PDJ Endre Pap, Arpad Takaci

Partial Differential Equations through Examples and Exercises
Kluwer Texts in the Mathematical Sciences

VOLUME 18
A Graduate-Level Book Series
http://avaxhome.ws/blogs/ChrisRedfield
Partial Differential
Equations through
Examples and
Exercises
by
Endre Pap
Arpad Takaci
and
Djurdjica Takaci
Institute ofMathematics,
University ofNovi Sad,
Novi Sad, Yugoslavia
..
SPRINGER -SCIENCE+BUSINESS MEDIA, B.V.
A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-94-010-6349-4 ISBN 978-94-011-5574-8 (eBook)

DOI 10.1007/978-94-011-5574-8
Printed an acid-free paper
AII Rights Reserved

1997 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 1997
Softcover reprint ofthe hardcover Ist edition 1997
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic Of mechanical,
incIuding photocopying, recording or by any information storage and
retrieval system, without written permis sion from the copyright owner.
Contents
Preface IX
List of symbols Xl
1 Introduction 1
1.1 Basic Notions . . . . . . . . . . 1
1.1.1 Preliminaries . . . . . . 1
1.1.2 Examples and Exercises 3
1.2 The Cauchy-Kowalevskaya Theorem 12
1.2.1 Preliminaries . . . . . . . . 12
1.2.2 Examples and Exercises .. 13
1.3 Equations of Mathematical Physics 15
2 First Order PDEs 17

2.1 Quasi-linear PDEs . . . . . . . 17
2.1.1 Preliminaries . . . . . . 17
2.2 Pfaff's Equations . . . . . . . . 32
2.2.1 Preliminaries . . . . . . 32
2.3 Nonlinear First Order PDEs .. 35
2.3.1 Preliminaries . . . . . . 35
3 Classification of the Second Order PDEs 49

3.1 Two Independent Variables .. 49
3.1.1 Preliminaries . . . . . . 49
3.2 n Independent Variables . . . . 64
3.2.1 Preliminaries . . . . . . 64
v
ci CONTENTS
3.3 Wave, Potential and Heat Equation . . . . . . . . . . . . . . . . . .. 69
4 Hyperbolic Equations 71
4.1 Cauchy Problem for the One-dimensional Wave Equation 71
4.1.1 Preliminaries . . . .. 71
4.1.2 Examples and Exercises . . . . .. 72
4.2 Cauchy Problem for the n-dimensional Wave Equation. 80
4.2.1 Preliminaries .. . 80
4.2.2 Examples and Exercises . . . 82
4.3 The Fourier Method of Separation Variables 89
4.3.1 Preliminaries .. 89
4.4 The Sturm-Liouville Problem 106
4.5 Miscellaneous Problems. 129
4.6 The Vibrating String 141
5 Elliptic Equations 143

5.1 Dirichlet Problem . . . . 143
5.1.1 Preliminaries . .. 143
5.2 The Maximum Principle . 163
5.2.1 Preliminaries . .. 163
5.3 The Green Function .. 167
5.4 The Harmonic Functions . 173
5.5 Gravitational Potential . .. 182
6 Parabolic Equations 183

6.1 Cauchy Problem ... . 183
6.1.1 Preliminaries . 183
6.1.2 Examples and Exercise 184
6.2 Mixed Type Problem . 193
6.2.1 Preliminaries . 193
6.3 Heat conduction. . . . .. .223
CONTENTS vii
7 Numerical Methods 227

7.0.1 Preliminaries . . . . . . 227
8 Lebesgue's Integral, Fourier Transform 249

8.1 Lebesgue's Integral and the L 2 ( Q) Space 249
8.1.1 Preliminaries . . . . . . .249
8.2 Delta Nets . . . . . . . . . . . . 256
8.2.1 Preliminaries . . . . . . 256
8.3 The Surface Integrals . . . . . . 260
8.3.1 Preliminaries . . . . . . 260
8.4 The Fourier Transform . . . . . 267
8.4.1 Preliminaries . . . . . . 267
9 Generalized Derivative and Sobolev Spaces 279

9.1 Generalized Derivative . . . . . 279
9.1.1 Preliminaries . . . . . . 279
9.2 Sobolev Spaces . . . . . . . . . 285
9.2.1 Preliminaries . . . . . . .285
10 Some Elements from Functional Analysis 303

10.1 Hilbert Space . . . . . . . . . . 303
10.1.1 Preliminaries . . . . . . 303
10.2 The Fredholm Alternatives .. . 313
10.2.1 Preliminaries . . . . . . 313
10.3 Normed Vector Spaces . . . . . 321
10.3.1 Preliminaries . . . . . . 321
11 Functional Analysis Methods in PDEs 329

11.1 Generalized Dirichlet Problem . 329
11.1.1 Preliminaries . . . . . . . 329
11.1. 2 Examples and Exercises . 330
11.2 The Generalized Mixed Problems 355
11.2.1 Examples and Exercises . 355
viii CONTENTS
11.3 Numerical Solutions . . . . . . . 366

11.3.1 Preliminaries . . . . . . 366
11.4 Miscellaneous . . . . . . . . . . 368
11.4.1 Preliminaries . . . . . . 368
12 Distributions in the theory of PDEs 373

12.1 Basic Properties . . . . . . . . . 373
12.1.1 Preliminaries . . . . . . .373
12.2 Fundamental Solutions . . . . . 390
12.2.1 Preliminaries . . . . . . .390
12.2.2 Examples and Exercises .390
Bibliography 397
Index 401
Preface
The book Partial Differential Equations through Examples and Exercises has evolved
from the lectures and exercises that the authors have given for more than fifteen years,
mostly for mathematics, computer science, physics and chemistry students. By our best
knowledge, the book is a first attempt to present the rather complex subject of partial
differential equations (PDEs for short) through active reader-participation. Thus this book
is a combination of theory and examples.
In the theory of PDEs, on one hand, one has an interplay of several mathematical
disciplines, including the theories of analytical functions, harmonic analysis, ODEs,
topology and last, but not least, functional analysis, while on the other hand there are
various methods, tools and approaches. In view of that, the exposition of new notions and
methods in our book is "step by step". A minimal amount of expository theory is included
at the beginning of each section Preliminaries with maximum emphasis placed on well
selected examples and exercises capturing the essence of the material. Actually, we have
divided the problems into two classes termed Examples and Exercises (often containing
proofs of the statements from Preliminaries). The examples contain complete solutions,
and also serve as a model for solving similar problems, given in the exercises. The readers
are left to find the solution in the exercises; the answers, and occasionally, some hints,
are still given.
The book is implicitly divided in two parts, classical and abstract. In the first (classical)
part, the necessary prerequisites are a standard undergraduate course on ODEs, on
Riemann's multiple and surface integrals and, of course, on Fourier series. For the second
(abstract) part, it would be desirable that the reader is familiar with the elements of
Lebesgue integrals and functional analysis (in particular, Hilbert spaces and operator
theory). We tried to make the book as self-contained as possible. For that reason, we also
included in the Preliminaries and Examples some of the mentioned mathematical tools
(see, e.g., elementary proofs of the Closed Graph Theorem, Adjoint Theorem and Uniform
Boundedness Theorem in Chapter 10).
Many different tools are presented for solving important problems with the basic three
partial differential equations: the wave equation, Laplace equation, heat equation and their
generalizations. We also give the usual three types of problems with PDEs: initial value
problems, boundary value problems and mixed type (eigenvalue) problems. For the
solutions of the stated problems, we discuss the three important questions: existence,
uniqueness, stability (continuous dependence of solutions upon data). We investigate also
three important questions for the solutions of PDEs mostly for applications: construction,
regularity and approximation.
We present, among other tools, the three principal methods for solving the stated
ix
x PREFACE
problems: Fourier method, Green's function and the energy (variational) method. One of
the very useful constructive techniques the Fourier method of separation of variables, is
applied first in Chapter 4 for hyperbolic equations with respect to the classical Fourier
series, where the eigenfunctions are the sine and cosine functions. In the next step, we
generalize this method through the Sturm-Liouville problem also with respect to other
systems of orthogonal functions, e.g., Legendre polynomials and Bessel functions. The
Fourier method is applied also in Chapters 5 and 6 to elliptic and parabolic equations,
respectivily. This theoretical background for these methods is obtained in Chapters 10 and
11 in the language of functional analysis through special spaces as, e.g., Sobolev spaces,
with generalized eigenvalues and eigenfunctions. The Fourier analysis is completed in
Chapter 8 by the Fourier transform.
Most of the book is devoted to second or higher order PDEs. However, for
completeness, Chapter 2 treats first order PDEs.
In the last Chapter we present a part of the distribution theory, which also covers the
theory of Dirac's delta distribution ("delta function").
The majority of the problems are of mathematical character, though we often give
physical interpretations (see sections at the ends of Chapter 1, 3, 4, 5 and 6). The
numerical approximations and computation of the stated problems are presented in
Chapter 7, with an abstract theoretical background in Chapter 11.
The book is prepared for undergraduate and graduate students in mathematics, physics,
technology, economics and everybody with an interest in partial differential equations for
modeling complex systems.
We have used Mathematica and Scientific Work Place 2.5. for some calculations and
drawings.
We are grateful to Prof. Olga Hadzic for her numerous remarks and advice on the text,
and to Prof. Darko Kapor on his useful suggestions on the physical aspects of PDEs. Dr
Dusanka Perisic made some contributions to Subsections 3.2 and 10.2 and has prepared
the Figures 4.1-4.4. It is our pleasure to thank the Institute of Mathematics in Novi Sad
for working conditions and financial support. We would like to thank Kluwer Academic
Publishers, specially to Dr Paul Roos and Ms Angelique Hempel for their encouragement
and patience.
Novi Sad, April 1997 ENDRE PAP

ARPAD TAKACI
DJURDJICA T AKACI
List of Symbols
N set of natural numbers

Z set of integers
Z+ set of non-negative integers
R set of real numbers
Rn n-dimensional real Euclidean space
C set of complex numbers
iRz = real part of a complex number z
~z = imaginary part of a complex number z
z A imaginary unit
AC complement of the set A
XA characteristic function of the set A
n intersection of sets
u union of sets
\ set difference
~ a-algebra of subsets of a set X
Q region of Rn
8Q border of the region Q
Q closure of the region Q
Q'ccQ closure of Q' is a subset of Q
x (XI, ... ,xn ) ERn
Ixl = /x? + ... + x~
a = (0'1, ... ,an) multi-index, where ai E Z+, i = 1,2, ... , n
z~ set of multi-indices
10'1 0'1 + ... + an
a! a1!" . an!
x'" Xfl ... x~n
liminf limes inferior
lim sup limes superior
xi
xii LIST OF SYMBOLS
Dj
o
OXj
D='V (D1, ... ,Dn )
02 02
~ --2 + ... + - - 2 (Laplace operator)
ox} oXn
(fIg) the scalar product
suppf support of a function or distribution
Ck(Q) space of continuous functions on Q with continuous
derivatives of order:::; k
COO(O) space of infinitely differentiable functions over an open set 0
C;'(O) space of infinitely differentiable functions over an open set 0
with compact support
L 2 (Q) space of measurable functions f with fQ If(x)i2 dx < 00
Ilfll2(Q) (JQ If(x)12 dx)1/2
Wk(Q) Sobolev space of order k
Ilfllwk(Q) ( L fQ IDCXf(xWdx)1/2
Icxl~k
ok ---:=;=:-
W (Q) Sobolev space which is C(f(Q)
(closure with respect to II IIwk(Q))
8 delta distribution (" delta function")
'D(O) space of test functions over the open set 0
'D'(O) space of distributions over the open set 0
Chapter 1
Introduction
1.1 Basic Notions

1.1.1 Preliminaries
A partial differential equation (briefly PDE) for a function U = U(XI, ... , Xn) IS a
relation of the form
(1.1)
where F is a given function of the independent variables Xl, ... ,X n , n > 1, and of
the (unknown) function U of a finite number of its partial derivatives. The order of
the PDE (1.1) is the order of the highest derivative that occurs.
A subset Q of R n is a region if it is open and connected. A function U =
U( Xl, ... ,X n ) is a solution of (1.1) on the region Q eRn if after substitution of u
and its partial derivatives in (l.1) the relation (l.1) is satisfied identically for all
(Xl, ... , xn) from the region Q.
The vectors (n-tuples) 0' = (O'lJ ... , O'n), whose components are non-negative
integers O'k, are called multi-indices. The set of all multi-indices is Z+. For 0' E Z+
and x ERn we define
We are using the convention 00 = 1. A partial order in Z+ is defined by
0' 2': (3 whenever O'k 2': (3k for k = 1, ... , n.
We introduce the differentiation symbol Dk = aD , and by a gradient vector of

Xk
differentiation D = (D I , ... , Dn) we define the gradient Du = \lu of a function
E. Pap et al., Partial Differential Equations through Examples and Exercises

Kluwer Academic Publishers 1997
2 CHAPTER 1. INTRODUCTION
U = U(Xl, ... , Xn) as the vector
The general partia.l differential operator DOl of order m = 10'1 is then
a
lOlI
DOl - DOl DOl" - _ _ _-.,-_
- l n - aXlOll ...
a
x~n
A PDE is linear if it is linear in the unknown function and its derivatives, with
coefficients depending on the independent variables Xl, . , X n . A linear PDE of
order m is of the form
L a",DOIu=F,
lal:5 m
where au and F are given functions depending on x = (XI, .. . ,xn ).
A PDE of order m is said to be quasi-linear if it is linear in the derivatives of
order m with coefficients depending on the independent variables Xl, ... , X n , and on
the unknown function and its derivatives of orders strictly smaller than m.
Classification of problems with PDEs.
1) Cauchy problem or initial-value problem. For a given PDE on a region Q,

some additional initial-values for the unknown function and its derivatives are
given on some subsets of Q.
2) Boundary value problem. For a given PDE on a region Q some additional

boundary conditions for the unknown function and its derivatives are given on
the boundary aQ.
3) Mixed type problem. For a given PDE on a region Q both some additional
initial-values and boundary conditions for the unknown function and its deriva-
tives are given.
A problem with a PDE is well-posed in a class of functions C, if the following

three conditions are satisfied
(i) there exists a solution in C;
(ii) the solution is unique;
(iii) the solution is continuously dependent on the given conditions, e.g., initial-
values, boundary conditions, coefficients, etc.
1.1. BASIC NOTIONS 3
1.1.2 Examples and Exercises

Exercise 1.1 Show that the function u( x, y) = y2 + x is the solution of the Cauchy
problem on R 2
au
ax = 1 and u(O,y) = y2 (y E R).
400
300
200
100
o
-20 -20
10
'2if20
Figure 1.1 u(x,y) = y2 +X
Exercise 1.2 Show that the function u(x, y) = cos v'X2 + yZ is the solution of the
Cauchy problem
(::f + (::f = l-u 2 and u(O,y) = cosy.
t
CUi
(1'
-0.5
-i-
..s . 6
4,.,.
-'" 0'
Y 2 4'.0 ."6 ..
88
Figure 1.2 u(x,y) = cosv'x 2 + y2
Exercise 1.3 Show that the function u( x, y) = -x 2/2 + y2/2 is the solution of the
equation
~ 4 4
Figure 1.3 u(x, y ) = -"21 x 2 +"21Y 2
Exercise 1.4 Show that the function u(x, t) = (x + ct)4 + (x - ct)3, where c is a
positive constant (Figure 1.4 for c = 3), is a solution of the wave equation
(Pu 2{PU _
ae - c ax 2 - o.
2020
Figure 1.4 u(x, t) = (x + 3t)4 + (x - 3t?
Exercise 1.5 Show that the function u(Xl) ... , X n , t) = (Xl + ... + Xn + vnct)lO,
where c is a positive constant, is a solution of the n-dimensional wave equation
which, using the Laplace operator 6 given by
can be written in the shorter Jorm

fPu
ot 2 - c2 6u = O.
Example 1.6 Let J : Q --. C be an analytic Junction, where Q is a region in C.
Prove that the Junctions u(x,y) = ~J(x+zy) and v(x,y) = ':SJ(x+zy) Jor x+zy E Q
(where z is the imaginary unit, Z2 = -1) satisJy
a) the Cauchy-Riemann equations
ou
and
oy
b) the Laplace equation

02U 02U
ox2 + oy2 = 0,
if, additionally, u and v are Jrom the class C 2(Q).
Solution.
a) Since the function J is analytic, there exists
1'(z) = lim J(z + h) - J(z)

h-+O h
for z = x + zy E Q. We evaluate the preceding limit in two ways. First let
h --. 0 for real h. We have for h =I- 0
J(z + h) - J(z) J(x+h+zy)-J(x+zy)
h h
u(x+h,y)-u(x,y) v(x+h,y)-v(x,y)
= h +z h .
Taking h --. 0 we obtain
J '( ) _ ou(x, y) ov(x, y)

(1.2)
z - ox +z ox .
Now let ih --. 0 for real h. We have for h =I- 0
J(z + zh) - J(z) J(x + z(h + y)) - J(x + zy)

zh zh
u(x,!} + h) -u(x,y) v(x,y + h) - v(x,y)

= -1 + h .

h ,
Letting h -+ we obtain
f '( ) - _ 8u(x,y) 8v(x,y)

z - Z 8y + 8y . (1.3)
Since both the real and imaginary parts of (1.2) and (1.3) must be equal, we
get the Cauchy-Riemann equations.
b) Differentiating the first Cauchy-Riemann equation with respect to x and the

second one with respect to y we obtain
82 u 82 v
and
8x 2 = 8x8y
Adding the obtained equalities we get that the function u satisfie~ the Laplace
equation. Differentiating now the first Cauchy-Riemann equation with re-
spect to y and the second one with respect to x and repeating the preceding
procedure we get that the function v satisfies the Laplace equation, too.
Example 1.7 Show that the function u = u( x, t), given by
u(x, t) = ae- t / a2 sin:::

a
for a real constant a i= 0, is a solution of the one-dimensional heat equation
8u 8 2u

8t = 8x2' which for t ~ converges to zero as a -+ 0, but does not for t < 0.
Solution. For the given function it holds
8u(x, t) 1 _t/a2 . x
= --e sln-
8t a a
and
au(x, t) = e- t/ a2 cos::: 8 2 u(x, t) = _~e-t/a2 sin:::
8x a' 8x 2 a a
. t he d . 8u ( x, t) an d 8 U ( x, t)
in t he heat equation, we obtain t h at
2
P uttmg envat!ves 2
8t 8x

it is satisfied identically for all x, t E R.
We have for t ~
lim u(x t)
a ........ a '
= lim e- t / a
a-+O
2
.
sm -
__
a . X
::..
x
= '
a
see Figures 1.5 and 1.6.
For t < 0 we have lim u(x, t) = +00, see Figures 1.7 and 1.8. Thus the last result
a~O
shows that the Cauchy problem for the heat equation is not well-posed.
10 20
x
Figure 1.5 Case a = 5 u(x t) = 5e- t / 25 sin - t >0
" 5'
-2
4
Y 6 2
8
10 4
Figure 1.6 Case a = 0.5, u(x, t) = 0.5e- t / O.25 sin 2x, t> 0
~20
. -10
30
Y 20
-10 10
020
Figure 1.7 Case a = 5, u(x, t) = 5e- t / 25 sin K' t < 0
-2
2
o <4
Figure 1.8 Case a = 0.5, u(x, t) = 0.5e- t / O.25 sin 2x, t < 0,
10- 8 :::; u:::; 108
Example 1.8 (Hadamard) Show that the function tt = tt(x,y) given by
eny _ c- ny
U(x,y) = 2n
2 sinn.x
for n E N is a solution on D = {(x, y) I x 2 + y2 < I} of the Cauchy problem for the

Laplace equation
OU(x,O) sm nx
tt(x,O) =0,
oy n
Prove that this problem is not well-posed.
Solution. It easy to check that the function Un for an arbitrary but fixed n E N
given by
e ny _ e- ny
ttn(x,y) = 2 sinnx
2n
is a solution of the given Cauchy problem. Letting n we obtain for x, y ED,

- t 00
x =I- 0, y =I-
On the other side, the given Cauchy problem for n - t 00 reduces to the problem
U(x,O) =0, OU(x,O) = 0,

oy
which has only a trivial solution U = 0. Therefore the considered solution of the
given Cauchy problem does not depend continuously of the initial condition. Hence
it is not well-posed (it is ill-posed).
Remark 1.8.1. In the contrast with this simple problem with a PDE which is
not well-posed, let us remind some well-known results from the theory of ordinary
differential equations where for general classes of problems with ordinary differential
equations the well-posedness can be ensured. For example, the well-posedness of
the Cauchy problem
y' = f(x, y), y(xo) = Yo
is ensured supposing that the function f
is continuous and satisfies the Lipschitz
condition in some region which contains the point (xo, Yo).
Example 1.9 Find the solution U = u(:1:, y) of the following problem on the set R2.
(1.4 )
au(O,y) = 3y2 u(x 0) = 0 (1.5 )

ay " .
Solution. The equation (1.4) can be written in the form
~ (au
ay ax
_ u) - 4
-,
which after integration with respect to y reduces on the equation
where FI is an arbitrary differentiable function. The solution of this equation is
u(x,y) = eX F2(y) - 4y + eX (X e-tF1(t) dt, (1.6)

Jxo
where F2 is an arbitrary differentiable function. Putting u from (1.6) in the first
initial condition in (1.5) we obtain
au(O,y) -F'(
- 2Y )-4-3 2
- y,
ay
which implies
(1. 7)
where C is an arbitrary real constant. Putting now u from (1.6) in the second initial
condition in (1.5) and using (1. 7) we obtain
u(x,O) = Ce x + eX (X e-tF1(t) dt = O.
Jxo
Hence the solution of equation (1.4) with the initial-conditions (1.4) is
Example 1.10 Find the general solutions of the equations

au
a) ay = 0;
0 2 11 011
C) ~+x~=O.
vxvy VY
Solution.
a) 11(X,y) = f(x), where f is an arbitrary function from Cl(R).
b) 11(X,y) = p(x) + q(y) for p,q E C 2 (R).
C) Taking the second variable fixed, say y = Yo, and introducing a function of
one varia ble w ()
x = 011(X, Yo) , the given
. . d d'
equatIOn re uces on an or mary
oy
differential equation
w' + xw = 0,
which can easily be solved by the method of separation of variables. Hence
w( x) = C e- x2 /2, where C depends from Yo. Therefore for arbitrary y we have
C = q(y), and
011= q()
- y e -x2/2
oy
Applying the integral with respect to y we obtain
11(X,y) = e- x2 / 2 J q(y) dy + f(x) = f(x) + e- x2 / 2 g(y),

where f and 9 are functions from C1(R).
Exercise 1.11 Prove that the following problem on the set R2

0 2 11
oxoy = 0,
011(X,O) _ ( )
u ( x,O ) = f ()
x , oy - 9 x ,
has a solution if and only if f E C 2 (R) and g(x) = const. Then the solution is given
by
u(x, y) = f(x) + F(y) - F(O) + y(g(O) - F'(O)),
where F is an arbitrary function from C 2(R).
Exercise 1.12 Prove that the equation

02U 02U a 011
ox 2 - ol + -; ax = 0,
taking the substitution v( x, y) = X a / 211( x, y), reduces on the equation
Exercise 1.13 Prove that the heat equation
au a au
2
at = i8x2 (x E R,t > 0),
a real constant, taking the substitution u( x, t) = exp( -v( x, t) / a) reduces to the BUTY-
ers equation
1.2 The Cauchy-Kowalevskaya Theorem

1.2.1 Preliminaries
Definition 1.1 A function f : 0 ....... R, where 0 is an open set in R n, is called real
analytic at XO EO if there exist COl E R (0: E Z+) and a neighbourhood V(xO) of XO
such that
f(x) = L c",(x - xO)"
"'EZ+
for all x E V(XO nO). The function f is called real analytic in 0 if it is real analytic
at every point xO from O.
Theorem 1.2 (Cauchy-Kowalevskaya) Let a~k and bj be real analytic functions

depending on z = (Xl, ... , Xn-l, Ul, . . . , us) at the point
Z - - (X1,... ,Xn_1'Ull
... 'U.0) E R n +s - l, . - 1 , ... ,n-,
%- ,) = 1, ... ,8,
l' k'
and let <pj = <Pj(XI,,,,,Xn-l) be real analytic functions at (x~,,,,,X?_I) for j

1, ... , s. Then the system of quasi-linear PDEs
for j = 1, ... , s, with the initial conditions
for j = 1, ... , s, has a unique solution (Ul, ... , us) which is real analytic at the point
xo.
1.2. THE CAUCHY-KOWALEVSKAYA THEOREM 13

Exercise 1.14 For a = (aI, ... , an) E Z~ and x = (Xl, .. , X n ) ERn prove
\
a) a!::::; lap::::; nl"la!;
Exercise 1.15 Prove thai for f,g E cm(Rn) it holds

,
D"(J g) = 2.:: 7' ,DiJ fD'g,
iJ",iJ+-y=a fJ,
where lal = a1 + ... + an = m.
Exercise 1.16 Let 0 be an open subset of the set Rn. Prove that a function f is
real analytic on 0 if and only if f E COO(O) and for every compact set J( c 0 there
exist M > 0 and r > 0 such that for every y E J( we have
IDiJ f(y)1 ::::; MlfJl!r- 1iJl (fJ E Z~).
Example 1.17 Prove that there is no real analytic solution in (0,0) of the equation
a u au
2
-----0
oxoy oy - ,
which would also satisfy the initial-condition
1
u(x,O) = --2'
l+x
Solution. If there were a real analytic (in (0,0)) solution
00 00
u(x,y) = 2.::2.::ai,jxiyj, (1.8)

i=lj=l
then the coefficients ai,j would be given by
_ (2s + 2k)!(_ )k+s

a2s,k - (s)!k! 1
But then the series (1.8) would not converge in any neighborhood of (0,0), since it
is divergent for all (0, y), YolO.
Remark 1.17.1. The reason for this is the fact that y = 0 is a characteristic of the
given equation (see more in Chapter 3.).
Example 1.18 Prove that the equation
au au au
ax +zay +2z(x+zY)az =f(z),
for a given function f which is not analytic, has no C1-solution u = u(x,y,z) m

any neighborhood of(O,O,O) in R3.
Solution. We give only the sketch of the proof. Suppose the contrary, i.e., that the
given equation has a solution u E C 1 (5), where
for some R >

and M > and
(-M, M). Introducing the function
f
Ul
is a non-analytical function on the interval
by Ul(P, tp, z) = u(p cos i.p, P sin tp, z), we have
that the function v defir,ed by
is a solution of the equation
av av
-ar + laz
- = 7rf(z)
on the set P =
{(r,z)i < r < R2, Izl < M}. Therefore the function w given by
w(r, z) = v(r, z) = v(r, z) + l7r f f(t) dt
is an analytic function on P and continuous on P with respect to r + lZ. By the

Schwarz Reflection Principle (see [21]) the function w as a function of complex
variable r+lZ can be analytically extended on {(r, z)ilrl < R2, Izl < M}. Therefore
the function
w(O,z) =m lZ f(t) dt
is analytic for Izl < M. Hence the function f is also analytic for Izl < M. A
contradiction.
Remark 1.18.1. It is interesting that the given linear equation of the first order
(Levy example from 1957) has no solution even in the space of distributions V' (see
Chapter 12). On the other side this equation has a weak solution in the Colombeau
space 9(Rn) of the so called new generalized functions (see [5]).
1.3. EQUATIONS OF MATHEMATICAL PHYSICS 15
1.3 Equations of Mathematical Physics

This section is devoted to a brief presentation of some problems in mathematical
physics which are modeled by some partial differential equations.
We start with some equations of motion from mechanics.
The equation of vibrating string is given by
where u = u(x, t) is the displacement of the point of the string with the abscissa x
at the time t, T is the tension, p is the linear density and q = q(x, t) is the external
load per unit length.
The equation of longitudinal oscillations of a rod of constant cross section is
given by
where u = u(x, t) is the displacement of the cross section of the rod with abscissa x
at the time i, E is Young's modulus and p is the density.
The equation of a vibrating membrane is given by
q(x, y, t)
T
where u = u( x, y, t) is the displacement of the point (x, y) of the membrane at the

time i, T is the tension per unit length of the boundary of the membrane, p is the
surface density and q = q(x, y, t) is the external load per unit area.
The equation of the flow of heat in a body is given by
where u = u(x,y,z,t) is the temperature at the point (x,y,z) at the time i, k the
thermal conductivity of the body, c is the specific heat, p is the heat density and Q
is the density of heat sources within the body.
The equations of the electrostatics are given by
flu = _ 47rp , E=-Vu,

E
where u is the potential of the electrostatic field E, p is the volume density of charge
at the considered point, E is the dielectric constant of the medium.
The Maxwell equations arc given by
div(c:E) 471" p, div(JiH) = 0,
18(fLH)
rotE -----
c at
rotH
where E = E(x, i) is the electric field vector, H = H(x, t) is the magnetic field
vector, c: is the dielectric constant, Ji is the magnetic permeability, c is the velocity
of the light, p is the charge and
d. a a 8
IV = -;-
UXI
+ -;-
UXz
+ -8
X3
'
for the definition of rot see Section 2.2.1. The Schrodinger equation in quantum
mechanics is given by
h a1jJ h2
l - - = ---,6.1jJ + V1jJ,
271" at 471"ma
where 1jJ = 1jJ(x, i) is the wave function of a particle of mass rna in the field of the
potential V and h is the Planck constant.
The equations of hydrodynamics are given by
~~ + div(pV) = f (continuity equation),
8-; + (VIVV)V + ~Vp = F (Euler movement equation),

ui p
where V = V(x,i), x = (XI,X2,X3) is the velocity vector of the fluid movements,
p = p(x, i) is the fluid density, f = f(x, i) the source density and where p = p(x, i)
is the pressure and F = F(x, i) is the massforce.
Chapter 2
First Order PDEs
2.1 Quasi-linear PDEs

2.1.1 Preliminaries
A first order PDE is an equation of the form
F ( XI,X2, ...
au au
,Xn,U,~,~, ...
au)
,~ =0, (2.1 )
UXI UX2 UXn
where F is a given and u = U(XI' X2, ... , xn) is the unknown function of n indepen-
dent variables Xl, X2, ... , X n A solution of (2.1) is called integral surface.
Let L be a smooth curve in the Euclidean space Rn+l given by
(7 E I), (2.2)
where I is a real interval. Cauchy's problem for the first order PDE (2.1) is a problem
to determine its solution (i.e., its integral surface) u which passes through the curve
given by (2.2). Such a solution is then called Cauchy's integral for the problem (2.1)
and (2.2).
Quasi-linear first order PDE is an equation of the form
au + au + ... + au
PI - P 2- Pn- = R, (2.3)
OXI OX2 OXn
where u = U(XI' X2,"" xn) is the unknown function, while Pi = Pi(XI, X2,"" X n , u),
i = 1,2, ... , n, and R = R(XI' X2,"" X n , u) are given continuously differentiable
functions on a region in Rn+1. If, moreover, for every i = 1,2, ... ,n the functions Pi
depend only on the independent variables .. , X2, . .. , X n , then (2.3) is a linear first
order PDE.
We associate to (2.3) the following symmetric system of ordinary differential
equations (shortly ODEs):
(2.4)
17

18 CHAPTER 2. FIRST ORDER PDEs
If
1jJi = 1jJi(Xl' X2, .. , x n , u) = Ci (i = 1,2, ... , n), (2.5)
are the functionally independent solutions of (2.4), the so called first integrals,
where Ci, i = 1,2, ... , n, are arbitrary constants, then the general solution of (2.3)
IS
(2.6)
where \lI is an arbitrary function from the class Cl(Rn). (See Example 2.1 below for
a relation between an arbitrary system of first order ODEs and its first integrals.)
The characteristic curve of a first order PDE is a curve L with the property
that infinitely many integral surfaces pass through it. In the case of quasi-linear
equation (2.3), the characteristic curves are given by the first integrals (2.5) of the
system of ODEs (2.4). One sees at once that in the geometrically most interesting
case n = 2, any integral surface is constituted of characteristic curves.
Throughout this chapter, in the case n = 2, we shall always denote by x and y
the independent variables, and by z = z(x,y) the dependent variable. Also, it is
usual to put
az az
p= - and q = -
ax ay
for the first order partial derivatives of the function z.

Tn this subsection, C, Cl , C 2 , . denote arbitrary real constants.
Example 2.1 Let Q be a region in R and suppose the following system of ODEs
dY
dx = f(x,Y) (x E Q) (2.7)
is given. In (2.7), Y = (Yl(X), Y2(X), ... , Yn(x)) is the unknown vector-function and
f = (!I(x, Y), h(x, Y), ... , fn(x, Y)) is the given vector function such that every fi'
i = 1,2, ... , n, is a continuous function which also satisfies the Lipschitz condition.
Then the first integral of (2.7) is any relation between the variables x and Y, provided
it is not identically equal to a constant, but is equal to a constant if Y = Y(x) is a
solution of (2.7). Prove
a) A continuously differentiable function 1jJ = 1jJ(x, Yl, ... ,Yn) t= 0 is a first integral
of the system (2.7) iff it holds
a1jJ + a1jJ fl + a1jJ h + ... + a1jJ fn =0 (2.8)

ax aYl aY2 aYn.
identically on Q, when (Yb ... , Yn) =(Yl (x), ... , Yn (x)) is a solution of (2.7).
2.1. QUASI-LINEAR PDEs 19
b) The general solution of (2. 7) is determined with n first integrals, say 1/;1, 1/;2, ... ,1/;n,
,
satisfying the condition
8(1/;1J'" ,1/;n) i
8(Yl,,,,Yn)
i.e., if they are functionally independent.
Solution.
a) Suppose 1/;(x, Yb Y2,"" Yn) = C is a first integral of the system (2.7). Then it
follows from the chain rule
d 81/; 81/; ' ) 81/; I ( )

O=-d 1/;(x,YI,,,.,Yn)=-8 +-8 Yl(X +"'+-8 YnX'
x X YI Yn
Since (YI,Y2,,,.,Yn) is a solution of (2.7), the relation (2.8) follows. Con-
versely, if (2.8) holds, then
d 81/; 81/; 81/; I ( )

-d 1/;(X,YI, ... ,Yn)
x
= -8x
+ -8 YI
~
I (
x) +". + -8 Yn x
~
= o.
Hence if (YI, Y2,.", Yn) = (YI(X), Y2(X),.", Yn(x)) is a solution of (2.7), then
it holds
b) The statement immediately follows from the Implicit Function Theorem.
Example 2.2 Solve the following Cauchy problems for the linear first order PDEs:
a)
8z
Y-
8z
8x - x8y
- -- , z(x,O) = X2;
8z 8z
b) x 8x + Y 8y = 0, z(x, 1) = Ixl
Solutions.
a) Let us determine firstly the general solution of the given equation. One of the
equations of the associated system of ODEs (2.4) is
dx
Y -x
dy
, or -J x dx = J
Y dy,
which gives the first integral
x 2 + y2 = CI (C1 an arbitrary real constant). (2.9)

20 CHAPTER 2. FIRST OR1)ER PDEs
Since clearly z = C2 is anothcr first integral of the given PDE, the general
solution is
z = cp(x 2 + y2),
where cp is an arbitrary function from C 1 (R). The initial condition gives
x 2 = z(x,O) = cp(x 2), hence cp(x) = x.
Thus the Cauchy's intcgral is z( x, y) = x 2 + y2.
b) In this case, one of the equations of the system (2.4) is
dx dy -
- = -, or In Ixl = In Iyl + In C
x y
for an arbitrary constant C. Hence Ixi/lyl = C I , C I a real constant, is a first
integral of the associated symmetric system of ODEs. As in part a), the
general solution is z = cp(lxl/lyi), where cp E C 1 (R) is an arbitrary function
and y =f=. 0. The initial condition implies
Ixl = z(x, 1) = cp(lxl/1), which gives cp(t) = t.
Thus the Cauchy's integral is
z(x,y)
Ixl (y
= TYT =f=. 0).
Example 2.3 Find the equation of the surface which satisfies the linear first order
PDE
4yzp + q + 2y = 0,
and passes through the ellipse
y2 + z2 = 1, x + z = 2. (2.10)
Solution. The corresponding system of ODEs is

dx dy dz
4yz 1 -2y'
whose first integrals are
(2.11)
Thus the general solution is
(Z+y2,X+Z 2) =0,
where is an arbitrary function of the class C 1 (R2). Eliminating the variables x, y
and z from (2.10) and (2.11), gives C1 + C 2 = 3, and thus the Cauchy integral is
x + y2 + z + z2 = 3.
Exercise 2.4 Let the first order linear PDE
xp + yq =z- x 2 _ y2
be given.
a) Find the general solution of this equation.
b) Find the Cauchy integral through the curve x = 1, z = _y2 + 2y - 1.
Answers.
a) z = r.p(xjy)..JXI+Y'l- x 2 - y2 (r.p E C 1 (R)) .
b) z=-x 2 -(y-l)2+1.
Exercise 2.5 Find the solution of the PDE
which passes through the parabola y = 0, x 2 = zj2.
Answer. The general solution of the given PDE is
and the Cauchy integral is
Exercise 2.6 Solve the Cauchy problem
x + y = 0, z = 1.
Answer. The general solution is
and the Cauchy integral is
2xyz + x 2 + y2 + 2 - 2z = O.
Example 2.7 Find the surfaces whose each tangent plane meets the x-axis in a
point whose x-coordinate is one half of the x-coordinate of the point in which the
tangent plane was set.
Solution. The equation of the tangent plane of the surface z = z(x,y) in its point
M(x, y, z) is
p(X - x} + q(Y - y) = Z - z,
where (X, Y, Z) are the coordinates of a point on the tangent plane. The given
condition implies X = x/2, Y = 0 and Z = 0, hence the PDE of these surfaces is
px + 2qy = 2z.
Note that it is a linear PDE of first order. Thus the required surfaces are of the
form
z=y<p(x2Iy),
for an arbitrary function <p E Cl(R).
Example 2.8 Find the surfaces whose tangent planes pass through the point (0,0,1).
Solution. Firstly, a plane

Ax +By+Cz = D (2.12)
passes through the point (0,0,1) iff it holds
A 0 + B 0+ C 1 = D, hence C = D.
Secondly, the plane (2.12) is a tangent plane of a surface z = z(x,y) iff the vectors
n(A, B, C) and (p, q, -1) are parallel, where the first order partial derivatives p and
q are evaluated in the meeting point of the surface and its tangent plane. Thus we
can take p = -AIC and q = -BIC, provided that C i- o. So we obtain the PDE
of these surfaces:
px + qy = z -1,
whose general solution is
y z-
cI> ( - , - -
1) =0, or z = 1 + x E C 1 (R2) and <p E C 1 (R) are arbitrary functions of their variables.
Example 2.9 Find the general equation of the surfaces S which are perpendicular
to each surface from the family
Z2 = kxy (k E R). (2.13)
Solution. If we denote by 0 = (p,q, -1) and 01 = (Pl>ql, -1) the normal vectors of
the surface S and one from the given family, then by the given condition they should
be perpendicular, i.e., 0 .1 01. The last condition holds if their scalar product is
zero, i.e.,
PPI + qql + 1 = O. (2.14 )
Now differentiating in x and y, respectively, equation (2.13), gives
ky kx
PI = - and qi = - .
2z 2z
Hence from (2.14) it follows
k(py + qx) = -2z,
and replacing the parameter k from (2.13) gives the PDE
pyz2 + qxz 2 + 2zxy = O. (2.15)
The first integrals of the associated system of ODEs (2.4) are
and the general solution of (2.15) is
(x 2 - y2,2x2 + Z2) = 0 ( E C l (R2)).
Example 2.10 Find the integral surface of the PDE
pxz + qyz = -xv,
which passes through the curve
(t E R). (2.16)
Solution. The first integrals of the given PDE are
x/y=Cl and z2+ xy =C2 , (2.17)
hence its general solution is (;., Z2 + x y) = O. Replacing x, y and z from (2.16)
into (2.17) gives
1
Cl = - and C 2 = t S + e.
t
Eliminating the parameter t from these equations gives the first integral
Example 2.11 Let M be a point on a surface, P its orthogonal projection to the

xv-plane and N the intersection of the xv-plane and the perpendicular line of the
surface at the point M.
a) Find those surfaces for which it holds LNOP = 45, where 0(0,0,0) is the origin.
b) From the surfaces obtained in a), find the one which passes through the x-axis.
Solutions.
a) The perpendicular line of a surface z = z(x, y) passing through its point M(x, y, z)
has the equation
X-x y -y Z-z
~,
p q
where X, Y and Z are the coordinates on the perpendicular line. Hence the
coordinates of the point N are (x + pz, y + qz, 0). Denote by 0: and (3, respec-
tively, the angles LNOx+ and LPOx+, where x+ is the positive direction of
the x-axis; by assumption, it should hold 0: = (3 + 45. Since it holds
Y y+qz y
tan 0: = - = ---, tan (3 = -,
X x+pz x
the required PDE is
p(zx + zy) + q(zy - zx) = _x 2 _ y2. (2.18)
One first integral of (2.18) is
Jx 2 + y2 earctan(y/x) = C1 ,
while the second first integral we get from the ODE
dz xdx + ydy
- x 2 + y2 - z( x2 + y2) ,
and it is C 2 = x 2 + y2 + Z2. Thus the general solution of (2.18) is
x 2 + y2 + Z2 = W( Jx 2 + y2 earctan(y/x) , (2.19)
or, in polar coordinates, x = p cos </>, y = p sin </>,
b) Since the equation of the x-axis is y = z = 0, from (2.19) it follows x 2 = W(x).

Hence the required surface has the equation
x 2 + y2 + Z2 = (x2 + y2). e2arctan(y/x).
Example 2.12 Find the equation of the surfaces which have the property that in
arbitrary point (x,y,z) their tangent planes determine on the x-axis a segment
which depends on a prescribed way from y/x. Compile a second order PDE which
these surfaces satisfy.
Solution. Similarly as in Example 2.7, the tangent plane of the surface z = z( x, y)

at the point (x, y, z) is given by
p( X - x) + q (Y - y) = Z - z,
where (X, Y, Z) is a point on the tangent plane. For X = 0 and Y = 0 by supposition

it follows Z = f(y/x), where f is a given function. Thus we obtain
z - px - qy = f CfJ . (2.20)
The general solution of (2.20) is z = x<p(x)+ f(y/x), where <p is an arbitrary function
of class Cl(R). If we differentiate (2.20) in x and in y, we get
rx 2 + 2sxy + ty2 = 0,
where
fj2z fj2z
s = oxoy and oy2'
Example 2.13 Let f E C 2 (R) be a given function.
a) Form a PDE satisfied by the function z = z(x, y), if the expression
oz +y ) dx+ ( Xoy
( xox oz +x-2xy+2yf(z-x)
, ) dy (2.21 )
is a total differential.
b) Find the solution of this equation which for x = 0 becomes z = y2. For such
function z find the function ( whose total differential is the expression (2.21).
Solutions.
a) From the condition
x-+y ) = -0
-o (oz ( x-+x-2xy+2yf(z-x)
oz ,)
oy ox ox oy
it follows
2y f "( z - x )oz
ox oz
+ oy = 2y ( 1 + f "( Z - x )) .
The general solution of the last PDE is

26 CHAPTER 2. FIRST ORDER PD&
b) The solution is
and the function ( is given by
x2
((x,y) = 2" + xy + f(y2) + C,
where C is an arbitrary constant.
Example 2.14 Let a family of surfaces be given with the equation
where a, band c are given real numbers. Determine all the orthogonal trajectories
for the given family of surfaces. Then, in particular, find those trajectories that pass
through the curve x = 1, z2 = ya.
Solution. Put
f(x,y, z) = ax 2 + by2 - ,\Z2 = c.
If z = <p(x, y) is the sought after trajectory, then it holds
of of of
p-+q---=O.
ox oy oz
This gives the PDE

axzp + byzq =c- ax 2 - by2,
where CP E C 1 (R) is an arbitrary function of the class C 1 (R2). The sought after
particular solution is
Example 2.15 Find the general solution of the PDE
oz oz
(1 + v'z - x - y) -
ox
+ -oy = 2,
and check whether there exits a solution which is not included in the obtained general
solution.
2.1. QUASI-LINEAR. PDEs 27
Solution. The associated symmetric system of PDEs is

dx dy dz
1 + Jz - :r - y 1 2
Then one of the first integrals is obtained as follows:
dy dz z
1
2' hence y = 2" + Gl ,
where Gl is an arbitrary constant. Further on we have
dx+ dy - dz dy
which gives
-d(z-x-y)
-7====:-'- = dy.
Jz - x - Y 1' Jz - x - y
Thus another first integral is
2(z - x - y)1/2 = -y + G2 ,
where G2 is also an arbitrary constant.
The general solution of the given PDE is
 (2y - z,y + 2(z - x _ y)1/2), (2.22)
where is an arbitrary function of the class C l (R2). However, note that
z=x+y
is also a solution of the given PDE, though it is not included in (2.22).

Remark 2.15.1 In general, the surfaces on which a partial derivative of some of
the coefficients P or Q in the linear PDE
Pp+ Qq =R
is not bounded, are good candidates for such "special" solutions.
Example 2.16 Prove that the integrability condition for the ODE
fl (P(x,y)dx+Q(x,y)dy) =0
is a first order linear PDE with the unknown function fl = fl(X, v). Then find some
cases when the integrating factor fl can be easily found.
Solution. The integrability condition gives the following PDE:
It is well known that this equation has a solution in fl if at least one of the functions
P or Q is not identically equal to zero and both P and Q are of the class Gl(R2).
Let us point out the following two cases:
a) If the expression X = -;1 (~~ - ~;) depends only on J:, i.e., X = X(x), then
the integration factor also depends only on x and is equal to
f.l(x) = exp (J X(x) dX) .

b) Similarly, if the expression Y = ~ (~~. - ~;) depends only on y, i.e., Y =
Y(y), then
f.l(y) = exp (J Y(y) dy) .
Exercise 2.17 Find the integration factor f.l = f.l(x, y) for the PDE
(2x 3 y - y2) dx - (2X4 + xY) dy = O.
Answer. f.l(x, y) = x- 2y-3.
Example 2.18 Find the solutions u = u(x, y, z) of the following linear PDEs:
au au au
a) + (z + u)-;:l
X-;:l
uX uy
+ (y + u)-;:l
uX
= y + z;
b) xy au _ ~ (y au _ z au) = xy au;
ox oy oz oz
c) (u - x)-
au + (u - y)-
au - zau- = x + y.
ox ay az
Solutions.
a) We start from the system of ODEs:
dx dy dz du
x z+u y+u y+z
Firstly we have
dz - du dy - dz z-u
which gives - - = C1 .
u-z z-y y-z
Secondly we have
dx d(y+z+u)
which gives
x 2(y+z+u)'
and, finally,
dz - du
dx
hence x(u - z) = C3 .
x u-z
Thus the general solution of the given PDE is
2
q>(Z-U, 2 x ,x(u-Z)) =0,
y-z y +z+u
where q> is an arbitrary function of the class C 1 (R3 ).
b) u = cp (xexp (arcsiny), 2yz + x (y + ~)) (cp E Cl(R2)).

Hint. The second first integral can be found from the following ODE.
(y + vr=?) dx (2Z + x - ~) dy _ 2y dz
xy (y + v'I=?) + (2Z + x _ xy ) (-yJf7) - 2y (zvr=? - Xy)"

vr=?
c) q> (
X - y
-z-, z(2u + x + y),
u- x-
Z2
y) '
Exercise 2.19 Solve by u = U(Xl, ... , xn) the first order linear PDE
n au
LXi- = XI X2 Xn
i=1 aXi
Answer. The general solution is
Example 2.20 Solve by U = u(x,y) the system of PDEs

au 2 au b 2u 2
ax = ay , ---+--ay
ay - 2y2 y ,
where a and b are given constants.

Solution. Let us solve (as an ODE) the first equation by the unknown function u
observed as a function of the variable x and treating y as a parameter. This solution
IS
u(x, y) = axy2 + G(y), (2.23)
where G is an arbitrary function from the class C 1 (R). Putting the right-hand side
of (2.23) in the second given PDE, we get the ODE
G'(y) - ~G(y)
y
=~
2y2
- ay2.
This linear ODE by the unknown function G = G(y) has the solution
where C is an arbitrary constant. Thus the solution of the given system is
u(x,y) = axy2 - ~- ay3 + Cy2.

6y
Example 2.21 Solve the following system of PDEs by the functions u = u(x,y)
and v = v(x,y) :
ou ov
y-+x-=4xy,
ox oy
and find the solution that satisfies the initial condition
(u, v) I Y =2X = (9x 2, 6x 2).

Solution. Adding the first PDE to the second one gives us
o(u+v) o(u+v)_12
y ox +x oy - xy, (2.24)
while subtracting the second PDE from the first one gives
o( u - v) _ o( u - v) _ -4
y ox x oy - xy. (2.25)
From (2.24) it follows
while from (2.25) it follows
where and III are arbitrary functions of the class Cl(R). Hence
and
1 1
v(x,y) = 4x 2 + 2'(x 2 - y2) _ 2'1lI(x 2 + y2).
The initial conditions give

and
v(x,2x) = 6x 2 = 4x 2 + ~ (<P(-3x 2) - 1lI(5x 2)).
Solving by 1> and III the last system we obtain
1>(x) = -3x and llI(x) = x,
hence, finally,
u(x,y) = x 2 + 2y2 and v(x,y) = 2X2 + y2
Exercise 2.22 Find the characteristic curves of the following PDEs:
a) z = px + qy;
b) (mz-ny)p+(nx-lz)q=ly-mz.
Answers.
a) Since the first integrals of the given equation are z/x = CI and y/x = C2 , the
two families of characteristic curves are obtained for different constants C I and
C2 .
b) The families of concentric spheres x 2 + y2 + z2 = C I and of parallel planes Ix +

my + nz = C 2 are the two families of characteristic curves. The intersections
of these spheres and planes are the circles that constitute each integral surface.
Example 2.23 Show that the Cauchy problem
2 cos x . p + 2y sin x . q = z sin x, z = 0, y = a cos x,
for a given constant a, does not have a unique solution.
Solution. One first integral is z2/ y = C I , while the other is y/ cos x = C2, where
C l and C 2 are arbitrary constants; these equations define the characteristic curves
of the given PDE. In view of the given condition, we see that the given Cauchy
problem does not have a unique solution. In fact, for any function c.p E Cl(R) such

that c.p(a) = the function z = z(x,y) given by
Z2 = yc.p (_y )
cos x
is a solution of the given problem.

2.2 Pfaff's Equations

2.2.1 Preliminaries
Pfaff's equation is an equation of the form
P dx + Q dy + R dz = 0, (2.26)
where z = z(x,y) is the unknown function, while P, Q and R are given continuously
differentiable functions of x, y and z in a region in R3. Equation (2.26) can be easily
solved in the following two cases.
1. The following integrability conditions hold:
oP = oQ oP _ oR d oQ oR
(2.27)
oy ox ' oz - ox an oz - oy'
Then there is a function u = u( x, y, z) such that
du = P dx + Q dy + R dz ,
i.e., the left-hand side of (2.26) is a total differential of a function u = u( x, y, z).
Hence for an arbitrary constant C the function
u(x,y,z) =C
is an implicit solution of (2.26).
2. The following complete integrability condition holds:
P (oQ _ OR) + Q (OR _ oP) + R (oP _ oQ) = o. (2.28)

oz oy ox oz oy ox
Then there are functions u = u(x,y,z) and v = v(x,y,z) such that
vdu = Pdx + Qdy + Rdz

(existence of an integration factor!).
It is convenient to put A = (P, Q, R). Then (2.28) can be written in the form
rotA A = 0 (2.29)
w!lere
J k
0 0 0
rotA = ox oy oz
P Q R
2.2. PFAFF'S EQUATIONS 33
If neither of the upper two cases hold, then there exist functions u, v and w such
that
Pdx + Qdy + Rdz = du + vdw, (2.30)
where v dw = Pldx + Qldy + Rldz = 0, and for the vector Al = (PI, Ql, R l ) the
complete integrability condition (2.28) holds.

Example 2.24 Solve the following Pfaff's equations.
a) (6x + yz) dx + (xz - 2y) dy + (xy + 2z) dz = 0;
b) yzdx+(XZ- yz3)dy=2xydz.
Solutions.
a) Since the equalities in (2.27) hold (integrability conditions), it follows that the
left-hand side of the given PDE is a total differential of some function u =
u(x,y,z). It can be found as follows:
u( x, y, z) = l
IO
x
(6x + y z) dx + l
Yo
Y
(xoz - 2y) dy + f\ xoyo + 2z) dz,
Jzo
or
u(x, y, z) = 3x 2 + xyz - y2 + z2 + C,
b) Let us observe first that the complete integrability condition (2.28) (or (2.29))
is satisfied, where A = (P,Q,R) = (yz,xz - yz3,-2xy) and
j k
f) f) f)
rotA ::::
f)x f)y f)z
yz xz - yz3 -2xy
The given PDE can be solved in the following way. Let us assume for a moment
that one of the variables, say y, is a constant; then we get
x
yzdx - 2xydz = 0; hence 2'
z
= cp(y).
Differentiation of the last equation gives
~ dx - 2x3 dz = cp'(y) dy. (2.31)

Z2 Z
Comparing the given PDE aIld (2.31) we see that the corresponding [unctions
(with the same differentials) must be proportional, i.e.,
2x
x:. ~ _ip/(y)
yz -2xy xz - yz3
The complete integrability condition (2.28) implies that now we should get an
equation containing only the variable y and the functions ip(Y) and ip/(y). In
fact, we have
ip' (y) X 1 ip(Y)

yz3 - xz
hence ip
I
(y) =1- -. - = 1 - --,
Z2 Y Y
and thus
<p(y) = ~ +~ (C E R).
C
x
So we obtain -Z2 - -y + -2y -- 0 , or
Exercise 2.25 Prove that for the following PDE
y dx + z dy + x dz = 0
the complete integrability condition does not hold, and find the functions u, v and w
from (2.30).
Answer. It holds
1 1
ydx+zdy+xdz=-d(xy+yz+zx)+-(x-y)
2 2
2
d --
x - y
(z - x) ,
and thus
1 1 z-x
u(x,y,z) = -2(xy + yz + zx), v(x,y,z) =
2
-2(x - y) and w(x,y,z) = 1--.
x-v
Exercise 2.26 Solve the following Pfaff's equations.
a) yzdx + 2xzdy + xydz = 0;

b) (eXy +e Z ) dx + (eYz + eX) dy +- (e Y - eXy - eYz) dz = 0;
c) (2x2 + 2xy + 2xz2 + 1) dx + dy + 2z dz = 0;

2.3. NONLINEAR FIRST ORDER PDEs 3.5
d) 2x(y + z) dx + (2yz - x 2 + y2 - z2) dy + (2yz - x 2 - y2 + Z2) dz = 0;

e) (2x 3 y + 1) dx + x 4 dy + x 2 tan z dz = 0;
f) 2(y+z)dx-(x+z)dy+(2y-x+z)dz=0;
Answers. In the solutions given below, C is an arbitrary constant.
a) xy2z=C.
Hint. Put y = const and solve the obtained ODE; then it follows t.p(y) =
C/y2, for some t.p E CI(R). Then continue as in Example 2.24 b).
b)
c) eX2 (x+y+Z2)=C.
d) x 2 + y2 + Z2 = C(y + z).
1
e) x 2 y - - -lncosz = C.
x
Hint. fleX) = -;..
X
f) y+z=C(x+z)2.
Hint. Put x = XIZ and y = YIZ. Then it holds
2dxI 2 dYI dz
-----+-=0.
Xl + 1 YI + 1 z
2.3 Nonlinear First Order PDEs

2.3.1 Preliminaries
A nonlinear first order PDE is an equation of the form
F(x,y,z,p,q) = 0, (2.32)
where z = z(x, y) is the unknown function of two independent variables x and y,

. 8u 8z
whIle p = 8z and q = 8y
Equation (2.32) has exactly three types of solutions, namely the complete, sin-
gular and general solution.
The complete solution is a solution of (2.32) which depends on two arbitrary and
independent constants, say a and b, and thus can be written as
veX, y, z, a, b) = O. (2.33)
The singular solution is a solution of (2.32) which depends neither from arbitrary
constants nor from arbitrary functions. When the singular solution exists, it is
obtained from the complete solution (2.33) by elimination of a and b from the
following system of three equations:
oV og
V(x,y,z,a,b) = 0, oa (x,y,z,a,b) =0 and ob(x,y,a,b) = o.
The general solution is a solution of (2.32) which depends from an arbitrary
function. If the complete solution of (2.32) is given by (2.33), then its general
solution is given by the system of two equations, namely
oV oV ,
V(x, y, z, a, b{a)) = 0 and oa (x, y, z, a, b(a)) + Tb(x, y, z, a, b(a)) . b (a) = 0,
where b = b( a) is an arbitrary function.

As we shall see in Example 2.27, both the singular (if any) and the general
solution of a nonlinear first order PDE can be obtained from the complete solution.
Thus our main task in solving (2.32) is to find a complete solution. It can be
obtained from the Lagrange-Charpite method exposed below.
The Lagrange-Charpite Method

Let a nonlinear first order PDE of the form (2.32) be given. The main idea of the
Lagrange-Charpite method for obtaining its complete solution, is to find a function
<P, functionally independent of F, of the form
<P(x,y,z,p,q) = a, (2.34)
where a is an arbitrary constant. The function <P should have the property that the
system of P DEs
F = 0 and <P = a (2.35)
can be solved in p and q :
p=cp(x,y,z,a) and q=,(x,y,z,a). (2.36)
Clearly, the functions cp and ' from (2.36) should satisfy the equality
(2.37)
Then Pfaff's equation
dz = pdx + qdy = cp(x, y, z,a) dx + '(x, y, z,a) dy

2.3. NONLINEAR FIRST ORDER PDEs 37
satisfies the complete integra,bility condition (2.28), and its general solution depends
on (another) arbitrary constant b. Thus it can be written in the form
V(x,y,z,a,b) = o.
This equation is then the complete solution of the given nonlinear first order PDE
(2.32).
We still have to find the function from (2.34), To that end, differentiating
both equations in (2,35) and putting p and q from (2.36) in the obtained equations,
we obtain a linear first order PDE with the unknown function , Its associated
symmetric system of ODEs (2.4) is
dx dy dz -dp -dq
(2.38)
of - of = of of - of of = of of'
op oq p op + q oq ox + p Oz oy + q oz
One can show that equality (2.36) holds iff the functions F and are in involu-
tion. By definition, this means that it holds
[F,] = O.
where
of of of of of of
-+p- -+q-
op ox oz oq oy oz
[F, ]
0 0 o + 0 o 0 (2,39 )
-+p- -+q-
op ox oz oq oy oz
In particular, if the function F from (2.32) does not depend on z, i.e., it has the
form
F(x,y,p,q) = 0
then (2.39) can be written as
of of of of
op ox oq oy
(F, o + 0 0 (2.40)
op ox oq oy
The expression (F, <I given by (2.40) is called Poisson bracket.


Example 2.27 Assume equation (2.32) can be written in the form
az ( az)
ax =f X,y,z, ay (2.41 )
aF
(for example, in a neighbourhood of a point where it holds ap =I- 0.) Then using
the method of variation of constants, prove that all the solutions of (2.32) ean be
obtained from the complete solution.
Hint. Firstly take a = a( x, y) and b = b( x, y) in the complete solution
z = g(x,y,z,a,b).
Then, by the definition of the complete solution, replacing z, a and b with the
functions g(x,y,z,a,b), :~(x,y,z,a,b) and :~(x,y,z,a,b), respectively, one gets
an identity in (2.41) for arbitrary functions a and b.
Conversely, a function z = g(x, y, z, a, b) is a solution of (2.41) only if
az ag and az ag
ax ax ay ay
Show that this will hold only if
ag aa ag ab
aa(x,y,a,b) ax + ab(x,y,a,b) ax = 0,
(2.42)
ag aa ag ab
aa(x,y,a,b) ay + ab(x,y,a,b) ay = O.
Putting :!(x,y,a,b) = and ~~(x,y,a,b) = 0, and, if possible, eliminating a

and b from equations (2.42), we obtain the singular solution. Note that it depends
neither from arbitrary constants nor from arbitrary functions.
Next, assume it holds a(a,b) -0 where
a(x, y) - ,
aa aa
a( a, b) ax ay
a(x,y) ab ab
ax ay
2.3. NONLINEtlR FIRST ORDER PDEs 39
and let a and b be some non constant functions. Then the functions a and bare
not functionally independent, i.e., there exists a differentiable function w such that
b = w(a). This leads to the general solution of (2.41).
Finally, check that any solution of (2.41) reduces to one of the following three
ones: singular, complete or general.
Example 2.28 Determine a PDE whose complete solution is a family of spheres

with the same radius R and with centers in the xy-planc, i.e.,
(2.43)
where a and b are arbitrary constants. Find then the general and the singular solution
of the obtained PDE.
Solution. Starting from the complete solution (2.43), we can reconstruct the PDE
as follows. Differentiating equation (2.43) in x and y, respectively, we get
2(x-a)+2zp=0 and 2(y-b)+2zq=0,
which after the elimination of the constants a and b gives the PDE
(2.44 )
The general solution of (2.44) is given by the system of equations
(x - a)2 + (y - b(a))2 + Z2 = R2 and (x - a) + b'(a)(y - b(a)) = 0, (2.45)
where b = b(a) is an arbitrary function of class C 1 (R). (The last equation was
obtained by putting b = b( a) in (2.43) and differentiation in a.) The geometric
interpretation of (2.45) is that it represents an envelope of the one-parameter family
of spheres given by (2.43) for b = b( a ).
The singular solution is given by the system
which, after the elimination of the constants a and b, gives simply Z2 = R2. The
geometric interpretation of the singular solution is that it represents a union of two
parallel planes, which are the envelopes of a family of surfaces (spheres) that make
the complete solut.ion.
Example 2.29 Find the complete and singular solution(s}, if any, of the following
nonlinear PDEs of first order:
a) 1 + p2 = zq;
Solutions.
a) Putting
F(x, y, Z,p, q) = 1 + p2 - zq,
we obtain
8F8F 8F 8F 8F 2
8x 8x + p 8z = -pq and 8y + q 8z = -q .
Then from system 2.38 we get

-dp -dq
_pq _q2
and thus we can take q = ap for an arbitrary constant a. Solving in p the

quadratic equation 1 + p2 - Z . (ap) = 0, we obtain
-az Va2z2 - 4
Pl,2 = 2 (2.46)
Let us take the sign + in (2.46); then it holds
J 2dz
---:--r=;;O=O;;====;'
-az + va 2z 2 - 4
= x + ay + b,
where b is another arbitrary constant, and thus the complete solution is
x + ay + b + -Va
z 2z 2 - 1
4 - -In azI + Va 2z 2 - 4 I+ -az 2
= O.
4 a 4
There is no singular solution.
Remark 2.29.1 The reader should check that any nonlinear first order PDE of the
form F(p, q) = 0 has a first integral q = ap, for an arbitrary constant a.
Remark 2.29.2 If we took the sign - in (2.46), we would get another complete
solution. This shows that the complete solution of a nonlinear first order PDE is
not unique; in fact, it even might have infinitely many complete solutions.
b) One easily gets the complete solution
z = ax + by + a 2 + ab + b2 (a, b arbitrary constants).
Taking the partial derivatives in a and b of the complete solution, respectively,
we obtain the system
x + 2a + b = 0, y +a + 2b = O.
Solving it in a and b, we get the singular solution
Z = 3"1 ( xy - x 2 - y 2 ).
Remark 2.29.3 The equation of the form z = px + qy + f(p, q), for some given
function f is called "generalized Clairaut's equation" .
c ) Therst
fi
mtegra I IS
. p= ~
~, an d t hus t h i
e comp ete'mtegraI IS
.
zy 1 + a2
-~= k(x+ay)+b.
1 +a2
The singular integrals are z = 1 and z = -l.
Example 2.30
a) Let M be a point on a surface, and let the point N be the intersection of the
xy- plane and surface's perpendicular line through M. Find the surfaces which
have the property that for every point M on the surface the segment M N is
equal to a given number a > O..
b) Find the Cauchy integral through the circle z = b, x 2 + y2 = R2, for given b,
o O.
Solutions.
a) Let z = z(x,y) be the sought after surface and M(x,y,z) its arbitrary point.
Then the equation of the surface's perpendicular line through M is
X-x Y-y Z-z
(2.47)
p q -1
Here X, Y and Z are the coordinates on the perpendicular line. If N(Xo, Yo, 0)
is its intersection with the xv-plane, then by supposition it holds
M N 2 = (x - Xo? + (z - O? = a2
+ (y - Yo?
From (2.47) we get Xo = pz + x and Yo = qz + y, which gives the PDE
a 2 = z2(p2 + q2 + 1).
As in Example 2.29 a), we obtain that a first integral is q = sp, for an arbitrary
constant s. Thus the complete integral is
s =f J a2 - z2 = X cos t + Y sin t,
where t is another arbitrary constant.
Clearly, the singular integrals are z = a and z = -a (give a geometric expla-
nation).
Finally, putting s = f(t), where f is an arbitrary function of the class CI(R),
we obtain the general solution given with the system of equations
f(t) =f -Ja 2 - Z2 = X cos t + Y sin t, f'(t) = -x sin t + y cos t.
b) By supposition we have
f(i) = Ja 2 - b2 + Rsin(s + t),

hence the Cauchy integral is
(R J a 2 - b2 =f J a 2 _ z2) 2 = x 2 + y2.
Example 2.31 Find the sU1faces which satisfy the PDE
z = px + qy + pq
and pass through the parabola x = 0, z = y2.
Solution. One easily finds the complete solution:
z = xa + yb + ab (a, b arbitrary constants).
If we put y = t, then for x = we have z = t2 , hence
t2 = bt + abo (2.48)
Differentiating this equation in t we get b = 2t, hence from (2.48) we obtain a =

-1/2. Thus
t
Z = --x + 2ty -
2
2
t ,
which after another differentiation gives t = Y - ~. Thus the solution is
Example 2.32 Find the complete, general and singular solution of the PDE
Moreover, find the surface which passes through the line y = 0, z = x - 1.
Solution. Firstly, we shall introduce a new unknown function u = u(x, y) by

u= Z2. Then we have
1
x2 8x
(8U)21
+ y2 8y = 8,
(8u)2
and the complete solution is
There is no singular solution.

The general solution is
ay2
~+f'(a).
2 - a2
where f is an arbitrary function of the class C 1 (R).

Let us find the sought after surface. By assumption we have
(x - l? = ax 2 + f(a),
or
1
2(x - 1) = 2ax, or x = - - .
I-a
In view of the equation /,(a) + x2 = 0, we obtain
1
f(a) = - - .
a-I
Finally, the parametric equation of the Cauchy integral is
2 ay2 1
x =---
2 - a2 (a - 1)2
Exercise 2.33 Using an appropriate change of variable (s), solve the following PDEs.
Answers.
a) The complete solution is
2>/Z = alnx + ~lny + b,

while the singular one is z + x 2 + y2
o. =
Hint. Put X = In x, Y = In y and Z = 2VZ. Then the given PDE transforms
az az
to p2 + Q2 = I, where P = ax and Q = ay'
b) z = ax 2 + by2 + ab (complete solution), z = _x 2 y2 (singular solution).

Hint. Put x = ,;x, y = JY.
Exercise 2.34 Lei the PDE pq = zm xy be given, mE Z.
a) Find the complete, singular and the general solution of this PDE.
b) For m = 0 find the Cauchy integral through the curve x = 1, z = Vf+Y2.
Answers.
a) If m = 2, then put Z = In z, and show that the transformed equation becomes

PQ = xy (P, Q are as in Exercise 2.33 a). The complete solution is
y2
21n z = ax 2 + -2a + b.
If m t 2, then put Z = _2_ zl - m / 2 , which gives again the PDE PQ = xv.

2-m
Hence
2 2 2
_ _ zl-m/2 = ~ + L + b.
2- m 2 2a
b) In a) we saw that for m = 0 the complete solution is
ax2 y2
z= -+-+b.
2 2a
The Cauchy integral is
z = xf1+;2.
Example 2.35 Find the complete solution of the PDE
yzp2 = q.
Solution. From the system (2.38) we have
dz dp dz
-- hence
yp3' p2yz
The solution is a = zp, hence p = a/z and q = yz (a/z)2. Thus we obtain Pfaff's
equation
a ya 2
dz = -dx
z
+ -dy,
z
and the complete solution
Example 2.36
a) Find anu. solve the PDE of the form z = F{x,y,p,q), satisfied by the con~c
surfaces with a peak in a fixed point V{a, b, c).
b) Assume now a = b = c = O. Then find a function cI> = cI>{x, y, p, q), such that it
is in involution with F, see (2.39).
Solution.
Z.3. NONLINEAR FIRST ORDER PDEs 45
a) The given condition can be written as a scalar product
(x - a,Y - b,z - c) (p,q, -1) = 0,

or
p( x - a) + q(y - b) = z - c. (2.49)
The solution of this linear first order PDE is
z-c y-b)
G ( - - , - - =0,
x-a x-a
where G is an arbitrary function of the class C 1 (R2).

b) For a = b = c = 0 we get from (2.49)
F(x, y,P, q) = px + qy.
We are looking for a function q> which is in involution with F, i.e., the Poisson
bracket (F, q, see (2.40), is equal to zero:
x p y q
(F,q
aq ay
aq> aq> aq> aq>

x--p-+y--q-
ax ap ay aq
= o.
Using this equality, we get a linear first order PDE in q>, whose solution is
easily found to be
q>(x,y,p,q) = III (xp,yq,x/y) =0 (Ill E C 1 (R3 )).
Exercise 2.37 Let the following first order PDE be given:
z2(2 + p2 + l) = x 2 + y2 + 2z(xp + yq).

Z2 az az . .
a) If Z = - , P = -a and Q = -a ,then prove that the transformed equatzon (zn
2 x y
Z) has the following two first integrals in involution:
F = y + Q+ a and q> = x + P + b, (2.50)
where a and b are arbitrary constants.

b) Using a), find the complete integrals of the transformed and the given equation.
c) Find the general and, if any, the singular solution of the given equation.
Answers.
a) Using the transformation Z = z2/2, check that one obtains
4Z + p 2 + Q2 _ x 2 _ y2 - 2(xP + yQ) = 0,
and hence its first integrals from (2.50) are in involution.
b) The complete solution is
V(x,y,z,a,b) = 2(x 2 + y2 + z2) - 4(ax + by) + a2 + b2 = 0.

c) The general solution is given by the system
av av
+ 8b b (a) = 0,
I
V(x,y,z,a,w(a) = 0, aa
where b E C l (R) is an arbitrary function.

From the system
av av
V=o, -=0,
aa 8b =0,
we obtain the singular solution
Example 2.38 Determine all surfaces z = z(x,y) with the property that in every
surface point M(x, y, z) the scalar product of the unit normal vector and the vector
OM is equal to
z - H(p,q)
(1 + p2 + q2)1/2'
where H is a homogeneous function of degree n -I- 1, i.e.,
H(tp, tq) = t n H(p, q) for all t > 0,

and, moreover, it is assumed that the direction of the normal vector has been chosen
so that it makes an acute angle with the positive direction of the x-axis.
Solution. If we denote by i, j and k, respectively, the unit vectors on the X-, y-

and z-axis, then by assumption it holds
.. -pi-qj+k z-H(p,q)
+ YJ + zk) . )1 + p2 + q2 = (1 + p2 + q2)1/2'
(Xl
Thus we get
px + qy = H(p, q).
Using the homogeneity of H, we obtain the complete solution
n - 1 (x + ay)n/(n-I)
Z= - - +b
n (H(l, b))n/(n-I)
It is left to the reader to find the general solution; clearly, there is no singular
solution.
Example 2.39 Solve the PDE
p2 + q2 + pq _ qx - py - 2z + xy = o.
Answer.
I mode. From the system (2.38) we have

dx dy dz
2p + q - y 2q + p - x 2p2 + 2q2 + 2pq - py - qx
-dp -dq
-q + y - 2p -p +x - 2q
Now we easily find one first integral, namely
x + a = p, and thus WI =p - x.
Another first integral is

W2 = q - y.
Since these two first integrals are in involution (check that!), it follows that
the complete solution is
II mode. As in mode I, we find that p = x + a is a first integral. From the given

equation it follows
q = --
a
2
+ J 2z + ay - (x + a)2 + -.
(12
4
Thus we obtain Pfaff's equation
or
a
dz+-dy-(x+a)dx
2 = O.
J2Z + ay - (x + a)2 + ~
Hence the complete solution is
Example 2.40 Find a one-parameter family of the first order PDE
pq = x+y+ z.
Solution. One easily finds that
p - q + x - y = 2a and
are two first integrals which are not in involution. Solving in p and q gives
1 1
p = b_ 1 (b(y - x) + 2ab - b + 1), q = b_ 1 (y - x + 2a - b + 1).
Bp Bq.
Since necessarily By = Bx' It follows that b = -1 and
2p =y- x + 2a - 2, 2q = x- y - 2a - 2.
Hence the sought after one-parameter family is
z = pq-x-y
(x y)2
= -
4
+a(x-y)-(x+y)+1-a 2 .
Chapter 3
Classification of the Second Order

PDEs
3.1 Two Independent Variables

3.1.1 Preliminaries
The quasi-linear second order PDE on some region Q C R2 is given by
(3.1 )
The functions A, B, C and F are the coefficients of (3.1). If, additionally, the
right-hand side function F has the form
for some functions Fj , j = 1, ... ,4, then (3.1) is linear.

