1 Eutiquio C Young Partial Differential Equations An Introduction Allyn and Bacon 1972

PARTIAL DIFFERENTIAL
EQUATIONS
An Introduction
1-:utiquio C. Young
The Florida Seate Unhersi~r
A I lyn an<l Bacon, Inc. Boston

t Copyright 1972 by Allyn and Bacon, Inc.
470 Atlantic AYenuc, Boston.
All rights reserved.
l'."o par! of !he 1110/erial

pro/ec!ed b.1 1his copyr(r;/11 no/ice
may he reproduced or 11li/i~ed in a11_1 j(1m1
or In any means, e/eclronic or 111echa11ical.
i11c/11di11g pho1ocop_1 i11g, recording, or b1 a11_1
i11fomwlio11a/ storage and relrie1 al .1Tsle111
1ri1ho11/ Tillen permission
from rhe copyrighl 01111er
Librar) of Congress Catalog Cnd Number: 70-166551
Printed in the United States of America.

T O M Y PA R EN T S A N D M Y F A ~I I L Y
Contents
Preface 1x
Chapter 1 Introduction
1. Some Properties of Functions of One Variable 2. Partial

Derivatives 3. Differentiation of Composite Functions; the
Chain Rule 4. Differentiation of Integrals Depending on a
Parameter 5. Uniform Convergence of Series 6. Im-
proper Integrals Depending on a Parameter 7. Directional
Derivatives 8. Green's Theorem and Related Formulas
Chapter 2 Linear Partial Differential Equations 36

J. Basic Concepts and Definitions 2. General Solutions
and Auxiliary Conditions 3. Linear Operators and Prin-
ciple of Superposition 4. Linear First-Order Equations
5. Solutions of Second-Order Equations with Constant Co-
efficients 6. Classification of Second-Order Equations
7. The Canonical Forms
Chapter 3 The Wave Equation 71

1. The Vibrating String 2. The Initial Value Problem
3. Interpretation of the Solution 4. Domain of Dependence
and Characteristic Lines 5. The Nonhomogeneous Wave
Equation 6. Uniqueness of Solution 7. Initial-Boundary
Value Problems 8. Nonhomogeneous Problems and Re-
flection of Waves 9. Method of Separation of Variables
vii
\iii Contents
Chapter 4 Green's Function and Sturm-Liouville Problems 121

1. Homogeneous Boundary Value Problems 2. Nonhomo-
geneous Problems; Green's Function 3. Modified Green"s
Function 4. Sturrn-Liouville Problems 5. Orthogonality
of Eigenfunctions 6. Eigenfunction Expansions; Mean
Convergence 7. Nonhomogeneous Sturm-Liouville Prob-
lems; Bilinear Expansion
Chapter 5 Fourier Series and Fourier Integral 163

I. Orthogonal Trigonometric Functions 2. Fourier Series
3. Fourier Cosine and Sine Series 4. Bessel's Inequality;
Riemann-Lebesgue Theorem 5. Convergence of Fourier
Series 6. Uniform Convergence of Fourier Series 7.
Fourier Integral 8. Fourier Transform
Chapter 6 The Heat Equation 212

1. Derivation of the Heat Equation 2. Initial and Boundary
Conditions 3. The Maximum Principle and Uniqueness
Theorem 4. Initial-Boundary Value Problems 5. Non-
hornogeneous Initial-Boundary Value Problems 6. The
Initial Value Problem 7. Initial-Boundary Value Problems
in Infinite Domain
Chapter 7 Laplace's Equation 250

1. Boundary Value Problems 2. Green's Theorem and
Uniqueness of Solutions 3. Maximum Principle for Har-
monic Functions 4. Dirichlet Problem in a Rectangle
5. Dirichlet Problem in a Disk 6. Poisson's Integral
Formula 7. Neumann Problem in a Disk 8. Problems
in Infinite Domains 9. Fundamental Solution and Green"s
Functions JO. Examples of Green's Functions 11. Neu-
mann's Function and Examples
References 311
Solutions of the Exercises 313
Index 343
Preface
This book is the outgrowth of an introductory course in partial differential

equations which the author has given for a number of years. The students are
mostly undergraduates who are majors of mathematics, engineering, or the
physical sciences.
The purpose of the book is to acquaint the students with some of the tech-
niques of applied mathematics and to provide them with basic material necessary
for further study in partial differential equations. As the needs and interests
or the students are varied, the author attempts to strike a balance in emphasis
between theory and application.
The book treats principally linear partial differential equations of the first
and second order involving two independent variables. Problems involving
:,econd-order differential equations are discussed with reference to the three
prominent classical equations of mathematical physics, namely, the wave
equation, the heat equation, and Laplace's equation. These equations serve
as prototypes for the three main types of linear partial differential equations of
second order. The kinds of problems that are treated for each of these equations
in two variables as well as the properties of the solutions are generally typical
of what can be expected with more general differential equations of the same
type in three or more independent variables.
Several techniques of applied mathematics-such as the method of eigen-
function expansion, the Fourier transform, and the use of Green's function-
are developed and their basic underlying theory are discussed. Along with
many examples, a sufficient number of exercises of varying degree of difficulty
appear in nearly every section. These exercises form an integral part of the text.
They are designed not only to test comprehension of the subject matter pre-
ix
x Preface
sented, but also to introduce other general ideas and procedures. Answers to
almost all exercises are given at the back of the book. A list of selected references
for further reading on the subject is also given.
As a prerequisite for a course based on this book, the student must have a
working knowledge of the topics usually covered in a standard calculus course
and must be familiar with the contents of a basic course in ordinary differential
equations. Many of the topics in the calculus which are extensively used in
later discussion are discussed briefly in the beginning chapter. For those
students who have an adequate background, this material can serve as a review.
Although the book is intended for use in a two-quarter course, it can be
adopted for a one-quarter or a one-semester course, depending on the back-
ground of students and the interest of the instructor. For example, topics
selected from Chapters 2, 3, 6, and 7 can be the bases of a one-quarter course
for students who are already familiar with the contents of Chapters 4 and 5.
A word about the numbering of equations and theorems: Unless a different
chapter is explicitly stated, the first number indicating an equation or a theorem
always refers to the section of the particular chapter under study. The exercises
are, however, numbered according to the chapters. Thus, the heading Exercises
5.1 refers to the first set of exercises in Chapter 5.
The author wishes to thank Professors Thomas G. Hallam and Howard E.
Taylor for testing some of the material in its early version and for making
helpful comments, and to Professor C. Y. Chan for reading the manuscript
and for making many invaluable suggestions. The author also acknowledges
with gratitude the secretarial help extended by the Department of Mathematics.
Last but not least, it is a pleasure to thank Mrs. Margaret Parramore for her
skillful typing of the manuscript.
Tallahassee, Florida EUTIQUIO c. YOUNG

Partial Differential Equations
An Introduction
Chapter 1
Introduction
In this chapter we shall review and discuss some topics from the calculus
concerning functions of a single variable and of several variables. These topics
will be needed in our later work. The consideration of functions of several
yariables will be limited mostly to functions of two independent variables, as
this will be adequate for our purpose.
1. Some Properties of Functions of One Variable
Let f be a function defined on the interval a ~ x ~ b and let x 0 be a point

or this interval. We define the left-hand limit of I at Xo, denoted by /(xQ"), to
be the limit off as x approaches x 0 from the left of x 0 ; that is,
/(x 0 ) = lim /(x)
x-+xo
x<xo
In a similar manner we define the right-hand limit of/at x 0 to be
/(xri) = Jim /(x)

x-+xo
x>xo
If bothf(x 0 ) and /(xri) exist and are equal to the value off at x 0 , then obviously
we have
Jim /(x) = /(x 0 ) = /(xri) = f(x 0 )
x-+xo
In this case the function/is said to be continuous at x 0 . On the other hand, if

both /(x 0 ) and /(xO) exist but are not equal, then f is said to have a jump
2 lntrod11ctio11 C!tap. 1
discontinuity at x 0 . The differencej(xci) -f(x 0 ) is called the jump of the

function at the point x 0 .
Example 1.1. The function

f(x) = Ix - i (x > 0)
\x + I (x < 0)
has a jump discontinuity at x = 0, since/(0-) = 1 and/(O+) = - I. The jump

of the function at this point is - 2.
A function f is said to be continuous on the interval a ::::; x :::::; b ifj is con-

tinuous at all points of the interval. We remark that at the end points x = a
and x = b, we require only that f(a+) = f(a) and f(b-) = f(b). It is well
known that if a function is continuous on a closed interval, then it is bounded on
that interval. This means that there is a number M such that
lf(x)I :::: M
for all x, a :::::; x ::::; b. The converse of this statement is, of course, not true. For
example, the function considered in Example 1.1 is bounded on the interval
[ - I, I J with M = 1, but it is not continuous there because it has a jump
discontinuity at x = 0.
A functionfis said to be piecewise continuous on the interval a ::::; x :::::; b ifit
is continuous on that interval, except possibly at a finite number of points
where it has jump discontinuities. Thus the function in Example I. I is piecewise
continuous. It is clear that every piecewise continuous function on a closed
interval is also bounded.
Letfbe a continuous function on the interval a ::::; x ::::; b. We say that.f has
a left-hand derivative at the point x 0 of the interval if the limit
(1.1) Jim f(x 0_+ h) - f(xi)

h-O fi
li<O
exists. We denote this one-sided derivative by f'- (x 0 ). In the same way, we say
that f has a right-hand derivative at x 0 , denoted by f~ (x 0 ), if
(1.2)
. f(xo
I1111 -
+ h) - f(xo)
h~O Ji
h> 0
exists. If both f'_(x 0 ) and f~(x 0 ) exist and are equal, then obviously f has a
derivative at x 0 , and we have
It should be noted that f'- (x 0 ) and f~ (x 0 ) are not the same as f'(x 0 ) and
f'(xci), which are the left-hand limit and the right-hand limit off' at x 0 , respec-
tively. The existence of the one-sided derivatives off at x 0 does not imply the
Sec. 1 Some Propenies of F1111ctio11s of One Variable 3
existence of the one-sided limits off' at x 0 . But, if both {'(x 0 ) and f' (x;) exist
and are equal to f'(x 0 ), then we have
lim f' (x) = f'(x 0 ) = f'(xci) = f'(x 0 )
x-+xo
\\hich means thatf' is continuous at the point x 0 .
Example 1.2. Consider the function
f(x) = l{.Ox2 sin xi (x # 0)
(x = 0)
This function is continuous at all points, including x = 0. Now, for h < 0, we

have
' O) . h 2 sin(l/h) - /(0-)
! _( = 11111 ""
11~0 h
2
= lim h sin(l/h) = 0
h~o h
Similarly, when h > 0, we find/~ (0) = 0, and so f'(O) 0. But when x cl 0,

we have
1 1
f'(x) = 2x sin - cos
x x
for which neither /'(0-) nor f'(O+) exist. In other words, f' is continuous at all
points x c;t 0, but is discontinuous at x = 0, although f'(O) = 0.
We say that a function f is smooth on the interval a s x s b if it has a

continuous derivative at every point of the interval. At the points x = a and
x = b, we require only that f'(a+) and f'(b-) exist. Thus, the function con-
sidered in Example 1.2 is smooth on any interval [a, bJ that does not include the
origin. Geometrically, the graph of a smooth function consists of a continuous
curve that has a tangent which turns continuously as the tangent moves along
the curve; that is, the curve does not have any "corners."
In the case where the function f is piecewise continuous on a s x s b, and
x 0 is a point of discontinuity, the definition of left-hand derivative and right-
hand derivative given by equations (I. I) and ( 1.2) remains valid, provided the
quantity f(x 0 ) in those equations is replaced respectively by the left-hand limit
and the right-hand limit off at x 0 ; that is, by f(x 0 ) and f(xri). rn this case,
equality between/~ (x 0 ) andf~ (x 0 ) does not mean existence off' at x 0 , since f is
not continuous at x 0 . We say that f is piecewise smooth on the interval
a s x s b ifit is piecewise continuous and has a piecewise continuous derivative
on that interval. It follows that the graph of a piecewise smooth function is
either a continuous curve or a discontinuous curve that has a continuously
turning tangent between any two consecutive points where f or f' has jump
discontinuity.
4 lntrod11ctio11 Chap. I
Example 1.3. The function

e-' (-1 :<; x :<; 0)
j(x) ~ l (0 :<; x < 1)
(
cos nx ( 1 < x :<; 2)
\\hose graph appears in Fig. I I, is piecewise smooth, since the function and its
derivative are piecewise continuous. The function has a jump discontinuity at
x = 1, \\hi le the derivati\ e /' has jump discontinuities at x = 0 and x = I.
FIG. 1-1 Pieceirise smooth f1111ctio11.
A function that is defined on the whole x-axis is said to be piecewise con-

tinuous or piecewise smooth if it is piecewise continuous or piecewise smooth on
every finite interval. This definition applies in particular to a periodic function:
that is, to a function/ ha\ ing the property
( 1.3) /(x + T) = /(x)
for all x and some constant T. The smallest value of Tfor which equation (1.3)
holds is called the period off
lffis a function defined only on a finite interval a ::; x ::; b, we can extend it
so that it yields a periodic function. In fact, let T = b - a and consider the
function </>, which satisfies the relation
</>(x + /1 T) = f(x) (a ::; x ::; b)
for all integers 11. Clearly, </> is defined for all x and is periodic with period T.
Moreover, it coincides withf on the interval a ::; x ::; b. We call the function <fJ
the periodic extension off with period T.
One important theorem of differential calculus, which we ~hall have occasion
to use later. is stated below.
Sec. 1 S'ome Properties o( F1111ctio11s of One Variable 5
THEORE\1 1.1. (Mean Value Theorem) If" f is continuous 011 the interrn!
a s x s
b andf' exists for each point x inside the interrn/, then thae is a point
x 0 het1ree11 a and b such that
(1.4) /(h) - /(a) = f'(xo)

b - a
The formula ( 1.4) is frequently written in the form
f(a + h) = f(a) + hf'(a + Oh)
where h = b - a and 0 < 0 < 1. If h is very small, then \Ve can approximate
f(a + h) by writing /(a + h) = f(a) + hf'(a).
We next recall some facts from integral calculus. Let f be a piecewise con-
tinuous function on the interval a s x s b. Then the integral of/ over [a, b J
exists and is the sum of the integrals over the subintervals whose end points are
the points of discontinuities off Moreover, from the geometrical interpretation
or definite integrals it is clear that the integral
( 1.5) f l/(x)I dx
exists. Jn this case we say that/ is absolutely integrable on the interval [a, b].
More generally, we say that a function/ is absolutely integrable on the interval
[a. b] (where a may be - oo or h may be oo) if/ is continuous on the interval,
except possibly at a finite number of points, and the integral (J .5) exists-
possibly as an improper integral. Over a finite interval we see that every piece-
wise continuous function is absolutely integrable. The converse of this statement
i~. however, not true. For instance, the function /(x) = !//-.; is absolutely
integrable on the interval 0 s x s 1, but it is not piecewise continuous there
because the right-hand limit/(0+) does not exist.
It is true, however, that every absolutely integrable function is integrable in
the ordinary sense. That 1s, the existence of the integral ( 1.5) implies the
existence of the integral
f /(x) dx
Further, if /is absolutely integrable and g is a bounded function on the interval

a 5 x s b, the product.fg is also absolutely integrable on the same interval.
We now state the fundamental theorem of calculus.
THEOREl\1 1.2. Let .f be a continuous.function on the interval a s x s band let

c be a fixed point of the interrnl. Define the function F by
( 1.6)
for each x, a s x s b. Then

F(x) = r f(t) dt
(1.7) F'(x) = d fx f(t) dt = .f(x)

dx c
6 Introduction Chap. I
(1.8) r
This theorem is also written frequently in the form
f(x) dx = F(b) - F(a)
where Fis any function such that F'(x) = f(x). It is easy to see that (1.6) and
(1.7) together imply (1.8), and vice versa.
Finally, we state a theorem that is the counterpart of Theorem 1.1.
THEOREl\1 1.3. (Mean Value Theorem for Integrals) If f is continuous 011 the
interi:al a ::::; x ::::; b, then there is a point x 0 betireen a and b such that
(1.9) f f(x) dx = f(x 0 )(b - a)
Geometrically, this formula says that there is a rectangle of height f(x 0 ) and
width (b - a), whose area f(x 0 )(b - a) is equal to the area determined by the
integral (1.9).
Exercises 1.1
1. Sketch the graph of each of the following functions and determine whether the
function is piecewise continuous, continuous, piecewise smooth, and smooth:
(a) f(x) =
{- ,1_:-x: (-1 < x < 0)
\x3 (0 < x < I)
(b) f(x) =
r-,,
0 ,-
vx
(x < I)
(x = I)
(x > I)
2. Let f be a function defined on the interval - L s x s L. State what properties

must be required off in order for its periodic extension to be piecewise continuous,
continuous, piecewise smooth, smooth.
3. Consider the function
. I
f(x) =
X Sll1 -
x
(x * 0)
(
0 (x = 0)
Show that/ is continuous but does not have a derivative at x = 0.

4. Consider the function f(x) = x - [x ], where [x] denotes the greatest integer
thatdoesnotexceedx. Showforeachintegernthatf'(n-) = f'(n+) = f~(n) =I,
but that f ~ (11) does not exist.
5. Let
(x S 0)
(x ;::::: 0)
Graph the function and show that f~ (0) = f'(O+) and !~ (0) = /'(0-). Does
f'(O) exist?
Sec. l P11rti11f Derfratives 7
6. Let
ex < 0)
= I\ -e - x
(x
/ (x)
(x > 0)
Show that f.'_(0) = f'(O-) = /~(0) = /'(O-). Does /'(0) exist? Graph the
function.
7. Show that the function
1
x 3 sin (x 1' 0)
/(x) x
(
.o (x = 0)
is smooth in any interval [a, b ].

8. Give an example to show that the product of two absolutely integrable functions
is not necessarily absolutely integrable. Hinr: Take /(_r) = 1/, /:\.
9. Show that the function /(x) = l /( l + x 2 ) is absolutely integrable on the interval
(- oc, oc).
IO. Use the result of Problem 9 to show that the integral
oc cos X dx
exists.
J o l + x2
1]. Give a geometrical interpretation of the mean value theorem (Theorem 1.1 ).
12. Let /(x) = Ax 2 + Bx + C, a ~ x ~ h, where A, B, Care any constants. Show
that x 0 = (a + h)/2 is the value that satisfies the mean value theorem.
13. Show that there is no continuous function f that satisfies
unless a = 0.
x" = r /(1) dt (11 > 0)
14. By using the mean value theorem for integrals, prove that the function defined by
( 1.6) is continuous on a ~ x ~ h. Hint: Consider F(x + /z) - F(x) =
J~+ h /(!) dt and show that this tends to zero ash _, 0.
15. Show that equations (l.6) and (1.7) imply (1.8), and vice versa.
2. Partial DerirntiYes
We now review some concepts concerning derivatives of functions of several

variables. Let u be a function of the two independent variables x, yin a domain
D and let (x 0 , y 0 ) be a point of D. If we set y = y 0 , then u becomes a function
of the variable x alone. Its derivative at x = x 0 is defined as the limit
.
IIm
u(x 0 + ~x, y0) - u(x 0
.
, y0 )
6.,~o ~x
whenever the limit exists. This derivative is called the first-order partial deriva-
tive of u with respect to x at the point (x 0 , y 0 ), and is denoted by (i3u/i3x)(x 0 , y 0 )
8 Introduction Chap. 1
or u)x 0 , y 0 ). Frequently, we shall also use the simpler notation ux whenever the
point at which the derivative is being evaluated is clear from the context. In the
same way, we define the first-order partial derivative of u with respect to y at
(x 0 , Yo) as the limit
ou ( ) . u(x 0 , Yo + ~y) - t1(xo , Yo)
-- Xo,Yo = 11m
oy ~y-o ~y
prO\ ided the limit exists. It follows from the foregoing definitions that, given a
function u of two variables x, y, the partial derivative of t1 with respect to either
one of the variables, whenever it exists, can be calculated by simply differentiat-
ing the function with respect to the variable concerned while treating the other
variable as a constant.
Example 2.1. Find the first-order partial derivatives of the function

u(x, y) = x cos xy + y
Sol11tio11: The first-order partial derivatives of the given function exist. To find 11x,
we differentiate with respect to x, treating y as a constant, and obtain
11_y(x, y) = cos xy - xy sin xy
Similarly, we obtain 11/x, y) = - x 1 sin xy + 1.
The partial derivatives ux and uY of a given function t1 can be given geometrical

interpretation if we regard the equation z = u(x, y) as representing a surface in
the xyz-space. In fact, when we set y = Yo, z = u(x, y 0 ) represents the curve of
intersection of the surface z = u(x, y) and the plane y = Yo. Thus, u)x 0 , y 0 )
is the slope of the tangent line to the curve z = u(x, y 0 ) at x = x 0 . Likewise,
uy(x 0 y 0 ) represents the slope of the tangent to the curve z = u(x 0 , y) at
y = y 0 , where z = u(x 0 , y) is the curve of intersection of the surface with the
plane x = x 0 . The tangent plane to the surface at the point (x 0 , y 0 ) can then
be shown to have the equation
z - z 0 = uJx 0 , y 0 )(x - x 0 ) + u,(x 0 , y 0 )(y - y 0 )
where z 0 = u(x 0 , )' 0 ).

It is clear that the value of u, and uY at a point depends on the coordinates
(x, y) of that point. This means that the partial derivatives ux and uY are also
functions of the variables x and y; hence, they, too, may admit partial derivatives
with respect to one of or both the variables. These derivatives, if they exist, are
called the second-order partial derivatives of 11. The four second-order partial
derivatives of 11, for instance, are
ou
2
o2 u
llxx =
ox
- 2 '
Uyx
:x e~) OX OJ
a ou 2
o2--u
II xy
oy (:~) oy ox
II !"J' -
0)'2
,<frc. 2 Partial Deriwttives 9
assuming these exist. Third-order partial derivati\es of 11 are obtained if the

preceding derivatives can be further differentiated with respect to x or J'.
Derivatives of order greater than the first are generally referred to as higher-
order derivatives: in particular. the derivatives uxy and urx together with all
higher-order derivatives of 11 with respect to both x and .r are called mixed
derivatives.
Example 2.2. Find the second-order partial derivatives of the function 11 =
x cos xy + y considered in Example 2.1.
Su/111io11: The second-order partial derivatives are

llxx = - 2y sin xy - xy 2 cos xy, -2x sin xy - x 2 y cos xy
"-rx = - 2x sin xy - x 2y cos xy, II.IT -x 3 cos xy
We notice in Example 2.2 that the mixed derivatives uxr and u,.x are equal.
This means that the order in which the differentiations are performed is im-
material. This is by no means true of all functions (see Problem 20, Exercise
1.2); however, whenever the mixed derivatives in question are continuous, it can
be shown that the order of differentiation does not matter. Thus, for the
function considered above, there also follow uxxy = u,.vx = llyx.n uxyy =
uyxy = urr.n etc .. as one easily verifies, since the function has continuous deriv-
atives of all order. Unless otherwise stated, we shall always assume in this work
that mixed derivatives are continuous whenever they exist.
Exercises 1.2
In Problems 1 through 5, find the first-order partial derivatives of the given function.
I. 11 = x 2 y + xy 2 .
2. 11 = e" cosy + xy.
3. 11 = ln(x 2 + y 2 ). (x. y) o;!c (0. 0).
x - y
4. 11 = - , (x, y) F (0, 0),
.\' + )'
x
5. 11 = tan , y > 0.
y
6. If 11 = x 3 + 2x 2 y + 3xy
2
+ y 3 , verify that :rnx + J'llr = 311.
7. If 11 arcsin ' - v). , (x.

(x+y y) # (0, 0), verify that xu, + yu,. = 0.
8. If /1 (ax + by)", 11 a real constant, verify that b11x - au, = 0.

9. If u = x y
2
+ y 2z + z 2 x, verify that llx + lly + llz = (x + y + z) 2 .
10. (Euler's Theorem). A function 11 of the variables x, y is said to be homogeneous
of degree 11 if 11(1x, ty) = 1 11 11(x. y). If 11 is homogeneous of degree /1 with con-
tinuous first derivatives, show that .rnx + y11" = 1111. Hint: Differentiate
11(tx, ty) = t"u(x, y) with respect to t.
10 /11troductio11 Chap. 1
In Problems 11 through 14, find the second-order partial derivatives of the given
function.
]], II= (xl + y1)1i2.
12. I{ = ellX + b)-.
13. 11 = e" sin y.

14. 11 = arctan x/y.
15. If 11 = ln(x 2 + y 2 ) 1 !2, (x, y) I= 0, show that llxx + Lin 0.
16. If 11 = arctan y/x, show that llxx + uY,. = 0.
17. If 11 = x", x > 0, show that u<J' = u,.x.
18. If 11 = sinh(x + at), show that 11" - a 2 uxx = 0.
19. If u =
x- - -
( x+y
y) , (x, y) I= (0, 0), show that llxy = u,.x .

x2 _ y1
xy [(x, y) I= (0, 0)]
u(x, y) = 0 x2 + y2
(
[(x, y) = (0, 0)]
Using the definition, show that ux<O, y) = - y, and 11/x, 0) = x; hence, deduce
that

u(x, y) = x 2 arctan y - y 2 arctan -~ [(x, y) I= (0, 0)]
x y
11(0, y) = u(x, 0) = 0
Show that for (x, y) I= (0, 0), llxy = llyx, but llx/O, 0) I= uyx<O, 0).
3. Differentiation of Composite Functions; the Chain Rule
Suppose that u is a function of the variables x and yin a domain D,
(3.1) u = U(x, y), (x, y) in D
and suppose that x and y are themselves functions of the variable t,
(3.2) x = f(f), y = g(t) (a s t s b)
such that for each t in [a, b], (f(t ), g(f )) is in D. Then, in effect, u is a functi.on
of the single variable t:
(3.3) u = U(x, y) = U[f(t), g(t)] (a s t s b)
The derivative du/dt, if it exists, can then be calculated by simply differentiating

the function (3.3) with respect tot. However, it is frequently desirable to obtain
the derivative of u with respect tot without an actual substitution for the variables
Sec. 3 Dijf'errntiatio11 o/' Composite F1111ctio11s; tlte Cltai11 Rule 11
x and y. One important and useful tool in this connection is called the '"chain
rule."
THEORE\I 3.1. (The Chain Rule) If rhefi111ction 11 in (3.1) and irsfirsr-order

partial deriwtires are continuous in D, and if rhe functions f and q in (3.2) are
differentiable in a < r < b, then u is a diff"erenriablefi111crio11 of r 11 hose deriwtire
is gii"en b_1' the formula
du i'u dx iu dy cU df iU dg
(3.4) = + - =- +
dt ax dr iy dr ax dt cy dt
Several modifications of this formula will be considered in the rest of this

section. We first illustrate its use in the following example.
Example 3.1. Let u(x, y) = e-' sin y, \\here x = Int and y = (1 - t 2 ), t > 0.
Find du/dt by substitution and direct differentiation and by the use of formula (3.4).
So/111io11: Substituting for x and yin 11, we obtain
11 = tsin(l - t 2 ), (I > 0)
\\hose derivative is easily found to be
du= sin(l - t 2 ) - 21 2 cos(l - 12 )

dt
On the other hand, \\e have

x . dx dy
llx = e Sill y, and -2t
di dt
so that by (3.4) we find
du , . dx- (e, cosy ) dy

(e sin y) +
dt dt dt
1
t[sin(l - 12 )) + 1[cos(I - 1
2
))(-21)
I
sin(! - 1 2 ) - 21 2 cos(! - 1 2 )
Formula (3.4) extends easily to the case where u is a function of three or more
independent variables. For example, if u is a function of the variables x, J', and z,
u = U(x, _r, z)
where
x = x(t), y = y(l), and z(t)
then
(3.5)
du au dx
+.au- dy
- + au
- - dz
-
dt ox dt ay dt cz dt
12 lntroduct ion Chap. I
pnnided U and its lirst-order partial deri\atives are continuous and x, y. z are
differentiable. In the special case where x = t. so thaty and z become functions
of the\ ariable x. formula (3 . .5) reduces to the form
(;u i' c_; dy i'U d:
(3.6) + (".\" +
i7x dx CZ dx
We shall h~l\e special need for this formula in the next section.
Next. suppose that 11 is a function of the\ ariablcs x. y where
x = x(s. t) and y = y(s, t)
that is. x and r are in turn functions of two independent variables s and t.
Then 11 may also be considered as a function of the variables s, t:
(3.7) 11 = U(x, y) = U[x(.1, t), )(S, t)]
and we may investigate the partial derivative of u with respect to sort when the
other variable is held constant. Under the condition that U and its first-order
partial derivatives are continuous and that the first-order partial derivatives of
x and .r exist, it can be shown that
cu au ux au rJ\"
(3.8) +
cs ex cs (r cs
and
(""}II cU ax au er
(3.9) + oy cit
(~( OX 1:1
These formulas are modification of the chain rule (3.4).
Finally, we consider a function iv of four variables x, y, u, and z:, defined by
w = ~V(x, J, 11, r)
where u and r arc, in turn, functions of the independent variables x and y. We

arc interested in obtaining formulas for the partial derivatives of 11 with respect
to x and r. To determine ('\\Rx, we hold y fixed and regard\\' as merely a func-
tion of x, 11, and r, remembering that both u and P depend also on x. By a
modification of formula (3.6) we have
,~\\' aw aw Du aw Du
ex + +
(3.10)
ax ex c11 Or ox
In the same way, we have
ow cw cw 011 aw av
c11 ay + ar: ay
(3.11)
ay cy +
These formulas can, of course, be established under the assumption that all
first-order partial derivatives of Ware continuous and that all first-order partial
derivatives of u and 1 exist. The student should carefully note the distinction
betv>een the terms cH/i'-x, 21r/i'y. and the terms iJ W/Dx, DW/ay.
Sec. 3 Differentiation of Composite Functions; the Chain Rule 13
Higher-order partial derivatives of a composite function can be calculated by

successive application of the appropriate formulas derived above. For instance,
the second-order mixed dcrivati\e of the function (3.7) with respect to sand t
can be obtained by applying formula (3.9) to the function defined by formula
(3.8). This gives
c2 u a (cu ex) a (c:u er)

ct as 21 2:, 1cs + ct cy ;;~.
o U ex
2
i'/U c1) ox+ iJU o x
2
(cx c'I 2
+ oy ex ct. cs ex ct OS
+ (. u2 u ex + u2 u a_r) cy + au a2 y
ex CJ ct oy 2 ct cs OJ ct OS
i/U ex 0 LI (ax cy
(;X
2
lX C_I")
<'x2 cs Dt
+ C_\' OX .cs cc t +
ct cs.
1
a u a_r ay cu c:x au a2y ~1
+ a_r2
cs ct + ex at e's + ay at as
\vhcre we have assumed that Uxr = Urx.
Example 3.2. Let 11(x, y) = x 2 + y 2 , where x = r cos 8 and y = r sin 8. Find the
partial derivatives of 11 with respect tor and 8 up to the second order.
Sof111io11: Using formulas (3.8) and (3.9). withs = rand I = 8, we find

11, = 2x cos {) + 2y sin {)
2r(cos 2 8 + sin 2 8)
2r
11 0 2x( - r sin 8) + 2y(r cos 8)
-2r 2 (sin ()cos 8 - sin 8 cos 8)
0
from which there follow
Example 3.3. Show that the function 11 = f(x 2y) + g(x + 3y),where/andgare
twice differentiable. satisfies the equation
So/111io11. Let 11 = x - 2y and r = x + 3y. Then w becomes a function of the

variable 11 and z:
w = /(11) + g(r) = W(u, i)
Using formulas (3.10) and (3. l I) and noting that W depends only on u and i-, we
find
\V x W. 11x +
1
11~1, = f'(u) + g'(io)
w;,u,. +
w;i-,. = ('(11)( - 2) + g'(v)(3)
- 2/'(u) + 3g'(i)
14 /11troduction Chap. I
Further application of the same formulas gives
ll'xx = /"(11) + g"(r)

)\'X) (- 2)/"(11) + (3)g"(c')
ll'J')' = (-
2
2) /"(11) + (3) 2g"(!')
Thus,
6wxx + Wxy - w,, = 6(/" + g") + ( - 2/" + 3g") - (4J" + 9g") 0
Exercises 1.3
In Problems I through 4, find d11/dt by use of the chain rule and check your result by
substitution and direct differentiation.
1. 11=xlny,wherex=1 2 ,y = (l - t) 11 2 .
2. 11 = x cosy + ye", where x = In 1, y = sin I.
3. 11 = arctan y/x, where x = cos t, y = sin 1.
4. 11 = In(x 2 + y 2 ), where x = e' - e-', y = e' + e-'.
5. If 11 = xyz, where y = In x and z = (1 +
x 2 ), find du/dx by the chain rule and
check your result by substitution and direct differentiation.
6. As in Problem 5, if 11 = x 2 + y2 + z 2 , where y = x cos x and z = x sin x, find
d11/dx.
In Problems 7 through 10, find the first-order partial derivatives of the given function
with respect to r and s.
7. u = x2 - 3xy + y 2 , where x = r + s, y = r - s.
8. u = excos}', where x = (r 2 + s 2 ) 112 , y = arctan r/s.
9. 11 = x 2 - y 2 , where x = r cosh s, y = r sinh s.
10. 11 = x ln(x - y), where x = r2 + s 2 , y = 2rs.
11. If 11 is a function of x and y, where x = s + t and y = s - t, express uxx - u,J'
in terms of the new variables s and t.
12. Express uxx + 11n in polar coordinates r and e, where r 2 = (x 2 + y 2 ) and
e = arctan y/x.
13. Let u = f(x + ct) + g(x - ct), where/ and g are twice differentiable functions.
Show that 11,, - c 2 llxx = 0.
14. Let II be a function of x and y, where x = r cos e and y = r sin e. Show that
(11J2 + (uy)2 = (11,)2 + _I, (119)2

,.-
15. If u and v are functions of x and y satisfying the equations ux = u,, 11, = - vx,
show that in polar coordinates
cu I cu I (;II OU
or r ae, r ce er
and
I
u,, + - u, + -;;- l/99 = 0
r r
Sec. 4 Dijfere11tiatio11 of Integrals Depending 011 a Parameter 15
4. Differentiation of Integrals Depending on a Parameter
In our later work we shall have occasion to deal with finding the deriYative of
a function cf> defined by a definite integral of the form
(4.1) cf>(x) = j'itxl f(x, t) dt

ll(X)
where the integrand! and the limits of integration u, v depend on the parameter
x. If the integration in (4.1) can be effected analytically, then the function cf> is
defined explicitly, and hence its derivative, if it exists, can be obtained by
straightforward differentiation. However, it is not always practical, nor possible,
to express the integral (4.1) in explicit form. Therefore, in such cases, it is
desirable to have an expression for the derivative cf>' of the function (4.1).
We first state the result in the special case where the limits of integration are
constants.
THEOREl\l 4.1. Consider the integral
(4.2) cf>(x) = f /(x, I) dt
here a and b are constants. /ff and f, are co11ti1111011s in the rectangular region
11
R: a s x s /3, a s t s b, then the fun ct ion (4.2) is continuous and possesses a
derirntice gicen by
(4.3) c/>'(x) = f JJx, t) dt
The fact that cf> is continuous on [:z, /3] implies that for each x 0 in [:z. /3]
Jim cf>(x) = c/>(x

