s/) f u(x,s)4>.(x) dx.

Then deduce that :t s. are simple poles of y(x,s) and that the sum of the residues of y
at the two poles :ts. is c.4>,(x)if s. :!:O, and if s. = O, the residue is c.4>.(x).
17. Verify or derive the solution of the initial-value problem
z" - S2Z = q(x)u(x,s), z(O,s) = sin a, z'(O,s)=-cos a,

where u is the solution of Probo (a) above, in the form
z(x,s)= sinacosh sx - !cosasinhsx+

s
!s f.ox q(~)u(~,s)sinhs(x - ~)d~.
Hence show that u satisfies the integral equation
su(x,s) = s sin a cosh sx - cos a sinh sx + f: q(~)u(~,s) sinh s(x - ~) d~.
18. Prove that [ef.condition (6), Seco103,and see Probo 17]

Isu(x,s)le-<7X < 2(1ssin al + Icosal + Icosal)
105 STEADY TEMPERATURE IN A WALL

Let U(x,y) be the steady-state temperatures in a semi-infinite wall bounded
by the planes x = O,x = 1, and y = O(Fig. 101),for which the face x = O
is insulated and surface heat transfer takes place at the face x = 1,whilethe
face y = Ois kept at temperature F(x), so that the boundary value problem
becomes
(1) u xx(x,y)+ Uyy(x,y)= O (O< x < 1, Y > O),
(2) UAO,y)= O, Ux(1,y)= -hU(l,y) (y> O),
(3) U(x,O)= F(x), lim U(x,y) = O (O< x < 1).
y ro
,-
,-O'
-
o U-}1x1 (1,0) Fig.101
The constant h is positive, and the function F is prescribed.

Although the variable y has a semi-infinite range, the problem is not
well adapted to solution by using the Laplace transformation for reasons
indicated in Seco52. The method of separation of variables does apply.
To find all functions of type X(x) Y(y), other than zero itself, that satisfy
all homogeneous conditions in the problem, we first note that, from condition
(1),
X"(x) Y(y) + X(x) Y"(y) = O.
Therefore X"(x) Y"(y)

X(x) = - Y(y) = A,
where the parameter A is independent of x and y (cf. Seco101). Then all the
homogeneous conditions are satisfied if X(x) is a solution of the Sturm-
Liouville problem
(4) X"(x) - AX(X) = O, X'(O)= O, hX(1) + X'(1) = O,
and if Y satisfies the conditions
(5) Y"(y) + AY(y) = O, lim Y(y)= o.
Y"" 00
When A = O,the general solution ofthe differential equation in problem

(4) is X(x) = Ax + B, and this satisfiesthe two boundary conditions in the
problem only if the constants A and B are both zero. Consequently zero is
not an eigenvalue.
When A :F O, the function that satisfies the first two of conditions (4)
can be written
X(x) = e coshx..ji.
It satisfiesthe third condition if
(6) h cosh ..ji + ..ji sinh ..ji = O.
According to Theorem 3, all eigenvalues of problem (4) are real. Ir A > O,
the left-handmember of (6)has a positive value. All eigenvaluesof problem

(4) are therefore negative.
We write A.= -CX2. Then Eq. (6) becomes
h
(7) tan cx= -,
cx
an equation that has an infinite set of roots :!:CXn(n = 1,2,. . .) that can be
approximated graphically. In fact the number CXn is only slightly greater than
mI:when n is large.
The complete set of eigenvalues of problem (4) is therefore
(n = 1,2,. ..),
where the numbers cxnare the positive roots of Eq. (7),and the corresponding
eigenfunctions are
Xn(X) = COS CXnX.
Let Pndenote the norm IIXnll. Then

I
Pn2 =
fo
l 2
COS CXnXdX = _2 1 +
(
sin CXncos
CXn
CXn
and, since CXn

is a root of Eq. (7), we find that
(8) Pn = 2-t( 1 + h2 : cx/r

The orthonormal eigenfunctionsof problem (4)are therefore
1
(9) 4Jn(x)= - COS CXnX (n = 1,2, .. .).
Pn
When A.= -cx/, the solution ofproblem (5) is

Y(y) = cn exp ( - cxny),
where the constant Cnis arbitrary, and therefore
cos cxx
(n = 1,2,...).
X(x)Y(y) = Cn~exp (-cxny)
The sum of any number of these functions also satisfies all homogeneous
conditions in problem (1) to (3); but unless F(x) is a linear combination of
the functions cos CXnX,
no finite sum will satisfy the remaining condition
U(X,O) = F(x) (O < x < 1).
The function represented by the infinite series
<X) C
L
n= 1 Pn
..!!. COS CXnX exp (-cxnY)
formally satisfies the homogeneous conditions. According to Theorem 4,

this function reduces to F(x) when y = O,provided F" is sectionally con-
tinuous, if the numbers Cnare the Fourier constants for F with respect to <fJn;
that is, if
(10) (1 1 (1
Cn= Jo F(x)<fJix)dx = Pn Jo F(x) cos OCnx
dx.
Note that the series used to represent F(x) here,
G() e
(11) F(x) = cos
n=1Pn
...!!. (l(nX (O< x < 1),
is a generalized Fourier series, not the Fourier cosine series for the interval
O<x<1.
The formal solution of our problem is therefore
G() e
(12) U(x,y) = exp ( - IXnY) COS IXnX,
n= 1 Pn
...!!.
where the constants Pnand en are given by formulas (8) and (10).
106 VERIFICATIONOF THE SOLUTION

In the preceding temperature problem F", and hence the prescribed func-
tion F itself, is sectionally continuous (O< x < 1). At each fixed point
x where F is continuous, the Sturm-Liouville series (11), Seco105, therefore
converges to F(x):
G() e
(1) ...!!.COS(l(nX =
n= 1 Pn
F(x) (O < x < 1).
We order the positive numbers (l(nso that (l(n+1 > (l(n'Then the functions
exp (-(I(nY) in our formal solution (12), Seco 105, are bounded, and non-
increasing with respect to n:
0< exp(-(I(nY) ~ 1, exp(-IXn+1Y)~ exp(-(I(nY) (y ~ O).
According to Abel's test (Prob. 6, Seco81), therefore, the series
G() e
(2) (y ~ O,O < x < 1)
n~1 P: cos (l(nXexp (-(I(nY)
converges uniformly with respect to y when y ~ O. Since its terms are con-
tinuous functions of y, its sum U represents a continuous function of y when
y ~ O. Qur formal solution therefore satisfies the condition
U(x, + O) = U(x,O) = F(x) (O< x < 1).
From Eq. (8), Seco105, we seethat Pn > l/fi. Also, a fixed number M
exists such that ICnl/Pn< M. Whenevery ~ Yowhere Yo> O,the absolute
values of the terms in series (2) are less than the constants M exp ( - anyo);
also the absolute values of the deriva ti ves of those terms, of first and second
order, with respect to x and y, are less than Man exp (-anYO) and
M an2 exp ( - anyo), respectively. The series of those constant terms converges,
according to the ratio test, since an - me O as n-oo. -
Hence series (2)
and the series of derivativesof its terms convergeuniformlywith respect to
x and Y over each strip O ~ x ~ 1, Y ~ Yo > O. The terms are continuous
functions, so the series converge to a continuous function U(x,y) and its
continuous partial derivatives.
In particular, U and Ux are continuous functions of x on the interval
O ~ x ~ 1,when Y > O. They satisfy the homogeneous boundary conditions
Ux(1,y) + hU(l,y) = O
because the terms in series (2) satisfy those conditions. The terms also satisfy
Laplace's equation (1), Seco 105; hence the sum U satisfies that equation
when Y > Oand O < x < 1.
FinaIly, we note that the sum U(x,y) satisfies the remaining condition
-
U O and y -
00 because each term satisfies that condition and the re-
mainder after N terms is arbitrarily smaIl in absolute value, uniformly with
respect to x and y(y ~ Yo),when N is sufficientlylarge.
Thus formula (12),Seco105, is established as a solution ofthe boundary
value problem in the temperature function U.
107 SINGULAR EIGENVALUE PROBLEMS'

If the differential equation in an eigenvalue 'problem has a singular point
on the interval, or if the interval is unbounded, the eigenvalue problem is
singular. A requirement of continuity or boundedness of eigenfunctions
replacesa boundary condition at a singular point or at an infinitelydistant
end of the interval. The representation theory for such problems may
paralIel that used for the Sturm-Liouville system, but the singular aspects of
the probems require variations in the development of the theory.2
Some singular problems have discrete eigenvalues like those in Sturm-
Liouville problems. Others have continuous eigenvalues, consisting of aIl
numbers on a bounded or unbounded segment of the axis of reals. When
eigenvalues are continuous, the representation of an arbitrary function in
terms of eigenfunctions takes the form of an integral with respect to A,instead
1 Further details, inc1uding proofs or references to proofs of representation theorems, and
applications, are given in the author's "Fourier Series and Boundary Value Problems," chaps. 6,
8, and 9, 1963.
2 See E. C. Titchmarsh, "Eigenfunction Expansions," 1946.
of a series, of eigenfunction~. Prominent examples of singular problems

follow.: :'
BESSEL'S EQUATlON
An example of an eigenvalue problem with Bessel's differential equation of

index zero, on an interval (O,e),is
[xX'(x)]' - ..txX(x) = O (O < x < 1),
(1)
X(O)= O, X and X' continous (O~ x ~ e).
Here the differentialequation is a Sturm..Liouvilleequation, but r(x) = x
and reO)= O. When we write it in the form
(2) X"(x) + .!.X'(x) - ..tX(x) = O (O< x < e),

x
we see that one of the coefficients in that form is discontinuous at the end
point x = Oof the interval, so that end point is a singular point of the equa-
tion, and the eigenvalue problem is singular.
The solution of Eq. (2) that is continuous, to~her with its derivative,
over the interval O ~ x ~ e, is (Sec. 83) X = J o(x"' - ..t),except for a constant
factor. Thus X(e) = O if Jo(e.::i) = O.As in Seco83, let an(n= 1,2,...)
denote the positive zeros of J oCa).Then the eigenfunctions and eigenvalues
of problem (1) are
(3) Xn = Jo( CXn~)'

The orthogonality of the eigenfunctions on the interval (O,e) with
weight function p(x) = x can be established by the procedure used in the
Sturm-Liouville theory. A representation theorem states that if.f' is sec-
tionally continuous on the interval, the generalized Fourier series of those
eigenfunctions converges to f(x) at each interior point wherefis continuous.
The representation can be written
(4) (O < x < e).
There are similar results if the differential equation in the problem is

Bessel's equation with a positive index v,
(O < x < e),

(xX')' - (..tx + ~)X = O
or when the boundary condition at the end point x = e reads
hX(e) + X'(e) = O (h ~ O).
But if the interval does not contain the origin X = O as an end point or as
an interior point, these eigenvalue problems in Bessel's equation, with a
homogeneous boundary condition at each end of the interval, are non-
singular special cases of the Sturm-Liouville problem.
LEGENDRE'S EQUATION
The eigenvalue problem
(5)
[(1 - X2)X'(X)]' - AX(X) = O (-1 < x < 1),
X and X' continuous on the closed interval - 1 ;;;; x ;;;; 1,
is another example of a singular problem. Legendre's differential equation
here has singular points at both end points x = :!: 1 of the interval. Its solu-
tion satisfies the continuity condition only if A = - n(n+ 1),wheren =
O,1,2, .... Corresponding to those eigenvalues, the eigenfunctions are the
Legendre polynomials
(6) )- 1
P( x--L.- ~ (-1}i (2n - 2j)!
x
n-2j (O!= 1)
n 2nj=O j! (n - 2j)!(n - j)!
wherem = tn ifn is even and m = t(n - 1)ifn is odd. They form an ortho-
gonal set with weight function p(x) = 1 on the interval (-1,1). The repre-
sentation of a functionj on that interval, such thatf' issectionallycontinuous,
as a generalized Fourier series of the eigenfunctions, can be established
rather easily by using properties of Pn. It can be written
(7) (-1 < x < 1).
FOURIER INTEGRALS
The singular eigenvalue problem

(8) X" - AX = O, x> O; X'(O) = O, IX(x)1< M
(O< x < ex),
where M is some constant, is one in which the interval is unbounded. The
solution of the differential equation that satisfies the condition X'(O) = O
is, except for a constant factor,X = cos x~, for any value of A.,including
zero. That solution is bounded over the half line x > O if and only if A is
real and A ;;;;O. We write A = - rx2,where rx~ O. Then the eigenfunctions
and eigenvalues are
(9) X(x) = cosrxx, (rx~ O).
In this example the eigenvalues are continuous, A ~ O, rather than

discreteo The eigenfunctions lack an orthogonality property, but a repre-
sentation theorem follows from the Fourier integral formula (Sec. 68) when
the function G there is even. Ir!, is sectionally continuous on each bounded
interval of the half line x > Oand iflis absolutely integrable over that half
line, then
2 '" '"
(10) I(x) = -
1t o f
cos ax
f.o
IR) cos <X~d~da (x ~ O).
Thus lis represented as an integral of the eigenfunctions cos <xx,with respect

to a, by the Fourier cosine integral formula
2 '"
A(a) = - I(~) cos <X~
d~ (x ~ O).
1tf.o
Further examples of singular eigenvalue problems will be cited in the
problems below and in the following chapters.
PROBLEMS
Apply the method of separation of variables to the boundary value problems in partial
differential equations that follow.
1. The steady-state temperature problem (cf. Seco105)
u ",,(x,y) + Uyy(x,y) = O, IU(x,y)1< M, (O< x < 1,Y > O)
U x(O,y) = U x~,y) = O, U(x,O)= F(x) (O< x < 1),
where M is some constant. Show that A.= Ois an eigenvalue of the associated Sturm-
Liouville problem and obtain the formal solution in the form
1 "" 1
U(x,y) = 1o F@d~ + 2 ft~1e-m'Ycosn7tx 1o F@cosn7t~d~.

Note that as y -+ 00, U tends to the mean value of F.
2. The problem in Seco90 on displacements Y(x,t) in a string when Y(x,O) = g(x),
,. Y,(x,O)= Y(O,t)= Y(c,t) = O.
3. Probo 1, Seco93, in the longitudinal displacements Y(x,t) of a stretched bar, where
Y(x,O)= Ax, Y,(x,O)= Y(O,t)= Yx(l,t) = O.
4. Prob.4, Seco93, for displacements Y(x,t) in a string with initial velocity Y,(x,O)=
g(x),O < x < C.
5. (a) The problem in Seco82 on temperatures U(x,t) in a bar when U(x,O)= g(x),
O < x < 1, U(O,t)= U(l,t) = O,and k = 1.
(b) When g' is sectionally continuous, verify the solution of the boundary value
problem in part (a) using the series representation
"" 1
U(x,t) = 2 ft~1exp(-n27t2t)sinn7tx 1o g(~)sinn7t~d~.

6. Give a physical interpretation of U(x,y) in the problem

u x(O,y)= U(l,y) = O, -KU,(x,O) = F(x),
where U is bounded over the region O < x < 1,Y > O,and K > O. Write m = (n - t)1t
and derive the solution in the form
l
-2 00
U(x,y)= K L -e-my cosmx

n= I m
1
fo
F(~)cosm~d~.
7. Let U(x,y) be the steady-state temperatures in a thin plate in the shape of a semi-
infinite strip, from which heat is transferred at the faces into a medium at temperature
zero, so that
, Uxx+U,,-bU=O (O< x < 1,Y > O,b > O).

If U is bounded and satisfies the conditions
U(O,y) = O, Ux(1,y) = - hU(l,y), U(x,O)= 1 (O< x < 1).

and h > O,derive the formula
00 A
U(x,y) = 2h L
n= 1 an
--!! exp [ - y(b + (Xn2)*] sin (XnX (b > O, h > O),
where An = (1 - cos (Xn)/(h+ cos2 (Xn)and (XI' (X2,. .. are the positive roots of the
equation tan (X= -(X/h.
8. Let U(x,y)be the steady'state temperaturesin an infiniteprism bounded by the
planesx = O,Y = O,x = 1,andy = 1. If U = 1onthefacey = 1andifUx = - hU at
x = 1,whereh > O,and U = Oon the other two faces(Fig.102),derivethe formula
where the numbers An and (Xnare those described in Probo 7.
Y/ U=1 0,1)
Fig.102
U=O
0
O U=O--X
-
=
-O.
.-
9. Writem = (n - t)1t and obtain the solution of the problem

(t + l)U,(x,t) = Uxx(x,t) (O< x < 1, t > O),
U(O,t)= Ux(1,t)= O, U(x,O)= 1 (O< x < 1),
in the form
U(x,t)= 2 f
..!..(t+ Wm2 sin mx.
n=1m
Verify that this function satisfies the condition U(x,+O) = 1(0 < x < 1).
Fig.103
10. The end x = 1 of a stretched string is elastically supported (Fig. 103)so that the
transverse displacement Y(x, t) satisfies the condition y"(l,t) = -hY(1,t), where h > O.
Also, let Y satisfy the conditions
Y(O,t)= O, Y(x,O) = bx, y'(x,O) = O.
Show that the solution of the equation y',(x,t) = y"x(x,t)is then

'" sin IX"sin 1X"x
Y(x,t) = 2bh(h+ 1).=L1 IX.2(h + cos2 IX. \ COSlX.t,
where 1X1,1X2,. . . are the positive roots of the equation tan IX= -IX/h.
11. The longitudinal displacements Y(x,t) in a certain nonuniform bar satisfy the
conditions
(O< x < 1, t > O),
Y(O,t)= Y(l,t) = y'(x,O)= O, Y(x,O) = F(x) (O< x < 1).

Derive the formula
ao
Y(x,t) = L c.l/J.(x)cos P.t,

n=1
where P. = n7te/(e - 1) and
l/J.(x)= j!jsin [P.(1 - e-X)],

12. The electrostatic potential V(r,z) in the space bounded by the cylinders r = a and
r = b above the plane z = O satisfies Laplace's equation (rV.)r+ rv.. = O and the
boundary conditions
V(a,z) = V(b,z) = O (z > O),
V(r,O)= F(r) (a<r<b);
also V(r,z) is bounded. Derive the formula

ao
V(r,z) = L A.
11=1
exp (-IX.z)t/!(rx.,r),
- Jo(aa)Yo(lXr).Jo and Yo are Bessel functions of the first

where t/!(ex,r)= Jo(rxr)Yo(aa)
and second kind, linearly independent solutions of Bessel's equation. The numbers IX.
are the roots ofthe equation tJ(a,b) = O,all reaP am > O,and An are given by theformula
An f r[tJ(an,r)]2 dr = f rF(r)tJ(a.,r) dr.
13. The temperature V(r,t) in a solid cylinder r ~ 1 of infinite length satisfies the
conditions
(O~ r < 1,t > O), V(r,O)= 1, V(l,t) = O;
also V and V. are continuous when O ~ r ~ 1 and t > O. Show that the method of
separation of variables involves a singular eigenvalue problem of type (1), Seco 107.
With the aid of the integration formula
[XJo(X)dX = cJ(c),
solve for V in the form
V(r,t) = 2 L.. - J
~ J o(a.r)
exp(- a. t), 2
.= a. (a.)
where the a;s are the positive zeros of J o(a).
14. The temperature U(x,t) in a semi-infinite solid x ~ O, whose initial temperature
function is a step function, satisfies the conditions
u,= kUxx; U(x,O) = 1 - So(x - 1),x> O; U x(O,t) = O,
and U is bounded. Show that the solution by separation of variables involves the
singular eigenvalue problem (8), Seco107, and derive the formula
2 "" sin a
U(x,t) = - -cosaxexp (-a2kt) da.
7t1o a
15. Prove that the eigenvalues and eigenfunctions of the singular problem
X" - J.X = O,O < x < co; X(O) = O; IX(x)1~ l,x > O;
are J. = -a2(a > O) and X = sin ax. If l' is sectionally continuous and absolutely
integrable over the half line x > O,note that, according to the Fourier integral formula
(Sec. 68) for odd functions, f is represented in terms of those eigenfunctions by the
F ourier sine integral formula
2 "" ""
f(x) = - 1 sinax 1 f(~) sin a~d~ (x > O).
7t o o
16. Use the results found in Probo 15 to solve the problem in Seco43 for the displace-
ments Y(x, t) of an initially displaced semi-infinite stretched string, namely,
x > O,t > O; Y(x,O) = (x),

x > O;
Y;(x,O) = Y(O,t)= O, IY(x,t)1 < M, 1(x)1 < M,
I See H. S. Carslaw and J. C. Jaeger, "Conduction of Heat in Solids," 2d ed., p. 206, 1959, and
their tables, p. 493.
in the form
y = ~ (00[sin ex(x+ at) + sin ex(x- at)] (00 cI>(~) sin ex~ d~ dex.
nJo Jo
Show how this reduces to the form (8), Seco 43.
17. Prove that the eigenfunctions of the singular problem
X"(x) - AX(X) = O, IX(x)1~ 1 (-00 < x < (0)

are X = cosex(x- ~), where A = -exz and ex~ Oand the parameter ~ has any real
value. Note that accordingto the Fourier integralformula(2),Seco68,
oo a>
f(x) =-
1
n o if -00
f(~) cos ex(x- ~)d~dex (-00 < x < (0).
18. Note that the first few Legendre polynomials are

Po(x) = 1, P1(x) = x, Pz(x) = !(3xZ - 1), P3(x) = !(5X3 - 3x).
Verify that they satisfy Legendre's equation in problem (5),Seco107,when A = -n(n +
1),n = 0,1,2,3.
19. Legendre's polynomials have the property
p.(-x) = (-l)"P.(x) (n = O, 1,2, . . .),
so PZnand PZn+lare even and odd functions, respectively. Deduce that X = PZ.+l(X)
and A = - 2(2n + l)(n + 1)are eigenfunctions and eigenvalues ofthis singular problem
on the interval (0,1):
[(1 - XZ)X')' - AX = 0,0 < x < 1; X(O)= O,
where X and X' are continuous on the c10sed interval O ~ x ~ 1.
20. If the condition X(O) = O is replaced by the condition X'(O) = O in Probo 19, show
that A = -2n(2n + 1) are eigenvalues and X = pz.(x) are eigenfunctions.
10
General Integral Transforms
108 LINEAR INTEGRAL TRANSFORMATIONS

Let (a,b) denote a bounded or unbounded interval of the x axis and let
K(x,A.) be a prescribed function of x on that interval and of a parameter A..
Then, as we pointed out in Seco 1, a general linear integral transformation
of functions F on the interval is
(1)
T{F} = f F(x)K(x)) dx = f(A.).
The resulting function f(A.) is the transform of F under that transformation
with kernel K.
Of course the class of functions from which F and K are chosen must
be restricted and a set A of values of the parameter A.must be specified so
as to ensure not only the existence ofthe integral (1) but also the uniqueness
of the inverse transformo That is, not more than one function F should
correspond to a given transform f(A.).
317
The class of functions, the function space, to which F belongs is to be a

linear space in order that AF(x) + BG(x) will belong to that space whenever
F and G belong to it, for each pair of constants A and B. Then T is a linear
transformation :
(2) T{AF + BG} = AT{F} + BT{G} = Af().) + Bg().).
Ir inverse transforms T-1 {J} are unique, the linearity condition (2) can be
written
(3)
that is, the inverse transformation is also linear.

When a = O,b = 00, and K(x,).) = e-J.x, the transformation T is the
Laplace transformation L{ F} = f().). In that case we noted that the function
space of F may consist of all real-valued functions of exponential order on
the half line x > O which are sectionally continuous on each bounded
interval, and the set A of values of)' may be all complex numbers in a right
half planeo We restricted the function space somewhat further in order to
ensure uniqueness of L - 1 {f}. Also, we found a general formula, the inver-
sion integral, for the inverse Laplace transformation.
We shall soon indicate how the transformation T may be specified
so that it transforms a given linear differential form in F into an algebraic
form in f().), ). and prescribed boundary values of F at the ends of the
interval (a,b). That gives the basic operational property of T for the reduc-
tion of boundary value problems in differential equations. As in the case
of the Laplace transformation, where that operational property is
L{ F'} = ).f().) - F(O),some further restrictions on the space of the functions
F are needed to ensure the validity of the property.
The operational calculus of an integral transformation becomes more
effective when its basic operational property and its inversion process are
supplemented by other operational properties and tables of transforms.
A convolution property of T is one that gives a representation of
T-1 {J().)g().)} directly in terms of the functions F and G whose transforms
are f and g. The representation is often simpler than one found by applying
a formula for inverse transformations. In some cases the property is modified
byreplacingfg bya weighted product oftransforms to give T-1{w().)f().)g().)},
where the weight function w is fixed for a given transformation. We ha ve
seen that the convolution is an important property of the Laplace trans-
formation. Convolution properties are also prominent operational pro-
perties ofvarious Fourier transformations. But for some ofthe less common
integral transformations such properties are either unknown or so com-
plicated that they are tedious to use.
GENERAL INTEGRAL TRANSFORMS [SECo 109 319
109 "KERNEL-PRODUCT AND CONVOLUTION PROPERTIES
For the Laplace transformation the kemel K(x) = e-.I.x has the simple
product property
(1) K(x,).)K(xO,A) = K(x + Xo, A).
Then if f(A) = L{F} and F(x) = O when x < O,it followedthat L has the
translation property (Sec. 11)
(2) f(A)K(xO,A) = L{ F(x - xo)} (xo > O).
This formula represents the product of the transform of F by the kemel as
the transform of a function related to F. We call it a kernel-product property
of the Laplace transformation. That property led us to the convolution
property (Sec. 16)
(3) f(A)g(A)= L{S; F(y)G(x - Y)dY} = L{F* G},

wheref(A) = L{F} and g(A) = L{G}.
Now, in the case of another transformation of the type
(4)
T{F} = f F(x)K(x,A) dx = f(A),
suppose we have a modified or weighted kemel-product property such that
for some specifiedfunction W(A)and for each function F there is a function
PF(x,xo) for which
(5) (a < xo < b).
Then a modified convolution property follows, formally.

For ifJ(A) = T{F} and g(A)= T{G}, we can write
(6)
f(A)g(A)W(A) = f g(A)w(A)K(y,A)F(y) dy
= f T{PG(x,y)}F(y)dy
= ff PG(x,y)K(X,A)dx F(y) dy.
When the order of integration is reversed, the iterated integral becomes
f K(X,A) f F(y)PG(x,y) dy d~
which is the transform of the function
(7)
XFG(X) = f F(y)PG(x,y)dy.
We call the function X FG the convolutionof F and G for the trans-
formation T, with respect to the weight function w. It is the inverse
transform of the product fgw. That is,
(8) f(A)g(A)W(A) = T{XFG(x)} = T{XGF(x)},
where the last form T {XGF}follows by interchanging the functions F and

G, and their transforms, in Eqs. (6).
Thus we have indicated how a convolution property (8)follows from a
kernel-product property (5). The convolution is described by Eq. (7) in
terms ofthe function PFin property (5). An example is given in the following
section. Other examples will be found in the problems and in the chapters
that follow. A corresponding result for a generalized convolution property
for iterated transforms of functions H(x,y) will also be noted in the problems
(Prob. 8, Seco 110).
Ir we replace F in the transformation (4) by the delta symbol, we ~nd,
formally,
(9) if a < Xo < b.
When F is replaced by l5(x - xo), the convolution (7) becomes
(10) X.JG(x)= PG(x,xo) [15= l5(x - xo), a < Xo < b],

and the convolution property (8) reverts to our kernel-product property
written in terms of g and G:
(a < Xo < b).
110 EXAMPLE
We illustrate the results in the preceding section by deriving a convolution
formula for the Fourier sine transformation on the half line x > O,
(1) (A > O).

SA{F} = 1'" F(x) sin AXdx = f.(A)
Other operational properties of SA will be developed in Seco116and Chap. 13.
We assume that our functions F are sectionally continuous on each bounded
interval and absolutely integrable over the half line x > O. It is convenient
to introduce the odd extension Fl ofF defined over the fullline - 00 < x < 00:
(2) Fl(X) = F(x) if x> O, Fl ( - x) = - Fl (x) for all X.
We introduce also the continuous function
(3)
H(x) = foo Ft(~)d~ = foo Ft(~)d~ + I: Ft(~)d~
(-00 < x < (0)
and note that it is even, H( - x) = H(x) or H(lxl) = H(x),and that it vanishes
as x --+ :t 00; also, H'(x) = Ft(x).
To find a icernel-product property of SJ..,we write
oo
2 1
lJ.<A) =l f -00 Ft(x) sin AXdx = f oo
-00
sin AX
x-H'(x)dx
00 oo
sin AX
[
= -H(x)
A ] f
-00-00
- H(x)cos AXdx.
The bracketed term here vanishes. Thus when Xo > O,
~.f.(A)sin AXo = - f:oo H(x) cos AXsin A.xodx
= t f:oo H(x)[sin A(X - xo) - sin A(X + xo)] dx'
= t f: 00 [H(y + xo) - HOy - xol)] sin AYdy.
The integrand of the final integral is an even function of y. Thus

oo Y+xO
2
l.f.(A) sin AXo = f f
o sin AY Iy-xolFt(~) d~dy.
Since F1(~) = F(~) when ~ > O, a kernel-product property for the
transformation SJ..can be written
(4) (Xo > O).
In terms of the notation used in formula (5), Seco109, then

2
(5)
m(A)= I
when K(X,A) = sin AX, a = O,b = 00, and T is SJ...
The corresponding convolution of two functions F and G is
oo oo X+Y
(6) XFG(X) = fo
F(y)PG(x,y)dy =
f
o
F(y)
fIx-yl G(~)d~dy,
according to formula (7), Seco 109. The convolution property reads
(oo F(y) {y+x G(~)d~dY (A.> O),

(7) fs(A.)g.(A.)~= SA{ Jo J,y-xl }
wherefs(A.) = SA{F} and g.(A.) = SA{G}.
ILLUSTRATION OF FORMULA (7)
In the particular case

(8) F(x) = G(x) = e-ax (o > O),
an elementary integration shows that
A.
(9) fs(A.) = g.(A.) =- (A.> O);
x+y 1
also, PG(x,y) =
f.Ix-yl e-a~ d~ = -[e-alx-yl
o
- e-a(x+y)].
Then the convolution (6) becomes
1 1
--
o o f oo
e-aYe-a(x+y)dy = -xe-ax.
o
Thus we have this useful Fourier sine transformation
(10) (o > O).
PROBLEMS
1. Use the convolution property (7), Seco110,when
F(x) = e-ax, G(x) = e-bx (a > 0, b > 0, a '# b)
to obtain this Fourier sine transformation:
(a> O,b > O,a '# b).
(This result can be found also by using partial fractions.) Evaluate the integral trans-
formation on the right to verify the result.
2. Interchange the order of integration in formula (6), Seco 110, to show that the
convolution X FGfor the Fourier sine transformation SA is symmetric in F and G:
XFG(x) = XGP(x).
GENERAL INTEGRAL TRANSFORMS 323
[SECo 110.
3. The exponentialFourier transformationof functions F on the entre x a~is is

(Chap 12)
(-00 <). < 00).

EA{F} = f:", F(x)e-UX dx = f.().)
We consider functions F that are sectionalIy continuous on each bounded interval,

absolutely integrable from - 00 to 00 and bounded. Show that EA has the kernel-
product property
f.().)exp(-iho) = EA{F(x - xo)} ( - 00 < Xo < 00)
and, formalIy, that the corresponding convolution is
XFG(x) = f: F(Y)G(x - y)dy = XGF(x),
so that if ge().)= EA{G}, then
f.().)ge().)= EA{XFG}= EA{f:", F(y)G(x - y)dY}.

4. Let So(x) be the unit step function: So(x) = Owhen x < O,So(x) = 1 when x > O.
Note that the exponential Fourier transforms of the two functions
F(x) = e-XSo(x), G(x) = ~So( -x) (- 00 < x < 00)
are f.().) = (1 + i).)-1 and ge().)= (1 - i).)-I. Use the convolution property of EA in
Probo3 to derivethis transformation:
).2
1
+ 1-
- E {1 -lxl }
A '[e .
Verifythe result by evaluatingthe integral EA{e-lxl}.
5. When K(x,).)= cos ).x, a = O, and b = 00, our transformation T becomes the
Fourier cosine transformation eA on the half line (Chap. 13),
(). > O).

eA{F} = L'" F(x) cos h dx = fc().)
Let the functions F be sectionalIy continuous on each bounded interval and absolutely
integrableover the half line. If F2 is the even extension of F, so that F2(X)= F(lxl)
-
when 00 < x < 00,then
fc().)cos ho = t L"'", F2(x) COSh cos ho dx.

Show that a kernel-product property of eA is
fc().)cos).xo = eA{t[F(x + xo) + F(lx - xol)]} (xo > O);
then that the corresponding convolution property is
fc().)gc().) = eA{ t 50'"F(y)[G(x+ y) + G(lx- yl)] dY};

that is, the convolution for C when W(A.)= 1 is
XFG(x) = t f' F(y)[G(x + y) + G(lx - yl)]dy.

6. Prove that the convolution for C found in Probo 5 is symmetric in F and G:
XFG(x) = XGF(x).
7. Let F be the step function 1 - So(x - k), where k> O,and let G be e-ax (a > O).
If C is the cosine transformation in Probo 5, then
C{F} = sinA.H '
Show that the convolution of F and G for C is the function

XFG(x) = a-l(1 - e-akcoshax) when x ;;;k
= a-le-axsinhak when x ~ k,
so that we have the transformation
asinH = C{XFG(x)}.
8. Generalized convolution For a function H(x,y) of two variables, let h(A.,y)denote
the transform T{H} with respect to x,
h(A.,y)= f H(x,y)K(x,A.)dx (a < y < b),

and let h(A.)denote the transform of h(A.,y)with respect to y when the same parameter A.
is used,
. h(A.)= f h(A.,y)K(y,A.)dy,
if H and T are such that this iterated transform exists.
In case T has a kernel-product property
h(A.,y)w(A.)K(y,A.) = f QH(x,y)K(x,A.) dx,
show formally that it has the generalized convolution property
h(A.)w(A.) = f [f QH(X,y)dY]K(X,A.)dX.
Note that this reduces to property (8), Seco 109, in case H(x,y) = F(x)G(Y); then
QH = PF(x,y)G(y). Examples are given in Probs. 9 and 10.
9. When the transformation in Probo 8 is the exponential Fourier transformatin E
describedin Probo3 and w(A.)= 1, show that the generalizedconvolution property
becomes
where Ji(l) = f: e-iAv I-OOXJH(x,y)e-iAx dx dy.

10. Ir H(x,y) = 1 when x > O and y > O and x + y < 1, while H(x,y) = O for all
other real x and y, show that the iterated exponential Fourier transfonn of H is
Ji(l) = (1 + iA)e-iA
l - 1
and verify that this agrees with the convolution property in Probo 9.
111 STURM-LlOUVILLE TRANSFORMS
We are now ready to present a method of determining a linear integral

transformation T that simplifies a given linear boundary value problem in
a function F. The problem is to be such that in the differential equation the
differential form in F with respect to some one of the independent variables
x can be isolated and, in the boundary conditions, boundary values with
respect to x can be isolated. The transformation T with respect to x is to
be designed so as to replace that differential form by an algebraic form in
the transform T {F} and the prescribed boundary values on F at the ends
of the interval over which the variable x ranges.
We begin with an important case that can be treated with the aid of
the Sturm-Liouville theory in Chap. 9.
Let the differential form with respect to x be of order 2,
(1) D2F = A(x)F"(x) + B(x)F'(x) + C(x)F(x) (a < x < b),
and let the interval (a,b) be bounded. For the sake of simplicity we assume
that aIl functions of x here, together with their derivatives of {he first and
second order, are continuous over the cIosed interval a ~ x ~ b, although
those conditions can be relaxed. Also, we assume that A(x) > O over that
cIosed interval. In applications to problems in partial differential equations,
F will be a function of x and the remaining independent variables, while
A, B, and e will be functions of x alone.
The boundary values with respect to x, to be prescribed as functions
of the remaining variables, are
(2) NaF = F(a) cos ex+ F'(a) sin ex,
(3) N (iF = F(b) cos /3 + F'(b) sin /3,
where the constants exand P are real.

In order to design a transformation
T{F} = f F(x)K(x,A)dx = J(A)

that transforms D2F into an algebraic form in f(A), NaF, and NpF, it is
convenient to write D2F in terms of a self-adjoint differential form
(4) !!)2F = [r(x)F'(x)]' - q(x)F(x).
We do that by writing D2F = Ar-1(rF" + A-1BrF') + CF, where

.
B()
]
X
(5)
r(x) = exp [Ia A() d(
so that r' = A -1 Br, and introducing the functions
r(x)
(6) q(x) = - p(X)C(X).
p(x) = A(x)'
Then
1
(7) D2F= _ !!)2F= ![(rF')' - qF].
px( ) . p
Note that r(a) = 1 here. Corresponding representations in terms of
self-adjoint forms are not always possible for forms of order higher than 2'-
In such cases the original linear differential form may be used; then the
adjoint of that form arises in the process of designing T.
We now write the kernel of our transformation T as
K(x,A.) = p(X)eI>(X,A)
where the weight function p is rlA, and the kernel el>and the range of the
parameter A are yet to be determined. Then the transform of D2F becomes
f~eI>!!)2F
dx and, either by integration by parts or by using Lagrange's
identity (5), Seco94, according to which
we can write
(8)
T{D2F} = f F(X)!!)2e1>(X,A)dx + [(eI>F' - eI>'F)rJ:.
In order to express F(a) and F'(a) in terms of the prescribed boundary

value NaF, we introduce the derivative of NaF with respect to a,
N~F = - F(a) sin a + F'(a) cos a,
and solve this equation simultaneously with Eq. (2) to get
F(a) = NaFcos a - N~Fsin a, F'(a) = N~Fcos a + NaFsin a.
At the lower limit the bracketed term in Eq. (8) can now be written
[eI>(a,A.)F'(a)
- eI>'(a,A.)F(a)Jr(a)
= -NaFN~eI>+ N~FNael>.
The value of this term is determined by NaF and $ if N/1> = O.