The unknown function u = u (x, y) is looked for in the set C 2 ( Q). Thus we assume
that the functions A, Band C are continuously differentiable on the region Q, while
F is assumed to be a continuous function on the set Q X R3.
The type of the second order quasi-linear PDE (3.1) depends on the sign of the
function
D(x, y) = B2(X,y) - A(x, 1/) C(x,y), (x, y) E Q. (3.2)
By definition, the second order quasi-linear PDE (3.1) is
1. hyperbolic at (x,y) E Q if D(x,y) > 0;

2. elliptic at (x,y) E Q if D(x,y) < 0;
3. parabolic at (x,y) E Q if D(x,y) = O.
49

50 CHAPTER 3. CLASSIFICATION OF THE SECOND ORDER PDEs
The equation is hyperbolic (respectively elliptic, parabolic) on the region Q if it is

hyperbolic (resp. elliptic, parabolic) at every point (x, y) E Q.
Note that the three most important second order PDEs: the wave, the Poisson
(hence also Laplace's) and the heat equation are respectively hyperbolic, elliptic and
parabolic in the whole plane R2 (sec Section 3.3 below).
Suppose now that the functions
~ = ~(x,y), 11 = 11(X,y), (3.3)
are from the class C 2 (Q), and, moreover, that the Jacobian
o~ o~
0(~,1]) ox oy
(3.4)
o(x,y) 011 011
ox oy
is nonzero for every (x,y) E Q. We also put
V(~,11) = u(X(~,11),Y(~,11 (3.5)
Then from the chain rule it follows

ou OV o~ OV 011 ou ov o~ OV 011
-=--+--, - oy -
- - - +011
o~ oy
- oy
-, (3.6)
ox o~ ox 011 ox
02V (0~)2 02v 0~011 02v (011 )2 OV02~ OV0211
oe ox + 2 0~011 ox ox + 011 2 ox + o~ ox + 011 ox 2'
2
02V o~ o~ 02V o~ 011 02V 011 o~ 02v 011 011

= - -- +---- +---- +---
oe ox oy 0~01] ox oy 0110~ ox oy 011 2 ox oy
(3.7)
OV OV 0211
02~
+ o~ oxoy + 01] oxoy'
Then the nonsingular transformation T : Q ----4 R2 of the independent variables

x and y, given by (x,y)
f-t (~,11), transforms (3.1) into the quasi-linear second order
PDE
02V 02V 02V ov ov
a(~'11)0~2 +2{3(~,TJ)0~011 +1(~'11)0112 = ~(~'11,V'OCOTJ) (3.8)
with new unknown function v = V(~,11) from (3.5). The reader should check that
the coefficients of (3.8) are:
(3.9)
3.1. TWO INDEPENDENT VARIABLES 51
f3( ~, 77) = A o~ 0'1/ + B (fJ~0'1/ + o~ 0'1/) + C o~ 0'1/, (3.10)

ax ax ax oy oy ax oy oy
0'1/ 2 0'1/ 0'1/ ( 0'1/ )2
I(~''I/) = A ( ax ) +2B ax oy + C oy (3.11 )
and

ov or}ov)
(~,T}, v, o(
(3.12)
The essential property of the quasi-linear second order PDE is the following
equality, which immediately follows from equations (3.9), (3.10) and (3.11):
D(x,y) = B2(X,y) - A(x,y) C(x,y) = f32(~,T}) - a(~,T}h(~,T}),
where on the right-hand side the variables ~ and T} are given by (3.3). The above
equality shows that a nonsingular transformation preserves the type of the equation.
Our main goal is to show that, depending on the sign of the function D from
(3.2), an appropriate change of variables reduces the quasi-linear equation (3.1) to
one of the following three canonical forms: :
(hyperbolic equation); (3.13)
(elliptic equation); (3.14)
(parabolic equation). (3.15)
To that end, let us analyze the quadratic equation in A :

A(x,y)A2+2B(x,y)A+C(x,y) =0. (3.16)
In the following, C 1 and C 2 are arbitrary constants.
1. If the function D from (3.2) satisfies D(x,y) > 0 for all (x,y) E Q, then
equation (3.16) has two real and distinct solutions, say Al = Al(X,y) and
A2 = Az(X,y). If we put
dy dy
dx = -Al(X,y) and dx = -A2(X,y),
and assume the upper two ODEs have solutions of the form
~(X,y) = C1 and T}(x,y) = C2, (3.17)
then reduces (3.1) to (3.13) (check that).
.12 CHAPTER 3. CLASSIFICATION OF THE SECOND ORDER PDEs
2. If D from (3.2) is negative on Q, then the solutions of (3.16) are complex

conjugate numbers:
Then assume the equations
~~ = -~A(X, y) and ~~ = -S'A(X, V),

have implicit solutions
~(X,y) = C1 and 17(X,y) = C2 (3.18)
Then the change of variables (x, y) f-t (~, 17) reduces (3.1) to (3.14).
3. In the parabolic case, D(x,y) = 0 for (x,y) E Q, the solutions of (3.16) are
equal and real, say A = A(X,y). Then assume the ODE
dy
dx = -A(X, y)
has an implicit solution
(3.19)
Then take ~ = ~(x, V); for the other independent variable 17 one can choose
any function 17 = 17(x,y), provided that the Jacobian (3.4) is nonzero on Q.
Then this change of variables reduces (3.1) to (3.15).
The two families of curves given by (3.17) are the characteristic curves for the
hyperbolic second order PDE. Also, the curves from the (single) family (3.19) are
the characteristics curves for the parabolic second order PDE. However, the elliptic
second order PDEs have no characteristic curves (why?).
Cauchy's Problem
Let a curve L be given in parametric form
(T E 1), (3.20)
where <p and tf; are Cl functions on some interval I in R. Then Cauchy's problem
for (3.1) is finding a solution of (3.1) that passes through the curve L and satisfies
the initial conditions
~: 1 = q( T) and aul = r(T),

[) (3.21 )
y
where p, q and r are given functions of the real parameter T.

:1.1. TWO INDEPENDENT VARIABLES 53

Example 3.1 Determine the type oj the Jollowing second order linear PDEs and
find their canonical forms.
a2u a2u a2u au au
a) -a 3 xy
x 2+ -a a - 4ay-2 +-ax +4-ay =0;
a2u a2u a2u au au
b) -a
x 2 +6- a
axy +10-ay 2 + -ax +3-
ay =0;
c)
Solutions.
a) The function D from (3.2) is equal to (3/2)2 - 1 (-4) = 25/4 > 0, and thus it
follows that the given PDE is hyperbolic in the whole xy-plane. In this case,
the quadratic equation (3.16) is
.\2+3'\-4=0,
and its solutions are .\1 1 and .\2 = -4. Thus we have the differential
equations
-=-1
dy
dx '
dy
dx
=4
'
whose solutions are
x + y = G1 and 4x - y = G2 ,
respectively. In order to obtain the canonical form (3.13) we introduce the
new variables ~ and 1J by ~ = x + y and 1J = 4x - y, see equation (3.17), and
we put v(~,1J) = u(x,y). Note that the above transformation (x,y) f-t (~,1J) is
nonsingular, since the corresponding Jacobian is nonzero:
a(~,1J)
a(x, y)
=11 11=-5
4 -1 .
Thus we obtain the canonical form

av 2 1 av
a~a77 + 5a~ = 0.
b) The solutions of the quadratic equation .\2 + 6.\ + 10 = 0 are conjugate complex
numbers, namely '\1,2 = -3 x, hence the given PDE is elliptic in the whole
plane. The characteristics are
y = 3x - xx + G1 , y = 3x + zx + G2 ,
where C 1 and C 2 are arbitrary constants. Thus we can put ~ = 3x - y and

TJ = x (the "real" and the "imaginary" part of the characteristics). Now the
transformation (x, y) 1--+ (~, TJ) is nonsingular, and applying the equations (3.6)
and (3.7), we get the canonical form of the given elliptic PDE
c) The given PDE is parabolic in the whole xy-plane, since both solutions of the
corresponding quadratic equation are equal to 1. Thus we can use the change
of variables ~ = x + y and TJ = x in order to obtain the canonical form of
parabolic equations. In our case we get
Remark 3.1.1 In case c), once we have put ( = x + y, for the new variable TJ we
could have chosen any function of x and y of the class C 2(R2), provided that the
transformation (x, y) 1--+ (~, TJ) is nonsingular.
Exercise 3.2 Determine the type of the following second order linear PDEs, de-
pending on the points in xy- plane, then use a suitable change of variables, ~ =
~(x,y), TJ = TJ(x,y) which, after putting v(~,TJ) = u(x,y), gives the canonical form
of the transformed equation.
fYu f)2u fPu
a) ox 2 + 2 oxoy + (1 - sgn y) oy2 = OJ
02U 02U
b) ox2 + x 2 oy2 =0 (x f- 0);
02U 02u
c) x2 y2 - O.
ox2 - oy2 - ,
02U 02U 02U ou

d) y2 ox2 + 2xy oxoy + 2X2 oy2 + y oy = 0;
02U 02U lou lou
e) x- - y-
ox 2 oy2
+ --
2 ox
- --
2 oy
=0 (x,y > 0);
02U 02U
f) Y ox 2 + x oy2 = 0;
4 202U 2x02U 20U
g) Y ox 2 - e oy2 - 4y ox = 0;
3.1. TWO INDEPENDENT vARIABLES 55
h)
i)
Answers.
a) For y > 0 the given equation is hyperbolic. Putting the new variables ~ =y- 2x,
'rf = y, we obtain the canonical form :e ;1] =

2
O.
However, for y < 0 the given equation is elliptic. Putting ~ = y - x, 'rf = x, it

cPv 8 2 v
becomes the Laplace equation 8e + 81]2 = O.
b) For x =f 0 the equation is elliptic. Putting ~ = y, 'rf = x 2/2, it becomes
8 2v 8 2v 1 8v
8e + 8'rf2 + 2'rf 8'rf = O.
c) For x =f 0, y f. 0, the equation is hyperbolic. Putting ~ = xy, 'rf y/x, it
becomes
8 2v 1 8v
8~8'rf - 2~ 81] = O.
Note that on the x- and on the y- axis the given equation is parabolic.
d) The PDE is elliptic in the whole xy-plane. Putting ~ = x 2 - y2, 'rf = x 2, it

becomes
8 2v 8 2v 1 8v 1 8v
8~2 + 8'rf2 + ~ - 'rf 8~ + 2'rf 8'rf = O.
e) The PDE is parabolic in the first quadrant x > 0, y > O. Putting ~ = Vi +.;y,
'rf = Vi - Vfj, we get
f) The PDE is elliptic in the first and in the third quadrant. Using the new variables
~ = x 3/\ 'rf = y3/2 for x > 0, y > 0,
and
~ = (_X)3/2, 'rf = (_y)3/2 for x < O,y < 0,
we get the canonical form
56 CHAPTER 3. CLASSIFICATJON OF THE SECOND ORDEn PDEs
However, this equation is hyperbolic in the second and in the fourth quadrant.
Then we put
( = (-x )3/2 _ y3/2, 1] = (-x )3/2 + y3/2 for x < 0, y > 0,
and
(= X3/ 2 _ (_y)3/2, 1) = X3/ 2 + (_y)3/2 for x > 0, y < O.
So we obtain
02V 1 1 (OV Ov)
0(01] + 31]2 - e 1] o( - (01] = O.
Finally, on the x- and on the y-axis the equation is parabolic.
g) The PDE is hyperbolic in the xy-plane. Putting ( = eX + y2, 1] = _eX + y2, it

reduces to the canonical form
h) The PDE is parabolic for every x and y. Putting ( = y tan(x/2) and 1] = y, we

get
i) The quadratic equation
(3.22)
has a discriminant equal to 4(x 2 + y2 - 1), hence the given PDE is elliptic
inside the unit circle, i.e., for x 2 + y2 < 1, and hyperbolic outside it. On the
circle x 2 + y2 = 1 it is parabolic. From (3.22) we get the differential equation
of the characteristics
(Xy jx2 + y2 - 1) dx + (1 - x 2)dy = 0,
which can be most easily solved using the new independent variable t given by
t = ~ and the new dependent variable z given by z = y/t. Thus putting
Y
( =x-
-1'
we obtain the canonical form

If X 2 + y2 > I, then we puL
Y Jx 2 +
y2 - 1
~=x-l' 1]= x-I
and obtain
[)2v [)2V
[)e - [)1]2 = O.
Exercise 3.3 Transform the given PDEs with the given changes of variables.
[)Zu . [)2U 2 [)2U [)u
a) [)x 2 - 2 Sill X [)x[)y - cos x [)y2 - cos x [)y = 0,
~ = x +y - cos x, 1] = x - y + cosx;
2 [)2U 2 [)2u
b) y [)x 2 +x [)y2 =0, ~=y2,1]=X2;
[)2U [)2U [)2U

c) x ox 2 + 2xy [)xoy - Y [)y2 = 0, ~ = y/x, 1] = y.
Answers. In all exercises we put v(~,1]) = u(x,y).

02V
a) o~[)1] = O.
[)2V [)2V 1 [)v 1 [)v
b)
02~ + [)1]2 + 2~ o~ + 21] [)1] = O.
[)2V
c) [)1]2 = O.
Exercise 3.4 Prove that using suitable changes of variables the given PDEs can be
reduced to the given canonical forms.
a)
b) [)2U _ 2x [)2U + X2[)2U _ 2[)u = 0, [)2V _ ov - O.

[)x 2 [)x[)y [)y2 [)y [)1]2 [)~ - ,
2 [)2U 2 [)2U [)u [)u [)2V [)2V

c) (1 +x ) [)x 2 + (1 + y ) [)y2 + X [)x + y [)y = 0, [)e + [)1]2 = 0;
d)
58 CHAPTER 3. CLASSIFICATION OF THE SECOND OIWER PDEs
Example 3.5 Prove that if the second order PDE with constant coefficients
a2 u a2 u a2 n (3.23)
A~+B~+C~=O
ux uxuy uy
is either hyperbolic or elliptic, (i.c., B2 - 4AC -:f. 0), then its solution has the form
where F and G are functions of the class C 2 (R), while mj, j 1,2, al'e are the
solutions of the quadratic equation
Am 2 + lim + C = O. (3.24 )
Solution. Let us assume A -:f. 0 and B2 - 4AC > 0; the other cases are handled
similarly. Then the solutions mj and m2 of (3.24) are real and different. Putting
~ = mjX + y, 1] = m2X + y, we obtain
Since mI + m2 = -BIA, mjm2 = CIA, we have
which implies
The last PDE can be written as
a (av)
a~ a1] = 0,
hence for some function G j from CI(R) it holds
av
a1] = G 1 ( 1] ).
This gives us that for some function FI from C2(R) it holds
J
1]
v(~, 1]) = FW + GI(T) dT,

1]0
or v(~,1]) == F(O + G2 (1]), and finally we get
u(x,y) = F(mIx + y) + G(m2x + y)

for some C 2 functions F and G.
Example 3.6 Prove that if the second or'der PDE with constant coefficients (3.23)
satisfies A2 + B2 + C 2 > 0 and B2 - 4AC = 0, i.e., is parabolic, in the xy-plane,
then its solution has the form
u(x,y) = F(mx + y) + xG(mx + y), (3.25)
where F and G are functions of the class C 2(R), while m = -BIA is the double
solution of (3.22).
Solution. Again we assume A f. 0; the other case A = 0 is left to the reader. By

supposition, it holds B2 - 4AC = 0, hence in order to obtain the canonical form of
(3.23), we can put e= mx + y, 11 = x. This gives
or :~~ = o. This gives (3.25).

Exercise 3.7 Find the general solution of the PDE
Answer. Since ml = m2 = 1, it follows from Example 3.6 that the general solution
IS
u(x,y) = F(x + y) + xG(x + y)

Exercise 3.8 Let a, f3 and, be some real constants. Simplify then the following
PDEs using the change of dependent variable v(x,y) = e AX+I'Y u(x,y) and choosing
suitable values for the parameters). and f-t.
a 2u a 2u au au
a) ax2 + ay2 + a ax + f3 ay +,u = 0;
a 2u 1 au au
b) -a f3
x 2 = 2-a
a y +au+ -a x ;
a 2u 1 a 2u au au
c) -a 2 - 2-a 2 = a-ax +f3-
a y +,u;
x a y
d)
Answers.
8 2v 1 8v
b)
8X2 a2 8y'
82 v 1 82 v ( a2 /Pa 4 )
c) 8X2 - a28y2 = a + f3 + I - 4" - -4- v.
82 v
d) 8x8y = (a + f3 - a(3)v.
Example 3.9 Find the general solution of the second order PDE with constant
coefficients
8 2u 8 2u 8 2u 8u 8u
A8~ + B 8x8y + C 8y2 + D 8x + E 8y = 0,
.
promded that
B + v'B2 - 4AC E
= -- and B2 - 4AC > O.
2A D
Solution. Let mj, j = 1,2, be the solutions of the quadratic equation (3.24). Then
putting ~ = mIx + y, '" = m2x + y and v(~,,,,) = u(x,y), we obtain
4AC - B2 8 2v 8v
A 8~8", + (Dm2 + E) 8", = O.
Following Exercise 3.8, let w(~,1/) = exp(AOv(~,,,,); then we obtain
8 2w \ _ A(Dm2 + E)
8~81/ = 0, provided that A - B2 _ 4AC .
Thus the general solution of the given PDE is of the form
A(Dm2 +
u(x,y)=exp ( B2_4AC
E) (m2+y)+G(m2 x +y).
Exercise 3.10 Find the general solution of the PDE
82u 1 (8U 8U)

8x8y = X - y 8x - 8y .
Answer. Putting v(x,y) = (x - y)u(x,y) we get
82 v
8x8y = 0,
hence
u(x,y) = f(x) + g(y).
x-y
Example 3.11 Solve the following Cauchy problems.

2 cPu 2 02U 02U 2y (011, OU)
a) 4y ox2 + 2(1 - y ) oxoy - oy2 - 1 + y2 2 ox - oy = 0,
ou
u(x, 0) = f(x), oy (x, 0) = g(x) (x E R),
where f E C (R) and g E C (R) are given functions.
2 1
02u 02u 02U OU OU

b) ox 2 - 2sinx oxoy - (3 + cos 2 x) oy2 + ox + (2 - sinx - cos x) oy = 0,
OU /2
U(x, cos x) = 0, oy(x,cosx) = e- x COSXj
02U 02U. 2 02U . OU

c) ox2 + 2 cos x oxoy - sm x oy2 - sm x oy = 0,
. ) = x + co~x, OU
U( x,smx oy ( x,smx
. ) .
= smx.
Solutions.
a) In our case, the solutions of the quadratic equation (3.16) are ),1 = 1/2 and
),2 = -1/(2y2), which gives
dy 1
dx = -),1 = -2" hence x + 2y = C},
and
dy 1 2y3
-dx = -),2
2y2'
hence x - -
= -
3
= C2
for arbitrary constants C1 and C2 . Thus the new variables e = x + 2y and

'" = x - 2y 3/3 lead to the PDE
02V
oeo", = 0, v(e,,,,) = u(x,y).
Thus the general solution of the given PDE is
v(e, "') = F(O + G(",), hence u(x, y) = F(x + 2y) + G (x - 2y3/3) ,

where F and G are arbitrary (but appropriately smooth) functions. The initial
conditions give
F(x) + G(x) = f(x) and 2F'(x) = g(x),
hence the solution is
J g(r) dr.
3 x+2y
u(x,y) = f (x - 2~ ) + ~
x-2y 3/3
b) The change of variables ~ =y- cos x + 2x, I] =y- cos x - 2x, v(~, 1]) = u(x, y),
gives the POE
[Pv 8v
4 8~8" = 8(
v(~,I]) = e'1/4F(~) + G(I]),

where F and G are arbitrary functions from C2(R2). Thus
u(x, y) = e(Y-Cosx- 2x l/4F(y - cos x + 2x) + G(y - cos x - 2x). (3.26)
The first initial condition gives the equation
e- x / 2 F(2x) + G(-2x) = 0
Differentiating the last equation in x gives
e- x / 2 ( -~F(2X) + 2F'(2X)) - 2G'( -2x) =0 (3.27)
The second initial condition gives
e- x / 2 ( -~F(2X) + F'(2x)) + G'( -2x) = e- x / 2 cos x,
and replacing G'( -2x) from (3.27) we obtain F(x) = sin(x/2) + C. Hence
G(x) = _e x / 4 sin(x/2) + C, and the final solution is
u(x,y) = SIn Y - cos x - 2x - sIn

e ( Y _ cosx -2 x l/4 (. . - - -x-+-2X)
y '- -cos -
2 2
= 2e(y-cosx- 2xl/4 cos x . sin ~ (y - cos x) .

2
c) Using the change of variables ~ = y - x - sin x, TJ = y +x - sin x we come to the

solution
u(x, y) = x + cos(x - y + sinx).
Example 3.12 Find the solution u = u(x, t) of the equation
82u 8 2 u . 2 82 u . 8u
8x 2 + 2 cos x 8x8t - sm x 8t 2 - sm x 8t = 0, (3.28)
on the set R x [0,00), satisfying the conditions
u(x, sin x) = <Po(x), ~~(x,sinx) = <Pl(X) (x E R), (3.29)
where <Po E C 2 (R) and <PI E Cl(R) are given functions.

Solution. The quadratic equation (3.16) is
). 2 + 2 cos x). - sin 2 x = 0,
whose solutions are

).1,2 = - cos x 1.
So we have
( : : ) 1,2 = cos x 1= 1, hence t 1,2 = sin x 1= x C1 ,2,

and thus we introduce the independent variables ~ and "I, and the dependent variable
v = v(t, "I) by
e=t-x-sinx, 'TJ=t+x-sinx and v(~,'TJ)=u(x,y).
Then equation (3.28) becomes
(3.30)
The solution of the equation (3.30) is
v(t,7]) = A(O + B(7]),

or
u(x, t) = A(t - x - sinx) + B(t + x - sin x),
where A and B functions from C 2 (R) are to be determined from the initial conditions
(3.29). Using them we have
u(x,sinx) = A(-x) + B(x) = cpo(x) and

(3.31 )
~~(x,sinx) = A'(-x) + B'(x) = CPl(X).
Differentiating the first equation in (3.31) and adding it to the second one gives
1
2B'(x) = cp~(x) + CPl(X), hence
B(x) =~ (cpo(x) - cpo(xo) + CPl(O d() ,

for some Xo E R. Then we have
A( -x) = CPo(x) - B(x),

which gives us
Therefore the solution of the problem (3.28), (3.29) has the form
( x,t ) -_ 1 (<Po(x+sinx-t)+<Po(x-sinx+t)) 1 x-sjinx+t

U "2 2 +"2 <Pl(d()d(.
x+sinx-t
3.2 n Independent Variables

3.2.1 Preliminaries
In Section 3.1 we exposed the classification and reduction to canonical forms of
quasi-linear second order PDEs, see (3.1), in two independent variables (n = 2). The
general case (n > 2) to be exposed below, is essentially analogous, but technically
more involved.
Let Q be a region in Rn, n > 2. The quasi-linear second order PDE is an equation
of the form
(3.32)
In (3.32), U = U(Xl' X2, ... , xn) is the unknown function, sought after in C2(Rn),
the coefficients ai,j are continuously differentiable functions on Q such that for all
i,j = 1,2, ... , n and x E Q it holds aij(x) = aji(x). Finally, the function is
assumed to be a continuous function of its variables. Note that (3.32) reduces to
(3.1) if n = 2.
Let us transform equation (3.32) by a change of variables ~ = Tx (x E Q); as
in Subsection 3.1, we put for the new dependent variable v(~) = u(Tx). Our goal is
to show that we can choose T so that it is a nonsingular transformation such that
equation (3.32) can be transformed at every point XO = (x~,xg, ... ,x~) E Q to a
canonical form, i.e., to an equation of the form
n
~bii(~O) aa + 1lI(~O,v,grad v) = 0,
a2v
(3.33)
for some function Ill, where bi; E {I, -1,0}, i = 1,2, ... ,n, and the function v( 0 =
u(Tx) and its derivatives are calculated at the point ~o = Txo.
3.2. n INDEPENDENT VARIABLES 65
To that end, let us quote some well known facts from the matrix theory. One
proves there that for every symmetric quadratic form
n n
LL aij(XU)YiYj (3.34 )
;=1 j=l
there exists a nonsingular linear transformation Y f--+ 7), Y = (Y!,"" Yn), 7)

(7)1,"" 7)n), which reduces (3.34) to the canonical quadratic form
n n
LL bij7)i7)j,
i=l j=l

where bii E {1,-I,O} for i = 1,2, ... ,n, and bi,j = for every pair (i,j) such that
i f. j, i, j = 1,2, ... ,n. Then it is well known that there exists a matrix B such that
Y = BT 1], where BT is the transposed matrix of B. In fact, B can be obtained as
a product of the following matrices: E ij , Ei(aj and Ejj(a), i, j E {I, 2, ... , n}, a E R.
These matrices are respectively obtained by transforming the identity matrix so that
the i-th and the j-th row of the identity matrix change places, the elements of the
i-th row is multiplied by a, and finally the elements of the j -th row are multiplied
and added to the elements of the i-th row.
Let now XO E Q and denote by A = [aij(xO)]i,i=l the matrix determined by
equation (3.32). Then choose matrix B as above; the product BABT gives the
diagonal matrix
i k E {-l,O,l}, k E {1,2, ... ,n}. Let us denote by 1" and 8 the number of positive
and negative, respectively, ones in the set {iI, i 2 , ... , in}; hence the number of zeros
on the diagonal of the above matrix is n - (1" + 8). In fact, the numbers 1" and 8
determine the type of equation (3.32) at a point x E Q in the following way.
The equation (3.32) is:
1. elliptic in XO if (1",8) = (n,O), or (1",8) = (O,n);

2. ultrahyperbolic in XO if 1" > 0, 8 > and 1" + 8 = n;
in particular, it is hyperbolic in XO if (1",8) = (n -1,1) or (1",8) = (l,n -1);
3. ultraparabolic in XO if 1" + 8 < n;

in particular, it is parabolic in XU if (1",8) = (n -1,0) or (1",8) = (O,n -1).

Example 3.13 Determine the canonical forms of the following equations.
02U 02U 02U 02U 02U
a) - + 2---2--+2--+6- u=u(x,y,z);
ox 2 oxoy OXOZ oyoy oz2 '
02U 02U
b) 4 - - 4 - - - 2 - -
OX2 oxoy
02U
oyoz
OU
+ -oy + -oz
OU
=' =u(x,y,z);
Ou
02u 02u OU ou ou
c) - - - - -
oxoy oxoz
+ -ox + -oy - -OZ = 0, U = u(x,y,z);
02U 02U 02U 02U 02U 02U 02U

a) oxz + 2 oxOY +20yz+2oyoz +2oyot+2oyoz +2 8T2 =O, u=u(x,y,z,t).
Solutions.
a) The matrix determined (in an arbitrary point XO E R 3 ) by the given equation is
A = [
1 1 -1
1 2 0
1
-1 0 6
Since BABT = E, where B = E3(~)E3,Z(-1)E3,1(1)Ez,1(-1)' the equation is

elliptic on R3. By the change of variables
[ ~l=B[:l=[ y~Xzl'
( z x--+-
Z 2
the equation transforms to
b) The matrix determined by the given equation is
-2 0
o -1 1
-1 0
Since
3.2. n INDEPENDENT VARIABLES 67
where B = E 1 (1)E 3 ,2(-1)E 2,lW' the equation is hyperbolic on R 3 , and by the

change of variables
x x
17=2+ Y' (= -2 - y + z,
it transforms to the canonical form
c) The equation is parabolic on R3. It transforms to the canonical form by the

change of variables:
( = x + y, 17 = -x + y, (= y + z.
d) The equation is elliptic on R4. It transforms to the canonical form by the change
of variables:
(=x, 17=-X+y, (=x-y+z, T=2x-2y+z+t
Example 3.14 Determine the canonical forms of the following equations:
a)
b)
c)
Solutions.
a) The canonical from of the equation is t ~2~ =

k=l (k
0, where
k
(k = L Xf, k = 1,2, ... , n.
=1
b) The canonical from of the equation is

n !l2 k
"(
L...J -1
)k+l u U _
8(2 - 0, (k = LXf, k = l,2, ... ,n.
k=1 k f=1
l]2u
L: 8(2 = 0, where
n
c) The canonical from of the equation is (1 = Xl, (k = Xk - Xk-l,
k=l k
k = 2,3, ... ,n.
Example 3.15 Show that the Laplace equation D.u =0 is invariant under an or-
thogonal transformation:
n
(Xl,'."X n) - (Yl, ... ,Yn), Yi = L:aijXj (i = 1,2, ... ,n),
j=l
where L:akiali = Dkl.

i=O
Exercise 3.16 The Lorentz transform is given by

n
(Xl,""X n) - (Y1,,Yn), Yi = L:aijXj (i = 1,2, ... ,n),

j=1
where
n n
ail - L:aii = 1, ak1all - L: akiali = Dkl (k,.e = 1,2, ... , n, k +.e > 2).
i=2 ;=2
a) Show that the wave equation is invariant under the Lorentz transformation
82 u _ (fPu + ... + 8 u) = o.
2
8x? 8x~ 8x~
b) Show that in the case n = 2 the Lorentz transformation can be represented in the
following from:
where B = -Inn 10:1 and 0: is a nonzero parameter.
c) Show that, in the case n ::::: 2, the Lorentz transformation can be represented as
a composition of the special Lorentz transformation (for the case n = 2) and
orthogonal transformations.
Example 3.17 Determine the characteristic manifold of the partial differential equa-
tion:
where u = U(Xl,' .. ,X n , t).

3.3. WAVE, POTENTIAL AND HEAT EQUATION 69
Solution. The characteristic manifold of a quasi-linear equation is a manifolds M

such that its unit normal." = ("'t,"'xw .. ,"'xJ satisfies for (t,x) E M the following
relations:
n n
.,,; + L"'~i = 1 and .,,; - L"'~i = o.

;=1 ;=1
The above is equivalent to "'t = 1/ >/2, i.e., a manifold M is characteristic for the
given equation if and only if the angle between M and t- axis is 7r / 4.
3.3 Wave, Potential and Heat Equation

The first of the three most important second order PDEs is the one-dimensional
nonhomogeneous wave equation, given by
fPu fPu
8t 2 - a 2 8x 2 = F(x,t) (x E [O,i], t E [0,00)). (3.35)
The wave equation describes, e.g., the displacement u = u(x, t) of a taut string,
fixed in its endpoints x = 0 and x = i, at a point x E [0, i] and at a moment t > o.
The positive constant a > 0 is depending on the density of the material the string
was made from. The function F on the right-hand side is the external force exerted
on the string. With (3.35) one usually imposes initial conditions
u(x,O) = f(x),
au
at (x,D) = g(x) (x E [O,i]), (3.36)
and boundary conditions
u(O, t) = u(i, t) = 0 (t ~ 0). (3.37)
In (3.36), f E G2 [0,i] and 9 E G1 [0,i] are the initial deflection and the velocity
of the string, while (3.37) merely expresses the fact that the string is fixed at the
points x = 0 and x = i.
The next important second order PDE is the Poisson's equation, given by
(3.38)
where u = u(x,y) is the potential at a point (x,y) from some region Q C R2, and G
is a given function on Q. In particular, if G = 0 on Q, then (3.38) is called Laplace's
equation (or potential equation). Let us denote by aQ the boundary of Q. Then
one imposes on the function u either the Dirichlet boundary condition:
(3.39)
which specifies the value of U OIl aQ, or the Neumann boundary condition:
aul
an 8Q
=f
'
(3.40)
which specifies the rate of change of u at points on aQ in a direction outwardly

perpendicular to aQ.
The third important second order PDE is the heat equation, given by
(3.41 )
where u = u(x,t) is the temperature at a point x E (O,C) and at a moment t > 0

of a wire of length C. Finally, H is t.he heat source affecting the wire and that it is
made from a homogeneous material; a > 0 is a constant depending on this material.
With (3.41) we impose the following initial condition
u(x,O) = f(x) (0 ~ x ~ C), (3.42)
i.e., f is the initial temperature of the wire, and, e.g.,

u(O, t) = u(C, t) = 0 (t:::: 0), (3.43)
i.e., the ends of the wire are being kept on the zero temperature.
Chapter 4
Hyperbolic Equations
4.1 Cauchy Problem for the One-dimensional

Wave Equation
4.1.1 Preliminaries
The one-dimensional wave equation given on the set f12 = {(x, t)1 x E R, t > O},
fPu fPu
ot 2 - a 2 ox 2 = F(x, t), (4.1 )
for a > 0, with conditions
u(x,O) = f(x), ou(x,O) () ( R) (4.2)

ot = g x x E
where f E C 2 (R), g E C1(R) and F E C 2 (f12) are given functions, is called the
Cauchy problem for one dimensional wave equation. If F = 0, then we are dealing
with a homogeneous wave equation, otherwise it is a nonhomogeneous wave equation.
The classical solution of problem (4.1), (4.2) is the function u = u(x, t) E C 2 (f12)
given by D 'Alambert 's formula
u(x,t)
f(x+at)+f(x-at)
'-'---"--2-"-''-----'- + 2a1
x+at
J ()d g s S
x-at
(4.3)
J
t x+at-av
+ 2~J F(w,v)dwdv.
o x-at+av
71

72 CHAPTER 4. HYPERBOLIC EQUATIONS

Example 4.1 Prove D'Alambert's formula (4.3) for homogeneous one dimensional
wave equation given by
(x E R, t > 0),
with initial conditions
u(x,O) = f(x), au(x,O) = g(x) (x E R),

at
where a is a positive constant.
Solution. Let us change the variables v = x + at, W =x- at, and denote by
U(v, w) = u(x(v, w), t(v, w)),

then the considered wave equation can be written as
The solution of the last equation is of the form
U(v, w) = F(v) + G(w),

where F and G are arbitrary continuous functions with continuous second deriva-
tives. Therefore the solution of the considered equation is
u(x, t) = F(x + at) + G(x - at).
Using the initial conditions we get
u(x,O) = f(x) = F(x) + G(x) (x E R),
au(x,O) () =a F'()
-----'':::--'--'-=gx x +a G'()
x (x E R).
at
From these two equations (after differentiation of the first one) we obtain
F'(x) = ;a (aJ'(x) + g(x)),

wherefrom it follows that the functions F and G have the forms
4.1. ONE-DIMENSIONAL CAUCHY PROBLEM 73
F(x) 1
= -f(x) + -1 lox g(Od~ + C, G(x) 1
= -f(x) - -1 lox g(Od~ - C
2 2a 0 2 2a 0
or
1 1 [x+at
F(x + at) = "2f(x + at) + 2a Jo g(Od~ + C,
1 1 [x-at
G(x - at) = "2f(x - at) - 2a Jo g(~)d~ - C.
The solution of the considered problem is
1 1 lx+at
u(x, t) = -2 (J(x + at) + f(x - at)) + - ~ _~
g(Od( ( 4.4)
Thus we obtain the D'Alambert's formula as in (4.3).
Example 4.2 Check the validity of the formal solution given by relation (4.4) for
the Cauchy problem for homogeneous one-dimensional wave equation.
Solution. Let us check that the solution given by relation (4.4)
J g(~)d~,
1 x+at
u(x, t) = ~ (J(x + at) + f(x - at)) + 2a
x-at
satisfies the wave equation with appropriate conditions. From

au
at
= ~ (J'(x + at) - J'(x - at)) + ~ (g(x + at) + g(x - at)),
a2 a
= 2" (J"(x + at) + f"(x - at)) + "2 (g'(x + at) - g'(x - at)) ,
~ (J'(x + at) + J'(x - at)) + 2~ (g(x + at) - g(x - at)),
au
2 1
-2 (J"(x + at) + f"(x 1
- at)) + - (g'(x + at) - g'(x - at)),
ax 2 2a
it follows that
Also, we have
au a 1
u(x,O) = f(x), at = "2 (J'(x) - f'(x)) +"2 (g(x) + g(x)) = g(x).
Example 4.3 Let the following Cauchy problem on the set {( 3;, t) I x E R, /, > o} bc
gzven:
u(x,O) = f(x), au(x,o) = g(x) ( R)

at x E ,
where a > 0, f E C 2 (R) and 9 E C1(R). Show that the following properties hold.
a) If the functions f and 9 are odd in x, for every fixed t > 0, meaning that f(x, t) =
- f( -x, t), then the function u(O, t), for every fixed t > is necessarily equal
to 0.

. au(x,O)
b) If the functwns f and 9 are even, for every fixed t > 0, then ax = 0.
Solution. From D'Alambert's formula
u(x, t) = f(x + at) + f(x -

2
at) +~
2a
J g(Od~,
x+at
x-at
one can conclude the following.

a) If the functions f and 9 are odd then
u(O, t) = f(at) +/( -at) + 2~ J g(~)d~ =

x+at
0.
x-at
b) If the functions f and 9 are even then

au(O,t) _ f'(at)+f'(-at) g(at)+g(-at)_o
ax - 2 + 2 -
(the first derivative of even function is the odd one).
Example 4.4 Solve the following problems
a 2 u= a 2 _a 2 u
_
a) (x> 0, t > 0),
at 2 ax 2
au(x, o)
u(O, t) = 0, u(x,O) = f(x), =g(x) (x > 0),
at
a2u a2u
b) _ = a 2 _ (x> 0, t > 0),
at 2 ax 2
au(O, t) au(x,O)
=0, u(x,O) = f(x), = g(x) (x> 0).
ax at
Solutions.
a) Let us introduce the functions
F(x) ={ f(x), x>O g(x), x>o

- J( -x), x < 0, G(x) = { _g( -x), x < O.
Then the Cauchy problem
(4.5)
U(x,O) = F(x), aU(x,t) = G(x) (-00 < x < (0),
at
has the solution given by
U(x, t) = F(x + at) ; F(x - at) + 21a J G(Od(

x+at
x-at
The functions F and G are odd and therefore it holds
U(O, t) =0
U(x,O) = F(x) = f(x),
au = g(x)
at (x> 0).
This means that the solution of the problem (4.5), for x > 0, t > 0, is in fact
the solution of the considered problem, i.e., U(x, t) = u(x, t), for x > 0, t > 0,
and it has the form
f(x+at)+f(x-at)
"-'------'--2----'-"'-----'- + 1
2a
J ()d
x+at
9 x x
x-at
u(x,t) =
f(x+at)-f(at-x)
'-'-----'--2-"-----'- + 2a1 J
x+at
()d
9 x X (x>o,t>~).
x-at
b) Introducing the even functions
F(x) = { f(x), x> 0, G(x) ={ g(x), x> 0,

fe-x), x < 0, g(-x}, x < 0,
and solving the Cauchy problem
02U 202U
Ot2 = a Ox2 (-00 < x < 00, t> 0),
(4.6)
U( x,O ) = F ()
x , at
OU(x,O)=G(x) (-00 < x < 00),
we obtain
aU(x,O) = 0
Ox
oU
U(x,O) = F(x) = f(x), at = g(x) (x > 0).
In this case we also have, U(x, t) = u(x, t) (x > 0, t > 0). So we obtain the
solution in the form
f(x + at) + f(x - at) 1

"---'----'----::-2---=-'------'- + 2a
J
x+at
g x
()d
X (x> 0, t < ~),
x-at
u(x, t) = f(x + at) - f(at - x)

2
(x> 0, t > ~).

Example 4.5 Let us consider the nonhomogeneous equation
with homogeneous conditions
u(x,O) = ou~,O) = 0 (-00 < x < 00).
Show the following states.
a) If the function f is an odd one by x, for every fixed t, then u(O, t) =0 .
b) If the function f is an even function by x, for every fixed t, then
ou(O, t) = o.
Ox
Solutions. Since the initial conditions aTe homogeneous the solution of the consid-
ered problem is
J J
t X+a(t-T)
U(X, t) = 2~ f(z, T)dzdT.

o X-a(t-T)
a) Therefore, if f is an odd function by x for every fixed t, then it holds
JdT J
t a(t-T)
u(O, t) = 21a f(z, 7 )dz = O.

o -a(t-T)
b) If f is an even function by x, for every fixed t, then it holds
8 t)
8u(O,
x
1
=~
2a
J(J(a(t - T),7) - f(-a(t - T),T))d7 =
t
O.
o
Example 4.6 Solve the following problems

8 2u 8 2u
a) [ii2 = a2 8x2 + f(x, t) (x> 0, t > 0),
8u(x,O)
u(x,O) = =0 (x > 0), u(O,t)=O (t > 0).
8t
82u 82 u
b) 8t Z = a2 8x 2 + f(x, t) (x> 0, t > 0),
8u(x,0) 8u(0, t) = 0
u(x,O) = =0 (x> 0),
8x
(t > 0).
8t
Solution.
a) Let us introduce the following odd function in x
F(x,t) ={ f(x,t), x> 0

- f( -x, t), x < o.
Usmg the solution of of the Cauchy problem
(-00 < x < 00, t > 0),
U(x 0) = 8U(x,O) = 0 (-00 < x < (0),

, 8t
and Example 4.5, we obtain the solution of considered problem as
LJ J
t x+a(t-r)
f(z, T)dzdT
o x-a(t-r)
J J
t-x/a a(t-r)+x
u(x, t) = 21a f(z, T)dzdT

o a(t-r)-x
J J
t x+a(t-r)
+21a f(Z,T)dzdT
t-x/a x-a(t-r)
b) The solution of this problem is
LJ J
t x+a(t-r)
f(z,r)dzdT
o x-a(t-r)
u(x,t) =
J J
t x+a(t-r)
+ 2~ f(z, r)dzdT (x> 0, t> ~) .

t-x/a x-a(t-r)
Example 4.7. Let us consider the homogeneous wave equation (4.1), (4.2) (F(x, t) =
0), i.e.,
(x E R, t> 0),
u(x,u) = f(x),
au
at (x,O) = g(x) (x E R),
where c > 0, f E C 2 (R) and 9 E C 1 (R) are given functions. Suppose there exist
constants A i- 0, B i- 0 and b > 0, such that
lim f(x) = A, lim g(x) = B.

Ixl->oo Ixl b Ixl->oo Ixl b- 1
Show then the existence of a constant C with the prope1"iy
lim u(x, t) = C.
t~= tb
Determine the constant C.
Solution. We know that u(x, t) given by (4.3) is the solution of the problem (4.1),
(4.2) with F(x, t) = o. Introducing the variable s by s = vt + x, the first integral in
(4.3) is transformed as
x+ct c
J g(s)ds=tJg(vt+x)dv.
x-ct -c
Fix x E R. Then we can write
. u(x,t)
1I m--

11m f (x + ct) + f (x - ct) + 11m -1 JC g(vt + x) dv
t~= tb t~= 2tb t~= 2c t&-1
-c
g(vt+x) )
Acb + ~ JC ( lim Ivt ~_~Ib-l dv = Acb + B JC Ivlb-1dv
2c t~= t C
-c a
Ivt + xl b- 1
BCb- 1
Thus C = Acb + -b-.
The interchange of the limit as t --4 00 and the integral was allowed in view of
the assumptions on g.
Example 4.8. Let the Cauchy problem for the homogeneous wave equation be given
on the set {(x, t)1 x E R, t > O}, i.e.,
u(x,O) = f(x), ou(x,O) = g(x) (x E R),

at
where f E C 2 (R) and g E C1(R) are given functions. Suppose the functions f and
g satisfy the following inequalities:
mlxl" ::; f(x) ::; Mlxl", (4.7)

Jor Ixl ~ 8> 0, 0' > 0 and 0 < m < M. Show then that Jor every Xo E R there exist
constants to, C1 and C2 > 0, such that it holds
(4.8)
Solution. Using formula (4.3) (for F(x, t) = 0), the solution of the given problem
at the point (xo, t) is
1 1 xJo+~
u(xo, t) = - (j(xo + at) + J(xo - at)) + - g(O d(
2 2a
Io-at
We shall only prove the right-hand side inequality in (4.8). Suppose Xo > 0 is
given; the case Xo ~ 0 is analogous and is left to the reader. Then for t sufficiently
large from the first assumption in (4.7) it follows
J(xo+at)+J(xo-at) < -M(I Xo + at 1

.:......:...---'-----'-.-'-".-----'- 01
+ IXo - at 1(1) .
2 - 2
Next we have for t > 0 with the property Ixo - atl ~ 8:
- J g(Ode ~ - J la>-lde = -
1 xo+at M xo+at M
2a 2a 20'
(Ixo + atl a -Ixo - atl a ).
xo-at Io-at
Since it holds
M (Ixo _
+ atl a + Ixo -
lim ( ~2,,----_ _ _ _ _ _ _W))
a
t-+ex> ta
= Ma Ol
,
and
M (Ixo + atl a - Ixo - at la )) M
lim ( 2 = _aa-l(1 - (-1)"),
t-+ex> t Oi 2
there exist constants C2 = C2 (xo, a, 0') and to = to(xo,a,O') such that for t > to it
holds
4.2 Cauchy Problem for the n-dimensional Wave

Equation
4.2.1 Preliminaries
We always suppose that the boundary 8Q of a bounded region Q C Rn is sufficiently
regular so that we can apply the Gauss-Ostrogradsky (divergence) theorem for u E
C 1 (Q) (this is satisfied if, e.g., Q is a locally quadratic region, see Chapter 8).
4.2. N-DIMENSIONAL CAUCHY PROBLEM 81
Theorem 4.1 (Gauss-Ostrogradsky) Let Q eRn be a bounded region with suf-

ficiently regular boundary aQ and u E CI(Q). Then
J
Q
Dku(x)dx = J
8Q
u(x)nkdSx,
where n = (nl,"" nn) is the exterior unit normal of aQ and dSx is the surface
element with integration on x.
The n-dimensional Cauchy problem n 2: 1, on S1 n +1 = {(x, t)i x ERn, t > o}
aatu -.6.u =
2
F(x, t), ( 4.9)
2

au
u(x,O) = f(x), Ft(x,O) = g(x)
where f E C 3 (Rn), g E C 2 (Rn), FE C 2 (S1 n +1), are given functions has the classical
solution u E C 2 (S1 n +1)'
The solution for two-dimensional case (n = 2) is given by the Poisson's formula
a JJ
[1
2
f(YI,Y2)dy l dY2 1 (4.10)
+ at 211" D(x,t) Jt - (YI - Xt}2 - (Y2 - X2)2
+ 1 JJ J
t
F(y!, Y2, t)dy l dY2 ds
211" o !y-x!<t-s VI(t - S)2 - (Yl - Xl)2 - (Y2 - X2)2'
where D(x, t) is a disc with the center in x = (xt, X2) and diameter t.
The solution for the three-dimensional case (n = 3) is given by Kirchoff's for-
mula
u(x, t) 1 Jg(x+tv)dSv + 47f"at

471" 1 a, [1t J f(x+tv)dS v1
8~ 8~
(4.11)
where St = {Yiix - yi < 1}, and aSI = {Yiix - yi = 1}.


Example 4.9 Let Q c Rn be a bounded region with sufficiently regular boundary
aQ. Provc thatforu E C 2 (Q),v E C 1 (Q) we have thcfirst (antisymmctric) Green
Jormula
j vLludx =- j \7u \7vdx+ j v~~ dS, (4.12)

Q Q 8Q
where n is the normal on aQ Jrom Q outside and ~~ is the derivative oj the Junction
u in the direction oj n.
Solution. Apply on j vLlu dx the product derivation formula and the Gauss-
Q
Obtrogradsky theorem.
Example 4.10 Let Q C Rn be a bounded region with a sufficiently regular boundary

aQ. Prove that Jor u, v E C 2 ( Q) we have the second (symmetric) Green Jormula
j(vLlu-uLlv)dx= j (v~~ -u~~) dS. (4.13)

Q 8Q
Solution. Changing the position offunctions u and v in (4.12) and subtracting the
obtained equality from (4.12) we obtain (4.13).
Example 4.11 Show that the D'Alamberl formula (4.3), giving the solution of the
problem (4.1), (4.2) for F = 0, can be obtained from Poisson's formula (4.10) for
n = 1, namely for the problem
(x E R, t > 0),
u(x,O) = f(x), u(x,O) = g(x) (x E R),

at
where f E C3(R), g E C 2 (R).
Solution. Starting from (4.10) we obtain

u(x, t)
a
1
27r J
-t
t
g(x
1
+ e)d~ + 2at /f(x + ~)d~
-t
t
1
2 (f(x + t) + g(x - t)) +2
1
J
x+t
x-t
g(r)dr.
Exercise 4.12 If h, h E C 2 (R) and gl, g2 E C 1 (R) are given functions, then find
the solutions two dimensional wave equation
aat2u= a (aax2u+ aay2u) '

2
2 2 2
with the initial conditions

au(x, y, 0) ()
u(x,y,O) = fl(X) + h(Y), at = gl X + g2 (Y) .
Answer.
u(x, t)
fl(X + at) + h(x - at) + f2(Y + at) + h(Y - at)
2
+ 2a1 J
x+at
gl{s)ds + 2a
1 J .
v+at
gl{s)ds.
x-at v-at
Exercise 4.13 Solve the following Cauchy problems for u = u(x, y, t);
a)
a2u
-
at 2
3
- D.u = x - 3xy
2
'
u(
x , y)
au(x,y,O)
, = eX cos y' a t = ev.sm x'
2 au(x,y,O) .
u ( x,y,O ) =x, at =smYj
a2u au(x, y, 0) 2 2
= x +y .
) 2 2 2
C at 2 - 2D.u = 0, u(x,y,O) = 2x - Y , at
Answers.
a)
b) u(x,y,t) = X2 + e + tsiny.
c)
Exercise 4.14 Solve the following Cauchy problems for u = u(x, y, z, t),-
cPu 2 2 3 f)u(x,y,z,O)
a) f)t 2 - flu = 2xyz, u(x,y,z,O)=x +y -2z, f)t =1;
b)
f)2u 2 2
f)t 2 - 8flu = t x,
(
u x, y, z, ) 2
= y ,
f)u(x, y, z,O)
f)t
2
= z .
Answers.
a)
b)
Example 4.15 The spheric mean value for a function w E C(Rn) and f)B(O, r) =
{yllx - yl = r} is given by
Mw(x,r)=u:n_l
n
J
8B(0,r)
w(y)dSy ,
27r n / 2
where Un = r(n/2)" Prove that
a) Mw(x,r)=~
Un
J w(x+rv)dSv ; (4.14)
8B(0,1)
b) Mw satisfies the Darboux equation
f) 2 u n - 1 f)u
flxu = !l
ur
2 + --~.
r ur
(4.15)
Solution.
a) Put y = x + rv for Ivl = 1. The equality (4.l4) enables to extend Mw also for
r ::; 0.
b) Differentiating (4.14) we obtain

a
~Mw(x, r) = -
ur
1
an'
BB(a,I) ,=1
J I: n
w(x + rv)v; dS v '
Applying the divergence theorem on the integral at the right-hand side we

obtain
!....
an
J .6. x w(x + rv)v; dv
B(a,I)
r
I
an
-
n
.6. x J w(y) dy
B(X,T)
J
T
r I - n .6. x sn-I Mw( x, s) ds.

a
Multiplying both sides by r n - I and applying the derivative with respect to r

we obtain
:r (r n- I :rMw(x,r)) = .6. x r n - I M w(x,r),
which implies that Mw satisfies (4.15).
Example 4.16 Prove that if a function u = u(.r, t) E C 2(Dn+d is a solution of the

problem

au
u(x,O) = I(x), at (x, 0) = 9 (x)
then the spheric mean value
Mu(x, r, t) =~
an
J u(x + rv, t) dSv (4.16)
BB(a,I)
is a solution of the Cauchy (initial) problem for the Allure-Poisson-Darboux equation
a2Mu a2Mu n -1 aMu

7fi2 = [i;2 + - r- Tr (4.17)
(4.18)
Solution. By (4.15) we have for Mu = Mu(x, r, t)
[PM" n -1 aM"
D.xMu = ar2 + -r-a;:-' ( 4.19)
Since the function u = u(x, t) is a solution of the wave equation we obtain by (4.16)
D.xMu = ~
O'n
J D.xu(x + rv, t) dSv
BB(O,l)
=
a 1
-a
t
2
2-
O'n
J U(X +rv,t)dS v =
a M"
-a
2
t
2 .
BB(O,l)
Now substituting the last right-hand side in (4.19), we obtain (4.17).
Example 4.17 Prove that the equation (4.17) for n =3

a 2v a 2v 2 av
(0::; R,O < t ::; T)
ae = ar2 + ;:- ar r ::;
v(r,O) = 0, av(r,O) = fer) (0 ::; r ::; R),

at
where f E Cl [0, R], has a unique solution
Jg(s)
rH
vCr, t) =~ ds,
2r
r-t
where
a
g(s) = at(sv(s,t))lt=o = sf(s).
Solution. The function w(r, t) = rver, t) satisfies the equation

a2 w a2 w
ae - ar2 = 0
w(r,O) = 0, aw(r,O) = r f(r)

at
and with the boundary condition w(O, t) = O.
The last problem has a unique solution given by D' Alambert's formula
Jsf(s) ds Jg(s) ds.

rH 1 T+t
w(r, t) = ~ = 2'
T-t T-t
Example 4.18 Suppose that a function U = U(XI' X2, X3, t) IS a solution of thc
Cauchy problem for the three-dimensional wave equation
U(X,O) = 0, ou(X,O) = g(x).

at
Prove that the function v = ~~ also satisfies the same wave equation and the initial
conditions
v(X,O) = 0, ov(X,O)
at
= .
Solution. We have
We have for the initial conditions
v (x, 0)
= at
ou(x,O) = 9 (x ),
Example 4.19 (Kirchoff formula) Prove that the Cauchy problem for three-di-
mensional wave equation
U(X,O) = f(x), AU
Ft(x,O)=g(x) (xER3) ,
where f E C3 (R3 ), 9 E C 2 (R3) are given functions has the solution
U(X,t) 417[" J
8S1
g( x + tv)dSv + :t [2~ J 8S1
g( x + tV)dSv ]
tMg(x, t)
a
+ at [tMj(x, t)].
Solution. By Example 4.18 and linearity of the wave equation it. is enough to prove
that the function
v(x, t) = tMj(x, t)
satisfies the equation
f]2v
~2 - llx v =0
at
and the initial conditions
v(x,O) = 0, av(x,O) = f(x)

ot
We have by Example 4.15 b)
llxv = tllxMj
Therefore using the equality

ov oMj
-=Mj+t--
ot or
we obtain
02V oMj o2Mj
ot Z = 2-----a;- + t~ = llx v .
We have for the initial conditions v(x,O) = 0 and
ov(x,O)
ot = Mjlt=o = 47r
1 J f(XI,X2,X3)dS v = f(XI,X2,X3).
BB(O,I)
Example 4.20 (Poisson formula) Prove that the Cauchy problem for two-dimen-
sional wave equation

ou
u(x,O) = f(x), ot (x,O) = g(x)
where f E C3(R2), 9 E C 2 (R2) has the solution
u(x,t) =
4.3. THE FOURIER METHOD OF SEPARATION VARIABLES 89
Solution. We suppose in Example 4.19 that the functions f and 9 are independent
of X3 as well as the solution U = U(XI,X2,t). Let (3 be the angle LABB', where
A(XI,X2,0), B(xI + Vlt,X2 + V2t,X3 + V3t) and B'(YI,Y2,0). We have
Therefore
4t7r J g(x + tv) dS

8B(O,I)
v 4~t J g(x+tv)dS
8B(O,t)
v
1
47r
J g(x + tv) -dS v
t
J J g(YI, Y2 )dYI dY2

8Biu8B;
1
27r D(x,t)
. ft2 - (YI - XI)2 - (Y2 - X2)2'
V
where oBt and oBt- are the half-spheres for Y3 :::: 0 and Y3 S 0, respectively.
We can obtain in an analogous way the second summand in the Poisson formula.
4.3 The Fourier Method of Separation Variables

4.3.1 Preliminaries
Fourier Series
A function f is piecewise continuous on an interval [a, b] if [a, b] can be divided into
a finite number of subintervals such that
(i) inside each of which f is continuous
(ii) the left-hand and right-hand limits exists at each point on the subintervals
including their end points.
The left-hand and right-hand limits are defined, respectively, by
f(x-) = lim f(x),

x~o-
f(x+) = x-+o+
lim f(x),
and the function f is continuous at the point x if f(x-) = f(x+) = f(x).
Definition 4.2 Let f be a 27r-periodic piecewise continuous function on the inter-

val [-7r, 7r]. The trigonometric series
00
A o + L:(An cos nx + Bn sin nx), (4.20)

n=l
90 CHA rTER 4. HYPERBOLIC EQUATIONS
is called the Fourier series of function f, if the coefficients An, n = 0, 1, ... , En, n =
1,2, ... , arc given by
Ao = 217r J f(x )dx,

-1'
An =:;1 J1' f(x)cos(nx)dx, n = 1,2, ... , (4.21)
-1'
En = ~ J
-1'
f(x) sin(nx)dx, n = 1,2, ....
The coefficients An, n = 0,1, ... , En, n = 1,2, ... , are called Fourier coefficients.
Theorem 4.3 The Fourier series of a periodic piecewise continuous function f on

an interval [-7r, 7rJ, with piecewise continuous first derivative 1', converges at any
point x E [7r, 7r] ( con verges pointwise for all values x). Then we have
f(x+) + f(x-)
+ ~(An cosnx + En sm nx),
00
2 = Ao
where An, n = 0,1,2, ... , En, n = 1,2, ... , are given by (4.21).
00
Theorem 4.4 (Weierstrass test) Let ~ u n ( x) be a series of continuous func-

n
tions un(x) defined on [a, b]. Suppose there exists a sequence {Mn}nEN such that
(x E [a, b]),
00 00
and the series ~ Mn converges. Then the series ~ un(x) converges uniformly on
n=l n=l
[a, b].
The following two theorems are giving two general criteria for uniform convergence
of Fourier series.
00
Theorem 4.5 Suppose the series ~ (IAn I+ lEn I) converges, where the Fourier coef-
n=l
ficients An and En are given by (4.21). Then the Fourier series converges uniformly
on every finite interval.
Theorem 4.6 The Fourier series of a continuous 27r-periodic function f, with

piecewise continuous first derivative 1', converges uniformly on every finite interval.
4.3. THE FOURIER. METHOD OF SEPARATION VAR.IABLES 91
The following two theorems are giving the conditions for termwise differentiation
and integration of Fourier series.
Theorem 4.7 If f is a continuous 27f-periodic function and both l' and 1" arc
piecewise continuous on the interval [-7f,7f], then the Fourier series (4.20) can be
differentiated term wise to the series
00
l' (x) = L n ( - An sin nx + Bn cos nx) .

n=l
The last series converges pointwise to l' at the points where f" exists.
Theorem 4.8 Iff is a piecewise continuous 27f-periodic function of period 27f then
its Fourier series (4.20) can be integrated termwise to the series
Jx
_~
f(t)dt
2
x+1r
= Ao-- +L
001
n=l
- (An sm nx - Bn cos nx + (-1) Bn).
n
n
The last series converges pointwise to the integral of f.
The change of variables x = 1rt/C, and f(x) = f(1rt/C) = get), imply the Fourier
series for the function g as
n7ft . n1rt
Ao + L(An cos -C- + Bn sm-C-)'
00
(4.22)
n=l
where the coefficients have the forms
Ao = 2C
1
J
-i
f
g(x)dx,
J
i
1 n1rX = 1,2, ... , (4.23)
An = g(x) cos -C-dx, n
-i
Bn =1 Ji
n7fX
g(x)sin-C-dx, n=l,2, ....
-i
Note that the functions cos n;x, sin n;x are the eigenfunctions of the Sturm-
Liouville problem
1" + )...j = 0,
with boundary conditions
f( -C) = f(C), J'( -C) = 1'(C)

(see Section 4.4).
The Fourier Method of Separation of Variables

The method of separation variables is the most often used techniques for obtaining
the solution of partial differential equations. Let us consider the following linear
differential homogeneous second order PDE
~u & ~u
A(x)-2 + B(x)- + C(x)u - D(y)- - E(y)- - H(y)u = 0,
& (4.24 )
8x 8x 8y2 8y
where 0 < x < , y > 0, and either D(y) > 0, or D(y) = 0, E(y) > 0, for hyperbolic
and parabolic equations.
We look for a particular solution of given equation which is a product of a
function of x alone with a function of y alone, namely we assume that the solutions
exist of the form
u(x, y) = X(x) . Y(y),
and try to obtain ordinary differential equations for X(x) and Y(y).
Since each factor depends on only one variable we have
8 2u2
8x = XI/( X )Y( y,) 88y2
2
u = X ( X ) Y 1/( y,
)
~~ = X'(x)Y(y), ~~ = X(x)Y'(y).
Substituting these expressions into (4.24) we obtain
1
X(x) (A(x)XI/(x) + B(x)X'(x) + C(x)X(x))
1
= Y(y) (D(y)YI/(y) + E(y)Y'(y) + H(y)Y(y)).
Let us remark that in previous equation the left-hand side contains only functions
depending on x and the right-hand side contains only functions depending on y,
meaning that left-hand side do not depend on y and the right-hand side do not
depend on x. This can happen only if both sides are equal to a common constant
-).. So we obtain two ordinary differential equations
X~x) (A(x)XI/(x) + B(x)X'(x) + C(x)X(x)) = -).,
y~y) (D(y)yI/(y) + E(y)Y'(y) + H(y)Y(y)) = -).,
where ). is a separation constant. Thus our task is to solve these two ordinary
differential equations.
The Mixed type Problem

The one-dimensional homogeneous wave equation given on the set
O2 = {(x, t)1 X2 ::::; x ::::; Xl, t 2: OJ, for a > 0
(4.25)
u(X,O) = fl(X), au(X,o) = h(x) (4.26)

at
(t > 0), ( 4.27)
where!t, h are functions defined on the interval (Xl,X2), and gl, g2 are functions
defined for t > 0, is called the mixed type problem.
Remark The solutions of one-dimensional wave equation (4.25) u(x,t) are the
transverse deflections of a string, which is stretched between fixed supports Xl and
X2. The length of a string is i!. If g1 == 0 and g2 == 0, in boundary conditions (4.27)
then we have zero displacement at the endpoints. The initial deflection of a string
is given by the function fl(X) and the initial velocity is given h(x) appearing in the
initial conditions (4.26).

Exercise 4.21 Prove the following orthogonal relations:
a) Jf
-l
brx n7rX
cos -g- cos -g-dx =
{ 0, kin
g, k = n i 0,
2g, k = n = OJ
J
kin
b) l . k7rX . n7rx
sm-g-sm-g- dx =
{~' -t, k= n i 0,
-l 2, k = n = OJ
c) Jl
-l
k7rx n7rX
cos -g- sin -g-dx =
{ 0'
kin
g, k = n i 0,
2, k = n = 0,
for all k E Z.
Hint: Use the trigonometric identities

1
cos a . cos 13 = 2" (cos( a - 13) + cos( a + 13)) ;
sin a . sin 13 = ~ (cos( a - 13) - cos( a + 13)) ;

sina cos 13 = ~ (sin(a - 13) + sin(a + 13))
Example 4.22 Determine the Fourier series for the following functions
a) f(x) = x 2;
b) f(x) = x, for x E [-f,f], and f(x + 2f) = f(x).
Specially determine their Fourier series when f = Jr.
Solutions.
a) The function f i~ even and therefore it holds En 0, n 1,2, .... The
coefficients An, n = 0,1, ... , can be written as
Ao =
1
2
-(
J i
x 2 dx =3
2f2
An -_ ~ Jf X 2 cos nJrx
0
d x -_ (-1)n42
2 2
f ~ n Jr
(
Thus, we obtain
f2 42 ()() (_l)n nJrx
f(x) = -+ -
3
L
Jr2 n=l
--coS-.
n2 f
(4.28)
If f = Jr, we have
(_l)n
+4 L
Jr2 ()()
f(x) = - -2-cosnx. ( 4.29)
3 n=l n
b) The function f is odd and the coefficients An = 0, n = 1,2, .... In this case we
have
J
i
1 nJrx
En x sin -f-dx
l
-2 ( - x f- cos -
f nJr
nJrx
f
- Ii + -nJrf Jf cos --dx
0
nJrx )
f
(4.30)
o
So the Fourier series [or the [unction f is
2 (-It+!
f(x) = - L
00
7r n=l n
sin
mfX
-0-'
{.
(4.31)
For = 7r we have
(_l)n+l
=2L
00
f(x) sin n7r. (4.32)

n=l n
Example 4.23 Find the formal solution of the problem
(0 < x < , t > 0), ( 4.33)
where a is a constant, with boundary conditions
u(O,t) =0, u(,t)=O (t>O), (4.34 )

and initial conditions
u(x,O) = f(x), au(x,O) = g(x) (0 < x < C). (4.35 )

at
Solution. The considered problem characterizes free oscillations of taut string with
fixed ends with zero displacement.
Let us construct the solution of this problem by using the method of separation
of variables. Taking
u(x, t) = X(x) . T(t),

we obtain two differential equations
X"(X) = _A T"(t)
X(x) , a 2 T(t) = -A,
or
XI/(x) + AX (x) = 0, (4.36)
T"(t) + A a 2 T(t) = o. ( 4.37)

From boundary conditions (4.34) it follows
X(O)T(t) = 0, X(e)T(t) = 0, I.e., X(O) = 0, X() = O. (4.38)

Let us first solve the first equation (4.36) depending on variable X with the boundary
conditions (4.38). This is Sturm-Liouville problem. First, we have to determine all
values of parameter A which allow nontrivial solutions of the given problem and then
to find the solution. These special values A are called eigenvalues and the solutions
of the considered problem are called eigenfunctions. Therefore we need the following
analyses.
(i) If A = -P < 0, then the separated solution is
X(x) = C1 ekx + C2 e- kx
Using boundary conditions (4.38) we have
X(O) = C1 + C2 = 0, C1 eki + C2 e- kl = 0 ,
wherefrom it follows C1 = C2 = o.
Clearly this gives us X(x) = 0, for all x E (-f,f), hence u(x,t) =0, for
x E (-f, f) (t > 0). But then the given problem has no solution if at least
one of the functions f and g are nonzero.
(ii) If A = 0, then from

X(x) = C1 + C2 x,
and from (4.38) we obtain the same conclusion.
(iii) If A = k 2 > 0, then

X(x) = C1 cos kx + C2 sin kx,
and required boundary conditions (4.38) lead to
X(O) = C1 = 0,
wherefrom, it follows that C1 = o. If C2 = 0, we obtain the trivial solution of

X again. Therefore let us take C2 f O. Then the second equation is equal zero
if sin kf = 0, which is true for k = nf1r, n = 1,2, .... (We do not have to take
n = -1, -2, ... , because they do not give any new solutions.)
So the eigenvalues of the considered problem are
(n EN), (4.39)
and the corresponding eigenfunctions have the forms
. n1rX
X = s1n--
X n ()
e (n EN), (4.40)
where we took C2 = 1.
For A given by (4.39) the solution of the ordinary differential equation (4.37) has
the form
na7rt na7rt
T(t) = An cos -f.- + Bn sin -f.- (n EN), (4.41 )
where An and Bn are arbitrary constants.

Multiplying (4.40) and (4.41) we obtain the solution of the considered problem
na7rt na7rt) . n7rX

un(x, y) = X(x) . T(t) = ( An cos -f.- + En sin -f.- sm -f.- (n EN), (4.42)
Using the superposition principle on the set of solutions we take the solution of the
given problem as an infinite series
na7rt na7rt) n7rX

u(x, t) = L
00 (
An cos -f.- + Bn sin -f.- sin -f.-. (4.43)
n=l
00
The series (4.43) converges uniformly if the series L(IAnl + IBn!), where An and
n=l
En are Fourier coefficients, converges.
00
Remark 4.23.1 It is known that the series L nklnl, where

n=l
for k = 0,1,2 converges, if the function has continuous second derivative and the
third derivative is piecewise continuous and it holds
(O) = (f.) = 0, "(O) = "(f.) = o.

00
The series L nkltPnl, where

n=l
for k = -1,0,1, converges if the function tP has continuous first derivative and the
second derivative is piecewise continuous and it holds
tP(O) = tP( f.) = 0, tP"(O) = tP"(f.) = o.

The solution u given by (4.43) has to satisfy the initial conditions, and therefore
for t = 0 we obtain two Fourier series
n?rx
E Ansin-e-;
00
u(x,O) = f(x) =
n=l
ou(x,O)_ ()_~na?rB . n?rx

~
vt
- 9 x - L"
n=l
e n sm i .
The coefficients can be determined from
J . -i-dx,
2
l
An = l f(x)
n?rx
sm n = 1,2, .... (4.44)
o
nC?r
-e- 2
Bn = l
Jg(x)sm-i-dx,
l
. n?rx n=1,2, .... (4.45 )
g(x),
o
If the functions f(x) and < x < i, given by the initial condition for wave
equation i.e.,
U (x,O ) = f ()
x , ou(at - 9(x ) (0 < x < {.D) .
x, 0) _
are continuous and their first derivatives are piecewise continuous and if
f(O) = r(O) = f(i) = r(i) = 0, g(O) = g(i) = 0,
then the Fourier series with coefficients
Jf(x)sm-i-dx,
. n?rx = l Jg( x) . -i-dx
l 1
nC?r 2 n?rx
An = l2 -i-Bn sm (n EN).
o o
converge uniformly for all periodic extensions of these functions.
Example 4.24
a) Prove D'Alambert's formula (4.3) for homogeneous wave equation given by
(0 < x < i, t > 0),
u(x,O) = f(x), ou(x,O) = g(x) (0 < x < i),

at
where a is a constant.
b) Explain the boundary conditions

u(O,t) = 0, u(C,t) =0 (t > 0).
Solution.
a) Similarly as in Example 4.l.
b) Let us now apply in relation (4.4) the boundary conditions. From the first one
it follows that
1
0= - (f(at)
2
+ f( -at)) + -2a1 jat
-at g(Od(,
implies
1
0= -(f(at)+f(-at)) and 0= -
1 jat g(Od(
2 2a -~
This means that the function f and g must be extended (from the domain
o < x < C) as odd functions.
The second boundary conditions gives us the following
0= -1 (f(C
2
+ at) + f(C - at)) and 0 = - 1
~
l
~~
l+ at g(~)d(
This implies that the odd extension of the functions f and g must be periodic
for all real arguments, with basic period 2. These extensions of functions f
and g are denoted usually as fa and go respectively.
Example 4.25 Show that the solution obtained by using the method of separation
variables, given by (4.43), can be written in the form of D'Alambert 's formula, (44),
for f E C 2 , g E C 1 .
Solution. The solution of the form (4.43) obtained by separation of variables can
be transformed as
u(x, t) =
00 na7rt
L An cos -C- n7rX na7rt . n7rX
. sin -C- + En sin -C- sm -C-
n=1
Since f E C2 , g E C 1 , this Fourier series converges uniformly. We can transform it

as follows
1 ~ A (. na7r(x + at) . na7r(x - at))
u(X,t) :2 L...,
n sm C + sm C
n=1
-:21 ~ E
L..., n
(
cos
na7r(x
C
+ at) - cos
na7r(x - at))
C .
n=1
From relations (4.44) and (4.45) it follows
u(x,t) = f(x+at)+f(x-at) + 1 ~ B
2 20 n
n7r
T
J . n7rx d
x+at
Sill -f- x
n-l x-at
f(x + at) + f(x - at) + -1

:......:...--'---:-....:......:'-----'-
x+at
J ~
~
B . n7rx d
-n7r Sill - x
2 2 n=l n f f
x-at
= f(x + at); f(x - at) + 2la J

x+at
g(e)d(
x-at
(0 < x < f, t > 0),
u(O, t) = 0, u(f,t) = 0 (t> 0),
u(x,O) = f(x), au(x,O) = 0 (0 < x < f),

at
where
a) f(x) = { x,
f
f - x, 2 < x < fj
b) f(x) = x{f - x) (0 < x < f).
Solution.
a) Using the method of separation variables we obtain the solution of the form
(4.43) with Bn = 0, because the initial condition is zero. So the solution has
the form
u(x, t) = L An cos -na7rt

00 . n7rX
f - sm -f-'
n=l
where in our case we have

lJ . -e-
I
2 n7rX
An = f(x) sm dx
o
-2
e
J 1/2
nr.x
X sin --dx
e + -2f J(e - .
l
n7rX
x) sm --dx
e
o l/2
An = -4 J.
l/2
nr.x
X sm --dx = -4e nr.xl
-2 cos - -
l
+ -4 J
l/2
X
n7rX
cos --dx
foe 2
n r. e l/2 nr. 0 e
4e . n7r 4e ()n+1
(n EN).
= n27r2 sm 2 = (2n _ 1)21"2 -1
So the solution can be written as
(X, t ) -_~~ (_I)n+1 ( (2n-l)a7rt. (2n-1)r.x

u 2 L..J (2 )2 cos stn
7r n=1 m
0
+1 0
~ ~
b) The solution of this problem has the form

na7rt . n7rX
=L An cos -e- sm -e-'
00
u(x, t)
n=l
The initial condition u(x,O) = x(e - x), lead us to Fourier series
n7rX
=L An sin -e-
00
x(e - x) (0 < x < e),
n=1
whose coefficients can be obtained from
An = l2 Jx(f - x) . -e-dx .
l
sm
n7rX
o
Integrating by parts we get
(n EN),
and therefore the solution of the considered problem has the form
_ 82 ~ 1 (2n - 1)a7rt . (2n - 1)7rX

u ( x, t ) - 3 L..J ( )3 cos n sm n .
r. n=1 2n - 1 ~ ~
Example 4.27 A hOl'izontal bar of length C is originally at rest and unstraint along
the x-axes. The end point x =
is fixed while the right end is subjected to a
constant elongating force F pel' unit area. Detennine the longitudinal displacement
of the cross section.
Solution. The problem
28 2 u 82 u
a-=- (0 < x < C, t > 0)
8x 2 8t 2
u(O, t) = 0, E 8u (C, t) = F (t > 0),

8x
u(x,O)=o, 8u(x, 0) =
8t
(0 < x < C),
describe this displacement of the cross section (E is Young's modulus of elasticity

of the material in tension and compression and a > is a constant depending on
the material the bar was made from).
Let us remark that in this problem the boundary conditions are not homoge-
neous. Thus we introduce the functions S, (depending on x) and v, (depending on
x, t,) such that
u(x, t) = S(x) + v(x, t). (4.46)

Then the considered problem can be written as
S"() 8 2v(x, t) _ _28v(x, t)
x + 8x2 - a 8t (0 < x < C, t > 0),
S(O) + v(O, t) = 0, ES'(C) + E 8v (C, t) = F (t> 0)

8x
S(x) + v(x,O) = 0, 8u(x,0) =

8t
(0 < x < C).
This problem can be treated as two problems
S"(x) = 0, S(O) = 0, S'(C) = f, (4.4 7)

and well known problem with homogeneous boundary conditions
8 2v(x,t) _28 2 v(x,t)
8x 2 = a 8t 2 (0 < x < C, t > 0),
8v(C,t)
v(O, t) = 0, --=0 (t > 0), (4.48)
8x
v(x,O) = -S(x), 8u(x,0) =

8t
(0 < x < C).
The solution of problem (4.47) is
P
S(x) = EX.
The method of separation variables leads the problem (4.4S) Lo the systems of ODEs
XI/(x) - '\X(x) = 0, X(O) = 0, X'(C) = 0,
TI/(t) - '\a 2 T(t) = 0, T'(O) = 0,

wherefrom, we obtain the solution of problem (4.4S) as
((2k-l)t) 2k-1 (0 < x < C).

v(x,t)=LCncoS
00
C tra sin--
C-7rx
n=l 2 2
The values Cn can be obtained by using the first initial conditions as the coefficients
of Fourier series
They are
Cn = - e2 J
o
F
f
EX
. (2n - 1 )
Sill
SCP (_I)n
---U- 7rX dx = E7r 2 (2n -1)2'
Therefore the solution of the considered problem is
( )_ F ( SC ~ (_I)n . ((2n - 1)7rX) (2n - 1)7rat))

U x, t - E x + 7r2 ~ (2n _ 1)2 Sill 2C cos 2C .
Example 4.28 Solve the nonhomogeneous problem
8 zu 8 zu +2 -3x
(0 < x < 1, t > 0),
8xz = 8t Z e
u(O, t) = 0, u(l, t) = (t > 0), (4.49)
u(x,O) =~ (e- 3X -1), 8u~,0) = (0 < x < 1).