x-xo
0) = Jb f(x 0, t) dt
a
that is.
~~~o f f(x, 1) dt = f ;~~a f(x, t) dt
We notice that this involves interchanging the order of the limit operation
x ....., x 0 and integration. Also, formula (4.3) indicates that the derivative of cf>
can be obtained by taking the derivative under the integral sign: that is,
(4.4) d- Jb f .(x, t) dt = Jb fJx, t) dt

dx a a
This also involves interchanging the order of the operations of differentiation

and integration. The validity of interchanging the order of two operations,
particularly those involving limiting processes, should not be taken lightly.
Such problems are of fundamental importance in mathematics and are of
frequent occurrence in practice.
We now state the result for the integral (4.1) involving variable limits .
.....
16 /11trod11ction Chap. I
THEORE\f 4.2. Let

'l(x)
(4.5) </J(x) = f(x. 1) dt
l
._,f/(XJ
1rhcrc 11 and r arc functions of' x. If" u and 1. are differentiable in the interrnl
a .:c:; x .:c:; b, and I and f~ are cont in11011s in the rectangle R: a .:c:; x .:c:; b,
11(x) .:c:; t .:c:; 1(x). then
def> I
1 ('rxl
dx dx

I.u(x)
f(x. 1) d1
t(x)
(4.6)
J 1(x)
. f~(x, I) dt
. dv du
+ .f(x, i) - - .f(x, 11)
dx dx
Formula (4.6) is commonly known as Leibnitz" rule. It is interesting to note

that this formula can be established by regarding the function cf> as a composite
function. We observe that the integral (4.5) depends on the limits u and v as
well as on x. Hence, we may write
1(x)
(4.7) cf>(x. u, t") =

J l(X)
f(x, t) dt
Since 11 and rare differentiable functions of x, and/ and/, are continuous, cf> is a
continuous and differentiable function of x alone. Then its derivative with
respect to x exists, and by formula (3.6) we obtain
de/> 8 ccf> du ccf> du

(4.8) = + + .
dx ex cu dx CV dx
Now the term i'rjJ/i'x in (4.8) is to be calculated with u and c treated as constants;
therefore, according to Theorem 4.1,
<'r/J r(x) .
(4.9) , . =
(..\
!,
u(x)
fx(x, t) dt
By the fundamental theorem of calculus (Theorem 1.2) applied to (4. 7), we have
cr/J
, = -f(x, cr/J
(4.10) 11) and " = f(x, i:)
Cli ci
Thus, substituting the results (4.9) and (4.10) in (4.8). we obtain formula (4.6).
Example 4.1. Let

cjJ(x) = {"'sin xi dt
Jo
By direct computation, we find
- cos x-'
rjJ(x) =
x
Sec. 4 Dijfere11riario11 of lutegra!s Depending 011 a Parameter 17
and therefore
,
cos :c
q/(x) 2 sin x 2
By Leibnitz' rule we have
(p'(x) ('I cos xi di + sin x 2

Jo
= rr_sin xi+ cos xt r + sin x2
I x x
2
Jo
11 hich gives the previous result.
Example 4.2. Let

u(x) = I
2
r'
.lo
(x - t) 2j(r) d1
11 here f is a continuous function. Show that 11'"(x) = f(x).
So/111io11: By (4.6) we find
u'(x) = C
0
(x - l)f{t) dr + ~(x -
2
1) f(l)],=x
r(x - l)f(t)dt
Differentiating once more, we obtain
1/'(x) = L' f(I) di
11hich, upon differentiating the third time. gives u~(x) = f(x).
More generally, when the integrand fin (4. l) involves the parameters x, y,
and the limits 11, rare functions of x. y. then the integral (4.1) defines </J as a
function of x, y; that is,
(4.11) <f>(x, _\') = J'(x.ri ('(x, r, t) dt

<(x.y)
If 11 and r have first-order partial deri\atives andf.fx, andfy are continuous, then
(pis a function of x and J' whose partial derivatives x and ),exist. Regarding
as a composite function c/J = c/J(X, y, 11, r). we obtain by means of formulas
(3.10), (3.11), and Theorem 1.2
t-(X,)')
(4.12)
ex J l(X,)')
j~(x, y, l) dt
+f(x,y,v)
ov -f(x,y,u _
, ) cu
ox ex
18 /11trod11ction Cltap /.
and
1(x,y)
c</J fy(x, y, t) dt
(4.13)
Dy J u(x,y)
+ f(x, y, v) ~
av '(
- ./ x, y, u)
au
~
oy oy
These are extensions of Leibnitz' formula (4.6).
Example 4.3. Let

11(x, )') = r+r f(t) dt
Jx-}
where f is a differentiable function. Show that 11 satisfies the equation
Solwion: By (4.12) and (4.13) \\e have
lix(x, )') = f(x + y) - f(x - y)

and
11,.(x, y) = f (x + y) + f (x - y)
Then. by the chain rule.

llu = f'(x + y) - f'(x - y)
11r,. = f'(x + y) - f'(x - y)
Thus Un - "-n = 0.
Exercises 1.4
I. Let <jJ(.Y) = .1 0 2 cos xi dt. Find dqJ/dx by formula (4.6) and check your result by
direct calculation.
2. Let cp(x) = J.~, arctan(t;x 2 ) dt. Find dcp/dx in two ways.
r.~" x
dt
3. Let </J(xJ = Find def!/ dx in two ways.
t +
4. Given that J: (x _ d~os 1J (x 2 ~ 1
J, x
2
> I. Evaluate
n dt
- .
o (x - cos t) 2
5. ,1,
Let ,,,(x) =
'
f t
x' sin xf dt. .
F111d d,1,/
'V dx.
"
6. Let </J(xl = l/k f~ f(I l sin k(x - t) dt, where/ is a continuous function and k a
constant. Show.that <V + k 2 </J = f
7. Let J 0 (x) = l ;r _( 0 cos<x cos 0) dO. Show that 10 satisfies the Bessel's equation of
zero order: y" + ( l /x)y' + y = 0. Hint: Find 16 and integrate by parts.
1
r' - I
8. Evaluate qJ(x) = -- dt. x > 0 .
0 In t
Hint: Find q/(.\) and note that qJ(x) -+ 0 as x -+ 0.
"
Sec. 5 L'nif'orm Co11veme11ce of Series 19
9. Let 11(x. I) 1/2c J~~:; f(s) ds, where/is differentiable.

(a) Find 11 1 and show that 11,(x, 0) = f(x).
(bl Show that 11 satisfies the differential equation 11,, - c 2 u"' 0.
5. Uniform Comergence of Series
Let u 11 ~ L be a sequence of functions defined on the interval a

11
~ x ~ b.
For each point x of this intenal, let
"'
S111 (x) L
11=
ull(x)
1
be the partial sum consisting of the first m-terms of the series

er
(5.1) ' L
11= I
u,,(x)
If the limit
lim S,,,(x) = Jim

"'
L u11 (x) = S(x)
(5.2)
m-oc m--+Cf~ ri= I
exists, then we say that the series (5.1) converges to S(x) at x, a ~ x ~ b.

Recall that equation (5.2) means that for any given number e > 0, there is an
111teger N > 0 corresponding to the point x, a ~ x ~ b, such that
"'
(5.3) [S(x) - S,,,(x)[ . L
,n=m+ 1
whenever 111 > N. In general, the integer N depends on the point x under
imestigation as well as on the given number i:. However, if corresponding to a
gi\en r,, there exists an integer N that is independent of x such that the inequality
( 'i.3) holds for all x in the interval a ~ x ~ b, then we say that the series ( 5.1)
converges uniformly to S(x) on that interval. It is clear that a uniformly
convergent series is convergent in the ordinary sense.
A convenient and practical method for determining uniform convergence of a
series is given by the following theorem.
TllLOREl\I 5.1. (Weierstrass !vi-Test) Let L,,';"= 1 u,,(x) be a series o/fu11ctio11s

defined on the interrnl a ~ x ~ b. fl L.;~= 1 M is a concergent series ol posit ice
11
consrants such that [11,,(x)[ ~ M,, for each 11 ~I and for all x in the interrnl
a ~ x ~ b, then the series L.';"= 1 u,,(x) conrerges 11niforml_1' 011 that interrnl.
Example 5.1. Show that the series I.,~ 0 x" converges uniformly to the function
1,1(1 - x) on the interval - a :o::; x :o::; a, where 0 < a < I.
So/111io11: For each x in the given interval we note that

20 /11trod11ctin11 Chap. I
Since the ,;cries 2,:,;. 0 a" is convergent \I hen .a < I. it follm1s by the \Veierotrass
\1-tcst that the -,eries2=,:~o x" converges uniformly for Ix! s a < I. To shm1 that
the series co/l\.crgcs to the function S(x) 1/(1 - x). we note that for 111 :> I.
\IC have
IJ/- I
,, I x111
S,,,(x) ~ .\ -
n::;:O I - x
0. it follm1s that Jim,,,~, S,,,(x) = l/(l -- x).
There arc se\eral important properties of a uniformly conYergent series which

\1e shall ha\e occasion to use later in our work. We shall state them here as
theorems to facilitate reference.
THEORE.\l 5.2. (Continuity of Limit Function) Let the fimctions 11,,(x)

(11 2': I) he cont/1111011.1 011 the interrnl a :::::; x :::::; h and let the series 2=;;'= 1 11,,(x)
com !'rgl' 11111/orm/y 10 S(x) 011 that in term/. T/1!'11 Sis a co11ti1111011sfi111ctio11 011 the
same inteffal.
This theorem implies that for each x 0 , a ::;; x 0 ::;; h,
lim S(x) = S(x 0 )

x- x 0
which means that

lim Jim S,,,(x) = lim lim S,,,(x)
Thus. under the conditions stated in the theorem. the order of taking the limits
with respect to x and 111 can be interchanged.
Example 5.2. For the series of Example 5. I. it follows that
Ii 111 I =
.., (I)"
~
X-+ A -- x 11~0 ,2
or
2 =I+ 1 + l + ~ +
THEOIU:\I 5.3. (Integration of Series) /,rt t!w fi111c1io11s 11,,(x) (11 2': I) he co11-
ti1111011s 011 the intenal a ::;; x :::::; band let the .1cries L:= 1 11,,(x) conrerge 11111form/y
to S(x) on that interrnl. fl x 1 and x 2 are anr fll'O poi11ts such that a :::::; x 1 <
x 2 ::;; h. then
I' S(x) dx = ,t .C' 11,,(x) dx
This theorem gi\es suftlcienl conditions under which a convergent series may
be integrated term by term to obtain the integral of the limit function. It
essentially invohes interchanging the order of the integration and summation.
Sec. 5 L'11ifon11 Convergence of Series 21
Example 5.3. Obtain the result

1 x11-:-1
- In( I -- x) ~
~ (xi < I J
n =~ 0 II
from the series or E-.;amplc 5.1.
Solurion: Since the series I_,;=o .1 11 conYcrges uniformly to the limit function
!/(! x), we can integrate term by term from 0 to any point x, 1\here < I,
and obtain
f 1 di ~
11=0
(' t" dr
.J 0
er_ xn+l
~ (ixl < I)
n=O II + I
The integral on the left-hand side gives -Jn(! - x).
THEORE:\I 5.4. (Differentiation of Series) !/the f1111ctio11s u,, (n 2 I) of the

co11n'rge11t series
S(x) = L u (x)
11:::: I
11
hare continuous dcrirntires 11,; 011 the interrnl a :S x :S b, and if the seril'.\
I,,;, 1 u:i(x) conrcrges un(j(Jrnzly on this interra/, then
'f.
C" '( ) =
,) ,\ d ( ~,
"--' !/" ( x )')' I u;,cx>
cfx 11= I 11= I
j(Jr a :::;; x :S b.
The theorem gi\es suft1cient conditions under which a comergent series may
he differentiated term by term to yield the derivative of the limit function. The
proce:-.s im olves interchanging the order of differentiation and summation.
l:xample 5.4. From the series of F-.;amplc 5. I. show that
= I + 2x + 3x 1 + ~ nx 11 - 1
(I X)2 n- I
!'or :.1: < I.
So/111/011: Consider the series "'[.,;= 1 11x"- 1 obtained by formally differentiati11: tbe
series of Example 5. 1. For each x such that lxl ~ a < 1 \\e note that
L'iing the ratio test, \\e sec that the series "'[_,~= 1 110 11 - 1 is convergent \\hen
a' < I. Thus, by Weierstrass M-test, series"'[_,~ 1 nx"- 1 converges uniformly for
.1, :: a < I. By Theorem 5.4, 1t follows that
2
(ix; < 1)
(! - x)
22 /11troduction Chap. I
Exercises 1.5
In Pro blerns I through 6, show that the series converge uniformly on the given
interval.
Y: z
sin /lX cos /IX
1. (aJ I:1
II= 112
forallx; (b) I:
II=} 112
for all x.
oc x'1
2. I:
11=1 11!'
'xi ~ a < oc
x
(-1)1'
3. I:
n= I :c
?
T'
?
11- '
for all x.
4. for all x ~ 0.
5. I; e- "", for 0 < a ~ x.

11= 1
6. ~ ( x" ) I for - I < a ~ x < :x.

11= I \1 + X 11 11 2
7. From the series of Example 5. J, obtain the series
In I + .r = 2 ( x + x3 + xs
+ .. ) ( x' < I)
I x 3 5
8. Let I; ( - 1)"x 2 " = ,x < 1.

n=O + x2 '
Show th;it the series converges uniformly and obtain the result
arctan x = x - + + ... (x < I)

3 5 7
9. If S(x) ~
~
sin ?nx ,
I
S lOW
that
ri= I ,,-
Irr
Jo S(x) dx
10. If S(x) I;
oc
11= I
COS /IX
11
2
, find a series for l" O
2
S(x) dx.
oc:
11. If ~ (-l)"x"for:x ~a< I, show that
l+x 11=0
~ ( - I J'' - 1 nx" - 1
(I + x)2 11= 1
Justify your reasoning.
12. If S(x) = ~ sin nx, verify that S'(x) i: cos nx

11:::: 1 n5 ' 2 n=l 113/2
Sec. 6 Improper Integrals Depending 011 a Parameter
1.3. If S(.\) ~ x", :;'1011 that S'(.\) = S(x) and /"' S(I) dr = S( 1) - l.
11=0 11 ! Jo
'l
14. Let 11(x, t) = ~ e -""' ----,

t.. sm nx , f or 0 s x s n. 0 < 10 s I.
11= I tr
(a) Show that 11 1 , u,.. 11xx can be calculated by differentiating the series term by
term.
(b) Thus verify that u satisfies the equation 11 1 - un = 0 and the boundary
conditions u(O, I) = u(n, I J = 0, I 2" 10 > 0.
6. Improper Integrals Depending on a Parameter
A discussion parallel to that of the previous section can be given for an

improper integral of the form
(6.1) f"' f (x. t) dt

11hcre the integrandfinvol\es the parameter x. We assume that/(x, r) i:-: con-
t111uous in lhe region D: a ~ x ~ b, c ~ I < oo. Then, for each x in the
i1itcnal a ~ x ~ hand for each d, c ~ d < x. the integral
(6.2) rf(x, 1) dt
certainly exists and thus defines a function of x. If the limit
(6.3) Jim (d f(x, I) di f,f f(x, I) dt = F(x)

d-+y:, Jc . {
c\i'.h for each x in the intcnal a ~ x ~ b. then we say that the improper
integral (6.1) coll\crges to F(x) on thal interval. This means that. for any g11en
r, 0. there is a number T corresponding lo each x such that
' d 1
(6...t) , F(x) - j f(x, t) dt [" /(x. t) c/1 < 1:

' ..,. c d
1\hc11c1er d > T Presumably. the number T depends on x as well as on the

gi1 en 1:. Ir a number T exists which is independent of x such that i neq uali ly (6.4)
holds for all x in the interval a ~ x ~ b, then the integral (6.1) is said to con1crgc
uniformly to Fon that interval.
We notice the similarity between the definition of uniform convergence of an
improper integral and that of uniform convergence of a series. As a consequence
of this similarity, we expect to obtain results for a uniformly com ergent improper
integral which parallel those obtained for a uniformly convergent series. \Ve
first 'late the analogue of Weierstrass M-test for integrals.
24 /11trod11ction Chap. I
THEORE\t 6.1. (\\eier,tra:,,; M-test of Integrals) Vt g(t) ;:::: 0 he a j1111crio11

dcfi11edfor c :S: t < r_r_, 111ch that the integral
.C g(t) dt
exists. lf lf(x. Ill :S: g(t)fiJr all x in the interrn/ a :S: .\: :S: h, rhen the integral
f"' f(x,
c
t) dt
com ages 1111ijiJrm!.r (a11d absolutely) 011 the in term! a :S: x :S: h.
Example 6.1. Sho\1 that the integral

if .
sin xi dr
r
o I + 12
con\erges uniformly for all \alues of x.

So!111io11: For any value of x. 11e have
sin xr
2
s
I + 1 + 1
2
Since J~ [I/( I ~ r ='I J dr converges (to r:/21. iL follm1 s from Theorem 6.1 that the
gi1en imrroper integral con\erges uniformly for all values of x.
The following three theorems correspond 1n order to Theorem:-, 5.2. 5.J. and
5.4 of the preceding o,ection.
THLORL\I 6.2. Ler I he a co11ti1111011s fi111ctio11 in the rcgio11 D: a :<::; x :S: b.

c :S: t < oc. a11d !er
I;
c
f(x. I) dt
co111crge 1111i/rm11h to F 011 t/1c i11tcrrn/ a :S: x :S: h. Then Fis co111i1111011s 011 the
i11tcrrn/ [a. b].
The continuity of F permits us to interchange the order or tah.ing limit and

integration. That is. for each x 0 where r1 :S: x 0 :S: h. 11c ha1c
(6.5) li'.11
.\ - .\Q ~
j'"'
(
f(x. t) dt = f" 1~11
., ( .\ .\Q
j(x. I) dt
This conclusion generalizes the first part of the result of Theorem 4.1 11hich
concerns a definite integral i111olving a parameter.
TI-IEORE\1 6.3. Lei the sw11c co11ditions as in Thcorrn1 6.2 hold. Jf _\ 1 and x 2
are a11y rirn points such rhat a :S: x 1 :S: x 2 :S: b, then
(6.6) .C,' F(x) dx = f" f' f(x, I) dx dt

Sec. 6 Improper /111egrals Depending 011 a Parameter 25
That is. we may interchange the order of the two integrations. one with respect
to x and the other with respect lot. Let us illustrate the use of this properly in
naluating certain integrals that otherwise cannot be effected by the usual
elementary method of integration.
Example 6.2. Obtain the result
., ('_,,, - e-1>1 dr b
In (0 < (/ < b)
fo I (/
b) integrating the function defined by the improper integral g e-xt d1.
So/111ion: For 0 <a:::; x, we note that
(t 2: OJ
Since the integral
exists. it fol!O\\S from Theorem 6.1 that the integral gc e-xi dt is uniformly con-
\ergcnt for a ~ x ~ b. By direct integration \IC find
1= 1 ,., e-xt dr (x > 0)

.\" 0
Integrating this 1\ith respect to x from .1 = a to x = band using (6.6), we obtain
(' :. d.1 = ('l Ce-" dx dr = J.,_ e-at

Ja - Jo ._a o
The integral on the left gives ln(b/a).
lllFORE:\t 6.4. Let the .fimctions f and(, he co11tin11011s in the region D: a ::;:
Y :::; h. c ::;: t < oc, and let the integral
r: f(x, t) cit
co111erge to Fon the interral a ::;: x ::;: b. If the integral
f" j~(x, I) dt
c
co111erqcs 11111/ormly 011 a ::;: x ::;: b, then Fis differentiable and
(6.7) F'(x) = d
dx c
f"' .
f(x, t) dt = f"'"c j~(x, l) dt
This generalizes the last part of the result of Theorem 4.1.

26 lntrod11ctio11 Chap. I
Example 6.3. Show that for x :::: 0,
(6.8)
J oc
c
e
-xr sin t d
--- t = - arctan x
t
So/11tio11: \Ve denote the given integral by F(x) and shO\\ that the integral converges
+ 7I-
2
uniformly to F(x) for x :::: 0. Theorem 6.1 cannot be applied here, so 1ve shall
resort to the definition. Let c > 0 be given. We \\ant to show that there is a
number T independent of x such that
(6.9) - - -t dt
I Jz e -xr sin .i
< c
: d t I
for x :::::: 0 \\ henever d > T. When x > 0, we have, by integration by parts with
11= e-"/t and du = sin t dt.
"' sin t
df =
e-xd cos d
d
"'(l + xt)e-"''
cos t dt
Jd C-Xl .. -{ -
Jd I2
Since !O + xt)e-xt cos ti:::; 1ex' e-xr cos If :s; l, it follows that
[f"' e
I d
-xr --t d t I :S: I
sin
t I d
1.
+ l"' -
d
1 dt = 2
t2 d
But the foregoing discussion also holds when x = 0. Thus, taking T = 2/c, it
follows that (6.9) holds for x :::::: 0 whenever d > 2/c. Therefore, the given im-
proper integral converges uniformly to F(x) for x :::: 0. Now, by Theorem 6.1,
the integral
Lx e-xr sin t dt
converges uniformly in the interval 0 < c :S: x for arbitrary number c, taking
g(I) = e-cr_ Thus, when x > 0, we have by Theorem 6.4
F'(x) = - L" e-xr sin t dt
+ x2
and hence
F(x) = - arctan x + C
where C is a constant. To determine the constant C, we observe that
'.F(x)j :S: J" 0

e-xt dt = _l
x
(x > 0)
since [sin t/tl :S: l for all t. This implies limx-oo+ F(x) = O; therefore, C = n/2.
This establishes (6.8). By Theorem 6.2, F(x) is continuous for x :::: 0 and so,
letting x - 0, we obtain the important result
(6.10)
Joc
0
sin_t dt = 7I
t 2
Sec. 6 improper lnte.qrals Depe11di11y 011 u Parameter 27
Exercises 1.6
In Problems 1 through 6. shmv that the improper integrals converge uniformly on

tr.e given interval.
dt
,,
l. rJ 2
() < {/ s; x.
.o x2 + I
sin t
2. , dt, for 0 < {/ ::; x.
0
x- + 12
cos xt
3. , dt. for all x. a -r- 0. _J_
.C er + 1-
0
4.
c
0
e __,., cos t d1, for 0 < a ::; x .
5.
6.
rI" e-' sin xi d1,
e-'t"- J dt.
for all x.
for 0 < {/ ::,-:; x ::; h.

I
7. (a) Show that the integral 1~ e-r cos xt dt converges uniformly for all x to
F(x) = 1/(1 + x 2 ) and l.hus prove that
" _, sin xi {
e ct = arctan x (x 2'. OJ
0 t
( b) By changing the variables show that

"
"- -xr sin t d
e --- - I arctan (x > 0)
0 t x
(cl Lelling x--> +er.;. deduce from the result of (b) that
"
" .
Sill I I Ti
<I =
o I 2
J;? e-' sin xt dt = x/(I + x 2 ) for all x, show that

8. .Assuming that
" cos bt) I 1 I + b1
c.t = - In (a < b)
2 I + a1
In particular, deduce the result
9. Consider the integral

i 'l
0
e- 1
1 - cos bt d
t
-- I
I
- In(!
2
+ b1 )
(x > 0)
x
28 l11trod11ctio11 Chap. I
\'crify that differentiation \\ ithin Lhe integral is permissible an) number of times.
and thus sho11 that
10. Let flxl = .1<; e_,. cos :11 dt.

(al Shm1 that the integral converges uniformly for all x.
(bi Shm\ that differentiation "ithin the integral is permissible and show by
integration by parts that F'(x) = - (x(2)F( <I.
(c) Solve the llrst-order differential equation in (b) and use the fact _1~ e-'' dr =
\ ;r/2 to obtain the result
11. Let 11(x. y) = 1(i' e- 1' " cos xt dr. y > 0.

(al Show that the integral converges uniformly.
(b) Show that the derivatives 11,. and uu can be obtained by differentiating \\ithin
the integral sign. and verify that "r - 11' = 0.
(c) Show that
u(x, y) = 1 ('T'r ) ; 2 exp ( - ,2)

2 y 4y,
Hint: Make the change of variable z 2 t 2 y and use results of Problem 10.
7. Directional DeriYatiws
In this section we review a generalization of the concept of derivatives of

functions of several \ariables. Suppose that u is a continuous function of the
variables x, y with continuous first-order partial derivatives in some domain
containing a point (x 0 _r 0 ). Let L be a directed line through the point (x 0 y 0 ),
which makes an angle 0 with the positive x-axis, and consider the function u on
this line. If L has the parametric representation
(7.1) X = Xo + S COS 0. r = ro + s sin 0

wheres denotes the distance of the point (x, y) from (x 0 , r 0 ), then clearly on the
line L the function 11 becomes a function of s. By the chain rule and from (7. l)
we have
du cu dx ou dy
Dx ds
+ Dy ds
ds
(7.2)
cu cu .
= - .. cos 0 + sm 0
ox c!y
When the derivatives ?u/<?x and tu/<":y are evaluated at the point (x 0 , y 0 ), then
equation (7.2) is called the directional derivative of u at (x 0 , y 0 ) in the direction
Sec. 7 Directioual Derivatives 29
of the line L. Geometrically. equation (7.2) represents the rate of change of 11

with respect to sat (x 0 _1 0 ) in the direction of L.
From the formula (7.2) we obsen e that the directional derivative of a function
u at a point depends only on the\ alue of the angle of inclination 0, inasmuch as
the \alues of the deri\ atives of 11, at a given point, are fixed. As a matter of fact.
when 0 = 0 so that Lis the positive x-axis, we find du/ds = cu/i'x; and when
() = rr/2 so that L is the positive _1-axis, d11/ds = cu/cy. Thus. the partial
derivati\ es c-11/i'x and ('11/?y of 11 are really directional derivatives in the direction
of positi\C x- and _1-axes. respectively.
More generally. let C be a curve with parametric representation
(7.3) _\ = x(s), y = _1(.1)
11 here .1 denotes arc length of C. 0 ::::; s ::::; L, L being the length of the curve.
\\'e assume that x(s) and _r(s) ha\e continuous derivatiws on the interval
() ::::; s ::::; L. which do not vanish simultaneously. The cune C is said to be
smooth. Let 11 be a continuous function with continuous first-order partial
deri\ atives on the curve C. The derivative
du cu dx (;u dr
(7.4) +
ds ax ds i1_1 ds
e\ al uated on the curve C is called the tangential derivati\e of 11 on C The terms

c1x ds and c~rlds are the direction cosines of the tangent line to C so that (7.4)
is a directional deri\ali\e along the tangent of C (Fig. 1.2). Here we adopt the
comcntion that the tangent is pointed in the direction wheres is increasing.
2 2
Example 7.1. Let 11(x. y) = x +
(a) Find the directional derivative of 11 at (I. 2)
y .
in the direction 0 = 45'; (b) find the direction in 11hich the directional derivative
of 11 at (I. 2) is maximum.
So/111io11. (a) By (7.2) \IC ha1e
du
= 2_1-:ct.zJ cos 45'' I 2y:c1.l> sin 45
ds
(bi The derivative of 11 at (I, 2) in any direction() is given by
du .
= 2 COS () + 4 Sill ()
ds
\lhich is a function of()_ To find the value of 0 for which d11/dl' is maximum (or
minimum), we differentiate d11/ds with respect to 0 and set the result to zero. We
ha1e
2( - sin {)) + 4 cos 0 = 0 or tan 0 = 2
If a0 is such that 0 s 00 <:- :rr/2 and tan 00 = 2, then() = 00 and 0 = 00 + :rr are
the two values of() less than 2n, which satisfy the above equation. Clearly, the
30 lntrotl11ctio11 Ch"P I
direction fl = fl 0 yields rna:-.imum value for cfuids, which is 2( 1/\ 5) + 4(2/ \ 5) =
2\ 5. and the direction () = () 0 + T:: gives minimum value for du/ds, \\ hich is
--2\ 5.
Of great importance in practice is the directional deri\atiw of 11 on the curve

C, which is in the direction of the normal to the cune. This deri\ati\e is called
normal deri\ati\e and is usually denoted by i'uJn. Referring to Figure 1.2 and
and using the definition (7.2), it is easy to obtain an expression for the normal
deri\ali\e of u on the cune (7.3).
d1'/
\':::: y l ')
\':::: l ( 1)
01
I
FIG. l-2 Sorma/ derirntire.
We note that
dx 0 - /[
cos 0, ,in 0,
ds 2
so that
dr dx
cos = sin 0 sin </> -cos 0
ds ds
Hence. according to (7.2),
I J 5)
cu cu .
- . cos q)
au . ,/,'I'
')J11
en OX oy
C!I dy au dx
ax ds C.1 ds
111 the case where C is a closed curve (not crossing itself), the derivative (7.5)
is called tl1e outward normal derivative, since the normal direction in such a case
Sec. 7 Directional Derivatives 31
is then pointing away from the domain enclosed by the curve. Of course. here
we observe the convention that as the parameters increases, the curve is traced
in the counterclockwise manner.
Example 7.2. Find the outward normal derivative of u(x, y) xy on the circle
x2 + y2 = 0 2_
Solution: An outward normal direction on the circle has direction cosines given by
x
cos and sin = Y
a a
Thus, according to (7.5),
cu
CII y (x) + (v)
a x ~
2xy
a
where x and y satisfy the equation x 2 + y 2 = a 2 . If we introduce polar co-

ordinates x = r cos , y = r sin cjJ, then cu/en = 2a sin <p cos <f> = a sin 2</J.
In fact, in polar coordinates we have 11 = r 2 sin cos. Since the outward
normal direction coincides with the radial direction, it follows that
(; /l (;/I .
, = , = 2r sm </J cos </> = r sin 2</J
c11 er
which gives the previous result when r = a.
Exercises I. 7
I. Compute the directional derivative of the following functions in the direction

indicated:
.. (a) x
(b) x 2
2
+
+
2xy + y 2 , at (1, 2), 0 = 45 .
yex, at (I, -2), 0 = 30'.
2. Compute the directional derivative of the function 11 x 1 y + y sin x at (2, 4)
in the direction of the tangent to the curve y = x 2 .
3. Find the directional derivative of /1 = x 3 + xy 2 - y 3 at ( - I, - 3) in the direc-
tion of the tangent to the curve x = I + t, y = 1 - t 2
4. Determine the direction and the maximum value of the directional derivative of
11 =xy 2 + x 2 y at (I, I).
5. Do as in Problem 4 for the function /1 = 9x 1 + 4y 2 at the point (I, I).
6. Compute the normal derivative of /1 = x2 + y 2 at (2, 0) on the curve 9x 2 +
4y 2 = 36.
7. (a) Find the normal derivative of /1 = ln(x 2 + y 2 ) on the circle x 2 + y2 a2 .
(b) Using polar coordinates, show that d11/dn = du/dr.
32 lntrod11ctio11 Chap. 1
8. Green's Theorem and Related Formulas
We conclude this re\ iew chapter \\ith a discussion of one of the important
and useful formulas for performing double integration in the plane. knmrn as
Green~ theorem. This theorem and its related formulas will be needed in
Chapters 3. 6. and 7.
Tl-IEORE\1 8.1. (Green's Theorem) Let C be a pieceirise smooth simple closed

cwTe---a piece11 ise smootlz closed cwTe irhich doe's not cross itself-in the .\T
plane hounding a domain D (Fig. 1.3). Let P and Q be tlloji111ctions ofx and_!".
irhich a,-e conti1111011s togC't/l('r irirh thC'ir/ir11-ord('J' 1wrtial derirntires in D. Then
(8.1) .
fc (P dx + Q dr) =
J' r ((~o
D. D.; - (r
ap) dx dy
irhac !he i111egratio11 011 the lcfi is taken along the curcC' C i11 the co1111terc/ock1rise
dirC'ctio11.
- - - , - 0 + - - - - - - - - - - - - - - - - - - - - - - - .\
FIG. 1-3 Gr!.'c11's lhC'orem.
Let us e\plain the meaning ol' the integral on the left of equation (8.1 ).
Suppose that the cune C is gi\en parametrically by the equations
(8.2) x = .\(/ ). _)' = y(I)

where x(I) and rU) ha\e continuous derivatives on the intenal t 1 s s t2 ,
which do not \anish simultaneously. Note that x(t 1 ) = x(f 2) and .i (t 1) = _1(1 2),
since the curve is closed. Further. obsene that as t increases from t 1 to t 2 , the
cune is traced in the counterclockwise manner. Then, by the integral
(8.3) Sc (P dx + Q dy)
Sec. 8 Green's Theorem and Related Formulas 33
1ve mean the ordinary integral

12
I P[x(t), y(t)] dx -
1 1 \ dt
Q[x(I), y(t)]
dyl
. dt
dt ,1
of the single variable 1 obtained when the functions (8.2) are substituted for x
and yin (8.3). The integral (8.3) is called a line integral on the curve C
Example 8.1. Compute the line integral Jc [x 2 y dx + (x 2 - y 2 ) dy] from (0, 0) to

(l. 2) along the curve C: y = 2x 2 .
Solution: We convert the line integral into an ordinary integral with x as the variable
of integration. We note that on C y = 2x 2 so that dy = 4x dx. Substituting
these for y and dy in the integral, we find
L 2
[x y dx + (x
2
-
2
y l dy J = L 1
[2x
4
+ (x
2
-
4
4x )4x J dx
L 1
(2x
4
+ 4x
3
- l 6x 5 ) dx
19
15
Example 8.2. Verify Green's theorem for the functions P(x, y) 2xy 2 and Q(x, y)
2x 2 y, where C: x 2 + y 2 = 4.
Sol111io11: Let us introduce polar coordinates, x = r cos t, y = r sin t, 0 -s r :<; 2,
0 r -S 2n. Then
P = 2xy 2 = 2r 3 cos r sin 2 r, Q 2x 2 y = 2r 3 cos 2 t sin t
Pr = 4xy = 4r 2 sin t cos t, Qx 4xy = 4r 2 sin t cos t
Hence, in the domain D bounded by the circle x 2 + y 2 = 4.
JJ<Qx - P,.) dx dy = 0
1J
On the other hand, evaluating the line integral on the circler = 2, we find
L (2xy 2
dx + 2x 2 y dy) = L17r [16(cos t sin 2
t)(-2 sin t)
+ l 6(cos 2 t sin t (2 cos t)] dt

= 0
Now suppose that 11 and v are functions of x and y with continuous partial
derivatives up to the second order in the domain D. If we replace the functions
P and Q in equation (8.1) by the functions - uuY and uvx, respectively, we then
obtain
(8.4) Sc u(1'x dy - VY dx) = JJ [u(vxx + uyy) + uxvx + uyvy] dx dy

D
34 Introduction Chap. 1
Let the equations (8.2) describing the curve C be given with the arc length s as
the parameter; that is, x = x(s), y = y(s), 0 ::;; s ::;; L, L being the length of
the curve. Then the line integral in equation (8.4) can be expressed as
(8.5)
Jc
r u(vx dy - Vy dx) = r ll ('vx ~~
Jc . ds
- Vy dx) ds
ds
CV
=
J c
u -ds
on
where ct:/011 is the outward normal derivative of z: on C as defined in equation
(7.5). Thus, equation (8.4) can be written in the form
Jcr ~ + +
(8.6) U ds = JJ (u LlV llxVx !lyl\.) dx dy
on
D
Here we have introduced the symbol i1 which stands for the Laplace operator
(8/ex) 2 + (o/8y) 2 . The formula (8.6) is often called Green's first identity.
If we interchange the role of u and v in equation (8.6) and subtract the result
from (8 6), we further obtain the formula
(8.7) Jc ( u ~~ - v ~~) ds = J J (u Llv - v i'.lu) dx dy

D
This formula is often known as Green's second identity.
Exercises 1.8
Using Green's theorem, evaluate the following integrals and check your answers by
evaluating the line integrals directly.
1. Jc (x 2 dx + y 2 dy) along the square with vertices (0, 0), (l, 0), (I, 1), (0, I).
2
2. Jc [-x y dx + (y 2 - 1) dy] along the triangle with vertices (0, 0), (I, 0), (I, I).
3. Jc [(x + y) dx + (y + x) dy] along the unit circle x 2 + y 2 = I.
2 2
4. Jc [(x + I) dx + xy 2 dy] along the closed curve formed by y = x and y = x2

from (0, 0) to (I, 1).
5. Evaluate directly the integral Jc (-y dx + x dy)/(x 2 + y 2 ) along:
(a) The circle x + y = I (use polar coordinates).
2 2
(b) The square with sides x = 1, y = 1.

(c) Why is Green's theorem not applicable here?
6. (a) Verify formula (8.6) for u = 1, v = x 2 + y 2 , and D the domain bounded by
the circle x 2 + y 2 = 4.
(b) Deduce the formula
Lun ds = JJ6v dx dy
D
from (8.6) or (8. 7).
Sec. 8 Green's Theorem and Related Formulas 35
7. (a) Suppose that ~v = 0 in D. Deduce the formula
Li U~) ds = JJ
D
[(vx)
2
+ (i:y}2] dx dy
from (8.6).
(b) Suppose further that v = 0 on C; show that v = 0 identically in D.
(c) If 21'/211 = 0 on C, instead of v = 0, show that v = constant in D.
Chapter 2
Linear Partial
Differential Equations
We now begin our study of partial differential equations. In this chapter we

first introduce certain basic concepts and definitions and then proceed to a brief
study of linear first-order equations. The rest of the chapter will be devoted to
the consideration of linear second-order differential equations with constant
coefficients. In particular, the classification of these equations into three
principal types and the corresponding canonical forms will be discysjed. Then,
in the next and the last two chapters, specific problems that aretd[a\,;n from
mathematical physics will be treated for each of the three types of second-order
equations.
1. Basic Concepts and Definitions
A partial differential equation is an equation that involves an unknown

function of two or more independent variables and certain partial derivatives of
the unknown function. More precisely, let u denote a function of the n inde-
pendent variables x 1 , , xn, n ~ 2. Then a relation of the form
F(x1, ... ,xn,u,ux1' .,ux",ux1x1,ux1x2') = 0,

where Fis a function of its arguments, is a partial differential equation in u.
The following equations are some examples of partial differential equations in
two independent variables x and y.
(a) xux + yuy - 2u = 0
(b) yux - Xlly = x
(1.1) (c) uxx - lly - u = 0
(d) llllx+ )!Uy - U = xy 2
(e) llxx + xu; + yu = y
36
Sec. 1 Basic Concepts and Definitions 37
As in ordinary differential equations, the highest-order derivative appearing

in a partial differential equation is called the order of the equation. Thus, in
(1.1) equations (a), (b), and (d) are all first-order partial differential equations
and the remaining two equations are both second order.
A partial differential equation is usually considered in a certain domain of
the independent variables. If there exists a function u in the domain under
consideration, such that u and its derivatives identically satisfy the differential
equation, then u is called a solution of the equation. It follows from this
definition that a solution of a partial differential equation must have partial
derivatives of orders equal to those appearing in the differential equation. fn
practice, however, this condition is often too restrictive. Later, we shall intro-
duce the notion of a generalized solution that will admit other kinds of functions
as solutions.
Example I.I. Verify that the function u(x, y) = x 2 + y2 is a solution of equation

(a) in (1.1 ).
So/11tio11: The given function can be differentiated and we find
llx = 2x, //}' = 2y

and so
xu, + y11,. - 211 = 2x 2 + 2y 2 - 2(x 2 + y 2) = O
for all x and y.
Example 1.2. Verify that the function u(x. y) e- 2 '" cos x is a solution of equation
(c) in (I. I).
Solution: Here we have

u, = -e- 2 " sin x, - e-ly COS X, 11,. - 2e- 2 " cos x
Thus,
for all x and y.
A partial differential equation in the function u is said to be linear if it is at

most of first degree in u and the derivatives of u. This means that the equation
should not contain any term that involves powers or products of u and deriva-
tives of 11. Thus, in (I.I) equations (a). (b), and (c) are all linear differential
equations. On the other hand, both equations (d) and (e) are not linear because
the former involves product of u and u" whereas the latter involves second
power of uY. Another way of defining linearity is given in Section 3 of the
pcesent chapter. A partial differential equation that is not linear is called a
nonlinear differential equation.
Throughout this book we shall be concerned only with linear partial
differential equations of the first and second order involving two independent
'
38 Linear Partial Differential Equations Chap. 2
variables. In particular. we shall limit our discussion of second-order equations

to tbe three prominent classical equations of mathematical physics, namely,
_,
(i)
o2 u o-u
= f(x, t)
c1 2 ox 2
_,
OU o-u
(1.2) (ii) g(x, 1)
eel cx 2
_,
c-u ou 2
(iii) 2 + ay2
= h(x, y)
cx
and their related forms. These equations are the basic partial differential
equations of wave propagation, heat conduction, and potential theory. Not
only are these equations of fundamental importance in many branches of
physics, but they also serve as prototypes for the three principal types of second-
ordcr partial differential equations. The kinds of problems that can be solved
for each of these equations and the properties of the corresponding solutions are
generally typical of what can be expected in the general case where more than
two independent variables are involved.
Exercises 2.1
1. Determine the order of each of the follov.ing equations and find which are linear.
(a) Uu + Xl/J' = 0. (b) lllly + llxx - llx = 0.
(c) 11.u - llyy + (ux) 2 = x 2 . (d) yu, + .my + II = 0.
(e) xu"' + yun' + 211
2
= I. (f) u"' + x11,.Y - /1 = xy.
2. Show that the given function satisfies the accompanying equation:
(a) 11(.\, y) ax + by; .rnx + y11,. = 11.
(bl u(x, y) ax 2 + by 2 ; xux + y11,. = 211.
(c) u(x. )') (x - a) 2 + (y - b) 2-; (11) 2 + (11)') 2 = 411.
2
(d) u(x. )') x2 + y ; Yllx - 11,.x = 0.
3. Show that the given function satisfies the accompanying partial differential
equation:
(a) 11(x, y) ex + sin y + xy; llxi = I.
(b) u(x. y) cos(3x + 2y); 6u,-x - 13uxr + 611,.r 0.
(c) 11(x, y) = x 2 + exy sinh x; 11,._,. - xu,. = 0.
(d) 11(x. y) = (cosh y) In x; 1111_". = (11J(u.. ).
4. Verify that the follov.ing functions are solutions of the \\ave equation 11 11 -
c2 u0 = 0 for some value of c.
(a) 11(x, f) = x 2 + t 2 . (b) u(x, t) = cos x sin 2t.
(c) 11(x, t) = ln(x + t) + (x - t) 2 (d) u(x. t) = sin 2 (x + bt) + ex-bt.
5. Verify that the following functions are solutions of the Laplace equation""' +
//}')' = 0.
(a) u(x, y) = ex cosy. (b) u(x, y) 3x 2y - y 3.
(cJ 11(x, y) = sin x cosh y. (d) 11(x, y) ln(x 2 + y 2).
Sec. 2 General Solutions and Auxiliary Conditions 39
6. Verify that the following functions are solutions of the heat equation 11, - c2 uxx
0 for some value of c.
(a) 11(x, t) = e- 41 cos x. (b) 11(x, t) = e-a 21 sin bx.
11
(c) u(x, t) = e4 s cosh sx. (d) u(x, t) = e' 21 sinhsx.
2. General Solutions and Auxiliary Conditions
A solution of a partial differential equation that includes every other solution

of the equation is called the general solution. We recall that in the theory of
ordinary differential equations, the general solution of an nth order differential
equation involves n independent arbitrary constants. These constants are
determined when the solution is required to satisfy certain conditions. For
instance, the first-order differential equation
du
(2.1) - + u = f(x)
dx
has the general solution
(2.2) u(x) = J: e-<x-t)f(t) dt + ce-x
involving an arbitrary constant C. If u is required to satisfy the initial condition

u(O) = 1, then we find C = 1.
In the case of a partial differential equation, the general solution is found to
involve arbitrary functions rather than arbitrary constants. The number of
these arbitrary functions is generally equal to the order of the differential
equation. Moreover, if' the differential equation involves m independent
variables, the arbitrary functions will depend on (m - I) variables. Thus, ifthe
function u in equation (2.1) also depends on the variable y, then the constant C
could be chosen as an arbitrary function of y. In other words, if u is a function
of the independent variables x, y, then the first-order partial differential equation
(2.3)
cu
~ + u = f(x)
ox
has for its general solution
(2.4) u(x, y) = J: e-(x-'lj(t) dt + g(y)e-x
where g is an arbitrary function of y. It is easy to see that for any choice of the
function g, u satisfies equation (2.3).
Example 2.1. Find the general solution of the second-order equation
Uxy = sin x + y
Sol111io11: The general solution can be obtained by partial integration. That is, we
integrate both sides of the equation \\ ith respect toy, treating x as a constant, to
obtain
yl
u,Jx, y) = y sin x + + h(x)
2
h being an arbitrary function of x. Next, integrating with respect to x, treating y

as a constant, we obtain
u(x. y) = - y cos x + xy
2
2
+ . r h(x) dx + g(y)
If \\e write
f(x) = J h(x) dx
then the general solution is !::iven by

X)'2
u(x, y\ = - y cos x + + f(x) + g(y)
2
involving two arbitrary functions f and p.
Example 2.2. Sho\\ that 11(x. Jl ~c f(x - 2r) + g(x + J), where f and g are twice
differentiable functions. is the general solution of the differential equation
Solution: We first verify that the !!iven function satisfies the differential equation.
Differentiating by the chain rule, \\ e find
llx = r + q', lly ~~ - 2f' + g', ll.u = (" + g"

Urr = 4_(" + g"
and so
To show that /1 is ti:e c-eneral solution. let us introduce the new Yariables
s = .r - 2.r, l=x+y
and write 11(s, r I iiLY. y). Then
11,.,. = 411".,_,
Substitution of these in the equation yields
9ws 1 = 0
Sec. 2 General Solutions and Auxiliary Conditious .ti
whose general solution is w(s, r J = f(.1J + g(I). Thus, the general solution of the
equation
2u'-' - u,._,. - 11,T = 0
is
u(x, I) = f(x - 2y) + g(x + y)
where f and g are arbitrary functions.
In certain special cases the general solution of a partial differential equation

may be obtained by one of the methods used to solve ordinary differential
equations, as in equation (2.3). We consider two examples to further illustrate
the procedure.
Example 2.3. Find the general solution of the equation

(x > 0)
Sof111io11:Writing v(x, y) = uy(x, y) and dividing the equation by x 2 , the equation

can be put in the form
..
I
i-., + r
x2 -v
x
which can be regarded as an ordinary differential equation in the unknown

function v, with y appearing as a parameter. The general solution of this equation
is
y In x
L' ( X, y) = + -g(___Y)
x x
\Vhere g is an arbitrary function of y. Since ur = 1-, we find the solution 11 by
integrating 1 with respect to y, obtaining
y 2 In x I
u(x, y) = + /z( y) + f(x)
2x x
where h(y) = f g(y) dy andf(x) are two arbitrary functions.
Sol111io11: Treating x as a constant, this equation may be regarded as a second-order

differential equation in 11, with y as the independent variable. We know that the
general solution of the homogeneous equation
is
u(x, y) = f(x)e-xr + g(x)exr
where/ and g are arbitrary functions of x. To find.a particular solution uP of the
nonhomogeneous equation, we use the method of undetermined coefficients and
assume //P(x, y) = A(x) sin y + B(x) cosy. Substituting this in the equation,
we have
-(1 + x 2 )A sin y - B(l + x 2 ) cosy = x sin y
so that
x
A(x) = - 0
B(x) = 0
+ x-
Hence the general solution is
x sin y
u(x, y) = f(x)e-xr + g(x)exr -
I + x2
A typical mathematical problem involving a partial differential equation

consists in finding in some domain of the independent variables a solution of the
differential equation that satisfies certain given conditions, called the auxiliary
conditions. For problems arising from physical applications, these auxiliary
conditions are usually suggested by the physical problems associated with the
differential equations. In a type of problem that involves time as one of the
independent variables of the differential equation, the solution is usually sought
in some domain of space and time, which assumes prescribed values at an initial
instant of time. Such a problem is often called an initial value problem and the
corresponding auxiliary conditions are called initial conditions. In contrast is a
problem where the solution desired is in a domain in space that is bounded and
in which the values of the solution are prescribed on the boundary of the
domain; this is called a boundary value problem. The auxiliary conditions in
such a case are called boundary conditions. A problem that involves both
initial conditions and boundary conditions is termed an initial-boundary value
problem or, briefly, a mixed problem.
We recall that in ordinary differential equations an initial-value problem can
be solved by first finding the general solution of the differential equation and
then determining the arbitrary constants so that the initial conditions are
satisfied. This procedure also applies in the case of boundary value problems,
as we shall see in Chapter 4. In both cases the determination of the constants
in order to satisfy initial or boundary conditions is a simple algebraic problem.
The situation is not so simple with initial value or boundary value problems
in partial differential equations. For one thing, it is not always possible to find
the general solution of a given partial differential equation. And when the
general solution is known, it involves arbitrary functions, and therefore it is
often too difficult, if not impossible, to determine those functions directly from
the auxiliary conditions. For this reason, the general solutions in partial
differential equations are not so useful as they are in ordinary differential
equations; hence, their determination is often not the main concern in a given
problem. This is a basic difference between ordinary differential equations and
partial differential equations.
In many initial value or boundary value problems involving partial differential
equations, we usually seek, as many as possible, particular solutions of the
Sec. 2 General Solutions and Auxiliary Conditions 43
differential equation that satisfy some of the auxiliary conditions. Then, by

suitably combining these particular solutions, we attempt to find a solution of
the equation that satisfies as well all the remaining auxiliary conditions. A very
useful and convenient method that accomplishes this is the so-called separation
of variables method, which will be introduced in the next chapter.
Exercises 2.2
]. Find the general solution of each of the follo\\ing partial differential equations:
(a) u,. = 3x + y. (bl fix = l/x + e'"-
(c) llx = y sm x. (d) u, = xy + tan x.
(e) 11,. = f(x), where f is an arbitrary function of x.
{f) ux = g(x), where g is an arbitrary function of x.
2. Find the general solution of each of the fol Im\ ing equations:
(a) llx}' = ex-}. (b) llxr = x + 2y.
(c) Uu = yex. (d) /In. = 3y 2 sin x.
(e) u," = f(x)g(y), where f and g are arbitrary functions.
3. In each of the following equations, introduce the new variables indicated to

reduce the equation to one of the forms in Problem 2 above; then find the general
solution of the equation.
(a) 4uu - 11,.,. = O; s = x + 2y, t = x - 2y.
(b) llxx 6uxy + Sur,. = O; s = 5x + )',I = x + y.
(c) Uu + 411xy + 411J'J' = O; s = - 2.\ + )', I = y.
(d) llxx 2u,.,. + lln. = O; s = X, I = X + )'.
(e) 11\'X + 411,.,. = O;s = 2x + iy, I= 2x - iywherei = \ 1 -l.
(f) llxx 4uxr + l 3urr = O; s = (2 + 3i)x + y, I = (2 - 3i)x + y.
4. Obtain the general solution of each of the following by regarding the equation as
an ordinary differential equation.
(a) u,. + yu = 2xy. (b) 11 (2/y)11 = (3 cos x)'y 2 .
1. -
(d) u'-' + 2y11x - 3y 11 = 0.

2 2
(c) uu - 4y 11 = 3x.
(e) urr - 2.rn,. + x u = 0.
2
5. Find the general solution of the partial differential equation
.\u,,. + ur = 2xe'
6. The general solution of the differential equation 11,r = 0 is given by u(x. y) =

((x) + g(y), where/and g are arbitrary functions. Determine/ and _q so that the
solution u(.r, y) satisfies the auxiliary conditions u(x, 0) = x(x - 1l and
11(0, y) = sin y.
7. Find the solution of the equation uxr = 0 satisfying the conditions 11(x, 0)
sinh x and u(O, y) = cosy - I.
2
8. Solve the problem u,,. = xy, u(O, y) = y u,(x, 0) = sm x.
2
9. Solve the problem ilxr = x cosy. u(x, 0) = x, 11,.(0, y) = e" - I.
3. Linear Operators and the Principle of Superposition
The definition of a linear partial differential equation given in Section I can

be made more precise if we introduce the notion of an operator. This facilitates
at the same time the statement of an important property concerning combination
of particular solutions of a linear differential equation.
We say that Lis an operator on functions if it transforms each function u of
one given class into a function denoted by L11 of another class. For example, the
"wa\'e operator''
(3.1)
transforms each function 11 that has second-order partial derivatives with respect
to x and t into the new function
Lu = u,, - c"uu
The particular operator (3.1) is called a partial differential operator because it

imolves basically the operation of partial differentiation.
An operator L is said to be linear if, for any constants c 1 and c2 and any
functions u 1 and u 2 for which both Lu 1 and L11 2 are defined, it is true that
(3.2)
It follows readily from the rules of partial differentiation that the \vave operator
(3.1) is a linear partial differential operator. Property (3.2) can be extended
immediately to any finite number of functions by induction. That is, if u 1 . . u"
are 11-functions and c 1 , . . . c,, are 11-constants. then
(3.3)
\Ve call the function :L7o 1 !';II; a linear combination of 11 1 . . . 1111

Let Land AI denote two linear operators. \Ve define their sum L + Mas the
operator defined by the equation
(3.4) (L + M)11 =Lu+ Mu
for all functions 11 for which both Lu and Mu are defined. Thus, if L 1s the
operator (3.1) and A! is the linear operator
M = 1
(i'xc'') + x (D)
c'y,
+ xy
then the operator L + Mis defined by

Sec. 3 Linear Operators and the Principle of Superposition 45
If we replace u in (3.4) by c 1 11 1 + c2 u2 we readily see that
L(c1ll1 + C2!12) + Af(t1ll1 + C2ll2)
(c 1 L11 1 + c 2 Lu 2 ) + (c 1 lvfu 1 + c2 JfoJ

c 1(L11 1 + Mu 1) + c2 (Lu 2 + lvfu 2 )
C1(L + M)111 + C1(L + M)112
which shows that the operator L + Mis also linear. Thus, we conclude that the
sum of any finite number of linear operators is a linear operator.
In similar manner we define the product LM of the linear operators Land M
by the equation
(3.5) (LH)u = L(Mu)
for all functions u for which both Mu and L(M11) are defined. From (3.2) it
follows that
(3.6) L(c 1 M11 1 + c 2 Mu 2 )

C1(LM)111 + C2(LM)112
v.hich establishes the fact that the product of linear operators is also linear.
Now let L denote a linear partial differential operator. An equation of the
form
(3.7) Lu =f
\I.here f is a given function, is called a linear partial differential equation. If

f = 0, the equation (3. 7) is said to be homogeneous; otherwise, it is called
nonhomogeneous.
Let u 1 , . . . , u" be 11-functions that satisfy the homogeneous equation
(3.8) Lu = 0
Then. by (3.2), it follows that any linear combination of these functions also
satisfies equation (3.8). This important fact is known as the principle of super-
po~ition. This principle is used extensi\'ely in the solution of linear differential
equations. Correspondingly, if the functions i: 1 , . . . , i,, are such that Lii = /i
(i = I ..... 11), then the function r = c 1 + + t,, satisfies the equation
fr=/1 ++f,,
In particular, if 11 is a solution of the homogeneous equation (3.8) and v is a
particular solution of the nonhomogeneous equation (3.7), then the function
w = 11 + 1' satisfies equation (3.7).
Under certain conditions on convergence, the principle of superposition can
also be applied in the case where there are infinitely many solutions of a linear
homogeneous equation. Suppose that 11 1 , . . . , u", . .. is a sequence of functions
such that each 11; satisfies equation (3.8). Consider the "infinite linear combina-
tion
f.
(3.9) I
i= 1
C,11;
This is actually an infinite series. If the cor.stants c; and the functions 11; are
such that the series (3.9) comerges to a function 11. and such that the series can
be differentiated term by term so that the resulting series comerges to the
corresponding deri\ati\e of 11. for all derivati\es appearing in the operator L,
then it is true that
Hence, Lu 0, since Lu; = 0 for all i ;;:: I.
Example 3.1. Consider the \I ave equation
It is easily seen that the functions u,,(x, y) = (x - y)", 11 = I, 2, ... satisfy the
equation for each 11. Hence, by the principle of superposition,
.\
u\(x, y) = L c,,(x - y)"
11= I
is a solution of the 11ave equation for any integer N 2: I and for arbitrary con-
stants c,,, 1 .::; 11 .::; N. In particular, if the constants en are chosen such that
en = 0 and c 2 k- 1 = ( - l )k- l /(2k I)!, for k = I, 2, ... , then by the Weier-
strass M-test the series
)2k- l
L
Cf_ (
<- i ik- 1 .\ -

.i
k=1 (2k - I)!
and its derivatives of any order converge uniformly in any bounded domain of the
.\-;-plane containing the origin. Hence. by the extended principle of superposition,
the preceding series is also a solution of the wave equation. As a matter of fact,
we notice that the series converges to the function u(x, y) = sin(x - y), which is
easily seen to satisfy the wave equation.
There is a variation of the principle of superposition concerning solution of a

homogeneous differential equation that depends on a parameter. Superposition
in such a case is achieved by integration with respect to the parameter rather
than by summation. Suppose that i is a solution of the linear homogeneous
equation (3.8), which depends on the parameter t. Let Ii be an arbitrary function
of t and consider the function u defined by the definite integral
(3.11) u(x, y) = f li(t)r(x, y; t) dt

Sec. 3 Linear Operators and the Principle of Superposition 47
\\here a and b are constants. If we can legitimately differentiate under the

integral sign for all the derivatiYes inrnhed in the operator L, then clearly
the function (3.11) is again a solution of equation (3.8). More generally, if
the improper integral
(3.12) r'I: /i(t)1(x, y: 1) dt

l) ~ 'l,
converges suitably with respect to x, y, and if the operator L can be applied

inside the integral sign, then the function defined by the integral (3.12) is also a
solution of the homogeneous equation (3.8).
Example 3.2. Consider the heat equation
It is clear that for any value oft, the function r(x, y; t) = e-r' sin xt satisfles the
heat equation. Let lz be an ab!>olutely integrable function on the interval 0 <
t < rx. \Ve shall show that the function 11 defined by the integral
(i) u(x, y) = L" 2

hU)e-"' sin xt dt
is a solution of the heat equation in the upper half-plane - er., < x < co, y > O.
First, \\e note that by Weierstrass M-test the integral (i) converges absolutely
and uniformly \\ith respect to x and y. since
(y ;:>: 0)
and his absolutely integrable on the interval 0 < t < x. Thus, 11 is a continuous
functionofx,yintheupperhalf-plane. Next, let h(r), <Mandy::>: y 0 > Oand
consider the integral
(ii) - Jor'l 1 h(1)e-n' sin xi dt

2
\\ hich is the formal deri\'ative of (i) \\ith respect lo y. Since

2
l1 /i(t)err' sin xi < i\.1t 2 e-'' 0 ' 2
and the function on the right is integrable from zero to infinity \\ith respect tot,
it i'ollm's by the M-test that the integral (ii) is uniformly convergent. Thus,
imcgral (i) is differentiable \\ith respect toy and
. . f' -
Cll
.
CT
=
(
.
C\' 0
/z(l)e
_ .,2
.i
.
sin xi dt = -
loo t h(t)e
0
2 . _. 2 .
'' s111 xt dt
Jn the same 11ay, integral (i) can be differentiated twice with respect to x under the
integrJ! sign. Therefore, for - x < x < x, y > 0,
= 0
Exercises 2.3
1. Let L = a(c,cxl b(c,cy) + c and .\f = c:tCC/cx) + /J((/cy) + i', \\here a, b, c

and <. [J. ;are constants. Sho\\ that LV/ = .'If L
2. Give examples of linear operators Land .\f to sho11 that LM and .\IL are not
ahvays equal.
3. Verify that each of the functions 11,,(x,t) = exp[-(11- ~) 2 1] sin (11 - }Jx,
11 = I. 2, ... satisfies the heat equation // 1 - llxx = 0 and the auxiliary condi-
tions 11(0, 1) = 0, un. 1) = 0. Then find a linear combination of these func-
tions that satisfies as 11ell the condition u(.Y, 0) = 2 sin ~x - 4 sin ix.
4. Verify that each of the functions
11 11 (x, t) = exp [- (k + 11
2
)1] cos 11x (n l, 2, . . )
satisfies the differential equation
11, - ll_'(X + ktt = 0 (k = const)

and I he auxiliary conditions 11)0, I) = 0 aEd 11);r. I) = 0. Then find a solution
of the given differential equation that also satisfies the condition u(x, 0) =
cos 3 x.
11
11
(x, I J = cos 11 cl + -112 t sin nx (11 I, 2,
satisfies the differential equation
(c = const)
and the auxiliary conditions 11(0. t l = 0, 11(n. I l = 0, 11,(x, 0) = 0. Then find a

solution of the given differential equation that also satisfies the condition
11(x, OJ = sin 3 x
11 11 (x, y) srn imx sinh nn (y - 1) (11 = l. 2, )

2 2
satisfies Laplace equation 11xx + 11rr = 0 and the auxiliary conditions 11(0, y) = O,
11(2, y) = 0, u(x, 1) = 0. Find a linear combination of these functions that
satisfies the condition u(x, 0) = sin nx -- 3 sin 2nx.
7. Let y 0 > 0. Show that the series ~~-~ 1 c 11 e-nr sin 11x, 11 here the coefficients c"
are all bounded, is twice differentiable \1ith respect to x and y and satisfies
Laplace's equation un + 11n. = 0 in the domain y > Yo
8. Show that the series
is twice differentiable with respect to x and 1, and satisfies the differential equation
11 11 - lixx + 2u, = 0 for all values of x and t.
Sec. 4 Linear Fint-Order Equations
9. Consider the Helmholtz equation 11-'-' + 11,, + k 2 11 = 0, (k = const).

(a) Show that an exponential solution of the type 11 = e'x,r (where .1 and jJ arc
constants) exists. provided 1 and /3 satisfy the equation 1 2 + /3 2 + k2 = O.
(b) Let Cl = ik cos 0 and jJ = ik sin 0. \\here 0 is an arbitrary real constant.
Verify that
cos [kh cos 0 + y sin Ol ]. sin [Id.I' cos 0 + .1 sin 0)]
are particular solutions or the Helmholtz equation.
(c) Verify that the integral
,.,.o,
j - h({i) e:xp [i/d \ cos 0 + y sin 0)] d()
e,
also satisfies the Helmholtz equation if his a function for which differentiation
under the integral is valid.
10. Obtain exponential solutions 11 = ,,,x~-111 of the wave equation 11 11 - 11,, = 0 and
show that the integral
is a solution if h is a function for which the integral converges and can be

differentiated under the integral sign.
4. Linear First-Order Equations
In this section we shall study linear first-order differential equations in t\\O

independent variables x and y. The general form of such an equation is
(4.1) a(x, y)u, + b(x, _1')11
1
+ c(x. _1')11 = d(x, r)
where the coefficients a. b. c. and dare functions in some domain D of the

.\T-plane. We assume that the functions a. b. c, and dare continuous and hme
continuous partial derivatives of the first order in D, and that a and bare not
hoth zero. We shall describe here a method for finding the general solution of
equation (4.1 ). The method is based on the possibility of transforming equation
(41) into an equation of the form
(..f.2) l\"~ + ,I(~. 1/)\1' = I(~. 1/)
where ~ and 17 are new independent variables. Equation (4.2) can now be
regarded as an ordinary differential equation with as the independent variable
anJ 11 as a parameter. Hence. its general solution can be found by the use of the
standard formula pertaining to such an equation. The integration constant in
the solution must, however, be replaced by an arbitrary function of 17. The
general solution of the original equation is then obtained by returning to the
variables x, r.
Indeed. if either one of the coefficients a, bis identically zero, then equation
(4.1) is readily reducible to the form (4.2). Let us assume that a =I= 0 and b =/=" 0.
50 Linear Partial Dijjerential Eq11atio11s Chap. 2
In order to transform equation (4.1) into the form (4.2) we introduce ne\\
\ariables sand 17 by the equations
(4.3) ~ = </!(.\. _r). 17 = 1/;(x. _1)
The functions</; and 1/;. \vhich will be subsequently delermined. are assumed to
be continuous and ha\e continuous first-order parlial deri\ati\es in D such that
the Jacobian
(4.4)
111 the neighborhood of some point in D. Writing 11(x. _1) 11k. 11 ), we then
ha,e by the chain rule.
llx +
ir::,r/; x h',,1/1_\,..
u,. H/P_, + ll',11/1)
and hence equation (4.1) is transformed into the equation

(4.5)
The coefficients a, b. c, and dare now to be considered as functions of sand 17.
If \le choose the function 1/1(x. r) in (4 3) such that
b
( 4.6) mflx + lJiflv = 0 or
a
then the term imolving 1r,1 in (4.5) drops out, and for any choice of r/; satisfying
the condition (4.4). equation (4.5) reduces to the form (4.2). Thus, it remains
for us to determine a function 1/1 that satisfies (4.6).
Suppose for the moment that such a function ijJ e\ists with 1/;r i= 0: set
ij;(x. r) = c. \\here c is any constant. Then. on taking the total differential of
i/J = c. \IC find
which implies that

cir lf1x
dx v1_..
Hence. if 1/; satisfies equation (4.6). then 1/; = c must be an imegral of the first-
order ordinary differential equation
dy b
(4.7)
dx a
Comersely, if ijJ = c \\ith I/Jr i= 0 is an integral of equation (4.7), then by
re,ersing the foregoing argument it follows that 17 = 1/;(x. y) satisfies equation
(4.6).
Equation (4. 7) is often called the characteristic equation of the partial
differential equation (4. I). It defines a one-parameter family of curves called the
Sec. 4 Linear First-Order Equations 51
characteristic curves of equation (4.1). These curves play an important role in

the consideration of initial \alue problems for equation (4.1 ). as we shall
soon see.
Thus. in order to reduce equation (4.5) to the form (4.2), we choose if; such
that 1/; = c is a characteristic cune of equation (4.1 ). The function </> may be
chosen arbitrarily. subject to the condition (4.4). We choose 1J(x, y) = x.
Then condition (4.4) is fulfilled and equation (4.5) takes the form
(4.8) 11~ + C) 11 C)
which is of the desired form (4.2).
To obtain the general solution of equation (4.8), \Ve multiply both sides of the
equation by the integrating factor
r(, 17) = exp(J ~di;)

and observe that
Integrating with respect to, treating 17 as a parameter, we then find
1r(, 17) = _
1
r((, 17)
rlJ{ ad 1'(, 17) d + /(17)1
.
where .f is an arbitrary function of 17. Hence, returning to the original variables

x. r, we obtain as the general solution of equation (4.1)
1
(4.9) 11(x, r) {W [x. 1/;(x, .r)] + f[i/;(x, y)]}
v[x, 1/;(x, y)]
where we have set
W((, 17) = J: i:((, 11) d
We note that the first term, W(x, ij;)/r(x, if;), in formula (4.9) is actually a
particular solution of equation (4.1), and the second term,f(x, ij;)/c(x, if;), is the
grneral solution of the corresponding homogeneous equation
(4.10) a(x, y)ux + b(x, y)uy + c(x, y)u = 0

In manv special cases, it may be easier to find a particular solution of (4.1) by
some other technique. For instance, the method of undetermined coefficients
used to determine particular solutions in ordinary differential equations may be
employed under similar conditions
Example 4.1. Find the general solution of equation
Gllx + /){{J' + Cl/ = 0

\\here a. h, c arc constants and a cf- 0.
So/11tio11: We see that the characteristic equation
dy b
dx a
defines a one-parameter family of characteristic curves
bx - ay = const
which arc straight Jines. Introducing the new variables ~ X, I/ bx - G)',
the given equation becomes

c
}1.:.: + )V 0
a
The general solution of this equation is
11'(-;, 17) = e-r~'"/(11)
where f is an arbitrary differentiable function. Thus, the general solution of the

given differential equation is
u(x, y) = e-cx "/(bx - ay)
Ir a = 0, then the differential equation reduces to 11,. + (c/b)11 0, whose

general solution is readily obtained as
Example 4.2. Obtain the general solution of the equation
)..."II_'( - Jlly + II = X
Su/111io11: \Ve consider the characteristic equation
dy y
dx x
By separation of variables \\e find that the characteristic curves are given by
xy =c c. We therefore introduce c; = x and 11 = xy. The transformed equation is
ll'~ + ll' = l
-;
\\hose general solution is
"
ll'(s, I/) = c + l f(lJ)
2 ~
Hence. the general solution of the original equation is
x: I
u(x. _1) = + f(xy)
2 .\'
where f is an arbitrary differentiable function.
In the study of the first-order ordinary differential equation di 1dx = /(x. r).
it is usually required to find a solution of the equation that assumes prescribed
\alue at a specified point on the .\"-a\is. Geornetricall:-. this means finding an
integral cune that passes through a specified point in the .\T-plane Under rather
general conditions on I such a problem possesses one and only one solution.
A similar problem for the lirst-order partial differential equation (4.1) consists
in determining a solution of the equation that takes on prescribed \alues at
points on a specified cur\"c in the xy-plane. This corresponds to finding an
integral surface that contains a specified cunc in .\T11-space. If the parametric
equations of the cune in the .\T-plane are
(4.11) C: x = x(s). r = r(s)
and if 11 <f;(s) on this cun e. then the problem of finding a solution 11 = 11(x. y)
of the equation (4.1). such that 11 = c/>(.1) on the cune C. amounts to the deter-
mination of the function/in the general solution (4.9). In general. the function
I can be uniquely determined in terms of the initial \aluc c/! and the other
specilic functions in (4 9) when the equations (411) of the cune Care substituted
for x. _1 in (4.9) and 11 is set equal to <f;.
Example 4.3. Find the solution of the equation

2u, - ?.11,. + 211 2x
2
\I hich assumes the value 11 = 1 on the liner = - '"..,
Sof11rion: B~ formula 14.9) the general solution of the differential equation is

u(x. y) = (x - Il + e-~((3x + 2y)
Substituting y -- -x 2 in the solution and setting 11 = x 2 11e have

,
2
= x - 1 + e-'/(2x)
or
f(2 I) = (\ 2 - .\ + ] )e-'
Let r 2.1 so that x ~ r 2. Then
I 2
2
./(11 = ( 4 2
1 ) e'
Thus. the solution of the problem is

~(3x +
2
u(x, y) = x - I + __ ,
e
2_1) (3x 1.- 2y) + ]] e(3x+ 2.n 2
..,
4
x - I ,- I (3x : 2y)2 - (3x : 2y) + I l ecx+hl 2
It should be pointed out that the cune Con which the values of 11 are pre-
scribed cannot be taken arbitrarily. Specifically. it cannot be a characteristic
cune of the differential equation. The reason for this restriction is apparent
when we look at the form of the general solution (4.9) of equation (4.1 ). On a
characteristic curve, say, ij;(x, y) = c 1 , the solution (4.9) assumes the form
ll =
Consequently, unless the prescribed function is of the special form
(4.12)
where k is a constant, the problem has no solution. On the other hand, if has
the form (4.12), then there exists infinitely many solutions given by (4. 9), where
f is any differentiable function such that f(c 1 ) = k.
Example 4.4. Show that the problem

2ux - 311,. + 211 = 2x, 11 = c/J(x), on y - 3x/2
has no solution unless has the form
c/J(x) = x - 1 + ke-x
where k is a constant. lf has the indicated form, give a solution of the problem.
Solution: The general solution of the differential equation is
u(x, y) = (x - 1) + e-xf(3x + 2y)
Substituting the prescribed conditions, we have

4J(x) = x - I + e-xf(O)
But for any function f /(0) is a constant. Hence, the problem has a solution only
if is of the form so indicated. In such a case,
ll(X, y) = X - I + e-xf(3x + 2y)
is a solution for any differentiable function f such that /(0) = k.
Exercises 2.4
In Problems 1 through 5, find the general solution of each of the equations.

1. llx + UY -= 0. II 2. 2ux - 311,. = X.
3. llx - 2uy + II = Sin X + y. 4. 311x - 411,- + 2u = x 2 y + 2ex + J.

5. llx - u,. - 211 = e 2x COS 3y.
6. Let Lu = au~ + bu,. + cu = 0, where a, b, c are constants. Show that by
introducing the new variables .; = ax + py, 17 = 1x + by, (:ic5 - P1 #- 0), the
equation is transformed into
II~ + CU = 0
\\hen the coefficients C\, /3, i', c5 are chosen so that
0) + b/3 = 1, Gi' + bc5 = 0

Hence, rederive the solution obtained in Exalllple 4.1.
In Problems 7 through 13, find the general solution of the given equation.
2
7. Xllx + Jllr = l. 8. Xllx - )'lly - xu = 2e" + x y.
9. Jllx + xu,. = y. 10. x u, - xyu,. + 2y11

2
= 0.
11. 2
y ux - xy11,. - 2x11 = x 2
+ y 2
.
12. (x + y)(u, - 11r) + 11 = e-x.(.YJ.
13. xyux - x 2 uy - yu = xy.

14. Let Lu = Au, + Bur + Cu= + Du = 0 be a linear equation in three variables
x, J', ::, where A, B, C, Dare constants. Introduce the new variables
~ = a 1x + h1Y + CI;:
I/ = a1 x + h2Y + C2Z
( = G3X + h3y + C3Z
where the deterlllinant of the coefficients a;. h;. c;. i = l, 2, 3 is not zero.
(a) Show that the coefficients a;. h;. c;. l :S i :S 3, can be chosen such that the
equation is reduced to the form
11~ + Du = 0
(b) Hence. obtain the general solution in the form
11(x, y . .::) = c-Dx'Aj(Bx - Ay, Cy B::)
assullling that none of the coefficients A. B. Care zero.

(c) Obtain the general solution \\hen one of the coefllcients A. B, C is zero.
15. Apply the result of Problem 14 to find the general solution of the following
equations:
(i) u, + '211.r 3u: + II = 0.
(ii) '2u, - 11,. + '211= + 311 = x + y + ::.
111 Problems 16 through 22. find the solution of the differential equation satisfying the
rrcsnibed condition.
16. llx + u,. 1. /1 = e" ''hen y = 0.
17. 11, , 11,. u = 0. 11 = I + cos x. when .1 = 2x.
18. 2u, - 511,. + 411 = x 2
, 11 = sin y + e-" + ~-\\hen x = 0.
19. 11" + (cos x)11r = sin x. 11 = y - 1 + cosy. when x = 0.
20. xu, .r11r + 11 = x. 11 = x when y = x. 2
21. ) 11, xu,. + xu = 0. 11 = y. when x 2 + 2y 2 = 4.

22. _rn, .rn,. + 2xy11 = 0, 11 = c" sin(x + I), when y 2 = 2x + I.
23. Show that no solution exists for the differential equation of Problelll 22, which
assullles the prescribed value ef>(_y) on the circle x 2 + y 2 = a 2 , unless cf> is of the
2
form qJ(x) = kc-x where k is a constant. If cf> has the indicated form. show that
the problem has infinitely many solutions.
24. Obtain the general solution of the equation y11, + x11,, - y11 = xex in the form
u(x, y) = e'y + e'f( y 2 - x 2 ). l f /1 is prescribed as q; on the upper portion of the
hyperbola y 2 - x 2 = 1 ( y ::::: I). shm\ that no solution exists unless qi is of
special form. Find this form and show that in such a case there are infinitely
many solutions.
5. Solutions of Second-Order Equations with Constant Coefficients
The general linear second-order partial differential equation in two inde-

pendent variables x. r has the form
(5.1)
where the coefficients A. B, C D, E. F, and Gare functions of x and yin some

domain of the xr-plane Unlike the first-order equation (4.1), it is impossible to
obtain the general solution of (5.1) except in some special cases. We shall
consider here the special case
(5.2)
where a, h, c. d. e. and /"are constants such that a, b, and care not zero simul-
taneously. It is col1\enient to \Hite equation (5.2) in the form
(5.3) Lu= 0
where L denotes the second-order differential operator
L = aD~ + bDxDy + cD;. + dDx + eDY +f

with D, = 1'/!'x and D"' = ?jr'.1'.
Consider the polynomial P(x, y) obtained by formally replacing D, and DY in
(5.3) by x and r. respectively. Suppose that P(x. J) can be factored as a product
of two linear factors: that is.
Then, accordingly, the operator L can be written in the factored form
where L 1 and L 1 are linear first-order operators gi\en by
(5.4)
Since the coefficients of L 1 and L 2 are all constants, we know (Problem I,

Exercises 2.3) that L 1 L 2 = L 2 L 1 and so we have
(5.5)
When this is the case, we say that Lis factorable or reducible.

In case the operator Lis factorable, a solution of equation (5.3) that involves
two arbitrary functions can be obtained by means of the result of the preceding
Sec. 5 So/11tio11s of Seconrl-Order Equations with Constant Coefficients 57
section. Such a solution constitutes the general solution of the equa1ion.

Indeed, suppose (5.5) holds. where L 1 and Li are given by (5.4) and L 1 i= Li.
Let 11 1 be the general solution of Lu 1 = 0 and let ui be the general solution of
Lui = 0. Set 11 = 11 1 + Iii. Then it follows from (5.5) and the linearity of L
that
(5.6) Lu L(u 1 + 11 2 )
Lu 1 + L11i
L 2 (L 1 u 1 ) + L 1(L 2 ui) = 0
~hO\\ing that 11 = 11 1 + ui satisfies equation (5.3). Now. according to Example

4.1, if a 1 i= 0 and a 2 i= 0, then
a I y)
and
u 2 (x, y) = e cixa'j~(bix - a 2 y)
Hence, if a = a 1a 2 i= 0, the general solution of equation (5.3) is
(5.7) u(x, y) = e-c,xfa'f1(b 1 x - a 1 y) + e-cix" 1{ 2 (b 2 x - a 2 y)
\\here the functions / 1 and/~ are arbitrary and twice differentiable.
If either a 1 = 0 or a 2 = 0, the corresponding term in (5.7) must be replaced
by e-c,x b, f1(X) Or e-c2x'hf~(X).
The general solution of the nonhomogeneous equation Lu = G is obtained
bv adding to the solution (5.7) any particular solution of the nonhornogeneous
equation.
l~Jcttmple 5.1. Find the general solution of the equation
un - u,r - llx + u, = 2 cos(3x + 2y)
So/111ion: The differential operator L = D~ - D; - DX + D, ean be factored with

L 1 = (D" +Dr - l)andL 2 = <Dx - Dr). Thegeneralsolutionoftheequation
L 1 11 1 = 0 is
and the general solution of the equation L 2 u 2 = 0 is
11 2 ix, y) = g(x + y)
Thus. the general solution of the corresponding homogeneous equation Lu = 0 is

11(x, y) = exf(x - y) + g(x + y)
where f and g are arbitrary functions that are twice differentiable.
We now look for a particular solution of the nonhomogeneous equation

Lu = 2 cos(3x + 2y). We use the method of undetermined coefficients and
assume a particular solution in the form
i(x, y) = A cos(3x + 2y) + B sin(3x + 2y)

58 Linear Partial Dijjerential Equations Chap. 2
Substituting this in the equation and collecting similar terms. \\e ha\e
-(SA + B) cosl3x + 2y) + (A - SB! sin(3x + 2y) = 2 cos(3_\ -' 2y)
Hence.
-(SA + Bi 2
A - 58 = 0
which gi\eS A = --/rand B = - -1\- _ Therefore. the general solution of the given
equation is
u(x, y) = e''j(x - y) + g(x + y) - -is cos(3x + 2y) - 1 's sin(3x + 2y)
Next \Ve consider the case when L 1 = L 1 ; that is, when L has repeated
factors. We wish to determine the general solution of the equation
(5.8)
If we set L 1u t\ then in order that u be a solution of equation (5.8). i must
satisfy the equation
(5.9) 0
When a 1 -/= 0, we know that

t(X, y) = e-c,xfa,g1(b1X - al y)
Thus, to determine the function 11, we have lo solve the equation

(5.10)
By the method of Section 4, we introduce new variables ( = x, 17
Then equation (5.10) is transformed into the equation
a 1 w~ + c1w = e-c,~/ 1 g 1 (17)
where ir((, 11) u(x, y). The general solution of this equation is
e-c,~., l~l I g1(1J) d( + f2(1J)J

I .. ,
= -- e-c,, a'(g1(11) + e-c";a'f2(17)
ai
Therefore, the general solution of equation (5.8) has the form

(5.11)
where / 1 and f 2 are arbitrary functions that are twice differentiable.

The procedure described above can also be used to find a particular solution
of the nonhomogeneous equation Lu = G, whenever Lis factorable. Suppose
L = L 1 L 2 Let v be a particular solution of L 1 v = G and let u be a particular
solution of L 2 u = i:. Then
Lu = (L 1 L 2 )u = L 1 (L 2 u) = L 1 v G
Sec. 5 Solutions of Second-Order Equations with Constant Coefficients 59
showing that u is a particular solution of the equation Lu = G. Here L 1 and

L 2 need not be distinct.
Solution: The operator L = D~ - 2DxD,. + D~. has repeated factors L 1 = L 2 =

(Dx - Dy). Hence, by formula (5.11), the general solution of the corresponding
homogeneous equation is
u(x, y) = xf(x + y) + g(x + y)
To find a particular solution w of the nonhomogeneous equation, we let v be a

particular solution of the equation
(i) Vx - v,. = xy
and determine w from the equation
(ii) Wx - Wy = L'
Making the change of variables .; = x, 17 = x + y, equation (i) becomes
v~ = (17 - )
with a particular solution
3
(iii) v = - - - + 211
---
3 2
Introducing the same new variables in (ii) and using (iii), we obtain
3 211
w~ = +
3 2
whose particular solution is
4
w= + 311
--
12 6
Thus, a particular solution of Lu = xy is
w(x, y) = -
x4
- - +- y)
+ x3(x
12 6
Combining this with the function 11, we obtain the general solution of the given
equation.
Exercises 2.5
In Problems 1 through 8, find the general solution of the differential equations.

I. U". + llx + 11,.+ ll = 0.
2. Uu - u,.,. - llx + lly = 0.
3. llxx + llyy = 0.
60 Linear Partial Dijj'erential Equations Chap. 2
4. u,-x + 2u,r + 211,.,, = 0.

5. llxy 211)';' + 3u,. = o_
6.
I c2 11
-- -
I (;2 '
( r - II) 0. (Let r
c 2
ct 2 ,.2(;r2
4
C II
7. = 0.
i:y4
04 U 04 11
8. 2 + = 0.
cx 4 2
ox oy 2
cy 4
9. Consider the homogeneous operator with constant coefficients
02 02 02
L=A-- 2 +2B- +C 2
ox ax oy cy
(a) Shov. that this operator is ah\ ays factorable with real coefficients if and only if
B 2 - AC 2: 0.
(b) Show that if f is an arbitrary twice-differentiable function, then 11 = f(mx + y)
is a solution of the equation L11 = 0 if and only if m satisfies the equation
Am 2 + 2Bm + C = 0.
(c) If 8 2 - AC = 0. verify that a second solution of L11 = 0 is given by
11 = xf(mx + y).
In Problems I 0 through 13, find a particular solution of each of the given non-
hornogeneous equations.
10. Uu 411)'}' = x + y.
11. Uu 61/X}' + 91/}" = A \'l
12. llx_, + llx + 11,. + II = e-'->'.
13. 11_'-' - 2u"' + 11,.Y + llx - 11,. = sin(2x + y).

14. ShO\\ that the equation with variable coefficients ax 2 11xx + hxy11x,, + cy 2 11,.,. +
dx11x + ey11,. + fi1 = 0, where a,_ .. , fare constants, is reduced to an equation
\\ ith constant coefficients by the change of variables = ln x, 17 = ln y.
15. Use the method of Problem 14 to obtain the general solution of each of the
following equations:
(a) x 2 uxx - y 1 un. + Xllx - y11,. = In X.
(b) xyun. + xu, + yu,. + 11 = xy.
2
(c) x 2 u,x + 2xy11x,. + y u,.y - llXllx - llYllv + nu = x + y, 11 const ct 0.
16. Find the general solution of each of the follO\\ing equations:
(a) (Dx Dy)(Dx + D,Ju = 11.n - u,Y + llxz - llyz = 0.
(b) (D_, D,. + Dz)(Dx + 2D,. + 1)11
= ll_u + lixy - 211,.Y + llxz + 2llyz + fix - lly + IL 0.
(c) (D_, + D, + D,) 2 11 = Uu + 11,., + llzz + 211xy + 211,.z + 211zx = 0.
6. Classification of Second-Order Equations
The study of a linear second-order partial differential equation is often

facilitated by a recognition of the type of the differential equation in question.
For, depending on the type of the equation, it is frequently possible by means of
Sec. 6 Classification of Second-Order Equations 61
a coordinate transformation to reduce the equation to one of three canonical

forms. These canonical forms correspond to different simple forms in which
the second-order derivative terms can appear in the equation. Moreover, the
type of a partial differential equation plays a decisive role in determining the kind
of auxiliary condition that can be considered with the equation so that the
resulting problem has a unique solution.
In this section we shall consider the classification of linear second-order
partial differential equations in the special case
(6.1)
where all the coefficients, A, ... , Fare constants, with A, B, C not all zero. In
an attempt to simplify the form of the second-order terms of this equation, we
introduce new variables ~ and 11 by means of the linear transformation
(6.2) ~ = O'X + {Jy, Y/ = yx + by
The coefficients c;, {3, y, and <5 are constants, which will be determined subject
to the condition ab - /31 f= 0. This last condition, which is the Jacobian
C:((, ry)/i}(x, y) enables us to solve for x and y in terms of ( and 17. Writing
u(x, y) = w(~, 17), we have by the chain rule
ux O'\\'~ + ya,,
uY (fil'.:, + bw,1
Cz'. li'~~
2
Uxx + 2~qn1~,1 + t12 H,,,1
llxy + (c;b + {Jy)ir,,1 +
c;{Jir.:,.:, y15wry,1
2
urr (l2w.:,:, + 2f3<5ll'.:,ry + i5 w,,ry
Substituting these in equation (6.1) and collecting similar terms, we have
(6.3)
\\here
a + 2Bc;f3 + Cf3 2
Ac; 2
(6.4) h Ar; + B(c;J + {J}) + C(N1
c = Ay2 + 2B115 + C3 2
The dots in (6.3) represent terms involving wand its first derivatives 11, and iv~.
It is easy to verify that
(6.5) b2 - ac = (B 2 - AC)(cxb - (fy) 2
by simply substituting the expressions (6.4) for a, b, and c.

Now, since the constants a, {J, /',and 3 are at our disposal, we shall choose them
in such a way that at least one of the second-order terms in (6.3) drops out. As a
matter of fact, we shall show that according as (a) B 2 - AC > 0, (b) B 2 -
AC = 0, (c) B 2 - AC < 0, the constants r1., {J, y, and i5 can be chosen such that
'
'
(a) a = 0, c = 0, (b) b = 0, c = 0, (c) a = c, b = 0, thus simplifying the

equation to one of the three canonical forms:
(i) H'~,,
+ ... = 0
(6.6) (ii) }t' ~,!' + ... = 0

"
(iii) u~~ + w,,,, + ... = 0
We say that the partial differential equation (6. J) is of the
(i) hyperbolic type, if B 2 - AC > 0
(6.7) (ii) parabolic t;pe, if B 2 - AC = 0
(iii) elliptic type, if B 2 - AC < 0
We see in the above classification that only the coefficients of the second-order
terms matter. The quantity B 2 - AC is called the discriminant of the equation.
In view of formula (6.5), we see that the sign of the discriminant remains un-
altered under the change of variables (6.2). Thus, the type of equation (6.1) is
invariant with respect to the linear transformation (6.2). This invariance
property is, of course, to be desired if the definitions (6.7) of hyperbolicity,
parabolicity, and ellipticity of partial differential equations are to have any
significance.
Example 6.1. The partial differential equation
is hyperbolic, since A = I, B = 0, C = - I and B 2 - AC= I > 0. This

equation is called the wave equation.
is parabolic, since A = B = O. C = - I, and B 2 - AC = 0. This equation is

called the heat equation.
is elliptic, since B 2 - AC = - 1 < 0. This is called Laplace's equation.
We remark that in the general case where the coefficients A, . .. , F of equation

(6.1) are functions of x, .r in a domain D, the equation is said to be of the hyper-
bolic type in D if B 2 - AC > 0 at all points in D. Similarly, equation (6.1) is
said to be of the parabolic or of the elliptic type in D according to whether
B 2 - AC = 0 or B 2 - AC < 0 at all points in D. The reduction of such an
equation to one of the canonical forms (6.6) is accomplished by means of a
transformation
~ = ~(x, y), 17 = 17(x, y)
Sec. 7 Tlte Canonical Forms 63
The determination of the functions ( and YJ, ~owever, involves solving first-
order differential equations, which we shall not pursii'e in this book.
It should be noticed that if equation (6.1) involves variable coefficients, it can
very well happen that the equation is of one type in one part of the domain and
of a different type in another part. For example, consider the equation
where A = 1, B = 0, and C = -yin the whole xy-plane. Since B 2 - AC = y,

it follows that the equation is of the hyperbolic type in the upper half-plane
(y > 0), of the parabolic type on the x-axis (y = 0), and of the elliptic type in
the lower half-plane (y < O).
Exercises 2.6
Classify each of the following partial differential equations.

1. 11.u + u,,. - /In, + llx = 0.
2. 2uu - 4uxr + lln - u,. = 11.
3. 3u.n - 2uxy + u,.,. = xy.
4. 4u,r + + lly + 3u = 0.
711}.,. - 211x
5. 2uu +
+ //}l' - XII = ex>
/lXJ,
2
6. 4uxx - 4ux,. + llyy + 3u, = x + y .
2
7. 4uxx + 12uxy + 9uH - 2ux + II = 0.

8. uxx - u,,. + Sun, + .Hix + yu,. = 0.
9. Determine the region in \\hich the equation
(x 2 - I )uu + 2yu,,. - llyy + llx + lly = 0

is (i) hyperbolic, (ii) elliptic, (iii) parabolic.
10. Determine the region in which the equation
Un + 2xuu + )'llyy - 11 = 0
is (i) hyperbolic, (ii) elliptic, (iii) parabolic.
7. The Canonical Forms
We now consider the problem of choosing the constants ix, {3, }', and c5 in (6.2)
so that equation (6.3) will reduce to one of the canonical forms (6.6).
(i) H1perbolic Trpe, B 2 - AC > 0. In this case we shall show that (6.2)
can be chosen such that the coefficients a and c in (6.3) vanish, while b i= 0. We
therefore consider the equations
(7.1)
a = Aa 2
+ 2Baf3 + C/3 2 = 0
c Ay 2 + 2Byb + Cc5 2 = 0
If A C = 0, then of necessity B -:/- 0 and thus equation (6.1) is readily put

in the canonical form (6.6i) by dividing out by 2B. Jn this case, the trans-
formation (6.2) may be chosen as the identity transformation
( = x, 1/ = y
with 'l. = b = l and f3 = ( = 0.
Suppose that A -:/- 0. If fJ = 0 or b = 0, then from (7.1) we would have
r.x = 0 or -/ = 0. In either case, the condition r.xb - fJy -:/- 0 is violated. There-
fore, fJ and 6 cannot be zero and hence equations (7.1) can be written as
+C=O
(7.2)
Thus, in order that the coefficients a and c will vanish, we shall choose rJ., fJ, y,
and c5 such that ry_//J and 1/b are roots of the quadratic equation
(7.3) Am 2 + 2Bm + C = 0
1
Since B AC > 0. equation (7.3) has two distinct real roots
-B + (B 1 - AC) 12 -B - (B 2 - AC) 112

1111 = - - 1112 =
A A
Let us choose rJ. = -B + (B - AC) 1 , fJ = A, I' = - B - (B 2 - AC) 1 12 ,
2 1 2
and 6 = A. Then 'Y.6 - /Jy = 2(B 2 - AC) 112 > 0 and the coefficients a and c
in equation (6.3) drop out. From (6.4), we see that
b = Ar/ + B('Y-6 + {3y) + C/Jr5

2
= -2A(B - AC)-:/- 0
Therefore, when equation (6. l) is of the hyperbolic type, the transformation
[ -B + (B 2 - AC) 112 ]x + Ay
(7.4)
1/ [ -B - (B 2 - AC) 112 ].\'. + Ay
will reduce the equation to the form

2bll'~,, + ... = 0
where b -:/- 0. We now have only to divide through by the coefficient 2b in

order to obtain the desired canonical form (6.6i).
The case A = 0, C -:/- 0 can be treated in a very similar way.
If we further introduce the variables ~' and 11' defined by the equations
(7.5) ~' = ~ + 11
2 2
Sec. 7 The Canonical Forms 65
the canonical form (6 6i) becomes
(7.6)
This is the alternative canonical form for equation (6.1) of the hyperbolic type.
This alternative form can. of course. be achie\ed directly from equation (6.1)
by means of the transformations
~I -Bx + Ay
(7.7)
17' (B 2 - AC) 112 x
and di\iding out the result by the coefficient -A(B 2 - AC). which is different
from zero.
Example 7.1. Sho\\ that the second-order partial differential equation
Uu + 611.n - J611J"J" = 0
is hyperbolic and find its canonical form and its general solution.
So/111io11: Here A = 1, B = 3, and C = -16 so that 8 2 - AC= 25. Thus, the

differential equation 1s of the hyperbolic type. By (7.4) we introduce the new
variables
~ = 2x + J. I/ 8x + y
Then
llx = 211 0 - 811,1
11,. = 11, + 11,,
Uu = 411,.: - 3211.:,1 + 6411,111
llxy 211::,c, - 611,,1 - 811,1,1
llyy = II~~ + 'J..t1S11 + 11,,,,
Substituting these in the differential equation we find the canonical form
-10011~,, = 0 or
If we have used the.variables given in (7.5), we find the alternative canonical

form
u l,..,
. ,, - II
,, ,,
= 0
The first canonical form has the general solution
II = /() + g(I/)
\\hile the second canonical form has the general solution
II = /(' + 11') + g(' - 1()
Either one leads to the general solution
u(x, y) = /(2x + y) + g( - 8x + y)
of the original differential equation.

66 Linear Partial Dijfere11tial Equations Chap. 2
This agrees with the result obtained when the method of Section 5 is used. In