Likewise at x = b we find that
($F' - $'F)r]x=b= -r(b)NpFN'p$

if N p$ = O, where
N'p$ = - $(b).) sin P + $'(b,A)cosp.
The integral in Eq. (8) becomes ,if(A) if $ satisfies the Sturm-Liouville
equation ~2$ = Ap$. Hence we have the desired transformation
(9)
provided that $ is a non trivial solution of the Sturm-Liouville problem

(10) (r$')' - (q + Ap)$ = O, Np$=O.
Since problem (10) has nontrivial solutions when and only when A is an
eigenvalueAn,the kernel $ must be an eigenfunction$n corresponding to
the eigenvalueAn;then
(11)
T{F} = f F(x)p(x)$n(x) dx = f(An)
(n = 1,2,...).
We call this the Sturm-Liouville transformation. Its parameter A ranges

through the discrete real values A.l,A.2,. . ., the set of all the eigenvalues
of the Sturm-Liouville problem (10). The transformf(An) is also an infinite
sequence of numbers.
The basic operational property (9) of T now reads
(12)
112 INVERSE TRANSFORMS
We select some convenient complete set of real-valued eigenfunctions $n,

not necessarily normalized, as the kernel in our Sturm-Liouville transforma-
tion (11) above. The generalized Fourier series for a function F on the
interval (a,b) can be written
ex>$n(X)
n~l II$nll lb
a F(~)p(~)
$n(~)
11$n 11d~ =
ex> $n(X)
n~l II$nlld(An),
where II$nIl2= s: p(~)[$n(~)F d~.

When F is of class C", that is, when F, F', and F" are continuous over
the closed interval a ~ x ~ b, that series converges to F(x) over the open
interval. Thus the series represents F in terms of its transform f(An).
A formula for the inverse Sturm-Liouville transform off(An) is therefore

co
(1) F(x) = L f(AnWl>nll-
n= 1 .
zcI>n(x)= T-1{f(An}} (a < x < b).
Uniqueness of inverse transforms among functions of class C" follows

from that inversion formula. For if two such functions F and H have
identical transforms, then T{F - H} = O, and formula (1) shows that
F(x) - H(x) == O.
Alternate ways of tinding inverse transforms are furnished by tables of
transforms, by devices for representingf(An) in terms of known transforms,
or by operational properties of T. The transform of any particular eigen-
function cI>m(x),where m is some positive integer, can be written at once
from the detinition of T and the orthogonality of the eigenfunctions. Thus
(2) if n :/=m, if n = m;
that is, the transform <Pm(An)
is the sequence O,O,..., IIcI>mllz,
O,O,....
We present now an operational property that gives T-l{f(An)/An} in
terms of the function F and a solution of a differential equation associated
with the Sturm-Liouville transformation T. We write Y(An) = f(An)/An if
An :/=O; then
(3)
Suppose tirst that T is such that no eigenvalue Anis zero. According

to our basic operational property, Eq. (3) is true if Y(An)= T{Y} and Y
is a solution of class C" of the boundary value problem
(4) Dz[Y(x)] = F(x), N"Y = NpY = O.
The notation is that used in Seco111; thus ~z Y = pF since
"" Y -- (r Y'Y
Dz Y -- P - 1 JJz - qy
.
p
The homogeneous problem corresponding to problem (4) is the Sturm-
Liouville problem in which A = O; ~z y = O,N"Y = N p Y = O. Since zero
is not an eigenvalue, that homogeneous problem has only the solution
Y(x) == O. Therefore Green's function G(x,~)for problem (4) exists,where
(rGx)x - qG =O (x :/=~), N"G = NpG = O, (a < ~ < b)
and Gx has the jump l/r(~) at the point x =~. The unique solution of
problem (4) is then
Y(x) = f F(~)p(~)G(x,~)d~
(a ~ x ~ b).
Thus when zero is not an eigenvalue of the Sturm-Liouville problem

associated with T, then
(5) T-I {fi~n)}= f F(~)p(~)G(x,~) d~;

that is, T {Y} = f()"n)/Anwhere Y is the solutionof problem(4).
1f one of the eigenvalues is zero, say Al = O,then AnY(An)= f(An) when
n = 2,3,... ; that is, if NaY = NpY = O and Y is of cIass C", then
T{Dz[Y] - F} = O when n = 2, 3, ... .
According to transformation (2), then, for some constant k, Y satisfies the
conditions
Dz[Y(x)] - F(x) = k<l>I(X),
Therefore AnY(An) - f(An) = kT{} and, when n = 1, it follows that

-f(O) = kll<l>lllz.
Thus Y, the inverse transform of f(An)/An, n i= 1, with f(AI)/AI undefined,
is a solution of class C" of the problem
(6) NaY= NpY= O
(Al = O),
if that problem has such a solution. Recall, however, that Green's function
does not exist for that problem, nor does the problem have a unique solution,
since A <1>
1(x) is a solution of the homogeneous form of the problem when A
is an arbitrary constant. Thus if Y = Z(x) is one solution of problem (6),
then Y = Z(x) + AI(X)is a family of solutions.
113 FURTHER PROPERTIES
Until special cases of Sturm-Liouville transformations are treated, cases in

which the kernels are fairly elementary functions, only a few operational
properties and transforms of particular functions can be presented.
In the general case, however, the basic operational property can be
extended easily to iterates of the differential form DzF. Let the function
DzF as well as F itself belong to the class C"(a ~ x ~ b). Then
Let D/ denote the iterated differential operator here. Then
(1) T{D/F} = A/f(An) + AnNiF]N~" - A"r(b)Np[F]Np<l>"
- r(b)Np[DzF]Np<l>".
+ Na[DzF]N~"
Thus T replaces the differential form D22F of fourth order by an algebraic

form in the transform j()"n) of F and the four boundary values NaF,
Na[D2F], NpF, and Np[D2F].
Further iterations of D2 can be treated in like manner.
Our transformation was developed in Seco 111 by expressing the
differential operator D2 in terms of a self-adjoint operator ~2' a procedure
that can be followed for operators of the second order. To proceed directly
without introducing the self-adjoint operator, we first write, as before,
T {F} = f F(x)K(x,A) dx = f(A)

and D2F = A(x)F"(x) + B(x)F'(x) + C(x)F(x). Then the transform
T{D2F} = fK(X,A)D2[F(X)] dx
is to be written in terms Ofj(A) and the two boundary values NaF and NpF.
In terms of the adjoint D! of D2, Lagrange's identity (3), Seco94, shows that
d
KD2F - FD!K = d)AKF' - (AK)'F + BKF]
where r(x) = exp f: [B(~)/AR)] d~, and r' = rB/A. Therefore when F and K
are of class C",
(2)
The remaining steps correspond to those used in Seco 111, where we

expressed the values of F and F' at the end points x = a and x = b in terms
of NaF, N~F, NpF, and N'pF. Details are left to the problems. We find that
Eq. (2) can be written as the basic operational property
(3)
provided that K = Kn(x) and A = An,where Kn and Anare the characteristic

functions and numbers of the Sturm-Liouville problem
(4) D!K - AK = O,
Note that
Na[ A:] = {AK cos a + [(AK)' - BK] sin a}x=a,

(5)
N{ A:] = [r(b)r1{AKcos P + [(AK)' - BK] sinP}x=b'
Then the Sturm-Liouville transformation has the form
(6) (n = 1,2,...).
Formulas (3) to (6) follow also from those in Seco 111 when the latter
are written in terms of the kernel Kn and the coefficients A, B, and e in D2,
where Kn = pCl>n.
The basic operational formula for T{D2F} must be modified in case
F or F' has jumps. Suppose, for instance, that both functions are eontinuous
over the interval a ;;;; x ;;;;b exeept at one interior point x = e where they
have jumps jc and j~:
jc = F(e + O)- F(e - O), j~ = F'(e + O)- F'(c - O)

(a < e < b),
and that F" is sectionally continuous. Then if we write
T{D2F} = f Cl>Q2F dx + f Cl>Q2F dx
and use Lagrange's identity, we find that

(7) T{D2F} = Anf(An) + N~[CI>n]NaF- r(b)Np[CI>n]NpF
+ r(c)[jcCl>~(e)- j~CI>n(e)],
where T{F} = f~FpCl>ndx and Cl>nand An are the characteristic functions and
numbers of the Sturm-Liouville problem (10), Seco111.
We cite one more property of Sturm-Liouville transforms f(An).
When F is sectionally continuous, its Fourier constants
en = IICI>nll-l f F(x)p(x)CI>n(x)dx
(n = 1, 2, . .. )
with respect to the orthonormal set of functions Cl>n/llCl>nll

on the interval
(a,b) tend to zero as n -+ oo. That is a consequence of Bessel's inequality,
C12+ e/ + ... + Cm2;;;;IIFII2 (m = 1,2,...),
for orthonormal sets.l Thus the transform of each sectionally continuous

function F must satisfy the condition
(8)
114 TRANSFORMS OF CERTAIN FUNCTIONS

As noted in Seco 112, for a fixed positive integer m the Sturm-Liouville
transform of the kernel $m is zero when n "' m, and JI$mJl2when n = m.
Thus if f()on) is the transform of a function F and k is a constant,
when n "' m,
(1)
when n = m.
Suppose Tis such that no eigenvalue Anis zero. Then Green's function
G(x,c)for problem (4), Seco112,exists. It is continuous, but its derivative
Gx has thejumpj~ = [r(c)r 1at the point x = c (a < c < x); also,
(2) (x"' c),
According to the modified operational property (7), Seco113, then
0= AnT{G(x,c)} - r(c)[r(c)r1$n(c).
Thus the Sturm-Liouville transform of G(x,c) is
1
(3) (a < c < b).
T{G(x,c)} = :f$n(c)
n
The continuity of G(x,c)and the sectional continuity of its first deriva-
tive are sufficient (Theorem 4, Seco 99) for the validity of our inversion
formula (1), Seco 112. Therefore we have the representation
(4) (a < x < b,a < c < b),
known as the bilinearformula for Green's function.

From transform (3) and property (5), Seco112, we find another Sturm-
Liouville transform when no eigenvalue is zero :
(a < c < b).

(5) T{f P(~)G(X,~)G(~'C)d~}= A:2$n(C)
A further consequenceof formula (3)is
(a < c < b).
(6) T{:cG(X,C)} = ;n$~(C)
I Churchill, R. V., "Fourier Series and Boundary Value Problems," 2d ed., p. 60, 1963.
The basic operational property can be used in various ways to obtain

transforms of particular solutions of differential equations involving our
differential operator D2. For instance, when a function H(x) of cIass C"
satisfies the differential equation D2H = O, it follows from property (12),
Seco 111, that its transform is
(7)
h(An) = ;n {r(b)Np[n]NpH - N~[n]N",H}
If constant values, not both zero, are assigned to N",Hand NpH here, then
H is the solution of the differential equation D2H = O that satisfies those
boundary conditions. We note that in case IX= P = O,formula (7)becomes
(8) T{H} = An-l[r(b)~(b)H(b) - ~(a)H(a)].
If F(x) == 1 (a ~ x ~ b), then D2F = C(x) and our formula for T{D2F}
reduces to
(9) - r(b)Np[n]cosP.
T{C(x)} = AnT{l} + N~[n]COSIX
Thus when <l>n,An, and either of the two transforms T{C(x)} or T{I} are
known, the other transform can be found. If C(x) = Co, a constant other
than one of the eigenvalues,then T{1}can be found,since
(10)
115 EXAMPLE OF STURM-LlOUVILLETRANSFORMATIONS

Let the differential form D2F be the simple self-adjoint form F" on the
interval (O,n),and let the prescribed boundary values be F(O)and F'(n).
Then
(1) ~2F = F"(x), O < x < n; N",F = F(O), NpF = F'(n);

thus IX = O,P = n/2, r(x) = p(x) = 1, and q(x) = O. In this case the kernel
<l>nof the Sturm-Liouville transformation is the family of all eigenfunctions
of the problem
(2) "(x)- A(X) = O, <1>(0)= '(n)= O.
We find that A = O is not an eigenvalue, and that

(3) <l>n(x)= sin (n - t)x, An = - (n - t)2 (n = 1, 2, .. . ).
Let Sm denote the transformation T here, where m = n - t. It is the

modifiedfinite Fourier transformation
(4)
Sm{F}= f F(x) sin (n - t)xdx =fs(m) (n = 1,2,...).
334 SECo 115] OPERATIONAl MATHEMATlCS
Since N~<Pn = <P~(O) = m and N'p<Pn

= -<Pn(n)= (-I)n, the basic opera-
tional property of Smfor functions of class C" is
(5) Sm{F"} = - m2f.(m) + mF(O)- (- l)nF'(n) (m = n - !).
We find that II<Pn112
an inversion formula for Tmis
2
-
= n/2, so the generalized Fourier series that furni$es
!.
<X>
(6) F(x) = -
n n= 1

f.(m) sin mx (m = n - O < x < n).
Transforms of a number of functions can be found by evaluating the

integral Sm{F} or by applying property (5) or formula (6). For instance,
(7) (m = n - !).
For other particular transforms see Probs. 5 to 8, Seco 116, and Table 2,
Seco 126.
We can derive kernel-product and convolution properties for Sm,
whose kernel is an odd function that is antiperiodic with the period 2n.
Let F3 be the extension of F which is odd and antiperiodic with period 2n;
~~ .
(8) F3(x) = F(x), O < x < n;

F3(X)= -F3( -x) = -F3(x + 2n), all x.
Then F3(X)sin mx, where m = n - !, is even and periodic with period 2n.
We introduce the integral
~ ~w=[~m~=[~m~-[~m~
which is even and antiperiodic with period 2n; also F(x) = - F3(x) and
Fo(n) = Fo(-n) = O. Then
2Sm{F}= f" F3(X)sinmxdx = - f" F(x)sinmxdx.

An integration by parts here shows that
2Sm{F} = 2f.(m) = m f" Fo(x) cos mx dx.

If Xo is a constant (O< Xo ~ n), then
f.(m)3... sin mxo = ~

m I
" Fo(x)[sin m(x + xo)
2 -"
- sin m(x - xo)] dx
beca use the integrand is periodic with period 2n. AIso, the integrand is an
even function of ~, SOthe last equation is the kernel-product property
(10) (m = n - t).
When O < Xo ~ n and O ~ x ~ n, we find that
(11)
if x + Xo~ n
if x + Xo~ n.
When Xo = n, Eq. (10) becomes
(12)
In terms of the notation used in Seco 109, w(A.)= 2/m and PF(x,xo) is
the function inside the braces in property (10). Hence the convolution of
two functions H(x) and F(x) corresponding to our kernel-product property is
XHF(X)
which can be written
= f H(y)[Fo(x - y) - Fo(x + y)]dy,
(13) X HF(X) = " H(y)

5.o
f" F(~)d~ -
[ )Ix-yl I " F3(~)d~ dy.
x+y J
A convolution property for Smis therefore
(14) (m =n- t),
where hs(m)= Sm{H}. Similar derivations of convolution properties can be

made for other Fourier transforms (cf. Seco110 and the chapters to follow).
Properties of Smare summarized in Table 2, Seco126.
Our transformation Sm reduces boundary value problems in partial
differential equations in which the differential form in the unknown function
U with respect to a variable x is Uxx' and the values of U and Ux are
prescribed at boundaries x = Oand x = n, respectively.
This problem in U(x,y) is an illustration:

a(y)U"" + b(y)U + c(y)Uyy = Q(x,y) (O< x < n, y > O),
U(O,y)= O, Ux(n,y)= p(y), U(x,O)= Uy(x,oo)= O,
assuming that the prescribed functions a, b, e, Q, and pare such that the
problem has a solution.
Let us(m,y)be Sm{U}with respectto x. Then according to our formula
for Sm{D2U},the transformed problem is, formally,
a(y)[-m2us - (-l)"p(y)] + b(y)us+ c(y)u;= qs(m,y) (y > O)
us(m,O) = u~(m,oo) = O,
where qs(m,y) = Sm{Q} and u~ = duJdyo If this problem in ordinary

differential equations can be solved for us(m,y), then U(x,y) can be written
with the aid of our inversion formula, or by using known transforms and our
convolution property.
If a term 04U/OX4,an iteration of the form Uxx' is added to the above
partial differential equation in U, and if the boundary conditions are
augmented by prescribing UxX<O,y)and Uxxx(n,y), the problem is again
reduced by Smsince
04U
Sm{ OX4} = -m2Sm{Uxx} + mUxx(O,y)- (-l)"UxxX<n,y),
and Sm{U",,} = - m2us(m,y) - (-l)"p(y)o Note that neither the Laplace
transformation nor separation of variables applies directly to these problems
in U.
116 SINGULAR CASES

Our method of setting up an integral transformation that applies to some
specified differential form and boundary values may require the kernel to
satisfya singular eigenvalue problem (Seco 107). Singular problems arise if
the interval of our variable x is unbounded, or if the differential equation in
the eigenvalue problem has a singular point on the interval, usually at an
end point of the interval. Conditions on the order of magnitude or regularity
of functions at singular points or infinitely distant end points may replace
prescribed boundary values at those points. We present two examples to
illustrate the procedureo
Example 1 Let the interval be unbounded, O < x < 00, and let the differ-
ential form and the prescribed boundary value and order condition
be these:
(1) D2F = F"(x), x> O; N"F = F(O), F(oo) = F'(oo) = O.
Here D2 is self-adjoint. The integral transformation
(2)
T{F} = fOF(x)eI>(x,..1.)dx = f(..1.)
requires additional conditions on F and el>to ensure the existence of

the improper integral.
We assume that F is of class C" and absolutely integrable over
the half line x ~ O,and that F((0) = F'((0) = O. Formula (8), Seco111,
obtained either by using Lagrange's identity or by integrating by parts,
now becomes
T {F"(x)} = {O F(x)eI>"(x,..1.)dx + [eI>F' - eI>'F]~
= ..1.IX) F(x)eI>(x,..1.) dx + eI>'(O,..1.)F(O)
if el>satisfies the singular eigenvalue problem

(3) eI>"= ..1.eI>,x > O; eI>(O,..1.)
= O, el>and <f)'bounded (x > O).
Ir k is a constant, the solution of problem (3) is
c1>(x,..1.)
= k sinh xfi (k :F O),
provided that ..1.is such that <f)and <f)'are bounded as x -+ oo. We
write fi= " + ip.,where " and Jlare real. Then sinh xfi is bounded
if and only if " = O, so that ..1.= - Jl2 and <f)= ik sin JlX; <f)'is also
bounded. Negative values of Jl yield no new eigenfunctions. Thus
..1.= - Jl2,Jl > O,and if we write ik = 1, then
(4) <f)(x,..1.)= sin JlX, ..1.= _Jl2 (jl > O).
Hence T is the Fourier sine transformation (Sec. 110)
(5) (jl > O)

T{F} = SIl{F} = IX) F(x) sin JlXdx = f.(jl)
with the basic operational property
(6)
Its inverse transformation is given by the Fourier sine integral formula

(Prob. 15, Seco 107):
2 "" 2
(7) F(x) = -
non f f.(jl) sin JlX dJl = -Sx{f.(Jl)}.
For other properties of Sil see Chap. 13 and Appendix D.

Example 2 As an illustration of a differential form with the end point

x = Oof the bounded interval 0-< x < b as a singular point, consider
the form
1
(8) D2F = F"(x) + -F'(x)
x
(O < x < b).
The form arises when the Laplacian operator is written in terms of

cylindrical coordinates. Let the prescribed boundary value and
regularity conditions be
(9) NpF = F'(b), F of class C"(O~ x ~ b).
In terms of the self-adjoint form ~2F = (xF')' and the notation
used in Seco 111, we write
D2F = x-l(xF')', p(x) = r(x) = x, /3= n/2,
T{F} = f: F(x)xCl>(x,A)dx = f(A).
Then formula (8), Seco 111, for T{D2F} becomes
I: (xF')'CI> dx = I: (xCI>')' F dx + [x(CI>F' - CI>'F)]g,
since r(x) = X. Ir CI>

satisfies the singular eigenvalue problem
(lO) (xCl>')'= AXCl>, CI>'(b))
= O, in class C" (O~ x ~ b),
CI>
then
T{D2F} = Af(A) + bCl>(b))F'(b).
The differential equation xCI>"+ CI>'- AXCl> = O in problem (10)
is Bessel's equation with index n = O, whose solution of class
C" (O~ x ~ b) is CI> = kJo(x~), where k is a constant (Sec. 83).
The condition CI>'(b))= O requires A to be zeros of the function
J1(b~). We write b~ = J1..The odd function J1(J1.)has an
infinite sequence of zeros J1.1= O, ::!:J1.2,::!:J1.3"'" all real; its non-
negative zeros J1.
yield all the eigenfunctions. Thus the eigenvalues are
,1= -J1.2/b2,and
[J 1(J1.)= O, J1.~ O,i = 1,2,...].

CI>(x))= Jo(J1.~)
Our transformation on the bounded interval (O,b) is then the
particular finite H ankel transformation
" (11)
T{F} = I: F(X)XJo(J1.~)dx = fo(J1.},
where J 1(f.l) = O,f.l ~ O,and f.lj-+ 00 as i -+ oo. It has the property
(12) T {F" + ~F'} = -b-2f.l/fo(P)+ bJo(Pj)F'(b).

Recall that f.ll = O and J 0(0) = 1.
We shall treat Hankel transforms more fully in Chap. 14 where
we shall note that Fourier-Bessel series furnish inversion formulas for
the finite transformations.
PROBLEMS
1. When D2F = F"(x), O< x < b, N.F = F(O), and NpF = F(b), show that the
Stunn-Liouville transformation becomes the .finite Fourier sine transformation on the
interval (O,b),
b mrx
T{F} = Sn{F} = f
o F(x) sin Tdx = fs(n)
(n = 1,2,. .. );
show also that if F of cIass C" (O ~ x ~ b), then
Sn{F"} = - (n;f.r.(n) + nb1t

[F(O)- (-l)nF(b),
and the inversion formula (1), Seco112,is the Fourier sine seriesrepresentationof F
on the interval (O,b).
2. (a) If F is of cIassC(4)(O~ X ~ b), show that the transformation Snin Probo l
has the property
Sn{F(4)(x)}= (n;)~(n) - (n;f[F(O) - (-l)nF(b)
+ nb1t[F"(O)- (-l)nF"(b).
(b) If F(4)(X) = O, F(O) = F(b) = F"(O) = O, and F"(b) = 6b, show that F(x) =
x(X2 - b2) and fs(n) = 6b4( -lf/(n1t)3.
3. When D2F= F"(x), O< x < b, N.F = F'(O),and NpF = F'(b), show that the
Sturm-Liouville transformation becomes the .finite Fourier cosine transformation on
the interval (O,b),
b n1tX
Cn{F} = fo F(x) cosT dx = fc(n) (n = O, 1, 2, . . . );
show also that if F is cIass C" (O~ x ~ b), then
(n = O,1,2, . . . )
Cn{F"}= -(n;ffc(n) - F'(O)+ (-l)nF'(b)
and that Cn-l{fc} is given by the Fourier cosine series
1 2 00 n1tX
(O< x < b).
F(x) = bfc(O)+ b n~lfc(n)cosT
4. Find the Sturm-LiouvilIe transformation on the interval (0,1) that resolves the
differential form F"(x) in terms of the transform of F and the boundary values F(O)
and hF(I) + F'(l), where h is a positive constant, and show that it has the operational
property
T{F"} = -/l/T{F} - F(O) + [hF(l)+ F'(l)]cOS/l.,
wherep. (n = 1,2,... ) are the positive roots of the equation tan /l = h//l (cf. Seco 105).
Ans. T{F} = f F(x) cOS/l.X dx

(n = 1,2,...).
5. Show that Green's function for the problem Y"(x) = 0, Y(O)= Y'(1t)= O is the
function
G(x,e) = - x (O~ x ~ e), G(x,e) = - e (e ~ x ~ 1t);
then apply formulas (3) and (6), Seco114, when T is the modified finite Fourier trans-
formation Smdefined in Seco115, where m = n - t. to obtain the inverse transforms
1 sin me l cosme
Sm- {--;;r-} = -G(x,e), Sm- { ;;;- } = So(x- e),
whereO< e < 1tand Sois the unit step function.
6. Showthat the transformation(Sec.115)
(m = n - t)
Sm{F} = s: F(x) sin mx dx = f.(m)
has the property
Sm -1{ f~7)} = J: r F(t) dt dr = J: tF(t) dt + x f F(t) dt,

and that
S -1 ~ = 1tX - X2
m { m3 } 2'
7. - When f.(m) = e-my where m = n - t and y > O and independent of x, inversion

formula (6), Seco115, for the modified finite Fourier transformation Smcan be written
F(x,y) = ~ I
1t.=1
e-my sin mx =~
1t
1m
.=1
I (exp iz).-t,
where z = x + iy and lexp izl < 1. Sum the series here to show that
Sm-l{e-my} = .!.Re ese:' = 1 (sinx/2)(coshy/2)

1t ( 2) 1tcosh2 y/2 - cos2 x/2
when y > O. [Since F(x,y) is the imaginary component of an analytic function of z,

it is a harmonic function. See Probo 2, Seco118, where e-my arises. Also note that F
satisfies conditions (8), Seco115, on F3.]
8. For the transformation S", in Seco115, where m = n - t, show that if y> O and
independent of x and z = x + iy, then
- e-my
S", 1
{- }
m
2
= - 1m
7t
L (expizr
00
n= 1 m
2 1 + exp iZ/2 2 sin x/2

= - I m l og = - arctan - (y> O).
7t ( 1 - exp iz/2 ) 7t ( sinh y/2 )
9. Complete the derivation of the basic operational property (3), Seco113, for the
Sturm-Liouville transformation in the form f:FK dx.
10. Write F(x) = e-kX(k > O) in the basic operational property (6), Seco116, of the
Fourier sine transformation SIl to obtain the transform
J.I
(k> O).
S/l{e-kX} = J.l2 + k2
Differentiate with respect to k to show that
S/l{xe-kX} = 2kJ.l(U2+ k2)-2 (k> O).

11. Derive the integral transformation on the half line x ;;;O that reduces F"(x) in
terms of the boundary value F'(O) when F is of cIass C", absolutely integrable over the
half line, and F(oo) = F'(oo) = O.This is the Fourier cosine transformation
(.t~ O)
C/l{F} = {') F(x) cos J.lXdx = fc(J.I)
(Prob. 5, Seco110 and Chap 13). Show that
C/l{F"(x)} = _J-l2fc(J-l)- F'(O).

12. Derive the integral transformation that reduces the form F"(x) + X-l F'(x) on
the interval(0,1)in termsof the boundary value F(1)when F is of cIassC"(O~ x ~ 1).
This is the finite Hankel transformation (Chap. 14)
T{F} = Jo(1 F(x)xJo(.tr)dx =f(J.I.)

J
(j = 1,2,... ),
where the real numbers J-ljare the positive roots of the equation Jo(.t)= O. Since
J(.t) = - J 1 (.t), show that
T{F" + x-1F'} = -J-l/f(J-lj) + J.I/l(J-l)F(l).
117 A PROBLEM IN STEADY TEMPERATURES

As an example ofboundary value problems that are well adapted to solution
by specific Sturm-Liouville transformations, we present the following one
on steady-state temperatures in a semi-infinite slab in which heat is generated
at a steady rate while surface heat transfer takes place at one face. Let the
temperature function U(x,y) satisfy the conditions ,
u xx(x,y) + Uyy(x,y) + Q(x,y) = O (O< x < 1, Y > O)
(1)
U(O,y) = P(y), U(l,y) + hUx(1,y)= O, U(x,O) = O,
where h is a constant (h ~ O),the functions P and Q are bounded, and we

require U to be bounded. .
Here the differential form with respect to x is Uxx' and the prescribed
boundary values at x = O and x = 1 are Na.U = U(O,y)a~d NpU =
U(l,y)cosP + Ux{l,y)sinpwheretanp = h(O ~ P < n/2). Hencethekemel
$n(x) and the parameter An of the Sturm-Liouville transformation that
applies to problem (1) are the characteristic functions and numbers of the
problem
(2) $" - A$ = O, $(O,A) = O, $(1) + h$'(l,A) = O.
Thus we find that
(3) tan J.ln= -hJ.ln (J.ln> O),
and we can see from the graphs of the functions tan J.l and - hJ.lthat
J.ln- (n - !)n -+ Oas n -+ 00 when h > O; when h = O,J.ln= nn. Our trans-
formation is
(4)
T{ U} = f U(x,y) sin J.lnXdx = un(Y)
(n = 1,2,...).
Now Na.U = P(y), IX= O,and N~$n = $~(O)= J.ln;also NpU = O,and
An= - J.l/. Therefore
T { U xx} = - J.ln2un(y) + J.lnP(y)
and if qn(Y) = T{ g(x,y)}, the transformed problem is

(5) 2un(y)+ J.lnP(y)+ qn(y) = O,
u:(y) - J.ln
where Unis bounded on the half line y > O. With the aid of the equation
tan J.ln= - hJ.ln,we find that
lI$nll2 = !(l + h COS2J.ln)'
After solving problem (5) for Un,we have the formal result
(6) U(x,y) = 2 f un(y)sin ~nx .

n=11 + hcos J.ln
The form of the solution can be improved in special cases. When
P(y) = O and Q(x,y) ==Qo, a constant, then '
and the solution of problem (5) becomes
(7)
1 - cosJ.ln 1 - cosJ.ln
un(y) = Qo 3 - Qo 3 exp (- J.lnY).
J.ln J.ln
The first fraction is -T{l}/An. According to formula (4), Seco112, this is

- T{Z} where .
Z"(x) = 1, Z(O) = O, Z(1) + hZ'(l) = O.
Hence we find that
l - COSJl.n
Z(x) = -~ 1 + 2h +
21+h
!X2 = -T-l
2 { Jl./ }
and the inverse of transform (7) can be written
+4 ~
- Qo 1 + 2h - 2
(8) U( X,Y) -
[
2 1 + hx x L...
n= 1
1 - cos Jl.n exp (- Jl.nY).
1 + h cos 2 Jl.n 3
Jl.n ].
SIn Jl.nx
Form (8) can be verified fully as the solution of problem (1) in this
special caseP = O,Q = Qo. Note that if only a finite number of terms of
the infinite series is used in formula (8), the resulting function satisfies all
conditions in problem (1) except the condition at the base y = O. Also, we
see that U has the property
. Qo l + 2hx - x 2 .
11m U(x,y) = -2 -
y-oo ( 1+ h )
The differential form Uyy and boundary conditions in problem (1) are
such that the Fourier sine transformation with respect to y, on the half line
y > O, applies to the problem provided that the transforms of both the
functions P and Q exist; but that is not the case when Q(x,y) = Qo. In case
the second method does apply, we have an advantage of obtaining a different
representation of the function U. Note that problem (1) is not adapted to a
Laplace transformation. Since the partial differential equation is not
homogeneous, the method of separation of variables does not apply.
For further properties of transformation (4) see Seco 128.
"'
118 OTHER BOUNDARYVAlUE PROBlEMS
Other special and singular cases of Sturm-Liouville transformations will be
applied to boundary value problems in the chapters to follow. Operational
properties of some of them are adequate to represent solutions of problems
in more than one form, sometimes in quite simple forms. But in each
application of our process an eigenvalueproblem must be solved to deter-
mine the kernel of the transformation; then the transformed problem must
be solved to find the transform of the unknown function. Failure of the
transformedproblem to have a unique solution suggeststhat conditions are
missing or incorrect in the boundary value problem.
Tomake some observations on types of problems to which the method

applies, we consider here problems in two independent variables and Sturm-
Liouville transformations. But when boundary conditions are properly
modified, our observations apply to other cases including singular cases,
differential forms of higher order, and problems in more than two indepen-
dent variables. In contrast with the method of separation of variables the
method of integral transformations is not restrieted to problems with homog-
eneous dijJerential equations or with eertain homogeneous boundary eonditions.
Let V(x,y) denote the unknown function in our problems and let the
differential form and prescribed boundary values on x be
D2 V = A(x)Vxx(x,y) + B(x)Vx(x,y) + C(x)V(x,y) (a < x < b)

N aV = V(a,y) cos rx + VAa,y) sin rx,
N (JV = V(b,y) cos /3 + VAb,y) sin /3.
Note that the coefficientsA, B, and e are functions of x alone; rxand /3are
real and constant. Then as before
T {V} = f V(x,y)p(x)c1>n(x) dx = vn(y)

(n = 1,2,...).
where ~n are the eigenfunctions of the Sturm-Liouville problem corre-

sponding to the operators D2, Na, and N(J'
To indicate some types of boundary value problems in V to which
transformation T applies, let L,y, i = 1, 2, 3, denote linear homogeneous
differential operators with respect to y whose coefficients are functions of y
alone. Those operators may be of any order; if the order is zero, the form
L,yVis simply a product E(y)V(x,y).
Consider a linear boundary value problem consisting of a part1al
differential equation and boundary conditions
(1) Ll,y[V] + L2,y[D2 V] = F(x,y),

(2) NaV = G(y), N(JV= H(y),
and some boundary conditions with respect to y of type
av amv
(3) el V + e2 - + .. . + e - = P(x e) when y = e,
ay m ayrn ,
where the e's are constants. When Tis applied to Eq. (1) and (3) with inter-
changes of operations with respect to x and y, then, in view of conditions (2),
the resulting differential equation is
wheref,,(y) = T{F(x,y)}. If Pn(C)= T{P(x,c)}, the boundary conditions on

Vnare of type
dVn d"'vn
(5) h
C1Vn+ C2 dy + ... + Cmdym = Pn(C) w en y = c.
Then V(x,y) = T-1{vn(Y)}. Unless tables or operational properties of
T can be used to simplify the inverse transformation, we may write the
formal solution as
ex>
(6) V(x,y) = L IIcI>nll-2vn(y)cI>n(x).

n=l
The transformation T may apply if the problem involves D2 and some

of its iterations. For example, the partial differential equation may have
the form
(7)
if it is accompanied by boundary conditions at x = a and x = b that

prescribe Na.V,NpV, Na.[D2V],and Np[D2V]. Then even ifEq. (7) and those
boundary conditions are homogeneous, the usual method of separation of
variables may not apply.
A special case of Eq. (7) is the biharmonic equation
04V 04V 04V
(8)
oy4 + 20y2ox2 + OX4 = O
satisfied by the static transverse displacements V(x,y) in an unloaded elastic
plate. Here D2V = Vxx, Ll.,y = o4joy4, L2,y = 202joy2, and L3,y = 1. If
displacements Vand bending moments are given along the edges x = a and
x = b, then V and Vxxare prescribed on those edges.
PROBLEMS
1. Note why the transformation (Sec. 115)
(m= n - t)
Sm{U(x,y)} = J: U(x,y) sin mx dx = u.(y)
applies to this problem in steady temperatures:
U",,(x,y) + Uyy(x,y)= O, U bounded, (O<x<x,y>O),
U(O,y)= U",(x,y)= O, U(x,O)= 1 (O< x < x).
Show that u.(y) = m-le-my and thus (Prob. 8, Seco116)
U(x,y) = -2x arctan sin x/2

--=
sm h y /2
(O< x < x, y > O).
346 SECo 118] OPERATIONAl MATHEMATlCS
2. Use the transformation Sm(Sec. 115) with respect to x to find the solution of the
problem
uxx(x,y) + Uyy(x,y)= O, U bounded, (O< x < 1t,Y > O),
U(O,y)= Ux(1t,y)= O, Uix,O) = -1,
in the form U(x,y) = Sm-1{m-2e-my}; thus show that
x
U(x,y)= -log
1t
x cosh y/2 + cos x/2
i
cos h y /2 - cos x/2 + o tF(t,y)dt
where (Probs. 6 and 7, Seco116)
F(t, ) = .!. Recsc t + iy =.!. (sint/2)(coshy/2) .

y 1t 2 1tcosh2 y/2 - cos2t/2
3. Determine the Sturm-Liouville transformation T, with respect to x, that is adapted
to the problem
(t + l)U,(x,t) = Uxx(x,t) (O< x < 1, t > O),
U(x,O)= O, Ux(O,t)= -1, U(l,t) = O.
Find T-1{1/ln} by solving and transforming the elementary problem Y"(x) = O,

Y'(O)= 1, Y(l) = O; then derive the formula
U(x,t) = 1 - x - 2 ~
L..
(t + l)-N (2n - l)1tX
cos-
n=1 N 2
4. Determine the integral transformation (a) with respect to y, (b) with respect to x,
that applies to the problem
U xx(x,y) + Uyix,y) + 2Ux(x,y) = O (O< x < 1, O < y < 1),
U(O,y)= G(y), U(l,y) = O, Uy(x,O)= O, Uy(x,l) = H(x),
and write the transformed problem in each case.
Ans. (a) uC<x,n)= f~ U(x, y) cos n1ty dy (n = O, 1,2, . . . );
u; + 2u~ - n21t2uc = (_l)n+1 H(x), uC<O,n)= gc(n), uc(l,n) = O.
(b) u(ln,y) = f~ U(x,y)~ sinn1txdx (ln = -1 - n21[2,
n = 1,2,...);
u" - (1 + n21t2)u= -n1tG(y), u'(ln'O)= O,u'(ln,l) = h(ln)'
5. The line y = O is a singularity of the equation
Vxx(x,y)+ y2Vyix,y) + y~(x,y) + F(x) = O.
If V is to be bounded over the rectangle O < x < 1t, O < Y < b, solve that partial
differential equation under the conditions
VAO,y)= V(1t,y)= V(x,b) = O.
n m
Ans. V(x,y) = fx io
'
F(r)dr dt - -
2 <X>
f(l
L -+ -b
1t n= 1 m
) y
( }
cos mx,
where f(ln) = So"F(r) cos mr dr and m = n - .