Solution. Taking u(x, t) = S(x) + v(x,t) the problem (4.49) can be written as
S"() 8 zv _ 8 zv
X + 8x z - 8t Z + 2e-3x (0 < x < 1, t > 0),
5(0) + v(O, t) = 0, 5(1) + v(1, t) = 0 (t > 0),
5(x) + v(x,O) = ~ (e- 3X -1), ou (x, 0) = 0 (0 <x< 1).

ot
The solution of the problem
5" = 2e- 3x , 5(0) = 0, 5(1) = 0,

IS
5(x) = 9
2 (3X
e- - 1) + 92 ( 1 - e- 3) x.
The solution of second part of the considered problem, i.e., of the problem
(0 < x < 1, t > 0),
v(O, t) = 0, v(1,0)=0 (t>O), ( 4.50)

v ( x, ) = -92 ( 1 - e-
3) x, ov(O, t) __
ot (0<x<1),
IS
L
00
v( x, t) = Cn sin( 7rnx) cos( tn7r).

n=1
Using the initial conditions

2
=L
00
v(x,O) = --(1- e- 3 ) Cn sin n7rX,
9 n=1
we obtain
4
Cn = - - (1 - e- 3 )
Jl x sin(n7rx)dx = 4(-1t
- - ( 1 - e- 3 ) (n EN),
9 9n7r
o
and the final solution of our problem has the form
u(x,t) = ~(e-3X_1)+~(1_e-3)x
-3) 4 ~ (_l)n sin(7rnx) ( )

+( 1 - e - L.-J cos -tn7r .
97r n=1 n
Example 4.29 Let us consider the following boundary problem for the wave equa-
tion
8 2u cPu
8x 2
=
8t 2
(0 < x < f 1 , o < t < ( 2 ),
u(O, t) = 0, u(ft, t) = 0 (0 < t < ( 2)
U(x,O) = 0, U(X,f2) = 0 (0 < x < fl)'
Show that the solution of this problem is not unique on the given rectangular.
Solution. The method of separation variables leads us to the systems
X"(x) + >"X = 0, X(O) = 0, X(fl) = 0,
T"(x) + >"T = 0, T(O) = 0, T(f2) = O.
The solution of the first system is
X(x) = Asin>..x + B cos >..x.

From the condition X(O) = 0 it follows B = 0 and from X(f 1 ) o it follows
Asin>..f1 = O. If A i 0, then we have
n = 1,2 ....
Also, from the condition T(O) = 0 it follows D = 0 and from T(f2) = 0 it follows
C sin >"f2 = O. If C i 0, then we have
k = 1,2 ....
From above it follows

>.. = n7r = n7r and f2 = 1:.
fl f2 fl n
So we get that the solution of this problem is not unique on a rectangle, if ~: is
rational. If f:
f
is irrational then there is no nontrivial separable solution.
Remark 4.29.1 This is the Dirichlet problem for the wave equation.
4.4 The Sturm-Liouville Problem

4.4.1 Preliminaries
Let us consider the homogeneous second order differential equa.tion
d~ (p(x)y') + p'(x)y' + (q(x) + Ar(x y = (a<x<b), (4.51 )
or p(X)y" + p'(x)y' + (q(x) + Ar(x y = 0, where p(x) > 0, rex) > 0, x E (a, b) and
p', q, a.nd r are continuous functions on [a, b]. The function r is called the weight
function. The equation (4.51) with homogeneous boundary conditions
Bdu] = Qlly(a) + Q12y'(a) = (Q~1 + Q~2 =f 0),
(4.52)
is called regular Sturm-Liouville problem. In (4.52), Q11, Q12, Q21, Q22 are arbitrary
constants.
The series
00
(a<x<b), (4.53)
n=1
where the set offunctions cPn(x) are orthogonal with respect to the given weighting
function r( x) > 0, satisfying
Jr(x)cPk(x)cPn(x)dx =
b
0, k =f n,
a
with the coefficients Cn given as
J
b
C n= IicPnll- 2 r(x)f(x)cPn(x)dx n E N, (4.54)
a
where
J
b
IlcPnll = rex) (cPn(X2 dx, (4.55)

a
are called Sturm-Liouville series.

Theorem 4.9 If a function f and its first derivative f' are piecewise continuous
on [a, b] then the series
(x E (a, b, (4.56)
n=1
where cPn(x) are eigenfunctions of the regular Sturm-Liouville system and Cn are of
the form (4.54), converges pointwise to f(x) at each point x E [a, b] of continuity of
the function f.
4.4. THE STURM-LIOUVILLE PROBLEM 107
Theorem 4.10 If the Junction f and its derivative f' are piecewise continuous on
[a, b] and f satisfies the boundary conditions oj the Sturm-Liouville problem on [a, b],
then generalized Fourier series (4.53) converges uniJormly to f on [a, b].
Theorem 4.11 (Fourier Integral Theorem) Suppose the function f and its first
derivative l' are piecewise continuous on every finite interval and that f is an ab-
solutely integrable function on R (i.e.,
J
=
If(x)ldx
-=
converges on the interval -00 < x < 00). Then it holds
f(x-) + f(x+) =;:1 J= (A(s) cos(sx) + B(s) sm(sx))

:......:....~-2~----"-
. ds, (4.57)
o
where
A(s) 11=
=;: _= f(t) cos(st)dt, (4.58)
B( s) =;: 11 00
-00 f( t) sine st)dt. (4.59)
Special Functions
The Legendre equation is given by
((1 - X2 )p l (x))' + AP(X) = 0

on the interval [-1,1] with the eigenvalues
A=n(n+1)
and with eigenfunctions - Legendre polynomials

108 CHAPTER 4. HYPERBOLIC EQUATTONS
-1 I
I
1
I
I
I
I
I
I
I
I
I
I
I
I
-1
Figure 4.1 Legendre polynomials for n = 0,1,2,3 with long dashed, thin, thick
and short dashed line, respectively.
We consider the more general equation (for m = 0 we obtain the previous Leg-
endre equation)
m
((1- x 2)y'(x)), + (A - --2)Y(X) = 0
I-x
for mEN. The eigenvalues are
An = n(n+ 1) for n = m,m+ 1, ...
and the corresponding eigenfunctions are the Legendre functions P:;' given by
where Pn is the Legendre polynomial.

The functions
y;(e,cp) = p;:'(cose)cosmcp,
y n- m ( e, cp) = P;:' (cos e) sin mcp,
for m = O,l, ... ,n = m,m + 1, ... , are the spheric functions. Using them we
introduce the ball functions
Specially for m = 0 the ball functions are axial symmetric, i.e., independent of cpo
The Bessel equation is given by
with the regular solution Bessel function of order n
In(x) = Er(n + (_l)k

<Xl
+ k l)r(k
(x)n+2k
+ 1) 2"
"-
0.8 \
\
0.6
0.4
0.2
-0.2
-0.4
Figure 4.2 Bessel functions for n = 0, 1,2,3 with long dashed, thin, thick and
short dashed line, respectively.

cPu cPu
ax2 = a- 2 at 2 - r(x, t) (0 < x < , t > 0),
Bdu] = 0, B 2 [u] = (t > 0), (4.60)
u(x,O) = f(x), au(x,O) = g(x) (0 < x < e).

at
Solution. Let us suppose that the solution of the problem (4.60) has the form of
generalized Fourier series
<Xl
u(x, t) = LTn(t)Xn(x), (4.61 )

n=l
where Xn are eigenfunctions of the Sturm-Liouville problem
X" + AX = 0,
If the termwise differentiation is allowed then we have
L T:(t)Xn(X),
00
n=l
00 00
L:Tn(t)X:(x) = An L:Tn(t)Xn(X),
n=l n=l
where An are eigenfunctions.

The partial differential equation (4.60) can be written as
L
00
a2 r(x, t) = (T:(t) + a 2 ASn(t)) Xn(x).

n=l
For fixed t this is a generalized Fourier series wherefrom we can write
Jr(x, t)Xn(x)
l
T:(t) + a 2 AnTn(t) = a 2 11Xn ll- 2 (n EN).

o
where IIXnll- 2 is given by (4.55) i.e.,
IIXnW = t(Xn(x)?dx (n EN).
Denoting by
J
l
vn(t) = IIXn(x)II- 2 r(x, t)Xn(x)dx,

o
we obtain the ordinary differential equation
with the solution
where An and En are arbitrary constants for n E N. The solution of the given
equation with boundary conditions is
u(x, t) = E(.k ! vn(r) sin ({i:.a(t - r)) dr

Using the initial conditions we obtain
L AnXn(x),
00
u(x,O) = f(x) =
n=l
wherefrom we get
An = IIXnll-2l f(x)Xn(x)dx,
r>:.Bn = IIX n ll- 2 foe f(x)Xn(x)dx.

Example 4.31 Determine the solution of the problem
cPu 8 2u
8x 2 = a- 2 8t 2 - R(x)sinwt (0 < x < e, t > 0),
u(O, t) = 0, u(e,t)=o (t> 0), (4.62)
u(x,O) = f(x), 8u(O, t) = g(x) (0 < x < e).

8t
Solution. This is nonhomogeneous problem and it describes the motion of an
elastic string with a time-dependent, sinusoidal, external force. The solution is the
sum of the homogeneous solution Uh(X, t), which is the solution of of the boundary
value problem
(0 < x < e, t > 0),
u(O, t) = 0, u(e, t) =0 (t > 0),

and the particular one up(x, t), which is the solution of the problem
82 u 82u
8x 2 = a- 2 8t 2 - R(x)sinwt (0 < x < e, t > 0),
(4.63)
u(O, t) = 0, u(e,t) =0 (t > 0),
I.e.,
u(x, t) = Uh(X, t) + up(x, t).
We have already obtained the homogeneous solution as
na7rt . na7rt) . n7rX

t) = ~ An cos -e- + En sm -e-
00 (
Uh(X, sm -e-' (4.64)
The particular solution can be found by using a modification of the method of

undetermined coefficient for ordinary differential equations. Let us suppose that the
particular solution can be written as
Up(x, t) = C(x) sinwt + D(x) coswt, (4.65 )
where G(x) and D(x) are the undetermined coefficients. Replacing
8~p =wC(x)coswt-wD(x)sinwt,
8;~p = -w2C(x) sinwt _ w2D(x) coswt,
in equation (4.63) we get
G"(x)sinwt + D"(x)coswt = (- (~r C(x) - R(X)) sinwt - (~r D(x)coswt,

wherefrom we obtain two equations
G"(x) + (~) 2 C(x) = -R(x), D"(x) + (~) 2 D(x) = O.
The solution of the first of those two equations is C(x), which satisfy the conditions
C(O) = G(e) = o. The solution of second equation we can take to be zero i.e.,
D(x) = o. Therefore the solution of the consider problem is
na~t . na~t) . n~x

u(x, t) = Uh(X, t)+up(x, t) = C(x) sinwt+ ~ An cos -e- + Bn sm -e- sm -e-
00 (
Using the initial conditions u(x,O) = f(x), 8u(x,0) __ g(x)

at ( 0)
O<x<.c.,we
obtain
L An(t) sin -e-'
00 n1rX
f(x) =
n=l
=L
00 na1r . n~x
g(x) -wC(x) -e-Bnsm-e-'
n=l
wherefrom
2J .l
An = -; f(x)sm-e-dx,
n1rX na1r J 2
1
.
-e-Bn = -; (g(x) -wG(x))sm-e-dx.
n1rX
o o
Example 4.32 Determine the solution of thc problem
(0 < x < C, t > 0),
u(O, t) = 0, u(C,t) =0 (t > 0), (4.66)
au(O, t) = 0
u(x,O) = 0, (0 < x < C),
at
where w, R are constants.
Solution. (Let us remark that the force R sin wt is proportional to the wind force.)
This is a special form of the previous example. The homogeneous solution has
the form given in relation (4.64). The particular solution given by (4.65) can be
determined from the problem
G"(x) + (~r G(x) = -R, G(O) = 0, G(C) = O.
It has the form
The solution of the considered problem has the form
u(x, t)
na7rt na7rt) n7rX

+L 00 (
An cos -C- + En sin -e- sin -C-
n=l
Using the initial conditions we obtain two Fourier series
u(x,O) = 0 = L
00
An sin (n7rx)
-C- ,
n=l
114. CHAPTER 4. HYPERBOLIC EQUATIONS
wherefrom the coefficients An = and from
au(x,o)
at
=
n7ra (n7rx)
+ ,; -f-
00
Bn sin -f- ,
we get
n7ra B __ 2Rw3
f n- a2f
J( (
l 1 - cos Wf)
- . (WX)
SIll
a
. (Wf)
-
a (WX) _
+ cos a
)
. n7rX d
1 SIll f x
I
o SIll -
a
4wR
n = 1,3,5, ... ,
nf (_n2_7r2 _ W2)
- f2 a2
- 0,
n = 0,2,4, ....
The solution of considered problem is
4Raw 00
. ((2n - 1)at) . ((2n -
sm f SIll f
l)X)
u(X, t) = -f-I: w2
n=! (2n - 1)27r 2 - -
a2
(4.67)
This solution can be obtained by using the other method. Namely let us consider
solut.ion in a form
= I: Tn(t) sin n7rx.
00
u(X, t)
n=I
After termwise differentiation we obtain the generalized Fourier series

4.4. THE STURM--LIOUVILLE PROBLEM 115
The coefficients of this series can be obtained as
T 11- 2 Rsinwt.If sin (nrrX)

2
T~(t) + na2rr2
21'r,(t) = 0.211 sin (nrrX) -- dx
o
2 {4RSinwt n=l,2,5, ... ,

= a2-Rsinwt(1 _ (_l)n) = a2 nrr
nrr
o n = 2,4,6, ....
The solution of previous differential equation has the form
4Ra Jt
Cn 2
.
sm
(narr(t -
C
T)).smWT dT
o
narrt) . (narrt)
+Ancos ( - f - +Bn sm - f - n=l,3,5, ... ,
narrt)
An cos ( -C- + Bn sin (narrt)
-g- n = 2,4,6, ....
(0 < x < 1, t> 0),
u(O, t) = 0, u(l,t)=l+t (t>O),
u(x,O) = x, au(x,O) = 0 (0 < x < f).

at
Solution. Taking u(x, t) = V(x, t) + W(x, i), we obtain the following problem
(0 < x < 1, i > 0),
V(O, t) + W(O, t) = 0, V(l, t) + W(l, t) = 1 + t (t > 0),
V(x, 0) + W(x, 0) = x, av~:, 0) + aw~:, 0) = 0 (0 < x < 1).
Let us first introduce the function
V(x, t) = (1 + t)x,
which satisfies the conditions V(O,t) = 0, V(1,t) = 1 + t, V(.r,O) = x. So we

have to solve the nonhomogeneous problem with homogeneous conditions
(0 < x < 1, t > 0),
W(O, t) = 0, W(l,t) = 0 (t > 0),
W( ) = 0,
x,O
aW(x,O)
at
=0 (O<x<l).
The function W(x, t) can be found in the following form W(x, t) = 5(x) +v(x, t),
where 5(x) is the solution of problem
5"(x) = -1, 5(0) = 0, 5(1) = o.

2
The solution of this problem is 5(x) = _ x ; x.
We still have to find the function v satisfying the equation
a v - -a v =
-
2 2
0 (0 < x < 1, t > 0),
ax 2 at 2
and the conditions
v(O,t) =0 v(l,t) =0 (t > 0),
x2 - X
av(x,O) = 0 (0 < x < 1).
v(x,O) = - 2 - '
at
The solution of this problem has the form
l: An sin(
00
v(x, t) = n7fX) cos( mrt),

n=l
2
Using the initial condition v(x, 0) = x - x, we obtain the Fourier series
2
x2 _ X 00
- 2 - = ~ Ansin(mrx),
where
An =2 J1 X2 -
--sm(mrx) dx
2
X
=
( _l)n - 1
n
3
7r
3 .
o
)_ ( ) _ x2 - X _ ~ ~ sin((2n -1)7rX) cos((2n - l)7rt)

u x, t - 1 + t x
(
3 L....J ()3 .
2 7r n=l 2n - 1
cPu 82u
-=--1 (0 < x < I, t > 0),
8x 2 8t 2
t2
u(O, t) = 2' u (1, t) = - cos t (t > 0),
u(x,O) = -x, 8u(x,O) = 0 (0 < x < 1).

8t
Solution. Introducing the function
V (x, t) =2
t -
2
( cos t + 2 t2) x,
t2
which satisfies the conditions V(O, t) = 2' V(1, t) = - cos t, V(x,O) = -x. If we
takeu(x, t) = V(x, t) + W(x, t) we get the problem
82 W 82W
-- = -- + x(cost - 1) (0 < x < I, t > 0),
8x 2 8t 2
W(O, t) = 0, W(l,t) = 0 (t > 0),
W() 8W(x,O) = 0
x,O =0, 8t (O<x<l).
The function W(x, t) can be written as W(x, t) = S(x) + v(x, t), where S(x) is
the solution of the following problem
SI/(x) = -x, S(O) = 0, S(l) = O. (4.68)
The solution of this problem is

x3 - X
S(x) = - 6 - '

82 v 82 v
8x 2 - 8t 2 =xcost (O<x<l, t>O), ( 4.69)
and the conditions

v(O,t) = 0 v(l,t) =0 (t > 0),
x3 - X _8v....:(,-x,--'0...:..) = 0
v(x , 0) = ---
6' (O<x<l).
8t
The solution of this problem is v(x, t) = Vh(X, t) + vp(x, t). The homogeneous
part of the solution is
00
Vh(X, t) = LAn sin(mrx) cos(mrt).

n=l
The particular solution can be taken as
up( x, t) = C( x) sin t + D( x) cos t,

where C(x) and D(x) are undetermined coefficients. The equation (4.69) implies
C" (x) sin t + D (x) cos t = -C (x) sin t - D( x) cos t,
wherefrom we obtain two differential equations
C"(x) + C(x) = -x, D"(x) + D(x) = O.
The solution of second equation we can take to be zero i.e., D(x) = O. The solution
o~ the first of those two equations and conditions C(O) = 0, C(l) = 0 is C(x) =
sm x
- - - x. So the solution of the problem (4.69) has the form
sin 1
sin x
= -.- - x + L Ansm(mrx) cos(n 2 7r 2 t).
00
v(x,y)
sIn 1 n=l
3
Using the initial condition v(x, 0) = - x ; x, we obtain the Fourier series
x3 x sin x
- - - - -.- + x = LAn sm(n7rx),
- 00.
6 sm 1 n=l
where
An = -2j 1(3 X . )
sm
- - - - -.- +
- X x
X
.
sm(n7rx) dx.
6 sm!
o
t 2 (1 - x)
u(x, t) = 2 + x cos t
2 (1- cosn7rt + 2(cost - cosn7rt)) (_l)n-l

+-L..
~
22 22
.
smn7rx.
7r n=l n 7r n 7r - 1 n
Example 4.35 Dclermine the transverse vibration of a uniform beam with simply
supported ends described by
fJ 2u
'!'l2
ut
fJ4 u
+ a2~
uX
= (0 < x < , t > 0),
u(O, t) = 0, 11(f, t) = (t > 0),

fJ 2 u(0,t) 0 (t> 0),
fJt2 =
fJu ( x, 0) ( fi)
fJt =0 O<X<L
Solution. The method of separate variables lead us to the following equations
X(4)(X) - '\X(x) = 0, T"(t) + '\a 2 T(t) = 0.

The first differential equation with boundary conditions in this case also, implies
trivial solutions for ,\ < and ,\ = 0.
Taking ,\ = k 4 , we obtain the general solutions of the differential equation
XIII'- k 4 X = 0,
in the form
X(x) = A cos kx + B sin kx + Ce kx + De- kx .
From boundary conditions
X(O) = X(f) = X"(O) = XI/() = 0,
we obtain that A = C ::-:- D = 0 and B # for k = n;. The solution of equation
Til + a2 n :
4 4
T =0 with condition T' (0) = 0,
IS
n 2 7f 2 at
T(t) = A cos - f -
2 -'
Using the superposition principle the solution of the considered problem is
From the second initial condition we obtain the Fourier series
=L
7fX 00 n7fX
xsmT Ansm--,
n=l
whose coefficients are
So we obtain the solution of the considered problem
Exercise 4.36 Solve the problem
[)2u a2 u
8 t 2 =a2~+F(x,t) (0 ::; x ::; , t > 0),
'.IX
au(o, t)
al ax + blu(O, t) = pet), (t> 0),
u(x,O) = f(x), au(x,o) = g(x) (0 ::; x ::; e),

at
where a~ + bi > 0, a~ + b~ > 0 and f, p, q, f, 9 and F are given functions.
Example 4.37 Determine the solution u = u(x, t) of the problem:
u(O, t) =0 (0 < x < 00,0 < t < 00),
u(x,O) = f(x), au(x,O) =g(x) (O<x<oo),

at
au
such that u and ax are bounded when x -+ 00. Let the functions f, /" g and g' be
piecewise continuous functions and absolutely integrable.
Solution. We determine the solution by the Fourier method of separation of vari-

ables in the form u(x, t) = X(x)T(t). The solutions of the problem
X" + AX = 0, X(O) = 0,
where X(x) and X'(x) are bounded when x -+ 00, are
X(x) = sin(sx), s >0 (4.70)

4.4. THE STURM-LIOUVILJ,E PROBLEM 12]
(s < 0 does not produce a new solution). We have here a continuum of eigenvalues
generated by ). = s2. The solutions of the problem
for these eigenvalues are
T(t) = A(s) cos(ast) + B(s)sin(ast) (s > 0), (4.71)
where A(s) and B(s) are arbitrary functions. Using the superposition principle,
from (4.70) and (4.71) we obtain
1(A(s) cos(sat) + B(s) sin(sat)) sin(sx)ds.

<Xl
u(x, t) = (4.72)
o
The functions A and B can be found from the initial condition, as in Fourier
series method. From
1A(s)sin(sx)ds,
<Xl
u(x,O) = f(x) =
o
it follows
2 <Xl
1
A(s) = :; f(x) sin(sx)dx.
o
From
u(x,O) 1<Xl .
--at" = g(x) = saB(s) sm(sx)ds,
o
it follows
saB(s)
o
=:;
2
1g(x)sin(sx)dx.
<Xl
The conditions that the functions f, f', 9 and g' are piecewise continuous func-
tions and absolutely integrable, enable the convergence of integral (4.72).
Exercise 4.38 Determine the solution u = u(x, t) of the problem
fPu 2fPu
at 2 = a ax 2 ' u(O, t) = 0 (0 < x < 00,0 < t < 00),
au(x,O)
u(x,O) = 0, --,-:---,.-:...=e
at
-x
(0 < x < 00),
au
such that u and ax are bounded when x ---+ 00.
Answer. This is a special case of the previous example. The solution of this
problem is
00
u(x, t) =j (A(s) cos(sat) + B(s) sin(sat)) sin(sx)ds.

o
From
00
u(x,O) = 0 = j A(s) sin(sx)dx,

a
it follows A( s) = o. From
u(~~ 0) = e- =
X 7
a
saB( s) sin( sx )dx,
it follows
2 00
saB(s) = ;: . j e- sin(sx)dx, X
a
and
2 1
B(s) - - - -
- a7r 1 + S2
Therefore the solution has the form
2 jOOSin(sat)sin(sx)d
u ( x,t ) = -a7r s.
1 +s
2
a
Example 4.39. Determine the transversal oscillation of a rectangular membrane,
whose length is a and width b, if the edges of the membrane are fixed, its initial shape
is given with a continuous function f and its initial velocity is zero.
Solution. The physical assumptions lead us to the following problem over the set
{(x, y, t)1 0 S x S a, 0 S y S b, t ~ o} ;
fPu _ 2 (fPu {)2U) (4.73)
{)2t - c {)x2 + {)y2 '
u(O,y,t)=u(a,y,t)=O (0 S y S b, t ~ 0),
{ (4.74)
u(x, 0, t) = u(x, b, t) = (0 S x S a, t ~ 0),
u(x,y,O) = f(x,y),
{)u(x,y,O) = o. (4.75)
{)t
Let us find the solution in the form
u(x,y,t) = V(x,y) T(t), (4.76)

which, in view of (4.73) takes us to the equation
Figure 4.3 u(x, y, 0) = sin(nx) sin(my) for (n, m) = (1,4), (1,3), (2, 2), (2,5).
Figure 4.4 u(x, y, t) = sin x sin 4y cos( VI7 t) for t = 0,1,2,3.

4.4. THE STURM~LIOUVILLE PROBLEM 125
82 V 82 V 82 T
-2 + -
8 y = 82t
8 x 2 = _A (4.77)
V c2 T .
Hence A must be a constant, which implies
(4.78)
The conditions (4.74) give
V(O,y) = V(a,y) = 0, for 0:::; y :::; b, and

{ (4.79 )
V(x,O) = V(x,b) = 0, for 0:::; x :::; b.
The PDE (4.78) is elliptic. In order to find the solution of the problem (4.78),
(4.79), we shall apply the Fourier method of separation of variables, i.e., let us put
V(x, y) = X(x) Y(y). (4.80)
Equation (4.78) with conditions (4.79) gives
and X(O) = X(a) = 0, Y(O) = X(b) = O. One easily checks that nontrivial solutions
exist only if A > /1 > O. Then we have
X n () . nJrx . mJrY
X = sIn - - , Ym(Y) = sm -b-'
a
n 2 Jr2 m 2 Jr2
/1n =7 and A - /1n = -,;z-.
The boundary conditions (4.74) give now the eigenvalues
and the eigenfunctions
(m,n EN).
The second ODE from (4.77) is TI/( t) - Aa 2 T = 0, and from the last condition
in (4.75) we get T'(O) = O. We know that we must take A = Am,n, hence the
corresponding Tm,n are of the form
Tm,n(t) = Am,n' cos (a JAm,n) (m,n EN).

We shall determine the constants Am,n from (4.75). From (4.76) we get
u(x,y,t) = L LA00
m=l n=l
00
m,n
n7rX
sin--sin--cos c
a
m7ry
b
(
(4.81 )
The first initial condition in (4.75) gives
n7rX . m7ry
=L
00 00
f(x,y) L Am,n sm -a- sm -b- (m,n EN),
m=l n=l
i.e., Am,n are the Fourier coefficients of the double Fourier series of the function f.
As in the case of the Fourier series, in our case we have (check!)
Am,n = ab!~
4 (
!
b
. mrx . m7ry
f(x,y) sm- )
a - sm-b-dy dx, (m,n EN).
Presumably the reader suspects that analysis of the convergence of the series
(4.81) and its formal derivatives is rather complicated. Of course, as in the one-
dimensional case, additional suppositions on the smoothness of the initial function
f might imply the convergence of the solution given by (4.81). In particular, if
f(x,y) = x on (x,y) E [O,a] x [O,b], then the solution of the problem (4.73), (4.74)
and (4.75) is
8a 00 00 (-1 )n-l n7rX m7ry (

u(x,y,t)=2"LL sin--.sin-b-cos C
7r m=l n=l mn a
Figure 4.3 represents some eigenfunctions Fm,n for a = 7r and b = 7r.

Figure 4.4 represents the solution u for special initial condition f( x, y) = sin x sin 4y
for t = 0,1,2,3, for a = 7r and b = 7r.
Example 4.40 Show that the Laplacian 6.u

y( x, y) = U(r, tp) can be written as
(4.82)
Solution. The connection between the rectangular and polar coordinates is
x = r cos tp, y = r sintp.
From
au aU or - aU atp aU - sin tp aU
ox = -or. -+
-
ox atp. -ox =costp-
or ----,
r atp
aU ar aU a<p
-au = - . aU cos <p aU
. - + - . - = SlfI<p- + ----,
ay ar ay a<p ay ar' '1' a<p
it follows
wherefrom the relation (4.82) follows.
Example 4.41 The Oscillation of the Round Membrane (Drum)

Solve the following mixed problem for the two-dimensional wave equation on the disc
D=D(D,l)
a
2u
- 2 - c2 6.u =
on D x (0,00) (4.83)
at
ui8D = 0, (4.84)
( ) =mlx,y,
() au(x,y,O)_ ( )
uX,y,O at -m2 x,y ((x,y) ED). (4.85 )
Solution. First we apply the Fourier method of the separation of variables and we
are looking for the solution in the form
X(x, y) . T(t) ((x,y) ED).
Putting this in the equation (4.83) we obtain
~T"X - T6.X =0
c2
and separating the variables
T" 6.X
-=-=-A
c T
2 X '
where A is a constant. Therefore we have
(tE(D,oo)), (4.86)
6.X + AX = 0 on D. (4.87)
The boundary condition (4.84) implies
Xi8D = O. (4.88)
We shall solve the problem (4.87), (4.88) in the polar coordinate system (r, cp) and
using again the Fourier method. We take
X(X,y) = R(r) . 4> (cp),

where x = r cos cp, y = r sin cpo Then separating the variables and introducing a
constant, we obtain the equation
4>" +,4> = 0, (4.89)
where, isa constant, and the Bessel equation
r(rR')' + (Ar2 - ,) = o. (4.90)
The equation (4.89) has its eigenvalues 'k = k2 (k E Z+). The corresponding
eigenfunctions are coskcp,sinkcp (k E Z+). Taking, = k 2 in the equation (4.90)
we obtain the Bessel equation (for x = r..,f). for A 2: 0) with solutions In(r..,f).). The
boundary condition R(1) = 0 gives In(..,f).) = 0, which implies that the eigenvalues
are given by
(j EN),
where Xnl < X n 2 < ... are the positive zeros of the function I n
The corresponding eigenfunctions are Rkj(r) = Jk(rXkj) (j EN). The bound-
ary condition (4.88) implies R(1) = o.
Therefore the (formal) solution of the problem (4.83)- (4.85) is given by
l: l:((akj cos cxkjt + bkj sin CXkjt) cos kcp

00 00
u(r, cp, t)
k=Oj=l
+ (Ckj coscxkjt + d kj sin CXkjt) sinkcp)Jk(rxkj),
where the coefficients akj, bkj , Ckj, dkj are given by

4.5. MISCELLANEOUS PROBLEMS 129
4.5 Miscellaneous Problems

Example 4.42. Let us consider the hyperbolic equation
fPu 1
L(u) = - - + -u = 0 ((x, y) E R2), (4.91 )
8x8y 2
u(x, x) = f(x), 8u(x,x) = g(x) (x E R) (4.92)

8y
where f and 9 are given functions from the class C 2 (R).

Let M (xo, Yo) be an arbitrary point in the first quadrant (xo > 0, Yo > 0), and let
us denote by A and B the points (xo, xo) and (Yo, Yo) respectively (see Figure 4.5).
Figure 4.5
a) Prove that for any u and v from C 2(R2) it holds
where T denotes the interior of the triangle ABM.
b) Prove the equality
8v J8v J(8v
u(M).v(M)=u(B).v(B)-_u8xdY+_U8ydx+_
J 8u) ,
u8xdx+v8ydy
AM BM AB
provided that u and v are the solutions of (4.42).

(We used the usual notation u(P) = u(x,y), when the point P = (x,y).
c) Let v = v(x,y) be a known particular solution of (4.42) such that v(xo,Yo) f= 0,

ov
ox =0 Jor (x,y) E EM
(4.93)
ov
oy =0 Jor (x, y) E AM.
Determine then the solution u = u(x,y) oj the problem (4.91), (4.92) at the
point (xo,Yo).
d) Show that the Junction v(x, y) = w((x - xo)(Y - Yo)) is a particular solution of
(4.91) and satisfies the conditions (4.93), provided that the function w = w(z)
is a solution of the ODE
zwl/(z) + w'(z) + ~w(z) = O.
Solution. We shall construct the solution using the Riemann's method. For sim-
plicity, we shall assume 0 < Yo < Xoj the other cases can be handled similarly.
a) By the definition of the operator L, we have
JJ (v Lu - u Lv) dx dy = JJ oxoy
2
(v. 0 1 - u 02V ) dx dy.
oxoy
(4.94 )
T T
Further on, Green's formula gives
f ( 1 ov dx
ox
+ v ou
oy
dY ) = JJ (v. ~
oyox
- 1 ~)
oxoy
dx dy. (4.95)
6~M T
The assumptions 1, v E C 2 (R2) imply that the mixed second order partial
derivatives are equal, hence the right-hand sides of (4.94) and (4.95) are equal.
This gives us the equality in a).
b) If 1 and v are two solutions of (4.91), then a) gives us
= 1
BM
1 ~~ dx + 1 ~~ +1
MA
v dy
AB
(1 ~~ dx + v ~~ d y ) .
Now using the equality
1
BM
1 1
~: dx =
BM
(:x (1v) - v ~~) dx = 1(M)v(M)-1(B)v(B)-l v :~ dx,
BM
we obtain b).
4.5. MISCELLANEOUS PROBLEMS
c) From b), using the conditioIls (4.92), we get the solution u as
(1 ) (J(xo) v(xo, Yo)

V Xo, Yo
Xo Xo
Note that all the quantities on the right-hand side are knowIl.
d) Left to the reader.
Remark 4.42.1 It is clear that Riemann's method can be applied also when the
segment AB from the line y = x from Figure 4.5 is replaced with a smooth increasing
curve C, see Figure 4.6.
YoIT
Y
M
,, ,
~ '
,
I ,
Xo
:A x
Figure 4.6
Example 4.43. Find the function u = u(x, t), 0 ~ x ~ C, t 2: 0, such that
ou(O, t) ou(C, t) = A. e-t

ox = 0, ox '
u(x 0) = A. a cosh(x/a) ou(x,O) = -A. a . cosh(x/a).

, sinh(C/ a) , ot sinh(C/a)
Solution. Let us put u(x, t) = v(x, t) + e-tw(x), where the function v satisfies the
homogeneous wave equation and the conditions
ov(O, t) _ ov(C, t) _ 0 for t> 0

ax - ox - - ,
v(x,O) = OV~t'O) = 0 for 0 ~ x ~ 1'..

Then the function wshould satisfy the conditions
w'(O) =0 and w'(C) = A.

Clearly the function v = v(;r, t) is identically equal to 0, while the function w is
easily found to be w(x) = A a, ~i:~~~j!)a). Thus the sought after solution is
-t a cosh(x/a)
u(x,t) = Ae . sinh(C/a) .
Examrle 4.44. Prove that the following Cauchy's problem on the set
{(x,y) Ixl<l, O<y<l}:
Bu(x, x 2 ) = g(x) (Ix I < 1)

By
has a solution only if 9 is an even continuous function on [-1, 1] such that the
function xg(x) is continuously differentiable on [-1,1]. Prove then that the unique
solution of the given problem is
J
VY
u(x, y) =2 z g(z) dz. (4.96)
x
Solution. Since the solution of the given PDE is of the form u( x, y) = A( x) + B(y),
for some functions A E C1 ([-I, I]) and B E C 1 ([0,1]), the initial conditions give
that
(x E [-1,1]),
hence 9 must be an even function. Multiplying the last equality with 2x gives
J
x
B(x 2 ) =2 zg(z)dz,
o
hence
J
VY
B(y) = 2 zg(z)dz if 0::; y::; 1.
o
Thus we obtain formula (4.96).
4.5. MISCELLANEOUS PROBLEMS 133
Example 4.45. Prove that the nonhomogeneous wave equation on R Z
82 u 82 u
-82 --8z =6(x+t),
x t
which satisfies the initial conditions
8u(x,x) ()
u(
x, )
x = 0, 8x = 9 x
has a solution only iJ g(x) = 3x 2 + C, Jor some constant C. P1'OVC then that the
solution is not unique, but all the solutions are oj the form
u(x,t) = x 3 - t3 + f(x - t) - 1(0) + (x - t)(g(O) - 1'(0)), (4.97)
where I is an arbitrary Junction J1'Om C 2 (R).

Solution. Using the change of variables (= x + t, 71 = x - t, v((, 71) = u(x, t), we
obtain
82v 3(
8(81] 2
Thus
3
v(CTJ) = 4 eTJ + rjJ(O + 1/;(71), Y,1/; E C Z (R1),
or
u(x,y) = ~(x + t?(x - t) + rjJ(x + t) + 1/;(x - t).
The first initial condition gives us
0= u(x,x) = rjJ(2x) + 1/;(0), hence rjJ(x) = -1/;(0), x E R,

which implies
3
u(x, t) = 4(x + t)2(X - t) -1/;(0) + 1/;(x - t). (4.98)
The second initial condition gives
g(x) = 8u(x,x) = ~(x + i)2 + ~2(x + t)(x - t) + 1/;'(x - i)1 = 3x 2 + 1/;'(0),

8x 4 4 x=t
which is, in fact, the sought after condition for the function 9 with C = 1/;'(0).
So we obtain from (4.98) the solution u as
u(x, i) = x 3 - i3 + 1/;l(X - i) -1/;1(0),

for some 1/;1 E C 2(R). Applying Taylor's formula we get
u(x, t) = x 3 - t3 + (x - i)(g(O) -1/;; (0)) + 1/;2 (x - t) -1/;2(0),

for some 1/;2 E C 2(R). The obtained form of the solution u is equivalent to (4.97).
134 CHAPTEH 4. HYPEHBOLIC EQUATIONS
Example 4.46. Lct the following two second order PDEs be given:
1
- < A < I, (4.99)
2
and
A OV
02 BOv
- -
T(A,B)(v) = _v_ _ ~ + -.!!JL = 0, (4.100)
oxoy x- y x- y
for 0 < A < I, 0 < B < I, A + B =11, where U = U(XI,YI) and v = v(x,y) are
the unknown functions, respectively.
a) Prove that in the half plane {(Xl, yd I YI < O} equation (4.99) is hyperbolic. Then
transform (4.99) with the change of variables X = Xl -2~, Y = Xl +2y'-YI.
b) Check that both functions f and g, where
t a real parameter,
and
J(t) fA,B(x, y, t) dt,

y
g(x, y) = an arbitrary continuous function on R,
are solutions of equation (4.100).
c) Prove that the function vA,B(x,y) = (y - X)l-A-Bvl-A,l-B(x,y) is a solution of

(4.100), whenever vl-A,l-B is.
d) Using b) and c), find all solutions of equation (4.100), and then solve also equa-
tion (4.99).
Hints. We leave to the reader to check that the given change of variables transforms
equation (4.99) into
T(A - 1/2, B - 1/2)(v) = 0, (4.101)
where v(x,y) = U(XI,YI). Also, parts b) and c) are omitted. However, they give us
the form of the general solution of (4.100), namely it is
J J
y y
v(x,y) = 1(t)fA,B(x,y,i)di+(y-x)I-A-B 2(t)!I_B,I_A(X,y,t)dt,

x x
4.5. M1SCELLANEOUS PROBLEMS 135
where <1>1 and <1>2 are arbitrary continuous functions on R. Thus t.he general solution
of equation (4.99) is
J
xl+2v=Y!
u(xl,Yd= <1>1(t) (t_xl+2,j_Ylr/2-A(xl+2~_tr/2-A dl
xl-2v=Y!
J
xl+2v=Y!
+ (4v=1ht2A <1>2(t) (t - XI + 2,j_YI/ I 2+A (XI + 2.../-YI - tf/2+ A dt.

xl-2v=Y!
Example 4.47 (Poincare's inequality on n-dimensional parallelepiped)

Prove that on n- dimensional parallelepiped
a) PI = (0,1) x ... x (0,1) and f E CI(PI ) we have the following inequality
j If(x)I' dx S; V f(x) dx ' + ~ j t,ID;J(x)I' dx, (4.102)
b) P = (aI, bl ) X ... X (an, bn ) and f E G I (P) we have the following inequality
(4.103)
Solution.
a) The basic integral formula gives us
JDd(xl' ... ,Xi-I,Si,Yi+I, ,Yn)dsi

n Y.
fey) - f(x) = L (4.104)

z=l Xi
for x, Y E Pl. Multiplying the equality (4.104) by the conjugate equality

(4.104) we obtain
(f(y) - f(x))(f(y) - f(x))
(
[;;!
n Yi
Dd(XI, ... , Xi-I, Si, Yi+1,, Yn) dS i .

Hence by the triangle inequality, the inequality

n n
(2: Iz;I)2 ::; n(2: I iI 2)Z
i=l i=l
and inequality IJI ::; J 11

If(y)1 2 + If(xW - f(y)f(x) - f(x)f(y)
::; n (t]
=1 0
ID;J(XI, ... , Xi-I, Si, Yi+I,, YnW dSi) .
Applying the integral with respect to X and y on parallelepiped PI we obtain

( using the iterated integrals - see also the Fubini theorem from S.l
i If(y)I' dy + i If(x)I' dx - U Q fly) dY) f(x) dX)
-U Q f(X)dX) f(Y)dY) S n t.jlDd(x)I'dX,
which gives the desired inequality (4.102).
b) The inequality (4.103) reduces on (4.102) by changing the variables

_ Xi - ai
Xi=-- (i=l, ... ,n).
bi - ai
Exercise 4.48 Prove that the solution u = u(x, t) of the mixed problem
au)
8 ( p(x)a
ax x + q(x)u = F(x, t) for
P E C 1 [0,b], q E C[O, b] (p 2: Po> O,q 2: 0),

8u(x,0)
u(x,O)=f(x), at =g(x) (0 ::; x ::; a),
(0 ::; t ::; T,)

where f(O) = rl(O), feb) = r2(0),g(0) = r~(O),g(b) = r~(O), depends continuously
from the initial conditions f and g.
4.5. MISCELLANEOUS PR.OBLEMS 137
Hints. Using the energy integral
E(t) = J(~~)2 + p(~~? + qU

o
2) dx
show that for two solutions U1 and U2 with the same boundary conditions and initial
conditions f,9 and f1, 91, respectively, for every > 0 there exists 8 > 0 such that
max(max If1(X) - f2(X)I, max If~(x) - f~(x)l, max 191(X) - 92(X)l) < 8
xE[a,b] xE[a,b] xE[a,b]
implies
IU1(X, t) - U2(X, t)1 < for x E [0, b] and t E [0, T].
Exercise 4.49 (Volterra's method) The equation
cPu + a(x,y) ou
oxoy
ou
ox + b(x,y) oy + c(x,y)u = F(x,y), (4.105)
where a, b, c, F E O(P), P = [Xo, Xl] x [yO, yd, with the boundary conditions
ou(x, Yo)
u(xo, y) = rl(Y) for Yo::; Y ::; Y1, 0
y.
= r2(x) for Xo ::; x ::; Xl>
where r1 E 01[yo, yd, r2 E 0 1[xo, Xl] and r1 (yo) = r2(xo), has a unique solution on
P.
ou
Hints. Let -
ox
v and ~~ = w. Then the equation (4.105) reduces on the
equation
-ov = -ow = F - av - bw - cu.
oy ox
Therefore we have
y y
v(x, y) = v(x, Yo) + j (F - av - bw - cu) dy = r~(x) + j (F - av - bw - cu) dy

YO Yo
and
x x
w(x,y) = w(Xo,Y) + j(F-av-bw-cu)dx=r~(y)+ j(F-av-bw-cu)dx.

xo xo
The third integral equation is (integrating ~; = w)

y
u(x, y) = r2(x) +j w dy.

YO
The system of three integral equations with three unknown functions u, v and w
is equivalent to the starting boundary problem. This system can be solved by the
method of iterated approximations.
138 CHAPTER 4, HYPERBOLIC EQUATIONS
Exercise 4.50 Solve the following qlJ,asi-linear equations
a) the Riccati equation oov - ~V2 = F(x)

x 2
where F is an aT'bitrary continuous function and v = v(x,y);
'
b) t Iw L wuvz'tt ,02U u
e equatwn oxoy = ce ,
where c is an arbitrary constant different from zero and u = u(x, V);
02U 02U
c) ox 2 + oy2 = 4ce u ,
where c is an arbitrary constant different from zero and U = U(x, y),
Hints.
a) Taking (the Schwartz derivative)
plll(X) 3 pll\X)
F(x) = pl(X) - 2' p/2(x)
for p E C 3 , we can easily see that

p"(x) 2p'(X)
v(x, y) = pl(X) - p(x) + q(y)'
for an arbitrary function q of the class C 1 , is the solution of Riccati equation.
b) Differentiating the Liouville equation with respect to x we obtain

02U u Ou
oyox = ce ax'
Eliminating ce from the preceding equation and Liouville equation and intro-
U
' a new f
d ucmg '
unctIon v by v = au
ax we 0 b ' htR
tam e "lccatl equatIOn.
. H ence
u 2 pl(X)ql(X)
e = - .,.......::,......:-....:....:..-'-:-:,-,.-
C (p(x) + q(y))2
c) Introduce new variables z = x + iV, z = x - iy and
z+z z-z
u(z, z) = U(-2-' 2i)
and use b) to prove that the desired solution is given by
u u 2 pl(Z)~
e=e=- ,
c (p(z) + p(z))2
where p is an arbitrary analytic function.
4.5. MISCELLANEOUS PR.OBLEMS 139
Exercise 4.51 Find a solution of the sine-Gordon equation

;:)2
(J U. .
fjxfjy = sm u
in the form ll(x, y) = p(ax + by), where a and b arc constants of the same sign.
Hints. The problem reduces to the ordinary differential equation
p "() 1smp,
z = ab .
where z = ax + by. Multiplying it by 2p' (x) and integrating we reduce it to the

ordinary differential equation of the first order
,2 2 2C1
P = - ab cos p + --;;;;'
where C 1 is an arbitrary constant. We obtain a real solution ( for C1 2 1) from
J
p(x)
ds _ 2 )1/2
v'C
1 - cos s
- ( -b
a
Z + C2 ,
o
where C2 is an arbitrary constant.
Example 4.52 Prove that the Cauchy problem

fj 2 u
oxoy =0 on f{ = {(x, v)llxl < 1, Ivl < I},
has a solution only for a = 0 or a 2 6. Prove that for these cases the unique solution
is given by u(x, V) = Ivl</3.
Solution. Since the general solution is of the following form 1l( x, V) = p( x) +q(V) for
arbitrary functions p, q E C 1 (1() we obtain by the initial conditions p( x) + q( x3 ) =
Ixl'" and p'(x) = O. Hence p(x) = C and q(x 3 ) = Ixl" - C, where C is an arbitrary
constant. Therefore q(x) = Ixl"/3 - C, which implies
This function u is the solution of the considered initial problem if u E C 2 (1<), what
is satisfied only for a = 0 or ~ 2 2, i.e., a 2 6.
Example 4.53
a) Consider the Jollowing Junctional equation
f(x) + kf(a(x)) = b(x), (4.106)
wher'e f is an unknown Junction, k a real constant and a and b continuou8

Junction8 defined on R. We denote the composition oj the function a with
itself in the following way:
actO) = Id, aO(l) = a, ao(m) = a(ao(m-l)).

IJ there exists < M <1 such that
(m E N,x E R), (4.107)
then the solution of the equation (4.106) is given by
L (_l)mk m b(ao(ml(x))
00
J(x) = (x E R). (4.108)

m=O
b) Using the result from a) find the solution of the problem
[Pu a2u
-2--2=0,u(x,0)=slnx,u x,-
. (X) =x
& ~ 4
on the angle {(x,y)1 x > 0,0 < y < ~}.
Solution.
a) The condition (4.107) implies the uniform convergence of the series (4.108) on
R. Hence (4.108) defines a continuous function on R. Taking it in the equation
(4.106) we verify that it is a solution of this equation.
b) The general solution of the given PDE is given by
u(x,y) = p(x + y) + q(x - y),
where p, q E C(R). The given conditions imply
p(x)+q(x)=smx,p . (5X)
4 +q (3X)
4 =x.
Taking y = 5: we obtain a functional equation of the form (4.106)
q(x)-qGx) =sinx- ~x, (4.109)

4.6. TI-IE VIBRATING STRING 141
where k = -1, a(x) = ~x, b(x) = sin x - 4x. It is easy to prove that ao(m)(x) =
5 5
(~)mx for m = 0,1, ... , and
If x belongs to some compact subset f{ of R, then we have
where M = M(K) E (0,1). Hence the condition (4.107) is satisfied for m ~

mo(K). Therefore by a) the solution of the equation (4.109) is given by
Then the solution u of the considered problem is given by
u(x,y) = sin(x + y) - fo (sin((~)m(x + y)) - ~. (~)m(x + y))

+ f: (sin((~)m(x
m=O 5
- y)) - ~. (~r(x - y)) .
5 5
4.6 The Vibrating String

Let us consider a perfectly elastic string (for example a violin string) which is stretch
with tension T between two fixed points x = 0 and x = f. on x axis. We assume
that the string is uniformly covered with constant density p.
If the string is somehow set into motion in the vertical plane, such that the
deflections u(x, t) are very small, then we shall analyze the subsequent motion of
the string. We assume that the gravity force is ignored.
Let x and x + b.x be two points on the string (b.x ~ 0) and assume that only
vertical components of the tension T are acting at these two points. Let al and a2
be the angles between T and the x-axis at the points x and x + b.x, respectively.
Since by assumption these angles are small, it holds sin a ~ tan a. So we have
(4.110)
.
T sm T au(x,x+b.x)
a2 ~ - tan a2 = - ax .
142 CHAPTER 4. HYPERBOLTC EQUATIONS
From the Newton's second low of motion it follows that the sum of the forces (4.110)
. equa1 to t 1
IS
. 1 f orce PLJ.X
Ie mertla A cPu
ot 2 :
T( OU(X,X+.6.X)
OX
_ ou)
Ox
= 02.U A
P ot2 LJ.X.
Let us divide the previous relation with T.6.x and take the limit as .6.x --t o. Then
we have
OU( x, x + .6. x ) OU
lim
ox - 7h _ f 02U
Ax->O .6.x - T ot 2 '
which leads us to the one dimensional homogeneous wave equation
02U _202u
ox2 = a ot2 .
If an external force F is acting to the string, then we obtain the one dimensional
nonhomogeneous wave equation
Chapter 5
Elliptic Equations
5.1 Dirichlet Problem

5.1.1 Preliminaries
We consider the equation given by
where aij = aji i,j = 1,2, .. . n, aij,bij,c E C(Q) and L is an elliptic operator, i.e.,
n n
LLaij(X)PiPj > 0,
i=1 j=1
for all (PllP2, ... ,Pn) E Rn \ {(O,O, ... ,O)} and for every x E Q, where Q is a
bounded region of R n with a boundary oQ, which has piecewise continuous normal
at each point.
Let us introduce the following problems where f E C( oQ).
1) Dirichlet problem (the first boundary value problem) Find u E C 2 (Q) n C(Q)
such that
L(u) = F on Q
2) von Neumann problem (the second boundary value problem) Find u E C 2 (Q) (~
C 1 (Q) such that
L(u) = F on Q, oul aQ -- f ,
on
where ~~ is the directional derivative of u along the outward normaL
143

144 CHAPTER 5. ELLIPTIC EQUATIONS
3) (The third boundary problem) Find u E C 2 (Q) n C1(Q) such that
au
L(u) = F OIl Q, on + a = f (x E oQ).
where a E C(oQ).
A function u E C2( Q) which satisfies the Laplace equation
flu=O (xEQ),
is called harmonic function on Q.

Let an arbitrary continuous function defined on the unit disc f be given in polar
coordinates f(x,y) = A('P), for x = cOS'P, Y = sin'P on interval [-7r,7rJ, such that
A(-7r) = A(7r).
The function u given by (the Poisson integral formula)
A('P), r =1
{
u(r,'P) =
---
27r11 -
1'" A(t)(1-r2)
2r sine t - 'P) + r2
dt
'
represent the solution of Dirichlet problem for flu = 0 on the unit circle
{(r,'P)IO:::; r:::; 1, -7r:::; 'P:::; 7r}.

Example 5.1 Determine the formal solution of the Dirichlet problem
02U cPu
ox 2 + oy2 = 0 (0 < x < i, 0 < y < i 1 ),
(5.1)
u(O,y) =0, u(i,y)=O (O<y<i t ),
u(x,O) = 0, u(x,e l ) = f(x) (0 < x < i).

Solution. We use the method of separation of variables. Taking
u(x,y) = X(x) Y(y),

we obtain two differential equations
X"(x) Y"(y)
X(x) = -A, Y(y) = A.
5.1. DIRICHLET PROBLEM 145
Using the conditions from (5.1) we get two problems
X"(x) + AX(X) = 0 (0 < x < e) X(O) = 0, X() = 0, (5.2)

Y"(y) - AY(y) = 0 (0 < y < e1 ), y(ed = O. (5.3)
Let us first solve the equation (5.2) depending on variable x. The eigenvalues of
the considered problem are
(n EN), (5.4)

X n () . n7rX
X = sln- e- (n EN).
For A given by (5.4) the solution of ordinary differential equation in problem

(5.3) has the form
(n EN),
where An and Bn are arbitrary constants. From the boundary condition Y(O) = 0,
we obtain that An = O.
The formal solution of the considered problem is
n7rX . n7ry
u(x,y) = L Bn sm -e-'
00
smh -e-'

n=l
for 0 < x < e, y e

0 < < 1.
This solution has to satisfy the condition u(x, ed = f(x) and therefore for y = e1
we obtain the Fourier series
where
Bn =
2
e
h n7r 1
Jf(x) . -e-
l
n7rX dx
sm (n EN).
o .
{;sm -e- 0

/)2u 02U
ox2 + oy2 = 0 (0 <' -- < e, 0 < y < ed,
with the conditions

u(O,y) = 0, u(e,y) = 0 (0 < y < e1 ),
u(x,O) = 2, u(x,ed = 0 (0 < x < e).
Solution. Using the method of separation of variables we obtain the problem (5.2)
with the eigenvalues
n 2 1r2
A = An = T (n EN),
and the corresponding eigenfunctions
X n () . n1rX
X = sln-- (n EN).
Also, we get the problem
ylI(y) - AY(Y) = 0 0 < y < 1,

The solution of the equation has the form
(n EN),
where An and En are arbitrary constants. From the boundary condition Y(d = 0,
we obtain
wherefrom it follows that
the solution Yn(Y) can be written as
n1rl
n 1ry cosh -- . n 1ry
An cosh -- - An . smh --
sinh n1r 1

1 ( n1ry. n1rl n1rl. n1r Y )

An cosh - - . smh - - - cosh - - , smh - -
, h n1rl
sm --
An 1 'h n1r (l-y)

D 'SIn D '
. h n1r{.l {.
sm --
The formal solution of the considered problem is
00 1 ,n1rx. n1r(l - y)
u(x,y) = LAn sm --' smh '
n=l sinh n1r 1

for 0 < x < f, 0 < y < fl.

This solution has to satisfy the condition u(x,O) = 2 and therefore for y = 0 we
obtain the Fourier series
00 n7rX
u(x,O) = 2 = LAn sm -f-'
n=l
where
2Je
.
An = 7 2 sm -f- dx = 4
1 + (_l)n+l
n7rX
n7r (n EN).
o
The formal solution of given elliptic equation has the form
( x, t ) -_ -8 L~ 1 . (2n -1)7rX . h (2n - 1)7r(1 - y)

U () sm sm .
SIn h 2n-17rf
-1(2n - 1)
7r n_ f f
1
f
Example 5.3 Find the solution of the problem
fPu fPu
ox 2 + oy2 = 0 (0 < x < 1, 0 < y < 2),
with the conditions
ou(O,y) _ 0 ou(l,y) =0 (0<y<2)

ox - , ox
u(x,O) = x2, ou(x,2)

oy
=0 (O<x<l).
Solution. Taking u(x,y) = X(x) Y(y), we obtain the problem
X"(x) + AX(X) = 0 (0 < x < 1), X'(O) = X'(f) = 0,

which eigenfunctions are
Xn(X) = cos(n7rx) (n EN).
Now, the solution of the equation
Y"(y) - AY(y) =0 (0 < y < 1),

is given by
Y. ( ) _ { Cy + D, n =0
ny - An cosh(n7rY) + En sinh(n7rY), n ~ 1.
= Cy + D + I: (Ancosh(mfY) + Bnsinh(nr.y)) cos(nr.)x.

00
u(x,y)
n=l
Using the condition u(x, 0) = x 2 we obtain
= x 2 = D + I: Ancos(nr.x),
00
u(x,O)
n=l
where
1
D = Ao = j
2 1
x dx = 3"'
o
1 (_l)n
An = 2jx 2 cos(nr.x)dx = 4n-2 -r. 2- (n ~ 1).
o
.
From t he con dltIons 8u(x,2)
8y = 0, we get
8u(x,2)
8
y
= C + I: -
=
00 1
n=l n r.
(An sinh( nr.2)
.
+ Bn sm cosh(nr.2)) . cos(nr.)x,
wherefrom we have
Hence
Bn = _4(-lt sinh (2nr.).
n 2 r. 2 cosh 2nr.
The solution of given elliptic equation has the form
u(x,y) =~ f (-It (cosh(nr.y) cosh(2nr.) - sinh(nr.y) - sinh(2nr.)) cos(nr.x)

r. 2 n=l n 2 cosh nr.
1 1 4 (-It
+ 2" I:
(Xl
+-3 = -3 h 2 . cosh(nr.(2 - V)) cos(nr.x).

r. n=l n cos nr.
Example 5.4 Determine the harmonic function u = u( x, y), on rectangle P( a, b) =

{(x,y)IO < x < a,O < y < b} satisfying the conditions
u(O,y) = u(a,y) = 0 (O<y<b),
u(x,O) = 0, u(x, b) = x(l - x) (0 < x < a).

Solution. The function u on P is harmonic if it satisfy

[Pu 8 2u
u E C 2 (R), 8x 2 + 8y2 = 0 (x,y) E P.
The eigenvalues of the problem

XI/(x) + AX(X) = 0 (O<x<a), X(O) = 0, X(a) = 0,
are of the form
(n EN),
and eigenfunctions are
. n7fX
X n ()
X = SIn--
(n EN).
a
For these values An the solutions of the problem
yl/(y) - AY(Y) = 0 (O<y<b), Y(b) = 0,
are given by
Yn(Y) = An cosh -n7fy
a
+ Bn smh
. n7fy
-
a
(n EN),
(
5.5
)
where An and Bn are arbitrary constants. From the boundary condition Y(O) = 0,
we obtain An = 0 and therefore the solution of the considered problem is
B . h n7fY . mrx
u (X,Y ) =~
L...J nsm - s m - - .
n=1 a a
This solution has to satisfy the condition u(x, b) = x(l - x) and therefore for
Y = b, we obtain two Fourier series
~ B . h n11"b . n11"X
U ( x, b) = x ( 1 - x ) = L...J n sm - sm - - (0 < x < a),
n=l a a
where
Bn . h -n 11"b =
. SIn
2
-
fa x(l- x) sm. -n11"X
dx
a a a
o
n = 2k +1
(k EN).
n = 2k
So the solution u can be written as

4a 2 00 1 2n11"Y . 2n11"x
u(x,y) = - ~ . sinh - - . sm--.
11"2 -1
n- n 2 .
SIn
h -n11"b a a
a

82 u 82 u
-8x 2 + -8y2 = u (0 < x < e, 0 < y < C1 ),
wdh the conditions

u(O,y) = 0, u(f,y) = 0 (0 < y < ed,
u(x,O) = f(x), u(x,C1 ) = g(x) (0 < x < C).
Solution. Taking u(x, y) = X(x) . Y(y), we obtain the problem
XI/(x) + AX(x) = 0 (0 < x < e), X(O) = 0, x(e) = 0,
with tbe solutions
n7rx
Xn(x) = sin -C- (n EN),
n 2 7r 2
for An = T' For these eigenvalues the solution of equation
ylI(y) - (1 - A)Y(y) =0 (0 < y < C1 ),

IS
/ n7ry . / n7ry
Yn(Y) = An cosh y 1 + -f- + Bn smb y 1 + -f- (n EN),
u(x,y) =];00 (
Ancosh y1 +
/ n7ry
-e- . /
+ B n smh y 1 +
n7r Y). n7rX
-f- sm-C-
Using the conditions u(x, 0) = f(x), u(x,f 1 ) = g(x) we obtain
f (x) = L (An cosh 1 + Bn sinh 1) sin -f-

00 ~x
(0 < x < f)
n=l
and
g(x)
~( AncoshVl~
=~ . . ~. n7rX
+ T-f- +smh y1 + T-f-) sm-f- (0 < x < e).
Therefore the coefficients An and Bn are given by
An cosh 1 + Bn smh 1 ="

. 2 Jf(x) . -e-dx,
l
n7rX
sm
o
fJ2u a2 U
aX 2 + ay2 - ku =0 (0 < x < i, 0 < y < J\),
for k > 0 with the conditions
u(O,y) = cp(y), au(i,y) = 1jJ(y) (0 < y < .ed,

ax
u(x,O) = f(x), U(X,i1) = g(x) (0 < x < ).
Solution. The equations appearing in this problem and in previous one are
called Helmholtz equations and for k = 0 it becomes Laplace's equation. Let us
introduce the functions v and w such that u(x,y) = v(x,y) + w(x,y), and the
function v satisfies the problem
(0 < x < i, 0 < y < i1)'
av(i,y) = 0 (0 < y < i 1 ),

(5.6)
v(O,y) =0,
ax
v(x,O) = f(x), v(x,lt) = g(x) (0 < x < i),
while the function w satisfies the following problem
(0 < x < i, 0 < y < i 1 ),
aw(i,y) = 1jJ(y) (5.7)

w(O, y) = cp(y),
ax
w(x,O) =0 w(x,id=O (O<x<i).
Let us first solve the problem (5.6). Taking v(x,y) = X(x) Y(y), we obtain the
problem
X"(x) + AX(x) = 0 (0 < x < i), X(O) = X'(l) = O.
The eigenfunctions of this problem are An = (2n (;i~t1f2, and the corresponding
eigenfunctions are
X() . (2n-l)1fx
n x = sm 2i (n EN).
The solution of equation
yl/(y) - (k - ,\) (0 < y < ed, Y(y) = 0
IS
(n EN),

The solution of the problem (5.6) is
v(x,y) = L
00
n=1
(
An cosh
J k+
(2n - 1)7rY
2
.
+Bn Slllh k+
(2n - l)7rY
2
. (2n - 1)7rx
. Sill 2
Using the conditions v(x,O) = f(x), U(X'I) = g(x) we obtain
Ancoshvk+Bnsinhvhk=e
J. /, 2 J f
f(x) sin
(2n -1)n7rX
2 dx, (5.8)
o
and
Je ( ) . (2n -
. J+
slllh k
1 -2
(2n - 1)7r1
2 0
9 x Sill
1 )n7rX dX
2
(5.9)
A h k (2n - 1)7r1
- n cos + 2 .
Let us find the solution of the problem (5.7). Taking w(x,y) = X(x) Y(y), we
obtain the problem
yl/(y) + >'Y(y) = 0 (0 < y < 1), YeO) = 0 = Y(I) = 0,
with solutions
y.n () . n7ry
y =sln-e;-.
The solution of equation
XI/(x) - (k - '\)X(x) =0 (0 < x < e),

IS
(n EN).
w(x,y) = L00 (
Ancosh ~7rX
k+ -C-+Bnsinh ~7rx. n7ry
k+ -C- .sm-C-
n=l 1 1 1
Using the conditions
= cp(y),
w(O,y)
ax = 1jJ(y)
ow(C,y) (0 < y < C1 )
we obtain
00
cp(y) = ~ (An cosh v'k + Bn sinh v'k) sin n~y (0 < y < Cd
and
n7r ( . ~7rC
~ An smh k+ T
00
1jJ(Y)=L
k n7rC
n=1 n
2f-1 +-C1
1
+Bn cosh ~ 7rC) n7ry

k + -C- sm -C-
.
1 1,
Therefore the coefficients are given by
Jcp(y)
l,
An cosh v'k + Bn sinh v'k = : sin n;y dy,
1 0 1
and
Example 5.7 Find the solution of the Poisson equation
with the conditions
u(O,y) =0, u(C,y) =0 (0 < y < C1 ),

u(x,O) = 0, u(x, C1 ) = 0 (0 < X < C).
Solution. If we take u(x,y) = Sex) + v(x,y), we obtain

S"(X) = -1, S(O) = 0, S(i) = O.
S(x) = ~ (c - D.
The function v satisfies the Laplace equation in the problem
lJ2v {J2v
ox 2 + oy2 = 0 (0 < x < i, 0 < y < il)'
v(O,y) =0, v(i,y) =0 (O<y<Ct), (5.10)
v(x,O) = u(x,i l ) = -~ (i -~) (0 < x < i).
The formal solution of this problem is

~ ( mry . mry). n7rX
vex, y) = L.J An cosh -e- + Bn smh -e- sm -e- (0 < X < e).
n=1
Using the boundary conditions we have
x ( X)
-- f-- = 2: An sm -n7rX dx .
00
2 2 n=1 i
x ( f - -X) = 2: An cosh - -l + B n smh
n7re . -l ) s
n7re .mn7rX
00
--
2 2
(

-
-
.
n=1
Therefore
42
f2 Jt "2x (
n = 1,3,5, ...
n7rX d _ { - n 37r 3 '
A -
n - - -"2X).sm -- x-
o 0, n = 2,4,6 ....
Also, we have
n7rfl
A cosh -
n
- + B n smh
. n7r1
-

2 -x ( f - -X).
- = --
22
J
sm -
l
n7rX
dx
'
o
n = 1,3,5, ...
0, n=2,4,6 ....
The solution of the problem (5.10) has the form
. ((2n - l)1rX)
) _ 4.e ~ Sill
2 f ( . h (2n -l)1rY . h (2n -1)1r(f2 - y))
V
(
x, t - - 3 L...J
1r n_- 1sm' h
((
2n _1) 1rq n) sm .e + Sill f .
f
The solution of the considered problem can be written as
sin ((2n - l)1rX)

u(X,y) x ( X) 4f2 00 f
2 .e- 2 -;J~
n_
.
sin h((2n-l)1rf
-'--_--'-_1 )
f
. h (2n - l)1rY . h (2n - 1)1r(f2 - y))

. ( sm f + Sill f .
Example 5.8 Determine the solution of the problem given in previous Example 5.7
by using double Fourier series.
Solution. Taking u(x,y) = X(x). Y(y) we obtain the problems
X"(X) + AX = 0, X(O) = 0 = X(f) = 0,

ylI(x) - AY = 0, Y(O) = 0 = Y(ft} = O.
Using the eigenfunctions of the previous problems we can write the formal solution
of the consider equation as
00 00 m1ry . n1rX
u(x,y) = L LAmnsm-f- .sm-.e-'
m=1 n=1 1
Using
82u 82u __ ((m1r)2 (m1r)2). m7rY . . n1rX

8 2
x
+ 8 y2 - L
00
L
m=ln=l
00
Amn f +.e sm f sm f
I 1
and the Fourier series for the constant 1
1=~t t l sin (2m - 1)1ry . sin ~(2_n_-~1)~1r_X

7r m=ln=l (2m - 1)(2n - 1) fl .e
we obtain that A 2m2n = 0, (m EN), (n EN), and
4 (( (2m f- 1)1r 2 (2n - 1)1r 2)-1

A(2m+I)(2n+I) = (2m - 1)(2n - 1)1r ) +( f )
for m = 1,2, ... , (n EN).

So the solution of the consider problem can be written as
. (2m - 1)1['Y . (2n - 1)7fX
4 00 00 4 sm . Sill
u(x, y) = -
1['
L L (1((2m - 2 2) .
m=! n=l ( (
2m - 1) 2n - 1 1['
)
1)1[') + ((2n -f. 1)1[')
Example 5.9 Determine the solution of the Dirichlet problem for the unit ball
B(O,R o) C R2
[PU 18U 1 8U
-81'2 + - - +-
l' 81'
-=0
1'2 8<p
(0 < l' < Ra, -1[' < <p < 1['),
U(Ro,<p) = f(<p) (-1[' < <p < 1[')

for f E C 2 (8B).
Solution. Separation of variables U(r, <p) = R(7' )H( <p) leads to
(PU + -""i.}
o= ~ 18U + 2"~
1 8U = R"() H (<p) + -R <p + 2"R
1 '() H () 1 () H "( <p.
l'
) l' l'
ur ur u<p
l' l' l' l'
Dividing by R(r)H(<p) and multiplying by 1'2 we get
R " + -R
1 ,
H"
1'2 r = -- = A
R H'
wherefrom we obtain the problem
H" + AH = 0, (5.11 )
and the equation

r2R" + rR' - AR = O. (5.12)
The solution of equation (5.11) for A = An = n 2 , n = 0,1, ... , is
n=O
(5.13)
(n EN).
The general solution of Euler equation (5.12), for A = An = n 2 , n = 0,1) ... , is
n=O
(5.14)
(n EN).
III order to avoid discontinuities of In rand r- n , n E N for r = 0, we put C2 = C4 =

O. So the solut.ion of considered equation is
1 rn
= 2Ao + L Rn (An cos ncp + Bn sin ucp) .
00
U(r, cp) (5.15)

n=l 0
Using the boundary conditions U(Ro, cp) = f(cp) i.e.,

1
2Ao + ~ (An cos ncp + Bn sin ncp) ,
00
U (Ro, cp) = 1 (cp) =
we shall find An and Bn. The last series is uniformly convergent, since 1 E C 2 (oB).
The Fourier coefficients can be written as
Ao = ~
271"
J1f f(O)dO
-1T
R~An = ~ j 1(0) cos(nO)dO (n EN),

-1f
R~Bn = ~ j 1(0) sin(nO)dO (n EN).