fact, v,e notice that the differential operator corresponding to the given differential
equation has the factorization
(D~ + 6DxD,. - 16D~) = (D, - 2D,.)(Dx + SD,)
Let us observe that the curves

~(x, y) = constant, ry(x, y) = constant
where the functions ~ and 17 are defined in (7.4), are integral curves of the
ordinary differential equations
dy dy
(7.8) -1111, = -m2
dx dx
respectively. They actually define two one-parameter families of straight lines,
which are called the characteristics of the second-order equation (6.1 ). We
notice that the equations in (7.8) can be combined into the single equation
2
A
(dy)
-
dx
- 2B (dy)
-
.dx
+ C = 0
or
(7.9) A dy 2 - 2B dx dy + C dx 2 = 0
This equation is called the characteristic equation of (6.1 ). Thus, when equation
(6. l) is of the hyperbolic type, its characteristic equation (7.9) has two families of
characteristic lines. These characteristics play a fundamental role in the con-
sideration of initial value or boundary value problems for partial differential
equations of the hyperbolic type.
(ii) Parabolic Type, B 2 - AC = 0. In this case, it is clear that the coefficients
A and C cannot be both zero. Suppose that A i= O; the case C = 0 can be
treated in similar manner. Then the quadratic equation (7.3) has only one
distinct root, m = - Bf A. If we choose y = B, <5 = - A, and let 'Y. and f3 be
any numbers such that rJ.b - [Jy i= 0-say, 'Y. = 1 and f3 = 0-then from (6.4)
it follows that a = A, b = 0, and c = 0. Here we have used the fact that
B 2 - AC = 0. Thus, in the parabolic case, the transformation
(7.10) ~ = x, 11 =Bx - Ay
reduces equation (6.1) to the form
A --
a1 w + = 0
a~2
which results in the canonical form (6.6ii) when divided out by the coefficient
A i= 0.
[n terms of the characteristic equation (7.9), we conclude that in the parabolic
case, there is only one family of characteristics given by Bx - Ay = constant.
Example 7.2. Reduce the partial differential equation
U<X + 21/X)' + ll>'Y = 0

to its canonical form and obtain its general solution.
Sol111io11: Here A = I, B = I, and C = I, so that B 2 - AC= 0. Thus. the given
equation is of the parabolic type. By formula (7.10), we make the change of
variables: ~ = x, 17 = x - y. Then
llx = II;;+ 11,1
lly = -11,,
u'" = us~ + 211,,1 + 11,1'1
llxy = - u~,, - u,,,.,
and thus the differential equation reduces to the canonical form
,,
II" = 0
The general solution of this equation is
Hence, the general solution of the given differential equation is

11(x. y) = xf(x - y) + g(x - y)
Again, this result can also be obtained by the method of Section 5.
(iii) E!liptiC', 8 2 - AC < 0. Here it is clear that neither A nor C can be

zero. Following the discussion in the hyperbolic case, we see that since
B 2 - AC < 0, the quadratic equation (7.3) has complex roots
-B + i(AC -B - i(AC
111 I = 1112
A A
2 112
The choice ry_ = - 8 + i(AC - 8 ) , fj = <5 = A, and /' ii. would then
make the coefficients a and c of equation (6.3) \'anish, with b 2A(AC - B 2 ) >
0. Thus, under the change of variables
[-B + i(AC 82)1;2Jx + Ay

(7.Jl)
Y/ [ -8 i(AC B2)1 2Jx + Ay
equation (6.1) would be reduced to the form
iJ 2 H 1
(7.12) 2b - - + ... = 0
8 OYf
However, from (7.11) we see that~ and 1J are now complex variables such that
~ = 11. In order that we may have a canonical form in real \'ariables, we further
make the change of variables
(7.13) ~' = ~ + Y/
2
Then. as is easily \eritled.
so that ( 7. J 2) reduces to the canonical form
(7.14)
after dividing through by A(AC - 8 2 ).

If we substitule ~and 11 from (7. I I) in (7. I 3). we see that
.;: = -Bx + Ay
(7.15)
1( = (AC - B 2 ) 112 x
Writing ~ and 11 again in place of ~' and 1( the transformations (7.15) corre-
spond to the choice CJ. = -B. f3 = A,/' = (AC - B 2 ) 112 , and <5 = 0 in (6.2).
With this choice, it is easily verified from (6.4) that a = c = A(CA - B 2 ) and
b = 0. Therefore. under the change of variables
c; -Bx + Ay
(7.16)
I/
equation (61) can be reduced to the desired canonical form (6.6iii) in the
elliptic case.
We notice that since B 2 - AC < 0. equation (7.9) has no real integral
cunes. This means that an equation (6.1) of the elliptic type has no real
characteristic.
Example 7.3. Reduce the elliptic partial difTerential equation
to its canonical form.
So/11tio11: We have A = I. B = -1. and C = 5 so that B 2 - AC= -4 < 0.

Making the changes of variables
= x + y, I/= 2x
\\eseethat
u, //, + 2u, 1
11,. //,
llxx = II~~ + 4u~, 1 + 411,,,,

llxy I I,,
" + 211~,,
Uyy 11~~
Substituting these in the differential equation, \\e have the desired canonical form
The general solution of this last equation is of the form
II= JW + g((}
involving arbitrary functions of a complex variable(, where ( = + i11. Thus,

the given differential equation in the variables x, y has the general solution
u = /[(x + y) + 2ix] + g [(x + y) - 2ix]
Exercises 2. 7
Reduce the following hyperbolic partial differential equations to the canonical

forms (6.1 i) and (7.6).
I. 3u""+ 2u<>. - 511)')' + "x = 0.
2. + 2u",. + llx + LI y = X)'.
11 rr
3. 811xx + 2u". - 311)'}' - X//r = 0.

0
4. 3uu + - 211)')'
UX}' II = c
5. 2llxx 3uq + "n + u" - "r = I.
Reduce the follm1 ing parabolic partial differential equations to canonical forms.
6. 4uxx - + l/J'J llx. = 0.
4u".v
7. llxx - 61/X)' + 911)')' + II; = xy.
8. 9uu + 6u,.,. + ii~._\' II 0.
9. llxx + 2u,,. + lln: - 411 0.
Reduce the follov,ing elliptic partial differential equations to the canonical form
(7.14).
10. Uu + + II = 0.
411n
1J. 211" - 2u, . . + 511rr - Xllr = 0.

12. lln + 2u,,. + 211yy - yu, = 0.
13. 3uu Su<>. + 611n, II = 0.
14. 411_"" 0.
sin y.
In Problems 16 through 19, find the general solution of the differential equation by
first reducing the equation to its canonical form.
16. 3uu Uxy - 2uH = 0.
17. 5u'-' + 2uxy + 2uJ'Y = 0.
18. Uxx + 4uxy + 411}'}' 0.
19. ll:cr: 4u,Y + 511)')' 0.
20. Consider the equation of the hyperbolic type
llxx - 11,.J' + Gllx + bu,. + Cit 0

\\here a, b. and c are constants. Shm\ that by introducing the new variable 1.
defined by r(x. y) = ex-. P)u(x, y), the equation can be put in the form
l"xx - Z\y + 'll = 0

where i-' = (b 2 - a2 + 4c)/4.
21. Consider the differential equation of the parabolic type
11 1 - Gllxx - bu, - Cll = 0
where a. b. and c are constants. Show that by introducing the new variable 1,
defined by r(x, t) = e'x'P 111(x, t), the equation can be put in the canonical form
1' 1 - ktxx = 0.
Chapter 3
The Wave Equation
Jn this chapter v.e shall study certain types of problems that are generally
associated with linear hyperbolic partial differential equations. We shall
consider these problems in connection with the equation
-2 ,z
c u - c2 o u = F( x t)
ot 2 ox 2 ,
where x and t are the two independent variables and c is a constant. This
equation, called the wave equation. serves as the prototype for a class of hyper-
bolic differential equations involving two independent variables. It arises in the
<otudy of many important physical problems involving wave propagation
phenomena, such as the transverse vibrations of an elastic string, and the
longitudinal vibrations or the torsional oscillations of a rod. The wave equation
i' certainly one of the most important classical equations of mathematical
physics.
J. The Vibrating String
We consider a uniform string of length L and constant density p, which is

stretched between two supports. The string is assumed to rest on the interval
[O. L J of the x-axis. with its ends fixed at x = 0 and x = L. Each point of the
string is thus determined initially by its coordinate x. If the string is displaced
from its resting position and then released, it will vibrate. Let u(x, 1) denote the
displacement of a given point x at any subsequent instant of time t. We assume
that other conditions of the string are such that only transverse vibration
71
72 The JVave Equation Chap. 3
(vertical motion) takes place in the xu-plane. The problem, then, is to determine
u(x, t) for 0 < x < L and I > 0.
We flrst show that under certain simplifying assumptions the displacement 11
satisfies the \\a\e equation. \Ve shall assume that the tensile force in the string is
very large so that the weight of the string can be ignored. Further, we shall
assume that the string is perfectly elastic and offers no resistance to bending;
this implies that the tensile force at any point of the string acts in a direction
tangential lo the profile of the string. Finally, \Ve shall assume that the string
has only small trans\erse \ibrations so that the slope 11" at any point of the
displaced string is small; in particular, we shall take u;
to be negligible compared
with unity.
Now Jet us consider an arbitrary segment AB of the string with length ~s
(see Fig. 3.1 ). Let x and x + ~x be the coordinates of the points A and B. and
let T 1 and 7~ be the tensile forces acting at these points, respectively. These
forces are tangential to the shape of the string at the points A and B, as we have
assumed abo\e. Since we ha\e only vertical motion, the algebraic sum of the
horizontal components of the tensile force must be equal to zero. Thus, we ha\e
where et. 1 and et. 2 are the acute angles between the tangents at A and B and the
x-axis. By the assumption on 11;.
we have
cos Cl.1
tan 2 172 I
2 1/2 I =
I
(I + et.i) (I + uJ . "
cos Cl.2
lan2 I 2 1;2 I
(I + et.2) i ;2 (I + ux) x+.6.x
B
I
I
I
I
I
I
I
: ll(X + 6.c. I)
I
I
---=o:-1'----------~x-----_,~.-+_,6-_,-.-------~L'-----'
FIG. 3-1 Vibrating string.

Sec. I The Vibrating String 73
and therefore
T 1 = T2 = T = const
That is, the tension in the string is constant.
The \ertical forces acting on the segment AB consist of T sin 'Y. 2 , \\hi ch is
directed upward. and T sin 'Y. 1 directed do\\ 11\\ard. Hence. the resultant
force is
R = Tsin 'Y. 2 - Tsin 'Y. 1
Since the \ibrations of the string are assumed to be small. \\e ma} replace
sin 'Y.; by tan 'Y.,. i = 1. 2, so that \\e ha\e
R = T(tan 'Y. 2 - tan 'Y. 1 ) = T f-~ 11 (x + Lix, t) - ~ 11 (x,1)1

lex ex
From the fact that

L'ls = (1 + 11~) 1 2
Lix ~ Lix
the mass of the segment AB is approximately (Lix)p. Therefore. in accordance

with Newton's second law of motion (mass times acceleration equals force). we
ha\e
~ t (,
11
(1.1) p(Lix) 1) = T f~u (x + Lix, t) - ~ 11 (x, 1)1
01- lex cx .J
\\here x < < x + Lix. Oi\iding both sides of equation (1.1) by Lix and
taking the limit as Lix tends to zero, we obtain
( 1.2) 0
\\here c = (T/p) 1 2 is a pos1t1\e constant. Equation (1.2) is called the one-

dirnensional wa\e equation. It is a hyperbolic partial differential equation with
constant coefficients. The constant c has the dimension of a \elocity, as the
reader can easily \erify by substituting appropriate units to T and p. This
constant represents the maximum \elocity of propagation of a disturbance in the
string, as we shall see later.
If there is an external \ertical force~sayJ(x. 1) per unit length~acting on the
'tring, then the terrnf(x. I) Lix must be added to the right-hand side of equation
(1. l ). In passing to the limit, we obtain the equation
o~z u ( ) '' (
, 0-11 f (x. I )
( 1.3) , x,t -c-~ x,t)= F(x. I)
a1- ox 1 p
This is called the nonhomogeneous wave equation.

Now, in order to determine completely the subsequent motion of the string,
it is necessary that we know initially the position or shape of the string, the
velocity with which the string is set into vibration, and the conditions at the ends
74 T/1e Wave Equation Chap. 3
of the string. Mathematically. this means that we must specify 11(x. t) and
cu/ct(x. t) at the initial time, say. I = 0: that is.
(1.4) 11(x. 0) = f(x), u,(x, 0) = g(x)
where/ and g are some given functions defined for 0::;, x :S L. Further. since
the ends of the string are fixed at x = 0 and x = L, the displacement function
at these points must ah\ ays be equal to zero: hence,
(1.5) 11(0. t) = 0, u( L. t) = 0 (I ~ 0)
It should be noted that physical considerations dictate that f(O) = g(O) and
f(L) = g(L) = 0.
\\'e call the conditions ( 1.4) the initial conditions, and the conditions ( 1.5),
the boundary conditions. In particular, the functions/and gin (1.4) are called
the initial data.
Therefore, to obtain the displacement of a vibrating string with fixed ends. we
must find a solution of the wave equation (1.2). or (1.3) if there is a forcing term
F. that satisfies the initial conditions ( 1.4) and the boundary conditions (I .5).
This type of problem is called an initial-boundary \'alue problem.
Of course, if the ends of the string are not fixed but are allowed to move
subject to some constraints. other boundary conditions arise. For example, if
the ends of the string are allowed to mo\'e vertically without a restraining force,
then the displacement function must always satisfy the conditions
(1.6) u,JO. 1) = 0. uJL, t) = 0
which are called free boundary conditions. On the other hand, if the ends are
attached to a spring, the boundary conditions assume the forms
(1.7) 11)0. I) + /111(0. 1) = 0, u,(L. 1) + hu(L. 1) = 0
where h is a constant. These are called elastic boundary conditions. The

boundary conditions ( 1.5). ( 1.6). and ( 1.7) are often referred to as boundary
conditions of the first. second, and third kind. respecti\ely.
In contrast to an initial-boundary value problem consisting of the wave
equation ( 1.2) or ( 1.3). the initial conditions (1.4), and one of the boundary
conditions ( 1.5). ( 1.6). and ( 1.7) in which the variable xis restricted to the interval
[O. L], we may seek a solution of the wave equation in the infinite domain
- :x < x < x. r > 0, \\hich satisfies only the initial conditions (1.4) given
for - oc < x < CD. Such a problem is simply called an initial value problem.
Ph:Jsically, an initial \'alue problem for the wa\e equation corresponds to the
problem of a \ ibrating string that is so Jong that the effect of its boundary
conditions can be neglected. As we shall see in the discussions to follow, our
solution formula for an initial value problem can be used to find a solution of an
initial-boundary \ alue problem with boundary conditions of type ( 1.5) or ( 1.6)
at either end point.
Sec. I The Vibrating String 75
Exercises 3.1
1. Let "and 1 be two solutions of the wave equation
(ix < oo)

each satisfying the corresponding initial conditions
u(x. 0) = f(x), u,(x, 0) = O; r(x. 0) = 0, v,(x, 0) = g(x)
Set 11 = 11 + 1 and shO\\ that u satisfies the \\ave equation Lii' = 0 together
\\ ith the initial conditions
w(x, 0) = f(x). w,(x, 0) = g(x)
2. Consider the initial-boundary value proble!ll
(0 < x < L, t > 0)

ll(X, 0) = f(x), ll 1 (x, 0) = g(x) (0 s x s L)
!1(0, t) = a(t), 11(L, I) = b(f) (1 ~ 0)
Let 1 be a t\1 ice-differentiable function satisfying the conditions
r(O, t) = a(t), 1(L, t) = b(t)
Deterllline the differential equation and the initial and boundary conditions
satisfied by w. where w = 11 - u.
3. Suppose that the initial value problelll
2
Lu= llrr llxx + C 11 = 0 (x' < CXl)
11(x. 0) 0 (jxl < CXl)
11,(x, 0) = 0 (jxj < CXl)
is known to have only the trivial solution " = 0. Prove that the problem
Lu= F, ll(X, 0) = f(x), ll 1(x, 0) = g(x) C.< < 00)

has at most one solution.
4. Verify that the function 11(x, I) = sin 2x cos 21 satisfies the equation 11 11 - uu =
0 and the initial conditions 11(x. 0) = 2 sin x cos x. and 11,(x, 0) = 0.
5. Verify that the function 11(x, I) = sin 2x sin 21 satisfies the wave equation of
Problem 4 and the initial conditions 11(x, 0) = 0, 11/x, 0) = 2 sin 2x.
6. Show that 11(x, 1) = t sin nx is a solution of the initial-boundary value problem
2
"" - llxx = 11 1 sin nx (0 < x < 1, t > 0)
11(x, 0) = 0, 11,(x, 0) = sin nx (0 s x s 1)
11(0, t) = 0, 11(1, I)= 0 (t ~ 0)
7. Show that 11(x, I) xt 2 + sin(x + I) is a solution of the initial value problem
llrr - llxx. = 2x (jxj < CXl)

11(x, 0) = sin x <Ix: < CXl)
11,(x, 0) = cos x (jx! < CXl)
'
76 The IVaYe Equation Chap. 3
8. Show that 11(x. r) (ex-i + e-'-')12 is a solution of the initial-boundary value

problem
(x > 0)
u(x, 0) = ex (x 2 0)
11,(x, 0) = 0 (x 2 0)
u(O, r) = cash t (r 2 0)
9. Show that u(x. I) = cos x sin t is a solution of the initial-boundary value

problem
ll11 - llxx = 0 (x > 0)
11(x. 0) = 0 (x > 0)
u,(x, 0) = cos x (x > OJ
uJO, r) = 0 (r 2 0)
10. Let
/z (r _') + f(x + ct) -

2
f (ct - x)
(0 :::: x < ct)
u(x, I)
{ j (x - ~) + f (x + cf)
(ct < x)
2
\\here h and fare twice continuously differentiable for x > 0. Show that 11
satisfies
(x > 0. t > 0)
u(x, 0) = f(x). 11 1 (x, 0) = 0 (x > 0)
11(0, I)= h(t) (t > OJ
1I. Let
( - c r-xic h(s) ds (0 :::: x < cf)

u(x. t) =
l 0
0
(ct < x)
Sho\\ that 11 satisfies

(x > 0, I > 0)
11(x. 0) = 0, 11,(x, 0) = 0 (x > 0)
u,(0. t) = h(f) (t > 0)
2. The Initial Value Problem
\Ve are concerned with determining a solution of the homogeneous wave

equation
iJ2u
(2.1) - -
c2 0 (- 00 < x < CIJ, t > 0)
21 2 <;x2
Sec. 2 The Initial Value Problem 77
satisfying the initial conditions
(2.2) ll(X, 0) = f(x). 11J:c 0) = g(x) (-x<x<a::;)

This problem is one special case where the general solution of the differential
equation actually leads to a solution of the problem.
Since the equation is hyperbolic we introduce the new \'ariables ::. IJ defined by
(2.3) ~ = x + cl. I/ = X - cl
and set u(x. r) 1r(, 17). Then equation (2.1) reduces to the canonical form
,J
c-11
(2.4) = 0
U, c17
which has the general solution
irk, 11) = F() + C(11)

Thus, the general solution of (2.1) is of the form
(2.5) u(x, 1) = F(x + cl) + G(x - ct)
where Fand Gare arbitrary functions of one variable and are twice differentiable.
Now we determine the functions F and G so that (2.5) may satisfy (2.2).
Setting t = 0 in (2.5) and using the \'alue of u(x, 0) from (2.2). we have
(2.6) f(x) = F(x) + G(x)
Let us differentiate (2.5) with respect tot to obtain
11,(x, r) = cF'(.Y + er) - cG'(x - ct)
Here the primes indicate differentiation with respect to the argument of the
frnction. If we set t = 0 and use the value of u,(x, 0) from (2.2), we find
(2.7) g(x) = cF'(x) - cG'(x)
The system of equations consisting of (2.6) and (2.7) enables us to determine F

and Gin terms of/and g. Indeed, by integrating (2.7) from 0 to any point x, we
obtain
F(x) - G(x) = J lx g(s) ds + C

c 0
where C is a constant of integration. Combining this result with (2.6) we obtain
1
F(x) = l f(x) + (' g(s) ds + C
2 2c Jo 2
Therefore, from (2.6), we ha\'e
G(x)

= -1 f(x) -
2
1-
2c
J"o g(s) ds c
2
78 The Wave Equation Chap. 3
If we replace the \ariable x in the expressions for F and G by x + ct and

x - ct, respectively, and substitute the results in (2.5), we nnally arrive at the
formula
(2.8) u(x, t) =
/(x - ct) + f(x

+- ct)
- +
l J,x+cr g(s) ds
2 2c x-cr
This is known as d'Alemberfs formula for the solution of the initial value
problem (2. I), (2.2).
It can be easily verined by direct differentiation (Exercises I .4, Problem 9)
that when f has a continuous second-order derivative and g has a continuous
first-order derivative, the function (2.8) is twice continuously differentiable and
satisfies (2.1) and (2.2). As a matter of fact, our derivation of d' Alem berfs
formula shows that any solution of (2.1 ), (2.2) that is twice continuously
differentiable must have the representation (2.8); hence, the solution is uniquely
determined by the initial data/andg. Thus, d'Alembert's formula represents the
unique solution of (2.1 ), (2.2). Formula (2.8) also shows that if small perturba-
tions are made in the initial data in an arbitrary fixed time interval 0 :S t :S
T < oo, then the resulting perturbation in the solution is also small; that is, 11
depends continuously on the initial data f and g. More precisely, suppose 11 1
is the solution of the problem (2.1 ), (2.2) corresponding to initial data / 1 , g 1 ,
and 11 2 is the solution corresponding to initial data f~, g 2 Then, for a given
c > 0, we can find a number b > 0 such that
whenever
l/1 (x) - / 2 (x)I < b,

for all x and 0 :S t :S T. Indeed, using the representation for u 1 and u 2 , we see
that
1
lu 1(x, t) - u,(x,
- t)I :S 2 l/Jx
. - ct) - f~(x - ct)I
l
+ - l/1 (x + ct) - f,(x + ct)!
2 -
] J,x+cr
+ c x-cr IB1(s) - 92(s)I ds
2
1 l
:S - (b + b) + 2ctb < (1 + T)b
2 2c
which can be made less than c if we take b = s/(l + T).
A problem that has a unique solution which depends continuously on the
initial data or boundary conditions is said to be well posed or correctly set. Th us,
the initial value problem (2.1 ), (2.2) is well posed. It is clear that well-posed
problems are meaningful in applications, since the initial or boundary data are
usually known only approximately; thus the corresponding solution represents

only an approximation to the exact solution of the problem. It is by no means
true, however, that an initial value or boundary value problem involving a
second-order partial differential equation is always well posed. In fact, whether
or not a mathematical problem is well posed depends on the 'kind of auxiliary
condition (initial or boundary, or both) imposed and the type of the equation
and domain considered. For instance, we have just seen that an initial value
problem is well posed for the hyperbolic wave equation, but (as we shall see
later) it does not lead to a well-posed problem for Laplace's elliptic equation.
Jn the light of this fact, our classification of second-order partial differential
equations into hyperbolic, parabolic, and elliptic types is all the more
significant and essential.
Returning to the d'Alembert"s formula (2.8), we note that if we assume only
thatfis continuous andf' and g are piecewise smooth, then there will be points
(x, t) where the first- or second-order derivatives of 11 may fail to exist. In such
a case, the function (2.8) cannot satisfy the wa\e equation, and therefore it
cannot be a solution of the problem (2. I), (2.2). However, except only those
points (x, I), the function 11 may still satisfy equation (2. I) and initial conditions
(2.2). In this case, we shall regard (2.8) as the generalized solution of the
problem. By this notion we really mean that u is the limit of a uniformly con-
vergent sequence of solutions u,,(x, t) of(2.I) satisfying the initial conditions
cu,.(x, 0) ( )
- = g,, x
cI
where the sequence of functions(,, have continuous second-order derivative and
the sequence of functions g 11 have continuous first-order derivative such that
Jim f,, = f; Jim g,, = g

11--+X
uniformly for - oo < x < oo. This concept will be further illuminated in
Section 9 of the present chapter.
Example 2.1. Find the solution of the wave equation (2.1) satisfying the initial
conditions u(x, 0) = sin x, 11,(x, 0) = 0, - x;; < x < x.
So/111io11: By the d'Alembert's formula (2.8) we have
u(x, t) = sin(x ~ ct_)_:J- sin(x + ct)

2
= sin x cos ct
It is easily seen that this function indeed satisfies the wave equation and the
given initial conditions.
Example 2.2. Find the solution of the problem (2.1 ), (2.2) when
u(x, 0) = 0 and 11,(x, 0) = sin 2x

80 The U'ave Equation Chap. 3
So/111io11: Again by d"Alembert"s formula, the solution is
u(x. t) =
l
~
lx+cr sm. 2s ds
-C x-ct
sin 2x sin 2ct

2c
as can be easily verified.
As examples in which the initial conditions are not differentiable, let us

consider the next two problems.
Example 2.3. Find the generalized solution of the problem
llu - 4un 0 (- oc < x < oc, t > 0)

(0 $ x <: l:l
u(x, 0) (:, - x) (l $ x :s I)
0 (othemise)
11,(x. 0) 0 ( - er: < x < r.ic)
and evaluate 11(!. lJ and 11(j. ~ ).
So/11tio11: The generalized solution is given by

((.\ - 21) T f(X + 2f)
ll(X. f) =
2
where/(.\) = ~x for 0 s x s l./(x) )(l - .\) for } s x :S l. and /(.1) = 0

for x s 0 and x 2: l. It is clear that except for x and t, such that x - 2t or
x + 21 is equal to 0 or l or I. 11 satisfies the \\ave equation and the initial condi-
tions. In particular. at (I, I) we find
f(x -- 21) o.u = / ( - ! ) = O. f(x + 21) j tl,IJ = /(3) = 0
and thus 11(!, I l = 0.

At(!,, ~) \\e have
f(x - 21) I(~.;)= f(li f(x + 2n I <+.:J = IW

and therefore 11( l. D = ~.
Example 2.4. Find the generalized solution of the problem
0 ( - X < x < r:J:, I > 0)

u(x, 0) 0 (-XJ<x<'D)
u,(x, 0) = g(x)
where
g(x)
Isin rcx (0 $ x $ l)
10 (otherwise)
and evaluate u(t t) and u(i, l ).

Sol11tio11: The generalized solution is given by
u(x, t) =
I
)
lx+1 q(z) dz
,
- x-t
and this satisfies the differential equation at all points except at those for \vhich
x - t or x + r equals 0 or 1, where the second derivatives of 11 fail to exist. At
(-}. tl we have
1- q(z) dz = I Ji sin nz dz =
2 J--} ..,
- 0 7[
At(~, }J \\e have
1
2
JJJ .
g(z) dz = I
2
J J
1
sin nz dz = 3
4n
' '
Exercises 3.2
1. The displacement of a string is given by the fonrnrd wave
u(x, f) = sin(.y - cl)
Find the initial displacement and velocity. By substituting your initial conditions
in the d'Alembert's formula, verify that u(x, f) = sin(x - ct).
2. Do as in Problem I when the displacement is given by the backward wave
11(x, t) = (x + cf) 2
3. Find the solution of the initial value problem
0 < oc. f > 0)

u(x. 0) 2 sin x cos x (]x] < oo)
11,(x, 0) cos x (]xi < oo)
11 11 - 9uu = 0 (]xi < oo, t > 0)

u(x, 0) = x sin x Cx! < oc)
11,(x. 0) = cos 2x (]x] < oo)
In Problems 5 to 8, find the solution of the initial value problem (2.1 ), (2.2) with the
giYcn initial conditions. Take c = 1.
5. f(x) = I /(I + x 2 ), g(x) = ex; ]x] < oo.
6. f(x) = e-x, g(x) = 1/(1 + x 2 ); !x' < oo.
7. f(x) = cos(n/2)x, g(x) = sinh ax; ]x] < oc.
8. f(x) = sin 3x, g(x) = sin 2x - sin x; ]x] < oo.
82 The Wave Eq11atio11 Chap. 3
Jn Problems 9 to 12 find the generalized solution of the initial value problem (2.1),
(2.2), (c = 1) \\ ith the given initial conditions, and evaluate the solution at the given
points.
9. At(~. Ji) and({. 1) with
11 ( x' -o; 1)
f(x) = l0 , (Xi > 1)
q(x) = 0 Cx! < cc)
10. At (-n/12, 7nil2) and (-n/2. 5;r/6) with
(sin x (1x -o; n)

/(x)
10 (x > n)
q(x) 0 (xi<oo)
11. At (0, ~),(-},ii, and 11(~. ~)with
( x(1 - x) (0 $ x $ 1)
f (x)
10 (othernise)
f1 x2 -<:; 1)
g(x) =
\0 (x > 1)
12. At (-n,'6, Sn/6) and (3n/S, Sn:/8) \\ith
f(x) (Ix: < cc)
g(x)
f cos x ( x! -<:; n/2)
10 (xi > n/2)
13. Let 11(x. t) be thrice continuously differentiable solution of the initial value
problem
(,x, < x., t > 0)
u(x, 0) = 0 11,(x, 0) = q(x) ( x < Y.)

where g is twice continuously differentiable. Set t"(x. t) = 11,(x, t ). Show that
dx. t) satisfies the same differential equation and the initial conditions r(x, OJ
g(x), 1,(x, 0) = 0.
14. Suppose that 11(x, t) satisfies the ''damped ha\e equation'
(.\ < X.)
together with the initial conditions u(.\, OJ = 0. 11,(x, 0) = g(r). Proceeding as

in Problem I 3, verify that L(x, t J 11,(x, t) satisfies the same equation and the
conditions r(x, OJ = g(x), r,(x. 0) = 0.
3. Interpretation of the Solution
Let us examine the solution (2.8) in more detail to get a clearer idea of the
behavior of the solution of the wave equation. Suppose that the string is
released with zero velocity after being given an initial displacement defined by
f(x). According to (2.8), the displacement of a point x at any time t is
(3.1)
f(x
u(x,t)='
ct) + f(x + ct)
2
Sec. 3 Interpretation of the So/11tio11 83
Consider the functionf(x - er). We observe that the graph off(x ct) is the
same as the graph of /(x) translated to the right by a distance equal to ct (Fig.
3.2). This means that as t increases,f(x - ct) represents a wa\ e of the formf(x)
traveling to the right with velocity c. We call the wave represented byf(x - ct)
a forward wave. Similarly, the functionf(x + ct) can be interpreted as repre-
senting a wave with shapef(x) traveling to the left with velocity c. This wave is
called a backward wave. With this interpretation. we see that the solution (3.1)
is a superposition of forward and backward waves traveling with the same
velocity c and having the shape of the initial profilef(x) with half the amplitude.
II
FIG. 3-2 Trareli11g 11aL"e form.
To illustrate this, suppose that the string has an initial displacement defined by
[Fig. 3.3(a)]
x +a ( - a ::;; x ::;; 0)
f(x) = -~ + a (0 ::;; x ::;; a)
(
(otherwise)
The forward and backward waves indicated by the dotted curve in Figure 3.3(a)
coincide at t = 0. At t = a/2c, both waves have moved in opposite direction
through a distance a/2, resulting in the shape of the string shown in Figure 3.3(b).
At t = a/c, the forward and backward waves are on the verge of separating
from each other. For t > a/c, the motion of the string consists of the forward
and backward waves traveling toward the ends of the string at the same velocity,
......
JI
84 The ~ave Equation Chap. 3
II
( t') { -:::: ~{/ (

~I
I ii\ a1
I~ I
\ I I I
I I I
I I
\ \
I I
I \ I
I \ I
\ \ I
I \ I
\ I I
tdJ r=3a~( I
\
I
I
I
I
I
\
I
I
fhl i a _(
la I I~ II a (J a
FIG. 3-3 The propaqation of wa1:e.1 due to initial displacement.
Figure 3.3(d) and (e). It is seen from the figure that each point of the string
returns to its original position of rest after the passage of each wave.
Jn the case when u(x, 0) = 0 and u,(x, 0) = g(x), the displacement is given by
(3.2) u(x, t) =
1
.
rx+cl g(s) ds
2L x-cr
Let us define
(3.3) <fl(x)
2
c f g(s) ds
Then (3.2) can be written as
(3.4) u(x, t) = -<fl(x - ct) + <fl(x + ct)

Sec. 3 Interpretation of the Solution 85
which again shO\vs that the solution is a superposition of a forward \\ave

-r/J(X - ct) and a backward wa\e </J(x + ct) traveling with the same velocity c.
The forms of these waves are determined by the function . which is related to
the initial velocity function g through the integral (3.3).
In illustration, suppose that
g(x)
pc (lxl < l)
10 (lxl > I)
Then
-cp(x - ct) - l fx-ct g(z) dz

2c o
-(x-ct) (Ix - < l)

ctl
J-1 (x - ct > l)
1 I (x - ct < -1)
and
rjJ(x-1-cl) r-+c1 g(z) d:

2c Jo
J~ + ct
(Ix + < l)
ctl
> [)
(x -1- Cl
l-1 (x + ct < - I)
fhe graphs of -qJ(X - ct) and q;(x + ct) together with the resulting shape of
the string for various values ol tare sho\\11 in Figure 3.4.
At t = 0, the forward wa\e -</J(x) and the backward \Va\e qJ(x) nullify each
other so that the string has zero displacement: that is, it is at rest. At t = -}c_
the forward and the backward \\a\es have mO\ed in opposite direction through a
di~tance} unit. The shape of the string at this instant is obtained hy adding
graphically the two wa\e forms. Alt = l/c. the point x = 0 allains its maxi-
mum displacement 11 = 2, and from this instant on, more and more points of the
string assume this maximum displacement and remain al rest in that position.
Thi.;; is evident from (3.2). since for a gi\en point x of the string.
j
2c dz = 2
2c f - I
as soon as t reaches a value for \vhich x - ct < - I and x + ct > I.

lul I= ~ / 0 '"\ ~
/
',
- ___________ / /
/
' '------------
- 0( \" ]/: l
II
( 1..: I f:..::::
1
,/C // I 0 I',
/
',
- -- - - - - - - - __ /
/
' ' - - - - - - - - - ---
qi(.\ 11
(1)\ .Y) / ---------------------

<Pl,.,
/
/
/
(a I I= 0 /
',,
///0
',
------------;;~~----------' I '----------------------
Isl
FIG. 3-4 Propaqario11 of ll'ares due to initial re/ocity.
4. Domain of Dependence and Characteristic Lines
We saw in the preceding section that the solution of the wave equation
consists of forward and backward waves. The forms of these waves can be
illustrated graphically for simple functions/ and gin the xu-plane. In order to
see how these wa\es are propagated, we shall now consider the xi-plane in which
each point can be taken to correspond to a definite position on the string at a
particular time.
Sec. 4 Domain of Dependence and Characteristic Lines 87
Let (x, t) be a point in the xi-plane. From formula (2.8) we see that the
solution at (x, t) depends only on the value off at the two points x - ct and
x + ct, and on the initial velocity g over the interval [x - ct, x + ct]. This
means that a change in the initial position and \e]ocity of the string at points
outside the interval [x - ct, x + ct] will have no effect on the displacement
at (x, t). Consequently, if/and g are known to vanish identically on an interval
a ::s; x ::s; b, then u(x, t) = 0 for all points (x, t) such that b < x - ct or
x + ct < a.
The interval [x - ct, x + ct] on which the value of u(x, t) depends is called
the domain of dependence of the point (x, t).
It is easy to see that the domain of dependence of a given point (, r) is
precisely the interval cut out of the x-axis by the two straight lines passing
through the point with slopes I/c (Fig. 3.5). These lines, with the equations
(4.1) X - Cf = ( - CT, x + ct = ( + er
are the characteristics (lines) of the wave equation (I .2). The triangle formed
by the characteristics and the interval of dependence of a given point (x, t)
will be called the characteristic triangle determined by (x, t).
( ~. r I
{ Cl = ~ - CT ,+c!=~+cr
---+----~-------------~----{
0 ((-cr.0) ((+n.OJ
FIG. 3-5 Characteristic triangle.
Now let us see the significance of the characteristics. We note that d'Alem-
bert's formula, which gives the displacement of a point((, r), can be written in
the form
(4.2) u(~, r) (~ - er) + If;(( + er)
where
(4.3) ( - er) l 5~-n g(z) dz

J f(t," - er) - --
2 2c o
and
(4.4) 1 /(( + er) + 1 r~+cr g(z) dz
2 2c .lo
Thus, along the line x - ct = ( - er, cp(x - ct) has constant value, namely,
cf>(( - er). This means that points on the characteristic x - ct = ~ - er have
the same displacement caused by the forward wave cf>(x - ct). Likewise,
t/!(x + cl) is constant along the line x + ct = ( + cr so that points on this
line have the same displacement due to the backward wave 1f!(x + ct). It is in
this sense and for this reason that we say the disturbance on the string is
propagated with velocity c along the characteristics.
In Figure 3.6 we illustrate graphically the concept of propagation of waves
along the characteristics. We consider a string that is initially at rest, with
isolated initial disturbance at the points ~ - cr and ~ + cr. According to our
discussion in Section 3, the disturbances at these points will each split into two
identical waves, the forward wave and the backward wave, both having the
same form as the original disturbance but with half the amplitude. Each of the
forward waves will tra,el along the characteristic
x - ct = const
that passes through the point from which the wave originated. Thus, referring
to the figure, points on the characteristics
X - Ct = ( - CT, x - ct ~ + CT
I
Y+cr=~ -(r\
rl
Y- u=~+cr
FIG. 3-6 Propagation of wai-es along characteristics.

Sec. 4 Domain of Dependence and Characteristic Li11c.1 89
experience the same displacement due to the fon\ard \\a\es originating from
the points - er and + cr, respectively. The same discussion can be made
for the back\rnrd \\a\es and the points on the characteristics
X + Cl = - CT. X + Cl = i Ct
through the respecti\e points - n and + n. Thus. the displacement at

the point (:. r) is the superposition of the forward ''a\e from - n and the
bach,ard wa\e from + n.
In Figure 3.7 we show how the other points of the \'I-plane are affected by a
disturbance confined to an interval [x 1 x 2 ]. \Ve obscne that the characteristics
through the points x 1 and x 2 divide the xt-plane into six regions. indicated in the
figure. For points in regions Y and YI. their domains of dependence nc\cr
o\erlap the intenal [x 1 xJ. This means that the points on the string that
correspond to the points in these regions are not afTected by the initial distur-
bance at the corresponding times. In region I the points are affected by both
fornard and backward wa\es originating from points in the interval [x 1 xJ.
In region II the points are affected only by backward waves, and in Ill only
by forward wa\"es. For points in region IV, the corresponding times are such
that the corresponding points on the string hme already experienced the passage
of the waves and thus remain at rest with a permanent displacement equal to
(see also Fig. 3.4)
2c I2 g(s) ds
X +(f = Y2 (I= Xi
FIG. 3-7 Region of" i11ft11e11ce of the interrnf [x 1 , x 1].

The region bounded by the characteristics x + ct = x 1 and x - ct = x 2

and the interval (x 1 , xJ is called the region of influence of the interval j:x 1 , x 2 ].
In particular, the region of influence of a given point consists of all points lying
above the two characteristics passing through that point.
5. The l'\onhomogeneous Wave Equation
We now consider the initial value problem
2 a2u
(5.1) c --2 = F(x, t) (t > 0)
ox
(5.2) u(x, 0) = f(x),

cu
~- (x, 0) = g(x) (-oo<x<oo)
ot
which involves the nonhomogeneous wave equation. As we noted before, this

problem corresponds to the problem of an infinite vibrating string with the
presence of an external force. Since the difference of any two solutions of (5.1)
is a solution of the homogeneous wave equation for which we have uniqueness,
it follows that the solution of (5.1) is again uniquely determined by the initial
data (5.2).
Let ((, T) be the point at which we wish to find the solution of the problem
(5.1 ), (5.2) and let i\ denote the characteristic triangle determined by this point
(see Fig. 3.5). We integrate the expression on both sides of (5.1) over the domain
i\ to obtain
(5.3) - ff F(x, t) dx dt = ff (c 2 uxx - 11 11 ) dx dt

~ ~
By Green's theorem (Section 8, Chapter I), lhe right-hand side of (5.3) can be
written as an integral around the boundary C of the domain i\ so that we have
(5.4) - ff F(x, t) dx dt = fc (u, dx + c2 ux dt)

~
The boundary C consists of segments of the characteristics L 1 : x - ct = ~ - er

and L 2 : x + ct = ( + er passing through the point ((, r), and the interval
J: [~ - CT, ( + er] that is cut off from the initial line t = 0 by the character-
istics. Now on I, t = 0 so that dt = 0, while on L 1 and L 2 we have dx = c dt
and dx = - c dt, respectively. Hence, it follows that
Sec. 5 The So11'101110!1e11eo11s ~Vave Equation 91
J (11, dx + c1 uxdt)
I
,,,~+( r
c-cr 11/x, 0) dx
r
L1
(11, dx + 2
C llx dt) c r
.., L1
(11, dt + u, dx)
c r
L1
du
(5.5)
c[u(( - CT, 0) - ll(s, T)]
lLi
(u, dx + c 2 u,. dt) -c L 2
(11 1 dt + llx dx)
-c J,, du
-c[u((, T) - 11( + CT, O)]

Substituting the initial data (5.2) into the results on the right side of (5.5), we
ha\e
[ (11 1 dx + c 2 u, dt) -2rn(, T) + cf( - er) + cf( + n)

c
+ r.~+cr g(x) dx
._,..;-er
Therefore, on substituting this result in (5.4) and solving for 11, we find
15.6) u(x, t) =
f(x - ct)+ /(x

+ ct)
+ 1 Jx+ct g(c:). de:,.
2 2c x-ct
+ , rr F(. T) d dT
2c
~
in \\ hich we have interchanged the role of 1. T) and (x, t). This represents the
unique solution of our problem (5.1 ), (5.2).
In fact, we obsene that the sum of the first two terms on the right of (5.6) is
precisely d'Alembert's solution of the homogeneous wave equation with the
initial conditions (5.2). Further, by writing the integral
(5.7) t{x, t) =
2
1
c JJF(, T) d dT
~
as an iterated double integral with appropriate limits determined from the

characteristic triangle fl. it can be verified directly that (5.7) satisfies (5.1) and
the homogeneous initial conditions
('l'
(5.8) i:(x, 0) = 0, (x, 0) 0
a1
(Exercises 3.3. Problem 10.) In' ie\1- of the linear nature of our problem. it
follo\\S from the principle of superposition that (5.6) satisfies (5.1) and (5.2).
Jn this connection we assume that f is twice continuously differentiable. g is
continuously differentiable. and F and F, are continuous for - x < x < L.
t ::0: 0.
Example 5.1. Find the solution of the equation
satisfying the homogeneous conditions 11LY. 0 I 11,(x, 0) 0.
Solution: By the formula (5.7) 11e have
11(x. I) = {' 1x+(t-r) d dr

2 Jo .x-(z-r)
4 .o I., [<x + I - T)
2
- (x -- I+ rJ 2 ] dr
1 ,., 4x(I - r) (h
4 . 0 2
It is easily seen that this function satislies the equation and the homogeneous
initial conditions.
Example 5.2. Find the solution of the initial va!ue problem
u(x. 0) x. //,(\'. 0) = 0
Sol11tio11: The solution can be obtained by the use of(5.6). However. since flx. t) = t
is independent of x. 11e readily find a particular solution of the differential equa-
tion (by integrating 11" = t Jin the func1ion
1.3
u(x. I)=
6
11hich satisfies the homogeneous initial conditions: 11 = 0 and 11 1 0 at I 0.

Hence, by superposition. the solution of the problem is
(x-11+(x+1)
u(x, I ) =
2 6
.
'
. '' 6
The reader may check that
(' r<+U-d T dC, dr = 13

2 Jo J,_U-r) . 6
As illustrated by the preceding example. the problem (5.1), (5.2) can some-
times be reduced to one inrnlving a homogeneous differential equation if we can
Sec. 5 The .Vo11ho111o!Jeneo11s Wave Equation 93
find a particular solution of' the nonhomogeneous differential equation by some

special method. This will, of course. a\oid the use of (5.6), which invohes a
term requiring double integration. For suppose 1 is a particular solution of
(5.1 ). Let 11 denote the solution of (5.1 ), (5.2) and set 11 = 11 - 1" Then. by
the principle of superposition. the function 1r satisfies the homogeneous wa\e
equation and the initial conditions
ll(X. 0) u(x, 0) - t(x, 0) = f(x) - t:(x, 0)
(5.9)
ir,(x, 0) u,(x. 0) - t",(x, 0) = g(x) - r,(x, 0)
Hence. 11 can be found by the use of d'Alembert's formula corresponding to the

ini!ial conditions (5.9). Then the solution of(5.I), (5.2) is gi\en by
u(x, t) = c(x, t) + 11(x, t)
A~ an illustration. consider again the problem in Example 5.1. A particular

solution of the equation is easily found to be 1(x, t) = -x 3 /6. Let
x3
ll(X. I)= u(x, I)+
6
Then
o. 11(x, 0) H',(x, 0) 0
6
Hence. by (2.8).
(x - 1)3 + (x + t)3
1r(x. t)
12
x3 xt 2
+
6 2
and therefore
ll(X, I)= t(X, I)+ \\'(X, t)

2
11 h1ch agrees 1\ith our earlier result.
Exercises 3.3
lil Problem-, I through 6, find the solution of each of the initial value problems on
f .1 < Cf::. t > 0. by using (5.6).
2
1. 11 11 -- llu = 2Y. 11(.1', 0) = x . 11,(x. 0) = 0.
2. 11 11 - u0 = 2x - t, u(x, 0) = sin x, u,(x, 0) = x.
3. 11 11 - u,x = .YI. u(x. 0) = 0. 11 1 ( 1. 0) = e-'.
4. 1111 411u ex, 11(x, 0) = cos x, u,(x, 0) = sin x.

5. llrr 4uu ex + sin t, 11(x, 0) = 0, u,(x, 0) = 1/(1 + x 2 ).
6. 1111 911.u = t sin x, 11(x, 0) = x 2 (1 - x). 11,(x, 0) = 0.
ln Problems 7 through 9 find the solution of each of the problems on - XJ < x < ctJ,
> 0 by finding a particular solution of the homogeneous differential equation and
reducing the problem to one involving a homogeneous equation.
7. l/ 11 - llxx = sin x. u(x, 0) = x, u,(x, 0) = 0.
8. 11 11 - llxx = x sin x - 2 cos x, 11(x, 0) = u,(x, 0) = 0. Hint: x sin x is a
particular solution.
2
9. 11 11 - llxx = x /(1 + t )2 + 2 In(! + t ), 11(x, 0) = 11,(x, 0) = 0. Hint: Take
i.:(x, t) = -x 2 In(! + t) as a particular solution.
10. Verify that formula (5.7) satisfies the nonhomogeneous wave equation (5.1) and
the homogeneous initial conditions (5.8).
1 l. (Duhamel's Principle) Let 11(x, t, r) be the solution of the initial value problem
(ix; < ctJ)
u(x, 0, r) 0, 11,(x, 0, r) F(x, r)

Set
r(x, t) I: 11(x, I - r, r) dr
(a) Show that r(x, I) satisfies the nonhomogeneous equation r 11 - c 2 1xx

F(x, t) and the initial conditions
i:(x, 0) = 0, r,(x, 0) = 0.
(b) Express u(x, t, r) by using d'Alembert"s formula and verify that 1(x, t)
agrees with (5.7).
( X < X, t > 0)
u(x, 0) = j(x), 11,(x, OJ = g(x)
Hint: Let u(x, t) = v(x, t)e"x+hr and choose the constants a and h in such a way
that the equation reduces to the wave equation.
6. Uniqueness of Solution
In the preceding section we deduced the uniqueness of the solution of the

initial value problem
(6.1) ( - J.) < x < J.), t > 0)
(6.2) u(x, 0) = f(x), u,(x, 0) = g(x) (-ctJ<X<W)

Sec. 6 Uniqueness of Solution 95
from the fact that the solution of the corresponding problem involving the
homogeneous \vave equation is uniquely represented by d'Alembert's formula.
It is important to note, however, that uniqueness can actually be established,
even without knowing that a solution of the problem exists. Jn this section we
shall present a method for proving uniqueness of the solution of an initial \alue
problem for the more general hyperbolic partial differential equation
(6.3) 11 11 -
2
c u'"' + h11 = F(x, t)
where c > 0 and h ~ 0 are constants. This equation is known as the damped
wave equation, which physically describes the motion of a vibrating string
under the action ofa restoring force -hu and an external force F. We shall show
that the initial value problem consisting of the differential equation (6.3) and
initial conditions (6.2) has at most one solution.
Suppose that u and u are two solutions of (6.3), (6.2), which are both twice
continuously differentiable. Set w = 11 - v; then, clearly, w is a solution of the
homogeneous problem
(- 00 < x < 00, 1 > 0)

(6.4)
w(x, 0) = 0, w,(x, 0) = 0 (-oo<x<oo)
let(, r) be any point in the xt-plane, T > 0, and denote by~ the characteristic
triangle determined by(, r), Figure 3.8. We shall prove that w vanishes identic-
ally throughout the region ~. and in particular at the vertex ((, r). Since the
point((, r) is chosen arbitrarily, this will mean that u = i at all points; that is,
the two solutions u and v are identical.
(~ T)
Ii
+----'------------------"-----x
0 (~ CT.0) I=() (~+CT.())
FIG. 3-8 Proof of uniqueness.

Let us multiply the differential equation in (6.4) by w, and observe that
W 1 \\'" =
I c ?
(w-)
2 Cf I
\\'x \V XI
1
') ~
a( \\' 2.)
x
- ct
Then
(6.5)
Let us draw the line t = t '. 0 :s; t' < T, intersecting the characteristics at the
points C and D (see Fig. 3.8). If we integrate equation (6.5) over the domain R
bounded by the line segments AB. BC. CD, and DA, and recall Green's theorem,
we obtain
(6.6)
= 0
where L denotes the boundary AB + BC + CD + DA of R. Now, on the line
segment AB, w(x, 0) = 0. ir,(x, 0) = 0, and hence also H$X, 0) = 0. Thus,
the line integral along AB vanishes. On the characteristic segments BC and DA,
we have dx = - c dt and dx = c dr, respectively. Hence,
Lc+DA l~ (w; + c
2
\\':
2
+ lnv ) dx + c
2
ll'x\\' 1 dtJ
J BC l
l (w I2
2
+ c 2 w .<2 + lnv 2 ) dx - cw x w dxJ I
.
+ LA l~ (w; + c w; + hw )(c dt) + c w,w, dt]

2 2 2
~ Le [(w, - c11$
2
+ hw 2] dx + ~ fvA 2 2
[(w, + cwJ + /Jw ] dt
Sec. 6 Uniqueness of Solutio11 97
which are nonnegati\e quantities. Finally, on CD we have dt = 0 so that the

integral there reduces to
which is also nonnegative. Now, according to (6.6). the sum of these integrals
is zero. Since the sum of nonnegative terms can never be zero unless each of the
terms is zero, we conclude in particular that
(6.7)
Since \\'and its first- and second-order partial derivatives are continuous, the
integrand in (6.7) is continuous. It follows that
(6.8)
identically on CD. Indeed, if (6.8) were positive at a point on CD, it would be
positive in some intenal containing that point, and thus the integral over the
intenal would be positive. contradicting (6.7). Now equation (6.8) is true if and
only if 11 vanishes identically on CD, since each of the terms in (6.8) is non-
negative. Therefore, 11(x, t ') = 0 for s-
c( r - r ') ::o; x ::o; + c( r - r ').
O ::o; t' < r. Thecontinuityof11 impliesthat11(, r) = Oalso. Hence,u((, r) =
11 . r), and consequently 11 and i are identical. Th us, we have established the
following uniqueness theorem.
THEORE.\I 6.1. The solution u(x, t) of the initial rnlue problem (6.3), (6.2) is
uniquely determined by the rnlues off and g on the interrnl [x - ct, x + ct] and
hr the rnlues of Fin the characteristic triangle determined by (x, t ).
The integral
.'.+c(r-1)
, , ,
(6.9) E(t) =
J
2 J
~-c(r-r)
(u; + cu, + hu-) dx
from (6.7) is often called the energy integral of the function u at time t over the
interval[ - c(r - t), ( + c(r - t)] for the damped wa\e equation (6.3). It
can be interpreted physically as representing the total kinetic and potential
energy of the portion [ ( - c( r - t ), ( + c( r - t )] of the string and the work
done by the restoring force on that portion. If h = 0, then (6.9) represents the
energy integral of u for the wave equation.
As a consequence of the uniqueness of the solution of the problem (6. I), we
can deduce the solution of the initial-boundary value problem
Ult - C2ll xx 0 (x > 0, t > 0)
u(x, 0) f(x) (x 2 0)
(6.10)
u,(x, 0) g(x) (x 2 0)
u(O, I) 0 (t 2 0)
"i!t,,
for a vibrating semi-infinite string x ;:::: 0 fixed at the end x = 0. Indeed,

consider the initial value problem
2
11 rt c u xx =0 (- 00 < x < 00, t > 0)
(6.11) 11(x, 0) = (x) (- 00 <x < 00)
u,(x, 0) = l/J(x) ( - 00 < x < 00)
where the initial data and l/J are the odd extensions off and g, respectively:
that is, cp(x) = f(x), l/J(x) = g(x) for x ;:::: 0, and (x) = -f(-x), l/J(x) =
- g( - x) for x < 0. Clearly, the solution u of the problem ( 6. I I) satisfies the
differential equation and the initial conditions of (6.10). It remains only to show
that 11 also satisfies the boundary condition 11(0, t) = 0. To this end, let v be
defined by r(x, t) = -u(-x, t). By direct differentiation, using the chain rule,
it is easily seen that i satisfies the wave equation. Moreover, from the definition
of and l/J it is seen that
i-(x, 0) = - 11( - x, 0) = - ( - x) = (x)
r,(x, 0) = -11,( - x, 0) = -1/1( - x) = l/J(x)
Hence, i is also a solution of the problem (6.11). Since the solution of (6.11) is
unique, we conclude that r(x, t) = u(x, t): that is,
-11(-x, t) = 11(x, t)
Setting x = 0, we obtain - u(O, t) = u(O, 1), which implies that 11(0, t) = 0,

t ;:::: 0. Thus, the solution of the problem (6.11) provides a solution of the
initial-boundary value problem (6.10) on a semi-infinite interval x ;:::: 0. Note
that in order for the solution of the problem (6.l I) to be twice continuously
differentiable for all x and t > 0, the extended functions and l/J must be twice
continuously differentiable and once continuously differentiable, respectively.
This will be the case if f and g are themselves functions with these properties
such that
/(0) = g(O) = f"(O) = 0
The foregoing discussion essentially shows that the solution of an initial

value problem with initial data that are odd functions is an odd function in the
variable x, and hence vanishes at x = 0 fort ;:::: 0. In an analogous manner it
can be shown that if the initial data/and g are even functions, then the solution
of. the initial value problem is also even in x; that is,
11(-x, t) = u(x, t)
This implies that
-11xC-x, t) u/x, t)
Sec. 6 Uniqueness of So/11tio11 99
from which we deduce u_,JO, t) = 0 for t ::=:: 0. This result provides a method
for solving the initial-boundary value problem
(x > 0, t > 0)
(6.12) u(x, 0) = f(x), u,(x, 0) = g(x) (x ::=:: 0)
u)O, t) = 0 (t :::::: 0)
on a semi-infinite interval x :::::: 0 with free boundary condition at x = 0. We

need only extend f and g for x < 0 as even functions and solve the resulting
initial value problem by d'Alembert's formula. Here, u is twice continuously
differentiable if, in addition to the differentiability conditions to be satisfied by
f and gas for (6.10), we assume that ('(O) = g'(O) = 0.
Example 6.1. Find the solution of the initial-boundary value problem
11,, - Uu = 0 (x > 0, t > 0)

u(x, 0-) = sin 2 x, 11,(x, 0) = 0 (x ~ 0)
u(O, t) = 0 (t ~ 0)
and evaluate u(n(l 2, n/4).
Solution: We extend the function f(x) = sin 2 x as an odd function for x < 0. By
d'Alembert"s formula (2.8), we have
u(x, t)
/(x - t) + /(x + t)
2
sin 2 (x - t J + sin 2 (x + r)
2
for x > t and
- sin 2 (f - x) + sin 2 (x + t)
u(x, t)
2
2 cos x sin x cos r sin t
1 . .
sm 2x sm 2t
2
for x < t. Thus, at (11/12, n/4) we have
u (fi, :) = -s~ (n/6) + sin (n/3)

2
2
2
= ~ ( - : + ~) 4
100 The WaYe Equation Chap. 3
Example 6.2. Sahe the problem in Example 6.1 when the boundary condition is
replaced b) the free boundary condition 11)0, t) = 0, and evaluate u(n/6, n/6).
Solution: Here \\e need to extend the initial data as even functions. Since /(x) =
sin 2 x is alread; an even function. \le immediately have for the solution of the
pro bl em
sin 2 (x - t) + sin 2 (x + t l
ll(X. I)
2
(x ?. 0, t ?. 0)
Thus. 11(n;6, " 6) 's.
Exercises 3.4
1. Let 11 be a solution of the \I ave equation
('x' < oo, r > OJ
111iich is twice continuously differentiable. and vanishes together with its first
derivatives as x' -> x for 0 <::;, 1 <::;, r 1 . By differentiating under the integral sign
and then integrating by parts, show that the "energy integral"
is constant for 0 :$ 1 <::;, t1

2. 'Lsing the result of Problem I. prove that the initial value problem (6.1 ). (6.2)
has at most one solution. assuming that its solution vanishes together with the
fast-order derivative as x - Cf2.
3. Let 11 be a solution of the damped wave equation
"hich is sufficiently smooth and vanishes together with its first derivatives as
-> ex;. Show. as in Problem I. that the energy integral
is constant for 0 s 1 s 11 .
4. From the result of Problem 3 deduce that the initial value problem (6.3). (6.2)
has at most one solution. assuming the same condition as in Problem 2.
5. Obtain the explicit solution of the problem (6.10) and verify that it satisfies the
differential equation and the initial and boundary conditions.
6. Prove the statement in the text that if f and g are two even functions, then the
solution 11 of the initial value problem (2.1 ), (2.2) is also even.
Sec. 7 lnitia/-Bo1111dary Value Problems IOI
7. Find the solution or the initial-boundary Yalue problem
ll11
-
('2 {IX.\ = 0 (x > 0, I > 0)
11(1. 0) - j(I). 11 1 (x, 0) g(x) ( .\ 2:: 0)
11,(0. I) = 0 (/ 2 0)
8. Find the solution of the initial-boundar\ value problem
"11 - llxx = 0 (X > 0. I > 0)

11(.\. '
0) = :c. 11/x. 0) 1(1 -
X) (x 2 OJ
u(O. I) = 0 (I 2 OJ
and evaluate the solution at the po111ts !}. 1) and(~.~). (U-,c the solution formula
of Problem 5.)
9. Sohe the initial-boundary 1alue problem
II rr -- llxx 0 (x > 0. I > 0)

//(X. 0) sin(rr.Y/2). 11J:c 0) .Y (x :->- 0)
11(0. I) 0 (r 2 0)
IO. Find the solution of the \\ave equation
flu
-
11.Y.\" - 0 {x > 0. r ' 0)
11 ith the foll011 ing conditions:
{al u(x, 0) = cos(rrx!2J. 11 1 lx. OJ= .\ 2 .\ 2 O; 11_0. 1) = 0, I 2 O;
(b) 11(.1. OJ= e-1-'1. 11Jx. OJ= cos x, x 2 0; 11)0, I)= 0, I 2 0.
11. Show that if the function flx. I) is odd (or even l in x. then the same is true of the
solution 11 of the initial value problem
F(x, I) ('xi< oc., I> 0)
u(x, 0) 0, 11,(x. 0) () ( x' < er:)
and thus deduce that 11 (or u) vanishes at x = 0.

12. Find the solution of the problem
Jill -- [{xx x (x > 0, I > ())

li(X, 0) sin x. 11,(x, OJ 0 (x 2 0)
11(0, I) 0 (t 2 0)
1111 - llxx = (x > 0, I > 0)

u(x, 0) = 0, uJx, 0) cos x (x 2 0)
uJO, I) = 0 (I 2 0)
7. Initial-Boundary Value Problems
We have seen in the preceding section that the solution of an initial value
problem for the wave equation with odd (or even) initial data, when restricted to
x ~ 0, t ~ 0, provides a solution of the initial-boundary value problem on the
semi-infinite interval x ;?: 0 with fixed (or free) boundary condition at x = 0.

The same idea can be used to solve an initial-boundary value problem on the
finite interval 0 s x s L with either fixed or free boundary condition at either
end point.
Let us consider the initial-boundary value problem
(7.1) (0 < x < L, t > 0)
(7.2) u(x, 0) = j(x), u,(x, 0) = g(x) (0 s x s L)
(7.3) u(O, t) = 0, u(L, t) = 0 (t ;?: 0)
which corresponds physically to the problem of a vibrating string of finite length

L with fixed ends and with no external force. We recall that the solution of an
initial value problem with odd initial data (about x = 0) is again odd: that is,
u( - x, t) = - u(x, t ). In a similar manner, we can show that if the initial data
are odd about x = L-that is,f(2L - x) = -f(x) and g(2L - x) = -g(x)-
then so is the solution 11. Consequently, if the initial data are odd about x = 0
and x = L, the solution 11 will then satisfy the relations
u(-x, t) = -u(x, t) and u(2L - x, t) = -u(x, t)
\v hich, in particular, imply
11(0, t) = 0 and u(L, t) = 0 (t ;?: 0)
Thus, the solution of the initial value problem with initial data that are odd
about x = 0 and x = L, and which coincides with (7.2) on [O, L], gives a
solution of the initial-boundary value problem (7.1 ), (7.2), (7.3). Therefore, to
find a solution of the problem (7. I), (7.2), (7.3), we extend the initial data/ and g
given for 0 s x s L as odd functions about x = 0 and x = L, and then
substitute the extended functions in d' Alembert's formula (2.8). The restriction
of this d'Alembert's solution to the region 0 s x s L, r ;?: 0 is a solution of
our problem. In this connection, the solution obtained will be twice continuously
differentiable if we require that the following compatibility conditions
j(O) = f(L) = 0, j"(O) = f"(L) = 0, g(O) = g(L) = 0
hold, in addition to the assumption that I and g have continuous second-order
and first-order derivatives on 0 s x s L, respectively. That such a solution is
uniquely determined will be proved at the end of this section.

1111 - llxx = 0 (0 < x < I, t > 0)
u(x, 0) = sin 11x (0 s x s I)
11,(x, 0) = 0 (0 s x :S I)
u(O, t) = 0 (u(l, t) = 0, t ~ 0)
Sec. 7 Initial-Boundary Value Problems 103
Solution: Since sin n( -x) = - sin nx and sin IT(2 - x) = -sin nx, the function
f(.1) = sin ITX is already odd about x = 0 and x = IT, and hence we immediately
have
u(x, t) = sin IT(x - I)__ + sin n(x + t)

2
sin ITX cos IT{
as the solution valid for 0 ::::; x ::::; 1, t 2: 0. This is easily verified.
Example 7.2. So]\e the problem in Example 7.1 when the initial displacement is
u(x, 0) = x(x - I) (0 ::::; x ::::; I)

and evaluate u(3/4, 1/2).
Solution: For points (x, t) such that 0 ::::; x - t < x + t ::::; 1, the solution is given
by
(x - t)(x - t - I) + (x + t)(x + t - I)
11(x, t)
2
= x2 - x + ,2
For points (x, t) such that x < tor 1 < x + t, the solution is given by
f(x - t) + f(x + t)
u(x, t) =
2
where I is the odd extension of 11(x, 0) = x(x - 1) about x = 0 and x = I.

Thus, at (3/4, 1/2) we h<l\e
/(3/4 - 1/2) = /(1/4) = (1/4)(1/4 - I)= -3/16

and
/(3/4 + 1/2) = /(5/4) = /(2 - 3/4) = -/(3/4)
= - (3/4)(3/4 - 1) = 3/16
so that 11(3/4, 1/2) = 0.
Example 7.3. Find the solution of the problem
llir - Uu = 0 (0 < x < I, t > 0)

u(x, 0) = x 2 (1 - x) (0 ::::; x ::::; I)
11,(x, 0) = x(x - I) (0 ::::; x ::::; I)
11(0, I)= 0, 11(1, t) = 0 ( ( 2: 0)
at the point (1/2, 5/4).
Solution: We extend the initial data here as odd functions about x = 0 and x = I.
Now, at (1/2, 5/4), we have
f( - 3/4) = -/(3/4) = -(3/4) 2 ( I - 3/4) = - 9/64

/(7/4) = /(2 - 1/4) = -/(1/4) = -(1/4) 2 (1 - 1/4)
= - 3/64
104 The l'l'ave Equatio11 Chap. 3
Further.
"
J
7 4
. .
- 3 4
q(.\) dx = f
J_
314
3;4
g(x) dx + f 5
3/4
14
g(x) dx + J
7 4
i
5/4
g(x) dx
-;4
= 0 + 0 +
J ."
5/4
.q(x) dx
since g(x) is an odd function about x = 0 and x = 1.

By the change of variable ' = 2 - s. we have
J
7;4 f3;4
g(x) dx = - s(s - 1) ds
5/4 I /4
s"! s 2 ! 314 11
+ i
3 2 I /4 96
Thus.
-9 - 3 1 11
2(64)
+ 2 96
-7
192
In the same way, we can show that the problem (7.1 ), (7.2) with free boLrndary
conditions
(7.4) ux(O, t) = 0, uxCL, t) = 0 (t > 0)
at both ends can also be solved by using d'Alembert's formula (2.8), provided
the initial data/ and g are extended as e\en functions about x = 0 and x = L;
that is,f( - x) = f(x), g( -x) = g(x),f(2L - x) = /(x), and g(2L - x) = g(x).
Then d'Alembert's solution 11 is itself an even function about x = 0 and x = L,
and hence satisfies the relations
11(-x, t) = u(x, 1), u(2L - x, 1) u(x. I)
These imply that

u.(-x, 1) = -uAx, I), uA2L - x, t) - uxCx, t)
which, in particular, give
uxCO, t) = 0 and u,(L, t) = 0

showing that u satisfies the boundary conditions.
The same method can also be used to solve an initial-boundary value problem
on the finite intenal 0 ::::; x ::::; L, which involves the nonhomogeneous wave
equation
(7.5) F(x, t)
Sec. 7 luitial-Boundary Value Problems 105
with fixed or free boundary conditions at either end point. In such a case. we
extend the function F together with the initial data as odd or even function
about an end point, according as the boundary condition prescribed at that
point is fixed or free. Then the same will be true of the solution given by ( 5.6)
and hence will satisfy the prescribed boundary conditions.
. 7r
11 11 - lln =
.. Slll
2 X (0 < x < L I > OJ
11(x, OJ = 0, 11,(x, 0) = 0 (0 5 x 5 1)
u(O, I) = 0, (t 2: OJ
So/111io11: Here we observe that the function F<x, I) = sin(n/2)x is already odd about
' = 0 and is even about x = I. Hence, by ( 5. 7), our solution is
u(x, I) = j'' J,x+(r-T) sin 71

.; d(, dr
2 0 x-(r-r) 2
71
Llcos;(x - I+ r) - cos;(x + 1- r)J dr
:
2
l sin ; (x - I + r) + sin ; (x + I - r) J'
0
7!2
4
sin
71
2
x ( 1 - cos '.1
-
71
I)
for 0 5 x s I, t 2: 0.
The uniqueness of solution of the initial-boundary value problem (7.1), (7.2),

(7.3) can, of course, be deduced from the uniqueness of d'Alembert's formula.
1-1 owever, it is worthwhile and not difficult to give a separate proof of uniqueness
which is independent of the manner of derivation of the solution of the problem.
We shall prove here the uniqueness of solution of the more general initial-
boundary value problem
ll Cl (0 < x < Lt > 0)
(7.6) u(x, 0) = j(x), u,(x, 0) = g(x) (0 :::; x :::; L)
u(O, t) = h(t). u(L. t) = k(t) (t ?. 0)
where we assume that the functions F.f, g, h, and k are sufficiently differentiable
and compatible at the end points. An explicit solution of this problem when
hand k are not both zero will be given in the next section.
Suppose u 1 and u 2 are two solutions of the problem (7.6) which are both
twice continuously differentiable, and let w = u1 - u 2 . Then w is a solution of

the homogeneous problem
0 (0 < x < L, t > 0)
(7.7) w(x, 0) = 0, w,(x, 0) = 0 (0 ~ x ~ L)

1r(O, t) = 0, tt:(L, t) = 0 (t 2 0)
Consider the energy integral (see (6.9))
(7.8) E(t) = 1- SL (w, 2

+ c 2 w;) dx (t 2 0)
2 0
for the function w on the interval [O, L]. Differentiating this with respect to t
and integrating the second term by parts, we have
(7.9) E'(t) = f0
L (w 1w11 + c 2 wxwx 1) dx
Since w is a solution of the problem (7. 7), we see that the integrand in the last
integral above vanishes, w,(O, t) = 0, and w,(L, t) = 0, so that (7.9) reduces
to E'(t) = 0 for t 2 0. This means that the energy integral (7.8) is a constant
fort 2 0. But at t = 0, we see from (7.7) that w_,,(x, 0) = 0 and w,(x, 0) = 0,
which imply that (0) = 0. Hence, the integral (7.8) is identically zero, and
since its integrand is nonnegative, it follows that
for 0 ~ x ~ L, t 2 0. By the same argument as in the proof of Theorem 6.1

of Section 6, we therefore conclude that w = 0 identically, which shows that
u 1 = u 2 , as we wish to prove.
Exercises 3.5
1. Show that if f is odd (or even) about the points x = 0 and x = L, then f is
periodic with period 2L; that is,f(x + 2L) = f(x).
2. Show that if f is odd (or even) about x = 0 and even (or odd) about x = L, then
f is periodic with period 4L.
3. Show that if f is periodic with period 2L, then
a+ZL JL
i a f(x) dx = -L f(x) dx
(This means that the integral off is the same over any interval of length 2L.)
Sec. 7 J11itial-Bo1111dary Value Problems 107
4. Let f and g be two odd functions about x = 0 and x = L. Verify that the
solution of (2.1 ), (2.2) is also odd about .1 = 0 and x ~ L; thus, deduce that
11(0, 1) = 11(L, tJ = 0
5. Let /and g be t\\O e\en functions about x = 0 and x = L. Pro\e that the
solution of (2.1 ), (2.2) is even about the same points and deduce 11x(O, t) =
llx(L, I)= 0.
6. Let f and g be two functions that arc odd about x = 0 and even about x = L.
Prove that the solution of (2.1 ), (2.2) is odd about x = 0 and even about x = L,
and deduce that 11(0, I) = 0, 11_L. t) = 0.
In Problems 7 through 12, find the solution of (7.1) with the given initial and
boundary conditions. Take c = I and L = I.
7. 11(x, 0) ~ sin nx, 11,(x, OJ = x(l - x 2 ); 11(0, 1) = u(l, t) = 0. Evaluate
u(5/8, 9/8J.
8. u(x, 0) = x(I - x), 11 1(.Y, OJ= x 2 (1 - x); 11(0, I) = 11(1, I)= 0. Evaluate
11(3/4, 2).
9. 11(x, 0) = x 2(1 - x 2 ), 11,(x, OJ = cos nx/2; ux(O, t) = 0, 11(1, t) = 0.
10. 11(x, 0) = sin 2 nx, 11,(x, 0) = O; 11)0, I) = 11x(I, t) = 0.
11.
u(x, 0) = .I
x (0 s x s 1/4)
\( - x + I )/3 (1/4 s x s 1)
11,(x, 0) = 0 (0 s x s I)
11(0, I) = 11(1, t) = 0
12.
11(x, 0) = 0, u,(x, 0) =
y (0 s x s 1/2)
{
-x + (1/2 s x s I)
u(O, t) = uI, I) = O.
13. Evaluate 11(1/2, 3/2) if u satisfies
(0 < x < I, t > 0)

u(x, 0) = 11,(x, 0) = 0 (0 s x s I)
llx(O, t) = uJI, 1) = 0 ( t 2: 0)
14. Evaluate 11( 1/4, 2) if
11 11 - uu = 1- x (0 < x < 1, t > 0)

u(x, 0) = x (1 - 2
x), u,(x, 0) = 0 (0 s x s I)
// x(O, t J = u( 1, t) = 0 (t 2: 0)
15. By using the energy integral (6.9) over [O, L], prove that the initial-boundary
value problem
l/ 11 - c 2uxx + h11 = F(x, t) (0 < x < L, t > 0)
u(x, 0) = f(x), 11.(x, 0) = g(x) (0 s x s L)
u(O, t) = a(t), u(L, t) = b(t) (t 2: 0)
has at most one solution.

108 The Wal'e Equation Chap. 3
8. :\onhomogeneous Problems and Reflection of \\a Yes
Jn the preceding two sections \\e showed how the solution of the initial\ aluc
problem for the \\ave equation can be employed to obtain the solution of the
initial-boundary\ alue problem on a semi-infinite interval or on a finite in ten al
when the boundary conditions are either llxed or free. When the boundary
conditions are not homogeneous. the method does not apply, and we have to
use a different approach. In this section we shall present a method for solving
problems involving the \\"a\e equation with nonhomogencous boundary
conditions of the first kind.
First, let us consider the initial-boundary value problem
(x > 0, I > 0)
(8.1) u(x. 0) = f(x). 11,(x, 0) = g(x) (x ~ 0)
11(0, t) = lz(f) (I ~ 0)
which corresponds to a vibrating semi-infinite string where the end point x = 0

is gi\en a displacement described by /i(I) fort ~ 0. For each point (x, I) such
that x > ct, we obscne that the interval of dependence [x - cl. x + cl] lies
entirely on the positi\e x-axis. Hence, for x > cl. the solution of problem (8.1)
is given by d'Alemberfs formula
(8.2) u(x.1) = cj;(x - ct)+ 1/1(x +ct)
where (/; and ~; are defined by
cf;(x) =
2
1 /(x)
2c
I q(s) ds
r
(8.3)
1/1(x) = I /(x) + 1 g(s) ds

2 2c
This is true regardless of the boundary condition at x = 0. Physically. this says

that at the point x on the string and at time I when no \\a\c from the end point
has yet arrived, the displacement u(x, 1) is caused simply by the initial disturbance
on the intenal [x - cl, x +- ct].
But \\hen x < ct, the portion [x - cl, OJ
of the intenal [x - ct. x + cl]
lies on the negative x-axis on which/and g are not defined (sec Fig. 3.9). Hence.
for x < ct the value of q; is not known and so (8.2) is not valid. However, by
using the boundary condition in (8.1 ). we see from (8.2) that
(8.4) u(O, 1) = h(I) = </;(-cl) + 1f1(ct)
which, on setting s cl. gives
(8.5) q;( - s) Ii (;) - ij;(s)

Sec. 8 Nonhomo_qeneous Problems and Reflection of J47 aves 109
This furnishes the value of r/; for negative values of ils argument. and so for
x < ct we have
(8.6) <j;(x - ct)= Ii (r ~) - t/l(ct - x)
Substituting this in (8.2). we obtain the formula
(8.7) u(x, 1) = lz (i ~) 1/f(ct - x) + t/f(x + ct)
h (t :)
+ f(x__+_ C2_~ f(ct - x) + I Jx-ct g(s) ds
2 2c ct-x
for x < ct. Thus, our complete solution of the problem (8. I) is given by (8.2),
(8.3) for x > ct and by (8.7) for x < ct. In either case, we notice that the solution
is the superposition of a forward wave cp(x - ct) and a backward wave
l/f(X + Ct).
By direct differentiation it is easy to verify that the function u defined by (8.2)
or (8.7) satisfies the wave equation and that (8.2) together with (8.3) satisfies the
initial conditions, while (8.7) satisfies the boundary condition of the problem.
In this connection, we assume that lz has continuous second-order derivative for
t > 0 and that f and g have continuous second- and first-order derivatives,
respectively, for x > 0 such that lz(O) = f(O) and /z'(O) = g(O).
Formula (8.7) has an interesting and important physical interpretation. Let
us consider (8.7) at the point (x 0 , t 0 ) in the domain x < ct of the xt-plane and
draw the characteristics through this point (Fig. 3.9). We see that the first term
in (8.7) is precisely the value of lz(t) at the t-intercept of the characteristic
x - ct = x 0 - ct 0 . It represents the displacement of the end point at time
t = t 0 - x 0 /c and travels along the characteristic x - ct = x 0 - ct 0 . The
forward wave - t/l(ct 0 - x 0 ) traveling along the same characteristic as
FIG. 3-9 Reflection of wave from the end x = 0.

110 The Ware Equation Chap 3
h(f 0 x 0 /c) represents the reflection of the backward wave that originated from
the point (ct 0 - x 0 0) on the initial line. This backward wave first traveled
along the characteristic x + ct = ct 0 - x 0 until it reached the end point and
was reflected. The min us sign in front of if1 indicates that there is a 180-degree
phase shift upon reflection. As a result of this phase shift, the contribution of
the initial velocity from the interval [O, ct 0 - x 0 ] canceled out. Thus, the
interval of dependence of a point (x 0 , t 0 ) in the domain x < ct is
[ct 0 - x0 , x0 + ct 0 J
which is the interval cut off from the positive x-axis by the characteristic
x + cf = x 0 + ct 0 and the reflection of the characteristic x - ct = x 0 - ct 0
at the end point x = 0.
The same method can be used to sohe the initial-boundary value problem
(0 < x < L, t > 0)
(8.8) u(x. 0) = f(x). 11,(x. 0) = g(x) (0 :.:;; x :.:;; L)
11(0. r) = h(t). u(L, t) = k(t) (t ~ 0)
on a finite intenal with nonhomogeneous boundary conditions. We note, as in

the case of problem (8. l ), that for points (x, t ). 0 < x < L, t > 0--for which
the intenal [x - ct, x + ct] lies inside the interval [O, L]--d'Alembert's
formula (8.2), (8.3) remains \alid regardless of the boundary conditions. It is
only \\hen the in ten al [x - ct, x + ct J extends beyond [O, L] that we need to
find for (8.2) the value of (/> for negative argument and the value of 1/> for
posi ti\ e a rgu men t.
Now, on the intenal -L < s < 0. the \alue of(/> is given by (8.5): that is,
(8.9) (-s) = h C) - 1/>(s) (0 < s < L)
To find the \alue of if1(s) for L < s < 2L. we note by the second boundary
condition 111 (8.8) that
11(L, t) = k(r) c/>(L - er) + 1/1(L + ct)
so that
(8.10) if1(s + L) = k (;) - q>(L - s) (0 < s < L)
This gives the \ alue of i/J(x) for L < x < 2L. 1n this way the functions cf> and if>
are then completely determined on the intervals ( - L, L) and (0, 2L), respec-
ti\ely. Thus, it is possible to write down the solution of problem (8.8) at any
point (x, t) such that - L < x - ct < x + ct < 2L (see Example 8. l ).
lfwereplacesbys + Lin(8.9)anduse(8.l0),wefind
(8.11) (-s - L) = Ii (s : L) - ijl(s + L)

/1 (s : L) - k {;) + (L - s) (0 < s < L)
Sec. 8 No11!10111ogeneo11s Problems and Reflection of Waves ll I
and this further extends (x) onto the interval - 2L < x < - L. Similarly. by
replacing s bys + Lin (8.10) and using (8.9), we have
(8.12) 1p(s + 2L) = k (5 : L) - hC) + lf;(s) (0 < s < L)
which defines 1p(x) on the interval 2L < x < 3L. Formulas (8. I I) and (8. 12)
are recurrence relations that together with (8.9) and (8. I 0), determine the value
of for any negative argument and the value of 1p for any positive argument.
If we know the values of and lf;, it is possible to write down the solution of
problem (8.8) at any point (x, t), for 0 < x < L, t > 0.
(()_I
/
/
,/
/
/
/
/
/
/
/
/
/
/
/
/
// 10.t +~-c
/ /
/
/
/
/
/ /
/ /
// //
/ // L
~ ------~-----~~--~-~------~-----~---.,;
~I L I 2L + \ u. 0 I I\ + 11 2 L 0 I 2L 3L
FIG. 3-10 Repeated reflection of wai-es.
For example. if (x, t) is a point such that -2L < x - ct < -L and
2L < x +ct< 3L(seeFig.3.10),thenaccordingto(8.Il),
(x - ct) = h (t x)c k (t - x +
c L) + (2L + x - ct)
where 0 < 2L + x - ct < L and, according to (8.12),
lf;(x + ct) = k (1 + x c L) - h(r + x -c

2
L) + lf;(x + ct - 2L)
with 0 < x + rt - 2L < L Hence, by (8.2), the solution at this point is
(8.13) u(x. t) = (x - cl) + lf;(x +cl)
x - 2L) + h. ( 1 + x - L)
(' . c .
.( x + L) f(x + ct - 2L) + f(2L + x - ct)

I' t - +
c 2
I JXTct-2L
+ g(s) ds
2c 2L+x-ct
where we ha\e used (8.3) for c/>(x) and t/J(x) for 0 < x < L.
It is interesting to note that the solution (8.13) can also be obtained geo-
metrically by using the idea of reflection of characteristics, which is analogous
to the reflection of \\a\es. We draw the characteristics through the point (x, t)
and their reflections at x = 0 and x = L, 1110\ing backward in time, until they
intersect the initial line on the interval 0 < x < L (Fig. 3.10). Then the
contribution of the boundary conditions to the displacement at (x, I) consists
of the \alues of hat the !-intercepts and the values of k at the intercepts with
x = L of the characteristics. counting a change of sign following each reflection.
Thus, from the end x = 0, we have h(f - x/c) and -h(t + (x - 2L)/c), in
which the second term has a negative sign because the wave it represents is
reflected once at x = L. Similarly, from the end x = L, we have
k(t + (x - L)!c) and -k(t - (x + L)/c)
where the \\ave represented by the second term is reflected once at x = L. The
contribution from the initial displacement consists of the terms f(2L + x - ct)
and f(x + cf - 2L), which represent waves that originated from the points
(2L + x - ct, 0) and (x + ct - 2L. 0). respecti\ely: these waves were re-
flected twice, once at each end before arriving at the point (x, t ). The net
contribution from the initial velocity comes from the interval
[2L + x - ct. x + cf - 2L].
Needless to say. the method described above applies in particular to the case
where the boundary conditions are homogeneous. In fact, in such case, formulas
(8.11) and (8.12) imply that and VJ must be extended as periodic functions of
period 2L. Using this fact and the formulas given in (8.3), it can be established
that the initial data f and g must then be extended as odd periodic functions of
period 2L. This is equivalent to extending/and gas odd functions about x = 0
and x = L (see Problem I, Exercises 3.5), which was the method we adopted in
Section 7.
Sec. 8 So11ho111oge11eo11s Prohfems and Rejfection of' Waves 113
Example 8.1. Find the solution of the problem 18.~J at points I 1. r) such that
- L < x - er < 0 and L < x + ct < 2L.
So!u1io11: Since - L < .1 - ct < 0. we ha1e by (8.9)
J,( \ - Cl) - /1 ( f :. ) - I//((/ - >:)
and since L < x + ct < 2L, 1\e lia1e by (8.10)
lj!(X-t-C/)=k
(I + ' - L) C dJ(2L - x - Cl)
Hence. by (8.2) and (8.3). the solution is giYen by
u(i. I ) = h (i ;.) + k (r + x ~ L) _ f(U - x) + 1;21_ - x - ct)
11121_-_\ -er
+
2c
J er - x
g(s) ds
Exercises 3.6
I. Evaluate 11(5/8, 19/8) for Problem 8 of Exercises 3.5, using the method of wave
reflect ion.
2. Evaluate 11(1 /4, 3/2) for Problem 7 of Exercises 3.5, using the method of wave
reflection.
3. Find the solution of the wave equation 11 11 - "-u = 0, x > 0, t > 0, satisfying
the given conditions:
(a) u(x, 0) = e-x - I, 11/t, 0) = 0, .1 ~ O; 11(0. t) = re-'. I ~ 0.
(b) u(x, 0) = sin x, 11,(x, OJ = x, x ~ O; 11(0, t) = sin 2t, t ~ 0.
4. Find the solution at the indicated point of the wa1e equation
Uu = 0 (0 < .1 < I, t > 0)

satisfying the given conditions
(a) at (1/4, 3/2) with 11(x, 0) = e' - I, 11,(.1, 0) = 0, 0 < x < I; 11(0, t) =re-',
11( 1, t) = t 2, I > 0.
(b) at (3/4, 3/2) with 11(.1, 0) = 0, 11,(x, OJ= x(I - x), 0 < x < I; 11(0, 1) =
sin nt, 11(1, I) = cos nt/2. I ~ 0.
5. Show that the solution of the initial-boundary value problem (8.8) 1vhen
f = g = k = 0 is given by
u(x, 1) = 2:
11:<:=0
h (_' _ x + 2nl) _
c
2:
11:2= l
h (r + x - c211L)
\\here h(t) = 0 fort < 0 and h has continuous second-order derivative.
6. Find the solution of the problem (8.8) when f = g = h = 0.
7. Solve the initial-boundary value problem
(x > 0, t > 0)
u(x, 0) = f(x), 11,(x, 0) = g(x) (x 2: 0)
u)O, t) = h(t) (t 2: 0)
8. Sohe the initial-boundary value problem
(x > 0, t > 0)
11(x, 0) = f(x), 11,(x, 0) = 0 (x 2: 0)
llx(O, !) - ku(O, t) = 0 (k a constant, t ;:::: 0)
Hint: Set r(x, t) = u)x, t) - ku(x, t) and consider the initial-boundary problem
for i.
9. Method of Separation of Variables
So far, we have been concerned with initial value and initial-boundary value
problems for the wave equation. We have seen that by an appropriate extension
of the initial data, it is possible to reduce an initial-boundary value problem to
an initial value problem for which d'Alembert's formula can be employed. For
a more general equation, however, we must use a different approach. Jn this
section we shall describe a rather convenient and powerful method for solving
initial value, initial-boundary value, or boundary value problems for a large class
of partial differential equations of hyperbolic, parabolic, or elliptic types.
Jn order to establish the basic steps involved in the method, we shall solve the
initial-boundary value problem
(9.1) (0 < x < L, t > 0)
where h = const > 0.
(9.2) u(x, 0) = f(x), 11,(x, 0) = 0 (0 ::::; x ::::; L)

(9.3) 11(0, t) = 0, u(L. t) = 0 (t > 0)
for the damped-wave equation. The method consists in assuming a particular

solution of (9. l) in the form
(9.4) u(x, t) = X(x)T(t)
where Xis a function of x alone and Tis a function oft alone. Moreover, the
function (9.4) will be required to satisfy all homogeneous conditions of the
problem, namely, the initial condition u,(x, 0) = 0 and the boundary conditions
(9.3). Now, substituting (9.4) in the differential equation (9.l), we have
X(x)T"(t) - c 2 X"(x)T(f) + hX(x)T(f) = 0

Sec. 9 ;1Jethod of Separation of Variables llS
where the primes denote ordinary differentiation with respect to the argument
of the function being differentiated. Dividing the last equation by c 2 X(x)T(t)
and "separating" variables, we find
T"(t) h X"(x)
(9.5) -2 + 2
= -
c T(t) c X(x)
Since the term on the left of equation (9.5) depends only on the variable t and
the term on the right depends only on the\ ariable x, it follows that the derivative
with respect to t of the term on the left is zero; likewise, the derivative with
respect to x of the term on the right is zero. Hence, both terms must be equal
to a constant, which we denote by -i.. Thus,
X"(x) T"(t) h
- = -), -? + ?=-).
X(x) c-T(t) c
which lead to the two ordinary differential equations
X"(x) + i.X(x) = 0
(9.6)
T"(t) + (}.c +
2
h)T(r) = 0
to be satisfied by the functions X and T, respecti\ ely. In this way we have

replaced the partial differential equation (9.1) by two ordinary differential
equations (9.6) involving a parameter /.. This is the essence of the method of
separation of variables.
Since the function (9.4) 1s to satisfy the homogeneous conditions of the
problem, we must have
X(x)T'(O) = 0
(9.7)
X(O)T(t) = 0, X(L)T(f) = 0
from which we conclude that T'(O) = 0. X(O) = 0, and X(L) = 0. The
alternate choice X(x) = 0 or T(I) = 0 would each lead to the trivial solution
u(x. t) = 0 of the differential equation. This. however, cannot be a solution of
the problem unless f = 0, too, in which case the whole problem is trivial.
Therefore, in order for (9.4) to be a nontrivial solution of equation (9. J) satisfy-
ing the homogeneous conditions (9. 7), the functions X and T must be the
solution of the problems
(9.8) X" + l.X = 0, .X(O) = 0, X(L) = 0
and
(9.9) T' + (i.c

2
+ h)T = 0, T'(O) = 0
respectively.
The problem (9.8) is a special case of a more general boundary value problem
in ordinary differential equations, known as the Sturm-Liouville problem, which
116 The II 1ne Equation Chap. 3
\I ill be seen that t Im type of problem has nou~

is discuso,ed i11 the next chapter. It
tri\ial solutions only for certain exceptional \alues of the parameter i.. called
the cigcmalucs of the problem. The corresponding nontri\ial solutio11s arc
called eigenfunctio11:,. For each cigcmalue i .. the product of the corresponding
eigenfunctio11 .\"and the corresponding function Tdeterrnined from the problem
(9.9) yields a particular solution or (9.1) satisf'\ing the conditions (9.7).
\\ c no11 proceed to the determination or nontri\ial solutions or problem
(9.8). \\"e consider three separate cases accord:ng to whether 1. is negati\C or zero
or positiw. Suppose first that i. < 0 and 1nite i. = -k 2 , where k is a nonzero
real 11umber. Then the differential equation in (9.8) has the general solution
Appl) ing the boundary conditions X(O) = XU.) 0. \le h~l\C
11h ich can he regarded as a system of algebraic equations for c 1 and c 2 Si nee
the determi11a11t of cocfllcients
= 2 sinh kL
does not \alllsh fork ~ 0. the only SOiution of the system is c 1 = c 2 = 0.

Thu' (9 8) has only tri\ial solution \\hen i. < 0: that is, the problem has no
negati\C eigell\ al ues.
Next. suppose that i. = 0. Then the diffe:ential equation becomes X" = 0.
\I ith general solution gi\en by /\"(x) = c 1 .\ c c 2 . Application of the boundary
condition,; readily sho\1s that c 1 = c 2 = 0. leading again to the tri\ial solution.

Hence. i_ = 0 is not an eigcmalue of" the problem.
Finally. \\C con~ider the c<hC / > 0. !11 this case the general solution of the
differential equation is
X(.\) c~ C 1 Sill"\/. X
Applying the boundar~ condition X(O) 0. \IC find c - 0: therefore a non-

tri\ial solution must be of the form .\"ix) = <1 .'>Ill, i. x. c 1 # 0. The boundary
condition X( L) = 0 then require:, that
Sill \ /_ L = ()
and this holds if and on!: if x i. L = 111c. n = 0. I. 2. + 3 ... _. Since

i > 0. it follows that the eigcm alue~ or the problem (9.8) arc
112 TC 2
(9.10) (I! I, 2, ... )
L"
Sec. 9 .\!ethod of Separation of Variables 117
and the corresponding eigenfunctions, apart from constant factors, are
(9.11)
Now, for each i. 11 n 2n 2 L2, n ~ J, the general solution of the differential

equation in (9.9) is
T(t) = Ci sin , i. 11 c2 +- fz t + c2 cos , i. 11 c2 + hr
The initial condition T'(O) = 0 requires that Ci = 0, thus leaving c2 arbitrary.

Hence, except for constant factors, we have
fn2rr2c2 J i:2
(9.12) T,.(t) = cos[ ~2 + Ii_ t (11 = 1, 2, ... )
as the corresponding solution of the problem (9.9) for each eigenvalue i 11 By

forming the product of the functions (9.11) and (9.12) in accordance with (9.4).
we obtain infinitely many particular solutions
(9.13) u,,(x, t) = sm
. 11llX
cos.
rl12T:2C2
+ h
J i:2 t (11 = 1,2,. .. )
L L2 .
of equation (9.1) satisfying the homogeneous conditions of (9.2) and (9.3). By
the principle of superposition, any linear combination of these functions
remains a solution of equation (9.1) and satisfies the same homogeneous initial
rnd boundary conditions. Hence, if we can find a suitable linear combination
of the functions (9. !..\)which satisfies as well the initial condition 11(x, 0) = f(x),
then such a combination will be a solution of the problem (9.1 ), (9.2), (9.3).
However. it is clear that no (finite) linear combination of the functions (9.13)
Lan satisfy (9.2) unless the initial datum f itself is a linear combination of the
eigenfunctions (9.1 l) We therefore try an infinite linear combinaiton of the
form
(9.14) 1/(X, t) = Ix
bll
.
'Ill
llllX
cos
l/1271:2C2
+ h
Ji 2
I
n=1 L . L .
Ckarly, if the constants b11 are such that the series (9.14) and its derivatives
comerge respecti\ely to the function u(x, t) and its derivatives for 0 < x < L,
I > 0, then by the extended principle of superposition the function u represented
by the series (9.14) is a solution of equation (9.1) safofying the homogeneous
conditions in (9.2) and (9.3). There remains the problem of determining the
constants b" so that (9.14) also satisfies the nonhomogeneous initial condition
in (9.2). Setting t = 0 in (9.14), we have
(9.15) . O) =. (( x )
u(x, f; b,, sin
= L.. . 11 rrx-
n= I L
Thus, the constants b,, appear as the coefficients in the series expansion of/ in
terms of the eigenfunctions (9. I I). To determine these coefficients, we make use
of the following fundamental property of the eigenfunctions (9. 11 ). If 111 and 11
are nonzero integers, then
(when /11 =/= n)

L . mnx . mrx d
(9.16)
Jo
Sl11 -
L
5111 - .
L
x
(when 111 = n)
Jn the first case (m =/= 11) we say that the functions Xm(x) = sin(nrnx/L) and
X,.(x) = sin(nnx/ L) are orthogonal on the interval 0 s x s L. This ortho-
gonality properly actually holds for every pair of eigenfunctions corresponding
to distinct eigenvalues of any Sturm-Liouville problem and is established in
Section 5 of the next chapter. In the present case, however. the validity of
(9.16) follows immediately from the trigonometric identity
. nrnx . nnx
= cos
(111 - 11 )nx
- cos
( 111 + 11 )nx
2 Sl11 . 5111 - .
L L L L
We therefore multiply both sides of equation (9.15) by sin(nrnx/L) and integrate
the result from x = 0 to x = L. Assuming that the series can be integrated
term by term, we obtain
(9.17) L1c ) . nm.: d x -- ~ b JL . nrn.: SJl1

. nnx d x.
J x
0
Sll1
L
J....,
n= I
II
0
Sll1
L L
According to (9.16), each term on the right-hand side of (9.17), except the term
corresponding ton = m, vanishes. Hence, equation (9.17) reduces to
L f( X) Sl11
. 111iIX - L bm
J o L
-
2
which gives the formula
(9.18) b,. =
2 JL f .(x) s111. lliIX
dx (11 L 2, ... )
L o. L
for the coefficients in (9.15).

Thus, our formal solution of the problem (9.1). (9.2), (9.3) is given by (9.14),