6. Note why the Fourier sine transform (Sec. 110) on the half line x > O applies to
this problem on transverse displacements of a stretched string with prescribed initial
velocity (Prob. 5, Seco44):
Y,,(x,t) = a2 Y",,(x,t) (x > O,t > O)
Y(x,O) = O, Y,(x,O) = G(x), Y(O,t)= O.
Use property (4), Seco110, to show that
1 "+a'
i
Y(x,t) = 2a ',,-a" G(r)dr
and verify that solution of the problem.
Without solving the following problems for V(x,y), determine procedures that
are adapted to their solution.
7. V",,(x,y)+ Vyy(x,y)- V(x,y)= O, (x> ,O < Y < 1),
V(O,y)= 1, Vy(x,O)= O, Vy(x,l)= - V(x,l) + e-".
Ans. Fourier sine transformation with respect to x (Sec. 110); the Sturm-
Liouvilletransformationin Probo4, Seco116,with respectto y, where h = 1.
8. V",,(x,y)+ X-l V,,(x,y)- Vyy(x,y)= Q(x,y) (O< x < 1,y > O),
V(l,y) = O, V(x,O)= O, Vy(x,O)
= O.
Ans. The finiteHankel transformationin Probo12,Seco116;the Laplacetrans-
formation with respect to y.
11
Finite Fourier Transforms
119 FINITE FOURIER SINE TRANSFORMS

A useful special case of the Sturm-Liouville transformation is that in which
the differential form is the simple self-adjoint form F"(x), and the prescribed
boundary values are the values of F at the ends of a bounded interval (Prob.
1, Seco116).
It is convenient to choose the origin and the unit of length so that the
interval is (0,1t). Then in terms of the notation used in Chap. 10,
O<x<1t; N"F = F(O), NpF = F(1t);
here oc= p = O,p = r = 1,and q = O. The kernel of the transformation T

consists of the eigenfunctionsof the problem
<1>" - A.= O, <1>(0)= (1t)= O.
= sin nx, and T becomes the finite Fourier

Thus A.= -n2(n = 1,2,.. .), <1>
348
FINITE FOURIER TRANSFORMS [SECo 119 349
sine transformation
(1) (n = 1, 2, .. .).
Sn{F} = J: F(x) sin nx dx = f.(n)
The basicoperationalproperty(12),Seco111,becomes
(2) Sn{F"(x)} = -n2f.(n) + n[F(O) - (-l)nF(n)]
when F is of class C" (O~ x ~ n), or even if F and F' are continuous while F"
is sectionally continuous over the interval. lt follows that
(3) Sn{F(4)(X)}= n4f.(n) - n3[F(0)- (-l)nF(n)]
+n[F"(O) - (-l)nF"(n)]
if F is of class C(4). Formulas for transforms ofhigher-ordered iterates ofthe
form F" can be written easily.
Transforms of many elementary functions can be found by evaluating
the integral Sn{F}. Some can be found readily from property (2); as examples,
when F(x) ==1, or when F is x, x(n - x), or x, that formula gives these
transforms :
(4)
Sn{~(n- X)} = 1 - (;-l)n,

Other entries in Appendix B, Table B.l, can be derived by those
methods. Some operational properties of Sn are included in that table of
finite sine transforms.
Following the procedure used in Seco 112, we write f.(n)fn2 = ys(n)
where ys(n) = Sn{Y(x)}. Then n2Ys(n) = f.(n), and this is true if Ysatisfiesthe
conditions
Y"(x) = - F(x), Y(O)= Y(n) = O.
Thus we find that, when F is sectionally continuous,
(5)
Sn -1 {f~~)} = J: ; (n - r)F(r) dr - I: (x - r)F(r) dr.
For Sn the series representation (1), Seco 112, of the inverse Sturm-
Liouville transformation becomes the Fourier sine series representation of
F(x):
2 DO
(6) Sn -1{f.(n)}= F(x)= - L f.(n) sin nx

n n= 1
(O < x < n),
"
1:
l.
Ij
~
since Ilsin nxll2 = n/2.

This formula for the inverse sine transformation is
valid over the interval if F and F' are sectionally continuous there and if, at
each point Xowhere F is discontinuous, F(x) is defined as the mean value of
its one-sided limits,
(7) F(xo) = !-[F(xo + O)+ F(xo - O)] (O < Xo < n).
For all real x the series (6) converges to this odd periodic extension F1 of F,
with period 2n:
F1(x) = F(x) when O < x < n, F1(0)= F1(n)= O;
(8)
F1(-x) = -F1(x), F1(x + 2n) = F1(x) ( - 00 < x < 00),
because sin nx is odd and periodic with period 2n.
120 OTHER PROPERTIES OF SR

Let F1 be the odd periodic extension, with period 2n, of a function F that is
sectionally continuous over the interval (O,n),and let k be real and constant.
Then
2f.(n) cos nk = f" F1(x) sin nx cos nk dx
since the integrand is an even function of x. The integral can be written in

these forms :
t f" F1(x)sin n(x + k) dx + t f" F1(x) sin n(x - k) dx
= t f" F1(r - k)sinnrdr + t f" F1(r + k)sinnrdr
because the last two integrands are periodic functions of r with period 2n.
The sum of those integrands is an even function, so
2f.(n) cos nk = f [F1(r - k) + F1 (r + k)] sin nr dr;

that is, if F is sectionally continuous and Sn{F} = f.(n), then
(1) 2f.(n) cos nk = Sn{F1(x - k) + F1(x + k)}.
When k = n, then, since F1(x- n) = -F(n - x) when O < x < n and

F1(x + n) = F1(x - n),property (1)becomes
(2) (n = 1,2,.. .).
For example, since S"{x} = 1tn-l( -1)"+ 1, it folIows that
(3)
Kernel-product and convolution properties can be written for S".

Again let Fl be the odd periodic extension of F with period 21t. Then the
function
(4) Fo(x) = J: Fl(r)dr (-00 < x < (0)

is periodic with period 21t and even, Fo(x) = Fo(lxl), and F(x) = Fl(X)
wherever Fl is continuous. Therefore
(5)
fs(n) = f F(x) sin nx dx = - n fa" Fo(x) cos nx dx.
If Xo is a constant, then
~f.(n)
n
sin nxo =_
2
1
IIt
-It
Fo(x)[sin n(x - xo) - sin n(x + xo)]dx,
and this kernel-product property folIows easily:
2 x+xO
- f.(n) sin nxo = S" .
(6)
n {f x-~
Fl(r) dr
}
The corresponding convolution of two sectionalIy continuous functions
F and H is (Sec. 109)
lt X+Y
(7)
XFH(X) = i
o H(y) fx-y F1(r)drdy,
with the convolution property, with weight function 2/n,

2
(8) S"{XFH(X)}= -f.(n)h.(n)
n
where h. = S"{H}. The integral (7) can be written in terms of H and the
function F (Prob. 13).
In case F is such that Fl is everywhere continuous and F' is sectionalIy
continuous, then Fl(:!::n) = Oand
l It
(9)
It
I
2f.(n) = -It Fl(x)sinnxdx I
= -;; -It F;(x)cosnxdx.
Thus a kernel-product property with weightfunction w = 2n can be obtained,
namely,
(10)
For Sn' property (8), Seco 113, becomes

lim.fs(n) = O
whenever F is sectionally continuous. Some further properties of Sn' and

transforms of particular functions, are given in the problems.
PROBLEMS
1. Use properties of S. to obtain these transforms:
(a) S.{X3} = n(-1)"(:3 - :2) ;

= (-1)"+I~ n - if k =F:t 1, :t 2, .. .;
(b) S.{sinkx} k2sinkn
n
(e) S.{coshex} = ~[1
n +e - (-1). coshen].
2. (a) If F(x) = sinmx wherem is a positiveinteger,showthatf.(n) = Owhenn =Fm,

andf.(m)= nj2.
(b) Derive: S.-I{:3(-1).+I} = :n(n2 - X2).
(e) Derive: S.-1 {:3} = :n (X2- 3nx + 2n2).

3. If F(n - x) = F(x),showthatf.(n) = Owhenn is even.
4. Let F and F' be continuouswhenO~ x ~ n exceptthat F' has a jump b at a point
x = e interiorto the interval,and let F" be sectionallycontinuous.Prove that
+ n[F(O)- (-1).F(n)] - bsinne.
S.{F"} = -n2f.(n)
5. Note the graph of the functionF(x) = Isin 2xl and use the results found in Probs.
3 and 4 to show that
f.(n) = O if n = 2,4, . .. , f.(n) = 4 4sin- nnj2

n2
if n = 1,3,5,.. ..
6. Derivethe inversetransform
- ifO ~ x ~ e < n
.
S -1 nsinne
{ n2 }
= (n e)x
.{ (n - x)e if e ~ x ~ n,
(a) with the aid of formula (3), Seco114, when T is S.;
(b) with the aid ofthe kernel-product property (6),Seco120,whenf.(n) = n-l.
7. If the real numbers al' a2, . . . are such that the series al + a2 + ... is absolutely
convergent, prove that
(n = 1,2,...).
8. When y> O,the numbers am = m-I(-I)m+le-my, where m = 1,2,..., satisfy the

conditions in Probo 7. If z = x + iy, showthat
'" (-Im+1 '" (-Im+1 .
L
m=1 m
e-my sin mx L
= 1m m=1 m
(elZm
= 1m [Iog (1 + eiZ)],
and thus derive the inverse transform
2 sin x
S -1 -e
( -1)n+ 1 -ny = - arctan if y> O.
n { n } n ( eY+cosx )
9. From the transform found in Probo 9 note that

1 2 sinx
Sn-1 -e-ny = -arctan-
{n } n eY- cosx
when y > O; then show that
1- 2
(-I)n e-ny = -arctan-,.-sin x
S -1 if Y > O.
n { n } n smh y
10. When -1 < b < 1 and -n/2 < arctan t < n/2, use transforms found in Probs. 8
and 9 to show that
bn 2 b sin x
- = S -arctan (-1 < b < 1);
n nn{ l-bcosx }
then differentiate with respect to b to get the transform
2 bsinx
bn
=
S
n{ n1 + b2 - 2b cos x } (-1 < b < 1).
11. Given that, when c =PO,the function
Y( ) = sin cx sinh c(2n - x) - sin c(2n - x) sinh cx

x 4c2(sin2cn + sinh2 cn) ,
satisfies the conditions
Y(O) = Y(n) = Y"(n) = O, Y"(O)= -1,
showthat
n
Sn{Y(x)} = n4 + 4c"
12. If Y1is the odd periodic extension of the function Y in Probo 11, with period 2n,
find the solutionof the problem .
2(4)(X) + 4c42(x) = 2F(x),

2(0) = 2(n) = 2"(0) = 2"(n) = O,
in the form
" x+'
Z(x) = i i
o F(t) X.-/ Y(r) dr dt.
13. Express the convolution integral (7), Seco120, in terms of the functions H and F
by the formula
Jo" H(y)i x+y

x-y
F(r)drdy = J "-x
o
H(y)
Jox+y F(r) dr dy
I i
+ " H(y) 2"-X-Y F(r)drdy -
"-X o i i
x
o
H(Y) X-Y
o
F(r)drdy
- f H(y) S:-X F(r)dr dy.

14. Ir F, the odd periodic extension of F with period 2n, is everywhere continuous
and if F' and the function H are sectionally continuous, show that a convolution
property for S" corresponding to the kernel-product property (10), Seco120, is
a "
i
Sn-{2nf.(n)h.(n)} = ax o H(y)[F(x - y) - F(x + y)] dy.
15. Establish the following relation between the Fourier sine transformation on the
interval (O,b)introduced in Probo 1, Seco116, and OUTtransformation Snon the interval
(O,n):
b . nnx b b
Jo F(x) SIDTdx = Sn{ ( )} . ,/;rx
121 FINITE COSINE TRANSFORMS

The Sturm-Liouville transformation that resolves the differential form
F"(x) on the interval (O,n)in terms of the transform of F and the boundary
values F'(O)and F'(n) has a kernel which satisfies the eigenvalue problem
"(x)- A(X) = O, <1>'(0) = <1>'(n) = 00
Thus A = - n2 and (x)

= cos nx, where n = O,1,2, . o.. In this case
(x)= 1 is an eigenfunction corresponding to the eigenvalue A = 00 The
transformation is the finite Fourier cosine transformation
(1) (n = O, 1,2,. o.)

Cn{F(x)} = I: F(x) cos nx dx = fc(n)
with the operational property
(2) Cn{F"(x)} = -n2fc(n) - F'(O)+ (-1)nF'(n) (n = O, 1,2, o. o)
when F and F' are continuous, and F" is sectionalIy continuous, over the
interval O~ x ~ no
I
When F is of cIass C(4), it folIows that
(3) Cn{F(4)(X)} = n4fc(n) + n2[F'(O) - (-lfF'(x)] - F"'(O)+"( -lrF"'(x).

The formula for Cn{F(6)}can be written by replacing F by F" in formula (3),
and so on for transforms of the higher even-ordered derivatives. The
boundary values that arise are those of the odd-ordered derivatives.
Since IIcosnxll2 = x/2 when n = 1,2,..., then if m = 1,2,...,
(4) Cn{cosmx} = O ifn.p m, Cm{cosmx} = !x;
(5) Cn{1} =O ifn = 1,2,..., Co{1}=x.
Ir A is a constant andfc(n) = Cn{F},then
(6) C"{F(x) + A} = fc(n) when n = 1,2,...
= fc(0) + 7tA when n = O.
AIso, fc(n) -. O as n -. 00 when F is sectionalIy continuous. Note that
fc(O)/x is the mean value of F on the interval (O,x). Thus Co{x}/x = x/2,
while property (2) shows that
C"{x} = -[1 - (-1)n]n-2 when n = 1,2,....
When F is sectionalIy continuous, a function Y whose transform
yc(n)is J.(n)/n2when n = 1,2,..., can be found by writing n2Yc(n)= fc(n).
That equation is satisfied,according to properties (2)and (6),if
Y"(x) = -F(x) + B, Y'(O)= Y'(x) = O,
where B is to be determined by the conditions on Y'. We find that
cxr 1
Y(x) = Jo J, F(r)drdt + 2xfc(O)(x- X)2 + C.
The constant C is determined by the value assigned to y.(O). We can simplify
the iterated integral to write
fc(n) X 1
(7) 7 = c" {Ix (x - r)F(r)dr + 2xJ.(O)(x- X)2 + A}
when n = 1,2,..., where A = C +

I
'0
x rF(r) dr.
For Cn the series representation of the inverse Sturm-Liouville trans-

form is the Fourier cosine series
1 2 co
(8) C"-1 {fc(n)} = - fc(0) + - L: fc(n)cos nx = F(x)
x 7t"=1
when O < x < x, valid if F and F' are sectionalIy continuous and if F is
defined as its mean value at each point of discontinuity. For alI real x the
L.
series represents the even periodic extension F2 of F with period 2n:

(9) F2(X) = F(x) when O< x < n,
= F2(lxl) = F2(x + 2n) for all x.
This kernel-product property for Cn can be written by using the even
periodic extension (9) of F:
(10) 2fc(n) cos nxo = Cn{F2(x - xo) + F2(x + xo)} (n = O,1,2,. . .).
Ir ~ Xo ~ n, then F2(x - xo) = F(lx - xol) here. It follows that
(11) fc(n)( -lt = Cn{F(n - x)} (n = 0,1,2,.. .).
The convolution property corresponding to formula (10) is
(12)
Cn -1{2fc(n)hc(n)} = s: [F2(x - y) + F2(x + y)]H(y) dy,
where F and H are sectionally continuous. The integral here will be written
in terms of H and the function F in Seco 123.
122 TABLES OF FINITE FOURIER TRANSFORMS

Tables of Cn{F} and Sn{F} will be found in Appendix B. Transforms of par-
ticular functions are found by evaluating the integrals Cn{F} or Sn{F}, by
using partial fractions or other decompositions of transforms, by differen-
tiating or integrating with respect to a parameter, by using operational.
. properties of the transformations, or by summing the series that represents
the inverse transformo Relations between sine and cosine transforms, given
in the following section, also aid in extending the tables.
An example of the use of the Fourier cosine series will be given here,
corresponding to Probs. 7 and 8, Seco120. Suppose
1
(1) fc(n,y) = n-e-ny when n = 1,2,..., and fc(O,y)= 0,
where y > O. Then the series with termsfc(n,y) is absolutely convergent and
fc(n,y) = Cn{F(x,y)} when
2001 2001.
F(x,y) = n=lnn
L: -e-ny cos nx = n-Re n=ln
L: -(e'z)n
where z = x + iy and y > O.It followsthat

2 . 1 2
F(x,y) = --Re[log(l
n - e'Z)]= --log(1
n + e- y - 2e-ycosx)
1
= --log[2e-Y(coshy
n - cosx)].
1
Thus we find that

(2) nF(x,y) = y - log 2 - log (cosh y - cos x) (y > O).
The method used to derive formula (2)is not sound if y = O. But in that
case the function F is still defined:
(3)
F(x,O) = -~log [2(1 - cos x)] = -~ log (2 sin~),
and our results suggest that
ifn =O
(4)
if n = 1, 2, . . . .
The improper integral Cn{F(x,O)}can be evaluated with the aid ofthe theory
of residues1 to prove that formula (4) is correct.
123 JOINT PROPERTIES OF Cn AND Sn

Some operational properties of our two finite Fourier transformations have
simple forms when stated in terms of both transformations.
Ir F is continuous and F' is sectionally continuous, integration by parts
shows that, when n = O,1, 2, . . . ,
(1) Sn{F'(x)} = -nCn{F(x)},
(2) Cn{F'(x)} = nSn{F(x)} - F(O)+ (-l)nF(n).
Alternate forms of those properties are
(3)
Sn{H(x)} = -nCn{J: H(r)dr},
(4)
Cn{H(X) - ~hc(O)} = nSn{f: H(r)dr- ;hc(O)},
when H is sectionally continuous.
Let F1 and F2 be the odd and even periodic extensions, with period 2n,
of a function F that is sectionally continuous on the interval (O,n), and let
k be real and constant. Then by the method used to derive property (1),
Seco120, we find that
(5) 2f.(n) sin nk = Cn{Fl(X + k) - Fl(X - k)},
(6) 2.fc(n)sin nk = Sn{F2(x - k) - F2(x + k)}.
I Cf. seco 73 of the author's "Complex Variables and Applications," 2d ed., 1960.
When P and Q are sectionally continuous on the interval ( -n,n) and

periodic with period 2n, the function
(7)
P * Q(x) = f" P(x - r)Q(r) dr
is periodic with period 2n and continuous. Let us call it the faltung integral
of P and Q on the interval (-n,n). We find that
(8) P * Q(x) = Q * P(x),

and that the faltung is an even function if P and Q are both even or both odd;
it is odd if either P or Q is even and the other odd.
Now let F2 and H 2 be the even periodic extensions, with period 2n,
of two functions F and H that are sectionally continuous on the interval
(O,n). By writing the faltung of F2 and H 2 on ( -n,n) as a sum of integrals and
introducing new variables of integration, we find that
" 4
(9)
F2 * H 2(X) = f o [F2(x - r) + F2(x + r)]H(r) dr = ~l l(x),
when O< x < n, where 1are theseintegrals involving F and H:
(lO)
11 = s: F(r)H(x - r) dr, 12 = f F(r)H(r - x) dr,
13= I:- x F(r)H(x + r) dr, 14 = f,,"-x F(r)H(2n - x - r) dr.

In view of Eq. (9) the convolution property (12), Seco121, takes the form
(11)
Similarly, if Fl and H 1are the odd periodic extensions of F and H, with
period 2n, it turns out that
(12)
Fl * H1(x) = I: [Fl(X - r) - Fl(X + r)]H(r)dr
= 11(X) - 12(x) - 13(x) + 14(x),
(13)
Fl * H2(x) = I: [F1(x.- r) + Fl(X + r)]H(r)dr
= Io" F(r)[H2(x - r) - H2(x + r)] dr i

I
I
Then froID the product properties (5) and (6) we find these joint convolution
properties:
(14) -2f.(n)hs(n) = Cn{F * H(x)},
(15) 2f.(n)hc(n) = Sn{F * H2(x)}.
PROBLEMS
1. Find these transforms, with the aid of properties of Cn:

n2
(a) Cn{x} = (-1~: - 1 if n = 1,2,..., Co(x)= 2;
2n n3
(b) Cn{(n - X)2} = "2n if n = 1,2,..., Co{(n - X)2} = 3;
(_1)n+1
(e) Cn{coskx} = ksink n r--- k 2
if k #: O,:tI, :t2,...;
e sinh en
(d) Cn{coshe(n - x)} = n2 + e2
if e#: O.
2. With the aid of transform (b), Probo 1, show that if

1
yc(n) = 4'
n when n = 1, 2, . . . , and Yc(O)= O,
then -1 n3 n 2 1 3 1 4
Cn
{yc(n)} = 45 - 6'X + 6X - 24nX .
3. If a #: Oand b #: O,derive the transforms
(a2 - b2)(_1)n cosh bx cosh ax
a =C -
( ) (n2+ a2)(n2+ b2) n { b sinhbn a sinhan}
3' h2 C -I
(-1)"
(b)
2a SIn an n { (n2 + a2)2 }
= (an cosh an + sinh an) cosh ax - ax sinh an sinh ax.
4. We noted in Probo 9, Seco120, that when y> O,
n sin x
-e-ny = S 2arctan (n = 1,2,.. .).
n n{ eY-cosx }
Use formula (2), Seco103, and its special case Co{F'(x)} = F(n:)- F(O)to prove that,
when n = O,1,2, . . . , and y > O,
if -1 00
n { n o eY-cosr }
(b) Differentiate with respect to y in formula (a), Probo 4, to show that
-n, - .!. cos x cosh y - 1

ne - Cn{n (cosx - cosh y )2} when y > 00
6. Use Cn to find the solution of the problem
Y"(x) - e2Y(x) = - F(x), Y'(O)= Y'(n) = O, (e #- O)
in the form 2e sinh en Y(x) = F2* H 2(X)where H(x) = cosh e(n - x), and reduce the
solution to the form
e sinh enY(x) = cosh ex f F(r) cosh e(n - r) dr
+ cosh e(n - x) I: F(r) cosh er dr.

7. If F and F' are continuous when O ;;;;x ;;;;n except at an interior point x = e of
that interval, where F(x) and F'(x) have jumps b and b', respectively, and if F" is sec-
tionally continuous, prove that
Cn{F"} = -n2fc(n) - F'(O)+ (-I)"F'(n) - bn sin ne - b' cos ne.
8. When p is not necessarily an integer, let us write
fc(p)= s: F(x) cos px dx, f.(p) = s: F(x) sin px dxo

When k is a constant, show that
(a) 2Sn{F(x)coskx} = f.(n + k) + f.(n - k);

(b) 2Cn{F(x) sin kx} = f.(n + k) - f.(n - k);
(e) 2Cn{F(x)coskx}= fc(n - k) + fc(n+ k);
(d) 2Sn{F(x) sin kx} = fc(n - k) - fc(n + k).
124 POTENTIALIN A SLOT
Let V(x,y) denote the electrostatic potential in a space bounded by the

planes x = O,x = n, and y = O,in which there is a uniform distribution of
space charge of density h/(4n). Then the function V satisfies Poisson's
equation
(1) v"x(x,y)+ V,,(x,y)= -h (O< x < n, y > O).
V=o h V=A
4r
Fig.104 o V=O "

(IT,O)
Let the planes x = Oand y = Obe kept at potential zero and the plane x =n
at another fixed potential V = A (Fig. 104). Then
(2) V(O,y)= O, V(n,y)= A (y > O),
V(x, O) = O (O< x < n),
(3)
IV(x,y)1 < M
throughout the region, where M is some constant.

The determination of V(x,y) is a type of problem that arises in the subject
of electronics. The function V here can also be interpreted as steady-state
temperatures in a semi-infinite slab containing a uniform source of heat with
strength proportional to h.
Our boundary value problem [(1) to (3)] is not adapted to either the
method of separation of variables or solution by Laplace transforms. The
finite Fourier sine transformation with respect to x does apply, because the
differential form in that variable is Vxxand the prescribed boundary values
on the intervalO < x < n are V(O,y)and V(n,y).
We write v.(n,y) = Sn{V(x,y)} and transform the members of Eq. (1),
using conditions (2), to find formally that
d2
-n2v.(n,y) - An( -1)" + dy2v.(n,y) = -hSn{1}.
We also write conditions (3) in terms of transforms. Thus the problem in
v.(n,y) becomes
d2v. 2
(n = 1,2,.. .),
dy2 - n v. = An(-1)" - hSn{1}
v.(n,O) = O, Iv.(n,y)1 < Mn.
The solution of that problem is
(4) V.(n,y )
= hSn{1} - 2An( -1)" (1 - e -ny)
n
= [h1 - ~;1)" + A(-l:+](1 - e-ny).

The formula for the potential can therefore be written

2 G()
(5) V(x,y) = -
nn=1
L vs(n,y)sin nx.
In addition to that form of the solution in terms of an infinite series we

can derive a closed form of the solution. Referring to Appendix B, Table
B.l, we find that
l - (-l)n - ~( - (-l)n+l ~
S
n
-1
{ n3 } - 2 n x
), Sn-l
{ n } =
n'
e-ny 2 sin x
S-
n
1
{
( - l)n+ 1 -
n } = - arctan
n eY + cos X ,
and
Sn-l{l - ~_l)n e-ny} = a(x,y),
where
2 sin x
(6) a(x,y) = - arctan -:-- (y ~ O).
n SInh y
From our formula for Sn- 1{.fs(n)/n2} it follows that
Sn -1
{ l - (-1)"
n3 e
-ny
} = U(
x,y ),
where
" x
f f
X
(7) U(x,y) = - (n - r)a(r,y) dr - (x - r)a(r,y) dr.
n o o
Our potential function is therefore
X A sin x
(8) V(x,y) = h -(n
[
2 - x) - U(x,y) ] + -n ( x - 2 arctan ey + cos x ) ,
where U(x,y) is given by Eq. (7) in terms of the function (6). Both inverse
tangent functions in volved here are known solutions of Laplace's equation;
consequently it is not difficult to verify that the function (8) satisfies all
conditions in our boundary value problem [(1) to (3)].
125 SUCCESSIVE TRANSFORMATIONS
A semi-infinite slab of finite thickness is initially at temperature zero, and

its base is kept at that temperature. One of its parallel faces is insulated and
the other is subjected to a prescribed constant and uniform inward flux of
heat. Units are chosen so that the boundary value problem in the temperature
function V(x,y,t) becomes
(O< x < n, y > O,t > O),
(1) V(x,y,O) = O, V x(O,y,t)= -1,
V x(n,y,t) = O, V(x,O,t) = O;
also IVI < Mt throughout the slab, for some constant M.
The presence of the differential form V xx together with prescribed
values of V x at x = O and x = n indicates that the finite Fourier cosine
transformation, with respect to x, can be used here. The form VI and a
prescribed value of V at t = O indicate the Laplace transformation with
respect to t. Let us begin with the cosine transformation.
The transform with respect to x,
(2) W(n,y,t) = Cn{V(x,y,t)} (n = O, 1,2, . . .),
satisfies this problem in partial differential equations:

(3) ~ = - n2W + 1 + J-Y,y, W(n,y,O) = W(n,O,t) = O,
where IW(n,y,t)1 < Mnt. The Laplace transform w(n,y,s) of W, with respect
to t, therefore satisfies a corresponding boundedness condition and the
conditions
d2w 1
(4) (s + n2)w - - =-, w(n,O,s) = O.
dy2 s
The solution of this problem can be written as
(5) w(nys) = 1 _! exp(-y~).

, , s(s + n2) s s + n2
With the aid of the inverse transform of S-l exp (- y,J;) and the
operational property on replacing s by s + n2, we find that
L-l ! exP(-y~) = En(y,t,)

{s s + n2 }
where y ~ Oand
1 y
(6) En(y,t) = (n = O, 1,2, . . .),
1o exp (- n2.) erfc 2y.;: d.
The inverse Laplace transform of the function (5) is therefore
1 exp ( - n2t)
W(n,y,t)= 2"
n - n? - En(y,t) (n # O,Y ~ O),
(7)
W(O,y,t) = t - Eo(y,t) (y ~ O).
The inverse cosine transform offc(n) = n-2,fc(0) = O,is!(x - n)2/n-

n/60 Thus Cn- 1{W} can be written
(x - n)2 n t Eo(y,t)
(8) U(x y t) =
" 2 - -6 + n- - - n
2 <X> exp (-n2t)
L
- -n n=1
[ n
2
]
+ En(y,t) cosnx.
An integration by parts in formula (6) shows that En(y,t) is of the order

of n-4 for large n, when t > Oand y > O. The series in our solution (8) can
be differentiated termwise, and the solution satisties all conditions in the
boundary value problem (1).
PROBLEMS
1 . A steady-state temperature function V satisfies the conditions
Vxx(x,y)+ y,y(x,y) = O (O < x < n, y > O),
V(O,y)= O, V(n,y)= A (y > O),
V(x,O)= B (O< x < n);
also V(x,y)is boundedoDerivethe formula

A 2A sin x 2B sin x
V(x,y) = -x ,
n - -n arctan e+cosx + -x arctan-=--
smyh
o
2. A boundedharmonicfunctionU satisfiesthe conditions

Uxx(x,y) + U,,(x,y) = O (O< x < n, y > O),
U(O,y)= U(n,y) = O, U(x,+O) = F(x) when O < x < 110
Show formally that u.(n,y) = f.(n)e-ny = 2f.(n)qc(n,y),where
e -1 { n } = ~ sinhy when y > O,

n qc(,y) 2n coshy - cos x
an even periodic function. of x with period 2110Use our joint convolution property to
represent U in this form, when y > O:
1 "
Ux = -
(,y)
r
i [F sinhy
- sinhy
2n o () coshy - cos (r - x) coshy - cos (r + x)Jd
ro
3. When F(x) = A sin x in Probo 2, note the values of u.(n,y) to show that
U(x,y)= Ae-Y sinx. Verifythat solutiono
4. Solvethis problemfor V whenO~ x ~ 1land y > O:
Vxx(x,y)+ Y,y(x,y) = G(x) (O< x < n, y > O),
V(O,y)= V(n,y) = V(x,+O) = O,
where V is bounded. Obtain these formulas for V:

2 00 1
V(x,y)= F(x) + - L "2g.(n)e-n, sin nx
11:n= 1 n
= F(x) - U(x,y),
where U is the solutionof Probo2 and
"
F(x) =
i x
o
(x - r)G(r) dr
X
i
- -11: o (n - r)G(r) dr.
5. Let V(x,y) satisfy the conditions (Fig. 105)

(O< x < n, O< Y < Yo),
V(O,y) = O, V(n,y)= 1, V,(x,O) = V(x,Yo) = O.
Derive the formula
x 2 00 (-l)n cosh ny sin nx.

V(x,y)= + ir n~l n coshnyo
YI v=o (1T',):)
=0 V=lo
Fig.105
VD
O - "
6. The steady-state temperature U(x,y) in a semi-infinite slab with insulated faces

Uxx + U" = O and IU(x,y)1< M when O< x < n,y > O,

Ux(O,y)= Uin,y) = O,U(x,+ O)= F(x).
(a) Obtain this formula for U(x,y):
"
= -1 sinhy sinhy
f F(r)[coshy - cos(x - r) + coshy - cos(x + r)Jd r.
"
U
21t o
(b) When F(x) = A, note the valuesof un,y)and deducethat U(x,y)= A.
(e) When F(x) = A cos x, note the values of un,y); thus show and then verify that
U(x,y) = Ae-' cos x.
7. The facesx = Oand x = n of a semi-infinite slab O ;;;;x ;;;;n, y ;;; O, are insulated
and the base y = O is kept at temperature V = O. A steady source of heat KQ(x) with
mean value zero, J~Q(x)dx = O, is distributed throughout the slab; thus Vxx+ Yy, +
Q = O,where V(x,y) is the steady-state temperature. If Vis bounded, derive the formulas
2 00 1
V(x,y) = F(x) L ,qn)e-n, cosnx
- -n n=1n
" = F(x) - U(x,y),

where F(x) = f;(X - r)Q(r)dr and U is the temperature function in Prob. 6 corres-
ponding to this function F.
8. Ir V is a bounded harmonic function in the semicircular region r < a, O < () < n,
whose values are prescribed as F() on the semicircIe and as zero on the bounding
diameter (Fig. 106),then
r2v,.,(r,() + rf;(r,() + V8(r,()= O, lV(r,()1 < M (r < a, O < () < n),
V(r,O)= V(r,n) = O, V(a - O, () = F(().
Derive this Poisson integral formula 1 for V:

n
1
i
V(r,() = 2n O F(a)[P(r, a - () - P(r, a + ()] da,
when r < a and O ;;;() ;;;n, where P is Poisson's kernel,

a2 - r2
P(r,() = a2 + r2 - 2ar cos
x Fig.106
9. In Prob. 8, if the condition on the bounding diameter of the semicircular region is

replaced by the condition that the normal derivative of V vanishes there,
V(r,O) = V(r, n) = O,
derive this Poisson integralformula for V:

1 "
i
V(r,() = 2n o F(a)[P(r,a - O)+ P(r, a + ()] da
(r < a, O ;;; () ;;; n).
10. The transverse displacement Y(x,t) in a stretched string with a steady transverse
force distributed along it satisfies the conditions
Y,,(x,t) = a2y':x(x,t) + F(x) whenO< x < n and t > O,
Y(O,t)= Y(n,t) = Y(x,O)= Y,(x,O)= O.
Find the solution, and verify it, in the form
1
Y(x,t) = 2a2[2G(x) - G1(x + at) - G1(x - at)]
where G1 is the odd periodic extension, with period 2n, of the function
x " x
G(x) = -
n o I
(n - r)F(r)dr -
o
(x - r)F(r)dr. I
I Cf. seco 101 of the author's "Complex Variables and Applications," 2d ed., 1960.
,
11. The ends of a stretched string are fixed at the origin and the point (n,O). The
string is initially at rest along the horizontal x axis, then it drops under its own weight.
Thus the vertical displacements Y(x,t) satisfy the equation Yr,= a2Yxx + g, where g is
the acceleration of gravity. If Q(x) denotes the odd periodic extension of the function
tx(n - x), where O < x < n, with period 2n, derive the formula
Y(x,t) = 2:2[2Q(X)- Q(x - at) - Q(x + at).
12. If a steady transverse load is distributed along a beam, the transverse displace-
ments Y(x,t) satisfy an equation
82y 84Y
af = -a2 8x4 + F(x).
If the ends x = Oand x = n are hinged so that Yand Yxxvanish there, and if the initial
displacement and velocity are zero, derive the formula
1 2 > f,(n) .
Y(x,t) = zG(x) - ~
a
L ~ cos n2at SInnx,
na 0=1 n
where G(4)(X)= F(x) and G(x) = G"(x) = Oat x = Oand at x = n.
13. Note that for the elementary problem
(O < x < n, t > O),
0(x,0) = 1, 0(0,t) = 0(n,t) = O,

the temperature function can be written
2 >
0(x,t) = L 8s(n,t)sin nx,
-n 0=1
where 8.(n,t) = So{l} exp(-n2t).
Use Fourier transforms to show that the problem
U,(x,t) = f(t)U xx(x,t) + g(t) (O< x < n, t > O),
U(x,O)= U(O,t)= U(n,t) = O,
a problem not adapted to the Laplace transformation, can be solved formally in terms
of 0 in the form
U(x,t)= L g(t)0[x, F(t,t)] dt,
where
F(t,t) = f f(r) dr.
14. Obtain the solution of the diffusion problem
U,(x,t) = Uxx(x,t) - h(t)U(x,t) + A (O< x < n, t > ( .,
U(O,t)= U(n,t) = U(x,O)= O,
in the fonn of a series. Also derive the fonn

'
A
i
U(x,t) = H(t) o H(t)0(x, t - 't) d't,
where 0(x,t) is the function described in Probo 13, and
H(t) = exp [E h(r)dr].