-1f
The expression An cos ncp + Bn sin ncp can be transformed as
J1(0) (cos(nO) cos(ncp) + sin(nO)

1T
_1_ sin(ncp)) dO
Rg71" -1T
1 J1f
R or.
n I( 0)( cos( n( 0 - cp) )dO,
-1T
and substituted in (5.15)
U(r,cp) = ~ J1T 1(0) (~+ f (~)n cos(n(O - cp))) dO.

r. -1T 2 n=l Ro
It can be proved that
1 00 ( r ) n R2 _ r2
2+ ~ Ro cos(n(O-cp)) = 2(R5+r 2 -;rRocos(0-cp))'
wherefrom the Poisson integral formula
U(r, cp) =;:

1 J1T r2 m-
1(0) 2(R6 + r2 _ 2r Ro cos(O _ cp )/( O)dO, (5.16)
-1T
for 0 < r < R is obtained.

Exercise 5.10 Determine the solution of the problem
a 2u 1 au 1 au
-ar2 + - -+-
r ar
- =0
r2 a<p
(0 < r < R, -7r < <p < 7r),
U(R, <p) =1- cos 2<p (-7r < <p < 7r).
Answer. U(r,<p) = 1 + (~r cos2<p.
a 2u 1 au 1 au
-+--+--=0
ar2 r ar r2 a<p
(0 < a < r < b, -7r < <p < 7r),
with the following boundary condition
a) U(a,<p) = f(<p), U(b,<p) = g(<p) (-7r < <p < 7r);
b) U(a,<p) = To, U(b,<p) = Tl
Solution. Separating the variables U(r,<p) = R(r)H(<p), the ordinary differential

equations (5.11) and (5.12) with solutions (5.13) and (5.14) are obtained.
So the solution of considered equation is
U(r,<p)
(5.17)
00
+ L (( An rn + Bnr- n) cos n<p + (Cnr n + Dnr- n)) sin n<p.

n=l
a) After applying the boundary conditions we obtain the following system of equa-
tion
Ao + Bo In a =;:
1 J" J(cp)dcp,
-"
Ao + Bolnb =;:1 J" g(cp)dcp,

-1r
JJ(cp) cos(ncp)dcp
1r
An an + Bna-n = ~
-1r
(5.18)
Cna n + Dna-n =~ j J(cp) sin(ncp)dcp

-1r
1
Cnb n + Dnb-n = ;: Jg(cp) sin(ncp)dcp,
1r
-1r
for n E N.
b) The boundary conditions are independent on cp and therefore we use the form
of the solution as
U(r) = Ao + Bolnr.
From boundary conditions we get two equations
1'[ = Ao + Bo In a, Ao + Bolnb,
with the solutions
Ao = 1'1 - (T2 - Tt}-b'

Ina B _ T2 - T}
0- b
In - In -
a a
The solution has the form
In ::.
U(r) = Tl + (T2 - Td--t
In -
a
(-oo<x<oo,o<y<oo),
with the boundary conditions
a)
u(x,y) - t
as x 2 + y2 - t 00
u(x,O) = f(x) (-00 < x < 00);
b)
u(x,y)-to + y2 - t 00
as x 2
u(x,O) = T (-00 < x < 00).
Solution. The solution will be constructed by using the method of Fourier integrals.
a) Separation variables leads to the
x (x) = A( s) cos sx + B( s) sin sx

Y(y) = D(s)e- sy + E(s)e sy ,
where s are the eigenvalues generated by ), = S2. Using the boundary condi-
tions we get
J
00
f(x) = (A(s)cossx+B(s)sinsx) ds,

o
wherefrom we get
11 00
A(s)=; _oof(x)cossxdx
B(s) = -11 00
f(x)sinsxdx.
7r -00
b) u(x,y) = T.
Example 5.13 Solve the following Dirichlet problem on the three-dimensional ball
B(O, 1) (in tht: spheric coordinate system)
1 a ( 2 au ) 1 a. au 1 a 2u
l-..u=2"-a r -a +~()a()(sm()a())+ 2 ' 2()a 2 =0 onB(O,l),
r r r r sm r sm 'P
Solution. The boundary condition is axially symmetric and therefore we suppose

that the solution of the problem is also axially symmetric. We represent the solution
by axially symmetric ball functions
00
u(r,O) = I>nrnPn(cosO),
k=O
where Pn is the Legendre polynomial. Then we obtain by the boundary condition
2:: anPn(x).
00
x2 =
k=O
Since
PO(x) = 1, Pl(x) = x, P2(x) = ~X2 - ~,
we obtain
1 2
(n E N \ {2}), ao = 3' a2 = 3'
The solution is given by
1
u(r,O) = 3(1 - r2) + r2 cos 2 0,
Exercise 5.14 Solve the exterior Dirichlet problem for 1 < r < 2 (in the spheric
coordinate system)
1
t.u = 0, u(1,0) = 2" cos 0, u(2,0) = 1 + cos 20 (0:::; 0 :::; 271").
Hint. Find the solution in the form of the series
2::( anr n + bnr-(n+l))P

00
u(r,O) = n ( cos 0).

k=O
Use that
5 3 3
P3(X) = -x - -x.
2 2
Answer.
4 1 1 8 32 1
+ 93(r
2
u(r,O) = 3(1- ;:)+ 14 (r 2 - r)P1 (cosO) - r3)P2(cosO).
Example 5.15 (Dirichlet problem on cylinder) Solve the Dirichlet problem on

the cylinder [0, i) x [0, 27r) X (0, H) (in the cylindric coordinate system (r, cp, z))
1 a au
t.u = --a (r-
r r a r r
1 a 2u
) + "2-a2
cp
+ -a
z
a2u
2 = on [O,l) x [0,27r) X (0, H),
with the boundary conditions

u(l,cp,z) = 0, u(r,cp, 0) = 0, u(r,cp,H) = m(r,cp),
where m is the given function which is 27r-periodic with respect to cpo

Solution. Applying the Fourier method of separation of variables first in the form
u(r,cp,z) = V(r,z) (cp)
and then
V(r,z) = R(r) Z(z),
where : R -+ R, R: (0,00) -+ R, Z: R -+ R and is a 27r-periodic
function, taking in the first step the constant i and then the constant A, we obtain
the differential equations
cp" + i = 0,
Z" - AZ = 0,
n2
(rR'(r))' + (Ar - -)R(r)
r
= o.
The last equation is the Bessel equation (for x = r.j). for A ~ 0) with solutions
In(r.j).). The boundary condition R(C) = 0 gives In(C.j).) = 0, which implies that
the eigenvalues are given by
A .(Xnj)2 (j EN),
nJ C
where Xnl < X n 2 < ... are the positive zeros of the function I n .
Therefore the solution of the considered Dirichlet problem can be represented by
the series
By the second boundary conditions we have Ckj = -d kj . Therefore taking A kj

2akjCkj and B kj = 2bkjCkj we obtain the (formal) solution
where A kj and B kj are given by ( the last boundary condition and the weighted
orthogonality, by the weight r)
A kj = 1
. x H
JlJ27r rm(r,cp)Jk(_Jr)coskcpdcpdr,
Xk
7r Mk J. smh :::J;;.::..
e 0 0
I'.
where
e
M kj = J rJk(ZXk j ) dr.
o
5.2. THE MAXIMUM PRINCIPLE 163
5.2 The Maximum Principle

5.2.1 Preliminaries
Theorem 5.1 Let L be a general elliptic operator from the preliminary on the
bounded region Q. If u E C2 (Q), L(u) 2: 0 and c(x) < 0 (x E Q), then the
function u has no positive maximum in the region Q.
Theorem 5.2 (Strong maximum principle) For a function u E C2(Q) n C(Q)

(where Q is a bounded region) which satisfies 6u 2: 0 (x E Q) either u = const.
or u(x) < max y E8Q u(y) (x E Q).
Definition 5.3 The function E = E(x) given by
for n 2: 3,
for n =2
27r n / 2
is the fundamental solution of the Laplace equation, where (Tn is the
r(n/2)
surface area of the unit sphere in Rn.
The exterior Dirichlet problem consists in finding a function u E C(O'1)' where

Q1 = Rn \ Q is a region for a given bounded region Q, which is harmonic on Q1
and for n 2: 3 we have lim u(x) = 0 (uniformly), and for n = 2 the function u is
x--++oo
bounded outside of a ball, and it satisfies the boundary condition Ui8Ql = f, for a
given continuous function f.

Example 5.16 Prove that the condition c(x) < 0 (x E Q) In Theorem 5.1 ~s
necessary.
Solution. The function u(x,y) = sinxsiny is a solution of the equation
on the square
s= {(x,y) i 0 ~ x,y ~ 7r},
but it attends its maximum inside of S at the point (~, ~).
Example 5.17 Prove that if for the elliptic operator L from the preliminaries for
u E C 2 (Q)
c = 0, L(u) > 0 (or L(u) < 0),
then the function u has no maximum ( minimum) in Q.
Solution. By Theorem 5.1 ( see Example 5.23) the function u has no positive
maximum in Q. The negative maximum can be reduced on the preceding case taking
the function u + k for enough big k that for u + k be positive maximum and apply
the preceding result on u + k since L( u + k) = L( u).
For the minimum take instead the function u the function -u and use min u =
- max( -u).
Example 5.18 Prove that for a function with the properties
~u = 0 on Q, u E C 2 (Q) n C(Q),
we have
min u(y) ::; u(x) ::; maxu(y) (x E Q).
yE8Q yE8Q
Solution. Since u E C(Q) the function u attends its maximum at some point from
Q. We introduce for c:. > 0 the function u< in the following way
which satisfies
~ux) = ~u(x) + 2nc:. > o.
Therefore by Example 5.17
Letting c:. -+ 0 we obtain

sup u(x) ::; maxu(y).
xEQ yE8Q
We can prove in an analogous way the left part of the desired inequality.
Remark 5.18.1. It is true the strong maximum principle - see Example 5.32.
Example 5.19 There can exist only one solution for the Dirichlet problem for the
Laplace equation.
Solution. Suppose contrary, i.e., that there are two solutions UI and U2. Then the
function u = UI - U2 is the solution of Dirichlet problem for the Laplace equation
with the boundary condition Ul8Q = O. Therefore by Example 5.1S we obtain u = 0,
I.e., Ul = U2.
5.2. THE MAXIMUM PRINCIPLE 165
Example 5.20 Lel {Un}nEN be a sequence of continuous functions on Q and har-

monic on bounded region Q. Prove that if the sequence {u,,} nEN conVC1'!]CS unifor'mly
on 8Q then it convergcs uniformly also on the whole Q.
Solution. Since the sequence {Un}nEN uniformly converges on 8Q it. is also a

Cauchy sequence, Le., for every c > 0 there exists noE N such that for every
m,n> no
(x E 8Q),
I.e.,
(x E 8Q).
Therefore applying Example 5.18 on the function Un - Um we obtain
(x E Q).
Since the space C( Q) is complete with respect to sup-norm, we obtain that the
sequence {Un}nEN uniformly converges on Q.
Exercise 5.21 Prove for the Dirichlet problem for the two dimensional Laplace
equation on the unit circle with the boundary condition u(1, 0) = A(O), -7r :s: 0 :s: 7r
a) the Poisson formula for A E C 2 [-7r, 7r);
b) limr-+l_ou(r,O) = A(O) (0 E [-7r,7r]) for A E C 2 [-7r,7r);
c) the Poisson formula for A is continuous on [-7r,7rJ.

Hints.
a) The Fourier series of the function A uniformly converges to A on [-7r,7r). Use
the Fourier method of separation of variables for the Laplace equation in the
polar coordinate system (See Example 5.9.)
b) Use the fact that the function A is uniformly continuous on the interval [-7r, 7r],
i.e., for every c > 0 there exists 5 > 0 such that 10 - tl < 5 implies IA(O) -
A(t)1 < c. For the case 10 - tl ~ 5 use that
lim 1 - r2 = O.
r-.l-O 1 - 2r cos(O - t) + r2
c) Use the Weierstrass theorem on the uniform approximation of continuous func-
tion with trigonometric polynomials
Tn(O) = ~o IJak cos kO + bk sin kO),

k=l
and consider Dirichlet problems with boundaries Tn (n EN). Then apply

a) and b) on the sequence {Tn(O)}nEN and the corresponding sequence of
solutions {Un }nEN and use Example 5.20.
Example 5.22 Prove that the solution 1l of the Laplacc equation in R" \ {OJ which
is spheric symmetric, i.e., with the property that it has the same value in the points
x which arc on the same distance of the origin is the fundamental solution.
Solution. Put
u = VCr) for r = Ixl > O.
Since we have
au = V'(r)~ = V'(r)x i
aXi aXi r
and so
~2 2 1 2
5!.....!!.... = V"(r)x; + V'(r)- _ V'(r)x i
OXi 2 r2 r r3
we obtain that V satisfies the ordinary differential equation
n-1
2.u = V"(r) + --V'(r)
r
= 0,
whose solution is
VCr) = C1 J -1
dr
r n-
+ C2 ,
where C1 and C2 are arbitrary real constants. Taking C1 = ...!...
ern
and C2 = 0 we obtain
Cl r 2-n
{ for n ~ 3,
for n = 2.
The function u(x) = VCr) for Ixi = r,O < r < 00 is the fundamental solution of the
Laplace equation.
Exercise 5.23 Prove Theorem 5.1 with an additional supposition (which is not nec-
essary!):
Let L be a general elliptic operator from the preliminaries on the bounded region Q.
If u E C(Q), L(u) ~ 0 and
c(x) :::: 0 and (L(u))2 + c2 > 0 (x E Q),

then the function u has no positive maximum in the region Q.
Hints. Suppose the contrary, i.e., that the function has its positive maximum at
some point XO E Q. Therefore
ou( XO)
.
.
- 0 - = 0 (z = 1, ... , n),
X
5.3. THE GREEN FUNCTION 167
for enough small t and fixed P = (PI,"" Pn). Apply Taylor formula on u at the
point .yO + ip and take t -+ 0, which will imply
Using that L(u(xO)) ~ 0 and applying the given conditions we can obtain a contra-
diction with the preceding inequality taking special coordinates Pi = akoi, where aij
are the coefficients in the transformation to a canonical form:
n
qi = LaijPj,
j=1
and k = ko is chosen such that
Example 5.24 Prove that the exterior Dirichlet problem
where Q = B(O, 1) C R3 and Ql = R3 \ B(O, I), has not a unique solution.
Solution. The functions u = 1 and u = I~I are solutions of the given problem.
5.3 The Green Function

5.3.1 Preliminaries
Definition 5.4 A function G = G( x, y) defined on Q x Q for a bounded region
Q C R3 is Green function for the Dirichlet problem if for an arbitrary but fixed
y E Q has the form
1
G(x,y) = 4 I I +g(x,y),
7rX-y
where the function 9 is harmonic with respect to x in Q and continuous on Q and
GlaQ = O.
168 CHAPTER. 5. ELLIPTIC EQUATIONS

Example 5.25 Prove that the functions 9 and G in Definition 5.4 satisfy
1
a) g(x, y) = - 47rX-y
/ / (x E 8Q,y E Q);
b) the function 9 is continuous on Q x Q;
1
c) 0< G(x, y) < 47rX-y
/ / (x E Q,x =I- V)
Solution.
a) Follows by Definition 5.4.
b) Take xO E Q and yO E Q. Applying the Maximum Principle we obtain for

x E Q and y E Q
/g(x,y) - g(XO,yO)/ < /g(x,yO) - g(xO,yO)/ + /g(x,y) - g(x,yO)/

1 1
< /g(x,y )-g(x ,y )1+~~~47rllx'_yOI-lx'-y/l.
1
Since the function 9 = g( x, y) is continuous with respect to x the last inequality

easily implies the desired continuity of 9 at (X O, yO).
c) The function G = G(x,y) with respect to variable x is harmonic on Q \ {V}

and continuous on Q \ {y}. Therefore by the Maximum Principle we have
G(x,y) > 0 (x,y E Q).
Since by a) g(x,y) < 0 (x E 8Q,y E Q) we obtain again by the Maximum

principle g(x,y) < 0 (x E Q,y E Q). Then by Definition 5.4 follows the
desired conclusion.
Remark 5.25.1. It can be proved that for enough smooth boundary 8Q always
. t s t he Green f unctIOn,
eXls . .
eXIsts 8G8n Q d G IS
( x, y) ,.lor every fi xed yEan . symme t flC,
.
i.e., G(x,y) = G(y,x) (x,y E Q).

Example 5.26 Find the Green function on the ball B = B(O, R) = {x Ilxl < R}.
Solution. We correspond to a point y =I- 0 from the ball B = B(O, R) = {x I Ixl <
R} a point y* with the spheric inversion, i.e.,
Figure 5.1
We shall find the Green function in the following form

1 A
I
G(x, y) = 47rX-y I
Using the condition G(x, y)lxE8B = 0 we obtain A = I~I. We obtain from Figure 5.1
that the triangles Oxy' and Oxy are similar. Hence the sides are proportional, i.e.,
R Ix-y'l
Iyl = Ix -YI'
where Ixl = R. Therefore we obtain
1 Rlyl
7rX-y I
G(x'Y)=41
Exercise 5.27 Find the Green functions on the following regions in R3
a) the semi-ball B(O,R) n {(x,y,z) I z > O};

b) the part of the ball in the first octant
B(O,R) n {(x,y,z)1 x,y,z> O}.
Hint. Use the spheric inversion and remember that for the ball B(O, R) by Exercise
5.26
~(_1__ R )
47r Ix - yl Iyllx - Y'l
HZ
for y' = lylZ Y.
Answers. Let
R
a)
Iyllx - Yoosi);
b) R ).
Iyllx - Yi'jsl
Example 5.28 (The Green formula) Prove that ifu E C 1 (Q) is such that u(x) =
o for x f. Q, then we have for x f. oQ
u(x) = 1 j .6.u(y) dy + 1 j( 1 ou(y)
(n - 2)O"n
Q
Ix - yln-Z (n - 2)cr n
DQ
Ix - yln-2 on
o
- u(y)~( I
1
In-z)) dS y (n 2 3), (5.19)
uny x - y
27r n / 2
where O"n = -n- is the surface area of the unit sphere in Rn, and
f(-)
2
u(x) = -~j.6.u(Y)ln _l_dy + ~ j(1n _1_. ou(y)

27r Q Ix-yl 27rDQ Ix-yl on
o 1
- u(y)~(ln -I-I)) dS y (n = 2). (5.20)
Uny x - y
Solution. For an arbitrary but fixed x E Q we take a ball B( x, e) such that
B(x, e) C Q. Applying the classical Green formula on the fundamental solution
E(y - x) and the function u E C 2 (Q) 011 the region Qe = Q \ B(x,e) we obtain
ou(y) oE(y - x)
j .6.u(y)E(x - y) dy = j (E(y - x)~ - u(y) ony ) dSy
Q, DQ
+ j (E(y-x)o~~)-u(y/u~n~x))dSy. (5.21 )
DB(x,e)
The second summand on the right side in (5.21) using the equality
(y E oB(x, e))
oly - xl
can be written for n ~ 3 in the form
(
n- 2
1
)
IJ"nc"
-2
J ou(y)
-",-
un
dS y
1
+ ~---1
IJ"ncn
J u(y) dS y
8B(x,e) 8B(x,e)
= u(x) -h J (u(y) - u(x)) dS

+ IJ"nc y - (
n - 2
~
IJ"nc
n-2 J oua(y) dS y .
n
88(x,e) 8B(x,e)
Putting this in the equality (5.21) and using the facts that the surface area of
. au(y)
aB(x,c) 18 equal to IJ"ncn-1 and u(y) - u(x) :::; kc and I~I :::; S ( k > 0 and
S> 0 are constants) and finally letting c -+ 0 we obtain the desired equality (5.19)
for n ~ 3.
Using again (5.21) we prove analogously the equality (5.20) for n = 2.
Exercise 5.29 Prove that for n =3 the boundary problem for the Poisson equation
flu = -F
for FE C 1 (Q)nC(Q) can be reduced on the boundary problem for a Laplace equation
flv = O.
Hint. Put
u = v + V, where V(x) = ~J IF(Y)I dy.
47r x- Y
Q
Then the function v E C 2 (Q) n C1 (Q) satisfies the Poisson equation fl V = - F and
v satisfies the Laplace equation.
Example 5.30 Supposing that the boundary aQ of the bounded region Q is enough
regular and that the solution u E C 2( Q) n C(Q) of the Dirichlet problem
flu = -F , Ul8Q = f,
where J F2(X)dx < 00 and f E C(aQ), and there exists the derivative with respect
Q
to the normal on the surface oQ, prove that u can be represented by the Green
function G in the following form
u(x) =- an y) f(y) dS + JG(x, y)F(y) dy

J aG(x, y (x E Q).
8Q y Q
Solution. Applying the Green formula from Example 5.28 011 the function u for
n 2: 3 we obtain
1
u(x) = 4-
7r
J(-8--1
8u(y)
n x-y
1
n (-I
- 1 - f(Y)-8 x-y
J x-y
F(y)
-1dS + 7r -I
8
- I dy
1
y
1
-4 (x E Q).
EJQ Y Y Q
(5.22)
By the definition of the Green function the function 9 = g(x, y) for a fixed x E Q
. harmomc
IS
.. Wit I1 respect to y, contmuous
. on Q an d on 8Q t h ere eXists
. 8g(x,y)
8 .
ny
Applying the classical Green formula we obtain
J(an9(x,Y)-f(y)
8u(y) 8g(x,y) J
8n )dS + F(y)g(x,y)dy=O y (XEQ). (5.23)
EJQ Y y Q
Adding the equalities (5.22) and (5.23) and using the definition of the Green function
G and Example 5.25 a) we obtain the desired representation of u by the Green
function.
Example 5.31 Prove that the function u for n = 3, given by
1
u(x) = 47rR J R2Ix _-lxl2
B(O,R) y
f(y)dS 13 y (x E B(O, R,
is the unique solution of the Dirichlet problem for n =3

6..u = 0, ul8B = f,
for f E C(8B), where B is the ball B(O, R) = {x Ilxl < R}.
Solution. The uniqueness follows by Example 5.19.

If u is the solution of the Dirichlet problem on B, then its restriction on B(O, r) for
r < R is the solution of the Dirichlet problem on the ball B(O, r) with the boundary
condition uI8B(O,r). Then this solution can be represented by Example 5.30 ( since
F = 0) in the form
( )- J
u x - 8G(x,y)
8
ny
u ( y )dSy. (5.24)
8B(O,r)
On the other side by Example 5.26 we have
8G(x'Y)1
8n y 8B(O,T) =
8
81yl
(1 rlYI)1
47rlx - yl - 47rlxlyl2 - yr21 /y/=r
5.4. THE HARMONIC FUNCTIONS 173
1 0 ( 1
47rop (IxI 2 +p2_2Ixlpcos,)1/2
(r4 + Ixl 2p2 - ~r2Ixlpcos,)1/2) Ip=r

Ixl 2 - r2 I
47rrlx - yl3 yEBB(O,r)
Putting the obtained equality in (5.24) we obtain
u(x)
1
= -47rr
J r2
I
-lxl 2
1 u(y) dSy
x- y3
(x E B(O,r)).
BB(O,r)
Letting r -+ R we obtain by the continuity of the function u in B(O, R) the desired

conclusion.
Example 5.32 (Strong maximum principle) Prove that for a function

u E C 2(Q) n C(Q) (where Q is a bounded region) which satisfies L).u 2: (x E Q)
either u = const. or u(x) < maxyEBQ u(y) (x E Q).
Solution. Since u E C 2(Q) and L).u(x) 2:

(x E Q) we can apply the Green
formula from Example 5.28. Therefore we have on the ball B(xO,Ro)
We consider now the partition of Q on two sets Ql and Q2 such that
Ql = {x I x E Q,u(x) = yE8Q
maxu(y)},
Q2 = {x I x E Q,u(x) < maxu(y)}.

yE8Q
Both sets Ql and Q2 are open sets. The first by the continuity of the function u
and the second by the inequality from the beginning. Since Q is connected either
Ql or Q2 have to be empty set.
5.4 The Harmonic Functions

Exercise 5.33 Find the harmonic function u on the ring
which satisfies the conditions
v(1,t) = 0, v(2,t) = 2sini (0 < t < 21f),
where u(x,y) = v(r,t).
Answer. v(r, t) = ~ . sinh (In 1') . sin t.
Example 5.34 (Mean value theorem) Prove that if a function u is harmonic

on the ball B = B(O, Ro) and continuous on B, then we have
a) for x tf. oB(O,r), l' < Ro,
1 j 1 ou(y)
u(x) = (n _ 2)O"n (Ix _ yln-2 ~
8E(O,r)
a 1
- u(y)~( I I 2)) dSy (n ~ 3),
uny x - y n-
(5.25 )
21f n / 2
where O"n = -n- is the surface area of the unit sphere of R 3 , and
f( -)
2
u(x) = ~ j (In _1_ . ou(y)

21f
BE(O,r)
Ix - yl an
a
- u(y)~(ln -I-I)) dSy
1
(n = 2). ( 5.26)
Uny x - y
b) the equality
u(O) = In_ljU(y)dS y ; (5.27)
O"nRo BE
c) the equality
Solution.
a) Since u E Cl(B(O,r)) and u is harmonic on B(D,r) we obtain by Example

5.28 the equalities (5.25) and (5.26).
b) For n :::: 3 taking x = 0 we obtain by (5.25)
u(x) = 1
(n - 2)o-n r n- 2
(_1_ J au(y)
an
dS
y
aB(O,r)
- J u(y) ana
aB(O,r)
(-
y y
1
I dS y ).
In-2) (5.28)
On the other hand by the classical Green formula we obtain
J au(y)
an
aE(O,r)
dS y = O.
Putting this in (5.28) we have
u(O) = _1_1-
rn - O-n
J u(y) dS y (5.29)
aE(O,r)
Letting in (5.29) r ---+ Ro we obtain by the continuity of the function u (5.27).

Using (5.26) we can prove in an analogous way (5.27) for n = 2.
c) Apply the integral with respect to r on (5.29) from 0 to Ro.
Example 5.35 Prove that if a function u has the mean value property on simple
connected region Q eRn, i.e., for every ball B(xo, R) = {y I Ixo - yl < R} such
that B(xo, R) C Q we have
u(xo) = 2:R J aB(xQ,R)

u(x) dSx ,
and u is continuous on Q then:
a) the function takes its maximum and minimum on aQ;
b) the function u is harmonic.
Solution. a) We shall prove that either u = const. or
min u(z) < u(x) < maxu(z) (x E Q). (5.30)

zE8Q zEaQ
Let M = max xEQ u( x). We shall prove that if there exists a point from Q where the
right part in the inequality (5.30) is not satisfied, then it have to be u = M = const.
on Q. Suppose that there exists such point. Hence there exists X O E Q such that
u( X O) = M. We connect an arbitrary but fixed point y E Q with X O by a broken
line L which does not intersed itself and which whole lies in Q. Let d > 0 be
the distance between Land 8Q. We cover the line L with a finite number of balls
( . d)
13 x"2 (i=0,1, ... ,N)withcentersx'ELnB(x"2)
..d (i=1,2, ... ,N)and
N d
y E 13(x , 2). We have
u(X O ) = 1
7r d
Ix-xol<d/2
J u(x) dx,
I.e.,
J (u(x) - u(X O )) dx.
Ix- xol<d/2
Since the function under the integral is not positive, we have u(x) = u(XO) on
d .
B(x O, 2) and speCially U(XI) = M. We conclude that M = U(Xl) = u(x 2 ) = ... =
u(y).
We prove in analogous way that either the left part in the inequality (5.30) is true
or u = const. on Q.
Example 5.36 Let a function u be harmonic on a bounded region Q in Rn and

bounded on it, i.e., there exists M > 0 such that lu(x)1 :<:::: M (x E Q). Prove that
for arbitrary x E Q
ID<>u(x)1 :<:::: M(~)I<>ljall<>l, (5.31)
r
where r = d(x,8Q) is the distance of the point x from 8Q.
Solution. The proof goes by induction.

Suppose that lal = 1. Since ~u
uX
(i = 1, ... , n) are harmonic functions on Q we
have by Example 5.34 c) (it holds also for the ball B(xO, Ro) for XO #- 0 since the
property of the harmonicity is translation invariant)for every r' < r
where -Pi is the angle between y - x and the coordinates line Yi. Therefore we have
0"
n
:,n J 8B(x,r')
lu(Y)1 dSy
< M_n_O"nr,n-1 = Mn.

anr'n r'
Letting r' ----) r we obtain (5.31) for lal = l.

Suppose now that the conclusion holds for every a such that /a/ :::; k - 1, k :::: 2.
We shall prove (5.31) for /al = k. By the inductive supposition we have for every
r'
y E B(x, 1),0 < r' < r, and every (3, IP'/ = k - 1,
ID~u(y)/ :::; M( ~ )k-l(k _l)k-l = M',

r-1:
where M' = M(~)k-le-l.

r'
By the first inductive step for the first derivative of D~u we obtain
Letting r' ----) r we obtain (5.31).
Example 5.37 Prove that a harmonic function on a bounded region Q C Rn is a

(real) analytic function.
Solution. Let u be a harmonic function on Q. Let XO be an arbitrary but fixed

point from Q. We denote by
8 8
8 = d(xO,oQ) > 0 and B(xo, 4) = {x I/x - xo/ < 4 j}.
We have u E C(Q5/2)' where Qe = {x I d(x,Q) < c:}. Hence
M = m~x /u(x)/ E R.
XEQ6!2
Since
8
d(B(xO,8/4),8Q5/2):::: 4
and for every multi-index a we have
IDoul :::; M(4;)lol/ aII01,

see Example 5.36. We obtain by the Stirling formula
k(k+ 1 )/2 1
lim ----:--
k ..... oo k!e k Vii'
that there exists c > 0 such that /a/ lal :::; ce1al/a /!.
Taking in the equality
(Xl + ... + Xn )k"" lal!, X a

= L.-t - (k EN).
lol=k a.
Putting Xi = 1 for all i we can obtain the estimation ~

a.
s; nl<>l. Therefore for every
5
X E B(xO, 4) and every a E z+ n we have
(5.32)
Hence the series

(5.33)
converges absolutely on B(.yO, ~). Therefore the sum (5.33) is an analytical func-
4n e
tion on B(xO, ~). We shall prove that the series (5.32) converges to the function
4n e
u on the ball B' = B(x O, _5_), i.e.,
2 8n e
() _ () ~l" D"'u(xO)(
R NX-UX-~~ , x-x
0)",_" D"'u(xO+O(x-xO))(
-~ I
0)'"
x-x,
k=O 1<>I=k a. I<>I=N a.
where 101 < 1, converges to zero as N --7 00 for every x E B'. Since XO + O(x - xO) E
5
B' c B(xO'4) we obtain by (5.32) for every x E B'
L cA{(4n2e)N(~)N
I"'I=N 5 8n e
cA{ N cA{
< --n =-.
(2n)N 2N
Therefore
(x E B').
Since xO was an arbitrary element of Q we conclude that the function u is analytical
on the whole Q.
Example 5.38 Prove that a function u is harmonic on simple connected region

Q C R2 if and only if there exists an analytical function f on Q (taking z = x + iy)
such that u = iRf.
Solution. If f is an analytical function and u = iRf, then 6u = 0 by Example 1.6.

Suppose now that the function u is harmonic on Q. We define a function v by
v(x,y) = J (_au~X'Y)dx+au~~Y))dy,
L((xO,yO),(x,y)) Y
where (XO, yO) is an arbitrary but fixed point from Q, (x, y) E Q and L((xO, yO), (x, y))
is an arbitrary path which connect the points (XO, yO) and (x, y) and completely lies
in Q. The function is independent of the choice of curve L since the region Q is
simple connected and by the classical Green formula. Since U E C 2 (Q) we obtain
that v E C 1 (Q). Since v satisfies the Cauchy-Riemann equations (check it!) we have
that the function f = U + iv is an analytical function with respect to x + iy, where
(x,y) E Q.
Example 5.39 Prove that from every infinite family of harmonic functions on a
bounded region Q, which are uniformly bounded we can select a sequence of harmonic
functions which uniformly converges on a subregion Q' of Q such that Q' c Q.
Solution. Let 1t be an infinite family of harmonic functions on region Q uniformly

bounded by a constant M > O. We choose a sequence of regions Q1, Q2, ... in the
following way: Q1 C Q2 C ... and Qi are relative compact in Q for every i E Nand
U~IQi = Q. Since 1t C C(Q) and the functions from 1t are uniformly bounded on
Ql there exists a constant C > 0 which depends on Ql such that for every U E 1t
we have
I \l ul ~ C
see Example 5.36. Hence 1t is equicontinuous on Ql' Therefore by Arze!a- Ascoli the-
orem we can select from the family 1t a sequence Un, UI2.UI3,' .. which is uniformly
convergent on Ql' Since this sequence satisfies also the conditions of Arzel-Acsoli
theorem on Q 2 there exists its subsequence U2l, U22, U23, ... which is uniformly con-
vergent on Q2' Continuing this procedure we obtain at the i-th step a subsequence
Uil, Ui2, Ui3, ... which is uniformly convergent on Qi' Then the diagonal sequence
Un, U22, U33, ... is the desired sequence.
Example 5.40 Let {uk} kEN be a sequence of harmonic functions on the region Q
which is uniformly convergent on every region Q' C Q such that Q' C Q. Prove
that the limit function u is also harmonic on Q and that for every multi-index
a = (a1,"" an) the sequence {D"'udkEN uniformly converges to D"'u on every
region Q' C Q such that Q' C Q.
Solution. First we have u E C(Q'). Let Q" be a region such that Q' cc Q" cc Q.
Since every function Urn is bounded on Q" for every multi-index a there exists a
constant C> 0 which depends only from Q', Q" and lal such that
(s,m = 1,2, ... ).

Since
lim
mls-+(X)
I\um - usllc(QII) = 0
we obtaill that all sequences {D"Um}mEN are Cauchy sequences in the space C(Q').
Therefore U E C=(Q') and for every a we have
J~oo IID"u m - D"Ullc(Q') = 0.

Taking m --t (Xl in .0.u m = (x E Q') we obtain .0.11 = (x E Q').
Example 5.41 (Harnack first theorem) Let {UdkEN be a sequence of harmonic
functions on the rcgion Q which belongs to C (Q). Prove that if the sequence { Uk laQ h E
N is uniformly convergent on the border 8Q, then the sequence uniformly converges
on Q to a harmonic function.
Solution. The first part follows from Example 5.20.

The function u is harmonic by Example 5.40.
Exercise 5.42 (Harnack second theorem) Let {ud kEN be a sequence of har-
monic functions on the region Q C R2 which belongs to C(Q). Prove that if the
sequence {uklaQ hEN satisfies the inequality
(Xi E Q,i EN),

then the convergence of the sequence {Uk} kEN in a point of the region Q implies its
uniform convergence on every compact set in Q.
Hints. Prove that u is a harmonic and non-negative function and using the Poisson
formula prove the inequalities on the disc of radius R
R~r R+r
- R u(O,O) s:; u(r,O) s:; - R u(O,O) (r E R).
+r -r
Exercise 5.43 Prove that a function u E C 2 (Q) for a bounded region Q C Rn
which is a solution of the Helmholtz equation
.0.u + Au = 0,
where A is a constant, is an analytical function on Q, i.e., all eigenfunctions for a
Laplace operator are analytical functions.
Example 5.44 We correspond to a point x f. 0 from the ball B = B(O, R) =

R2
{y I Iyl < R} a point x' with the spheric inversion, i.e., x = xF1 2 . Prove that
if a function u is harmonic on Rn \ B, then the function u' defined by (Kelvin
transformation)
(5.34)
is harmonic on B \ {OJ.
Solution. Suppose n > 2 ( for n = 2 we proceed analogously). Let n; be a region

strictly inside of B \ {OJ and Bl is its original with respect to the illversion. Then
BI is a bounded region strictly inside of Rn \ B. The function u is harmonic and
belongs to C 2(Q I) an d therefore
where
u(Z)
v(z) = - (n _ 2)O"n 18B1
and these are continuous functions on aBI. By (5.34) we have for all x E B;
u *( x *) -- J(Ix*ln-21]{Rfj(z)
2
x' _ Zln-2
(5.35)
8B1 Ix'12
a R
+v(z)an (I *In-2IR2x'_ In_2))dSz.
z X Ix'I' z
Since the function
Ix*1 2 -nR 2 x* 2-n
- R - I Ix*12 -z 1
is harmonic we have that
is harmonic function with respect to x* when ~2*~~ f. z. Therefore the function under
integral in (5.35) is continuous with respect to x* and z as well as its derivatives
with respect to x* and it is harmonic with respect to x* for z E aB and x* E B\ {OJ.
Exercise 5.45 Prove that the exterior Dirichlet problem has maximally one solu-
tion.
Hint. For two solutions

UI and U2 consider the function U = Ul - U2 which satisfies
the homogeneous Dirichlet problem. Apply the inversion transformation and use
Example 5.44.
Exercise 5.46 The Poisson equation .0.u = -F has has not more than one regular
solution in the infinite for n ;::: 3. For n = 2 regular solutions in the infinite differ
up to a constant.
Hint. For n ;::: 3 use the maximum principle on the ball B(O, R), and for n = 2 the
inversion and the maximum principle.
Exercise 5.47 Prove that the solution of the Dirichlet problem in R2, R3
~u = -F (Ixl < R) and UlaB(o,R) =f

is given by
R2
where y' = Y1Yf2 is obtained by inversion.
5.5 Gravitational Potential

Let the function
c
u(x, y, z) = -, (5.36)
r
defines the gravitational potential. In equation (5.36) c = ,Mm,and , is gravita-
tional constant; M is the mass of a particle at a fixed point (~,.", e), m is the mass
of a particle at a fixed point, and r is the distance between two points
r = V(x -~)2 + (y -."F + (z - (p.

If PI is particle of mass M and P2 is particle of mass m, then PI attract P2 with a
gravitational force that is the gradient of the function -=.
r
Let u(x,y,z) be a potential u of continuous body at a point (x,y,z) outside the
body. Then the potential u is defined by
where p is the density of a mass at (~,."e) and k is a positive constant. The function
u satisfies the Laplace equation
In Laplace equation, u may represent electric or magnetic potential functions at

points not filled with electric charges or magnetic poles. Laplace equation is associ-
ated with incompressible fluid flow problems, too.
Chapter 6
Parabolic Equations
6.1 Cauchy Problem

6.1.1 Preliminaries
The problem for heat equation given as
au .
at = a2~u+F(x,t) (x E R n t > 0), (6.1)
with condition
u(x,O) = I(x) (x ERn), (6.2)
where a > 0, 1 is bounded and 1 E C(Rn), F E C 2(Rn x [0,00)), and all its
second order partial derivatives are bounded on every set of the form Rn X [0, T] is
called the Cauchy problem.
If F = 0, then we are working with homogeneous heat equation otherwise this is
nonhomogeneous heat equation. The classical solution of problem (6.1), (6.2) is the
function u = u(x, t) E C2(Rn x [0, T]) U C(Rn x [0, T]) given by
u(x, t) = (47ra 2t)-n/2 j exp (_Is4a- x12)

2t
I(s) ds
Rn
(6.3)
+
j j (47raF(e,r)
t
(t_r))n/2 exp
2
(
o Rn
The Maximum Principle

For an arbitrary but fixed T > 0 we introduce the set Q by
Q = {(x,t)1 x E 0,0 < t < T},
183

184 CHAPTER 6. PARABOLIC EQUATIONS
where 0 is an open bounded set of R n. The boundary oQ we can divide on two

disjoint parts "lower" boundary o'Q and "upper" boundary o"Q given by
o'Q = {{x,l)! either x E 00,0 'S t 'S T or x E O,l = O},

a"Q = {{x,l)! x E 0, l = T}.
Figure 6.1 For n = 1 and 0 an open interval

Theorem 6.1 (Maximum Principle) Let
u E C(Q),
ou 02U
at' OX;OXj E C(Q)
and
au -
ot
- t:.u
-< O.
Then
ma~ u(x, t) = max u(x, t).
(x,t)EQ (x,t)E&'Q
6.1.2 Examples and Exercise

Example 6.1 Suppose that the function f : R ---r R is continuous and bounded,
Z. e.,
!f(x)! 'S M (-00 < x < 00).
Prove that the function
u(x, t) =
1
c;
lco fey) exp ( - (x -
2
y)2) dy, (6.4)
2ay7rt 4a t
-co
is bounded and salisfy the equation in the problem

6.1. CAUCHY PROBLEM 185
u(x,O) = f(x) (x E R).
Solution. The integral (6.4) converges absolutely for arbitrary (x, t) E R X R+,
because
lu(x, t)1 S; M,G" Joo exp (_ (x -
2ay 7rt -00 4ta
;)2) dy,
where M = sup If(x)l. The change of variables, z = x - ~, gives
xER 2ayt
lu(x, t)1 S;
M
..Ji J00
-00
2
e- Z dz = M.
Differentiating the integral (6.4) by x and t under the integral the obtained integrals
converge absolutely. In an arbitrary compact neighbourhood of a point (x, t) E
R X R+ the integral (6.4) with its partial derivatives converge uniformly. The
partial derivatives are
au
at
1;00 fey) (X_y)2

+ Vi -00
((X_ y )2))
4a2t2 exp - 4ta2 dy
1 r.;.
2 ay7r
1 ~ Joo f(y)exp
4a 2 yt 5 -00
( (X-
4 2
y
ta
?) ((x-y)2-2ta 2)dy;
au 1
2aV7rt
(;oo
leo fey) exp ((X_y)2)
- 4ta2
-(x-y) )
. 2a2t dy ;
ax
a2u _
ax 2 -
_1_ (Joo
loo
2aV7rt
fey) exp (_ (x - y)2) .
2a 2t 4ta 2
(-1 + (x - y)2) dY)
2a 2t '

au 2a 2U
-=a- (-00 < x < 00, t > 0).
at ax 2

au 2 a2u
Dt := a ax2 (x E R, t > 0),
u(x,O):=
Solution. In this case formula (6.4) reduces to

r T,
0,
for
for
for
c
x < c,
< x < d,
x> d.
u(x, t) =
T
r=;
Jd exp ((x_
-
y )2)
dy. 2
2av-rrt c
4a t
After changing of variables, ~ = x ~, we get

av2t
u(x, t) =T vk J (x-d)/a-/2i
(x-c)/a-/2i
e-e/ 2 d(
Remark 6.2.1 Since the normal distribution function is defined by
J e-
x
(x):= t2 / 2dt,
-00
we have shown that the solution of this problem can be written as
Exercise 6.3 Determine the solution of the problem

au 2a 2U
at = a ax2 (-00 < x < 00, t > 0),
u(x,0)=T1 (x < 0),
u(x,O):= T2 (x> 0).

Answer.
u(x,t) = Tl (1 - C~)) + T2 C~) .

It holds lim (x)
x~o
:= (O) := ~.
2
Exercise 6.4 Show that the solution of Cauchy problem
u(x,O)=f(x) (xER),
where the function f E C 2 (R), satisfies the following
a) u(O, t) = 0, if the function f is odd;
ou(O, t)
b) ax = 0, if the function f is even;
Hint. It can be done similarly as in Example 4.3.
Example 6.5 Solve the following problem
aU 202U
a- 2 (x > 0, t > 0),
-
at
=
ox (6.5)
u(O,t) =0 (t>O), u(x,O)=f(x) (x > 0).
Solution. Let us define the function W as
f(x), for x> 0,

W(x) = { _ f( -x), for x< 0,
and consider the problem
au 202U
- =a - 2 (x E R, t > 0),
at ox
U(x,O) = w(x) (x E R).
U(x,t) = 1,-," J=W(y)exp ( (x - y)2) dy

2ay1rt -00 4ta 2 '
wherefrom it follows that the solution of the considered problem (6.5) is
u(x, t)
1
= 2a,jirt
Joo (
f(y) exp -
((X_y)2)
4a t
2 - exp -
((X+y)2))
4a t dy.
2
o
Exercise 6.6 Solve the following problem
au 2 fJ211,
~ = a - (x > 0, t > 0),
at ax 2
(6.6)
au(O, t) =0 (t > 0), u(x,O) = f(x) (x> 0).
ax
Answer.
Exercise 6.7 Introducing the function v such that u(x, t) = ehtv(x, t), determine
the solution of the equation
au 2a 2U
-- = a - - hu (x> 0, t > 0),
at ax 2
with the conditions
a) u(O,t) = 0 (t> 0), u(x, 0) = f(x) (x> 0);
b) au~~, t) = 0 (t > 0), u(x, 0) = f(x) x> 0;
where h is a constant and the function f is continuous and bounded on the interval
(0,00).
Answer.
a) u(x, t) = e- ht
r:::i. Joo fey) ( exp ((x
- -2 y)2) - exp ((x +2y)2)) dy.
2ay 7rt 4a t 4a t
o
b) u(x, t) = e- ht r:::i.
JOO (
fey) exp -
((x y)2) + exp ((x
2 -
+ y)2)) dy. 2
2ay7rt 4a t 4a t
o
Exercise 6.8 Let Ui be the solution of the Cauchy problem on the set {(x, t)lx E
Rn, t> O},
au 2
at =a ~u,
u(x,O) = fi(x;), for fi : R ---) R, i = 1,2, ... , n.

Prove that then the function 11, given by
u(x, t) = Ul(X, t) ... un(x, t)

represent the solution of the Cauchy problem on the set {(x, t) Ix E Rn, t > O},
au
7ft = a
2
llu,
Exercise 6.9 Determine the solution of the Cauchy pmblern on the set {(.T, i)lx E
R" , t > O} ,
au = a llu,
at 2
u(x,O) = f(x),
for
a) f(x) = COS(Xl + X2 + ... + x n )
Answers.
b) u(x, t) = (1 + 4 t t n/2 exp (-lxI2) .

1 + 4t
J
t
Exercise 6.10 Show ihai the function u(x, t) = v(x, i, r)dr, where
o
v(x, i, r) = J
1 Joo g(x, y) exp ( - (X_y)2)
( ) dy,
27r(i-r)_00 4t-r
and the fu nction g( x, y), (x, y) E R 2 , is continuous and bounded on R2 and satisfies
the nonhomogeneous equation
au
T = !l2
a u + g(x, t).
2
ut ux
Exercise 6.11 Suppose that the functions f E C (R n) and F E C 2 (R n X {t It 2: O} )
are bounded and the derivatives of F arc bounded on Rn X {tiO ::; t ::; T}). Prove
that the function
u( X, t)
1
= (2a#)" J f(y) exp
(- Ix - Y12)
4a 2t dy
Rn
(6.7)
+ JooJ F(y,s) n exp (-l x -
YI2 )dd
y s,
-ooRn (2aJ7r(t-s)) 4a 2 (t-s)
represent the solution of the problem
au
at = a .6.u + F
2 (
x E Rn t > ,
)
u(x,O) = f(x) (x ERn).

Exercise 6.12 Determine the solution of the following Cauchy problems
au cPu 2
a) at = ax2 + 3t (x E R, t > 0), u(x,O) = sin x (x E R),
au
b) at =.6. u + cos t (x E R2,t > 0), v(x,O) = X1X2 exp (-xi - xD (x E R2)
au t
c) at = 3.6.u +e
Answers.
a) u(x, t) = t3 + e- t sin x
b) . X1X2 (xi + :l:~)

u(x,t)=smt+( )3exp
1 + 4t 1 + 4t
c)
Exercise 6.13 Determine the solution of the following problems
au a 2u
a) (x E R, t > 0), u(x,O) = xe- x2
(x E R)
at ax 2
b) au = .6.u (x E R2, t > 0), u(x,O) = COSX1X2 (x E R2),

at
au
c) - = a.6.u (x E R 3 , t> 0), u(x,O) = COS(X1X2)X3 (x E R 3 ).
at
Answers.
_X2 )
a) u(x, t) = x(l + 4t)-3/2 exp ( - -
I +4t
b) u ( Xll X2, t ) -_ (
1 +t )
2 -1/2 x1 x 2
cos - - 2 exp
(-t(Xi
(
+ 2)x~)) ,
l+t 21+t
c) u(xllx2,x3,t)- smx3(1+4t)
_ . 2 -1/2 x2 x 2
c o s - -2 exp
1 + 4t
(-t-t 2(21 ++ Xt 2)'
Xl 2) 2
Exercise 6.14 Determine the solution of the Dirichlet problem for parabolic equa-
lion
EPu a2 u
+ x 2 _a u au au
2
t2 - - 2xt-- - x - - tx- - u = 0,
ax 2 axat at 2 ax at
u(x\-I) = u(x,l) = x 2, u(-I,t) = u(1,t) = t 2.

Answer. The given equation in polar coordinates (r, cp) has the form
a2 lj
acp2 - lj = 0,
and the solution can be written as
u(x,t) = 0, for x 2 + t2:::; 1,
ex p (arCSin ~) +exp (arccos ~))

u(x,t)=(x 2 +t 2 -1) ( yx-+t- vx-+t- ,
exp (arcSin ~)
x +
+ exp (arccos ~)
2 t2 x + 2 t2
Example 6.15 Let
u E C(Q),
Prove that u is determined uniquely in Q by the value of
au
- - ~u on Q
at
and of u on the boundary a'Q (see Preliminaries).
Solution. It is sufficient to consider the case

au
- -~u
at
= on Q
and ulalQ = 0. Then applying the Maximum Principle Theorem 6.1 on u and -u we
obtain
ma~u(x,t) = ma~(-u(x,t)) = 0.
(x,t)EQ (x,t)EQ
Therefore u = on Q.
Exercise 6.16 (\Vidder) Prove that if
u E C({(.r,t)1 x E R,O ~ t ~ T}),
where V = {(x, t)1 x E R,O < t < T} and
u(x,O) = f(x) (x E R), u(x,i) 2: 0 ((x,t) E V),

then u = u(x, t) is deiermined uniquely on V and it is a real analytic function given
by
u(x,t) = ( 1) /2 exp (ls-x

41ft
- -n
J
I2 )
f(s)ds. 4t
Rn
Hints. We define for a > 1 the functions za and va in the following way
forlxl~a-l,
,"(x)~ {~-IXI for a-I < Ix I < a,
for Ixl 2: a,
and
Using the Maximum Principle prove that for c: > 0 we have
va(x, i) ~ e + u(x, i)
for Ixl < r, 0 ~ t < T, where

2Ma
r>a+ ~
ey 21fc
for M = maxf(x). Letting r - t 00 and e -t 0 prove that
Ixl:<;a
o ~ v(x,t) ~ u(x,i) ((x,t) E V),
where
v(x,t) = lim va(x,t) = (
a ..... 00 41ft
\ n /2 J exp (_IS - x12) f(s)ds.
4t
Rn
Prove the regularity of v = v(x, i) using va and that
v(x,O) = f(x).
6.2. MIXED TYPE PROBLEM 193
Introduce a function h by h =u- v, and prove that
8h _ 8 2h _ 0 V
8t 8x2 - on,
h(x,O) = f(x), (x E R), h(x, t) ;::: 0 ((x, t) E V).

Then prove that h = 0, using the function H given by
Jh(x,s)ds,
t
H(x,t) =

which has the same properties as h and which satisfies additionally the inequality
H(x,s):::;
1r(t-S)1/2
2
(X2) H(O,t).
exp -
x t- s
Prove by this inequality that H = o.
6.2 Mixed Type Problem

6.2.1 Preliminaries
The mixed (initial and boundary) value problem (for a > 0 and l > 0) is given by
the equation
8u 282 u
8t = a 8x 2 ' (0 < x < l, t > 0),
u(O, t) = gl(t), u(l, t) = g2(t), (t > 0),

and initial condition
u(x,O) = f(x), (0 < x < l),
where the function f is defined on (0, l), g1, g2 are functions defined for t > O. This
is one dimensional heat equation (with two variables).
Remark. The solution of this equation u(x, t) means the temperature distribution
on a finite thin rod or wire of length l. Usually it is supposed that this rod made of
uniform material, has a uniform cross section and it coincides with a part of x-axis.
The homogeneous boundary (gl (x) = g2( x) = 0) conditions express the situation
that the both ends of the rod are maintained all the time at the temperature OoC.
The initial conditions express that the initial temperature of the rod given by f
depends on x namely on the distance from one end of the rod (x = 0).
The solut.ion of this problem is given by applying Fourier method of separation

of variables. It has the following form
u(x, t) = ,?; Cn exp ((7rna)2)

00
-t -/!- 7rnx
sin(-/!-),
where
Cn =~ ) f(x) sin C~X) dx (n EN).

o

(PU _ ~ au =0 (0 < x < /!, t > 0),

ax 2 a2 at
where a >0 is a constant, with
(i) boundm'Y conditions u(O, t) = 0, u( /!, t) =0 t > 0,
(ii) initial conditions u(x,O) = f(x) (0 < x < /!).
Solution. Let us suppose that the solution of given equation has the form
u(x,y) = X(x) T(t),
then we obtain the problems
X"(x) + AX(X) = 0, X(O) = 0, X(/!) = 0, (6.8)

and the equation
T'(t) +.,\ . a 2 T(t) = O. (6.9)

The eigenvalues of the problem (6.8) is the considered problem are
(n E N) (6.10)
Xn(x) = sin -/!-

nJrx (n EN),
where we have taken C z = 1.

For A given by (6.10) the solution of equation (6.9) has the form
(n EN),
where en are arbitrary constants. The solution of the considered problem is
This solution has to satisfy the initial condition and therefore for t = 0, we obtain
the Fourier series
n7rX
= f(x) = L
00
u(x,O) en sin -l-
n=l
The constants en can be found as the coefficients of Fourier series
en = p.2 Jf(x)sm-l-dx,
l
. n7rX
(n EN).
o
Example 6.18 Solve the following problem
ou 202U
at = a ox2 (0 < x < 7r, t > 0),
u(O, t) = 0, u(7r,t) =0 (t > 0),
u(x,O) = To (0 < x < 7r),

where To is a constant.
Solution. By the method of separation of variables we obtain two differential

equations
X" + AX = 0, X(O) = 0, X(7r) = 0, (6.11)
(6.12)
Eigenvalues for (6.11) are An = n 2 and eigenfunctions are Xn(x) = sinx. Therefore
solutions of the equation (6.12) are of the form
(n EN).
Then the solution of the problem is of the form

L: An sin(nx)exp( _a
00
u.(x, t) = 2 n 2 t).
n=l
By the initial condition

00
u(x,O) = To = L: Ansin(nx),
n=l
therefore one can obtain coefficients An as coefficients of a Fourier's series
An = -7f2 In. 2To n

Tosm(nx) = -(1- (-1) )
n7f
o
4To
{
n = 1,3,5, ...
n7f
0 n = 2,4,6, ... ,
and the solution is
u ( x, t ) = -4To ~
L.J
sin((2n - l)x exp ( -a 2(2 n - 1)2) t ) .
7f n=l 2n - 1
Exercise 6.19 Determine the solution of the following problem
(0 < x < , t > 0),
u(O, t) = u(, t) = 0 (t > 0), u(x,O) = f(x) (0 < x < f),

where
a) f(x) = x;

O<x<-
- 2
b) f(x) = { x,

- x, 2" ~ x < .
Answer.
2 (_l)n . n7fX 2 2 2
a) u(x, t) = - L: -
7r n=l
00
n
- . ( a 7f n )
sm-"-' exp ---2-t
~
t.
b) u ( x, t ) -_4i~
2 L..J (
(-I)n . (2n+l)7rX
n exp -
(a 7r 2(2n+l)2 t )
2
n .
7r n=O 2n + 1)2 . Sill
{. {.
aaxu = at
2 au + 4u - 20 (0 < x < 7r, t > 0),
2
u(O,t) = 5, u(7r,t) = 5 (t > 0)
U(X,0)=5+2X (O<x<7r).
Hint. Introduce a new function z as
u(x, t) = 5 + z(x, t)e- 4t .

Then the considered problem becomes the problem appearing in Exercise 6.19 a)
for f(x) = 2x. The solution is
(-It+!
u(x, t) = 5 + 4 L + 4)t).
00
sin(nx) exp( _(n 2
n=I n
Example 6.21 Solve the problem
(0 < x < 7r, t > 0),
u(O,t) = TI , (t > 0),
u(x,O) = x(7r - x) (0 < x < 7r).
Solution. Let us introduce the functions S and v such that
u(x, t) = Sex) + vex, t). (6.13)

So, the considered problem can be written as
S "() a2 v(x, t) _ -2 av(x, t)

(O<x<7r, t>O),
x + ax 2 - a at
S(O) + v(O, t) = TIl S(7r) + v(7r,t) = T2 (t > 0),
Sex) + v(x,O) = x(7r - x) (0 < x < 7r).

Let us first solve the steady-state problem
198 CHAPTER. 6. PARABOLIC EQUATIONS
S"(x) = 0,
The solution of the previous problem is

x
S(x) = Tl + (T2 - Td-
11"
Now, we still have to solve the well known problem with homogeneous boundary
conditions
(Pv(x,t) _28v(x,t)
=a
--=-'-:--'-
2 (0 < x < 11", t > 0),
8x 8t
v(O, t) = 0, v(11",t) =0 (t > 0),
x
v(x,O) = x(11" - x) - S(x) = x(11" - x) - Tl - (T2 - Td-.
11"
LC
00
u(x, t) = n exp( -t(na)2) sin nx,

n=l
where
.".
en = ~J(x(11"-X)-Tl-(T2-Td~)sinnxdx
11" 11"
o
(n EN).
Therefore the solution is
E
+ OO( (2n -1)311"
8 2
+ ;((-lt T2 - )
T1 ) exp(-t(na?)sin2nx.
Physically, the boundary conditions express that the temperature of the rod at
the end x = 0 is maintained at value Tl and the temperature at the end x = 11" is
maintained at value T 2
Exercise 6.22 Solve the problem

iPu -2 i}u
-=a - (0 < X < 1, t > 0),
Ox 2 01,
u(O, t) =1 u(l,t) = 2 (t> 0)
u(x,O) = 0 (O<x<l).
Answer.
u(x, t) = 1+ x + ~
rr
f
n=l
(~(( -It - 1) exp( -t(na71f) sin mfX.
n
Example 6.23 Solve the following nonhomogeneous equation
ou
= at + u,
02U
ox 2 (0 < x < 1, t > 0),
with homogeneous boundary
u(O, t) = 0, u(1,t) = 0 (t > 0),

and initial conditions
u(x,O) = To (0 < x < rr),
where 1'0 is a constant.
Solution. Let us introduce a new function v such that u(x, t) = z(x, t)e- t . Then
the above problem is equivalent with
02 z Oz
(0 < x < 1, t > 0),
Ox 2 ot
z(O, t) = 0, z(l,y)=O (t > 0),
z(x,O) = To (0 < x < rr),

which solution is of the form
=:L An sin(nrrx) exp(-n rr t),

00
z(x,t) 2 2
n=l
where
4To, n = 1,3,5, ...
A n -- { n?r
0, n = 2,4,6 .... ,
The solution is
u(x,t) = -4To ~ --sm((2n

L
1 . -l)?rX)exp(-(l + (2n -1 )2 rr 2) t.)
rr n=l 2n - 1
200 CHAPTEH 6. PAHADOLIC EQUATIONS
Example 6.24 Detenninc the solution oj the Jollowing problem
cPu -2 iJu
iJx2 =a iJt (O<x<I!, t>O)
iJu(O, t) =0 iJu(,t)
--=0 (t > 0),
iJx ' iJx
u(x,O) = x (0 < x < e).

Solution. Using the method of separation variables we obtain the following problem
X" + AX = 0, X'(O) = 0, X'(f!) = 0,

whose eigenfunctions are
n7rx
Xn(X) = cos -f!-'
Therefore the solution in of this problem can be written in the form
u(x, i) = 2Ao + LAn exp (7rna)

00
-i2 cos (7rnx)
-f!- (0 < X < , 0 < i < 00),
n=l
where the coefficients are obtained from the initial condition as the coefficients of
corresponding Fourier series and have the forms
Ao =f!
An = e2 Jx cos (7rnx)
l 2
-f!- dx = 7r n 2 2 (cosn7r -1), II, = 1, ...
o
or
4f!
II, = 1,3,5, ...
0, II, = 2,4, ....

The solution of the considered problem has the form
_ ~
u ( x, t ) - 2 +~ ~
2L (
1
)2 exp
(_ (7r(2n -
t p
l)a) 2) cos (7r(2n - 1)x) p ,
7r n=l 271, - 1 t- ~
for 0 < x < f!, 0 < t < 00.

6.2. lHIXED TYPE PROBLEM 201
Exercise 6.25 Determine the solution of the following problems
au a2 u
a) -;:) = - 2 (0 < < 1, t > 0),
vt ax X
au(O, t)
ax = 0, u(I, t) = (t > 0),
u(x,O)=x2-I (O<x<l).
au a2 u
b) at = ax 2 - u (0 < x < e, t > 0),
au(O, t)
ax = 0, u(i, t) =0 (t > 0),
u(x,O)=l (O<x<i).
Answers.
32~
a ) 3 L..J (
(_l)n
exp
(((2n+l)7r)2)
- t
2n+l
--7rX.
2n + 1
)3 COS
7r n=O 2 2
b) -4~ 1 ((7r2(2n+I)2)).
L... - - exp - + 1 t sm (2n-I) 7rX.
7r n=O 2n + 1 i l
Example 6.26 Solve the problem
(0 < x < i, t > 0)
au(O,t) = 0, u(i, t) = uo (t> 0),

ax
u(x,o) = f(x), lim f( x) = uo (0 < x < l),
x-l-
where uo is a given constant.
Solution. Let us write the solution in the form
u(x, t) = uo + vex, t),

such that v is the solution of the problem
av(O, t) = 0
ax ' vel, t) = 0,
V(J:,O) = f(x) - Uo.
The solution of the previous problem is
2n + 1
v(x, t) ~ An cos ( --uxr.) exp
co
(-t ((2n +21)r.na )2) '
where
e
2J (2n+l) 4(-1)"uo
An=C f(x) cos 2C- x r. dX-r.(2n+l) (n=O,l, ... ).
o
202u ou
a-=- (0 < x < , t > 0),
ox 2 at
ou(O, t)
ox = qo, u(, t) = 110, (t > 0),
u(x,O) = f(x) (0 < x < C),

where Uo and qo are arbitrary constants.
Answer. The solution has the form
u(x, t) = qox
2n + 1
+ Uo + ~ An sin (-U~xr.) exp (-t
00
((2n +21 )r.a ) 2) '

where
A - ~
n -
Je f( ) . (2n + 1 ) d _ 4 (2n + l)r.uo + Cqo
x sm \ 2 xr. x r.2 (2n + 1)2 , n = 0,1, ....
o
20 2 11 OU
a ox2 = at (0 < x < , t > 0),
ou(O, t) =
ox
'
o11(f, t)
- - =qo
ox
()
t >
u(x,O)=o (O<x<),
where qo is an arbitrary constant.
Answer. The solution has the form
U(X, t) = qo ( -
a2t + 3X2 .- 2 + -2 L 00 (_It+l
cos -
(n7rX) exp( -t (l!7ra)
-
2) .
e 6e 7r 2 n=O n2 e e
Example 6.29 Solve the following p1'Obiem
2 fPu AU
a ox2 = at (0 < x < f!, t > 0),
ou(O, t)
ox = 0, (t > 0),
u(x,O) = f(x) (0 < x < e).

The values of of KI and K2 are given.
Solution. The boundary conditions for this heat equation express that for time
t = 0, at the end of the rod x = 0 there is no heat flow and the heat is exchanged
at the other end with an environment at temperature OOC.
The method of separation variables bring us to the problem
X"(x) + AX(X) = 0 (0 < x < f!), X'(O) = 0,

and to equation T' - Aa 2 T = 0 (t > 0).
In order to obtain nontrivial solution of this equation the eigenvalues must be
positive i.e., A = k 2 Using the boundary conditions we obtain
K2
tankJi = --,
KIA
wherefrom we obtain the values for the eigenfunctions An. The corresponding eigen-
functions (it may differ only in a constant and we take it to be 1) are of the form
(n E N)
(they may differ only in a constant and we put it to be 1).
For these eigenvalues the solution of differential equation
T' +a 2 '\T = 0 (t > 0),

IS
T(t) = Cn exp( -aAnt).

The final solution of the considered problem is given by
L
00
u(x,t) = Cn COS(AnX) exp(-aAnt ),

n=l
where the Cocfl1ciellts are of the form
Jf(x) cos(Anx)dx.
l
Cn =
o
Exercise 6.30 Solve the followng problem
2E)2u au
(0 < x < 1, t > 0),
ax
a-=-
2 at
au(O, t) _ T au~~, t) + u(1, t) = T2 (t> 0),
ax -"
u(x,O) = TJ (0 < X < 1).
Answer. The solution can be written as
( x, t ) = TJ + (T2 - 1 ~ cosknsinknx
T 1 ) ( -x + 2 L.. k (
(2 2))
2 k ) exp -a k n
1 + cos
U ,
2 n=l n n
where kn is the solution of equation
(n EN).
Exercise 6.31 Determine the solution of cquaton
(0 < x < , t > 0),
GU(O, t)
ax - hu(O, t) = u(, t) = 0 (t > 0),
u(X, 0) = Uo (0 < x < f).
where h > 0 is a given constant.
Answer. The function v(x, t), introducing as u(x, t) = e-Btv(x, t), satisfies the
equation from previous example with a = 1. The solution has the form
where the eigenvalues An are the positive solutions of the equation
h tan A = -A.
Example 6.32 Solve thf- jollowing rni.rcri problem JOT the heat equation j01' a > 0
and > 0
ou 2(]2u
- = a -2 (0 < .T < , t > 0), (6.14 )
at (J.T
u(:r,O)=.f(x) (O<x<), (6.15)
ou(O, t)
ox -hJ(u(O,i)-UJ) =0 (t > 0), (6.16)
ou( , t)
~ - h2(U(, t) - U2) =0 (t > 0), (6.17)
where hi >0 and Ui > 0 are given constants.
Solution. We try to find the solution of the given mixed problem in the following
form
u(x, t) = v(x) + w(x, t),
where v is a solution of (6.14) which satisfies the boundary conditions (6.16) and
(6.17), i.e., v is the solution of the equation v" (x) = 0 which is of the form
where
The function w satisfies the equation (6.14) and the initial condition (6.15), i.e.,
w(x,O) = u(x,O) - v(x) = f(x) - v(x) (6.18)
and the homogeneous boundary conditions
ow~~, t) _ hJ w(O, t) =0 (t> 0), (6.19)
ow(,t) I (P)_ (t > 0). (6.20)

oX - L2 W {.,t - 0
Applying the Fourier method of separation of variables on the problem (6.14), (6.18),
(6.19), (6.20) we obtain
Wn ( x, t ) T
= Cn e- a2J1.2t/t> (/l-n
n cos T
/l-n X eX
+ h1 Sill /l-n ) ,
where /l-n are the positive solutions of the following equation
Then t.he series
2 2t f2 (lfn
w(X , t) = "C
00
D n
e _a 1'-,.1 fin
e ' e"
. fi" )
-cos-:r+hjsm-xe'
n=l
satisfies the equation (6.14) and the boundary conditions (6.16) and (6.17). By the
orthogonality of the system of functions on the interval [0, e]
and the initial condition (6.18) we obtain
jU(x)-v(x)) (fJencoslfenx+hjSinlfenx) dx (nEN).

o
u(O, t) = 0, u G, t) = (t> 0), (6.21)
Solution. Taking u(x,t) = S(x) +v(x,t) the problem (6.21) can be written as
5" fJ2v _ ov _ 3x
+ ox2 - ot 2e,
5(0) + v(O, t) = 0, (t > 0),
S(x) + v(x,O) = ~ (1- e 3X )
The solution of the problem
S"(X) = _2e 3x 5(0) = 0, 5(1) = 0,

IS
S(x) = ~ (1 -e 3X ) - ~ (1 - e3 ) x.
The solution of second part of the considered problem, i.e., of the problem
v(O, t) = 0, (t > 0), (6.22)
IS
L Cr, sin(2'lmx) exp( -2tn 7r

00
v(x, t) = 2 2 ).
n=1
Using the boundary conditions we obtain
Cn = -4 (1 - e3 ) x
Jx sin(n7rx)dx = -'---~--'-
1/2
4(1-e )(-lt 3
(n EN).
9 9n7r
o
The solution of our problem has the form
(_l)n sin(27rnx)
2(
u(x, t) = - 1 - e3x
)
-
4(
-
)
1 - e3 x + 4(1 - e3 )
L
00
exp( _2tn 2 7r 2 ).
9 9 971" n=1 n
(0 < x < e, t > 0),
u(O, t) = 0, u(e, t) = (t > 0), (6.23)
u(x,O) = f(x) (0 < x < f).
Solution. Let us suppose that the solution of the problem (6.23) has the form of
generalized Fourier series
00
(6.24)
n=1
where Xn are eigenfunctions of the Sturm-Liuoville problem X" + >-X = 0, with

associated homogeneous boundary conditions, with eigenvalues >-n.
The right-hand side function r = 7"(x, t) call be expanded also in the generalized
Fourier series as
00
7"(X, t) = :L r,Jt)X,,(.r),
n=l
where the coefiicients 7"n can be written
J
f
7"n(t) = IIXn(x)II- 2 7"(x, t)Xn(x).

o
Assuming that termwise differentiation of the series (6.24) is permitted we obtain
OU
at = :L T~(t)Xn(t),
00
n=l
~U
= :LTn(t)X:(x) = - :L ASn(t)Xn(x).
00 00
ox2
n=1 n=1
The partial differential equation from (6.23) can be written as

00 00
:L (T~(t) + a 2 AnTn(t)) Xn(x) = :L rn(t)Xn(x).

n=1 n=1
So we obtain the differential equation
(6.25)
i
with the solution
Tn(t) = (Cn + exp(a 2 AnT)rn(T)dT) exp( _a 2Ant).
If t = 0, then Tn(O) = Cn. Taking the initial condition as

00
n=1
where in are the coefficients of the Fourier series i.e.,
J
f
in = IIXn ll- 2 i(x)Xn(x)dx,

o
we can say that
Tn(O) = in = Cn.
The solution of equation (6.25), with corresponding condition can be written as
Jrn( T)
t
Tn(t) = in exp( -Ana2t) + exp( -Ana2(t - T))dT.
o
Example 6.35 Find the solution of the following problem
-811. - -8 11.
2 X
= -cost (0 < x < 1, t > 0),
8t 8x 2 2
(6.26)
11.(0, t) = 0, u(l,t) =0 (t > 0),
u(x,O) =0 (O<x<l).
Solution. The eigenvalues for the corresponding Sturm-Liuoville problem are An =

n 2 7r 2 and eigenfunctions are Xn = sin n7rX (n EN). In this case we can write
x OCJ
'2 cos t = ; Tn(t) sin nn,
where
Tn(t) =
.
Ilsmn7rxll
_;1 -cosismn7rxdx
x. (_l)n
= --cost.
2
o
2 n7r
Let us consider the solution in form
L Tn(t) sin nn.

OCJ
u(x, t) =
n=1
After termwise differentiation we obtain differential equation
whose solution is
From u(x, 0) = 0, it follows that en = 0 and the solution of the considered problem
IS
Example 6.36 Find the so/uion
au 2 a2u
~-a~=r(x,t) (0 < x < f!, t > 0),
ui ux
(6.27)
u(O, t) = A(t), u(f, t) = B(t) (t > 0),
u(x,O) = f(.r) (0 < x < g).
Solution. Let us introduce the function vas v(x, t) = u(x, t) - V(x, t), where
.r
V(x, t) = A(t) + e(B(t) - A(T)),
satisfying the conditions V(O, t) = A(t), V(f!, t) = B(t) (t > 0).

The function v then satisfies
av 2 a2 v aV
at - a ax 2 = r(x, t) - 8t (0 < x < , t > 0),
v(O, t) = 0, v(f,O) = (t > 0),

(6.28)
u(x,O) = f(x) - V(x, 0) (0 < x < g).

The function v(x, t) can be found as in the Example 6.34.
(0 < x < f!, t > 0),
(6.29)
u(f!, t) = bo + blt (t> 0),
u(x,O)=o (O<x<f!).
Solution. Taking v(x, t) = u(x, i) - V(x, i), where
we obtain that the function v satisfies the nonhomogeneous equation
av
- - a -
2a 2V = - ( a1 + -X) (b 1 - aj)
at ax 2 f! '
with initial condition
v(x,O) =- (ao + :Z(bo - a o)) ,

and homogeneous boundary conditions
v(O, t) = 0, v(f, t) = O.
The solution of previous problem is treated as the generalized Fourier series
n?TX
v(x, t) = L vn(t) sin -C-,
00
n=1
n?TX n 2?T 2
where sin -f- are eigenfunctions for eigenvalues An = - f - for the corresponding
Sturm-Liuoville problem.
The right-hand side function - (a + 7(b ad) can be expanded also in the
l l -
generalized Fourier series as
where the coefficients Tn can be written
~ (al(I_ (-It) + bl -al(_lt+ I ) .

?T n n
Using differential equation (6.25) we have
(6.30)
where In are the coefficients of the Fourier series
X) n?TX
- ( ao + e(bo - ao) = L In sin -f-'
00
n=l
and have the for:n
In = . n?TX 2
-II sm-f-W J
l
(aoX . n?TX
+ e(bo - ao) ) sm- d
f- X,
o
The solution of equation (6.30), with corresponding condition is
+ J;;:2
t (
al bl - al
-;;:-(1 - (-It) + -n-(-I)
n+l ) 2 a2(t - T) ) dT
n 7r
exp (--C-
2
o
l
2 (a-;;:-(1-
+;;: (-1) n) + -n-(
bl - al -1) n+l) (1 - 22
exp ( --f.-a
n 7r 2 t) ) )

x
u(x, i) = ao + ali + f. (bo - ao + (b l - ar)t)
+2- L ~. - 1 (a l(
-1-(-1) l -- (
n ) +b- al- 1 )n+l) ( 1-exp ( -n-2 7ra
2
2 t ))) .
a 2 7r 2 n=l n 2 n n
02U ou
(0 < x < 1, t > 0),
ox 2 ot'
u(O, t) = 2t + sin t, u(l, t) = 2t (t > 0), (6.31)
u(x,O) = 0 (0 < x < 1).
Solution. Taking v(x,t) = u(x,t) - V(x,t), where V(x,t) = 2t + (1- x)sint, we

obtain that the function v satisfies the nonhomogeneous equation
OV 02V
- - - = - (2 + (1 - x) cos t) (0 < x < 1, t > 0)
ot ox 2 (6.32)
v(x,O) = (0 :s; x :s; 1), v(O, t) = 0, v(l,t) = (t > 0).
The solution of this problem is treated as the generalized Fourier series
L
00
v(x, t) = vn(t) sin mfX,

n=l
where sin n7f are eigenfunctions for eigenvalues An = n 27f2, for the corresponding
Sturm-Liuoville problem.
The function - (2 + (1 - x) cos t) can be expanded also in the generalized Fourier
senes as
L
00
- (2 + (1 - x) cos t) = rn(t) sin n7fX,

n=l
where the coefficients rn(t) can be written
J(2 + (1 -
1
rn (t) -II sin n7fX 11- 2 x) cos t) sin mfxdx

o
- ~ (cos t + 2 (1 -
n7f
( -1 n) , n = 1,2,3, ....
From the partial differential equation we have

Un(O) = O.
Therefore the solution of the problem (6.32) has the form
2 ( 2 2( ( 2 2) ) 4 ( exp ( n 2 7f 2 ) - 1)
( 44) n 7f exp -n 7f t -cos)-smt -
1 + n 7f
33
n7f n 7f
The solution of our problem is

Example 6.39 Find the solution
[Pu au 2
- 2 = - + 1- 3xt + x (0 < x < 1, t> 0),
ax at
(6.33)
u(O, t) = t, u(1,t) = t3 (t > 0),
U(X,0)=X2 (0<x<1).
Solution. Taking u(x, t) = V(x, t) + W(x, t), we obtain the following problem
a2 v a2 w av aw
at + 1 -
2
ax2 + ax 2 = at + 3xt +x (0 < x < 1, t > 0),
(6.34)
V(O, t) + W(O, t) = t, V(I, t) + W(I, t) = t 3 (t > 0),
V(x, 0) + W(x, 0) = x2 (0 < x < 1).

The function
V(x, t) = t + (e - t)x, (6.35)

satisfies the conditions V(O, t) = t, V(I, t) = t 3 , V(x,O) = O. Therefore, we have
to solve the nonhomogeneous problem
a 2 w _ aw 2
ax2 - at + (0 < x < 1, t > 0),
W(O, t) = 0, W(1,t) =0 (t> 0), (6.36)
W(x,O) = x 2 (O<x<I).
The function W = W(x, t) can be found as in previous two examples. The
second way is to introduce two functions S and v such that W(x, t) = S(x) +v(x, t),
where S(x) satisfies the following problem
S"(x) = 2,
S(O) = 0, S(I) = o. (6.37)
The solution of this problem is S(x) = x 2 - x.
a2v _ av =0 (0 < x < 1, t > 0),

ax 2 at
and the conditions
v(O, t) = 0, v(1,t) =0 (t > 0), v(x,O)=x (O<x<l).