where the coefficients b11 are given by (9.18). The series (9.14) is called the
eigenfunction expansion of the solution u.
The series in (9.15) for the function f with b,, gi,en by (9.18) is called the
Fourier sine series off on the interval 0 s x s L. and the constants b11 are
known as the Fourier coefficients off with respect to the eigenfunctions (9.11 ).
It will be shown in Chapter 5 that whenf is continuous and piecewise smooth on
0 s x s L with f(O) = f(L) = 0, the Fourier sine series of f converges
uniformly and absolutely to f on 0 s x s L. Under such conditions. the series
(9.14) then converges uniformly to u(x, t) for 0 s x s L, t : '.".: 0. Since each
term of the series is continuous for 0 s x s L, I ::'.".: 0, the same is true of 11,
and hence u satisfies the initial conditions (9.2) as well as the boundary conditions
(9.3). However, in order for u to be twice contiP11ously differentiable, we 1eed
Sec. 9 1\Jethod of Separation of Variables 119
more stringent conditions on/ because otherwise the partial derivatives of u may
fail to exist. In such a case, it is comenient to regard 11 as represented by (9.14)
as a generalized solution of the problem in the sense described in Section 2. We
illustrate these remarks by the following example.

11,, - Uu + II = 0 (0 < x < n, I > 0)
u(x, 0) = f(x), u,(x, 0) = 0 (0 :S x :S n)
11(0, t) = 0, 11(n, /) = 0 (t ~ 0)
where
(0 :S x :S n/2)
/(x) = jx
\n - x (n/2 :S x :S n)
So!11tio11: We compute the Fourier coefficients b,, of I By (9. J 8), with L = 11, we
have
b,, = 2 Jrr j(x)

. sin. nx dx
71 0
2 frri2 x sin nx dx + 2 Jrr (n - x) sin nx dx

71 0 n rr/2
~ + ~~IX]rr/
2
[-=X"_COS /IX
71 II II- o
+ -2 [ - (n - x) cos nx _ sin :zx]"

,
71 11 11- rr/2
4 . /lT[
... - sm
1111 2 2
in which sin(im/2) = 0 v,hen 11 = 2k, and sin(im/2) = ( - JJ'+ 1 when /1 =

2k 1, k = I, 2,. . . . Thus, by (9.14), our formal solution is
4 oc (_I l'+ 1
~ + 1]
2 12
(9.19) u(x, t) = - , sin(2k - l)x cos [(2k - I) 1 t
11 k= 1 (2k - I)-
The series solution (9.19) of the preceding example can be shown to converge
uniformly for 0 :o:; x ~ 11, t ;::: 0 by the Weierstrass M-test, using the convergent
~eries L.:'= 1 I /(2k - I ) 2 . Hence, the limit function u is continuous for
0 ~ x ~ rr, t ;::: 0, and satisfies the initial and boundary conditions of the
problem. However, the second partial derivatives of u do not exist because the
second derivatives of the series fail to converge. For this reason, 11 cannot be a
solution of the differential equation. In this case, it is appropriate to regard
(9.19) as a generalized solution of the problem in the sense of Section 2.
Indeed, for each integer n ;::: I, let us define
4 n (-J)'+l
u11 (x, t) = -
rr
L (2k
k= 1 - 1) 2
sin(2k - l)x cos[(2k - 1) 2 + 1] 1 12 t
...
120 The JYave Equation Chap. 3
Then 11 is the limit of the uniformly convergent sequence u,,. Further, for each
n, is twice continuously differentiable for 0 ::::; x ::::; L, t ~ 0, and satisfies
11 11
the differential equation together with the initial conditions
. 4 n ( _ I )k + 1 .
u,lx, 0) = JnCx) =
7'[
L
k~1 (2k -
0
1)-
s111(2k 1)x
ou,, (x, 0) = g,,(x) = 0

ct
and the boundary conditions 1111 (0, t) = u,,(TC, t) = 0. Since the sequences f,,
and g,, com erge uniformly to the respective initial data f and g, it follows by
definition that (9. I 9) is a gen~ralized solution of our problem.
Exercises 3. 7
In Problems I through 5, find the solution of the given problem by the method of
separation of variables.
1. 11,, - ilxx = 0, 0 < x < t > O; u(x, 0) = sin 3 x, u,(x, 0) = 0, 0
iT, s x s ir;
11(0, 1) = 0, ll(iT, t) = 0, t 2: 0.
2. 11 11 - ilxx + 11 = 0, 0 < x < ir, t > O; u(x, 0) = 0, 11,(x, 0) = sin 3 x, 0 s x :S
ir; 11(0, I) = 0, il(iT. t) = o. t 2: 0.
3. ilrr - llxx + 11 = 0, 0 < x < ir, t > O; 11(x, 0) = 0, u,(x, 0) = I + cos 3 x,
0 :S X S ir; = 0, llx(iT, 1) = 0, f 2: 0.
llx(O, t)
4. llrr - u'-' - 2u, = 0, 0 < x < I, I > O; 11(x, 0) = e-x(2 sin 2rrx - 3 sin 5nx),
0 :S x :S 1 ; 11,(x, 0) = 0, 0 :S x s l ; u(O, t) = 0, 11(1, t) = 0, t 2: 0.
5. ilrr + 11, - llxx =0, 0 < x < n, I > 0: 11(x, 0) = 0, 11,(x, 0) = sin 3 (x/2),
0 s x s ir; 11(0, 1) = 0, uJn, t) = 0, t 2: 0.
In Problems 6 through 9, find all particular solutions of each of the following partial
differential equations satisfying the given initial and boundary conditions.
6. 11 11 - x 2u'-' - Xllx = 0, I < x < e. I> O; 11(x, OJ= 0, 11(1, !) = 0, uJe, 1) = 0.
7. llrr - 11.u + 211 1 - 211x + II = 0, 0 < X < ir, 1 > O; 11,(x, 0) = 0, u(O, t) = 0,
U(iT, 1) = 0.
8. ilrr - x 2 ilxx = 0, I < X < 2, I > 0; u(x, 0) = 0, fl( 1, I) = 0, 11(2, I) = 0.
2
9. ilrr -
2
[x /(I + I) ](ux.J 0, I < x < 2, I > O; u(x. 0) = 0, 11(!, t) = 0,
11(2, t) = 0.
Chapter 4
Green's Function and

Stur111-Liouville Prob] en1s
In Section 9 of Chapter 3 we introduced the method of separation of variables

to solve an initial-boundary value problem for the damped wave equation. We
saw that the method leads to the consideration of two ordinary differential
equations, both involving a parameter i.. In contrast to an initial value problem
in which values of the unknown function and its derivatives are prescribed at a
single point, we found that the solution of the differential equation in the space
rnriable has to satisfy conditions prescribed at each end of the interval in which
the solution is sought. Such conditions are called boundary conditions, and
problems of such a type are generally known as boundary value problems. Jn
this chapter we shall study some of the properties of boundary value problems
for linear second-order differential equations. Our discussion of these problems
will lead us to some important concepts. such as Green's function and eigen-
function expansion of arbitrary functions. which play a fundamental role in the
application and extension of the method of separation of variables to more
general problems in partial differential equations.
I. Homogeneous Boundary Value Problems
Consider the general linear second-order differential equation

(1.1) A(x)u" + B(x)u' + C(x)11 = g(x)
on the interval a < x :-:::; b, where we assume the functions A, A', B, C, and g
are continuous and A(x) > 0 for all x in the interval. If A' = B, then we can
write the equation in the form
d- lA(x) du] + C(x)u g(x)

dx _ dx
121
122 Green's Function and Sturm-Liouville Problems Chap. 4
called the self-adjoint form. With our assumptions on the functions A, B, and C,
it is always possible to transform equation (I. I) into an equivalent differential
equation that is in self-adjoint form. Indeed, if we multiply both sides of
equation ( 1.1) by the function
1
A
exp(J B
A
ilx).
the equation becomes
(1.2) Lu
d
dx
lp(x) d111 +
dx
q(x)u = f(x)
where
~ dx) , ( , g(x)
p(x) exp(J q XJ = C(x) p ( X,) f(x) = - p(x)
A(x) . A(x)
Clearly, the functionsp,p', q. and/are again continuous andp(x) > 0 on the

interval a :::; x :::; b. The self-adjoint differential equation (1.2) is known as an
equation of Sturm-Liouville type. For the most part in this chapter, we shall be
concerned with this equation.
In this section we consider the problem of finding a solution of the homo-
geneous differential equation
(1.3) Lu = dxd l p(x) duJ

dx
+ q(x)11 = 0 (a :::; x :::; b)
which satisfies the homogeneous boundary conditions
B 1 (u) = xu(a) + f3u'(a) = 0

(1.4)
B 2 (u) = )'U(b) + <5u'(b) = 0
The coefficients x, /3, /', and <i are real constants such that Y. 2 + /3 2 i= 0 and
/ + <5 2 i= 0. Clearly. the problem ( 1.3), ( 1.4) has at least one solution, namely,
the trivial solution u = 0. However, this solution has little practical importance.
The significant question is whether the problem (1.3), (1.4) has any nontrivial
solution and. if so, to find all such solutions.
Thus, let 11 1 and u2 be any two linearly independent solutions of equation (1.3)
on the interval a :::; x :::; b. That such solutions exist follows from the theory of
ordinary differential equations, with our assumptions on the functions p and q.
Suppose that u is a nontrivial solution of the problem (I .3), (1.4). Then there
exist constants c 1 and c2 , not both zero, such that
(1.5) (a :::; x :::; b)
Since 11 satisfies the boundary conditions (1.4) and because of the linearity of
these conditions, we see that
B1(11) C1B1(ll1) + C2B1(ll2) = 0

( 1.6)
Sec. I Homogeneous Boundary Value Problems 123
a system of homogeneous algebraic equations in c 1 and c 2 . The fact that c 1

and c 2 are not both zero requires that the determinant of coefficients must
vanish; that is,
, )I
(1.7) 'B1(U1) B1(U2 ' = 0
B 2 (u 1 ) Bi(u 2 )
Hence, if this condition does not hold, then problem (1.3), (1.4) has only the
trivial solution.
On the other hand, if condition (I. 7) is satisfied, then the two equations in (1.6)
differ only by constant factor. Hence, we can use either one of the equations to
derive a relationship between the constants c 1 and c 2 , and in this way obtain a
nontrivial solution of the problem. For instance, from the first equation we may
take c 1 B 1 (u 2 ) and c 2 = -B 1 (u 1 ). Then
(l.8)
is a nontrivial solution of the problem (1.3), (1.4). From the linear homogeneous
nature of the problem it follows that r(x) = Cu(x) is also a solution for any
value of the constant C. Thus, we obtain an infinite family of nontrivial solu-
tions. Jn fact, it is true that ifthe problem (1.3), (1.4) has a nontrivial solution u,
then all other solutions of the problem are of the form v(x) = Cu(x), where C
is a constant.
To verify the last statement, suppose that v is any other solution of the
problem. Since both u and t' satisfy the first of boundary conditions (1.4), we
have
r111(a) + /Ju'(a) = 0
C1.u(a) + f3v'(a) = 0
which is a system of homogeneous equations for the constants Cf. and /3. By
assumption, Cf. and f3 are not both zero. Hence, the determinant of the system
must vanish; that is,
!u(a) u'(a)!
(1.9) w (u, v; a) = Ii (a) v'( a) ' = o
1
This determinant is precisely the Wronskian of u and v evaluated at x = a.

Now we recall that whenever the Wronskian of two solutions of equation (1.3)
\anishes at a point in a :S x :S b, it vanishes identically on that interval.
Therefore, (1.9) implies that W(u, v; x) = 0 for a :S x :S b, which in turn
implies that u and v are linearly dependent; that is, i:(x) = Ci(x) for some
constant C.
We summarize the results we have obtained above in the following theorem.
THEOREM 1.1. A necessary and sufficient condition for the problem (1.3), (1.4)
to hare a nontrii'ial solution is that for any two linearly independent solutions of
(1.3). the condition (1.7) holds. Jn such a case, all solutions of the problem are
giren by i(x) = Cu(x), where u is any nontrivial solution of the problem and C is
m1 arbitrary constant.
124 Green's F1111ctio11 and St11r111-Lio11ville Prohlem.1 Chap. 4
Example I. I. Consider the boundary \alue problem
// = 0. 11(0) = o. 11( ;y) 0

The general ,o!ution or the differential equation is
11(x) = c, sin x C2 COS X
Since the determinant

sin 0 cos 0
= 0
sin Ti co~ 7i.
the problem has nomri\ ial solutions. In fact, applying the boundary conditions
to the general solution. we find that c 2 = 0 and c 1 is arbitrary. Hence, the
solutions are gi\en by
11(x) = C sin x
\\here C is an arbitrary constant.
Example 1.2. Consider the problem
11" + ~II= 0, 11(0) = o. 11( ;y) 0

The general solution of the equation in this case is
11(x) = c 1 sin }x + c 2 cos lx

Since the determinant
sin 0 cos 0
-I
cos
2 2
the problem has only the trivial solution 11 = 0. In fact, if we apply the boundary
conditions to the general solution, we find i;rnnediately that c 1 = c 2 = 0.
Exercises 4.1
Jn Problems I through 5, reduce the given differential equation to the self-adjoint

form (1.2).
1. + bu'+ rn = 0, 0 S x S 1, band
11" care real numbers.
2. xu" + 2u' + x11 = 0, 0 < a S x S b.
3. xu" + x11' + u = 0, 0 < a S x S b.
4. x 2 u" + x1/ + u = 0, 0 < a S x S b.
5. xu" + (1 - x)1/ + 11 = 0, 0 < a S x S b.
In Problems 6 through 14, determine if the problem has a nontrivial solution and,
if so, find the solution.
6. 11" = 0, 0 s x s I; (a) 11(0) + u'(O) = 0, 11(1) = O; (b) u(O) + u'(O) = 0,
u'(I) = 0.
7. 1/' + u = 0, 0 S x S rr/2; 11(0) = 0. 1/(11/2) = 0.
Sec. 2 So11ho1110.qeneous Problems; Gree11's Functio11 125
8. 11" II= 0.0 s x s 1;11(0) + 11'(0) = 0;1111) = 0.

9. 1/' 411 = 0. 0 :o: x s 1; 11(0) + 1/(0). 11( 1) = 0.
10. 1/' + 411 0. 0 s x s n; 1i'(OJ 0. 11(nJ = 0.
11. xu" + 1/ 0, 1 ~: \' :::= 2; 1111) 0, 11(2) - (111 4)11'(2) = 0.
12. x 11" + xu' - 411 = 0. 1 s .\ s 2; 211(]) +
2 1/(I) 0, 11(:?.) + 1/(2) = 0.
13. x 2 u" - 2.rn' + 211 0. 1 s .Y s 2; 11(1) 1/(l) 0, 11(2) 21/(2) = 0.
2
14. (d/dxJ[(l - .\ )1/] 0. 0 s .\ < I; 11(0) = O. 11(1) is fimte.
15. (Nonhomogeneous boundary conditions.) Consider the boundary value problem
U11) = 0, B 1(!1) = c 1 , B 2 (11) == <'>\\here Lis the operJtor (1.2) and B 1 . B 2 are
the boundary operators gi\en in (l .3). Let 11 1 and 11 2 be two linearly independent
solutions of the equation L{11) = 0.
(a) Show thJt a necessary and suf!lcient condition fLlr the problem to ha\e a
unique solution is that the determinant
D = !B1<111J B1C112J le O
B2<111 J B2(112)
(b) If D = 0. sho\\ that the problem has a solution only if the constants c 1 and
c 2 satisfy the equations
Ci B2ill2) C2B1(ll2) 0
Ci B2(ll1) C2B1ill1) 0
In Problems 16 through 2:1. use the result of Problem 15 to determine if the problem
ha-; solution and, if so. find the solution.
16. 11" + 11 = 0. 0 s .1 s_ ;:; (a) 11(0) = 1, 11lrr) = -1; (b) 11(0) + 1/(0) I,
11lr:) = 1: (c) 11(0) - 11'(0) = l, 11lr:J = 0.
17. 1/'-11=0, Os.1sl; la) 11(0)=1, 11(1)-1/(lJ=O; (bJ 1/(0) 0,
11(1) + 11'(1) ~I; (c) 11(0) = -J, 11(1) + 11'(1) = 0; (d) 11(0) - 1/(0) ")
~,
1dl) + 11'(1) = 0.
18. 11" - 411 = 0. 0 :S .\ l;la)1110) =- 1.11(!) = O;ibl11WJ = 1,211(1) + 1/il)
2c 2 .
19. 1/' 411 0.0 s x::; ;:;la)11(Q) = -l,11(;:) = l;(b)1110) = -l,11irr) = 1;
le') 11(0) 1/(0) -' I. 111;:) - 1/I;-:) = 0.
20. \II"+ 11' = 0.1 :c; x s 2;1a)11(J) == 0.1i'r2) = J:(b)1111) = 1.1112) - 1/12)
0, iC) 1/(I) = 1. 1112) = 0.
21. .\' 2 1/' + .rn' + 11 = 0, I::: x:::: e;(a)11(l) = l,11'(e) O; (b) 11( I) - ;/(I) 1.
11IC') ~cos l;(c)11llJ + 1i'(l) = l,11(e) =sin I.
2. ~onhomogeneous Problems; Green's Function
We now consider the nonhomogeneous differential equation

d - du- .
(2.1) Lu = p(x) + q(x)u = f(x)
dx 1_ dx 1
126 Green's Function and St11rm-Lio11ville Problems Chap. 4
where f is a given continuous function on the interval a s x s b_ We seek a

solution of this equation that satisfies the homogeneous boundary conditions
(2.2) u(a) = 0, u(b) = 0

We have taken here the simpler boundary conditions (2.2) mainly for con-
' enience. The results that will be obtained remain valid for the general boundary
conditions (1.4), as we shall indicate at the latter part of this section_
We first consider the problem (2.1 ), (2.2), assuming that the corresponding
homogeneous problem
(2.3) Lu = 0, u(a) = 0, u(b) = 0
has only trivial solution. As we shall see in the next section, when the problem
(2.3) has a nontrivial solution, (2.1), (2.2) either has no solution or has infinitely
many solutions_ Our procedure here is to find a particular solution of equation
(2.1) by the method of variation of parameters and then to determine the
constants of integration in such a way that the boundary conditions (2.2) are
also satisfied. With this purpose in mind, we shall choose two linearly inde-
pendent solutions of the homogeneous equation Lu = 0 in such a way that each
satisfies one of the boundary conditions (2.2)_ Thus, let 11 1 be a nontrivial
solution of Lu= 0 that satisfies the condition u 1 (a) = 0, and let 11 2 be a non-
trivial solution of the equation that satisfies the condition 11 2 (b) = 0. It is clear
that such solutions exist. In fact, if y 1 and y 2 are any two linearly independent
solutions of the equation Lu = 0. then we may take, for example, 11 1 (x) =
y 2 (a)y 1 (x) - yJa)Yl(x) and uh:-) = y 2 (b)y 1 (x) - y 1 (b)y 2 (x). Since the
corresponding homogeneous problem (2.3) has only trivial solution, we have
u 1 (b) # 0, u 2 (a) # 0 and the functions 11 1 and 11 2 are linearly independent.
Hence, the Wronskian W(u 1 , u 2 ; x) of u 1 and 11 2 does not vanish; that is,
(2.4)
for a s x s b. Now let

(2.5)
where r 1 and i- 2 are functions yet to be determined. Differentiating (2.5) and
setting
(2.6)
we have
(2.7)
Differentiating (2.7) and substituting the result together with (2.7) and (2.5) in
the differential equation (2.1 ), we obtain
f(x)
(2.8) v;(x)11;(x) + v;(x)u;(x)
p(x)
Sec. 2 No11ho111oge11eo11s Problems; Gree11 's F1111ctio11 .127
for which we have used the fact that Lu 1 = 0 and Lu 2 = 0. In view of (2.4),
we can solve for i; and z:;
from equations (2.6) and (2.8). We find
i:',(x) = -u 2(x)f(x)
p(x)W(u 1 , u 2 ; x)
r;(x) = u 1(x)f(x)
p( X) W ( U I , U 2 ; X)
and therefore
V1(x) = -Ix c,
Uz(Of(() d( -
p(')W(11 1 , 11 2 ; ()
(2.9)
v2(x) = 'x 11,(()/(() d(
~ c p( () W ( 11 I , 11 2 ; ()
2
\\here c 1 and c 2 are arbitrary constants. Substituting these functions in (2.5),

\\e obtain a particular solution of (2.1) given by
(2.10) 11(x) = U1(x) fx ~U2(0f(() d_ + 11z(X) fx - ~1(~)[(0 d( y

c p( S) w(U I , !/ 2 ; S)
1 c2 p( S) J1 (U J , U2 ; S)
We now determine the constants c 1 and c2 in (2.10) so that u satisfies also the
boundary conditions (2.2). Setting 11 = 0 for x = a and noting that 11 1(a) = O,
we find c 2 = a. In the same way, setting 11 = 0 for x = b and noting that
u2 (b) = 0, we ha\e c 1 = b. Substituting these values for c 1 and c2 in (2.10), we
finally have
(2.11) ( ) - rx" p(---

ux- u,()u2(x)f(O cc+ /" J" 111(x)112()/(() d" l
() W (LI I , 11 z; () - p(O W ( 111 , U2; ') -
x
for the solution of the problem (2.1 ), (2.2).

If we define
!11(X)ll2()
(a :s; x :s; ()
. pk)W(u 1 , 11 2 : ()
(2.12) G(x: ) = f
then the solution (2.11) can be written simply as

l 111(()11z(X)
p(()W(11 1 , u 1 ; ()
(2.13) 11(x) =

r
(J
G(x; ()/(() d(
The function G defined in (2.12) is called the Green's function for the differential
operator L, corresponding to the boundary conditions (2.2).
It is left as an exercise to verify that (2.13) with (2.12) indeed satisfies equation
(2.1) and the boundary conditions (2.2). The uniqueness of this solution follows
from the assumption that the corresponding homogeneous problem (2.3) has
only trivial solution.
The definition of the Green's function given above holds for the general
second-order differential equation (I. I). However, the fact that the operator L
is self-adjoint leads to the important result that the quantity p(x)W(u 1 , u 2 ; x)
is a constant. To see this, we note that
d [p(x)u 1(x)u;(x)] - d [p(x)u;(x)u 2 (x)]

dx dx
u 1(x) { d [p(x)u;(x)] + q(x)ui(x)lS

dx
- u 2 (x) {~ [p(x)u;(x)] + q(x)u 1(x)}

dx
= 0
since both u 1 and u 2 satisfy the equation Lu = 0. Hence, for the self-adjoint
operator L, the Green's function (2. I 2) becomes
(2.14)
G(x; 0 ~ 1u,(x~,m (a ~ x ~ ~)
u 2 (x)u 1 (c;)
(~ ~ x ~ b)
K
where K = p(x) W(u 1 , u2 ; x) is a constant.
As can be easily verified from (2. I 2) or (2. I 4), the Green's function possesses
the following properties:
(i) For each . the Green"s function satisfies the equation
d p dG)- + qC = 0 (a < x < b, x of ~)

dx (. dx
(ii) Greens function is continuous at x = ;that is,
(2.15) cc.; + O; ) = G( - O; )
(iii) Green's function satisfies the boundary conditions of the problem;

that is, G(a; ) = 0, G(h; ) = 0.
(iv) The derivative of the Green's function is discontinuous at x = , where
it has a jump of magnitude
:t(J ( + O; ~) _ dG ( _ O; ~) = _1 _
dx dx p(~)
(v) For a self-adjoint operator, Green"s function is symmetric with respect

to the variables x and ~; that is,
G(x; ~) = G(; x)
Sec. 2 So11ho111oge11eo11s Problems; Gree11 's F1111ctio11 129
Actually. properties (i) through (iv) determine the Green's function uniquely.
This means that if \\e construct a function that satisfies all the properties
(i)-(iv), the function will be of the form (2.12) or of the form (2.14) if the
differential equation is self-adjoint.
Let us summarize in the following theorem the result we have obtained above.
THEOREI\1 2.1. If the homogeneous problem (2.3) has 011ly the tricial solution,
then the Green's jimctio11 (2. I 2) exists and the 11011homoge11eo11s problem (2.1 ),
(2.2) has a unique solution gil'en by the formula (2.13).
Example 2.1. Find Green's function and the solution of the problem
d 2 11
f (x) (0 s x s ])
dx 2
11(0) = 0, 11(1) = 0
Solution: It is easily verified that the corresponding homogeneous problem has only
the trivial solution, and so the Green's function exists. The differential equation
1/' = 0 has the general solution
11(x) = Ax+ B
A solution that satisfies the condition 11(0) = 0 is easily found to be 11 1 (x) - '
and a solution that satisfies the condition 11( I) = 0 is given by u 2 (x) = x - 1.
These two functions are linearly independent, with the Wronksian being equal to
Hence, by (2.12), Green's function is
G(x; )
fx<.; - 1) (0 s x s )
$x - I) ( s x s 1)
and the solution of the problem is
11(\') =
0
c
0
G<x; )/(.;) dC,
In particular, if/(.$ 1, we find u(x) = x(x - I )/2, which is readily verified as

the solution.
Example 2.2. Find Green's function and the solution of the problem
d 2 11
+ 11 = f(x) (0 S x S I)
dx 2
11(0) = 0, u(I) = 0
Solution: The general solution of the corresponding homogeneous equation is
11(x) = A cos x + B sin x
For this solution to satisfy also the boundary conditions, we find A = B = O;

thus, the corresponding homogeneous problem has only trivial solution. To
130 Green's F1111ctio11 and Sturm-Liouville Problems Chap. 4
construct the Green"s function, \\e take 11 1 (x) = sin x and 11 2 (x) cos 1 sin x -
sin I cos x = sin(.\ - 1) so that u 1 (x) = 0 and 11 2 (1) = 0. The Wronskian of
these functions is
W(11 1 . 11 2 ; x) = 11 1(x)11;(x) - u;(x)u;(x)
sin x cos(x - I l - cos x sin(x - I)

sin [x - (x - 1)] = sin I
Hence. by (2.12), the Green"s function is
C(x; ,) = {sin x ::~(; - 1)

(0 :<'. x :<'. c;)
sin sin(x - 1)
(, :<'. x :<'. I)
sin l
and the solution is
u(x) = f C(x; c;J/() di;
In particular, if/(x) = I, we find
sin(x - I) - sin x
(X
/I )= - - - + 1
sin l
The solution of the boundary value problem consisting of equation (2.1) and
the more general boundary conditions (1.4) can also be given in the form (2.13),
where G is the corresponding Green's function for the problem, provided the
related homogeneous problem has only trivial solution. In such a case, the
Green's function can likewise be constructed from properties (i), (ii), and (iv)
of (2.15) with property (iii) replaced by the conditions
cxG(a; 0 + {3G'(a: () 0
yG(b; () + 15G'(b; () 0
Here we use prime to denote differentiation with respect to the variable x. Thus,
the Green's function in this case is also defined by (2.12), provided u 1 and u2
are linearly independent solutions of Lu = 0 such that 11 1 satisfies the boundary
condition at x = a and u 2 satisfies the boundary condition at x = b.
Example 2.3. Find the Green's function for the problem
q_~u2 = f(x) (0 :<'. x :<'. I)

dx
11(0) = 0, u(l) + 11'(!) = 0
Solution: It is easy to verify that the related homogeneous problem has only the
trivial solution u = 0. Now, from the general solution u(x) = Ax + B of the
homogeneous equation 11" = 0, we determine two linearly independent solutions
Sec. 2 So11/wmoge11eo11s Problems; Green's F1111ctio11 131
11 1 and 11 2 that satisfy the boundary conditions of the problem al x = 0 and

x = I, respectively. We find 11 1 (x) = x and u 2 (x) = x - 2 for which
Hence, by (2.12), the Green's function is
(0 <:'. x <:'. )
G(x; )
( <:'. x <:'. 1)
It is easily seen that this function satisfies properties(!), (ii), and (iv) in (2.15) and
the boundary conditions of the problem. Moreover, G(x; c;) = G(; x), since
(d/dx) 2 is self-adjoint.
The solution of the boundary value problem
(2.16) Lu = f(x), u(a) = A, u(b) = B
which involves nonhomogeneous differential equations and boundary condi-

tions. can be found by combining the solution of problem (2.1), (2.2) and the
solution of the problem
(2.17) Lu= 0. u(a) = A, u(b) = B
Thus, we need only to find the solution of problem (2.17). With the functions 11 1
and 11 2 used to construct the Green's function (2.12), let us assume a solution in
the form
u(x) = C1U1(X) + C2ll2(X)
where c 1 and c 2 are constants to be determined. Applying the boundary

conditions and noting that u 1 (a) = 0, u 2 (b) = O. we obtain
u(a) = C2ll2(a) = A. ll(b) = C11l1(b) = B

Since
ll1(b) -:/- 0 and 11 2 (a) # 0
we find
B A
cI - -- C2
u 1(b) u 2(a)
Therefore, the solution of problem (2.16) is
(2.18) u(x) = Jb G(x; cJf(O d~ + B 11 i(x) + - _A u 2(x)

a 11 1(b) ll2(a)
Example 2.4. Find the solution of the nonhomogeneous problem
d211
+II=
I
dx 2
uiO) =A, 11(1) = B
132 Green's F1111ctio11 a11d Sturm-Liouville Problems Chap. 4
Solution: Referring to Example 2.2, we rake u 1 (x) = sin x and u 2 (x) sin(x - I).
Then, according to (2.18), the solution of the problem is
11(x) =o .t " 1 (<:;)

C;(x; <;). " c<;
/ + .B . x -
sm A .
. sm(x - I)
J0 sm 1 sin 1
\\here C is the Green's function determined in Example 2.2.
Exercises 4.2
1. find the Green's function for the operator L = d 2 /dx 2 , 0 ::; x :S I, subject to
the given boundary conditions: (a) 11(0) = 0, u'(OJ = 0; (b) 11'(0) = 0, 11( I)
u'(l) = O; (c) u(O) + 1/(0) = 0, 1/(I) = 0. '
2. Find the Greens function for the operator L = (d 2 /dx 2 ) + 1, 0 s x s IT,
subject to the given boundary conditions: (a) 11(0) = 0, 1/(n) = O; (b) 11'(0) = 0,
u(rr) = 0.
Jn Problems 3 through 6, find the Green's function for the problem and give the
solution.
3. 11" - 2u' + s x :S n/2; u(O) = 0, u(n/2) = 0.
211 = f(x), 0
4. x 2 1/' - + 2u = f(x), 1 s x :S 2; u(l) - u'(l) = 0, 11(2) = 0.
2.11/
5 . .\ 2 11" - .rn' + u = /(x), 1 s x :S 2; 11(1) - 1/(l) = 0, 11(2) + 21/(2) 0.
6. (d/dx)[(x + l)u'] =/(x),O s x s l;u(O) = O,u'(l) = 0.
7. Verify that the function u defined by formula (2.13) satisfies the differential
equation (2. J) and the boundary conditions (2.2).
8. Determine the Green's function (2.12) from the properties (i)-(iv) in (2.15).
Hint: Let u 1 and u 2 be two linearly independent solutions of L11 = 0 such that
u 1 (a) = 0 and u 2 (b) = 0, and set
(as x < ,;)

(,; s x :S b)
9. Use the definition (2.12) and verify that
(a) (cG/c,;)(x, a) satisfies the equation (2.1) and the conditions
cG -I cG
- (a, a) = - - (b, a) = 0
c,; p(a), c,;
(b) (cG/c,;)(x, b) satisfies the equation (2.1) and the conditions
cG (b, b) = +I , oG
- - (a, b) = 0
a,; p(b) a,;
10. From the results of Problem 9 show that
u(x) = Jb. G(x, ,;)/(,;) d,; - Ap(a) ~~ (x, a) + Bp(b) ~~ (x, b)

is a solution of the nonhomogeneous problem (2.16).
Sec. 3 \Jodijied Gree11 's F1111ctio11 LB
11. \erif:;, that the solution gi\en in Proble111 10 agrees 11i1h (2.18).
12. Find the solution of the proble111
II -f- 411 /(1) IO::::xs I)

u(O) A, 11(1) B
13. F111d the solution of the problem
11" + /.: 2 11 = f(x) (0 .S 1 S I)
11(0) - 1/(0) == A, u(!) B
11here A and Bare not both zero. and determine the 1alucs of k for 11hich the
problem has no solution.
In Problem 14 through 16, find the Green's function for the gi1en operator subject
to the gi\en boundary conditions.
14. L x(d 2 /d1 2
) + (d d1l. 0 < .1 < I. 11(0) is finite. 11(1) == 0.
2
15. L x (d ,'d.r 2 )
2 2
, . .r(d,d.1) - 11 , 11 > 0. 0 < x < l; 11(0) is finite. 11(1) 0.
2 2 2
16. (I x )(d 1d1 ) - 2x(d/d1l. 0 < .r < l; 11(0) = 0. 11(1) is finite.
17. Show that the Green's function for the operator
d2
L = - s2 (- ~;: < .1 < x. (s real constant > 0))
with the boundary conditions 11 0 as x _, x is gi1en by
3. :vtodified Green's Function
In the preceding section 11e were able to obtain a unique solution of the
problem (2.1 ). (2.2) under the assumption that the related homogeneous
problem (2.3) has only trivial solution. Linder that assumption. the Green's
runction (2.12) for the problem is uniquely determined. We now consider the
case 11 hen the homogeneous problem (2.3) has a nontrivial solution. In such a
case. the Green's function (2.12) docs not exist. since the functions 11 1 and u 2
can no longer be linearly independent. In fact, we shall show that the problem
(2. l ). (2.2) in this case either has no solution or has infinitely many solutions.
We first establish the following lemma.
LE\ll\JA 3.1. (Green's Formula) !f 11 and t are t11ice co11ti1111011sly diflerentiah!c

on the intl'rrnl a ::; x ::; b, then
(3.1) f [uLr: - iLu] dx

l ( cfr
p 11
dx
- z: du)Jb
dx a
134 Green's Function and Sturm-Liouyil/e Problems Chap. 4
Proof: We ha\e
a
b
l
l
u -d _ (. p
dx .
dr) + qr J dx
dx
_
(3.2)
b 11 - d- (p cit:) dx + rb q11r dx
-a l dx , dx. a
r
Integrating by parts twice the first integral term on the right of (3.2). we find
b u d_(P di:) dx = p fu d11b - p d11 ch dx

l
a dx . dx l dx a a dx dx
lp(u cir_ r du)Jb + f\- _d (.p du)

dx dx a a dx dx
dx
Substituting this in (3.2) we obtain

b
/ 11Lr dx
a
rb I (- dr t111)Jb
.I a rLu dx + Ip 1/ dx - [' dx L
which yields (3.1 ).
Now let 11 0 be a nontri\ ial solution of the problem (2.3) and suppose that the
problem (2.1 ). (2.2) has a solution u. Let us multiply both sides of equation (2.1)
by 11 0 and consider the integral
(3.3)
By Lemma (3.1) we ha\e
(3.4) f 11 0 L11 dx = f 11L11 0 dx + l P (u ~~0 - 11 ~~o) I

Since both 11 and 11 0 \anish at x = a and x = b, and Lu 0 = 0. the right-hand
side of (3.4) reduces lo zero and hence, by (3.3), we have
(3.5)
a
r llof dx = 0
This shows that when (2.3) has a nontrivial solution 11 0 equation (3.5) must hold
in order that the problem (2.1 ). ( 2.2) may have a solution. Therefore. if condition
(3.5) is not satisfied. then the problem (2.1 ). (2.2) has no solution whatsoever.
In other words, (3.5) is a necessary condition for the existence of a solution of
the problem (2.1 ), (2.2) in the event that the related homogeneous problem (2.3)
has a nontrivial solution 11 0 .
The condition (3.5) turns out to be also sufficient to ensure the existence of a
solution of the problem (2.1), (2.2). As a matter of fact, if (3.5) holds, we can
Sec. 3 Jl,fodijied Gree11's Function 135
construct a modified Green's function analogous to (2.12) such that a solution of

(2. I), (2.2) may be given in the form (2. I 3). Observe, however, that there is no
longer uniqueness of solution, since we can always add to the solution any
constant multiple of 11 0 . Before we proceed to the determination of a modified
Green's function, let us consider a simple example to illustrate the various
points pointed out above.
Example 3.1. Find the solution of the nonhomogeneous problem
(3.6)
11(0) = 0, 11(1) = 0
Solution: A particular solution of the differential equation is easily seen to be

up(x) = 1/n 2 ; therefore the general solution of the equation is
I .
11(x) = - + c 1 sm nx + c 2 cos nx
n2
Applying the boundary conditions of (3.6), we find
1 I
2 + Cz = 0, 2 - Cz = 0
n n
\\ hich yield contradictory results for c2 . This shows that the problem (3.6) has
no solution. As a matter of fact, we observe that the corresponding homogeneous
problem 11" + n 2 u = 0, 11(0) = 0, 11(!) = 0 has a nontrivial solution u 0 (x) =
sin nx for which J6 I sin nx dx # 0, violating condition (3.5).
On the other hand, if we take the non homogeneous term in (3.6) to be f(x) =
2x - I so that
L 1
(2x - I) sin nx dx = 0
then by a similar elementary method we find infinitely many solutions given by
I 2x - I .
11(x) = cos nx + - + C sm nx
112 n2
where C is an arbitrary constant.
Incidentally, the preceding example showed that whenever a particular solu-

tion of the differential equation can be found by some other method, it is
possible to treat a nonhomogeneous boundary value problem without resorting
to Green's function.
We now proceed to the construction of a modified Green's function for the
problem (2.1 ), (2.2) in the case that u0 is a nontrivial solution of the related
homogeneous problem (2.3). We assume that condition (3.5) holds. Let u 2 be a
solution of the equation Lu = 0, which does not satisfy any of the boundary
'
conditions (2.2). Then u0 and u 2 are linearly independent (see Problem 7).
Hence, by (2.10),
(3.7)
is a solurion of equation (2.1) for arbitrary constants c 1 and c2 . Here K =

p(x)W(11 0 , 11 2 : x); this is a constant, since L is self-adjoint. Applying the
boundary conditions (2.2) and noting that 11 0 satisfies those conditions, we find
'a Uo(J/() d
(3.8) 11o(a) = 0,
- l
c2 K
Since u 2 (a) of. 0 and u 2 (b) /' 0. equations (3.8) will hold if and only if the integral
terms \'anish for some value of c2 . In \iew of condition (3.5), it suffices to take
c 2 = a or c 2 = b. leaving the other constant c 1 arbitrary. Let us choose
c 1 = c2 = a. Then the solution (3.7) becomes
(3.9)
.
u(x) =
Ix 11 0 ()uh:) -- 1lo(x)u2(~) /'("1..) d,c..
a K . . -
It is easy to verify that (3.9) is a solution of the problem (2.1). (2.2), subject to
the condition (3.5). This solution is not unique because \\e can add to (3.9) any
constant multiple of u 0 . In fact, if we add the term
( ) J'b 112((}((;)
u0 x -- - f" =- u (' x)
l '= 0
rx Li2(~)f(() d"c;
a K a K
~b
c
... /'(.
j -=-U-~l d~
ti, (
+ u 0 (x)
x K
to the solution (3.9). we obtain
u (x ) -- 1x 11o(()112(X)
- ~-- 1("') d"c,
a K
(3.10)
b
J G*(x;cJ/(()d(
a
where
(3.11)
(x :::; ~ :::; b)
The function G* defined in (3.11) is called a modified Green's function. It is

easily seen that G* possesses all the properties of a regular Green's function
listed in (2.15). except property (iii). Here, L must be self-adjoint. and hence K
is a constant. It is clear that G* is symmetric with respect to x and~.
Let us summarize the result we have obtained above in the following theorem.
Sec. 3 .'vfodified Green's Function 137
THEORE.l\I 3.1. fl rhe homogeneous problem (2.3) has a 11011/ricial solution 11 0 .

the11the11011homoge11eous problem (2.1 ), (2.2) has a solution if and 011ly if condition
(3.5) holds. If (3.5) holds. rhen a solution of the problem (2.1 ). (2.2) is gicen bP
(3.10).
Example 3.2. Find the solution of the problem 11" + n 2 11 = /(x), 11(0) = 0, 11(1) = 0,
assuming that
jI
0
/(x) sin nx dx = 0
Sol111io11: The function 11 0 (x) = sin nx is clearly a nontrivial solution of the corre-
sponding homogeneous problem 11" + n 2 u = 0, 11(0) = u(I) = 0. Let us seek a
solution of the homogeneous differential equation that does not vanish at x = 0
and x = 1. We find 11 2 (x) = cos nx, for which we have K = uau; - 11 011 2 = - n.
Hence, according to (3.1 J),
1 . -
- ; Sll1 l1X COS m:;
G*(x; ) f
1 . ..
1 l[
Sll1 n<;; COS l1X ( s: x s: I)
and the solution of the problem is
11(x) = Lr G*(x; )/() d~ + C sin nx
for any constant C.
Exercises 4.3
In Problems 1 through 5, find a modified Green s function for the given operator L
subject to the gi\en boundary conditions.
1. L d 2 /d.\ 2 , 0 s: x s: I; 11'(0) = 0, 11'(1} = 0.
2. L 2
d /dx 2
, 0 s: x s: I; 11(0) + 1/(0) = 0, 11(1) = 0.
3. L d 2 /dx 2 + J, 0 S: x S: rr/2; 11(0) = 0, u'(n/2) = 0.
4. L 2
x(d /dx 2
) + (d/dx), 0 < x < I; 11(0) is finite, u'(l) = 0.
5. L = (I - x 2 )(d 2 /dx 2 ) - 2x(d/dx), - I < x < I; /1 is finite at x = 1 and
x = -1.
6. Suppose the problem Lu = 0, 11(0) = 0, 11(b) = 0 has a nontrivial solution.
Let 11 1 and u 2 be two nontrivial solutions of Lu = 0 such that 11 1(a) = 0 and
u 2 (h) = 0. Show that u 1 and 11 2 must be linearly dependent.
7. Let Ila be a nontrivial solution of the problem L(u) = 0, u(a) = 0, 11(b) = 0, and
let u be any solution of the equation U11) = 0. Prove that 11 and 11a are linearly
dependent if and only if u satisfies one of the conditions 11(a) = 0, 11(h) = 0.
8. Verify that for the modified Green's function given in (3.1 I), G* satisfies pro-
perties (i), (ii), and (iv) of (2.15) but not property (iii).
138 Green's F1111ction and St11r111-Lio11ville Problems Chap. 4
9. Let 11 0 be a nontrivial solution of the problem Lu = 0, 11(a) = 0, u(b) = 0

(a s x s b) such that J~ 116(x) dx = I. Show that the problem Lu = 11 0 ,
11(a) = 0, u(b) = 0 has no solution.
10. Determine \Vhether the problem r/' - 211' + 11 = 1, 0 < x < I ; 11(0) = 0,
2u(l) - 1/(1) = 0 has a solution.
11. Obtain the necessary condition for a solution of the problem
A(x)z/' + B(x)z/ + C(x)u = f(x) (a s x s b)

u(a) = 0, u(b) = 0
to exist, assuming that the related homogeneous problem has a nontrivial
solution.
4. Sturm-Liomille Problems
ln Section 9 of Chapter 3 we introduced the method of separation of variables

to find particular solutions of type X(x)T(t) of the damped-wave equation that
satisfy homogeneous initial boundary conditions. The method leads to the
boundary value problem
(4.1) X"(x) + l.X(x) = 0, X(O) = 0, X(L) = 0
involving a parameter). for the function X. We saw that nontrivial solutions of

this problem exist only for certain exceptional values of the parameter )..
When the method is applied to more general boundary value problems in
partial differential equations, there frequently arise boundary value problems
imolving ordinary differential equations of the type
(4.2) d
dx
rl p(x) ~dx ]
11
+ [q(x) + l.r(x)]u = Lu + l.r(x)u 0
together \Yith the boundary conditions

(4.3) 'Y.u(a) + {Ju'(a) = 0, '/U(b) + 6u'(b) = 0
Here, 1. 1s a parameter and r is a function that is assumed continuous and
positive on the interval a :s; x :s; b. As before, the functions p, p', and q are
assumed continuous, and pis positive on this interval. Under these conditions,
the boundary value problem (4.2), (4.3) is known as a regular Sturm-Liouville
problem. This is distinguished from the case when p or r \anishes at some point
in the interval [a, b J or when the interval is of infinite length, in which case the
problem is called a singular Sturm-Liouville problem.
Other important boundary conditions that may also arise with the differential
equation (4.2) are
(4.4) u(a) - u(b) = 0, u'(a) - u'(b) = 0
These are called periodic boundary conditions, since both the solution u and its
derivative are required to have the same value at the end points of the interval
[a, b].
Sec. 4 Sturm-Liouvil/e Problems 139
In this discussion we shall be concerned mainly with regular Sturm-Liouville

problems. We shall establish several important though elementary results in the
general theory, which will be useful in later discussion. As will be seen in the
exercises, some of these results hold also in the singular case and in the case when
the boundary conditions are periodic.
We note that problem (4.2), (4.3) constitutes a homogeneous boundary value
problem of the type considered in Section 1 and is therefore satisfied by the
trivial solution u = 0 for any value of the parameter /.. However, as we saw
in the special case (4.1), the existence of nontrivial solutions depends on the
particular value of/.. Indeed, let u 1 = u 1 (x, }.) and u 2 = u 2 (x, I.) be two linearly
independent solutions of equation (4.2). Notice that u 1 and u 2 depend on both
x and /. Then, by Theorem 1.1, nontrivial solutions of (4.2), (4.3) exist if and
only if the determinant of coefficients ( 1. 7) vanishes: that is,
cxu 1(a, ).) + f3u'1(a, ).) cxu 2 (a, ),) + /3u~(a, ).)! =
0
(4.5)
yu 1(b, /,)
1 + bu'1(b, ),) yu 2 (b, ).) + bu~(b, i)
Hence, the problem (4.2), (4.3) has nontrivial solutions if and only if;, satisfies
the determinantal equation (4.5). The values of). satisfying equation (4.5) are
called eigenvalues of the problem (4.2), (4.3) and the corresponding nontrivial
solutions are called eigenfunctions. It follows from Theorem 1.1 that an
eigenfunction corresponding to an eigenvalue is uniquely determined up to a
constant factor. This means that to each eigenvalue there corresponds only one
linearly independent eigenfunction. An eigenvalue having this property is said
to be simple.
It is possible to show that a regular Sturm-Liouville problem has infinitely
many real and simple eigenvalues/.,., n = l, 2, , which can be arranged in a
monotonic increasing sequence ). 1 < 1. 2 < < ).,, such that lim i.,, = co as
n tends to co. These properties are clearly exemplified in the particular problem
(4.1) for which we found infinitely many real and increasing eigenvalues
which are all simple. We consider two more examples as further exhibits of
these properties.
Example 4.1. Find the eigenvalues and.eigenfunctions of the problem

u"(x) + ).11(x) = 0
11(0) = 0, 11'(11) = 0
So/11tio11: As in problem (4.1 ), it is readily shown that the present problem has only
trivial solution when Jc :S 0. When )_ > 0, the differential equation has the
general solution
u(x) = c 1 sin v ). x + c 2 cos\ ). x
Applying the first boundary condition, we obtain c 2 = 0. Thus, any eigen-

function of the problem must be of the form 11(x) = c 1 sin v'). x.
Applying the second boundary condition, we find c 1 VJ. cos V ), n 0.

Since we do not v,ant c 1 = 0 and since i. > 0, we must then have cos "\ 1 J. n 0.
This implies that
(11 I, 2, ... l
Hence, the eigenvalues of the problem are
which are real and increasing, and the corresponding eigenfunctions are
un(x) = . (211 -
Sll1 .
2
I) X (11 = 1, 2, .. )
except for constant factors.
Example 4.2. Find the eigenvalues and eigenfunctions of the problem 11"(x) +
i.11(x) = 0, 11(0) = 0, 11(11) - 1/(11) = 0.
Solution: The reader can easily verify that the problem has only trivial solution when
2
i. :; 0. Let i. > 0 and write i.= k , where k is a real nonzero number. Then the
general solution of the differential equation is
u(x) = c 1 sin kx + c 2 cos kx
The first boundary condition requires c 2 = O; therefore an eigenfunction of the

problem must be of the form
u(x) = c 1 sin kx
Since v,e do not \\ant c 1 = 0, application of the second boundary condition

leads to sin kn - k cos kn = 0. Thus, the eigenvalues of the problem are
square of the roots of the preceding transcendental equation. Let us write the
equation in the form
tan kn = k (k =f. 0)
Then the roots can be determined approximately by a graphical method in the

folio\\ ing manner: We plot the graph of the functions y = tan kn and y = k in
the same k, y coordinate system and determine the abscissas of the points of
intersection of the two curves (see Fig. 4.1 ). We exclude k = 0 in this considera-
tion because we have assumed k to be nonzero. Thus, from the figure, we see that
k 1 ;:: 1.3 so that the first eigenvalue is approximately i. 1 ;:: 1.69. For very large 11,
we see that k" is approximated reasonably well by (211 + I )/2; hence, ).,, ;:
2
(211 + 1) ;4. The corresponding eigenfunctions are given by lln(x) = sin J.nx.
Note that while the eigenvalues i." may also be determined from the negative
values of kn, no new eigenfunctions are derived by considering -kn, since the
eigenfunctions are determined to within a constant factor.
Sec. 4 Sturm-Liouville Problems 141
FIG. 4-1 L'iqe11rn/11es determined hy graphical method.
Exercises 4.4
111 Problems 1 through 'i. find the eigenvalues and eigenfunctions of each of the
gi\en Sturm-Liou\ ille problems.
1. I/ " + I.I/ 0. 0 s x s: L; 1/(0J = 0. u(L) 0.
2. // " + I.Ii o. 0 < .\ s: n; 1/(Q) - 0. 1/(;:) 0.
3. 11 " + I.I/ 0. 0 s: x s: j( ; 11(0) + 1/(Q) o. 11( Ti) T 1/(n) 0.
4. II + /.// o. 0 x 5- ;:-; u(O) 1/(0) 0. //(Ti) 1/(7r) 0.
5~ 1/i\) - I.I/ - 0. () :S \' :S L; 11(0) = u"(OJ = 0, u(L) - 1/'(L) 0.
For each of Problems 6 through 10. find the equation (4.5) and estimate graphically
the eiger1\alues for sufficiently large /1 and give the form of the eigenfunctions.
,.
6. Ii + /.// -- 0,0 s: x s: 1; u(O) + u'(O) = 0, u(l) = 0.
7. [/ " + /.Ii 0, 0 x s: IT; 11(0) - 0, ll(li) T 11'(n) = 0.
8. II " T }u 0, 0 < x < 1; 1/(0) - 0, 11( I) - 1/(I) - 0.
9. ll " + /Ji 0, 0 s: x s: 7I; 11(0) + 11'(0) = 0, u'(n) 0.
10. // " + )_// 0. 0 s x s: "' 1/(0) + 11'(0) = 0, u(n) u'(n) = 0.
11. Find the eigenvalues and eigenfunctions of the Sturm-Liouville equation
d ).
(xu') + 11 = 0 ( l S: x s: e)
dr x
ror each of the boundary conditions (a) 11(!) = 0, 11(e) = O; (b) 11(!) = 0,
1/(e) = O; (c) 1/(l) = 0, 1/(e) = 0.
12. Find the eigenvalues and eigenfunctions of the Sturm-Liouville problem
x 2 11" + 2x11' + ).11 = 0 (I :s: x :s: e)
II( I) = 0, u(e) = 0
d (x 3 11') + ).XII = 0
dx
11(!) = 0, u(e") = 0
d
dx
(1 ') +
- II
x
t. II
_\3
= 0 (I :s: x :s: e)
11(!) = 0, 11(e) = 0
15. Consider the boundary value problem
11" + ;_// = 0 (-L :s: x :s: L)

11(-L) - u(L) = 0, 1/(-L) - 1/(L) = 0
\\ ith periodic boundary conditions. Find the eigenvalues of the problem and
shO\\ that to each eigell\alue). F 0, there corresponds two linearly independent
eigenfunctions.
16. Find the eigenvalues and eigenfunctions of the problem
11" + i.11 = 0 (0 < _\ :s: L)
11(0) - u(L) 0. 1/(0) - 11'(L) 0
\\ ith periodic boundary conditions.

17. The function
( - I )k(tl2)2-cp
k!l(k+p+1)
is known as Bessel's function of the fast kind, of order p, and satislies the
differential equation
0 ( p a real number)
called Bessel"s equation.

(a) By introducing the new variable x defined by l = \ 1. x, show that the
equation becomes
2
Af11 =dxd (xu') + (i.x - P ) 11 = O
x
where 11(x) = y(\ ). x).

Sec. 5 Orthogo11ality of Eige11fimctio11.1 143
(b) Thus. show that 11 11 (x) = Jr(\ i.,, x) is an eigenfunction of the Sturm-Lioll\ illc
problem
Afu = 0. 11(0) = o. 11(/) = 0 ( p > 0. 0 < .\ < I )
when /. 11 is a root of the equation Jr(\ /.) = 0.

18. \\hat are the eigell\ alues and eigenfunctions of the singular Sturm-Liom ille
problem
xr/' + 1/ -,- i.xu = 0 (0 < x < I)
1/(0) = 0, 11(!) = 0
19. Let J 1 ; 2 (1) be the Bessel's function of order p = -i and let u(x) = xJ 112 (('' l./2)x" )-
Show that 11 satisfies the differential equation
Lu = dxd ( x1 1/) + J.
x
11 = 0 (x > 0)
and thus determine the eigell\alues and eigenfunctions of the singular Sturm-
Liouville problem Lu = 0, 11(0) = O. 11( I) = 0.
5. Orthogonality of Eigenfunctions
In relation to their corresponding eigenvalues, the eigenfunctions of a regular

Sturm-Liouville problem possess a basic property that is of fundamental import-
ance in application. We have obsened this in Section 9 of the preceding chapter
i11 connection with the eigenfunctions X (x) = si n(mrx)/ L and eigell\ al ues
in = (nIT) 2/I.3. n ~ I. In this section we shall establish this property in the
general case and in later sections consider some of its implications.
We now introduce the definition of orthogonal functions. Let u and c be two
functions that arc integrable on the intenal a ::;: x ::;: b. and let a be a positive
continuous function on this interval. The functions u and c arc said to be
orthogonal on [a. hJ with respect to the \\eight function u if and only if
(5.1) rb a(x)11(x)1(x) dx = 0
Thus, the functions u(x) = sin x and c(x) = cos x arc orthogonal on the
interval [-IT, rr] with respect to the weight function O"(x) = I, since the integral
of their product O\er this integral vanishes. Whene\er (5.1) holds for the
runctions u and r with O"(x) = I. we shall simply say that 11 and care orthogonal
on [a, b].
When u c, the quantity
(5.2) I rill =
(I Jb a(x)u 2(x) dx)\ ,_
a
1n
is called the norm of the function u. Notice that Ii ul! = 0 if and only if u(x) = O
for a ::;: x::;: b. If 111111 = I, then the function u is said to be normalized. It is
easy to see that a given function that is not identically zero can always be
144 Green's F1111ctio11 a11d Sturm-Liouville Problems C!tap. 4
normalized. Thus, if 11 is a function such that llull # 0, then the function

v = 11/lluil has a unit norm. Indeed, we see that
rb o-(x)v 2(x) dx = {b o-(x)u 2(x) dx

Ja lluli 2 Ja
1; 11 !12
2
=I
If 11 i
We now establish the orthogonal property of eigenfunctions of the Sturm-

Lioll\ ille problem (4.2), (4.3).
THEORP'I 5.1. Let u and 1 be tHo eige11f11ncrions of the Sturm-Lioui:il!e problem

(4.2), (4.3) corresponding to the eigenrn!ues 11 and\'. If 1i and v are distinct, then
u and i are orthogonal on [a, b J 11ith respect to the 1\'efqht fimction r; that is,
f r(x)u(x)v(x) dx = 0
Proof: Since u and i: satisfy equation (4.2) corresponding to ). = p and

/. = 1, we have
Lu = -1-;ru, Ll' = -1'1'L'
If we multiply the first equation by 1 and the second equation by u, and then
subtract one result from the other, we obtain
(J1 - 1)rul' = uLz. - rlu
Let us integrate both sides of the last equation from x = a to x = b and use
Lemma 3.1. We find
(5.3) (1-1 - v) f r(x)u(x)i:(x) dx

[.
p ( !I -dv
. dx
- v du)]b
dx . a
p( b) iv (u, r : b) - p( a) W ( u, r: a)
where W(11. 1: x) = 11(x)1"(x) - u'(x)l'(x) is the Wronskian of 11 and r. From

(l.9)weknowthat W(u, l': a)= O,andinasirnilarmannerwefind W(11, c: b)=
0. Therefore, the right-hand side of equation (5.3) reduces to zero and we obtain
b
(5.4) = 0
(11 - v)
J r(x)u(x)v(x) dx
a
Since Ji - v # 0. we conclude that
f r(x)u(x)v(x) dx = 0
and the theorem is proved.

It follows from this theorem that the eigenfunctions of the Sturm-Liouville
problem (4.2), (4.3) form a set of mutually orthogonal functions on the interval
Sec. 5 Ortho_qonality of Ei{fe11ji111ctio11s 1-45
a ::::; x :::::. b with \\eight function r. Such a set of functions is called an ortho-
gonal system on the interval [a. b] with respect to the weight function r. If
r = I. then the system is simply said to be orthogonal on [a. b]. Thus. the set of
functions X,,(x) = sin(nnx)/L. n = I. 2... forms an ortl1ogonal system O\er
the interval 0 :::::. x :::::. L.
If each eigenfunction 11 11 of (4.2). (4.3) 1s normalized--that is. if we replace
each u,, by c/J,, = 11,,11 11,,:i-then the
1 alternate set of eigenfunctions c/J,'" 11 :2: J,
satisfies the relation
(5.5) f r(x)c/J 111 (x)c/J 11 (x) dx = !~ (iI

(if
/11
/11
11)
=I= n)
In such a case the set of functions c/> 11 n :2: J. is called an orthonormal system
over [a. bJ with respect to the weight function r. Equation (5.5) is oftentimes
written as
(5.6) L'" r(x)</>,,,(x)</>,,(x) dx = c5 11111
where the symbol 611111 , known as the Kronecker delta, is defined by the equation
(i c /11 11)
(5.7)
(ifm =I= 11)
Example 5.1. Determine the normalized eigenfunctions of the problem t4.1).
So/11tio11:The eigenfunctions of problem t4.I) are X,J\) = sin(ni7.\)/L /1 2: I. for

which we find
'("nnx)
l
L
o
.
s11r -
L
I
1x
= 11. _!___=-._~)_'~2nn:\}_L dx = ~
0 - -
Thus, l!X,,11 = \.0_(2 and the normalized eigenfunctions are gi\en by
J (2) I 2 . 17 Tl.\"
u 11 (.\) = - Sill - (11 2 ])
L L
\Ve stated in the preceding section that the Sturm-Lioll\ille problem (4.2),
t4.3) has only real eigenvalues. This fact can be established by using the proof
of Theorem 5.1.
TIIEORE.\I 5.2. The eige11rn/11es of the St11r111-Lio11rille proble111 (4.2). (4.3) are
all real and the corresponding eiqenfunct ions are real except f(Jr a constant
( po1.1ibh complex) factor.
146 Green's F1111ctio11 and St11rm-Lio11ville Problems Chap. 4
Proof: Suppose i. = s + it is an eigenvalue of the problem (4.2), (4.3) and

suppose t/J = U + iV is the corresponding eigenfunction. Then i/1 satisfies the
(5.8) Lt/I + i.rt/J = 0

Since the function rand the coefficients of the differential operator Lare real-
valued, it follows by taking the complex conjugate of each term in (5.8) that
(5.9)
Similarly, by taking the complex conjugate of each of the boundary conditions
(5.10) ry_t/J(a) + /N'(a) = 0,

we obtain
(5.11) :J.lji(a) + (Jlji'(a) = 0,
since :J., fi, /',and c) are real constants. Now equation (5.9) and (5.1 I) imply that
17f is also an eigenfunction of (4.2). (4.3) and I. is the corresponding eigenvalue.
Hence, Jetting 11 tjJ and r = lji in the proof of Theorem 5.1, from (5.4) we have
(5.12) (). - I.) r

a
r(x)t/J(x)17f(x) dx 0
Since 1/1(x)17f(x) U 2 (x) + V 2 (x) > 0 and r(x) > 0 for a ::;; x ::;; b, the
integralin(5.12)isalwayspositive. Therefore.wemustha\e i. - I.= 2it = O;
in other words, r = 0. Thus, i. = s, a real number.
If we consider the real and imaginary parts of equations (5.8) and (5. I 0), we
see that both U and V are eigenfunctions corresponding to the eigenvalue
i. = s. Hence, C = c V for some constant c so that
i/1= U+iV=(c+i)V
Thus. 1/; is a real-valued function except for a multiplicative (possibly complex)

factor. This completes the proof of the theorem.
Theorems 5. I and 5.2 can also be shown to hold in the case when the boundary
conditions (4.3) are replaced by the periodic boundary conditions (4.4), provided
p(a) = p(b) (Problems IO and 13.)
Exercises 4.5
I. Determine the normalized eigenfunctions of Problem .1. Exercises 4.4.

2. Determine the normalized eigenfunctions of Problem 5, Exercises 4.4.
3. Determine the normalized eigenfunctions of Problem 15, Exercises 4.4. (See
Problem 11 be!Olv.)
Sec. 5 Orthogonality of Eigenjimctions 147
4. By applying Theorem 5.1 to the eigenfunctions of the problem 11" + ).11 = O,

= 0, u(n) - 1/(n) = 0, show that
0 ::::; x :S n; u(O)
when }. 1 and /. 2 are roots of the equation tan ).n = }..

5. Verify by direct integration that the eigenfunctions of Problem 1 l(a), Exercises
4.4, are mutually orthogonal; that is,
e 1 . ( .
- sm mn In x) sm(mn In x) dx = 0
Jxl
when m and 11 are distinct integers.

6. Verify that Theorem 5.1 holds for the singular Sturm-Liouville problem
d (x1/)
-- + ( }.x - p2) u = 0 (0 < x < 1, p > 0)
dx x
u(O) = 0, 11(1) = 0
and thus deduce that
when }. 1 and .Jc 2 are distinct roots of the equation Jp(Vl) = 0. (See Problem 17,
Exercises 4.4.)
7. Do as in Problem 6 for the singular Sturm-Liouville problem
d
- -
dx
(1x ') + x).
II ll =
0 (0 < x < 1)
11(0) 0, 11(1) = 0
and deduce that
when and v are distinct roots of Jl/2(J.. 112 /2) = 0. (See Problem 19, Exercises
4.4.) .
8. The singular Sturm-Liouville problem
(I - x 2 )u" - 2x11' + }.u = 0 (-l<x<I)

11 finite at x = I
has eigenvalues }." 11(11 + I), n = 0, I, 2, ... with the corresponding eigen-
functions
I d"
-- - (x 2 - I)" (11 = 0, I, 2, ... )
2"11! dx"
148 Green's Function anti 5jt11r111-Lio11vifle Problems Chap. 4
called Legendres polynomials. Veriry that Theorem 5. 1 holds in this case, and
thus deduce that
j_
'j
1
P,,,(xJP,,(xJ dx 0
\\hen m fc 11.
9. A sequence or functions (/;I' </!2, '', ,, .. ' dellned on [a, bl is said to be

linearly independent if, for each 11, any linear combination I:L 1 c1J/; 1_(.Y) cannot
\anish identically on [a, b] unless c, ~~ 0 for each k, I <; k::; 11. Show that a
sequence of eigenfunctions is necessarily li1~eariy independent.
10. Show that Theorem 5. 1 remains \a lid when the bouml<uy conditions (4.3) are
rerlaccd by the periodic boundary conditions (4.4) \\ith p(a) ~ p(b).
11. Verify the result of Problem 9 for the eigenfunctions of Problem 15, E\en:ises 4.4,
and then obtain the orthogonal system of eigenfunctions
I 1. Sill
' lliTX
. cos
lliTX I
111 2 IJ
\ L L I
12. Do as in Problem 10 for the eigenfunctions of Problem 16. Exercises 4.4, and
obtain an orthogonal system of eigenfunctions.
13. Show that Theorem 5.2 remains \alid \\hen the boundary conditions (4.3) are
replaced by (4.4), pro\ided p(a) = p(h)
14. Show that if q is nonpositi\e on the intcnai u <; x < h, then the eigemaiucs of
equation (4.2) are nonnegati\ c under each of the folio\\ ing bound;.iry conditions:
(i) Ilia) = 0, 11(/JJ = 0.
(ii) 1d11) = 0, 1dh) ~ 0.
(iii) 1/(a) - Ci 11(a) == 0, 11'(/JJ + C211(h) = 0. Ci 2 0, Cz 2 0.
Hint: /\1uitiply the d1fTerential equation (4 2) by 11 and integrate from x 11 to
.\' =c /J.
15. Show that if II and I' arc eigenfunction-; or the boundary \alue problem
1/"J - i.11 = () (0 \. < LJ

0. 11( L) ~ II" ( L) = 0
corresponding to distinct cigcn\alue'i. then 11 and 1 arc orthogonal on [O. L ].

(Problem 5. Exercises 4.4) "<ote that
/ff{i\') -- rr/ivl =---

d
(111''" - ii [' - 1/1'" -i- 1/'1')
dx
6. Eigenfunction Expan<;ions: l\ lean Conwrgence
In Section 9 or the preceding chapter WC \\Cre faced with the problem of

expanding a gi\en function I in an inflnite series of the form
(6.1) /(x) = I ' '

/J,, S111
lli1.\
n:::: 1 L
Sec. 6 Eigen.fimction Expansiom ,- .\fean CouverlJence 149
where the functions X (x) = sin(nnx/L). (11 2: 1). are eigenfunctions of the
11
particular Sturm-Liouville problem (4.1 ). We found that in order for the

expansion (6.1) to hold, the coefficients b must be given by the formula
11
(6.2)
2
h,, = -
rl. j(x)
. sm. _ 11nx
dx (11 = 1, 2, ... )
L. 0 L
This formula was deri\ed by making use of the orthogonal property of the
eigenfunctions X (x) = sin(11nx/ L), 11 2: I.
11
The series (6.1) is actually just a special case of the problem of expanding a
function fin an infinite series of eigenfunctions of the more general Sturm-
Liouville problem (4.2), (4.3). For if f satisfies appropriate conditions. it is
possible to expand the function in a series of the form
Cf.
(6.3) f(x) = L cn<fJ,/x) (a :<::: x :<::: b)

11= I
where c/J," 11 2: l. are the eigenfunctions of the problem (4.2). (4.3). Here we
assume that the eigenfunctions <P have already been normalized. If the expan-
11
sion (6.3) is possible, the coefficients c,, can then be determined in the same way
we determined h,, for the series (6.1 ). We multiply both sides of equation (6 3)
by r(x)c/J'"(x) and integrate the result over the interval [a. h]. Assuming that the
series can be integrated term by term. we obtain
b x 1b
r
a r(x)f(x)</J,,,(x) dx = ~ c,,.
1 1 0
r(x)c/J 11 ,(x)</J 11 (x) dx
(6.4)
Cf:
== L
,1=1
c11<3"'"
where we have used equation (5.6). Since <5,,,,, 0 when m =I= 11 and ri,,, 11
when 111 11, equation (6.4) yields the formula
(6.5) c,, = f r(x)f(x)</J 11 (x) dx
The series in (6.3) with the coefficients c,, given by (6.5) is called the generalized
Fourier series of the function f with respect to the orthonormal system {c/J,,J.
The coefficients c,, are called the Fourier coefficients. We observe that the series
in (6.3) can always be formed for a given function f whenever the coefficients
(6.5) can be determined. However, there is no guarantee that the series obtained
\\ill converge at any point in the interval a :<::: x :S: b. In fact, e\en when the
series does converge, there is no assurance that it will converge to the function f
on that interval. Until the question of convergence has been established, we
shall simply regard the series in (6.3) as the formal eigenfunction expansion of
/with respect to the orthonormal system {</.i and write 11 }
00
f(x) ~ L c,,,,(x)
11= 1
150 Green's Functions and St11r111-Liouville Problems Chap. 4
For reference, we present a basic theorem that gives sufficient conditions for
the pointwise convergence of the series in (6.3) to.f The theorem will be proved
in the next chapter in the special case where L = (d/dx) 2 for which the ortho-
normal system {,,} consists of trigonometric functions.
THEOREI\1 6.1. Let {,,}, 11 ::=:: I, be an orthonormal system of e(qenfunctions

of the Sturm-Liou1'i!le problem (4.2), (4.3), and let/be a function that is piecell'ise
smooth on the intenal a :S: x :S: b. Then the series in (6.3), whose coefficients c"
are given by (6.5), conrerges to the wlue
f(x + 0) + f(x 0)
2
at each point x for a < x < b. If, in addition, f is continuous on a :S: x :S: band
satisfies the boundary conditions (4.3), then the series (6.3) concerges uniformly and
absolutely tofjor all x, a :S: x :S: b.
Example 6.1. Determine the series (6.3) for the function f(x) x, 0 S x < 71 with
respect to the orthonormal system {(2/71) 112 sin nx].
Solution: By (6.5) we find
(2/71) 112 "

J x sin nx dx
0
= (2/71) 112 l-x cos nx
11
sin
+ __
112
II.\']"
0
,--- cos 1171

- v' 271 (II :2'.: ] )
II II
Thus, we have
x ~ 2
n= I
r (-1)"+ 1 sin nx
X
According to Theorem 6.1, the series converges to the function /(x) = x for
0 < x < 71. At x = 71, we notice that the series converges to zero while/(71) of- O.
At x = 0 both the series and the function converge to zero. Hence, the series
expansion above is valid for 0 S x < 71.
In the study of convergence of the series in (6.3) or, for that matter. of any
series involving orthogonal functions, there is a different kind of convergence
that is often more convenient and appropriate to use. especially when pointwise
convergence can nowhere be attained. We refer to the notion of convergence in
the mean square sense or, briefly, convergence in the mean. We say that the
sen es
00
(6.6) f(x) ~ L c,,/x)

n=l
with coefficients given by (6.5) converges to fin the mean if and only if
(6.7) Jim
m-oo f l b
a
r(x) f(x) - "'
L
n=l
c'n 11 (x) J
2 dx = 0
Sec. 6 Ei_qenfimction Expansion; /Hean Convergence 151
Here, we shall continue to regard the system {"} as the set of normalized
eigenfunctions of the Sturm-Liouville problem (4.2), (4.3). although it can be
any given set of orthonormal continuous functions on a :s; x :s; b relative to a
weight function r, r(x) > 0.
The integral
(6.8) E,,, = f r(x) rf (x)

"'
1 ~1 c .,(x)
11
]2 dx
is a measure of the average error in approximating/ over the interval a :s; x ::::; b
by the mth partial sum
m
(6.9) S,,,(x) = L cn<f> (x)
11= 1
11
and is called the mean square deviation of S"' fromf The vanishing of (6.8) as
m tends to infinity implies that S,,, is close to f for all points in [a, bJ except
possibly for points in a set of intervals whose total length is small. Thus, it is
possible for the series (6.6) to converge in the mean to f without actually con-
verging at every point in the interval [a, b]. This indicates that convergence in
the mean does not imply pointwise convergence. It is also true that pointwise
convergence does not imply mean convergence.
It is significant that if we are to approximate the function f by any linear
combination I:~'= 1 a" 11 (x) of the normalized eigenfunctions 1 , . . . , ,,, in the
sense that the mean square deviation (6.8) is minimum, the coefficients a 11 must
be chosen precisely as the Fourier coefficients (6.5) off with respect to the
system { 11 } relative to the weight function r. Indeed, since the integral
rf(x) ,,
- "~ a 11 ,.(x) J2 dx
1
b
(6.10) = r(x)
E(a 1 , .. , a 111 )
J a
is a function of the coefficients a 1 , , a,,,, we see that in order for E(a,, ... , a,,,)
to be minimum, it must satisfy the equation
~' = 2 'rb r(x) rf(x) - ~

m
anc/Jn(x) J c/J;(x) dx 0
cai ~a n-1
:s; i ::::; m. Now. in view of the orthonorrnality of the system {<fJ 11 }, this gives
a; f r(x)f(x)c/J;(x) dx
= C; (i=l, ... ,111)
which are the Fourier coefficients of/ relative to the system {</J 11 } with weight
function r.
152 Gree11's F1111ctio11 and 5:i't11rm-Lio1nille Problems Chap. 4
To see that these coefficients render the integral (6.10) minimum, let us
expand the integrand in (6.10) and integrate the terms. We obtain
E(a I ' ... a,,,) =

a
r r(x)/2(x) dx
2 n:s=I a,, '" j'" r(x)f(x)qi,.(x) dx

a
(6.11) + 11=1
m
L r
'a
r(x)cp,,(x)cf>,(x) dx
b
J,, r(x)f 2 (x) dx 2 L"'
n::: !
a,,c,,
"'
+ L
11= 1
a~
rb m
L"'
= I r(x)/2(x) dx + ~ (a,, c,,)2 ~ c~
.. a 11- 1 n::::: 1
where we ha\e again used the orthonormality of the system {</>,,]. It is now
apparent from the last expression abo\c that (a 1 , , a,,,) has minimum value
when a; = c; for i = I. , 117. Thus. the mthpartial sum (6.9) really provides
the best approximation to f with the eigenfunctions qJ 1 , cf>"' in the mean
sq uarc sense.
\Vhen a,, c,,. \\e see from (6.8) and (6.11) that
(6.12) 111 = E(c 1 , ... , c,,,) = C

a
2
r(x)( (x) dx L'"
11::;:; I
c~
Since by its definition. ,,, ~ 0, it follo\\S that
( 6.13)
"'
,,:s= c;,::::;
rb r(x)/ 2(x)dx
I
1 a
for any 111. This important result is knO\\n as Bessel's inequality. \Ve notice
that as m increases. the sum on the lef't is always bounded abO\e by the integral
of r/2 O\er [a. b]. Hence. if we assume that/is a function such that the integral
on the right of ( 6.12) exists (e.g .. I is piece\\ ise continuous). then on letting m
tend to infinity. Bessel's inequality becomes
(6.14) ~
"' c;, : : ; {/> r(x)(1(x) dx
II-] ., a
This implies that the series L.::'~ 1 c; comerges and therefore

( 6.15) Jim c,, = Jim
11-+ :1, n_. 'l~
I
..
b
a
r(x)f(x)cp,,(x) dx = 0
No\\, suppose that for a given function/the series (6.6) comerges in the mean.
Sec. 6 Ei_qe1(/imctio11 Expa11sio11; .Hean Com'er_qe11ce 153
Then. in view of (6.12). it fol!O\\S that

x
lirn E,,, = rh r(x)j 2(x) dx L c~ 0
II
n= l
r
or
(6.16) r(x)f 2 (x) dx

a
This is a partieular case of the inequality (6.14) and is called Parseval's equation.
It is readily seen that if Parseval's equation holds. then the series (6.6) converges
in the mean. Therefore, Parseval's equation is a necessary and sufficient condi-
tion for the mean convergence of the series (6.6).
We say that the system of orthonormal functions {c/> 11 ] with weight function r
is complete with respect to a class of functions (e.g., the class of piecewise
continuous functions) if for every function fin this class, the series (6.6) con-
verges in the mean, or equivalently. Parseval's equation (6.16) holds. In terms
of this notion of completeness, it is known that the set of eigenfunctions of the
Sturm-Liouville problem (4.2), (4.3) is complete with respect to the class of
functions that is square integrable on a ~ x ~ b. (A function f is said to be
square integrable on a ~ x ~ b if both f and / 2 are integrable on [a, b].)
J
Thus. if/is a square integrable function on [a. b (e.g . .fis piecewise continuous).
then its expansion in series of eigenfunctions of the problem (4.2), (4.3) with
coefficients given by (6.5) converges in the mean. In this connection, it is interest-
ing to note that a function satisfies significantly much weaker condition for
mean convergence than it ordinarily does for pointwise convergence.
Exercises 4.6
I. If a11 (11 2: I) are constants such that the series of eigenfunctions L,,~= 1 a 11 c/J 11 (x)
converges to/(x), where/(x) = 0 for each x on the interval a S x S b, show that
a 11 = 0, 11 2: 1.
2. Let 11 11 (11 2: 1) be a sequence of eigenfunctions not necessarily normalized and
suppose that
f(x) = ~ a 11 11 11 (x) (a S x S b)
11:...-:J
Obtain formally the formula for the coefficients a11 , 11 2: I.

3. Find the formal eigenfunction expansion of the function /(x) = x 2 , 0 s x < n,
with respect to the system of orthonormal functions {(2/n) 112 sin 11x}.
4. Find the formal eigenfunction expansion of /(x) = x, 0 s x s L, with respect
to the eigenfunctions of Problem I, Exercises 4.4.
5. Find the formal eigenfunction expansion of f(x) = x, 0 s x s n, with respect
to the eigenfunctions of Problem 2, Exercises 4.4.
6. Find the formal eigenfunction expansion of j(x) = 1, 1 s x s e, with respect
to the eigenfunctions of Problem 11 (a), Exercises 4.4.
I54 Green's Function and Sturm-Liouville Problems Chap. 4
7. Obtain the formal eigenfunction expansion of a function I with respect to the

eigenfunctions of Problem 3, Exercises 4.4.
8. Show that the series
~ cf,,,(.1)
n=1 vn
cannot be the eigenfunction expansion of any square integrable function. (Use

Bessel's inequality.)
9. Show that Parseval's equation can be obtained formally from equation (6.3)
by squaring both sides, multiplying by r(x), and then integrating term by term.
IO. Let {</>,,}be a complete orthonormal system on a ~ x ~ b relative to a positive
weight function r. Show that any continuous function f that is orthogonal to all
functions of the system must be identically zero. (Use Parseval's equation.)
I I. Let {</>,, J be a complete orthonormal system of continuous functions on a ~ x ~ b
(relative to the weight function r = I). Show that if f is a continuous function
that is orthogonal to all functions of the set {</> 1 + </> 2 , </> 2 + </> 3 , qJ 3 + </> 4 , . . . ],
then f must be identically zero.
I2. Let {</>11 1 be a complete orthonormal system of continuous functions on a ~
x ~ b, where :</>,,! ~ M for all 11. Find a continuous function f, not identically
zero, that is orthogonal to all the functions cjJ 1 + </> 2 , </> 2 + 2</> 3 , </> 3 + 3<f>4, ... ,
c/J,1 + 11</Jn+l'
7. Nonhomogeneous Sturm-Liouyille Problems; Bilinear Expansion
One important application of eigenfunction expansions concerns finding

solutions of boundary value problems for nonhomogeneous differential equa-
tions. We consider here the nonhomogeneous Sturm-Liou\ille problem
(7.1) Lu + i.r(x)u = /(x)

where / is a given function on the interval a ::;: x ::;: b, together with the
boundary conditions
'Y.u(a) + f3u'(a) 0
(7.2)
}'li(b) + bu'(b) 0
We suppose that the normalized eigenfunctions qJ,, corresponding to the eigen-
values i. 11 (11 ~ l) of the related homogeneous Sturm-Liouville problem (4.2).
(4.3) are known. Then, for ~ach 11 ~ I, we have
(7.3)
We assume that a solution u of the problem (7.1), (7.2) exists and can be
expressed as a series of the form
y;
(7.4) u(x) L CncPn(x)

n=l
(a ::;: x ::;: b)
Sec. 7 .\'onho111ogeneo11s St11r111-Lio11ville Problems; Bilinear Expansion 155
If the series in (7.4) con\erges suitably for appropriate values of the coefficients
c,,, then the function u represented by the series automatically satisfies the
boundary conditions (7.2), since each of the eigenfunctions,, does. It remains.
then, for us to determine the constants c,, such that the series in (7.4) satisfies the
differential equation (7.1 ). Proceeding rather formally, we substitute (7.4) in the
differential equation (7.1) and use (7.3) to obtain
if. y_,
Lu + h(x)u = I c11 Lcf;,.(x) + l.r(x) I c,/p 11 (x)

n= 1 n= I
CfO
I c11 [ - ) . 11 r(x)</J,,(x)]
n= l
if;
+ lr(x) I c,, 11(x)

n= 1
r(x) I"'' c,,(l - ), 11),,(x)

.. n=l
=f(x)
This can be written as
00
(7.5) I c,,(), - ),,,)<fJ.(x) = F(x)

n=l
where we have set F(x) = f(x)/r(x). Let us further assume that the function F
can also be expressed in a series of the eigenfunctions 11 Then
00
(7.6) F(x) = I
n:::; 1
b,,,,(x)
where, by (6.5),
(7.7) b,, = f r(x)F(x)<fJ,,(x) dx = f f(x) 0 (x) dx
Substituting (7.6) in (7.5) and combining terms, we obtain

00
(7.8) I [c.(), - A0 ) - b,,],,(x) = 0

n=I
Since this holds for all x in the interval [a, b], each coefficient of the series must
rnnish; hence, we have
(7.9) (11=I,2, ... )
Therefore, if the parameter)_ in the differential equation (7_ J) is not equal to an

eigenvalue of the corresponding homogeneous problem (4.2), (4.3), we can solve
for en from (7.9) to obtain
(n=l,2, ... )
156 Green's F1111ctio11 and Sturm-Liouville Problems Chap. 4
Substituting this in (7.4). we finally ha\e
cc b11
(7.10) u(x) =
..
L .
~ (!. -
1

/.II)
,,(x)
where the constants h,, are given by (7.7). This is our formal solution of the
boundary value problem (7.1), (7.2). The series in (7.10) is called the eigen-
function expansion of the solution of u. If the function f(x)/r(x) is continuous
and piecewise smooth on [a, b], it is possible to show that (7. l 0) is the one and
only solution of the problem (7.1), (7.2).
Now, if /. is equal to an eigenvalue of the corresponding homogeneous
problem (4.2), (4.3)-say, /. = !.k for some fixed integer k-then when n = k.
equation (7.9) becomes ck 0 = bk. If bk =/= 0, then the coefficient ck cannot be
determined. This means that the problem (7.1). (7.2) has no solution in this case.
On the other hand, if bk = 0, that is, if
(7.11) b, = rb f(x)cpk(x) dx 0
~a
then ck becomes arbitrary and a formal solution of the problem (7. l ), (7.2) is
given by
Cf b11
(7.12) u(x) ,i.,,(x) + cr1'k(x)
11~1 (A - A,J ~
"
'
nk
for any constant C.

Hence, if/. is equal to an eigenvalue i., of the homogeneous problem (4.2),
(4.3), then a necessary and sufficient condition for the non homogeneous problem
(7.1), (7.2) to have a solution is that condition (7.11) holds. This result is in
agreement with our discussion in Section 3.
Let us summarize in the following theorem the results we have obtained
above. (Compare with Theorems 2. l and 3.1.)
THEOREJ\I 7.1. The nonhomogeneous Sturm-Liourille prob/I'm (7. l ), (7 .2) has a

unique solution ghen by (7.10), if and only if!. is not an eigenrnlue of thl' corre-
sponding homogeneous problem (4.2), (4.3). If I. is equal to an eigenralue }.k,
for some fixed k, then the nonhomogeneous problem (7.1), (7.2) has a solution if
and only if f is orthogonal to the corresponding eigenfunction ,, in ll'hich case a
solution is gil'en by (7.12), 1rhich is not unique.
Example 7.1. Find the solution of the nonhomogeneous problem 11" + J,.11 = x,
0 < x < 11;11(0) = 0,11(11) = O;where). fc. 11 2 ,11 = 1,2, ....
Solution: We recall from Example 5. I that the related homogeneous boundary value
problem has the normalized eigenfunctions rp,,(x) = v 2/ir sin nx corresponding
Sec. 7 So11ho111oge11eous .S't11r111-Lio11ville Problems; Bilinear Expu11sio11 157
to the eigenvalues i.,, = 11 2 , 11 2: I. From L\amrle 6.1. the Fourier coetlicients of'
the function/(.1) x \\ith respect to the orthonormal system : rp,,: arc given by
\ !;r
b,, = _, (- I)''' I (11 = I, 2. .l
II
Since i. is not an eigenvalue. by (7.10) the solution is

(- 1 l"~1
11(x) = 2 !; sin nx
2
IJ- l n(i. - 11 )
We have described above the method of eigenfunction expansions for solving

nonhornogeneous boundary \alue problems. It is. of course. also possible to
approach such problems from the standpoint of our discussion in Section 2 and
Section 3, which uses the concept of Green's function. Indeed. let us write
L"' =L+1.r=
. d (p d) +q *
dx dx
where q* = q + i.r. and note that [}' is also self-adjoint, thus possessing
properties similar to those oft he operator L Suppose that i. is not an eigemalue
oft he homogeneous problem (4.2 ). ( 4.3 ). Then. according to Theorem 2. I. the
Green's function corresponding to the operator L'' and boundary conditions
(7 2) exists. This function can be determined by using the formula (2.14) with
11 1 and u 2 now depending on both x and i .. being two linearly independent
solutions or the equation [}'11 = 0 such that they satisfy the boundary conditions
al x = a and x = b. respecti\ely. If C(x: .;: i.) denotes the Green's function so
determined. then by Theorem 2.1 the solution of the nonhomogeneous problem
(7.1), (7.2) is given by
h
(7.13) t1(x) = j G(x: .;: i.)f(;) d.;
u
Thus. in the case\\ here i. 1s not an eigcmaluc. the solution of the problem (7.1 )_
(7.2) has the representations (7.10) and (7.13). If we substitute the formul:.i
(7.7) for h,, in (7.10) and formally interchange the order of summation and
integration. (7.10) becomes
( 7. 14) ) _ ,.,, ~ qi,,(x)rp"(,;) _/.(") I"

u ( .\ - L.. .. :---- - --. " ( "
... u 1! ~- 1 /. - ).11
Jn\ iew of the Uniqueness Ol the solution II in the present Case. We conclude that
the representations (7.13) and (7.14) arc identical_ Hence_ we formally deduce
that
(7.15) G(x: .;: i.) = I

11= J
rf;"(x)q>"(,;)
/. - J. 11
(a .::;; x .::;; b)
The series on the right-hand side of (7.15) gives the eigenfunction expansion of
the Green's function G and is called the bilinear expansion of C. In this form.
158 Green's Function and St11rm-Lio11vil/e Problems Chap. 4
we have a composite expression for the Greens function on the whole interval
a :::;; x :::;; b. and its symmetric property with respect to the variables x and ~
becomes apparent.
It is natural to expect that the bilinear expansion in (7.15) can be established
from Theorem 6.1, it being the generalized Fourier series of G with respect to the
orthonormal system {cp,,]. with coefficients depending on ( and i.. In fact, from
the properties of a Green's function, we know that G is continuous, has piecewise
continuous derivative. and satisfies the boundary conditions (7.2). Hence,
according to Theorem 6.1, G can be expressed in generalized Fourier series of
the form
oc
(7 .16) G(x;(;/.) L c,,cp,,(x)
n=J
where
( 7 .17) c11 = f r(x)G(x; (; l.)c/J,,(x) dx ( /1 :::'.'. 1)
Thus, the bilinear expansion (7.15) will be established if we can show that
(7.18) c ~ (11 :::'.'. 1)

" ). - ).II
Now. from their definitions we have
Lc/J 11 = - i,,r(x)c/J,,(x) (11 :::'.'. 1)

and
LG= -i.r(x)G(x; ~: i) (x -:f. ()
If we multiply the first equation above by G and the second equation by </>,,,
and then subtract one result from the other, we obtain
(7.19) (i. - i. 11 )r(x)G(x: (; i)</>,,(x) = C.iL</> 11 ,,LG
Let us integrate both sides of equation (7.19) over the interval a :::;; x :::;; b.
making an intermediate stop at the point x = (because of the discontinuity of
dG/dx at that point. to obtain
(7.20) (/. - I.,,) r

a
r(x)G(x; (; l.)c/J,,(x) dx r-o (GLc/J,, - c/J 11 LG) dx
+ f" (GLc/J,, - ,,LG)dx

~+ 0
Sec. 7 Nonlwmogeneous Sturm-Liou ville Problems; Bilinear Expansion 159
Applying Lemma 3.1 to each of the integrals on the right side of (7.20), we find
(}. - l 11 ) r
a
r(x)G(x; (; 1.),,(x) dx
'
rp (c d</J,, - cPn dG)l~-o + f p ('c cij>,, - 11 ~G)Jb

l dx dx a l . dx dx ~+o
(7.21)
p(b)W(G, 11 ; b) - p(a)W(G, 11 ; a)
+ p(() 11 ( ( ) r~~ (~ + O; ~; ).) - ~~ (~ - O; ~;Jc)]

where we have noted that the functions p, d</J,Jdx, and G are all continuous at
x = (. Since both G and 11 satisfy the boundary conditions (7 .2), it follows
(as in the proof of Theorem 5.1) that the Wronskian W(G, 11 ; x) vanishes at
x = a and x = b. Moreover, by property (iv) of (2.15),
-dG("c + O; t;;
" /.') - dG(,
- ( - O; t;;
" ).') = - 1
dx dx - p(O
Hence, (7.21) reduces to
which yields (7 .18) in view of (7 .17). This completes the derivation of the
expansion (7.15) from Theorem 6.1.
It follows from (7.15) that if/. = 0 is not an eigenvalue of the homogeneous
problem (4.2), (4.3), then the Green's function G(x; () for the operator L
corresponding to the boundary conditions (4.3) has the bilinear expansion
(7.22)
In the case where i. is equal to an eigenvalue (say, i. = i.k for some integer
k > 0) for which condition (7.11) holds, it is possible to construct a modified
Green's function G ''(x; (; /) along the procedure described in Section 3 so that
a solution of the problem (7.1), (7.2) may also be given in the form (7.13).
However, we shall not discuss this possibility.
Example 7.2. Find the Green's function and its bilinear expansion for the operator
L* = (d/dx) 2 + i. with boundary conditions 11(0) = 0,11(11) = 0, where ..1. is not
an eigenvalue.
So/111io11: We recall that the eigenvalues of the problem L*u = 0, 11(0) = 0,

11(11) = 0 are i. 11 = 11 2 , with the corresponding normalized eigenfunctions
.....
160 Green's Function a111/ Sturm-Liouvi/le Problems Chap. 4
,/x) = \ 2/rr sin nx, 11 2: I. Hence, if ). -,t 11 2 and G is the Green's function
sought, then by (7.15) we ha\e the bilinear expansion
.. , 2 ~ sin 11 sin 11x

G(x; i;: l) = '-' . ,
1[ 11= 1 I~ - II-
On the other hand, using formula (2.14), we t1nd
7I)
(''" '. i. " ' " ' i<(
(0 Sc x Sc ")
\ ;, sin \ 1. n
G(x; ~;}.)
sin \ ) " sm \ /.(x - 7I)
(~ Sc x Sc 7I)
\ i. s 111 \ /. 7I
For i. = 0, which is not an eigenvalue, the Green's function for the operator
L = (d 1dx) 2 can be obtained from C(x: (: i.) by letting i. tend to zero. In fact.
for 0 _::;; x _::;; (, we see that
. sin , i. x sin . Jae.- 7[)

11m.
i.-O
1
, 1i.sin, /.n ;.----1 0 ,, ). . I.. ) , I ..
( 5111 \,' I. 7r / \ I.
Similarly, for ~ _::;; x _::;; rr, we have

-
x(( -
7[
n)
r1111
i. - O
sin \,1 ). f2in ' /.(x
--- r-
,;,sin, i. rr
-
7[)
-
((x -
7[
n)
Hence, the Green's function corresponding to;_ 0 is
j
x((_~)
(0 _::;; x _::;; ~)
7[
G(x: c) =
((x - n)
-~-~--
7[
which has the bilinear expansion

sin_1_i( sin nx
G(x; 0 =
Exercises 4. 7
In Problems I through 7, find a formal eigenfunction expansion of the solution of

each of the follO\\ing problems and state the values of)_ for which the solution exists.
1. 11" + ).u = 2 sin x - 4 sin 3x (0 Sc x s 7<)

11(0) = 0. u(rr) = 0
Sec. 7 .\011ho111oge11eo11s S111r111-Lio11vi/le Problems; Bilinear Expansion 161
2. LI " + /.LI = 2 cos 2x - _, cos Sx 10 ~ .\ ::; ;:)

1/(0) - 0. 1/( 77) - ()
/TX 777x
3. II " I di - 3 cos i- 2 cos (0 < 1 , 11
2 2
11'(0) o. 11(1) - 0
4. II " /.// !:
" I
sin (2k
-
1) x
k=I k 2
LliOI = 0. II' ( 77) 0
5. 1/' + i.u = /(x) iO :S x :S 77)
L1(0) + 1/(0) 0, L1(77) + 1/(77) 0

(See Problem 3, Exercises 4.4.)
6. d (.rn') -t- I. 11 = f(.r) (1 :S x -c eJ

dx x
11(1) = 0. u(e) 0
7. d
dx
(I ')
x
II
.
i- 1.
x3
II= [3 sin(77 In x) - 4 sin(377 In x) ]/x
3
(I :Sx:Sel
11( Il = 0, u(e) = 0
8. Obtain the bilinear expansion of the Green's function for the problem
11" + ).11 /(x) (0 s x s L)
LICO) 0. 11(L) = 0
when ). is not an eigenvalue.
9. Find the Green's function for the operator L = (d 2 /dx 2 ) + i. subject to the
boundary conditions L1(0) = 0, 11'( I) = O. and obtain its bilinear expansion. In
particular, show that
~ ~ sin(11 i}77.; sin( /1 - -})77x fx (x :S ')

n- 11= i (11 - })2 \ (' s x)
10. By using formula (7. I 5), show that the Green's function for the operator
(0 < x < 1)
subject to the conditions u(O) is finite and 11(1) = 0 has the bilinear expansion
E <h().nx) cf,(}.,/,)
n=l ).
2
- }.~
where
1 0 (}.,,x)
cf,().,,x) = and
10 ().,.x)
162 Green's Function and Stur111-Lio11ville Problems Chap. 4
11. Find the Green's function G(x; s; }.) for the operator
L =
d
dx
.
(-i dx
d) + ).x 2
subject to the boundary conditions 11(1) = 0, 1/(e) = 0, and by (7.15) show that
it has the bilinear expansion
G(x; ; }.)
where
sin [( 11 -:\ );r In _\]
J,,(x) =
-sin(n 1J;rlnx,;
provided). #- (11 - 1/2);r. In particular, show that
~ <b,,()m,,(x) = fln x <t s x s sl

;r 2 11=1 (11 - !,-l 2 \In i; ( s x s e)
12. Find the bilinear expansion of the Green's function for the operator
L = d ("
dx dx.
d) - n2x + ;_2x ( 11 positive integer)
subject to the boundary conditions 11(0) = 11(1) = 0, and show in particular
1
[(x/)" - (x)"] (0 s x s i;J
~ m(i.,i;Jmhxl
211
).I
[(~/x)"
!
k= 1 1
- (.r)"] ( s x s I)
211
where
, . ) J,,( ;_,x)
1v(1.,x = and
I J,,(i.kx)
13. Find the bilinear expansion of the Green s function for the operator
d 1d) ;_ 2
L= dxCdx + -~
subject to the boundary conditions 11(0) = 0, 11( I) 0, and show in particular
~ ,iCxl<h,,() f +xi(l s2J (0 s x s i;)

k= J
... ?
1.;. I ~s 2
(1 - x2) (i; s x s 1)
where
_, xl 112 U,,/2)(x 2 )
1v,,(x) =
''x J 112 0.,,/2)(x 2 )'
and 11 2
C;') 0

Chapter 5
Fourier Series and Fourier

Integral
In this chapter we shall be concerned with expansion of functions on finite

intervals in series of orthogonal trigonometric functions. This represents an
extremely important special case of the eigenfunction expansions, which we
discussed rather formally in Section 6 of the preceding chapter. We shall
establish here conditions under which such an expansion is possible. We shall
also be interested in the representation of functions on infinite intervals in
terms of eigenfunctions of certain singular eigenvalue problems.
1. Orthogonal Trigonometric Functions
We recall that the Sturm-Liouville problem

(1.1) u" + ;_11 = 0, 11(0) = 0, u(L) = 0
has the eigenfunctions

. nnx
(1.2) u,,(x) = sm -
L
corresponding to the eigenvalues
L2
(11 = I, 2, ... )
If we replace the boundary conditions in ( l. I) by u'(O) = 0 and u'(L) = 0, then

the new Sturm-Liouville problem has the eigenfunctions
1111X
(1.3) lln(x) = COS --
L
163
164 Fourier Series and Fourier Inteqrnf Chap. 5
\Vhich corre,pond to the eigenvalues
/.II (ll = o. 1. 2, ... )
A.ccording to Theorem 5.1 of the preceding chapter, each of the sets of eigen-
functions giwn b\ ( 1.2) and by ( 1.3) forms an orthogonal system on the interval
0 :::;; x :::;; L. That is.
0 ( /)) =/= 17)
and
'L _ Ill T:X 11 i!.\ d
cos -- cos - - x 0 ( l1l =/= II )
l
0 L L
Now let us consider the set of functions
Ill:.\ 11 lT .\
( 1.4) cos----. Sin - - (11 = I, 2, ... )
L L
which is the collection of the eigenfunctions ( l .2) and (1.3). These functions are
all periodic and have the common period 2L: that is.
cos l1Ti (x + 2L) = cos lliIX

L L
sin l17I (x + 2L) = s111 l1r:x
L L
Cor all 17. We shall show that these functions also form an orthogonal system on
the larger interval - L :::;; x :::;; L. Indeed. for each integer 17 :2: I, it is clear that
/. ri- .
I
1117.\ 11TtX
( 1.5) l cos - - dx = I I Sill - -- dx = 0
'-L L -L L
which shows that the function I is orthogonal to the functions cos(mrx 1L) and
sin(mrxl L) for /1 :2: I on the interval [ - L. L].
Next. let m and /1 be two distinct integers. From the trigonometric identities
cos
))) JI x
L
-- cos
11 Tix
L 2
I cos (Ill
-
L
II )n:x
+ cos
( /)J
+L11)n:xJ
Sill
111 TLX
L
--~ Sill
11;:x
L 2
l (
cos /Jl
-
L
11 )n:x
cos
(m +L11)rrx J
we immediately obtain
(1.6) J L
cos
mnx
-~cos
nn:x
--
d
x 0
-L L L
(1.7) f L . 1111IX . 11TCX

Sill -- - Sll1 - -
dX
0
-L L L
Sec. 2 Fourier Series 165
Formulas (l.6) and ( 1.7) show respectively that orthogonality of the eigen-
functions (1.3) and (I .2) remain valid on the extended interval - L s x s L.
Finally, from the identities
. 17li1X
Sll1 -
L
cos
lli1X
L
=
1
2
.
l
Sll1
(111 - n)n
L
+ .
Sll1
(111 +Ln)i1x]
J
we see that for all integers m and /1
mnx nnx
I
L .
(1.8) s111-- cos dx=O
-L L L
This shows that each of the functions in (l.2) is orthogonal to each of the
fun ct ions in ( l.3) on the interval - L s x s L, and vice versa. Therefore, by
definition, it follows that the functions in (1.4) form an orthogonal system on the
interval - L s x s L.
Note that because of the periodicity of the functions ( 1.4), formulas ( 1. 5)-( 1.8)
also hold over any other interval of integration that is of length 2L. Thus, the
set (I .4) is also orthogonal on the interval a s x s a + 2L for arbitrary
constant a.
It is worthwhile to note that the functions in ( 1.4) actually constitute a set of