15. Derive the fonnula for the temperatures U(x,y,t) when
u, = UJCJC
+ Uyy + A (x > O, O < Y < 11:,t > O),
U(x,y,O)= U(O,y,t)= U(x,O,t)= U(x,1I:,t)= O,

and U(x,y,t) is bounded, in the fonn
i r
U(X,y,t) = A O0(y,'t) erf 2J"i d't,

X
where 0 is the function described in Probo 13.
126 A MODIFIED SINE TRANSFORMATION

In Seco115 we presented the modified finite Fourier sine transformation
(1) Sm{F(x)} = I: F(x) sin mx dx = f.(m)

t
where m = n - and n = 1,2,..., as an example of Sturm-Liouville trans-
formations. It resolves F"(x) on the interval (O,n)in terms of f.(m) and the
boundary values F(O)and F'(n), when F is of class C":
(2)
We introduced the odd antiperiodie exteflsion F3 of F with period 2n;
that is,
(3) F3(X)= F(x) when O < x < n,
F3(- x) = - F3(x), F3(x+ 2n)= - F3(x) forall x.
In terms of F3 the properties of Smrepresented by Eqs. (10), (13), and (14),
Seco 115, can be written
x+e
2
(4) - f.(m) sin me = Sm
m x-e
F3(r)dr ,
} {f x+y
2 lt
(5) -
m
f.(m)hs(m) = Sm
{fo H(y)
fx-y F3(r)dr dy} .
Table 2 summarizes properties of Sm and lists transforms of some
functions.
Table 2 Modified sine transforms' !,(n - !)
J.(m) (m = n - t,n = 1,2,...) F(x) (O < x < n)
F(x)
s: F(x) sin mx dx = Sm{F}
2 00
2 I J.(m) (m = n - t) L f.(m) sin mx

-1t 11-1
3 I -m2f.(m)+ mF(O)- (-I)"F'(n) F"(x)
1
4 I -mf.(m)(-1)"+ 1
f-x F(r) dr
x+e
5 2 .
I - J.(m) Sinme
m f x-e
F3(r) dr
6 I
1
2m f.(m)
2
r O
"
'F(r)dr+ x
x+y
rx F(r)dr
7 -mf.(m)h.{m) f O
H(y)
fx-y
F3(r) dr dy
8 -1
m
cosme O ir O<x<e
9 I --m (O<c<n)
{1 ir c<x<n
1
10 I _(_1)"+1 x
m2
sin me X ir O~x~e
11 I (O < e < n)
m { ir e~x~n
e
2
12 I
m3
6
13 I -( -1)"
m4
14 I
sinhex
(-lr+1
m2 + e2 coshen
15 m - ee""(-I)"
I
m2 + e2
k(-lr+1 sin kx
16 (k,p :!:t.:!:f,...)
m2 -p cos kn
17 1 (sin x/2)(cosh y/2)

e-my (y > O)
; cosh2 y/2 - cos2 x/2
1 2
18 I -e-my (y > O) -arctan -sin x/2
m n sinhy/2
1 For details see Secs. 115 and 126 and the problerns rollowing Seco 116.
369
Note that the modified cosine transformation Cmthat corresponds to

Smis not essentially different from Smbecause
(6) Cm{F(x)} = s: F(x)cosmxdx = (-lt+lSm{F(n - x)}.

Thus Cm transforms F"(x) in terms of the boundary values reO) and F(n).
That variation of boundary values is taken care of by Sm itself when we
replace the coordinate x by n-x.
127 GENERALlZEDCOSINE TRANSFORMS

The transformation
(1) Ch{F(x)} = f F(x) cos qnx dx = fc(qn)'

where qn (n = 1,2,...) are the po sitive roots of the equation
h
(2) tan q =- and h > O,
n qn
is the special case (Prob. 4, Seco 116) of the Sturm-Liouville transformation

on the interval (0,1) that has the property
(3) Ch{F"(x)} = -q/fc(qn) - F'(O)+ [hF(l) + F'(l)] cos qn'
The boundary value hF(l) + r(1) arises in applications involving elastic
supports or surface heat transfer.
Since qnsin qn = h cos qn, we find that
(1 h + sin2 q
IIcosqnxll2 = Jo cos2qnxdx = n.
Thus the inversion formula becomes
00 cos qnx
(O< x < 1).
(4) I
F(x) = 2h n=1 fc(qn)h + sm. 2 qn
From Eq. (2) we can see graphically that qn - nn -+ O as n -+ oo.

Let F2 denote the even periodic extension of F with period 2:
(5) F2(x) = F(lxl) if -1 < x < 1,
When O ~ Y ~ 1, the formulas
2fc(qn)cos qnY = Ch{F2(x + y) + F2(x - y)}
-2 sin qn I: F(x + 1 - y) sin qnx dx,
(6) sin qn
iy
o
F(x) sin qnx dx =h fl-yl e-hx
fl-yx ehrF(l - r) dr cos qnx dx
can be derived or verified by making several elementary manipulations.

From them we can find this kernel-product property:
(7)
where z =2 - x - y and
R(z) = f e-hrF(r) dr
if z < 1, R(z) = O if z > 1.
The corresponding convolution property for Ch is
(8)
in terms of this convolution of F and G:
(9)
XFG(x) = f G(y)[F2(x + y) + F2(x - y)] dy
- 2hG(1 - x) * F(l - x) * e-hx.
Here the asterisk denotes the convolution integral used with Laplace trans-
forms (Sec. 16); thus
(10) G(l - x) * F(l - x) * e-hx = e-hx s: ehtG(l - t) s:-t ehrF(l - r) dr dto
Unfortunately, our convolution is not a simple one.1

When y = 1, formula (7) becomes
(11)
fc(qn) cos qn = Ch{ F(l - x) - he-hx s: ehrF(l - r) dr} .
From property (3) we find in the usual way that
(12)
J;(q ) (1 1 r
~J = Ch{ Jo ( 1 + h - r) F(r)dr + Jo (r - x)F(r)dr} .
These transforms can be found in various ways:
(13)
(14)
1Another derivation of the property can be found in the author's paper, Generalized Finite
Fourier Cosine Transforms, Micho Matho Jour., vol. 3, pp. 85-94,1955-56.
A RELATED TRANSFORMATION
Since the above kernel cos qnx is an even function of X, the transformation
(15)
Ch{F(x)} = fl F(x)cosqnxdx =!c(qn)
of functions defined on the interval (-1,1), where qn (n = 1,2,...) are again
the positive roots of Eq. (2), is expressed in terms of Ch by the equation
(16)
It follows from property (3) that, when F is of class C" on the interval
-1 ;;;;x ;;;; 1, Ch resolves F"(x) in terms of the boundary values hF( -1) -
F'( -1) and hF(l) + F'(l):
(17) Ch{F"(x)} = -qn2!c(qn) + [hF(-l) - F'(-l) + hF(l) + F'(l)]cosq..

Since h > O,the kernel ofCh is the set ofall eigenfunctions oftheproblem
~"(x) = A~(X), h~( -1) - ~'( -1) = h~(l) + ~'(1) = O,

corresponding to the eigenvalues A = - qn2. The inversion formula for Ch is
(18) (-1 < x < 1).
128 A GENERALlZED SINE TRANSFORM'

The transformation
(1) Sk{F(x)} = s: F(x) sin PnX dx = !s(Pn),
where Pn(n = 1,2,...) are the positive roots of the equation

Pn
(2) and k> O,
tan Pn = - k
was used in Seco117 to solve a problem involving surface heat transfer. It
has the property
(3)
Sk{F"(x)} = - P/!s(Pn) + F(O)Pn+ [kF(l) + F'(l)] sin Pn,
and the inversion formula
C() sin PnX
(O< x < 1).
(4)

F(x) = 2k n=1 fs(Pn)k + cos 2 Pn
1 The transformations in Secs.126to 128are among the generalizedfinite Fourier transforma-
tions treated by Ida Roettinger (Kaplan) in Quart. Appl. Math., voL 5, pp. 298-319, 1947, and
in lour. Math. Physics, vol. 27, pp. 232-239, 1948.
The transformation has the further properties
(5) s -I J.(Pn) = 1- ~ (1 rF(r) dr + I (x

Sx
- r)F(r)dr,
k { p/ } ( 1 + k ) Jo
1 l l
(6) -Sk{F(x)}=
Pn f o
cosPnx
f x
F(r)drdx,
(7)
Sk{F(l - x) - ke-kxI: rF(l - r)dr} = sinPn f F(x)cos p,.xdx.
Let FI be the odd periodic extension of F with period 2:
= F(x) ifO < x < 1,

FI(x)
(8)
Fl(x) = - Fl ( - x) = FI (x + 2) for all x.
When O~ Y ~ 1, we can obtain the integration formula

x+y
2
(9) - J.(Pn)sin PnY= Sk Fl (r) dr
Pn {fx-y }
+ 2 sin Pn (y cos pnX
Jo f l
x+l-y
F(r) dr dx,
from which we find the kernel-product property

x+y
2
(10) - J.(Pn)sin PnY= Sk
Pn {fx- y Fl (r)dr + 2z R(z)}
where z = 2 - x - y and
R(z) = r e-krF(r) dr if z < 1,

R(z) = O if z > 1.
The corresponding convolution property is

2
(11)
p/'<Pn)g.(Pn)= Sk{XFG(X)}
when g.(Pn)= Sk{G(x)}and

. l x+y
(12) XFG(x) =
f o
G(y)
f
x-y
Fl(r) dr dy + 2G(1 - x) * F(l - x) * e-kx,
where the final term, the iterated convolution used with Laplace transforms,
is given by formula (10), Seco 127.
We note some particular transforms.
(13)
(14) Pn .
Sk{sin rx} = r SInPn
Sk{e-kx} = k2 + p/' r2 - Pn - .
PROBLEMS
1. Use the transformations Sm(Sec. 126) to solve for Y(x) when
d4y
dX4 - y = 2x, Y(O)= Y"(O)= Y'(n) = Y"'(n) = O.
Ans. Y(x) = sechn sinh x - 2x - sinx.

2. Let the steady-statetemperatures V(x,y)in a semi-infinitewall O;;;x ;;;n, y ;;;O,
satisfythe conditions
y"x(x,y) + y,y(x,y) = O, lV(x,y)1 < M (O< x < n, y > O),
V(O,y)= O, y"(n,y) = 1, V(x,+O) = O,
where M is some constant. Find the solution of that boundary value problem in the
forms
2 ex>(-1)'
L1 my- e-my sin mx
V(x,y)= x + -n ,= (m = n - !)
2 x
=x - -
I
n n-x
sin r/2 dr.
arctan --;--
( SInh Y/2)
3. A harmonic function V(r,O)in the semicircular region r < 1, O < O< n, satisfies
these conditions :
r2Y.r(r,0) + ry'(r,O) + J.ok,O)= O (r < 1, O < O< n),

V(r,O)= 1, J.o(r,n) = O, V(l,O) = O;
also, Vis bounded over the region. Derive the formula
V(r,O)= 1 - ~
n arctan 20 1-r'
sin0/2
4. Ir V(r,O)satisfies the conditions stated in Probo 3 in the part of the upper half plane
exterior to the semicircIe, the unbounded region r > 1, O < O< n, show that
V(r,O)= 1 - ~
n arctan 20 r-1
sin(0/2) (r > 1, O < O < n).
I
I
J
5. For the transformation Ch in Seco127 show that, if O ~ e ~ 1,
1 + -l - e if O~ x ~ e,
(a ) Ch- 1 --
cos qnc - h
{ q/ } 1 .
1 f
{ +-x I c~x~l;
if O < x < c,
ifc<x<1.
6. The longitudinal displacements Y(x,t) in an eIastic bar with an elastic support at

one end and constant pressure applied at the other end satisfy the conditions
Y,,(x,t) = a2Yxx(x,t) (O< x < 1, t > O),
Y.,(O,t)
= - B, Yx(I,t)= -hY(I,t), Y(x,O) = Y,(x,O) = O,
where h > O. Derive the formula
Y(x,t) = B 1 +
[ !h - x - 2h f
n=1
cos qnat) C~S}qnX) '
qn (h + sm qn) J
where ql' q2'... are the positive roots of the equation tan qn = h/qn' Also,use trans-
form (a) in Probo 5 to show that, when at ~ 1,
Y(x,t) = B(at - x) if x ~ at, Y(x,t) = O if x ~ ato
7. Use the transformation Ch in Seco127 to solve the following problem for steady-
state temperatures in a semi-infinite wall with heat transfer at one face into a medium
at uniform temperature A :
vxx(x,y) + Y,y(x,y) = O (O< x < 1, Y > O),
Vx(O,y)
= O, - Vx(l,y)= h[V(I,y)- A] (y > O),
V(x,O)= O, Y,(x,oo)= O (O< x < 1),
whereh > O. Obtain the solution in the form
V(x,y)= A 1 - 2h2
[
f cosqn~o~qnX exp(-;qnY>l,J
n= 1 h + sm qn qn
where tan qn = h/qn and qn > O.
8. Let the temperatures U(x,t) in a slab with surface heat transfer at one face, and
with heat generated within the slab, satisfy the conditions
U,(x,t) = Uxx(x,t) + F(x) (O< x < 1, t > O),
U(O,t) = O, - Ux(I,t) = bU(I,t), U(x,O) = O,
where b > O. Derive the formula
U(x,t) =
(
1 -~
1 + b) o f l rF(r)dr +
Ix
I (x - r)F(r)dr
~ f.(PJ sin p"x2 exp ( - Pn2t),

- 2b Lo..~
n=1 Pn b + cos Pn
where p, P2'... are the positive roots of the equation tan P. = - pJb and
J.(P.)= f F(x) sin P.Xdx.
129 FINITE EXPONENTIALTRANSFORMS E.{F}

A linear integral transformation of functions F(x) on an interval (-n,n) that
resolves F'(x) in terms of the cyclic boundary value F(n) - F( -n) is often
useful in solving problems in which x is an angle, such as the polar coordinate
e, or in finding periodic solutions of problems in differential equations. For
a transformation
T{F(x)} = f" F(x)I>(x..l)dx = f(..l)

we find that, when F and el>are of class C' (-n ~ x ~ n),
T{F'(x)} = Fc1>]':."
- f" dx = ..lf(..l)+
Fc1>' c1>(n,..l)[F(n) - F(-n)]
if I>and ..l satisfy the homogeneous conditions
(1) I>'(x) = - ..lc1>(x), I>(n,..l)= I>(-n).
This last equation is a periodic boundary condition on I>.The non trivial
solutions of problem (1) are I>(..l,x) = e-inx and ..l= in where n = O,Il,
I 2, . . .. Thus T is the finite Fourier exponential transformation
(2) (n = O,:t 1, :t2,.. .),

En{F(x)} = f" F(x)e-inxdx = fe(n)
with the basic operationalproperty
(3) E.{F'(x)} = inEn{F(x)} + (-l)n[F(n) - F(-n)].
When F is of class en on the interval - n ~ x ~ n, or even if F and F'
are continuous there and F" is sectionalIycontinuous, we find, by replacing
F and F' in formula (3),that
(4) En{F"(x)}= -n2fe(n) + in(-l)n[F(n) - F(-n)]
+( -l)n[F'(n) - F'(-n)].
The formula can be extended easily to transforms of derivatives of higher
order in terms offe(n) and cyclicboundary values of F and its derivatives.
The set of functions e- inx(n = O, Il, I2,...) is orthogonalin the
Hermitian sense on the interval (-n,n); that is, when m = O,:t 1, :t2,...,
(5) = O ifm i=n.

f" exp (-inx)exp (-imx) dx = f" ei(m-n)xdx
That set is the basis for the exponential form of the Fourier series on that
intervaJ.l In terms offe(n) that series representation of F(x) can be written as
1 n=m
(6)
F(x) = 2n l~ n];.m fe(n)einx
( - n < x < n), .
valid if F and F' are sectionally continuous over the interval and F is defined
as its mean value from the right and left at each point of discontinuity. Note
that the sum in Eq. (4) ineludes the term corresponding to n = O. The
equation is a formula for En-l{fe(n)}.
In case the infinite series of terms lfe(n)l, where n runs from - 00 to 00,
converges, then
(7) (-n<x<n);
that is, the sum of the series here exists, and so it is the same as the principal
value used in formula (6). In fact, whenever the series of terms lfe(n)1con-
verges, then series (7) converges uniformly to a continuous sum F(x) whose
transform is fe(n).
In terms of Fourier sine and cosine transforms
En{F(x)} = f [F(-x) + F(x)] cos nxdx
+ s: [F(-x) - F(x)]sin nx dx
where n = O,:t 1, :t 2,. . .. In particular, when F is an evensectionallycon-
tinuous function on the interval (-n,n), we have a useful formula for
obtaining exponential transforms from cosine tran~orms:
(8) if F( -x) = F(x) (n = O,:t 1, :t2,.. .).
When m = :t 1, :t 2, . . . , we find easily that

(9)
(n = O, :t 1, :t 2, . . .).
From formula (5) and the linearity of En it follows that

(10) if n = :t 1, :t 2, . . . , but
(11) En{F(x)+ A} = fe(n) if n .. O,

Eo{F(x) + A) = fe(O)+ 2nA.
I Churchill, R. V., "Fourier Series and Boundary Value Problems," 2d ed., seco43, 1963.
From property (3) we find that, when F is sectionally continuous,
(12) if n =FO.
130 OTHER PROPERTIES OF En

Let F" denote the periodic extension of F with period 21t,
(1) F,,(x) = F(x) if -1t < x < 1t, F,,(x+ 21t)= Fix) for all x.
In case F" is of cIass C' on every interval, or if F" is everywhere continuous
and F' is sectionally continuous over the interval (-1t,1t), then according
to property (3), Seco 129,
(2) (n = O, :t 1, :t 2, .. .).
Ir F" has a continuous derivative of order m everywhere, then
(3) (m = 1,2,.. .).
This kernel-product property for En{F} is easily found:
"+Y
(4) e-inYfe(n)= f-,,+yF(x - y)e-inXdx = En{F,,(x- y)}
for all real y. When y = 1t, it becomes
(5) [2"
(-1)nfe(n) = En{F,,(x- 1t)} = Jo F(x - 1t)e-inxdx.
Note that F,,(x - 1t)= F(x + 1t)if -1t < x < O,and F,,(x - 1t)= F(x - 1t)
if O < x < 1t. The corresponding convolution property is
(6) fe(n)ge(n)= En{XFG(x)}

where gin) = En{G(x)} and
(7)
XFG(x) = f" G(y)F,,(x- y)dy.
In terms of G and F directly, that convolution has the form
"+X "
(8)
XFG(x) = f-" G(y)F(x - y) dy + I"h G(y)F(x - y + 21t)dy
when - 1t < X < O; but when O < x < 1t,
XFG(x)= f~,,"h G(y)F(x- y - 21t)dy+ f"h G(y)F(x- y)dy.

In form (7)the convolution has the properties .

(9) X FG(X)= X GF(X) = X FG(X + 2n).
Table 3 lists transforms of several functions, transforms that can be
derived by using our properties of En including the relation (8), Seco 129,
between Enand Clnlwhen F is even. Sme ofthe properties of Enare included
in the table.
Table 3 Finite exponential transforms E.{F:
f.(n) (n = O, i: 1, i: 2, . . .) F(x) (-n < x < n)
r -n F(x)e-inXdx = En{F(x)}
F(x)
1 m
2 f.(n) (n = O, i: 1, i:2,...) -1'
2n..~ n=L-m f.(n)einx
3 inf.(n)+ (-I)n[F(7f)- F(-n)] F'(x)
4 J~(n - m) (m = i: 1, i:2,...) eimxF(x)
5 e-in'f.(n) F.(x -
y) (Sec. 130)
6 f.(n)g.(n)
f. G(y)F.(x- y)dy
7 o ir n,p O,f.(O)= 2n 1
8 o ir n,p i: l,f.(i: 1)= n cos x
9 o ir n,p i:1,f.(1) = -ni,f.(-I) = ni sin x
(-I)n
10 x
21ti - ir n,p 0,f.(0) = o
X+ n ir -n<x<O
11 2ni
n ir n,p 0,f.(0) = o { x-n ir o < x < n
(-1). 2 eCX
12 :- (e ,p O, - 1, - 2,.. .)
e - In 2 sinh en
e<.ec;" ir - n < x < O
13 2 sinhen (e2,p O,-1, - 2, . . .)
e - in { e-".e'" ir O< x < n
14
(-Ir
4n~n ir n,p 0,f.(0)= ~n3
(-I)n cosh ex
15 (e2 ,p O, -1, -2,...)
n2 + e2 2e sinh en
(_I)n+1 cos ex
16 (e,p O, i: 1, i:2,...)
n2 - e2 2e sin en
(_l)n+1 1 .
17 -(2x SIOx + cosx)
~ ir n,p i: I,f.(i: 1) = O 4n
18
1 1 - b2
blnl (-1 < b < 1, b ,p O)
2n 1 + b2 - 2bcos x
Applications of En are illustrated in the set of problems that follows.

Finally, we note that by writing x = (2n/c)t - n,
(10)
En{F(x)}= (-lt s: e-inw1G(t)dt
where J = 2n/c and G(t) = JF(Jt- n). The integral in Eq. (10) is the
exponentialtransformation of G(t)on the intervalO < t < c. Intervals other
than (-n,n) have been usedas a basisfor finiteexponentialtransformations.l
PROBLEMS
1. Ir F(x) is sectionally continuous over the interval (-n,n) andJ.(O) = O,prove that,
when n = :t 1, :t 2, . . . ,
K
f.(n) X X
-::r =
n
En
{f-n (r - x)F(r) dr - _21tf-n rF(r) dr} .
2. Use En with respect to t to find the periodic solution Y(t), with period 2n, of the
equation
Y'(t) - eY(t) + P(t) = O,
where P(t + 2n) = P(t) for all t, and e2 =f.O,-1, - 2, . . . .
'+2K
Ans. Y(t)(l - e-2<K) = e<t I i P(r)e-<r dr.
3. Ir the external force acting on the mass m shown in Fig. 15 (Sec. 29) is periodic,
the unit of time can be chosen so that the displacement X(t) of m satisfies an equation
mX"(t)= -kX(t) + P(t) where P(t + 2n) = P(t) and k > O.
Write Jo= jki;. When Wo =f.1,2,..., and P is sectionally continuous over its
period, obtain these formulas for the periodic displacements with period 2n:
X( t ) -- ~ ~ p.(n)
'-- 2 2e
in' -- f:K P(t + n - . t) COSJot dt
2mnn=_""WO-n 2mwosmwOn
'+2n
= (2mwosinwon)- f, P(r)coswo(t+ n - r)dr.
4. Let P(x) be a prescribed periodic function with period 2n such that P" is everywhere
continuous. Use En to find the periodic function Y(x) of period 2n that satisfiesthe
integral equation
f~K Y(x - r) cosh er dr = P(x) wheree > O.

Ans. 2eY(x)sinhen = e2p(x - n) - P"(x - n).
Kaplan, W., "Operational Methods for Linear Systems," chap. 4, 1962; Doetsch, G.,
Integration von Differentialgleichungen vermittels der endlichen Fourier-Transformation,
Mathematische Annalen, vol. 112, pp. 52-68, 1935.
5. If P(t) and Y(x,t) are both periodic in t with period 211:, derive the solution of the
problem
Y(O,t) = P(t), y"(1I:a,t) = O
in the form
6. In Probo 6, Seco85, let the unit of time be chosen so that the temperature V(x,t)
in the semi-infinite solid x ~ Osatisfies the conditions
V(O,t)= A sin t, V(oo,t) = O;
also V(x,t + 211:)

= V(x,t). Showthat the exponentialtransformve(x,n)vanishesexcept
when n = j: 1, and that
V(x,t) = A exp ( - p) sin(t- p)'

7. Heat is generated at a steady rate uniformlyalong the radius 9 = 11:of a thin,
circular plate r < 1, - 11:< 9 ~ 11:(Fig. 107)whose faces are insulated. If the edge
r = 1 is kept at temperature zero, the steady temperatures U(r,9) in the plate are such
that
r2 Urr(r,9) + rU,(r,9) + Uoir,9) = O (r < 1, - 11:< 9 < 11:),
1
-[Uir,1I:)-
r Uir, -11:+ O)]= A, U(r,1I:)= U(r, -11: + O), U(l,9) = O.
Also, U is bounded over the region r < 1, - 11:< 9 < 11:,and U and its partial deriv-
atives of the first order are continuous there. Show that, with respect to 9,
ifn *" j:1,
A
= -rlogr ifn = j:l.
2
Thus derive the formula
A r
[ - r9 sin 9 - (2- r log r) cos 9
U(r,9) = 211:1 -
co (-1)"
2 "~2 n2 - 1r"cos n9 ].
Fig.107
I
I'
8. If V is a bounded harmonic function interior to the circle r = a that assumes

values F(O)on the circle,then
r2 Vrr(r,O) + .y'(r,O) + V88(r,O) = O - 1t ~ O~ 1t),
(r < a,
V(a - O,O) = F(O) when - 1t < O< 1t;
also, V and its partial derivatives of first and second order are continuous everywhere
inside the circle and thus periodic in Owith period 21t. Show that, with respect to O,
when F is sectionally continuous. Thus derive Poisson's integralformula for the har-
monic function in the disk (cf. Probs. 8 and 9, Seco125),
V(r,O)= ~
x (a2- r2)F(IX) L. (r < a, -1t ~ O~ 1t).
21tI -x . . -. ua
12
Exponential Fourier Transforms
The exponential Fourier transformation is a prominent singular linear

integral transformation of functionsF(x) on the entire x axis. Conditions of
validity on its operational properties are severe. For most properties weuse
the requirement that F be absolutely integrable from x = <X)to x = 00 -
in order to ensure convergence or uniform convergence of improper integrals
that arise. Proofs and statements of some properties of the transformation
can be simplified by using Lebesgue integrals instead of our Riemann integ-
rals and requiring F to belong to certain classes of Lebesgue integrable
functions;1 but this calls for familiarity with the theory of Lebesgue integrals.
131 THE TRANSFORMATION E.{F}

Let F denote a real- or complex-valued function of a real variable x such that
F(x) is defined over the entire x axis. A linear integral transformation of such
1 See, for inslance, books by Wiener and Titchmarsh on the theory of Fourier integrals, listed
in the Bibliography.
383
L
functions is
T{F} = f:oo F(x)K(X,A.)dx = f(A).

Let us determine the kernel K and the domain of the parameter A so that T
transforms derivatives F' into products .if(A).
For the purpose of designing the transformation so that it has that
property, we shall assume that, for each fixed value of A.,the kernel K has a
continuous derivative K' with respect to x and that K and K' are bounded
for all x. We assume that the functions F are everywhere continuous with
sectionally continuous derivatives F'(x) on each bounded interval (a,b), that
F(:!:(0) = O,and that F is absolutely integrable from 00 to oo. Then -
f F'(x)K(x,A) dx = F(x)K(x,A)]~ - f F(x)K'(X,A) dx,
and the right-hand member here has a limit as a -. - 00 and b -. 00, so the
integral on the left has the same limit, and it followsthat
T {F'} = f:oo F'(x)K(x,A.)dx = - f:oo F(x)K'(X,A) dx.

Thus T{ F'} = Af(A) if K is a solution other than zero of the horno gen-
eous equation
K'(X,A) = - AK(x,A)
and if K satisfies our requirements of continuity and boundedness. We may
select the solution K(x,A.) = e-''x. This function is bounded for all real x
only if A = iexwhere exis any real number; then K' is also bounded with
respect to x. Thus T becomes the exponential Fourier transformation
(1) (- 00 < ex< (0),

E,,{F(x)} = Loooo F(x)e-iOtX dx = fe(ex)

(2) E,,{ F'(x)} = iexfe(ex).
If F' is everywhere continuous and F" is sectionally continuous, and if

F(:!:(0) = F'(:!:(0) = O and f~ ooIF(x)1dx exists, then
(3) f F"(x)e-i"X dx = [e-iOtX(F' + iexF)]~- ex2 f F(x)e-i"x dx

and, when a -. - 00 and b -. 00, we find that
(4) (- 00 < ex< (0).
..
EXPONENTIAL FOURJER TRANSFORMS [SECo 132 385
Likewise, the operational property (2) can be extended as follows:
Theorem 1 If F(x) is absolutely integrable from - 00 to 00, if its derivative

F(m-l) is everywhere continuous while F(m) is sectionally continuous
on each bounded interval, and if
F(:too) = F'(:too) =... = F(m-l)(:too) = O,

then
(5) (m = 1,2,. ..).

In case F has a jump discontinuity at a point x = e but otherwise
satisfies our conditions for the validity of property (2), we find that
(6) E..{F'(x)} = iafia) - e-iCZC[F(c
+ O)- F(c - O)].
The literature on exponential Fourier transforms contains minor varia-
tions in the choice of the kernel and in notation. The kernels eiax and
(21t)-teiax, for instance, are sometimes used.
132 THE INVERSE TRANSFORMATION
In Seco68 we stated conditions under which a function is represented by its

Fourier integral formula
1
f f
p o oo o
(1) F(x) = - lim e"'" F(-r)e-IIItd-rda (- 00 < x < (0).

21t(J-+00 -p -00
Since the inner integral here is.fe(a),the formula represents the inverse ofthe
exponential Fourier transformation:
1
f
(J o
(2) F(x) = - lim .fe(a)e"'"da = E..-l{.fe(a)}.

21t (J-+oo -(J
Sufficient conditions for its validity can be stated as follows, even if F is a

complex-valued function (Probs. 9 and 10, Seco133).
Theorem 2 Let F(x) be absolutely integrablefrom - 00 to 00 and let F' be

sectionally continuous over each bounded interval. AIso let F(x) be
defined at each point e of discontinuity as t[F(c + O) + F(c - O)]. Then
its traniform fJ..a)exists, and the inversionformula (2) is validfor all real x.
In casef J..a)is also absolutely integrable from - 00 to 00, or in case the
integral J~oo.fe(a)eiax da exists, formula (2) can be written
1 1
(3) F(x) = -
21t f oo
-00
.fe(a)eiu:da = - E-A.fe(a)}
2 1t
(-00 < x < (0).
By replacing a by - a in the integral here, we can write

1
(4) F(x) = 2nExCfe(-a)}.
Thusunderappropriateconditionson F andfe the inverseof the operator

Ea can be written in terms of Ea itself.
That the principal value of the improper integral E_x{f.(a)} in the
inversion formula (2) may be needed is iIIustrated by the following example.
Consider the function F such that
F(x) =O if x < O, F(x) = e-X if x> O, F(O)= t,
which satisfies the conditions in Theorem 2. Here F'(x) = - F(x) when
x =FOand, according to formula (6),Seco131,
Ea{F'} = - fe(a) = iafe(a)- 1;
thus fe is not absolutely integrable from - 00 to 00 since
1 1 - ia 1
and lfe(a)1= :===--;,.
fe(a) = 1 + ia = a2 + 1 1 + a2
When x = O, the integral E-Afe(a)} does not exist because (a2 + 1)-1 is
integrable from - 00 to 00 while a(a2 + 1)-1 is not. But
- i fJ a - i fJ a
- f~da
21t - fJa + 1
=O and lim -
fJ-+00 21t f
- fJ a
--y--
+1
da = O.
'
therefore
ti
lim _21 f fJ
fe(a)da = 2 _1 f
oo
~
da
1
1
=_
2
= F(O).
fJ-+oo 1t -fJ n -ooa +
No two functions F and G that satisfy the conditions on F in Theorem 2
can have identical transforms. For if F and G are absolutely integrable from
-
- 00 to 00, then F - G is also, since IF GI ;;;IFI + IGI. Thus we see that
the function F - G satisfies all those conditions. But Ea{F - G} = O if F
and G ha ve identical transforms and, according to the inversion formula
(2), F(x) - G(x) = Ofor all X. In that cIass of functions, therefore, there is a
one-to-one correspondence between functions and their transforms.
Theorem 2 gives conditions on F that are sufficient for the validity of
inversion formula (2). We now establish sufficient conditions on the func-
tion fe(a) instead, conditions that also establish fe as a transform and as an
inverse transform; but one of our conditions is apt to be difficult to apply.
Theorem 3 Let a function f(a) have a sectionally continuous derivative f'

over each bounded interval, and let f be absolutely integrable from - 00
EXPONENTIAL FOURIER TRANSFORMS [SECo 132 387
to OO.Then its transform ExU} exists. We write
(5) ( - 00 < x < (0).

27t4>(x) = ExU(a)} = f~oo f(-r)e-irxd-r
lfin addition 4>(-X) has a transformfor all real a, then
(6)
and,Jor all real x, the inversionformula

1
(7) 4>(-x) =-
27t I oo
-<X>
f(a)eixda = -E_xU(a)}
1
27t
is valid. Also,Jrom Eq. (5) it follows that

(8) EU(x)} = 27t<l>(a).
I
Under the conditions on f the integral (5) converges uniformly to a
function 27t4>(x)that is continuous for all real x. Since the integral Ea{4>(- x)}
exists, by hypothesis, we can write it as E_{(x)}, or in the forms
oo . 1 oo . oo .
f - <X>
(x)el~
dx = -27t f - <X>
elx f- IX)
f(-r)e-ItX d-r dx.
But the right-hand member here representsf(a) for all real a, according to the
Fourier integral formula (1), sincefsatisfies the conditions ofvalidity ofthat
formula. Thus
f(a) = f~oo<1>(
-x)e-ix dx = E{(
-x)}.
Equation (5) shows that 4>(-x) is represented by the inversion formula (7).
To illustrate Theorem 3, let us find the inverse transform of
f(a) = exp (-Ial),

a function that is everywhere continuous, absolutely integrable from - 00
to 00,for whichf' is continuous except for a jump at the point a = O.For that
function
The integrals here can be written as simple Laplace transforms (s = 1),and

we find that
1 1
(x)= --z = (-x).
7t1+x
Since ct>is absolutely integrable over the x axis, Ea{ct>}exists and
(9) exp ( -Ial) = Ea

{! ~ }.
nl+x
AIso, according to formula (8)

2
(10) Ea{exp (-Ixl)} = 1 + '
133 OTHER PROPERTIES OF E.

We now point out several properties of the exponential Fourier transforma-
ton that follow readily from properties of integrals noted in Secs.14, 15, and
56. It is convenient to assume throughout this section that F(x) is seetionally
eontinuous over eaeh bounded interval on the x axis.
Suppose that, for all real a, the integral
Ea{F(x)} = f~", F(x)e-iax dx = fe(a)

exists. Then if e is real and constant, we find that
(1) E.{F(x - e)} = e-ic.fe(a),
(2) Ea{F(-x)} = f( -a),
(3) if e ::f=O,
Ea{F(ex)} = 1~lfe(~)
(4) Ea{eicxF(x)} = fe(a - e).
AIso the complex conjugates offe and F are related:
(5)
Ir F is an even or an odd function, its transformfe is even or odd, respectively,

andfe can be written in terms ofFourier cosine or sine transforms (Chap. 13):
(6) if F(-x) = F(x),

fe(a) = 2 1'" F(x) cos ax dx = fe( -a)
(7) if F(-x) = -F(x).

fe(a) = -2i 1'" F(x) sin ax dx = -fe( -a)
Property (1) is the kernel-produet property of Ea that we noted in Probo 3,
Seco110.
Suppose now that F is absolutely integrable from - 00 to oo. Then the
integral Ea{F} converges uniformly with respect to a, and we find that
(8) fia) is eontinuous for all real a.
Also,fe is bounded; in fact, the Riemann-Lebesgue theorem noted in Probo 8,

Seco65, shows that
lim firx) = o.
a-;t co
In case IxF(x)1 is also integrable from - 00 to 00, then

(10) f~(rx) = E~{ - ixF(x)}
since the member on the right represents a uniformly convergent integral
(Sec. 15).
Example Ir F(x) = exp (- X2),then
fe(rx) = f:ex> exp [- (X2 + irxX)]dx
= exp ( - :) f:ex> exp ( - r2) dr = ~ exp ( - :),

since the integral of exp ( - Z2)over each horizontalline z = x + irx/2in
the complex plane is equal to the integral over the real axis. Thus
(11)
E~{exp(_X2)} = ~ exp ( - :).
Then from properties (3) and (10) we find these transforms:
(12)
E~{exp ( - C2X2)} = ; exp ( - ::2) (e real, e =1=
O),
E~{xexp(_X2)} = -if rxexp ( - rx:).
PROBLEMS
1. If F(x) = 1 when -b < x O,we found in Seco132that
1 l-ia.
E{e-XSo(x)}= 1 + ia = 1 + 11.2
Use properties of E. in Seco133 to deduce that
ife > O;
1 .
(b) E.{xe-XSo(x)} = (1 + io,? ,
1
(e) E.{ e"So(- x)} = 1 - ia'
3. Use transform(10),Seco132,of exp(-Ixl) to prove that
2e
(a) E.{exp(-elxl)} =--r-2
e +a ife > O;
2 e-ic. if e is real;
(b) E.{exp (-Ix - el)} = 1 + a2
-4ia
(e) E.{xexp(-Ixl)} = (1 + (2)2'
4. From transform(9),Seco132,of (1 + X2)-1 deducethat
(a) E. a 2 +a x 2 = n exp(-alal) ifa > O;

{ }
(b) E. = nexp(-alal- ica) if a > Oand e is real.

{(x-e ~2 +a 2 }
5. Apply Theorem 3 to the function f(a) = exp (-(2) to establish the transform
(a) 2fiexp(-a2) (- x;)};

= E.{exp
then deduce that
(b) fi exp ( - ~) = E.{exp(- X2)}.

6. (a) Show formally that
E.{xF'(x)} = - ;a[a/.(a)].
(b) Apply E. formally to find a solution Y, for which E.{ Y} exists, ofthe ditTerential
equation
2Y"(x) + xY'(x) + Y(x) = O.