00
v(x, t) = 2:= An sin(mrx)exp(n 2 7r 2 t),

n=1
where
An = 2 J 1
X
2
sin(mrx) dx = - ( _l)n+l.
n7r
o
2 (_It+1
+ - 2:=
n
u(x, t) = t + (t 3 - t)x + x2 - X sin(n7rX) exp(n 2 7r 2 t).
7r n=l n
Example 6.40 Determine the solution of two dimensional heat conduction problem
ou = a 2(02U + 02U) (0 < x < 1, 0< Y < 2, t > 0),

ot ox 2 oy2
u(O,y,t) =0, U(I,y,t) = (0 < y < 2, t > 0),
u(x, 0, t) = 0,
u(x,y,O) = f(x,y)
Solution. The solution can be found by using the method of separation variables
Taking
u(x,y, t) = X(x) Y(y) T(t),
we obtain
X" 1'" T'
-
X
= ---+
l'
-=,\
a 2T '
where'\ does not depend on variables x, y, t. Therefore we obtain two Sturm-Liuoville
systems X" +,\2 X =
(0 < x < 1), 1'" + 1121' = 0 (0 < x < 2),
1'(0) = 1'(2) = 0, with corresponding eigenvalues
respectively and eigenfunctions
The solution of the first order differential equation
(t> 0),
216 CHAPTER- 6. PAHABOLIC EQUATIONS
lS
T(t) = Cn,m exp(-a 2 (.\;' + fJ;n)t).
The solution of this heat equation with homogeneous boundary conditions has the
form
u(x, y, t) = 2:= 2:= Cm,n exp( -a

00 00 2 2
(An
2 n7rX .
+ fJm)t) sm -f..- Sln -f..-'
n7ry
(6.38)
n=1 m=1 I 2
For t = 0 we obtain the double Fourier series
n7rX . n7ry
2:= 2:= cm,n sm -f..- sm -f..-'
00 00
f(x, y) = (6.39)
n=lm=1 1 2
The coefficients Cm,n can be found by
It is known that if f and l' is arc continuous functions in a rectangle 0 < x < 1,
o < y < 2, then the double Fourier series converges to f.
Example 6.41 Determine the solution of the problem given in Example 6.40 for
the initial function f(x, y) = 2.
Solution. The solution ofthis problem is given by relation (6.38) and the coefficient
Cn,m can be written as
IT JJ . --
, 2
Cn,m 8 n7rX . n7ry
sIn sm --
1200 I 2
8 (1- (-l)n)(1- (_1)m) 32 1

=
12 mn - 1f.. 2 (2m + 1)(2n + 1)'
Therefore the solution can be written as
32 00 00 1
u(x, y, t) = 12 ];];0 (2m + 1)(2n + 1)
'
Th IS . h . h . . 1 d' . ou ou 02U fYu ou
senes converges toget er Wlt Its partJa envatJves ox' oy' ox2' oy2' ot'
for each t > 0 and 0 < x < 1, 0 < y < 2.
Exercise 6.42 Show thal the solution of two dimensional problem
ou(O,y,t)
-a} Ox + f3lU (O,y,t ) = 0 (0 < y < 2, t > 0),
(0 < y < 2, t > 0),
ou(x,O,t)
-a3 oy + f33 U(X, 0, t ) = 0 (O<X<l, t>O),
(0 < x < ll t > 0),
u(x, y, 0) ::= f(x)g(y) (0 < x < }, 0 < y < 2)'

is the product of the solutions of the following one dimensional problems
ou(O, t)
-a} ox + /31 u(O, t) = 0 (t > 0),
+ f32U(1l t) =
OU(l, t)
a2 ax (t > 0),
U(x,O) = f(x) O<X<l'
(0 < y < 2, t > 0),
au(O, t)
-a3-[}y- + f33U(O, t) = (t > 0),
aU(2, t)
a4 oy + f34U(2,t) = 0 (t > 0),
U(y,O) = g(y) (0 < y < 2)'

Exercise 6.43 Determine the solution of the two dimensional problem
ou = a2 (02U + 02U) (0 < x < 1, 0 < y < 2, t > 0),

ot ox 2 oy2
u(O, y, t) = 0, U(I,y,t) =0 (0 < y < 2, t > 0),
ou(x,O,t)
oy = 0,
U(x,y,O)=xy (O<X<1,O<y<2).
Answer. The solution follows from Exercises 6.42, Exercises 6.19, Example 6.24
and has the form
(0 < x < 1> 0 < y < 2, t > 0),
u(O,y,t) = 0, U(l,y,t) =0 (0 < y < 2, t > 0),
OU(x,O,t) OU(X'2,t) = g(x)

oy = 0, (O<X<l' t>O),
oy
U(x,y,O)=xy (0<X<1,0<y<2).
Hint. Take u(x, y, t) = S(x, y) + v(x, y, t).

Exercise 6.45 Prove that the three dimensional heat equation
D..u = Ou (x E R 3 , t > 0),

ot
can be transformed in a system
D..W+.\W=O, T'+.\T = 0
by using the separation variables method
u(x,y,z,t) = W(x,y,z) T(t).

Example 6.46 Determine the temperature distribution on a very very long rod,
(semi-infinite rod), iJ one end is always kept on the zero temperature while the initial
temperature distribution is given by the Junction f.
Solution. In this case we consider the heat equation with semi-infinite domain,
i.e., the problem
(0 < x < 00, t > 0),
u(O,t)=O (t > 0), u(x,O) = f(x) (0 < x < 00).
au
Further on, let us suppose that u and ax are bounded when x --+ 00.
Using the separation variables u(x, t) = X(x)T(t) we obtain the problem
XI/(x) +'\X = 0, X(O) = 0,
which solution is
X(x)=sinsx, .\=S2.
Since the solution must be bounded when x --+ 00, we need .\ = S2 > o. Let us
remark that in the case on bounded intervals the eigenvalues are not discrete, in
this case they are arbitrary. The solution of the equation T' + .\a 2 T = 0 for .\ = S2,
IS
The solution of considered problem can be written as
J
00
u(x, t) =exp( _a 2 s 2 t) . B(s) sin(sx)ds. (6.40 )

o
In relation (6.40) we have the integral as the suprerposition of separates functions
over all values of s and for arbitrary function B. Using the initial condition we have
JB(s) sin(sx)ds
00
f(x) = (0 < x < (0).

o
This is the Fourier sine integral representation of f and the function B can be
obtained from
B(s) = -;
2
Jf(O sin(sOd(
00
o
(6.41)
Let us replace B form (6.41) into (6.40). So we obtain the following form of
solution
u(x,t) = II f(Osin(sOsin(sx)exp(-a2s2t)d~ds
~
o 0
= ;: !')00
[f(O
(00
d~.
sin(sOsin(sx)cxp(-a 2s 2t)ds
)
This can be transformed as
200
u(x,t) = ;[f(O (00
[(coss(x-0-coss(x+O)exp(-a2s2t)ds ) d~
Compare with Example 6.5. We used the following rezult
Example 6.47 Determine the solution of the following problem

au = a -a u 2 2
(-00 < x < 00, t > 0),
at
-
ax 2
u(x,O) = f(x) (-00 < x < (0).

au
Let us also, suppose that u and ax are bounded when x ---t 00.
Solution. The solution of considered problem can be written as
J
00
u(x,t) = exp( _a 2s 2t). (A(s) cos(sx) + B(s) sin(sx))ds. (6.42)
o
where
A(s) =; J f(O cos(sOd~,

2 00
-00
B(s) = ; Jf(O sin(sOd~.

2 00
-00
Example 6.48 Determine the solution of problem on an injinil.c domain
u(x,O) =e x2
(-00 < x < (0),
u and ax
au arc bounded when x ---+ 00.
Similarly as in Example 6.47 by using the separation variables u(x, t) = X(x)T(t)

we obtain
X(x) = A(s) cos(sx) + B(s) sin(sx), ,\ = s2 > O.

So, the solution of considered problem can be written as
00
u(x,t) =; exp(-a 2s 2t). (A(s)cos(sx) + B(s)sin(sx))ds, (6.43)
o
Using the initial condition we get
00
e- x2
=; (A(s)cos(sx) + B(s) sin(sx))ds,
-00
where
00
1;00 2
A(s) = ;- e- x cos sxdx B( s) = ~ ; e- x2
sin sxdx. (6.44 )
In this case, it is allowed to make differentiation under the integral.
A'( s ) = -;-1;00 xe- x Slnsx dx 2
-00
1 e- X2 . 1
00 s 00 _x2 S
- - - sm(sx) - - ; e cos sxdx =--A(s).
7r 2 -00 27r 2
The solution of the following differential equation

s
A'(s) + 2"A = 0,
is A( s) = Al exp ( - :2) , and it satisfies the initial condition

A(O)
1;00 e-
= -; x
2
dx
1
= Vi.
-00
Therefore
A(5) 1
= -exp
Vi
(52)
-- .
4
Since e- x2 sin(sx) is an odd function, it holds that B(s) = o. So the solution of the
considered problem with infinite domain can be written as
u(x, t) 1;00exp (_a s t -"4S2) . cos(5x)ds,

= Vi 2 2

Exercise 6.49 Solve the mixed problem for r > 0 and t > 0
with the initial condition

u(r,O) = f(r) r> 0
and boundary condition
u(Ro, t) = 0 (t > 0),
where the function f and the constant Ro > 0 are given.
Hints. Use the Fourier method of separation of variables in the form
u(r,t) = R(r) T(t).
The solution is of the form
where Ifn are the positive zeros of the Bessel function Jo, and an are determined by
the initial condition. The Bessel function J" (n E Z+) has the following represen-
tation by series
6.3. HEAT CONDUCTION 223
6.3 Heat conduction

Our task is to develop the mathematical model of the heat conduction both in
the one-dimensional homogeneous thin rod or wire and in the three-dimensional
homogeneous body, in which heat can flow freely.
Let us consider a long thin homogeneous rod of length made of a uniform
material. For simplicity, we assume that the road coincides with the segment [0, ]
on the x-axis. If u(x, t) is the temperature measured at the point x and at the time
t, then the rate of heat flow Q = Q(x, t) at the point x and at the time t (the heat
flux) is, in view of the Fourier's low of heat conduction, equal to
Q = -k au (6.45 )
ax
In (6.45), k is a constant denoting the heat conductivity of the wire. The sign "-"
in relation (6.45) shows that the heat flows from hotter to cooler parts.
Let S be the cross section area of the uniform rod. Then, the amount of heat
f'lQl, entering any cross section S during the time interval (i, t + f'lt). equals
Thus during the time interval [iI, t 2] the amount of heat QI, written in the ir uegral
form is
(6.46)
From the other hand, in order to raise the temperature of the rod for f'lu, one
needs the amount of heat f'lQ2 equal to
where c > 0 is the specific heat constant and p is the constant mass density.
Thus [Xl, X2] is a part of the rod, then Q2, equals to
a
Ja~
X2
Q2 = Scp dx. (6.47)

X,
Finally, if a heat source F = F(x, t) at the point X and at the time t is present
in the rod, then the corresponding amount of heat f'lQ3 is equal to
f'lQ3 = S F(x, y) ~x ~i,

or in integral form
X2 t2
Q3 = sJ J F(x,t)dtdx. (6.48)
The low of conservation of energy implies
or
X2
= Cp J(U(~,t2) -U(~,tl))d~
XI
Applying the mean value theorem, we obtain
or
where t 3 , t 4 , t s , and X3, X4, Xs, are the points from the intervals (tl, t 2 ) and (Xl, X2)
respectively. After dividing with tlt tlx, we can let Xl -+ x, X2) -+ X and tl -+ t,
t 2 ) -+ t, so we obtain the heat equation
a ( au) au
ax kax +F(x,t)=cPat'
Usually, the last equation is written in the form
au 2a 2u
at =a ax 2 + lex, t),
where a 2 = ~ and lex, t) = F(x, t).

cp cp
We turn now to three dimensional case. Let P( x, y, z) be a point of the three-
dimensional body, and u(P, t) = u(x, y, z, t) be the temperature measured at the
point P and at the time t. Let Q = Q( x, y, z, t) be the heat flux at the point (x, y, z)
and the time t. Now, the Fourier's low of heat conduction implies
6.8. HEAT CONDUCTION 225
Q = -kVu. (G.49)
Let V be the volume of the body bounded by surface 0". Then, applying the low
of conservation of heat energy, similarly as in the one-dimensional case, we have
kp JJJ(U(P,i2) - u(P,i
v
1 )) dVp 1(J !J F( P, i) dVp ) dt
In the last multiple integral, Wn = Q . n is the heat flux in direction n. Applying

the divergence theorem, we obtain
JvJJ(U(P,i2) - u(P,i1)dV JJvJJF(p,t)dtdx

t2
kp p
t1
Similarly as in one-dimensional case we apply the mean value theorem
where i 3 , i 4 , t s , are the points from the interval (tl' t 2 ) and PI, p 2 , p 3 , is a point
from V.
So we obtain
. ) ) ou(x,y,z,t)
-dzvW(x, y, z, t + F(x, y, z, t = cp ot '
or
o ( ou) a ( au) a ( ou) au

ax kax +ay kay +az kaz +F(x,y,z,t)=cP{jt.
Thus we obtain the three dimensional heat equation. Mostly it is written as
(6.50)
or
k
where a 2 =- is positive const.ant which is called the dijJllsivity of the mat.erial of
p
F
the rod and Fl = -.
cp
Chapter 7
Numerical Methods
7.0.1 Preliminaries
The most commonly used method for obtaining the approximate solutions of certain
partial differential equations is the finite differences method. In order to employ this
method we replace the continuous independent variables x, Y, z, i, ... , by a finite
numbers of discrete variables Xi, Yj, Zm, in,"', namely we determine suitable
mesh points
Xi = Xo + ih, i = 0, l, 2, ... ,
Yj=Yo+jk, j=0,l,2, ... ,
and so on. Replacing each of derivatives by a suitable difference quotient, a dif-
ference equation for i,j = 0, l, 2, ... , is obtained. It represents a system of
algebraic equations, whose solutions can be treated as the approximate solut.ions of
the considered problem at the mesh points.
As usual, we consider the following partial differential equation
L( U) =A( X,y )fJ2u(x,y) B( )02 U(X,y) C( )ou(x,y)

ox2 + X,y oy2 + x,y ox
(7.1)
OU(X, y)
+D(x,y) oy + G(x,y)u(x,y) = F(x,y),
where A,B,C,D,G,F are continuous functions on the set Q C R2, and on its
boundary oQ.
We denote by Ui,j, Ui-l,j, Ui,j-l, Ui+l,j, Ui,j+l the corresponding values for
u(x;,Yj), U(Xi-l,Yj), U(Xi,Yj_l), U(Xi+l,Yj), U(Xi,Yj+l), respectively. The deriva-
tives are replaced by the corresponding difference quotients
forward difference
and OU ~,
Ui J"+1 - Ui ,J" .
oy ~ k '
227

228 CHAPTER 7. NUMERICAL METHODS
backward difference
au ~ Ui,j - Ui-1,j au
ay U '-1 - u',).j
ax ~ h and ~ ',J k
centered difference
au ~ Ui+1,j - Ui-1,j d au +1 - U'' ,'-1
U'1,1
ax
).
ay
_~
~ 2h an 2k'
centered difference for second derivative
r--r'-r
(x,t+k)
! !(X,t)!
(x - h,t) t-__--t__-t (x + h,t)
........_..... .!
I :
..... (x,t-k)
I
Figure 7.1
If we denote by Ai,j, Bi,i> Ci,j, Di,i> Gi,j, and Fi,j, for i = 0, 1, 2, ... , J =
0, 1, 2, ... , the values of the functions A, B, C, D, G and F, respectively, at the
points Xi, Yj, then we obtain the following difference equation
A- .. Ui+1,j - 2Ui,j + Ui-l,i B-. Ui,j+1 - 2Ui,j + Ui,j-l
',) h2 + >,J
k2
(7.2)
+c ... ui+1,i 2h
>,)
- Ui-l,j + D . .. Ui,j+1 - Ui,j-l + G u' . =
2k >,) >,) ',)
f ..
t,)'
for i = 0, 1, 2, ... , j = 0, 1, 2, ....
The Error of Approximation

The error of approximation for the solution of difference equation (7.2), which can be
treated as the approximate solution for differential equation (7.1). can be obtained
by using the Taylor formulas. Namely, if we suppose that the solution of a given
problem has continuous derivatives up the order four, then there exist X, fj i, y
such that
Xi - h S; X S; Xi + h, Yj - k S; fj S; Xi - k,
Xi - h S; i S; Xi + h, YJ - k <
-
y"Z- <
-
Xt - k,
229
satisfying
U(Xi,Yj + k) - U(Xi,Yj - h) 811(Xi,Yj) k2 8 3 u(Xi,fj)

2k = 8y + (3 . 8 y3 '
If we take the same denotation as in (7.2), in the neighbourhood of the points

(Xi,Yj), it holds
8U(Xi,yj) D 8U(Xi,Yj) G ( )
+Ci,j 8x + i,j 8y + i,jll Xi, Yj
(7.3)
for i = 0, 1, 2, ... , j = 0, 1, 2, .... In (7.3), Ri,j is the error of approximation

and has the form
(7.4)
Taking
230 CHAPTER 7. NUMEIUCAL METHODS
we obtain the estimation of the error of approximation as

h2 k2
iRi,ji ::::; 12 (Ai,jM4 + 2Ci,jM3 ) + 12 (Bi,jM4 + 2Di,jM3), (7.6)
for i = O,1,2, ... , j = O,l,2, ....

Exercise 7.1 Dete1'mine the difference equation for Poisson's equation
cPu 8 2u
8x 2 + 8y2 = F(x,y) (x,y E R),
with the condition
uiaQ = </>.
Solution. Using the finite difference method, we obtain the difference equation
Ui+I,j - 2Ui,j + Ui-l,j Ui,j+l - 2Ui,j + Ui,j-l _ P .
h2 + k2 - I,J'
and the values at the boundary points satisfy the boundary conditions given by the
the function </>.
Example 7.2 Using the finite difference method determine the approximate solu-
tion of the Dirichlet problem on a square, which is given by
8 2u 8 2u
8x2 + 8y2 = 0 (0 < X < 1, 0< y < I),
U(O,y) =0, u(l,y) = lOy (0 < y < 1),
u(x,O) =0, u(x,l) = lOx (O<x<l).

Solution. If we choose the same increment h = k on the x- and y- axes, then we
obtain the following difference equation corresponding to the Laplace equation
4Ui,j - (Ui+I,j + Ui-l,j + Ui,j+l + Ui,j-l) = o. (7.7)

Taking n = 5, i.e., h = 1/5, we obtain i,j = 1,2,3,4,5. in equation (7.7) and the
mesh of the given square is shown in Figure 7.2
The conditions of Dirichlet problem correspond to the following

UO,j = 0, US,j = 2j,
(7.8)
Ui,O = 0, Ui,5 = 2i,
231
for i = 0,1,2,3,4,5, j = 0,1,2,3,4,5, The equations (7.7) and (7.8) can be express
in terms of grid points in Figure 7.2 as
4Ul - U2 - Us = Ul,O + UO,I, 4U2 - U3 - UI - U6 = U2,O,
4U3 - U4 - U2 - U7 = U3,O, 4U4 - U3 - Us = U4,O + US,I,

4us - U6 - UI - Ug= UO,2, 4us - U2 - Us - U7 - UlO = 0,
4U7 - U3 - U6 - Us + Un = 0, 4us - U4 - U7 - U12 = US,2,
(7.9)
4ug - Us - UlO - Ul3 = UO,3, 4UlO - U6 - Un - Ul4 = 0,
Ug -
4un - UlO - U12 - U7 - UIS = 0, 4Ul2 - Un - Us - Ul6 = US,3,

4U13 - Ul4 - Ug = UO,4 + Ul,5, 4Ul4 - U13 - UlO - UIS = U2,S,
4UIS - Un - Ul4 - Ul6 = U3,S, 4Ul6 - Ul2 - U15 = US,4 + U4,S,
o 1
Figure 7.2
Using the boundary conditions we can evaluate the terms on the right-hand side
UI,O = U2,O = U3,O = U4,O = us,o = 0,

UO,! = UO,2 = UO,3 = UO,4 = uo,s = 0,
u!,S = 2, U2,S = 4, U3,S = 6, U4,S = 8,
US,! = 2, US,2 = 4, US,3 = 6, US,4 = 8.
The values Ul, U2, , Un are found by solving the system (7.9). We obtain the
solution of this system by using the MAPLE subroutine in Scientific WorkPlace 2.5
and they have the values:
UlO = 1. 8269, UIS = 4.4637, U4 = 1. 2234, Us = .46368,

U7 = .22339, U12 = 4.3403, UI3 = 1.4701, U2 = .34034,
Us = 2.4468, Ul4 = 2.9402, U6 = .71357, Un = 2. 7136,
Ul6 = 6.201, Ug = .94016, U3 = .44678, UI = .20101.
These are the approximate solutions of the considered problem at the corresponding
points.
Example 7.3 Solve the Laplace equation

fPu 82 u
-+ -=0
Dx 2 8y2
in a triangle 0 < x < y < 1, with boundary conditions
u(O, y) = 0, u(x, 1) = 5x(1- x), u(x, x) = O.
Solution. Let us choose the same increment h =k =~ on x- and y- axes, i.e.,

n = 5. From boundary conditions we have
UO,O = 0, UO,l = 0, UO,2 = 0, UO,3 = 0, uO,4 = 0, UO,5 = 0,
4 6 4
U 1,5 -- -
5'u 2,5 --u 3,5 -- -
5' U4,5 -
- -
5'
Ul,l = 0, U2,2 = 0, U3,3 = 0, U4,4 = 0, us,s = O.
The solutions of the system
4Ul,4 - Ul,3 - UZ,4 = UO,4 + Ul,S,

4U2,4 - Ul,4 - UZ,3 - U3,4 = U2,S, 4U3,4 - UZ,4 = U3,3 + U4,4 + U3,S,
are obtained from the MAPLE subroutine in Scientific WorkPlace 2.5 and they are
U2,4 = .54545, Ul,4 = .37273, Ul,3 = .14545,

U3,4 = .43636, u2,3 = .17273, Ul,2 = 3.6364 X 10- 2.
Example 7.4 Determine the solution of the Dirichlet problem on the region Q,
8zu 8zu
Dx z +Dy 2=0 (x,yEQ)
Ui8Q = r(x, y),

a) where Q is the triangle with vertices (-4,0), (4,0), (0,3) and the function r is
given by
DQnR,
r(x,y) = { (7.10)
(x,y) E DQ \ Rj
b) where Q is the ellipsis
and the function r is given by

r(x,y)=1-x 2 +y, xEDQ.
233
Solution.
a) Let us choose the same incrcmcnt h = k = 1, on x- and y- axes, i.e., n = 8.
'18
I
__ --~-~yt
isTI6
1 2 3 4 5:
13~j~.J",>
6 7 8 9
Figure 7.3
The points (-4,0), (-3,0), (-2,0), (-1,0), (0,0), (1,0), (2,0), (3,0), (4,0),
and (0,3) are denoted respectively by 1, 2, 3, 4, 5, 6, 7, 8,9, 18. The values
of the function U at these points are determined directly by the equality
Ui = r(x;, Yj).
The values at these points are
Ul = Ug = 4, UIS = 0, U2 = Us = 11,
U3 = U6 = 16, U4 = U7 = 19, Us = 20.
The values of the function U at the points (-1,1), (0,1) (1,1), and (0,2), de-
noted in Figure 7.3. by 11, 12, 13, 16 respectively can be determined directly,
because they together with their neighboring vertices do belong to Q.
Figure 7.4
The values at the points (-2,1), (2,1), (-1,2), (1,2), denot.ed by 10, 14, 15,
and 17 respectively can not be determincd directly because senne of their neigh-
boring vertices do not belong to Q. Their va.lucs are given by the following
formu lae (Figure 7.4)
where A(Xll yd is one of the mentioned points, 11B = 11(Xl - h, Y1), and dis
the distance from the boundary.
The line Y = ~x + 3 connects the points (-4,0) and (0,3) and the line y =
3 4 .
-4"x + 3 connects the pomts (4,0) and (0,3). Therefore
2
for the points 10 and 14 the distance from the boundary is d = 3;
1
for the points 15 and 17 the distance from the boundary is d = 3'
Therefore we can write
2
31113 +0
1114 = 2 ,
1 +-
3
(7.11)
1117 =
1
31116
1
+
1+-
3
Finally we have the following system
(7.12)
The solutions of system (7.11) and (7.12) are
1112 = 9.5293, 1111 = 8.1139, 1113 = 7.2805, U16 = 2.7226,

1110 = 3.2455, 1114 = 2.9122, U17 = .68066, UIS = .68066.
Example 7.5 The Poisson equation

82 11 82 u
8X2 + 8y2 = -1 (O<x<l,O<y<I),
235
U(O,y) =0, u(l,y) =0 (0 < y < I),

u(x,O) = 0, u(x,l)=O (0 < x < I),
characterizes a squa7'e membrane unde7' constant tensions on all sides and supporting
a uniformly distributed load, Dcte7'mine the approximate solution by using finite-
difference method for n = 4, h =k = ~
4
Solution. Firstly, note that the given equation is nonhomogeneous and on the
right-hand side is a constant function. The boundary conditions are equal zero
Ui,O = Ui,4 = UO,j = U4,j = O.

Therefore the values of U are symmetric, namely it holds
This means that it is enough to find the values of u at the points (1,1), (2,1), (1,2)
and (2,2),
The difference equation corresponding to the given nonhomogeneous differential
equation is
Ui+1,j + Ui-1,j + Ui,j+1 + Ui,j-1 - 4Ui,j = h 2 ,
So we have the system of equations
1 1 1
4U11 -
, , = -16'
2U1 2 - 4U2 1 -
. 2U1 1 -
, . = -16'
U2 2 - 4U2 2 -
, , = - -
4U1 2
16.
The solutions of this system are
U1,1 = 0,0429, U1,2 = 0,0547, U2,2 = 0,0703.
Example 7.6 Determine the approximate solution by using the finite-difference

1
method for n = 4, h = k = 4' for PDE
[Pu [Pu
ox2 + oy2 =0 (0 < x < I, 0 < y < I),
U(O,y) = 0, u(I,y) = 0 (O<y<I),

u(x,O) = 0, u(x, 1) = 10 (O<x<l),
236 CHAPTER 7. NUMEIUCAL METHODS
Solution. In this case the boundary conditions imply
Ui,O = UO,j = U4,j = 0, Ui,4 = 10,
and the following symmetry
1L1,2 = U3,2,
From the system
4U1,l - U1,2 - U2,l = 0,
4Ul,3 - Ul,2 - U2,3 = 10,
4U2,2 - 2Ul,2 - U2,3 - U2,1 = 0, 4U2,3 - 2Ul,3 - U2,2 = 10,
we obtain
U2,3 = 5.2679, Ul,3 = 4.2857, Ul,2 = 1. 875,

U2,2 = 2.5, u2,1 = .98214, Ul,l = .71429.
Example 7.7 Let us consider the mixed type problem for heat equation.
au a u 2
(t > 0, 0< x < L),
at ax 2
U(x,O) = f(x) (0 < x < L),
u(O, t) = <fo(t), u(L, t) = 1j;(t) (t > 0).

Using the method of finite differences determine the difference equations by and
estimate the error of approximation.
Solution. Taking Xi = ih, tj = jk, i = 0,1,2, ... , j = 0,1,2, ... , and replacing
both the spatial derivative
aax2u
2
d h . d' .
an t e time envatIve
au
at
by their difference quotients (taking the forward difference for time derivative and
the central one for the second space one) we obtain the following difference equation
Ui,j+1 -
k
Ui,j
=
Ui+1,j - 2Ui,j
~~--~h72~--~~
Ui-l,j + (7.13)
The difference equation

Ui,i - Ui,j-l Ui+l,j - 2u;,j + Ui-l,j (7.14)
k h2
237
k
is obtained for the backward difference for time derivative. Taking (J" = h2 ' the
equation (7.]3) can be written in the form
(7.1 5)
while equation (7.14) becomes
(7.16)
One can prove that

1
the difference equation (7.15) is stable for 0 < (J" :S 2;
the difference equation (7.16) is stable for every (J".
If (J" = ~, then the difference equation (7.15) has the form
Ui,j+l =
Ui-l,j + Ui+l,j (7.17)
2
1
If (J" = 6' then the difference equation (7.15) has the form
Ui-l,j + 4Ui,j + Ui+1,J

Ui,j+1 = 6 (7.18)
The error of approximation for the approximate solution obtained by using equa-
tions (7.17), (7.18) and (7.16) are respectively
Iu - ill :S ~Mlh2,
lu - ill :S 1~5 M 2 h\ (7.19)
for 0 :S x :S L, O:S t :S T, where il is the exact solution of the considered problem

and u is the approximate one and
Example 7.8 Determine the difference equations corresponding to the nonhomoge-

neolls parabolic equation
au a u 2
(t > 0, 0 < x < L),
at = ax2 + F(x, t)
1 1
for a = 2 and a = 6' Estimate the error of approximation for 0 :::; t :::; T.
Solution. Taking the forward difference for the time derivative and central one for
the second space derivative we obtain the following difference equation
Ui,j+1 = (1 - 2a)ui,j - a(ui+l,j + Ui-l,j) + kFi,j' (7.20)

1
If a = 2 then the difference equation (7.15) has the form
Ui,j+l =
ui-l,i +
2
Ui+1,j
+ kF.i,j (7.21 )
1
If a = 6" then the difference equation (7.15) has the form
Ui,j+l =
Ui-l,j + 4Ui,j
6
+ Ui+l,j + k r.,}" D ..
(7.22)
The error of approximations for the approximate solutions obtained by using

equations (7.21) and (7.22) are
Iu - iii :::; '4 M4)

T ( M2 + 3 h,
2
(7.23)
IU-U-I :::;
T
72
(M3 M6) h4 '
3+"'5
where ii is the exact solution of the considered problem and U is the approximate
one and
i = 2,4,
i = 3,6.
Example 7.9 Determine the numerical solution for the following problem
au a 2u
(t > 0, 0 < x < 1),
at ax2
U(O, t) = u(1, t) = 0, (t > 0),

u(x,O) = x, (0 < x < 1),
by using the finite difference method for the
239
1 1
a) mesh h = -, and k = -;
5 50
1
b) mesh h = k = -
5'
Solution.
a) Taking Xi = ih, tj = jk, h = !,

5
k = ~,
50
we have that ()" = !2 and therefore
the difference equation corresponding to this problem is
From the conditions we get
UO,j = X5,j = 0, j = 0,1,2, ... , UiO

, =~,
5 i = 0,1.2,3,4,5.
So we get
UO,O + U2,O 1 Ul 0 + U30 2 3

Ul,l = 2 = 5' U2,l = ' 2 ' - 5' 5'
U30 + U5 0 4 UO,l + U2,1 1 Ul,l + U3,l 2
U4 1
, ==' 2 1
- 5' - Ul 2 -
, - 2 5' U2,2 =
2 5'
3 U31 + Us 1 3 UO,2 + U2,2 1
5' U4,2= ' 2 ' -10' Ul,3 =
2 5'
Ul2 + U32 2
= '
U22 + U42 7 U3,2 + U5,2 3
U2,3 = ' 2 ' = 5' U3,3 2 ' -
20"
U43 =
2 10
b) In this case we have (}" = 5, and the difference equation has the form
Ui,jH = -9UiJ + 5(Ui-l,j + UiH,j) (7.24)
Show that this system is not stable.
au fPu
(t > 0, 0 < x < 1),
at ox2
U(x,O) = x (O<x<l),
aU(o, t) = 0 _OU-,:(1_,...:...t) =0 (t > 0).

at ' ot
Solution. If h =I I I
-5' k = -,
50
then c> = - and
2
therefore we have
(7.25 )
From the given conditions we get
UO,l - Ul,j U4,1 - US,j

--'----:-h-.:.:.. = 0, -----:--'-'- = 0,
h
wherefrom
Z
UO,j = Ul,j, U4,j=US,j, j=0,1,2, ... , Ui,O = 5' i = 0,1,2,3,4,5.
So we obtain
Uoo + U20 1 UI,O + U3,O 2
Ul,l = ' 2 ' = UO,l = 5' U2,l = 2 - 5'
3 U30 + Us 0 4
5' U4,1 = ' 2 ' = US,1 = 5'
Uo 1 + U21, 3
=' 2
UII + U31, 2
, =
U02 UI 2
, =' 2 10'
U22
, 5'
U42 = Us ,2 =
U3,l + US,1 7
, 2 10'
7 Ul2 + U3 2 9
20' u2,3 = ' 2 ' - 20'
11
U4,3 = US,3 =
U32
' 2
+ US2' 13
20' 20
(t > 0, 0 < x < y < 1),
U(O,y,t) = u(x,O,t) = u(x,x,t) = 0,

and initial condition u( x, y, 0) = x( 1 - y).
Solution. The given PDE is the two dimensional heat equation. Let us take
Xi = ih, Yj - kj, h = k,
241
[or i = 0,1,2,3,4,5, j = 0,1,2,3,4,5. Then after replacing the second derivatives

with corresponding difference quotients we obtain
8 2u + cPu ~ 11;+1,j + Ui-1,j + Ui,j+1 + U;,j-1 - 411i,j

8x 2 8y2 h2
Let us replace the time derivative ~~ with
'(l
U'I.,J + 6t) - (t)
Ut,l
6t
Then the difference equation corresponding to the two dimensional heat equation is
Ut ,}(t + 6t) - Ut,l(t) Ui+1,j + Ui-1,j + Ui,j+1 + Ui,j-1 - 4U;,j
6t h2
or
U;,j(t 6t (Ui+1,j(t) + Ui_l,j(t) + Ui,j+1(t) + Ui,j_1(t)) +

+ 6t) = h2 6t) Ui,j(t).
( 1 - 4h2
1 1 6t 1
Let us use the mesh h = 4' 6t 64' Then h2 - 4' and the corresponding
difference equation has the form
Using the initial and boundary conditions together with the previous equation we
obtain
i 4- i
UO,j = U;,4 = 0, U,",1 = -.
4 -4- for t = 0,
UO,j = Ui,1 = Ui,i = 0, i = 0,1,2,3,4, j = 0,1,2,3,4,

113 1
for each value t = -,
64
t = -, t = -, t = -
32 64 16
We have
1) 1 1
U2,3 ( 64 =4 (U3,3(0) + U1,3(0) + U3,2(0)) = 8;
U1,2 C12) = U2,3 (312) = ~ . U1,3 (614) = 614;

U,1,3 G2) = ~ ( U,2,3 (614) + 1l1,2 C:J) = }16;
Ul,2 (:4) = U,2,3 (634) = ~ . Ul,3 C1J = 611;
Ul,3 (t+
1
64) = ~ (U2,3 (:4) +U,1,2
3
(6 4)) = 1~8;
1 1)
Ul,2
( 16 + 64 = U2,3 (1
16 + 641) = 4"'
1
ul,3
(1)
16 =
1
512;
Ul,3 C16 + 61J = ~ (U,2,3 C16) + C16) )

U,1,2
1
= 10 24'
Exercise 7.12 Determine the numerical solution for the following problem
aU, a 2u 02U
-
ot
=ox-2+oy2
- (t>o, O<x<l, O<y<l),
u(O,y,t) = u(x,O,t) = u(1,y,t) = u(x,l,t) = 0,

and initial condition u(x,y,O) = xy. Use the mesh h = k = ~, flt = _1_.
5 125
Exercise 7.13 Determine the difference equation for three dimensional Laplace
equation
(t > 0, 0< x < 1, 0< y < 1, < z < I),

by using the finite differences method.
Exercise 7.14 Determine the difference equation for three dimensional heat equa-
tion
ou
-
at
=ox
02U
-2+oy2
- +OZ2
-
02U 02U
(t > 0, 0< x < 1, 0< Y < I, < z < 1),
by using the finite differences method.

243
Example 7.15 Determine the difference equation and the error of approximation
corresponding to the mixed type problem for the hyperbolic equation
(t>O,O<x<L),
u(x,O) = f(x), ou~,O) = g(x) (0 < x < L),
u(O, t) = rfJ(t) , u(L, t) = 1jJ(t) (t > 0).
Estimate the error of approximation for 0 ::; t ::; T.
Solution. Taking Xi = ih, tj = j k, i = 0, 1,2, ... , j = 0,1,2, ... , and replacing

both spatial derivative
by their difference quotients we get
Ui,j+1 - 2Ui,j + Ui,j-1 UH1,j - 2Ui,j + Ui-1,j (7.26)

k2 h2
for i = 0,1,2, ... , j = 0,1,2, .... Denoting by a = ~, we obtain
Ui,j+1 = 2Ui,j - Ui,j-1 + a(ui,j+1 - 2Ui,j + ui,j+d (7.27)
One can prove that the difference equation (7.27) is stable for < a ::; 1.
The error of approximation for the approximate solution obtained by using equa-

tions (7.27)
(7.28)
where it is the exact solution and U is the exact solution of the considered problem
and
i = 3,4.
If we put
OU(x,O) _ ( )
at -gx,
where g is from initial conditions
U;,1 - 1/;,0 _ ( )_
k - 9 Xi - gi,
then from the relations

Ui,l = fi + kg;,
we obtain the values of the function U at the points (i,j), i = 0,1, ... , j = 0,1.
The error of approximation can be determined as
ah
IU"l- ull < -2M2 .
I, I, -
(7.29)
where
Exercise 7.16 Determine the approximate solution of the problem
(O<x<l, t>O)
U(O, t) = u(l, t) = 0 (t > 0),
u(x, 1) = x(l - x), ou(x,O) =0 (O<x<l).

ot
Answer. Let us choose the same increment h = k = 0,1 on the x- and the t-
axes, ( a = 1.) From the initial conditions we have
i = 0,1, ... , 10
Ui,l = Ui,O = 1iO (I - lio) , i=0,1, ... ,10.
We still have to solve the following system of equations
Ui,j+1 = Ui-l,j + U;,j - Ui,j-b i=I,2, ... ,10, j=1,2, ... 10.
Example 7.17 Determine the numerical solution of the problem
02U 02U
(O<x<l, t>O),
ot 2 = ox 2
U(x,O) = 0, OU(x,O) = I (0 < x < 1),
ot
U(O, t) = 0, u(l,t)=O (t>O).
Estimate the error of approximation for 0 ~ t ~ 1.

245
Solution. Taking h = k = ~, we obtain the difference equation

10
Ui,j+1 = Ui+l,j + Ui-l,j - Ui,j-l' (7.30)
If we replace au~~, 0) with Ui,l : Ui,O, then from the initial conditions we get
1
Ui,O = 0, Ui,l = 10 . 1 i = 0,1,2, ... , 10.
Also from the boundary conditions we have
UO,j = U10,j = 0, j = 0,1, ... 10.
From the previous equations we get

1 2
U 1,2 -u
- 2,1 +u0,1 -u 1,0 - -
- 10' U2,2 = U3,1 + Ul,l - U2,0 = 10'
2 2
U 3,2 -
- u 4,1 + u 2,1 - u 3,0 -
- - 10' U4,2 = US,l + U3,1 - u4,O = 10'
Similarly we obtain
2
uS,2 = U6,2 = U7,2 = UB,2 = U9,2 = 10'
1 2
Ul,3 = U2,2 + UO,2 - Ul,l = 10' u2,3 = U3,2 + Ul,2 - U2,1 = 10'
3 3
U 3,3 -
- u 4,2 + u 2,2 - u 3,1 -
- - 10' U4,3 = US,2 + U3,2 - U3,1 = 10'
3
uS,3 = u6,3 = U7,3 = UB,3 = U9,3 = 10'
1 3
Ul 4 , = U23 , + Uo,3 - U1 2, = -10' U2,4 = U3,3 + U2,3 - U2,2 = 10'
4 4
U3 , 4 -- U 4, 3 + U2 ,3 - U3 ,2 -
- - 10' U4,4 = US,3 + U3,3 - U3,2 = 10'
2
US,4 = U6,4 = U7,4 = UB,4 = U9,4 = '5'
3 1
U1,S = 10' U2,S = U3,S = U4,S = us,s = U6,S = U7,S = us,s = U9,S = 2'
3 6
U1,6 = 10' U26
I
= U36
I
= U46
,
= US6
,
= U66
,
= U76
,
= US6
,
= Ug6
I
= 10
-,
3 7
UI,7 = 10' u2,7 = u3,7 = U4,7 = U5,7 = U6,7 = U7,7 = u8,7 = u9,7 = 10'
4 8
UI,8 = 10' U2,8 = U3,8 = U4,8 = U5,8 = U6,8 = u7,8 = u8,S = U9,8 = 10'
5 9
UI,9 = 10' U29
,
= U39
I
= U48
,
= US9
,
= U69
,
= U79
I
= US9
,
= U99
1
= 10
-.
Exercise 7.18 Determine the numerical solution of the problem
(t>O,O<x<I),
U(x,O) = x 2 aU~,O) = x, (0 < x < 1),
U(O,t) = 0, u(l,t) = 0, (t > 0).
Answer. Taking h = k = 110' we obtain the difference equation
i j
Ui,i+l = Ui+l,i + Ui-l,i - Ui,i-l + Xi + ti = Ui+l,i + Ui-l,i - Ui,i-l + 10 + 10
Example 7.19 (von Neumann's criteria for stability) Examine the behavior
of the special exponential solution
U1.,) . -- e,iSe'i S, (7.31)
where B E R, z = A and A = Al + ZA2, of the difference equation (correspoding

to the wave equation)
Ui,j+I - 2Ui,i + Ui,j-l _ Ui+l,j - 2Ui,j + Ui-l,j = (7.32)

k2 h2 '

Ui,O = Ii (7.33)
and
ff'
+ gi k + 2k
2
Ui,l = 1; , (7.34)
where Ii = I(ih),gi = g(ih) and If' = f"(ih) for
a) k/h> 1;
b) k/h ~ 1.
247
Solution. Putting the special form of the solution (7.31) in (7.32) we obtain the
equation
P. 2 0 (7 35)
sm 2" = h 2 sm 2"
2).
.
For i = 0 we obtain from (7.31) the initial condition Ui,O = e,iO.
The solution (7.31) is continuously dependent from the initial condition if for
every real 0 the solution from (7.31) remains bounded as j - t 00. This will be
satisfied if the zeros). of of the equation (7.35) have nonnegative imaginary part for
every O.
a) If klh > 1 we can always find 0 such that the right part of the equation (7.35)
is greater of one. Then for such 0 the zeros). of the equation (7.35) are complex
and conjugate to each other. Therefore one of them is with negative imaginary
part and by the general statement from the beginning it is not stable as h - t 0
and k - t O.
b) If kl h :::; 1, then all zeros of the equation (7.35) are real. Therefore Ui,j from
(7.31) are bounded as i - t 00. Hence the solution Ui,j of (7.32), (7.33) and
(7.34) will converges to the solution U of the corresponding Cauchy problem
for one-dimensional wave equation.
Exercise 7.20 Let A,B,C,D be the corners of the parallelogram whose sides are
the characteristics of the homogeneneous one-dimensional wave equation
(7.36)
Prove that a function U E C 2 (R2) is a solution of the equation (7.36) if and only if
the function U satisfies the functional equation
u(A) + u(C) = u(B) + u(D) (7.37)
for all parallelograms ABC D obtained by the given construction.
Exercise 7.21 Apply the preceding result for the construction of the solution of
the mixed type problem for the one- dimensional wave equation on the strip S =
{(x,y)IO < x < f.,t > O} with the usual initial conditions
u(x,O) = f(x), au(x,O) = g(x) (0 < x < f.),

at
u(O,t) = m1(t), u(f.,t) = m2(t) (t > 0),
where f,m1,m2 E C 2,g E C 1 and ml(O) = J(O),m~(O) = g(O),m~(O) = 1"(0),

m2(0) = J(f.), m~(O) = g(f.), m~(O) = 1"(f.).
Hints. Divide the strip S with characteristics of the one-dimensional wave equation
in the following way. First take the characteristics at the endpoints (0,0) and (f!, 0)
of the interval [0, f!] till the borders x = 0, x = f!. Then at the points of the
intersections with the borders take characteristics at these intersection points, and
continue this procedure. At the first part, which is a triangle on the interval [0, C],
we can find all values of the solution by the D'Alambert formula. Then in the next
two neighbourhood triangles we can find all values of the solution by the functional
equation (7.37). Continuing this procedure we can reach every point since it belongs
to some part of the constructed partition of the strip.
Chapter 8
Lebesgue's Integral and the

Fourier Transform
8.1 Lebesgue's Integral and the L 2 (Q) Space

8.1.1 Preliminaries
Let A c Rn be a (Lebesgue) measurable set. For a measurable function
f : A ~ [0,00] its Lebesgue integral on a measurable subset E of A is given by
Jf(x) dx
E
= sup Jsex) dx,
E

where the supremum is taken over all simple functions s, s = 2::;=1 aiXE;, which
satisfy the inequality ~ s ~ f (we are using the convention 000 = 0) and XE. are
the characteristic functions of sets Ei E E (i = 1, ... , n) and UEi = E, where E is
the a-algebra of measurable sets.
Theorem 8.1 (Beppo-Levi) Let {fn}nEN be a sequence of measurable functions

on the set A such that

1) ~ f1(X) ~ h(x) ~ ... < 00 for every x E A,
2) limn ..... co fn(x) = f(x) for every x E A.
Then f is measurable and limn..... co J fn(x) dx = J f(x) dx.
A A
Theorem 8.2 (Fatou lemma) If {fn}nEN is a sequence of measurable functions

defined on the set A and with values in [0,00], then
Jli..~inffn(x)dx ~ li...~inf Jfn(x)dx.

A A
Let L1 (A) denote the set of all measurable complex functions defined on the
measurable set A with the property f If(x)1 dx < 00.
A
249

250 CHAPTER 8. LEBESGUE'S INTEGRAL, FOURIER TR.ANSFOR.M
Theorem 8.3 (Lebesgue theorem on convergence) Let {fn}nEN be a sequence

of measurable functions on I.he set A such that J(x) = limn_OQ In(x) exists almost
everywhere on A. rr
9 is an integrable Junction such that Ifn(x)1 :::: Ig(x)1 a.e.,
n = 1,2, ... , then J is integrable and
lim
n~OQ
JIJn(x) - J(x)1 d.T = O.
A
Consider now the Lebesgue measure on the real line. Let function f be integrable
on the interval [a, b] and define
J
x
F(x) = f(t) dt (a::::x::::b).
Then F has a finite derivative almost everywhere on the interval [a, b] and F' = f
almost everywhere on [a, b] and F is absolute continuous on the interval, i.e., for
every E; > 0 there exists b > 0 such that
n-l
L !F(Xk+l) - F(xk)1 < E;
k=l
for
n-l
L IXk+l - xkl < b and a = Xl :::: X2 :::: ... :::: Xn = b.
k=l
Theorem 8.4 (Fubini) If f is an integrable function on a measurable set
A = Al X A2 from Rn, where Al and A2 are measurable subsets of R m (m:::: n)
and Rn-m, respectively, then the functions It and g defined by
h(Xl"",Xm) = J f(xl, ... ,xn)dxm+l ... dxn

A2
J
and
g(Xm+l,""Xn) = f(Xl, ... ,Xn)dxl ... dxm
A,
are integrable and
J f( Xl, ... , Xn) dXI ... dXn J(J f( Xl, ... , Xn) dXm+! ... dxn)d.rl ... dXm
A A, A2
J(J f(Xl, ... , Xn) dXm+l ... dXn) dXl ... dXm.
A, A2
Let X be a vector space. A function II . II : X --7 [0,00) is a norm on X if for

every X, y E X and every scalar a it satisfies the following conditions:
8.1. LEBESGUE'S INTEGRAL AND THE L 2 (Q) SPACE 251
(i) Ilxll = 0 if[ x = 0;

(ii) Ilaxll = lal'llxll;
(iii) Ilx + yll ~ Ilxll + Ilyll
Then the pair (X, II II) is called normed space.
Let p be a fixed real number grea.ter or equal one. Then we say that a function
defined on a measurable subset Q of Rn belongs to Lp( Q) if J If(xW dx < 00, where
Q
functions are equal if they are equal almost everywhere, i.e., they differ only on a
set of a measure zero.
A complete normed space is called Banach space. The set Lp( Q) endowed with the
following norm
IIfIlLp(Q) = (J If(xW dx)~
Q
is a Banach space.
Let f,g E Lp(Q) and l/p+ l/q = 1. Then we have the Holder inequality
I
Q
Jf(x)g(x) dxl ~ IlfIILp(Q)llgliLq
and the Minkowski inequality
If Q = R n, then we denote Lp = Lp (Rn). In the special case p = 2 the space L2 (Q)

is a Hilbert space, i.e., a Banach space whose norm is induced by the following scalar
product ( see more in Chapter 10.)
Ulg)L2(Q) = Jf(x)g(x) dx,

Q
where !J is the conjugate of the complex valued function g.
Theorem 8.5 (Continuity of the integral) Every function f from L 2 (Q) is con-
tinuous in mean, i.e., for every c: > 0 there exists 0 > 0 such that
IIf(x + v) - f(x)IIL2(Q) < c: for Ivl < 0, x + v E Q.

We denote by Ljoc(Q) the set of all functions which belong to LI(K) for some
compact subset K of Q.
252 CHAPTER 8. LEBESGUE'S INTEGRAL, FOURIER TRANSFORM
Change of variables
Let Q C Rn be a bounded open set with a boundary aQ of measure zero which is
mapped with a function <1> : Q -+ Rn, (<1>1' ... ' <1> n) of class C 1 as a bijection on a
bounded open set n with a boundary an of measure zero with the Jacobian
Then the following holds.

1) The functions <1> and <1>-1 map every set of measure zero on a set of a measure
zero;
2) The function f : n -+ R is (Lebesgue) integrable on the set n if and only if
the function (J 0 <1IJI is (Lebesgue) integrable on the set Q. Then the following
equality holds
Jf(y) dy = J(J
n Q
0 <1(x)IJI dx.

Example 8.1 Prove that the Dirichlet function defined by
D(x) = { ~ for x rational,

for x irrational.
is Lebesgue integrable on the interval [0,1] but not Riemann integrable.
Solution. The function D is equal almost everywhere zero, which implies that its
Lebesgue integral is zero. On the other hand, for this function all upper Darboux
sums are equal one and all lower Darboux sums are equal zero, which imply the
non-existence of the Riemann integral.
Exercise 8.2 Prove that the following functions are Lebesgue integrable and find
the corresponding Lebesgue integrals:
a)
for x irrational and x > ~,
for x irrational and x < ~,
for x rational.
b)
R(x) = { 5 for x = and n have no common divisors,
~, m
for x irrational
(the Riemann function).
8.1. LEBESGUE'S INTEGRAL AND THE L 2 (Q) SPACE 253
c) g(x) = sgn(sin;:).
A nswers. a) 35
108' b) 0, c)1-2In2.
Exercise 8.3 For which real number a the Junction
1
f(x) = -
Ixla
is integrable on B(O, 1) = {x Ilxl < 1 }?
Answer. a < n.
Exercise 8.4 Show that for the integral
JJ
00 00
e- xy sin x sin y dxdy

o 0
there exist the iterated integrals but not the double integral.
Exercise 8.5 Show that for the integral
there exist the iterated integrals but they are different.
Example 8.6 Show that for the integral
there exist the iterated integrals and they are equal, but the double integral does not
exist.
Solution. Both iterated integrals are zero. If we suppose that there exists the
double integral then it would exist also on the square {(x, y) I 0 ~ x ~ 1, 0 ~ y ~ 1 }
and it would be possible to apply the Fubini theorem, but for x = 0 the function
J
1
xy d =~_ X
(x 2 + y2)2 Y 2x 2(x2 + 1)
o
is not integrable.
254 CIIAPTEH 8. LEBESGUE'S INTEGHAL, FOURIER THANSFORM
Exercise 8.7 Let J be a bounded measurable Junction on the region Q. Prove thai
the Junction
g(x) =
J(z)
Ix _ zla dz J
Q
(Hilbert transform) belongs to Ck(Rn) if k < n - [aJ, where [a] is the integer' part oj
a.
Exercise 8.8 !"ind which functions belong to the space Ck(Q) or Lp(Q):
SlilX
a) f(x) = X 3/ 4 on Q = (0,1);
b)
X-l/3 cos X for x irrational,
{
g(x) = ~-1/3 for x rational and x -=I- 0,
for x =
on Q = [0,1];
c)
sin(xy)
for x 2 + y2 -=I- 0,
~2 + y2
h(x) = {
for (x,y) = (0,0).
Exercise 8.9 Find the real number a such that the function Ixl- a belongs L 2 (Q) if
a) Q = {x Ilxl < I}, b) Q = {x Ilxl > I}, c) Q = Rn,
Answers. a) a <~, b) a >~, c) there is no a.
Exercise 8.10 Let (J' be a continuous and positive function defined on a region Q.
Denote by L 2.,,(Q) the space of functions such that (J'1f1 2 E L 1 (Q). Prove that
(fIg) = J(J'(x)f(x)g(x) dx
Q
is a scalar product (see more details in Chapter 10.) on L 2.,,(Q).
Exercise 8.11 IJ (J' is a bounded function on Q (see Exercise 8.10) then
Example 8.12 Show that the following systems of functions are orthogonal in the
corresponding spaces oj functions.
a) 1,sinx,cosx,sin2x,cos2x, ... in L 2 (0,27r);
b) Pn(x) = 2n1n ! J2n; 1 dd: Jx2 - 1), n = 0,1,2,... in L2( -1,1) (Legendre
B.1. LEBESGUE'S INTEGRAL AND THE L 2 (Q) SPACE 255
polynomials);
c) Hn(x) = (-1te x2 ddx: e-x2, n 0,1,2, ...
polynomials) .
d)
1
n-l n-"2
{or--<x< - -
J' 2m - 2m '
1
n--
{or _ _ 2 <x < ~
J' 2m - 2m '
n -1 n
o forxrJ.[-- - ]
2m '2 m
for mEN, 1 ~ n ~ 2m in L 2(0, 1) (Haar's system).
Exercise 8.13 Let A eRn be a (Lebesgue) measurable set and f a function

f: A x (a, b) -+ C such that for almost every (fixed) x E A the function t t-+ f(x, t)
is continuous on the interval (a,b), and for t E (a,b) the function x t-+ f(x,t)
is measurable, and there exists an integrable function h on A such that for every
t E (a, b) we have If(x, t)1 ~ h(x) for every x E A. Prove that the function F
defined by
F(t) = J
A
f(x,t)dx
is continuous on the interval (a, b).
Hints. Apply the Lebesgue theorem on convergence on the sequence of functions

{J(x, ti)}iEN, where {t;}iEN is a convergent sequence from the interval (a, b).
Exercise 8.14 Let A c Rn be a (Lebesgue) measurable set and f a function
f : A x (a, b) -+ C such that for almost every (fixed) x E A the functions t t-+
f(x, t) and t t-+ Dd(x, t) are continuous on the interval (a, b), and for t E (a, b) the
functions x t-+ f(x, t) and x t-+ Dd(x, t) are measurable and there exist integrable
functions hI and h2 on A such that for every t E ( a, b) we have If( x, t) I ~ hI (x) and
IDd(x, t)1 ~ h2(x) for almost every x E A. Prove that the function F defined by
F(t) = J
A
f(x,t)dx
belongs to C I (a, b) and
F'(t) = J
A
Dd(x, t) dx (t E (a, b)).
Hints. Apply the Lagrange Mean value theorem and Lebesgue theorem on conver-
gence.
256 CHA.PTER 8. LEBESGUE'S INTEGRAL, FOURIER TRANSFORM
8.2 Delta Nets

8.2.1 Preliminaries
A continuous function 'P has a compact support in region Q if 'P( x) = 0 outside of
some compact subset of Q. \Ve denote
supp 'P = {x I 'P(x) # O}
and by C~ (Q) the set of all functions which have continuous derivative till the order
s on Q and they have compact support contained in Q.
We introduce a function 81 defined by
E, (x) ~ { {I ~ t' for - 1 < t < 1,

for It I ?: 1.
81 has the following properties:
a) it belongs to COO(Q);
b) 81 (t) ?: 0, 81 (t) = 0 for It I ?: 1.
We choose the constant c such that the following condition is satisfied
c) J 81 (t)dt = l.
R
In the following considerations we can take instead the function bJ any other function
which satisfies the conditions a), b) and c).
For h > 0 the function bh defined by
bh(X) =
1 Ixl
hn bl( h) (x ERn)
has the following properties
1) it belongs to COO(Q);
2) bh(t) ?: 0, bh(t) = 0 for It I ?: h;
3) f 8h (t)dt = 1;
Rn
4) For every a E Z~ and every x ERn we have
ID<>8h (l x lll ::; h::I<>I'

where c" > 0 is a constant. We shall call {bhh>o delta net.
If f E L 2 (Q) ( or L l (Q)) then
Ilh - fIIL 2 (Q) -+ 0 (or Ilfh - flILdQ) -+ 0) as h -+ 0,
where fh is defined by
fh(X) = Jf(Y)bh(lx - yl) dy (x E Q)

Q
and fh is called mollifier. If f E C(Q), then Ifh(X) - f(x)1 -+ 0 as h -+ 0 uniformly

on each compact subset of Q.
8.2. DELTA NETS 257

Example 8.15 Let
fh(X) = Jf(y)5h(lx - yl) dy.
Q
Prove
a) If f E L 2 (Q), then
b) If f E L 1 (Q), then
c) If f E C(Q), then
lim lih(x) - f(x)1
h--+O
=0
uniformly on every compact subset of Q.
Solution.
a) Using the properties 2) and 3) of Dh and the Cauchy-Schwartz inequality we

have
2
Ifh(X) - f(xW = J f(Y)Dh(lx - yl) dy - f(x) J 5h(lx - yl) dy

x-yl<h Ix-yl<h
< J 5Wx-yl)dy J If(y)-f(x)1 dy 2
Ix-YI<h Ix-YI<h
< ~ j If(x+v)-f(xWdv
Ivl<h
for M > O. Applying the Fubini theorem 8.4 we obtain further
Ilih - fIIL(Q) = ~ j dx j If(x + v) - f(xW dv

Q Ivl<h
~ j dvjlf(x+v)-f(xWdv.
Ivl<h Q
Using now the continuity of the integral we obtain the desired conclusion.
b) The proof is analogous to a).

258 CHAPTER 8 LEBESGUE'S INTEGRAL, FOURIER TRANSFORM
Example 8.16 Prove thal the set Cgo(Q) is dense in the spaces L 1(Q) and L 2 (Q).
Solution. We shall prove only the case L 2 ( Q) since for L1 (Q) the proof is analogous.
For an arbitrary c: > 0 by the absolute continuity of the Lebesgue integral there exists
8 > 0 such that
J If(xW dx < c:;,
Q\Q6
where Q5 is a subset of Q which contains all points whose distance till the boundary
GQ is greater than 8. We introduce a function 9 E L 2 (Q) in the following way
for x E Q5
g(x) = { ~(x) for x E Q \ Q5.
It is obvious that
c:
Ilg - fIIL2(Q) ~ 2".
By Example 8.15 there exists ho > 0 such that Ilgh - gIlL2(Q) < ~ for 0 < h ~ ho
For enough small h we obtain gh E CO'(Q), and we have
Hence Cgo(Q) is dense in L 2 (Q).
Example 8.17 Let f E L1 be a function which is almost everywhere zero outside

of some compact subset J( of Q. Prove that the function
fh(X) = J f(y)8 h (lx - yDdy

Q
belongs to Coo (Q) for h > 0 which is smaller than the distance of the set K to the
border GQ (for Q = Rn we can take for h any positive number).
Solution. Since 8h E COO( Q) we obtain by the property of Lebesgue integral on

exchange with the derivative that fh E COO( Q).
If h > 0 is smaller than the distance d of the set K to the border GQ, then we have
for every x from Q whose distance from K is greater than d
fh(x) = J f(y)8h(lx - yl) dy

K
since for y E ]{ and Ix - yl ~ d always 8h (lx - yl) = 0 holds.
Example 8.18 Let Q C Rn be an open set and ]{ C Q an compact set. Then

there exists a function r.p from CO'( Q) such that r.p( x) = 0 for every x in some
neighbourhood of the set K and 0 ~ r.p ::::; 1.
8.2. DELTA NETS 259
Solution. Let. d> 0 be the distance between the sets f{ and 0(2. We introduce the
set
d
K d/ 2 = {x I d(x,K):::: "2}
and a function 9 by
g(x) ={ ~ for x E f{d/2,

for x E Q \ J{d/2'
Using Example 8.17, since the distance bet.ween J{d/2 and oQ is greater or equal
with d/2, we have for h = d/2 that gh E Co(Q). We have for x E f{d/2
gh(X) = J g(y)bh(lx - yl) dy = J bh(lx - yl) dy = J bh(lx - yl) dy = l.

Q Kd / 2 !x-y!<;;d/4
We take 'P = gh
Example 8.19 (Partition of the unity) Let f{ C Rn be a compact set covered
by open sets 0 1 , ... , OS) i.e., K C UfO;. Then
a) There exist open sets O~, ... , 0: such that 0; are compact, 0: C Oi (i = 1, ... , s)
and f{ C Ui=lO;.
b) There exist functions 'ljJi E Co(Rn) (i = 1, ... , s) such that supp'IjJi C Q, 0::::
'ljJi :::: 1 and 2:::=1 'ljJi = 1 in some neighbourhood of the set f{.
Solution. a) We choose O~ such that O~ = {x I d(x,1(') < d/2} where f{' =
f{ \ (Ui=20i) and d = d(f{',OOl)' We construct O~ in an analogous way starting by
0~,02""'Os.
b) Applying Example 8.18 on the compact set K; = 0:
and open set Oi (i = I, ... , s)
o(
we obtain that there exist functions 'Pi E C Oi) with the properties 'Pi = 1 in some
neighbourhood of the set K and 0 :::: 'Pi :::: 1. We define the functions 'Pi equal zero
outside of Oi, respectively. We define
'ljJ1='P1, 'ljJi='Pi(1-'P1)oo(1-'Pi-I) (i=2,oo.,s).
It is easy to check that 'ljJi satisfy the prescribed properties. Namely, by the con-
struction 'ljJi E Co(Rn), suPP'IjJi C Oi and 0 :::: 'ljJi :::: 1. We have to prove only that
2::i=I 'ljJi = 1 in some neighbourhood of the set K. We can represent the functions 'ljJi
in the following form
'ljJ1 = 1 - (1 - 'Pr)
'ljJi = (1 - 'PI) 00. (1- 'Pi-I) - (1 - 'Pr) 00' (1 - 'Pi-I)(1- 'Pi)

for i = 2, ... ,s. This implies
s
L 'ljJi = 1 -
(1 - 'Pl)(l - 'P2) (1 - 'Ps), 00
i=l
Then by the fact that 'Pi = 1 in some neighbourhood of the set K; we obtain the
desired equality.
Exercise 8.20 Let 0 eRn be an open set. Prove that the set of functions C3"( 0)
is a normed space with the norm
Ilfll = L sup IDQ f(x)1 (f E C({'( 0)),

1<>1~k xED
but it is not a complete space!
Exercise 8.21 Let Q c Rn be a bounded region. Prove that the set Coo(Q) en-
dowed with the norm fmm Exercise 8.20 which takes the following form
IIfll = L ma~ IDQ f(x)1 (f E C({'(Q))

l"'l~k xEQ
is a Banach space.
B.3 The Surface Integrals

8.3.1 Preliminaries
Let U, VeRn be bounded regions and ([> a bijection of the class C k from fJ onto
V such that <1>-1 is also of the class Ck. Such function ([> is called dipheomorphism
between U and V.
Let Q c Rn (n 2 2) be a bounded region Q is locally quadratic if for every point
Xo on the boundary 8Q there exists region U which contains the point Xo and a C 1
dipheomorphism from U onto V for some region VeRn such that it maps the
set Un Q on a n-dimensional parallelepiped P = (aI, bd x ... x (an, bn ) and the set
U n 8Q maps on a side or union of sides of P.
A compact subset S of Rn (n 2 2) is a locally quadratic (n - I)-dimensional
part of surface if for every point Xo E S there exists a region U which contains the
point Xo and there is a C 1 -dipheomorphism from U onto V, where V is a region,
such that it maps the set S n U on some side or union of sides of P.
A compact subset S of Rn,n 2:: 2 is locally quadratic (n -I)-dimensional part
of surface if for each point Xo from S there exist a region U which contains the
point Xo and a C1-dipheomorphism ([> from U onto V, where V is a region and
 maps S n U on some side or union of sides of some generalized parallelepiped
P = (aI, b1 ) X x (an, bn ). By compactness of S it can he covered by finite number
of regions Ub ... , Us and there are dipheomorphisms (;) which map Ui onto Vi,
for some regions 11; (i = 1, ... , s) such that S n U i is mapped on some side or
union of sides of some generalized parallelepiped P CUi. For U1 , .. , Us we take
the corresponding partition of the unity ( see Example 8.19) 'Pi E C({'(U;) with
l::=l 'Pi(X) = 1 in some neighbourhood of S. A function f : A ~ C is Lebesgue
integrable on the surface S if for every i, i = 1, ... , s, the function 'P;j 0 (([>(i)t 1
8.3. THE SURFACE INTEGRALS 261
is Lebesgue integrable on the sides <1>(i)(S n U i ). Then the corresponding Lebesgue

surface integral of f on S is given by
Jf dS = t J rpJ 0 ((i)t1IJd dx, (8.1 )

s .=l<l>(i)(snV'.)
where the integration goes through parallelepipeds from Rn-l, <1>(i)(S n U i ) is a part
of some
II X ... x I m- 1 x {am} X 1m+! X .,. X In
for m = mi and Ji is a vector valued determinant of the following form
Ji = Dm_1(((i))-lh Dm_1(((i))-lh Dm_1(((i)t1)n ,
Dm+!((<1>(i))-lh Dm+1((<1>(i))-lh Dm+1(( (i))-l)n
where
el = (1,0, ... ,0),e2 = (O,l,O, ... ,O), ... ,en = (0,0, ... ,1)
and
((i)t 1 = (((i)t1h, ... ,(((i)t1)n).
Lp(S), 1 :::::; p < 00, be the set of all functions f : S -t C such that for all
i,i = 1, ... ,8, we have

Example 8.22 Let Q be a bounded region of Rn. Prove that if 8Q E Ck, k ~ 1, is
a (n -I)-dimensional surface (manifold), then Q is a generalized quadratic region.
Solution. Let
2fi2 CHAPTER 8. LEBESGUE'S INTEGRAL, FOURIER TRANSFORM
for <1> E Ck(P n ) and Pn = (al,b l ) x ... x (an-I,bn-d. Taking enough small Pn we
can obtain that
a~ = ~f <1> > an and b~ = sup <1> < bn
Pn Pn
Let
Then the Ck-dipheomorphism <1> = (<1>1, . , <1>n) defined by
maps, for a~ = a~ - an and b~ = bn - b~, the region
on the closure of the region
such that U n Q is mapped on the generalized parallelepiped
and Un GQ is mapped on the side P n X {O}.
Exercise 8.23 Let Q C Rn be a locally quadratic bounded region. Prove that

(a, b) X Q, for a, bE R, a < b, is a locally quadratic region in Rn+l.
Exercise 8.24 Let Q eRn be a locally quadratic bounded region. Prove that GQ is
a locally quadratic part of surface.
Example 8.25 Prove that the cylinder
v= {(x,y,z)1 x 2 + y2 < R2,0 < z < H}.
for R > 0 and H > 0, is a locally quadratic region and GQ is a locally quadratic part
of the surface.
8..3. THE SURFACE INTEGRALS 263
Solution. We introduce the cylindric coordinate system x = rsincp, !J = rsincp

and z = z. Then we can cover the border by
R 3R 3r. 37r }
y z)1
{(x " 3
- < r < - 2' - -
4
< 1/'
r
< -4' -1 < z < If + 1 ,
R 3R r. 7r.
{(x,y,z)l- < r' < - , - < '-P < -,-I < z < II + I},
3 2 4 4
R R H
{(x, y, z)llxl < 2' Iyl < 2' Izl < 2}'
R R If 3H
{(x,y,z)llxl < 2,lyl < 2'2 < z < T}'
We introduce for i = 1, 2 the functions a(i) = (aii ), a~i), a~i), a(i) : Vi --7 U i, in the
following way
(8.2)
where
R 3R 3r. 3r.
r < -4' -1 < z < H + I} ,
{(r ,.,..,
In z)1 -
3 < r < - 2' - -
4 < ""
R 3R r. 7r.
{(r,<p,z)l- < r < - , - < '-P < -,-1 < z < 1l + I}.
324 4
In this way we obtain a CI-dipheomorphism from Vi onto Vi (i = 1,2), such that
<li(l) maps the set UI n V onto the parallelepiped
R 3r. 3r.
PI = {(r,'-P,z)l"3 < r < R'-4 < <P < 4,0 < z < H},
and <li(2) maps the set Uz n V onto the parallelepiped
R r. 7r.
Pz = {(r, <P, z)1 "3 < r < R, 4" < <P < 4,0 < z < H}.
On the borders <li(1) maps U I n av onto Pll U P l2 U P13 , where

R 3r. 3r.
Pn = {(r, '-P, z)1 "3 s:; r s:; R, -4 s:; '-P s:; 4' z = O},
R -3r. 3r.
P l2 = {(r,'-P,z)l"3 s:; r s:; R, -4- s:; '-P < 4'Z = H},
-3r. 3r.
PI3= {(r, <P, z)1 r = R, -4- s:; '-P s:; 4,0 s:; z s:; H},
and i{l(Z) maps U z n av onto P21 U P zz U PZ3 , where
R r. 7r.
P21 = {(r, <p, z)1 "3 s:; r s:; R, 4" s:; '-P s:; 4' z = O},
R 7r 77r
P22 = {(r,cp,z)1 3 :S; r:S; R,t;:S; cp:S; 4'z = Il},
7r 77r
P23 = {(r,!p,z)1 r = R,t;:S;!p:S; 4'0:S; z:S; Il}.
We choose for Cl-dipheomorphisms cll(3) and cll(4) the identical mappings. Then cll(3)
maps U3 n aVon the side
R R
P3l = {(x, y, z)llxl < 2' Iyl < 2' z = oJ,
and cll(4) maps the set U4 n aVon the side
R R
P4l = {(x,y,z)llxl < 2' Iyl < 2'z = Il}.
Since V is a locally quadratic region we have by the preceding Exercise that av is
a locally quadratic part of a surface.
Example 8.26 Find the surface integral J xds, where V is the cylinder from Ex-
8V
ample 8.25.
Solution. By the Example S.25 we have
4
U Ui:J avo
i=l
We take the corresponding partition of the unity !Pi E CO'(Ui ) (i = 1,2,3,4),
n
L !pi(X) = 1 (x E aV). (S.3)
i=l
We shall find the surface integral by (S.l). For that purpose we shall first find the
absolute value of the determinant Ji by on sides Pn, P12 , P13 , P2l , P22 , P23 , P3l , P4l .
We have on sides Pij (i,j = 1,2)
Ji = Dl((cll(i)-lh Dl((cll(i)-lh Dl((cll(i)-lh
D 2 ((cll(i)-lh D 2 ((cll(i)-lh D 2 ((cll(i)-lh
el e2 e3
cos cp sin !P 0
-r sin cp r cos !P 0
8.3. THE SURFACE INTEGRALS 265
Hence IJ;! = T. We have on sides Pi3 (i = 1,2)
Ji D 2 ((<li(i))-lh D 2 ((<li(i))-lh D 2 ((<]i(i))-lh
D 3 ((<li(i))-lh D 3 ((<li(i)t 1 h D 3 ((<li(i)t 1 h

eJ e2 C3
-Rsincp Rcoscp
001
= R cos cp . eJ + R sin cp . e2'

j
Hence IJil = R2 cos 2 cp + R2 sin 2 cp = R. It is obvious that on sides P31 and P4J
we have IJ;I = 1.
Using (8.1) we obtain
2 3 4
j f ds = LL j (cpJ 0 (<li(i)t 1 )lJil dcpdr +L j (cpJ 0 (<1>(i)t 1 )IJ;I, (8.4)
av .=1 J=IPij .=3Pil
By (8.3) we have
j (cpt! 0 (<1>(I)(I)IJ1 1 + j (cpz! 0 (<1>(2)(I)IJ21 + j (CP3f 0 (<1>(3)(I)IJ3 1

P ll P2l P3l
= (1
f
R
) It
a -.
~
4
3n-
(cpd)(rcoscp,rsincp,O)r dcp)dr
+ ill (1: (cp2f)(r

R l1!:
cos cp,rsincp,O)r dcp)dr
3
+1:(1: (cpd)(r
R R
cos cp, r sin cp, O)r dy)dx
2 2
R 211'
= j(j(cpd)(rcoscp,rsincp,O)r dcp)dr
o 0
R 211'
+j (j (cp2f)(r cos cp, r sin cp, O)r dcp )dr

o 0
R 21r
+ 1(I(rp3.f)(7.cosrp,rsinrp,0)r drp)dr.
o 0
Hence
1 (rpI! 0 (4)(I))-I)!Jll +1 (rpd 0 (4)(2)t 1 )!J21 +1 (rp3f 0 (4)(3))-I)IJ31
I'll 1'21 P31
R 2r.
= I(r 1 f(rcosrp,rsinrp,O) drp)dr. (8.5)

o 0
We obtain in an analogous way
1 (rpI! 0 (<1>(I)t 1)IJ11+ 1 (rpd 0 (<1>(2}t 1)IJ21+ 1 (<p4f 0 (<1>(4})-1)IJ41

~2 ~2 ft1
R 27r
= I(r 1 f(rcosrp,r sinrp, H) drp)dr. (8.6)

o 0
Finally, we have
1(rpI! 0 (4){1})-I)IJ1 1+ 1(rpd 0 (4)(2})-I)IJ21

P13 P23
2" R
= R 1(1 f(Rcosrp,Rsinrp,z)dz)drp. (8.7)
o 0
Putting (8.5), (8.6) and (8.7) in (8.4) we obtain
R 21r R 21r
1 f ds = I(r 1 f(rcosrp,rsinr.p, 0) drp)dr + I(r 1 f(rcosrp,rsin<p,H)drp)dr

av 0 0 0 0
2" H
+R 1(1 f(Rcosrp,Rsinrp,z)dz)drp. (8.8)
o 0
Hence
R 21r 2" H
1 xdS=2 I(r 1 rcosrpd<p)dr+RI(1 Rcosrpdz)d<p.

av 0 0 0 0
Exercise 8.27 Find the surface integml
1(x +
av
2 y2) dS,
where V is the cylinder from Example 8.25.