eigenfunctions for the eigenvalue problem
(1.9) 11" + lu = O; 11(-L) - u(L) = 0: u'( - L) - u'(L) = 0
whose orthogonality property can be deduced from Theorem 5.1, Chapter 4.

J;1deed, we observe that although the problem (1.9) is not of the Sturm-Liouville
type because its boundary conditions are periodic, nevertheless it has real eigen-
values given by i.n = 11 2 n 2 / L2, 11 = 0, 1, 2, . . . . (Problem 15, Exercises 4.4.)
For each integer /1 ::::: I, the functions un(x) = cos(nnx/L) and L"n(x) =
sin(mrx/L) constitute two linearly independent eigenfunctions, corresponding
to the same eigenvalue i.n = 11 2 rr 2 / L2, which are orthogonal on the interval
- L s x s L. The function 11 0 = 1 is the only eigenfunction associated with
the smallest eigenvalue }. 0 = 0. Now, for the differential equation in (1.9), we
have p(x) = I. Hence, by Problem I 0, Exercises 4.5, the orthogonality of these
eigenfunctions corresponding to distinct eigenvalues follows from Theorem 5.1,
Chapter 4. Thus, the set ( 1.4) is an orthogonal system of eigenfunctions for the
problem ( 1.9). Notice that except for i. 0 = 0, the eigenvalues of the problem
are not simple, as is the case with a regular Sturm-Liouville.
The orthonormal system corresponding to the set (1.4) is given by
I IJITX . llITX
(l.10) - cos
~ . - SIJ1 ( 11 ::::: I)
'L L ~L L
2. Fourier Series
We now consider the problem of expanding an arbitrary function in infinite

series of the form considered in Section 6 of Chapter 4, using the orthonormal
system ( 1.10).
166 Fourier Series nnd Fourier Integral Chap. 5
Let f be a function defined on the interval - L ~ x ~ L. Then the formal

eigenfunction expansion off with respect to the system ( 1.10) is given by
(2.1) f( .\) ~
c0___ _ -,-, ~
/... (
-c" -_COS mrx + c,;; _Sill
. mrx)
-
'\ 2L "= I '\ L L '\. L L
where the Fourier coefficients are given by
C0 = - j- -- J'L f(x) dx
-,,2L -L
L
. mrx
(2.2) .f (x) cos - dx
J-L L
I
L llTCX
c'
II
= f (x) sin dx
-,,! L -L L
for /1 = I, 2, .... lf we incorporate in the formulas (2.2) the common factor
1
1/-,, L appearing in the series (2.1). then we can write the series in the convenient
form
(2.3) f(x) ~ a
-0 + L
"' (
a" cos 11nx
+ b" sin ''- 11-x)
2 11=! L L
where the coefficients a11 and b11 are now given by
an ==
1
L
IL -L -
f(x) cos nnx dx
L
(ll 0, I, 2, ... )
(2.4)
b,, = I- IL f (x) /l7[X
- dx
sin (11 = I, 2, ... )
L -L L
The series in (2.3), with its coefficients given by (2.4). is called the Fourier series
off on the interval - L ~ x ~ L. We notice that this series is precisely the
eigenfunction expansion off with respect to the set ( 1.4) of eigenfunctions, with
the constant I replaced by -}: hence, the coefficients (2.4) are the Fourier
coefficients off with respect to that set.
If f is periodic, of period 2L. then the integraritl-s-in the integrals ( 2.4) are also
periodic, of period 2L. Hence. the interval of integration [ - L. L J can be
replaced by any other interval of length 2L: that is. (2.4) can also be written as
a,, =
J fa+ 2L .
f (x) cos
llTCX
- dx (n = 0, I. 2, ... )
L 0
- L
(2.5)
bII = I r~2L.1c).
x Sill ll7IX,( x
- (II 1, 2, ... )
L 0 L
for any choice of the constant a.
In Section 5 of the present chapter. we shall show that if f satisfies certain
restrictions, then its Fourier series converges pointwise to f on - L ~ x ~ L.
In such case, the function can then be represented by its Fourier series, and thus
the symbol ~ in (2.3) can be replaced by the equality sign.
We observe that whenever the series in (2.3) converges on the interval
- L :::; x :::; L, it converges for all x to a periodic function of period 2L. This is
so because each term of the series is periodic, of period 2L. Therefore, if the
series converges tofon -L:::; x:::; L, it will converge to the periodic extension
off for all x with period 2L. Consequently, iff is defined for all x and is not
periodic, it cannot have a Fourier series representation that is valid for all x.
Example 2.1. Find the Fourier series of the function
/(x) = fO ( - 7[ s x s 0)
\x (0 < x s n)
Sol11tio11: We compute the Fourier coefficients an and bn according to the formulas

(2.4), noting that here L = n. We find
an = ; L' x COS nx dx
1_ (~in /IX + COS /IX) In

7r II 112 ' 0
J COS Im - (- 1)"
7[
II II
for 11 = I, 2, . . . . Thus, the Fourier series off is
/(x) ~ ~ + "- l( - J" -

~
I
?
1
-- cos nx +
( - 1J" +
1
sin nx
]
4 11= 1 mi- n
7[
4
oc
~
11= 1
l n(211 -
2
1) 2
cos(211 - l)x + ( - --
I)" sin
n
11x
]
In Section 5 it will be shown that this Fourier series does converge to the
function on the interval - n < x < n. Hence, outside this interval, the series
converges to the periodic extension off with period 2n. The graph of that
extension is shown in Figure 5.1, where it is seen that the extension is discon-
tinuous at the points x = (211 - I )n, 11 = I, 2, .... At these points, the series
will be seen to converge to the value n/2, which is the average of the left-hand
and right-hand limits of the extended function at the points of discontinuity.
168 Fourier Series and Fourier Integral Chap. 5
i( \]
FIG. 5-1 Periodic f1111ctio11.
Example 2.2. Find the Fourier series of the function /(x) = - 7[ ~ x ::;; 7r.
Sol11tio11: Computing for the Fourier coefficients of the function. we find
I ,-
IT
f 0
__ .. --rr
- x dx + ,-,. x dx
""'o
J n
2
+ n
2
n
a,, J rso
n __ "
-x cos//_\" dx + rrr x
o
cos nx dxj
/IX sin /IX + cos llX '

0
/IX Sin /IX + COS /IX Irr
+
0
2
(( - I)" -- I)
J \_
iI
lO
.. - rr
- x sin 11x dx + Jrr
0
x sin 11x dx l
0
11x cos nY - sin 11x /IX cos /IX - sin /IX
1127l I
0
= 0
for n 1, 2, , . . . Hence,
'le
2 (-I)" -
:x: n
+ l: n-?
cos llX
2 1I n '.:'.: 1
cos(211
l:"'
n 4 - J)x
(-n ~ x ~ n)
2 TC n= I (211 - I )2
lt will follow from Section 5 that this series converges to the function lxl on the
interval indicated and hence to the periodic extension of that function with
period 2n for all x outside that interval (Fig. 5.2). Notice that here the periodic
extension is continuous at all points.
;'( y l
-3rr
FIG. 5-2 Periodic extension of f(x) ~
The Fourier series in (2.3), with coefficients given by (2.4), can be written in
the equivalent form
'fJ
(2.6) f(x) ~ L c,,einrrx'L

n= - x
where the coefficients c,, are given by
(2.7) 'L f(x)e-i11r.x t dx (n 0, l, 2, ... )

2L J-L
Indeed, (2.6) implies that
oc 00
(2.8) f(x) ~ Co + L c,,ei11rrxJ + L c _,,e-inu.'L

n= 1 n:::: I
By the Euler's formula, e;' = cos t + i ~in t: this becomes
.( ~ (. 111LX . . IJTIX)
} x) ~ Co + ~ c,, cos + I Sill
11=1 L L.
(2.9) ~:, . ( COS_11/LX . . lllIX)

+ ~ C -n - I 5111
11=1 L L.
~ c0 + L
":, ,-
(c,, + c _ 11 ) COS
IJ IIX
+ Ic c,, - C -n
) 5111. IJITXJ
L
11=1 _ L
Now, by (2.7), we see that for 11 ::2:: 1,
c,, + c_n = I
2L
ft_ _
f(x)(e-inrrx!L + einrrx/L) dx
1
L _f(x) einrrx/L +'1 e-inrrx/L dx

L J -L "-
1 IL f(x). cos 117IX

dx
L -t L
=a"
and
2L I L j(x)(e-inrrx/L _ einrrx,'L) dx
-L
L ein"x/L _ e-inrrx/L
L I -L
f(x)
2i
-- dx
[ JL J(x) sin lliTX dx

L -L L
and
c0 = J JL f(x) dx = a0
2L -L 2
So the series in (2.6), with its coefficients given by (2. 7), is precisely the Fourier
series off. This series is known as the complex form of Fourier series.
Exercises 5.1
In Problems I through 7, find the Fourier series of the given function on the interval
indicated and describe graphically the periodic function to which the series can
CO!l\'erge.
(-rr<x<O)
1. f(x)
(0 ,::; x < n)
(~
( - 7[ ,::; x ,::; 0)
2. /(x) = (0 ,::; x ,::; rr/2)
7[ - x (rr/2 ,::; x ,::; rr)
3. f(x)=
I \' + l
11-x (-1 ,::; x ,::; 0)
(0,::; x,::; I)
4. /(x) = e"" (-L < x < L)

fO (-rr ,::; x ,::; 0)
5. f(x)
\sin x (0 ,::; x ,::; rr)
nx
6. f(x) = x cos (-L,::; x,::; L)
L
7. f(x) = x + x2 (-L<x<L)
8. Using the result of Problem 4, find the Fourier series of the function cosh ax =
(eax + e-ax)/2 on the interval [ - n. n ].
9. As in Problem 8, find the Fourier series of the function sinh ax = (ea" - e-x)/2
on the interval ( - n, n).
Sec. 3 Fourier Cosine and Sine Series 171
10. Let f be a periodic function with period 2L and set g(x) = f(x - L) for all x.
Show that
g(x) ~ ~
a 0 + .... ( - I) n (' a,, cos -nnx + lJn SID
. -nnx)
2 11=1 L L
where an and b,, are the Fourier coefficients off (x).

11. Using the result of Problem 10, obtain from Problem 5 the Fourier series of the
function
. _ 1-sin x ( - 7r :::; x :::; 0)
/( x) - \0 (0 :::; x s rr)
I (x) = f (x + 2rr)
12. As in Problem 11, obtain from the result in Example 2.2 the Fourier series of the
function
f(x) =
.
{x-x+ +n: n: (-n: s x :::; 0)
(0 :::; x s rr)
f (x) = f (x + 2n:)
13. If f(x - L) = f(x), show that a 211 _ 1 = b 211 _ 1 = 0 for n ~ l and that
a 211 =
2- JL f(x)cos
. 2nn:x
- - dx (n = 0, I, ... )
L 0 L
b 211 =
2 JL f (x) sm. ---
2nn:x dx (11 = I, 2, ... )
L 0 L
14. Apply the result of Problem 13 to obtain the Fourier series of the function
f(x) = x, 0 :::; x < n:, f(x - rr) = f(x)
15. Using the complex form of Fourier series (2.6), show that the function in Problem
4 has the expansion
00
sinh aL ~ (-I)" _!1_L____!/1n: einnx!L

11= -oo 0 2 L2 + 11 27!2
3. Fourier Cosine and Sine Series
We recall that a function defined on an interval - L :::;; x :::;; L is said to be

even if
(3.1) f(-x) = f(x)
and odd if
(3.2) f(-x) = -f(x)
for all x in the interval. Iff is an odd function, then it follows from the definition
that f(O) = 0. Geometrically, the graph of an even function is symmetric with
respect to the y-axis, and that of an odd function is symmetric with respect to
the origin (Fig. 5.3).
I I
I I
I I
I I
I I
I I
I I
I I
I _____ '
.._____LI _ _ -
~
-
_
_
_
L
~------
IJ L I .
/1
I
I
I
I
I
I
I
(d I IhI
FIG. 5--3 (a) Eren function. (b) Odd function.
If f is even and integrable on [ - L, L]. then

L (L
(3.3) j f(x) dx = 2 I f(x) dx
-L 0
This follows readily from the geometrical interpretation of the integral as the
area under the curve y = f(x) bounded by the lines x = L. Analytically. it is
easily proved by writing
Jro IL f(x) dx
IL
-I. f(x) dx = _L f(x) dx +
0
and setting x = - I in the first integral on the right. Using the definition (:U).
we thus obtain
Ir ((x)dx
(L 0
I j(x) dx = I
{
f(-t)(-dt)+
-L L 0
C
()
f(I) dt + C
0
f(x) dx
/
2 1 f(x) dx
0
On the other hand. if f is odd and integrable on [ - L, L ], then by a similar

argument we have
(3.4) rL f(x) dx = Q
L
It is easily shown from the definition that the product of two even or two odd
functions is an even function and that the product of an even and an odd func-
tion is an odd function. For example, if both f and g arc even and if we set
h(x) = f (x)g(x), then
h(-x) = f(-x)g(-x) = f(x)g(x) = h(x)
Sec. 3 Fourier Cosine am/ Sine Series 173
which shows that h is even. On the other hand. if.f is even and g is odd. then
h(-x) = f(-x)g(-x) = f(x)[ -g(x)] = -f(x)g(x) = -h(x)

so that his odd. Similarly, the ratio of two functions that are both even or both
odd is an even function, and the ratio of two functions of which one is even and
the other odd is an odd function. Here it is assumed. of course, that the function
in the denominator never vanishes.
Now suppose that f is an even function on the interval - L s x s L. Let us
consider its Fourier series
_ (lo
f(x) - + ~
L.
(
an COS
mrx
+ b . 11r:x)
n Sll1
2 Fl . L L
where
a,, IL f(x) cos mrx dx (11 = 0, I, 2, ... )

L -L L
b,, IL f(x) sin dx llirX

(11 = 1, 2, ... )
L -L L
For each integer 11, we note that cos(nnx/L) is even and sin(nrrx/L) is odd; hence,
f(x) cos(nrrx/L) is even and f(x) sin(mrx/L) is odd. Therefore, according to
(3.3) and (3.4), we have
I
L nnx
an == f(x) cos - dx
L -L L
2 JL f(x) cos
nnx dx
(n=0,1,2, ... )
L 0 L
and
b,, = 1 IL .
f(x) sm ~ dx
/l1[X
= 0 (11 = 1, 2, ... )
L -L L
respectively. Th us, iff is an even function on [ - L, L J, its Fourier series reduces
to
ao oc 'lirX
(3.5) j (x) ~
+ L a,, cos , .
2 n= 1 L
where
(3.6) a" = "L JL f(x) cos -r;

2 nnx
dx (n = 0, 1, 2, ... )
0
This series is called the Fourier cosine series off on the interval 0 s x s L.
We note that formula (3.6) for the coefficients a. involves only the values off in
the interval 0 s x s L. Hence, even when f is defined only on the interval
[O, L ], one can formally form the Fourier cosine series off. If the series con-
verges to the function on [O, L ], then the series automatically extends the func-
tion into the interval '--- L s x s 0 as an even function, and extends it outside
the interval [ - L, L J as an even periodic function with period 2L.
~-.
174 Fourier Series and Fourier Integral Cliap. 5
In terms of the idea of eigenfunction expansion, we observe that the Fourier

cosine series (3.5) on [O, L] is precisely the eigenfunction expansion of the
function f with respect to the system of eigenfunctions { 1/2, cos(mrx/ L)} of the
Sturm-Liouville problem
z( + i.u = 0, u'(O) = 0, u'(L) = 0
On the other hand, if f is an odd function on the interval - L s x s L, then

f(x) cos(mrx/L) is odd and f(x) sin(mrx/L) is even for each integer n. Accord-
ingly, we have
an = 1- fL f(x) cos nnx

-- dx = 0 (11 = 0, 1, 2, ... )
L -L L
and
b
n
= -J
L
f L
-L
j(x) sin -11nx dx
L
=
2- JL f(x) sm. nnx
--- dx (11 = I, 2, ... )
L 0 L
Therefore, for an odd function f defined on [ - L, L ], the Fourier series reduces

to
00
(3.7) J(x) ~ L b sin

II
'!_TrX
L
n=I
where
2 JL . --L-
nnx
(3.8) b,. = L J(x) sm dx (11 = I, 2, ... )
0
This is called the Fourier sine series off on the interval 0 s x s L.

The sine series extends f into the interval - L s x s 0 as an odd function
whenever it converges to f on the interval 0 s x s L. Outside the interval
- L s x s L, the series represents the odd periodic extension of the function
with period 2L. Here we notice that the Fourier sine series (3.7) is the eigen-
function expansion off with respect to the eigenfunctions (1.2) of the Sturm-
Liouville problem (I. I).
In summary, given a function f on the interval 0 s x s L, it is possible to
represent the function on that interval by either a Fourier cosine series or a
Fourier sine series. In the former case, the function is extended as an even
function into the interval - L s x ::; 0 and as an even periodic function for all
x with period 2L. In the latter case, the function is extended as an odd function
into [ - L, OJ and as an odd periodic function for all x with period 2L.
Sec. 3 Fourier Cosine and Sine Series 175
Example 3.1. Find the Fourier cosine and sine series of the function /<x) x.
0 :::: x s Tl.
Sof111io11: The Fourier coefficients for the cosine series are
an
2
7I
I'"
... 0
x cos 11x dx
2 ('11x sin nx + cos 11x) !"

Tl 112 0
2 (-1)"
Tl 11
2
Hence, the Fourier cosine series is
Tl 2 "' ( - 1J" - 1
X ~ -1- ~ COS /IX (0 :::: x :::: Tl)
2
2 Tln=I 11
This series is the same as the series we obtained in Example 2.2, as should be
expected. (Why?) The series cornerges to the function x on (0, n] and represents
the even periodic extension of the function for all x \Vi th periodic 2n. The periodic
extension coincides with the extension of the function considered in Example 2.2.
The Fourier coeftlcients for the sine series are
") ,_
b11 = ~
Tl
J"x sin 11x dx
0
(-1)"+1
2
/[
so that we have
'f (-11"+1
x ~ 2 ~ sin nx
n=l II
It will be seen that this series converges to the function x for 0 :::: x < n. For
all x outside the interval [O, n ], the series extends the function as an odd periodic
function with period 2n (Fig. 5.4). The periodic function is discontinuous at the
points x = nn, 11 = 1. 3, 5, ... , at which the series has the value zero. We note
in passing that on the interval 0 S x :::: n, the function /(x) = x can also be
represented by the Fourier series obtained in Example 2.1.
Example 3.2. Find the Fourier cosine series of the function
j . ) = f cos x (0 :::: x :::: n/2)

(.\ \0 (n/2 :::: x :::: n)
176 Fourier Series and Fourier lute_qra/ Chap. 5
i( ' l
I I
---1-,-
1
I
.'ii 10
I
I
FIG. 5-4 Odd periodiC' extension off(Yl.
So/111io11: The Fourier coeflicients for = 0 and = I arc
cl
11 11
Irr. 2
2 2 2
ao = cos x dx -- Sll1 x
n 0
7[
'0
2
2
ll1
/[
r"
0
cos x dx
2
il 0 C' (cos 2x + 1 I dx
2
For > I. \le ha\e
r
11
2
2
a,, cos \"cos /IX rfx
7i
/[
cl
0
[co'>!n + llx + cos( n -- l)x] dx
~ .-r .' 2
j sin( n + lh sin( 11 - I Ix\
+
;r I I/ + I II -
I I 0
Since
7i
- 1) s111 nn cos n cos im sin n
2 2 2 2
l/li
:!: cos
2
and cos(nll/21 = 0 \1hen 11 is odd. and cos(mi:j2) (- I/ 1. hen 11 2k, A
I, 2,. _., 11e obtain
(-I )k f 1 l 2(-]Jk+I
n 2k + 2k t I n( 4k 2 - 1)
Thus, 11e have
2 "' (-I )k+ I
f(x) + cos x + ~ cos 2/...x
2 "k=1 (4k 2 - !)
Sec. 3 Fourier Cosi11e all(/ .";i11e .5eries 177
If we assume that 1he series converges to f on 0 <: x s n. then outside this

interval the series converges to the continuous even periodic e:xtension off for all
x. The graph of /and its periodic e:xtension is shown in Figure 5.5.
/(.\)
t
I
& oi
... \
FIG. 5-5 Eren periodic extension ojf(x).
Exercises 5.2
In Problems I through 5, find the Fourier cosine and sine series of the given function.
and graph the periodic function to which the series can converge outside the given
interval.
1. f (x) fx, 0 s x s -}
\I - x, -i s x s l
2. f (x) 7[2 - x2, 0 s x < 7[.
3. f (x) I -
x, 0 < x < 2.
0 s x s
4. f(x) = (x.
t I. s x < 2'
5. f(x) = eX, 0 < x < 7[.
In Problems 6 through I 0, find the Fourier cosine series of the function on the given
interval.
6. /(x) = cos 2x, 0 s x s 7[.
7. /(.\) =
f sin nx, 0 s x s 1
\I -1- s x s I
8. f(x) = sinh x, 0 s x s 7[.
9. /(x) = f x,
0 s .\' <
\0, < x < 2
10. f(x) = sin x, 0 s x s 7[.
In Problems 11 through 15, find the Fourier sine series of the function on the given
interval.
f sin x, 0 s x < n/2
II. /(x)
\0, n/2 < 0 S n
12. f(x) = x( 1 - x), 0 S x S 1.
13. f (x) = {'1,2, 0 s x .:; I

I s x s 2.
14. f (x) = co sh x, 0 s x < n.
15. f(x) = cos nx, Osxsl.
4. Bessel's Inequality; Riemann-Lebesgue Theorem
Jn preparation for proving pointwise convergence of Fourier series, to be

presented in the next section, we derive here Bessel's inequality for the Fourier
coefficients a,, and b,, given in (2.4) of a function f with respect to the orthogonal
system
f1 nnx . nnx \
12 , cos -I.- , s111 L I
(4.1)
on - L :::; x :::; L. We shall assume throughout this discussion that f is a

function that is piecewise continuous on the interval - L ::::; x :$ L. Then,
clearly, the Fourier coefficients a,. and b,, off exist. Now, by using (2.4), we see
from (2.2) that
, L fl IL f(x) dx]2 = L a6
c0
2 LL . -L 2
c7, L f1 IL j(x) cos mrx dx]2 La;,

IL -L L
,2
c,. =
Lil IL/'(). 117TX I ]2
LL -L. x Sll1 L ( x =
Lb'
;,
for n 1, 2,. . Hence, by the Bessel's inequality (6.13) of Chapter 4, we have
c6 + L
11=
Ill
1
(c7, + c;, 2 ) L fa6
L2
+ I
11= I
(a;, + b~)l
or
(4.2)
a20
+ L"' (a;, + b7,) :$
I IL f 2 (x) dx
2 11= I L -L
for any m. This is the form of the Bessel's inequality we seek for the coefficients
an and bn.
Sec. 4 Bessel's /11eq11a/ity; Rie111a1111-Lebesg11e Theorem 179
By assumption, f and hence also f2 is piecewise continuous on [ - L, L so J

that the integral on the right of (4.2) exists. Since the integral is independent of
111 and (4.2) is true for any 111, no matter how large, we find on letting /11 --+ co,
(4.3) q6 +
2
I
11= t
(a;, + b;,) s 1
L
Jt
-t
/2(x) dx
This says, in particular, that the series
a2 oc
(4.4)
2
0
+ .L ca;,
11= I
+ b7,)
converges. Thus we have established the following result.
THEOREl\I 4.1. If f is a piffeuise continuous function on the interrn/ - L <

x s L, then the series ( 4.4) of squares of the Fourier coefficients off 11ith respect
to the system (4.1) conL"erges and the inequality (4.3) holds.
The fact that (4.4) converges implies that
(4.5) Jim an = lim b,, 0

11-+oo rr-+oc
that is, the Fourier coefficients of a piecewise continuous function approach

zero as n --+ oo. This result is actually a special case of the following theorem,
which is due to Riemann and Lebesgue.
THEOREM 4.2. (Riemann-Lebesgue) If g is a piecewise continuous function on

the interrnl a s t s b, then
(4.6) J~n; f g(t) sin l.t dt = 0
Proof: We first prove the theorem in the case where g is continuous on the
interval a s t s b. Set
(4.7) I= r a
g(t) sin l.t dt
and make the change of variable t = r + (rr/).), assuming that}, is so large that
b - rr/} > a. Since sin /.t = sin /.(r + (rr//.)) = -sin h, the integral (4.7)
becomes
b-rr/). ( rr)
(4.8) I - g r + -~-
J a- rr/). I.
sin /.r dr
....
.180 Fourier Series and Fourier Integral Chap. 5
Using again t as the variable of integration in (4.8), and adding the equation to
(4.7), we obtain
I = ~ Ir g(l) sin i.t dt - f~:,; g (1 + ;) sin l.t dtl

(4.9)
2
(f-rr; [~g(I) - (1 + ;) j 'm i.t dt
g
+ Jbb-rr/i g(t) sin I.I dt - .c_rr/i g (1 +;)sin I.I dt)

Let M denote the maximum value of lg(l)I on the interval a :::; t :::; b. Then the
last two integrals on the right in (4.9) yield
1 {b i Afr.
(4.10) I. g(t) sin i.t dt! :::; -
2 b-rr/A 2).
and
(4.11)
respectively.
Now let i: > 0 be any given small number. Since g is continuous on the
closed interval a :::; t :::; b, it is uniformly continuous there; that is. correspond-
ing to the given c;, there exists a number() such that
''g(t) - g
(I t + "')- .: < f,
(whenever ~ < c5)
. J. b - a !. .
where I is any point in the interval [a, b]. Then the first term on the right of
(4.9) gives
:~ .c-rr/;. rg(t) - g (1 + ;) l sin i.t dt
( 4.12) : :; r-rrli ,g(t) - g (1 + ~): dt

2 a ' !.
1 r: (b - Ti") < 1:
< a - . ')
2 b - a J '
Therefore, if we choose i. sufficiently large so that Tt/i. < <> and M-:c/i. < 1:/2,
then on taking absolute values in (4.9) and using the bounds in (4.10), (4.11).
and ( 4.12), we get I/I < r,, which precisely implies (4.6).
Jn the general case, let a = t 0 < t 1 < < t" < tn+ 1 = b be the points
where g is discontinuous. Then
(4.13) I
IL>
{b
a
g(t) sin i.t dt = -~
"
1-0
ri+I
._ fj
g(t) Sin i.f dt
Sec. 4 Bessel's b1eq11aliry; Rie111a1111-Lebes_que Theorem 181
Since g is continuous on each of the subintervals I; s r s I;+ 1 0 s i < 11. 1t

follows from the preceding case that each of the integrals on the right in (4.13).
and hence their sum. vanishes as i -> oc. This completes the proof of Riemann-
Lebesgue"s theorem.
With additional condition on the function the Riemann-Lebesgue theorem r

also holds when the integral in (4.6) is improper.
THEOREM 4.3. ~fg is a piece11 ise continuous function on the interrnl a s t < oo
s11ch that the integral
(4.14) f 00
a
Ig(I JI dt
exists (i.e., g is absolutely integrable on a s x < CIJ ), then
(4.15) lirn
.I.----> oc
f'"
.. a
g(t) sin i.t dt = 0
We leave the proof of this theorem as an exercise (Problem 10).
Riemann-Lebesgue theorems (Theorems 4.2 and 4.3) will be essential in the

convergence proof of Fourier series (Section 5) and of the Fourier integral
formula (Section 7). respectively
Exercises 5.3
l. From the Fourier series of the function f!x) = x/2. - rr < x < n, show by
Bessel"s inequality that
II= 1
(See Examrlc 3.1 .)

, From the Fourier series of/(x) '
x-. - :5: x s; 7Y ~ show that
"" h
iI4
~ s;
n=l 114 90
3. From the Fourier series of.f(x) x' - iI s; x s; 7r' show that
if
~
n= I (211 1)4 96
<See Example 2.2.)
4. From the Fourier series of the function f(x) = e"', - ;-; < x < rr, show that
I + 2 ~ s; n coth rr
ri=l 112 +
(See Problem 4. Exercise 5. I.)
182 Fourier Series and Fourier lnte_qral Chap. 5
5. Let f be piecewise continuous on 0 s: x s: n and let a,, (11 0, 1.... ) be the

coefficients in the Fourier cosine series off. Show that
00
2
2
+ I:
"
11~ I
a,7 s: ;-
7r
J" f
O
2
(x) dx
6. Let f be piece\\ ise continuous on 0 :S x s n and let h,, be the coefficients in the
Fourier sine series off. Show that
I:'"
11= l
b~ S 2
7t
J" f
0
2
(x) dx
7. Under the stronger condition that g is continuous and piecewise smooth on

a s x s h, prove the Riemann-Lebesgue theorem by first performing an
integration by parts.
8. If g is piece\\ ise continuous on (a, b), show by the Riemann-Lebesgue theorem
that
. Jb g(t) sm. (m +
l1m
m--+:r a 2
I). r dt = 0
9. Show that
Jim rb g(t) cos Ill( dr = 0

111--+ oc J a
IO. Prove Theorem 4.3. Hint: Absolute integrability of g on a -s t < ex: implies that
there is a constant B such that J;
_q( f) dt I < c/2, whenever b > B, corresponding
to a given r > 0. I.et h > B and write
.C 9(1) sin ).t dt = ib q(f) sin ).t dr + L"" g(t) sin l.t dt
Take absolute values on both sides of the equation and use Theorem 4.2 in the
first integral.
5. Comergence of Fourier Series
We ~hall now present conditions under which the Fourier series of a given
function can converge to the function. Without loss or generality, we shall
assume in our discussion that L = rr. We do this mainly for the sake of con-
venience as the theorems that we shall prove hold for arbitrary value of L.
Thus, let f be a function defined on the interval - ;r :<.::;; x :<.::;; rr and let the
J
function be extended outside [ - rr. rr as periodic function with period 2rr. Then
the Fourier series off is given by
f.
(5.1) f (x) ~
00
2
+ L
11= 1
(a" cos nx + b" sin nx)
Sec. 5 Convergence of Fourier Series 183
where
a,, = Jr J~rr f(x) cos nx dx (n = 0, I, 2, ... )

(5.2)
b,, = ~ J~" f(x) sin nx dx (11 = 1, 2, ... )
Our goal is to establish conditions on.funder which the symbol~ in (5.1) can be
replaced by equality.
Consider the mth partial sum of the series in (5.1); that is,
a0 m
(5.3) S 111 (x) = + L (a,, cos nx + b 11 sin nx)
2 11= I
Substituting the formulas (5.2) for a 11 and b,, in (5.3), we obtain
+; ,,t l(f r 1
_/(()cos nf, df,) cos nx
(5.4) + (J:rr .fW sin nf, df,) sin nx J
JrJ-rr" .f(() l2r~ + t 11-I

(cos nf, cos nx + sin 11( sin nx)J df,
= J Jrr f(f,) l)- + ~ "' cos n(f, - x) J df,

Jr -rr 2 11- l
The sum inside the bracket in (5.4) can be expressed as
I ~ (" ) sin(m + !-)((, - x)

+ '-' cos /1 ( - x =
2 11= 1 - 2 sin[(f, - x)/2]
This can be proved as follows: Let f, - x = u and set
J m
(5.5) S =- +
2
L
11= I
cos nu
Multiplying both sides or the above equation by 2 sin(l/2)u, we find
2S sin
1
2
u = 2 (sin l
2
u) (l2 + I 11= I
cos nu)
. 1 "' . 1
= sm- u + 2 I sm - u cos 11!1
2 11= I 2
184 Fourier Series and Fourier Inte.qral Chap. 5
Since
2 Sl11
. 1
2
U COS llU sin(n + ~) u - sin(n - ~) u.
the last equation reduces to
. l
2 S srn u
2
. l]
Sll1 U + r.Sl11
. 3
2
11
. 1
5111 2 111l + ...
sin ( 111 - ~) ll]
I
sin ( 111 +
2) u
or
sin(111 + l/2)t1
s --------
2 sin(u/2)
Hence, we have
"' sin(m + l /2)11
(5.6) + L cos nu
2 n= I 2 sin( u /2)
Therefore, (5.4) reduces to
(5.7) S,,,(x) = l Jrr Jm sin(m .+ _1/2)(~ - x) d~

1
rr -n 2 srn(s - .x)12
If we set ~ x = t, (5.7) becomes
(5.8) Sm(x) =
1 Jn-x ./.(x + sin(m + l /2)1 I
I ) - - - - - .- {f
7r -n-x 2 sin(t/2)
where the intenal [ -rr - x, rr - x] of integration remains of length 2rr. Now,
by assumption and from (5.6), the functionsf and [sin(m + l/2)1J'si11(1 12), and
hence their product, are periodic of period 2n. Since the integral of a periodic
function is the same over any interval whose length is equal to a period. we
finally have, from (5.8),
(5.9) S,,,(x) = ! J" f(x + t) sin(112__-_1_@1 dt

n -n 2 sin(t/2)
This is called Dirichlet's formula for the partial sum of a Fourier series. This
formula will enable us to prove pointwise convergence of a Fourier series to its
corresponding function.
Sec. 5 Conver_qe11ce of' Fourier ,S'eries 185
In the particular case where f(x) = I. so that the Fourier series of( consists
only of the single term I. (5.9) yields the special result
(5.10) 1 = l f"
n . __ rr
sin(111 + l/2)t cit
2 sin(l/2)
for all /11 (see also Problem 2 in the exercises). Clearly. the integrand in formula
(5. 10) is even: hence. it follows that
(5.11) I ro sin(m + 1/2)1 dt = I I'" sin(m _+ 1/2)1 dt = I

n -rr 2 sin(t/2) i! 0 2 sin(t/2) 2
With the preliminary results obtained above. we are now in position to state
and prove our main theorem on convergence of Fourier series.
THEOREJ\I 5.1. Let f be a fimction that is pil!Ce11ise smooth on the intcrrnl

- n ~ x ~ n and periodic with period 2n. Then f can be represented by its
.. Fourier series; that is .
a
(5.12) f(x) ...Q + I" (a 11 cos nx + b,, sin 11x)
2 ti~\
where the coefficients a,, and b,, arc gicen b_i ( 5.2). The Fourier series concerges to
f(x) at all points 1rhere f is continuous. and to the acerage [f(x - 0) + f(x + O)] 12
at all points ll'here f is disconti11uo11s.
Proof: At a point x where f is continuous, we know that f(x - O) =

f(x + 0) = f (x). Hence. it is enough to prove that the Fourier series converges
to [j(x - 0) + f(x + O)J/2 for all x. We do this by showing that
(5.13) Jim S,,,(x) = f(x - 0) + f(x + 0)
m-x 2
where S,,, is the mth partial sum of the Fourier series.
.. From the integral formula (5.9), we have
S,,,(x) = 1 Jo f(x + I) sin(m + _1/2)1 dt

n _" 2 sm(t /2)
+ I J" f(x + 1) si11(111 + l/2)1 dt

n 0 2 sin(t/2)
Thus, (5.13) will be established if we can show that
(5.14) .
I1m -1 Jo 1.( x + t ) sin(m + . - d t -_ l f( x - O)
l/2)t
m-acn -n 2 sin(t/2) 2
and
(5.15) .
IJill -1 J" f( x + t ) sin(m
+
l /2)1 cl I 1
- f(x
2
+ 0)
,,,_ oo n 0 2 sin(t/2)
In view of the identities in (5.11), equations (5.14) and (5.15) are equivalent to
(5.16) Jim }_
m~oo n
J
0
-rr
[f(x + t) - f(x - O)] sin(m __+
2 sm(t/2)
1i 2
)t di = 0
and
(5.17) .
I Im
m~oo
l
n 0
f" - - - + --l/2)t
[f( X_ + t ) - j'( X_ -r. O)] sin(m
2 sin(t/2)
- -- d t -_ 0
respectively.
Let us consider (5.17) and set
(5.18) g(t) = /(x + 1)- f(x +O) (0 < t ::.:; n)

2 sin(t/2)
Since f is piecewise smooth on the interval - n s x s n and is defined by

periodicity outside this interval, g is piecewise smooth for 0 < t ::.:; n. At
t = 0, we note that
+ . f(x + t) - f(x + 0)
g(O ) = 11m - ----- ------
r~o 2 sin(t/2)
r> O
= lim J(x __:t-_ tL--::_ f(~ q) Jim - t/ 2

1~ o t r~o sin(I /2)
r>O r>O
= Jim f (x_+_t )_-=JC:< __:t- ~)

r~o t
r>O
= f~(x)
which exists, since f is piecewise smooth. Hence, g is piecewise smooth, and

consequently piecewise continuous, on the interval 0 s t ::.:; n. By the Riemann-
Lebesgue theorem (Theorem 4.2), it follows that
Jim
m-oo
f" g(t) sin(m + 1/2)1 dt = O
0
which proves (5.17). In exactly the same manner, we can show that (5.16) is also
true. Thus, (5.13) is established and so the theorem is proved.
Returning to the examples of Section 2, we see that in Example 2.1 the

function f (x) = 0 when - n s x s 0 and f (x) = x when 0 ::.:; x ::.:; n is
continuous and piecewise smooth on the interval -n ::.:; x ::.:; n. Hence, by the
theorem just proved, it follows that
n-
4
00
n~1
l
- L --- -2-
(2n - 1) 2 n
cos(2n - 1)x + (-_-1)"-
n
. ] = (0x
sm nx
n/2
(-n < x ::.:; 0)
(0 ::.:; x < n)
(x = n)
Further, the series converges for all x outside - n s x s n to the periodic

extension off, except at x = nfi_n = I, 2, ... , where the series converges to
the value n/2.
To illustrate the use of Fourier series in evaluating the sum of series of
constants, Jet us set x = n/2 in the preceding series. We obtain
n 1 1 + ...
-=l--+-
4 3 5 7
Similarly, setting x = rr, we find
n2 1 1
- = 1 + + + -- + ...
8 32 52 72
In Example 2.2, the function f(x) = lxl is continuous and piecewise smooth,
including its periodic extension outside [ - n, n] for all x. Hence,
lxl = 1! _ ~ f ~osqn - _l)x (-n s x s n)

2 n 11= i (2n - 1) 2
By the comparison test and Weierstrass M-test (Chapter 1, Section 5), we see
that this series converges absolutely and uniformly to lxl on the interval
indicated. This important result is actually true of every piecewise smooth
continuous function that is periodic, and will be proved in Section 6.
The convergence of the Fourier cosine series or the Fourier sine series
corresponding to a given function that is piecewise smooth on the interval
0 s x s n can be deduced from the main theorem. Indeed, we note that if f is
piecewise smooth on the interval 0 s x s n, its even or odd extension into the
interval - n s x s 0 is likewise piecewise smooth so that the extended function
is piecewise smooth on [ - n, n]. According to Theorem 5. I, the Fourier series
of either of these extended functions converges to f on the interval 0 s x s n
in the manner stated in the theorem. But the Fourier series of an even function
reduces to cosine series and that of an odd function to sine series. Thus, the
following theorem follows from Theorem 5. I.
I.
THEOREM 5.2. Let f be a piecewise smooth function on the interval 0 s x s n.

Then f can be represented by a Fourier cosine series
a oo
(5.19) f(x) = _() + L an cos nx
2 n= I
tt11ere
(5.20) an = -2 f" f(x) cos nx dx (n 0, 1, 2, ... )

n o
or by a Fourier sine series
00
(5.21) f (x) = L
n=l
bn sin nx
188 Fourier Series and Fourier Integral Cliap. 5
where
(5.22) b,, =
2
7[
f n
0
f(x) sin nx dx (11 = 1, 2, ... )
The cosine series and the sine series converge rm the intervals 0 ::::; x ::::; n and
0 < x < n, respecticefy, to f(x) at points where f is continuous and to
[f(x - 0) + f(x + O)J/2 at points where f is discontinuous. Moreover, outside
the interval 0 ::::; x ::::; n, the cosine series cotm:rges to the even periodic extension
off, and the s1i1e series to the odd periodic extension off.
Notice that the even periodic extension of the function f is necessarily con-
tinuous at the points x = nn, n = 0, 1, 2,. . .. Hence, the Fourier cosine
series in (5.19) converges at the points x = 0 and x = n to f(O+) and f(n-),
respectively. On the other hand, the odd periodic extension off is continuous
at x = nn, 11 = 0, 1, 2, ... if and only if f(O+) = f(n-) = 0. This is evident
from the sine series in (5.21), which converges to zero at the points x = nn.
Exercises 5.4
I. Consider the geometric series
(a) Show that

iu eimll - 1
s = e - - -
eiu - I
(b) By multiplying the numerator and denominator by e-iu1 2 , show that

ei(m+l/2)11 - eiu/2
s =
(c) By using the formula ei = cos u + i sin 11, take the real and imaginary
parts on both sides of the result found in (b) and obtain the results
sin(m + I /2)11 I
cos 11 + cos 211 + + cos mu = - -- - --- --- - -
2 sin(u/2) 2
. . . I u cos(m + l/2)u
sm 11 + sm 211 + + sm mu = - cot - - - ---- - - - -
2 2 2 sin(u/2)
This gives an alternate method of proving formula (5.6).

2. Obtain (5.10) by integrating the equation (5.6) with respect to u from - n ton.
3. Establish the result (5.16).
4. Let /(x) = x 2 when - n ~ x ~ n and /(x + 2n) = f(x) for all x. Prove that
1
n ro cos nx
f(x) = -- + 4 I; (-1)" - -
3 11=! n1
for all x, and deduce that

00 ( - 1)11- l 00 T[2
l:1
ri= n 2
12
l: I
II= ll
2
6
5. Let f(x + 2n) = f(x) for all x, where
f(x) = fO ( - T[ s x s 0)
\sin x (0 s x s n)
Prove that for all x,
00
1 . 2 cos 211x
f (x) = + - sm x - - l:
Ti 2 Tin=I 411 2 - I
Deduce that
00
1 1
l: --=-
11=1 (411 2 - I) 2
6. Let f(x) = 0 for - I < x < 0, /(0) = 1, and f(x) = cos nx for 0 < x < I.
Prove that for -1 < x < 1,
1 4 <:.() /1
f (x) = cos nx + l: sin 2nnx
2 Ti n=1 411 2 -
7. Let
f(x) = {x
(0 s x s n/2)
T[ - x (n/2 s x s n)
(a) If /(x) = f(x + n) for all x, show that
f (x) = n _ 2 ~ cos 2(211 _- !)x

4 Ti n=I (211 - 1) 2
(b) If f(x) = -f( -x) for - n s x s 0, and f(x + 2n) = f(x) for all x, show
that
8. Let the function

f(x) = { ~ - x/2h (0 s x s 2h)
(2h s x s Ti)
be extended as an even periodic function with period 2n. Show that, for all x,
f(x) = ~h
n
[! + 2
f (si!2_'
11=1 nh
!!!)
2
cos nx]
9. Letf(x - L) = f(x) for all x, where
f (x = {cos(nx/L) (0 s x s L/2)
) - cos(nx/L) (L/2 s x s L)
Show that, for all x,
f(x) = -
2
+
4 f (- !)"+ 1 co~(2n7!x)jL
T[ T[ n=l 4n 2 - 1
(See Problem 13, Exercise 5.1.)
10. Let f(x - n) = f(x) for all x, where f(x) = cos x when 0 < x < n. Show
that, for all x, x #- 11, (11 = 0, l, 2, ... )
8 ~ n_ s_in 21_1x
f(x) = - ..,,
n 11 = 1 4n 2 - I
11. Let the function
f(x) = {(~/L)x (0 ::; x ::; L/2)

sm(nx/L) (L/2 ::; x ::; L)
be extended as odd periodic function with period 2L. Show that, for all x,
f (x) = (4 + 2I) . nxj,"

;-2 sm
+ _ ~ (-!)" f~n_[gn_Jrx)/LJ + ~sin[(2n_:+_nnx/~]}

n n=l l 2
11(411 - I) n (2n + 1) 2
Hence, deduce
oo I
2: - -
8 II= l (211 - 1) 2
12. Prove that, for - n ::; x s n,

sin na ~ ( !)11 2a sin na
cos ax=-----+ ..,, - 2
- --cos
2
nx
na n= 1 n(a - 11 )
where a is not an integer. Hence, deduce
6. Uniform Convergence of Fourier Series
In the preceding section we proved that the Fourier series of a piecewise

smooth and periodic function converges at all points to the average value of the
left-hand and right-hand limits of the function. If, in addition, the function is
continuous, then the Fourier series converges to the function at all points. It
turns out that in the latter case the series actually converges absolutely and
uniformly to the function. We shall establish this fact in this section.
Letf be a continuous piecewise smooth function on the interval - n 5 x 5 n
such thatf ( - n) = f(n). Sincef is piecewise smooth, f' is piecewise continuous,
and hence its Fourier coefficients
a;, = Ir f'(x) cos nx dx (n 0, 1, 2, ... )
lr
1[ - "'
(6.1)
b'n = - f'(x) sin nx dx (n 1, 2, ... )
1[ - "'
Sec. 6 Uniform Convergence of Fourier Series 191
exist. Since f is continuous and f( - n) = f(n), we see by integration by parts

that
a~
1 rrr f'(x) dx = --I [f(n) - f( - n)] = 0
7[-n 7[
a;, = l [f(x) cos nxJ~rr + 11

Jrr f(x) sin nx dx
7[ 7[ -n
= cos nn [f(n) - f(-n)] + nb 11

7[
11bn (11=1,2, ... )
b;, = ~ [f(x) sin nx]~,, - : f" f(x) cos nx dx
-na 11 (11=1,2, ... )

where an and bn are the Fourier coefficients off. Thus,
b'
(6.2) a
n
= ---"
'
(11 1, 2, ... )
ll
From the obvious inequalities
(1a~I ~r a,,
,2 21a:.I + -I > 0
2 -
11 11
21b:.I
it follows that
(1b;,1
lf = b;,2
11
+ 11
1
2 -
> 0
la~I lb;,1 .::;; l ( a11'2 + b'1) + 1

+
ll 11 2 n 112
and, in view of (6.2),
Ian! + lb,,I .::;; I ( all,2 + b;.2) +

2 112
Hence,
(6.3)
for arbitrary integer m. Since f' is piecewise continuous, it follows from

Theorem 4.1 that the first sum on the right of (6.3) converges as m --+ co. The
second sum also converges since I:'= 1 (1 /11 2 ) is a p-series with p = 2. Therefore,
the series
oc
(6.4) L (Ian!
n=I
+ lbnl)
converges. Using this result, we now prove the following theorem on uniform
convergence.
THEOREM 6.1. Let f be a continuous piece1rise smooth function 011 the interrn!
-n S: x S: n such that f( - n) = f( n). T/]('11 the Fourier series off
(6.5) ao
2
+ f
11= 1
(a 11 cos 11x + b 11
sin 11x)
conrerges absolutely and uniformly to f(x) jiJr all x on the i11tenal - n S: x S: n.
Proof: Since f( - IT) = f(IT), the periodic extension of I is continuous and

piecewise smooth for all x. Hence, by Theorem 5. L the series (6.5) converges to
f(x) for all x on the interval -TI S: x S: IT. To show that the convergence is
absolute and uniform, we observe that
la,, cos nx + b,, sin nxl S: la,, cos nxl + lb,, sin 11xl
S: la,,I + lb11I
where, according to the previous result, the series I.,':= 1 ( la,,I + lb,,I)
converges.
The theorem then follows by the comparison test and Weierstrass M-test.
It should be pointed out that the conditions stated in the theorem actually
ensure absolute and uniform convergence of the series (6.5) on any interval to
the periodic extension of the function. In this connection, the condition
f( - IT) = j(IT), which guarantees continuity of the periodic extension off; is
essential because the sum of a uniformly convergent series of continuous
functions must be continuous.
The next theorem on uniform convergence of Fourier cosine and sine series
can be treated as special case of Theorem 6.1.
THEORE;\I 6.2. Let f be a continuous piece1rise smooth f1111clio11 on the interrn!

0 S: x S: IT. Then the Fourier cosine series off
Go oc
2
+ L
11= l
a,. cos nx
(a 11 = (2/IT) J~ .f (x) cos nx dx, 11 0, I, ... )

coni-erges absolutely and uniformly to f(x) for 0 S: x S: n and to the eu'n
periodic extension of.( for all x.
If, in addition, .f satisfies the conditions f(O) = f(IT) = 0, then like1rise the
Fourier sine series o.f f,
CfC
L b,, sin nx
n= 1
(b,, = (2/TI) J~ f(x) sin nx dx, 11 :?: I)

coni-erges absolutely and uniformly to f(x) on the interrnl 0 S: x S: IT, and to the
odd periodic extension off.for all x.
Notice that here the condition.f(O) = f(IT) = 0 ensures continuity of the odd
periodic extension off, thus making it possible for the Fourier sine series to
converge uniformly to f (x) for all x.
Sec. 6 U11iform Conrergence of Fourier Series 193
Now suppose that f is a function that satisfies the conditions of Theorem 6.1.
Then we have
(6.6) j .(x) = Go
+ ~
L.
(
a 11 cos. nx + h . nx)
Sill (-rr:S::x:S::1r)
11
2 n~ I
where the convergence of the series tofis absolute and uniform. Let us multiply
both sides of(6.6) byfto obtain
(6.7) ./2(x) = ao f(x) +

2
f
11= I
[a,J(x) cos nx + b11 f(x) sin nx]
The series on the right of(6.7) converges uniformly tof 2 (x) on -n :s:: x :s:: 1r.
To see this, let t: > 0 be an arbitrary small number, and let B denote the
maximum value or If I on - n :S:: x :S:: n. By the definition of uniform con-
vergence, there is an integer M such that
{;
JS,h) - f (x)I < (whenever - n :s:: x :s:: n)
B
for all m > M. Here, S,,, denotes the mth partial sum
ao m
S,,,(x) = + L (an cos nx + h 11 sin nx)
2 11= I
Then, clearly,
(6.8) JS,,,(x)f(x) - / 2 (x)I lf(x)J JS"'(x) - f(x)J
whenever - n :s:: x :s:: n and for all 111 > M. This shows that the sequence
S 111 (x)f(x), which is the mth partial sum or the series on the right of (6.7),
converges uniformly to .f2(x) on [ - n, n]. Therefore, term-by-term integration
of (6.7) is possible (Chapter I, Section 5) and we obtain
In .f2(x) dx = a; In f(x) dx
+ t rl
11 l
G 11 Jn
-rr
f(x) COS llX dx + bn Jn
-n
f(x) sin IJX dx]
+ I~ + b;,) J
or
n
l a2
2 11= 1
(a;,
(6.9)
al0
+ Ioc:
(a~ + b~) =
I In f 2
(x) dx
2 11=1 n -n
which is Parseval's equation [see (6.16), Chapter 4].

Hence, from (6.10), (6.11 ), and ( 6.12) of Chapter 4, it follows that
1
Jim f" [f(x) - S,,,(x)] 2 dx
m--+XJ n J _rr
(6.10) = lim j l
111-.-J_ lrr
J" -n:
f 1
(x) dx - la~+
l2
I
n=l
(a;,+ b;,)lJI
J
=0
which says that the Fourier series (6.6) converges to fin the mean. Thus, we see
that uniform convergence of a Fourier series implies convergence in the mean.
In terms of the idea of completeness that we introduced in Section 6 of Chapter 4,
we can therefore conclude that the orthogonal system (4. I) of eigenfunctions is
complete with respect to the class of functions that is continuous, piecewise
smooth, and periodic of period 2rr. Actually, the system (4.1) is complete with
respect to the class of functions which is square integrable (e.g., piecewise con-
tinuous) on the interval [ - rr, rr]. This means that the Fourier series of a func-
tion f converges in the mean whenever the integral f"..." .f2(x) dx exists. Then, for
such a function, Parseval's equation (6.9) is satisfied.
Let f and g be two square integrable functions on [ -rr, rr], with Fourier
coefficients denoted by a,1' b,, and a:, b:, respectively. From the obvious
inequality
[.f(x) + g{x)] 2 2[.f 2(x) + g 2(x)] - [f(x) - g(x)] 2

:S: 2[.f2(x) + 2
g (x)]
it follows thatf + g is also square integrable, with Fourier coefficients given by

a,, + a: and b,, + b:. Hence, according to Parseval's equation, we have
(6.11) If"
rr -rr [f(x) + g(x)] dx
2
+ a6) 2
=
(a 0
2
- -- +
11=1
L:c; [(a,, + a,';) 2 + (b,, + b:)2J
Writing the corresponding Parseval's equation for f - g and subtracting it from

(6.11), we then obtain
(6.12) If"
rr -rr f(x)g(x) dx
2
00
I
n=l
This result is known as the generalized Parseval's equation.

If we define the function gin (6.12) by the equation
f1 (-n<x<t)
g(x) = \0 (t < x < rr)
Sec. 6 Uniform Co11vergence of Fourier Series 195
where -rr :s:; t :s:; rr,then(6.12)becomes
(6.13) f,, .f(x) dx = ~a 0 (t + rr)
+ ~ la,, f' cos nx dx + b,, f' sin nx dx]

" l -n -n
This is precisely what we shall get if we formally integrate the series (6.6) term
by term from - rr to t, where - rr :s:; t :s:; n. Jn view of this result, we can there-
fore assert that if f is square integrable (say, piecewise continuous), then whether
or not the Fourier series of J converges, the integrated series converges to the
integral of .f. (See also Problem 8.)
Exercises 5.5
1. Let a,, and b,, be real numbers and consider the quadratic equation
m m m m
~ (a,, = ~ a~ + 2). ~ a,,b,, + ). 2 ~ b~
2
+ J.b,,) 2 0
n=l ri=l 11=1 n= I
in the parameter ).. Prove that
( "~I
111
a,,b,,
)2 (n~I a;,
S
111 ) (
"~I
m
b;,
)
This is called the Cauchy-Schwarz inequality.

2. Consider the identity
m m
~ (a;, + b;,)112 = ~ 1 (a~2 + b~2)112
n=J n=l n
from equation (6.2). By using the Cauchy-Schwarz inequality and Theorem 4.1,
prove that the series L::= 1 (a~ + b;) 112 converges. Hence, deduce the conver-
gence of the series (6.4).
3. Show that the Fourier series of the function/(x) = x 2, -n s x s n (Problem 4,
Exercises 5.4) can be integrated term by term from 0 to x when - n s x s n,
and obtain the result
~ {- l)" ~in 1~ = _1 x(x2 _ n2)

3
n= I 11 12
Deduce
00 (-1)"+1 7[3
n~I (211 - 1) 3
32
By Parseval's equation, show that
"' 7[6
~ 6
n=I n 945
4. Sho\\ that the Fourier series of lhe function given in Problem 5, Exercises 5.4,
can be integrated term by term from 0 to x when 0 < x < n, and obtain the
result
8 oc 11 sin 211x
cos x = ~ (0 < x < n)
l[ rJ:::: 1 411 2 - I
Hence, by Parseval's equation, shov. t hal

(f fl1
~ -
2
11=1 (411 - 1) 2 64
5. Justify the term-by-term integration of the Fourier series in Problem 7b, Exercises
5.4, and show that
{ ~ ;:,
2x)(n + 2x) (0 :S x ::S rr/2)
~ (_ )" + 1 cos(2n - I )x
1
11= i (2n 1) 2
- n)(2x - 3rr) (rr/2 ::S x :S rr)
32
6. Let /(x) = I - x, 0 ::; x ::; 2. Show that the Fourier series of/ converges
uniformly to the function for 0 :S x :S 2, and that
8 ~ 1 (211 - I )rrx
I - t = ~ . cos
. rr 2 n= I (211 - 1)2 2
Deduce
00
~
II= [ (211 96
7. Obtain from Problem 6 the series
16 ' . (211 - I )rrx

x - ~ . Sin - (0 :S x :S 2)
2 rr 2 n= I (211 I )3 2
and deduce
"'
~ - -
11= I (211 960
8. Let I be a piecewise continuous function on the interval - rr ::; x ::; rr and con-
sider its Fourier series
if)
f(x)
00
+ ~ (a 11 cos nx + h 11 sin 11x)
2 11=!
Set
F(x) = f l/(t) - a;] dt
where - rr ::; x ::; n and a is any constant.

(a) Show that for - rr :S x :S rr, F(x) is continuous, piecewise smooth, and
F(rr) - F(-n) = 0
Hint: Note that
ao
F'(x) = f(x) - -2 and F(n) - F(- rr) = jrr
-"
(
f(t) -
a )
} dt
Sec. 6 Uniform Co11ver.qe11ce of Fourier Series 197
Thus F satisfies the conditions of Theorem 6.1 and therefore can be represen-
ted by an absolutely and uniformly convergent Fourier series \\ith Fourier
coefficients given by
A11
71
f:" F(x) cos 11x dx (11 O. I, 2, ... )
B11 =
I
n
J:. F(x) sin nx dx ( /1 I, 2, ... )
(bl By integration by parts, show that
A= - b11 B,, (11 1, 2, ... )

11 II
Hence,
F(x)
Ao
2
+ ~
11= 1
(- b 11 cos 11x
11
+ a,, sin
/1
11x)
and
II~) 11a)
Ao
0 = + b,, cos 11a + a 11 sin
2 (-
II II
r
(c) Thus, obtain the result
/(I) dt = a o (x - a)
+
2
~
C
ll= I
l Gii
sin nx - sin na _ b cos 11x - cos
II
II
/1
11aj
This proves that '-" hether or not the Fourier series of a piecewise continuous
function f converges to/, the series can be integrated term by term to yield a series
that converges to the integral off
9. Integrate the series
00
x ~ (-1)"+1
sin nx
( - n < x < n)
2 n=l n
from 0 to x, term by term, to obtain successively

7[2
+ ~
x2
(a) (-I)" cos nx (-n:S:x:S:n)
4 12 II= l ll 2
00
(x2 (-I)" sin 11x
~
(b) _x_ - 712) (-n:S:x<n)
12 n=l n2
10. Integrate the series

n ~ sin(211 - I )x
(0 < x < n)
4 11=1 (211 - I)
from O to x, term by term, to obtain successively
00
cos(2n - I )x
(a) ~ (n - 2x) ~ (0 :s: x :s: n)
8 n=l (2n - 1)2
00
sin(2n - l)x
(b) ~ (nx - x2) ~ -----~
(0 :s: x :s: n)
8 n=l (211 - 1)3
~>.
7. Fourier Integral
So far, we have been concerned with the representation of a function by its

Fourier series. We have seen that if the Fourier series converges to the function
in some finite interval, then the series will represent the periodic extension of the
function outside that interval. This means that in case a function is defined for
all x, it cannot have a Fourier series representation that is valid for all x unless
the function is periodic. In an attempt to find a representation analogous to
Fourier series for a nonperiodic function that is defined for all x, we shall
consider the Fourier series of the function on an arbitrary interval [ - L, L J
and investigate its limit as L tends to infinity.
Let f be a piecewise smooth function on the interval - co < x < oo. Then,
on every subinterval [ -L, L ], f has the Fourier series representation
00
(7.1) j(x) = a- 0 + L ( ll77X

a,, cos --- + .
b,, srn nnx)
--
2 11=1 L L
where
a = -1 JL f(x) cos nnx

- dx (11 = 0, 1, 2, ... )
n L -L L
(7.2)
b = 1 JL f(x) .
Sill - - -
nnx dx
(n = 1, 2, ... )
n L -L L
We understand that at a point x where f is discontinuous, f(x) is to be replaced

by [f(x - 0) + f(x + 0)]/2. Substituting (7.2) for an and b,, in (7.1) and
simplifying, we obtain
(7.3) f(x) = -1- IL f(O d~ + L L1- JL_

oo
f(O cos -nn (~ - x) d~
2L _ [, n= I L L
As we Jet L --> co, the first term on the right of (7.3) will vanish, provided the
integral s~ 00 f (~) d~ exists; which it certainly does if we assume that f is abso-
lutely integrable on the interval - co < x < co. We assume that this is the case.
Then, on letting L--> co, from (7.3) we obtain
(7.4) f(x) = Jim

L~oo n=l
f 1
L
- JL
-L
Jm cos nn
L
(~ - x) d~
In order to investigate the limit on the right of (7.4), Jet us set
nn n
Sn= -- ' ~s,, = S11 + 1 - Sn = - (11 1, 2,. .. )
L L
(7.5) j(X) = !~ ~ JI rJ~L f(O COS S,,(~ - X) d~ J ~S11

Sec. 7 Fourier Integral 199
The sum in equation (7.5) reminds us of the sum defining the definite integral of
the function
F(x, s) =
1
f(~) cos s(~ - x) d JL
71: -L
over the interval 0 s s < oo. Thus, it seems natural to expect that as L--> oo,
(7.5) leads to
f"' ds J .rm cos s(( -

00
(7.6) f(x) = nJ -w x) d
0
It should be emphasized that the argument used above in arriving at the

formula (7.6) was purely heuristic and did not constitute a proof. In the theorem
to follow, we shall show that under the assumptions made on f, the function
can indeed be represented by (7.6). This is the representation we seek for a non-
periodic function that is defined for all x. The formula (7.6) is called the Fourier
integral formula for the function f.
THEOREM 7.1. Let f be a function that is piecewise smooth and absolutely

integrable on the interrnl - oo < x < oo. Then, at every point x in ( - oo, oo ),
(7.7) .f( ,x - 0) +
2
f(x + 0) = l
71:
f "'j
0
ds Joo f mcos s( -
-oo
x) d
Proof: By the definition of improper integrals, we have
J J d~
00
(7.8) oo ds .f(O cos s( - x)

71: 0 -oo
= Jim
).--4-C()
1
1[
J). ds J"' .rm cos s(( -
0 -!"fj
x) d~
The inner integral on the right side of equation (7.8) is uniformly convergent
with respect to s for 0 s s s /., since
Jim cos s Cs - x)J s 1.f()1
and .f is absolutely integrable on ( - oo, oo ). Hence, the order of integration with
respect to and scan be interchanged (Chapter I, Theorem 5.3), and we obtain
I
7!
f 00
O
ds J' "'
-~oo
Jm cos s(~ - x) d = ~im
A-too
I
JI
J""
-oo
Jm J;. cos s(~
0
- x) ds d~
= Jim I Joo /(0 sin /,(~ - x) d~

;~oo n _ 00 (~ - x)
lf we set t = ~ - x, then this becomes
(7.9)
71:
1
J 0
00
ds f"'
-oo
f(O cos s(~ - x) d~ = Jim
,l~oo 71:
1
Joc
-oo
.f(x + t) sin l.t dt
t
.....
In Chapter I, Example 6.3, we found that JO' (sin t)/1 d1 rr12. From this it is
easily shown that
(7.10)
o si11 ; I
. dt =
Joc sin ).I dt = 7l
(). > 0)
J
-"" I 0 I 2
Hence, the Fourier integral formula (7.7) will be established if we can show that
(7.11) hm 1
1.-+oc7IL-r_
j [/(x + t) - f(x - O)] sin It dt
f
= 0
and
(7.12) hm I
). --4 'X: rr L
f0
oo [ .f( x + t ) - .f( x + O)] sin )t dt
t
= 0
in view of(7.9) and (7.10).

Consider the integral in (7.12) and let us write
I = I j'" f (x +_ t) - f (x + 0) sin It dt
7l o I
where
I1
I l'h
f(x

+ t) - f(x
-
+ 0) . . I
sm 1.t ct
TI o 1
I2
1 ('
I f (x + t)
sm
. . d
1.t t
TI b I
13 = .f (x + 0) J"' sin i.t dt

7l b I
with b being an arbitrary positive number. Since [f(x + !) - .f(x + 0)] 1t is

piecewise continuous for 0 ~ I ~ b and .f (x + t )/ t is piecewise continuous and
absolutely integrable on b ~ t < CD, it follows from Riemann-Lebesgue
Theorems 4.2 and 4.3 that / 1 and / 2 tend to zero as I. --+ CD. respectively. The
change of variable i.t = z transforms / 3 into
IJ
_
-
f (x + 0) Joc sin z -- (
Iz
TI ib Z
which clearly goes to zero as }. --> oo. Thus,
Jim I = Jim (1 1 + 12 - I 3) = 0
A--+XJ ;.-oo
and so (7.12) is proved. By the same method we can also show that the limit in
(7.11) is zero. This completes the proof of the theorem.
Sec. 7 Fourier /11tegral 201
Example 7.1. Find the Fourier integral formula of the function
. JI (x s 1)
j(x) = \0 > 1)
Solution: It is clear that the function here satisfies the conditions of Theorem 7.1.
Thus, for - oo < x < oo, we have
f'(x - 0) + f (x + 0)
2 n
{" ds
Jo
f 1 cos .1( -
-1
x) d
r'l"~ins(-x)Jl ds
n Jo l .1 - l
n L" sin s(l - .\') : sin s(I + x) ds
2 {"' co~sx sin .1 di

n Jo s
It follows that
.2
n
l"'
0
cos sx sin--s ds
s
= \.-~ -
( xj < I)
(xi 1)
(xi > I)
In particular, at x = 0, we have
00
1 0
sin s
--
S.
ds =
n
?
-
a result we also obtained in Example 6.3, Chapter I. The result at x = 1 can be

verified directly. lndeed, at x = 1 we see that
2
n
l"" cos s sins
s
- ds= 1
n
J.'l) sin 2s ds
s 2
0 0
by (7.10).
By expanding the function cos s(( - x) in (7.6). we see that Fourier integral
formula can be written as
(7.13) f(x) = f' [A(s) cos sx + B(s) sin sx] ds
where
(7.14)
A(s)
h. J oo
- cc
f(() cos s( d
B(s) = ~ ~ f 00
f(O sin s~ d
In this form, (7.13) is analogous to a Fourier series representation and (7.14)
to the formulas for the Fourier coefficients. It is significant in this connection
202 Fourier Series anti Fourier Integml Chap. 5
to note that the functions cos sx and sin sx are linearly independent eigenfunc-
tions, corresponding to the same eigenvalue i. = s 2 , of the singular eigenvalue
problem
u" + i.u = 0 (-oo < x < oo)
u bounded as Jxl ---> oo
The eigenvalues }, = s 2 in this case are no longer discrete but continuous, and
consist of all real nonnegative numbers. Thus, the representation (7.13) is a
generalization of eigenfunction expansion where summation with respect to the
eigenvalues is now achieved by integration.
Now suppose thatf is an even function that satisfies the conditions of Theorem
7. I. Since f (O cos s and f () sin s( are respectively even and odd functions
of, the formulas (7.14) yield
1
A(s) = -
n
J Im00
-co
cos s( d~ = ~ roc
n Jo
f(() cos s( d~
B(s) = ~ J~ co f() sin s d~ = 0
so that (7 .13) reduces to
(7.15) f(x) = ~
00
{ cos sx (I.() Jm cos s d~) ds

This is called the Fourier cosine integral formula. In case f is defined only on
the interval 0 s x < oo, (7.15) extends f for all x as an even function.
Similarly, if f is an odd function, then from (7 .14) we have
2
A(s) = 0, B(s) =
n
f"'o f<O sin s d
so that (7.13) reduces to the Fourier sine integral formula
(7.16) f(x) = ~ f" sin sx (f' f() sins d() ds
This extends f for all x as an odd function when f is defined only for x :?: 0.
Let us state these results in the following theorem.
THEOREM 7.2. Let f be a Junction that is piecewise smooth and absolutely

integrable on 0 s x < oo. Then f can be represented by the Fourier cosine
integral formula (7.15) on 0 s x < oo, or by the Fourier sine integral formula
(7.16) on 0 < x < oo. Each of the integrals com:erges to f(x) at all points where
f is continuous and to [f(x - 0) + f(x + O)J/2 at all points ll'here f is discon-
tinuous.
Notice that the Fourier cosine integral necessarily converges to f(O+) at

x = 0 and the sine integral converges to zero at x = 0.
Sec. 7 Fourier l11tegral 203
As in the case of (7.13), it is worth noting that (7.15) is a generalized eigen-

function expansion in terms of the eigenfunctions cos sx of the singular eigen-
value problem
11" + ;_,, = 0 (0 ~ x < co)
11'(0) = o. 11 bounded as x ---+ co
whose eigenvalues ;_ = s 2 consist of all nonnegative real numbers. Similarly,

(7.16) is a generalized eigenfunction expansion involving the eigenfunctions
sin sx of the singular problem
u + l.u
11
= 0 (0 ~ x < co)
11(0) = 0, 11 bounded as x ---+ co
2
with eigenvalues ). = s > 0.
Example 7.2. Find the Fourier cosine integral formula for the function
f(x) = f sin x (0 s: x s: n)
\0 (x > 0)
Solutions: Note that the function here is continuous for all x ~ 0. By (7.15) we have
f(x) = ~ L' cos sx (L" sin f. cos sf, df.) ds
The inner integral with respect to f. yields
f"
0
sin f. cos sf. df. = I
2
f" 0
[sin(! + s)f, + sin(l - s)f.) df.
f
_ I cos(l + s)f. + cos(! - s)f,l"
l+s2l 1-s 0
I p_ _- ~os( I + s )n + 1 - cos(! -__!lir]

2 l I + s I - s
= 1 f I + cos sn + I + cos snl

2l l+s 1-s
I + cos sn
I - s2
Hence, the Fourier cosine integral formula is
f(x) = -~ {"'
1 + C5>S s_ir cos SX dx
n Jo I - s2
(0 s: x S: n)
(x > n)
The integral converges for all x to the even extension of the function (Fig. 5.6).
! ( _\ J
FIG. 5-6 Effn extension off(x).
Example 7.3. Find the Fourier sine integral formula for the function given in Example
7.2.
Solution: By (7.16) we have
.
j (x) =
2 oo sin sx (foo
Jo sin(, sin sf,
d(,) ds
n: J 0
Since
Ln sin(, sins(, d(, = i f [cos(! - s)(, - cos(! + s)(,] d(,
I [sin( I - s); sin(! + s)(,l rr

2 I - s I + s j0
sin( 1 - s)n: _ sin(] + s)n:l
2 I 1-s l+s
I
2
I sin sn:
I - s
+
sin sn:
I - s2
the Fourier sine integral formula is

00
f (x) = 2 ( sin .rn sin sx ds
n: Jo I - s2
f sin x (0 s; x s; n:)
)0 (x > ;r)
Here, the integral converges to the function for all x ( - CfJ < x < oo ), since the
function is odd for all x (Fig. 5.7).
Sec. 7 Fourier Integral 205
!( \)
FIG. 5-7 Odd exre11sion ojf(x).
Exercises 5.6
1. Let
(0 < x < k)
I .(xl =
11
. \0 (othernise)
Show that the Fourier integral formula aprl ies to f and evaluate
{"' sin s(k - ,) + sin sx ds

Jo s
for all x.
2. Let
f(x) = jO (x s 0, x > 1)
\x (0 s x < I)
Show that the Fourier integral formula applies to /and evaluate

" s sin s(l - x) +cos s(l -
. x) - cos 1,
for all x.
r
0 s-'
ds
3. Represent the function

sin x (0 S x S n)
j(x) =
\o
'
(otherwise)
by Fourier integral formula and verify that

.,. cos s(x - n) ~ cos 1x
when 0 S x S n.
sm x =
n i 0 1 - s
" ds

(cos x Cx < n)
f (x)
\o (]xj > n)
by Fourier integral formula and verify that
cos x =
2
n
f"'-
o
s cos sx sin sn d
I - s2
s
when - n < x < Tt. What is the value of the integral at the points x n?
f(x) = fO (x < 0)
\e-x (x > 0)
by Fourier integral formula and evaluate
s sin sx + cos- sx
CS1
I + s2
for all x.
6. Show that the Fourier cosine integral formula applies to the function
f(x) = e-x (x 2 0)
to give the representation
e-x 2 f"' cos ds s.: (x 2 0)

n 0 I + s-
-x (0
/(x) = f 1
:$ x :$ I)
to (x > I)
can be represented by Fourier cosine integral formula and thus evaluate
- cos.\')
.I 2 COS SX ds
for all x 2 0.
8. Express the function/(x) e-x cos x. x 2 0, by Fourier cosine integral formula
and sho\\ that
e-xcosx= 2
n
f 0
s
s4
1
+ 2 cossxds
+ 4
(x 2 0)
9. Express the function/(.\) xe-x, x 2 0, by Fourier cosine integral formula and

ShOllthat
2
? (" I - 1
xe-x = ~ Jo (I + ~ 2 )2 cos sx ds (x 2 0)
10. Sho11 that the Fourier sine integral formula applies to the function
{~
(0 < x < I)
f(x) =
(X > l)
and evaluate
- ~os ~ sin sx ds
s
for all x 2 0.
Sec. 8 Fourier Tramform 207
I I. Establish the result
e-x sin x 4 [Cf' i

s sin sx cs
(x :0: 0)
n .o s2 + 4
'( x) ,. cos x (0 < x < n)
! =
lo (x > n)
can be represented by Fourier sine integral for!llula, and thus evaluate
" s(coss sn + J) .
for all x.
J0
2 _ I- sm sx ds
13. Establish the result

00
e-x = 2 { s sin sx ds
(x >OJ
n Jo I + s2
14. Express the function f(x) = xe-x, x :0: 0, by Fourier sine integral forlllula and
show that
_ x 4 [~ s sin sx ds
xe = (x :0: 0)
n .o (I + s2)2
8. Fourier Transform
The Fourier integral formula (7.6) can be written m a complex form if we

replace the function cos s(~ - x) by the expression
2
In fact, upon making this substitution for cos s( - x) in (7.6), we find
. I Joc ("' 1(,e) eisr~-x1 + e-i1(~-xl de" d.c

.f (x) = ~ o oc . ~J- 2 . ,
(8.1) I
2n:
J"' f"'
0 _ ""
f(s)eis(~-xl di;, ds
+ 71 f"' f"' f(()e-is(~-x) d( ds

_n o - oo
If in the first integral on the right we make the change of variable from s to -s
and then combine the result with the second integral, we obtain
(8.2) f(x) =
1
J"' f
2n _ oc _
00
'fj
f(()e-is(~-.x) d(, ds
This is the complex form of the Fourier integral formula.

Let us set
.f(x) f"'~ F(s)eisx ds

"\; 2n - ,~
which is now analogous to the complex form (2.6) of the Fourier series. Using
these symmetric formulas, we can restate Theorem 7 .1 as follows:
THEOREM 8.1. Let f be piecewise smooth and absolutely integrable on ( - oo, oo ),

and let
(8.3)
Then
(8.4) f(x) =
"\i2n
J:~
- ,_
F(s)t!isx ds
(11here f(x) is to be replaced by the arerage off(x - 0) andf(x + 0) at points

;rhere .f is discontinuous).
The function F defined by (8.3) is called the Fourier transform of .f, and .f
defined by (8.4) is the inverse Fourier transform of F. Along with other integral
transforms (e.g., Laplace transform), the Fourier transform is an invaluable tool
in applied mathematics.
Example 8.1. Find the Fourier transform of the function
/(xJ = j1 (jxj :o; al

\0 (I.\: > a)
Solution: By (8.3) the Fourier transform is
F(s) = . I r-e-i'']"
v' 2rr \/27! l is -a
ei.\a - e-istJ
2
is ?"
_/
112
= (;) sins a!.
Thus, by Theorem 8.1, we have
][ J
CL
- 00
sins-e
.as isx ds =
l ~
-,
0
(ixi < a)
(!xi = a)
(ix! > a)
Sec. 8 Fourier Transform 209
Next, we define the Fourier transform of a function that is defined only for
x :2': 0. In the Fourier cosine integral formula (7.15). let us set
(8.5) F,.(s) = (~)

112
J: f(O cos s~ d(
(8.6) f(x) = e) 112

J: Fc(s) cos sx ds
The function FJs) is called the Fourier cosine transform off. In view of the
symmetry between (8.5) and (8.6), we see that I is in turn the Fourier cosine
transform of F,(s); thus, f and Fe are cosine transforms of each other.
Similarly, from the Fourier sine integral formula (7.16), we define the Fourier
sine transform off as the function
(8.7) F,(s) = CJ
'2' 1/2 J"' f (() sin s( d(
0
This, in turn, has the Fourier sine transform
(8.8) f(x) = (2)- J"' f (s) sin sx ds

,7[
1/2
0
5
When f is a given function, the formulas (8.5) and (8. 7) can be regarded as the
solutions of the integral equations (8.6) and (8.8), respectively. Applications of
these various transforms will be illustrated in the chapters to follow.
Example 8.2. Find the Fourier cosine and sine transforms of the function
f(x) = e-ax [x 2'.: O; (a > 0))
Solution: It is clear that the function is smooth and absolutely integrable on

0 :<'. x < oc. By (8.5), its Fourier cosine transform is
Fh) = G) L" 112

e-a::, cos sC, dC,
112
2
) [ e-a::, (ssinsC, - acossC,)] "'
(n a2 + s2 0
(n2)
2
1/ a
a2 + s2
and. by (8.7), its Fourier sine transform is
F,(s) = G) 112
.C' e-a~ sin sC, dC,
e-~ 2
112
2 00
(n
) [ , (s cos sC, + a sin sC,)]
a- + s 0
(2)
1 2
1 s
11 a2 + s2
By (8.6) and (8.8), it then follows that
n:e-ax
J
oo COS SX
ds = (x :2: 0)
0 a 2
+ s 2
2a
and
("' s sin sx
(x > 0)
Jo a2 + s2 2
respectively.
Exercises 5.7
1. Find the Fourier transform of the function
f(x) =
I Ix/ (!xi s I)
{Q (1x > I)
1
2. Find the Fourier transform of the function f(x) = e-aJxl, where a > 0.
3. Find the Fourier transform of the function f(x) = I /(I + x 2 ). (Use the result of
Problem 6, Exercises 5.6.)
4. Show that the Fourier transform of the function f(x) = e-x' 12 is F(s) = e-'' 12 ;
that is, e-x'J 2 is its own Fourier transform. (Use the result of Problem 10,
Exercises 1.6.)
5. Let F(s) be the Fourier transform off(x). Show that-
(a) the Fourier transform of f(x)ei> is F(s - a).
(b) the Fourier transform off(ax) is (l/a)F(s/a).
(c) the Fourier transform of g(x), where
g(x) = { f (x - a) (x > a > 0)

. 0 (0 < x < a)
is e-''"F(s).
6. If tl1e Fourier transform off and/' exist andf(x)--> oo as !xi --> oo, show that the
Fourier transform of f'(x) is F(s).
7. Find the Fourier cosine transform of each of the following functions:
f(x) = f cos x (0 S x S rr/2)

(a)
\o (x > rr/2)
(b) f(x) = xe-x (x :2: 0)
8. Find the Fourier sine transform of each of the functions given in Problem 7.
9. Solve the following integral equations for the function f:
(a) Jg' f(s) sin sx ds = e-x cos x, x > 0.
(b) JO' f(s) cos sx ds = e-x sin x, x :2: 0.
Sec. 8 Fourier Transform 211
10. Let f and g satisfy the conditions of Theorem 8.1 and Jet F and G denote their
respective Fourier transforms. Define
h(x) = fxix f(x - y)g(y) dy
= f(x) * g(x)
This is called the convolution off and g.
(a) Show thatf(x) * q(x) = _q(x) * f(x); that is,
L: f(x - y)g(y) dy = J_~' f(x)g(x - y) dy
(b) Show formally that H(s) = (2n) 112 F(s)G(s), where His the Fourier transform
of h. (Take the Fourier transform of h and interchange the order of integration.)
11. Solve the integral equation
(x < 0)
h(x) =
(x ~ 0)
for the function f, and then verify part (b) of the preceding problem.
Chapter 6
The 1-leat Eguation
In this chapter we shall be concerned with some problems dealing with the
typical linear parabolic partial differential equation
cu F(x, t)
c:t
in two independent variables x and I. This equation, known as the one-dimen-

sional heat equation, arises in the analysis of conduction of heat in solids as
well as in a variety of diffusion phenomena. We shall formulate and study
various types of problems for this equation by reference to the conduction of
heat in a thin rod.
1. Derirntion of the Heat Equation
Consider a heat-conducting homogeneous rod, extending from x = 0 to

x = L along the x-axis (Fig. 6.1 ). Let us assume that the rod has uniform cross
section A and constant density p. Further, let us assume that the rod is insulated
laterally so that heat flows only in the x-direction, and that the rod is sufficiently
thin so that the temperature at all points of a cross section is constant. Let
u(x, I) denote the temperature of the cross section at the point x at an instant of
time t, and let c denote the specific heat of the rod (that is, the amount of heat
required to raise the temperature of a unit mass of the rod by I degree). Con-
sider a segment of the rod between the cross section at x and the cross section
212
Sec. 1 Derivation of the Heat Equation 213
0 \" + /',_\"
FIG. &-1 Heat co11ductio11 i11 a thin rod.
at x + Ax. According to the theory of heat flow, the amount of heat in this
segment is
x+Ax
(1.1) cpAu(s, t) ds
Q(t) =
Jx
On the other hand, the rate at which heat flows into the segment across the
cross section at xis known to be proportional to the product of the cross section
and the gradient of the temperature of the cross section; that is, it is equal to
OU
(1.2) -KA ~ (x, t)
ox
where K denotes the thermal conductivity of the rod. The negative sign in
( 1.2) is used to indicate the fact that heat flows in the direction of decreasing
temperature. Similarly, the rate at which heat flows out of the segment through
the cross section at x + Ax is given by
(1.3) -KA -
au (x + Ax, t)
ax
The difference between the amount of heat that flows in through the cross
section at x and the amount of heat that flows out through the cross section at
x + Ax must be equal to the change in the heat content of the segment x :s;
s :s; x + fix. Hence, by subtracting (1.3) from (1.2) and equating the result to
the time derivative of (I. I), we obtain
00 rx+Ax CU
-:::-,- = cpA - (s, t) ds
ct x at
(1.4)
KA au
~ - (x + Ax, t) - ~ - (x, t) J
au
=
[ox ox
We assume that the integrand in ( 1.4) is a continuous function of s. Then, by

the mean value theorem for integrals (Chapter I, Theorem 1.3), we have
x+Ax au OU
J x
;.;- (s, t) ds
ot
= -:;- (s, t) Ax
ot
(x < ~ < x + Ax)
214 The Heat Equation Chap. 6
so that (I .4) becomes
cp cu
_ (c;:, t L'i.x
v )
= K rau_ (x + ox,
A I) - _ (x, t )J
au
ct .ex ex
Dividing both sides of this equation by cp L'i.x and taking the limit as L'i.x -> 0,
we finally obtain
Du k
i?u
(x, t)
- (x, l)
ct iJx2
or
au (x, a2 u (x, I) K_)
(1.5)
ct
t) - k
<ix 2
= 0 (k =
cp,
by recalling the definition of partial derivative. Equation (1.5) is called the one-
dimensional heat equation. Jt is an important example of a partial differential
equation of the parabolic type. The constant k is usually called the diffusivity.
If heat is supplied to the rod from an external source at a rate, say, f(x, t)
per unit volume per unit time, then the term .f:~+tix f(s, t) ds must be added to
the right-hand side of (1.4). Thus, in passing to the limit as L'i.x -+ 0, we get
(1.6)
cu_ (x, t) - k ;:
a2 u (x, 1) = F(x, t)
ct cx 2
where F(x, t) = f(x, t)/cp is the source density. This equation is called the
nonhomogeneous heat equation.
On the other hand, if heat is being radiated from the rod into the surrounding
medium at a rate proportional to the temperature of the rod, then we shall have
the so-called radiation equation
(1.7)
OU
_ - (x, l) -
c1 u
k _ 2 (x, I) - hu(x, t) = 0
ct cx
where h is a positive constant. A combination of ( 1.6) and (1. 7) will, of course,
lead to the nonhomogeneous radiation equation
(1.8)
cu (x, t)
c2
u
/.,: (;x 2 (x, t) + hu(x, t) = F(x, 1)
2t
2. Initial and Boundary Conditions
In each of the cases of heat conduction in a thin rod discussed above, we saw
that the temperature function u satisfies a partial differential equation of par-
abolic type. From physical considerations, we know that the differential equation
alone cannot determine the temperature distribution in the rod at any subsequent
time. We need to have additional information regarding the initial temperature
of the rod and the conditions imposed at its two ends. This means that we must
Sec. 2 lnitial and Boundary Co11ditio11s 215
specify u(x, t) at some initial time (say. at time t = 0) and describe how the
two ends of the rod exchange heat energy with the surrounding medium. Thus,
with reference to the case in which there is no heat source, if the temperature
of the rod at t = 0 is described by the function f(x), 0 ::;: x ::;: L, and the two
ends are maintained at zero temperature at all time, then the temperature
distribution u in the rod at any later time t > 0 is found by solving the differential
equation
(2.1) u1 - ku.u = 0 (0 < x < L, t > 0)
subject to the conditions

(2.2) u(x, 0) = f(x) (0 ::;: x ::;: L)
and
(2.3) u(O, t) = 0, u(L, t) = 0 (t ;:::: 0)
This problem is called an initial-boundary value problem for the heat equation.
The auxiliary conditions (2.2) and (2.3) are called initial condition and boundary
conditions, respectively.
We notice in the problem (2.1 ), (2.2), (2.3) that only the value of u is prescribed
initially and not both u and u,, as in the case of the wave equation. This is,
of course, dictated by the physical problem being considered. But even from a
mathematical standpoint, 11 1 cannot be prescribed arbitrarily, since it is related
to u.u initially through the differential equation. As a matter of fact, we shall
see in the next section that the set of auxiliary conditions (2.2) and (2.3) is
appropriate for the heat equation or for any of the other related differential
equations given in the preceding section, in the sense that the problem is well
posed.
The conditions (2.3) correspond to the fixed boundary conditions for a
vibrating string, which we considered in Chapter 3 for the wave equation.
Other forms of appropriate boundary conditions corresponding to the free and
the elastic boundary conditions for a vibrating string can also arise in our heat
conducting problem. Indeed, if both ends of the rod are insulated so that there
is no heat flow across the ends, then according to (1.2) the boundary conditions
assume the form
OU
;;-- (0, t) = 0, ~u (L, t) = 0 (t ;:::: 0)
(2.4)
OX ox
.. On the other hand, if there is radiation of heat at the ends of the rod into the
surrounding medium, which is kept, say, at zero temperature, then the boundary
conditions will be of the form
OU
- - (0, t) + hu(O, t) = 0
(2.5)
ox
OU
-- (L, t) + hu(L, t) = 0
OX
for t ~ 0, where h is a constant. Notice that all these boundary conditions arc
linear. As in Chapter 3, we shall refer to these conditions, (2.3), (2.4), and (2.5),
as boundary conditions of the first, second, and third kind, respectively.
It is also possible to consider the problem of finding a solution of the heat
equation or any of its related equations in the infinite domain - oo < x < oo,
t > 0, which satisfies only the initial condition (2.2) prescribed for - oo <
x < oo. Such a problem physically represents conduction of heat along an
infinite thin rod; accordingly, the problem is called an initial value problem.
It will be seen in later sections and exercises that an initial value problem for the
heat equation and its related differential equations is well posed.
Exercises 6.1
1. Verify that each of the given functions satisfies the heat equation ( 1.5) for O <
x < re and the accompanying initial and boundary conditions.
(a) u(x, t) = e-kt sin x; u(x, 0) = sin x, u(O, f) = u(rc, t) = 0.
(b) u(x, t) = e-kc cos x; u(x, 0) = cos x, ux(O, t) = ux(rc, t) = 0.
(c) u(x, t) = ++ 1e- 4 kt cos 2x; u(x, 0) = cos 2 x, uO, t) = ux<rr, t) = 0.
2. Verify that the function
] -x2/4t
u(x, f) = - e
Vt
satisfies the heat equation ( 1.5) for k = l, t > 0 and that
Jim u(x, I) = 0 (provided x "# 0)

t--+O+
3. Consider the initial-boundary value riroblem
11, - ku.u = F(x, f) (0 < X < L, I > 0)

u(x, Q) = f(x) (0 :S x ::; L)
u(O, t) = a(t), u(L, f) = b(t) (! > 0)
(a) Determine a function of the form </J(x. f) = A(t) + xB(f) such that
</J(O, /) = a(t) and </J(L, t) = b(f ).
(b) By introducing a new function v, defined by v = u - , reduce problem
(a) to one in which the boundary conditions are homogeneous.
4. Consider the initial-boundary value problem in Problem 3, with the boundary
conditions uxCO, t) = a(t), ux(L, t) = b(t). Determine a function such that
</Jx(O, t) = a(t) and </Jx(L, t) = b(t) and thus reduce the problem to one with
homogeneous boundary conditions.
5. Reduce the problem
u, - kuxx = 0 (0 < x < 7T, f > 0)
u(x, 0) = sin x (0 :S x :S rr)
u(O, t) = t 2 , u(rr, t) = e' (t > 0)
to one with homogeneous boundary conditions.

Sec. 3 The ,V/aximum Principle and Uniqueness Theorem 217
11, - kuu = xt (0 < x < I, 1 > 0)

11(x, 0) = x (0 < x < I)
ux(O, 1) = sin 1, 11(1, t) = ln(l + t) (1 > 0)
to a problem in which the boundary conditions are homogeneous.

7. Consider the problem
fl, - ku.u = Ae-ax (a > 0, 0 < X < L, t > 0)

11(x, 0) = f(x) (0 ~ x ~ L)
11(0,1)=0, 11(L,t)=O (t~O)
where A is a constant. Introduce a new function v, defined by u(x, t) v(x, r) +

cp(x), where is a function of only the variable x. (a) Determine so that v
satisfies v, - kvxx = O; (b) determine so that z: also satisfies v(O, t) = 0 and
v(L, t) = 0, and thus reduce the given problem to one with homogeneous
differential equation and boundary conditions.
11, - kuu = 0 (x > 0, t > 0)
u(x, 0) = 0 (x > 0)
u(O, t) = h(t) (t > 0)
to one in which the boundary condition is homogeneous.

11 1 - kuxx + bu = 0 (x > 0, t > 0)
u(x, 0) = 0 (x > 0)
ux(O, t) = h(r) (1 > 0)
to one in which the boundary condition is homogeneous.

10. Let 11 denote the solution of the initial-boundary value problem
11, - kuxx = 0 (x > 0, t > 0)

11(x, 0) = f(x) (x ~ 0)
ux(O, t) + /111(0, I) = 0 [1 ~ O; (h = constant)]
where u is sufficiently differentiable. Set v(x, t) = uxC>:, 1) + hu(x, t ). Show

that v satisfies the heat equation, the initial condition v(x, 0) = f'(x) + hf(x),
and the boundary condition v(O, t) = 0. Assuming that v has been found, what
will be a solution u of the original problem?
3. The Maximum Principle and Uniqueness Theorem
Before we proceed to the consideration of an initial-boundary value problem

for the heat equation, we shall first establish an important property that is
possessed by every continuous solution of the heat equation in a rectangular
region. This property, known as the maximum principle, enables us to prove
that an initial-boundary value problem is indeed well posed for the heat equation.
'
218 The Heat t.q11atio11 Chap. 6
THEORE!\J 3.1 (Maximum Principle) Let u be a so/11tio11 of the heat equation

(1.5). which is continuous in the rectangular region 0 s x s L 0 s t s T.
Then u assumes its maximum rnlue either at the base t = 0 or at the sides x = 0
and x = L (Fig. 6.2).
I= T
"= 0 .1 ~I
0 ! = Q
FIG. 6-2 The maximum principle.
Proof: We first prove that the maximum principle holds for a continuous
function 1" that satisfies the differential inequality
(3.1)
for 0 s x s L, 0 s t s T. Indeed. suppose v assumed its maximum value at

(x 0 , 10 ). where 0 < x 0 < Land 0 < 10 s T. Then, from elementary calculus,
\\e know that
and
Since k > 0, these lead to the inequality
\Vhich is contrary to the fact that 1' satisfies (3.1) at (x 0 , t 0 ). Hence, i: cannot
assume a maximum at (x 0 , 10 ). Since i: is continuous in the closed region
0 s x s L, 0 s t s T, 1 has a maximum value. It must therefore occur
either at t = 0 or x = 0 or x = L.
Now consider a continuous solution 11 of the heat equation (1.5) in the region
0 s x s L, 0 s t s T. Let u s 1V/ at t = 0 for 0 s x s L, and at x = 0
and x = L for 0 s t s T. We shall show that 11 s M at all points in the
rectangular region. Define the function i: by the equation
r(x, I) = u(x, t) + BX
2
Sec. 3 The /\1ax:i11111111 Principle a11d Uniqueness Theorem 219
where c: is an arbitrary positive number. Then
1' 1 -- k1\., = 11 1 - k11xx - 20 = -28 < 0
so that z; satisfies (3.1 ). Hence, by the previous result, 1' must assume its maximum
value at t = 0 or x = 0 or x = L. On all these, we see that
1: = 11 + ex
2
s M + c:L2
Therefore, at all points in the region 0 s x s L, 0 s t s T, it is true that
u = v - r:x 2 s v s M + cL2
for any c: > 0. Letting e --. 0, we find that 11 s M for 0 s x s L and 0 :::::;
t s T, as we wanted to show. This establishes our theorem.
We remark that the maximum principle stated in Theorem 3.1 is often referred
to as the weak maximum principle, since it permits the maximum of the solution
to occur both on the boundary and at interior points. A stronger version of
this principle, which contains Theorem 3.1 as a special case, asserts that if the
maximum occurs at an interior point, then the solution must be constant in a
certain region. The proof of this assertion is so involved that we shall not bother
ourselves with it.
By applying the maximum principle to the negative of the function 11, we
obtain a similar result for the, minimum value of 11. That is, the minimum value
of a continuous solution of the heat equation is also taken either at t = O or
x = 0 or x = L. Physically, the maximum-minimum principle says that the
temperature in the rod at any instant of time cannot get higher (or lower)
than the highest (or lowest) temperature that occurs initially or which is yet to
be observed at the ends of the rod.
As a consequence of the maximum principle, we can prove a uniqueness
theorem for the following initial-boundary value problem for the heat equation:
11, - kuu = F(x, t) (0 < x < L, t > 0)

... s s
(3.2) u(x, 0) = f(x) (0 x L)
u(O, t) = a(t), u(L, t) = b(t) (t :;:o: 0)
THEOREl\1 3.2 A solution of the initial-boundary value problem (3.2) which is

continuous for 0 s x s L and t ::::: 0 is uniquely determined.
Proof: Suppose that i: and ware two solutions of the problem (3.2). Then
11 = v - w satisfies the homogeneous heat equation and vanishes at t = 0,
x = 0, and x = L; that is, u(x, 0) = 0, u(O, t) = 0, and u(L, t) = 0. Hence,
by the maximum principle, the value of u inside the region 0 s x s L, t :::0: 0,
cannot be greater than zero; likewise, the minimum value of u cannot be less
than zero. Therefore, u must be identically zero throughout the region and so
z: and w are identical solutions.
Another consequence of the maximum principle is the following theorem on

the continuous dependence of the solution of (3.2) on the initial and boundary
conditions.
THEOREM 3.3 Let u; be the solution of the problem (3.2) corresponding to

initial-boundary conditions u(x, 0) = f;(x), u(O, t) = a;(t), u(L, t) = b;(t) (i =
1, 2). If lf1 (x) - fz(x)I s e, la 1 (t) - az(t)I s e, and lb 1 (t) - h 2 (t)I s e,
then
for all (x, t) where 0 s x s Landt :::::: 0.
Proof: Set u = u 1 - u 2 ; then u satisfies the homogeneous heat equation

(1.5). The hypothesis of the theorem implies lul s e at t = 0, x = 0, and
x = L; that is,
(3.3)
on the base and on the sides of the region 0 s x s L, t :::::: 0. By the maximum
principle, the inequality (3.3) holds throughout this region. Thus, the theorem
is proved.
In a later section we shall see how a solution of the problem (3.2) can be
constructed. Then, in view of Theorems 3.2 and 3.3, we can conclude that the
problem (3.2) is well posed; that is, the problem has a unique solution that
depends continuously on the initial and boundary data.
We conclude this section by establishing a corresponding uniqueness theorem
for the initial value problem
u, - kuxx = F(x, t) (-oo < x < oo, t > 0)

(3.4)
u(x, 0) = f(x) (-oo<x<oo)
We shall require that our solution of this problem be bounded in the infinite
region - oo < x < oo, t :::::: 0. Then we have-
THEOREM 3.4 A solution of the initial value problem (3.4) which is continuous
and bounded for - oo < x < oo, t ;::::: 0, is uniquely determined.
Proof: Suppose v and w are two continuous solutions of the problem (3.4)
such that
Jvl s M, lwl s M
for - oo < x < oo, t ;::::: 0, where M is a constant. Set u = v - w; then,
clearly, u satisfies the homogeneous heat equation (1.5), the initial condition
u(x, 0) = 0, and the inequality
lul s Iv - wl s lvl + lwl s 2M

Sec. 3 The Maximum Principle and Uniqueness Theorem 221
Let us consider the finite region - L s x s L, 0 s t s T, where Land Tare

arbitrary numbers, and define
V(x, t) = 4M
L (I x-
2
7
+ kt ')
2
It is easily verified that, in this region, V satisfies equation (1.5) and that, at
t = 0,
2Mx 2
V(x, 0) = - ; - 2 u(x, 0) = 0
L
and, at x L,
4Mkt
V(L, t) = 2M + L 2 2 2M 2 u(L, t)
Hence, by the maximum principle, it follows that
V(x, I) 2 u(x, I) (- L s x s L, 0 s I s T)
Similarly, by considering the function - V(x, 1), we obtain
u(x, t) 2 - V(x, 1) (- L s x s L, 0 s 1 s T)
Consequently, for each point (x, t) in the region - L s x s L, 0 s t s T,
we have
(3.5) lu(x, t)I s V(x, t) = 4M

L2 (12 x
2
+ kt )
for arbitrary numbers L and T, no matter how large. Letting L -+ oo, we then
find lul = 0, which means that v(x, t) = w(x, t) for - oo < x < oo, t 2 0.
Exercises 6.2
1. Show that if u and v are two solutions of the heat equation (1.5) for 0 < x < L
such that u(x, 0) S v(x, 0), u(O, t) S v(O, t), and u(L, t) s v(L, t), then
u(x, t) S v(x, t) for 0 S x S L, t 2. 0
2. Let t1 be a solution of the heat equation ( l.5) which is sufficiently differentiable
for 0 S x s L, t 2. 0, and set v(x, t) = ux(x, t ). Show that v satisfies the
maximum principle.
3. Let 11 be a solution of the nonhomogeneous heat equation (1.6), with 0 < x < L,
t > 0, which is continuous for 0 s x s L, t 2. 0. Prove that if F(x, t) < 0,
then the maximum of u is attained at t = 0, or x = 0, or x = L.
4. Show that if t1 is a c9ntinuous solution of the heat equation (1.5) for 0 s x s L,
t 2. 0, such that ux<O, t) = 0, then the maximum of u is attained either at t = 0
or at x = L. Hint: Apply the maximum principle to the function v over - L s
x s L, 0 s t s T, where v(x, t) = u(x, t) for 0 s x s L and v(x, t) =
u(-x, t) for-Ls x s 0.
5. Let 11 be a solution of the heat equation for 0 < x < L, t > 0, such that 11(0, t) =
u(L, t) = 0. Consider the identity
0 = (OL
J
11(11, - kuu) dx = I_ ~-
2 ct
(L
Jo
11
2
dx - k r,
Jo
llUu dx
(a) By integrating by parts the second integral on the right, deduce that
, lL
~ 11
2
dx s 0 (t ? 0)
ct 0
This means that the function r(/) = J~ u 2 (x, t) dx is nonincreasing fort ? 0.

(b) If u(x, 0) = 0, 0 s x s L, obtain the result u(x, I) = 0 for 0 s x s L,
t ? 0, and thus give an alternate proof of Theorem 3.2.
6. Let 11 be a solution of the initial value problem
( - 00 < x < oo, f > 0)

u(x, 0) = f(x) (-00 < x < 00)
where f is continuous on every finite interval. If 11 tends to zero uniformly in t

as Jxl tends to oo, show that Ju(x, t)! s maxl/(x)J. Hint: Apply Theorem 3.1
to the strip-Ls x s L, t ? 0, and let L--> oo.
7. Formulate and prove the corresponding maximum principle for the two-
dimensional heat equation 11, - k(uxx + u,) = 0 in the cylindrical region
n = {(x, y, I) I (x, y) ED + c. 0 s I s T}, where D is a domain in the xy-
plane bounded by a continuous curve C.
8. Prove that the following initial-boundary value problem
111 - k(uxx + llyy) = 0 [(x, y) in D, 0 < f < T]

u(x, y, 0) = f(x, y) [(x, y) in D]
u = g(x, y, t) [on C, 0 s t s T]
h<Js at most one solution. Here, D is a domain in the xy-plane bounded by the
continuous closed curve C.
9. Let /1 be a solution of the two-dimensional heat equation of Problem 8 such that
u = 0 on C for 0 s t s T. By considering the identity
0 = ff 11[111 - k(uxx + llyy)] dx dy = 0
over the domain D for 0 s t s T and using Green's identity [(8.6), Chapter l ],
show that
Hence, deduce that u(x, y, t) = 0 in D for 0 s t s T if u(x, y, 0) = 0. This

provides an alternative proof of the uniqueness of solution for Problem 8.
Sec. 4 bdtial-Bo1111dary Value Problems 223
4. Initial-Boundary Value Problems
Let us now consider the problem of finding a solution of the homogeneous