Verify that solution. Ans. Y(x) = A exp(-x2f4).
7. Ir a > O and
F(x) = COS2ax and F(x) = O

whenalxl ~ I' whenalxl~ I'
show that F"(x) = 2a2[1 - 2F(x)] when alxl < 1t/2and hence that
8. When c > Oand F(x) = Oir Ixl> c while

F(x) = -1 ir -c < x < O, and F(x) = 1 ifO<x<c,
showthat
2
E{F(x)} = -;-(1
IIX - cos CIX).
9. Write F(x) = U(x) + iV(x) where U and V are real-valued functions of x

(- 00 < x < (0). Let F be sectionally continuous on each bounded interval of the
x axis (Sec. 56). If F is absolutely integrable from - 00 to 00, point out why U and V
also have that property.
10. Theorem 2, Seco132, gives conditions under which a function F is represented
by the Fourier integral formula, conditions that were stated in Seco68 when F is real-
valued. With the aid of the observation in Probo 9, show why Theorem 2 applies when
F(x) = U(x) + iV(x) where U and V are real-valued functions.
134 THE CONVOLUTION INTEGRAL FOR E
The kernel-product property (1), Seco133, for eaeh fixed real t is

e-itl'fe(rx)= E{F(x - t)}.
As noted in Probo 3, See. 110, the eorresponding eonvolution of two fune-
tions F and G with transforms .fe(rx)and girx) is
(1)
XFG(x) = t: F(t)G(x - t) dt = F * G(x)
(-00 < x < (0).
Note that the eonvolution or faltung operation here, F * G(x), is not the one
used with Laplaee transforms sinee the integral has limits - 00 and 00 rather
than O and X. Both eonvolutions are the same in case F(x) = G(x) = O
whenever x < O.
Before establishing eonditions on F and G under whieh Ea{XFG} =
Je(rx)ge(rx),
we need some properties of the operation F * G and information
on the nature of the function X FG(X).
We assume in this seetion that our functions F(x) and G(x) are sectionally
continuous on each bounded interval of the x axis, and that they are bounded
and absolutelyintegrableJrom - 00 to oo.
Sineea eonstant M exists such that lG(x - t)1< M, then IF(t)G(x - t)1<
MIF(t)l. But MIF(t)1is independent of x and integrable from - 00 to oo.
"
Hence the convolution integral (1) converges absolutely and uniforrnlyfor all
real x. Moreover,
f F(t)G(x - t) dt = r~: G(y)F(x - y) dy

and since the first integral tends to XFG(X)as a -+ - 00 and b -+ 00, then the
second one has the same limito It follows that the operation F * G is commu-
tative :
(2)
XFG(x) = f: G(y)F(x - y) dy = XGF(x).
If Fl and F2satisfyour conditions on F, it is clear that

(3) [Fl(X) + F2(X)]* G(x) = Fl * G(x) + F2 * G(x).
To establish the continuity of XFG(X),first assume that G(x) is con-

tinuous on an interval Ixl ~ 2e. Then G(x - t) is continuous in x and t
together when - e ~ x ~ e and - e ~ t ~ e. But F(t) is continuous over
the intervalltl ~ e except possibly for a finite number of jumps at points ti'
Thus the integral
(4)
le(x) = fe F(t)G(x - t) dt
can be written as a sum of integrals from -e to tI' tI to t2, etc., whose in-
tegrands are continuous in x and t. ConsequentIy le is continuous when
- e ~ x ~ e.
Next let G be the unit step function So(x - k) where - 2e < k < 2e.
Then G(x - t) = So(x - t - k) = 1 ift < x - k, and it vanishesift > x-k.
If - e ~ t ~ e, then t ~ x - k when e ~ x - k, and so
if x ~ k + e.
le(x) = fe F(t)So(X- t - k)dt = fe F(t)dt
Since So(X - t - k) = Owhen t > - e> x - k, then
if x ~ k-e.
Thus IAx) has constant values on each of the half lines x ~ k + e and
x ~ k-e. On the remaining interval of the x axis, k - e < x < k + e, we
note that both t and x - k lie on the interval (- e,e),and thus
e "-k
f
le(x) = -e F(t)So(x- k - t)dt = -e F(t)dt f
(k - e ~ x ~ k + e).
This continuous function of x assumes the above constant valuesat the end
-
points k e and k + e of the interval. Therefore le is continuous for all x
when G(x) = So(x - k).
Finally, let G be sectionally continuous over the interval (- 2e,2e), with

jumpsjl,j2,...,jn at points k1,k2,...,kn (k1 < k2 <... <kn) Then it is
c1ear graphically that the function
Gn(x) = G(x) - j1So(x - kd - ... - jnSo(x- kn)
is continuous on the interval when properly defined at the points x = k.
Thus G is a linear combination of the continuous function Gnand unit step
functions. It follows that the function le represented by the integral (4) is
continuous when - e ~ x ~ e if F and G are sectionally continuous over
each bounded interval.
The continuity ofthe convolution integral (1)for all x now follows from
the uniform convergence of that improper integral. We write
(5)
X FG(X) = fe F(t)G(x - t) dt + Re(x)
and the absolute value of the remainder RAx) can be made arbitrarily small,
uniformly for all x, by making e sufficiently large. Let x be fixed and its
increments .1x be such that l.1xl < b, where b is some positive constant, and
select e such that e > Ixl + b. Then
IXFG(x + .1x) - XFG(x)1~ Ile(x + .1x) - le(x)1+ IRe(x + .1x)1+ IRe(x)1

and the right-hand member can be made arbitrarily small by first taking e
large, then l.1xl small because le is a continuous function.
We ha ve now established these properties of F * G(x):
Theorem 4 Let twofunetions F and G be bounded and absolutely integrable

from - 00 to 00, and seetionally eontinuous over eaeh bounded intervalo
Then their eonvolution integral
(-oo<x<oo)
F * G(x) = J:CX}F(t)G(x - t) dt
converges absolutely and uniformly to afunetion XFG(x) that is bounded
and eontinuous for all real X. Also, F * G(x) has the eommutative and
distributive properties (2) and (3).
135 CONVOLUTION THEOREM

We now prove that the product of two transforms is the transform of the
convolution of the original functions.
Theorem 5 The eonditions on funetions F and G stated in Theorem 4 are
sufficient not only for the existenee of the transforms fe(a.)and ge(a.),but
also for the existence of the transform of the convolution
(1) (- 00 < X < (0),

X FG(X) = f~ 00 F(t)G(x - t) dt
and the transform of that convolution is the product fe(a)ge(a):
(2) ( - 00 < a < (0).

Also, X FG(X)is absolutely integrable from - 00 to oo.
The existence and even the continuity of the transformsfe and ge were
noted earlier (Sec. 133).
Since the functions IFI and IGI satisfy the conditions on F and G in
Theorem 4, the convolution IFI * IG(x)1is a continuous function
(3)
Q(x) = t: lF(t)IIG(x - t)1dt
( - 00 < x < (0),
and the integral here converges uniformly with respect to x. With an in-
version of order of integration (Sec. 15),we can write
rb Q(x) dx =
Jo f-00
oo IF(t)1 rb IG(x
Jo
- t)1dx dt.
That integral is bounded for all positive numbers b, since
s: Q(x) dx = {0000IF(t)1 f:' IG(Y)I dy dt
~ f~oo IF(t)1dt f~oo IG(Y)Idy.
It is a monotone nondecreasing function of b because Q(x) ~ O. Therefore it

it has a limit as b -> oo. Likewise
or
a~i~ooLO Q(x) dx, f: 00 Q(x) dx,
exists, and hence the improper integral
(4)
f~ 00 Q(x) dx = f~ f~ 00 00 IF(t)IIG(x - t)1dt dx
converges. Since le-iuXFG(x)1 ~ Q(x) for all a, it follows that the Fourier
integral
(5)
converges absolutely and uniformly with respect to tl. In particular, X FGis

absolutely integrable.
To prove that the transform (5) is fJ..tl)gJ.tl), we may use the kernel-
product property
(6) e-i"'ge(tl) = E,,{G(x - t)} (-00 < t < 00),
which is satisfied for each fixed t under our conditions on F and G. Then
fe(tl)gJ.tl) = f~", F(t)e-i"'ge(tl) dt
= f:", F(t)E,,{G(x - t)} dt.
The integral here is the limit as e -+ 00 of the integral
(7)
fe f:", F(t)G(x - t)e-i"X dx dt.
Here the integral with respect to x converges uniformly in t when Itl < e
because F is bounded, IF(t)1 < M, and the remainderfrom x = b to 00
I L'" F(t)G(x- t)e-iaxdx = L~, F(t)e-"'G(y) dy I

I I
~ M L~e IG(Y)Idy.
The last member is independent of t and vanishes as b -+ 00; similarly for

the remainder from - 00 to x = a as a -+ - oo.
It follows (Sec. 15) that the order of integration can be interchanged to
write the iterated integral (7) as
f:", q(c,x)dx
where q is the function
(8) q(c,x) = e-iax fe F(t)G(x - t) dt.

The integral here, and hence q, is a continuous function of x (Sec.134). We
have now shown that
(9)
fe(tl)ge(tl)= !~~ f~", q(c,x) dx.
We note that q(c,x)-+ e-i"xXFG(x) as e -+ 00, and that limit is uniform with
respect to x because the improper integral XFGconverges uniformly for all x.
Since
Iq(c,x)1~ fe lF(t)G(x - t)1dt ~ f~oo IF(t)G(x - t)1dt = Q(x) -

and Q is integrable from - 00 to 00, then the integral in Eq. (9) converges
uniformly with respect to e (e > O). Our earlier observation [Eq. (5)] that
Ea{XFG} exists completes the facts that will enable us to interchange the
limit and the integral in formula (9).
To prove that
lim
c-+oo foo
-00
q(c,x)dx=
f oo e-iaxXFG(x)dx,
-00
we write the difference of the two integrals here in the form
(10)
f [e-iaxXFG(X) - q(c,x)]dx + foo e-iaxXFG(x)dx
-
fa
-00
q(c,x)dx+ oo e-iaxXFG(x)dx - ooq(c,x)dx.
Jb Jb
The improper integrals here converge absolutely; those involving q con-
verge uniformly with respect to c. Thus their absolute values can be made
small, independendy of e, by taking a large and negative and b large and
positive. Then by taking e large and positive, the absolute value of the first
integral can be made small beca use the integrand tends to zero uniformly
in x as e ~ oo. This completesthe proof of Theorem 5 since it followsthat
fe(a)ge(a) = f:oo e-iaxXFG(x)dx = Ea{XFG(x)}.

The convolution property here was illustrated in Probo 4, Seco110. In
Probs.9 and 10,Seco110, we noted and illustrated a generalized convolution
property for Ea, one that represents the iterated transform h(a) of a function
H(x,y) of two variables as the transform of the function
Loo:(X - t, t)dt.
136 TABLES OF TRANSFORMS

Appendix e consists of a table that summarizes properties of Ea and lists
exponential Fourier transforms of several functions.
Tables listed in the Bibliography under the names Erdlyi (vol. 1) and
Oberhettinger give exponential Fourier transforms of some other functions.
AIso, chap. 5 of W. Kaplan's "Operational Methods for Linear Systems,"

1962, presents short tables of transforms of functions and generalized func- .
tions and a few additional properties of Ea'
Tables of Laplace transforms furnish a source of exponential Fourier
transforms. In correspondence with the Laplace transform of a particular
function, we can list a certain exponential Fourier transformo For if F(t) is
defined for all real t and has a Laplace transform L{ F(t)} = f(s) when
s = y + irx,then
f(y + irx) = f" e-(y+ia)IF(t)dt
= 5:00 e - alSo(t)e - ylF(t)

/ dt.
Thus, in terms of the Laplace transformf(s) of F(t),

Ea{So(x)e-YXF(x)} = f(y + irx).
137 BOUNDARY VALUE PROBLEMS

Problems in differential equations that are adapted to solution by means of
the exponen tial Fourier transformation are singular in that the variable,
called x here, with respect to which the transformation is made, ranges from
- 00 to oo. But they are singular boundary value problems with rather severe
boundary conditions since the formula for the transform of a derivative with
respect to x assumes that the function itself vanishes as x ~ :t oo. That
condition is more severe than the condition of exponential order as t ~ 00
used in initial-value problems treated by the Laplace transformation. Con-
ditions of integrability with respect to x from - 00 to 00 further restrict the
variety of problems to which Ea applies.
Of course, formal procedures may yield verifiable solutions of some
problems in which the implied conditions are not satisfied. But the variety
of problems that can be solved by using Ea is small compared with the
varieties that are adapted to solution by finite Fourier or Laplace trans-
formations.
Here and in the set of problems to follow we give some examples of
problems that can be solved readily by using Ea.
Example 1 Let F(x) be such that F' is continuous and F itself is absolutely
integrable, from - 00 to 00, and such that F(:t 00) = O. To find a
solution of the boundary value problem
Wx(x,y) + Jv,,(x,y) + 2yW(x,y) = F(x) (- 00 < x < 00,y> O),
(1)
W(x,O) = O, W(:t oo,y) = O,
we let we(oc,y)denote the exponential Fourier transform of W with

respecttoxandwritefe(ex) = Ea{F(x)}. Then,formally,Ea{Wx} = iexwe,
and the transform of problem (1) becomes
. d
(y > O),
(iex+ 2y)we(ex,y) + dy we(ex,y)= fe(ex)
(2)
We(ex,O) = O.
The solution of problem (2)in linear ordinary differentialequa-

tions of first order is
(3) we(ex,y)= exp(-y2) I: !e(ex)eia(t-y)exp (t2)dt.

According to property (1), Seco133, the inverse transform offe(ex)ei'x(t-
y)
is F(x + t - y). By formally inverting the order of the operators Ea-1
and integration with respect to t, we can write
(4)
W(x y) = exp ( - y2) I: F(x - y + t) exp (t2) dt.
It is not difficultto verifythat the function (4)satisfiesall condi-
tions in problem (1).
Example 2 To obtain a formula for the bounded harmonic function
V(x,y) in the halfplane y > Othat assumes values F(x) on the boundary
y = O,we solve this problem:
Vxix,y) + Vyy(x,y)= O (- 00 < x < 00, y > O),
(5)
V(x, +0) = F(x), lV(x, y)1 < M,
where M is some constant and, over the entire x axis, the function F is
bounded and continuous except possibly for a finite number of finite
jumps.
To solve the problem using the transformation Ea with respect to
x, we shall assume some auxiliar y conditions on F, conditions associated
with our method of solution but not necessarily required in order that
the resulting formula for V(x,y) will be valido We assume that F(x) is
such that its transform f..(ex)exists and that the problem will have a
solution V(x,y) whose transform vJex,y)with respect to x exists and is
bounded when y > O for each fixed ex. Further conditions on V are
needed to write Ea{Vxx} = -rx2ve and to justify steps in transforming
problem (5) into the problem
2 d2
(y > O),
(6) - + dy2ve(rx,y)= O
ex vAex,y)
ve(rx,O)= fe(rx), Ive(rx,y)1< N(rx),
where N is independent of y.
The solution of problem (6) is

(7) V.,(a,y) = f.,(a.) exp ( - yla.l)
We found earlier that
exp ( - yla.1)= E", ! 2 Y 2 when y > O.

+y } { 1tX
Thus our convolution property indicates that
1 '"' yF(t)
(8) (y > O).
J
V(x,y) = -1t -'" (X - t)2 + y 2dt
That formal solution of problem (5) is known as the Poisson integral
formula for the half plane y > O,or the Schwarz integralformula. It can
be fully verified as a solution under the conditions on F first stated with
problem (5), without the auxiliary conditions.l
If F(x) is a constant Fo, for instance, it satisfies the conditions
stated in problem (5), butJ.,(a.) does not exist unless Fo = O. In that case
formula (8) reduces to
'"
(9) V(x,y) =
F.
: t - x
arctany
Fo 1t 1t
]-'" = -; 2 + 2 = Fo,
( )
i I
which is c1earlya solution of problem (5).

In case F(x) = So(x), formula (8) becomes
(10)
1 '"
i y dt 1 1t
2
V(x,y) = -1t o (t-x )2 +y 2 = -1t( _
X
+ arctan -
Y)
.
This function is bounded and harmonic in the half plane y > O; it

vanishes as y -+ +0 if x < O,and V(x,y) -+ 1 as y -+ +0 if x> O.
PROBLEMS
1. Let F(x) be sectionally continuous over an interval -c < x < c, while F(x) = O
when x < -c and when x > c. Use the exponential Fourier transformation to find
a solution of the differential equation
Y"(x) - Y(x) + 2F(x) = O (-00 < x < (0)
such that Y' is everywhere continuous and Y(i: (0) = Y'( i: (0) = O,in the form
Y(x) = e-X fe e'F(t)dt + e" [ e-'F(t)dt.

Verify that solution. Note that the first integral here vanishes when x < -c and the
second one vanishes when x > c, since F(t) = O when Itl > c.
1 For a full verification see the author's "Complex Variables and Applications," 2d ed., seco l.
108, 1960.
I
q
2. Use E. and the inversion formula to solve the problem
Vxx(x,y)+ Yy,(x,y) + x exp ( - X2) =O ( - 00 < x < 00, y > O),
V(x,O) = O, lV(x,y)1 < My when y > O,

where M is some constant, in the form
1 <X> . (2
V(x y) =-
2-ic o f
(1 - e-.y)sm ax
a exp ( -"4 ) da.
3. Let h(y) be continuous when y ;; O,and let F(x) have a continuous derivative F'
for all x and vanish as x --+i: oo. Derive the solution of the problem
Wx(x,y) + J-Y,(x,y)+ h(y)W(x,y) = O ( - 00 < x < 00, y > O),

W(x,O)= F(x), W(i: oo,y)= O,
in the form
W(x,y) = F(x - y) exp [- I: h(t) dtJ.

Verify that solution.
4. The transverse displacements Y(x,t) in a string stretched along the entire x axis
( - 00 < x < 00, t > O),
Y(x,O) = F(x), Y,(x,O) = O;
also F(x) --+O and Y(x,t) --+O as x --+i: oo. Transform with respect to x to derive the
formula
Y(x,t) = t[F(x - at) + F(x + at)].

5. Transformwithrespectto x to solvethisproblemformallyfortemperaturesU(x,t):
U,(x,t) = kU xx(x,t) (- 00 < x < 00, t > O),
U(x,O)= F(x).
Ans. U(x,t)= ~
2.jidZi f-
<X> F(~) exp
<X> [- (x - ~)2 d~
4kt J
O>
1
=
f
j1r - <X>F(x + 2r.jki) exp(- r2)dr.
13
Fourier Transforms on the
Half Line
Fourier sine and cosine transforms of functions defined over the half line
x > O were introduced in Chap. 10 as examples of general integral trans-
forms. They transform the differential form F" in terms of the transform of
F itself and the initial values F(O)and F'(O),respectively. Operational prop-
erties of the two transformations are noted here in greater detail and sum-
marized in Appendix D; some applications are given. AIso, we present a
modified transformation that resolves F" in terms of a boundary value
F'(O) - hF(O).
138 FOURIER SINE TRANSFORMS f.((1,)

In Seco116 we derived a linear integral transformation that transforms F"(x)
in terms of the boundary value F(O)and the transform of F when F is defined
over the positive x axis. It is the Fourier sine transformation
(ex> O),
(1) Sa{F(x)}= La:> F(x) sin exxdx = f.(ex)
401
whose basic operational property is

(2) Ser{F"(x)} = _(1.21.((1.) + (1.F(O).
The inverse transformation Ser-l{J.} can be expressed in terms ofthe trans-

formation itself:
2 2
(3) F(x) = -
non i OO
1.((1.)sin (1.Xd(1.= -S,,{f.(a)} (x > O).
The transformf.(a) exists as a continuous function of the parameter ex

if F is sectionalIy continuous over each bounded intervalO < x < e and
absolutely integrable from O to oo. Our conditions are sufficient, but not
necessary for the existence ofJ.(ex);for example, Ser{1/x}= n/2 when a > O.
Under those conditions it folIows from the Riemann-Lebesgue theorem
(Prob. 8, Seco65) that
(4) lim f.(a) = o.
er-+oo
The operational property (2) is valid if F" is sectionalIy continuous over

each bounded interval while F and F' are continuous when x ~ Oand vanish
as x -+ 00, and if F itself is absolutely integrable from Oto oo. The condition
that F be absolutely integrable can be replaced by the condition that J.(a)
exists for all positive a because integration by parts shows that, whenever
e> O,
(5)
s: F"(x) sin ax dx = -a2 s: F(x) sin ax dx
+ [F'(x)sinax - aF(x)cosax].
As e -+ 00, the bracketed term tends to (1.F(O);thus the integral on the left
has the limit Ser{F"}when the integral on the right has a limit.
By replacing F by F" in Eq. (5) and integrating by parts, we find that
(6)
under these conditions : F and its derivatives up to F'" are continuous when
x ~ Oand vanish as x -+ 00, F(4)is sectionalIy continuous on each bounded
interval, and J.(ex)exists; similarly for transforms of further iterates of the
operator d2/dx2.
Sufficient conditions for the inversion formula (3), the Fourier sine
integral formula, are that F be absolutely integrable from O to 00 and that
F' be sectionalIy continuous on each bounded interval. Then the formula
represents F(x) where F is continuous, and the mean value of F(x + O)and
F(x - O)where F has a jump.
FOURIER TRANSFORMS ON THE HALF LlNE [SECo 138 403
The sine transformation is a special case of the exponential Fourier

transformation. For if F1 denotes the odd extension of F to the entire x axis,
that is,
F1(x) = F(x) ifx>O, for all x,
then as noted in Seco133,
Eer{Fl(X)}= -2i {"> Fl(X) sin ocxdx.

Therefore
(7)
Some properties of Serfollow from properties of Eer. In particular, according

to Theorem 3, Seco132, the inversion formula (3) is valid iff~(oc)is sectionally
continuous on each bounded interval, if!s is absolutely integrable from O
to 00, and if Sx{fs(oc)}exists.
Property (2) is modified in case F and F' have jumps b and b', respec-
tively, at a point x = c:
(8) Ser {F"(x)} + ocF(O) +
= - oc2fs(OC) ocb cosocc - b' sin occ (c > O).
From property (2) and the inversion formula we find that
oc x 1t
Ser{e-X}= 1 + OC2' Ser{ 1 + X2} = "le-er.
From property (8) we find that the function
G(x) =c- x if O ~ x ~ c, G(x) = O ifx;;; c,
has this transform :
Ser {G(x)} = oc- 2(OCC- sin occ).
Under our conditions for the existence of fs(oc)we find that
(9) if k > O.
Ir in addition x2F(x) is absolutely integrable from Oto 00,we can differentiate

integral (1) with respect to (Xto see that
(10)
It is convenient to introduce the cosine transformation on the half line
before taking up further properties of Ser.
139 FOURIER COSINE TRANSFORMS 'c(a.)

In Probo 11, Seco116, we noted that the integral transformation that resolves
F" on the half line x > O in terms of the transform of F and the boundary
value F'(O)is the Fourier cosine transformation
(1) (a > O).

Ca{F(x)} = 100 F(x) cos ax dx = fc(a)
Its basic operational property is
(2)
The Fourier cosine integral formula furnishes an inverse transformation
Ca-1 in terms of CAfc(a)}:
2 00 2
(3) F(x) = - fc(a)cos ax da = -Cx{fc(a)} (x > O).
1t 5.o 1t
Sufficient conditions for the existence offc(a), or for the validity of the
inversion formula (3) or the operational properties of Ca, are those used for
the corresponding formulas for Sa in the preceding section.
We find, corresponding to property (6), Seco 138, that
(4)
Property (2) can be extended easily to formulas for the transform of higher
derivatives of F of even order. In case F and F' are continuous except for
jumps b and b', respectively, at a point x = c, we find that property (2) is
modified as follows:
(5) Ca{F"(x)} = -a2fc(a) - F'(O) - ab sin ac - b' cos ac (c > O).
We find that, corresponding to other properties of Sa,
(6) if k > O,
fc(ak) = ~Ca{F(I)}
(7) f;(a) = Ca{ -x2F(x)},
(8) lim fc(a) = O.
The relation between Ca and the exponential Fourier transformation (Sec.

133) can be written
(9)
where F2 is the even extension of F over the entire x axis.
Conditions of validity of the kernel-product property
(10) 2fc(a)cos ak = Ca{F(x + k) + F(lx - kl)} (k > O)
FOURIER TRANSFORMS ON THE HALF UNE [SECo 140 405
were noted in Probo 5, Seco 110. The corresponding convolution property

for Ca will be established in Seco 141.
140 FURTHER PROPERTIES OF S. AND C.

Some useful operational properties of the sine and cosine transformations
involve the two transformations jointly.
If F is continuous when x ~ O and F((0) = O,if F' is sectionallycon-
tinuous on each bounded interval, and if either F or F' is absolutely integrable
from O to oo~then upon integrating by parts between limits O and e and
letting e tend to infinity, we find that
(1) Sa{F'(x)} = -ocCa{F(x)},
(2) Ca{F'(x)} = ocSa{F(x)} - F(O).
If G is sectionally continuous and absolutely integrable from O to 00, then
the function
(x ~ O)
F(x) = {") G(r)dr
satisfies our conditions since F'(x) = - G(x) and F( (0) = O. Thus property
(1) can be written
(3)
By differentiating the integrals Sa{F} and Ca{F} with respect to oc,we

find that
(4) /;(oc) = Ca{xF(x)},
(5) /;((1.) = Sa{ -xF(x)},
assuming the sectional continuity of F and the absolute integrability of
xF(x) from O to oo.
Let Fl and F2 be the odd and even extensions of F:
X
F2(X) = F(lxl), when -00 < x < oo.
Fl (x) = IxlF(lxl),
Assuming the sectional continuity and absolute integrability of F, we find
these properties corresponding to the kernel-product property (10) of the
preceding section:
2fs(oc)sinock = Ca{F1(x + k) - F1(x - k)},
(6) 2fs(oc)cos ock = Sa{F1(x + k) + Fl(X - k)},
2fc(oc)sin ock = Sa{F2(x - k) - F2(x + k)},

where k is a real-valued constant.
/
406 SECo 141) OPERATIONAL MATHEMATICS
The existence of the trbnsforms!s and h for all positive a, and the
natural agreement that fs(-;a) = - fs(a) and fc(-a) = fc(a), enables us to
write the relations
fs(a + k) + fs(a - k) = Scz{2F(x)cos kx},
(7)
fs(a + k) - fs(a - k) = Ccz{2F(x)sin kx},
fc(a + k) + fc(a - k) = Ccz{2F(x)cos kx},
(8)
fc(a - k) - fc(a + k) = Scz{2F(x) sin kx}.
141 CONVOLUTION PROPERTIES

Let F and G be two bounded functions that are sectionally continuous on
each intervalO < x < e and absolutely integrable from O to oo. Then
where F2(x) = F(lxl) and G2(x) = G(lxl) when - 00 < x < oo. According to
Theorem 5, Seco 135, the product
is the exponential Fourier transform of the convolution
That integral is easily written in the form
which is an even function of x, so its exponential transform is twice its cosine

transformo
It follows that Cczhas the convolution property
(1)
2fc(a)gc(a) = Ccz{L'"F(r)[G(x + r) + G(lx - rl)]dr}.
Under the same conditions on F and G, and in the same manner, we
can establish these joint properties of sine and cosine transforms:
(2)
2fs(a)g.(a) = Ccz{L'" F(r)[G(x + r) - Gt(x - r)]dr} ,
FOURIER TRANSFORMS ON THE HALF LINE [SECo 142 407
(3)
2.fs(IX)gc(lX)
= S..{foa>F(r)[G(lx - rl) - G(X + r)] dr}
= S..{foa>G(t)[F(x + t) + F1(x - t)]dt}

where F1 and G1 are the odd extensions of F and G.
To establish a convolution property for the sine transformation alone,
we impose an additional condition on the function G, namely, that its
integral
(x ~ O)
H(x) = la> G(t) dt
be absolutely integrable from x = O to x = oo. Now H is continuous when

x ~ O,and H(oo) = O. According to property (3),Seco140,then
gs(lX)= IXC..{H(x)} = IXhc(lX)

and it follows from formula (3) that
2f.(IX)gs(lX)= 21Xf.(IX)hc(lX)
= IXs..{foa>F(r)[H(lx - rl) - H(x + r)] dr}
Thus we have the convolution property that was obtained formally in

Seco110:
2 a> x+r
(4) - f.(IX)gs(lX)= S..
IX {f.o F(r)
i Ix-rl
G(t) dt dr
}.
142 TABLES OF SINE AND COSINE TRANSFORMS
Appendix D consists of tables that summarize properties of Fourier sine

and cosine transformations on the half line and list transforms of particular
functions.
Some transforms can be found by evaluating integrals with the aid of
Laplace transforms (Chap. 3). The results shown in Probs. 14 and 15,Seco35,
for example, can be written as
(1) s..{c ~ x} = (~- SiIXC)COSIXC

+ CilXCsinlXc,
(2)
C..L ~ A = (~ - Si IXC)sin IXC- Ci IXCcos IXC,
--
where e > O. The evaluation of improper integrals by using contour integrals

and residue theory, illustrated in seco72 ofthe author's "Complex Variables
and Applications," 2d ed., 1960, is another method of deriving some of the
transforms listed in Appendix D. Other transforms can be found by direct
integration, by using operational properties of Sa.and Ca.,or by differentiating
or integrating a known transform with respect to a parameter.
Extensive tables of sine and cosine transforms will be found in vol. 1
of the "Tables of Integral Transforms" edited by A. Erdlyi. Those tables,
listed in tbe Bibliography, give transforms of several hundred special func-
tions. The book by Ditkin and Prudnikov, also listed in the Bibliography,
includes extensive tables of those transforms.
143 STEADY TEMPERATURES IN A QUADRANT

Let V(x,y) denote steady temperatures in a solid that fills the space x > O,
Y > O, - 00 < Z < 00 (Fig. 108). Ir the face x = O is insulated, if V(x,O)=
F(x), and if a steady source ofheat varying only with x is distributed through-
out the solid, then
(1) Vxx(x,y) + v,y(x,y) + H(x) = O (x> O,Y > O),
(2) J-:(O,y)= O, V(x,O) = F(x);
also, Vis to satisfy boundness conditions as x -+ 00 or y -+ oo.
To help keep temperatures steady, we assume that the total rate of
generation of heat throughout the solid is zero, that is,
(3)
{' H(x) dx = o.
Let us assume also that whenever x ~ O, the integral
(4)
P(x) = {>OI H(t) dt dr
exists and that P has a cosine transformo Since P"(x) = - H(x) and P'(O) = O,
the cosine transforms of P and H satisfy the relation
(5) pc(a) = h.(a).
y
...-;
--
Hb:)
O V=F(x) x Fig.108
FOURIER TRANSFORMS ON THE HALF LINE [SECo 143 409
Our problem in Vis adapted to the Fourier cosine transformation with

respect to x on the half line x > O, in view of the differential forro Vxx in
Eq. (1) and the prescribed boundary value Vx(O,y). Let vc(ex,y)denote the
transform Ca{V} where the transformation is made with respect to x, and
writefc(ex)= Ca{F(x)}. Then, formally,
d2
(6)
- ex2viex,y) + dy2 vc(ex,y)+ hc(ex)= O,
where y > O. The bounded solution of problem (6) is
vc(ex,y)= :2hiex) + [fc(ex) - :2hc(ex~ e-ay
or, in view of Eq. (5),

(7) (ex> O,Y > O).
From Appendix D, Table D.2, we see that
e-ay = C ~ y
a { 7t X2 + y2 }
and, according to the convolution property (1), Seco 141, it follows fram
Eq. (7) that
ex>
1 1
(8) V(x,y) = P(x) + -
y
7t Of [F(r)- P(r)J
[ (x+r )
2
+y
2+
(x-r
2
) +y ]
2 dr.
We note some special cases. First, if F(x) = P(x), then V is a function

of x alone:
(9)
V(x,y) = P(x) = {ex>L H(t) dt dr,
a result that is easily verified as a solution.
In case no source is present, then H(x) = O,and
I
1 Y Y
(10) fex>
[
V(x,y)= ; Jo F(r) (x + rf + y2 + (x - r)2+ y2 dr ] 11
when x ~ Oand y > O. That is an integralformula for the harmonic function
V in the quadrant such that Vx(O,y)= O and V(x,+O) = F(x). In this case
where H(x) = O, the problem can be solved also by using the sine trans-
formation with respect to y.
When H(x) = Oand F(x) = 1,wefindfromformula(10)that V(x,y) = 1,
and this is cIearIy a solution of the problem even thoughfc(x) does not exist.
ln cas.:: H(x) = D, and .F(x) ~ 1 W'h.::n O <' x <' e W'hil.:: F(x) - O ovhen
X > e, formula (10)gives the verifiablesolution

1 e + x e- X
V(x,y) = - arctan- + arctan- .
n( y y )
144 DEFLECTIONSIN AN ELASTIC PLATE

As noted in Seco 118, the sta tic deftections or transverse displacements
Z(x,y) in a thin elastic plate with no transverse load satisfy the biharmonic
equation
V2(V2Z) = V4Z =
where V2Z = Zxx + ZYY' The bending moment about an edge x = e is
proportional to Zxx(c,y), so if that edge is simply supported, then Z(c,y) =
and Zxx(c,y) = 0, and similarly for the other edges.
Let the plate have the form of a semi-infinitestrip x ;;;0, ;;;; y ;;;; n.
H there is no load and the edgesy = and y = n are simplysupported,
whilethe edge x = has a deftectionbsiny but no bendingmomentis
applied about the axis x = 0, then deftections Z(x,y) satisfythe conditions
04Z 04Z 04Z
(1) -+2-+-=0 (x > 0, < y < n),
OX4 ox20y2 oy4
(2) Z(O,y) = b sin y, Zxx(O,y) = 0,
(3) Z(x,O) = Zyy(x,O)= Z(x,n) = Zyy(x,n) = O;
also Z and Zx should tend to zero as x -+ oo.
The differential operators with respect to x in this problem are 02/0X2
and its iterate 04/0X4, on the half line x > 0, and the prescribed values at
the boundary x = are the values of Z and 02Z/0X2. Thus the sine trans-
formation with respect to x may apply. We write
zs(et,y) = S",{Z(x,y)};
then, formally,
FOURIER TRANSFORMS ON THE HALF LINE [SEC. 144 411
Then our boundary value problem in Z transforms into the following

problem in z.(a,y):
d4z. 2d2z. 4 b 3'
O)
dy4 - 2a dy2 + a z. = (2a + a ) SinY (O< Y < n, a > ,
(4)
d2z
z =-1=0 when y = Oand when y = n.
. dy
We find that A sin y is a particular solution of the differential equation here
if (1 + (2)2A = b(2a + (3), and that particular solution satisfiesthe condi-
tions at the edges y = Oand y = n. Thus
(5)
Referring to Appendix D, Table D.1, we see that

(6) Z(x,y) = b(l + tx)e-X sin y.
That function does satisfy conditions (1), (2), and (3) and it vanishes, together
with its derivatives, as x ~ oo.
Such problems in Z are also adapted to the finite Fourier transforma-
tion Snwith respect to y, or to successive transformations S. with respect to
x and Sn with respect to y.
PROBLEMS
1. Derive properties (6), Seco140.

2. For the function
G(x) = c - x when O~ x ~ c, G(x) =O when x ~ c,

prove that C.{ G(x)} = (1 - cos !XC)a-2.
3. The static transverse displacements Y(x) in a string stretched upon an elastic
support along the positive x axis, with its end x = Ofree to slide along the Y axis and
with a uniform force over the span O < x O),

Y'(O)= O, Y(<X)
= O,
where F(x) = A when O < x b, A, b, and h being positive
constants. Also, Yand Y' are to be continuous when x ~ O. Show formally that
A sin!Xb !X .
C.{ Y(x)} = - h2 ( ---;-- h2 + !X2 SIn!Xb )
and hence that
A(1 - e-bh cosh hx) when O ;; x ;; b,

-h2y(x) ={
Ae-h" sinh bh when x ~ b.
Verify that solution.
4. Use a transformation with respect to x in the boundary value problem
Y,,(x,t) = Yxx(x,t) (t > O,x > O),
Y.,{O,t)= -1, Y( oo,t) = O, Y(x,O) = y,(x,O) = O,
to derive the solution
t - x when O ;; x ;; t,
Y(x,t) = { O whenx ~ t.
5. Use a transformation with respect to x to find the solution of the boundary value
problem
(t > O,x > O),
Y(x,O) = II(x), Y,(x,O) = O, Y(O,t)= O, Y(oo,t) = O,
in the form (8), Seco43. Also, use superposition of the solutions of this problem and
Probo 6, Seco118, to write the solution when the initial conditions are
Y(x,O) = II(x), Y,(x,O) = G(x) (x > O).
6. Find the solution of the problem

Y,,(x,t) = Y",,(x,t)+ 2F(x) (t > O,x > O),
Y(x,O) = Y,(x,O)= Y(O,t)= O,
in terms of the odd extension Flor F, in the form
7. (a) Transform Laplace's equation Yx" + Y,p= O with respect to x to derive the
integral formula
y y
1 f'"
[
V(x,y) = -;)0 F(r) (r - X)2 + y2 - (r + X)2 + y2
dr
]
for the harmonic function in the quadrant x ~ O, Y > O, such that V(O,y)= O and
V(x,+O) = F(x).
(b) In case F(x) = 1 whenever x> O,then, even thoughf.(cx)does not exist, show
that the formula in part (a) gives the valid solution
V(x,y) = ~ y = 1- ~
1tarctan ~ 1t1m [Log (x + iy).
(e) Verify the solution given by the integral formula in part (a) when
F(x) = 1 if O < x < 1, and F(x) = O if x > 1.
8. The steady temperatures W(r,x) in a semi-infinite cylinder satisfy the conditions

1
J-Y,,(r,x)+ -r J-Y,(r,x)+ "Yxx(r,x)= O (O< r < 1, x > O),
W(l,x) = O, W(r,+O) = 1,
and W is to be continuous when O ~ r ~ 1 and x > O. Use a transformation with
respect to x to derive the formula
2 ">
Wrx)=-
(
, i[
1t o
1-- JO(iar) -
sin (XX
J o(ia)] a
d a = 1 --
i
2 ">Jo(iar) sin ax
--
1t o Jo(ia) a
da.
9. Temperatures in a semi-infinite solid satisfy the conditions

U,(x,t) = Uxx(x,t) + H(x,t) (t > O,x > O),
U(x,O) = F(x), U(O,t) = G(t).
Derive formalIy the solution

2 ">
where
U(x,t) = -
i
1t o
u.(a,t)sin (XXda
u.(a,t)exp (a2t) = f.(a) + {[aG(-r) + h.(a,t)) exp (a2T)dT,

f. and h. being sine transforms of F and H with respect to x.
10. (a) When G(t) = H(x,t) = Oin Probo9, obtain the formulafor U(x,t)in the form i :
given in Probo lO, Seco47. I
(b) When F(x) = H(x,t) = O in Probo 9, obtain the formula for U(x,t) in the form l'
derived in Seco47.
11. The static deflections Z(x,y) in a thin elastic plate in the form of a quadrant
04Z 04Z 04Z
(x > O,Y > O),
OX4 + 2ox2oy2+ oy4 = O
Z(O,y) = Zxx(O,y)= O (y > O),
bx
Z(x,O) = 1 + x2' Zyy(x,O) = O (x > O); I
also Z and its derivatives are to vanish as x -> 00 and y -> oo. Derive the formulas
I
b (">
Z(x,y) = '2Jo (2 + ay) exp [- (1 + y)et)sin ax da }
.
- bx + bx 1+ y .
- X2 + (1 + y)2 y [X2+ (1 + yf)2 I
The first form is helpful in verifying the solution. I
I
I
12. In Probo 11 let the conditions on the edge y = O be replaced by the conditions
Z(x,O)= O, (x > O),
then obtain the solution in the forros

b ""
Z(x,y) = -: f. o exp [-(1 + y)cx]sin cxxdcx
- b xy
- - x2+ (1+ y)2'
145 A MODIFIED FOURIER TRANSFORMATION T,.