8.4. THE FOURIER TRANSFORM 267
Hint. Use the equality (8.8) from Example 8.26.
Exercise 8.28 Pmvc that the space Lp(S) is independent of the C01Jer VI, ... ,US)
C 1 -dipheomorphisms q,(i) and partition of the unity <Pi (i = 1, ... , s).
Hint. Use the facts that a C 1 -dipheomorphism maps a measurable function on a

measurable function and it maps an integrable function on an integrable function.
Exercise 8.29 Prove that the space Lp(S), 1 ~ p < 00 is a normed space equipped
with the following norm
IlflILp(S) = (J Ifl
S
P ds)~ = t J
.=I4>(i) (sni'7;J
(<p;lfI P 0 (q,(i)rl)IJil)~.
Remark 8.29.1. The space Lp(S) is a Banach space with respect to the introduced
norm iifIILp(s).
8.4 The Fourier Transform

8.4.1 Preliminaries
The Fourier transform of a function f E L1(Rn) is defined by
(FJ)(x) = j(x) = (2~)- J

Rn
e-' zx f(z) dz (x ERn),
n
where z . x =L ZiXi
;=1
Theorem 8.6 Let f E Ll(Rn). Then:
1) j E C(Rn);
2) if additionally Zk f(z) is also a function from L1 (Rn), then there exists Dkj(x)
and
Dd(x) = -zzk/(x);
more general, if additionally Z'kk . f(z) for k = 1, ... ,n are also functions from
L 1(Rn), then there exists DC. j(x) and
268 CIlAPTEH 8. LEBESGUE'S INTEGRAL, FOURIER TRANSFORM
3) if additionally Dd E L 1(Rn) U C(Rn), then

Dd(x) = zXki(x);
more gene/'al, if additionally D;k f E Ll(Rn) U C(Rn) for k = 1, ... ,n, then
Do J(x) = (-zxtf(x)
Jor a = (at, ... ,an) E Z~.
Theorem 8.7 For a linear differential operator
L(D) = L aoDo
lol$m
and a function f E cm(Rn) with DO f E L1(Rn) for 10'1:5 m the following equality
holds
L(i5)f(x) = L(zx)f(x).
Definition 8.8 The convolution oj two functions f, g E L1 (Rn) is defined by
(J*g)(x) = J
Rn
f(y)g(x-y)dy.
The following exchange formula allows us to transfer the convolution in the usual
product.
Theorem 8.9 If f,g E L1(Rn), then
/-;g(x) = (27r)~ f(x). g(x).
Definition 8.10 The inverse Fourier transform ;:-1 oj a function f E L1 (Rn) is
defined by
;:-l(J)(X) = fe-x).
Theorem 8.11 IJ J E L1(Rn),j E L1(Rn) and fez) is a continuous function at

the point z = zo, then f(zO) = ;:-l(j)(ZO), i.e.,
f(zO) = (27rt~ J e-zo':j(y) dy.

Rn
Definition 8.12 The equation

L(D,Dt)u = L aoDOu(x, t) =0 (8.9)
lal$m
is hyperbolic (with respect to the manifold t = 0) if the Cauchy problem (8.9) with
the conditions
fPu(x,O) =
J at f(x)
{O forO:5j < m
for j = m
-1, (8.10)
always has a solution u(x, t) E cm(Rn+1) for every f E q(Rn) for enough big s.

Example 8.30 Find the Fourier transJorm oj the Jollowing Junctions
a) e(x) = (27rt~ exp(_I:rt);

b) J(x) = (2tt~exp(-I:~\
Solution.
a) We have
(Fe)(x)
J
+00
where we have used that exp( _z2 /2) dz = ..j2;.
-00
b) We obtain by a)
Example 8.31 Solve the Cauchy problem for the heat equation
au
- = Llu (x E Rn,t > 0),
at
u(x,O) = f(x)
using the Fourier transform.
Solution. Applying the Fourier transform F on the both equation we obtain
au(z,t)
at + /z /2"( ) 0
U z, t = ,
u(Z,O) = J(z),
supposing that we can exchange the order of the derivative with respect to t and the
Fourier transform. The solution of this ordinary differential equation is given by
Applying now the inverse Fourier transform ;::--1 on both sides we obtain
u(x, t) = (27r)-~ Jj(z)e-lzI2te'ZX dz.

Rn
using the exchange formula and Example 8.30 we have
u(x, t)
Exercise 8.32 Prove that if Ql C Rn and Q2 C Rm are bounded regions, {cpihEN

an orthonormal complete system of functions in L 2( Ql) and {~j LEN an orthonormal
complete system of functions in L 2 (Q2), then the system of functions {CPi' ~ih,jEN
defined by
(i,j E N,x E Ql,y E Q2)
orthonormal and complete in the space L 2( Q1 X Q2)'
Hints. For the proof of the completeness of the system of functions {CPi' ~j} i,iEN use
the denseness of C(QI x Q2) in L 2(QI X Q2) and the fact that a system offunctions
{hi hEN is complete in the space L 2 ( Q) if for every function f the Parseval's identity
holds
f: IUlh )L
;=1
i 2 (Q) 12 = 11fl2 dx,
Q
and use the Fubini theorem.
Exercise 8.33 Prove that for f,g, hE Ll(Rn) a) f * 9 = 9 * Ij

b) U * g) * h = 1* (g * h)j c) III * gllLI :::; IIfllLI llgIIL 1
Example 8.34 A function f E L 2 (Q) is continuous in mean (quadratic) , i.e., for

every c > 0 there exists {j > 0 such that
III(x + v) - I(x)liL2(Q) < c
for each v which satisfies Ivl < {j (x + v E Q).

Solution. Let f E L 2 (Q) and a> 0 such that Q CC B(O, a). We define the function
F in the following way
for x E Q
F(x) = { ~(x) for x E B(O,3a) \ Q.
F belongs to L2(B(0,3a)). Since the set C(Q) is dense in L2(B(O,3a)), for c >
there exists a function F E C(B(0,3a)) such that
- c
IIF(x) - F(x)IIL 2 (B(O,3a)) < 3' (8.11)
We can suppose that F(x) = 0 for x E B(O, 3a) \ B(O, a). Namely, we can always
multiply F with a "cutoff" function on B(O,a). We have for Izl:<::: a
- - c
IIF(x + z) - F(x + z)IIL2(B(O,3a)) = IIF(x) - F(x)IIL2(B(O,a)) :<: : 3' (8.12)
Since the function P is uniformly continuous on B(O,2a), there exists 5> 0 (5 < a)
such that
- c
IIF(x + z) - F(x)IIL 2(B(O,2a)) :<: : 3' Izl < 5. (8.13)
Then we have by (8.12) and (8.13) for Izl < 5
Ilf(x + z) - f(x)IIL2(Q) = IIF(x + z) - F(x)IIL 2(B(O,2a))
:<: : IIF(x + z) - F(x + z)liL2(B(O,2a)) + IIF(x + z) - F(x)IIL,(B(O,2a))

- c c c
+IIF(x) - F(x)IIL 2 (B(O,2a)) :<: : "3 + "3 + "3 = c.
Exercise 8.35 (G~rding hyperbolicity condition) Prove thal if
L(zx,z)..) =f
for all complex numbers such that S')" :<: : -c, then the equation (8.9) is hyperbolic.
Remark 8.35.1 The given condition is also necessary for the hyperbolicity of the
equation (8.9).
Hints. Apply the Fourier transform on (8.9) and (8.10) to obtain the formal solution
u(x, t) = (2rrtR-
JRn
r c'XY K(y, t)j(y) dy,
where the function J( satisfies with respect to the variable t the following ordinary
differential equation
L(ty, DdK(y, t) = 0
272 CHAPTER 8. LEBESGUE'S INTEGRAL, FOURIER TRANSFORllI
and conditions
for 0 ~ j < Tn - 1
DfIy,t) ={ ~ for j = Tn - 1.
Prove that the function K has the following representation
K(y,t
)
= -1 i c zt dz
27fi c L(iy,z)
= -1 i e'At
27f c L(zy,z>..)
d>..,
where C is a closed countor which goes one times around all zeros of the polynomial
L(zy, z>..) = O. Use this result to prove the absolute convergence of the integrals
[ e'XYDfyCtK(y,t)j(y)dy
JRn
for lal + j ~ Tn which will imply that the given formal solution is the solution of the
problem.
Exercise 8.36 If all zeroes >"k of L( zy, z>..) = 0 are different, then J( (y, t) given by
K(y, t) = -27f
1 i C
e'At
L( zy, Z>..) d>..,
where C is a closed path which goes only one times around each zero of L(zy, z>..) = 0,
can be represented in the following form
m elAlt
K(y, t) 2: Dt L( zy, Z>..).
= j=l
Hint. Use theorem on residiums (see [21]).
Exercise 8.37 Prove that the wave equation

n
Dtu - 2:DJu =0
j=l
u(x,O) = 0 and Dtu(x, 0) = g(x) for 9 E c;+3(Rn),
is hyperbolic and that the solution is of the following form
u(x,t) = (27ft'
n 1. Rn
e ozy
sin(lylt) g(y)
Iyl
. dy.
Hints. Use the Gartiing condition and Exercise 8.36.

Example 8.38 Let r.p E Cg"(Rn) such that suppr.p c B(O,R). Then its Fourier
transform r.p = Fr.p can be extended from Rn onto en such that it will be analytical
function on en with the property that for every sEN there exists Cs > such that
where Z = (Zl, ... , zn).
Solution. For every Z E en the following integral is well- defined
and it is an analytical function ( since we can differentiate under the integral). Since
supp r.p C B(O, R) we have
Applying the partial integration we obtain
ZCXr.p( z) =
n
(211")-2
1 DCX(e-'ZX )
( )1 1 r.p(x) dx
B(O,R) -z
= ( -z)'O'
--n
(211") 2
1 B(O,R)
e- UX Dr.p(x) dx.
The last equality give us the following estimation
IZIlCX1 IZnIOnli(z)1 = Iz<>Ilcp(z)1

< (211"r~ f le-ozxlIDr.p(x)1 dx.
JB(O,R)
< (211")-~ f e9 (z.x) IDr.p(x) Idx

JB(O,R)
(211")-~ f ex .9z . IDr.p(x) Idx

JB(O,R)
~ (211"t~ sup ex .9z f ID<>r.p(x) Idx

xEB(O,R) JB(O,R)
= (211"r~eR19zl f IDr.p(x)ldx
iB(O,R)
Since for sEN the expression

274 CHAPTER 8. LEBESGUE'S lNTEGRAL, FOURIER TRANSFORM
consists of the finite sum of the expressions IZ1101 .. IZn lan,

we obtain that by the
preceding estimation there exists Cs > such that
Remark 8.38. 1. The opposite statement is also true. Therefore the following
theorem is true.
Theorem 8.13 (Paley-Wiener) Entire analytical function g is a Fourier trans-

form of a function from the space Cg"(Rn) with a support in B(O, R) if and only if
for every sEN there exists a constant C s > such that for every Z E en we have
Example 8.39 Let S be the set of all rapidly decreasing functions, i.e., f E S if f
has derivative of any order on Rand Jor every pair (k, I) E No we have
sup(1 + Ixl)k IJ{I)(x)1 :::;: Ck . (8.14)

xER
a) Prove that the Junction J defined by

(8.15)
belongs to S iJ J E S.
b) Prove that the Junction u = u( x, t) defined by
= -1 /00 dy /00 J(z)exp(zzy -zty2 -zxy) dz

u(x,t)
27r -00 -00 (8.16)
for every t E R belongs S.
c) Prove that the function u given by (8.16) is the solution of the Cauchy problem
Jor the Schrodinger equation
au a 2u
at = zax2 x,t) E R x (0,00)), u(x,O) = J(x) (x E R), (8.17)
where J E S.
d) Prove that if the functwn u is given by (8.16) then the function v(x, t)
u(x, T - t) is a solution of the following problem
av a 2v
at =-zax2' (xER,tE[O,T]) andv(x,T)=g(x) (xER), (8.18)
where g is a given function such that f = g.

e) Let WR, R > 1, be a function defined on R which has derivative of arbitmry or-
der such that wdx) = 1 for Ixl < R-1,wR(x) = 0 for Ixl > Rand Iw};l(x)1 ~
Ck, kEN, C k > 0 independently of R. Starting from the equality
(8.19)
prove that if u is a solution of the problem (8.16) and v a solution of the
1:
problem (8.14), then the following is true
1: f(x)v(x, 0) dx = u(x, T)g(x) dx. (8.20)
f) Prove that there is a unique solution of the problem (8.17) in the set of slowly
increasing functions u, i.e., there exist C> 0 and kEN such that for t > 0
Solution.
a) The condition (8.14) implies
-
If(y)1 ~ Ck,o
1 00
-00 (1 +dxIxl)2 < 00

and also absolute and uniform convergence of the following integral
Then by the property of Fourier transform we have J(ll(y) = F/(y) for every
lEN. Hence the function J given by (8.15) has derivative of arbitrary order.
Further, we have for y =I 0 and kEN
1 00
-00
e-'YXj(x) dx = (':)kl e-'YXj(kl(x) dx.
y -00
From this easily follows f E S.

Remark 8.39. 1. a) implies that the inverse Fourier transform is an inner
operation on S. It can be proved that it as well the direct Fourier transform
are isomorphism of the space S on itself.
b) Since we have
u(x,t) = - 1
27r
1 00
-00
f(y)exp(-zty2
- -zxy) dy,
we can prove the desired equality in an analogous way as for a).

c), d) Can be checked directly using a) and b).
e) Using the partial integratioll and the properties of the function WR we obtain
j R
-H
lT
0
au
v(x, t)WR(X)"!}(X, t) dtdx
ui
= j R WR(X)(U(X, T)v(x, T) - u(x, O)v(x, 0) - lT u(x, i)"!}(x,

ov
t) dt) dx,
-R 0 vt
and
(-z) [TjR v(X,t)WR(X)~2~(X,t)dxdt
Jo -R vX
T
-_ (_ z)l jR U (x, t )02(V(X,t)WR(X))
2
dX dt.
o -H AX
1:
Adding the last two equalities we obtain by (8.14)
WR(X)(U(X, t)v(x, t) - u(x,O)v(x, 0)) dx
R lT OV 02V oV
=j u(x, t)( ~(x, t)+Z'(-2 (x, t)wR(x)+2!:j(x, t)WR(X )+w" R(X ))) dtdx.
-R 0 vt ax vX
(8.21)
Since
lim WR(X)
R---+-oo
= 0, lim W'R(X)
R----.oo
= 0, lim W"R(X)
R-+oo
= 0,
uniformly on every compact set and v is a solution of the equation (8.18) the
left part in (8.21) converges as R ---+ 00 to
1: u(x,T)v(x,T) dx -1: u(x,O)v(x,O) dx

= 1: u(x, T)g(x) dx -1: f(x)v(x, 0) dx.
Since the right part in (8.21) converges to we obtain the equality (8.20).
f) For a given f E S the function U given by the integral (8.16) is a solution

of the problem (8.17). Suppose that an other slowly increasing function Ul is
also a solution of the problem (8.17). Then the function z defined by z(x, t) =
u( x, t) - Ur (x, t) is a solution of the problem
oZ 02 Z
at = Zox2 (x E R,t > 0), z(x,O) = 0.
If v is a solution of the problem (8.18), then we obtain by (8.20)
1: z(x, T)g(x) dx = for all T > 0, (8.22)

8.4. THE FOURTER TRANSFORM 277
and all functions 9 E S. The integral in (8.22) exists for all T > 0 since
the function z is slowly increasing and g rapidly decreasing, which implies
that there product is an integrable function. The equality implies z(x, t) ==
o for alll > O. Hence U = UI'
Example 8.40 Prove that the solution of the integral equation
U(x) = f(x) + 1: k(x - y)u(y) dy (8.23)
for given functions f and k, has the following representation ( if the expressions are
defined)
u(x) = f(x) + lax s(x - y)f'(t) dt, (8.24 )
where
(8.25 )
Solution. We apply the inverse Fourier transform F- 1 on (8.23)
F- 1 (u)( z) 1 foo
V'h _oo(J(x)+ foo -00 k(x-y)u(y)dy)e,zxdx
F-I(J)(z) 1 foo
+ V'h u(y) dy foo k(x - y)e- m dx
1: 1:
-00 -00
F-1(J)(z) +~ u(y) dy k(i)e,(YH)Z dt
F- I (J)(z) + y'2;F- I (u )F- I (k).

This implies
F- 1 (u)(z) = F-l(J)(Z)
1 - V'hF-l(J)(Z)'
Applying on the last equality the Fourier transform we obtain
u(x) = - -
1 foo F-I(f)(z)
e-'xz dz.
,j2; -00 1 - V'hF-l(J)(Z)
Using that f = F(F-I(J)) we obtain by the preceding equality
U(x) _ f(x) = _1_ foo ( F-1(J)(z) _ F- 1 (J)(z))e-' xz dz

V'h -00 1 - V'hF-l(J)(Z)
= foo F- 1 (J)(z) F-1(J)(z) e-' xz dz. (8.26)

-00 1 - V'hF-l(J)(z)
On the other hand, from the equality F- 1 (u * v) = v'27rF- 1 (u)F- 1 (v) we obtain
u * v = F(v'27rF- 1 (u). F- 1 (v)). Applying the last equality on (8.26) we obtain
u(x) = f(x) + 1: s(x - t)f(t) dt,
where the function s is given by (8.25).
Exercise 8.41 Let f E L2(R) and k E Ll(R). Prove that there exists a solution in
the space L2(R) of the integral equation
f(x) = 1: k(x - y)u(y) dy
if and only if
F- 1 (f)(z)
F-l(k)(z) E L2(R).
Hint. Using the procedure from Example 8.40 prove that
()
u x =
~2
VZ7r
1"" F-F-l(k)(z)
-00
1
(f)(z)
e
-,.:z d
z.
Chapter 9
Generalized Derivative and

Sobolev Spaces
9.1 Generalized Derivative

9.1.1 Preliminaries
Let a = (a},, an) E Z'+- and Q c Rn is a region.
Definition 9.1 A function f{OI) E L 2 (Q) is a-generalized derivative on the region
Q of the function I E L 2 (Q) if for every function r.p E cbal(Q) we have
k f(x)Darp(x) dx = (_l)la l k f{OI)(x)rp(x) dx.
For more general case, we can take the space Lroc( Q) instead of L 2 ( Q). Instead of the
notation f(OI) we are often using also the usual notations for the classical derivative,
alalf
such as Da I and a a, a OIn
Xl .. Xn
If IE Clal(Q) then there exists the generalized derivative I(a) and it is equal to the
classical derivative D(a) f.

Example 9.1 Prove that the generalized derivative 1(-1) of the function f given by
for X < 0,
f(x) ={ ~ for x;:::: 0,
on R is the function g given by
for x < 0,
g(x) ={ ~ for x;:::: 0.
279

280 CHAPTER 9. GENERALIZED DERIVATIVE AND SOBOLEV SPACES
Solution. For an arbitrary but fixed 0 such that
= 0 and d~~x} = 0 for every x 2': R. Therefore
<p( x)
j +OO f(x)-d-
-00
d<p(x)
x
dJ.; =
lR x-d<p(x)
0
IR
d-x dx = x<p(x) - 0
lR -
0
1 <p(x) dx =- lH-
0
<p(x) dx.
1: t?
On the other side we have
00
g(x)<p(X) dx = <p(X) dx.
The preceding two equalities imply
j +oo (d<P(X)
-00
-)
f(x)~ + g(x)<p(x) dx = 0
for every <p E CJ(R). This implies that the generalized derivative f{1} of f is equal 9
on R, although the classical derivative does not exists on R ( it exists on R \ {O}).
Example 9.2 Prove that if for a function f E L 2 ( Q) there exists the a- generalized
derivative f(OI} then it is unique and it is independent of the order of the differenti-
ation.
Solution. Suppose the contrary, i.e, that for f E L 2 ( Q) there are two a-derivatives
fi and fJOI}. By Definition 9.1 we have for an arbitrary but fixed <p E cb"I(Q)
Ol
}
kU }OI}(X) - fJOI}(X))' <p(x) dx = O.
Since fi Ol
} - fJOI} E L 2 (J{) for every compact subset of Q, we obtain by Example
8.16
II"} - IJ"} = 0
almost everywhere on J{. Since J( was an arbitrary compact subset of Q it follows
fi Ol } - fJ"} = 0 almost everywhere on Q.
The second part of example follows by the same property of functions from CbOlI(Q)
and Definition 9.1.
Example 9.3 Prove that if a function f E L 2 ( Q) has the generalized derivative

DOl f E L 2 (Q), then for every y E Q and for enough small h > 0 we have
where
/h(x) = k f(y)8 h ([x - y[) dy.
Moreover, for a compact subset J( of Q we have
lim liD" fh - D" f[[L 2 (K}

h-+O
= O.
9.1. GENERALIZED DERIVATNE 281
Solution. We have
h D" f(X)Oh(lx - yl) dx
(_1)1"1 hf(x)D~6h(lx - yl) dx
h f(x)D;Oh(lx - yl) dx.
We shall need the following consequence of Lebesgue theorem on convergence 8.3

(which we can easily prove).
Theorem 9.2 Let Q1 be a region of Rn and Q2 a region of Rm. If for some k ~ 0
the function F defined on Ql x Q2 belongs for almost every x E Q1 to C k (Q2) and
for some integrable function g defined on Q1 for every y E Q2
ID;F(x,y)1 ~ g(x) (Ial ~ k),
almost everywhere on Q1! then
By this theorem and the property 4) (section 8.2) of the function 6h we obtain
h f(x)D;6h(lx - yl) dx = D;fh(Y).

We shall prove now the second part of this exercise. Let K be a compact subset of
the region Q. Then there exists ho > 0 such that for every h ~ ho we have
(D" J)h(Y) = D cx h(Y) for every Y E I<.

By Example 8.15 we obtain
Remark 9.3.1.
(i) The preceding procedure can be applied also on an arbitrary subregion
Q' cc Q ( instead of K) such that we obtain
(ii) By the preceding remark (i) and the fact that every bounded subset of a
Hilbert space is weakly compact (see Example 10.1 ) it can be proved that
the generalized derivative Dcx f of a function f E L 2 (Q) exists if and only
if for every subregion Q' ceQ there exist constants c = c( Q') > 0 and
ho = ho(Q') > 0 such that liDo fhIlL2(QI) ~ c for every h < ho
Example 9.4 Prove that if all derivatives of the first order of a function f are zero,
then this function f is constant almost everywhere.
Solution. We shall use Exercises 9.3. We have for every subregion Q' Cc Q and
enough small h
(Ddh = 0 (i = 1,, n).
By Example 9.3 we have
Ddh = (Dd)h = 0 (i = 1,, ,n).
Hence !h is equal to a constant C = C(h). We have
(9.1 )
On the other hand the following inequality is true
&i. IC(h l ) - C(h2)1 IIC(h l ) - C(h 2)IIL,(QI)

< IIC(h l ) - fI1L2(QI) + II! - C(h 2 )IIL2(QI),
where IQ'I = r
lQI dx. By (9.1) the right part of the preceding inequality converges
to zero as hI, h2 ---+ O. Hence C(h) converges uniformly as h ---+ on Q' to some
constant. This means that f is equal to a constant on Q' and since Q' was arbitrary
it follows that f is equal to a constant on the whole Q.
Example 9.5 The function f(x) = sgnxi has no generalized derivative ~f on the
UXI
unit ball B(O, 1) = {x Ilxl < 1}.
Solution. Suppose the contrary, i.e, that there exists a function 9 E L 2 (B(O,1))
such that for every cp E CJ(B(O, 1))
r
lB(o,l)
g(x)cp(x) dx =- r
lB(o,l)
sgnxi a~(x) dx.
UXI
Hence
J g(x)cp(x) dx
B(O,])
J __
aCP(__
x)dx+
B(O,I)n{x I Xl >O}
aXI
J
B(O,I)n{x I xl <a}
2 J
B(O,l)n{x I XI=O}
Since the preceding equality holds for every cp E CJ(B(O, 1)), taking in a special
function cp which is zero on B(O, 1) n {x I x] < O} we obtain
J g(x)cp(x)dx=O,
B(O,I)n{x I XI>O}
9.1. GENERALIZED DERIVATNE 283
which implies 9 = 0 almost everywhere on B(O,l) n {x I Xl > O}. In an analogous

way we obtain that 9 = 0 almost everywhere on B(O, 1) n {x I Xl < O}. Therefore
9 = 0 almost everywhere on B(O, 1). Then we have
J <p( x) dX2 dXn = 0

B(O,I)n{x I Xt=O}
for every <p E CJ(B(O, 1)), which is impossible.

Example 9.6 Prove that the function f : B(O, 1) -t R defined by
f(x) = sgnxl + sgnx2
af af a 2f
has not the derivatives ~ and ~ on B(O,l), although there exists ~ on
UXI UX2 UXI UX2
B(O,l).
Solution. The non-existence of the derivatives ~f and ~f follows by Example

VXI UXI
9.5.
We shall prove that there exists the generalized derivative !l
UXIUX2
!
a2 on B(0,1).
We have that for every <p E C~(B(O, 1))
Hence there exists the generalized derivative
~ =
af2
0 on B(O,l).
VXI UX 2
This example shows the difference with respect the classical derivative for which it
can not happen that there exists the derivative of higher order although it does not
exist the derivatives of lower order.
Example 9.7 Prove that a function f E L 2 ( -1,1) has a generalized derivative if
and only if it is absolutely continuous and f' E L 2 ( -1,1).
Solution. Suppose that a function f E L 2 ( -1,1) has a generalized derivative

l' E L 2 ( -1,1),
i.e.,
II f(x)'f"(x) dx = -ll f'(x)'f'(x) dx for all 'f' E Gri[-1, 1J.

Applying the partial integration we get
j l
-I f(x)'f"(x)dx =
j l jf'(x)dx'f"(t)dt.
-1 -1
t_
Introducing a function u = u(t) in the following way
u(t) = f(t) - ill f'(x) dx (9.2)
we obtain u E L 2 ( -1,1) and
II u(t)'f"(t) dt = 0 for all 'f' E GM-I, IJ.

If we denote by F the class of all functions r.p E GM-I,lJ with the property
J~I 'f'(x) dx = 0, then such functions 'f' E F satisfy
II u(t)tp(x) dx = o.
F is dense in the subspace of L2(-1,1) consisting of the functions g E L2(-1,1)
such that J21 g(t) dt = o. Hence u is almost everywhere constant. Therefore by
(9.2) the function f is equal to the function t ~ J21 f'(x) dx almost everywhere
up to a constant. Hence f is absolutely continuous function, since f is absolutely
continuous if and only if
f(t) - f( -1) = II f'(x) dx.

Using now the preceding procedure in the opposite direction it can be easily proved
that each absolutely continuous function f with f' E L 2 ( -1,1) has a generalized
derivative ( Starting from (9.2) we have that u is absolutely continuous and u(t) = 0
almost everywhere on [-1,1]).
Remark 9.7.1. It is obvious that in the preceding example we can take any finite
interval [a,bJ instead of the interval [-1, 1J. The supposition of the existence of the
derivative f' does not imply the absolutely continuity of the function f (we have
the both conditions in Example 9.7). For example the function f given by
f(x) = { 0x 2 sinx- 2 for 0 < x < 1,

for x = 0
has the derivative on the interval [O,lJ but f' f/. LIfO, 1J and so f' f/. L 2 [0, 1J. Even if
for some function f we have f' E LIfO, 1J this does not imply always the absolutely
continuity of r
9.2. SOBOLEV SPACES 285
Exercise 9.8 Prove that for the function
for x 2 + y2 < 1, y > 0

f(x,y) ={; for x 2 + y2 < I, y < 0,
there exist the generalized derivatives of the first order on both semidises, for y > 0
and y < 0, but for a =I- b there is no generalized derivative on the whole disc with
respect to y.
9.2 Sobolev Spaces

9.2.1 Preliminaries
Definition 9.3 Let k be a non-negative integer. Sobolev space Wk( Q) consists of
all Junctions J E L 2(Q) which have all generalized derivatives D" J Jar 0:; lal :; k
and they belong to L 2 (Q) equipped with a sealar product (see Section 10.1)
The scalar product (Jlg)wk(Q) induces the norm
IIJIIWk(Q) =(2: lIDaJI2dx)~.

lal:Sk Q
Theorem 9.4 IJ Q is a bounded region, then
( k -- 0 " 1 ').... , ... ) ,
where the closure of Ck(Q) is taken with respect to the norm II IIwk(Q).
ok
Definition 9.5 Sobolev space W (Q) is defined by
ok --
W (Q) = cg(Q) (k = 0,1,2, ... ),
where the closure ofCk(Q) is taken with respect to the norm 1IlIwk(Q).
Theorem 9.6 ( Sobolev lemma) Let Q be a bounded region with fJQ E C k. IJ

ok _
k> s + ~,s E N U {O}, and JEW (Q), then J E C5(Q).
Theorem 9.7 (Rellich) IfQ is a bounded region oJRn andm > k (m,k E Z+),
om 0 k
then the embedding map of the space W (Q) in the space W (Q) is a compact
operator.

Exercise 9.9 Find for which a the given function f belongs to the given Sobolev
space
a) f(x) = lxi-a sin lxi, space W2(B(0, 1)), where B(0,1) is the unit disc in R2
with center in (0,0) and radius 1;
= racp, the space W2(B(0,1)) for x = =

: :;
b) f(x) (XI,X2) (rcoscp,rsincp) for
cp < 27r
Answer. a) a < 0.
Exercise 9.10 A Junction f E L 2 ( -7r, 7r) belongs to the space WI( -7r, 7r) if and
only if the series
00
L n2(a~ + b~)
n=O
converges, where
an = ~jlr f(x)cosnx dx, bn = ~jlr f(x)sinnx dx forn = 0,1,2, ....

7r -lr 7r -lr
Exercise 9.11 For f E WI (-7r, 7r) we have
"
IIfIIWl(-lr,lr) = j_)P(x) + r(x)) dx = 7r ;(n 2 + l)(a~ + b~) + 7r( 2)2.
00 a
Example 9.12 Prove that the Sobolev space Wk(Q),k:::: 0, is a Hilbert space (see
Section 10.1).
Solution. It is easy to check the properties of the scalar product for the bilinear
1
functional
Ulg)wk(Q) = L
D a f Dag dx.
lal::ok Q
We shall prove the completeness of the space Wk(Q) endowed with the induced
norm
IIfIlWk(Q) = (L
IDa f(xW dx)t. 1
lal::ok Q
Let {Jm}mEN be a Cauchy sequence from Wk(Q), i.e.,
Ilfs-fmll~k(Q)= L lIDafs(x)-Dafm(XWdX-->O (9.3)

lal::ok Q
as s, m --> (Xl. This implies that for every fixed a, lal : : ; k,

as s, m -+ 00 and so also specially for a =0

10 Ifs(x) - fm(xW dx -+ 0
as s, m -+ 00. Since the space L 2 ( Q) is complete we obtain by the last relation

that there exists f E L 2 ( Q) such that fm ~) 1 as m -+ 00. We obtain in an
analogous way by (9.3) that for every a, lal :::; k, there exists f(C>!) E L 2 (Q) such
that DC>! 1m L~) f(C>!) as m -+ 00. Since DQ fm E L 2 (Q) (Ial :::; k, mEN), we have
for every cp E C~(Q). Letting m -+ 00 we obtain using the Lebesgue convergence

theorem
(JIDC>!cp)L2(Q) = (-1)/C>!/(J(<lcp)L2(Q)
The last equality shows that f(Q) is the a-generalized derivative of f and so f E
Wk(Q). It is easy to check that
IIfm - fllwk(Q) -+ 0 as m -+ 00.
Exercise 9.13 Prove that the function f(x) = Ixl belongs to the space Wl(-l,l)
but not W2( -1,1).
Example 9.14 Suppose that for a bounded region Q and A> 0 the region
is starshaped. For A > 1 and 1 E Wk( Q) we introduce a function f>. by

x
f>.(x) = f(),l (x E Q(A)).
Prove that
1 x
a) (Dcxf>.)(x) = ~(DcxJ)(-:\), which implies f>. E Wk(Q(A));
lim 11f>. - fllwk(Q) =

b) .\ .... 1+0
o.
Remark 9.14.1. A bounded region Q is starshaped if there exists Xo E Q such that
for every A > 1 the set
{ x IXo + -A-
x - Xo
E Q
}
is a subset of the set Q.

Solution.
a) We have for cP E Cgo(Q(A))
r
lQP,)
D" f>..(x)cp(x) dx (-1)1"1 r
lQp)
f(~)D"cp(x)dx
A
(-1)1<>110 f(z)(D"cp)(AZ) dz
(~~lll"l 10 f(z)D;(cp(Az))An dz
A~al 10 DC> f(Z)(cp(AZ))An dz
1 r x-
~ lQp..) D" f().. )cp(x) dx.
b) First we shall show that lim), ..... l+o II!>. - fIIL 2(Q) = O. Co(Q) is dense in L 2 (Q).
Therefore for every E > 0 there exists a function g E C o( Q) such that II f - gil <
E. Using the substitution in the integral we have
Therefore
11f>. - fliL2(Q) -:; II!>. - g),IIL 2(Q(),)) + Ilg)' - gliL2(Q) + Ilg - fIIL 2(Q(),
-:; E(l + A~) + Ilg), - gIIL2(Q)
Taking A -+ 1 + 0 we obtain lim),..... l+O IliA - fIIL 2(Q) = 0, since
lim sup Ig),(x) - g(x)1 = O.

), ..... 1+0 xEQ
Hence lim),..... l+O Ilg), - gIIL2(Q) = O.

In a quite analogous way we can prove that
lim
), ..... 1+0
liD"!>. - DO fIIL 2 (Q) = O.
Finally by the definition of the norm in the space Wk(Q) it follows
lim
),-->1+0
II!>. - fllwk(Q) = O.
Exercise 9.15 Let Q be a starshaped locally quadratic bounded region. Iff E C 1 (Q)
o 1
and fl8Q = 0, then fEW (Q).
Hints. Take the function f>. from Exercises 9.14 for >., 0 < >. < 1. Then Q(>.) C Q.
Take f>.(x) = 0 for x E Q \ Q(>.). Hence f>. E C 1 (Q(>.)) and 1>, E C 1 (Q \ Q(>.)). Use
Example 9.14.
Example 9.16 Let Q be a bounded region in Rn and the function <.Ii = (<.IiI'' <.lin)
is a C k - dipheomorphism from Q onto IT, where 0 = <.Ii (Q).
If f E Wk(Q), then f 0 <.Ii-I E Wk(O) and there exists C > 0 such that
Ilf 0 <.Ii- 1 1Iwk(fl) ::; Cllfllwk(o).
Solution. C k(Q) is dense in the space Wk( Q). Therefore for f E Wk( Q) there exists
a sequence {fdiEN of functions from Ck(Q) such that limi..... oo II/; - fllwk(Q) = O.
Hence {li}iEN is a Cauchy sequence in Wk(Q). Since fi 0 <.Ii-I E Ck(IT) (i EN) we
can apply the classical Leibniz rule on the product uv
D"'(uv) = L ap(D"'fjP)(DPv ).
P$.'"
Therefore by compactness of Q the sequence {Ii 0 <.Ii-I }iEN is a Cauchy sequence in

the space Wk(O). By completeness of the space Wk(Q) the sequence {/; 0 <.Ii-I hEN
is convergent in Wk(O). Hence there exists g E Wk(O) such that
Jim
...... 00
II/; 0 <.Ii-I - gllwk(fl) = O. (9.4)
Using the rule for changing the variable in a Lebesgue integral we have that if
fE L 2 (Q) then f 0 <.Ii-I E L 2 (0) and so
Jim
...... 00
II/; 0 <.Ii-I - f 0 "<.Ii- I IIL2 (fl) = o.
By (9.4) we have 9 = f 0 <.Ii-I almost everywhere and so f 0 <.Ii-I E Wk(O).
Since fi E Ck(Q), /; 0 <.Ii-I E Ck(IT) and using the boundedness of the functions <.Iii
and the chain rule we obtain that there exists C > 0 such that
Hence by (9.4) we obtain the desired inequality.
Example 9.17 The system of functions

"P .X}PEZ+
{(21r)-'n e";;"'
for f3 . x = f3IXI + ... + f3nxn is a complete orthonormal system of functions (see

Section 10.1) in the space L 2 (Pa ), where
Pa = {xllxil < a, i = 1, ... ,n} and a > o.

Solution. Since {(27r)-lel~XikhEZ+ is a complete orthonormal system of functions

in the space L 2 ( -a, a) (one-dimensional case), we obtain the desired conclusion by
example from Chapter 8. on the product of orthonormal systems.
Example 9.18 Let j E Cg:'(Rn) be such a function thai
suppj C P = {xllx;1 < 7r,i = l, ... ,n}.

Prove that the Fourier series of the function j wilh respect to the orthonormal system
uniformly converges to J. Moreover,prove that for any arbitrary but fixed multi-index
0'applying the differential operator DO on this Fourier series on every summand the
new series converges uniformly on P to DOt J.
Solution. By Example 9.17 the system {'0,6} ,6EZ+ is complete and orthonormal.
Therefore every function f E Cgo(P) c L 2 (P) has a representation with a Fourier
series convergent in L 2 ( P)
f = a ,6'0,6. "2:.
,6
We shall show that the series L,6 a,6DO:'0,6 converges absolutely and uniformly. Since
where (3Ot = (3f' ... (3~n, we have
Hence it is enough to prove that the series L,6la,61 . 1(30:1 is convergent. For that
purpose we shall give an estimation for the Fourier coefficients a,6. Applying the
partial integration with respect to Xk we obtain for (3k =I- 0
a,6 = (27rt- J
p
f(x)e-,6x dx = (27rt- J
p
Dd(x)(z(3kt 1e -,6x dx.
Let I E Z~. Applying the preceding procedure lk-times with respect to the variable
Xk, k = 1,, n, and the convention "0 = 1" we obtain
This equality implies for every I E Z~

Taking a. special multi-index I = (,1,' .. , In)
for (3k = 0
for (3k :/:- 0,
and using the last inequality we obtain that there exists a constant (with respect to
(3) C > 0 such that
(9.5)
Since we have
L{3
and the series on the right side converge we obtain that the series
also converges. Hence by (9.5) the series E{3la{3II(3"'1 converges, too.

Remark 9.18.1. We can prove the same statement in a quite analogous way for
a more general case taking Pa = {xl IXil < a, i = 1"", n} instead of P and the
system
t
{(21l' ~ e~i{3.x} (3EZ+.
Exercise 9.19 Prove that for every f E Wk(Q) and every subregion Q' cc Q we
have
lim IIfh - fIlWk(QI) = O.
h-+O+O
Moreover if the function f is complementary finite in Q, then
lim
h-+O+O
11th - fllwk(Q) = O.
Example 9.20 Prove that the space Wk(P), where P = {xllxil < 1l', i = 1,' ,n}
is a cube in Rn, is separable, i.e., it has a countable dense subset.
Solution. By Exercises 9.19 every function f E Wk(P) can be approximated

by functions fh from the set Cgo(P) with respect to the Wk(P)_ norm. Every
function fh for enough small h can be represented by Exercises 9.18 by its uniformly
convergent Fourier series (and so convergent also in Wk(P)) whose derivatives are
also uniformly convergent. Therefore the function fh can be approximated with
respect to Wk(P)-norm by partial sums of the corresponding Fourier series, i.e.,
fh can be approximated with respect to Wk(P)-norm by linear combinations of

t
the system {(271" ~ ei{J.x} and with coefficients whose real and imaginary parts are
rational numbers. Therefore the countable set of all such linear combinations with
rational coefficients of the system {(271"t~ei,6.xL3EZ+ is dense in the space Wk(P).
Example 9.21 Let ai > 0, i = 1, ., n,

p = {x I x ERn, ai < Xi < 2ai, i = 1, .. , n},
P' = {xl x E Rn,O < Xi < 3ai,i = 1, ... ,n}.

Prove that for every function f from the space Wk(P) there exists its extension
F E WI (P') such that
IIFIIW 1(pl) :::; Cllfllw 1(p),
where the constant C is independent of f. Specially, if f E C(P) then FE C(P').
Solution. First we shall consider the special case f E CI(P). We shall extend this
function on P' and denote this extension by F such that it will be symmetric in all
coordinates i = 1, .. ,n with respect to ai and 2ai (i = 1, .. ,n), i.e., F(x') = f(x)
where the mapping t : P' --t P is defined by
for 0 :::; x: :::; a;

for ai < x: :::; 2ai
for 2a; < x; :::; 3ai.
Since the function t continuously maps P' onto P we have F 0 t E C(P,). The
function F and the translation function for (C1, . .. ,cn ), where Ci is either 0 or ai or
-ai have continuous derivatives on P. Since these derivatives of first order coincide
with the generalized derivatives of first order we obtain FE WI(P'). Since
JIF(xW dx = 3n JIf(xW dx
p p
JID;F(xW dx = 3n JID;J(xW dx (i =
and
1, ... , n)
pi p
we obtain
IIFllw1(pl) = 3nllfll~1(P)'
i.e., the desired inequality reduces on the equality for this special case.
We shall show now that the inequality is true for the general case, i.e., for f E
W1(P). We remark that the map T: Cl(P) --t Wl(P') given by f I--t F is a linear
and bounded operator on a dense subset C1(P). Then we can extend the operator
T on the whole space WI (P) ill a way that it remains to be bounded. We denote
this extension also by 1'. Then for every I E WI (P) and a sequence {Ij LEN from
- WI(?) Wl(l'/)
C 1 (P) with the property Ij ---t I we have 1'(Jj} ---t 1'(J) as j ---t 00. Since the
restriction of T(Jj) on P is equal Ii it follows that the restriction of T(J) on P is
almost everywhere equaJ I and that the extension T(J) is obt.ained in the same way
as for the case I E C 1 (P) (excluding a set of a measure zero).
Remark 9.21.1. Exercise 9.20 can be used in the proof of the following general-
ization of Exercise 9.20.
Theorem 9.8 If Q and Q' are bounded r'cgions such that Q cc Q' and 8Q E Ck,
then for every function f E Wk(Q) there exists a finite extension F E Wk(Q') such
that
1!PIIWk(Q/) :::; Cllfllwk(Q)'
wherc the constant C > 0 depends only from Q and Q'.
Example 9.22 Prove that the set C(f'CQ) is a dense subset of the space Wk(Q),
where Q is a bounded region whose boundary 8Q belongs to C k. This implies that
Ck(Q) is also a dense set in the space Wk(Q), i.e.} Ck(Q) = Wk(Q).
Solution. Let 0 be a region with Q cc O. By Exercise 9.21 and corresponding
Remark for every function f E Wk(Q) there exists its extension F on 0 which
belongs to the space Wk(O). Then by Exercise 9.19 we have
h-O
lim l!Ph - fllwk(Q) = h_O
lim l!Ph - Fllwk(Q) = 0,
J
where
Fh(X) = F(y)bh(lx - yl) dy.
Q
Since Fh E C(f'( Q), we have proved the desired statement.

a k
Example 9.23 Let I EW (Q) and Q is a subregion of the region Q'. Prove that
the extension by zero of the function f on Q' given by I(x) = 0 for x E Q' \ Q
ok
belongs to W (Q').
a k
Solution. By the definition of the space W (Q) there exists a sequence {<pj LEN
from C(f'(Q) such that
lim lI<pj - III k
}~OO
= O. 0
W (Q)
Since for every function <pj its extension by zero on Q' belongs to C:f( Q') and
o k
{<pj LEN is a Cauchy sequence in the space W (Q') we obtain
lim
}-OO
II<pj - fll 0 k
W (Q/)
= O.
ok
Hence f EW (Q').
o k
Example 9.24 Prove that if a function fEW (Q) is extended by zero on the whole
Rn, then for every multi-index a, lal ~ k, we have
a) DO: j E L2(Rn);
b) Do f(y) = (zy)O j(y);
c) IIfIlW k (Q) = l LI<>ISk ly<>1 2 Ij(y)1 2 dy.
Solution.
ok
a) It is true for f E C~(Q), but C~(Q) is dense in W (Q).
ok
b) By Example 9.23 the extension by zero of a function fEW (Q) on Rn belongs
ok
to W (Rn) C L2(Rn). Therefore we have for every cp E GO'(Rn)
(Do flcp)L2(Rn) = (-1 )1<>1(}IDOcp )L2(Rn) = J

p
(_i)lol y<> j(y )cp(y) dy.
Since by Example 9.22 Go(Rn) is dense in L 2 (Rn) we obtain that C8"(Rn) is

also dense in L2(Rn) ( f 1-4 j is an isometrically isomorphism). Hence b).
ok
c) Since W (Q) is a Hilbert space we obtain c) by Parseval's identity.
Exercise 9.25 For which sequences {an}nEN the function f given by
L
00
f(x,y) = ane-nYsinnx
n=l
forO ~ x ~ 7r andy> 0 belongs to the space W 1 ({(x,y): 0 < x < 7r,y > O})?
Answer. The desired sequence {an} nEN has to satisfy the condition that the series
f
n=l
nla n l2 converges.
Example 9.26 Prove that for every function f E GI[a, b]
J J
b b
If(aW ~ b~ a If(xW dx + (b - a) If'(xW dx.

a a
Solution. We have by Newton-Leibniz formula for f E CI[a, b]
J1'(z) dz.
x
f(x) - f(a) =
a
Applying the Cauchy-Schwartz inequality we obtain
J1'(z) dz l)2
x
lJ(aW < (If(x)1 +1

a
2lf(xW + 21 JI'(z) dzl

x
< 2
2lf(xW + 2(x - a) J1f'(zW dz

x
<
a
2lf(xW + 2(x - a) JII'(zW dz.

b
<
a
Integrating over the interval [a, b] we obtain
JIf(xW dx + (b - a)2 J1f'(zW dz,

b b
(b - a)lf(a)12 ::; 2
a a
which implies the desired inequality.
Example 9.27 Let P be the n-dimensional parallelepiped
with sides
for i = I, ... ,n,

Pn = (al1bl) x X (an-I,bn- 1 ) X {an}.
Prove that for f E Cl(P)
a)
JIfIi ::; b. ~ a JIf(xW dx + (bi - ail JIDJ(xW dx;

Pi P P
b) for a n-dimensional parallelepiped P, whose edges have the length

E( bi - ai) (i = 1, ... , n), 0 < E < 1, we have
JIfl2 ~ ~l J
BP P
If(xW dx + C2E t,=1 JIDJ(xW dx,
P
where C 1 > 0 and C 2 > 0 are constants.
Solution.
a) Using the same procedure as In Exercise 9.26 only starting now from the
equality
J
Xi
= DJ(Xh, Xi-I, Zi, Xi+l,, Xn) dZi

a,
we obtain the inequality
J
bi
~ 21f(xW + 2(Xi - ai) IDJ(Xl, ... , Xi-I' Zi, Xi+l, ., XnW dzi
ai
Integrating with respect to X on P we obtain the desired inequality.
b) Applying the inequality from a) on n-dimensional parallelepiped P, instead of

P, holding the side Pi we obtain
jlf12 < E(b i ~ ai) / If(xW dx + E(bi - a;) / IDJ(xW dx
< E(b
,
~ a) PJIf(xW dx + E(bi - a;) J
P
IDJ(xW dx
for i = 1, ... ,n. Summing up all n inequalities we obtain
JIfl2 ~ ~l J
BP P
If(xW dx + C2E t,=1 JIDJ(xW dx.
P
Example 9.28 (The trace of a function) Let Q ~ Rn (n 2: 2) be a bounded

region and S ~ Q a locally quadratic (n - I)-dimensional surface with the property
that for every Xo E S there exist regions U(xo) and V and C1-dipheomorphism it>
from U onto V such that some subset of Q n U is mapped on some n-dimensional
parallelepiped P C V, and S n U is mapped on a side or union of sides of P. Then
there exist constants C 1 > 0 and C 2 > 0 such that for every function f E Cl(Q) and
10,0 < 10 < 1 we have
JIJI 2dS ~ ~1 JIf(xW dx + e t JID;J(xW dx.

s Q
C2
.=I Q
(9.6)
The preceding inequality (9.6) can be written also in the following form
1
(/1f12 dS) 2" ~ C 1Ifllw

3 1 (Q)
Jor some constant C3 > O.
Remark 9.28.1. If we define the operator T: C 1 (Q) -> L 2 (S) in the following way
T(f) = floQ
then by Exercise 9.28 there exists a unique linear and bounded extension l' : WI ->
L 2 (S) of T. The function T(f) E L 2 (S) is the trace of the functionf E W 1 (Q) on
the surface S, which we will denote by fls.
Solution. Since S is a compact set we can cover it with a finite number of regions
Ui which Cl-dipheomorhisms CPi map on the regions Vi, respectively, and a subset
of Q n Ui is mapped on n-dimensional parallelepiped Pi C Vi, and S n Ui is mapped
on a side or union of sides of the parallelepiped Pi. Let 'Pi E Cgo(Ui ), 2:: 'Pi = 1 be
the partition of the unit in a neighborhood of S. By Exercise 9.27.b) we have for
f E C1(Q)
Js Ifl2dS t J
.=l<Pi (snu;)
('PilfI 2 0 cp;-I)IJi l dx
< c 4 t J
.=l<Pi(SnU;)
Ifocp;-11 2 dx
< t, (~' / If <>,' I' dx + G,E t.j ID;(f <>,' )1' dX)
0 0
t, (0,' / If <>,' I' dx + G,E t./ I <>i' <1>,),' I' dX)

0 f,(D,f 0 )D;(
< t, (~' Q1, IfI' dx + G,E t.Q1. ID;iI' dX)

< C;s JIfl2 dx + CSSe t JIDjfl2 dx.
Q J=I Q
298 CIlAPTER 9. GENERALIZED DEIU"ATIVE AND SOBOLE" SPACES
Remark 9.28.2. The trace of a function f E W1( Q) we can interpret also in the
following way. Since C 1 (Q) is dense in the space W 1 (Q) there exists a sequence
{I;1iEN from C 1 (Q) such that limj~oo III; - fIIH!I(Q) = o. Then by (9.6)
Ilfk - fmll2(S) 5. C1 11fk - fmIIWI(Q) (k, Tn EN).

Hence {I; LEN is a Cauchy sequence in the space L2(S). Since the space L2(S) is
complete there exists fs E L2(S) such that
lim
) .... 00
III; - fsll2(s) = o.
The function f 5 is the trace of the function f E W 1 (Q) on the surface S.
For a continuous function f on Q its trace on a surface S c Q is a continuous
function from C(Q) on S which coincides almost everywhere with f.
ok
Example 9.29 Prove that Wk(Q) #W (Q), k 2: 1, for some region Q # Rn.
ok
Solution. By Exercise 9.28 the trace flaQ of a function fEW (Q) is zero (Remark
9.28.2). Taking the continuous function equal 1 on Q we have that it belongs
ok
Wk(Q), k 2: 1, but since it trace on 8Q is 1 it does not belong to W (Q).
ok
Remark 9.29.1. A function from Wk(Q) belongs to W (Q) if and only if its trace
on the border is zero.
Remark 9.29.2. If Q is a locally quadratic bounded region and f E W 1 (Q x (O,T))
is a function whose trace on Q x [0, T] is zero, i.e., flaQx[o,11 = 0, then there exists
a sequence UdkEN from C1(Q x (0, T)) such that
WI(Qx(O,T))
fklaQx[O,Tj = 0 and fk ~ f as k ---> <Xl.
Example 9.30 Prove that

a) for f E CJ(P), P = (all br) x ... x (an, bn ), there exists a constant C > 0 such
that
JIf(xW dx 5. C Jt
p P )=1
IDjf(xW dx;
o 1
b) the preceding inequality holds also for every fEW (P):
o 1
c) for every bounded region Q C Rn and fEW (Q) there exists a constant
C > 0 such that
JIf(xW dx 5. Jt
Q
C
Q )=1
IDif(xW dx;
o 1
d) the usual norm in the space HI (Q) and the norm
1/2
(
II fill = / ~ IDj/(xW dx )
a1'e equivalent.
Solution.
a) By the Newton-Leibniz formula

XI
f(Xl, x') = J Dd(zI, x') dzl ,

al
where x' = (X2, ... , xn). Then by Cauchy-Schwartz inequality we obtain

Xl Xl bl
If(Xl, x'W :::; J IDd(zl, x')1 dZl J dYl :::; (Xl - ad J Ddh, x') dzl
a] al al
Therefore
XI
Jlf(xl,x'Wdxldx' < J(xl-al)dxl

P al
b2 bn bl
J . .. J (J IDd(zl, x'W dZ 1 )dX2 ... dX n
(bl ~ al)2 J IDd(xW dx.

P
o 1
b) Since Ca (P) is dense in the space HI (P) there exists a sequence {lj LEN from
0 1
Ca(P) such that fk H::J!;) f as k -+ 00. By a) we have
J Ih(xW dx :::; C 2 J
P
t
3=1
IDjfk(xW dx.
Since for each fixed j, j = 1, ... , n, we have
taking in the preceding inequality k -+ 00 we obtain the desired inequality.

300 CHAPTER 9. GENERALIZED J)l!;HIVATIVE AND SOBOLEV SPACES
c) Let P = (aI, bd X ... X (an, iJn ) be a n-dimensional parallelepiped with the

o I
property that Q c P We extend the function lEW (Q) on P in such a way
that it is zero almost everywhere on P \ Q. We denote this extension a.lso by
J. Since
1II(xW dx = 1II(xW dx 1t
P Q
and
P J=1
IDjl(xW dx = 1t
Q J=1
IDd(xW dx,
we obtain by b) the desired inequality.

o 1
d) By c) we have for lEW (Q)
Since trivially
(1t IDd(xW dX)

Q j=l
1/2 :s 11/11 01 ,
W (Q)
we have proved the equivalence of the norms 111110W 1

(Q)
and 11/1h.
Exercise 9.31 Prove that lor IE Wk(R)

lim I(x) = lim I(x).
x-++oo x-+-oo
Hints. Suppose the opposite of the statement. Take a sequence {/dkEN from
C1(R) such that Ik W~) I as k -+ 00 and use
1I~(x) dx.
b
Ik(b) - Ik(a) =
a
Exercise 9.32 Prove that lor I E W 1 ([a, bJ)
1J'(x) dx.
b
I(b) - I(a) =
a
Hint. Use Exercise 9.31.
Exercise 9.33 Does there exist any lunction I E C[a, b] such that I (j. W 1 ([a, b])?
Exercise 9.34 Prove that for every real function f E Wl([O,27r])
7
o
f(x? dx ~ 7f'(X)2 dx + (7 J(x) d:r;)
0 0
2
o 1
Exercise 9.35 Prove that for every real function fEW ([O,7r])
Jf(X)2 dx ~ Jf'(X)2 dx.

". ".
o 0
Exercise 9.36 Let Wo be a subspace ofWl([O,21r]) which contains functions f with

the property
J
2".
f(x)dx = O.
o
Prove that the scalar product in the space Wo has the form
Jf'(x)g'(x) dx
211"
(f,g E Wo).
o
Example 9.37 (Poincare inequality) Prove that for an n-dimensional paralle-

lepiped P = (al>b 1 ) X ... x (an,bn ) and f E Wl(P) we have
Solution. Each function f E W1 (P) can be approximated in the space W 1 (P)

by a sequence {fm}mEN of functions from Cl(P), i.e., fm ~) f as m -+ 00. By
Exercise from Chapter 4. for fm holds the Poincare inequality
Ifm(x)12 dx < (b 1 _ al)'~. (b n _ f

an) I fm(x) dxl 2
+ i Jt(bi - ai)2IDdm(xW dx.

p .=1
Therefore we have
II/ fm(x) dxl- 1/ f(x) dX11 < I Jfm(x) dx - Jf(x) dxl

p p
< JIfm(x) - f(x)1 dx

p
< (J Ifm(x) - f(xWdx)~. Jdx

p p
( b1 - ad' .. (b n - an) . (J If
p
m ( x) - f ( X Wdx) L
ok
Exercise 9.38 (Sobolev lemma for W (Q)) Let Q be a bounded region. Pr'ove
ok _
that for k > s+~, s E Nu {O}, every function f from W (Q) belongs also to GS(Q).
Hints. Prove first that for every f from GO' (Q) there exists G > 0 such that such
that
sup If(x)1 ~ Gllfll ok (9.7)
xEQ W (Q)
This inequality (9.7) will imply
supIDaf(x)1 ~ G11Daf11 ok
xEQ W (Q)
ok _
for lal ~ s, and s < k-~. Therefore the embedding of W (Q) in GS( Q) is continuous
on a dense subspace GO'(Q).
Exercise 9.39 Prove that
(k E N U {O})
and that the embedding is continuous.
Hint. Use Sobolev lemma 9.6.
Exercise 9.40 Prove the formula for partial integration for f, 9 E Wl( ( a, b))
b b
J f'(x)g(x) dx
a
flbglb - flagla - J f(x)g'(x) dx
a
b
f(b)g(b) - f(a)g(a) - J f(x)g'(x) dx.
a
Hints. Use Example 9.28, since by Exercise 9.39 f(a) = fla.

Chapter 10
Some Elements from Functional

Analysis
10.1 Hilbert Space

10.1.1 Preliminaries
Definition 10.1 A Hilbert space is a vector space (real or complex) H endowed
with a scalar product (1), i.e., (real or complex) valued bilinear functional defined
on H x H with the properties for all x, y, z E H :
(h1) (axly) = a(xly) for every scalar a;
(h2) (x + ylz) = (xlz) + (ylz);

(h3) (xly) = (ylx);
(h4) (xix) > 0 for x =I- 0;
(hS) (xix) = 0 for x = 0;

and H is a complete metric space with respect to the metric IIx - yll induced by the
norm IIxil = [WJ.
A subset B of H is bounded if and only if for every sequence {xn }nEN from
B and every sequence of numbers {O::n}nEN which converges to zero, the sequence
{O::nxn}nEN converges to zero.
Let HI and H2 be Hilbert spaces.
Definition 10.2 A linear operator T : D(T) -+ H 2 , D(T) C HI, is bounded if there

exists M > 0 such that
(x E D(T)).
303

304 CHAPTER 10. SOME ELEMENTS FROM FUNCTIONAL ANALYSIS
We denote by L(H1' H 2 ) the vector space of all bounded linear operators from HI
into /{2 endowed with the norm 111'11 = sUPllxlllf, Sl 111'( x) II1l2'
Theorem 10.3 (Riesz representation) For every continuous linear functional f

on a Hilbert space H there exists a unique element Yf E H such that
f(x) = (xIYf) (x E H),

and IIfll = IIYfliH.
Definition 10.4 An orthonormal family {Ci} ieI in a Hilbert space H is an orthonor-
mal base in H if for every x E H the (Fourier) series
l)xlci)ci
ieI
converges to x, where in the sum there are not more than countably many nonzero
members.
We have for x E H the Parseval identity (for a countable base)

00
1: l(xlciW = Ilxll 2
i=1
If H is separable, i.e., has a dense countable subset, then a linear operator l' : H -+
H has a matrix representation [tij]i,jeN in the following way
00 00
Tx = 1:(1: tijX;)Cj,
j=1 i=1
00
where x = 1: XkCk Then the adjoint operator 1'* has a matrix representation
k=1
[tj;)i,ieN.
A sequence {Xn}neN from a Hilbert space H weakly converges to an element
x E H if
lim (xnly) = (xly)
n-+oo (y E H).
Definition 10.5 A linear operator l' : D(T) -+ H2 , D(T) c HI, is closed if its
graph
G(T) = {(x, T(x)) I x E D(T)}
is a closed set in HI x /{2 with respect to the topology induced by the norm
II(x,y)11 = /ll xll1I, + Ilyll1I 2,
which is induced by the scalar product

10.1. HILBERT SPACE 305
We have the following characterization of the closed operators.
Theorem 10.6 A linear operator T : D(T) --t H 2 , D(T) C HI, is closed iJ and only
iJ Jor every sequence {xn}nEN Jrom D(T) with the property that iJ it converges to x
and T( Xn) --t y as n --t 00, then x E D(T) and T( x) = y.
Definition 10.7 Let T be a linear operator T : D(T) --t I1 2 , where D(T) is a dense
subspace oj HI. Then the adjoint operator T* oj the operator T has the domain
D(T*) = {y E H2 I there exists t E HI, (yIT(x))H2 = (tIX)Hl (x E D(T))}
and T* : D(T*) --t HI is defined by T*(y) = t.

Theorem 10.8 (Closed Graph Theorem on Hilbert spaces) Let HI and H2
be Hilbert spaces. IJ T : HI --t H2 is a linear closed operator, then T is bounded.
Theorem 10.9 (Uniform Boundedness Theorem) Let A be a family of addi-

tive and continuous operators from a Hilbert space HI into a Hilbert space H 2 If the
family A is pointwise bounded on HI, i.e., for x E HI there exists M(x) > 0 such
that
IIA(x)11 :S M(x) (A E A),
then it is also uniformly bounded on every bounded subset B of HI, i. e., there exists
M > 0 such that
IIA(x)11 :S M (x E B, A E A).
Definition 10.10 A linear operator T : HI --t H2 is compact iJ T(B), for every

bounded subset B of X, is a subset of some compact subset of H2

Example 10.1 Prove that every bOllnded sequence in a Hilbert space has a weakly
convergent subsequence.
Solution. Let {Xn}nEN be a bounded sequence in a Hilbert space H, i.e., there exists
M > 0 such that Ilxnll < M (n EN). Let L({xn}) be a closed subspace spanned by
the sequence {Xn}nEN. The sequence of numbers {(xllxn)}nEN is bounded, since
Therefore by Bolzano-Weierstrass theorem there exists a convergent subsequence

{(xllx~)}nEN. The sequence {(x2Ix~)}nEN has also a convergent subsequence
{(x2Ix~)}nEN. Continuing this procedure, after k steps we get a convergent subse-
quence {(xklx~)}nEN' We choose the diagonal sequence {X~}nEN. Then the sequence
{(xklx~)}nEN converges for every k, since for n > k this is a subsequence of the se-
quence {(xklx~)}nEN' We shall show that {X~}nEN is the desired subsequence of
{xn}nEN, i.e., limn .... oo(xlx~) = f(x) for every x E Hand f is a cont.inuous linear
functional.
It is obvious that the preceding limit exists for In = Lk=l akxk. Since the set of
all linear combinations of the sequence {Xn}nEN is dense in L( {x n }) there exists a
sequence of linear combinations {1m} mEN such that
lim 1m =
m~oo
X and there exists m_~
lim (lmlx~).
Therefore by the equality
and the inequality l(x-lmlx~)1 < Mllx-lmll we have that there exists limn .... oo(xlx~)
for every x E L; {x n } ). Since every element x from H can be represented in the form
x = y + h for y E L({x n }) and h orthogonal on L({x n }) and (hlx n ) = 0 for every
n E N, we obtain that for every x E H there exists limn .... oo(xlx~) and it defines a
linear functional f on H. This functional f is continuous, what easily follows from
the inequality
If(x)1 = Inlim
.... oo
(xlx~)1 < Mllxll
for every x E H. Therefore by Riesz representation theorem 10.3 there exists y E H

such that
lim (xlx~) = (xIY)
n-+oo
(x E H).
Example 10.2 (Diagonal Theorem) Let [Xijji,jEN be an infinite matrix of non-

negative real numbers such that
lim Xij
J~OO
= 0 (i EN), lim Xij
t-OO
= 0 (j EN), lim Xii
t~OO
= O.
Prove that there exists an infinite subset I of N such that
LLXij < 00.

iE! jE!
Moreover, the elements of the set I can be ordered in a increasing sequence {Pi}iEN
such that 00 00
lim L xp;p.J =0 and lim L xp;p = O.

J=l
1-+CX) .
)-+00 i=l )
Solution. Let io = O. We shall choose a sequence {in}nEN of natural numbers such

that
(i) in- 1 < in for every n EN,

(ii) Ximi. < 2- k - m for every 1 ::::: m ::::: n, 1 ::::: k ::::: nand n E N.
The proof goes by induction. Since limi~CXl Xii = 0 there exists an index r such that
Xii < 2- 2 for i 2: 1'. Let i 1 = r. Then (i) and (ii) hold for n = 1. Suppose now that
we have already find i 1 , ... ,ip such that (i) and (ii) hold for 1 ::::: m ::::: p, 1 ::::: k ::::: p.
Since Ximi -+ 0 and Xii", -+ 0 as i -+ CXl for m = 1, ... , P and :rii -+ 0 as i -+ CXl,
there exists an index ip+l > ip such that
for m = 1, ... , p + 1. Therefore (i) and (ii) hold for 1 ::::: m ::::: p + 1 and 1 ::::: k ::::: p + 1.
So we have constructed a sequence {in}nEN with the properties (i) and (ii). By (ii)
we obtain
<Xl 00
m=1 k=1
Taking I = {iI, i 2 , . . } we obtain the desired conclusion. Since the sequence of

summands of a convergent series converges zero, we obtain
Take Pk = ik for kEN.

Exercise 10.3 Let T : D(T) -+ H 2, for D(T) C HI, be a linear operator with dense
domain. Prove that
a) the adjoint operator T' of T is a linear and closed operator;
b) if additionally T is closed operator, then D(T') is dense in H 2.
Example 10.4 (Closed Graph Theorem on Hilbert spaces) Let HI and H2

be Hilbert spaces and T : III -+ H2 be a linear closed operator. Then T is a bounded
operator.
Solution. Since T is a closed operator we have by Example lO.3 that D(T') is dense
in H 2 Hence D(T*) -=I- {OJ for non-trivial Hilbert spaces HI and H 2 Let {Yn}nEN be
an arbitrary sequence from D(T'), {Xn}nEN an arbitrary but fixed sequence from
HI such that Ilxnll : : : 1 and {an}nEN an arbitrary sequence of numbers such that
limn _ oo an = O. We can represent the sequence {an} nEN as a product an = tn . Un
where tn 2: 0 and both sequences {tn}nEN and {l1 n }nEN converge to zero.
We introduce an infinite matrix of nonnegative numbers [Xij]i,iEN such that
for i -=I- j, Yi -=I- 0,

for i -=I- j, Yi = 0,
for i = j.
We shall show that the matrix [Xijji,jEN satisfies the conditions from Example 10.2.
Since UjXj ....... 0 as j ....... 00 we obtain by the continuity of the scalar product that
Xi) ....... 0 as j ....... 00 for i E N. By the definition we have Xii = O. It remains to prove
that Xij ....... 0 as i ....... 00 for j E N. Since we have
letting i ....... 00 we obtain Xij ....... 0 as i ....... 00 for arbitrary but fixed j E N.
Hence by Diagonal Theorem - Example 10.2 there exists an increasing sequence of
integers {Pn}nEN such that
L
00
lim X piP1 = O. (10.1 )

,-co .
)=1
Since UjXj ....... 0 as j ....... 00, we obtain by the completeness of HI that there exist a
subsequence {Sj LEN of {Pj LEN and an element X from HI such that
L U S1 X S1
00
= X.
j=1
We have for every pEN and every YS i -I- 0

ts l(usixsiIT*(Ys.))1 < ~ t .1(usjxsjIT*(Ys'))1
IIYsi II j=l,jii s. IIYsi II
+ts I Lj~i(usjxs1IT*(Ys'))1
IIYsi II
[or every i E N. Letting P ....... 00 in the preceding inequality we obtain by the
continuity of the scalar product
ts I(usix si IT*(Ys,))1 < ~ + t l(xIT*(Ys,))1

. IIYs i II ~ X Sis1
)=1
Si II Ys..11
LX SiSj + tsiIIT(x)11
00
<
j=1
for every i E N. Letting i ....... 00 we obtain by (10.1)
Therefore, by the Urysohn property of numbers: if for every subsequence {Zn}nEN

of a given sequence of numbers {Tn}nEN there exists a subsequence {Vn}nEN such
that Vn ....... 0 as n ....... 00, then Tn ....... 0, we obtain
as n ---+ 00. Since the sequences {an}nEN, {xn}nEN and {Y,,}nEN were arbitrary se-
quences with the prescribed properties it follows (by the boundedness) the existence
of M > 0 such that
l(xIT*(y))1 M
sup < (y E D(T*), y ::j.: 0).
IIxll~! IIYII
Therefore for an arbitrary clement x' #- 0 from H! and y#-O we have
Taking x' = T*(y) we obtain

IIT*(y)11
lIyll ::; M,
I.e,
IIT*(Y)II ::; Mllyll (y E D(T*)).
Since D(T*) is dense in H 2 , the adjoint operator T* has a bounded extension on the
whole space H 2. Therefore T* is a bounded linear operator from H2 into H!. Hence
T** = (T*)* is a bounded operator on H! and since it is an extension of the operator
T we have T = T**. Hence T is a bounded operator on HI.
Example 10.5 Let H! and Hz be Hilbert spaces. Prove that if T : HI ---+ Hz is an

injective linear operator then T- l is a closed operator if and only if T is a closed
operator. Moreover, if T maps H! onto Hz, then T- l is a bounded linear operator.
Solution. We have that G(T) is closed in HI X Hz if and only if G(T-l) is closed

in Hz x HI with respect to the same norm v'llxlllk, + Ilxzllk2.
If T is a closed operator from HI onto H 2 , then T- l is a closed operator from Hz
onto HI and by the Closed Graph Theorem - Example 10.4 the operator T- l is
bounded.
Remark 10.5.1. The preceding result can be interpreted in the following way. If
we know that the equation T( u) = f, where T is a closed linear operator, has for
every f E Hz a unique solution, then this solution continuously depends from f.
Exercise 10.6 Let H be a Hilbert space and T : If ---+ If be a linear operator. Prove
that
a) If there exists m> 0 such that IIT(x)11 : : : mllxll (x E D(T)), then T is closed
operator if and only if the range R(T) is a closed set;
b) IfT is a closed operator, then T- l is a bounded linear operator on H if and only

if R(T) is dense in H and there exists m > 0 such that IIT(x)11 ::::: mllxll (x E
D(T)).
Hint. Consequences of Example 10.5.
Example 10.7 A linear opcrator T : HI -+ H2 is compact if and only if T maps

every wcakly conveT'fJcnt sequence {xn}nEN from HI on a strongly (norm) convergent
sequence {Txn}"EN in H 2 .
Solution. Suppose that the sequence {xn }nEN from HI weakly converges to x E HI.
Then by theorem on Uniform Boundedness there exists M > 0 such that Ilxnll ::;
M (n EN). By the definition of the compact operator the sequence {TX"}"EN
belongs to a compact subset of H 2 Therefore for every subsequence of {Txn}nEN
there exists a strongly convergent subsequence {TXnJiEN with the limit v E H 2 .
We remark that v is independent of the choice of the subsequence of {TXn}nEN. We
shall show that lim n ..... oo TX n = v. Namely, we have for every y E lI2
(vly) = lim(Txnily) = lim(xniIT*y) = (xIT*y) = (Txly)

1-+00 1"""""00
Therefore v = Tx. By the Urysohn property of the convergence in the Hilbert space
we obtain that the sequence {TXn}nEN converges to v.
Suppose now that T maps every weakly convergent sequence {xn }nEN from HI on a
strongly (norm) convergent sequence {TXn}nEN in H 2 Since every bounded subset
of a Hilbert space is weakly compact, see Example 10.1, we obtain that for every
bounded subset B of HI the set T(B) is compact, i.e., T is a compact operator.
Exercise 10.8 Prove that if a sequence {Tn}nEN of compact operators from L(HI , Hz)
converges in the norm of operators to an operator T, then T is a compact operator.
Exercise 10.9 Let 0 be an open subset of Rn and H = Lz(O). Prove that the
operator T defined by
(TJ)(x) = r f(y)dy (x EO)

Jo Ix - yl"
for n - 1 < 0: < n is a compact operator on Lz(O).
Hint. Use Exercise 10.8.
Exercise 10.10 Let I< be a compact subset of Rn and H = Lz(I{). If the function
k is continuous on the set I< x I< prove that the operator T defined by
Tf(x) = iKk(x,Y)f(y)dy
is a compact operator on L 2 (I{).
10.1. I-IIr,BERT SPACE 311
Exercise 10.11 Let 0 be an open subset of R nand H = L 2( 0). rr thc function k

is measurable and satisfies
J1 OxO
Ik(x, yW dxdy < 00
prove that the operator l' defined by
Tf(x) = 10 k(x,y)f(y) dy
is a compact operator on L2(0).
Exercise 10.12 A subset A of a n-dimensional IIilbert space is compact if and only

if it is bounded.
Hint. Show that an unbounded set A can not be compact, constructing a sequence
in A which has no Cauchy subsequence.
Exercise 10.13 Let H be a separable Hilbert space, T : H - t H an linear bounded

operator and b}, b2 , . a base of the space H. Prove that the operator T is completely
given by the matrix [tijJ;,jEN, where tij = 1'(bilbj ).
Hint. Show that L~1 Itij l

2 :::; IITII (j EN), and then
00 00
T(x) = 2]2:tijxi)bj,
j=1 i=1
where x = L~1 xsbs

Exercise 10.14 A subset A of a separable Hilbert space H is compact if and only
if it is bounded and for every c > 0 there exists no E N such that II P~'o (x) II <::: c for
every x E A, where P~o is the projection operator on a no-dimensional subspace of
the space H, and P~o = I - P~o .
Exercise 10.15 A linear bounded operator T : H - t H, where H is a separable

Hilbert space, is compact if and only if for every c > 0 there exist n E N and linear
operators T1 and T2 such that 1'1 is n-dimensional and IIT211 <::: c and l' = I; + T2.
Hints. The necessity follows by the decomposition
T(x) = P~T(x) + P:T(x)

and Exercise 10.14. Take T1 = P~T and T2 = P~T.
In order to prove the sufficiency of the condition, we have to prove that for
the sequence {T(Xk)hEN, for IIxkll < M for some M > 0, there exists a Cauchy
subsequence (use Exercise 10.12).
Exercise 10.16 Let a linear operator T : If --+ II, where 1I is a separable Hilbert
space, be a contraction, i.e., IITII < 1. Prove that there exists (I - Ttl: H --+ H
and
-1 1
11(1 - T) II:.::; 1 -IITII
Hint. Prove that the unique solution of the equation (I - T)x =y for arbitrary
y E H is of the following form
Exercise 10.17 Prove that if T : H --+ H is a compact operator on a separable

Hilbert space H, then its adjoint operator T* is also a compact operator.
Hints. Using Exercise 10.15, it follows
Prove that TI* is a finite dimensional operator.
Example 10.18 A linear operator T on a complex Hilbert space H is symmetric if

its domain D(T) is dense in Hand
(T(x)ly) = (xIT(y)) (x,y E D(T)). (10.2)

A linear operator T on H is positive if (T(x)lx) ~ 0 (x E D(T)).
Prove that
a) T is symmetric if and only if (T(x)lx) E R (x E D(T)).
b) ifT is symmetric (positive) operator, then all its eigenvalues, i.e., A E C for
which T(x) = AX for some x E H, x - 0, are real (positive) numbers and the
eigenvectors which corresponds to different eigenvalues are orthogonal.
Solution.
a) If (T(x)lx) E R (x E D(T)), then we have for x, y E D(T)

1
?RT(y)lx) - (T(x)ly)) = ~-T(x + zylx + zy) - (T(x)lx) - (T(y)ly)) = 0
z
(10.3)
and
8(T(y)lx)-(T(x)ly)) = 8(T(x+y)lx+y)-(T(x)lx)-(T(y)ly)) = 0 (lOA)

10.2. THE FREDHOLM ALTERNATIVES 313
We obtain by (10.3) and (10.4)
(T(x)ly) = ?R(T(x)ly) + l8'(T(x)ly) = ?R(T(y)lx) - z8'(T(y)lx)
= (T(y)lx) = (xIT(y)),
i.e., T is a symmetric operator.
If we suppose that T is a symmetric operator, then we have by (10.2)
(T(x)lx) = (xIT(x)) = (T(x)lx) (x E D(T)),
i.e., (T(x)lx) E R for x E D(T).
b) Let A E C be an eigenvalue of a symmetric operator T and x its corresponding

eigenvector. Then we have
(T(x)lx) = (Axlx) = Allxl12 = A. (10.5)
By a) (T(x)lx) is always real (positive) and therefore (10.5) implies that the
eigenvalue is also real (positive).
If Al and A2 are two different eigenvalues for operator T and Xl, X2 the corre-
sponding eigenvectors, respectively, then by (10.2) we have
Since Al =1= A2 the preceding equality implies (xllx2) = 0, i.e., Xl and X2 are
orthogonal vectors.
10.2 The Fredholm Alternatives

Theorem 10.11 (Fredholm alternative) Let T : Ii -+ Ii be a linear compact
operator, where Ii is a Hilbert space. Then for A =f 0, the equation
(>..J-T)x=O (10.6)
either has the only trivial solution x = 0, and then the equation
(>..J - T)x =y (10.7)
has exactly one solution X = (AI - T)-I y ,

or the homogeneous equation (10.6) has a solution X =f 0, and then the nonhomoge-
neous equation (10.1) has solution if and only if (ylx*) = 0 for every solution x* of
the equation
(XI - T*)x* = O.