beat equation
(4.1) u, - kuxx = 0 (0 < x < L, t > 0)
which satisfies the initial condition
(4.2) u(x, 0) = .f (x) (0 ::::; x ::::; L)
and the boundary conditions
(4.3) u(O, t) = 0, u(L, t) = 0 (t z 0)
We shall solve this problem by the method of separation or variables presented

in Section 9 of Chapter 3. Accordingly, we seek particular solutions of (4.1)
in the form
(4.4) u(x, t) = X(x)T(t)
which satisfy the boundary conditions (4.3). Substituting (4.4) in (4.1) and
separating the variables, we obtain
T' X"
-)
kT X
This leads to the two ordinary differential equations X'' + IX = 0 and
(4.5) T' + J.kT = 0

for the functions X and T, respectively,
In order for (4.4) to satisfy the boundary conditions (4.3), we must have
X(O) = 0 and X(L) = 0. Thus, the function X is determined by solving the
Sturm-Liouville problem
X'' + i.X = 0 (0 ::::; x ::::; L)
(4.6)
X(O) = 0, X(L) = 0
This is the same problem we encountered in Section 9 of Chapter 3, for which

we found the eigenvalues
(4.7) (11=1,2,. .. )
and the corresponding eigenfunctions

. IJJTX
(4.8) X n (X ) = Sll1 (n = 1, 2,. .. )
L
Now, for each }"n = 11 2 n 2 / L2, n z I, a solution of the equation (4.5) is given by
(4.9) T,,(t) = e-ki,,r

224 Tlze Heat Equation Chap. 6
Thus, the functions
(4.10) u 11 (x, t) = e - u,,r sin n nx (11 = I, 2, ... )

L
with An given by (4. 7), are all particular solutions of (4.1) satisfying the homo-
geneous boundary conditions (4.3).
To obtain a solution of our problem, we consider a series of the functions
( 4.10) in the form
(4.11) u(x, t) = f
n=I
b e-u."' sin nnx
n L
and determine the coefficients bn so as to satisfy the initial condition (4.2).

Thus, setting t = 0 in (4.11 ), we see from (4.2) that the coefficients b" must
satisfy the relation
00
nnx
(4.12) f(x) = L bn

Sll1 - - (0 s x s L)
n= I L
This is precisely the Fourier sine series representation of f on the interval

[O, L]. Therefore, the coefficients bn are given by the formula
(4.13) b = 2-
n L
f
o
L
f (x) .
sm -117TX
L
-- dx (n = I, 2, ... )
The function u represented by the series (4.11 ), with b" given by (4.13), is the
solution of our problem. In order to verify this, we now assume that the func-
tion fin (4.2) is continuous and piecewise smooth on [O, L] and that f(O) =
f(L) = 0. Then, according to Theorem 6.2 of Chapter 5, the Fourier sine series
(4.12) off converges absolutely and uniformly to the function on [O, Now, LJ.
for t z 0,
00 00
I . . nnxl
L ! b11 e-ki.,,t Sll1 - - 1 S lbnl L
11= I [ L I n= I
where the series on the right converges. Therefore, the series (4.11) converges
absolutely and uniformly to u(x, t) for 0 s x s Land t z 0. Since each term
of the series is continuous and satisfies (4.3), it follows that u, too, is continuous
and satisfies (4.3). Moreover, on setting t = 0, we see from (4.12) that u satisfies
the initial condition (4.2) as well.
There remains to be verified only the fact that u satisfies the heat equation.
To this end, we need to show that the series (4.11) can be differentiated term by
term, once with respect to t and twice with respect to x. Let us consider the
sen es
(4.14) - ~ k'An bne -Ont SITI

1.... . 117TX
---
n =I L
which is the formal derivative of(4.l l) with respect tot. Sincefis continuous on
[O, L ], it is bounded there and so there is a constant M such that Jbnl :::; M for
all n ~ l. Then, clearly, for any t 0 > 0,
whenever t ~ t 0 . Since the series with the general term k/."M exp(-kJ.nto)
converges, it follows by the Weierstrass M-test that the series (4.14) converges
uniformly for 0 :::; x :::; L and 0 < 10' :::; t. This shows that (4.11) has con-
tinuous derivative with respect to t for 0 :::; x :::; L, t > 0, which can be ob-
tained by differentiating the series term by term.
In the same way we can establish that ( 4.11) has continuous second-order
derivative with respect to x for 0 :::; x :::; L, t > 0, which is obtainable by
term wise differentiation. Thus, substituting (4.11) in equation (4.1 ), we see that
-u nnx , ,
u, - kuxx = L00
11=1
b,,e
r .
" sm - -- [ -k1.
L
11 + kJ. 11 ] = 0
This completes the verification that (4.11) with (4.13) is a solution of the
problem ( 4.1 ), ( 4.2), ( 4.3) under the conditions thatf is continuous and piecewise
smooth on [O, L] and vanishes at the end points. That this solution is uniquely
determined follows by Theorem 3.2.
It is noteworthy that by applying repeatedly the procedure we have described
above, the series solution (4.11) can be shown to have continuous derivatives
of all orders with respect to x and t for 0 :::; x :::; L and t > 0. This result is
quite in contrast with the result we obtained in an example problem in Chapter
3, Section 9, involving the wave equation. Although there the function! satisfies
the same conditions as given above, yet the series solution cannot be differentiated
twice, term by term; we have had to resort to the concept of generalized solution.
This is a basic difference between solutions of hyperbolic and parabolic differen-
tial equations.
If we substitute (4.13) for b,, in (4.11) and interchange the order of summation
and integration, we can write ( 4.11) in the integral form
(4.15) u(x, t) = IL G(x, t; )!(~) d

where
2 00
nn _ nnx
(4.16) G(x, t; ) = -- L e
-u .
..,r sm - sm - (t > 0)
L n=1 L L
with )." = n 2 n 2 / L2 , n ~ 1. Notice that for t > 0, the series ( 4.16) is uniformly
convergent in so that the interchange of summation and integration in ( 4.15)
is perfectly valid. The function G defined by (4.16) is known as the Green's
function for the heat equation corresponding to the boundary conditions (4.3),
and the series is called its bilinear expansion [Chapter 4, Section 7]. Physically,
G(x, t; i:) represents the temperature distribution in the rod for 0 :<o: x :<o: L,
I > 0, due to a concentrated heat source at the point x = ( at time t = 0,
with the temperature at the end points maintained at zero degree.
Jt is easy to verify from (4.16) that: (i) G satisfies the heat equation with respect
to x and t for t > 0. provided x -=/= (; (ii) G is continuous at x = (; (iii) G
satisfies the boundary conditions (4.3) with respect to the variable x; and (iv)
G is symmetric with respect to x and (. These properties are analogous to the
properties possessed by a Green's function for an ordinary differential equation
[see (2.15), Chapter 4].
11, - uu = 0 (0 < X < IT, { > 0)

u(x, 0) = sin x (0 ::<; x ::<; rr)
11(0, 1) = 11(rr, !) = 0 (t 2 0)
Solution: By (4.13) \\e have
b,, = -? ln sin x sin nx dx

n o
(11 = I)
(11 > 1)
Hence, the solution is u(x, t) = e-' sin x.
11, - I/xx= 0 (0 < x < n, t > 0)

u(x, 0) = x(rr - x) (0 ::<; x ::<; n)
11(0, t) = 0, 11(rr, t) = 0 (I 2 0)
Solution: According to (4.11 ). the solution is given by

<fO
u(x, I) = E bne-"
21
sin nx
fl=l
where, by (4. 13 ),
b,, = ~ I' x(n - x) sin 11x dx
J - (-1)"
= 4 -- -
77:113
Hence, we have
8 ro -(211- 021
u(x, t) = E e_ _ _ . sin(2n - l)x
7r 11 ~ I (211 - J) 3
In case both ends of the rod are insulated so that the initial-boundary value
problem consists of (4.1 ), ( 4.2) and the boundary conditions
(4.17) 11)0. 1) = 0, u,(l.. I) = 0 (t ~ 0)
the method of separation of variables leads to the Sturm-Liouville problem
(4.18) X" + i.X = 0, X'(O) = 0, X'(L) = 0
for the function X together with equation (4.5) for the function T. We recall
that the eigenvalues of the problem (4.18) are )"n = n 2 n 2 / L2, n = 0, 1, 2, ... ,
with the corresponding eigenfunctions X,,(x) = cos(nnx/ L), 11 = 0, 1, 2, ... .
Thus, the functions
_,-, nnx
( 4.19) II,, ( X, t) = e ,_,, COS (n = 0, I, 2, ... )
L
are all particular solutions of ( 4.1) satisfying the boundary conditions ( 4.17).
Now, to obtain a solution of the problem ( 4.1 ). ( 4.2), ( 4.17), we consider a
series consisting of the functions (4.19) in the form
a0 ~ _ u , n nx
(4.20) u ( x, t) = + L a,,l' .,, cos --
2 ,, i L
The initial condition (4.2) then requires that the coefilcicnts a,, satisfy the relation
a nnx
f (x) = ---0 +
2 n=1
L
oc,
a,, cos
L
which is the Fourier cosine series representation of the function f on the interval
0 :S x :S L. Hence, the coefficients are gi\'en by the formula
(4.21) a,, = 2 JL f(x) cos nrrx

'_ dx (n = 0, 1,2, ... )
L 0 L
Thus, a solution of the problem (4.1), (4.2), (4.17) is given by (4.20), with a,,
(n ~ 0) given by ( 4.21).
The verification that (4.20) with (4.21) satisfies (4.1 ), (4.2), and (4.17) under the
conditions thatf is continuous and piecewise smooth on the interval 0 :S x :S L
is done in the same manner as in the previous case. That the solution is uniquely
determined follows from a modified maximum principle corresponding to the
boundary conditions ( 4.17) (Problem 2, Exercises 6.2).
If we substitute (4.21) for the coefficients a,, in (4.20), the solution can be
written in the integral form
(4.22) u(x, t) = f G(x, t; ()f (O d(
where, as in the previous case,
1 2 ~ -u nn( nnx
(4.23) G(x, t; () = -L + ~ e "1 cos ----- cos -
L n=I L L
is called the Green's function corresponding to the boundary conditions (4.17).

Here, G is again represented by its bilinear expansion.

(0 < x < 71, t > 0)
u(x, 0) = x (0 s x s 71)
ux(O, t) = 0, (t 2 0)
Solution: By the formula (4.21) we find
ao =
2
71
2
rl"
0
xdx=n
x cos nx dx = l
2 x sin nx + cos nx J
a11 =
71 O 71 n II
2
1:
2 (-1) 11 - l
(11 2 I)
Thus, a" = 0 when 11 is even and a11 = -(4/nn 2 ) when 11 is odd; therefore
-4
(k 2 I)
n(2k - I )2
Hence, by (4.20), the solution is

oo e-(2k-1)2r
71 4
u(x, t) = :2 2: - cos(2k - l)x
7r k=I (2k - 1) 2
Exercises 6.3
In Problems I through 6, find the solution of the initial-boundary value problem

for the heat equation (4.1 ), 0 < x < n, t > 0, satisfying the given initial and boundary
conditions.
1. u(x, 0) = A, A a constant; u(O, t) = 0, u(n, t) = 0, t > 0.
2. u(x, 0) = sin 3 x; 11(0, t) = 0, u(n, t) = 0, t 2 0.
3. u(x, 0) = cos 2 x; llx(O, t) = 0, ux(n, t) = 0, t 2 0.
4. u(x, 0) = x(n - x); u(O, t) = 0, 11(11, t) = 0, t 2 0.
5. u(O, t) = 0, u(n, t) = 0, t 2 0,
u(x, 0) = {x71 - x
(0
(n/2
s x s rr/2)
s x s n)
6. u(x, 0) = x2 - n 2 ; ux(O, t) = 0, u(n, t) = 0, t 2 0.
!11 - kuxx - U = 0 (0 < x < 1, t > 0)
u(x, 0) = x(I - x) (0 s x s I)
u(O, t) = 0, (t 2 0)
Sec. 5 Nonl10111ogeneo11s Initial-Boundary Value Problems 229

11, - kuxx +II= 0 (0 < x < n/2, t > 0)
u(x, 0) = cos x (O .s x .s n/2)
11x(O, I) = 0, 11(n/2, t) = 0 (I ? 0)

llr - llxx - LI = Q (Q < X < 77:, t > Q)
u(x, 0) = sin 3 x (0 .s x ::; n)
u(O, t) = 0, u(r., t) = 0 ( 1 ? 0)
10. Find the solution of the initial-boundary value problem

i, - /a-.u + hv = 0 (0 < x < L, t > 0)
r(x, 0) = f(x) (0 .s x .s LJ
t)O, t) = 0, i;_.( L, I) = C (I ? 0)
where h is a positive constant, and show that i"(x, I) = u(x, t )e-h', with u given
by (4.20).
11. Find the solution of the initial-boundary value problem
/1 1 - Uu + LI = Q (0 < x < I, t > 0)
2
u(x, 0) = x (0 s x s 1)
u)O, 1) = 0, 11(1, t) = 0 (t ? 0)

11 1 - kuxx = 0 (0 < X < L, t > OJ
u(x, 0) = f(x) (0 S x S L)
11(0, t) = 0, LlxCL, t) + hu(L, t) = 0
(t ? 0; lz a positive constant)
Ll 1 - kuu = 0 (0 < x < L, t > 0)
u(x, 0) = f(x) (0 s x s L)
u/O, t) + hu(O, t) = 0, u(L, t) = 0
(t ? O; lz = const > 0)
5. Nonhomogeneous Initial-Boundary Value Problems
The problems we have considered so far all involved homogeneous differential

equation and homogeneous boundary conditions. In this section we shall
consider problems in which either the differential equation or the boundary
conditions, or both, are nonhomogeneous. Such problems will be called
nonhomogeneous problems. The method that we shall describe here for solving
these types of problems is a variation of that described in Section 7 of Chapter
4 for ordinary differential equations.
.
230 Tlie Heat Equation Chap. 6
We first consider the problem consisting of the nonhomogeneous heat

equation
(5.I) u, - ku'"' = F(x. I) (0 < x < L, t > 0)
together with the initial condition (4.2) and boundary conditions (4.3). We
assume a series solution or this problem in the form
. _ ,, . ll71X
(5.2) 11(x. I) - w 11,,(t) c,111-
" I L
where the coefficients 11., are functions or I. This is actually the eigenfunction
expansion of the yet unknown function 11. in terms of the eigenfunctions (4.8)
of the associated Sturm-Liouville problem (4.6). Thus, the coeftlcients 1111 must
be related to the function u by the formula
(5.3) 11 11 (1 ) = 2 JL u(x, t) s111-~-

. 1171X Jx (n = I, 2, ... )
L 0 L
Now let us a~sume that 11 is twice continuously differentiable for 0 s x s L,
t ;:::.: 0, and that the function Fin (5.1) can be represented in Fourier sine series
(5.4) f(x, t) = L" F 11 (f) sin !25..:.

11=1 L
so that
') J'L ll71X

(5.5) F.,(t) = :. F(x, t) sin - dx (11 = 1, 2, ... )
L 0 L
Then, by differentiating (5.3) with respect tot and using (5.1), we obtain
(5.6) 11;,(1) = L2 JL
11, sin
.
L
lliTX
dx
0
= 2k
-L J'L Un
. lliTX
Sin - - dx + F,,(t)
.. L
0
where F 11 (t) is given in (5.5). By using the boundary conditions (4.3), the first
term on the right of equation (5.6) can be integrated by parts twice, to give
2/.: r.L Uu
. 1171X d
sm -- x = - k. ( )
1. 1., t
L O L
where }. 11 1s as given 111 (4.7). Thus, (5.6) leads to the ordinary differential
equation
(5.7) 11:,(t) + k).,,u,.(t) = F,.(t)

Sec. 5 Nonhomogeneous lnitial-Bo1111dary Value Problems 231
for the functions 11", n = I, 2, .... The initial condition (4.2) implies that
(5.8) 11,,(0) =
2 JL f(x)sin
. . l/7L'(
dx = b,, (n = I, 2, .. )
L 0 L
Hence. by solving the initial value problem (5.7), (5.8), we find
(5.9)
for /1 = 1, 2, . . . . Substituting this in (5.2), we thus obtain
(5.10) u(x, t) = f: JJ'

l
11=! O
e-U,,(1-rlF,,(T) ch ( sin
(
nn~ +
L
f
n=I
b,,e-u .. i sin nnx
L
for a solution of the problem ( 5.1 ), ( 4.2), (4.3 ).

To verify that (5.10) satisfies (5.1 ). (4.2), and (4.3), we need show only that the
first term on the right of(5. IO) satisfies (5.1), (4.3), and vanishes at t = 0 because
the second term is just the solution (4.11) of the homogeneous initial-boundary
value problem (4.1), (4.2). (4.3). Thus, set
~ . nnx
(5.11) i:(x, t) = ... u,,(t) sm -
11=1 L
with
(5.12) v,,(t) = [' e-k;.,,u-rlF,,(T) ciT (n ~ 1)

0
If we assume that F is continuous for 0 s x s L, t ~ 0, then for any fixed

T we see from (5.5) that
2 rL
IF,,(1)1 s I. \F(x, t)i dx s 2M
L O
where M denotes the maximum value of IFI in the region 0 s x s L, 0 s

t s T. Hence, from (5.12). we have
Since the series of constants 2M /kl.,, converges, it follows by Weierstrass M-test

that the series (5.11) converges uniformly in the region 0 s x s L, 0 s t :::;: T
for arbitrary T. This implies that v is a continuous function for 0 s x s L,
t ~ 0, and thus satisfies (4.3) and vanishes at t = 0 because each of the terms
of the series does. Under more stringent conditions on F [say, F, Fx, and Fx:x:
are continuous for 0 s x s L, t ~ 0, and F(O, t) = F(L, t) = OJ, we can
232 The Heat Equation Citap. 6
verify that the series (5.11) can be differentiated term by term. Then, by (5. 7)
and (5.4), it follows that
00
nnx
vt - kv.n = L [v;,(t) + knv(t)] sin
11=1 L
00
nnx
= L F,,(t)

SJl1.
11=1 L
= F(x, t)
so that v satisfies (5.1 ).
If we substitute (5.5) for F,, in (5.11) and formally interchange the order of
summation and integration, we can write (5.11) in the integral form
(5.13) v(x, t) = LJ: G(x, t - r; ~)F(~, r) di; dr

where G(x, t; !;) is the Green's function given in (4.16). In this form, it can be
verified that (5.13) gives a solution of the problem (5.1), (4.3) with homogeneous
initial conditions under the less restrictive conditions that F and Fx are con-
tinuous and F(O, t) = F(L, t) = 0. Indeed, from the discussion in Section 4,
we see that the inner integral in (5.13) is a solution of the problem
U1 - kuxx = 0 (0 < x < L, 0 < T < t)

(5.14) u(x, O; r) = F(x, r) (r > 0)
u(O, t; r) = u(L, t; r) = 0 (0 < T :S: t)
Our assertion then follows by the Duhamel's principle (Problem 10, Exercises
6.4). The uniqueness of our solution follows by Theorem 3.2.
u, - llxx = (sin x (0 < x < n:, ( > 0)

u(x, 0) = 0 (0 s x s n:)
u(O, r) = u(n:, t) = 0 (! 2: 0)
Solution: By (5.5) we find
Fn(t) = -2r
n:
in . .
0
Sll1 x 5111 nx dx = {ot (11 = 1)
(11 > I)
Hence, by (5.11 ), the solution is
u(x, !) = (J: e--t>r dr) sin x
= (e-' + t - 1) sin x
Sec. 5 Nonlto111oge11eo11s /11itiul-Bo11nd11ry Value Problems 233

11 1 - llxx = F(x, 1) (0 < x < rr, I > 0)
u(x. 0) = 0 (0 :S _\- :S rr)
11(0, I)= u(n, I)= 0 (I 2: OJ
where F(x, I) = x when 0 :S x :S rr/2, and F(x. I) = rr - x when rr/2 :S x :S rr_
Solution: The coefficients in the Fourier sine series expansion of Fare given by
F,,(t) = 2 Jn/2 x sm
_ nx dx + 2 J" (rr - x) sin nx dx
Tr 0 n rr/2
4 . /17[
2
sm (11 = J, 2, ... )
nn 2
or
4( - 1)11 - 1
F2n-1 (I) = - -- (n = I, 2, __ )
rr(211 - 1)2
(See Example 9.1, Chapter 3.) Hence, by (5.12),
V 2 n_ 1 (1) = -- -
4( - l)n-1
- -
J' exp[-(211 - I)2(r - T)] dT
rr(211 - 1)2 o
= 4(--_I)"_~_ [I _ e-<211-1)11]
n(2n - I )4
and therefore, by (5.11), the solution is

4 oo (-1)11-I
u(x, t) = -- E - ~- [I - e-< 2 n- 011 ] sin(211 - l)x
7r n=I (211 - 1) 4
Next we consider the initial-boundary value problem
(0 < x < L, t > 0)

(5.15) u(x, 0) = 0 (0 < x < L)
u(O, t) = a(t), ux(L, t) = b(t) (t > 0)
which involves nonhomogeneous boundary conditions. This problem can, of

course, be reduced to a problem of the kind considered above, in which the
differential equation is nonhomogeneous and the boundary conditions are
homogeneous. Jn fact, if we introduce the new dependent variable w, defined
by w = u - v, where v(x, t) = a(t) + xb(t), then the problem (5.15) becomes
w, - kw xx = - [a'(t) + xb'(t)]
(5.16) w(x, 0) = -a(O) - xb(O)
w(O, t) = 0, wx(L, t) = 0
which can be treated by the method described above.

It is possible, however, to apply the method directly to the problem (5.15)

without reducing it to (5.16). That is. we assume a solution of (5.15) in the
form of a series involving the eigenfunctions of the Sturm-Liouville problem
X" +IX= 0, X(O) = 0, X'(L) = 0
which arises if the method of separation of variables is applied to the problem

when the boundary conditions are homogeneous. This problem has the
eigenvalues
2
). = (211 - 1 rr)
" 2 L
with the corresponding eigenfunctions
X (x
11
. (211
. ) = s111 - --
2
1) nL..~ (11 = 0, 1, 2, ... )
Therefore, we assume a solution u in the form
(5.17) u(x, t) = Lro .

u 11 (t) s111 (2n - 1) rrx-
- -
n=I 2 L
with
(5.18) u,,(t) = 2- JL u(x, I) sin ('-) 11 - 1) rrx

- dx (11 = 1,2, ... )
L 0 2 . L
Assuming that u is twice continuously differentiable for 0 c:;; x c:;; L, t :;::: 0, we
obtain, by differentiating (5.18) with respect to t and using the differential
equation,
u;,(t) = - 2 JL u,(x, t) sin (211 ')- -1) rrx- dx

L 0 - . L
2k
= -
L
JL uxx(x, t) . (211
.
Sll1
-
2
-
1) nxL dx
-
0
Using the boundary conditions in (5.15), this can be integrated by parts twice
to give
(5.19) u;,(t) + kl.,,u 11 (t) = g,,(t)

2 2
where ). 11 = [(211 - l)rr] /4L and
2k -
g,,(t) = -z_- [,/;,,, a(t) + (-1)'- 1 b(t)] (11 = 1, 2, ... )
From the initial condition in (5.15) we see that u,,(O) = 0. Hence, by solving
the differential equation (5.19) with the initial condition u,,(O) = 0, we find
(5.20) 11,,(t) = { exp(-k).,,(t - r))g,,(r) dr (11 = 1, 2, ... )

Sec. 5 So11lro111oge11eous lnitial-Bormdary Value Problems 235
Substituting 1his in (5.17), we thus obtain
(5.21) u(x, t) = ~ ~ f'exp(-kX 11 (t - c))g,/1) (h1) sin( 211 2- 1

) nx
11= I l. () L
The verification that this gives a solution of the problem (5.15) is not so straight-
forward as in problems we encountered before. We shall illustrate the situation
in this case by means of the following example.
Example 5.3. Find and verify the solution of the problem
(0 < x < n, t > OJ

u(x, 0) = 0 (0 < x < rr)
11(0, 1) = 0, uJn. I) = I (t > 0)
11 1
So/111io11: Here we have g,,(t) = (2/rrl(-1 ) - so that by the formula (5.21) our
solution is
u(x, I)=~ 1 ~
1 (-1)"-
1
(L exp(--~ 211
-
2
~) (1= r))dr} sin(~i 1

) x
8 I:.,_ (- I)"- 1
I- e 211
< -
-
1
>'t1 4
- sm -
. (211~- - I) x
7f n= I (211 - l )2 . 2
To verify that this gives a solution of the problem, we observe that the function
f(x) = x has the Fourier series representation
x =
8 I: oo ( - 1--)"-I sm
. (211 - 1) x (0 < x < 7!)
7! 11=1 (211 - 1) 2 2
in terms of the eigenfunctions X 11 (x) = sin [(211 - 1l/2 ]x, n = I, 2, . . . . Hence,
we can write our solution above in the form
u(x, I) = x -
8
-
00
I: (-1)"-
-
1
? exp
(-(211
- - - 1)
2
1) sm. (211 - 1) x
7T 11= I (211 - 1)- 4 2
The series on the right converges uniformly in the region 0 :S x :S n, t ;::: 0. so

that 11 is a continuous function there. Further, for 0 < x < 11 and t > 0, the
function u has continuous derivatives of all orders with respect to x and t, which
can be obtained by termwise differentiation. Moreover, it is clear that 11(x, 0) = 0,
u(O, t) = 0, and u,(n, t) = 1. By a straightforward calculation it is readily
seen that u satisfies the differential equation for 0 s x :S n, t > 0. This com-
pletes the verification that 11 is a solution of the given problem.
Exercises 6.4
In Problems 1 through 5, reduce the problem to one involving homogeneous boundary

conditions and then find the solution.
l. u1 - kuxx = 0, 0 < x < L, t > O; u(x, 0) = 0, u(O, I)= u(L, t) =A, A a constant.
2. 11 1 - ku_u = 0, 0 < x < L, t > O; u(x, 0) = 0, ux(O, t) = A, u(L, t) = 0, A a
constant.
236 Tlze Heat Equation Clzap. 6
3. 11, -ku'"' = 0. 0 < x < L, t > O; u(x, 0) = 0, 11(0, t) = sin UJt, u(L, t) = 0, (J) a
constant.
4. u, - kuxx = 0, 0 < x < L, t > O; u(x, 0) = 0, 11(0, t) = 0, uJL, t) = cos UJt.
5. u, - uu + hu = 0, lz a positive constant, 0 < x < rr, t > 0; u(x, 0) = 0, u(O, t) =
0, 11(71, t) = J.
6. Solve the problem
u, - ku'"' = e-x (0 < x < n, t > 0)
u(x, 0) = f(x) (0 ~ x ~ n)
11(0, t) = 0, un, t) = 0

11 1 - kuxx = A cos UJt (0 < x < n, I > 0)
11(x, 0) = f(x) (0 ~ x ~ n)
(A a constant)
(Observe that any function oft satisfies the boundary conditions.)

u, - kuxx = g(x, 1) (0 < x < L, I > 0)

u(x, 0) = 0 (0 < x < L)
u(O, I) = 0, u(L, t) = h(r) (I > 0)
11 1 - kuu = g(x, I) (0 < x < L, f > 0)

uO, I) = h(I), ux(L, 1) = 0 (t > 0)
u(x, 0) = 0 (0 < x < L)
10. (Duhamel's Principle) Let 11(x, t, T) be the solution of problem (5.14) and set
1 (x, I) == s~ u(x. I - T, T) dT. Show that v agrees with (5.13) and that it satisfies
(5.1), (4.3), and the initial condition r(x, 0) = 0.
6. The Initial Value Problem
We now consider the problem of heat flow in an infinite rod without heat
source, whose lateral surface is thermally insulated. This problem leads to the
initial value problem
(6.1) u, - kuu = 0 ( - Cl) < x < Cl), t > 0)
(6.2) u(x. 0) = f (x) (-oo<x<oo)
for the heat equation. We shall solve this problem by means of the Fourier
transform presented in Chapter 5.
We suppose that the problem (6.1), (6.2) has a solution 11, which has the
property that 11, 11,, ux, and uu are all piecewise smooth in x and t, and absolutely
integrable in x for - oo < x < oo. Then, according to Theorem 8.1 of Chapter
5, the Fourier transforms of all these functions exist; in particular, we have
(6.3) U(s, t) = J~n J:00

u(x, t)e-i.sx dx
and
(6.4) u(x, t) = /
v 2n
- J00
-oc
U(s l)ei.sx ds
. '
Thus, if we can find the transform (6.3), then the solution of the problem will
be given by the formula (6.4).
To determine the function U, let us differentiate (6.3) with respect to t and
use the differential equation (6.1) to obtain
au J~
00 u,(x, t)e-isx dx
1
- (s, t) = - -,=
ot J2n
(6.5)
k
,/2;
J: 00
uxix, t)e-isx dx
Note that because 11, and 11.u are absolutely integrable on - oo < x < oo for
all s and t > 0, differentiation with respect to t under the integral sign is valid.
Integrating by parts twice the last integral above, we have
au k . . 1 oc
---;;- (s, t) = -r (uxe-sx + isue-sx):
ot --.J2n i-oo
- ---=
ks2 Joo u(x, . dx
t)e-sx
J2n -oo
If we further assume that u and u, vanish as lxl --> oo, then equation (6.5)
reduces to
au
-~ - (s, t) = -ks 2 U(s, t)
(6.6)
ct
where we have used (6.3).
We now require for this discussion that the function fin (6.2) be piecewise
smooth and absolutely integrable on ( - oo, oo) so that its Fourier transform
F(s) exists. Then, from (6.3) and (6.2), we have
U(s, 0) = ~l~n J: 00
u(x, O)e-isx dx
(6.7)
,-- J_"'oo J(x)e-isx
'\/ 2n
= F(s)
Thus. as a function of the variable t, the Fourier tramform of the solution u

satisfies the first-order differential equation (6.6) and the initial condition (6.7).
In this way we have therefore transformed the problem (6. I), ( 6.2) for the func-
tion 11 into the simpler problem (6.6), (6.7) for the Fourier transform U o[ u.
The solution or the differential equation (6.6), which satisfies the initial con-
dition (6.7). is given by
(6.8) U(s, t) = F(s)e-" 21
Consequently. by (6.4) and (6.7), we have
u(x, t) = 11.
, 2rr
I"'
"- oc
F(s) exp( - ks 2 t + isx) ds
(6.9)
Now the double integral corresponding to the iterated integral in (6.9) is

absolutely convergent because both f(x) and e-ks 21 are absolutely integrable.
Therefore, \Ve can interchange the order of integration in (6.9) and write
(6.10) u(x, 1) = J~ G(x - (. t)fm d(

00
where
(6.11) G(x, t) = 1 II" exp(isx - ks 2 t) ds

2rr _"'
The integral on the right of (6. I I) can be evaluated as follows: By using the
Euler formula
eix = cos x + i sin x
(6. I I) can be written as
(6.12) G(x. t) = II"' e-ks''(cos sx + i sin sx) ds

2rr .. - oc
2 2
where the functions e-ks ' cos x and e-ks ' sin sx are both absolutely integrable
on - oo < s < oo, uniformly in x and t for t > 0. Since e-ks'r cos sx is even
ins and e-ks 21 sin sx is odd. it follows that
oc
so that (6.12) becomes

J - 00
e-ksir sin sx ds = 0
G(x, t) I
2rr
J"'_"' e - ks'r cos sx ds
1 Joo e -ks't cos sx cs/
J[ 0
Jntroducing the new variable z = s, k1, we obtain
By Problem 10, Exercises 1.6, we find
,_;rr e-x'/4kt
so that
G(x, I) =
1
f"' e-k>'r cos sx ds
rr o
(6.13)
Substituting this in (6.10), we finally obtain
(6.14) u(x, t) = - 11'---- J'"' (-(x _ ()2) f(c;). de;"

exp ----
2vrrkt -x 4kt
for a solution of the initial value problem (6.1), (6.2).

The function
(6.15) G(x - (, t) = I
1
..
(-(x -
exp - - c)
2
)
(t > 0)
2,i rrkt 4/.:t
is called the fundamental solution or the Green's function for the heat equation
in the infinite domain - oo < x < oo, t > 0. Physically, the Green's function
(6.15) represents the temperature at a point x at time I due to a concentrated
heat source at a point at time zero.
It is easily verified from (6.15) that (i) G has continuous partial derivatives
of all orders with respect to x and t for - oo < x < oo, I > 0, and satisfies the
heat equation for all x and I > 0, provided x # ( (see Problem 2, Exercises
6.1); (ii) G is continuous at x = (; (iii) G vanishes as !xi -> oo for t > 0; (iv)
G has the property
f~0 G(x - (, t) d( = 1
and, finally, (v) G is symmetric with respect to x and (. These properties are
analogous to those of a Green's function for an ordinary differential equation.
Now, to verify that (6.14) actually gives a solution of the problem (6.1),
(6.2), it is enough to require that f be continuous and bounded on ( - oo, oo ).
Then, for arbitrary constants L > 0 and t 0 > 0, the integral (6.14) together
with its partial derivatives of any order obtained by differentiating under the
~: ..
integral sign with respect to x and t converges uniformly in x and t for - L s;

x s; L, t 0 s; t. Therefore, the function 11 and its derivatives of all orders exist
for - oo < x < oo, t > 0, and these derivatives can be found by differentiating
under the integral sign. Since C(x - (, t) satisfies equation (6.1) whenever
x # (, it follows that u does also.
Finally, set z = (~ - x)/(4kt) 112 in (6.14) to obtain
u(x, t) = - l_
Jn
J"'
-oo
f(x + J4kt z)e-= 2
dz
Since f is bounded for all values of its argument, the preceding integral con-
verges uniformly with respect to x and t for - oo < x < oo, t ;:::: 0. Hence,
Jim u(x, t) = - :
t->O"
1
'Ii1n
J 00
-oo
f(x)e-= dz
2
= f(x)
which verifies (6.2). From the boundedness off it follows that u is also bounded.
If we define u(x, 0) = f (x), then u is bounded and continuous for - oo <
x < oo, t ;:::: 0, so that by Theorem 3.4, u is uniquely determined.
Example 6.1. Find the solution of the problem (6.1 ), (6.2) if the function f is given by
/(x) = {~ (x < 0)
(x > 0)
Solution: According to the formula (6.14), the solution is
1
u(x, t) = 2vnkt
{"'
Jo
exp --
(-(x4ki-
- ~) )
2
d<;

Jn terms of the error function,

erf(r) = }
v; Jo
r e-zl dz
which has the property that erf( ro) = 1, this solution can be written as
u(x, t) = -1 + -1 erf ( -_--_

x )
2 2 v4kt
Then, by the Leibnitz rule, it is easily verified that this satisfies the heat equation
(6.1). Also, as t-> o+, u(x, t) _, 1 if x > 0, and u(x, t) _, 0 if x < 0. At
x = 0, we have u(O, t) = I /2, which is in accordance with the theory of Fourier
integral transform, since f has a jump discontinuity at that point.
In the case of the initial value problem

U1 - kuxx = h(x, t) ( - 00 < x < 00' t > 0)
(6.16)
u(x, 0) = 0 (-oo<x<oo)
which involves the nonhomogeneous heat equation, the method described above
leads to the differential equation
(6.17) U, + ks 2 U = H(s, t)
for the Fourier transform (6.3) of the solution 11. Here.His the Fourier trans-
form of the function h; that is.
(6.18) H(s, I) = -
I
If z h(x, t)e- 1" ' dx
.
, 2n - x
From (6.3) and the initial condition in (6.16). we see that U(s, 0) = 0. By
solving the differential equation (6.17) with the initial condition U(s, 0) = 0,
we thus find
(6.19) U(s, t) = J~ e-' 1' 11 -r

1
H(s, r) dr
Therefore, according to (6.4). the solution of the problem (6.16) is given by
(6.20) u(x, t) = Tl Jx e"-'. f' e-s'<t-r H(s, r) dr ds 1
v 2n - oc o
By substituting (6.18) for Hand formally interchanging the order of integrations,
this solution can be written as
(6.21) u(x, t)
J' Jx
0 - ""
G(x - (, t - r)h(~, r) d( dr
where, by (6.13),
G(x - ~, t - r) = I
2n
J"'_ oc
exp(is(x - () - ks 2 (t - r)) ds
exp(-(x - ) 2 /4k(t - r))

I
2, 1 nk(t - r)
is the Green's function for the problem.
To verify that ( 6.2 J) gives the solution of (6.16), we assume that his continuous
and bounded for all x and t ~ 0. Then, for 0 < r < t, we know from the
previous case that the function
w(x, t; r) = f"' G(x (,I - r)h(, r) di;

. - "'
is the solution of the initial value problem
H' 1 - /.;11.u = 0 ( - oo < x < oo, t > r)
1r(x, r: r) = h(x, r)
Hence, by Duhamel's principle (Problem 8, Exercises 6.5), it follows that (6.21)

is the solution of the problem (6.16).
242 The /feat Equation ('hap. 6
Exercises 6.5
I. \\'rite the solution or the problem
lixx 0 (-oc<x<x,l>OJ
(' ixl
11(x. OJ (-X < X < X)
2. So\\e the problem

0 ( - X < x < X. I > 0)
11(.r. 0)
{1 ix1 :s: L)
lo ( x > /_)
and express the solution in terms of the error function
erf x = -') f'X !'- '

2
ds
\ 7l .. 0
3. As in Problem 2, find the solution of the problem
11, - kuu 0 ( - 00 < x < CIJ' I > 0)
11(x, OJ
{A (x < 0)
\B (x > 0)
4. Show that if the function fin (6.2) is odd (or even), then the solution (6. 14) is
also odd (or even) with respect to the \ariable x. Thus, deduce that 11(0, I) = O
(or 11x(O. I) = 0).
5. \'erify the result in Problem 4 \1 ithout resorting to the formula (6. 14) and using
only the fact that a solution of the problem (6. 1), (6.2) is uniquely determined.
( - x < x < x. I > 0)
11(x. 0) 0 (-w<x<x)
and sho11 that (6.21) yields 11(1. 1) ~ t 2 '2.

( - oc < x < :::r:. I > 0)
11(x. 0) 0 (-x<x<oc)
and show that (6.21) ~ields 11(x. r) = 1 -- !'-'.
8. (Duhamel':, Principle) Let 11(.1', I, T) be the solution or the initial value problem
// 1 - kun 0 ( - CIJ < x < x, I > 0)
u(x. 0, r) f(x. r) (- x < x < x. r > 0)

and set
r(x, t) = f~ 11(x, I - r, r) dr
Pro\e that z is the solution of the problem
r, - kt'.n = f (x, t) (- x < x < x, t > 0)

l'(x. 0) = 0 (-oo < x < oo)
Sec. 7 lnitial-Bo1111dary Value Problems in Infinite Domain 243
(h = const; 1 > 0)
1(.\", OJ= /Ix) (-:x: < x < :X:)
and shO\\ that z(x. 1) = 11(.1. t)e- 111 , where 11 is given by (6.14).
1
IO. By the sub'.;titution 11(\. I) = r(x, I )e ", show that Problem 9 can be reduced
to the problem (6. I). (6.2).
11. By using the result of Problem 9 and the corresponding Duhamel's principle for
the differemial equation, obtain the solution of the initial value problem
11 1 - k11_u + '111 = /(x, t) ( - oc < X < 02, t > 0)

u(x. 0) = 0
( - oc < x < oc ; h = canst > 0)
7. Inilial-Boundary Value Problems in Infinite Domain
In this last section of the chapter we shall consider initial-boundary value

problems for the heat equation in the domain 0 < x < oo, t > 0. The solutions
of our problems will be obtained by means of Fourier sine or cosine transform,
as this becomes appropriate for this type of problem. We first consider the
problem
(7.1) u, - ku'-' = 0 (x > 0, t > 0)
(7.2) 11(x, 0) = f (x) (x ?: 0)
(7.3) 11(0, t) = 0 (t ?: 0)
which corresponds to the problem of heat flow in a semi-infinite rod x > 0,
with initial temperature distribution described by the function f(x) for x ?: 0,
and whose end point x = 0 is maintained at zero temperature. To solve this
problem. \\e shall use the Fourier sine transform. This choice is based on the
fact that the singular eigenvalue problem that is related to the problem at hand is
X" + l.X = 0, X(O) = 0 (X bounded as x--> oo)

which has the eigenfunctions XJx) = sin sx corresponding to the eigenvalues
2
; = s , s > 0. Accordingly, we assume that the problem (7.1), (7.2), (7.3)
has a solution u, which has the property that 11 and its partial derivatives up to
the second order are all piecewise smooth and absolutely integrable on 0 <
x < oo for t > 0. Then the Fourier sine transform of 11,
(7.4) U,(s, t) = (n,2) 1/2 f"'

o
.
u(x, t) sm sx dx
exists and
(7.5) u(x, t)
2')1,2
-
Joc UJs, t) sin sx ds
( 7[, 0
244 The Hear Equation Chap. 6
Here, Us does not mean the partial derivative of U with respect to s. Following
the procedure described in the preceding section, we proceed to find the trans-
form Us or our solution. Once this function is found, then our solution is given
by (7.5).
By differentiating (7.4) with respect to t under the integral sign and using the
differential equation (7.1 ), we have
au s (s,
ct
t) = (2):" 1 2
1 J"' u,(x, t) sin sx dx
0
112
k (.;) {" u,.Jx, t) sin sx dx
Arter integration by parts twice, this gives
cU
- -
ct
5
(S, t) k ; (2) 1/2
(llx sin sx su cos sx) '
io
(7.6)
ks 2 (n,2)
-
1/2 J"'
o
u(x, t) sin sx dx
ff we assume as before that u and ux vanish as x --+ oo, then because or the
boundary condition (7.3), the boundary terms on the right of (7.6) vanish,
and by (7.4) we are led to the differential equation
(7.7) cl/
- (s,5 .
t) + /.:.s 2 U,(s, t) = 0
.
ct
for the function Us. Let us assume that the function f in (7 .2) is piecewise
smooth and absolutely integrable on (0, oo ). Then, from the initial condition
(7.2) and (7.4), we find
U,(s, 0) = (;
2)1;2 oc
.t u(x, 0) sin sx dx
(7.8) 2)1/2 J"' j(x) sin sx dx

(-
n. o
= F,(s)
which is the Fourier sine transform of the function f. The solution of equation
(7.7) that satisfies the initial condition (7.8) is
(7.9) U,(s, t) = F 5 (s)e-k 521
Hence, by (7.5) and (7.8), we obtain
u(x, t) = (-2)1/2Jx F (s)e-ks-i, sin sx ds

Ji. 0
5
(7.10)
=
n o o
.3 c r"
f(~)e-ks't sins~ sin sx d~ ds
This gives the solution of the problem (7.1), (7.2), (7.3).

Sec. 7 lflitial-Bo1111dary Value Problems ;,, lfljiflite Domain 245
If we replace the function sins sin sx in the integrand of the integral in

(7.10) by
}[cos s(x - ~) - cos s(x + )]
and formally interchange the order of integration. we can write (7. l 0) in the
form
(7.11) u(x, n= 1
. --
2, nf\1
J, Ir
o
exp(. -(x -
.
4k1
)2) - exp(-(x + ) I /{) d
4k1
2
)
.
in which we have used the resu It (6.13 ). In this form, it can be verifled that u
is the solution of our problem under the weaker condition that f is continuous
and bounded for x ~ 0 with f(O) = 0.
We remark that the solution (7.1 I) of our problem (7.1), (7.2), (7.3) can also
be obtained by extending the initial datum fin (7.2) as an odd function for
- oo < x < oo and then solving the resulting initial value problem by the
formula (6.14). [See Problem 4, Exercises 6.5.J Using the fact thatJ(-x) =
-f(x), the solution as given by (6.14) then reduces to (7.11). Observe that this
is precisely the method we employed in Section 6 of Chapter 3 for finding the
solution of an initial-boundary value problem for the wave equation. The
method works because both the wave and the heat equations remain unchanged
in form when the variable x is replaced by - x.
In the same manner, we can find the solution of the problem (7.1 ), (7.2)
with the boundary condition
(7.12) u)O, t) = 0 (t ~ 0)
by extending the function fas an even function for - w < x < oo. In this
case, the solution is given by
(7.13) u(x, t) = _l_ f"' lexp(_-_(x - 02) + exp(:-(x + )2) l f(() d

2,' rrkt 0 l 4kt . 4kt J
We note here that to solve the problem (7. l ), (7 .2), (7 .12) by the method of
Fourier transform, it will be necessary to employ the Fourier cosine transform
of the solution u. This is also the case even in the more difficult problem where
the boundary condition (7. l 2) is nonhornogeneous. We illustrate the pro-
cedure in connection with the problem
u, - ku'-' = 0 (x > 0, I > 0)
(7.14) u(x, 0) = 0 (x ~ 0)
uAO, t) = h(t) (1 ~ 0)
As usual, we suppose this problem has a solution that is piecewise smooth

and absolutely integrable on (0, oo) including its first- and second-order
derivatives. We consider the Fourier cosine transform of u,
uces, l) '')) J/2 J'f) u(x, t) cos sx dx

(7.15) = ( ~
/., 0
..,...,_
246 The Heat Eq11atio11 Chap. 6
Differentiating this with respect to t and proceeding as in previous cases. we

have
au,
- - (s. I )
ct (-re2)1/2 f"' u,(x, 1) cos sx d.x
0
k (.::
7)1;2 I.(" t) COS dx
Uu(X, SX
.n. . o
2) l !2 Ioc
k - ('IT.
(u, cos sx + su sin sx)'
'0
112
- ks
2
(~) {''' u(x, t) cos sx dx
Under the assumption that u and llx vanish as x --+ oo, and taking note of the
boundary condition in (7.14), this leads to the nonhomogeneous differential
equation
(7.16)
cu
~- c (s, t) + ks 2 uces, t) = -k (2)1/2
- h(t)
ct n
The solution of this equation that, by the initial condition in (7.14), satisfies
the condition U,(s, 0) = 0 is given by
(7 .17) UJs, 1) = -k (D 112 J~ e-ks"(t-T)h(T)<h

Hence, by taking the inverse transform, we obtain
u(x, t) =
2) 1/2 j'"' uces, 1) cos sx ds
(-
'IT 0
(7.18)
- 21
__'.: f"" f' e-ks'u-' 1h(T) cos sx dT ds
'IT 0 0
1f we formally interchange the order of integration and use the result in (6.13),
this formula simplifies to the form
(") l/l j'I

2
exp(-x /41,\(t - T))ii(T) dT
(7.19) u(x. I) = - - - , -
'IT. 0 '\l-T
This can be shown to provide the solution or the problem (7.14) under the
assumption that h is differentiable for t :2:: 0.
We remark that we cannot verify that (7. J 9) satisfies the boundary condition
of (7.14) by simply differentiating (7.19) under the integral sign and setting
x = 0. In fact, such a formal manipulation would lead to the false result
u)O, t) = 0. We illustrate the situation in this case by means of the following
example.
Sec. 7 /11itial-Bo11ndary Value Problems in Infinite Do111ai11 247
Example 7.l. Find the solt1t1on or Liie problem
11, II_\-.\ 0 (_\ > 0, t > 0)

11L1, OJ 0 (x .2: 0)
11)0. I) ( t .2: 01
Solution: B) the formuL1(7.19)11e have
11(.1, I)~
,., C';fl( -- X 2 /4(t - T) I lfT
-
1 T: I
,O \f-r
On setting z x. \ 411 - rl. this becomes

_\" ,,,'l. e~z2
II(\". I) dz
\ 7[ J,. '41
_2
for 11 hich it is readily see11 that
11).\", /) = -
\ 7I
\ nl
4r
11,(x, I)
\ 7'.'{
for x > 0 and 1 > 0. Thus. 11, - u'-' 0. sho11 ing that 11 satisfies the heat
equation for x 0. I > 0. As I > 0 it is clear that u(x. t) ' 0. Now, to
1erify the boundar1 condition 11,!0. 11 = I. 11c perform an integration by parts
to put 11 in the form
I'"/
,r; ( --- e- :')

x I 2x
1,4,
~:2
11(x. II -- ~ e dz
\ X, '\ 4f \ n
e
.. 41
'
2x Jy e :' d::
\ ,, X,'- 41
Then
2
u)x. I)
\ -
which immediately yields 11)0. r I l, since Jg e-=' d::
Exercises 6.6
I. (a) Obtain the solution formula (7. 11) from (6.14) by extending the function fin
(7.2) as an odd function for - T < x < T ..
(bl Obtain the solution formula (7.13) of the problem (7.1), (7.2). (7.12) by
extending the function fas an el'en function for - 'XJ < x < C1J.
248 The Heat Lq11ation Chap. 6
2. Obtain the solution !7.13) by using the Fourier cosine transform.

0 (x > 0, t > 0)
11(\", 01 :re-x (x 2: OJ
11(0, /) 0 (t 2: 0)
11, - kuu: = 0 (x > 0. I > 0)

(a > 0, x 2: 0)
(! 2: 0)
11 1 - kuu: = 0 (x > 0, I > 0)

11(x, 0) 0 (x > 0)
11(0, I) = A (A a cons tan!; t > 0)
and express !he solution in terms of the error function. Verify that it gives the
solution of the problem.
11, ku'"' = 0 (x > 0, I > 0)

11(.1. OJ= 0 (x 2: 0)
11(0, I) = h(t) (t 2: 0)
11, - kuu = f(x, I) (x > 0, t > 0)

11(x. OJ 0 (x 2: 0)
11)0, t) lz(t) (I 2: 0)
8. Find the solulion of the problem
11 1 - ku'"' + /111 = 0 (h = const > 0)

11(.\, 0) = f(x) (x 2: 0)
11)0. /) = 0 ( f 2: 0)
9. Obtain the solution of Problem 8 from the resull of Problem 9, Exercises 6.5,
by extending the function fas an even function for - oo < x < oo.
11, - kuu + bu 0
(b const > O; x > O; t > 0)
11(x, 0) 0 (x 2: 0)
11(0, f) h(t) (t 2: 0)
Sec. 7 Initial-Boundary Value Problems in Infinite Domain 249
1 l. Find the solution of the problem
u, - kuu + bu = f(x, t) (h = canst > 0)

u(x, 0) = 0 (x ~ OJ
ux(O, t) = h(t) (t ~ 0)
12. By using the Fourier cosine transform, obtain the solution of the problem
l/ 1 - kuu + fll = 0 (x > 0, t > 0)

u(x, 0) = e-x (x ~ 0)
11x(O, t) = 0 (t ~ 0)
11 1 - kuu = 0 (x > 0, t > 0)
11(x, 0) = f (x) (x ~ 0)
u/O, t) - hu(O, t) = 0 (h = const > O; t ~ 0)
Hint: Let v(x, I)= u/x, I) - h11(x, t). See Problem JO, Exercises 6.1.
11 1 - k11xx = 0 (x > 0, t > 0)
u(x, 0) = 0 (x ~ 0)
ux(O, t) - hu(O, t) = g(t) t ~ OJ
111 - kuxx + bu = 0 (x > 0, t > 0)
11(x, 0) = f(x) (x ~ 0)
llx(O, f) - h11(0, t) = 0 (t ~ 0)
2
11 1 - llxx - - llx = 0 (0 < a < x, t > 0)
x
u(x, 0) = 0 (x ~ a)
11(a, t) = A (A constant; t ~ 0)
Hint: Set x11 = c, :: = x - a and use the result in Problem 5.
".
Chapter 7
Laplace's Equation
In the previous chapters we studied initial value and initial-boundary value

problems for the wave and the heat equations. We saw that these problems
are well posed for the wave and heat equations in the sense that the problems
have unique solutions that depend continuously on the initial or boundary
data. Jn this last chapter we shall be concerned with certain problems that are
well posed for Laplace's equation in two independent \"ariables
This equation serves as the prototype for elliptic differential equations in much
the same way as the wave and the heat equations do for hyperbolic and parabolic
differential equations, respectively.
1. Boundary Value Problems
In order to sec how Laplace's equation may arise in a physical problem, let
us consider the temperature distribution u in a uniform thin plate occupying
a domain D in the xy-plane (Fig. 7.1 ). We assume that D has a continuous
boundary C. By making assumptions similar to those made in deriving the one-
dimensional heat flow equation, it can be shown that the temperature distribu-
tion u on the plate satisfies the homogeneous equation
(1.1) l_u _ k (c"u O~u) = 0

ct - '
ex" + oy-.
- '
250
Sec. I Boundary Value Problems 251
i
--ur ----- -------- \
FIG. 7-1 Plane domain and bo1111dary curre.
in D. This equation is called the two-dimensional heat equation. Suppose now

that the heat flow in D is steady: then the temperature distribution u is inde-
pendent of time (that is, i":u/ot = 0) and hence equation (1.1) becomes
(1.2)
which is the two-dimensional Laplace's equation.

If there is a heat source in D, the temperature function u will satisfy the non-
homogeneous heat equation
(1.3) cu - k (c -- u + o u)
2 2
= f(x, \' t)
ot .iJx 2 ox 2 . ,
Consequently, vvhen the heat flow is steady so that f does not depend on I,
(1.3) reduces to the nonhomogeneous Laplace\ equation
.f (x' y)
(1.4) q(x, y)
k
Equation (1.4) is known as Poisson's equation. Both equations, ( 1.2) and (1.4),
are prominent examples of elliptic differential equations that are of great
importance in mathematical physics. We shall often write these equations in
the compact form '1.u = 0 and '1.u = q, where i1. denotes the Laplace's operator
From the preceding discussion, it appears that in a steady state, the temper-
ature distribution in a uniform heat-conducting body satisfies Laplace's equation.
Correspondingly, when there is a known heat source, the temperature function
satisfies Poisson's equation. From this physical consideration it seems natural
"l!J; ..
252 Laplace's Equation Chap, 7
to consider boundary value problems for the Laplace's operator rather than
initial value or initial-boundary value problems. As a matter of fact, boundary
value problems are typical for elliptic differential equations because initial
value or initial-boundary value problems are in general not well posed for this
type of equation. In this connection, let us examine an example devised by
Hadamard for the Laplace's equation. Consider the following initial value
problem:
Un + UYY = 0 (- 00 < X < 00, )' > 0)
(1.5) u(x, 0) = 0 (-oo < x < oo)
11/x, 0) = e - ,';, sin nx ( - oo < x < oo)
By the method of separation of variables, we readily obtain the explicit solution
(1.6) u(x, y) = e_-," sin nx sinh n y

11
of this problem. Now we observe that as the parameter /1 becomes infinite,

the initial data of the problem tend to zero uniformly, whereas for x =I= 0, the
solution ( 1.6) becomes arbitrarily large. This shows that the solution of the
problem does not depend continuously on the initial data, and hence the problem
is not well posed.
A typical boundary value problem for a given elliptic differential equation
consists in finding a solution of the differential equation in a given domain D
that satisfies certain conditions on the boundary C of D. In this chapter we
shall be concerned with boundary value problems for the Laplace's equation
with boundary conditions that are typical for elliptic differential operators.
Our discussion will be confined to the two-dimensional xy-plane, although
many of the results that will be obtained and the procedures used can be im-
mediately extended to the case of three-space dimensions.
The first type of boundary value problem for the Laplace's equation consists
in finding a solution of the equation in a given domain D, which assumes
prescribed values on the boundary C of D. That is,
(1.7)
1!11 = 0 in D
ll =f on c
where f is a known function on C. This type of problem is called a Dirichlet
problem, and the boundary condition is often called Dirichlet condition or
boundary condition of the first kind. Physically, the problem can be interpreted
as finding the equilibrium temperature distribution in D when a fixed temperature
distribution is given on the boundary of D.
The second type of problem for equation (1.2) consists in finding a solution
of the equation in the domain D satisfying the boundary condition
OU
(1.8)
Oil
=f
Sec. 2 Green's Theorem and Uniqueness of Solutions 253
on the boundary C of D. Here <Ju/c~n denotes the outward normal derivative

of u on C. This problem is called the Neumann problem, and the condition
(I .8) is called Neumann condition or boundary condition of the second kind.
Such a problem arises, for example, when we determine the equilibrium tem-
perature in a uniform body D for which a given amount of heat is supplied to the
boundary C of D.
The third type of boundary value problem involves the condition
(1.9)
au
_ + hu = g
l'/1
on the boundary C where h and g are known functions. Such a problem is

called the Robin problem or a mixed boundary value problem. The boundary
condition (1.9) arises when, for instance, we allow for radiation of heat from the
boundary of the body into the surrounding medium.
It is of interest to observe that the boundary conditions (1.7), (1.8), and (1.9)
correspond to the three types of boundary conditions we encountered in con-
nection with our discussion of the vibrating string and the conduction of heat
in a thin rod.
Analogous problems can, of course. be considered for Poisson's equation
(1.4). Oftentimes, however, it may be possible to reduce a problem involving
Poisson's equation to one that involves Laplace's equation. The procedure
consists in first looking for a particular solution v of Poisson's equation /1.11 = q
and setting w = 11 - 1. Then w satisfies Laplace's equation and a correspond-
ing boundary condition that will involve r. The function z; must, of course,
be such that it is defined and differentiable on the boundary of the domain under
consideration. In the general case, the problem is be.st approached by using
certain kernel functions, which are discussed in Section 9 of this chapter.
Each of the various problems mentioned above is called an interior-boundary
value problem inasmuch as the solution is required to satisfy the differential
equation inside the domain D. If, instead, we seek a solution of the differential
equation outside the domain D with boundary condition on C, then the problem
is called an exterior-boundary value problem. In this work we shall be con-
cerned primarily with interior problems, although some exterior problems will
be presented in the exercises.
2. Green's Theorem and Uniqueness of Solutions

In this section we shall seftle the question of uniqueness of the solutions of
the Dirichlet problem, the Neumann problem, and the Robin problem for the
Laplace equation, with the help of Green's identity. Let us consider first the
Dirichlet problem (I. 7). Suppose that 11 1 and u 2 are two solutions of this problem
and set w = 11 1 - 11 2 . It is clear that
(2.1) in D
and
(2.2) w=.f-f=O on C
254 Laplace's Equation Chap. 7
Let us consider G1ee11's Jir:,t identity O\er the domain D:
(2.3) r
c
II ~I' ds
(II
=

rr
D
(11 !11 + 11,rx + uyi-.J dx dy
Setting ii - r in (2.:\). we ha\e
(2.4)
.c
r 11 (~1/n ds .ror(11 '111
<
= + 11: + 11~) dx dr
Jj'\\e substitute the function 11 = 11 1 11 2 for 11 in (2.4) and use (2.J) and (2.2).
\\e then obtain
(2.5) JJ
D
(w: + \\~.) dx dy 0
Since the integrand in (2.5) is continuous and nonnegative, we conclude that

11~ + ir~ = 0 identically in D. This implies that 11 must be constant in D + C.
But 11 = 0 on C. Hence. it follows that \\. = 0 identically ir; D + C: that is,
11 1 = 11 2 . Thus. \\e have pro\ed our llrst theorem in this section.
THFORL\T 2.1 A solution of the Dirichlet problem (1.7) is 1111iq11e!r determined.
Jn the case or the Neumann problem
(2.6) "'111 = 0 in D. (l/ =f on C

C/1
if' 11 1 and 11 2 are t\\O solutions. then their difference ll' = 11 1 - 11 2 will be a
solution of the homogeneous problem
(2.7) in D. 0 on C
Applying the identity (2.4) to the function l l and using (2.7), we are again
led to the conclu;,ion that 1r must be constant in D + C: that is. ll'(x. t) = K.
ln thi, case. ho\\ever. it Lannot be e:;tablishecl that K = 0. since w does not
ncce-;sarily \anish on C. Therefore. \\e can have only
which says that any t\\O solutions or the problem (2.6) differ by a constant.
We state this result as our second theorem.
THEORE\I 2.2 A solution of the Xeumann problem (2.6) is unique up to an

additire constant.
Finaily, for the Robin problem
(2.8) '111 = 0 in D. Cl/ + /111 g on C

en
Sec. 2 Green's Theorem and U11iq11e11ess of' Solutions 255
uniqueness of the solution can be established if we assume that h > O and It

is not identically zero on C. Indeed. suppose u 1 and u 2 are two solutions of the
problem (2.8). Then \\ = 11 1 - u 2 satisfies the condition
Cl\'
+ /11\' = 0
211
or
(2.9) C\\' = -hw on C

011
Hence, by the identity (2.4), we have
(2.10) -L hll' ds =
2
JJ(w; + II';.) dx dy
D
Since h ~ 0 on C, the term on the left of (2.10) is not positive and the term on rhe
right is nonnegative. Therefore, both terms must be equal to zero, and so
JJ(\\'; + 11~) dx dy = 0
D
"hich implies that l\' = const in D + C. From (2.9) it follows that w 0 in

D + C; that is, 11 1 = u 2 . This proves our next theorem.
THEOREM 2.3 A solution of the mixf'd boundary ralz1C' problem (2.8) lrith
h ~ 0 and h not idf'nlical!y zero is uniquely dC'termined.
It is important to note that Green's identity (2. 3) also enables us to establish

a necessary condition for the existence of a solution of the Neumann problem
(2.6). If we interchange the role of u and 1 in (2.3), and set u = 1, we obtain
II j t.i1
(2.11)
I c
cu
~
l/1
ds =

D
dx dy
Thus, if the problem (2.6) has a solution u, then application of (2.11) yields
J'
(2.12)
f c en
?u
~ ds =
c
f ds = 0
This shows that if the problem (2.6) is to have a solution, then the integral off
over the boundary C must vanish. Jn terms of our physical interpretation of the
problem, condition (2.12) simply says that in a steady-state heat flow, the net
supply of heat at the boundary of D must be zero.
Jn the case where problem (2.6) involves Poisson's equation, !J.u = q in D,
application of (2.11) gives the necessary condition
(2.13) JJq(x, y) dx dy
D
= L f ds
for the existence of a solution of ~uch a problem. Consequently, if the condition

(2.12) or (2.13) is not satisfied, then the Neumann problem for Laplace's equation
or Poisson's equation cannot have a solution.
Exercises 7 .1
1. Reduce the Dirichlet, the Neumann, and the Robin problems for Poisson's
equation to a problem involving Laplace's equation and indicate the correspond-
ing boundary conditions.
2. Show that if 11 is a solution of the Robin problem
fi.u = q(x, y) in D
rn + hu = f (x, y) on C
011
then
L hu ds = L f ds - ff
/)
q dx dy
3. Prove that the Dirichlet problem
/!;u + k11 = q(x, y) in D

II = f(x, y) on C
has a unique solution ifk-~ .i.

4. Prove that the Neumann problem
/!;u + ku = q(x, y) in D
(11
= f(x, y) on C
(;17
has a unique solution if k < 0.

5. Prove that the Robin problem
611 + /, ll = q(x. y) in D
(II
+ hu = f(x, y) on C
CII
has a unique solution if k < 0 and h > 0.

6. Show that if u is a solution of the problem
L Li =
_ c ( .\. 2 cu)
ex ex,
+ .c
cy
(e' ~u)
oy
0 in D
Li = 0 on C
then u = 0 throughout D. Hint: Consider 11L11 = (C/cx)(x 2 1111") + (a/oy)

2
(e'"1111,.) - x u; - e,.u~ and use Green's theorem.
Sec. 3 Jfaxi11111111 Principle for Harmonic F1111ctio11s 257
7. Sho\\ that the problem
[(!
ex
+ 1
x )11,] +
(y
[(I + 2
y )uy] + ku = q(x, .J)
in D, !, :2: 0. and 11 = f(x. y) on Chas at most one solution.

8. Show that the problem
C.,,u = q(.r, y) in D
II = f(x, y) on C 1
C II
+ h11 = 0 on C 2
(;II
\\here C 1 is a part of the boundary of D and C 2 the remainder. and where h is a

positi1e constant. has at most one solution.
9. Formulate and establish the theorems of Section 2 in three dimensions. Note
that Green's first identity in three dimensions is of the form
III
v
(11 C.,,i- + llxLx + 11,r,. + ll:l':) dV = If
s
II::; ds
where S is a smooth surface bounding the domain V, and ti is the Laplacian in

three-space dimensions.
3. Maximum Principle for Harmonic Functions
A function u is called a harmonic function in a domai1i D if u has continuous

second-order derivatives and satisfies Laplace's equation at each point in D.
In this section we shall prove a maximum principle for harmonic functions,
which is much the same as the maximum principle we proved for functions
satisfying the heat equation.
THEOREJ\13.1 (Maximum Principle) Lf'l ll be a harmonic function in a bounded

domain D with boundary C. If u is continuous in D + C, then the maximum
ralue of u oier D + C is attained on thf' boundary C of D.
Proof: We prove first that any function u that is continuous in D + C and

satisfies the differential inequality
(3.1) in D
must attain its maximum value on the boundary C. Let u be a continuous
function in D + C that satisfies (3.1) and suppose that u assumed its maximum
value at an interior point (x 0 , y 0 ) of D. Then, at such a point, we would have
so that
258 Laplace's l:q11atio11 Chap. 7
This contradicts the fact that 11 satisfies (3.1) in D. Therefore, a function con-
tinuous in D + C and satisfying (3.1) attains its maximum value on C.
Now suppose that 11 is a harmonic function in D and continuous in D + C.
Let .\f denote its maximum value on C and consider the function
2 2
1(x. y) = u(x, r) + 1:(x + y )
where 1: > 0 is an arbitrar) number. lt is clear that r i~ continuous in D + C

and
~r = ~11 + 4i; = 41; > 0 in D
Hence, by the previous result, the maximum value of vis attained on C. Thus,
at any point in D, we have
r(x, y) :::;: max i.:(x, y) :::;: M + i;R

on C
where R = max(x 2 + y 2) for all (x, y) in D + C. Since u(x, J') c(x, y) -

s(x 2 + y 2 ) :::;: 1(x, y), it follows that
u(x, y) :::;: c(x, y) :::;: M + cR
for all (x, y) in D + C. and for any number > 0. Letting D --+ 0, \Ve then find
u(x, F) :::;: A1, which establishes the theorem.
By applying the maximum principle to the harmonic function i: = - 11, we

obtain an analogous result for the minimum value of u. Thus, a harmonic
function in a bounded domain that is continuous throughout the closure of the
domain attains its maximum and minimum values on the boundary of that
domain. Of course, if u is a constant. its maximum (which coincides with its
minimum) is attained at every point in D. As a matter of fact, a constant is the
only harmonic function that can attain a maximum at an interior point of a
domain. bounded or unbounded. This is the strong version of the maximum
principle. which we shall establish in Section 6.
By using the maximum principle for harmonic functions, it is possible to give
an alternative proof of Theorem 2. l for the uniqueness of solution of Dirichlet
problem for Laplace's equation. In fact, if 11 1 and 11 2 are two solutions of the
problem (l. 7), then their difference 11 = u 1 - 11 2 is a harmonic function that
\anishes on the boundary C. Then, by the maximum and minimum principle,
11 cannot be greater than or less than zero in D. Hence, 11 must vanish identically
in D + C; that is, 11 1 = u2
The maximum principle also enables us to establish continuous dependence
of the solution of Dirichlet problem for Laplace's equation on the boundary
data. We state this result as a theorem.
THEORE\l 3.2 ff 11 and L' are f\t'o harmonic functions in a bounded domain D
with boundary C such that 111 - rl :::;: e on C, then
lu - rl :::;: in D
The proof is simple and is left as an exercise.

Sec. 3 .Haximum Principle for Harmonic Functions 259
In sections to follow, solutions of the problem (1.7) in rectangular and circular

domains will be determined. Thus, for these types of domains, Theorems 3.1
and 3.2 imply that the Dirichlet problem is well posed for the Laplace's equation.
Exercises 7.2
1. Express the Laplace's equation 6.11 = 0 in polar coordinates rand 0, where
x = r cos. 0, y = r sin()
2. Show that if u is a harmonic function in D, then
v(x, y) = II
.x
(r2 ' r
y) 2
is also harmonic at all points in D, where r tc 0.

3. Show that by introducing the new variables .; = ln(a/r ), 11 = 0, Laplace's
equation in polar coordinates is transformed into the equation u~~ + 11,,ry = 0.
Thus, determine the problem into which the Dirichlet problem for a sector
/11; = 0 (r < a, 0 < 0 < h)

il(r, 0) = f(r) (0 < r < a)
u(a, ()) = 0
u(r. b) = 0
is transformed by the change of variables.

4. Under the change of variables .; = In( I /r ), 11 0, show that the Dirichlet
problem for an infinite sector
!1il = 0 (r > 0, 0 < 0 < b)
u(r, 0) = 0 (r > 0)
u(r. b) = f(r) (r > 0)
(u is bounded for r 2: OJ is transformed into the problem

e~s + v,111 == 0 (-w <.; < oc,O < 11 < h)
r(,;, 0) = 0 (- a:; < c; < a-:)
r(,;, h) = /(e- 0 ) (-:x: <.; < 0'.l)
5. Let u be a harmonic function in D and continuous on C. Show that if u > 0

on C, then 11 > 0 in D.
6. Let 11, L and w be three harmonic functions in D and continuous on C. Show
that if v s 11 on C. then 11 :::; v :S w in D.
11 :::;
7. Let ii and i be continuous in D and on C such that 11 is harmonic and 11v 2: 0 in D.

Show that if u = i: on C, then ii 2: r in D.
8. Show that a solution of the Poisson's equation /111 = q in a domain D cannot
attain its maximum at any point in D if q > 0, nor its minimum if q < 0.
9. Let u be a solution of
!lu - hu = 0 in D
where h > 0. If 11 is continuous in D + C and 11 :S on C, show that 11 s 0 in

D. Hint: Show that u cannot attain a positive maximum in D.
10. Prove Theorem 3.2 of this section.
11. Show that if 11 is a solution of the three-dimensional Laplace's equation
is a domain D bounded by a surface S, and 11 is continuous in D + S, then 11

satisfies a maximum principle.
4. Dirichlet Problem in a Rectangle
Beginning with this section we shall consider solutions or the Dirichlet and
Neumann problems for the Laplace's equation in simple domains such as a
rectangle and a disk. It will be seen that for this type of domain, the method of
separation of variables can be used to obtain a solution of the Dirichlet or the
Neumann problem. We first consider in this section the Dirichlet problem in a
rectangular domain R (Fig. 7.2): 0 < x < a, 0 < y < b. We seek a solution
of the Laplace's equation
(4.1) u'-' + uYY = 0 in R
which satisfies the boundary conditions
u(O, y) = 0, u(a, y) = 0 (0 ::;; y ::;; b)
(4.2)
u(x, 0) = 0, u(x, b) = f(x) (0 ::;; x ::;; a)
This problem may be interpreted physically as finding the equilibrium temper-

ature throughout a thin rectangular body R when its sides x = 0, x = a,
y = 0 are kept at zero temperature and the sider = b is maintained at given
temperature distribution j(x), 0 ::;; x ::;; a.
II = (I.\ I
Io. hi
"
R II= 0
II= 0
--~------------...__ ____ x
0
0 II=
"
FIG. 7-2 The Dirichlet prohlem in a rectangle.
Sec. 4 Dirichlet Problem in a Rectangle 261
In accordance with the method of separation of variables, we assume non-

trivial solutions of (4.1) of the form
(4.3) u(x, y) = X(x) Y(y)
which satisfy the three homogeneous boundary conditions in (4.2). Substituting

(4.3) in (4.1) and separating variables, we obtain
X" Y" .
= - - = -).
x y
This leads to the two ordinary differential equations
(4.4) X" + ).X = 0, Y" - l.Y = 0

which are to be satisfied by the functions X and Y, respectively. In order for
(4.3) to satisfy the three homogeneous boundary conditions for (4.2), we must
have
(4.5) X(O) = 0, X(a) = 0, Y(O) = 0
Thus, the function X and Y must be determined from the now familiar eigen-
value problem
(4.6) X" + l.X = 0, X(O) = 0, X(a) = 0

and the initial value problem
(4.7) Y" - l.Y = 0, Y(O) = 0
When we encountered problem (4.6) before, we found the eigenvalues )."
2
11 n 2 /a 2 with the corresponding eigenfunctions
. nn
x,, ( x ) = 5111 - x (11 1, 2, ... )
a
For each /. 11 n 2 n 2 /a 2 , a solution of (4.7) is given by
. h 117[
5111 y (11 1, 2,. .. )
a
It folio\\ s that the functions
11
U 11 ( X, }' ) = Sln ~~ sinh 11_7:f}' (11 = 1, 2,. .. )
a a
are all particular solutions of the Laplace's equation vanishing at x = O,

x = a, and y = 0. These functions are called rectangular harmonics.
To obtain a solution of the problem (4.1), (4.2) we now form the series
00
nrrx . nny
(4.8) u(x, y) = L c,,

s111 - s111h -
,,~ 1 a a
262 Laplace's Equation Clrap. 7
where the coefficients c,. are constants yet to be determined. Formally setting
y= band using the boundary condition in (4.2), we obtain
J(x) = u(x, b)
(4.9)
= f (en sinh '!_'T[b)
11= 1 a ,
sin ~_7[X
a
ro
11
= L: b sin ~~
II
n= I a
where
, mrb
(4.10) b,. = c11 sinh - -- (11 = 1, 2, ... )
a
The last expression in (4.9) is the Fourier sine series expansion off over (0, a),
and therefore
(4.11) b11 = 2--

a
s .0
f (x) s111 nnx
-- dx
a
(n 1, 2, .. ')
Substituting c,. from (4.10) in (4,8), we then obtain
~ b sinh(nrrJ/a) . nnx
(4.12) U(X y.) = 1.... - - Sll1
' 11= 1 "sinh(nnb/a) a
for the solution of the problem (4.1 ), (4.2). The coefficients b" are given by
(4.11).
If we assume that f is continuous, piecewise smooth on 0 :::;; x :::;; a, and
vanishes at x = 0 and x = a, we can verify that (4.12) indeed gives the solution
of our problem (4.1 ), (4.2) by the same method used in Section 4 of Chapter 6.
Notice that the inequality
sinh(mry/a) e""Yi - e-,,ny/a
sinh(nnb/a) e""b a - e-irnb a
- 1m(b - y),'a J - e - 2ny 'a

e ---,---:---
1 - e-.11rrb;a
e-1m(b-y)'a
:::;; -----
1 - e-2rrb:a
may be used to prove uniform convergence of the series (4.12) and the second-
order partial derivatives of that series with respect to x and y for 0 :::;; x :::;; a,
0 :::;; y :::;; J'o with any Jo < b.
The general Dirichlet problem
llxx + = 0
llyy in R
( 4.13) u(x, 0) = f 1(x), u(x, b) = f 2 (x) (0 < x < a)
u(O, y) = f3(y), u(a, y) = f4(y) (0 < y < b)
Sec. 4 Dirichlet Problem i11 a Rectangle 263
in the rectangular domain R can be solved by superposition of the four func-

tions u 1 , u 2 , u 3 , and u4 , where each u; (I ::::; i ::::; 4) denotes the solution of (4.13)
when all boundary data, except f;, are zero.
Example 4.1. Find the solution of the problem (4.1), (4.2) when R: 0 < x < n,
0 < y < n, and /(x) = sin 3 x.
Solution: We recall from trigonometry that
sin 3 x = :i sin x - { sin 3x
so that from (4.9) we see that b 1 = 3/4, b 3 = - 1/4, and b" = 0 for all other
values of 11. Hence, by (4. 12), the solution is
3 sinh y . I sinh 3y .
u(x y) = - sm x - -- - sm 3x
' 4 sinh n 4 sinh 3n:
The general Neumann problem for the Laplace's equation in the rectangle R,
Uxx + u\')" = 0 in R
(4.14) uy(x, 0) f(x), u/x, b) = h(x) (0 ::::; x ::::; a)
ux(O, y) = k(y), ux(a, y) = g(y) (0 ::::; y ~ b)
can also be solved in the same manner, provided the boundary data satisfy
the necessary condition (2.12), which means in the present case that
(4.15) J: [f(x) - h(x)] dx + I: [g(y) - k(y)] dy = 0
We illustrate the situation in this case by means of the following example.
Example 4.2. Find a solution of the Neumann problem
flu = 0 (0 < x < n:, 0 < y < n:)
u,.(x, 0) = x - (0 :S x :S re)
2
u,.(x, n:) = 0 (0 :S x :S re)
llx(O, y) = u,(n:, y) = 0 (0 :S y :S n:)
Solution: Here we note that condition (4.15) is fulfilled by the boundary datum
/(x) = x - re/2. The method of separation of variables u(x, y) = X(x) Y(y)
leads to the Sturm-Loiuville problem
X" +IX= 0, X'(O) = 0, X'(n:) = 0
for the function X. This has the eigenvalues A." = n 2 and the corresponding
eigenfunction X"(x) = cos nx, 11 = 0, 1, 2, . . . . For each eigenvalue )," = 11 2 ,
the initial value problem for the function Y,
2
Y" - 11 Y = 0, Y'(n:) = 0
gives Y"(y) = cosh n(n - y). Therefore, to solve the problem, we assume a
series solution in the form
11(x, y) = A0 +L A,, cash n(y - n) cos nx
consisting of the functions Xn Yn, 11 2: l. Differentiating this formally with respect

toy and applying the boundary condition at y = 0, we find
x-
71
- :E"' nAn sinh 1m cos nx
2 n=l
00
:E G 11 COS /IX
n=l
where an = -nA" sinh 1111, n = 1, 2,.... In view of the fact that .fO (x -
(11/2)) dx = 0, we see that the constant term a 0 in the Fourier cosine series
expansion of the function /(x) = x - (11/2) is zero. Therefore, the series given
above is precisely the Fourier cosine series of the function, and so
an = ; f( x - ~) cos nx dx
2 ( -1)" - 1
- - - -
(n = 1, 2, ... )
71
Hence,
2 (- l)n - 1
A,, 3
11 11 sinh 1111
or
4
Aik-1 = - -- -
(k 1, 2, ... )
11(2k I )3 sinh(2k - I )11
so that a solution of the given Neumann problem is
u(x, y) = A
0
+ ~ f ~sh( 2 k - l)(y -
3
71
) cos nx
11 k= 1 (2k - l ) sinh(2k - 1)11
and A 0 is an arbitrary constant.
Exercises 7 .3
1. Solve the Dirichlet problem

flu = 0 (0 < x < a, 0 < y < b)
u(x, 0) = f(x) (0 s x s a)
u(x, b) = 0 (0 s x s a)
u(O, y) = u(a, y) = 0 (0 s y s b)
In particular, find the solution when f(x) = sin(11x/a).

Sec. 4 Diriclilet Problem i11 a Rectangle 265
2. By interchanging the role of x and y, deduce from (4.12) the solution of the
problem
!1.u = 0 (0 < x < a, 0 < y < b)
11(x, 0) = u(x, b) = 0 (0 :S x :S a)
11(0, y) = 0, u(a, y) = _q( y) (0 :S y :S b)
Jn particular, find the solution when _q(y) = sin(2ny)/b.

!1.u = 0 (0 < x < 7!, 0 < y < 7!)

u(x, 0) = sin 2x, u(x, n) = 0 (0 :S x :S n)
u(O, y) = 0, u(n, y) = sin 2y (0 :S )' :S 7!)
4. Solve the nonhomogeneous problem
!1.u = -1 (0 < x < 7!, 0 < y < 7!)

u(x, 0) = 11(x, n) = 0 (0 :S x :S 7!)
u(O, y) = u(n, y) = 0 (O :S y :S n)
Hint: Introduce the new variable

x(n - x)
w(x, y) = - u(x, y)
2
5. Solve the nonhomogeneous problem
!111 = - sin x (0 < x < n, 0 < y < l)

11,.(x, 0) = 11(x, I) = 0 (0 :S x :S n)
u(O, y) = 11(n, y) = 0 (0 s: y s: I)

!111= x - y (0 < x < I, 0 < y < I)
u(x. 0) = 111.(x, l) = 0 (0 :S x :S I)
ux(O, y) = 0, 11(1,y) = 0 (0 :S y :S I)
7. Let u be a solution of the problem
!111= 0 (0 < a < x, 0 < y < a)

u(x, 0) = 0, u(x, a) = - f(x) (0 :S x :S a)
u(O. y) = 0, u(a, y) = f(y) (0 :S y :S a)
Show that u(x, y) = - u(y, x); thus, deduce u(x, x) = 0.

8. Use the result of Problem 7 to find the harmonic function for the right isosceles
triangle bounded by the lines y = 0, x = n, x = y which satisfies the boundary
conditions
u(x, 0) = 0, u(n, y) = sin 3 y, u(x, x) = 0
9. Find the harmonic function for the right isosceles triangle of the preceding
problem which satisfies the boundary conditions u,.(x, 0) = 0, u(n, y) = /(y),
(cu/cn)(x, x) = 0.
1'111 - hu = 0 (ha positive constant; 0 < x < rr, 0 < y < rr)
u(x, 0) = 0, u(x, rr) = f(x) (0 :S x <:: re)
u(O, y) = 11(n, y) = 0 (0 :S y :S re)
I 1. Solve the Neumann problem
!iu = 0 (0 < x < a, 0 < y < b)

uy(x, 0) = u,.(x, b) = 0 (0 :S x :S a)
llx(O. y) = g(y), u)a, y) = 0 (0 :S y :S b)
assuming that sg g( y) dy = 0.
!iu = 0 (0 < x < a, 0 < y < b)

ux(O, y) = 0, u/a, y) = g(y) (0 :S y :S b)
u,.(x, 0) = f (x), u/x, b) = 0 (0 :S x :S a)
assuming that
f f(x) dx = 0 and J: g(y) dy = 0
13. Find the solution of the Neumann problem
1'111 = 0 (0 < x < rr, 0 < y < re)

u.JO, y) = u_Jn, y) = 0 (0 :S y :S rr)
u,.(x, 0) = x/rr, (0 :S x :S re)
14. Find the solution of the Neumann problem
!iu = 0 (0 < x < re, 0 < y < re)

ux(O, y) = cosy, u.x(n, y) = 0 (0 :S y :S n)
u,(x, 0) = 1 - 2x/n, lly(x, re) = 0 (0 :S x :S rr)
15. Solve the mixed boundary value problem
!iu = 0 (0 < x < rr, 0 < y < re)

u(x, 0) = 0. u(x, re) = f(.y) (0 :S x :S rr)
uJO, y) = ux(n, y) = 0 (0 :S y :S rr)
In particular, find the solution when f(x) = 4 cos 2x.

16. Solve the mixed boundary value problem
!iu = 0 (0 < x < re, 0 < y < re)

ux(O, y) = g(y), u(rr, y) = 0 (0 :S y :S rr)
u(x, 0) = f(x), u,.(x, rr) = 0 (0 :S x :S n)
Sec. 5 Dirichlet Problem in a Disk 267
5. Dirichlet Problem in a Disk
Let D and C denote respectively the interior and boundary of the disk x 2 +
2
y S: a2 (Fig. 7.3) with radius a and center at the origin. We seek a harmonic
function u in D, which assumes the prescribed values u = f on the boundary
C; that is,
(5.1) in D, u=f on C
FIG. 7-3 Dirichlet problem in a circular region.
In order to be able to apply the method of separation of variables to this problem,

we introduce polar coordinates
x = r cos 8, y = r sin 8 (0 s: 8 s: 2n, 0 < r s: a)
to transform the Laplace's equation into the form
I 1
(5.2) urr + - u, + 2 U99 = 0
r r
The boundary condition in (5.1) becomes
(5.3) u(a, 8) = f (8) (0 s: 8 s: 2n)
Thus, the problem (5.1) is reduced to the equivalent problem (5.2), (5.3), where
now the separation-of-variables method can be applied. Accordingly, we
assume particular solutions of (5.2) in the form
u(r, 8) = R(r) 0(8)

268 Laplace's Eq11atio11 Chap. 7
Substituting this in (5.2) and separating variables, we obtain
,. 2 li"(r) + ,. ~'(r) = _ 8"(8) = }.

R(0 R(0 0(~
where i. is our separation constant. This leads to the two ordinary differential
equations
(5.4) r 2 R"(r) + rR'(r) - J.R(r) = 0

and
(5.5) 0"(0) + ).0(8) 0
for the functions R and 0, respectively.

In order for the solution of the problem (5.2), (5.3) to be single-valued, it is
necessary to require the solution to be periodic in e,
of periodic 2n:; that is,
u(r, 0 + 2n:) = u(r, 0)
This requirement is clearly met if 0 is periodic, of period 2n. Hence, we impose

the "periodic" boundary conditions
(5.6) 0(-n:) - 0(n:) = 0, 0'(-n:) - 0'(n:) = 0
for the function 0. Equation (5.5) together with the conditions (5.6) constitute
an eigenvalue problem [Problem 15, Exercises 4.4] for which the eigenvalues
are J. = n 2 , n = 0, I, 2, ... , and the corresponding eigenfunctions are
When ), = 0, equation (5.4) has the general solution

R 0 (r) = A 0 + B 0 Jn r
and when /. = n 2 11 = I, 2, ... , it has the general solution
Therefore, all the functions
u"(r, 0) = A 0 + B 0 In r
(5.7)
+ (A"r 11 + B,,r-")(C,, cos 118 + D" sin 118)
n = 0, I, 2, ... , r > 0 which are periodic of period 2n, satisfy Laplace's

equation (5.2) in polar coordinates. These functions are called the circular
harmonics.
We shall form a suitable infinite combination of the functions (5. 7) for our
solution of the problem (5.2), (5.3). Since the solution must be continuous in
D, which contains the origin r = 0. we must exclude from our consideration
Sec. 5 Diricltlet Problem in a Disk 269
the logarithmic term and the terms involving negative powers of r. We there-
fore choose B" = 0, 11 = 0, I, 2, ... in (5.7) and consider the infinite series
(5.8) u(r, 0) = 1Xo

2
+ I
11= I
r 11 (1X 11 cos 110 + /3 11
sin 110)
where we have relabeled the constants. To determine the constants 1Xo, IX",
and {3 11 , 11 ~ 1, we set r = a and apply the boundary condition (5.3) to obtain
(5.9) f(O) = ~o +
2
f
11= I
a"(1X 11 cos:o + /3 11
sin nO)
This is the Fourier series expansion off on the interval [ -rr, rr ], for which the
coefficients are given by
a" = a cx 11 =
11 I
II" /(8) cos 110 de (11 = 0, 1, 2, . ' . )
rr - rr
(5.10)
b11 = a"/3 11 =
1
rr - rr
j" f(e) sin 11e de (n = 1, 2, ... )
Substituting for cx 11 and /3 11 in (5.8), we obtain
11
(5.11) u(r, 8) =
a0 + L -. (a
"' ( ,.) 11 cos 110 + b" sin 118)
2 11= I Q
where a11 and b11 are the Fourier coefficients off on [ - rr, rr ], given by the
integrals in (5.10).
Formula (5.11) together with (5.10) gives the solution of our problem (5.2),
(5.3). To verify this, we now assume thatf is continuous, piecewise smooth, and
periodic of period 2rr on [ -rr, rr]. Let
M = ~ f" lf(O)I dO
so that Ja 11 I s M and lb I s 11 M, and define

11
11 11 (r, 0) = (. ') (a 11 cos 11e + b11 sin nO) (11=1,2, ... )

a,
with u0 = a0 /2. Then, for r s r0 with any r0 < a,
(5.12)
The series consisting of the terms 2M (r /a)" converges uniformly for r s r 0 .

By Theorem 6.1 of Chapter 5, the series (5.11) also converges uniformly for
r = a. Therefore, the series ( 5.11) converges uniformly to u(r, 0) for r s a,

and thus u(r, 0) is continuous for 0 s r s a, 0 s 0 s 2rr. Since
a oo
u(a, 0) = - 0 + L (a,, cos 110 + b,, sin nO) = f(O)
2 11= I
the boundary condition (5.3) is met.

To show that u satisfies the Laplace's equation (5.2), we note that the series
(5.11) and its first and second partial derivatives are dominated by some con-
stant multiple of the series :L 2Mn 2 (r /a)" - 2 . This series converges uniformly
for r s r0 < a. Hence, u has continuous first- and second-order derivatives
for r < a, which can be obtained by termwise differentiation of the series
( 5.11 ). It follows that
00 1
n= I
L -_, u,,(r, O)[n(n
,.-
- I) + n - n2] 0
thus verifying that u satisfies (5.2).
Example 5.1. Solve the Dirichlet problem
1
u,, + - u, + (r < a)
r
u(a, 0) = a cos 2 0 (0 ~ 0 ~ 2n)
Solution: We note that

a cos 2 0 = a(~ + } cos 20)
Therefore, by (5.11), we have
u(r, 0) = ~ (a + r: cos 28)

Example 5.2. Find the solution of the Dirichlet problem
llxx + llyy = 0 for x 2 + y2 <

u = y2 for x 2 + y2 = I
Solution: Considering the problem in polar coordinates, we see that the boundary
condition becomes u(I, 0) = sin 2 0 = -W - cos 20). Thus, by (5.1), we have
u(r, 0) = HI - r 2 cos 20)
Reverting back to rectangular coordinates, we find
u(x, y) = 1(1 - x2 + y2)
which is the solution of the problem.

Sec. 5 Dirichlet Problem in a Disk 271
Exercises 7.4
1. Find the solution of the Dirichlet problem

I I
13.u = /Irr + - 11, + 11 66 = 0 (r < a)
r r2
u(a, 0) = a 3 sin 3 () (0 s () s 27T)
!111 = 0 (r < 1)
u(I, (J) = sin 0 + cos 2 8 (0 S (J s 2rr)
3. Solve the Dirichlet problem
u(x, y) = .x2 , x2 + y2 = a2
!lu = 0 (x2 + yl < 4)
u(x, y) = y 4 (x2 + y2 = 4)
5. Show that if the boundary data/ in (5.3) is an odd function of 0 (that is,/( - {}) =
- /({})),then so is the solution 11(r, (})of the problem (5.2), (5.3)-that is, 11(r, -8) =
-u(r, 0)-and thus u(r, 0) = 0 and u(r, rr) = 0. Hint: Set z;(r, (}) = u(r, 0) +
u(r, -8).
6. Use the result of Problem 5 to find the solution of the problem
!111 = 0 (r < a, 0 < 0 < rr)

11(r, 0) = u(r, 8) = 0 (0 s r s a)
u(a, 0) = 4 sin (} cos 0 (0 S 0 S rr)
7. Show that if the boundary data fin (5.3) is an odd function of() and is periodic
of period rr, then so is the solution u(r, 0) of the problem (5.2), (5.3 ). Thus,
deduce that 11 vanishes for (} = 0, rr, and rr/2.
8. Use the result of Problem 7 to find the solution of the problem
!111 = 0 (0 < r < a, 0 < 0 < T<i2)

u(r, 0) = u(r, 7i/2) = 0 (0 S r s a)
u(a, 0) = 2a sin 20 cos 20 (0 s 0 :S 7r/2)
!lu = 0 (r < 2. 0 < 0 < T</2)

u(r, 0) = u 0 (r, rr/2) = 0 (0 < r < 2)
u(2. 0) = 4 sin 3 (}
10. Solve the Dirichlet problem for an annular domain
/111 = 0 (a < r < b)

11(a, 0) = 0 (0 S 0 S 2rr)
u(b, 0) = g(O) (0 S (J S 2n)
11. Solve Problem I 0 when the boundary conditions are

u(a, 8) = /(8) (0 s 8 s 2n)
u(b, 8) = 0 (0 s 8 s 2n)

f:i.11 = 0 (a < r < b, 0 < 8 < n)
u(r, 0) = u(r, n) = 0 (a S r s b)
u(a, 8) = 0, u(b, 8) = f (8) (0 s 8 s n)
f:i.u = 0 (a < r < b, 0 < 8 < n)
u(r, 0) = /(r), u(r, n) = 0 (a s r s b)
u(a, 8) = u(b, 8) = 0 (0 s 8 s n)
Hint: ri~ = eii. ln' = cos(). In r) + i sin(), In r ).
f:i.u = 0 (a < r < b, 0 < 8 < n)
uo(r, 0) = 0, ue(r, n) = f(r) (a S r s b)
u(a, 8) = u(b, 8) = 0 (0 s 8 s n)
6. Poisson's Integral Formula
The solution (5.11) of the Dirichlet problem in a disk can be expressed in the
form of an integral from which we can deduce several important and useful
properties of harmonic functions. This is accomplished by substituting the
formulas (5.10) for the Fourier coefficients an and bn in (5.11). We obtain
u(r, e) = J"_,, f(</J) d</J

-1
2n
+ 1 n~l -~r)" f" /()(cos n</J cos ne + sin n</J sin ne) d</J
00
(
7r -n
(6.1)
= 2In J"_,, f(</J) d</J + n1 J ~ f" f(</J) cos n(8 - ) d</J
00
( ,.) n
_"
1
00
= -1
7r
J" -,,
f(</J) [ -1
2
+ ~
n-1
( -")"
a
cos n(8 - ) ] d</J
where we have interchanged the order of summation and integration. This

interchange is justified, since by ( 5.12) the series
00
(6.2) -1 +
2
L:
n= 1
-
( ")"
a
cos n( e- )
converges uniformly for r < a and - n s es n.

Sec. 6 Poisson's llltegral Formula 273
Let us calculate the sum of the series (6.2). We Jet z = pei = p(cos Cl. +
i sin rx), where jzj = p < 1, and consider the series
x
l
2
+ 11=
I. 1
z'1
Since jzj < I. we have
+
"' -"
I. ,!,
+ -- l +-
2 11= 1 2 z 2(1 - .z)
+ p cos rx
- -
+ ip sin rx
-
2(1 - pcosrx - ipsinet.)

(6.3)
(I + p cos rx + ip sin rx)(I - p cos Cl. + ip sin rx)
2[(1 - p cos rx) 2
+ p 2
sin 2
Cl.]
1 - p 2
+ 2ip sin rx
2( I - 2p cos :J. + p2)
On the other hand, since
z" = p"ei" = p"(cos nrx + i sin 11C1.)

we have
(6.4) + I. -" + I, p"(cos llCI. + i sin 1m)

2 n=l 2 11= I
Thus, by equating the real parts of the right-hand sides of (6.3) and (6.4), we
find
oc
l - p2
(6.5) + I. p" cos ll'.J.
2
2 11= l 2( I 2p cos '.J. + p )
It follows that (6.2) has the sum
- + I,
]
2
00
n= l
(r)"
a
cos n(O
2(a 2
a2 _
2ar cos(O - >)

1
.2
+
-
r 2)
so that ( 6.1) becomes
(6.6) u(r, 0) = l J" (a2 - ,.2)

- f(<f>) d<f>
2n _,, a 2 - 2ar cos(B - >) + r2
This is known as Poisson's integral formula for the solution of the Dirichlet
problem in a disk.
We recall that in (5.11 ). f was assumed continuous and piecewise smooth in
addition to being periodic of period 2n. In the integral formula (6.6), it is
sufficient to require that f be continuous and periodic. For then the function
11 defined by (6.6) for r < a and by u(a, 0) = f(B) for r = a is continuous
throughout the disk r :s; a. Moreover, u has continuous derivatives of all
orders with respect tor and 13 for r < a, which can all be obtained by differentia-
tion within the integral sign. By a straightforward calculation, it is easily seen
that the function
1 al _ ,.2
(6.7) K (a r 13 - </J) = -- -
, ' 2rr a 2 2ar cos(O - ) + r2
satisfies (5.2) for r < a, 0 ~ 0 ~ 2rr. Thus, the integral formula (6.6) with
u(a. G) = f(G) represents the unique solution of the Dirichlet problem (5.2),
(5.3). Frequently, this formula is more convient to use than the series (5.11 ).
The function (6. 7) is often called Poisson's kernel. This function is positive
for 0 ~ r < a, -rr ~ 13 ~ rr. Indeed, since a 1 - r 2 > 0, we see that
a2 - 2ar cos(O - ) + r 2 = (a r) 2 + 2ar[I - cos(8 - )]
=(a r) 2 + 4ar sin 2 ( 8 - )

2
>0
Further, Poisson's kernel satisfies the interesting property
(6.8)
I - I Jn (a2__- __,.2) dj_ --
2rr -n a2 - 2ar cos(O - ) + r2
This result follows from (6.6) when f(</J) = I and the fact that in such a case
u(r, 8) = I by the maximum principle.
If we set r. = 0 in (6.6), we then obtain the very important result
1
u(O, 0) = -- Jn f(</J) d</J
2rr -n
or
I {n
(6.9) u(O, 0) = - I u(a, ) ds
2rra.-n
since u(a, </J) = f(</J) and ds = a d</J. This formula expresses the well-known
mean-value property of harmonic functions. Let us state this property as a
theorem in a more general setting.
THEORE:\l 6.1. Let u be a harmonic function in a domain D. Then the z:alue

of 11 at the center of any disk lying in D is equal to the mean or average of the
values of u on the boundary of the disk.
It is noteworthy that the mean-value property of a harmonic function can be

established by using merely the harmonicity of the function without resorting
to Poisson's integral formula. To see this, let D, be a disk of radius r and
center (x, y) lying in D. Then each point(~, 17) in D, can be expressed in polar
coordinates
.; = x + p cos 8 (0 ~ p ~ r)
17 = y + p sin 8 (0 ~ 8 ~ 2n)
Sec. 6 Poisson's Integral Formula 275
with pole at (x, y). Applying the identity (2.11) to the domain D,, we obtain
(6.10) f" ~~-

0
(x + r cos (}, y + r sin (})r d(} = JJf..up dp d8
D,
Since u is harmonic, f..11 = 0 and thus (6.10) becomes
lrr CU .
+ +
(6.11)
J
0
~
or
(x r cos 8, y r sm 8) d{) = 0
Let us set
lrr
(6.12) U(x,y;r)= u(x+rcos(},y+rsinO)dO
J 0
Since u and its derivatives are continuous in D,, so is U together with its first
derivatives. It follows that we can calculate cU /er by differentiating (6.12)
under the integral sign. In view of (6.11), we find
cU
~ - =
or
f
2" cu
~ (x + r cos 8, )' + r sin 8) d8 = 0
or
0
This implies that U is independent of r; in_ particular, we have
(6.13) U(x, y; 0) = U(x, y; r) (0 :S: r :S; a)
Now, from (6.12), we see that
U(x, y, 0) = Jim
r-o
f 0
2" u(x + r cos (}, )' + r sin 8) d8
2rc
J 0
u(x, y) d(}
= 2nu(x, y)
Therefore, by (6.12) and (6.13), we finally obtain
(6.14) u(x, y) = }; f" u(x + r cos(}, y + r sin 8) d8
which proves Theorem 6.1.

This yields the special case (6.9) when the point (x, y) happens to be the origin.
It is interesting to note that the converse of Theorem 6.1 is also true. That is,
if u is continuous and satisfies the mean-value property (6.4) for any r in a
domain D, then u is harmonic in D. As a matter of fact, under the continuity
requirement it can be shown that u, as expressed by the formula (6.14), has
continuous derivatives of all orders in D which themselves also satisfy the rnean-
value property. This implies that a harmonic function necessarily has derivatives
of all orders. Here, we shall show only that f..u = 0 in D.
Suppose that there is a point (x, y) in D at which f..u of. O; say, f..u(x, y) > 0.
Since u is continuous, there is a disk D, about (x, y) with radius r and lying in
~'
D such that 611 > 0 throughout D,. Then. applying the formula (2.11) to D,.
we find
0 < JJ6u d~ d17

D,
I' I
,; 0 0
2rr /J..up dp dO
CU
J er
2rr
r dO
0
2
r o f rr u(x + r cos e. y + r sin 8) dB
or Jo
which, by (6.14), yields
0 < JJ!J..u d~ di) = r :r [u(x, _r)J 0

D,
a contradiction. Therefore, 611 = 0 throughout D.

Because of Theorem 6.1 we are now in position to establish the statement
made in Section 3 to the effect that a constant is the only harmonic function
that can attain a maximum inside a domain. We shall state and prove this
fact as a theorem.
THEOREM 6.2. Let 11 be a harmonic function in a domain D. If 11 attains its

maximum i:alue at a point inside D, then 11 is a constant.
FIG. 7-4 The strong maximum principle.

Sec. 6 Poisson's Integral Formula 277
Proof: Let M denote the maximum value of u in D (Fig. 7.4), which is

attained at a point P0 : (x 0 , y 0 ) inside D. We shall prove that u = M identically
in D. Let Q be any other point inside D and join P 0 and Q by a smooth curves
lying in D. Beginning from the point P0 and moving toward the point Q, we
shall construct a finite number of disks lying in D anc.l with centers on S such that
the center of each disk lies inside the preceding disk. We shall prove that
u = M throughout each of the disks. Thus, let D0 be a disk with center at p 0
and radius r0 such that D 0 lies in D. Then, by the mean-value property of u,