We follow the procedure used in Chap. 10 to design a transformation
T {F(x)} = 1X> F(x)cf)(x,A)dx = f(A)

of functions F on the half line x > O that resolves F" in terms of f(A) and
-
the boundary value F'(O) hF(O),where the constant h is po sitive.
Suppose that F is of class C" and absolutely integrable over the half
line x ~ Oand that F(oo) = F'(oo) = O. If for each fixed ..1.the kernel <1>
and
its derivative <1>'
with respect to x are bounded and if <1>
is of class C", we can
integrate by parts from Oto e and let e tend to infinity to write
T{F"} = 1X> F"dx + [F' - F']O'
= i oo"
o F(x) (x,A) dx - /
F'(O) -
<1>'(0,..1.)
hF(O)
- h<l>(O)
F(O)
<1>(0,..1.)
I .
Thus if <1>andA satisfy the singular eigenvalue problem

(1) "(x) = A(x) when x > O, - h<l>(O,A)= O,
<1>'(0,..1.)
where <1>and <1>'are bounded, then
T{ F"(x)} = .lf(A) - (O)[F'(O) - hF(O).

A solution of that eigenvalue problem is ..1.= _0(2, where O(is real and
positive, and (x) = ..(x)
where
(2) ,,(x) = O(COS o(x + h sin <XX (O(> O).
Our transformation T is therefore the modified Fourier transformation

T..:
(3) (O(> O),


(4) 1;,{F"(X)}= -a2fy{a) - a[F'(O)- hF(O)].
From formulas (2) and (3) we see that 1'..can be expressed in terms of
Fourier cosine and sine transformations:
(5) 1;,{F(x)} = aC{F(x)} + hS{F(x)};
that is, fT(a) = afc(a) + hJ.(a).
Thus particular transforms and some properties of 1;,can be obtained from

tables and properties of C and S; in fact, property (4) can be verified by
writing the transformations there in terms of sine and cosine trans-
formations.
If we write
a 11:
(6)
O(a) = arctan h where 0< O(a)< 2'
then 41= Jh2 + a2 sin (ax + O),and 1;,takes the form
(7)
1;,{F(x)} = .Jh2 + 11.2100F(x) sin [ax + 0(11.)]
dx.
In Seco 140 we noted conditions on F under which
aC{F(x)} = -S{F'(x)}.
Under those conditions form (5) for 1;,can be written
(8) 1;,{F(x)} = fT(a) = S{hF(x) - F'(x)}.
Formally then we can invert the sine transformation to write
oo
2
(9) hF(x) - F'(x) = -
f
11: o
fT(a)sin ax da (x > O).
An inverse of the linear differential operation

(10) hF(x) - F'(x) = P(x)
is found by solving the differential equation here for F. The particular
inverse that is bounded when P is bounded can be written
(11) F(x) = ehx r

Jx
e-hrP(r) dr = (00e-htp(x + t) dt.
Jo
When P(x) is the right-hand member of Eq. (9), then
oo oo
2
F(x) =-
11:of e-ht fo fT(a)sina(x + t)dadt.
We invert the order of integration here and use the known Laplace trans-
I
form of sin a(x + t) with respect to t, with parameter h, to write this formula
for the inversetransformation T,.-l{JT(a)} :
f
(12) F(x) =~
nJo flX) fT(a) a cos ~~ + a.
h~sinax da.
(x > O).
The inversion formula (12) is valid when F satisfies our conditions

(Sec. 138) under which F(x) is represented by its Fourier sine integral
formula. 1
Note that if h = O,formula (12)reducesto the inversionformula for the
Fourier cosine transformation.
146 CONVOLUTION FOR r.

The following development of kernel-product and convolution formulas
for our modified Fourier transformation is formal. The formulas are more
complicated than those for the sine and cosine transformations, so they are
apt to be troublesome to apply. We illustrate in the following section how
our expression (8), Seco145, for T,.in terms of S" may be used to circumvent
our convolution formula for T,..
The expression just referred to can be written
(1) if P(x) = hF(x) - F'(x),
wherefT(a) = T,.{F(x)}. Now P(x) = -Q'(x) if
(2)
Q(x) = F(x) + h {IX)F(r) dr
and, since S,,{-Q'} = aC,,{Q}, we have this alternate representation of
T,.{F} :
(3)
To obtain a kernel-product formula, we write
~fT(a)~ik)
a = 2aqia) cos ak + 2hqJa) sin ak
1 The proof, with the aid of Fourier sine and cosine integral formulas, is not difficult even for
a somewhat more general integral formula. See R. V. Churchill, Generalized Fourier Integral
Formulas, Mich. Math. Jour., vol. 2, pp. 133-139, 1953-54.
j
where k > O,and use the kernel-product properties (AppendixD)

2qc(a.)cosa.k = Ccx{Q(lx - kl) + Q(x + k)},
2qc(a.)sin a.k = Scx{QUx- kl) - Q(x + k)},
to write
2
(4) -fT(a.)ik)
a. = J:{Q(lx- kl)} + a.Ccx{Q(x + k)} - hScx{Q(x+ k)}.
In formulas (1) and the definition offT we can replace h by - h and F
by Q.Thus
a.qc(a.)- hqs(a.) = -Scx{hQ(x) + Q'(x)}.
Ir we replace h by - h in the definition (2) of Q and write
R(x) = F(x) - h L~ F(r) dr,

we find that
hQ(x) + Q'(x) = F'(x) + h2 L~ F(r) dr = - hR(x) + R'(x). I (

I
Then, in view of Eqs. (1), I
-Scx{hQ(x) + Q'(x)} = J:{R(x)},

and the right-hand member of Eq. (4) becomes .
L
J:{Q(lx - kl) + R(x + k)} = J:{PF(x,k)}
where
PF(x,k)= FUx- kl) + h
JIx-kl
oo F(r)dr + F(x + k) - h
f oo
x+k
F(r) dr.
Qur kernel-product formula is therefore
(5) (k > O),

where
X+k
(6) PF(x,k) = F(lx - kl) + F(x + k) + h

JIx-kl F(t)dt.
The corresponding convolution (Sec. 109) for J: is
(7)
X FG(X) = 100 G(r)PF(x,r) dr
= [<X>G(r) F(lx - rl) + F(x + r) + h x+r F(t) dt

JIx-rl dr,
Jo [ ]
and if gT(a) = T.,{G(X)}, the convolution formula is
(8)
147 SURFACE HEAT TRANSFER

Let V(x,y) denote steady temperatures in a quadrant x ~ O,Y ~ O,
- 00 < Z < 00,whose two facesare subject to the linear law of surfaceheat
transfer. Ifthe medium outsidethe face x = Ois at temperature zero and that
outside the face y = O is at a temperature that varies with x, then
Vxx(x,y)+ y"y(x,y)= O (x > O,Y > O),

(1)
V",(O,y)= hV(O,y), y,,(x,O)= kV(x,O) - F(x);
also, V and F are to satisfy appropriate condtions of regularity. The con-
stants h and k are positive.
Our transformation T.,with respect to x applies here in view of the
differential form Vxx,on the halfline x > Oand a prescribed value of V", hV -
at the boundary x = O. If vT(a,y)= T.,{V(x,y)}and fT(a) = T.,{F(x)},then
problem (1) transforms into the problem
d2
(y > O),
-a2vT(a,y) + dy2VT(a,y) = O
(2)
dVT
- = kv - fT(a) when y = O.
dy y
The bounded solution of this problem can be written as
(3) VT(a,y) = fT(a) e-ay = fT(a.)Y

k+a. f y
oo e-(k+a)t d7:.
According to our inversion formula (12), Seco145, then
2 roo a.cos ax + h sina.x

(4)
V(x,y) = ; Jo fT(a)e-ay (k + a.)(h2+ a2) da.
That form of the solution of problem (1) can be verified.
To present another form ofthe solution, let us express the transforms in
terms of sine transforms. We write
(5) U(x,y) = hV(x,y) - Vx(x,y).
Then according to formula (1), Seco146,
(6)
SimilarIy,
(7) if P(X) = hF(x) - F'(x).

We use formula (11), Seco145, to recover V and F ftom U and P:
V(x,y) = {X' e-h~U(x + ~,y) d~,

(8)
F(x) = f" e-h~p(x + ~)d~

The second representation of Vr in Equation (3) can now be written in
the form
(9)
Since
2 t
e-at - -C
- 1t a{ t2 + X2 } '
it followsfrom the joint convolution property (3),Seco141,that
e-atps(ex)= ~Sa {IX' P(r)[t2 + (: - r)2 - t2 + (: + rfJ dr}.

Formally then, the inverse of the sine transform (9) is
oo oo
(10)
eky
U(x,y) = -1ty f f e-kt o P(r)[ t+x-r
2 (
t
)2 2 (
t+x+r
t
)2J dr dt.
Formula (8) then gives a representation of V(x,y).
The solution represented by Eqs. (8) and (lO) can be simplified and
verified in case F(x) = 1,in whichcase P(x) = h, eventhough the transforms
fT(ex)and p.(ex)do not exist.
14
Hankel Transforms
148 INTRODUCTION
Transforms produced by linear integral transformations whose kernels are

Bessel functions are called Hankel transforms.1 They are sometimes re-
ferred to as Bessel transforms. There is a great variety of those transforma-
tions, first because of the variety of Bessel functions, solutions of Bessel's
differential equation
(1)
containing two parameters J1and v. Furthermore, as in the case of Fourier

transforms, the transformation may be made over either a bounded interval
and involve various boundary conditions, or over a half line.
1 Friedrich W. Bessel (1784-1846), German aslronomer and mathemalician; Hermann Hankel
(1839-1873), German mathematician.
420
HANKEL TRANSFORMS [SECo 149 421
The origin is a singular point of Bessel'sdifferential equation. Ir v ~ O,

the solution of the equation that is continuous everywhere, including the
origin, is
(2) y(x) = CJv(jlx)
where C is any constant and Jv is Bessel's function of the first kind with the
series representation
<X) (-1}i t V+2j
(3)
Jv(t) = j~O j!r(v +j + 1)( 2) .
In the more common applications the parameter v is a nonnegative
integer. Then Eq. (3) becomes
t n <X) (-l)j t 2j
(4) (n = O, 1, 2, . . . ).
Jn(t) = ( 2) j~Oj!(n + j)! ( 2)
We restrict our attention to that case and first treat finite transformations,
then transformations on the half line x > 0.1
Useful operational properties of Hankel transformations are much
fewer than those for Fourier and Laplace transformations. But again,
applications to the solution of linear boundary value problems do not depend
on the homogeniety of differential equations or boundary conditions.
149 FINITE HANKEL TRANSFORMATIONS
We consider functions on a bounded interval extending to the right of the

origin and take our unit of length as the length of the interval, and this
differential form in such functions :
(1) D2F = F"(X) + .!.F'(x)

x - n:F(X)
x (O< x < 1),
where n is zero or a positive integer. Tht form arises when the Laplacian
operator V2 is written in terms of cylindrical coordinates; in that case our
variable x here is the radial coordinate usually called r or p (cf. Seco83).
In terms of a self-adjointform we write
1f n2
(2) D2F = x1[xF'(x)' - x-F(x)} (O< x < 1,n = 0,1,2,...),
so in the notation used in Seco 111, p(x) = r(x) = x and q(x) = -n2jx.
1 As noted in Seco 83, basic properties and representations of Jn are developed in the author's
"Fourier Series and Boundary Value Problems." For extensive treatments ofBessel functions
see G. N. Watson, "A Treatise on the Theory of Bessel Functions," 2d ed., 1944, or A. Gray,
G. B. Mathews, and T. M. MacRobert, "A Treatise on Bessel Functions and Their Applications
to Physics," 2d ed., 1952. For a shorter account see F. Bowman, "Introduction to Bessel
Functions," 1958.
Let us first find the integral transformation
(3)
T{F} = f F(x)X<l>(X,A)dx = f(A),
with weight function p(x) = x, that transforms D2Fin terms of A,j(A), and
the boundary value F(l). According to Eq. (8), Seco111, we can write
Thus if <1>
T {D2F} = f
and A are such that
F(X{ (x<l>')'- : <1>]dx + [x(F' - <1>'
Fm.
n2
(4) - -<1> and <1>(1)) = O,
(x<l>')'
x = AX<l>(X))
then
(5) T{ D2F} = Af(A) - '(l,A)F(l).
We assume, say, that <1>and F are of cIass C" on the interval O ~ x ~ 1.

It is convenient to write A = - J1.2.The differential equation (4) is
Bessel's equation with index n,
(6) x2<1>"
+ x<l>'+ (J1.2X2- n2)
= O.
Its solution of cIass C" (O~ x ~ 1) is <1> = Jn(J1.x),and the condition

(l,A) = Ois satisfied if J1.is such that Jn(J1.)= O. For each fixed n (n = 0,1,
2,. . .), the equation Jn(J1.)= Ohas an infinite sequence of roots, all real, and
only the positive roots need be considered beca use they correspond to the
full set of linearly independent eigenfunctions of the singular eigenvalue
problem (4). Let J1.j,or J1.j(n),denote those roots:
(7) Jn(J1.j)= O where j = 1,2,..., J1.j> O,and J1.jdepends on n.
Our transformation T then becomes the finite Hankel transformation
on the interval (0,1),
(8) HnAF(x)} = f F(x)xJiJ1.jx)dx = fH(J1.j)

(j = 1,2,.. .)

HnA D2F} = - J1./fH(J1.)- J1./~(J1.)F(l).
Here J~denotes the derivative of Jn with respect to its argument, so dJn(J1.jx)

dx = J1./~(J1.jx).Note that for each fixed n the transforrnfH(J1.)is a sequence
of real numbers.
In view of the known formula
(9) tJ~(t) = nln(t) - tJn+l(t)

and the fact that JnVJ)= O,it foIlowsthat J~(/l)= -Jn+l(J1.j)'Thus the
basic operational property of Hnj can be written
(10) Hnj{F"(X) + ~F'(X) - ::F(X)} = -/l/fH(J1.j) + /l/n + 1(J1.j)F(l),
when F is of cIass C" (O~ x ~ 1) and /l1, /l2,'" are the positive zeros of
J n(J1.).Numerical values of the first four zeros of J o(J1.)
were given in Seco83.
-
When n = O, D2[1]= O and D2[X2 1] = 4. Thus these particular
transforms foIlow from formula (10):
1
(11) HOj{l} = -J1(J1.j),
/lj
(12) HoA1 - X2} = ~Hoj{l} = ~Jl(J1.j),

/lj /lj
where Jo(/lj) = O and /lj > O.

Also from property (10) we find that
(13) Hnj{xn} = .lJn+l(J1.j),

/lj
/l'
(14) Hnj{Jn(cx)}= Jn(c) 2 J 2Jn+l(J1.j) (e > O),
/lj - e
where /lj are the positive zeros of Jn(/l) when n is some nonnegative integer,
and e is not one of those zeros.
150 INVERSION OF H.j
For a fixed n the set of functions Jn(J1.jx),where j = 1,2,..., is orthogonal

on the interval (0,1) with weight function x, and norms of the functions are
known. We can express that orthogonality in this form:
(i = 1,2,...)
(1)
ifj = i,
where /lb /l2,' .. are the positive zeros of Jn(/l).
For a function F on the interval the generalized Fourier series in terms
of the functions of that set is the Fourier-Bessel series
That series converges to F(x) on the intervalO < x < 1 if F' is sectionally
continuous there and if F is defined as the mean value of its limits from the
right and left at each point where F is discontinuous. Then we can write
00
(2) F(x) =2
j= t
[Jn+t(.uj)r2fH(.uj)Jn(.ujx) (O< x < 1),
wherefH is our Hankel transform of F corresponding to the zeros J.ljof Jn(.u).

Equation (2) is an inversion formula for the transformation Hnj'
Kernel-product and convolution properties are not known for finite
Hankel transformations. Some general properties of Sturm-Liouville
transforms noted in Chap. 10 can be extended to this singular case. The
basic formula for Hnj{D2F} when F or F' has a jump on the interval
O < x < 1, for instance, can be written by the method used to obtain formula
(7), Seco 113 (see Probo 3, Seco 152, for the case n = O). The following is
another such property. ~
IrfH(J.lj)= Hnj{F(x)},thenfH(.uj)/J.l/is the transform YH(J.lj)
ofa function
Y(x) such that D2Y = -F(x) and Y(l) = O,according to our basic opera-
tional property for Hnj' Thus
(3) x2Y"(x) + xY'(x) - n2y(x) = -x2F(x), Y(l) = O,
and Yand Y' are to be continuous and Y" sectionallycontinuous over the
interval O ~ x ~ 1. A particular solution of the homogeneousform of the
differentialequation here is xn. If we write Y = xnZ, problem (3)is reduced
to one of first order in Z'. In that way we find this formula for Y(x), the
inverse transform offH(.uj)/J.l/:
(4)
where J n(.uj)= O,J.lj > O,n = O, 1,2,. . . , a result that is valid if F is sectionally
continuous on the interval (0,1).
151 MODIFIED FINITE TRANSFORMATIONS Hnh
Ir our differential form D2F is to be reduced by the transformation on the

interval (0,1) in terms of a boundary value hF(l) + F'(1), where h ~ O, we
denote the transformation by Hnh' The special case n = h = O was treated
briefly in Seco 116.
Ir (x,A.)
is the kernel, with weight function x, then as in Seco 149 we
can write
i,
f

_J
1
We now require that

2
(1) (X~')' - ~~ = A.X~, and when x = 1,
x
so tha t
(2)
If we write A.= - p2, a solution of the singular eigenvalue problem (1) on

the interval (0,1) is ~ = Jn(PiX)where Pl' P2"" are the roots ofthe equation
(3)
where n is somefixed nonnegative integer.

For each pair of the numbers h and n the roots of Eq. (3) are all real.
A full set of eigenfunctionsof the singular problem (1) arisesfrom the non-
negative roots, roots that make up an infinite sequence Pl' P2' . .. where
O ~ Pl < P2 < The sequence depends on the values of h and n. In
case n = h = O, we find that Pl = O, then the first eigenfunction is the con-
stant Jo(O)= 1; otherwise Pl > O. 1
Our transformation is then the modified finite H ankel transformation ,

I!
(4) (i = 1, 2, .. .),
where Pi are the nonnegative zeros of the function hJiP) + pJ~(P). The
transformfH(P)is a sequenceof numbers, a sequencethat depends on the
valuesgivento n and h. The basicoperationalproperty(2)now takes the form
(5)
Hnh{F"(X) + ~F'(X) - ::F(X)} = -P/fH<Pi)
+ JiP)[hF(l) + F'(l)],
valid if F is of class C" on the interval O ~ x ~ 1.
Again, for each fixed pair of constants n and h, where n = O,1,2, ...
and h ~ O,the set of functions Jn(PIX),Jn(P2X),,,,,is orthogonal on the
interval (0,1),with weight function x. Using known formulas for norms of
Jn(PiX),we can write
O if i ~ j, where j = 1,2,. . . ,
~ I
(6) = P/ + h2 -
Hnh{Jn(PjX)} n2 2
if i = j,
{ 2P/ [Jn(Pj)]
except that in case n = h = O,then P1 = O and J 0(0) = 1 so
(7) Hoo{l} = f xJo(Pix)dx =!

when i = 1.
The Fourier-Bessel series, the generalized Fourier series of those

orthogonal functions corresponding to a function F, converges to F(x) under
the conditions stated for the representation (2~ Seco 150, and serves as the
inversion formula
(8)
but if n
F(x) - 2 ;t, Po' +~: - n' [~:gJ],J.(P,X)
= h = O,then P1 = O,and the first term in the summation is the
constant 18(0) = f~ tF(t) dt.
r < x < 1);
Some further properties of finite Hankel transforms and transforms of

particular functions are included in the problems folIowing Seco 152.
The development of Hankel transforms has only minor modifications
when the integer n is replaced by an arbitrary positive number V.
152 A BOUNDARV VAlUE PROBlEM

In a half cylinder of semi-infinite length with internal heat generation, let
steady temperatures U(r,<p,z)satisfy the conditions
1 1
(1) U" + -Ur + 2U,p,p + Uzz + Q(r) = O
r r
(O < r < 1, O < <p< 1t,Z > O),
(2) U(l,<p,z) = 1, U(r,O,z) = U(r,1t,z) = U(r,<p,O) = O.
Equation (1)is Poisson's equation V2U = -Q(r) in cylindricalcoordinates

r, <p,z (Fig. 109). Unless its nonhomogeneous term Q(r)vanishes,that term
has no Fourier sine transform on the half line z > O.
Let u,(r,n,z)denote the finite Fourier sine transform of U with respect
to O),
(%,y,z)
z
y
------------------
% Fig.109
(4) us(1,n,z) = Sn{l}, u.(r,n,O)= O,

where Sn{l} = [1 - (-l)"]n-l and n = 1,2,....
The differential form with respect to r in Eq. (3) and the bound~~
condition at the cylindrical surface r= 1 are adapted to the finite Han~
transformation Hnj with respect to r. We write
usH{Jlj,n,z)= Hnj{us(r,n,z)} (j = 1,2,.. .),
where Jlj are the positive zeros of Jn{Jl). Then according to property (10),
Seco 149,
d2
-Jl/UsH + Jl!n+l{Jlj)Sn{l} + dz2USH + qH{Jlj)Sn{l} = O
when z > O, and usH{Jlj,n,O)= O, where qH{Jlj)is the transform Hnj{Q(r)}.

The bounded solution of this problem is
1.. qH{Jlj) + JljJn+l{Jlj)
(5) ) Sn{l}
usH\fA'j,n,z= 2 [l - exp ( - JljZ)].
Jlj
The inverse transformations are represented by series:
then
(6)
,1,.
)
4 ~ ~ usH{Jlj,n,z)
) ',1,.
L.J L.J [Jn+l (Jlj WJ n{Jljr SIO n'l'
U(r,'I"z = -1tn=lj=l
where UsHis given by formula (5) and Jlj are the positive zeros of J n{Jl),which
depend on the value of n.
In case the condition U(l,cjJ,z)= 1 is replaced by
Ur(1,cjJ,z)= -hU(1,cjJ,z) + P(cjJ,z),
a condition of heat transfer at the cylindrical surface, the modified finite

Hankel transformation, treated in Seco151, can be made with respect to the
variable r.
The more common applications of Hankel transforms involve the
special case n = O and the Bessel function Jo.
PROBLEMS
1. Derive the finite Hankel transfonns (13) and (14), Seco 149, of the functions X'
and Jn(cx).
2. Obtain the finite Hankel transfonns
J.ll-4
(a) HoAx2} = -r-J(J1.j),
J.lj
I
d
(b) HOj{l - X4} = 1~HoAx2},

J-Ij
where JO(JJ.j)
= Oand J-Ij> O (j = 1,2,.. .).
3. If F and F' are continuous on the interval O ~ x ~ 1 except possibly for finite
jumps at an interior point x = e, and if F" is sectionally continuous on the interval,
show that
HOj{F"(X) + ~F'(X)} = -J-I/fH(JJ.j) + J-I/l(JJ.j)F(l)
- eJ-l/l(JJ.}.c)[F(e
+ O)- F(e - O)] - eJo(JJ.}.c)[F'(e
+ O)- F'(e - O)]
where J-Ijare the positive zeros of Jo(JJ.)and
(j = 1, 2, . . .).
fH(JJ.j) = HOj{F} = s: F(x)xJo(JJ.r)dx
4. Let G(x,e) be Green's function (Sec. 95) for the differential form D2 when n = O,
such that
xGxx(x,e) + Gx(x,e) = O (O< x < 1, x ,p e),

1
G(l,e)= O, Gx(e+ O,e) - GAe- O,e) =- c
and G is continuouswhen O~ x ~ 1, whereO< e < 1. Showthat

G(x,e)= logc whenO~ x ~ e,
= logx when e ~ x ~ 1;
then use the operational property found in Probo 3 to obtain the transform
1
HOj{G(x,e)} = -~Jo(JJ.}.c) (j = 1, 2, .. .),
J-Ij
where Jo(JJ.j)= O and J-Ij> O. Compare formula (3), Seco114.

5. If n and h are not both zero and Pl' P2, . . . are the positive zeros of hJ .<P) + pJ~(P),
show that
l h+ n
fo x.+ lJ.<PX)dx = H.h{X.} = yJ.<P) (n = O, 1, 2, . . .).
6. When n = O and h > O and p are the positive zeros of the function hJo<P)+
- pJ 1 (P),and if
pJ'o<P),or hJ o(P)
(i = 1,2,. ..),
fH<P)= HOh{F}= s: F(x)xJo<Px)dx
derive the formula
l l
1 y t 1
p/fH<P) = HOh {Jx f.o yF(t)dtdy + h f.o tF(t)dt } .
Also, from the transform HOh{l} found in Probo 5, show that

h 1 1 - X2
P/Jo(/J)= HOh { 2h + - } .
7. When n = h = O in the transformation Hnh, note that
ifi = 2,3,...,
HOO{F"(X)+ ~F'(X)} = -P/Hoo{F} + F'(l)Jo(/J)
= F'(l) ifi = 1, since p = O,
where P, P2, . . . are the nonnegative zeros of J (/J). Show that
Hoo{1} =O if i = 2, 3, . . . , if i = 1;
ifi = 2,3...,
if i = 1.
8. If e> O and Jo(e) t= O, show that the function Y of class C" (O~ x ~ 1) that
1
Y"(x) + -x Y'(x) + e2 Y(x) = x2 - 1 Y(l) = O,
has the transform
9. Solvethe problem
1
Y"(x) + -x Y'(x) - e2Y(x) + F(x) = O, O<x<l; Y(l) = O,
in the form
(j = 1,2, . . .).
fHJ.j)= I: F(t)tJ oJ.t) dt
In case F(x) = JoJ.x),show that the solution becomes
L
.
430 SECo 152] OPERATIONAL MATHEMATICS ~
/
10. Use the Hankel transformation HOj with respect to r to derive, in only a few steps,
the formula (9), Seco83, for temperatures U(r,t) in a circular cylinder of infinite length,
when
1
U,(r,t) = Urr(r,t) + -Ur(r,t)
r (O ;;i; r < 1, t > O),
U(r,O)= O, U(I,t) = 1.
11. Use results found in Probo 7 to show that the appropriate Hankel transform of
U(r,t), where
1
U,(r,t) = Urr(r,t) + -:Ur(r,t) (O;;i;r < 1, t > O),
U(r,O)= O, Ur(l,t) = A,
has the value
ifi=2,3,o..,
1[P~zJo(fJ) - zJo(fJ)exp(-p/t)]
if i = 1,
4[~ +M4t +~)]
and thus obtain the formula for U(r,t) found in Prob:S, Seco 830
12. In Probo 8, Seco144, an integral representation was found for the temperature
distribution W(r,x) in a semi-infinite circular cylinder whose axis is the positive x axis,
when ~
(O< r < 1,x > O),

W(I,x) = O ifx > O, W(r,O)= 1 ifr < 1;
also, W is to be continuous and bounded when O ;;i;r ;;i; 1 and x > 00 Use a Hankel
transformation to derive this series representation of W:
.., 1 J o(p.jr)
W(r,x) = 2 L --exp(-JljX)
j= Jlj J(p.j)
when O ;;i; r < 1 and x > O,where Jo(Jlj) = O and Jlj > 00
13. Let transverse displacements Z(r,t) in a membrane stretched over a circular frame
1
Ztt = Zrr + -Zr - 2kZ, (O;;i;r < 1, t > O),
r
Z(I,t) = Z(r,O)= O, Z,(r,O) = Vo o
Ifthe coefficient of damping 2k is such that O < k < Jl where Jl, Jlz," oare the positive
zeros of J o(p.)in order of ncreasing magnitude, derive the formula
- -k' ;, Jo(p.I)sin[t(p./ - kZ)t]

Z(r,t) - 2voe L.
j= Jll(p.j)
z zt
(p.j - k )
o
14. Let Jlj denote the positive zeros of J (J1.)and write .
HAF(r)} = f F(r)rJ1(J1.jr)dr = fH(J1.j)

u = 1, 2, .. . ). \i
Derive the formula
~fH(J1.j)
Jl)
f ro !.P 2F(t)dtdP } '
= Hlj { , ~
P { }
and since H1j{r} = Jlj-IJ2(J1.j), according to transform (13), Seco 149, show that
1 r- r3
-J2(J1. j ) = Hl" { - .
Jlj3 ) 8 }
15. If Y = r sin ,p and V2V + Ay = O iriside a half cylinder 0< r < 1, O < ,p < n,
z > O,and if V = Oon the boundaries,then V(r,4>,z)
satisfiesthe conditions
1 1
+ V.Z+ ArslD'I' = O
"~
v" + -v, + -V4>4> (O< r < 1, O < ,p < n, z > O),
r r2
V(I,,p,z) = V(r,O,z)= V(r,n,z) = V(r,,p,O)= O.

Reduce the problem to one in W(r,z) by writing
V(r,,p,z) = A sin ,p W(r,z)
/
and, withthe aid ofthe transformfoundin Probo14,derivethe solution
r-r3 ) J1(J1.y)
W(r,z)=
g- - 2j=IJlj 3 J 2Jlj
( ) exp ( - JljZ)
where J (J1.j) = O and Jlj > O.
153 NONSINGULAR HANKEL TRANSFORMATIONS

Ir our differential form
1 n2
(1) D2F = F"(x) + -F'(x)
x - zF(x),
x
where n is a fixed nonnegative integer, applies to functions F on a bounded
intervalO < a ~ x ~ b, one that does not have the origin as an end point
or an interior point, then D2 has no singular point on the interval. The linear
integral transformation that resolves D2F in terms of the transform of F
itself and prescribed values of F or F', or of a linear combination of F and F',
at the points x = a and x = b is a nonsingular special case oJ the Sturm-
Liouville transJormation introduced in Chap. 10.
The kernel eI>(x,- Jl2)satisfiesBessel'sequation
n2
(2) (xel>')'
x = - p.2xel>
- -el> (a < x < b)
and homogeneous boundary conditions at the end points of the interval.

The weight function p(x) in the integral transformation is x itself. To satisfy
both end conditions, the general solution of Eq. (2),
(3) <1>= AJ n(Px) + B y"(JLx)
is needed, where Y"is Bessel's functiOir of the second kind.1 The ratio of the
constants A and B, and the discrete values J1.1,J1.2,... of J1.,are determined
by the homogeneous boundary conditions on <1>.
In Probo 12, Seco107, we used a generalized Fourier series to solve the
following problem for the electrostatic potential V(r,z) in the charge-free
space between cylindrical surfaces r = a and r = b, above the plane z = O.
1
V,r(r,z) + -r V,(r,z)+ ~z(r,z) = O (O< a < r < b, z > O),
(4)
V(a,z) = V(b,z) = O, V(r,O) = F(r).
The solution found there can be obtained by using the nonsingular Hankel
transformation
(5)
where
H o{F(r)} = f F(r)rtf(ai,r)dr
(i = 1,2,. ..),
(6)
and ai are the positive zeros of the function tf(a,b).But the transformation
H o applies also to problems in which the differential equation and all
boundary conditions in problem (4) are made nonhomogeneous.
154 HANKEL TRANSFORMATIONS H.. ON THE HALF UNE (x > O)

The best known of the various Hankel transformations are singular trans-
formations of the type
T {F} = 1"" F(x)X<l>(X,A)

dx = f(A);
that is, they apply to functions F(x) defined over the positive half of the
x axis. The transformation T corresponding to a fixed integer n (n = O,1,
2, . . .) replaces our differential form
D2F = F"(x) + !F'(x)

x - n:
x F(x) = x~ (XF')' - n2x F
[ ]
I The function y"(x)is of the order of x-o as x ->Oif n = 1,2,..., while Yo(x)is of the order
of log x as x -> O. A surnrnary of representations and properties of Bessel functions, including
y., with references, is given by A. Erdlyi et al., "Higher Transcendental Functions," vol. 2,
chap. 7, 1953.
I
i
~ J
on the half line x > O by A.f(A.). That operational property is the principal
one for applications of such transforms to problems in differential equations.
To construct T so that it has that property, we write
(1)
T{D2F} = LX) [(XF')' - : FJ~dX
= {X> [(~')' - : ~JF dX-:' [x(~F'- ~'F)]o'

We assume, say, that F and ~ are of class C" on the half line x ~ O and
that the functions F and ~ are such that the last bracketed term vanishes
as x -+ 00 and the second improper integral exists. Then we require thar-
2
(2) (x~/)' - ~~
x = A.~ (x > O)
so that T{D2F} = A.T{F}. Equation (2) is Bessel's equation. We write

A.= - rx.2; then if rx.is real, the bounded solution of that equation that is of
class C" when x ~ Ois ~ = AJn(rx.x).Alllinearly independent solutions are
represented if A = 1 and rx.~ O.
Thus T becomes the Hankel transformation
(3) (rx.~ O),

Hna{F(x)} = LX>F(x)xJn(ax)dx = fH(rx.)
where n has one of the values O, 1,2, . . .. Its basic operational property is
(4)
Here the Hankel transformfH(rx.)is a function of rx.that depends on the choice

of n. The notationfHn(rx.) is more descriptive but rather cumbersome.
An asymptotic representationl of Jn(t) for large values of t shows that
Jit) is 19(ct) as t -+ 00; that is, .JtIJn(t)1 is bounded. From the formula
(5) tJ~(t) = nJn(t) - tJ n+ l(t) (n = O,1,2,. . .)
it follows that J~(t) has that same order as t -+ oo. Now in Eq. (1), ~ = Jn(rx.x)
and d~/dx = rx.J~(rx.x)where J~ is the derivative of Jn with respect to the
argument of Jn; also, (x~')' - n2x-l~ = -rx.2xJn(ax). Then if F(x) is
19(X-k)as x -+ 00,where k > t, the integrand ofthe second improper integral
in that equation is 19(x-k+t)and, sincek - ! > 1, that integral is absolutely
convergent. Ifin addition F' is 19(X-k+l),the final term in Eq. (1)vanishes;
thus the right-hand member of the equation exists, and so T{D2F} exists
and is equal to -rx.2Hna{F}.
I Erdlyi, A., et al., "Higher Transcendental Functions;; vol. 2, p. 85.
Operational property (4) is therefore valid if F is class C" when x ~ O,

if F is (9(X-k), and if F' is (9(X-k+l) as x -+ 00, where k > !.
The Hankel transform (3) of F exists if F is sectio~lIy continuous on
each bounded interval of the positive x axis and if F s (9(X-k) as x -+ 00,
where k> t.
The Fourier-Bessel integral representation of F(x) corresponding to
the Fourier sine or cosine integral formula, furnishes an inverse transforma-
tion
(6) (x >,0).
It is valid if F(x) is (9(X-k) as x -+ 00, where k > !, if F' is sectionally con-

tinuous over each bounded interval, and if F(x) is defined as the mean value
of its limits from the right and left at each of its points of discontinuity.
Our conditions under which the above formulas are valid are sufficient.
They can be relaxed. The transform (3) of F(x) may exist, for instance, when
F is unbounded at the origin but such that xn+1F(x) is (9(XC)as x -+ O,where
e > -1. No major changes are involved when the integer n in this section
is replaced by v where v > - t.
155 FURTHER PROPERTIES OF Hna
Ir fH(a) = Hn".{F(x)},we find that
(1) when k > O.
A property on division of Hankel transforms fH(a) by a2 follows from

our basic operational property (4), Seco 154, by solving the differential
equation
Y"(x) + !.
X
Y'(x) - x2
n2 Y(x) = - F(x) (x > O)
for Y(x), since -a2YH(a) = - fH(a). The formula is

1 A
x n r' t n+ 1
(2)
a2 fH(a) = Hna {f
x ( y) JB ( Y) F(t) dt dy}
where n is a nonnegative integer and A and B are nonnegative constants or

infinity, to be chosen if possible, so that the transform on the right exists.
Ir F satisfies our conditions under which the operational property (4),
Seco 154, is valid except for finite jumps jc = F(c + O) - F(c - O)and j; =
F'(c + O)- F'(c - O) in F and F' at a point x = e, the property takes the
modified form
1 n2
(3) Hna; {F" }

+ ~F' - x2F = -a.2fH(a.)+ a.cJ~(a.c)jc
- cJ;;cac)j~.
A kernel-product property for Hna;arises from the formula1
(4) J (x)J (y) =

n n
x"y"
2n~r(n + t) i n Jn[R(x,y,O)] sin 2nO dO
o [R(x,y,o)]n
where R(x,y,O) is the length of the side of a triangle opposite the angle O
included between two sides of length x and y; that is --
(5) R(x,y,O) = (X2 + y2 - 2xy cos 0)+.
If we use formula (4), the product Jn(ay)Hna;{F(x)} can be written as an

iterated integral with respect to x and O. It is an integral over the upper
half plane if x and Oare interpreted as polar coordinates. When that integral
is written in terms of R and </>,polar coordinates about the point (y,0) on
the axis O = O, we obtain the kernel-product property
1
(6) a"yJ..{a.Y)fH(a.) = Hna;{PF(x,y)}
wherefH(a.) = Hna;{F(x)}and
(7) PF(x,y) = x"y"+1 r" F[R(x,y,</>