Exercise 10.19 Let T be a mapping H --+ H, where H is a scparable Hilbert space
and let I : H --+ If be identical mapping. Then the equation
(1 - T)x =y
can be written in the form:
where T = Ii + T 2, h = (1 - T 2)x, x = (1 - T 2)-lh and T1 and T2 are defined as in

Example 10.15 for adjoint operator T* of the operator T. The adjoint equation
(1 - T*)x* = y*
can be written in the form
x - [(I 'T' )-l)*T*

-.L2 1X* -- Z*,
where (I - T;t1y* = z',y* = (1 - T;)z* and T* = T1 + T;. (Note: according to

Example 10.16, there exists (1 - T;)-l = [(1 - T2)-1]*.)
Hints. Use Examples 10.15 and 10.16. See the solution of Example 10.20.
Example 10.20 (First Fredholm alternative) Let H be a separable Hilbert space,

T : H --+ H a compact opemtor on H, I : H --+ H the identical opemtor and
(1 - T)x = y (x, y E H), (10.8)
(1 - T*)x* = y* (x',y' E H*), (10.9)

where T* is the adjoint opemtor of the opemtor T. If one of the above equations has
a solution for an arbitmry right-hand side, then the other equation has a solution
for an arbitmry right-hand side of the equation, and those solutions are unique, ~.e.,
the homogeneous equations
(I - T)x =0 (x E H)
(I - T*)x* = 0 (x* E H),

have only zero solutions.
If one of the above homogeneous equations has only zero solution, then the other
one has only zero solution, and equations (10.8) and (10.9) are uniquely solvable for
arbitmry y, y* E H, i.e., the inverse operators of the opemtors (1 - T) and (1 - T*)
are defined on H and they are bounded.
Solution. Example 10.15 implies that for each t > 0 there exist n E N and the
linear operators T1 and T2, such that T1 is n- dimensional, IITzll < E and T = 7~ +T2
Therefore we obtain that the following equations are equivalent with (10.8)
(1 - T2)X - T1x = Y, (10.10)
h - T1(1 - T2t1h = Y, (10.11 )

where h = (1 - T2)X (from Example 10.16 it follows that x = (1 - TZ)-lh). The
operator Tl(I - Tzt 1 is n- dimensional operator (since it is T1 ). Let [tijj be its
matrix representation in an orthonormal basis ek, k = 1,2, .... Under the assumption
that the space generated by ell ez, ... , en is equal with the range of the operator
T1(1 - Tzt1 it holds tij = 0, j ~ n + 1, i ~ 1 and for each j we have f: Itijl2 ~
i=1
IIT1 (1 - T2tl1l2. It follows that (10.11) is equivalent to:
Lhjej - L(Lh;t;j)ej = LYjej,

j j i j
i.e., to
00
hj - Ltijhi = Yj,j ~ n
;=1
hj = Yj, j > n.
Since
hj = Yj, j > n, (10.12)
the last system of equations reduces on the system for j ~ n
n 00
hj - Ltijhi = Yj - L tijY;, j = 1,2, ... ,n. (10.13)

;=1 i=n+l
The equality T = T1 + T2 implies T* = Tt + T;. Therefore the following equations

are equivalent with (10.9)
x* - [(1 - TZ)-lj*T1*X* = (1 - Tn-1y' (10.14)
x* - [(1 - Tzt1j*T1*X* = z* (10.15)

(from. Example 10.16 it follows (1 _T;)-l = [(I -Tzt 1]*), where z* = (I -T;)-ly*,
and Y' = (I - T;)z*). Since [(1 - T2)-lj*Tt" is adjoint operator for the operator
T1(1 -T2t1 its matrix representation is [tjij. Therefore equation (10.15) is equivalent
to
j = 1,2, ... ,n. (10.16)
n
xj = z; + Ltj;xi, j > n, (10.17)
i=l
The matrices of the system of equations (10.13) and (10.16) are Hermit-conjugate,
implying that the absolute values of theirs determinants are equal. Hence for them
the analogue of the finite Fredholm theorem holds.
Let equation (lO.S), i.e., (10.10), (or (10.9)) be solvable for each Y E 11 (or
Y' E II). This assumption is equivalent to the assumption that equation (10.11) (or
(10.15)) is solvable for each Y E H (or Y' E 11). Specially it is solvable for each y
from the space induced by CI, C2, ... , Cn , and therefore system (10.13) (or (10.16 )) is
solvable for the arbitrary right hand side. Thus, the determinant of the system is
different of zero and the same is true for the determinant of system (10.16). It follows
that (10.13) and (10.16), with an arbitrary right hand side, has one and only one
solution and therefore (10.17) has one and only one solution. As the system (10.17)
and (10.16), is equivalent to (10.15), i.e., (10.9), we conclude that (10.9) with an
arbitrary right hand side, has one and only one solution. Therefore the homogeneous
equations (10.13) and (10.16) have only zero solution. Then by (10.12) and (10.17)
the homogeneous equations (lO.S) and (10.9) have only zero solutions.
The opposite statement follows analogously.
Let us prove that operators (I - Ttl and (I - T*)-I are bounded. Let the
system (10.13) has one and only one solution (the determinant is nonzero) and let
(hI, h2' ... , hn ) be a solution. Then (on the base of the Cramer rule) it follows that
there is a constant c > 0 such that
n n 00
L Ih lj 2 ::; c2 L IYj + L tijYil 2 . (10.lS)

j=l j=l i=n+1
Since
n n
< 2 L IYjl2 + 2 L(L It;jI2) L ly;J2
00 00
i=l j=l ;=1 i=n+I
< 211YII2 + 2nllTd1 - T2)-111211Y1I2
< IIY1I2(2 + 2n1lT1 (1 - T2)-III 2 ) = cf11Y112,
n
L Ihl12 ::; (ccd 21IYI12,
j=l
and therefore
n
IIhl12 L Ihj l2 + L
00
IYjI2::; (1 + c2c~)IIYI12 = c;IIYI1 2 ,

j=l i=n+I
(since h j = Yj for j > n). Since x = (J -1~)-lh, it holds
where C3 > 0 is a constant which does not depend on y.

Example 10.21 (Second Fredholm alternative)
Let H be a separable space, let T : h ........ II a compact operator, and
(I - T)x = 0, (10.19)
(I - Tr x * = 0, (10.20)
where I : H ........ H is the identical operator and x, x*, y, y* E H. If equation {1 0.19}
has nonzero solutions, only finite many of them are linearly independent, and equa-
tion {10.20} has the same number of linearly independent solutions.
Solution. The matrices B = [b ij ] and B* = [bji], where
blJ.. = {-tii> i.=f..j, ..

2,]="
1 2 ... ,n,
1 - tii, z = ] ,
have the same rank. Therefore, the homogeneous systems (10.13) and (10.16) have
the same number k, k ::::: n of linearly independent solutions. From Example 10.20,
it follows that there are k linearly independent solutions of homogeneous equations
(10.19) and (10.20).
Example 10.22 (Third Fredholm alternative) Let H be a separable Hilbert spa-

ce, T : H ........ H compact operator and
(J - T)x =Y (10.21 )
(I - T")x* = y*. (10.22)

Equation {10.21} has a solution if and only if y is orthogonal on all solutions of the
homogeneous equation {10.22} (y* = 0). Among the solutions of equation {l0.21},
there exists a unique solution x, which is orthogonal on all solution of homogeneous
equation {10.21}. Every solution of equation {l0.21} is the sum of x and a solution
of a homogeneous equation {10.21} (i.e., y = 0). There exists a constant c, which
does not depend on y, such that Ilxll : : : cllyll.
Solution. Let xl, x 2 , ... , xk be linearly independent solutions of homogeneous equa-

tion (10.21), and Xh, x 2 *, ... , xk< the linearly independent solutions of homogeneous
equation (10.22). The system xl, x 2, ... , xk, as well as xI-, xh, ... , xh, can be consid-
ered as orthogonal. Suppose that the solution of equation (10.8) exists. This implies
the existence of the solution of the system
n co
(10.23)
318 CHAPTER 10. SOME BLEMBNTS FROM FUNCTIONAL ANALYSIS
i,j = 1,2, ... ,n.
The rank of the matrix B is equal to n - k. Let us denote by R n - k the subspace of

the n-dimensional vector space, which is generated by the columns of the matrix
B, i.e., by the vectors Bi = (b il , ... , bin). The system (10.23) has a solution if and
only if the vector on the right-hand side belongs to Rn-k, or equivalently, if it is
orthogonal on the space R~_k'
Let us characterize R~_k' The vector x* = (xi, ... , x~) belongs to R~_k if and only
if (X*IBi) = 0, for i = 1,2, ... , n, i.e., if and only if L:~l bjiXi = 0 for j = 1,2, ... , n.
The solutions of homogeneous equations given by (10.16) belong to R~_k' If x* is
an arbitrary solution of homogeneous equation (10.22)
where x* = (xr, ... , x~) is a solution of the homogeneous system
j=l,2, ... ,n, (10.24)
(then x* E R~_k)' and xj = L:i=ltjiX';, for j > n. Since x* E R~_k it holds
n 00
L(Yj + L tijYi)xj = O.
j=l i=n+1
It follows
n (X) n n 00
o= L(Yj + L iijYi)xj = LYjxj + L( L tijYi)xj

j=l i=n+l j=l j=1 i=n+l
n 00
= LYjxj + L Yixi = (ylx*),

j=1 i=n+l
i.e., Y is orthogonal on all solutions of the homogeneous equation (10.22).
Let us prove the opposite statement. If Y is orthogonal on all solutions x* of
the homogeneous equation (10.22), then vector Yj + L:~n+l iijYi,j = 1,2, ... , n, is
orthogonal on all solutions x' of the homogeneous system (10.24), which implies
that systems (10.23) and (10.21) have solutions.
Let Xo be a solution of homogeneous equation (10.21), and Xl, x 2 , ... , xk be an
orthonormal system of solutions of homogenous equations (10.21). It follows that
x = Xo - (xolx)x 1 - . - (xolxk)x k is a solution of equation (10.21). It is unique
(prove that). Let Xf be an arbitrary solution of equation (10.21); then Xf - x = Xff
is a solution of the homogeneous equation, i.e., Xf = x + Xff.
Let. us prove the inequality Ilxll ~ cllyll. Let h be an element of II such that
(1 - T2)X = h; then h is a solution of the equation
h -1"1(1 -12)-lh = y,
where TI and T2 are linear operators, such that Tl is n( E)-dimensional, IIT211 ~ f
and T = TI + T 2 Then h satisfies the following k conditions:

(10.25)
where i = 1,2, ... , k.

Since the rank n - k of the extended matrix of system (10.23) is the same as the
rank of the matrix B, if follows that in system (10.23) there are k equations, which
are linear combinations of the rest of n - k equations, they can be excluded from
the system.
The vector (hI, h2' ... , h n ) is a solution of the system of n linear equations (n - k
of them are linearly independent equations from the system (10.23) and k of them
from the system (10.25)), whose coefficients are independent of the right hand side in
system (10.23). The uniqueness of x implies that (hI, h2' ... , h n ) is a unique solution
of the system, i.e., the determinant of the system is not equal to zero. The vector
(hI> h2' ... , h n ) can be evaluated by the Cramer rule, and therefore it holds
n n 00
L Ih j l2 ~ c2 L IYj + L t ij y;j2. (10.26)

j=1 j=l i=n+l
which implies Ilxll ~ cllyll

Example 10.23 (Fourth Fredholm alternative) Let H be a separable Hilbert
space and T : H ~ H compact operator. For an arbitrary constant M > 0 in
the circle {{l E ClI{l1 < M} of the complex plane, there exist only finitely many
characteristic values (i.e., they are of the form t,
where >. is an eigenvalue) of the
operator T, i. e. outside the circle {{l E GII>'I < 1/ M} there can be only finitely
many eigenvalues.
Solution. Let us suppose that there exist infinitely many numbers {ll, {l2, ... , {In, ... ,
which are characteristic values of the operator T, such that {li f:. {lj for i f:. j,
which are elements of the set {{l E CII{l1 < M}. By ei we denote the eigenvector
determined by the characteristic value {li, i = 1,2, ....
For arbitrary n 2:: 1, the system e}, ... , en is linearly independent. We will prove
this assertion by induction. For n = 1 the assertion is obvious. Let us suppose that
the assertion is true for n = m - 1, and that el, e2, ... , em are linearly dependent
vectors. It follows that there are nonzero constants CI, C2, ... , Cm-I, such that em =
cle! + ... + Cm-Iem-l, and therefore:
Tern = -em = CI-

el
+ ... + Cm-l--,
em-l
{lm {ll {lm-l

320 CHAPTER 10. SOME ELEMENTS FROM FUNCTTONAL ANALYSIS
and
c1(1 - 11m )Cl + ... + c",_1(1 - ~ )C m-1 = O.
fl1 11",-1
Therefore, 1 - 11m/11k = 0, k = 1,2, ... , Tn - 1, which is a contradiction. So
the assertion holds also in the case n = Tn. Denote by Rn the space generated by
{C],C2,""C n }. Then we have Rl C R2 c ... C Rn c ... and Rn oF R n- 1 , for each
n E N. For each n E N there exists Xn ERn, such that Xn is orthogonal on R n- 1
and Ilxnll = 1. Since {XJ,X2, ... ,Xn, .. '} is a bounded set and T compact operator,
the sequence TXl, TX2, ... Tx n, ... has a Cauchy subsequence. We will prove that
this contradicts to the our assumption that there exist infinitely many numbers
/11,/12,'" .
If Tn < n, then
where t7n ERn-I. This holds since TX m E Rm C Rn- 1 and
Therefore
IITx n - TXmll2 = lI~xn + t7 n 1l 2 = (~Xn + t7nl~Xn + t7n )

fln fln fln
1 1 1
= -I-12 (xnlxn) + -(xnl t7n) + -(t7n lxn) + (t7nl t7n)
fln 11n fln
1 12 2 II X nl1 2 1
= Iflnl211Xnl + li nll ;:::
t7 Iflnl 2 ;::: M2'
The above inequality implies that the sequence TX1, ... , Tx n , ... has not a Cauchy
subsequence. A contradiction.
Exercise 10.24 The equation
x - flTx = y, (10.27)
has for each y E H a solution if and only if p, is not a characteristic value of the
operator T.
If f1 is a characteristic value of the operator T, its multiplicity is finite and Ti is
characteristic value of the operator T*, with the same multiplicity. Equation (10.27)
is in that case solvable if and only if y is orthogonal on all eigenvectors of the
1
operator' T*, which correspond to eigenvalue -::-. If equation (10.27) is solvable then
fl
10.3. NORMED VECTOR SPACES 321
there exists unique solution of the equation, which is orthogonal on all eigenvectors
1
of operator T, which correspond to the eigenvalue -.
P
A compact operator has not more than countably many characteristic values. Let
I/iil < IPi+ll, i = 1,2, ... , (10.28)

be a sequence of characteristic values of a compact operator (if they exist), where each
mtmber Pi appears in the sequence k( i) - times, where k( i) is its multiplicity. If the
sequence (10.28) is infinite, then IPnl--+ 00, as n --+ 00. Then the sequence (10.28)
determines a sequence of eigenvectors e1, e2, ... , which are linearly independent.
Hint. Follows by the first, second and third Fredholm alternatives.
10.3 Normed Vector Spaces

Let X and Y be normed vector spaces.
Definition 10.12 A linear operator T : D(T) --+ Y, D(T) c X, is bounded if there
exists M > 0 such that
(x E D(T)).
We denote by L(X, Y) the vector space of all bounded linear operators from X into
Y endowed with the norm IITII = sUPllxllx9I1T(x)lIy.
Definition 10.13 A linear operator T : D(T) --+ Y, D(T) c X, is closed if its graph
G(T) = ((x,T(x)) I x E D(T)}
is a closed set in X x Y with respect to the topology induced by the norm
lI(x, y)1I = Vllxllk + lIyll}

We have the following characterization of the closed operators.
Theorem 10.14 A linear operator T : D(T) --+ Y, D(T) c X, is closed if and only
if for every sequence {xn}nEN from D(T) with the property that it converges to x
and T(xn) --+ y as n --+ 00, then x E D(T) and T(x) = y.
Definition 10.15 Let T be a linear operator T : D(T) --+ Y, where D(T) is a dense
subspace of X. Then the adjoint operator l' of the operator T has the domain
D(T*) = {y' I y' E Y',y'T is continuous on D(T)}

and T* : D(T*) --+ X' is defined by T*(y') = y'T, where X' and Y' are the dual
spaces (spaces of continuous linear functionals) of X and Y, respectively.
Vie are using for Xl E XI and x E X also the notation < :1:', X >= XI(X).
Definition 10.16 A sequence {.r" }nEN fmm a normed vector space X converges
weakly to x E X if
lim < X/,Xn >=< XI,X >
n~oo
for every Xl E XI.
Definition 10.17 A sequence {<}nEN from the dual space XI
(i) converges weakly to Xl E XI if
lim
n~oo
< X",X~ >=< X",XI >
for every x" E X";
(ii) converqes *-weakly to Xl E XI if
lim < x,x~

n~oo
>=< x,x >l
for every x E X C X".
A Banach space X is reflexive if X = X" ( in the sense ofthe canonical map x f-7 x").
Theorem 10.18 (Closed Graph Theorem) Let X and Y be Banach spaces. If

T : X --+ Y is a linear closed operator, then T is bounded.
Theorem 10.19 (Uniform Boundedness Theorem) Let A be a family of ad-

ditive and continuous operators from a Banach space X into a normed vector space
Y. If the family A is pointwise bounded on X, i.e., for x E X there exists M(x) > 0
such that
IIA(x)11 ~ M(x) (A E A),
then it is also uniformly bounded on every bounded subset B of X, i.e., there exists
M > 0 such that
IIA(x)11 ~ M (A E A, x E B).
Theorem 10.20 (Banach-Steinhaus) Let X and Y be Banach spaces and {An}nEN

a sequence f;-Qm L(X, Y). The sequence {An}nEN converges strongly to an operator
A E L(X, Y), i.e., limn-+oo An(x) = A(x) for every x E X, if and only if
a) M = sUp{IIAnlll n E N} < 00,
b) {An}nEN is a Cauchy sequence for every x from a set E c X such that L(E) =
X, where L(E) is a vector space spanned by E.
Definition 10.21 A linear operator T : X -+ Y is compact if T(B) for every

bounded subset B of X is a subset of some compact subset of Y.
Theorem 10.22 (Banach fixed point theorem) If X is a Banach space and T :

X -+ X is a contraction, i. e., there exists d,O :::; d < 1, such that
IIT(x) - T(y)11 :::; dllx - yll (x,y EX),
then T has a unique fixed point Xo, i. e., T( xo) = Xo.

Example 10.25 Let T be a linear operator with the domain D(T) in a normed
vector space X and with values in a Banach space Y. Prove that if T is bounded and
closed operator, then D(T) is a closed set with respect to the norm.
Solution. Let x be an arbitrary but fixed element of D(T) and {Xn}nEN a sequence
from D(T) which converges to x. The inequality
implies that {T(Xk)hEN is a Cauchy sequence. Since Y is complete the sequence

{T(Xk)hEN converges to some element y from Y. By the closedness of the operator
T it follows x E D(T) and y = T(x). Hence D(T) is a closed set.
Example 10.26 Let X and Y be normed vector spaces. If T : X -+ Y is a linear

operator, then prove that
a) its adjoint operator T* : D(T*) -+ X' is closed,

b) D(T*) is weakly *-dense in Y'.
Solution.
a) Let {Y~}nEN be a convergent sequence from D(T*), i.e., limn ..... "" y~ = y', and
T*(Yn) = <
-+ z'. We have by the definition of the adjoint operator
< T(x),y~ >=< x,z~ > (xEX,nEN).
Hence by the continuity of the functionals < T(x), > and < x, >
< T(x),y' >=< x,z' > (x EX).
Therefore y' E D(T*) and T*(y') = z'.
b) Follows by the definitions of D(T*) and weak *-convergence.

324 CHAPTER 10. SOME ELEMENTS FIWM FUNCTIONAL ANALYSIS
Example 10.27 (Adjoint Theorem) Let X and Y be normcd vector spaces. If

X is a Banach space and T : X -7 Y a linear operator, then p1'Ove that its adjoint
operator T* : D(Y*) -7 XI is a bounded operator.
Solution. Since Y* is a closed operator the domain D(Y*) of the adjoint operator
T* is dense in yl and therefore D(T*) =I {O} for non-trivial spaces X and Y.
Let {Y~}l1EN be an arbitrary sequence from D(T') with the property
IIY~II :::; 1. We shall prove that the sequence {Y*(Y~)}nEN is bounded what will imply
the desired conclusion.
We choose a sequence {X n }l1EN from X such that IIXnl1 = 1 and
(n EN). (10.29)
Let {an} nEN be an arbitrary sequence of numbers such that limn~oo an = O. We
can represent the sequence {an}nEN as a product an = tn . Un where tn ::: 0 and
both sequences {tn }nEN and {un }nEN converge to zero.
We introduce an infinite matrix of nonnegative numbers [Xij]i,jEN such that
for i =I j,
for i = j.
We shall show that the matrix [Xiili,jEN satisfies the conditions from Example 10.2.
Since UjXj -7 0 as j -7 00 we obtain by the continuity of the functional that Xij -70
as j -7 00 for i E N. By the definition we have Xii = O. It remains to prove that
Xij -7 0 as i -7 00 for j E N. Since we have
letting i -7 00 we obtain Xij -7 0 as i -7 00 for arbitrary but fixed j E N.

Hence by Diagonal Theorem - Example 10.2 there exists an increasing sequence of
integers {Pn}nEN such that
00
lim LX piP =0. (10.30)

t---+oo . J
J=l
Since UjXj -7 0 as j -7 00, we obtain by the completeness of X that there exist a

subsequence {SJjEN of {Pj}jEN and an element X from X such that
00
L USjx sJ == X.
j=l
On the other side, we have for every pEN and every YSi =I 0
i+p
tSiIT*(y:.)(uSiXsJI:::; L tSiIT*(Y:J(uSjxsJ
j=l,#i
i+p
+tsi IT*(Y:i)(L( us,xsJI

j=l
for every i E N. Letting p -+ 00 111 the preceding inequality we obtain by the

continuity of the functionals
00
ts,IT*(us,xsJI < L XS,Sj + ts,IT*(y~.)(x)1

j=l
00
< LXs,s, + ts,IIYs,IIIIT(x)1I

j=l
for every i EN. Letting i -+ 00 we obtain by (10.30)
Therefore, by the Urysohn property of numbers: if for every subsequence {Zn}nEN

of a given sequence of numbers {rn}nEN there exists a subsequence {Vn}nEN such
that Vn -+ 0 as n -+ 00, then rn -+ 0, we obtain
as n -+ 00. Therefore by (10.29) we obtain
as n -+ 00. Since the sequences {an}nEN, {Xn}nEN and {y~}nEN were arbitrary
sequences with the prescribed properties it follows that T* is a bounded operator
on its domain D(T*).
Example 10.28 (Hellinger-Toeplitz) Let H be a Hilbert space. Prove that if a

linear operator T : H -+ H is selfadjoint, i.e., (T(x)IY) = (xIT(y)) (x, y E H),
then T is a bounded operator.
Solution. Since T = T* we obtain the desired conclusion by Example 10.27.
Example 10.29 (Closed Graph Theorem for normed spaces) LetX be a Ba-
nach space and Y a reflexive Banach space. If a linear operator T : X -+ Y is closed,
then prove
a) that D(T*) = Y',
b) that T is a continuous operator.
Solution.
a) By Example 10.26 b) D(T*) is weakly dense in Y', since Y is a reflexive space.
By Example 10.26 a) and Example 10.27 the adjoint operator T* is closed
and continuous. Therefore by Example 10.25 D(T") is a closed subspace with
respect to the norm. Since D(T*) is a subspace the closures for weak topology
and norm topology coincides and therefore D(T*) = Y'.
b) By a) and the inequality
IIT(xll = sup 1< y',T(x) > I = sup 1< T*y',x > I

IIY'II~l Ily'119
::::; Ilxll sup IIT*(y')11 = IIT*llllxll

IIY'II~l
we obtain the continuity of the operator T.
Example 10.30 (Banach-Steinhaus Theorem) Let X and Y be Banach spaces

and {An}nEN a sequence of operators from L(X, Y). Prove that the sequence {An}nEN
strongly converges to an operator A E L(X, Y), i.e., limn--+ex> Anx = Ax (x E X) if
and only if the following conditions are satisfied
(i) M = sup{IIAnllln E N} < 00;
(ii) the sequence {Anx }nEN is a Cauchy sequence for every x from a subset E of
X such that L(E) = X, where L(E) is the vector space generated by E.
Solution. Suppose that limn_ex> Anx = Ax (x E X). Therefore
sup{IIAnxllln E N} < 00 (x EX).
Therefore by the Uniform Boundedness Theorem 10.19 follows (i). Since {An x }nEN
is a convergent sequence it follows (ii).
Suppose now that (i) and (ii) hold. Let x E X and c > O. We choose x' E L(E)
such that Ilx - x'il < c. Since {Anx'}nEN is a Cauchy sequence in Y, there exists
no E N such that for every n, m 2: no
Therefore we have for every n, m 2: no
IIAn x - Amxll < IIAnx - Anx'il + IIAnx' - Amx'il + IIAmx' - Amxll

< cllAnl1 + c + cllAml1 < (2M + l)c.
Hence {Anx }nEN is a Cauchy sequence in Y. Since Y is a Banach space the sequence
{Anx }nEN converges to v E Y. Denote by A the correspondence x 1-+ v. The operator
is obviously linear and by Uniform Boundedness Theorem 10.19 there exists M > 0
such that IIAnl1 ::::; M (n EN). Then for x such that Ilxll ::::; 1 we have
Exercise 10.31 (Riemann-Lebesgue lemma) The Fourier cocIFcients
an = ~1 j1f f(x)cosnx dx and bn = ~1 j1f f(x)sinnx d.T (n E N)

1[" -7f' 7r -7r
for a function f E L1 [~'if, 'if] converges to zero as n -+ 00.
Hints. Consider the Fourier coefficients as sequences of bounded linear functionals

on Ld~'if, 'if], e.g.,
an(J) = -;1 j"
-1f f(x) cos nx dx,
and apply Banach-Steinhaus theorem - Exercise 10.30 , where.
E = {1,cosx,sinx,cos2x,sin2x, ... }.
Example 10.32 Prove that in the Banach theorem on fixed point
a) the inequality can not be changed to strict inequality, i.e., to I/T(x) - T(y)11 <
IIx ~ yll (x,y E X,x f:. y),
b) if Tk is a contraction, then T have not to be continuous.
Solution.
a) Counterexample: X = Y = Rand T is given by T(x) = i +x - arctanx.
b) Counterexample: X = Y = [0,2] and
T(x) = { for x E [0,1]'

for x E (1,2].
Exercise 10.33 The integral equation
u(x) = 1 + A [u(t - x)u(t)dt (10.31 )
for x E [0,1] and A E (0,3/8) has a solution in the space C[O, 1].
Hint. Show that the solution u of equation (10.31) satisfies
U(U)2 ~ 2I(u) + 2 = 0, (10.32)
where I(u) = J~ u(x) dx. Examine (10.31) with respect to the parameter A. Then
apply the Banach fixed point theorem for
1
X = {'Ill u E C[O, 1], lu(x)1 ~ 1 (x E [0, l]),I(u) = >:(1 ~~)}
and operator T given by
T(u)(x) = 1 + A[ u(t ~ x)u(t) dt.

Example 10.34 (Uniform Boundedness Theorem) Let A be a family oj addi-

tive and continuous operators Jrom a Banach space X into a normcd vector space
Y. Prove that iJ the Jamily A is pointwise bounded on X, i.e., fOI x E X there exists
M(x) > 0 such that IIA(x)11 :<: M(x) (A E A), then it is also uniformly bounded
on every bounded subset B oj X, i.e., there exists M > 0 such that IIA(x)11 :<: M
(A E A, x E B).
Solution. Let {An}nEN be a sequence of operators from A, {Xn}nEN a sequence of
elements from a bounded subset B of X and {an}nEN a sequence of numbers which
converges to zero. We have to prove that anAn(xn) -+ 0 as n -+ 00.
There exists a sequence {rn} of natural numbers such that rn -+ 00 and anrn -+
O. We have by the additivity of An
(n EN). (10.33)
Suppose the theorem were not true. Then there exist c > 0 and two increasing
sequence of natural numbers {m;}iEN and {ndiEN such that
(i EN), (10.34)
where we have used (10.33).
We introduce an infinite matrix [xijkjEN in the following way Xij =
Ilam;rm;Am;(r;;]xnJII for i -:I j and Xi; = O. By the suppositions we obtain
limXij=O (jEN) and limxij=O (iEN).
t-+oo J~(X)
Therefore by the Diagonal Theorem - Example 10.2 there exists an increasing se-
quence of natural numbers {pdiEN such that
L
00
lim
:---+00 .
xp;p
J
= O. (10.35 )
J=1
Since X is a Banach space, there exists a subsequence {SdiEN of {pdiEN such that
n
lim LYs)= Y for some Y E X, where Yj = r;;:-1xn'
n--+oo J J
We have
j=1
~p ~p
lIas;As;(YsJII:<:: L lias; As;(ysJ II + II L as; As; (YsJII

j=1,#i j=l
for pEN. Letting p -+ 00 we obtain
L
00
lIas;As;(YsJII:<:: xs;s) + lIas;As;(y)1I (i EN).

j=1,jf.i
Letting i -+ 00 we obtain by (10.35) lim

...... 00
lIas;As;(YsJII = o. A contradiction with
(10.34).
Chapter 11

Functional Analysis Methods In
PDEs
11.1 Generalized Dirichlet Problem

The equation
is uniformly elliptic in the region Q eRn if there exist constants C1 > 0 and C 2 > 0
such that
C1 1z12:::; (A(x)z,z) :::; C21z12 (x E Q,z ERn)
where A(x) = [aij)nxn.
A differential operator of order 2k is given by
L(g) = L (_1) laID a(aapDP g) (ILl)

l"I.If3\$k
where aap E COO ( Q), a"p =I 0 for some a and 13 such that lal = 1131 = k. The
corresponding adjoint operator L * is given by
L*(g) = L (_1) laID"(ap"DPg) (11.2)

l"I.If3\Sk
The corresponding bilinear form is given by
B(j,g) = L (aapD'" fI DP g)L2 (Q)'

1<>1.1.BISk
329

330 CHAPTER 11. FUNCTIONAL ANALYSIS METHODS IN PDEs
The generalized Dirichlet problem for the equation L( u) = f E L 2 ( Q) means that

ok a k
for a given hw EW (Q) (corresponding to F) we have to find u EW (Q) such that
a k
B(u,g) = (hwlg) ok (g EW (Q)).
W (Q)
The bilinear form B is coercitive if there exists a constant c > 0 such that
a k
(g EW (Q)).
Theorem 11.1 (Garding inequality) Let the operator L given in Preliminaries

be strongly (uniformly) elliptic, i.e., there exists c' > 0 such that for every z E Rn
~ L zO' aO'13 (x)z13 ~ c'Izl2k (x E Q).

IO'I=I13I=k
Then there exist constants c >0 and a E R such that

Example 11.1 Let L be a linear differential operator defined by
L(u) =L aO'(x)DO'u, k ~ 1,
1<>I~k
defined on bounded region Q of R n.

Prove
a) that if aO'(x) are continuous on Q, then the operator L : C(Q) ~ C(Q) is not
bounded, but as an operator L : C k ( Q) ~ C( Q) is bounded.
b) that if a O' are bounded measurable functions, then the operator

L : L 2 (Q) ~ L 2 (Q) is not bounded, but as an operator L : Wk(Q) ~ L 2 (Q) is
bounded.
Solution.
a) We prove first that the operator D'" is not a bounded operator from C(Q) to
C(Q). Namely, if we take the sequence of functions
fs(x) = exp(is(Xl + ... + xn)) (s E N)

11.1. GENERALIZED DIRICHLET PROBLEM 331
which belongs to Ck(lJ) and which is bounded in C(Q), then the operator D'>
maps it on the sequence (is)I<>lexp(is(XI + ... + xn)) (s EN), which is not
bounded in the space C( Q). This follows by
lI(is)I<>lexp(is(xl+ .+xn))llc(Q) = m~l(is)I<>lexp(is(Xl+ +xn))1 = sl<>1 -+ 00

xEQ
as s -+ 00.
On the other hand, since the functions a", are continuous on Q, they are
bounded functions in the space C(Q). Hence there exists a constant C > 0
such that
where
Ilulb(Q) = L m~ ID"u(x)l
1",I;<;k xEQ
b) Taking the same sequence of functions {i.LEN as in a), we obtain that the
operator D'" is not a bounded operator from L 2 (Q) into L 2 (Q).
On the other hand, the inequality
IIL(u)IIL2(Q) ::; CIIUllwk(Q) (u E Wk(Q))

implies that L is a bounded operator from Wk(Q) into L 2 (Q).
(L(f)lg)L 2(Q) = (fIL*(g))L 2(Q) (f,g E Cg"(Q)),

where Land L* are given by (11.1) and, (11.2), respectively.
Solution. Applying few times the partial integration we obtain
(L(f)lg)L 2(Q) = L
l"I,I.6I;<;k
(_1)1"'1 J
Q
D"(aCtf3 Df3 g)J dx
L
l"'I,If319Q
J (a"'f3 Df3 f)DCtg dx
L
1",1,1.6I;<;k
(_1)1f31 Jf
Q
Df3(a"'.6 D"'g) dx
Example 11.3 Let L be a linear differential operator defined by
L(u) = L a",(x)Dau, k ~ 1,
lol$k
defined on a bounded region Q ofRn with coefficients a", E C,-k(Q), where s ~ k.

Prove that L is a continuous linear operator from C'(Q) into C-k(Q).
Solution. Since L is a linear operator it is sufficient to prove that it is bounded,

i.e., that there exists M> 0 such that
We have by the Leibniz formula
Therefore, taking
since lall ~ lal ~ s - k, we have
L m~IDaL(u)1 <
lal$s-k xEQ
< M' E E E m~IDa+P-<>lul

lol$s-k IPI$k 01 $<> xEQ
Therefore by the inequality
we obtain
E m~IDOL(u)1 ~ M E m~ID'YL(u)l
101$6-k :z:EQ 1"11$8 xEQ
Exercise 11.4 Prove that the differential operator from Example 11.3 is a contin-
uous operator from WS(Q) into WS-k(Q).
Hint. Prove that there exists M > 0 such that

Example 11.5 Let Q be a bounded region of Rn with an enough regular boundary

DQ. Prove that the bilinear form
L
ok
B(j,g) = (aa{3IY' fI D{3g)L 2(Q) (f,g E W (Q))
lal.161Sk
ok 0 k
is bounded on W (Q) X W (Q), i.e., there exists M > 0 such that
ok
IB(f,g)l:::; Mllfll
w
ok
(Q)
IIgll Wok
(Q)
(f,g E W (Q)).
Solution. Since aa{3 E C(Q) (0:,(3 E Z't-), we have lIar>{3liL2(Q) < C for some
o k
C> O. Therefore we have for f,g E W (Q)
IB(f,g)1 < L lIa,,{3D" fIIL2(Q)IIDf3 g II L2 (Q)

1,,1,1{3IS k
< C L liDO' fIlL2(Q) II Df3g Il L2 (Q)
lal.l{3ISk
< Mllfllw\Q)iigllw\Q)'
ok
Example 11.6 Prove that for every FE L 2 (Q) there exists a unique hw E W (Q)
such that
ok
(g E W (Q)).
Solution. For a fixed but arbitrary F E L 2 ( Q) we have that the functional h

defined by
ok
h(g) = (Flg)L2(Q) (g E W (Q))
ok
is continuous on the Hilbert space W (Q), since we have by Cauchy-Schwartz in-
equality
o k
Therefore by Riesz representation theorem there exists a unique hw E W (Q) such
that
(hwlg) ok = h(g) = (Flg)L2(Q).
W (Q)
Example 11.7 Prove the equivalence of the following two probZ,~m.8
ok
(i) For a given F E L 2 ( Q) find u E W (Q) such that
ok
B(u,g) = (Flg)L2(Q) (g E W (Q));
334 CHAPTER 11. FUNCTIONAL ANAJXSJS METHODS IN PDEs
a k a k
(iij For a given hw E HI (Q) find 11 E HI (Q) such that
ok
B(11.,g) = (hwlg) ok (g E HI (Q)),
VI' (Q)
where B is fmm Example 11.5.
Solution. Follows by Example 11.5.
Example U.B (Lax-Milgram) Pmve that for every bounded bilinear form B on
a k ok
W (Q) X W (Q)
a k ok
aj there exists a unique linear continuous operator l' : W (Q) -t W (Q) such
that
a k
B(f,g) = (TIlg) ok (f,g E W (Q));
W (Q)
bj il additionally B is coercitive, then there exists 1'-1 and it is a continuous

a k
operator on W (Q).
Solution.
ok
a) Since for an arbitrary but fixed fEW (Q) the functional h(g) = B(f,g) is
antilinear (for complex case) and continuous by Example 11.5, there exists by
a k
Riesz representation theorem a unique function w E W (Q) such that
ok
(wig) ok = h(g) = B(f,g) (g E W (Q))
w (Q)
and Ilwll W k
(Q)
= Ilhll. Then the desired operator l' is defined by
ok
T(f) = w (f E W (Q)). The operator T is linear and B(f,g) = (TIlg) ok
W (Q)
The boundedness of the operator l' follows by
for some M > o.

b) Since B is coercitive there exists e > 0 such that
111'III W 0 k
(Q)
Ilfll.W k
(Q)
:::: IB(f, J)I :::: ?RB(f, J) :::: e11f112.W k
(Q)

Therefore
111'111 W ok
(Q)
:::: ellIl1 W
ok
(Q)

If 9 is orthogonal on the range R(T) then
B(g,g) = (Tglg) W ok
(Q)
= O.
ok
Hence by the coercitivity of B we have 9 = O. Therefore R(T) W (Q).
This implies by Closed Graph Theorem that T- 1 is a continuous operator on
ok
W (Q).
Example 11.9 Let the corresponding bilinear form B to the differential operator L
of the order 2k in the generalized Dirichlet problem from Exercise 11.7 be coercitive.
Prove that for every F E L 2 (Q) there exists a unique solution.
Solution. By Example 11.7 and Example 11.8 we obtain

ok
(Tulg) ok = (hwlg) ok (g E W (Q)).
w (Q) W (Q)
Therefore T(u) = hw. Since by Example 11.8 b) the operator T- 1 is continuous and
ok
linear on the space W (Q). Therefore u = T-1h w gives the unique solution of the
considered generalized solution of the generalized Dirichlet problem.
Remark 11.9.1 It is important the question of the regularity of the solution of the
generalized Dirichlet problem, i.e., when the obtained generalized solution from the
ok
space W (Q) is also the classical solution. We give here only a general theorem in
this direction.
Theorem 11.2 Let Q be an open set of Rn and the operator L of the order 2k is
given by
L(u) = aoD"'u, L
lol:52 k
where a", are constants for lal = 2k and a o E COO(Q) for others a. If L is strongly
elliptic, FE WS(Q) and u is the generalized solution in L 2 (Q), then u E W 2k +s(Q')
for every bounded open subset Q' of Q.
If additionally m < 2k + s - ~ then U E cm(Q'). If additionally F E COO(Q),
then U E COO(Q').
Example 11.10 Prove under same suppositions as in Theorem 11.1 that the bilin-
ear form Ba for a E R given by
is coercitive.
336 CIIAPTER 11. FUNCTIONAL ANALYSIS METHODS IN PDEs
Solution. By Theorem 11.1 we obtain
SRba(g,g) SRB(g,g) + allgIIL(Q)

> cllgll
W (Q)
allgIIL(Q) + allgl1 2W
0 k - 0 k
(Q)
= cllgll 2W
0 k
(Q)
Example 11.11 The solution u, of the generalized Dirichlet problem from Example
11.7 continuously depends on F E 2(Q).
ok
Solution. We denote by U the embedding operator U : W (Q) - t 2(Q) and by
U* its adjoint operator. The operators U and U* are continuous. We have
ok
(F E 2(Q),g E W (Q)). (11.3)
By Example 11.7 and Example 11.8 the solution is T-l(hw), where

ok
(hwlg)
w
ok
(Q)
= (Flg)L 2 (Q) (g E W (Q)).
Therefore by (11.3) we have u = T-1U*(F). Since T-1U* is a continuous operator

we obtain the desired conclusion.
Remark 11.11.1 We shall call the operators
o k
G = T-1U' : 2(Q) - t W (Q),
a = UT-1u. : 2(Q) - t 2(Q)
Green operators. If the corresponding bilinear form is coercitive then the Green
operators are continuous.
Example 11.12 Prove that for the bounded region Q the Green operator aa :
2(Q) - t 2(Q) which corresponds to the bilinear form Ba from Example 11.10
is a compact operator.
ok 00
Solution. By Theorem 9.7 the embedding U : W (Q) - t W (Q) = 2(Q) is a
compact operator. Since Ta- 1 and U are continuous operators, where Ta is the
corresponding operator to Ba from Example 11.8, we have that aa = UT;;lU* is a
compact operator.
Example 11.13 Prove the following analogy to Fredholm alternative.
If is a strongly elliptic operator from Preliminaries on a bounded region Q, then
either the generalized Dirichlet problem from Example 11.7 has exactly one solution
for every F E 2(Q),
or the zero is the characteristic value, i. e., >. is the characteristic value if there exists
ok
a function uf:.O from W (Q) such that
ok
B(u,g) = >'(Ulg)L (Q)2 (g E W (Q)).
Solution. By Theorem 1l.1 and Example 11.10 there exists a real number a such
that the Green operator Ga is a compact operator. Since the generalized Dirichlet
problem is equivalent with the equation
where FGa(F) a.nd u = U(u) ( to prove that it is enough to add a(ulg) to both
sides in the equality B(u,g) = (hwlg) ok and apply Remark 11.11.1), we obtain
w (Q)
the desired conclusion by theorem on Fredholm alternative.
Example 11.14 Pr'ove that the following bilinear form
B(j,g) = 10 (V f(x)Vg(x) + (k + p(x))f(x)g(x)) dx

for k E R, p E CCO(Q) and k ~ 1 - Po, where Po = infxEQ i'Rp(x) , is coercitive.
Solution. The coercitivity of B follows by
i'RB(j, J) i'R 10 (V f(x)V f(x) + (k + p(x ))f(x )f(x )) dx

Ilfllivl(Q) -llfliL2(Q) + i'R((k + p)flJ)L (Q) 2
> Ilflliv1(Q) + (k + Po - l)llfIIL(Q) ~ Ilflliv1(Q)'

since k ~ 1 - Po.
Example 11.15 !Vith the same notations as in Example 11.14 prove that for the
generalized Dirichlet problem for L = -,0. + k + p ther'e exists a unique solution for
k ~ 1 - Po, where Po = infxEQ i'Rp(x).
Solution. By Example 11.14 we have that for the operator L the corresponding
bilinear form
B(j,g) = 10 (V f(x)Vg(x) + (k + p(x))f(x)g(x)) dx

is coercitive. Then by Example 11.9 the given generalized Dirichlet problem has a
unique solution.
Exercise 11.16 Let Q be a bounded region with smooth boundary oQ. If a function
u E C(Q) n C 2 ( Q) is the solution of the Dirichlet problem
n n 02u n OU
L(u) = L L aik(X)~ + Lai(X)~ + au = f on Q
i=l k=l UXiuXk i=l uXi
Ul8Q = cp,
where
aik, ai, a E C(Q), aik = ak; (x E Q),
cp E C(aQ), a(x) ~ 0 (x E Q)
and L is strongly (uniform) elliptic, then there exists a constant C > 0 such thai we
have the following a-priori inequality
where
IlfllQ = sup
xEQ
If(x)1 and Ilcpll&Q = max
xE&Q
Icp(x)l
Hints. Take Xl ~ 0 in Q. Choose a number z > Xl (X E Q) and a number 0: >0
enough big that the following inequalities hold
co: 2 - k(o: + 19) ~ 1 and eOlz > 2m~eOlXI.

xEQ
Then introduce the function
and prove that -L(h) ~ IlfiIQ' Then show by the Maximum Principle that lu(x)1 ~
h(x) (x E Q). Taking
C= m~(eOlZ _ eaxl )
xEQ
prove by the last inequality the desired inequality.
Example 11.17 Let

n n a a
L = - t;~ aXj (PaXi) + q,
be a differential operator on a bounded region Q, where p E C 1 (Q),q E C(Q);
p(x) > 0, q(x) ~ 0 (x E Q). The domain D(L) of the operator L consists of the
functions u E C 2 ( Q)nCl(Q) and L( u) E L 2 ( Q) and on the boundaryaQ they satisfy
the condition
(11.4 )
where 0:, f3 E C(aQ) and
o:(x) ~ 0, f3(x) ~ 0, o:(x) + f3(x) > 0 (x E aQ).
Prove that
a) the operator L is symmetric and positive;

b) eigenvalues of the operator L al'e nonnegative and the con'esponding eigen-

functions to different eigenvalues are orthogonal.
Solution.
a) We obtain by Green identity
1(vL( u) - uL(v)) dx = 1 P( u-;:;-av - v-;:;-)

Q
au dS.
~ ~ un
Therefore taking v E D(L) instead of v and u E D(L) we obtain
= 1aQ
Ov au
P( U-;:;- - V-;:;-) dS.
un un
(11.5)
The functions u and v satisfy the boundary condition (11.4), i.e.,
The condition a( x) + f3( x) > 0 (x E aQ) enshures that the preceding ho-
mogeneous system of linear equations has nontrivial solution (a, 13). Therefore
the determinant of this system is equal zero, i.e.,
u au
an Ov
= (u- au o.
v ail an - v-)laQ
an =
an aQ
Putting this in (11.5) we obtain (L(u)lv)L2(Q) = (uIL(v))L 2(Q), i.e., the operator
L is symmetric.
We shall prove now that the operator L is positive. We obtain by the Green
formula
(L(u)lu)L2(Q) = 1Q L:nau
p1!:J12 dx - la
uxj=l aQj
aau
pU- dS
n
+ 1Qqlul dx.
2 (11.6)
Since (11.4) implies
au
- = --u
a
for f3(x) > 0 (x E aQ)
an 13
and
u=O for f3(x) =0 (x E aQ),
we obtain by (11.6)
(L(u)lu)L2(Q) = . :;- . au I +qlul )dx+ J

J(pL.-lax 2 2
p,a1u
a
l2 dS (u E D(L ) ).
Q J=l J 8Q
a>O,!3>O
(11. 7)
Since all summands in the preceding equality are nonnegative we obtain
(L(u)lu)L2(Q) 2: 0, i.e., the operator L is positive.
b) By a) the operator L : D(L) -+ L 2 ( Q) is symmetric and positive and therefore

the desired conclusion follows by Example 10.18.
Remark 11.17.1 By (11.7) we obtain
since the function p is continuous on the compact set Q.
Example 11.18 (Hormander) If a linear differential operator
P(D) = P( -i~, ... , -i~)

aX1 GX n
with constant coefficients is hypoelliptic , i.e., for every generalized solution u E
L 2,loc(Q) of the equation P(D)u = F for F E coo is almost everywhere equal to a
function from Coo, then for every constant C1 > 0 there exists a constant C2 > 0
such that every solution z = ~ + iT/ = (Zl' ... ,zn) of the algebraic equation P(z) = 0
satisfies the condition:
n n
if IT/I = (2: IT/jI2)1/2 < C2, then Izl = (2: IZjI2)1/2 < C 1 (11.8)
j=l j=l
Solution. Let U be the set of all generalized solutions u E L 2 (Q'), where Q is an

open subset of the region Q, for the equation P(D)u = 0, i.e.,
(uIP*(D)<p)L 2 (Q') = i[Q, U P*(D)<pdx = 0 for every <p E C;'(Q'),

where P' is the adjoint operator for P, P*(z) = P( -Zl, ... , -zn). Since P(D) is a
linear operator U is a vector subspace of the space L 2 ( Q'). We shall show that U is
a closed subspace. Let {Un}nEN be a sequence from U which converges in the space
L 2 (Q') to u E L 2 (Q'). Since we have
i[Q, Un' P*(D)<p dx = 0 (n EN),

11.1. GENER.ALIZED DIR.ICI-lLET PR.OBLEM 341
we obtain by the continuity of the scalar product (I)L2(QI) that
Hence (uIP*(D)cp)L2(QI) = 0, i.e., u E U. Therefore U as a closed subspace of the

Banach space L 2 ( Q') itself is a Banach space.
By the hypoellipticity of the operator P(D) it follows that every function u from
the space U belongs to C=(Q'). If Q~ cc Q', then aauXj E COO(Q~) (j = 1, ... , n)
for every function u E U. The operators T j : U -4 L 2 (QD, j = 1, ... , n, given
by Tj(u) = ::. are closed operators, which by the Closed Graph theorem (U is
J
a Banach space and L2 (QD is a Hilbert space) are bounded operators. Therefore
there exists a constant C > 0 such that
(u E U). (11.9)
If z = ~ + iTJ is a solution of the algebraic equation P(z) = 0, then applying the

inequality (11.9) on the special function u(x) = exp(i2:i=l XiZi) we obtain
This implies that if for some C 2 > 0 we have ITJI < C2 , i.e., it is bounded, then there
exists C1 > 0 such that Izl < C1
Remark 11.18.1 It is true also the statement in the opposite way, which gives
a characterization of the hypoelliptic operators. As a special case of this opposite
statement we obtain
Theorem 11.3 (Weyl lemma) Every generalized solution u E L2 of the Laplace

equation
/).u = FE L2
is a function from COO except on the set of measure zero from the region where
FE Coo.
Follows by the fact that the zeroes of the equation - 2:7=1 zJ = 0 satisfy the condition
(11.8).
Exercise 11.19 Let A be a closed subspace of a Hilbert space H. Prove that for
every x E H there exists a unique XA E A such that
IIx - xAIl = inf

zEA
IIx - zll = d.
Hints. Consider the set l; - A and a sequence {xn} nEN from x - A such that
lim
n-+oo
Ilxnll = d,
and prove using the parallelogram law
(X,y E H),
that {Xn}nEN is a Cauchy sequence with the limit x - XA, where XA is the desired
unique clement.
Remark 11.19.1 This exercise gives as a consequence
Theorem 11.4 (Projection theorem) Every clement x E H can be represented

in the following form
x = XA + Y for some y E AJ.,
where AJ. is the subspace of all orthogonal elements on A.
Example 11.20 Let Q be a bounded region with 8Q E Coo. Let p E COO(Q) and
FE L 2 (Q). Prove
a) that
o 1
(u,g E W (Q))
o 1
is a scalar product on W (Q) which induces an equivalent norm with the norm
II'IIW'(Q);
o 1
b) if u E W (Q) is a solution of the equation
o 1
(g E W (Q)), (11.10)
o 1
then for an arbitrary closed subspace A of W (Q) the function UA E A ( see
Exercise 11.19) is the unique solution of the equation
(g E A). (11.11)
If A is finite dimensional we call UA the Ritz approximate solution.
Solution.
a) See for more general case Example 11.17.

11.1. GENERALIZED DIRICHLET PIWBLEl\,f 343
o 1
b) I3y Exercise 11.19 we have for u E Hi (Q)
u = UA + (u - UA) for UA E A and u - UA E A.L. (11.12)
Putting this function U from (11.12) into (ILlO) we obtain that for every
gEA
(F[g)L2(Q) = (UA + U- UA[g)E = (UA[g)E + (u + UA[g)E = (UA[g)E,

i.e., (11.11). If we suppose that u~ is another solution of the equation (11.11)
in A, then we have (UA -U~[g)E =
(g E A). Hence UA -u~ E A.L. Therefore
UA - u~ = 0, i.e., UA = u~.
Remark 11.20.1 For n-dimensional A we have
Theorem 11.5 (Ritz) If b1 , , bn is a base of the space A, then the system of
equations
n
L cj(bj[b;}E = (F[bi)E (i=I, ... ,n)
j=1
has a unique solution and the approximate Ritz solution UA in A for the correspond-
ing Dirichlet problem has the following form
n
UA = LCibi
i=l
and the following estimation holds
(g E A).
Exercise 11.21 Prove that the following systems of functions {tpn}nEN are or-
thonormal bases in the space L2 (1), where I is the corresponding interval in R
e" nx
a) tpn(x) = ICC' n = 0,1,2, ... , for x E 1= (-7r,7r) (the Fourier system);
v27r
b) tpn(X) = (n + ~)1/2. Tn [~J(_lr(~) (2n - 2i)x n- 2i ,

2 i=O l n
n = 0, l, 2, ... , for x E 1= (-1,1), (the Legendre system);
c) tpn(X) = (2nn!.jirtl/2e x2 / 2Hn(x),
n = 0,1,2, ... , for X E I = (-00,00), where Hn is the Hermite poly-

nomial given by
[n/2] (_I)i(2x)n-2i
Hn(x) = n! ~ i!(n _ 2i)! '
(the Hermite system);

d) r.pn ( x ) -_(2)1/2'
- _
smnx, n -1,2, r
... , jorx E J -_ 0,7r.
()
7r
Exercise 11.22 Provc that the systcms of functions {r.pn}nEN from Exercise 11.21
a)- d) are the corresponding eigenfunctions for the following differential operators
with the corresponding eigenvalues {,\n}nEN :
a) A = -iD, An = n, where D = ~;
dx
b) A = D(x 2 - I)D, An = n(n + 1);
x 2 + 1, An
%2 2:r2
C) A = e> De- x De> = D2 - = -2n;
d) A = D2, An = _n 2
Exercise 11.23 Let Q C Rn be a bounded region with smooth boundary aQ. Prove
that the equation
(11.13)
where aij E Cl(Q), bj , c E C(Q) and F E C( Q), can be written also in the following
form
n n a au n a(ai U) n au
~f; ax/aij(x) aXj + ~ ~ + ~BiaXi + Cu = F, on Q, (11.14)
where ai E C 1(Q),Bi ,C E C(Q). Prove that the equation (11.14) can be written in
the form (11.13).
Exercise 11.24 Let Q be a locally quadratic bounded region and

aij, ai E C 1(Q), bi , c E CeQ), F E C( Q) n L 2( Q)
and g E C( aQ). Prove that the (classical) solution u E C 2 ( Q)nC( Q) of the Dirichlet
problem for the uniform elliptic PDE
n n a ( au ) n a( aiu ) n au
L L ;lx. aij(X)~ + L-;l-. + Lbi~cu = F, UlaQ = g, (11.15)
.=1 ;=1 U UX) .=1 UX. .=1 uX,
which belongs also to C 1(Q) satisfies:

a) for every r.p E CJ (Q) the following equality
-LL
nn
i=1 j=1
1
Q
aij(x)--dx-L
aXj aXi
au~
;=1
n
1
Q
ai u -
mp
aXi
dx
+ tl bi(x)~Ucpdx+ 1
i=1 Q uXi Q
cucpdx = 1Q
Fcpdx. (11.16)
o I
b} for every 'P E W (Q) the equality {11.16};
c} the trace of the function u on 8Q {see Remark 11.17.1} is equal to the function
g.
Remark 11.24.1 (The generalized solution) Let a;j,a;,b;,c E Loo(Q),aij =

aji, F E L 2 (Q),g E L 2 (8Q) in (11.15) ( the generalized Dirichlet problem). A
o I
function u E WI(Q) for which the equality ( 11.16) holds for every 'P E W (Q) and
its trace on the boundary 8Q coincides with g is called the generalized solution of
the problem (11.15).
Exercise 11.25 Prove:
a} that the classical solution u E CI(Q) of the problem {11.15} is also a general-
ized solution;
b} that the generalized solution u of the problem {11.15} which satisfies u E

CI(Q) n C2(Q) is also the classical solution.
Hints. a) follows by Exercise 11.24 a), c).
Example 11.26 Formulate the generalized Dirichlet problem for ulaQ = o.

Solution. Since the conditions u E WI(Q) and ulaQ = 0 are equivalent with
o 1
u E W (Q), we obtain that the Dirichlet problem (11.15) for g = 0 reduces on
0 1
the problem of finding a function u E W (Q) which satisfies (11.16) for every 'P E
o 1
W (Q).
Example 11.27 Which of the following differential operators are uniform elliptic
a) -D.. = -Ei=I D?; c) Dl - D~;

b) - (Xl D12 + D2)2 ; d') (_I)k L..l.
A k?
Solution.
a) The equality
n n
- L Zi(-I)Zj = LZ?,
;=1 ;=1
the differential operator -D.. is uniform elliptic.
b) The operator -(xID~ +D~) is uniform elliptic on {(Xl, X2)1 Xl ~ a} for a > 0,
but not for a = o.
c) Since Z2 1 Z2 ::; z:
IzI2 = +z~ the heat transfer operator is not uniform elliptic.
346 CHAPTER 11. FUNCTIONAL :\NALYSIS METHODS IN PDEs
d) This operator is uniform elliptic.
Example 11.28 Let us consider the following PDE on a bounded region Q

n n
L: L: D;(aijDju) + cu = F, (11.17)
i=1 j=1
where a;j, c E Coo(Q), aij = aji, F E L 2 (Q) and c(x) ~ 0 (x E Q) and the
differential operator in the equation (11.17) is uniform elliptic. Prove that
a) the bilinear functional
o 1
(J,g E W (Q))
o 1
is a scalar product on the space W (Q) which induce a norm equivalent to the
norm induced by the usual scalar product (JIg) .1 ;
W (Q)
b) the functional
o 1
(cp E W (Q))
o 1 0 1
is linear and continuous on W (Q) and there exists a function hw from W (Q)
such that h(cp) = (cplhw)E and for some M > 0 we have
(hwlhw)E = IIhll2 ~ M21IFIIL(Q)

o 1
c) there exists a unique solution u E W (Q) of the generalized Dirichlet problem
o 1
for the equation (11.17), i.e., we have for every cp E W (Q)
-ttli=lj=1 Q
a;j(D;u)(Djzp) dx + 1
Q
fZP dx = 1 Q
Fzp dx.
d) there exists C > 0 such that
Ilull w.1
(Q)
~ CIIFIIL2(Q).
Solution.
a) It is easy to check that
o 1
(J,g E W (Q))
o 1
is a scalar product on the space W (Q) ( the property (fl!) ~ 0 will follow
from (11.18)).
We shall prove that the induced norm lilliE = V(fIJ)E is equivalent to the
usual norm
11/11., =([ 1/1 2dx+t [IDJI 2dx)t.
W (Q) JQ J
;=1 Q
Since the differential operator in the equation (11.17) is uniform elliptic and
c ::; 0 we obtain
IIIII~ = (f1!)E = t t l a;j(Dj/)(DJ) dx - 1 cl/1 2dx

;=1 j=l Q Q
~ C 10 ~ IDj/12 dx -
1 10 cl/1 2dx ~ C 10 ~ IDj/12 dx.
1 (11.18)
Since by Example 9.14 d) the norm
is equivalent to the usual norm 11/11 .1 there exists C2> 0 such that
W (Q)
11/1I2~C211/112.1
W (Q)
.
Therefore by (11.18) we have
II/II~ ~ C1 C211/11 2
1 .
W (Q)
(11.19)
We shall prove that there exists a constant I< > 0 such that
II/II~ ::; I<11/112.W 1(Q) .

Since a;j, c E Coo (Q) thre exists C~ > 0 such that laij (x) I ::; C~ and Ic( x) I ::; C~
for x E Q. Therefore we have by the Cauchy-Schwartz inequality
lilliE = ~ ~ 10 aii(Dd)(DJ) dx - 10 cl/1 dx 2
< C~ ~ ~ 10 IDj/llDJI dx + C~ 10 1/12 dx

< C~ ~ ~(1o IDj/12 dx)t . (10 IDJI2 dx)t + C1' 10 1112 dx
= C~ (t(1IDJ,2 dx)t)2 + C~ 11/12 dx .

=1 Q Q
Therefore, using the inequality (L:~1 S;)2 ~ n 2:i'=1 S[ for rea.! numbers Sj 2:
0, (i=l, ... ,n),weobtain
for ]( 2: Cl/n.
o 1
b) It is obvious that h is a linear functional on the space W (Q). We shall show
that it is a bounded functional. Using the Cauchy-Schwartz inequality and
(11.19) we can find a constant M > 0 such that
Ih(cp)1 =I k Fcpdxl ~ IIFIIL (Q)lIcpIlL (Q) ~

2 2 IIFIIL2(Q)lIcpIIW 1 (Q) ~ IIFIIL2 (Q)llcplJE.
(11.20)
Therefore applying Riesz representation theorem on the scalar product (cpICP)E
o 1
we obtain that there exists a unique hw E W (Q) such that h( cp) = (cplhw).
Then by (11.20) we obtain
(11.21)
c) Using the scalar product introduced in a) we can rewrite the Dirichlet problem
for the equation (11.17) in the following form
o 1
(cp E W (Q)).
o 1
By b) there exists hw E W (Q) such that the previous equality can be rewrit-
ten in the form
(Ulcp)E = (cplhw), i.e., (u - hwlcp) =0

o 1
for every cp E W (Q). Hence u - hw = 0, i.e., u = hw, what means that the
considered generalized Dirichlet problem has a unique solution u = hw.
d) The inequality (11.21) implies
Remark 11.28.1 By Example 11.28 and Exercise 11.24 a) we obtain also the
uniqueness of the classical solution for F E L 2 (Q) for a locally quadratic region
, in C 1 (Q).
11.1. GENER.ALIZED DTRICHLET PR.OBLEM 349
Exercise 11.29 Prove that for the generalized Dirichlet problem
ulaQ = g, (11.23)
for aij,C E COO(Q),aij = aji,F E L2(Q),c(x) ~ 0 (x E Q) and g E L2(8Q) such a
function that there exists gw E W 1(Q) such that the trace gwlaQ = g, always there
exists a unique generalized solution u E W 1(Q) and a constant M > 0 such that
Hints. Use the preceding Example 11.28, since u E W 1 (Q) is a solution of the
generalized Dirichlet problem (11.22) and (11.23) if and only if the function U1 =
o 1
U - gw is the generalized solution from the space W (Q) of the equation
= 1Q
F'j5 dx + ttl
;=1 j=l Q
a;j(Djgw ) (D;'j5) dx -1
Q
cgw'j5 dx ()O E H/ (Q)).
Taking the right side as a linear continuous functional -h we can obtain the solution
in an analogous way as in Example 11.28.
a) for every bounded sequence {fm}mEN of functions from the space W1(P), where
P = (a1, bd x ... x (an' bn ), there exists its subsequence which is convergent in
the space L 2( Q)j
b) for a bounded region Q eRn every bounded sequence {fm}mEN from the space
o 1
W (Q) there exists its subsequence which is convergent in the space L 2( Q),
o 1
i.e.} the embedding mop of the space W (Q) into the space L 2 (Q) is a compact
operatorj
c) for a locally quadratic bounded region Q C Rn the embedding map of the space
W 1(Q) into the space L 2(Q) is a compact operator.
Solution.
a) We divide the parallelepiped P = (aI, bd X '" X (an, bn) on smaller paral-

lelepipeds with sides
s
" .. , s
for some sEN. The total number of these parallelepipeds is sn, and therefore
we denote them by PI, ... , Fsn .
We shall show that the sequence {!m}mEN has a Cauchy subsequence in the
space L 2 (P). For that purpose we apply the Poincaire inequality from Chapter
9. on the function !m - !k on one of the parallelepipeds, say Pj
Adding all these inequalities we obtain
(11.24)
By the boundedness of the sequence {!m}mEN there exists a constant C > 0

such that
Therefore for every > 0 there exists So ( the number from the dividing the
parallelepiped) such that
(11.25)
On the other side, since the sequence Um}mEN is bounded in the space Wl(P)
it is bounded also in the space L 2 (Q). By Example 10.1.2 on weak compactness,
there exists a subsequence {!rm}mEN of Um} "EN which weakly converges in the
space L 2 (P), i.e., the sequence of numbers {L
!rmXj dx }mEN (j = 1, ... , sn)
converges, where Xj is the characteristic function of the parallelepiped Pj'
Therefore there exists rno E N such that for every rn ;::: rno
(11.26)
where
sn sn
M-L~~-~-~
]=1 (bi - aD", (b~ - a~)
Applying the inequalities (11.24), (11.25) and (11.26) on the subsequence

{Jrm}nEN we obtain
t IJrm - fr.1 2 dx < E for m, k ~ mo
Hence {Jrm} mEN is a Cauchy sequence in the space L2 (Q) and therefore also
a convergent sequence.
b), c) It is enough to take a parallelepiped P :J Q, since the bounded sequence can

be extended with {Fm}mEN on P such that it remains bounded on P ( taking
Fm(x) = 0 for x E P \ Q and l"m(x) = fm(x) for x E Q).
Example 11.31 Prove that the problem of finding the generalized eigenvalues for
homogeneous boundary problem for uniJorm elliptic equation
n n
LLDi(aijDju) + cu = F,
i=1 j=1
where aij, c E COO(Q), aij = aji, c(x) ~ 0 (x E Q), F E L 2 (Q), which consists in
o 1
finding nontrivial solutions U E W (Q) Jor the equation
n n
L L Di(aijDju) + cu + AU = F (11.27)
i=1 j=1
( A E C is the eigenvalue) reduces on the equation
u-AT(u)=h (11.28)
01 01 01
Jar hEW (Q), where T W (Q) --+ W (Q) lS a compact selJadjoint positive
operator.
Solution. The generalized problem (11.27) reduces on finding a function U E

o 1
W (Q) such that
-t t ir
;=1 j=1 Q
aij(Dju)(D;zp) dx +
iQ
r cuzp dx + A ir uzp dx = ir Fzp dx.
Q Q
Using the solution of Example 11.28 introducing the scalar product

o 1
(1,g E W (Q)) (11.29)
o L
and then we can rewrite the preceding equation (h = hI\' E W (Q) from Example
11.28)
o 1
(cp E W (Q)). (11.30 )
We shall represent JQ uip d.T also by the scalar product (11.29). Namely, take the
linear functional hI for an arbitrary but fixed u E L 2 (Q)
o 1
(cp E W (Q)).
It is continuous, since we have
o 1
By Riesz representation theorem there exists a unique element HI E W (Q) such
that
(11.31)
and
IIh l l1 2 = (HIIHdE ~ M21IuIlL2(Q) (11.32)
By the definition of the functional hI and (11.31) we have
(11.33)
o I
Let TI : L 2 ( Q) --+ W (Q) be an operator which maps the element u on the eJement
HI. By (11.33) TI is linear, and by (11.32) it is bounded, since we have
By (11.33) we obtain
o I
(cp E W (Q)). (11.34 )
Therefore the equation (11.30) reduces on the form

o 1
(cp E W (Q)).
Hence u - )"T( u) = h, where we denote by T the restriction of the operator Tl to
o I
W (Q).
The operator T is selfadjoint, since by (11.34) we have
(T(u)lcp)E = k uipdx = k cpudx = (T(cp)lu)E = (uIT(cp))E.

l1.1. GENERALIZED DIR.ICHLET PROBLEM 353
On the other side, since
if the function U is not almost everywhere equal to zero, we obtaiIl that the operator
T is strictly positive.
We shall prove that the operator T is compact. Namely, since l' = 1'11'0, where
a 1
To : W (Q) --+ L 2 (Q) is the embedding operator, which is by Example 11.30 com-
o 1
pact, and Tl : L 2 (Q) --+ W (Q) is a bounded operator, we obtain that their compo-
sition T is a compact operator.
Example 11.32 Let Q be a bounded region in R n. Prove:
a) that the equation (11.27) from Example 11.31 with the same conditions has
countable many positive eigenvalues AI, A2,"" which as a sequence {An}nEN
tends to +00;
b) that the corresponding eigenfunctions Ul, U2, ... , form a base of the space
W 1 (Q);
c) that the eigenfunctions Ul, U2, ... , form a complete orthogonal system in the
space L2(Q);
d) that holds
(11.35)
Solution.
a), b) By Example 11.31 the problem of eigenvalues of the equation (11.27) is reduced
on the equation (11.28) with a selfadjoint compact positive operator T. Then
there exist a sequence of eigenvalues {a;}iEN of the operator T which converges
to zero and a sequence of eigenfunctions. Taking Ai = .L we obtain the desired
",
conclusions.
c) By (11.34) we have
10 UiUj dx = (T(Ui)luj)E = (ai u ;\71,\g = ai(uiluj)E = ~i (U;\Uj)E. (11.36)
Since for i i- j we have (uiluj)E = 0 we obtain by (11.36) for i i- j
10 UiUj dx = 0,
i.e., {U;}iEN is a orthogonal system of functions in the space L 2 (Q). The com-
pleteness of the system {u;} in the space L 2 (Q) follows by the facts that the
a 1
space W (Q) is dense in the space L 2 (Q) and that the set of all linear combi-
nations of the sequence {U;}iEN is dense in the space L 2 (Q).
d) Follows by (11.36).
Example 11.33 Let {..\;}iEN be the sequence of generalized eigenvalues and {U;}iEN
the sequence of generalized eigenfunctions as a base in L 2 ( Q) corresponding to Ex-
ample 11.32 and u is the solution of the generalized boundary pmblem . Pmve:
a) that for A i- Ai (i E N)
n c; r
U=LA_A.Ui, wherec;= jc.QFuidx,
i=1
o 1
and the series converges in the space W (Q);
b) that for A = Ai (i E N)
where Uil" .. ,ui, are the eigenfunctions for A = A. The series converges in
a 1
the space W (Q).
a 1
Solution. We expand the function U E W (Q) in the following convergent series in
0 1
the space W (Q)
00 (UIUi)
U = L aiui, where at = -.1 (.)'
.=1 U, u,
and we omit the index in the scalar product. By (11.30) and (11.34) we obtain
a 1
(hl<p) = (ul<p) - A(T(u)I<p) (:Ui) = >:(ului).