we have
(6.15) M = u(P 0 ) = I J2n u(r 0 ,8)d8

2n 0
Now, on the circumference C 0 of D 0 , we have u(r 0 , 8) ::; M. If u(r 0 , 8) < M

for some point on C 0 , then by the continuity we would have u < Mover some
arc of C 0 containing that point; consequently,
I J2n u(r 0, 8) d8 < M

2n 0
which contradicts (6.15). Therefore, u = M identically on C 0 . But the preceding

argument holds as well for any concentric disk with radius r for 0 ::; r _.::; r0 .
Hence, u = M identically in D 0 .
Now let P 1 be the point of intersection of C 0 with the curve S so that u(P 1 ) =
M. Let D 1 be a disk about P 1 with radius r 1 such that D 1 lies in D. By repeating
the preceding argument, we again conclude that u = M throughout D 1 .
Proceeding in this manner, it follows that after a finite number of steps we shall
reach a disk Dn that contains the point Q and for which u = M identically in
D". Thus, u(Q) = M, and since the point Q is arbitrary in D, it follows that
u = M throughout D.
Exercises 7 .5
1. Deduce from (6.3) and (6.4) the identity
oc n . p sin t
~ p Sill II( = - - - - -
n~ I (I - 2p COS ( + p 2)
2. Verify directly the relation (6.8) by making use of the formula
CIAi < l)
3. Let m and M be the minimum and maximum values of u, respectively, on the

circle r = a, 0 s 8 s 2n:. Using (6.6) and (6.8), obtain the inequality m s
u(r, 8) s M for 0 s r s a, 0 s 8 s 2n:.
4. By using (6.6), prove that if 11 is a nonnegative harmonic function for r < a;

then 11 satisfies the Harnack"s inequality
a - r
u(O, 0) :S u(r,
a
0) :S -
+ r
u(O, 0)
a + r a - r
5. By using Harnack's inequality, prove Lhat if 11 ~ 0 is harmonic in the entire

xy-plane, then u must be a constant.
6. Show that if the function fin (6.6) is odd with respect toy (that is, /(2rr - 8) =
- /(8)), then (6.6) can be written as
u(r, 0) = f [K(a, r, 0 - ) - K(a, r, 0 + </JJ]/() drp
where K(a, r, 8 - </J) is defined by (6.7). Show that this solves the Dirichlet
problem liu = 0, 0 < r < a, 0 < 8 < rr; u(r, 0) = u(r, rr) = 0, 0 :S r :S a;
and u(a, 8) = f(8), 0 :S 8 :S 7L
r
7. Show that if the function/ is even with respect to, then (6.6) can be written as
u(r, 0) = [K(a, r, 0 - ) + K(a, r, 8 + </JJJ /( ) drp
which solves the Dirichlet problem for the semicircular domain 0 < r < a,
0 < 8 < rr, satisfying the conditions: u8(r, 0) = 11 8 (r, rr) = 0, 0 :S r :S a;
u(a, 8) = /(8), 0 :S 8 :S rr.
8. Obtain from the result of Problem 6 an integral formula for the solution of the
problem
liu = 0 (0 < r < a, 0 < 8 < rr/2)
u(r, 0) = u(r, rr/2) = 0 (0 :S r :S a)
u(a, 8) = /(8) (0 :S 8 :S rr/2)

9. Show that the exterior Dirichlet problem for a disk
liu = 0 (r > a, 0 < 8 < 2rr; ll bounded as r tends to 00)

u(a, 8) = /(8) (0 :S 8 :S 2rr)
has the solution
u(r, 8) -I
= --
2rr
f"
-n a2 - 2ar cos(O - ) + r2
10. By multiplying both sides of equation (6.14) by r dr and integrating from Oto a,
prove that the value of a harmonic function at the center of a disk is equal to
the average value of the function on the disk. That is,
u(x, y) = rr~ 2 ff D
u(.;, 11) d.; d17, D: (x - .;)
2 + (y - 11) 2 :S a 2
Sec. 7 Neumann Problem in a Disk 279
11. Let S(x, y, z; a) denote a sphere with center (x, y, z) and radius a. For each
point (, 17, (} within the sphere, introduce the spherical coordinates ~ = x +
r sin qJ COS 0, I/ = y + r Sill qy sin (}, ( = Z + r COS, Q :S qJ :S 'lr, Q :$ (} :$
2rr, r ::S a. Let 11 be a function defined on the sphere S(x, )', z; a). The integral
2
U(x, y, z; a) = I-, ( " {rr 11(,;, 17, ()a 2 sin q, dcp dO
4rra- Jo Jo
is called the mean value of 11 on the sphere S. Prove that if u is a harmonic

function in the xyz-space, then u(x, y, z) = U(x, y, z; a). (This is the mean-
value property of harmonic functions in three-dimensional space.)
12. Let 11 be a continuous function of x, )', z, with continuous first- and second-order
derivatives in a domain D. Prove that if 11 has the mean-value property stated
in the preceding exercise, then 11 is harmonic in D.
7. Neumann Problem in a Disk
We now consider the Neumann problem

cu
+
on =f
(7.1) Uu Uyy = 0 in D, on C
where D denotes the interior and C the boundary of the disk x 2 + y2 :::;; a2 .
In polar coordinates (r, 0) this problem is transformed into
I
u,, + u, + ,.--:; U99 = 0 (r < a)
r
(7.2)
~u (a, 8) = f(8) (0 :::;; 8 :::;; 2n)
er
We require the function f to satisfy the condition
(7.3) f" f(O) dO = 0
which is necessary if the problem (7.2) is to have a solution.

As in Section 5, we assume a series solution of the form
(1. Cf)
(7.4) u(r, 0) =
0
+ L r"(ct. 11 cos 118 + /3,, sin 118)
2 ll= 1
Differentiating (7.4) with respect to r and applying the boundary condition in

(7 .2) for r = a, we find
00
f(O) L:
n;::: 1
11a"- 1(a,, cos 118 + /3,, sin 118)
(7.5) 00
I. (a,, cos 118 + b,, sin 118)

n= 1
280 Laplace's Equarion Chap. 7
where we have set na"- 1 ex,. = a,. and na"- 1 /3,. = b,. for 11 ;::;. I. This is precisely
the Fourier series expansion off on the interval -n ::::: () ::::: n, with its constant
term a0 being zero because of the condition (7.3). Therefore,
(7.6)
Cf.fl=
na"- 1 n
r -rc
f(8) cos 17{) d()
/311 =
11a 11 - 1 n
J_"." f(8) sin 118 d()
where n = 1, 2, .... Substituting these in (7.4), we obtain
u(r, 0) = .ex 0 + a ~ 1
2 nn-111
00
(r )" J"
a

-re
[cos 118 cos mp + sin n8 sin 11 JJ( ) d</J
(7.7) = ex () + a
2 n
00
L
11= 1 11
l (r) Jn
a
11
-re
f(</J) cos n(O - ) d</J
= r:1.~o
2
+ a Jn
n _"
l ~
00
,._ 1 n a
J ( -,.)" cos 11(8 - ) J .f(</J) d</J
where ex 0 remains an arbitrary constant. J'\ otice that the interchange in the
order of summation and integration in (7.7) is justified, since for r < a, the
sen es
(7.8)
11=
I
Cf)
I
l (.
/1
r)
G
11
cos n(8 - )
converges uniformly with respect to .

We now calculate the sum of the series (7.8). From (6.5) we recall that, for
IPI <I,
ro " ] - p2 j
L p cos /lex = - - - -- - 2 -
11=1 2(1-2pcosr:1.+p) 2
,
p cos ex - p-
-- --- --
(1 - 2p cos a + p 2 )
and hence
00
L p"-1 cos na = c_os_a_-:-- _p___
"= 1 1 - 2p cos a + p 2
The last series converges uniformly for IPI < I so that, by termwise integration,
we find
n~l ~ p" cos na = 11~! cos na I: ,n- I d(
(cos a - d( 0
- 2( cos a + ( 2
= -! Jn(l - 2p cos a + p 2 ) (IPI < 1)

Hence, the series (7 .8) has the sum
(7.9)
00
1 1
~ J a
/1
(r)" cos n(O - </J)
and the solution (7.7) becomes
(7.10) u(r, 0) = ~o - a f"_,, f(</>) ln[a 2

- 2ar cos(8 - ) + r 2 ] d</>
2 2rc
where we have used (7.3) to drop the factor l/a 2 in the argument of the
logarithm.
When f is continuous and satisfies condition (7.3), formula (7.10) provides
a solution of the Neumann problem (7. l ), (see Problem 6). In accordance with
Theorem 2.2 of this chapter, this solution is unique up to an additive constant
C = rt. 0 /2.
t>.u = 0 (r < I)
OU
sin 8 (0 :'.".: 8 :'.".: 2rr)
on
Solution: We note that
Jorh sin 0 d8 = 0
so that condition (7.3) holds. From (7.6) we see that a" = fl" = 0 for n ~ J,
except /3 1 = 1. Hence, by (7.4), the solution is
u(r, 8) = r sin 8 + C

t>.u = 0, x2 + y2 < 4
au
- . = xy on x
2
+ y
0
= 4
en
Solution: In polar coordinates we note that the boundary condition becomes
(i:Ju/or)(2, 8) = 2 sin 28, for which condition (7.3) holds. From (7.6) it follows
that the only nonzero coefficient in (7.4) is [3 2 = 1/2. Hence, the solution is
,.2
u(r, 8) = - sin 20 + C
2
which, in rectangular coordinates, becomes u(x, y) = xy + C, with arbitrary

constant C.
It is interesting to note that the Neumann problem in the xy-plane can be

reduced to a Dirichlet problem by utilizing certain elementary results from the
.
theory of complex variables. 11. is kno\vn that if r is a harmonic function in a

simply connected domain D, there is another harmonic function 11 in D, called
the harmonic conjugate of 1, such that
(7.11) u< = r".
These equations are known as the Cauchy-Riemann condition~ for analytic

functions. From (7.11) it follows that
du =
cu dx +
cu dr
c1 er
dx - -:. dy
ax cy CJ OX
Thus, if i is a harmonic function in a domain D. then its harmonic conjugate 11
is given by the line integral

('cv,, d~--:_d11
(7.12) u(x,y)=
J(X,)')
(xo,Yo) ll)
CV )
cc;
along any path lying in D that joins an arbitrary point (x 0 y 0 ) to (x, y).
Now suppose that 11 is a solution of the Neumann problem
(7.13) L'lu = 0 in D, au = f' on C

en -
which is continuous together with its first derivatives in D + C. Since u is
harmonic, we know that its harmonic conjugate v exists. Let the boundary
C of D be given parametrically by the equations x = x(s), )' = y(s), 0 ~ s ~ s0 ,
wheres denotes the arc length of C. By the Cauchy-Riemann conditions (7 .11 ),
we see that on C,
CV av dx CV dy -au dx cu dy cu
(7.14)
ox ds + cy ds (r ds
+ ex ds en
OS
That is. the tangential derivative of 1' on C is equal to the normal derivative of
11 on C. Hence, the harmonic conjugate i of the solution of the problem (7.13)
satisfies the equation
CT
(7.15) -.::- =f on C
cs
It follows that r is given on C by the formula
(X,J)
(7.16) r(x, y) =
J(xo.yo)
I ds = (x, y)
with (x 0 , y 0 ) being an arbitrary point on C. Thus, r is a solution of the Dirichlet

problem
(7.17) M = 0 in D, r = on C
where </> is given by (7.16). Having found r from (7.17), we can then determine
its harmonic conjugate u by (7.12). Because of(7.14) and (7.15), it follows that
u is a solution of the problem (7.13). As an illustration, we consider the next

example.
Example 7.3. Solve the problem

6u = 0 inx 2 + y 2 <a 2
cu x2 _ y2 on x2 + yz = az
on
by first reducing it to a Dirichlet problem.
Solution: In polar coordinates we see that on the circumference r = a, (7.16) gives
i-(a, ()) = J: 2
a 2 (cos - sin 2 ,)a d
9
a 3 sin 2fJ
a Jo
3 (
cos 2 d =
2
Therefore, the corresponding Dirichlet problem is
6v = 0 in x 2 + y 2 < a2
v(a {}) = a~in 28 on x 2 + y 2 = a2

, 2
which, by (5.11), has the solution

2
ar
v(r, 8) = - sin 28 = axy
2
Hence, by (7 .12), the solution of our original problem is
(x.y)
u(x, y) =
J (x.J")
(vy dx - Vx dy)
a(xz - y2)
a (x dx - y dy) - + c
J 2
Exercises 7 .6
1. State why the Neumann problem

611 = 0 (r < 1)
ou = sm. z 8
-~~ (0 ::; 8 ::; 2n)
er
does not have a solution. Does the problem
611 = l (r < 1)
ou
- = sm z 8 . (0 ::; 8 ::; 2n)
or
have a solution? If so, find the solution.

"'1.11 = 0 (r < 2)
oil
2 sin 3 0 (0 S IJ S 2n)
er

"'1.11 = 0 (r < I)
cu sin 20 (0 S 0 S 2n)
or
llxx + u,.y = 0 (x2 + J'2 < 4)
Oil
(x2 + y2 4)
en
llxx + llyy = 0 (x2 + y2 < I)
cu = x
3
- 3xy 2 (x2 + y2 1)
(m
6. By differentiating (7.10) with respect to r within the integral sign, show that
OU
or
a Jn K(a, r, 0- dJ)f( </;) d@
r -n
where K is the Poisson's kernel given in (6.7), and thus deduce that
. (II
IIm = /tOJ (r < a)
r- a (: r
7. (a) Let/be a periodic function such that j"_rrf(fl) dO = 0. Define
fldJJ = r<P j(OJ dO

"0
Show that F(n) - F( - n) = 0.

(b) Replace l(cp) in (7.10) by F'(q>l and integrate by parts to show that (7.10)
can be written as
u(r O) = -a r
2
f" 2
F(d,) sin<8 ~ @) d@
' n .-rra -2arco<;((J-<f;)+r 2
8. Find a solution of the exterior Neumann problem for a disk
"'1.11 = 0 (r > a)
(11
f(O)
er
(0 s es 2n; 11 bounded as r ~ o::;)
Sec. 8 Problems in Infinite Domains 285
9. Show that if the function fin (7.10) is odd with respect to y, then (7.lOJ can
be written as
u(r, 0) =
- a J" /(ef,) In
1
a--- - Zar--cos(O - )--+ r d<P
----
2
Zrc 0 a 1 - Zar cos(() + ) + r 2

which solves the mixed boundary value problem fi.u = 0, r < a, 0 < 8 < re;
u(r, 0) = 11(r, re) = 0; cu/or = /(8), 0 :S 8 :S re, r = a.
10. Deduce from (7.10) the solution of the boundary value problem
!!.u = 0 (r < a, 0 < 8 < re)
u6(r, 0) = u0(r, re) = 0
OU
or
/(0) (O :S e :S TC, r = a)
where JO /(8) d8 = 0.
8. Problems in Infinite Domains
In this section we shall consider boundary value problems for the Laplace's
equation in an infinite domain of the xy-plane. We shall again use the method
of Fourier transform, as we did in the case of the heat equation. We consider
first the Dirichlet problem in the upper half-plane y > 0, namely,
(-oo < x < oo, y > 0)

(8.1)
u(x, 0) = f(x) (-oo < x < oo)
We take the Fourier transform of the solution u with respect to the variable x,
treating y as a parameter. Following the by-now familiar procedure, we assume
that the problem (8.1) has a solution u that, together with its first- and second-
order partial derivatives, is piecewise smooth and absolutely integrable on
- oo < x < oo for y > 0. Then the Fourier transform of 11,
(8.2) U(s, y) = J~~ J: 00

u(x, y)e-isx dx
exists and
(8.3) u(x, y) = ,A~ f:"' U(s, y)eisx ds
Differentiating (8.2) under the integral sign twice with respect toy and using
Laplace's equation, we obtain
f~
00 uyy(x, y)e-isx dx
-1
,/)_-~
f: 00
Uxx(x, y)e-isx dx
As usual, we require that both 11 and ux tend to zero as lxl --+ oo. Then integra-
tion by parts twice yields the equation
(8.4) V_Js, y) = s 2 U(s, y)
for the transform (8.2). This has the general solution
(8.5) V(s, y) = C 1 (s)e-sy + Ci(s)e 5 Y
involving two arbitrary functions C 1 and C 2 From the boundary condition
in (8.1) we get one condition to be satisfied by (8.5). We therefore need another
condition in order to be able to determine the functions C 1 and C 2 uniquely.
A physically plausible condition is that our solution u be bounded as y becomes
infinite. This requires (8.5) to be bounded for large values of y. Consequently,
when s > 0, we must choose C 2 = 0 so that (8.5) becomes
(8.6) (s > 0)
Setting y = 0 in (8.2) and (8.6) and using the boundary condition given 111
(8.1 ), we see that
U(s, 0) = C 1 (s) = F(s)
where F denotes the Fourier transform off. Hence, whens > 0, (8.6) becomes
(8.7) U(s, y) = F(s)e-sy
In the same manner, when s < 0, we find

(8.8) U(s. y) = F(s)e'Y
The formulas (8.7) and (8.8) can be jointly written as
(8.9) U(s, y) = F(s)e- lslY
for all values of sandy ~ 0. Substituting this in (8.3), we then obtain
u(x, y) = -1= J
, 2rr
00
- "'
F(s)eisx- lsly ds
(8.10)
= 1 Joo JW J" eis(x-~)- lsJy ds d(
2rr -'YJ -oo
where we have introduced the definition of F(s) and formally interchanged the
order of integrations. The inner integral with respect to s can be evaluated to
give
(8.11)
oo
= 2
J
2y
0
e-sr cos s(x - 0 ds
(x _ ()2 + y2
Sec. 8 Problems in Infinite Domains 287
Here we have observed that 2 cos Cl. = e;" + e-;._ Thus, (8.10) simplifies to
(8.12) u(x y) = Y f"'- JW ~ -- (y > 0)

' , n oc (x - 02 + yz
This is known as Poisson's integral formula for the half-plane y > 0.
Under the assumption that f is continuous and bounded for - oo < x < oo,
it can be shown that the integral (8.12), together with its partial derivatives of
any order taken under the integral sign, converges uniformly in x and y for
y ;:::: y 0 , - I/Yo :::;: x :::;: l/y 0 with any Yo > 0. This means that u has continuous
partial derivatives of all orders. By a straightforward calculation, it is readily
shown that u satisfies Laplace's equation for y > 0. Moreover, by making the
change of variable x - ( = y tan t, (8.12) becomes
u(x, y) = 1 f"'2 f(x - y tan t) dt

n -rr/2
which can be shown to have the limit

ir/2
. u(x, y) =
IIm
y--+O..
rIm -1
y~O"' 7f. J -rr/2
f (x - y tan t) dt
= f(x)
If we define u by (8.12) for y > 0 and u(x, 0) = .f (x) for y = 0, then u rep-
resents the unique bounded solution of the problem (8.1 ).
It should be pointed out that the requirement that u be bounded, which was
imposed in the determination of (8.9), is an essential condition if the problem
(8.1) is to have a unique solution. For we see that the function
i'(x, y) = (eY - e-y) sin x

which is unbounded for y > 0, satisfies Laplace's equation and vanishes at
y = 0. Hence, the superposition of (8.12) and vis also a solution of the problem
(8.1), and so the problem does not have a unique solution.
It is noteworthy that the problem (8.1) can also be solved by considering the
Fourier sine transform of its solution with respect to the variable y. That is,
we set
(8.13) Us(x, s) = e) + I" u(x, y) sin sy dy
Then, differentiating (8.13) twice with respect to x and using the differential
equation together with the boundary condition in (8.1), we obtain
(2)t f"' u,jx, y)

;
0
.
s111 sy dy
e) + sf(x) - s2 U.(x, y)
Here we have assumed as usual that 11 and ur vanish as y -. oo. Thus, the
transform Us satisfies the equation
(8.14)
Further, we assume that u is bounded for lxl < oo, 0 < y < oo, and vanishes
as lxl -. oo. This requires that u.. satisfy the boundary condition
(8.15) u. (x, s) __, 0 as lxl -. oo
Equation (8. I 4) together with the condition (8.15) constitutes a singular Sturm-
Liouville problem. By the method of Section 2, Chapter 4, the Green's function
(Problem 17, Exercises 4.2) for this problem is given by
G(x; ~) = - ;s e-six-~I
Therefore,
U,(x, s) = - (rr2)-''s
- Jae-oc f(~)G(x; ~) d~
11 J" J(~)e-slx-~I d~
Y 2rr - CYJ
Substituting this in formula (8.8) of Chapter 5 for the inverse Fourier sine
transform, we find
(8.16) u(x, y) = (;2)t f"" U,(x, 0

s) sin sy ds
= l
rr
J 00
-oo
JC() f" e-slx-~I sin sy ds d~
o
after an interchange in the order of integration. Since
J oo
0
e-st sin sy ds =
t2
_Y__
+ y2
(t > 0)
We see that (8.16) leads back to the solution (8.12).

Next, we consider the Dirichlet problem in a semi-infinite strip D; 0 <
x < oo, 0 < y < b; that is
= 0
~u in D
u(x, 0) = f(x) (x ~ 0)
(8.17)
u(x, b) = 0 (x ~ 0)
u(O, y) = 0 (0 ~ y _:-: ; b)
where we require that u be bounded as x --> oo. It is appropriate for this

problem to use Fourier sine transform in the variable x, treating y as a parameter.
Sec. 8 Problems i11 lnfim'te Domains 289
Thus, we set
(8.18) U.,(s, y) = ; (2) t Joc u(x, y) sin sx dx
0
(s > 0)
Then, differentiating (8.18) twice with respect to y and using the differential
equation, we obtain
a-::,2 u/ = - (2)-t
- Joc uxix, y) sin sx dx
oy n 0
This leads to the differential equation for U,,

~1u
(8.19) ~
~ 2
s -
s 2u s = o
oy
under the usual assumption that u and ux vanish as x ~ oo. From the boundary
conditions in (8.17), we have
(8.20) U,(s, 0) = F,(s), U,(s, b) = 0

where F,(s) denotes the Fourier sine transform of f(s).
Now equation (8.19) has the general solution
(8.21)
which satisfies the second condition in (8.20) if we choose
and
for arbitrary function C(s). Thus, (8.21) can be written as
(8.22) U.(s, y) = C(s) sinh s(b - y)
Using the first condition in (8.20), we readily see that
C(s) = F.(s)
sinh sb
and hence (8.22) becomes
(8.23) U (s }") = F (s) sinh s(b -~ y)

' ' s sinh sb
By inverting (8.23), the solution of the problem (8.17) is therefore given by
(8.24) u ( x, y ) = -2
n
Ix - .s(b- - y) sm
sinh
smh sb
. sx f"' f( c;") sm. .,,s;: d;:.,, ds
0 0
If the boundary conditions in (8.17) are such that the nonhomogeneous

part is prescribed on the side x = 0, the solution can be obtained by the method
described m Section 5 of Chapter 6. For example, suppose we wish to solve the

problem
/:,,u = 0 (0 < x < 00, 0 < y < b)
(8.25) u(x, 0) = u(x, b) = 0 (x ~ 0)
u(O, y) = g(y) (0 ~ y ~ b)
We observe that the method of separation of variables u(x, y) = X(x) Y(y)

leads to an eigenvalue problem for the function Y, whose eigenvalues are
).
11
= n 2 n 2 /b 2 ; the corresponding eigenfunctions are Y,, = sin(nn/b)y, n =
I, 2, .... Accordingly, we assume a solution of the problem in the form
(8.26) u(x, y) = I
11= l
B,,(x) sin
11
b
~y
where
(8.27) Bn(x) = b2 Jb u(x, . nn
y) sm b y dy
0
Then, differentiating (8.27) twice with respect to x and using the differential
equation together with the homogeneous boundary conditions in (8.25), we
obtain the equation
(8.28)
for the functions 8 11 From the boundary conditions in (8.25) we have
(8.29) B,lO) = b
2 Jb0 g(y) S111. -b-
llTi y dy = b,, ( 11 = 1, 2, ... )
which are the Fourier sine coefficients of g on (0, b). Now equation (8.28)
has the general solution
(8.30)
Jn order that (8.26) is bounded as x tends to oo. B,i(x) must be bounded; hence,
we choose C 2 = 0 in (8.30). By the condition (8.29) we thus see that C 1 = b11
Therefore, the solution of the problem (8.19) is
(8.31) u(x. y) = f
n= 1
blle-(nrrb)x sin /Jn
b
y
where the constants b11 are given by (8.29).

For example, when g( y) = sin y and b = n in (8. 19). we have the solution
u(x, y) = e-x sin y
In the case of the Neumann problem
flu= 0 ( - 00 < x < 00, )' > 0)

(8.32)
~~ (x, 0) = f(x) (-oo<x<oo)
CJ
Sec. 8 Problems ill /11fi11ite Domains 291
in the half-plane, it is possible by a simple device to reduce the problem (8.32)

to a Dirichlet problem and thereby obtain a solution. In fact, suppose u is a
solution of (8.32). Let r(x, y) = cu(x, y)/cy. Then, clearly,
L'i.v(x, y) = L'i.uy(x, y)
i)
- iiu(x, y) = 0 (y > 0)
oy
and
i"u
r(x, 0) = , (x, 0) = f (x) (-oo<x<oo)
oy
so that v solves a Dirichlet problem in the half-plane, y > 0. Therefore, by
(8.12), we have
(8.33) i(x y) = Y
, n
J"'- oc (x - 02
.rm d(
+
-
y2
(y > 0)
r
and from the definition it follows that
11(x, y) = z(x, 17) d17
)' IJ Joc rm d~
(8.34)
J n -w (x _--()2 + t/2 d17
I I"' f m J)' --
IJ" dr7 d~
nJ- oc (x - r;:)z + t/2
1 Joc f(() ln[(x - ~)2 + y2] d~

2n _x
The interchange in the order of integration in (8.34) is valid in view of the fact
that the integral (8.33) is uniformly convergent in y for I.Fl ::=:: e, as we observed
earlier in connection with (812).
Under the assumption that f is continuous and satisfies an order relation
lx'f(x)I < /ad for - oo < x < oo, where k > 1 and Mis a constant, it can be
verified that (8.34) is a solution of the Neumann problem (8.32), which is unique
up to an additive constant.
Exercises 7.7
1. Verify that
1 J'OC df,
n - ~ (x _--6:z + y2 = 1
( - 00 < x < oo, 0 < y < 00)
2. Show that if f is an odd function, formula (8.12) can be written as

1 1
u(x, y) = :>' {'"' ( - - ) f (f,) df,
n Jo (x _ f,)2 + Yi (x + f,)2 + y2
~- ..
which solves the Dirichlet problem f..11 = 0 in the first quadrant x > 0, y > O
with the boundary conditions: 11(0, y) = 0, y > O; u(x, 0) = /(x), x ;::: 0. In
particular, when /(x) = I for x > 0 and /(x) = - 1 for x < 0, show that
u = (2/n) arclan(x/ y).
3. Deduce from formula (8.12) the solution of the problem
f..11 = 0 (x > 0, y > 0)

u,(O, y) = 0 (y;::: 0)
u(x, 0) = f(x) (x ;::: 0)
where f is bounded and continuous for x ;::: 0.

4. By interchanging the role of x and yin (8.12), a solution formula for the Dirichlet
problem 6.u = 0 for the right half-plane x > 0, - oo < y < oo, with boundary
condition u(O, y) = f(y), - oo < y < oc, is obtained. As a special case of this
formula, deduce the solution of the problem
6.11 = 0 (x > 0, y > 0)

11(0, y) = f(y) (y ;::: 0)
u(x, 0) = 0 (x ;::: 0)
11 bounded for x ;::: 0, y ;::: 0.

5. Find the solution of the Dirichlet problem
6.u = 0 (x > 0, y > 0)

u(x, 0) = f(x) (x ;::: 0)
u(O, y) = g(y) ( )' ;::: 0)
by assuming it is bounded for x > 0, y > 0.

f..11 = 0 (0 < x < oc, 0 < y < b)

11(0, y) = 0 (0 ::; y ::; b)
11,.(x, 0) = 0, 11(x, b) = f(x) (x ;::: 0)
u bounded as x -> oc.

7. Sol\e the problem
flu= 0 (x > 0, 0 < y < n)
11..(0, )') = g(y) (0 ::; )' ::; 7t)
uJx, 0) = 11(x, n) = 0 (x 2 0)
11 bounded as x -> er:.
6.u = 0 (x > 0,0 < y < 7t)
11x(O, y) = 0 (0 ::; y :5 n)
u(x, 0) = 0, 11,.(x, 1:) = /(x) (x > 0)
11 bounded as x -> 'X;.

Sec. 8 Problems i11 Jnfi11ite Domains 293
9. Find the solution of the problem for sector
u,, + -I 11, +
l
1199 = 0 (r < I, 0 < e< b)
r r2
uo(r, 0) = 0, u(r, b) = /(r) (0 ::; r ::; I)
u(I, OJ = o (0 ::; 8 ::; b)
by introducing the new variable c; = ln(l/r), 17 = e. (See Exercises 7.2.)

!i11 - h11 = 0 (x > 0, 0 < y < n)
u(O, y) = 0 (0 ::; y ::; n)
11(x, OJ = 0, u(x, rr) = /(x) (x 2: 0)
11 bounded as x ___, oo; h a positive constant.

/i11 - h11 = 0 (x > 0, 0 < y < n)
11(0, y) = g(y) (0 ::; y ::; n)
11(:c OJ = 11(x, ;r) = 0 (x 2: 0)
11 bounded as x -> ex ; h a positive constant.

12. Deduce from (8.34) the solution formula
u(x, y) = -
I
2n
f cr_ (>.: -
In
(x
cJ2
-
+ c;J 2 +
+- y z
y2
f(c;) de;
0
for the problem !iu = 0, x > 0, y > O; 11(0, y) = 0, y 2: O; u,.(x. 0) = f(x),

x 2: 0.
13. Deduce from formula (8.34) a solution of the Neumann problem
!iu = 0 (x > 0, y > 0)

11x(O. y) = 0 ( y 2: 0)
11,.(x, 0) = /(x) (x 2: OJ
14. Find a solution of the mixed problem
!iu = 0 (x > 0, y > 0)

u(x, OJ = 0 (x 2: 0)
ux(O, y) = g(y) (y 2: 0)
11 bounded for x 2: 0, y 2: 0.
15. Solve the Neumann problem
!iu = 0 (x > 0, y > 0)

u/x, 0) = 0 (x 2: 0)
ux(O, y) = g(y) (y 2: 0)
16. Find a solution of the Neumann problem

f..11 = 0 (x > 0. J > 0)
11)0, y) = g(y) (y 2: 0)
uy(x, 0) = j(x) (x 2: 0)
11 bounded for x 2: 0, y 2: 0.
9. Fundamental Solution and Green's Function
Let ((, IJ) and (x, y) denote an arbitrary fixed point and a variable point,
respectively, in the xy-plane. We recall that in polar coordinates with pole at
(~, 17), Laplace's equation takes the form
iJ2u I cu I cu2
(9.1) flu=
a,.2 +--+---=0
r or r 2 c0 2
where
r = ''(x - 0 2 + ()' - 11) 2
and
1f 11 is a solution of (9.1) that depends only on r, then 11 satisfies the ordinary

(9.2) ~~i~ + ; ~~ = ~ f,. (r ~~) = 0

Thus, u must be of the form
(9.3) u(r. 8) = C 1 In r + C2
where C 1 and C 2 are arbitrary constants. If we choose C 1 l/2n and C 2 = 0,

then the particular solution
1
(9.4) U(r, 8) = - In r
2n
is called the fundamental solution of the two-dimensional Laplace's equation.
The fundamental solution has the property of being harmonic throughout the
xy-plane except at the pole ((, 11), where it becomes logarithmically infinite.
It is the simplest conceivable solution having this type of singularity at a point.
The essential role played by the fundamental solution (9.4) is that it leads to
an integral formula that expresses the value of any harmonic function u inside
a domain in terms of the values of u and its normal derivative cu/on on the
boundary of the domain. This makes it possible for us to obtain explicit
representations for the solutions of the Dirichlet problem and the Neumann
Sec. 9 F1111da111e11tal So/11tio11 and Gree11 's F1111ctio11 295
problem in terms of kernel functions known as Green's function and Neumann"s

function, respectively.
Our basic tool is the Green's second identity
(9.5) v cu) ds
~
CIJ
over a domain D with a smooth boundary C. The functions u and i together

with their first- and second-order derivatives are assumed continuous in D.
We shall take 1 in (9.5) to be the fundamental solution U with the pole((, 17)
in D. Since U is singular at the pole, we delete from D a small disk about
(~, 17) with radius r: and boundary C 0 (Fig. 7.5).
FIG. 7-5 The f1111dame11ta/ solution.
Applying (9.5) to the punctured domain D' bounded by C and C 0 , and noting
that !iU 0 throughout D', we find
(9.6) - IrD (In r) !iu dx dy = J fu <.n~(} (Jn r) -

c+c 0
In r ~ 11 l
cnl
ds
where we have dropped the common factor 1/2n. This relation holds for all
values of c: no matter how small. We shall show that the integral along C 0
yields the value -2nu((, 17) when e is allowed to approach zero.
On C 0 we note that r = 8 = const and the outward normal vector is opposite
in direction to the radius vector. Thus,
a
~(In r)
I d
= - - In r I
1
=
1
Oil I r =' dr :r = f. 8
so, by introducing polar coordinates, we see that

2
Jim r II :) (In r) ds = Jim J " u(~ + 8 COS 0, 11 + 8 sin 0)
f,-O JCo C/1 ,~o 0
(9.7)
-2nu(~, 17)
Since the first derivatives of u are continuous in D, there is a constant 1\1 such'
that (:u/cn ~ ,\1 on C 0 Therefore,
(9.8) 'J,
1 Co
CU
~ In r ds,
011 I
j
~ M J
2
o
" Jin cl e dO ~ 2nMe Jin 1:;J
which tends to zero with e. Thus, letting i; approach zero in (9.6) and usmg
(9.7) and (9.8), we obtain the integral formula
(9.9) u(~, 11) = .

2n
1
J Lf ~-
c
u
on
(In r) - In r ~]
cnJ
ds +
2n
ff
D
(Liu) In r dx dy
This expresses the value of u at any point in D in terms of the values of u and
ou/cn on the boundary C and of Liu in D. Jn particular, when 11 = I, we obtain
an important property of the fundamental solution, namely,
I ' c-
(9.10) ~ (In r) ds = I
2n Jc c11
Formula (9.9) does not immediately yield a representation for the solution
of the Dirichlet problem or the Neumann problem, since it involves the values
of both 11 and its normal derivative ('ujcn on C. We shall therefore convert
it into an integral formula from which it will be possible to eliminate u or
ou/cn.
Let g = g(x, y: ( '1) be a function of (x, y) that depends on the pole ( ~. 11)
such that g is harmonic throughout the domain D. Applying (9.5) for the
function u and g. we then have
(9.11) r (u ('g
Jc h1
- g lll) ds + Jr g Liu dx d_r
(n.
0
0
Combining this with (9.9). we thus obtain
(9.12) u(~, 17) = Jc ('u en~.(}_ - G ~!')

en ds + ff
D
G Liu dx dy
where we have \Hitten G = G(x. y; (, 17) with
(9.13) G(x, y; (, 17) = ~n In r + g(x, y; ~' 17)

Sec. 9 F1111dame11tal Solution and Green's Function 297
Now. if the harmonic function g can be determined in such a way that G or

?G.!c11 vanishes on C. the term involving ?11/c11 or u 111 (9.12) disappears, and the
resulting formula yields a representation for the solution of the corresponding
Dirichlet or Neumann problem.
We focus our attention here on the problem of determining the function g
such that G vanishes on the boundary C of the domain D. It is clear that if g is
a solution of the boundary value problem
(9.14)
-I
g(x, y; , 11) In r on C
2n
for each point(~. 17) in D, then the function G defined by (9.13) vamshes when
(x. y) lies on C. Thus, we obtain from (9.12) the specific representation formula
(9.15) 11((, 11) = r u(x, J) ~G (x, y: (, Y/) ds + ff G(x, y; (, 11) t.u dx dy

C C/1 D
for the solution 11 of the Dirichlet problem
(9.16) t.11 = q(x. y) in D. 11 = f(x, y) on C
in terms of the function (9.13) and its normal derivative on the boundary.
Let us state our result as a theorem.
THEOREM 9.1. If 11 denotes the solution of the problem (9.16). then u has the
representation
(9.17) u(x, r) = r /((, 1/) (~G ((, 17:

C C/l
X, J') ds
+ fJ
/)
q((, 11)G((, Y/: x. J) d( dYJ
frhere G is the .fi111ctio11 defined by (9.13) and (9.14) Hith the role of (x, y) and
(, 17) interchanged.
The function G defined by (9.13) and (9.14) is called the Green's function.
It is clear that the Green's function, whenever it exists for a given domain, is
uniquely determined. This follows from the uniqueness of the function gas a
solution of (9.14), which is a Dirichlet problem. It might also be seen from the
fact that if G 1 and G 2 are two distinct Green's functions for a given domain,
then their difference G 1 - G 2 is a harmonic function throughout that domain
which vanishes on the boundary, and so, by the maximum principle, must
vanish identically.
Let us note the following important properties that characterize the Green's
function G:
(i) For each (~. 11) in D, G satisfies the Laplace's equation
for all (x, y) in D
except at (r;, 17).

(ii) G(x, y; , 17) = 0 when (x, y) lies on the boundary C.
(9.18)
(iii) G has a logarithmic singularity at(~, 17) such that
cG
(iv)
l ;- ds = 1
c en
G is symmetric with respect to the points (x, y) and(~, 17); that is,
G(.;, 17; x, y) = G(x, y; .;, 17)
The first two of these properties clearly follow from the definition of the Green's
function. The third property is a consequence of (9.10) and (2.11 ), g being
harmonic in D. The last of the properties can be proved by setting
u(x, y) = G(x, y: (, 17), v(x, y) = G(x, y; (, T)
in Green's second identity (9.5) over the domain D' bounded by C, C 1 , and C 2 ,
where C 1 and C 2 are small circles of radius e about the poles(~, 17) and ((, T),
Figure 7.6. We note that since !:lu = 0 and .i1v = 0 in D', and u and v vanish
on C, (9.5) yields
Jc, (.u on~~ - v ~u) ds Jc, (u, on~ - 011

(9.19) + v ) ds = O
on 011.
0 c I
l)'
0 )
c,
----------- y
0
FIG. 7-6 Proof of symmetry of the Green's function.

Sec. 10 Examples of Green's Functions 299
By using the definition (9.13) together with (2.11 ), and by a calculation quite
like our derivation of (9.9), we see that
lim
,~o
J (u en~v ~)
c1
- v
011
ds = v(, 17)
(9.20)
lim
.~o
J (u ~~ - ;)
c2 on
v
on
ds = -u((, r)
Thus, from (9.19), we have u((, r) = v(, 17), or G((, r; , 17) G(, 17; (, r).
10. Examples of Green's Functions
In this section we shall present examples of Green's functions for the Laplace's
equation for special domains in the xy-plane. We determine first the Green's
function for the interior D of the disk (Fig. 7.7),
FIG. 7-7 Inverse points with respect to a circle.
(10.1) x2 + y 2 :::;; a2
Let P: (, 17) denote the pole of the Green's function G = G(x, y; ~. ~1) in D,
with Q: (x, y) being a variable point, and let
;- - ----- ---
r = v (x - )2 + (y - 17)2
denote the distance of the point Q from the pole P. We wish to determine a
function g = g(x, y; , ri), which is harmonic throughout D and which assumes
the value -(l/2rr) In r when x 2 + y 2 = a 1 . Ir((, 17) = (0, 0), then clearly
g = ( - l/2rr) In a fulfills the requirement. Hence, by (9.13), the Green's
function for the disk ( 10.1) with pole at (0, 0) is given by
1 I
G(x,y:0,0)= Jnr- -Ina
2n 2rr
(10.2)
I r
In
2rr a
Suppose now that((. 17) =I= (0, 0) and denote by p = " c +

/ v? 2
11 , the distance
of the pole P from the center of the disk. Let
a2 .. az )
R: ( ,c- IJ
p- pl
denote the inverse of the point P with respect to the circle C: x 1 + y 2 = a 2

(that is, OP OR = a 1 with P and R lying on the same line from 0). When the
point Q is located on the circle C, we see that the triangles 0 PQ and OQ R are
similar. This is true because the two triangles have an angle at 0 in common,
and the corresponding sides forming that angle are proportional; that is,
OP OQ
(OQ a)
OQ OR
--
in view of the relation OP OR = a 1 . Hence,
PQ OP a
QR a OR
from which we obtain
OP - a
(10.3) PQ -- . QR QR
a OR
Thus,
(10.4) -
a1 ..
zs
)2 +
(_r- a1
zl7
)zJ 1;2
p . p
whenever x 2 + y 2 = a 1
Now consider the function
1 p rL(' x - p2a2 c;y) 2

+
. a2 )2] 1;2
(10.5) g(x, y; , 17) = - 2~ In a (y - -- IJ
pl
It is easily verified by direct differentiation that this function satisfies the Lap-
lace's equation for all (x, y) in D. Moreover, in view of (10.4), the function
assumes the value -(l/2rr) In r whenever (x, y) lies on the boundary C of D.
Hence, (10.5) is the solution of the Dirichlet problem (9.14) for the disk ( 10.1 ).
Sec. JO Examples of Green's Functions 301
By (9.13) the Green's function for the disk (10.1) is therefore given by
I 1 p 1 ar
(10.6) G(x, y; ~' 11) = - In r - -- In r* = In -
2n 2n a 2n pr*
where we have written
(10.7) r*
a2
= [( x--~ )2 +(. a2 )21112
y--17
p2 p2
Since we have found the Green's function, the solution of the non homogeneous
Dirichlet problem (9.16) for any disk is now immediate, by Theorem 9.1. In
particular, when q = 0, we have
(10.8) u(x, y) = Jc f(~, 11) ~(;_

on
((, 11; x, y) ds
where we have interchanged the role of (x, y) and U:, 11) in (10.6). Let us show
that this agrees with the Poisson's integral formula (6.6), which was previously
obtained in Section 6. We have to calculate the normal derivative of the Green's
function (10.6) for ~ 2 + f/ 2 = a 2 For this purpose, it is convenient to introduce
polar coordinates to describe the points (x, y) and(~, IJ). Thus, let x = p cos 0,
y = p sin 0, and = v cos </>, 11 = v sin </>. Then
,.2 = ( _ x)2 + (IJ _ y)2
= v
2
2pv cos( - 0) + p2
r
-
(10.9) ,.*2
= (~ - ~: x + ( 1J - ~-: yr
a2v a4
= v2 - 2 - cos( - 0) +
p p2
Since the outward normal derivative of a function on a circle coincides with the
derivative in the radial direction, it follows from (10.6) and (10.9) that
ac
~ - (v, <P; p, 0)
en
=
2n ov
[a
1 ~-- (In r) - -~ (In r *)
ov
a l
= 1-
4n ov
[a~ (In ,.2) -
ov
a (In ,.*2)
~ l
(IO.IO)
= 1 [v - p cos(_~~) _ 2
v -_(a /p) cos</> - 0)1
2n ,.2 ,.*2
This yields
oG
- (a,; p, 0)
1
= -- a2
-~---
- p2
(I0.11)
an 2na a 2 - 2ap cos( - 0) + p2
upon setting r = a and noting that r* = (ar)/p. Substituting this result in

(10.8) and noting that ds = a d1J, we indeed obtain the Poisson's integral
formula (6.6).
We take up next the determination of the Green's function for Laplace's
operator for the half-plane y > 0 (fig. 7.8). Again let P: ((, 17) denote the
{) ~ y \ I
f': ii; O) I
I l
+--~----------------~-
!
:
I
I
I :
I :
I I
/:!, I~ i()
FIG. 7-8 Gree11's ji111ctio11 i11 the half-plane y > 0.
pole of the Green's function in the half-plane y > 0 and let Q: (x, y), y > 0,
be any point whose distance from the pole is denoted by r. Consider the re-
flected image R: ( (, -1)) of the point P with respect to the x-axis. It is clear
that when Q is located on the boundary y = 0 of the half-plane y ;;::: 0, Q is
equidistant from P and R. Thus. we take
(10.12) ~ I/ )
g ( X, .J": (, = - I I 11 1*
2:r
where
--
(10.13) r * = ,:(x - c)2 + (y + 17)2
Clearly, this function is harmonic for all (x. y) in the half-plane y ;;::: 0 and
reduces to -(l/2n:) In r on y = 0. Therefore. the Green's function for the
half-plane y ;;::: 0 is given by
I l l r
G(x r: C. IJ) =
(10.14)
' - . _, 2n:
Jn r -
2n:
Jn r* = - In -
2n: ,.*
Thus, we can formally state the following theorem, which corresponds to
Theorem 9.1.
Sec. JO Examples of Green's Functions 303
THEOREM 10.1. if u denotes the solution of the Dirichlet problem
Liu = q(x, r) ( - 00 < x < 00' y > 0)

(10.15)
u(x, 0) = f (x) (-oo < x < oo)
then 11 has the representation
u(x, y) = - Joc /(0 ~G (~, O; X, y) d~

- oc OYf
+ J"' f"'
0 - cc
q(~, 17)G((, 17; x, y) d~ d17
(10.16)
y f"' f(() d~
rr _"'(~ _ x)2 + y2
irhere G is the Green's function (10.14).
When q = 0, (I 0.16) leads back to the Poisson's integral formula (8.12)

for the half-plane y > 0.
Exercises 7 .8
1. Establish the results given in (9.20).

2. Verify that the function g defined in (10.5) satisfies Laplace's equation in the
variables x and y.
3. Verify directly that the Green's function ( 10.6) is symmetric with respect to the
point (x, y) and the pole(~. 17).
4. Shm\ that by introducing the polar coordinates x = v cos 0, y = i sin 0 and
~ p cos (/J. 11 = p sin q1. the Green's function (10.6) takes the form
~
G(l' 0 p r/J) =
1
ln
a 2 (1' 2 -
-
2l'p cos<O -
-
) +- p 2..)
' ' ' 4n (a 4 - 2a 2 vp cos(O - <h) + l' 2 p 2 )
and thus verify that G = 0 when i = a.
5. Determine the harmonic function ( 10.5) by actually solving the Dirichlet problem
22 q 1 c..q 1 22 g
. '2 + + 2
= 0 (i < a)
C1' l' l l' 1' (;8 2
1
g(a. e; p, f) = ln[a 2 - 2apcos(0 - di)+ p 2 ] (1 = a)
4n
for fixed p and . Hint: Assume a solution g analogous to (5.8) and use the
series (7 .9) for the boundary data.
6. Let G(v, 8; p, ) and G(v, 8; p, -) be the Green's functions for a disk of

radius a with poles at the points (p, ) and (p, - ), respectively. Show that
G*(v, 8; p, cj;) = G(v, 8; p, cj;) - G(v, 8; p, -)
is the Green's function for the semicircular domain r s: a, 0 s: 8 s: 7r. What is
the harmonic function g(i', 8; p, )in this case?
7. By using the Green's function G* obtained in Problem 6, and by a similar pro-
cedure, obtain the Green's function for the quarter-disk 0 S: r s: a, 0 :::; 8 s: n/2.
8. Find the Green's function for the sector 0 s: r S: a, 0 S: 8 :::; n/3. Hint: Let
v' = r 3 , 8' = 38, and use the results of Problem 6.
9. Prove that the Green's function G(p, ; \', 8) satisfies the relation
a2 -
-
1'- 2 =
J2n Jora G(p, ; v, 8)p dp d
4
Hint.- Take/= 0 and q = I in (9.16) and obtain a solution by inspection; then
use (9.17).
10. Determine the harmonic function (I 0.12) by actually solving the Dirichlet
problem
/!,.g = 0 ( y > 0, - 00 < x < 00)
g(x, y; ~, 17) - _I In [(x - c;)2 + 1/2] (y = 0)

4n
for each fixed (c;, 17), 17 > 0.

11. By a procedure similar to that used in Problem 6, find the Green's function for
the quadrant x ~ 0, y ~ 0.
11. Neumann's Function and Examples
In Section 9 of the present chapter we obtained from the basic formula (9.12)
an explicit representation for the solution of the Dirichlet problem (9. I 6) in
terms of the normal derivative of the Green's function (9.13). In this section
we shall derive from (9.12) a similar representation formula for the solution of
the nonhomogeneous Neumann problem
(11.1) L'1u = q(x, y) in D,

au- = f (x, y) on C
011
Let us rewrite (9.12) in different notation as
(l l.2) u(~, ry) = f c

(u iJH - H
an
qu)
an.
ds + JJL'1uH dx dy
,1
where
I
(11.3) H(x, y; ~, ry) - - Jn r + h(x, y; ~' ry)
2n
Sec. 1 I Seumann's Function and Examples 305
with h being an arbitrary harmonic function throughout D. In order for (11.2)

to represent a solution of the problem (I I. I), it seemed natural to choose the
function h so that cH /en vanishes on C, thus eliminating from (l 1.2) the term
that involves the values of u on the boundary C. This would mean that h is a
solution of the special Neumann problem
!ih = 0 in D
ch I o (In r) on C
011 2n on
But this problem cannot have a solution because, by (9. JO),
oh
f
c on
~ ds = -l
which violates the necessary condition (2.12) for solvability of the Neumann
problem.
Let us determine h so that cH /en = const on C; say, cH /en = b # 0.
Then, from (11.3), we have
0
cH ds = J (In r) ds + J oh ds
Jc on 2n c en c 011
which, by (9. l 0) and (2.11 ), yields
b Jc ds
This implies that
(11.4) b =I
L
where L denotes the perimeter of the domain D. This means that we must
determine h as a solution of the Neumann problem
!ih = 0 in D
(11.5)
ch I <' (Jn r) on C
011 L 2n l 11
Notice that here the condition (2.12) is satisfied. So, if the function h in (11.3)
is a solution of (11.5), then substitution of (11.3) in (11.2) yields the specific
representation
" Du
u(~, 11)
Jc H(x, y; s, 17) ds ~
011
+ JJH(x. ~. 17) !iu dx dy + l J u ds

(11.6)
y;
D C
for a solution of the Neumann problem ( 11. I). The last term on the right of
(I 1.6) is simply an additive constant, representing the average value ii of the
solution. Thus, we have the following representation theorem for (11. I).
THEOREl\I 11.1. Let u denote a solution of the nonhomogeneous Neumann

problem (I I.I) for H hich the compatibility condition (2.13) holds. Then u has the
representation
u(x, y) = - I H((, 17; x, y)f((, I)) ds
(11.7)
+ JJq(~, IJ)H((, I); x, y) d( di] + 11
\\'here H is defined by (11.3) and (11.5) iiith the role of (x, y) and((, IJ) inter-
changed, and where ii is an arbitrary constant equal to the araage rnlue of 11.
The function H defined by (11.3) and (I 1.5) is called a Neumann's function

for the Laplace's equation. Unlike the Green's function, a Neumann's function
is unique up to an additive constant. This is due to the fact that a solution of
(11.5) is uniquely determined up to an additive constant. However, if we choose
the additive constant in such a way that the normalization condition
(11.8)
Jc H (x, y; (, 17) ds = 0
holds, then H becomes uniquely determined. In this discussion we shall assume
that our Neumann's function satisfies (11.8).
It is of interest to note the following characteristic properties of the Neumann's
function:
(i) H satisfies Laplace's equation for all (x, y) in D, except at (<!, 11).
(ii) oH/cn = 1/L on C, where Lis the perimeter of D.
(iii) H has a logarithmic singularity at (<!, 17) such that
(11.9)
l
oH
~ds = 1
C C/1
(iv) H is symmetric with respect to the points (x, y) and (<!, 17); that is,
H(x, y; .;, 11) = H(<!, 17; x, y).
The symmetric property (iv) can be established with the help of (I 1.8) in
exactly the same manner we established the symmetry of the Green's function,
and is therefore left as an exercise (Problem I).
Let us now consider the construction of the Neumann's function (11.3) for
the interior of the disk x 2 + y 2 ~ a 2 . Using the same notation as in (10.6)
we set
1 I
(11.10) H(x, y; ~, 17) = -- In r + - In r* + A(~, 17)
2n 2n
Sec. 11 Neumann's Function and Examples 307
where A is a harmonic function. It is clear that the function (11.10) is harmonic

in the interior of the disk for all points (x, y) except at the pole (e, 17). More-
over, from the calculation in (10.10), we readily verify that when (x, y) lies on
the circumference of the disk, we have
8H
an (x, y; e, 11 ) = 2n~1 = i1
Thus, the first three properties listed for the Neumann's function are fulfilled.
To ensure symmetry of the Neumann's function, we have introduced the har-
monic function A(e, 17) in (11.10). The function A will be determined by
requiring (11.10) to satisfy the normalization condition (11.8). In this way, we
shall also have ensured uniqueness of our Neumann's function.
Thus, applying (11.8) to (11.10) and noting that on the circumference C of
the disk, r = pr*/a, we have
f c
H(x, y; e, 17) ds = 2n
_!__ f c
(In rr* + 2nA) ds
(11.11)
= _l_
2n
f c
(in I! r* 2 + 2nA) ds
a
=0
Now the integrand in the last integral of (11.11) is a harmonic function of
(x, y) throughout the disk. Hence, by the mean-value property of harmonic
function (Theorem 6.1 ), the integral in (I l.11) is equal to the value of the
integrand at the center (x, y) = (0, 0) of the disk. Thus, we find
p a4
In - 4 (e2 + 17 2 ) + 2nA = O
ap
or
1 p
(11.12) A(e, 11) = - In - 3
2n a
Therefore, our Neumann's function for the disk x 2 + y2 :S; a2 is
H(x, y; e, 17) = _!__ [1n r

2n
+ In r* + In P3 ]
a
(11.13)
= _!__In rr*p
2n a3
where r = [(x - e) 2 + (y - 17) 2 ] 1 12 ,
r* =
a2
[( x - --p2 e)2 + (y a2
- p2 11
)2] 112
and p = (( 2 + 17 2 ) 112 . Thus, a solution of the Neumann problem (I I.I) for a

disk is immediately known by Theorem I I. I. In the particular case where
q = 0, it is easily shown that, in polar coordinates,
(11.14) u(x, y) = - L H((,, 17; x, y)f((,, 17) ds
where His given by (11. I 3) with the role of (x, y) and ( (, 17) interchanged, agrees
with formula (7.10) previously obtained.
Neumann's function for the half-plane y > 0 is given by
(11.15) H(x, y; <:,, 17) = ~rr [In r +In r*J = 2.~ In rr*
where we have used the same notation as in (I 0.14). Clearly, this function is
harmonic throughout the upper half-plane except at the pole((, 17). Moreover,
oH/oy = 0, as to be expected, since the boundary here is of infinite length.
The ful)ction (11.15) is unique up to an additive constant. Thus, a solution
of the nonhomogeneous Neumann problem
Au = q(x, y) ( - CX) < x < oo, y > 0)

(11.16)
cu = .f(x) (-oo < x < oo)
Dy
has the representation
1
u(x, y) =
2rr
Jco_ oo
f(() ln[(x - (,) 2 + y 2 ] d(
(11.17)
+ 1 f"' f"' ln(rr*)q((, 17) d( d17 + C
2rr 0 _ oc
where C is an arbitrary constant. When q = 0, this formula reduces to (8.34).
Exercises 7.9
1. Show that the Neumann's function (11.3) satisfying the normalization condition
(11.8) is symmetric with respect to the points (x, y) and (~, 17).
2. Verify that the function (I I. I 0) satisfies the relation
2H (X, y; \;,I/
-
) = -
1
an 2na
when x 2 + y 2 = a2
3. Verify directly that the Neumann's function (I 1.13) for a disk is symmetric with
respect to (x, y) and (,;, IJ).
Sec. 11 .Ve11111ann 's Function and Examples
309
4. Introduce the polar coordinates x = i cos 8. y = 1 sin () and .; = p cos,

11= p sin , and shO\\ that the formula ( l L 14) agrees with (7.10) except for an
additive constant.
5. Show that the Neumann's function (11.13) satisfies the relation
CJ-{ 1
- [K(a, p,8 - </>) - l]
ca 2na
where K is the Poisson's kernel given in (6.7).

6. Find a Neumann's function for the quadrant x > 0, y > 0.
Son1e Selected References
1. Berg, P., and J. L. McGregor. 1966. Elementary Partial Differential Equations.

San Francisco: Holden-Day, Inc.
2. Boyce, W. E., and R. DiPrima. 1969. Elementary Differential Equations and
Boundary Value Problems. New York: John Wiley & Sons, Inc.
3. Churchill, R. 1963. Fourier Series and Boundary Value Problems. New York:
McGraw-Hill Book Co., Inc.
4. Dennemeyer, R. 1968. Introduction to Partial Differential Equations and Boundary
Value Problems. New York: McGraw-Hill Book Co., Inc.
5. Dettman, J. 1969. Mathematical Methods in Physics and Engineering. New
York: McGraw-Hill Book Co., Inc.
6. Duff, G. F. D., and D. Naylor. 1966. Differential Equations of Applied Math-
ematics. New York: John Wiley & Sons, Inc.
7. Greenspan, D. 1961. Introduction to Partial Differential Equations. New York:
8. Hildebrand, F. 1962. Adrnnced Calculus for Applications. Englewood Cliffs,
N .J.: Prentice-Hall, Inc.
9. Petrovskii, I. G. 1967. Partial Differential Equations. Philadelphia: W. B.
Saunders Co.
10. Sagan, H. 1961. Boundary and Eigenvalue Problems in Mathematical Physics.
New York: John Wiley & Sons, Inc.
11. Sneddon, I. 1959. Elements of Partial Differential Equations. New York:
12. Tikhonov, A., and A. Samarskii. 1955. Equations in Mathematical Physics.
New York: Pergamon Press.
13. Tolstov, G. 1965. Fourier Series. Englewood Cliffs, N.J.: Prentice-Hall, Inc.
14. Weinberger, H. 1965. A First Course in Partial Differential Equations. Waltham,
Mass.: Blaisdell Publishing Co.
311
Solutions of the Exercises
Chapter 1
Exercises 1.1
1. (a) Continuous and piecewise smooth.
(b) Piecewise continuous and piecewise smooth.
2. The extension off is piecewise continuous if f is piecewise continuous on [ - L, L];
the extension is continuous if f is continuous on [ - L, L] and/( - L) = f (L).
The periodic extension off is piecewise smooth if f is piecewise smooth; the
extension is smooth if f is smooth on [ - L, L] and/(- L) = f(L).
5. No. 6. No.
Exercises 1.2
x1 + 2xy.
2. llx = ex cos \' + r~ lly = -ex sin y + x.
2x 2y
3. llx = , "r =
.\"2
+ y-J .\"2
+ y2
2y -2x
4. u, = , /{-'
-
(x + y)2 (x + )')2
I x x x
sec 2
?
5. Ux = - sec- lly ?
y y r y
313
314 Solutions of the Lxercisn
x r- )'
11. -
[{_\_\"
(x2 + r2 )312
12. ll,;x
13. "xx - e' sin y.
-2.YJ' x-' -- 2.\')'

14. ii_\")"= '
(IX_\.
(.12 + y2)2' (x2 + y2 )2 (x2 + y2)2
x2 y2
21. For(\. y) cfc (0. OJ,
x2 + v2
Exercises 1.3
d11 1-'
1. t In(! - t) -
dt 2(1 - I)
2. d11 1d1 [COS( sin I) + 1 sin t ]/1 [(In t) sin(sin I) - t l cos f.
3. du/dt 1. 4. du/dt = 2 tanh 2t.

5. du/d\ (I + 3x 2 ) In x (! +
6. d11ld\ 4x. 7. 11, -- '2r; lls = !Os.
8. fir= 0: u, = e'. 9. 11, 2r; lls = 0.
10.
2(r
2
+ s
1
)
+ 4r ln(r - .1): 11, =
- 2(r 2
+ s 1
)
-r 4sln(r s).
11,
r - s r - .1
12. Uu + Uq
x y y x
14. = - 11 8 : -- lie.
fix
r
II,
,.- fl_\
r
11,
' ,.-
0
15. u, u, cos 0 + "r sin O: 11 8 = - rnx sin 0 -i- my cos 0.
Exercises 1.4
TLX cos(i:i2)x -- 2 !>in(IT/2)x

1. <f/!1) =
2. q/(.\) = x(i:/2 - In 2).

3. </J'(x) = (1 -i- cos xJ'( I ,- x + sin x).
4. (2TLx)/(x 2 - 1) 2
. 5. q/(x) (3 sin 3 x - 2 sin x 2 J/x.
8. cp(x) = Jn(! + x).
315
Exercises t .5
8. For each a 2 ). Then

x 'ff x X_2n+ I
-- ' -- " = ~ ( - I l"
( - J)"+ I
r !+1-
0 11=0 2n+l
JO. ~ -
,, = 1 (211 - I )3
14. The series for 11 1 11x, u'-' are all uniformly convergent for 0 ::; x ::; rr, 0 <
f0 s I.
Exercises 1.6
1.
jx2 + 121
i ::;
a2 + - [
?
for all t, and Lx di
a2 -1- (2
i[
2a
I sin t 1
2. I ? I ::;
1
x- + r- x2 + t2
4. ie--' 1 cos ti s e-ai for 0 < a s x, 0 s I < oc and J;;' e-t dt = 1/a.
6. ForO <a::; x::; b,e- 11"'- 1 ::; e-r,b-i::; A/t 2 for some constant A.
7. Integrate under the integral sign from 0 to oo.
10. Forall x, \e- 12 cos xtf ::; e- 12 and J;;' e- 1 ' dt = v'rr/2.
Exercises 1.7
I. (a) du/ds = 6\il; (b) du/ds = v3 - (I+ \ 3)e.
1
du 32 4
2. -1- ::__(sin 2 + cos 2J.
ds '\
1
17 17
3. du/ds - 72/\ 1 17. 4. () = 45; d11/ds
5. du/ds = 194/\ 97 (max). 6. cu/en = 4.
7. cu/en = 2/r.
Exercises 1.8
I
1. 0. 2. 4
3. 0. 4. 5/42.
5. (a) 2n; (b) 2n.
Chapter 2
Exercises 2.1
I. (a) and (f) are second order, linear; (bl, (c), and (e) are second order, nonlinear;
(d) first order, linear.
..... _
316 Solutions of the Exercises
Exercises 2.2
1. (a) u = 3xy + y 2 /2 + f(x). (b) u = ln x + xeY + f(y).
(c) u = - y cos x + f(y). (d) 11 = xy 2 /2 + y tan x + f(x).
(e) 11 = yf(x) + g(x). (f) u = s~ g(I) dt + f(y).
2. (a) u = -ex-y + f(x) + g(y). (b) u = x 2y/2 + xy 2 + f(x) + g(y).
(c) u = yex + xf(y) + g(y). (d) u = y 4 /4 sin x + yf(x) + g(x).
<el u = s~ t<~l d~ g(11) d11 + h(x) H + k(y).
3. (a) u = f(x + 2y) + g(x - 2y). (b) u = f(x + y) + g(Sx + y).
(c) u = f(2x - y) + xg(2x - y). (d) u = f(x + y) + xg(x + y).
(e) u = f(2x + iy) + g(2x - iy).
(f) u = ![(2 + 3i)x + y] + g[(2 - 3i)x + y].
4. (a) 11 = 2(x - (I/ y)) + e-xyf(y). (b) u = - (cos x/ y) + y 2f(x).
(c) u = f(y)e 2 xy + g(y)e- lxy - 3x/4y 2.
(d) u = f(y)exy + g(y)e- 3 xy. (e) u = f(x)exy + g(x)yexy.
5. u = xeY + (l/x)f(y) + g(x). 6. u = x(x - 1) + sin y.
7. u = sinh x + cosy - 1. 8. u = I - cos x + y 2 + x 2y 2 /4.
3
9. u = x - y + eY - I + x sin -y
3
Exercises 2.3
3. u = 2e- 9 ' 14 sin(3x/2) - 4e- 25 ' 14 sin(5x/2).
4. u = ie- 0 +kJr cos x + ie-< 9 +klr cos 3x.
5. /1 = i cos(YJ +~~ 2 1) sin x - t cos(V'9+-?t) sin 3x.

sinh n(y - I) sin 7L\ 3 sinh 2n( y - I) sin 2n
6. u = - +
sinh n sinh 2n
Exercises 2.4
1. u = exf(x - y). 2. 11 = f(3x + 2y) + x 2 /4.
sin x - cos x
3. u = + y + 2 + e-xf(2x + y).
2
4. u = x2 - 6x + -}x 2 y - hY + ty + ~ex + 14 + e- 2 x13f(4x + 3y).

5. u = - le 2 ' sin 3y +
+ y). e 2 xf(x
7. 11 = ln x + f(y/x). 8. u = 2ex In x - xy + exf(xy).
9. u = x + f (x2 - Y2 ). 10. u = eY 1Xf(xy).
11. u = x + x3/y2 + (l/y2)f(x2 + y2).
12. u = x e-x/(x+)') + e-x/(x+J)f(x + y).
x + y
13. 11 = x ln x + xf(x 2 + y 2).

14. If A -.p 0, then u = e-Dx!AJ(Bx - Ay, Cy - Bz). If A = 0, but B # 0, then
u = e- 0 Y18f(Cy - Bz, x). If A = B = 0 and C # 0, then u = e-Dz!Cf(x, y).
317
15. (a) 11 = e-xf(2x - y, 3y + 2z ).

(b) 11 = e- 3 x 12f(x + 2y, 2y + z).
17. u = e1 x-y[l + cos(x - y)].
18. u = e-zx sin Sx + 2>' + e<J/ll<x+Yl + tx 2 - }x + -k.

2
19. 11 = - cos x + cos( sin x - y) - sin x + y.
20. II = x/2 + \/~\" y 312 - y/2.
21. II = e'"(4 - x2 - y2)1;2e-(4-x'-,.'J'f2.
22. u = ey'-(x'+y'J'1' sin(x2 + y2)112.
23. 11= e-x'f(x 2 + y 2 ), where f is any differentiable function such that /(a 2 ) = k.
24. <fi(x) = e'[(x 2 + ]) 1i 2 + k].
Exercises 2.5
1. u = e-if(x) + e-xg(y). 1. II = f(x + y) + exg(x - y).
3. 11 = f(x - iy) + g(x + iy). 4. 11 = f(x - y) + xg(x - y).
5. 11 = /(x) + e- 3 xg(2x + y). 6. r 2 11 = f(r +cl)+ g(r - ct).
7. 11 = f(x + iy) + g(x - iy) + h(x + y) + k(x - y).
8. 11 = f(x + y) + xg(x + y) + h(x - y) + xk(x - y).
IO. 11 = x 3 /6 - y 3 /24. 11. 11 = xy 4 /I08 + y 5 /810.
13. 11 = --![cos(2x + y) + sin(2x + y) ].

15. (a) 11 = f(xy) + g(x/ y) + } 3
ln x.
(b) u = (ljy)f(x) + (ljx)g(y) + xy/4.
(c) 11 = xf(x/y) + x"g(xjy) + [(x + y)/(l - 11)] ln x, 11 of. I;
11 = xf(xjy) + (x In x),q(xjy) + [(x + y)/2] ln 2 x, 11 = 1.
16. {a) 11 = f(x + y) + g(x - z).
{b) ll = f(x + )'. y + z) + e-xg(2x - y).
(c) 11 = f(x - y, y - z) + xg(x - y, y - z).
Exercises 2.6
9. Hyperbolic for x 2 + y 2 > I, parabolic for x 2 + y 2 = 1, and elliptic for x 2 +
y2 < l.
JO. Hyperbolic for x 2 - 4y > 0, parabolic for x 2 - 4y = 0, and elliptic for
x 2 - 4y < 0.
Exercises 2. 7
1. 48(11~, - 11ryry) + u{ + 4ury = 0. 2. u,, - llryry - llry = -17(~ + r/).
3. 7511~ry + // = -(~ - 2
ri) /25. 5. 2u,ry + ury + l = 0.
6. 411~, - u, - 2u,1 = 0. 7. u,, + 11,, = ~(17 - 3~).
8. 911,~ - II = 0. 9. 11,~ - 411 = 0.

IO. 4(11,~ + 11ryryl + u = 0. 11. 27(u~~ + uryry) - 'Ill~ = 0.
12. "cc +u,,.1 + (~ - 11)(u~ + ury) = 0.

13. + u,,.,) - u = 0.
6(11(;
14. J2(u~~ + 11 1111 ) + 7u( + V3 /lry + II = 0.

15. u,,
+ u,,.1 - eryu = sin(217 - ). 16. u= f(x + y) + g(2x - 3y).
17. = /[(! - 3i)x -- 5y] + g[( I+ 3i)x - Sy].
u
18. u = f(2x - y) + xg(2x - y).
19. u = /[(2 + i)x + y] + g[(2 - i)x + y].
Chapter 3
Exercises 3.1
2. ll'1r -c 2 wxx = - v" + c 2 vxx; w(x, 0) = f(x) - v(x, 0); w,(x, 0) = g(x) - v,(x, 0);
w(O, t) = 0, w(L, t) = 0.
3. Suppose u 1 and 11 2 are two solutions and set u = 11 1 - u2 .
Exercises 3.2
1. u(x, 0) = sin x; u,(x, 0) = - c cos x.
2. u(x, 0) = x 2 ; u,(x, 0) = 2cx.
3. u = sin 2x cos 4t + 1- cos x sin 2t.
4. /1 = x sin x cos 3t + 31 cos x sin 3t + t cos 2x sin 6t.
2 2
5. /I= - - I -+- x + t
+ ex sinh t.
[I + (x + t) 2 )[1 + (x - 1) 2 ]
6. u = e-x cash t + 1-[arctan(x + t) - arctan(x - t)].
7. u = cos(rr/2)x cos(rr/2)t + (I /a) sinh ax sinh at.
8. u = sin 3x cos 31 + 1- sin 2x sin 2t - sin x sin t.
9. u(l/8, 11/8) = O; u(l/4, I) = 1/2.
10. u(-rr/12, 7rr/12) = 1/2 - v3/4; u(-rr/2, 5rr/6) = v3;4.
11. u(O, 3/4) = 45/64; u(-1/2, 3/4) = 211/384; 11(9/8, 7/8) = 117/384.
e-1 + e-213
12. u( - n/6, Srr/6) = - - - - - + I.
2
e-114+e-1 I v2
11(3rr/8, 5n/8) = - - - ---- + - + - - .
2 2 4
Exercises 3.3
1. u = x 2 + t 2 + xt 2 . 2. u = xt(t - 1) - t 3 /_ + sin x cost.
3
3. 11 = ex sinh I + xt /6. 4. 11 = cos(x - t) + (ex/4)(cosh t - 1).
5. u = (ex/4)(cosh t - 1) + t[arctan(x + t) - arctan(x - t)] + t - sin t.
6. 11 = - x 3 - 3xt 2 + x 2 + t 2 - ~ sin x sin t + tt sin x.
Solutions of tlze Exercises
319
7. 11 = x - sin x cos t + sin x.

8. 11 = - x sin x cos t - t cos x sin t + x sin x.
9. u = :\-(3x 2 t + 13 ) - x 2 ln(l + t).
10. Write
r(x. I) = -1 J' rx+c(!-r) F(~, r) d!;, dr
2c o .x-c(z-r)
Then
v, = -1
2
j''
0
[F(x + c(t - r), r) + F(x - c(t - r), rJ] dr
z;x = --I
2c
J' 0
[F(x + c(t - r), r) - F(x - c(t - r), r)] dr
z; 11 = F(x, t) + ~ f [F.(x + c(t - r), r) - Fp(x - c(t - r), r)] dr
ixx = -
2c
I J' 0
[F,(x + c(t - r), r) - Fp(x - c(t - r), r)] dr
where a = x + c(t - T) and fJ = x - c(t - r).
12. {{ = ~ [f (x - t) + e"'f (x + t)]

2
+2
J e-<ax,2h(>lf2)
. Jx-r r'+r e"l [ g(s) - ; f(s) J ds.
Exercises 3.4
ee
1. E'(t) =
f -x
x CLO
2
c 2 u 0 J dx +
I - cc
u,(u,, - c ux11,:
- oc
= 0
by the differential equation and the assumption on llx and 111 as lxl --> oo. Thus
E(t) = const for t 2 0.
2. Tf and u 2 are two solutions, then u = 11 1 - u 2 satisfies llzr - c 2 u'"' = 0,

11 1
11(x, 0) = 0 and vanishes together with its first derivatives as lxl --> oo. Since
E(t) = cons! and (0) = 0, it follows that E(t) = 0 for t 2 0. Therefore,
uf + c 2 u; = 0, which implies u 0. =
r::
{
~ [f(x + ct) - f (ct -- x) l + ~~ g(s) ds (x < ct)
5. u(x, t) =
~ [j(x + ct) + f(x - ct)] +
2~ f_+c:' g(s) ds (x > ct)
320 So/11tio11s o( the Exercises
6. Let 11 be the solution of (2.1 ), (2.2) where f and g are even. and set r(x, I)
u( - x. I). Then z is also a solution of (2.1 l, (2.2). Since the solution is unique,
u(x, t) = r(x, I), or 11(1, I)= u(-x, r).
7. u(x, !) = ~ [!(., + cl)+ f(cl - x)] + c J:r-x g(s) ds
+ ( lx+n g(s) ds (x < Cl)

2c cr-x
f3x1 - s(x 3 + 3xt 2 ) (x ~ I)

8. u(x, I) = .
\x 2 + xr + 1- -
7
1(3x 2 1 + 1
2
) (x > t)
11(1/4, 1) = 47/128.
9. u(x, I) = xi + sin(nx/2) cos(nt /2).
10. (a) u(x, t) = cos(nx/2) cos(m/2) + x 2 1 + t 3 /3;
(b) u(x, t) = -~[e-lx-rl + e-lx+<I] +cos x sin t.
12. u(x, t) =sin x cost+ tx1 2 . 13. u(x, I)= cos x sin t + 1 3 /6.
Exercises 3.5
1. If f is odd about x = 0 and x = L, then f(x + 2L) = f[2L - (-x)]
-f( - x) = f(x).
2. If f is odd about x = 0 and even about x = L, then
j(x + 4L) = f[2L - (-2L - x)]

=f(-2L-x)= -f(2L+x)
= -f(-x) = f(x)
3. Write ia+2L g(x) dx = 1-L g(x) dx + J:L g(x) + J:+ZL g(x) dx.
The first and the third integrals on the right cancel out. In fact, set s = x + 2L.
Then
ra-L g(x) dx = (L g(s - 2) ds = (L g(s) ds

J, la+2L la+2L
since g is of period 2L.
4 Let 11 be the solution of (2.1), (2.2) where/ and g are odd about x = 0 and x = L.
To show 11 is odd about x = L, set v(x, t) = - u(2L - x, t). Then v is also a
solution of (2.1 ), (2.2). Since u is unique, u(x, r) = v(x, t) = -u(2L - x, t).
7. Let g denote the odd periodic extension of u,(x, 0) = x(I - x 2 ) of period 2.
Then
u(x, 1) = sin nx cos nt + l ix_+,' g(s) ds
and
-v2 11os
4 2048
321
8. 11(3/4, 2) = 3/16.
9. 11(x, f) =(1/2)[/Lx - f) + f(x + t)] + (2/n) cos(n/2lx sin(n/2)1 where/ h
. o f 11(x. 0) = x~(l
extension 0 0
- x-) sue I1t1at
I / 1s
even about x = o and odd ISb t c
x = l. ' a out
10. 11(x, t) = j(l - cos 2nx cos 2nt).
13. 11(1/2, 3/2) = l[3e - 2e 1 12 - l].
14. 11(1/4, 2) = 87/128.
Exercises 3.6
1. 11(5/8, 19/8) = 273/32. 2. 11(1/4, 3/2) = -11/128.
x)e-<r-x) - e- 1 sinh x
3. (a) 11(x, t) = f (t - (x < I)
\ e-x cosh t - 1 (x > t)
(b) u(x, t) = {
si.n 2(t - x) + sin x cost (x S t)
sm x cost (x 2 t)
4. (a) 11(1/4, 3/2) = j[e 314 - e 114 - 2e + I].

(b) 11(3/4, 3/2) = -11/192 + cos(5n/8).
6. u(x, t) = ~ k (r + x - _( 2 n____llL) - ~ k (r - ~~_Qn_+_I)[,')

"~o c 11~0 , c
with k(t) = 0 fort< 0.
7. u(x, f) = - c s~-x/c h(s) ds + v(x, t) for x < ct where v is the solution of

Problem 7, Exercises 3.4.
-H/Cx + ct) + /(ct - x) J - kek(x-cr) Jrcr-x ek~f(?;,) d~

0
8. ll(X, I)= (0 s x s ct)
{
-![/(x - ct) + f(x + cf)) (x 2 ct)
Exercises 3.7
1. 11(x, t) = i sin x cos t - sin 3x cos 3t.
2. 11(x, !) = (3V2./8) sin x sin YZ t - c/J0/40) sin 3x sin\! l-0 t.
3. u(x, t) = sin r + (3v'2./8) cos x sin \ 12 t + (\ 1 10/40) cos 3x sin YlO t.
4. u(x, f) = e-x[2 sin 2nx cos(4n 2
+ 1) 112
r - 3 sin 5nx cos(25n 2
+ 1) 112 t].
5. 11 (x, t) 34 te-'
= r'2 .
sm (x) -
2
gv2 e -r/2 sm. \ ;-2 t sm. (3x)
2
.
6. un(x, f) = sin (1(211 - 1)n In x] sin [1-(211 - I )nt ].

7. 11
11
(x, f) = e-<x+t) cos(n 2
+ 1) 112
t sin nx .
8. 11 11 (x, . (mrlnx)
t) = x 112 sm -
ln 2
. (n n -
sm
ln 2 2
2 2
+ - 1)
4
112
t.
9. un(x, t) = (I + t 11-')'sin [nn ln(t + 1)/ln'-J sin(nn In x/IQ;i).

Chapter 4
Exercises 4.1
1. + cebx, = 0.
(d /dx)(ebxz/) 2. (d/dx)(x 2 u') + x 2 u = 0.
3. (d/dx)(ex11') + (e-'/x)11 = 0. 4. (d /dx)(x11') + (I /x)u = 0.
5. (d /dx)(xe-x11') + e-"11 = 0.
6. (a) u(.Y) = C(x - 1).
(b) II= 0.
7. u(x) = C sin x. 8. = 11(x) 0.
9. u(x) = 0. 10. 11(x) = 0.
11. u(x) = C In x. 12. u(x) = c_,-2,
13. u(x) = Cx. 14. u(x) = 0.
16. (a) No solution. (b) u(x) = 2 sin x - cos x. (c) u(x) = - sin x.
17. (a) u(x) = e". (b) u(x) = e- 1 cosh x. (c) u(x) = e-x.
(d) u(x) = e-x.
18. (a) u(x) = eh/(! - e 4 ) + e- 2"/(I - e- 4 ). (b) u(x) = cosh 2x.
19. (a) No solution. (b) u(x) = ~ sin 2x - cos 2x.
(c) u(x) = { sin 2x + -J cos 2x.
2 In x
20. (a) 11(x) = 2 In x. (b) u(x) = I + (c) 11(x) = In x - In 2.
- 2 In 2
21. (a) u(x) = cos(ln x) + (tan I) sin(ln x). (b) u(x) = cos(ln x).
(c) u(x) = sin(ln x).
Exercises 4.2
1. (a) C(x, ) = / --'

(0 s x s )
I- ( S x s I)
f (0 :S .\- :S J
(bJ C(x, J =
Ix ( s x s I)
J -(\ ll (0 :S x :S )
(cJ C(x, l =
\-( - l) ( s x :S I)
/-sinxcos (0 :S x :S )
2. (a) G(x, ) =
\-sin cos x ( s x s 7!)
(b) G(x. J = /cos x sin co :s: x :s: sJ

\cos sin x ( :S x :S 7!)
f -e'-; sin x cos (0 :S x s )

3. G(x, ) =
\-ex-; sin cos x ( S x S n/2)
f (x - 2x)jl,
3
(1 s x s )
4. G(x, ) =
\ (\
2
- 2xi1e ( :S x s 2)
f x In ,; - x( I + 2 Jn 2)/.;3 (l ::; x::; )

5. G(x, ,;J = \
x Jn x - x(l + 2 Jn 2)/C, 3 (C, ::; x ::; 2)
f-ln(l + x) (0::; x::; .;)

6. G(x, C,J =
\-ln(l + ,;) (C,::; x::; 1)
1 B . ? A sin 2(x - 1)
12. u(x) =
1 o
C(x, )f(C,) dC, + - sm _x
sm 2 sin 2
where
sm 2x sin 2( - I)
l
(0::; x::; )
2 sin 2
C(x, C,J =
sin 2 sm 2(x - I)
( ::; x ::; 1)
2 sin 2
13. u(x) = Jo1 G(x, C,)f(C,) dC, + B sill kx + '5_cos_~'._ - A~/((~= !_)
sin kx + k cos k sin k
where
(sin_!'.\"__+! cos k:-J ~n k(~ -
l
I)
(0 ::; x ::; )
k(sin k + k cos k)
G(x, ) =
(sin kC, + k cos kC,) sin k(x -
-~- - -- .
1)
( ::; x ::; 1)
/dsin k + k cos k)
with sin k + k cos k #- 0.
. /In (0 ::; x ::; )

14 . G(x, <;) = \
In x (C, ::; x ::; 1)
15. G(x, ) = l x"CC - C"l/211 (0::; x::; )

\ C,'1(x" - x-")/211 (C,::; x::; I)
16. G(x, ) = I - } Jn[< I + x)/(1 - x)] (0 ::; x ::; c;)

\ -1 ln[(l + C,)/(1 - )] (C, ::; x ::; 1)
17. A solution of the differential solution that vanishes as x -> - oo is given by

11,(x) = e'x. On the other hand, u 2(x) = e-sx vanishes as x-> oo. Then
W(ll 1 , ll 2 ; x) = -2s. Formula (2.12) gives the result.
Exercises 4.3
I. 11 0 (x) = I, u 2 (x) = x so that
G*(x, J =
/ (0 ::; x ::; )
Ix ( ::; x ::; J)
2. G*(x, ) = J(1 - x)C,

(0 ::; x ::; )
\(1 - )x (C, ::; x ::; I)
(0 ::; x ::; )
3 _ C*(x, ) = /-sin x cos C,
\ - sin cos x ( ::; x ::; n/2)
Inc; (0 ~ x < ')

4. G*(x, ') =
{In x (c; ~ x ~ I)
5. G*(x, c;) =
-!ln[(l + c;)/(1 - m (0 ~ x ~ c;)
{ -! ln[(l + x)/(1 - x)] cc; ~ x ~ 1)
6. Let u0 denote a nontrivial solution. If 11 1 and u2 are linearly independent on

[a, b ], there exist constants c 1 and c 2 , not both zero, such that u 0 (x) = c 1 u 1 (x) +
c 2 u 2 (x). Then c 2 u 2 (a) = 0 and c 1 u 1 (b) = 0, which imply either u 2 (a) = 0 or
u 1 (b) = 0. If u 1 (b) = 0, then the Wronskian of 11 1 and u 2 vanishes at x = b,
while if u 2 (a) = 0, the Wronskian vanishes at x = a. Either case contradicts
the assumption that 11 1 and 11 2 are linearly independent.
7. Consider the Wronskian W = u0 (x)u'(x) - u0(x)u(x), at x = a and x = b.
9. Consider s~ u6 dx = s~ Uo Lu dx and use Lemma 3.1 of this chapter to get a
contradiction.
10. No solution.
11. J~ u(x)[(Au 0 )" - (Bu 0 )' + Cu 0 ] dx = J~ f(x)u 0 (x) dx.
Exercises 4.4
1. }.n = (211 - 1) 2 n 2 /(4L2 ), un(x) = cos \!"f,, x; II = 1,2, ....
2. An = n 2 , un(x) = cos nx; 11 = 0, I, 2, ....
3. Ao = -1, 11 0(x) = e-x; A,,= 11 2 , 11.(x) = sin nx - n cos nx; n = 1, 2, ... .
4. Ao = -1, 11 0 (x) = ex;).,, = 11 2 , u (x) = sin nx + 11 cos nx; n = 1, 2, ... .
11
5. An = 114 n 4 /L4 11,,(x) = sin(nrrx/L); n = 1, 2, ....

6. An~ (211 - 1) 2 n 2 ,'4, 11,,(x) = sinvI,,x - vl,;cosvI,,x.
7. A"~ (211 - 1) /4, u.(x) 2
= sin \
1
i,; x.
8. An ~ n n 2 2
, u.(x) = cos \ ). 11 x.
9. }.n ~ 11 2
' 11.(x) = sin \ ).,, x - \ x~ cos vi,; x.
An ~ 11
1
10. 2
, 11,,(x) = sin\ /. 11 x - \ /. 11 cos \ J:: x.
11. (a) ).11 = n 2 n 2 , u11 (x) = sin(11;-r Jn x); n = 1, 2, ....
(b) },n = (2n - 1) 2 n 2 /4, u,,(x) = sin[(2n - l)n In x/2]; n = 1, 2, ....
(c) }.n = n 2 ;-:". u,,(x) = cos(nn In x); n = 0, 1, ....
12. An = (4n"n 2 + 1)/4, u,,(x) = x- 112
sin(mr Jn x).
13. }.0 = 11
2
+ 1, 11,,(x) = x- 1 sin(nn In x); n = 1, 2, ... .
14. An = 11 2 rr 2 + 1, u11 (.y) = x sin(mr Jn x); /1 = 1, 2, ... .
15. An = n n /L , 11,lx) = A,, cos(nnx/L) + B11 sin(mrx/L). A 0 and Bn are arbitrary
2 2 2
constants, 11 = 0, I, 2. ....
16. An = 4n 1 rr 2 /L2 , u,,(x) = A" cos(2nn:'(/L) + Bn sin(211nx/L). A 0 and Bn are ar-
bitrary constants, /1 = 0. 1, 2, ....
18. u.(x) = 10 (\' ).,, x), where 10 ( \ ).n) = 0.
Exercises 4.5
1. </Jo(x) = (2/(1 - e- 2") ]112e-x; ,,(x) = v2 (sin nx - 11 cos 11x)/ [n(n 2 + 1) ]112;
11 = 1, 2, ....
2. 11 (x) = (2/L) 112 sin(mrx/L); 11 = l, 2, ....

3. {!/YU, (l//i) cos(nnx/L), (!/\IL) sin(nnx/L)}.
5. Make the substitution t = In x.
9. Let 1 , qJ 2 , . . . , be a sequence of eigenfunctions with weight function r on
[a, b ]. Suppose c 1 1 (x) + + c11 11 (x) = 0 for some n. Multiplying this by
r(x)cp,,,(x) and integrating from a to b, there results c,,, = 0 form = 1, 2, ... , 11.
14. Consider - ). J~ v(x)11 2 (x) dx = J~ u(x)Lu dx = [pu(x)u'(x) ]~ - J~ p(x)u''(x) dx +
J~ qu 2 (x) dx. If u(a) = u(b) = 0 or u'(a) = u'(b) = 0, then the left-hand side is
:50 since q :5 0. Therefore, A:::>:: 0. Under (c), pu(x)u'(x)]~ = -c 2 p(b)u 2 (b) -
c 1p(a)11 2(a) :5 0, so that also). : :>: 0.
15. Let /1 and r be eigenfunctions corresponding to distinct eigenvalues tl and v
respectively. Then (v - ) J~ u(x)r(x) dx = s~ (uv(iv) - Vll(iv)) dx = [uv"' -
u"'i - u'v" + u"v']6 = 0. Since v I- , u and v are orthogonal.
Exercises 4.6
1. Multiply the series by r(x)cpk(x) and integrate from a to b.
2. 011 = {J~ r(x)f(x)1111 (x) dx/;u i
11 1
2
}.
oc (- I )11 + 1 . 8 ~ s_in(2_n~ I__)~

3. x 2 ~ 2n E - sm nx - - '-' - - (0 :5 x < n).
11=1 /1 7[ 11=! (211 - 1) 3
4. x ~ -
SL
n2
00
"~1 (2~ -
I I)i cos
(211 -
2-
I) z n
x (0 :5 x :5 L).
n _ 4 ~ cos(2n -:--- I )x
5. x ~ '-' (0 :5 x :5 n).
2 7[ = l (211 - 1) 2
6
_
1
~ ~ ~ sin [(211 - IJn ln_.x-] (I < x < e).
7[ n=I (211 - J)
X'
7. f (x) ~ a 0e-x + E a 11 (sin nx - /1 cos nx) (0 < x < n).

n= 1
where
2
- e-2rr
a. =
n(n 2
2
+ 1)
J f(x) (sin
0
" nx - 11 cos nx] dx
8. If the series were the eigenfonction expansion of a square integrable function,

1 2
then by Bessel's inequality the series L:(l/v ;;) must converge. But L:(l/11) is a
divergent series.
9. If f is orthogonal to all the functions ., then c. = 0 for all n. By Parseval's
equation, J~ /2(x) dx = 0. This implies that f = 0, since f is continuous.
326 Solutions ol the Exercises
11. If f is orthogonal lo all the functions </>" + ef;"T 1 11 :0: 1, then c 11 + c,,+ 1 = 0
for112: 1. Thusc,,+ 1 = (-l) 11 c 1 \\l1ichviolatesBessel'sinequalityunlessc11 = O
for all 11. If all the c,, are zero. then by the preceding result f must be identically
zero.
x (-1)11+1
12. f(x) = r, cp 11 (.\).
11=1 (n - 1)!
Exercises 4.7
2 sin 2x 4 sin 3x
1. u(x) = u of. 4, 9).
}. - 4 I. - 9
2 cos 2.\" 3 cos 5x

2. u(x) = (i. of. 4, 25).
}. - 4 }. - 25
3 cos(n r/2) 2 cos(7r:/2)

3. 11(x) = -
2
+ --
i. - n /4 }. - (49n 2 /4)
4. u(x) = t
k=l
sin [(2k -:_ l)x/2]
k[}.-(2k-1) 2 /4]
l. 2, ... ' 11).
S. u(x) = aoe-:" + ~ _a,,i_si1111x - n cos nx)

(}.+!) 11=1 i.-11 2 '
where a0 and a,, are the Fourier coeAlcients off given in Problem 7, Exercises 4.6.
w
6. u(x) = r, --~" sin( 11n Jn x),
n == 1 }. - 1127!2
where b11 = 2 Jr (l/x)/(x) sin(nn In x) dx.

3 sin(n Jn x) 4 sin(3n Jn x)
7. u(x) =
}, - n2 i. - 9n 2
2 ~ sin_(_nn__/LJ sin(nnx/L)
8. G(x; ) = '-' (11 = I, 2, . . ).
L n=l }. - (n 2 n 2 /L2) '
9
.
2
i': sin( 11 - I /2)n sin(n - I /2)nx
n= 1 ). - (11 - l/2) 2 n 2
f - (sin X x cos v';: ~)/\/A (O :S x s )

l- (sin vJ.,; cos \/). x)jv'}. (~ :S x :S 1)
~in(}'-- lr2....::~) c_os_ll.(1 _-::_ In 5J]

- ). cos }
11. G(x; ~;I.) =
1 sinU In c;) cos[).( 1

- }. cos ).
ln ~)]
(c; :S x :S e)
Sol11tio11s of the Exercises
327
Chapter 5
Exercises 5.1
1. f(x) ~
n
4
~ ~C_-:-IJ ?- -l
+ '-'
n=l
11
rm-
cos llX + I +2(-1) +
- - -- --
II
11 1
.
sm ILi:
l .
2. f(x) ~ n + - l: C l(-1) -
-
11
-
1
- cos 211x + 2(-1)"+ 1 .
- sm(2n - l)x
J.
(211 - l) 2
0
8 n 11= 1 2n-
~ 4 ~ c~s(2!1 -::- _I )nx

3. f(x) '-' 2
n2 n= 1 (211 - I )
sinh aL
4. eax ~ -- -'- 2(sinh aL) l:x
aL 11=1
cos 2nx
5. f (x) ~
I
-
2_ ~
'-' 2
- + -1 sm
.
x.
n n n= 1 411 I 2
w
nx L . nx 2L l: ( -
6. x cos -- S111 - + 1 )"
2
n sin nnx .
L 2n L n 11=2 11 - 1 L
l! 4L2 "'(-1)
11
1mx 2L "'(-1)"+ 1 nnx
7. x + x2 ~ + 2
l: - cos -L- + l: ~-- - sm -
3 n 11= 1 112 n n= 1 n L
sinh an ?a Cfo (-1)"

8. cosh ax + - sinh an l: COS llX.
an iT 11=1 a2 + n'2
. 2 sinh an ~ ( .,. 1 n .
9. sm 11 ax ~ ------- "-' - 1)11 --- sm nx.
1[ n= 1 a2 + n2
Jl. f (x) ~ _1 _ 2 ~ cos 2nx sin x.

n n 11= 1 4n 2- 1 2
12 _ f(x) ~ n + 4 ~ cos(211 :-- l)x

2 n 11= 1 (211 - I )2
13. Consider
a"=
L J o
-L
. mrx
j(.c)cos-- dx+-
I
L L
JL
0
.
f(x)cos
11nx
dx
L
and set x = t - L in the first integral on the right and use f(t - L) = f(t).
14. f (x) ~ n/2 - Li~= 1 (sin 2nx)/ 11.
Exercises 5.2
1. f(x)
1 _ 3 ~ c~s_2(21! -::- I )nx
4 n2 n= 1 (2n - 1) 2
4 00 (-1)11+1 .
l: ---- ----2 sm(211 - 1)nx.
n2 n=l (2n - 1)
2n 2 1
2. n2 - x2 ~ - + 4 :E"(-1)"+
2
-- cos 11x
3 11= 1 11
~ r-2n + 4 I (- l)"j .
- --- Sill nx.
11= 1 11 7[ 113
Cf)
8 cos [(211 1)nx/2]
3. 1 - .t ~
2 :E1
n 11= (211 - 1)2
oc-
2 sin mrx
n 11=1
:E II
4 . f(x) ~ 3 _ 2 ~ ~os(~~- - l)nx 2

4 n n=I (211 - 1) 2
__4 ~ cos [(2n_ -= I )nx/2]

n2 n=I (2n - 1)2
00
1 2
5. ex ~
e7[ -
- + :E ( - l)"e" -
- -l cos nx
n 11 11= 1 1 + n2
~ ~
n n=I
(-ill__
n + 1
_ .) [e"(- J)"+ I] sin nx. I
7. f(x) ~
J ( 2) I
l + - - - cos nx - -- :E
2 cos nnx -x
2
2 n. 71 71 n= 2 11 - ]
00
2 4
8. sinh x (cosh 11 - 1 ) + co sh n :Ex -(2--I)"- cos nx -
4
:E cos /IX
n 11 11= 1 11 + 1 n ri= 1 n2 + 1
9 _ f(x) ~ I _ __22 ~ cos(2_'1--=-~ ~nx

4 71 n= I (2n - 1)-
+
4"'
71 2
j11(-l)"+ 1
n~I [ 2 (-2~--=-i) (211
l
l) 2 j cos (211 ---
2
1) 11x.
10. sin x ~
2 4 :E
Cf
cos __7 11x .
2
1[11=1411 -1
I 4 "' n
11. f (x) ~ Sill x - :E (- l )11 --- sin nx.
2 n 11= 1 411 2 - I
~ 8 ~ sin(2_n__=_ l )nx .
12. x(l - -x) '-'
11 3 n= I (211 - I )3
13. f(x) ~ J 1
[ b 211 sin 1mx + b 2 ,,_ 1 sin(
2
n__=-J_)
2
11xl
where
16 2
------ -3 +
n3(2n - l) n(2n - l)
14. cash x ~ 2
~ (. ;
1
~ -) [(-I ) 11 ~ 1
cash n + I] sin nx.
n 11= 1 n- + I
8 cc
cos nx ~ ~
11
JS. - sin 211nx.
2
7r II= I 411 - I
Exercises 5.7
(~) 1;2 I - cos s 2) 1

.2 a
1. F(s) =
s2
2. F(s) =
(11
-
a2
~.
+ s2
5. (c) G(s) = - I= lx
,/211
'Y) e-is(a+l)f(t) dt
v2n Jo
e-iasF(.1).
6. J:., e-is'f'(x) dx
I e-i'''j(x) I oc + is
\ 271 : - 00 \ in
(112)
112
cos(5._n/2).
7. (a) F,(s) =
I - 52
2)1/2 1 - .1'2
(b) Fc(s) =
( n. (I + s2)2
(n2)
112
s - ,sin(sn/2).
8. (a) F,(s) = ,
s- - I
(b) FJs) = .:. :-. (. "~)112 (l
2 2 - s2
(bl f(s) =
4
7r 5 + 4
11. f(x) = 2xe-' (x 2: 0).
Chapter 6
Exercises 6.1
3. </>(x, t) = [(L - x)/L)]a(t) + (x/L)b(t).
4. <f>(x, I) = xa(I) + 2
(x /2L)[b(t) - a(t)].
330 Solutions of the Exercise!.
5. c/>(x, I) = ((;r - x)/Ti)t 2 + (x/rc)e'

6. 1, - lffxx = xi - (x - 1) cos I - (1/U + t)); r(x, 0) = x, r)O, I) = r(], t) = O;
c/>(x, I) = (x - 1) sin t -1- ln(l + /).
7. (a) efJ(x) = -(A/ka 2 )e- 0 -'.
IJ A
(b) efJ(x) = - x +
ka 2
8. Set i(x, I) = u(x, t) lz( !). 9. Set r(x, t) u(x, t) - xh(t).

10. 11(x, t) = e-hx s~ e '~1(', t) d'.
1
Exercises 6.2
1. Let 11(x, t) = i:(x, t) - u(x, t). Then w(O, t) 2: 0 and w(L, t) 2: 0 for t 2: 0,
and 1r(x. 0) 2: 0 for 0 ::; x ::; L. By the maximum principle, w(x, t) 2: 0 for
0 :o; x :o; L, I 2: 0.
3. If 11 assumed a maximum at a point (x 0 , 10 ), where 0 < x 0 < L, 0 < t 0 :o; T, T
an arbitrary number, then 11,(x 0 , t 0 ) 2: 0 and u~)x 0 , t 0 ) ::; 0 so that u,(x 0 , t 0 ) -
kux/x 0 , 10 ) 2: 0, contradicting the fact that F(x 0 , 10 ) < 0.
5. (a) Integration by parts yields
! .~ IL 11 2 dx - k llllx iL -1- k roL u; dx 0

201J 0 o Jc
Since 11(0, t) 11(L. t) = 0 and k > 0,
-J (;. jL 11 2 dx = - k r.L u; dx ::; 0

2 01 a a
6. Since 11 tends to zero uniformiy in r as Ix[ -> w, there is a number L sufficiently

large such that '.u( L, t ): ::; maxi/(x)[ for Ix[ ::; L. Then, by the maximum
principle, [11(x, t ll :S: max[/(x)[ for [xi :S: L, 0 ::; t ::; T for sufficiently large
Land T. Hence, [11(x, I)[ ::; maxl/Cxll for all (x, r).
7. If 11 is a solution of the heat equation 11 1 - k(uxx + 11,.,.) = 0 that is continuous
in n, then 11 assumes its maximum either in D at t = 0 or on C for 0 ::; t ::; T.
Let r(x, y, t) = u(x, y, t) -1- cx 2 . Then r, - k(ixx + vr) < 0 and hence i
cannot attain a maximum in the interior of n.
8. Suppose 11 1 and 11 2 are two solutions. Take 11 = 11 1 -u 2 . Then 11 vanishes in
D and on C for 0 ::; t ::; T, T being arbitrary. By the maximum principle,
11 =
0 for all (x, y, t).
9. By Green's identity,
JJ
D
1111, dx dy - k JI
D
u t.u dx dy
-I ;;-
2 ct
, II u 2 dx dy k
II
au
u-ds+
on
11~) dx dy = 0.
D c
Solutio11s of the Exercises 331
Since 11 0. it follows that

1 (
2 21
JJ D
11
2
dx dy - k f5f (u; + u;) dx dy :'.:'.: 0
Exercises 6.3
4A x e-kc211- 1 l'
l. 11(x, t) ~ - - sin(211 - l)x.
7[ II= J (2fl - 1)
2. 11(x, t) = le-" sin x - }e- 9 kt sin 3x.
3. ll(X, I) j (1 T e- 4 ' ' cos 2x).

'l_ e-kC211-1 )2t
8
4. 11(x. I) ~ 1)3
sin(211 - J)x.
7[ n= 1 (211 -
lf e-kc211-1J2t
4
5. 11(x, I) -
~ - sin(211 - - l)x.
l[ n:: 1 (211 - 1)2
32 ~
Cf
II e --k(211-J)21.'4~;
, _// - 1)
6. 11(x, t)
8 11~1 ( - I J (2---;;--_ I )3 "", 2 - _ x.
8 x ,- ( - I J" 4 J
7. ll(X, I)
n-? ~
11=1 _(211 - ])-' + n(211 - I) 3
8. 11(x, tJ e-Ck+J)r cos x. 9. u(x, t) = i sin x - e- 3 ' sin 2x.
11. 11(x, t)
~ - (-
-32 '-" JJ"+l
e -[(211-lJ'n'A+l)t CO>
-('211 -
-~--
J) n.x..
n3 11=1 (211 - 1) 3 2 .
12. u(x. t) = L, 1~:,,, 1 b 11 e-k~r sin 11 x, where b = (2/LJ J~/(x) sin 11 11

x dx with p 11
- h tan p 1,L.
13. 11(x, I) = L,;;'= 1 h,,e-' 1 '~'(11 sin p 11 x - p 11 cos 11 11 x), where b11 (2/ L) J~ /(x) x
(h sin p 11 x - p 11 cos p 11 x) dx, with p 11 =' h tan p 11 L.
Exercises 6.4
4A x e-k(211-1 )2rr2r,'L2 2 I'
~
1
I. 11(x, t) = A - - - -- sin(-'- L=- ) nx.
7[ 11= I (211 - I) '
oc e-k(211- I ) 2 rr 2 tC(2L) 2
cos(2~ - -1)
2
SAL nx
2. 11(x, I) (x L)A + --
~ - --- --
7[2 11= 1 (211 - 1)2 L

oc
L x 2w sin(11nx/L)
3. ll(X, f) sin CJ/ -- ~
L 7[ n=l 11(k 2 x;, + (!)2)
x [k). 11 cos C'Jf + w sin <Jf - kl.,,e-k;_,,t]

4. u(x. I) = H'(X. 1I + x cos wr where
w(x, I) = 81:JL
7
~
oc --
( - I)"+
1
, (
f'' e-ki,,(-ri sin en ch ')
Ii.- 11= l ( 211 1)- 0 !
x sin \ ;. 11 x
(211 - 1) 2 7i. 2
(11 l. 2, ... )
4L2
5. 11(x, r) H(x, 1) + x/7i., where
11
2/z ~ (-1)
1r(x. I) = ~ [1 - e-(li'+l)f] sin nx
7i. II= 1 //(11 2 + h)
(-1 )II
+ -J ~
if_
--- '
e- 111 +/l)r sin nx.
n n=I 11
6. 11(x. I) r(x. 1) + cf>(x) + ~ b11 sin(~'l_-:___l) x.

n::::: 1 . 2
where
1(x, I) =
16
7i.
.,
~
n= 1
2e-rr(-1)"+ 1
(211 -
-
1) 2 [(211
+ 211 -
1)2 + 4]
l
e -u_,,, sin
(''
-II;--_}I' x
i_ _
c/>(x)
b11 ~ C
- 0
f(.Y) sin(2
-
11
; - ) x dx
1
},11 (211; If
7. u(x. t) = (a 0 /2J + L,,";'~ 1 a,,e-k"'' cos nx + (A/w) sin oJt where a,, = (2/n) x
J~ f(x) cos nx dx. 11 = 0, I, 2,.
8. 11(x, 1) = ~"'- J' _.

n= I 0
e-k,_,,(t-rl l (-1)"+ 1 -2knn- h(r)
L2
+ l
g 11 (r)_ dr
where gll(r) = (2/L) s~g(x. !) sin(mrx/Ll dx and J.,, = 11

2
7i. 2 /L2 , II= 1, 2, ....
9. u(x, t) = (11 0 (!)/2) + L,,';'= 1 11 11 (1) cos(1mx!L) where , ' =

11 ,. (/) 0 e-ki,,(r-r) x J'
[g 11 (r) - (2k/L)h(r)] dr with ). 11 = n 2 n 1 /L2 and g 11 (1) = (2/L) J~ g(x, t) x
cos(nnx/L) dx, 11 = O. 1. 2, _
Exercises 6.5
1. u(x, t)
2. u(x, I) I
2 er
r(x2\Ffr,
+ L) 1 r(x - L)
- 2er -zVfi .
3. u(x, t) HE - A) erf(x/2\ kt) + -1-(A + B).

5. Let u(x, t) denote the solution of the problem (6.1 ), (6.2), where f is odd, and
define r(x, t) = -11( -x, t ). Then v is also a solution of (6.1 ), (6.2). Since the
solution is unique, u(x, t) = r(x. f); that is, u(x, t) = -11( -x, t ). Tf f is even,
set i(x, t) = 11( - x, t).
6. Set s = (x - c;)/ [21 k(t - r) ].
9. Let V(s, t) denote the Fourier transform of the solution L (x, 1). Then cV/ct 0 +
(ks 2 + h) V = 0, V(s, OJ = F(s), where F(s) is the Fourier transform of f
Thus, V(x, t) = F(s)e-(ks'+"l' so that
v(x, t) =
e-IH
= Joo F(s)e-ks't+i.1x ds
v 271 - -c
- e
---=-
-hr J'lo e
-(x-~)2,'(4kr)j"(") d"
c; c;
2vkrrr -w
11. 11(x, t) = s~ f".'oc e-h(t-r)G(x - c;, t - r)/(c;, T) d!; dr where (,'(x - <:. f - r)
is as it appears in (6.21) of the present chapter.
Exercises 6.6
4. u(x, r) = e"'k'cosh ax - ~ea'kr [e-axer[(a'\lkt - ; )

2 0
+ e"" erf(a'\ 1kt + 2vk1
.:':. )l
5. u(x, t) = A [I - erf(x /2 ,If,. r)].
6. u(x, t) (2k/rr) J~ s sin sx J~ e-ks'(r-r)fi(r) dr ds.
e-x'/[4k(t-rllh(r) dr
7. u(x, t)
'\lf~-T
[G(x + c;, r - r) + G(x - <;, r - r)]/Ci;, r) di; dr.
8.
10. u(x, r) = 2k/rr s~ s sin sx J~ e-(1-r)(ks'+b)h(r) dr ds.
11. u(x, t) = - (k)112 (' exp(-x2/[4k(t_ _._r)] =-bU= ~! h(r) dr.

Tr Jo Vt-T
2e-r': 2 ("'. e-k.''' cos sx
12. u(x, t) = - - ds.
rr o I + s2
13. 11(x, t) = - J.~ e-lr(~-xlr(!;, r) d~. where
r(x, t) = 1_= ('"' [/'(~) - hf(,;)][e-<x-;l'!<4kl) - e-<x+{l'!<4ktl] di;

2'\ 1Im t a J
14. u(x, t) = - J.~ e- 1<;-x)r(!;. t) di;, where i is the solution of Problem 6 of these
exercises.
334 So/11tio11s of the Exercise.1
15. 11(x. I) = e- 1"dx. 1). 1\here d.1. 11 is the solution of Problem 13 of these
exercises.
16. 11(.1.11 = (aAix)[I - erW.\ - a)2\ k11].
Chapter 7
Exercises 7. 1
2. Appl) formula (2.11) of this sect ion.
3. Let 11 1 and 11 2 be t110 solutions and set u = 11 1 - 11 2 . Then 11u + ku = 0 in
D and 11 = 0 on C By formula (2.4). Jv [-ku 2 + 11~ + u;J dx dy = 0. J
Since /; < 0 the integrand is nonnegative. The vanishing of the integral means
-/,11 2 + 11; + 11?: = 0, 1\hich implies 11 0. =
5. Let 11 1 and be t110 solutions and set 11 = 11 1 - 11 2 . Then !111 + /,11 = 0 in D
11 2
and cu/( II + h11 =
0 on c.
By formula (2.4 ). -- fc /111 2 ds = [ D [ ( - k 11 2 +
u; + u?:) dx dy. If k < 0 and h > 0. then the integr~l on the left is 1~onpositive
and the integral on the right is nonnegative. Thi<;. means that both integrals must
vanish, 11hich implies u 0. =
6. By Greens theorem
v'
rI11L11 dx dy JJ D
(1
2
11; + e''u~.) dx dy
+
Jc.r u(-e-"11, dx + x 2 11, d_l') = 0
"hich gives [v [ (x 2 11~ + e'11;1 dx (~V 0. since 11 = 0 on C. Since the integrand

is nonnegative u 0. =
7. If 11 1 and u 2 are t110 solutions and 11 11 1 - 11 2 , then
L11 i: [ (I + '
.c)11, ] + < ,(r l + y z)11, ] k11 0
ex ey
and
11L11 [ii +x 2 J1111,]+

ex
Bv Green's theorem.
JJ
D
11Lu dx dy ,. ,. [(!, x 2 1u~ +
ri'
(1 + v 2 )u;J c1x di 0.
which imp[ ies 11 = 0.

8. If 11 1 and u 2 are t110 solutions and u = 11 1 - u 2 , then !111 0 in D. 11 0 on
C 1 and c11/c11 + h11 = 0 on C 2 . By formula (2.4).
r r(111:\11 T II~+ II~) dx dy

v
Hence.
I
c,
hu
2
d1 r r(11; + 11~) dx dy
D
\\hich. as in Problem 5. implies 11 =:o 0.
Exercises 7.2
2
1. !111 = 11,, + (1/r)u, + (l/r )u 00 = 0.
2. Let = x r 11 = _r/r 2 and introduce polar coordinates.
2
Then i(r, 0) =
u(~, 17) and r,, + (l/r)1', + (1/r 2 )1' 00 = (1/r )(11~ 0 + 1111 ,1) = 0.
4
3. let r(, 11) = 11(r, 0). Then
rec + i,,,, - 0 (0 < < x, 0 < I/ < b)

r(~, 0) = f(ae< l (0 s < oc)
r(O, I/) 0 (Q s
"
I/ s b)
l(~. b) 0 (Q 0:: ~ s X)
6. Since w - 1 ?: 0 and 1 -- 11 ?: 0 on C Theorem 3. l implies that 11 - 1. ::=: O

and r - 11 ?: 0 in D. Hence. 11 :-; i o:: 1v in D.
7. Let 11 = r - 11. Then !iH = lu: ?: 0 in D. Since w = 0 on C. the maximum of
w is zero. Hence. 11 s 0 in D: that is. 1. ::; 11.
8. If 11 attains a maximum in D, then /'i.u ::; 0, contradicting the fact that q > O.
9. H has a positive maximum at a point in D. then at that point /'i.u ::; 0, con-
11
tradicting the fact that h11 > 0.

11. Let 1: = u + r.x 2 \I ith c > 0. Then lir = 2i: > 0 so that r cannot attain its
maximum in D.
Exercises 7 .3
~ sinh [m:(b -- y)/ b J . mrx
1. u(x, y) l.... 1111 . Sill -
11= 1 o,1nh 1m a
where
2 ,.,, . . wrx
b,, = j L1) s111 dx (11 =I, 2, ... ).
a 0 a
~ b sinh(mrx/bl . mry
2. ll(X, y) = l.... II Sill -
11= 1 sinh(mra/b) b
where h11 = (2/b) J~ .lJ< y) sin(nnylb) dy, (11 = J, 2, ... ).

sinh 2(:rr - y) _. sinh 2x .
3. 11(x. y) = -- - - - - sill 2x + - - - sm 2y.
sinh 2n sinh 2n
4. 11(x, y) = x(n - x)/2 - w, where
4 sinh(211 - I )(7T__- }')__~siri_h(2n -=_ ~) y sin( 2 n _ l)x

w(x, y) = - ~
71 n=l (2n I )3 sinh(2n - l)n
5. Set w(x, y) = Ll(x, y) - cos x and solve the problem Aw = 0, w(O, y) = -1,
w(n, y) = I, wy(x, 0) = 0, w(x, 1) = - cos x.
6. Set w(x, y) = LI - (x 3 - y 3 )/6 and solve the problem Aw = 0, wx(O, y) = 0,
w(I, y) = (y 3
- 1)/6, w(x, 0) = -x 3 /6, w,(x, I) = -!.
7. Define v(x, y) = - Ll(y, x). Then /':,.v = 0, v(O, y) = 0, v(a, y) = f(y), v(x, 0) =
0, and v(x, a) = -/(x). Thus, v is also a solution of the problemc Since the
solution LI is uniquely determined, u(x, y) = v(x, y); that is, Ll(x, y) = - u(y, x).
In particular, setting x = y, there follows u(x, x) = - u(x, x) = 0.
8. u(x, y) = i(sinh x sin y - sinh y sin x)/sinh n
-i(sinh 3x sin 3y - sinh 3y sin 3x)/sinh 3n.
9. ucx, Yl = J 1
cosh [c ::_ lln/
2 21
fcosh(~ ; I) x cos(2n; !) y
1
+ cosh -2n ---

(
-
2
1) y cos (2n
- - -
2
1) ] x
where
A"= ~ f f(x) cos(2n;

1
) x dx (11 = 1, 2, ... )
10. u(x, y) =
00
~
sinh(h
b - - - - - - - sm nx
+ n 2 ) 112y .
n= 1 n sinh(h + n 2 ) 112 n
where b,, = (2/n) s~ f(x) sin nx dx (n = 1, 2, ... ).

00
b cosh[nn(x - a)/b] nny
11. u(x, y) = A - - ~ an ------:- - - - - cos - -
n n= 1 n smh(nna/ b) b
where an = (2/b) J~ g(y) cos(nny/b) dy (n = I, 2, ... ) and A arbitrary.

b ~ a_n cosh(nnx/ b) nny
12. u(x, y) = '-' cos - -
n n= i n sinh(nna/b) b
_ a ~ b,, co~h [nn( y--=-- b )/a cos nnx J

n 11= 1 n sinh(nnb/a) a
v.here an= (2/b) s~ g(y) cos(nny/b) dy and b,, = (2/a) J~f(x) cos(nnx/a) dx
(n = 1, 2,. .. ).
13. u(x, y) = A + !y + ~2 ~ rcosh[(2n -:::___!)(n --:___YJL+ cosh(2_11_----=----1_L.v]

2 n n=l (211 - 1)3 sinh 117!:
x cos(2n - l)x.
cosh(x - n) 8 00 cosh(2n - l)(y - n)

14. u(x, y) = - --. - - - cosy - ~ --- --- ---- ---- cos(2n - l)x.
smh n 2n n=l (211 - 1) 3 sinh(2n - 1)
00
y sinh ny
15. u(x, y) = a 0 - + ~ -.--cos nx
2n n= 1 smh nn
where an = (2/n) s~ f(x) cos nx dx (n = 1, 2, ... ).

16. u(x, y) = ~ an ~os~[( 2 n - l)(y - nJ/ 2_J cos(?~- 1 ) x

n=l cosh[(2n - l)n/2] . 2
~ h,, sin[(2n - l)(x - n)/2] . ("211 - ])

2
+ 11=1 (211 - l) ~~sh-[(211 - -o~/2) sm -2- - Y
where
a 11 = 2J{" f(x) cos (211 - --I) x dx (n=l,2, ... )

71 0 2
b,, = Jo
2 {" g(y) sm
. (211
--- - -
2
1) y dy (n=l,2, ... )
71
Exercises 7.4
1. u(r, 8) = ~a 2 r sin 0 - lr 3 sin 30.
2. u(r, 8) = t + 2
tr cos 20 + r sin 8.
2 2
3. 11(x, y) = t(a + x - y ).
2
4. u(x, y) = i(x 4 - 6x 2 y 2 + y 4 ) - 2(x 2 - y

2
) + 6.
5. Set i(r, 8) = u(r. 8) + u(r, - 8). Then Llv = 0 and v(a, 8) = /(8) + f( - 8) =
/(8) - f(8) = 0. Hence, v =
0; that is, u(r, - 8) = - u(r, 8). In particular,
u(r, 0) = - u(r, 0) = 0.
6. u(r, 8) = (2r 2 /a 2 ) sin 20.
7. If f is an odd function of 8, then according to Problem 5, the same is true of the
solution u and thus it vanishes at 8 = 0, n. Set v(r, 8) = u(r, 8 + n) - u(r, 8).
Clearly, Llv = 0 and v(a, 8) = f(O + n) - /(8) = 0. Hence, v 0, that is, =
u(r, e + n) = u(r, 8). Setting 8 = - n/2 yields u(r, n/2) = u(r, - n/2) =
- u(r, n/2); thus, u(r, n/2) = 0.
8. u(r, 8) = (2/a 3 )r 4 sin 2fJ cos 20.
9. u(r, 8) = (3/2Jr sin () - (r 3 /8) sin 38.
10. u(r,8) = (o: 0 + [J 0 lnr)/2 + I,;;~ 1 [(o. 11 r" + /3,,r-")cos11() + (y/' + b r-")sin118],
11
\\here
o. 0 + /3 0 In a = 0 rxna" + /3,,a- = 0 11
et 0 + /3 0 In b = a0 etnb" + f311b-n = an
}'
11
/J" + b11 b-n = b 11
with et,, = (1 /71) J~" /(8) cos 110 d8, /3,, = (I /71) g /(8) sin n() d8, n = 1,2, ....
12. Define /(8) = - f( - ()) for - 71 ~ 8 ~ 0 and use the solution formula of
Problem JO.
13. u(r, ()) = -

n= 1
~ - bn -
sinh [1171/\n(b/ a)]
sinh _!1_
ln(b/a)
(8 -
lf
71
71)] sin f - mr
lJn-(b/a)
In 'J
~
where
2 Jln(b/a) ~ . nnf; ..
b 11 = l;;-(b~) f(ae ) sm ln(b/~) de;
0
....
338 So/11tio11s of the Exercises
Exercises 7.5
3. Multipl) \he inequality 111 S 11(a, 0) S .\1 by the sum of (6.2) and integrate
over [O, 2i1 ].
4. For r < a,
a + -r 11<0. 0)
a r
and
11(r, @) 2: I a2 - r 2
2i1 (a -'- r) 2
f"
, -rr
u(a, c/JJ def;
a - r
2: 11(0. 0)
a+ r
5. Let r ---> oc in Harnack's inequality.

8. 11(r,O) = 2
\3
[K*(a,r,0,</J) - K*(a.1,0.1[ - c/!l]f(<j;)dqJ.whereK*(a,r,O.c/>) =
K(a, r, 0 _ qJ) - K(a, r, 0 + </Jl.
11. Since ~11 = 0 inside the sphere, it follows from the Green's identity
1 ~'11 ds
J Cll
s'
JJJ D
/';.u d~ d11 d(
that
Define
U(x, y, z; r) =
r
4n o2;r , ""
o
u(~. 11. ()sin de/! dO
Then
~~~1 (~.
()U
11. () sin d</J dO 0
or er
so that U is independent of r. Therefore, 111 particular, U(x. y. z; a) =