)] sin 2n </>d</>.
~2nr(n + t) lo [R(x,y,</]n
The corresponding convolution XFG(x)of two functio~s F and G such
that
(8)
is then formally, as indicated in Seco 109,
(9)
X FG(X)= {ex>F(y)PG(x,y)dy.
Formulas (4) to (9) are presented here to display their complexity. In
case n = O,the results are included in Table 4, but even in that case the
kernel-product and convolution properties are difficult to use. The integer
n can be replaced by v, where v ~ O,in those formulas.2
1 Watson, G. N., op. cit., p. 367.
2 Derivations and conditi~ns of validity of those kemel-product and convolution properties
are inc1uded in J. S. Klein's doctoral dissertation, "Some Results in the Theory of Hankel
Transforms," University of Michigan, 1958. Earlier, J. Delsarte, J. Math. Pures Appl. (9), vol.
17, pp. 213-231,1936, and Acta Math., vol. 69, pp. 259-317,1938, gave convolution properties
associated with Hankel transforms.
-
436 SECo 156]
OPERATlON\l MATHEMATICS
156 TABlES OF TRANSFORMS H",,{F}

Table 4 summarizes properties of the transformation Hna on the half line
x > O and lists transforms of some particular functions, for the important
special case n = O. Entry 5, giving a possible inverse transform of fH(rx)jrx2,
is the special case of formula (2), Seco155, when n = O,A = 00, and B = oo.
Derivations of some of the transforms listed in the table are indicated in
problems following Seco 157.
Table 4 Hankel transforms (n = O)
fH(a) (a > O, n = O) F(x) (x> O)
F(x)
f""
o F(x)xJo(ax)dx = Ho.{F}
2
{"" fH(a)aJO(ax)da = Hox{fH}
I
3
F"(x) + 'XF'(x)
4 fH(ka) (k > O)
"" ""t
5 -
f f x
-F(t)dtdy
y Y
6 1tJo(ak)fH(a) (k> O)
r o
F[(X2 + k2
"-.
- 2kx cos O)t]dO
7
fo""yG(y) s: F[(X2 + y2 - 2xy cos O)t]dOdy
1 I
8
a x
I I -ex
9 (e> O) -e
J a2 + e2 x
10
II (e> O)
12 (e> O)
.jX2 + e2
13 1 .j X2 + e2 - e
(e> O) -Iog
2 .j X2 + e2 + e
I ir O< x < e
14 (e> O)
{ O ir x > e
15 (e> O)
~Jo(~)
An extensive table of Hankel transforms on the half line x> O,given

by A Erdlyi el al., "Tables of Integral Transforms," voL 2, 1954. In those
tables the transformation is written, except for variations in the letters used,
as
{X>G(x)~Jv(ax)dx = g(a;v).
/
Thus in terms of the notation we use here,
G(x) . g(a;v).
and
F(x) = .fi fH(a) = .fi
Those tables list several properties of transforms g(a; v)in which different
values of the index v occur in the same property. Such properties follow
from relations such as Eq. (5), Seco 154, between J~, Jn, and Jn+l' or the
recurrence relation
The tables list transforms of particular functions in cases v = O and v = 1,

and for the general index v. Many of the functions and transforms involve
higher transcendental functions.
Another table will be found in the book by Oitkin and Prudnikov;
other treatments and applications of Hankel transforms are included in
books by 1. N. Sneddon ("Fourier Transforms") and C. J. Tranter, alllisted
157 AXIALLYSYMMETRIC HEAT SOURCE
The following is an example of a boundary value problem for which a Hankel

transformation Ho",with respect to r is the natural method ofsolution. Find
a formula for steady temperatures V(r,z) in a semi-infinite solid z ~ O
throughout which heat is generated at a steady rate that varies only with
the radial coordinate r, when the face z = O is kept at temperature zero.
Then
1
Y,.r(r,z) + -r y"(r,z)+ ~%(r,z)+ F(r) = O
(1) (r > O,O ~ O),

V(r,O) = O,
where r, <p,and z are cylindricalcoordinates (Fig. 109);also F and V are to

satisfy order properties as r -. 00 and z -. 00 such that a solution exists.
/'
Let vH(a,z)denote the Hankel ttansform of V with respect to'r, for which
n = O:
(2) v~a,z) = HOIX

{V(r,z)} = LX)V(r,z)rJ o(ar)dr (a > O),
and writefH(a) = HoIX{F(r)}.Then the transformed problem is, formally,
d2
(a > o, z > O),
-a2VH(a,z) + dz2 VH(a,z)+ fH(a) = O
(3)
I
The bounded solution of that problem is
1 - e-IX:
(4) vH(a,z)= fH(a) a .
Thus a formal solution of problem (1) is

1 - e-IX:
f
a>
(5) V(r,z)= fH(a) Jo(ar)da.
o a
Another form of the solution can be written by using Table 4, first to
represent the inverse transform of a-2fH(a) as an iterated integral in terms
of F, then by using the convolution integral (entry 7) to write an inverse
transform of either the product [a-2fH(a)]e-:IXor fH(a)(a-2e-:IX). But those
forms are complicated, even when F is specified as the simple step function
F(r) = A when O< r < e, F(r) = O when r > e, representing the case in
which heat is generated uniformly throughout a cylindrical core of the s9lid.
In the special case
k
(6) F(r) = (r2 + k2)3/2 where k > O,
thenfH(a) = e-k, and formula (5)can be verified as a solution ofproblem (1).
But another form of the solution can be written since
(7) VH(a,z)= ~e-kIX - ~exp [-(z + k)a].

a a ..
From Table 4 it follows that
W (r O) - 1 W(r,z) - 1
(8) , - log
2V(r,z) = log W(r,O) + 1 W(r,z) + 1
where
r2 1/2
(9)
W(r,z) = [ (z + k)2 + 1J
/'
Formula (8) can be verified as a solution of problem (1) when F is the

function (6). According to that formula, the temperatures on the z axis tend
to infinity as z -. 00, since
(10) V(+O,z) = logZ +k k.
PROBLEMS
1. UsetheknownLaplace transformation /
L{ J O(lXt)} = (S2 + 1X2) -1/2 (s > O, IX> O)
to establish the transform H o. {x -1 e -ex} listed in Table 4. Also, differentiate with respect
to the parametere to showthat, if e > Oand m = 1,2,...,
2. Note that our conditions of validity of the inversion formula (6), Seco 154, are
satisfied by the function F(x) = e-ex whene> O.Thus
when x > O.
From the transform Ho.{e-ex} established in Probo 1, deduce transform Ho.{e(x2+

e2)-3/2} listed in Table 4.
3. Use transformations 5 and 14 in Table 4 to indicate that, when e > O,
where G(x)= O
e2 - x2 + 2c210g~ if O< x ~ e,
{ e if x ~ e.
4. Let V(r,z) be bounded in the half space z > O, r ~ O, O ~ f/>< 21t, where r,z,
and f/>are cylindrical coordinates, and satisfy the conditions
1
V2V(r,z)= v" + -v,
r + V.z= O (r > O, z > O),
1
V(r,+O) = 'r2 + e2,1I2'
where e > O.Derive and verify the solution

V(r,z) = [r2 + (z + e)2r 1/2.
5. If V(r,f/>,z)
satisfies Poisson's equation
v" + !v,
r + r12V",,,,+ V.Z+ F(r)sin f/>= O
J
/
in the region r > O,O < cP < n, Z > O,and if V = Oon the boundaries cP = O,cP = n,
and Z = O,derive formally the formula
a:>
a:>1 - e-U
V(r,cP,z)= sincP
f o ex
J1(exz)
f o
F(P)J1(exp)dpda..
6. Let V(r,z) denote steady temperatures in a slab r ;?;O, O ~ cP< 2n, O ~ z ~ 1,

where r, cP,and z are cylindrical coordinates. If the face z = O is kept at temperature
V = O and the face z = 1 is insulated except that heat is supplied through a Ci{cular
region, such that
if O<r<c,
y'(r,l) = { ~ if r> c,
then V2V = Owhen r ;?;Oand O < z < 1, and V(r,O)= O. Derive the formula
V(r,z) -c
-
f
a:>
--
sinh
o cosh z
exz J 1(exC)
ex
( ) dex.
J otXr
15
Legendre and Other Integral
Transforms
We have presented the operational calculus of various integral transforma-

tions, with emphasis on properties that are useful in solving problemNn
differential equations. We gave spedal attention to the Laplace transforma-
tion because a great variety of prominent problems, especialIy in partial
differential equations, can be solved or simplified by using Laplace trans-
forms. A different variety, not so extensive, can be solved by using various
Fourier transforms, and still more special types of problems are adapted to
Hankel transformations. Of course those and other transformations may
be useful outside the field of differential equations, including applications to
theory of functions, integral equations, or difference equations.
Unless a transformation has a substantial variety of applications, a
development of its operational properties here is hardly justified. Procedures
given in Chap.l0 may be used to design transformations needed for particular
problems. Legendre transforms, treated below, have some good applica-
tions, especialIy to problems in potential theory that involve spherical polar
coordinates.
441
158 THE LEGENDRETRANSFORMATION T. ON THE

INTERVAL(-1,1)
Let r, tjJ,and e denote the sphericalcoordinatesof points (X, y,z) such that
(Fig. 110)
y
Z = reos e
X = tan tjJ,
(o~e~1t
- "--.-
) .
In terms of those coordinates the Laplacian of a function V(r,tjJ,O)takes the
form
2 1 2 1 1.
(1)
[
V V = 2r (r V,)r+ ~Sin OV.. + --=--
Sin e
(VSin O)0
]
.
In case V is independent of tjJ, that is, V = V(r,O)so the values of V have
axial symmetry about the Z axis, the differential form with respect to O
that appears in V2 V is
1. o 2 Ov(r,e)
sin O(VSin 0)0 = sin 000 [(1 - cos O)sin O00] .
Ir we write x = cos O, the form becomes
o 2 OV
(2) ~2 V = -ox
[-
(1 x )-
ox] (-1 < x < 1).
That self-adjoint form ~2 V is singular at the points x = :!:1. We can

construct a transformation of functions F on the interval (-1,1),
T{F(x)} = fl F(X)~(X,A)dx = f(A),
such that T {~2F} = Af(A) by first writing
T{~2F} = f 1 ~~2F dx = f 1 ~(x,A)[(l - x2)FT dx
= fl F~2~dx+ [(1 - x2)(~F' - ~'F)]~l'

z
cx,y,z)
:
I
I
i
I
x Fig.110
LEGENDRE AND OTHER INTEGRAL TRANSFORMS [SEC. 158 443
then requiring that ~2~ = A~ and that ~ and F be of class C" on the
interval - 1 ~ x ~ 1. Thus
(3) [(1 - X2)<l>'(X,A)]'
= A~(X,A) when - 1< x< 1
and ~ is to be of class C" on the closed interval - 1 ~ x ~ 1.
That singular eigenvalue problem in Legendre's differential equation (3)
has as its set of eigenvalues
(4) A = -n(n + 1) (n = O, 1,2, . . . ).
"
The corresponding eigenfunctions are the Legendre polynomials Pn(x) of
degree n:
(5) - - 1 m (-lY
~(X,A) - Pn(X)- 2
(2n - 2j)! -2]
n ]=0 -:--
J. L (n - 2")1
'.] .(n - J.")Ix"
I
where m = !n if n is even or zero and m = !(n 1) if n is odd.1- . 1I

Thus T {F} is the Legendre integral transformation
(6)
T,,{F(x)} = f 1 F(x)Pn(x) dx = fL(n)
(n = O, 1, 2, . . . )
on the interval (-1,1), with the operationalproperty

(7) T,,{[(1 - x2)F'(x)]'}= -n(n + l)fL(n).
According to Eq. (5), the first few Legendre polynomials are
Po(x) = 1, P1(x)= x, P2(x)= !(3X2 - 1),
P3(x) = !(5X3 - 3x), P4(x) = !(35x4 - 30x3 + 3).
We note that Pn is an even or odd function according as n is even or odd:
(n = O, 1,2, . . . ).
In case F and F' are continuous on the interval -1 ~ x ~ 1 except

at an interior point x = e where they have jumps jc and j~, respectively,
while F" is sectionallycontinuous, property (7)is replaced by
(8)
The set of funccions Pn is orthogonal on the interval (-1,1). That
orthogonality and the norms of Pn are expressed by the transform of Pm,
where m has one of the values O,1,2,. .. :
if n :p m
(9)
T,,{Pm(x)}= { ~m + !)-l if n = m.
1 For basic properties of Legendre polynomials see the author's "Fourier Series and Boundary
Value Problems," 2d ed., chap. 9,1963.

The corresponding generalized Fourier series representation of a function F

on the interval is Legendre's series
00
(10) F(x) = L: (n + t)fdn)Pn(x) (-1 < x < 1).

n=O
That inversionformula for T,.is valid if F' is section\lIy continuous over the
interval and if F is defined as the mean value of its limits from the left and
right at each point where F is discontinuous.
When F is a polynomial of degree m, it is a linear combination of
Po, P1,..., and Pm,represented by series (10) in whichfL(n) = Owhen n > m.
Under the substitution x = cos e, our transformation
T,.{F(cos e)} = f F(cos e)Pn(cose) sin e de = fL(n)

transforms the <;lifferentialform
(O < e < n)
into -n(n + l)ji:,<n).
159 FURTHER PROPERTIES OF T.

By replacing n(n + 1) in property (7) above by (n + t)2 - l, we find that
(1) (n + t)2fdn) = T,,{!F(x)- [(1 - x2)F'(x)]').
Since Po(x) = 1 and Po is orthogonal to Pnwhen n = 1,2,..., then, if
C is a constant,
fdn) when n i= O
(2)
T,.{F(x) + C} = { fL(O)+ 2C when n = O.
Suppose that a sectionally continuous function F has been adjusted

by the addition of a constant so that fL(O)= O;that is,
f1 F(x)dx = O.
Then by solving the differential equation
[(1 - X2)Y'(X)]' = -F(x)
for Yand requiring Yand Y' to be continuous on the interval -1 ~ x ~ 1,
we obtain the property
(3) A(n) =
n(n + 1)
T,.
{fo +-
X
s - 1 5-1
F(t) dt ds +
s C
}
(n = 1, 2, .. . ).
LEGENDRE AND OTHER INTEGRAL TRANSFORMS [SECo 159 445
The value of the transform on the right when n = O depends on F and the
choice of the constant C.
Legendre polynomials have the product propertyl
n
(4)
where
Pn(cos f3)Picos O) =
1
-
1t oi Pn(COSv) da
(5) ~v=~f3~O+~f3~O~~
The following convolution property of 1'"has been established with the aid
of that property.2
Let F(x) and G(x) be sectionally continuous functions with Legendre
transforms fL(n) and gL(n). Then
(6)
A(n)gL(n) = T,,{XFG(x)} = 50" XFG(cos O)Pn(cos O) sin O dO
where XFGis this convolution of F and G:

n n
(7) X FG(COSO) =
1
-
1t oi F(cos 13)sin 13
io
G(cos v) da df3,
cos v being given by formula (5). The convolution here is not a simple one
to use.
Some useful transforms of particular functions can be found from the
representation
00
(8) (1 - 2tx + t2)-t L t"Pn(x)

= n=O (-1 < t < 1)
of a generating function for Pn. When t is fixed, the series converges uniformly
with respect to x, and it follows that
2tn
(9) T.{(1 - 2tx + t2)-t} = -
n 2n+1
when -1 < t < 1. Equation (9) is also valid when t We can = :t 1.

differentiatethe members of the equation with respect to t to write
I MacRobert, T. M., "Spherical Hannonics," p. 138, 1927.

z Churchill, R. V., and C. L. Dolph, Inverse Transforms of Products of Legendre Transforms,
Proc. Amer. Math. Soc., vol. 5, pp. 93-100, 1954. In formula (5), p, 6, and v can be interpreted
as sides of a spherical triangle on the hemisphere XZ + yZ + zZ = 1, Z ?; O,and a as the
angle between the first two sides.
Then with the aid of transform (9) it follows that

2t"
(10) T,,{(1 - 2tx + t2)-t} = _
1 -t 2 (-1 < t < 1).
Transforms of a few other functions and some applications of T" are

given in the author's paper, The Operational Calculus of Legendre Trans-
forms, Jour. Math. Physics, vol. 33, pp. 165-178, 1954. Brief treatments of
Legendre transforms are given in the books by Sneddon and Tranter listed
160 LEGENDRETRANSFORMS ON THE INTERVAL(0,1)

Let F(x) be defined over the interval (0,1) and let F1 be the odd extension of
F to the interval (-1,1). Then the product F1(X)Pix) is an even function
when n is an odd integer; it is odd if n is zero or even. Thus the Legendre
transform of F1 on the interval (-1,1) can be written
T2n+{F1(x)} =2 f F(X)P2n+l(X)dx, T2n{F1(x)} = O,
wheren = 0, 1,2, . . .. Let us use the symbol1;n+ 1 {F} to denote the integral
transformation of F that appears here; that is,
(1)
T2n+l{F(x)} = f F(x)P2n+ l(X) dx = JL(2n + 1)
(n = 0, 1, 2, .. . ).
We ca1lJL(2n + 1) the Legendre transform of F on the interval (0,1) with odd

indices 2n + 1.
In terms of T" on the interval (-1,1)
(2)
Since T2n{F.} = 0, the Legendre series representation (10), Seco 158, of the
function F1 becomes
00
F1(x) = L
n=O
(2n + 1 + !)T2n+1{F.}P2n+1(X) (-1 < x < 1).
This becomes an inversionformulafor 1;n+1:

00
(3) F(x) = L (4n + 3)JL(2n +
n=O
1)P2n+1(x) (O < x < 1),
because F1(x) = F(x) when < x < 1. The formula is valid if F' is s~c-
tionally continuous over the interval (0,1).
The orthogonality of the set of polynomials P211+1(X) on the interval

(0,1) follows from Eq. (2) by writing F(x) = P2m+l(X) = Fl(X) and recalling
that the set PII(X)is orthogonal on the interval (-1,1).
In caseF and F' are continuous on the intervalO ;:; x ;:;1, then at
the point x = O the function F1 has the jump 2F(0), but the jump in F
there is zero; elsewhere on the interval -1 ;:;x ;:; 1 the functions F1 and
F are continuous. Then if F" is sectionally continuous, formula (8),
Seco158, shows that .
(4) T,,{~2F1} = -n(n + 1)T,,{F1} + 2P~(0)F(0),

where ~2F1 = [(1 - x2)F'1]' = (1 - x2)F~ - 2xF. Since ~2F1 is an odd
function, then T211+d~2F1}= 2T211+1{~2F},while both members ofEq. (4)
vanish when n = O,2, 4, .... It follows that
From the formula

(5) (n = 1,2,...)
we find that Pll+ 1(0) = (2n + l)P211(0). Therefore
(6) 1 {[(1
'1'211+ - x2)F'(x)]') = -(2n - 1)(2n + 2)JL(2n + 1)
+ (2n + 1)P2n(0)F(0).
Formula (6), the basic operational property of '1'211+

1>shows how the
form ~2F on the interval (0,1) is transformed in terms of T211+dF}and the
initial value F(O).
In like manner we find that the Legendre transformation with even
indices, on the interval (0,1),
(7) (n = O, 1, 2, . . . ),
is related to the transformation of the even extension F2 of F on the interval

(-1,1) by the formula
(8) (n = O, 1,2, . . . ),
and T211+
1{F2} = O. It follows that '1'211
has the inversion formula
00
(9) F(x) = L
n=O
(4n + 1)JL(2n)P211(x) (O < x < 1),
and the basic operational property

(10)
that resolves the differential form ~2F on the interval (0,1) in terms of
1;n{F} and the initial value F'(O) of F'. AIso, the set of functions P2n is
orthogonal on the interval (O,1).
161 DIRICHLETPROBLEMS FOR THE SPHERE

The problem of determining the harmonic function in a region which
assumes prescribed values on the boundary is called a Dirichlet problem.
If instead the normal derivative of the harmonic function is prescribed on
the boundary, the problem is called a N eumann problem.
Consider the Dirichlet problem for the region interior to the unit
sphere r = 1 such that the harmonic function V(r,cos ()) is independent of
the spherical coordinate <p. Then
[
~
V2V = r12 (r2v,)r+ sin () (Sin ()J.-9)o
J
=O (r < 1, O < () < n),
and V -> F(cos ()) as r -> 1; also V and its partial derivatives of first and
second order are to be continuous interior to the sphere. We write x = cos ().
Then
(1) [r2 V,(r,x)]r + [(1 - x2)Vx(r,x)]x= O (r < 1),
(2) V(1 - O,x) = F(x) (-1 < x < 1).
Ir vL(r,n) is the Legendre transform of V with respect to x on the
interval (-1,1),
(3) vdr,n) = f 1 V(r'X)Pn(X)dx = T,,{V} (n = O,1,2, . . . ),

and fL(n) = T,,{F(x)}, the above problem transforms into the problem
2 d2vL dVL
(4)
r dr2 + 2r-: - n(n + l)vL = O,
In problem (4) VLis to be a continuous function of r when O ~ r < 1.

The solution with that continuity is
(5) (O~ r < 1, n = O,1,2, . . .),
and so the formal solution of the problem in V is
00
(6) V(r,x) = (n + t)fL(n)rnPix)

n=O
(r ~ 1, -1 < x < 1).
This form of the solution is also obtained when the problem in Vis solved
by the method of separation of variables.
Another representation of the solution follows from our convolution

property for T". According to transform (lO), Seco159,
1 - r2
r" = ~T,,{(1 + r2 - 2rxt3/2} (O ~ r < 1).
The productfL(n)r" in Eq. (5) is the transform of the convolution (7), Seco159,
of the functions F and T"-1{r"}; that is,
1 - r2 " "
(7) V(r,cosO)= -
2n I o
F(cosP)sin P
I o
(1 + r2 - 2r cos v)- 3/2da dp
where cos v = cos P cos O + sin p sin O cos a.

Formula (7) is known as the Poisson integral formula for the harmonic
function interior to the sphere r = 1 in this special case where V is pre-
scribed on the boundary as a function F (cos O) of O only. It is of some
interest to note, especially when V is interpreted as steady temperatures in
a solid sphere r ~ 1,that either representation (6) or (7) gives the value of V
at the center r = O of the sphere as the mean value of F over the surface.
In the corresponding Dirichlet problem for the bounded harmonic
function W(r,cos O) in the region r > 1 exterior to the unit sphere, where
W(l + O,cos O) = F(cos O),the solution of the transformed problem is
n
1 l
(n = O, 1,2, . . . ).
wL(r,n) = -;:f(n) ( -;:)
Therefore, in terms of the function V found above,

i I
(8) (r > 1).

W(r,cosO)= ~V( ~,cosO)
I
PROBLEMS I
1. Using orthogonality properties of Pn(x)on the interval (-1,1), display the ortho-
gonality and norms for the set of polynomials P2n+I(X),n = 0,1,2,..., on the interval I
(0,1) by showing that for each m (m = O,1,2, . . .)
- O if n # m
T2n+1{P2m+1(x)} = { (4m+ 3)-1 ifn = m
where(Sec.160) T2n+I{F} = f~F(x)P2n+l(X)dx.
2. Display the orthogonality and normsoftheset ofpolynomialsP2n(x),n = O,1,2,...,
on the interval (0,1) by showing that for each m(m = 0,1,2,.. .)
- O ifn#m
T2n{P2m(x)}
= { (4m + 1)-1 ifn = m
where (Sec. 160) T2n{F} = gF(X)P2n(x)dx.

3. Use properties (6) and (10), Seco160, to show that

- P2n(0)
(a) T2n+l{1} = 2n + 2'
- P2n(0)
(b) T2nX
{ }= .- -.'
4. The Neumann problem in a function U(r,cos O) independent of the spherical

coordinate C/>, for the interior of the unit sphere,is
v2 U(r,x) = O when r < 1, Ur(l - O,x) = F(x) (-l<x<l),
where x = cos O. Since U is to be continuous interior to the sphere, at the center r = O
in particular, its Legendre transform udr,n), with respect to x, must be continuous
when O ~ r < 1. Show that, as a consequence, the prescribed values F(x) of the normal
derivative of U on the surface r = 1 must be such thatfL(n) = O when n = O; that is,
the mean value of F(x) must be zero, or
(a) I: F(cos O)sin OdO = O.
Also if U is interpreted as steady temperatures and Ur(l,x) as flux of heat, note why
condition (a) is to be expected in order that temperatures remain steady.
Under condition (a) derive the formulas
ifn = 1,2,...,uL(r,0) = C,
1 1""2n+l
U(r,cosO)= -2 C + 2 L - _n=1 n
fL(n)r"Pn(cosO)
where r < 1, O ~ O~ n, and C is an arbitrary constant.

5. In the Dirichlet problem in V(r,x) represented by Eqs. (1) and (2), Seco 161, let
F(cos O) satisfy condition (a) in Probo 4 above; then vL(r,O)= O, and vdr,n) = fdn)r"
when n = 1, 2, . . .. From the transform of the solution U(r,x) of the corresponding
Neumann problem with the same function F, found in Probo 4, show that
uL(r,n)=
i r1
-vL(p,n) dp
op
if n = 1,2,..., uL(r,O)
= C,
and deduce this relation between U and V:

r 1 1
i
U(r,cos O)= op- V(p,cos O)dp + '2C J .
=i t
1 1 1
(r < 1),
O V(rt,cos O)dt + '2C
where C is an arbitrary constant. The relation also follows from the series representa-
tions of U and V. It can be verified even if F, U, and V are not independent of C/>,
provided that
I: F(c/>,cos O) sin O dO = O.
"I
.EGENDRE AND OTHER INTEGRAL TRANSFORMS [SEC. 161 451
i. Let U(r,cos O) be harmonic interior to the sphere r = 1 and satisfy this condition
It the surface:
U,(1,x) + (k + 1)U(I,x) = F(x)

vhere x = cos Oand the constant k + 1 is positive. Show that the Legendre transform
'L(r,n) on the interval -1 < x < 1 is related to the transform vdr,n) = fL(n)r" in the
:orresponding Dirichlet problem (Sec. 161) in V(r,cos O) by the equation
r-k :r[~+ luL(r,n)] = vL(r,n).
fhus deduce the relation
U(r,cos O)= f V(rt,cosO)tkdt,

Nhich can be verified even when F, U, and V are not independent of the spherical
:oordinate cp.
7. Heat is generated at a steady and uniform rate throughout a solid hemisphere
) ~ r < c, ~ O< n/2. The entireboundaryis keptat temperature zero. Thusthe
teady temperatures V(r,x), where x = cos O,satisfy the conditions
V2V(r,x) + A = (r<c,O<x<I),
V(c,x) = V(r,O)= 0,
where A is a constant. With the aid of the transform 1'2n+1 {1}found in Probo 3, derive
the formula
'"
V(r,cos O) = I (4n
n=O
+ 3)iL(r,2n + I)P2n+ I(COSO)
Nhere r ~ e, ~ O~ n/2 and

Ae2p (O) r 2 r 2n+1
iL(r,2n+ 1) = 4(n + 1)(2/':r 3n - 1)[( ) e - ( e) J .
~. Let U(r,cos O) be the steady concentration of a substance diffusing through a
)orous hollow hemisphere e < r < 1, < O < n/2 whose hemispherical boundaries
ife kept at zero concentration. Ir the flux of the substance through the base O = n/2,
; < r < 1, is uniform and constant, then
,
OU
roOJ 8=.12
= sinO ~
r sinOoOJ 8=.12
= _!U (rO)= B
r x , , J . ~
f
,
Nherex = cosOand B is a constant. Thus ,
'11
V2U(r,x) = (e < r < 1, < x ~ 1),

U(I,x) = U(e,x) = 0, Ux(r,O)= -rB.
.
'lote that 1'2n{x} is given in Probo 3 and show that
2n+1
BT (x ) e
T { U(r,x) + Brx }
2n
=
,
2n
An+ 1 [
(1 - e2n+2)r2n + (e - e2n)-r
() J
.
In the special case e = O,show that

00
U(r,cos 6) = - Brcos 6 + B (4n +
n=O
1)T2.{x}r2.P2.(cos6).
162 LAGUERRE TRANSFORMS
Operational properties have been developed for another integral trans-

formation whose kernel is a family of orthogonal polynomials, the Laguerre
polynomials.1 The Laguerre transformation is one over the half line x > O,
with weight function e-x. Its properties are presented only briefty here
because the differential form associated with this transformation is not one
that is frequently encountered in differential equations.
Laguerre polynomials of degree n can be written
eXd. . (-lYn!xj
(1) L.(x)= -. d
n. ",,(x.e-X) = }=o
.L ('1)2(
J. n - J.')1'
where n = O,1,2, . . .. In particular
Lo(x) = 1, L1(x) = 1 - x,
They satisfy the differential equation

(2) xL;(x) + (1 - x)L~(x)+ nLix) = O
which has the self-adjoint form
d
dx[xe-XL~(x)] = -ne-XLn(x).
Thus Ln(x) are eigenfunctions and A.= -n are eigenvaluesof the singular
eigenvalue problem consisting of the Sturm-Liouville equation
[xe-Xy'(x)]' - A.e-Xy(x)= O (x > O)
and the requirement that y be of class C" on the half line x ~ O.
We find that the Laguerre integral transformation
oo ~
(3)
Gn{F(x)} = fo F(x)e-xL.(x) dx = fG(n) (n = 0,1,2,...)
has the operational property
(4) Gn{(xe-XF')'} = Gn{xF"(x) + (1 - x)F'(x)} = -nfG(n),
valid if F is of class C" when x ~ Oand if F is (!)(e<l1as x -. 00, where IX< 1.
I McCully, Joseph, The Laguerre Transform, SIAM Review, vol. 2, pp. 185-191, 1960. The
author includes a convolution property for the transformation.
LEGENDRE AND OTHER INTEGRAL TRANSFORMS [SECo 163 453
The set of polynomials Lix), n = O,1,2,..., is orthonormal on the

half line x > O with weight function e - x; that is, when m has one of the
values O,1, 2, ... , then
The corresponding generalized Fourier series representation of a function F

can be written
""
(5) F(x) = L
fG(n)Ln(x)
n=O
(x > O).
That inversion formula for the transformation Gn is valid where F is con-

tinuous if F' is sectionally continuous on each bounded interval and if F
is C9(eU)as x --+00, where IX< t.
The convolution of two functions, whose transform is the product of
the transforms, can be represented by an iterated integral; but the integral
is a complicated one.
163 MElLlN TRANSFORMS

Our formal procedure of designing an integral transformation shows that if
the differential form xF'(x) on the half line x > O is to transform into
- SfM(S),where fM(S) is the transform of F itself, then the kernel of the trans-
formation can be taken as the function XS- 1. The transformation is the
M ellin transformation
(1)
Ms{ F(x)} = 1"" F(x)xS-1 dx = fM(S)
(2) Ms{xF'(x)} = -sMs{F} = -SfM(S).
Formally, from property (2) we find that

(3) Ms{x(xF')'} = Ms{x2F"(x) + xF'(x)} = s2fM(S)
and so
(4)
For further iterations of the operator x d/dx, property (2) leads to the
formula
(5) (n = 1,2,...).
The differential form p(p Vp)pwith respect to p can be isolated in

Poisson's equation in V(p,cf where p and cf>are polar coordinates in the
planeo The forms r2v,r and rV, with respect to r can be isolated in Poisson's
equation in V(r,cf>,lJ)
where r, cf>,and lJare spherical coordinates. Thus the
MeIlin transformation with respect to the radius vector p in the first case,
or which respect to r in the second, will remove those differential forms,
provided that p and r range through all positive values. That range of values
of p and r and conditions under which the MeIlin transforms exist restrict
the types of problems to which M. applies.
Let us assume as usual that F is at least sectionally continuous over
each bounded interval of the positive x axis. Ir s = a + ia where a and a
are real, then
x' = exp (s log x) = exp (alog x) exp (ia log x)
and thus Ix'l = xa when x > O. The integrand of the Mellin integral (1) is
then 19(xa- 1) as x -+ O and we shall assume a > O to ensure the convergence
of the integral at its lower limit. Suppose that F(x) is 19(X-k) as x -+ 00,
where k> O. Then the integrand is 19(I/xk-a+l) as x -+ 00, so the Mellin
transform J.(a + ia) exists when k - a > O and a > O, that is, when
O < a < k, provided that F(x) is 19(x-k) as x -+ 00, where k > O.
When F is continuous (x ~ O) and F' is sectionally continuous, an
integration by parts shows that M.{xF'} exists and that property (2) is valid
if O < (T< k.
The Mellin transformation can be described in terms of the exponential
Fourier transformation Ea. Writing s = (T+ ia and substituting x = e-t,
we find that
that is,
(6)
Thus, known exponential transforms furnish corresponding Mellin trans-

forms, and some properties of M. follow from properties of Ea. ButJhe
relationship also suggests the substitution x = e-r in order to change the
differential form xF'(x) and its iterates into forms to which Ea applies.
According to relation (6) and our inversion formula for Ea (Sec. 132),
when F' is sectionaIly continuous, then
1 p .
F(e-r)e-ar =- lim
2np-oo -p f fM(a + ia)e"arda
if IF(e-r)le-ar is inte.grable from t =- 00 to t = 00, and we find this is true

""
f
lIIIII!IMI"'f'I'I!'!'!!'ftlH!I"!!It!H"
if F(x) is (9(X-k) as x --+00 and ifO < u < k, where x = e-l. Thus by writing
e-I = x, we have an inversionformula for Ms:
1 p .
(7) F(x) = - lim
f
2n p oo -{I
fM(U
U+iP
+ iex)x-u-,,, dex
i
1
= _ . lim fM(S)X-Sds.
2nI p '" u-iP
The kernel-product property
(8) (y > O)
is easily derived. The corresponding convolution of two functions F(x) and

G(x) is (Chap. 10)
(9) XFG(x) = {'" G(Y)F(~) d;

such that, if Ms{ G} = gM(S),then
(lO) Ms{XFG(x)} = fM(S)gM(S),
Tables of Mellin transforms, including someof the properties of Ms,
are given in A. Erdlyi et al., "Tables ofIntegral Transforms," vol. 1. Other
treatments of the theory of the transformation can be found in E. C. Titch-
marsh, "Theory of Fourier Integrals" and G. Doetsch, "Handbuch der
Laplace-Transformation," vol. 1. Other books listed in the Bibliography,
under authors Sneddon and Tranter, include some applications of the
transformation.
Bibliography
Boehner, S.: "Lectures on Fourier Integrals," Prineeton University Press, Princeton,

N.J., 1959.
Carslaw, H. S., and J. C. Jaeger: "Operational Methods in AppliOOMathematies,"
Oxford University Press, London, 1941.
-: "Conduetion ofHeat in Solids," 2d oo., Oxford University Press, London, 1959.
Churehill, R. V.: "Fourier Series and Boundary Value Problems," 2d oo., MeGraw-
Hill Book Company, New York, 1963.
Doetsch, G.: "Theorie und Anwendung der Laplace-Transformation," Springer-
Verlag, Berlin, 1937.
-: "Handbueh der Laplace-Transformation," vol. 1, Verlag Birkhauser, Basel,
1950; vol. 2,1955; vol. 3, 1956.
-: "Einfhrung in Theorie und Anwendung der Laplace-Transformation."
2d ed., Verlag Birkhauser, Basel, 1970.
Kaplan, W.: "Operational Methods for Linear Systems," Addison-Wesley Publishing
Company, Ine., Reading, Mass., 1962.
Lighthill, M. J.: "Fourier Analysis and Generalized Funetions," Cambridge University
Press, London, 1958.
Mikusinski, J.: "Operational Caleulus," Pergamon Press, New York, 1959.
van der PoI, B., and H. Bremmer: "Operational Caleulus BasOOon the Two-sidOO
Laplace Integral," 2d ed., Cambridge University Press, London, 1955.
Sneddon, 1. N.: "Fourier Transforms," MeGraw-Hill Book Company, New York,
1951.
-: "Funetional Analysis," Handbueh der Physik, vol. 2, pp. 198-348, Springer-
Verlag, Berlin, 1955.
Titchmarsh, E. c.: "Theory of Fourier Integrals," Oxford University Press, London,
1937. '
-: "Eigenfunetion Expansions," Oxford University Press, London, 1946.
Tranter, C. J.: "Integral Transforms in Mathematical Physics," 2d oo., Methuen and
Co., Ltd., London, 1956.
Voelker, D., and G. Doetsch: "Die Zweidimensionale Laplace-Transformation,"
. Verlag Birkhauser, Base!, 1950.
Widder, D. V.: "The Laplace Transform," Prineeton University Press, Princeton,
NJ., 1941.
Wiener, N.: "The Fourier Integral," Cambridge University Press, London, 1933.
456
1
BIBLIOGRAPHY 457
Tablas
Campbell, G. A., and R. M. Foster: "Fourier Integrals for Practical Applications,"

D. Van Nostrand Company, Ine., New York, 1948.
Ditkin, V. A., and A. P. Prudnikov: "Integral Transforms and Operational Calculus,"
Pergamon Press, New York, 1965.
Doetsch, G., H. Kniess, and D. Voelker: "Tabellen zur Laplace-Transformation und
Anleitung zurn Gebrauch," Springer-Verlag, Berlin, 1947.
Erdlyi, A., W. Magnus, F. Oberhettinger, and F. Tricomi: "Tables of Integral
Transforms," vols. 1 and 2, MeGraw-HilI Book Company, New York, 1954.
Oberhettinger, F.: "Tabellen zur Fourier Transformation," Springer-Verlag, Berlin,
1957.
...
Appendix A
Tables of Laplace Transforms
Table A.1 Operations'
f(s) (Res> 0:) F(t) (t > O)
F(t)
Y+IP
~
I
2 f(s) -lim
2ni P-., I,-1' et%f(z)dz
3 sf(s) - F(O) F'(t)
4 5"f(s)- r I F(O)
- 5"-2F'(O)- .. . - F(n-I)(O) F(n)(t)
I
5 I -f(s)
s f~ F(T)dT
F(t - b) if t > b
6 I e-bY(s) (b > O)
{O ift O)
f(s
f" e-"F(t)dt
13 I o
I -e -as
>O) F(t) if F(t + a) = F(t)
I See Chaps. 1, 2, and 6 for details and other operations.