11.2. TIlE GENERALIZED MIXED PROBLEMS 355
Putting this ill (11.37) we obtain
- 10 F<j5dx = (ului) - ;i (11l u i).
Therefore for A f- Ai
(ulu;) = _Ai_
A - Ai
r FUi dx.
iQ
This implies by (uilu,) = Ai ( see (11.35)) and the expansion L~l aiui the conclu-
sions in a) and b).
11.2 The Generalized Mixed Problems

Example 11.34 Let Q c Rn (n 2: 2) be a bounded region and S c Q a locally
quadratic (n - I)-dimensional surface with the property that for every point Xo E S
there exist regions U(xo) and V and a C1-dipheomorphism q, from U onto V such
that a subset of Q n U is mapped on a n- dimensional parallelepiped P C V, and
S n U is mapped on a side or union of sides of P.
Prove that:
a) there exist constants C1 >0 and C2 >0 such that for every f E Wl( Q) and
every c, 0 < c < 1,
(11.38)
l. e.,
b) If additionally Q is a locally quadratic region, then the operator T : W1(Q) -+

L 2 (Q) given by T(J) = fls is compact.
Solution.
a) There exists a sequence UdkEN from C1(Q) such that
(11.39)
By the definition of the trace of the function Jon S (see Chapter 9.)
fk L2(S)
~
fl
s (11.40)
t
We have (see Chapter 9.)
is IJkl2 dS :::; ~1 kIIkl2 dx + C2 c IDiJkl2 dx.
Letting k -+ 00 we obtain by (11.39) and (11.40) the inequality (11.38).
b) We have to prove that the operator T maps every bounded sequence {Ik} kEN
from W 1 (Q) on a sequence from the space L 2 (Q) which has a convergent
subsequence. Since the embedding operator of the space W 1 (Q) into the space
Lz(Q) is compact there exists a subsequence {In. hEN of the sequence {hhEN
which is convergent in the space L 2 ( Q). We shall prove that Un. Is hEN is the
desired subsequence, which converges in Lz(Q).
By (11.38) we have
1s lIn. - In m12 dS :::; C1 f lIn. - Inm 12 dx+Czc

c iQ
t
j=1
IDj(fn. - InmW dx. (11.41)
Since Un. hEN is a bounded sequence in the space W 1 (Q) we have that for
every 8 > 0 there exists c' > 0 such that
(k,m EN). (11.42)
On the other side, since {In. hEN is a convergent sequence in the space L 2 ( Q)
there exists ko E N such that
eC if IIn. - Inm 12 dx < 2"8 for k,m 2: ko

1
Q
(11.43)
We obtain by (11.41), (11.42) and (11.43)
is lInk - Inml 2 dS < c.

Since L2(5) is a complete space, we have that {Ink Is }nEN converges in the
space L2(S),
Example 11.35 Let Q be a locally quadratic bounded region in Rn and for 0 <
T < +00 we introduce
r T = 8Q x [0, T), Qo = Q x {OJ, QT = Q x {T}.
We suppose that for the uniform elliptic differential operator

n n
L(v) = L, L, Di(aijDjv) + cv
;=1 j=1
11.2. THE GENERALIZED MIXED PROBLEMS 357
foraij = aji E C1(Q),c E C(Q),c S 0, there are known all eigenvalues {AihEN and
eigenfunctions {VdiEN for the classical problem
L(v)+).v=O onQ (11.44 )
vlaQ = o. (11.45 )
Using the Fourier method of separation of variables construct the classical solution
(supposing that there exists)
u E C 2 (Q X (O,T)), u,Dtu E C(Q X (O,T)UQoUfT)
with F E C( Q X (0, T) U Qo U f T ) for mixed problem
D;u - L(u) =F on Q X (O,T) (11.46)

ul Qo = f E C(Qo), (11.47)
Dtul Qo = g E C(Qo), (11.48)
with boundary condition
(11.49)
as a series
I': Ti(t)Vi(X),
00
u(x, t) =
i=1
supposing the "good" convergence of the series, i.e., that we can exchange the order
of the infinite sum and double differentiation with respect to t.
Solution. Since u E C( Q X (0, T) U Qo U f T ) we can expand the function x I-t

u(x, t) into the Fourier series with respect to the eigenfunctions {V;}iEN on Q for an
arbitrary but fixed t,O S t < T,
00
u(x,t) = I':T;(t)Vi(X) (11.50)

i=1
and this series converges in the space L 2 (Q) for every fixed t. Applying the operator
Lon (11.50) we obtain by (11.44)
= I': Ti(t)L(Vi)(X) = - I': Ti(t).iVi(X),

00 00
(L(u))(x, t) (11.51)
;=1 i=1
Applying the operator D; on (11.50) we obtain
00
D;u(x,t) = I':T!'(t)Vi(X). (11.52)

;=1
We expand the function x 1--+ F(x, t) E C( Q) also in a Fourier series for an arbitrary
but fixed t,O ::; t < T,
F(x,t) = fa;(t)v;(x), where ai(t) =

G1
1F(x,t)v;(x)dx.
Q
(11.53)
Putting (11.51), (11.52) and (11.53) in (11.46) we obtain

00 00 00
LT!,(t)Vi(X) + LAiT;(t)Vi(X) = La;(t)v;(x).

i=1 ;=1 ;=1
The preceding equality holds if and only if
Tnt) + AiTi(t) = ai(t) (t E (0, T), i EN). {11.54)
We have by the initial conditions (11.47) and (11.48)

00 00
LT;(O)v;(x) = u(O,x) = f(x) = Lb;v;(x),

;=1 i=1
00 00
LT!(O)Vi(X) = Dtu(O,x) = g(x) = LCiVi(X),

;=1 ;=1
where we have taken the expansions of the functions f and 9 with respect to the
eigenfunctions {Vi hEN. Hence
Ti(O) = b;, T!(O) = Ci (i EN), (11.55)
where
b; = k /v; dx and c; = k gF; dx.
The initial problem (11.54) and (11.55) for A = Ai -I- 0 has the following solution
T;(t) = bicos(';>::t) + %--sin(';>::t)+ ~ ra;(T)sin(';>::(t-T))dT (11.56)

vA; VAi Jo
and for Al =0
(11.57)
So we have obtained the classical solution of the mixed problem (11.46),(11.47),(11.48)

and (11.49) as a series (11.50), where T; is given by (11.56) and (11.57).
Example 11.36 Let Q, f T , Qo, QT and L be same as in Example 11.35. Let a;j =
E C 1(Q),c ~ 0 and /,g E L 2(Q), FE L 2(Q x (O,T)). We shall call a/unction
aji,C

u E WI (Q X (0, T)) the solution oj the generalized mixed problem Jor the hyperbolic
equation (11.46)- (11.49) for' l' = if it satisfies
1 d= t
Q x(O,T) i=1 j=1
aij(x)(Djll)(Dizp) - (Dtu)(Dtzp) - cuzp) dxdt
= ( Fzpdxdt+ ~gzpdS, (11.58)

JQX(O,T) lOa
for every function cp E WI (Q X (0, T)) whose traces on QT and r T satisfy

CPIQT = and CPlrT = 0, (11.59)
and
ul QT = f and ul rT = 0. (11.60)
Prove that
a) ai E L2 (0, T) for ai( t) = fQ F( x, t)Vi( x) dx (i EN);
b) functions TiVi belong to the space W 1 (Q X (O,T)) and
TiVi/Qa = b;vi, TiVdrT = and Ti(O) = Ci;
c) the function TiVi satisfies the generalized mixed problem for F = aiVi, f =
bivi, g = CiVi and l' = 0;
d) the function Uk = 2:::7=1 TiVi (k E N) satisfies the generalized mixed problem
for
k k k
Fk = L aivi, fk =L bivi, gk = L CiV; and l' = 0;
i=l i=1 i=1
e) if the series 2:::~ITivi converges in the space W 1 (Q X (O,T)) to a function u,
then u is the solution of the generalized mixed problem.
Solution.
a) Since
FE L 2 (Q X (0, T)), Vi E L 2 (Q) and ai(t) = k F(x, t)v;(x) dx (i EN).
we have by Cauchy-Schwartz inequality
/ai(t)/:s: (k IF(x,tWdx)~(k IVi(XWdx)~.

Then by Fubini theorem
(T /ai(tW dt:s: ( IF(x, tW dxdt ( /Vi(X W dx,

lo lQX(O,T) lQ
and so ai E L 2 (0, T) for i E N.

b) We have for '\i # (see Example 11.35)

T;(t) = bicos(r;:t)+ ~sin({>:;t)+ ~
v'\, V'\i
l0
ai(r) sin({>:;(t-r)) dr, (11.61)
and for '\1 =

(11.62)
Since ai E L 2 (0, T) there exists a sequence offunctions {a7hEN from C[O, T]

such that
k L2(O,T)
ai ~ ai as k ......; 00. We have for '\i #
Tik(t) = bi cos( {>:;t) + ~ sin( {>:;t) + ~
V'\i V'\i
r a7(r) sin( {>:;(t -
Jo
T)) dr,
and so Tik E C2 [0, TJ, and for '\1 =
We have for '\j #
and for '\1 = I

T 1k (t) = Cl + Jrto ak1 (r) dr.
Applying once more the preceding procedure we obtain
and for '\1 = T;~(t) = a~(t).

Letting k ......; 00 in the preceding equalities and using the Cauchy-Schwartz
inequality we obtain Ti E W 2 (0, T) and
T.ik W2(O,T)
~
T.i as k ......; 00
(.l E N) . (11.63)
Hence for '\i #

T:(t) = -bi,\j sin( {>:;t) + Ci cos( {>:;t) + l ai( r) cos( {>:;(t - r)) dT, (11.64)
and for '\1 =

(11.65 )
and
TI'(t) = -biAi COS( ji:;t)-Ciji:; sin( ji:;t)-ji:; l ai( r) sin( ji:;(t-r)) dr+ai(t)
(11.66)
and for Al = 0
(11.67)
Since for the function Vi E WI (Q) there exists a sequence {Vik} kEN from C l (Q)
such that
W 1 (Q)
Vik - - t Vi as k - t 00,
we conclude by (11.63) that
1 WI (Qx(O,T))
TiVi E W (Q x (O,T)) and TikVik -t Tivi as k - t 00.
Since Tik E C 2[0, T] and Tik(O) = bi we have
We obtain in an analogous way
Since Ti E W2(0, T) we obtain by Example 9.39 Ti E ClIO, T] and then by

(11.64) and (11.65)
T:(O) = Ci. (11.68)
c) Since {Vk hEN are the generalized eigenvalues for the problem
Lv + AV = 0 on Q and VI8Q = 0,
ok
we have (see Examples 11.31, 11.32) for every cp E W (Q)
-ttl aij(DjVk)(Di<P) dx +1CVk<P dx + A1 Vk<P dx = o.

i=l j=l Q Q Q
(11.69)
On the other side we have by (11.66), (11.67), (11.61) and (11.62) for Ti E
W2(0,T)
Tf = -AkTk + ak
By partial integration (see Example 9.40) we obtain
IT(-AkTk+ak)<pdt = lT Tf<pdt = THT)cp(T)- faT T~cp'dt (cp E Wl(O,T)).

(11.70)
We take now a function cP E W 1 ((0, T) x Q) such that it satisfies (11.59). By

Remark 2 from Example 9.28 there exists a sequence {CPm}mEN of functions
from C1((0,T) x Q) such that
'Pm
WI(Qx(O,T
--+
d I
cP an CPm r T =
0
.
Since the trace of the function x 1-+ CPm(x, t), which belongs to C 1 (Q), on DQ
is zero and the function t 1-+ 'Pm(x, t) belongs to GI(O, T] we have by (11.69)
and (11.70)
1= j (t t 1aijDj(TkVk)(Di'Pm) - Dt(TkVk)(DtCPm)
(O,T)xQ ;=1 j=l Q
-1 Q Jo T~(t)DtCPm(x,t)dt) dx - JQo
(Vk(X) fT ~CkVk(X)'Pm(x,O)dS
= Ak loT (Tk( t)VkCPm(X, t) dX) dt - k Vk(X) (loT (AkTk( t) - ah(t) )CPm(X, t)) dt
+T~(T)CPm(x, t)) dx - T~(O)'Pm(x, 0)) dx - k CkVk(X)CPm(x, 0) dx.
By (11.68) we obtain
1=1 akVk'Pm dxdt - T~(T) f__(l. Vk)CPm dS.

QX(O,T) J"QX(O,T)
Taking m -400 we obtain by cplQT = 0 the desired conclusion of c).
d) Follows by c).
e) By the first example in this section the trace of the function

00
u= LT;Vi
;=1
on the set Qo is a limit of the sequence {UdkEN from d) as k -4 00 in the

space L 2 ( Qo). In an analogous way we have that ul rT = O. Since by d) the
function Uk satisfies (11.58), letting k -4 00 we obtain by the first example in
this section and Cauchy-Schwartz inequality that u satisfies (11.58).
o 1
Example 11.37 Prove that if in the generalized mixed problem fEW (Q), then
a) the series I:~l Ti(t)vi(x) converges in the space W1(Q x (0, T)) to function
which is the generalized solution of the generalized mixed problem (11.58))
(11.60) for r = 0;
b) there exists C >0 such that
(1l.71)
Solution.
a) Since 9 E L 2(Q) we have for its Fourier series I:~1 CiVi for Ci =- JQ 9V;dx by
the Parseval equality
00
(11.72)
i=l
We have for the Fourier series

00
F(x, t) =L ai(t)vi(X)
i=1
again by the Parseval equality that almost everywhere with respect to t from
[O,T]
f lai(tW = i IF(x, tW dx.
i=1 Q
Therefore we have by Beppo-Levi theorem
f
i=1
{T
lo
la;{tW dt = {
lQx (O,T)
IF(x, tW dxdt. (11.73)
ok
We shall use that 9 E W (Q) and Examples 11.31, 1l.32 and that the system
{VdiEN is orthogonal in the space L 2 ( Q). Then the system
ok
is orthonormal in the space W (Q), where the scalar product (v Iw) is given
by
(Vlw) = tti
i=1 j=1 Q
aij(Djv)(DiW) dx - iQ
cvw dx (11.74)
(see Examples 11.31, 11.32). The function f has the following }<ourier series
expansion in the space W 1 (Q)
Hence by the Parscval equality
L Ib;(Vil v i)I/212.
00
(fl!) =
i=l
Since by Example 11.32

(11.75)
we obtain
L Ib;l2 Ai =
00
(flf) (11.76)
i=1
We introduce a new scalar product [,1,] in the following way
[Vlw] = 1 (t t aij(Djv)(DiW) - cvw + (DtV)(DtW)) dxdt,

Qx (O,T) i=1 j=1
(11.77)
which by the uniform ellipticity of the considered differential operator induces

a norm which is equivalent to the usual norm in W 1 (Q x (O,T)). The scalar
product (11.77) can be written also in the following form
[Vlw] = faT(v(., t)lw(-, t)) dt
+ faT (10 v(x, t)w(x, t) + Dtv(x, t)Dtw(x, t) dx) dt.

The system offunctions {Vi LEN is orthogonal in the space WI (Q) with respect
to the scalar product (,1,) given by (11.74) and therefore we have for i =I j
[TiV;lTjVj] = faT Ti(t)Tj(t)(VilvJ dt
+ (T(T;(t)Tj(t)+T:(t)Tj(t)).( ( ViVj dx) dt+ (T T;(t)Tj(t)( ( ViVj dx) dt = 0,

Jo JQ Jo JaQ
since vilaQ = 0.
Therefore the system {TiVi} is orthogonal with respect to the scalar product
[,1,] and the series
00
converges if and only if the series

00
L[Tj V iI 7i v ;] (11.78)
i=1
converges. By the defillition of the scalar product [1] we have
[Tiv;JTiv;J = (vilvi) faT 11i(tW dt + (10 IVi(XW dx)(faT(IT;(tW + IT:(tW) dt).

This implies by (vilvi) = ),i and the normdness of the system {V;}iEN in the
space L 2 (Q)
(11.79)
To prove the convergence of the series (11.78) we have to find by (11.79) the
estimations for ITil and mi.
By (11.61), (11.62), (11.63) and (11.64) and the Cauchy-Schwartz inequality
we obtain first for)' =J 0
ITi(tW < 3(lbiI2 + ;i

1~;2 + I faT ai(7)Sin([\;(t-7))d7 1)
< 3 (lbil2 + 1~;2 + ;i (faT lai( 7 W d7)T) ,
and
IT:(t)12 < 3 (lb;J2),i + Ic;J2 + I faT ai(7) cos( [\;(t - 7)) dTI)
< 3 (lbi l2 + ICil2 + (faT lai(7W d7)T) .
We have for ),1 =0

IT1(tW :=; 3 (lbl l2 + ICll2T2 + (faT lal(tW dt)~2) ,
IT;(tW :=; 2 (IClI2 + (faT lal (tW dt)T) .

Putting these estimations in (11.79) and using the fact that the sequence
{I-.f. }iEN is bounded ( since),i ~ +00 as i ~ (0) and (1l.72), (11.73) and
(11.76) we obtain
~ ((),i + 1) faT ITi(tW dt + faT IT:(tW dt)

00
L[TiViITiVi]
i=l
< G' (~lb;J2()'i + 1) + ~ Jcil 2+ ~ faT lai(7)12 d7)

< Gil (1Ifll~'(Q) + IlgllE,,(Q) + 1IFIIL2(QX(O,T))) ,
which implies the convergence of the series (11.78).
366 CIIAPTER 11. FUNCTIONAL ANALYSIS METHODS IN PDEs
b) The inequality (1l.71) follows by the last inequality in a), the Parseval equality
and the inequality
00
lIulI?v'((O,T)XQ) :::; C1 L[TiViITi V ;].

i=1
11.3 Numerical Solutions of PDEs in the Frame-

work of Functional Analysis
Let Bl and B2 be two Banach spaces and A : Bl ~ B2 a linear operator. We shall
consider for f E B2 the equation
A(u) = f. (1l.80)
We introduce two another Banach spaces B 1 ,n and B 2 ,n (n EN), which are

usually finite dimensional, and linear (discretization) operators
Figure 11.1
Then we reduce the equation (11.80) on the approximate equation
(11.81 )
where in = D 2 ,n(f). Supposing the existence and continuity of A;;-1 we can solve
(11.81) Un = A;;-I(fn).
Definition 11.6 (i) A and An are compatible if
11.3. NUMERICAL SOLUTIONS 367
(ii) The numcl'ical pmccdure is it stable if the sequence {A;:;l }nEN is uniformly
bounded.
(iii) the sequence {un} nEN converges to U if
(iv) the procedure is convergent for f E B 2, if for every f and a convergent sequence
{fn}nEN the sequence {Un}nEN, where Un = A;:;1fn' converges to u.

Example 11.38 Provc that if A and An are compatible and the numerical procedure
is stable, then the procedure is convergent.
Solution. Since the procedure is stable there exists a constant M > 0 such that
IIA;:;111 < M (n EN). Therefore we have
IID1,nU - unll IIA;::1(An(D1,nU - un))11
< MIIAn(Dl,n - un)11
< M (1IAn(Dl,nU) - D2,nA(U)11 + IID2,nf - fnl!).
Since the right side tends to zero as n -> <Xl we obtain the desired conclusion.
Exercise 11.39 For the mixed problem for the heat equation
au - aax2u = F
at
2
on (O,rr) x (O,T)
with the conditions

U(x,O) = f(x) (x E [0, IT])
u(O,t) = u(IT,t) = 0 (t E [O,T]),
give the approximate solution by the notions from Preliminaries for
B1 = Co([O, IT] x [0, T]), B2 = C([O, T] x [0, T]) x C[O, rrJ,

B1,n = RS, where s is the number of the points (mh, nk) of the lattice M for the
parallelogram [0, IT] X [0, T] and B 2,n = RS S X RS', where 31 is the number of the
points oj the lattice M in [0, rr].
Hints. Take A = (~- ::2 ,I) for J(u) = u(x,O) and An = (Th,k,h), where
1 2k k k
n,k(U) = k(u(x, t + k) - ((1 - h2 )u(x, t) + h 2 u(x + h, t) + h 2 u(x - h, t)))
and h(u) = u(x,O), x = mho

Exercise 11.40 Prove that for b :.:; ~ the procedure from Eur-cise 11.39 convcr-ges.
Hint. Prove the compatibility and stableness of the preceding procedure and then
apply Example 11.38.
Example 11.41 Let D 2 ,n (n E N) be continuous operators. If
(11.82)
then there exists a constant M > 0 such that
(u E Bz,n EN).
Solution. By (11.82) we obtain the pointwise boundedness of the family of function-

als {IIDz,nl!}nEN. Then by theorem on uniform boundedness we obtain the desired
conclusion.
11.4 Miscellaneous
Let Q be a bounded region in Rn. We denote by C>'(Q) the Banach space of all
Holder continuous functions on Q, i.e.,
ID"'u(x) - D"'u(y)1 :.:; Hlx - yl>'

for H > 0, 0 < ,\ :.:; 1, 10:1 = 1, endowed with the norm
lu(x) - u(y)1
Ilull>. = sup
Ix-yl>'
x,yEQ,x#y
We introduce another Banach space of functions C2+>'( Q), which contains all func-
tions from C 2 (Q) for which the second derivative belongs to C>'(Q). This space is
endowed with the norm
where
IluIIC(Q) = sup lu(x)l.
xEQ
We consider the problem
L(u) =F on Q and ulaQ = 0, (11.83)

1l.4. MISCELLANEOUS 369
where
for aij, ai, a E C(Q), a(x) :s 0 (x E Q) and L is uniformly elliptic, i.e., there exists
c' > 0 such that for all Z E Rn
n n
L: L: a;j(x)z;Zj ~ c'lzl2.
;=1 j=1
For the solution of this problem in C2+A( Q) the following Schauder a-priori estima-
tion holds
(11.84)
where C > O.
The special case L = .6. is uniquely solvable in C2+A(Q).
Let {S( t) h~o be a family of linear continuous operators defined on a Banach
space X and with values in the same Banach space.
Definition 11.7 A family {S(t)h~o is a semigroup of operators if it satisfies:
(i) S(O) = Ii
(ii) S(tl +t 2) = S(tdS(t2) = S(t2)S(t1) for every t l , t2 ~ O.
The semigroup {S(t)h~o is strongly continuous at zero, or Co-semigroup if for

every x E X we have
(iii) 1imt_o+o IIS(t)x - xii = o.

Example 11.42 Consider the equation
La(U) = (1 - a).6.(u) + aL(u) = F, (11.85)
where 0 :s a :s 1, with the boundary condition ulaQ = O.

Specially for a = 0 we obtain the problem
.6.u =F on Q and ulaQ = O.

Let
A= {aIO:S a:S I}, FE CA(Q)
implies that u E C2+ A(Q) is the solution of (11.85). Prove that the set A is
a} closed set;
b) open set.
Solution.
a) Let {ai} iEN be an arbitrary convergent sequence from A. Denote by a the limit
of this sequence. We have to prove that a E A. We denote by Ui the solution
of (11.85) which corresponds to ai. The estimation (11.84) implies
(i EN).
Hence the sequences {Ui};EN, {DUi}iEN and {D 2u;}iEN are equicontinuous.
Then by the Arzela-Ascoli theorem there exists an uniformly convergent sub-
sequence
{Ui,} jEN of {Ui hEN for which also the sequences of first and second deriva-
tives are uniformly convergent. Denote the limit by U E C2+.\(Q). Therefore
we have on Q
F = 3-+
Jim L,,(Ui)
00J J
= L,,(u) and U\8Q = O.
This implies a E A.
b) We shall prove that A is the neighborhood of every its point. For that purpose
we have to find for an arbitrary but fixed ao E A a number c: > 0 such
that (ao - c:,ao + c:) C A. We introduce the family {Wa}aE[O,I] of functions
Wa : C2+.\(Q) -+ C2+A(Q) defined by
where v is the unique solution of the problem
Lao(v) = (a - ao)(~u - L(u)) + F on Q and V\8Q = O. (11.86)
If we prove that W" for a E (ao - c:, ao + c:) has a fixed point, i.e., W" (u,,) = u"
for some a, then it would be u,,\aQ = 0 and (11.86) would imply
Then it would follow L,,(u,,) = F, i.e., the fixed point U a would be the solution
of (11.85).
To prove that Wa has a fixed point we shall find c: > 0 such that for a E
(ao - c:, ao + c:) the map W" is a contraction and then we can apply Banach
fixed point theorem.
Let
UI,U2 E C 2+.\-
(Q), VI = W,,(UI) and V2 = W,,(U2),
where VI and V2 are the corresponding unique solutions of (11.85). Then we
have for their difference
11.4. MISCELLANEO US 371
Therefore by Schauder estimation (11.84) we obtain
IlllIa(ud - llIa(uz)IIH>' IIVI - vzllz+>.

< Cia - aolll.6.(uI - uz) - L(UI - uz)II>.
< CMla - aoiliul - u211HA,
where M > 0 is a constant independent of UI, U2 and C. Taking la - aol < c
for c = 2dM
we obtain
Hence ilia is a contraction for la-aol < c. Therefore ilia by Banach fixed point
theorem has a fixed point Ua for every a E (ao - c, ao + c), and this is the
solution of the problem (11.85). Therefore (ao - c,ao + c) C A, i.e., A is an
open set.
Example 11.43 Prove that for every F E CA(Q) the problem (11.83) is uniquely
solvable.
Solution. Consider instead of the equation L( u) = F the perturbed equation

(11.85), i.e.,
La(u) = (1 - a).6.(u) + aL(u) = F,
where 0 ::::; a ::::; 1, ulaQ = O. Take the set
A={aIO::::;a::::;l}.
F E CA(Q) implies that u E CHA(Q) is the solution of (11.85). The set A is by
Example 11.42 open and closed set. Therefore A = [0,1]. Hence 1 E A, what implies
the desired conclusion.
Exercise 11.44 Every Co-semigroup {S(t)}t~O is continuous for any t > O.
Hints. The right continuity follows easily by the definition.

For the left continuity prove first the pointwise boundedness and then use theorem
on uniform bounded ness .
Exercise 11.45 For the mixed type problem for heat equation
au a u 2
< x < 7r,t > 0),
at - ax z (0
u(O,t) = u(7r,t) = 0 (t > 0),

and initial condition

u(O,x) = f(x) (0 < x < 1r)
for f E L 2 (0, 7r) consider the family of operators {S( t)}t~O as a map t f-+ S( t) with
the domain [0,+00) and range L(L 2 (0,7r),L 2 (0,7r)) given by
S(t)f(x) = u(x, t),

where u is the solution of the considered mixed problem obtained by the Fourier
method of separation of variables (see Chapter 6) given by
=L
00
u(x, t) cn e- n2t sin nx,

n=l
where Cn are the Fourier coefficients of the function f.

Prove:
a) {S(t)}t~O is a semigroup of operators;
b) {S(t)h~o is a Co-semigroup.
Exercise 11.46 Prove that for a Co-semigroup there exist two real con-

{S(t)h~o
stants M > and w > such that
IIS(t)11 ~ Me wt (t E [0,+00)).
Chapter 12
Distributions in the theory of

PDEs
12.1 Basic Properties

In this chapter 0 denotes an open set in R n.
The support of a continuous function c.p : 0 -+ C, denoted by supp c.p, is the
closed set defined by
suppc.p = {x E 01 c.p(x)::f OJ.
An infinitely differentiable function c.p : 0 -+ C is in the set C[(' if it has a
compact support. Then we shortly say that c.p is a test function. The space D( 0) is
the set C[('(O) endowed with the convergence defined below.
A sequence {c.pJ JEN from D( 0) converges to the zero function c.p = 0 if
(i) there exists a compact set J( C 0 such that for all j E N it holds supp c.pj C J(;
(ii) for every multiindex a E Z't and every x E J( it holds lim :Ci
) .... +00 uX"
c.pj(x) = O.
Definition 12.1 A distribution T on 0 is a linear continuous functional on D(O),

where the continuity of T means that for every sequence {c.pj} JEN which converges
to zero in D( 0) it holds
lim T(c.pj) = O.
J ..... +OO
The set of distributions on 0 will be denoted by D'(O).
The set of distributions is a vector space. The value T( c.p) of a distribution T at

a test function c.p is also denoted by (T, c.p).
373

374 CHAPTER .12. DISTRIBUTIONS IN THE THEORY OF PDEs
Definition 12.2 A sequence of distributions {1j}jEN from 1)'(0) converges to an

element l' E V'(O) iJ Jor every t.p E V(O) it holds
(12.1 )
Then we say that l' is a weak limit of the sequence {Tj} jEN .
Every locally integrable function Jon 0 defines a unique 1'1 E V'C 0), such that
(1'1, t.p) = JJ(x)t.p(x) dx (t.p E V(O)). (12.2)
o
Such distributions are called regular. However, the functional 8a , a E 0, given by
(12.3)
is not regular. If in (12.3) a = 0, it will be denoted simply by 8; this is the well

known delta distribution ("delta function").
The support of a distribution l' is the smallest closed set [{ C 0 such that for
every t.p E V( 0), with support in 0 \ [{ it holds (1', t.p) = O. Clearly, the support of
8a is the single point a.
Let a = (aI, a2,"" an) of- (0,0, ... ,0) be a multiindex from Z~. In order to
define the distributional partial derivative of a distribution 1', let us assume first
that a function f = f( Xl, ... ,x n ) has a continuous partial derivative in Xl on R n.
Then for r.p E VCR n) it holds
J af(x) r.p(x) dx = f(x) r.p(x)

-::l-
UXI
100
-00
- J f(x)-::l-
at.p(x) dx = - J f(x)-::l-
UXI
ar.p(x) dx.
UXI
Rn Rn Rn
(See also Section 9.1.) Since both f and its derivative ~f define unique regular
UXI
distribu tions 1'1 and T ~, the obtained equality can be written as
aX1
( D(I,O, ... ,O)TI, en)

r
= _ \1'1> ::lar.p) .
uXI
where D(1,o, ... ,O)Tf is, in fact, the distributional partial derivative in Xl of the distri-
bution TJ . Thus for an arbitrary element l' E V' (0) and a multiindex a E Z't, the
a-th distributional derivative of T, denoted by D"T, is defined by
(12.4)
Since r.p E COO ( 0), we get the essential property of the space of distributions, namely
that every distribution has a distributional derivative of arbitrary order. For the
relation between the distributional and "classical" derivatives,see Example 12.9.
12.1. BASIC PROPERTIES 375
Definition 12.3 The space S(Rn) of rapidly decreasing functions on Rn is the set
of infinitely differentiable functions on Rn such that for all multiindices Q and f3 it
holds
lim Ixc> uX"
Ixl-->+oo
!{3/J'P(x)1 = 0,
endowed with the following convergence:
A sequence {'Pi} iEN converges in S(Rn) to the zero function 'P = 0 iff for all
multiindices Q and f3 and every x ERn it holds
lim x'" !(3/J'Pi(X) = O. (12.5)

)-->+00 uX"
The space of tempered distributions S'(Rn) is the space of linear continuous

functionals on S(Rn), the continuity being defined analogously to (12.1). The space
'D(Rn) is dense in S(Rn). Thus the space S'(Ftn) can be considered as a subspace
of the space of distributions 'D'(Rn). In fact, it holds
The important property of S(Rn) is that the Fourier transformation is a topolog-

ical isomorphism of S(Rn) (see Chapter 8). Thus, in view of the Parseval equality,
the distributional Fourier transform :FT of a tempered distribution T is defined by
and it is also a tempered distribution.

Let f and 9 be locally integrable functions on Rn such that the improper integral
JIf(r)g(x-r)ldr
00
-00
converges for almost all x E Rn and defines a locally integrable function on Rn.
Then the convolution f * g is defined by
(J*g)(x) =
Rn
J f(r)g(x-r)dr (xERn). (12.6)
One can prove that the function f *g is locally integrable and the convolution (12.6)
is commutative.
In the following section, we shall have to deal with the convolution in the space
of distributions. To that end, let us take 'P E 'D(Rn) and then, using the Fubini
376 CHAPTER 12. DISTRIBUTIONS IN THE THEORY OF PDEs
theorem, calculate the following integral:
J(I * g)(x) cp(x) dx

Rn
JJf(T)g(X-T)cp(x)dTdx
R~R~
JgeT) Jf(x-r)cp(x)dxdT
R~ R~
J Jf(X)g(T)cp(x+r)dxdT.
R~R~
Let now f and 9 be two distributions on Rn. Then the upper calculation suggests
us to define the convolution of f and 9 by
(I * g, cp) = (I(x), (g( T), cp(x + T)}} (cp E V'(R n )), (12.7)
provided this relation defines an element from V'(Rn). The problem of existence
of the convolution of two distributions is rather involved. Let us just say that
(12.7) exists if at least one of the distributions f and 9 has a compact support. In
particular, if 9 = Ii, then it holds
for every f E V'(Rn).

(Note that supp Ii = {O}, hence a compact set.)

Example 12.1 Prove that the following sequences converge to the delta distribution
{j (given by (12.3) for a = 0,) in V'(R) :
a) { (1
7r
J
+ J2x
2)} jEN
j
Solutions.
a) Let cp E VCR). For every j E N, the function ( J. 2 2) jf, locally integrable

7r1+Jx
on R (in fact, it is infinitely differentiable on R), hence it defines a unique
regular distribution via the formula (12.2). Thus we have
( j cp) = .!. Joo j cp(x) dx = .!. Joo cp(tfj) dt.

7r(1 + j2x 2 ) , 7r
-~
1 + j2 x 2 7r
-00
1 + t2
By supposition, the support of cp is contained in some interval [- L, LJ, which

implies that the last integral is equal to
L ~
.!. jJ cp(t/j) dt = 2CP(O) . arctan(jL) + .!. jJ cp(t/j) - cp(O) dt. (12.8)

7r 1 + t2 7r 7r 1 + t2
~L ~
Since it holds
~im 2CP(O) . arctan(jL) = cp(O) = (c,cp),
J-OO 7f
we have yet to prove that the last integral in (12.8) tends to zero as j --+ 00.
To that end, we use the mean value theorem and obtain
.!. ji cp(t/j) -
~
cp(O) dt 2 , jJ tdt
7r 1 + t2 < -; 'irlt1 lcp (e)1 . 0 j(1 + t 2 )
-j
= 1 I '()I
-maxcp t In(1 +.PL2)
7r Itl::; J

Slllce l' -
.1m lnj
. = 0, t h i ' ten d s to zero as J. --+
e ast expreSSlOn 00.
J-+OO J
b) Left to the reader.
Remark 12.1.1 One can prove that every distribution can be obtained as a weak
limit of a sequence of test functions.
Exercise 12.2 Construct the sequences {h} jeN and {gj} jeN of locally integrable
functions, which both converge almost everywhere {a.e.} to zero, and the first con-
verges to the delta distribution C in 1)'(R), while the other does not converge at all
in 1J'(R) .
.A:nswer. Let us put, e.g.,
j /2 if Ixl ~ ~, if Ixl ~ ~,
h(x) = { J and J
o otherwise, otherwise.
We have for every x =l 0 :

~im
]-+00
h(x) = ~im
,-+00
gj(x) = 0
which means that these two sequences converge a.e. to zero.

If cp E V'(R), then, using the mean value theorem for definite integrals, we obtain
where (j E [-Iii, Iii]. This implies
Further on, if cp E V'(R), then
J
IJj
(gj, cp) = j2 cp(x) dx. (12.9)
-I/j
Assume that cp is identically equal to 1 in some neighbourhood of zero; then the

expression on the right-hand side of (12.9) does not converge as j - t 00.
Example 12.3 The functional x-Ion V(R) is defined by
(
x
-1 ) _
,cp -
;00 cp(x) - cp(o) H(1 - x) d
x (cpEV), (12.10)
x
-00
where H is the Heaviside function given by
I if x > 0,
H(x) ={
o if x :S O.
Prove that X-I is a distribution on V(R) with the property
X-I. X = 1, (12.11)
where the last equality is in the sense of V'(R).
Remark 12.3.1 In general, there does not exist a definition of the product of ar-
bitrary two distributions, which would generalize the usual product of continuous
functions and would also preserve the commutative and the associative law. How-
ever, it is possible to define the product f . g of a distribution f and an infinitely
differentiab:e function 9 by
(j. g,cp) = (j,g. cp) (cp E V(R)).

Then f . 9 is also a distribution.
Solution. The linearity of X-I is obvious. If a sequence of functions {cpj} JEN from
D(R) tends to zero in the sense of D(R), then, by definition, there exists a compact
set I( C R such that for every j E N it holds supp r.pj C J(. Then we have
Thus we get
l(x-I,r.pj)l::; (maxlr.pj(x)1 +maxlr.pj(x)l) .m(I(),

xEK xEK
where m(l() is the measure of the compact set 1(. Hence, by the the definition of
the convergence in D(R), the right-hand side tends to zero as j -+ 00.
Let us prove now the equality (12.11). If r.p E D(R), then from (12.10) it follows
= I xr.p(x) - (xr.p(x~)(O). H(l - x) dx

-00
Jr.p(x)dx = (1,r.p).
00
=
-00
Example 12.4 Show that the following two distributional products exist:
(x-1.x).0 and x-1.(xo), (12.12)
but are nonequal. In (12.12), the distribution x-I is given by (12.10), and the delta
distribution 0 is given by (12.3) (for a = 0).
Solution. Firstly, let us calculate the distributional product 9 . 0, for 9 E COO(R).

If r.p E D'(R), then it holds
(g. o,r.p(x)) (o,g(x) r.p(x)) = (0, (g. r.p))

(g r.p )(0) = g(O) r.p(0) = g(O) . (0, r.p),
which means that

g. 0 = g(O) o. (12.13)
in the distributional sense.
The function g(x) = x is in COO(R), hence by (12.13) it holds
x 0 = xl x=o 0= 0.
Since for any distribution T it holds 0 . T = 0, we have

X-I. (x . 8) = X-I. 0 = o. (12.14)
In view of relation (12.11), see Example 12.3, and equation (12.13) (for g(x) = 1),
it holds
(X-I. x) 8= 1 8 = 8. (12.15)
Hence from (12.15) and (12.14) we obtain the inequality
(x-I. x) . 8 =f. X-I. (x. 8). (12.16)
Remark 12.4.1 The inequality (12.16) shows that the associative law does not
always hold in 'D'(R). In fact, if the multiplicative product in 1Y(R) is defined as a
generalization of the usual product of continuous functions, the space of distributions
cannot be an algebra.
Example 12.5 Find the distributional products

a) xl' 8(q) (p, q EN); b) eax . 6(q) (a E R, q EN),
where 6(q) is the q-th distributional derivative of 6 for q E N.
Solutions.
a) For cp E 'D(R) it holds
(x p 8(q),cp) = (8(q),x P cp(x)) = (-1)q(8,(x Pcp(x))(q)).
Assume first p > q. Then it holds
(xPcp(x))(q) = t
)=0
(~)
J
(xp)(j)cp(q-il(x)
(12.17)
t
]=0
(~)p(p - 1) (p - j + 1) x p- j cp(q-j)(x).
J
So we have
(-l)q ~ \ 8, e)p(p - 1) (p - j + 1) x p- i cp(q-i) (x) )
(-l)q t (;)P(P - 1) .. (p - j + 1) (x p - i cp(q-i) (x)) Ix=o
o= (0, cp).
Assume next p :S q. Then for all x E R it holds for
for j > p,
for J = p,
and therefore we have from (12.17)
(x".8(q),cp) = (-1)q(8,(x Pcp(x))(q))
(-l)q (8, %(;)p(p - 1) (p - j + 1) x p- j cp(q-j)(X))

+ (-l)q (8, (;)p! cp(q-P)(x))
+ (-lq) (8, .t
]=p+1
(~)(XP)(j)cp(q-j>(X))
J
0+ (-1 )qq( q - 1) ... (q - p + 1) (8, cp(q-p)) + 0
q(q - 1) (q - p + 1)( -1)"(8(q-p), cp).
b) Similarly as in a), we get

q
eax . 8(q) = ~]-a)j8(q-j).
j=O
p(x) 8' = P'(O) 8 + p(O) 8',

where p is an arbitrary continuously differentiable function on R. In particular,
prove that in "D'(R) it holds
H'(x) = 8.
Remark 12.6.1 Assume additionally p(O) = 0 and p'(O) -=I- O. Then note that the
function p is equal to zero on the support {O} of the dis~ribution 8, but still their
product is nonzero.
Example 12.7 Let p be an infinitely differentiable function on R with simple zeros

aI, a2, . .. , am. Prove that the equation with the unknown distribution T
p(x) . T =0 (12.18)
has the same solutions as the equation
(12.19)
Moreover, the solutions of (12.19) (hence alsQ of (12.18)) are of the form
m
T = 'I:.,Cj oai'
j=l
see (12.3), 1IJhere C j , j = 1,2, ... ,m, are real constants.
Solution. Clearly, it is enough to analyze the case when P has only one simple zero
at some point a. In that case, we have to prove that equation (12.18) is equivalent
with equation
(x-a)T=O, (12.20)
and its solution is the distribution T = A oa, for some constant A.
Let us introduce the function PI by
p(x) = (x - a)PI(x). (12.21 )
Then it holds Plea) =I- 0 and the mapping cp f-4 'l/J = PI cp is a bijection from VCR)
onto itself. Putting (12.21) into (12.18) we obtain
(p(x) T,cp(x)) (T, (x - a) PI(X) cp(x)) = (T, (x - a) 'l/J(x))
((x - a) T, 'l/J(x)).
which implies the equivalence of the equations (12.18) and (12.20).

Let us find now the solution of (12.20). To that end, note that the mapping
'l/J f-4 (x - a) 'l/J from the set
A = N E V(R)I a i supp'l/J}
into itself is, in fact, a bijection. Thus for every test function cp with the property
a i supp cp it holds
(T, cp) = 0,
which is equivalent with the statement supp T = {a}. Any distribution T whose
support is a single point a is necessarily of the form
p
T = A 8a + 'I:., Ak 8~k) , (12.22)
k=l
for some constaut.s A and A k , k = l,2,oo.,p. Choose now k E {l,2,oo.,p}. Then

[or cp E D(R) it holds:
(Oikl, (x - a) cp(x))
(-l)k (Oa, ((x - a) cp(X))(k l ;
(-l)k (Oa, (x - a) cp(kl(x) + kcp(k-ll(X);

(_l)k ((x - a)cp(kl(x) + kcp(k-ll(x))lx=a
(-l)k k cp(k-ll(a).
Since there exists a cp E D(R) such that cp(k-1l (a) f. 0, it follows that. the dist.ribution
oikl is not a solution of (12.20), hence also not of (12.18). Thus T from (12.22) is a
solution of (12.18) (for m = 1 and aI = a) if and only if Al = A2 = ... = Ap = 0,
which finally gives us the solution
T = A oa for some constant A.
Example 12.8 Let us denote
ej(x) = exp(2j7l'zx) (j E Z),
DOl the derivation operator in the sense of D'(R) and assume that for the sequence
{Cj} jEZ of complex numbers there exist a positive constant A and a natural number
k such that
(12.23)
a) Prove that the sequence of functions

m
fm(x) = L cjej(x) (m EN),
j=-m
converges in D'(R) as m -7 CXJ to the distribution
f(x) =L cjeJ(x). (12.24)

jEZ
b) Prove that for a E Nand f from (12.24) it holds
DOl f(x) = L(2j7l'Z)OI cjej (X), (12.25)

jEZ
where the convergence in (12.25) is in the sense ofD'(R).

c) Find the sum of (12.24) in VI(R), if Cj = 1 for all j E Z.

Solutions.
a) Let us us start from the sequence of functions {jm,k+ 2 (x)}mEN (k from (12.23)),
where
m c.
fm,k+2(X) = 2:= (2 J)k+ 2Cj(x) (m EN).
O#jeo-m J 7rl
In view of (12.23), this sequence uniformly converges on every compact set
J( C R, hence its limit is a continuous function on R; let us denote it by Fk+2.
The distributional derivative of order k + 2 of the function Fic+2 is
In other words, the distribution f, given by (12.24), is the limit of the sequence
{aD + fm,k+2(X)}mEN in VI(R).
b) Since for Q' E Nand m E Z+ it holds
m
D" fm(x) = 2:= (2j7rl)"Cj(X),
jeo-m
part a) implies that the sequence offunctions {D" fm}mEN converges in "VI(R)
to the distribution D" f and thus (12.25) holds.
c) From part a) it follows that the sequence Ltm Cj(x) } mEN converges in "VI(R)
to a distribution which we denote by g. Then we have in the sense of VI(R) :
m
(1- Cl(X))g(x) = lim (1 - Cl(X))

.
2:= Cj(X) = lim (C_m(X) - Cm+l(X)).
Jeo-m
m~~ m-~
(12.26)
Let us prove next that the last limit is equal to 0 in VI(R). To that end, let
us analyze the difference Lm (x) - C m +l (x) :
Lm(X) - Cm+1(x)), cp(x)) = j c- 21rmxcp(x) dx - j c 27r (m+l).xip(x) dx

R R
= -l
27rml
jc- 21rmxcpl(x)dx+ 27r( m1+ l)l jc 21r (m+l).xcp'(x)dx
R R
Thus we have
I(Lm(X) - cm+l(x),ip(x))1 :; - 12
7rm
j Icpl(X)1 dx+ 27r (1m + 1) j lip/(X)1 dx :; mC,
R R
for some constant C = C( rp). The last right-hand side tends to 0 as m -+ 00,
which implies that the right-hand side of (12.26) tends to zero in 1)'(R) as
m -+ 00. Thus we obtained that the sought after distribution 9 is the solution
of the equation
(12.27)
For Ixl < m, the solutions of the equation e- 27r x = 1 are the integers j such
that Ijl < m. Example 12.7 tells us that the solutions of (12.27) in 1)1 ( -m, m)
are exactly the solutions of the following equation
('ifJ=-m+l
(x - j)) .g(x) = o. (12.28)
The same example gives us the solution of (12.28) in 1)1 ( -m, m)

m-l
g(x) = L Aj c5j (x),
j=-m+l
while in 1)'(R) the solution of (12.27) is
g(x) = L Aj c5j (x).

jeZ
We next show that all constants Aj, j E Z, are equal to a single constant C.
To that end, note that for a test function rpm such that
supp rpm C (m - 2/3, m + 2/3) and rp(x) = 1
on the interval (m - 1/3, m + 1/3), it holds
(g, rpm) = / L Aj c5 rpm) = (Am c5m, rpm) = Am. (12.29)

\ez
j
From the other hand, 9 is I-periodic, i.e.,
(g,rp) = (g(x),rp(x -1))
for every test function rp. But then it follows from (12.29):
Am = A m- 1 = C for every mE Z.
So we get
g(x) = C L c5j (x).
jeZ
Example 12.9
a) Let f be a contin uous function on R \ { a }, which is also continuously differentiable

on the inter'vals (-00, a] and [a, +00). Prove that
where D f denotes the distributional derivative of the function f, while Tf' is

the l'egular distribution defined by the classical derivative l' of f, see equation
(12.2). As usual,
[fJa = f(a+) - f(a-)
is the jump of f at the point a.
b) If f is a piecewise continuously differentiable function on R with isolated dis-

continuities at the points aj, j E J, J a finite or infinite subset ofN, then
Df = TI' + L [flu, b
aj .
jEJ
Remark 12.9.1 In a), by assumption, l' exists and is continuous on the set R \ {a},
but not in the point a. Since a point is a set of measure zero, the classical derivative
l' of the function f defines a locally integrable function on R.
Solution. Clearly it is enough to prove part a). To that end, for <p E V(R) we
have using (12.4)
J f(x)<p'(x)dx - J f(x)<p'(x)dx
a +00
(Df,<p} = -(j,<p'} = -
-00 a
J1'(x) <p(x) dx - f(x) <p(x)i: + J1'(x) <p(x) dx

a +00
- f(x) <p(x)[~ +
-00 a
(Tj',<p) + (J(a-) - f(a-)) <p(a).
Since <p(a) = (ba,<p(x)), we obtain the statement.
Example 12.10 Find all distributions y such that

a) y' = 0; b) y(m) = 0, mEN.
Solution. a) Let y E V'(R) be a solution of the given equation, i.e.,
(y',<p) =0 forevery <pEV(R). (12.30)

Let X be a fixed test function with the property
J x(x) dx =
+00
1.
-00
Now every cp E V(R) can be written in the form
J cp(x)dx + 1j/(x)
+00
cp(x) = X(x) (x E R)
-00
for some test function 1jJ, depending on cp (prove that). Then we have
(y, cp) = (y, X(x) . T

-00
cp(x) dx + 1jJ'(X)) = T)
-~
cp(x) dx (y, X) + {y,1jJ'}.
By assumption we have
0= -(y',1jJ) = (y,1jJ');
J 1jJ(x) dx, we obtain

+00
then putting C =
-00
J cp(x)dx=(C,cp)
+00
(y,cp)=C
-00
for every cp E V' (R). Thus the sought after solution is

y = C,
Answer. b) y(x) = Co + Clx + ... + Cm_lXm-l, where C l , C 2 , ... ,en are
arbitrary constants.
Exercise 12.11 Show that the linear first order ODE
y'(x) + p(x) y = q(x),
where p and q are infinitely differentiable functions on an interval (a, b) has only
classical solutions in 1)' (a, b), which are also infinitely differentiable functions.
Exercise 12.12 Let A(x) = [ai,j(x)]i,j=l denote an infinitely differentiable matrix
on R, i.e., for every i,j E {1,2, ... ,n} the function ai,j = ai,j(x), x E R, is in
COO(R). Prove then that the system of ODEs
dy
dx = A(x)y(x), (12.31)
where y( x) = (Yl (x), Y2( x), ... , Yn( x)) is the unknown column-vector, has only clas-
sical solutions in the space of distributions 'D'(R).
Exercise 12.13 Prove that the .qeneral solution of the equation

xny(m) = 0, n > m, ( 12.32)
in 1)'(R) is the distribution
m-l n-l
y(x) = L (ai H(x) x m- i - 1 + bi xi) +L Ck H(x) tr- k ,
i~ ~m
where aj, bj, j = 0,1, ... , m - 1, and Ck, k = m, m + 1, ... , n - 1, are arbitrary
constants.
Example 12.14 Prove the following equalities for every f E 1)'(R) :
a) 6a *f = f (a E R)j b) Dm6 * f = D(m) f, mEN.
Remark 12.14.1 One often meets the equality in a) in the (formal) form
Jf(y) 6(x - y) dy
+00
= f(x) (x E R).
-00
Solutions.
a) Let cp E 1)(R). Then it holds
(6 * f,cp) = (J(x), (6(T),cp(X + T))) = (J,cp).
b) For cp E 1)(R) and mEN it holds

(D m 6*f,cp) = (J(x), (D m 6(T),cp(X+T))) = (-lr (J(x),cp(m)(x)) = (Dmf,cp).
Example 12.15 Let P(x, D) be the linear differential operator
m
P(x,D) = Lai(X)Di, (12.33)
i=O
where aj E COO(R), j = 0,1, ... , m, and D is the distributional derivation operator.
a) Prove that the solution of the following equation in 1)'(R)
P(x, D)6 * u = (~ai(X)Di6) * u = 6

is the function
u(x) = H(x) v(x),
where H is the Heaviside function, while v E cm(R) is the solution of the
initial value problem
P(x, D)v = 0, v(O) = v'(O) = ... = v(m-2)(0) = 0, v(m-l)(o) = 1.
b) Suppose that each aj, j = 0,1, ... , m, is a constant; then for the operator given
by (12.33) we put simply P(D). Assume F is a di8tribution with support in
[0,00). Prove then that the solution of the equation
P(D)o*u=F,
is the distribution
u = (H v) * F,
where v is the Junction defined in part a).
Solution.
a) Firstly we find the distributional derivatives of u = E = Hv.
E'(t) H(t)v'(t) + (H(O+) - JI(O-))o(t) = H(t)v'(t),
E"(t) H(t)v"(t),
E(m-l)(t) H(t)v(m-l)(t),
E(m)(t) H(t)v(m)(t) + oCt).
Therefore
P(D)E = H(t)P(D)v + oCt) = o(t).
Specially for m = 1 we have
E(t) = H(t)e-aot/al.
b) Left to the reader.
Exercise 12.16 Let a cylindrical, homogeneous isotropic rod of length f with insu-
lated sides be on initial temperature 0 C. For times t > 0, there is located a cross
section x = b (0 0, < x < e.
Hint. The physical assumptions give the following mathematical problem:
2 f} 2 u f}u
a -
f}x 2
= U
~t - ql Db (0 < X < e, t > 0),
u(O,t) = u(e,t) = (t > 0), (12.34)
u(:.::, 0) = (0 < x < f).

where Ob is the delta function with support in b. The solution of this problem is
n 2 7r 2 a 2 t
u(x t) -
2qe 1- 00 exp( - e ) . sm. -
n7rb . n7rX
sm-
, -
_1_ ' "
a 2 7r 2 ~
n=!
n2 e e'
where q1 = J( q, J( a constant depending on the material the rod was made from.
390 CHAPTER 12. DISTIUBUTIONS IN THE THEORY OF PDEs
12.2 Fundamental Solutions

A distribution E is the fundamental solution of a differential operator P( x, D), where
P(x, D) = L aQ(x)D'\
IQISm
if it satisfies the equation
P(x, D)E = o.

Example 12.17 Lei the function f : R2 ---+ R be given for (x, y) = x + zy by
1 1
f(x+zy)=_'_-'
7r x + zy
Prove that
a) the function f is locally integrable on R2, and so it defines a distribution on R2;
b) the function f is the fundamental solution for the Cauchy-Riemann operator
Solution.
a) Since we have If(z)1 = ~ . I~I for z = x + zy, it follows that f E Lfoc(R2).
b) We have to prove that

< lif,ip >= ip(O). (12.35)
To that end, we start from the equality
< -o,'P >= - < -

f,o'P 1
>= - -
27r
J- (o'P
ox
+ zo'P)
- - dxdy
-.
oy X + zy
R2
Changing the variables x = rcost, y = rsint, 'Pl(r,t) = 'P(rcost,rsint), we

obtain
< -of,'P >= - - 1 00
1271" 1
(O'Pl
- - -.Z -O'Pl) drdt.
27r 0 0 or r at
Therefore we have
1 1211"
< lif, 'P >= - -
27r 0
+ -27rz 1211"
('P(oo, t) - 'Pl(O, t)) dt
0
('P(r,27r) - 'Pl(r, 0)) dr.
By the compactness of the support of the function 'P and hence also of 'Pl and
the periodicity of 'Pl, we obtain (12.35).
12.2. FUNDAMENTAL SOLUTIONS 391
Example 12.18 Prove that the function E : R x R -+ R given by
E(x, t)
1
= -H(t -Ixl) = {~2 if t > Ix I, (12.36)
2 0 if t < lxi,
is the fundamental solution of the one-dimensional wave equation. In other words,
it holds
a2 E a2 E
at 2 - ax 2 = 5.
As before, H is the Heaviside's function, i.e., the characteristic function of the
interval (0, +(0).
Solution. Let us note first that E from (12.36) is a locally integrable function on
the set 0 = {(x, t)1 x E R, t > O}, and thus defines a regular distribution which we
simply denote also by E. For any cp E V(R) it holds
/ a2E E)a2 +00 +00 fJ2 +00 t a2

\ at2 - ax2' cp = ~ j j at~ dt dx - ~ j j ox~ dx dt
-00 Ixl 0 -t
~ +j()()(oCP(X,t)jt=oo) dx-~ +jOO(acp(x,t)jx=t) dt

2 at t=lxl 2 ox x=-t
-00 0
+00 0
_~ j ocp(x, x) dx _ ~ j ocp(x, -x) dx
2 at 2 at
o
+00 +00
_~ j ocp(t, t) dt +~ j ocp( -t, t) dt.
2 ox 2 ox
o 0
Putting y = -x in the second integral and taking y instead of x in the other three
integrals of the last right-hand side, we get
( 12.37)
From (12.37) it finally follows
/0 2 E a2 E ) 1 1
\ at 2 - ox 2 ,cp = "2CP(O,O) + "2CP(O,O) = (5,cp).
Exercise 12.19 Find the fundamental solution of the two-dimensional wave equa-
tion.
Hint. Take the function

E2(X, t) = H(t - Ixl) ,
27rJt 2 -lxl 2
where for x = (X!,X2) E R2, we put Ixl 2 = xi + x~.
Example 12.20 Let
1
E(x) = O"n
(2 - n )1 x In- 2 for n ~ 3,
where O"n = 27r n / 2 /f(n/2).

Prove that
a) E is a locally integrable function;
b) E is the fundamental solution for the Laplace operator .6..
Solution.
a) There exist M > 0 and a < n such that
M
IE(x)1 ::; r;;F'
where the integral 1 dx
B(O,s)
- converges f or arb'Itrary s > 0, Slllce
xa
.
1
B(O,s)
dx
xa
j ( 1 dS
a ~B(O,r) x
x)
a
dr = j ( 1 dS) dr
a ~B(O,r)
= O"n I s
r
n-l-a
dr = O"n--'
sn-a
n-a
a
b) By definition of the distributional derivative and Lebesgue convergence theorem

we have for every r.p E V and 0 < E; < S
< .6.E,r.p >=< E,.6.r.p >= 1E.6.r.pdx = lim

e-tO+
1 E(X).6.E;(X)dx
Rn e$lxl$s
and supp r.p C B(O, s). Applying the classical symmetric Green formula on the
region B(O,s) \ B(O,E;), since E E C=({xllxl ~ E:}), we obtain
1 (E.6.r.p - r.p.6.E) dx = 1 (Er.pon - OE)

on
dS r.p
e$lxl$s 8B(0,s )U8B(O,.)
88(0,e)
J (E--r.p-
or.p OE)
on on dS.
12.2. FUNDAMENTAL SOLUTIONS 393
Therefore we have
J ED.<p dx J <pD.E dx + J (E:: - <p ~!) dS,

<SlxIS-
=
.Slxls- 8B(O,<)
(12.38)
since <p = 0 on oB(O,s). Let now Ixl = r. Then
~ (_1_) = (2 _ n)Xir -
OX; r n - 2
n
and
~ (_1_)
n 2
= (2 _ n)r- n + (2 _ n)x.(-~)2x.r-n-2.
OX;2 r - ' 2 '
Therefore we have
n
D.2 - n)anE) = (2 - n)nr- n - n(2 - n)(L:xDr- n- 2 = O.
;=1
Then we obtain by (12.38)
< D.E,= lim

<--+0+
J (E~ . Using the polar
coordinates we have that dS. = c: n - l dSl and so
J (E<Por - <p OE) dS< = J 2- o<p c: n c: n - l dSl

or or
8B(O,.) r=.
- f <p(c:, Ot, ... , On-l)(2 - n)c: l - n )-1 dSl
r=E
=
r=.
Jc: ~~ dSl + (n - 2)r=.J<p(c:, 0 1 , ... , On-d dSl. (12.39)
Since
we obtain Jc: o<por dS

r=e
l --t 0 as c: --t 0 + . We have by Lebesgue theorem on
convergence
as c: --t 0 + . Therefore letting --t 0+ in (12.39) we obtain
< D.E,= hm . J

<--+0+
(o .
8B(O,.)
a) H(t)
-(-)-
2Ti n
J exp( -a
2Izi 2i + zzx) dz = H(t)
fit
(2a\ Tit)n
exp
(lxI2)
--2
4a t
;
Rn
b) E( x, t)
H(i)
= (2av;;:t)n exp
(lxI2)
- 4a t ;2
is the fundamental solution for the heat equation, i.e., it satisfics the cquation
au
at - a /lu = 6(x,t).
2
Hint. Apply the distributional Fourier transform, which gives the equation
au -
- a 2A _ CE
at tiU - U
Using the equality 6(x, t) = 6(x)6(t), the form of a solution of the obtained
ordinary differential equation in the space of the distributions (see Example 12.15),
the inverse Fourier and a) deduce the desired form for E.
Exercise 12.22 Assume that >-j((), j = 1, ... , k, are continuous real functions in
( = (~l' ... , ~n-d ERn-I. Prove that there exists a measurable function
<l> : Rn-l --+ [-k - 1, k + 1'], such that
min {1<l>(O - >-j((m 2:: 1 for all ( ERn-I.

l~j~k
Example 12.23 (Malgrange - Ehrenpreiss) Every linear partial differential op-

erator with constant coefficients P(D) has a fundamental solution, i. c., there exists
a solution of the equation
P(D)u = 6.
Solution. Let us define
K(x) = J J exp(2Tizx'O(P(O)-ld~nd(,
Rn-I ~(Cn)=q;(e)
where, as before, ~ = (~l'''' '~n-ll ~n) = ((, ~n)' while <l> was defined in Example
12.22.
We shall show that J( is well defined, and, moreover, is the desired fundamental
solution of P. To that end, put first
12.2. FUNDAMENTAL SOL UTIONS 395
where N is a (sufficiently big) natural number. Next, put
I<N(X) = J J exp(27flX O(PN(~)tl d~n de',

Rn-J !S(n)=~W)
where the function n/2.
Next, let us prove
PN(D)I<N = o.
Assume rp E veRn). Since the adjoint operator to PN(D) is PN( -D), it holds
J J (PN ( -Xi))-l
Rn-J !S(n)=~(e)
.J exp(27flX O(PN ( -D))rp(x) dx d~n d~

Rn
Rn-J
J J
!S(n)=~(')
Frp( -0 d~n d~
JFrp(-Od~ = rp(O) = (o,rp).

Rn
Note that we used the equality
rp(O) = JFrp(O d~,

Rn
which follows using the inverse Fourier transformation.

So we obtain
0= PN(D)I<N = PN(D)(1 - 6.)N I<N.
Now I< = (1 - 6.)N [(N, hence [( is the fundamental solution of the operator P(D).
Exercise 12.24 (L. Nirenberg) Let
P(D) = L a"D" (12.40)

iai:Sm
be a linear partial differential operator with constant coefficients. Then for any .given
function f from Cgo there exists a solution u in C= of the equation
P(u) =.f.
Answer. The solution u is given by t,he convolution
u = J( * f,
where J( is the fundamental solution of the partial differential operator P{D) from
Example 12.23.
Bibliography
[1) Andrews, L. C., Elementary Partial Differential Equations, Academic Press

College Division, Orlando, Florida, 1986.
[2) Bitzadze A. V., Equations of Mathematical Physics, Moscow, 1982 (in Rus-
sian).
[3) Budak, B. M., Samarski, A. A., Tikhonov A., The Exercises in Mathematical
Physics, Nauka, Moscow, 1972 (in Russian).
[4) Carol, R. C., Abstract Methods in Partial Differential Equations, Happer and
Row, 1969.
[5) Colombeau, J. F., Elementary Introduction to New Generalized Functions,

North Holland, 113, 1985.
[6] Courant, R., Hilbert, D., Methods of Mathematical Physics I, II, Interscience
Publishers, 1953, 1962.
[7) Epstein, B., Partial Differential Equations, McGraw- Hill, New York 1962.
[8] Friedman, A., Partial Differential Equations, Holt, Reinhart and Winston,
1969.
[9] Garabedian, P. R., Partial Differential Equations, John Wiley & Sohns, Inc.,
1964.
[10] Gelfand, I. M., Shilov, G. E., Generalized Functions I, II, III, Fizmatgiz, 1958
(in Russian).
[11] Gilberg, D., Trudinger, N. S., Elliptic Partial Differential Equations of Second
Order, Springer-Verlag, Berlin, 1977.
[12] Greenspan, D., Introduction to Partial Differential Equations, McGraw-Hill,

New York, 1961.
[13] Hellwig, G., Partielle Differentialgleichungen, Teubner Verlag, Stuttgart,

1960.
397
398 BIBLIOCHAPIIY
[14J Hormander, L., The Analysis of Linea!' Partial D~fJerenlial Operators I-IV,
Springer-Verlag, Berlin, 1983-1985.
[I5J Hubson, V. C. L., Pym, J. S., Applications of Functional Analysis and Oper-
ator Theo!'y, Academic Press, 1980.
[16] John, F., Partial Differential Equations, Springer-Verlag, 1982.
[17] Ladyzcnskaya, O. A., Uraltseva N. N., Linea7' and Quasilinear Elliptic Equa-
tions, Academic Press, New York, 1968.
[18] Lakshmikantham, V., Leela, S., Differential and Integral Inequalities, Aca-
demic Press, New York, 1969.
[19] Mihajlov, V. P., Partial Differential Equations, Nauka, Moscow, 1983 (in Rus-
sian).
[20] Pap, E., Null-Additive Set Functions, Kluwer Academic Publishers, Dor-
drecht, 1995.
[21] Pap, E., Complex Analysis through Examples and Exercises, Kluwer Academic
Publishers, Dordrecht, in print.
[22] Pinski, A. M., Partial Differential Equations and Boundary Values with Ap-
plications, McGraw-Hill, New York 1991.
[23] Renardy, M., Rogers, R. C., An Introduction to Partial Differential Equations,

Springer-Verlag, 1993.
[24] Rubinstein, Z., A Course in Ordinary and Partial Differential Equations, Aca-
demic Press, 1969.
[25] Schmeelk, J., TakaCi, Dj., TakaCi, A., Elementary Analysis through Examples
and Exercises, Kluwer Academic Publishers, Dordrecht 1995.
[26] Schwartz L., Theorie des distributions I, II, Paris, 1950-195l.
[27] Smoller, J., Shock Waves and Reaction-Diffusion Equations, Springer-Verlag,

1983.
[28] Sobolev, S. L., The Equations of the Mathematical Physics, Nauka, Moscow,
1966 (in Russian).
[29] Stepanov, V. V., Ordinary Differential Equations, Fizmatgiz, 1959 (in Rus-
sian) .
[30] Strang G., Introduction to Applied Mathematics, Wellesley-Cambridge Press,

1992.
BIBLIOGRAPHY 399
[31) Taylor M.E., Partial Differential Eql1,ations I,II, III, Springer-Verlag, 1996.
[32) Tikhonov A., Samarski A. A., Equations of Mathematical Physics, Pergamon

Press, New York, 1963.
[33) 'rreves, F., Basic Linear Partial Differential Equations, Academic Press, 1975.
[34) Vvcdensky, D., Partial Differential Equations with Mathematica, Addison-

Wesley, 1993.
[35) Vladirnirov, V., S.,Equations of Mathematical Physics, Nauka, Moscow, 1976.
[36) Wloka, J., Partielle Differentialgleichungen, Teubner, Stuttgart, 1982.
[37] Zeidler E., Nonlinear Functional Analysis and its Applications, Vol. IV: Ap-
plications to Mathematical Physics, Springer-Verlag, New York, 1988.
Index
Co-semigroup, 369 for a second order PDE, 52

of a first order PDE, 18
Adjoint Theorem, 324 Closed Graph Theorem
ball function, 108 on llilbert space, 305, 307
Banach fixed point theorem, 323 on normed space, 322
Banach space compact support, 256
reflexive, 322 complete solution
Banach-Steinhaus theorem, 322 of first order PDEs, 35
Beppo-Levi theorem, 249 convergences pointwise, 90
Bessel convolution
equation, 108 of distributions, 376
functions, 109 of functions, 268, 375
bilinear form
coercitive, 330 D'Alambert's formula, 71
boundary value problem, 2 Darboux equation, 84
first, 143 delta distribution, 374
second, 143 delta net, 256
third, 144 Diagonal Theorem, 306
differential operator
canonical form hypoelliptic, 340
for a quasi-linear second order PDE, of order 2k, 329
51 dipheomorphism, 260
for a second order PDE, 51 Dirichlet boundary condition, 69
Cauchy integral Dirichlet problem, 143
for a first order PDE, 17 exterior, 163
Cauchy problem, 2 on cylinder, 161
n-dimensional, 81 distribution, 373
for a first order PDE, 17 tempered, 375
for a second order PDE, 52 distributional
one-dimensional, 71 derivative, 374
parabolic equation, 183 Fourier transform, 375
Cauchy-Kowalevskaya theorem, 12 divergence theorem, 80
Cauchy-Riemann equations, 5
change of variables, 252 eigenfunctions, 96
characteristic curve eigenvalue, 96, 312
401
402 INDEX
eigenvector, 312 mixed problem, 359

elliptic second order PDE, 49 solution, 345
energy integral, 137 solution of mixed problem, 359
Euler movement equation" 16 generalized derivative, 279
Euler- Poisson- Darboux equation, 85 Green
exchange formula, 268 operators, 336
Green formula, 170
first integral first (anti symmetric), 82
of a first order PDE, 18 second (symmetric), 82
first order PDE, 17 Green function, 167
characteristic curve, 18
first integral, 18 Hormander theorem, 340
linear, 17 Hadamard example, 9
quasi-linear, 17 harmonic function, 144
Fourier Harnack
coefficients, 90 first theorem, 180
integral theorem, 107 second theorem, 180
inverse transform, 268 heat equation, 70
method, 92 one-dimensional, 6
series, 90, 304 heat flux, 223
transform, 267 Heaviside's function, 391
Fredholm alternative, 313 Helmholtz equation, 151, 180
first, 314 Hilbert space, 303
forth, 319 hyperbolic second order PDE, 49
second, 317
integral
third, 317
Lebesgue, 249
Fubini theorem, 250
Lebesgue on surface, 261
fundamental solution, 390
integral surface, 17
of Laplace equation, 163
of the one-dimensional wave equa- Kirchoff formula, 87
tion,391
of the two-dimensional wave equa- Lagrange-Charpite method, 36
tion, 391 Laplace equation, 5, 69
Lax-Milgram theorem, 334
G~rding Lebesgue theorem on convergence, 250
hyperbolicity condition, 271 Legendre
G~rding equation, 107
inequality, 330 polynomial, 107, 161
Gauss-Ostrogradsky theorem, 80, 82 linear first order PDE, 17
general solution linear operator, 303, 321
of first order PDEs, 36 adjoint, 305, 321
generalized bounded, 304, 321
Dirichlet problem, 330 closed, 304, 321
INDEX 403
compact, 305 quasi-linear

symmetric, 312 first order PDE, 17
linear second order PDE, 49 second order PDE, 49, 64
Liouville equation, 138
locally quadratic (n - 1)-dimensional rapidly decreasing function, 274, 375
part of surface, 260 real analytic function, 12
locally quadratic region, 260 region, 1
starshaped, 287
maximum principle regular distribution, 374
elliptic equation, 163 Rellich theorem, 285
parabolic equation, 183 Riccati equation, 138
strong, 163, 173 Riemann's method, 130
Maxwell equations, 16 Riesz representation theorem, 304
mean value theorem, 174 Ritz approximate solutioll, 342
mixed type problem, 2, 93, 193
scalar product, 303
mollifier, 256
Schauder a-priori estimation, 369
multi-index, 1
Schrodinger equation, 16,274
Neumann boundary condition, 70 second order PDE
nonlinear first order PDE, 35 elliptic, 49
norm, 250 hyperbolic, 49
normed space, 251 linear, 49
parabolic, 49
order of a PDE, 1 quasi-linear, 49
semigroup of operators, 369
parabolic second order PDE, 49 sine-Gordon equation, 139
Parseval identity, 270, 304 singular solution
partial differential equation, 1 of first order PDEs, 36
PDE slowly increasing function, 275
hyperbolic, 268 Sobolev
linear, 2 lemma, 285
of first order, 17 space, 285
quasi-linear, 2 solution of a PDE, 1
uniformly elliptic, 329 spheric
Pfaff's equation, 32 coordinate system, 160
piecewise continuous function, 89 function, 108
Poincare's inequality, 135 inversion, 168, 180
Poisson equation, 69 Sturm-Liouville
Poisson formula, 88 series, 106
for wave equation, 81 Sturm-Liouville problem, 96, 106
Poisson integral formula, 144 regular, 106
potential equation, 69 superposition principle, 97
Projection theorem, 342 support
404 INDEX
of a continuous function, 256, 373

of a distribution, 374
test function, 373

trace, 297
Uniform Boundedness Theorem, 305,

322, 328
Urysohn property, 308
Volterra's method, 137
wave equation
n-dimensional, 4
homogeneous, 71
nonhomogeneous, 71
one-dimensional, 4, 69, 71
weak convergence, 322
in Hilbert space, 304
weak limit of a sequence, 374
Weierstrass test, 90
well-posed problem, 2
Weyllemma,341

PDJ Endre Pap, Arpad Takaci

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PDJ Endre Pap, Arpad Takaci

Uploaded by

Copyright:

Available Formats

Partial Differential Equations through Examples and Exercises

Kluwer Texts in the Mathematical Sciences

A Graduate-Level Book Series

ISBN 978-94-010-6349-4 ISBN 978-94-011-5574-8 (eBook)

Printed an acid-free paper

AII Rights Reserved

2 First Order PDEs 17

3 Classification of the Second Order PDEs 49

3.3 Wave, Potential and Heat Equation . . . . . . . . . . . . . . . . . .. 69

5 Elliptic Equations 143

6 Parabolic Equations 183

7 Numerical Methods 227

8 Lebesgue's Integral, Fourier Transform 249

9 Generalized Derivative and Sobolev Spaces 279

10 Some Elements from Functional Analysis 303

11 Functional Analysis Methods in PDEs 329

11.3 Numerical Solutions . . . . . . . 366

12 Distributions in the theory of PDEs 373

Novi Sad, April 1997 ENDRE PAP

N set of natural numbers

1.1 Basic Notions

We are using the convention 00 = 1. A partial order in Z+ is defined by

0' 2': (3 whenever O'k 2': (3k for k = 1, ... , n.

We introduce the differentiation symbol Dk = aD , and by a gradient vector of

E. Pap et al., Partial Differential Equations through Examples and Exercises

U = U(Xl, ... , Xn) as the vector

The general partia.l differential operator DOl of order m = 10'1 is then

Classification of problems with PDEs.

1) Cauchy problem or initial-value problem. For a given PDE on a region Q,

2) Boundary value problem. For a given PDE on a region Q some additional

A problem with a PDE is well-posed in a class of functions C, if the following

(i) there exists a solution in C;

(ii) the solution is unique;

1.1.2 Examples and Exercises

(::f + (::f = l-u 2 and u(O,y) = cosy.

Figure 1.3 u(x, y ) = -"21 x 2 +"21Y 2

which, using the Laplace operator 6 given by

can be written in the shorter Jorm

b) the Laplace equation

if, additionally, u and v are Jrom the class C 2(Q).

a) Since the function J is analytic, there exists

1'(z) = lim J(z + h) - J(z)

J '( ) _ ou(x, y) ov(x, y)

Now let ih --. 0 for real h. We have for h =I- 0

J(z + zh) - J(z) J(x + z(h + y)) - J(x + zy)

u(x,!} + h) -u(x,y) v(x,y + h) - v(x,y)

f '( ) - _ 8u(x,y) 8v(x,y)

b) Differentiating the first Cauchy-Riemann equation with respect to x and the

Example 1.7 Show that the function u = u( x, t), given by

u(x, t) = ae- t / a2 sin:::

Solution. For the given function it holds

Figure 1.7 Case a = 5, u(x, t) = 5e- t / 25 sin K' t < 0

Example 1.8 (Hadamard) Show that the function tt = tt(x,y) given by

for n E N is a solution on D = {(x, y) I x 2 + y2 < I} of the Cauchy problem for the

U(x,O) =0, OU(x,O) = 0,

au(O,y) = 3y2 u(x 0) = 0 (1.5 )

which after integration with respect to y reduces on the equation

where FI is an arbitrary differentiable function. The solution of this equation is

u(x,y) = eX F2(y) - 4y + eX (X e-tF1(t) dt, (1.6)

Example 1.10 Find the general solutions of the equations

11(X,y) = e- x2 / 2 J q(y) dy + f(x) = f(x) + e- x2 / 2 g(y),

Exercise 1.11 Prove that the following problem on the set R2

Exercise 1.12 Prove that the equation

Exercise 1.13 Prove that the heat equation