U(x, y, z; 0) = u(x, y, z).
12. Suppose ~11 > 0 at a point (x, y, z). There is a sphere S of radius r about this
point throughout which ~11 > 0. Consider
u(x, y, z) = - l f Irr u sin </J dc/J dB

2rr
4n o o
over the sphere S. Then
c11
I (21[ {" ~11 sm d d(J
er 4:rr J J
0 0 cr
'j' ~ 11
dS = ] 6.11 df,, d17 d( > 0
Js
{ { {
4:rrr 2 011 4nr-

0
I vI .)

contradicting the fact that cu,!(;r = 0.
Exercises 7.6
I. u(r, 8) = (r 2 /4J(l - cos 20) + C.
2. 11(r. ri) = (3/2)r sin(} - (r 3 /24) sin 3{i + C.
3. u(r, ri) = (r 2
/2) sin 20 + C.
4. u(x, y) = -!,-; 3 - -!x 2 y + 6y + C.
5. u(x, y) = 1x 3
- xy 2 + C.
8. u(r, ri) = (C1' 0 /2) + (a/2:rr) J~" /(</!) Jn [a 2 - 2ar cos(O - J + r 2] d</J.
10. u(r, ri) = (:1 0 /2) - (a/2n) J~f(</!) Jn[(a 2 - 2ar cos(8 + </!) + r 2 )
(a 2 - 2ar cos(8 - ) + r 2 )] ddJ.
Exercises 7.7
3. u(x, y) = y
:rr
J"' lf
o (x +
--I
f,,)2 + y2
+
(x -
_\
<;) + y2
J /(<',,) df,,.
4. u(x, y) = x
:rr
J"' f
o [x2 +
I
(y - 11)2
-
x2 + (y
l
+ 17)2_
l /(11) d17.
6. u(x, y) = -
,., Jx cosh sy .
-- Sill SX
J"' f(<;)_ . , d.;_ SIO .1.; ds.
:rr 0 cosh sb 0
7. u(x, y) - 2 ~
..._, bn - e -(2n-l)x/2 cos - - - (211-]) y
11~1 (217 - 1) 2
where
b,. =;2 j~" g(y)cos (211 ---1) y dy

0 2
8. u(x, y) =
2 Joo sinh sy
- cos sx
Joo /(<;)cos
_ s;_d.;_ds.
n 0 s cosh s:rr 0
9. u(r, 8) =
2
:rr
Jexc c~h s_~ sin(s In
0 cosh sb
1
r
) J0
1
1
p
/(p) sin(s Jn
1
P,
) dp ds.
10. u(x, y) =
2-
n
J"'
0
sinh(s
- - -
2
+ h) y sm
sinh(s 2 + lz)n
.
sx
112
5 00
0
_
/(;) sm s; d<; ds.
" _
00
11. u(x, y) = E bne-\ 111 +/i)li'x sin 11y

11=1
where b11 = (2/n) J~ g(y) sin 11y dy. II= J. 2, ....
13.
1
u(x, y) = - - J'" f(C,) ln{[(x + d + 2
y ][(x - ~) 2 + y
2
]} dC,.
2n 0
14. u(x, y) =
1
2n )o
f'"
2
g(17) Jnl:~___(Y_::-
lx2 + (y +
l/)
17)2
2
1 d17.
15. u(x, y) = - f " g(17) In{ [x 2 + (y - 17J

2
][x 2 + (y + 2
1/) ]} d17.
2n Jo
16. Superpose solutions of Problems 13 and 15.
Exercises 7.8
3. In the expression for G(C,, 17; x, y); set <J = (x 2 + y 2 )1! 2 and note that
(x 1
+ y
1
) r(c, - :~ xf + (11- :~ )rl
= (x2 + y1)p2 - 2a2(x.; + )'I/) + a4
= p
2 r X
2
+ Y- - 2a2
-- 0
pl
(X<;

+ }"I/) + Q4
p4
eO
(<.,, + I/ )
2 j
- a2 I/ ).
p2
21
5. By the formula (7.9), write the boundary condition in the form
g(a, e; p, ) = -
I
In a + --l E
x l
( p)" [cos 11 cos 118 + sin 11 sin 118].
2n 2nn=lll a
Assume q(1, 8) = o: 0 /2 + L::'= 1 v"(a,, cos 118 + b 11 sin 118). Equating this to the
boundary condition when 1 = a yields a 0 = - ( l /n) Jn a, a,, = (I /2nn) x
(p/a)" cos 11 and b11 = (l/2m1)(p/a)" sin 11, 11 2: I. Hence.
q(l', 8) = - _I In a - I_ Inl a4 - 2a2pv cos<B ~_cP)__P2l'21

2n 4n I_ a4
-I In a - 1- ln [ a 4 - '
2apv cos({} - ,1.,
'+') + ' ']
pi-
2n 4n
I
6. q(r, 8; p, c/y) = ln[a 4 - 2a 2 pv cos(B - c/y) + v2 p 2 )
4n
1
In{[a 4 - 2a 2 pi cos(O - ) + v2p2 ]
4n
x [v 2 - 2vp cos(O + c/y) + p 2 )}.

7. C(1. (J; f!, 1Jl C*(r, e; fl. (Pi - C*(1, 8; p. ;r - l.

8. C*(1, O; f!.1JJ = C(r 3 , 30; p 3 , 31!J - C(r', 30; p3, - 3\.
1 2 2
II. CLY, J; , 17) = ln:[(x - ) 2 + (1 - 17) j[(x + 02 -i- (y + 1n ]j
4n
Index
A Cauchy-Riemann conditions, 282

Chain rule, 11-13
Absolutely integrable function, 5, 7 Characteristic equation
Auxiliary conditio.ns, 42 (See also for first order equations, 50
Boundary conditions and Initial hyperbolic equations, 66
conditions) parabolic equations, 66
ChJracteristic lines, 87-89, 108-112
Characteristic triangle, 87, 90, 95
Classification of second order equations, 6 l
B Completeness of orthogonal system, 153,
194
Backward wave, 83, 85, 88, 110
Composite functions, I 0
Bessel equation, 18, 142
Continuous dependence on initial or
function, 142
boundary data, 78, 220
Bessel's inequality, 152, 178
Continuous function, 2
Bilinear expansion, 154, 157, 224
piecewise continuous, 2, 4
Boundary conditions, 42, 121
Continuity of limit function, 20, 24
elastic, 74
Convergence or uniform convergence
first kind, 74
of series, 19
fixed, 74
improper integral, 23
free, 74
in the mean, 150
periodic, I 38, 165
Convolution, 211
second kind, 74, 215-216
third kind, 74, 215-216
Boundary value problems, 42, 121,
252-253, 278, 284
Bounded function, 2 D
d'Alembert"s formula, 78, 87, 91, 95, 102,

104, 108
c Derivatives
directional, 28-29
Canonical forms of second order left-hand and right-hand, 2
equations, 62 mixed, 9
343
..
344 Index
Derivatives (cont.) Fourier sine transform, 209

normal, 30-31 Fourier transform, 208, 236-237, 241,
partial, 7, 8 285286
Differentiation Fundamental solution
or improper integrals, 25 heat equation, 239
integrals depending on a parameter, 15-18 Laplace's equation, 294
series, 21 Fundamental theorem of calculus, 5
Dirichlet condition (See Boundary
condition of first kind)
Dirichlet for mu la, J 84
Dirichlet problem, 252 G
in annulus region, 271
disk, 267 General solution, 39-41
exterior problem, 278 first-order equation, 51
infinite domain, 285 second-order equation, 57
rectangle, 260 Generalized solution, 79, 119
Domain (or interval) of dependence, 87, Green's formula, 133
108, I JO Green's function for
Duhamel's principle, 94, 236, 242 a disk, 299, 301
heat equation, 225, 239
Laplace's equation, 297-298
ordinary differential equations, 127-128,
E 136
the half plane, 302
Eigenvalue, 116, 139 Green's theorem, 32, 253
Eigenfunction, 116, 139 first identity, 34, 254
Eigenfunction expansion, 148-149, 225 (See second identity, 34
also Bilinear expansion)
Elliptic equation, 62
reduction to canonical form, 67-68
Energy integral, 97 H
Error function, 240, 242
Even function, 98, 171 Hadamard's example, 252
extension, 104, 173, 202 Harmonic conjugate, 282
Harmonic function, 257
circular harmonics, 268
rectangular harmonics, 261
F Harnack's inequality, 278
Heat equation, 214
Forward wave, 83, 85, 88, 109 Helmholtz equation, 49, 256
Fourier coefficients, 118, 149 Homogeneous differential equation, 45
Fourier cosine integral formula, 202 Homogeneous boundary value problem, 121
Fourier cosine series, 173 Hyperbolic equation, 62
convergence, 187 reduction to canonical form, 62-65
uniform convergence, 192
Fourier cosine transform, 209
Fourier integral formula, 199
complex form, 207 I
convergence, 199
Fourier inverse transform, 208 Improper integral, 23
Fourier series, 166 Initial-boundary value problems, 42
complex form, 169-170 wave equation, 97-98, 102
convergence, 185 heat equation, 225, 227, 243
generalized, 149 Initial conditions, 42
uniform convergence, 192 first order equations, 53
Fourier sine integral formula, 202 heat equation, 215
Fourier sine series, 118, 174 wave equation, 74
convergence, 187 Initial value problems, 42
uniform convergence, 192 first order equations, 53-54
Index 345
Initial value problems (cont.) Neumann problem (cor1t.)

heat equation, 216, 236-239 necessary condition for solution, 255
wave equation, 74, 77, 90 rectangle, 266
Integration of Nonhomogeneous boundary value
Fourier series, 195, 196-197 problems, 125-127, 131, 154-156
improper integral, 24-25 Nonhomogeneous initial-boundary value
series, 20 problems, 108-112. 229-235
Nonhomogeneous initial value problems,
90-92, 240-241
Jump discontinuity, 1 0
Odd function, 98, 171

extension, 98, 102, 174, 202
K Orthogonality of eigenfunctions, 118, 143,
144
Kronecker delta, 145 Orthogonal system, 145, 163
completeness, 153, 194
Orthonormal system, 145
L
p
Laplace's equation, 250
Legendre's equation and polynomials, Parabolic differential equation, 62
147-148 reduction to canonical form, 66-67
Leibnitz rule, 16 Partial differential equation, 36, 45
Limit, Jefl-hand and right-hand, first order equation, 49
Line integral, 32-33 linear equation, 37, 45
Linear combination, 44 second order equation, 37, 56, 62
Linear independence, 148 Parseval's equation, 153, 193
Linear operator, 44--45 generalized, 194
Period, 4
Periodic function, 4
extension, 167-168, 173-174
M Poisson's equation, 251, 253. 256
Poisson's integral formula, 273, 287
Maximum principle Poisson's kernel, 274
heat equation, 218-219 Principle of wperposition, 45-46
Laplace's equation, 257, 276 Propagation of waves, 88-89
Mean value property of harmonic functions
two variables, 274
three variables, 279
Mean value theorems, 5, 6 R
Radiation equation, 214
Reflection of waves, 109-110, 112
N Region of influence, 89-90
Riemann-Lesbegue theorem, 179-181
Neumann condition (See Boundary Robin problem, 253
condition of second kind)
Neumann's function, 306
in a disk, 307
in the half plane, 308 s
Neumann problem, 253
in a disk, 279 Self-adjoint equation, 122
exterior problem, 284 Separation of variables method, 114, 223,
half plane, 291 227, 230, 234, 260, 267
346 Index
Simple closed curie, 32 Neumann problem, 254

Smooth runction, 3 Robin problem, 255
piece11 ise smooth. 3
Square integrabk function, J 53
Sturm-Liouville equation, 122
problem, 115, 138-139, 154-156 w
Wa1e equation, 71
nonhomogeneous, 73, 90
u damped, 95, 107, 114
Weierstrass M-test for
Uniqueness or solution integrals, 24
Dirichlet problem, 254 series, 19
heat flow problem, 219, 220 Weight function, 143
initial \alue problem ror 11ave equation. Well-posed problem, 78, 220, 259
95-97, JOO, 105-106 Wronskian, 123, 126

1 Eutiquio C Young Partial Differential Equations An Introduction Allyn and Bacon 1972

Uploaded by

Document Information

Original Title

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

1 Eutiquio C Young Partial Differential Equations An Introduction Allyn and Bacon 1972

Uploaded by

Copyright:

Available Formats

PARTIAL DIFFERENTIAL

A I lyn an<l Bacon, Inc. Boston

All rights reserved.

l'."o par! of !he 1110/erial

Librar) of Congress Catalog Cnd Number: 70-166551

Printed in the United States of America.

1. Some Properties of Functions of One Variable 2. Partial

Chapter 2 Linear Partial Differential Equations 36

Chapter 3 The Wave Equation 71

Chapter 4 Green's Function and Sturm-Liouville Problems 121

Chapter 5 Fourier Series and Fourier Integral 163

Chapter 6 The Heat Equation 212

Chapter 7 Laplace's Equation 250

Solutions of the Exercises 313

This book is the outgrowth of an introductory course in partial differential

Tallahassee, Florida EUTIQUIO c. YOUNG

1. Some Properties of Functions of One Variable

Let f be a function defined on the interval a ~ x ~ b and let x 0 be a point

In a similar manner we define the right-hand limit of/at x 0 to be

/(xri) = Jim /(x)

In this case the function/is said to be continuous at x 0 . On the other hand, if

discontinuity at x 0 . The differencej(xci) -f(x 0 ) is called the jump of the

Example 1.1. The function

has a jump discontinuity at x = 0, since/(0-) = 1 and/(O+) = - I. The jump

A function f is said to be continuous on the interval a ::::; x :::::; b ifj is con-

(1.1) Jim f(x 0_+ h) - f(xi)

\\hich means thatf' is continuous at the point x 0 .

Example 1.2. Consider the function

f(x) = l{.Ox2 sin xi (x # 0)

This function is continuous at all points, including x = 0. Now, for h < 0, we

Similarly, when h > 0, we find/~ (0) = 0, and so f'(O) 0. But when x cl 0,

We say that a function f is smooth on the interval a s x s b if it has a

Example 1.3. The function

FIG. 1-1 Pieceirise smooth f1111ctio11.

A function that is defined on the whole x-axis is said to be piecewise con-

( 1.3) /(x + T) = /(x)

</>(x + /1 T) = f(x) (a ::; x ::; b)

(1.4) /(h) - /(a) = f'(xo)

The formula ( 1.4) is frequently written in the form

f(a + h) = f(a) + hf'(a + Oh)

Further, if /is absolutely integrable and g is a bounded function on the interval

THEOREl\1 1.2. Let .f be a continuous.function on the interval a s x s band let

for each x, a s x s b. Then

(1.7) F'(x) = d fx f(t) dt = .f(x)

f(x) dx = F(b) - F(a)

(1.9) f f(x) dx = f(x 0 )(b - a)

2. Let f be a function defined on the interval - L s x s L. State what properties

Show that/ is continuous but does not have a derivative at x = 0.

is smooth in any interval [a, b ].

We now review some concepts concerning derivatives of functions of several

Example 2.1. Find the first-order partial derivatives of the function

The partial derivatives ux and uY of a given function t1 can be given geometrical

where z 0 = u(x 0 , )' 0 ).

assuming these exist. Third-order partial derivati\es of 11 are obtained if the

Example 2.2. Find the second-order partial derivatives of the function 11 =

x cos xy + y considered in Example 2.1.

Su/111io11: The second-order partial derivatives are

7. If 11 arcsin ' - v). , (x.

8. If /1 (ax + by)", 11 a real constant, verify that b11x - au, = 0.

12. I{ = ellX + b)-.

13. 11 = e" sin y.

18. If 11 = sinh(x + at), show that 11" - a 2 uxx = 0.

20. Consider the function

~o f f(x, 1) dt = f ;a f(x, t) dt