458
-
APPENDIX A 459
I(s) (Res> ex) F(t) (t > O)
f
14 "o e-si F(t)
1 + e -as
dt (a > O) F(t) if F(t + a) = - F(t)
00 a
15 2:
.~o s"
~ (k > O)
Tabla A.2 Laplaca transforms'
I(s) F(t) (t > O)
1
3 (n = 1.2 )
s" (n - 1)!
1
4
fi J:;
5
1
s,fi 2jf
6 (n = 1.2 )
1 x 3 x 5...(2n - l)fi
r(k)
7 (k > O) rk-I
T
8 e"'
s-a
1
9 te"'
(s - a)2
1 1
10 (n = 1.2 )
(s - ar (n - l)(-Ie"I \
11 r(k)
(k> O)
(s - at
1 1
-(e"'-eI")
(s - a)(s - b) a-b
s 1
-(ae"' - beb')
(s - a)(s - b) a-b
I For more extensive tables of Laplace transforms see the books by A. Erdlyi et al.. vol. l.
G. Doetsch et al.. or V. A. Ditkin and A. P. Prudnikov. listed under Tables in the Bibliography.
2 Here a. b. and (in entry 14) e represent distinct constants.
460 OPERATIONAL MATHEMATICS
Tabla A.2 (continuad)
f(s) F(l) (1) O)
14
1 (b - e)e'" + (e - a)e" + (a - b)e"
(s - a)(s - b)(s - e) (a - b)(b - e)(e - a)
1 1 .
15 - sm al
S2+ a2 a
S
16 cosat
S2 + a2
1 .
17 - smhal
a
S
18 cosh at
S2 - a2
1 1
19 2"(1 - cos at)
s(s2+ a2) a
1 1 .
20 "3(at - smat)
S2(S2+ a2) a
1 1 .
21 3(sm al - al cosal)
(S2+ a2)2 2a
S 1 .
22 -smat
(S2 + a2)2 2a
S2 1
23 -(sin at + at cosat)
(S2 + a2)2
2a
S2 - a2 t cos at
24
(S2 + a2j2
S cos at - cos bt
25
(S2 + a2)(s2 + b2) b2 - a2
1 1
26 -e"'sinbt
(s - a)2 + b2 b
s-a
27 e'"cosbt
(s - a)2 + b2
3a2 atJ3 r;. at Ij

28
s3 + a3
e-a' - e"'/2
(cos~ - ...;3sm-T )
4a3
29 sin al coshal - cosal sinh al
S4 + 4a4
S I
30 ~2a sin al sinh al
S4 + 4a4
1
31 32a (sinh al - sin al)
1
APPENDIX A 461
I(s) F(/) (1) O)
s 1
32 1 s----
- a 22a (eosh al - eos at)
8a3s2
33 1
(S2 + a2)3
e' dn
34
'1 s t-T
--
s Ln(/) = - -(tne-')
n! dtn
s 1
35 I
(s-a) )mea'(1 + 2al)
1
36 1 -
2fit3(eh' - e"')
1 1
371-
,fi+a )m - ae""erfe(ajt)
38 1 ,fi
s - a2
39 I ,fi
s + a2
40 1 1
JS<s - a2)
1 2 fajl
41 1
JS<s + a2) -e-a"
a..fi J,o eA'dA
b2 - a2
42 1 e""[b - a erf(ajt)] - beh" erfe (bJi)
(s - a2)(b + ,fi)
1
43 1 e"" erfe (aJi)
JS<,fi + a)
1
441
(s + a).;+b
b2 J
45 1 - a2
JS<s - a2)(';; + b)
n!
462 1 (1 - sr
s"+t
(2n)!)mH 2n(Ji)
47 I -(1 - s) n!
, H ~
sn+l,fi 1t(2n + 1)! 2n+ 1('" t)
1 Ln(t) is the Laguerre polynornial of degree n (See. 162).

2Hn(x) is the Herrnite polynornial, Hn(x) = e"'(dn/dxn)(e-X').
Tabla A.2 ("continuad)
/(s) F(t) (t > O)
Js + 2a - 1
-fi
1
49
~fi+b
50 r(k)
(k >0)
(s + 4(s + b)k fiL ~ br\-tt4+b1. Ik_t(a ~ bt)
1
51
~.fi+b<s + b) te-tI4+bIt[Io(a~ bt) + Il(a ~ bt)]
52 ~-.fi
~+.fi
53
(a- bY' (k >0)
(~ + fi+b)2k
54 (~ + .fi)-2Y (v> -1)
.fi~
1
55
J S2 + a2
56 (~-st
(v> -1) aYJ y(at)
JS2 + a2
1
57 (k > O)
(S2 + a2t'
58 (k> O)
59 (v> -1) aY1y(at)

-
60
1
(k> O)
fi !...
k-t It-t(at)
(S2- a2t' r(k) ( 2a)
I
O when O < t < k
61 Sk(t) = {1 when t > k
e-U O when O < t < k
62
SZ {t - k when t > k
O when O< t < k
e-o'
63 (Jl > O) (t - k)"-'
s" when t > k
{ r(Jl)
II.(x) = i-.J.(ix), where J. is Bessel's function of the first kind.

APPENDIX A 463
I(s) F(t) (t > O)
1- e-b I when O < t < k

64
s {O when t > k
65
1 I + coth !ks 1 + [t/k] = n when (n - I)k < t < nk
s(1 - e-Os)= ~ (n = 1,2,...) (Fig.5)
O when O < t < k
1 1 + a + a2 + ... + a"-I
66
s(eh - a)
{ when nk < t < (n + I)k (n -- 1,2,...)
1 M(2k,t) = (-1)"- I
67 - tanh ks
s
when 2k(n - 1) < t < 2kn
(n = 1,2,... )(Fig. 9)
1 1 1 - (-1)"
-M (kt ) + - =
68 2 '2 2
s(1 + e-k')
when(n - I)k < t < nk
1
69 "2 H(2k,t) (Fig. 10)
s tanh ks
70 F(t) = 2(n - 1)
s sinh ks when(2n - 3)k < t < (2n - I)k (t > O)
71 M(2k,t + 3k) + 1 = 1 + (-1)"

5 cosh k5 when(2n - 3)k < t < (2n - I)k (t > O)
72 : coth k5
5
F(t) = 2n - 1 when 2k(n - 1) < t < 2kn
k 1tS
73 -coth- Isin ktl
S2+p 2k
1
74 Hsin t + Isin ti)
(52 + 1)(1 - e-ns)
1
75 -e-Ik/')
5
I
1
76 1"
.Jii cos 2ft
I
1
77
.Jii cosh2ft
1
78
"j;k sin2ft
1
79
J;k sinh2ft
1 t I.-I)/2
80 (/l > O)
~e-Ik/') ( k} J.-I(2ft)
464 OPERATlONAL MATHEMATICS
I(s) F(t) (t > O)
81 (JI> O)
82 (k> O)
83 (k ~ O)
erfe (2~)
1 P
84 (k ~ O)
jmexp ( -41 )
85 (k ~ O) 2 f!...exp-~ -kerfe ~
V; ( 4t) ( 2Jr )
86 (k ~ O)
s(a + ./S) -e"ke""erfe(a.j + 2~) + erfe(2~)
e -k./S
87 (k ~ O)
fl.a + ./S) e"ke""erfe (aJr" + 2~)
e- k,/$+.i O when O < t < k
88
Js(s + a) { e-t"loHaJt2=k2) when t > k
when O < t < k
89
\ ~ { ~o(aJt2=k2) when t > k
e-k./sl-G1 when O < t < k

90
J - S2 a2 {~ o(aJt2=k2, when t > k
e-k(~-S)
91 (k ~ O)
~
92
when O < t < k !

93
when t > k I 1
when O < t < k

J
94 (v> -1) when t > k !

~(~+s)V
1
I
95 -Iogs
s
f'(1) - log t [f'(I) = -0.5772]
1 k- 1 r'(k) log t
96
l< log s
(k> O)
t {[r{kJF - f(k) }
APPENDIX A 465 .
f(s) F(t) (t > O)
97' logs (a>O) e"'[loga + E1(at)]

s-a
logs eos I Si I - sin I Ci I

s2 + 1
s logs -sinlSil - eoslCil

S2 + I
1
100' -Iog(l + ks) (k> O)
s
1 J>t r
101 logd -(e- - e")
s-b I
102
103 (a> O) 210g a - 2 Ci (ar)
2
104 (a > O) -[al log a + sin al - al Ci (al)]
a
2
\ 105 -(1 - eos ar)
I
2
106 -(1 - eoshae)
e
k 1 .
107 aretan -s -Sin
e
kI
1 k
108 -aretan - Si (ke)
s s
1 (2
109 e"'" erfe (ks) (k> O) -exp
k-fi ( - 4k2)
110 ~e"'" erfe (ks) (k > O)

s erf (;k) I
111 (k> O)
-fi
e'" erfe .Jks
nJt<e + k)
1 \ O when O< e < k
112
fierfe (.Jks)
{(no-t whene > k
1 1
113 (k> O)
Jse'" erfe (.Jks) Jn(t + k)
I The exponential-integral funetion El(t) is defined in See. 35. For tables of this funetion and
other integral funetions, see, for instance, Jahnke and Emde, "Tables of Funetions."
2The eosine-integral funetion is defined in Seco35. Si t is defined in Seco22.
I
/ L..
I(s) F(t) (t > O)
1
114 I erf (~) ;t sin (2kfi)
115 I -
1, /Serfc - k -e1 -2e"".'t
Js (
Js ) Fe
[t(2k - t)]-t when O < t < 2k
116 I ne-kslo(ks) {O when t > 2k
k - t
117 e-ksll(ks)
{ ~kJt(2k - t) when t > 2k
when 0< t < 2k
1
118 e'''EI(as)
t + a (a > O)
1
119 I -1 - se"EI(as)
,
a (t + a)2 (a > O)
1
120 I
( - Sis) coss + Ci s sin s t2 + 1
t
121 I
(i - Sis) sin s - Ciscoss t2 + 1
/
Appendix B
Tables of Finite Fourier
Transforms 1
Table 8.1 Finite sine transforms
f/.n) (n = 1.2....) F(x) (O< x < n)
ro F(x)sinnxdx = S"{F}
F(x)
2 '"
2 I f.(n) - I f.(n) sin nx
1t n= 1
3 I -n2f.(n) + n[F(O) - (-I)"F(n)] F"(x)

1 x " x
4 I 2f.(n)
n -
n ~ f
(n - r)F(r) dr - fo
(x - r)F(r) dr
5 I (-Ir+ 11.(11) F(n - x)

6 I 2f.(n)cosnc F1(x + e) + F1(x - e)
2
7 I - f.(n)h.(n)
n f f " H(Y) x+, F1(r)drdy
o x-y
8 I n-x
n
1 x
9 I -(-lr+1
n n
1
10 I -[1
n - (-1)"]
-x ifx<c
11 I
n
cosnc (O< e < n)
{n-xifx>c
- c)x ifx ~ c
(n
12 I ..;
n sin nc
(O< c < n)
{(n - x)c if x ~ e
1 See Chap. 11 for details and operations not tabulated here.
467
468 OPERAT/ONAL MATHEMATICS
1I
Tabla B.1 ,I
(continuad)
fs(n) (n = 1,2,...) F(x) (O< x < n)
6n
13 1 - x(n - x)(2n - x)
n3
2
14 I -[1 - (-1)"] x(n - x)
n3 .
15 I n2 +I 2
-(-1)" --[1-(-1)"]
n n3
16 I 1t(-I)" - --
( n3
6 n
n2)
17 1 [I - (-O"eCO]
n2 + el
n sinh e(n - x)
181 (e #<O)
n + c sinh en I
n
19 I [1 - (-1)" cosh cn] cosh ex
n + c
201
n
(k #< O, + 1, + 2,.. .)
sin k(n - x)
n - sin kn
n
21 I Oirn #<m;f.(m) =- 2 sinmx (m = 1, 2,. ..)
n
22 1 [1 - (-l)"coskn] cos kx (k ,, :t 1, :t 2, .. .)
n-k
1 - (-O"'+'. cos mx
23 I n n-m
2 2 Irn #<m;f.(m) = O (m = 1, 2, .. .)
2kn !... sin k(n - X)

24 1 (n2\- k2)2
(k #< O,:t 1, :t 2, . . .) ak [ sin kn ]
25 I 2cn (e #<O) !... sinh e(x - n)

(n2 + C2)2 oe [ sinh en ]
26 1 4c2(sin2cn + sinh2 en)n

n4 + 4e4
(e #<O) sin ex sinh e(2n - x)
- sin e(2n - x) sinh tx
sinh kx sin kx
27 I 2k2n(-1)"
4 4 (k #<O, :t 1,:t 2,. . .) ---
n-k sinh kn sin kn
e2(- O" sinh ex

--- x
28 1 (e #<O)
n(n2 + e2) sinh en n
k2( -1)" x sin kx

29 I / (k #< O, :t 1, :t 2, . . .)
---
n(n2 - k2) 7t sin k7t
2 bsinx
30 I b" (-1 O)
n cosh y
sin x
- cos x
bn 2 bsinx
32 I - (-1 O) - arctan
n n e'-cosx
2 2bsinx
34 I 1-(-1)n bn (-I<b<l) - arctan-
n n 1 - b2
---- 2 sinx
35 I 1 - (-I)n e -n, (y> O) - arctan-
n n sinhy
Table B.2 Finite cosine transforms
fM (n = O,1,2.00 .) F(x) (O< x < n)
F(x)
I: F(x)cosnxdx = Cn{F}
2 I .fc(n) .fc(0) +
1t
~ ~
1t..=1
.fc(n) cos nx
3 I -n2/.,(n)- F'(O)+ (-lrF'(n) F"(x)
. /.,(0)
4 I
n
/.,(n)ifn = 1,2.... f. (x - r)F(r)dr+ -(x~. - n)2+ A
5 (-Ir/.,(n) F(n - x)
6 2.fc(n)cos ne Fix + e) + F2(x- e)
7 28n {F} sin ne F(x + e) - F(x - e)
1
8 -Sn{F} irn = 1,2,...
" - s: F(r)dr+ A
9 2/.,(n)ht(n)
I:. F2(x- r)H2(r)dr
10 /.,(n) ir n .. 0;/.,(0) + An ir n = O F(x) + A
11 I O irn = 1,2'00' ;(O) = n 1
I
12 I - ir n = 1,2.... ;.fc(0) = O
n -~ log (2 sin~)
2 l ifO < x < e
131 -n sin ncir n ..O;/.,(0)= 2c - n { - 1 ire < x < n
I
J
f,(n) (n = O, 1,2,...) F(x) (O < x < n)
nI
14 x
(-1)"n2 - 1 ifn 'fi 0;.fc(0) = 2"
21t 1t3
15 2( - 1)"if n 'fi O;h(O) = -
n - 3
16
1. (n
--- - X)2 n
2" If n 'fi O; .fc(0) = O
n 2n 6
241t
17 --:-(-I)n fn 'fi 0;.fc(0) = O
n
18 cosh e(n - x)
~'fi~
e sinh en
(-I)"e'n-I 1
19 ~'fi~ -e'X
n2 + e2 e
1 cos k(1t - x)
20 (k 'fi O. :t 1, :t 2, . . .)
n2 - k2 k sin k1t
(-1)" cos k1t - 1 sinkx

21 (k #< O, :t 1, :t2,...)
n2 - p k
(-I)m+n-I sinmx
22 (m = 1,2,...)
n2 - m2 fn 'fi m;.fc(m) = O m
1 1 .
23 -x Sin X
I n2 - 1 fn 'fi l,h(l)
(-Ir! = -4 1t
f 24
Oifn'fim;.fc(m)=z
1t
cos mx (m = 1,2,...)
25
2e(_I)n+! a cosh ex
~'fi~
(n2 + e2)2 OC( e sinh e1t )
26 ~
1 1 - b2
(-1 ~
1tcoshy - cos x
bn 1
28 -n ifn 'fi O;h(O) = 0(-1 O) y - log(2coshy - 2cos x)

n
30
1a sinh y
~>~
1t iy( cos x - cosh y )
n bn
31 -2 -n! if n -# O.' j,(0)
e = 1t exp (b cos x) cos (b sin x)
j
Appendix e
Table of Exponential Fourier
Transforms 1
Table C.1 Experimental Fourier transforms
j.(rx) (-oo<rx<oo) F(x) (-oo<x<oo)
F(x)
f"" F(x)e-1OXdx= E.{F}
2 I J.(rx)
(m = 1,2,...)
lim -1 -p
p"" 27t
P"')(x)
r J.(rx)e'OXdrx
.
43 I (irx)'"I.(rx)
if(rx) xF(x)
5 I eic"fe(rx) (e real) F(x + e)
6 I J.(rx - e) (e real) e'uF(x)
7 I 1.(crx) (e real, e #c-O)

F( )
F(-x)
981 f.(-rx)
1.(-rx) F(x)
10 I J.(rx)ge(rx)
1
r-"" F(t)G(x - t) dt
112 ---
e + irx (e> O) e-e",so(x)
1
12 - (e > O) I "'So(-x)
e - irx
131
(e + irx)2
(e> O) I xe-e",So(x)
1 See Chap. 12 for details and Sec. 136 for references to other tables.
2S0(X) = O if x < O, So(x)= 1 if x> O.
471
472 OPERATlONAL MATHEMATICS
Tabla C.1 (continuad)
f.(a) (-oo<a<oo) F(x) (-oo<x<oo)
1
14 (e> O) - xe""So(- x)
(e - ia)2
2e
15 (e> O) exp (-elxl)
16 (e real) exp(-Ix - el)
4ica
17 (e> O) -x exp (-elxl)
(e2 + a2)2
n!
18 (Rep > O) (n = 1,2,...)
(p + ia)n+1
e
19 7t exp (-elal) (e> O)
e2 + X2
20 (e> O)
2ex
21 i7ta exp ( - elal) (e> O)
- (e2 + x2)2
22 (e> O)
2 .
23 -sIDea (e> O) So(x + e) - So(x - e)
a
-1 if -e < x < O,
24 21-cosea
ia (e> O)
{1 ifO < x < e, Oflxl > e
25 (e> O)
COS2ex iflxl < ~
{O iflxl >-
2e
a1
i
i
i
J
Appendix D
Table. of Fourier Sine and
Cosine Transforms 1
Table D.1 Sine transforms on the half IIne
f.(a) (a> O) F(x) (x> O)
F(x)
(' F(x)sinax dx = S.{F(x)}
2 '" 2
2 f.(a) -Itof f.(a) sin ax da = -SAf.(a)}
It
3 F"(x)
i
4 -E.{ (E. in Appendix C) F(x)
2 F.(xJ}
5 f.(ak) (k> O)
6 !;la) -x2F(x)
2
7 - I.<a)g.(a)
a f'"o F(r)r+r
I%-rlG(r)dr dr
8 2f.(a) cos ak F.(x + k) + F.(x - k)
9 f.(a + k) + f.(a - k) 2F(x) cos kx
10 f').a) (1. in Table D.2) -xF(x)
11 a/.{a) -F'(x)
12 ~.(a) sin :rk F2(x- k) - F2(x+ k)
13 f.(a - k) - fcta + k) 2F(x)sin kx
. For details and propertics not listed here see Chap. 13. Rererences to other tables are given
in Seco 142
473
474 OPERATlONAL MATHEMATlCS
Tabla D.1 (continued)
f.(a) (a> O) F(x) (x> O)
14 2f.(a)g,(a)
{O F(r)[GUx - rl) - G(x + r)]dr
= {O G(t)[F(x + t) + F(x - t)] dt

1t
15
2 x
a
16 (e> O)
e2 +
a
17
1 + a4
2ea
18 (e> O)
(e2 + )2
8c3a
19 (e> O) x(l + ex)e-CX
(e2 + (2)3
1 fil
20
.fi ,r~ fi
2 x
21 (e> O)
;-e2 + X2
4e x
22 (e> O)
-; (e2 + x2j2
4x 3e2 - X2
23 (e> O)
1t (e2 + X2)3
24 (e> O) --2 e2
1tX e2 + x2
2 e
25 (e> O) - aretan -
a 1t X
26 (e> O)
x
27
a (e> O) erfe =
2Jc
1 - eosea 1 ifO < x < e
28
a (e> O) {Oifx > e /
4 x
29
sin ea I x +e
30 (e> O) -Iog-
a 1t Ix - el
APPENDIX D 475
Table D.2 Cosine transforms on the half Une
fc(rx) (rx> O) F(x) (x> O)
fO F(x)
o F(x)cosrxxdx = C.{F(x)}
2 O> 2
2 fc(rx) -non
s fc(rx)cosrxxdrx = -Cx{fc(rx)}
3 -rx2fc(rx) - F'(O) F"(x)

4 tE. {F(lxIH F(x)
5 fc(rxk) (k> O)
6 - x2 F(x)
7 2fc(rx)gc(rx)
{O> F(r)[G(x + r) + G(lx - rl)] dr
8 2fc(rx)cos rxk (k > O) F(lx - kl) + F(x + k)
9 fc(rx+ k) + fc(rx- k) 2F(x) cos kx
10 f~(rx) (fs in Table D.I) xF(x)
1
11 -f.(rx)
rx fO>
x F(r) dr
12 2f.(rx)sin rxk (k > O) F(x + k) - F1(x - k)
13 f.(rx + k) - f.(rx - k) 2F(x) sin kx
14 2f.(rx)g,(rx)
so> o F(r)[G(x+ r) - G1(x- r)]dr.
e
15 (e> O)
e2 + rx2
2
16 (l + x)e-X
(1 + rx2)2
1
17
1 + rx4
exp ( - fi) sin ( + fi)
rx2
18
1 + rx4
exp ( - fi) cos ( + fi)
1
19 [21
.fi. V~vx
20 2 e
(e> O)
~'e2 + X2
21 2 1 - X2
n (l + X2)2
/
Table D.2 (continued)
fc(a) (a> O) F(x) (x> O)
e-a. - e-ea.
22 I (e> O)
a
23 I -1 -e -c
(e> O)
1
-Iog 1 + -X2
e2
a 1t ( )
1 X2
24 I exp (-ea2) (e> O) c: exp --
...1te ( 4c)
1 . 1 2
25 I -e-' SIna - arctan -
a 1t X2
1 . l ifO < x < c

26 I -SIn ea
a
(e> O)
{O if x > c
1
27 I - arctan a E.(x) (E. in Seco 35)
a
2 2
28 I arctan -e-X sin x
x
Index I
I
Abel, Niels, 117 Branch of function, 177

Abel's integral equarion, 53 Branch cut, 178
Abel's test, 234, 308
Adjoint, 277
Analytic continuation, 179 Cauchy, A. L., 3
Analytic function, 165 Cauchy-Riemann conditions, 164
Analytic transforms, 188 Cauchy's integral formula, 170
Angular displacements (see Vibrating shaft) extensions of, 181
Antiperiodic function, 68 Cauchy's integral tbeorem, 170,
182
Characteristic function (see Eigen-
Bar (see TemperatUre Vibrating bar) functions)
Basic operational propeny, 8, 318 Characteristic number, 288 '
Beam: (See also Eigenvalues)
deflection of, 113, 120-122 ClassC', C", 281,327
shearing force in, 113 Concentration, 144
vibrating (see Vibrating beam) (See also Diffusion)
Bessel, F. W., 42On. Continuity:
Bessel's equation, 62, 63, 420 sectional, 5
Bessel's functions, 63, 421 by subregions, 39
product of, 435 Contour integral, 169
zeros of, 241,422,425 Convolution, 318-320
Biharmonic equation, 345,410 for cosine transforms, 406
Bilinear formula, 332 for exponential transform, 391-396
80undary value problem, 127, 397 for finite cosine transforms, 356, 358,
linear, 159 371
in ordinary differential equations, 278-290 for imite exponential transform,
uniqueness tbeorem for, 283 378
superposition of solUtions of, 139, 234 for imite sine transforms, 335, 351,
uniqueness of solution of, 228-230,23411., 354, 359, 373
263. generalized, 83, 324
verification of solutions of, 128, 161, 222- for Hankel transform, 435
226,238-240,257,259,308 - joint, 359, 406
(See also Diffusion; Temperature under for Laguerre transform, 452n.
Vibrating) for Laplace transform, 45-48
477
478 INDEX
Convolution: Exponential function, 32, 166

for Legendre transform, 445 Exponential-integral function, 110
for Mellin transform, 455 Exponential order, 6
for modified Fourier transform, 417
for sine transform, 407
Cosine-integral function, 110 FaltUng integral, 358
Cylinder, diffusion in, 159 Final-value theorem, 83
(See also Temperature) Flux of heat, 143
Formal derivation, 8
Fourier-Bessel integral, 434
Damping, viscous, 86 Fourier-Bessel series, 423, 426
Difference equation, 30, 36 Fourier constaMS, 285
Differential equation, 20-24, 50-54 Fourier integral formula, 198, 312, 315,391
boundary value problem in, 278 Fourier series, 285, 288, 302
existence of solution of, 278
generalized, 285
initial-value problem in, 278 representation theorem for, 291
partial, 123
Fourier transform, 2
(See also Boundary value problem)
cosine, 323, 341,40+408
self-adjoint, 277
table, 475, 476
with variable coefficients, 61-64
Differential form, 277 exponential, 323, 3~397
table, 471, 472
adjoint of, 277, 285
finite cosine, 354-360
transform of, 2, 325
Diffusion, 143, 144 table, 469, 470
in composite solid, 157, 159 finite exponential, 376
table, 379
equation of, 144
finite sine, 339, 348-354
in hemisphere, 451
in plate, 154, 159 table, 467-469
in solid, 245, 249 generalized cosine, 370
Diffusivity, 144 generalized sine, 372
Dirac delta function, 34 on half !ine, 4Oi-418
modified, 414
Dirichlet problem, 448, 450, 451
Distributions, 34 modified finite, 333
Duhamel's formula, 249, 252, 271 convolution for, 325
modified sine, 368
table, 369
sine, 320, 337,401-407
Eigenfunctions, 288
orthogonality of, 290 table, 473, 474
Function:
Eigenvalue problem, 288, 303
singular, 309 even extension of, 405
Eigenvalues,288,290,296 of exponential order, 6
continuous, 309, 312 generalized, 34
Elastic displacements (see Vibrating bar; multiple-valued, 177
Vibrating beam; Vibrating membrane; odd extension of, 135, 405
Vibrating sbaft; Vibrating string) periodic extensioyof, 378
Elastic support, 133, 270, 279 sectionally continuous, 5
Elecnic circuit, 87, 102-106
Kirchhoff's laws for, 102
Entire function, 165 Gamma function, 11
Error function, 46 Green's function, 237, 244, 279-283
complementary,80 construction of, 281, 286
Evaporation, 154, 159, 245 transform of, 332
INDEX 479
Hankd, Hermann, 42On. Inverse transformation:

Hankd transform, 420-437 sine, 337,402
finite, 338, 341,422 Sturm-Liouville, 328
on half line, 432-437 Inversion integral, 193-201
modified imite, 425 aIteration of, 213-216
nonsingular, 431 derivatives of, 202-204
table, 436 series representation of, 206-213
Harmonic function, 177 Iterated transforms, 83, 252
in circular regions, 366, 382
in half plane, 398
Heat conduction, 143,219 Kernd, 2, 317 .
(See also Temperature) Kernel-product property, 319-324

Heat equation, 144 Kirchhoff's laws, 102
fundamental solutions of, 150
Heaviside, Oliver, 3
Heaviside's expansion, 71 Lagrangeidentky,277,285
Hooke's law, 86 Laguerre polynomials, 452
Hyperbolic functions, 166, 167 Laguerre transform, 452
Laplace, P. S., 3
Laplace transformarion, 3
Image, 3 basic operational property of, 8
Impulse function, 30 convolution for, 45-48
Impulse symbol, 33-39 of derivatives, 8-10
lnitial-value problem, 278 generation of, 24
existence theorem for, 278 i[verse of, 14-17, 193-202
Initial-value theorem, 83, 191 operational propenies of, table, 458, 459
Inner product, 25, 284 translation property of, 28
Integral equation, 52-54, 116, 297 Laplace transforms, 3
of convolution type, 52 analytic, 186-189
Integral transformation, 2 derivativesof, 54-56, 188 .
general linear , 317 integrarion of, 66
kernd of, 2, 317 order properties of, 189-193
singular, 336 series of, 57-60
Integrals: tables, 18,458-465
evaluation of, 106-112 Laplace's equation, 177
improper, 13,41,50 Laplacian in spherical coordinates, 442
limit of, 82 Laurent's series, 172
principal value of, 193 Legendre polynomials, 311, 316,443
uniform convergence of, 41 product property of, 445
Inverse transformation, 14 Legendre transform, 442-448
cosine, 404 on interval (-1,1), 443
exponential, 385-387 on interval (0,1),446,447 ./
finttecosine,339,355,370 Legendre's equation, 311,443
imtteexponential,377 Legendre's series, 444
finite Hankd, 424, 426 Leibnitz formula, 39
finite sine, 334, 349, 372 Linear combinarion, 2
Hankel, 434 Logarithmic function, 179
Laguerre, 453
Laplace, 14-16, 193
Legendre, 444, 446, 447 Mellin transform, 453-455
Mdlin, 455 Mortaltty of equipment, 119-122
modified Fourier, 416 Motion of a particle, 11 5, 122
480 INDEX
NatUral frequency, 86 Semi-inf'mite solid:

Neumann problem, 448, 450 diffusion in, 245
Norm, 284 temperarure in, 145-151, 155, 250, 413
Separation of variables, 293, 306
Servomechanisms,117,121
Object function, 3 Sbear, modulus Es' 126, 275
/ Orthogonal functions, 284 Sifting propeny, 33
Orthonormal set, 284 Sine-integral function, 66
Oscillation: Singular point, 62, 173, 310, 311
stable, 76 isolated, 174
unstable, 210 Slab, temperarure in (see Temperarure)
Square-wave funcrion, 67
Stable oscillations, 76
Partial differential equation, 123 Staircase funcrion, 30
(See also Boundary value problem) Steady temperarure, 312
Partial fractions, 17, 70-78 in cylinder, 413, 426
Periodic funcrion, 66,265, 356 in hemisphere, 451
rectification of, 70 in plate, 313, 381
transform of, 67-70 in prism, 313
Plate, deflections in, 410, 413, 414 in quadrant, 408,.418
Poson~equabon,360,426 in sIab, 341, 364, 365, 374, 375, 440
Poson's integral fomrula, 366, 382, 399 in solid, 437
Pole, 174 in sphere, 4:49
Potential: in waIl, 305
in cylindrical region, 314 Srep function, 6, 31
in slot, 360 String, transverse displacements in, 124
Producr: static,279,411
of temperatUre functions, 251 (See also Vibrating string)
of transforms (see Convolution) StUrm-Liouville expansion, 294
StUrm-Uouville problem, 288, 304
reduced fnrm of: 292
Reflection, principIe of, 180 Srurm-Liouville rransfonn, 327
for Laplace transforms, 188 inverse of, 328
Replacement of equipment, 119 Successive transformations, 161, 362, 427
Residuetheorem, 174,183 Superposition of solutions, 139, 234, 293
Residues, 174 Surface heat rransfer, 144
series of, 206-213
Resonance, 92, 259
in bar, 258-266, 273 Tautochrone,115 /
in beam, 269, 273, 274 Taylor's series, 171
in membrane, 274 Telegraph equation, 126, 272
in propeUer shaft, 274, 275 Temperarure, 143
in string, 264 in bar, 153, 158, 220, 233
in vibrating spring, 92, 96,100 in composite solids, 155, 159
Resonance type, 210 in cooling f'm, 153
Riemann-Lebesgue theorem, 192 in cylinder, 240-244, -315, 430
periodic, 250
in slab, 149, 151,158, 233-237, 242,
Schwarz integral formula, 399 243, 368
Sectionally continuous derivative, 198, 206 in soU, 251
Sectionally continuous function, 5 in salid$, 145-149, 220-252, 315,400,
Self-adjoint form, 277 413
j
I
INDEX 481
TemperatUre: Vibrating bar:

steady-state (see Steady temperatUre) with mass attached, 266
TemperatUre functions: resonance in, 259-2~
product of, 251 Vibrating beam, 268, 367
propenies of, 226-228 resonance in, 269, 273, 274
Torque, 126 Vibrating membrane, 274,430
Transfer function, 104 resonance in, 274
t Transform, 2 Vibrating shaft, 126, 142,263,272
Transformation of functions: resonance in, 274, 275
Iinear,2 Vibrating spring, 85-101
r linear integral, 2, 317 Vibrating string, 126-135,264, 314, 366,
Translated function, 28 367, 400
Transmission line equations, 272 equations for, 124, 125
! Triangular wave function, 68 resonance in, 264
j
Vibration, forced, 92, 95
Vibration absorber, 96-98,101
1 Uniqueness:
of inverse transform, 1S, 201, 328
of solutions, 229, 278, 283
l. Wave equation, 123-126
Unit step function, 6, 31
Units, 153, 242 Weierstrass test, 41
Unstable oscillations, 210 Weight function, 284
I
Vibrating bar, 132-142, 253-268, 314, 375 Young's modulus, 113

equations for, 125 for steel, 142

Churchill, R. v. - Operational Mathematics

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Churchill, R. v. - Operational Mathematics

Uploaded by

Copyright:

Available Formats

Operational Mathematics

McGraw-HiII Book Company

Copyright O 1958, 1972 by MeGraw-Hill, lne. Al! rights reserved.

Library ofCongress Catalog Card Number 70-174611

78910 KPKP 7832109

Chapter 1. The Laplace TransformatioR 1

Chapter 2. Further Properties of the Transformation 27

23. Periodic Functions 66

Chapter 3. EIemeotary ApplicatioDS 85

Chapter 4. Problems in Partial Differeotial EquatioDS 123

Chapter 5. FunCtiODS01 a Complex Variable 162

60. Branches of Multiple-valued Functions 177

Chapter 6. The Inversion Integral 186

Chapter 7. Problems in Heat Conduction 219

Chapter 8. Problems in Mechanical Vibrations 253

Chapter 9. GeneraUzed Fourier Series 276

95. Green's Functions 278

Chapter 10. General Integral Transforms 317

Chapter 11. Finite Fourier Transforms 348

Chapter 12. Exponential Fourier Transforms 383

132. The Inverse Transformation 385

Chapter 13. Fourier Transfonns 08 the HaJf Line 401

Chapter 14. Hankel Transfonns 420

Chapter 15. Legendre and Other Integral Transfonns 441

Table A.l Operations 458

This is an extensive revision of the second edition of "Operational

value problems in partial differential equations: the operational methods of

I{F(t)} = I: F(t) dt.

(2) T{F(t)} = rK(t,s)F(t) dt.

When a = O and b = 00 and K(t,s) = e-sr, the transformation (2)

2 DEFINITION OF THE LAPLACE TRANSFORMATION

L{F} = foaJe-SIF(t) dt.

L{F} = fo ~e-st dt = -e-(s-k)1

hence, when s > k, L{e"'}=~.s-k

L !e"I- !e-lct =!~ - !~.

L{sinhkt} = s - ~ (s > Ikl).

(a) L{a + bt} = asS2

(b) L{a + bt + et2} = L{(a + bt) + et2}

(s> a and s> b).

T{F(t) + G(t)} = T{F(t)} +T{G(t)},

3 SECTIONALL y CONTINUOUS FUNCTIONS. EXPONENTIAL

is an example of a function that is sectionally continuous on the interval

thus whenever s > O,

. for the convergenceofimproper integrals (seeProbo14,Seco5). For in view

But the integral from O to 00 of M e- (s-a)' exists when 8 > OC;consequently,

is ensured when 8 > OC.

L{F'(t)} = 1'" e-S'F'(t)dt

= e-S'F(t)J~ + 8 {'" e-S'F(t)dt.

Theorem 1 Let the function F(t) be continuous with a sectionally continuous

Now F(t) is continuous so that F(t 1 - O) = F(t 1 + O),etc., and hence

(3) {T e-stF'(t)dt= -F(O) + e-STF(T)+ s{T e-stF(t)dt.

Theorem 2 Let F and each of its derivatives of order up to n - 1 be con-

5 EXAMPlES. THE GAMMA FUNCTION

Example 1 Find L{t}.

L{F'(t)} = sL{F(t)} - F(O) (s > O),

Example 2 Find L{sin kt}.

Solving for L{sin kt}, we see that

L{sinkt} = s + ~ (s > O).

Example 3 Find L{ tm} where m is any positive integer.

(1) (s > O).

This formula can be generalized to the case in which the exponent

The integral on the right represents the gamma function, or factorial

(2) L{tk} = r(ks + 1) (k > -1, s > O).

Formula (1) is a special case of (2) when k is a positive integer (see