Professional Documents
Culture Documents
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.
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
v
ci CONTENTS
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
4.3.2 Examples and Exercises 93
4.4 The Sturm-Liouville Problem 106
4.4.1 Preliminaries .. 106
4.4.2 Examples and Exercises 109
4.5 Miscellaneous Problems. 129
4.6 The Vibrating String 141
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.
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)
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
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.
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.
(iii) the solution is continuously dependent on the given conditions, e.g., initial-
values, boundary conditions, coefficients, etc.
1.1. BASIC NOTIONS 3
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
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 CHAPTER 1. INTRODUCTION
~ 4 4
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
1.1. BASIC NOTIONS 5
Solution.
Since both the real and imaginary parts of (1.2) and (1.3) must be equal, we
get the Cauchy-Riemann equations.
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.
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.
1.1. BASIC NOTIONS 7
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
8 CHAPTER 1. INTRODUCTION
~20
. -10
30
Y 20
-10 10
020
-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
1.1. BASIC NOTIONS 9
eny _ c- ny
U(x,y) = 2n
2 sinn.x
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
Example 1.9 Find the solution U = u(:1:, y) of the following problem on the set R2.
(1.4 )
~ (au
ay ax
_ u) - 4
-,
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
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
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).
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
Z - - (X1,... ,Xn_1'Ull
... 'U.0) E R n +s - l, . - 1 , ... ,n-,
%- ,) = 1, ... ,8,
l' k'
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 1.16 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
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
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.).
14 CHAPTER 1. INTRODUCTION
au au au
ax +zay +2z(x+zY)az =f(z),
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
av av
-ar + laz
- = 7rf(z)
on the set P =
{(r,z)i < r < R2, Izl < M}. Therefore the function w given by
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
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,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
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.
16 CHAPTER 1. INTRODUCTION
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
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
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.
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
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
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,
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)
(2.11)
Thus the general solution is
<I>(Z+y2,X+Z 2) =0,
where <I> 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.
2.1. QUASI-LINEAR PDEs 21
xp + yq =z- x 2 _ y2
be given.
Answers.
b) z=-x 2 -(y-l)2+1.
x + y = 0, z = 1.
Answer. The general solution 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.
22 CHAPTER 2. FIRST ORDER PDEs
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).
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 <p (;) ,
x x
where cI> 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
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 )
2.1. QUASI-LINEAR PDEs 23
ky kx
PI = - and qi = - .
2z 2z
Hence from (2.14) it follows
(t E R). (2.16)
hence its general solution is <I> (;., 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
a) Find those surfaces for which it holds LNOP = 45, where 0(0,0,0) is the origin.
24 CHAPTER 2. FIRST ORDER PDEs
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
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) ,
x 2 + y2 + Z2 = W( Jx 2 + y2 earctan(y/x) , (2.19)
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.
2.1. QUASI-LINEAR PDEs 25
p( X - x) + q (Y - y) = Z - z,
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'
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.
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 )) .
b) The solution is
x2
((x,y) = 2" + xy + f(y2) + C,
where C is an arbitrary constant.
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
where CP E C 1 (R) is an arbitrary function of the class C 1 (R2). The sought after
particular solution is
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
2(z - x - y)1/2 = -y + G2 ,
where G2 is also an arbitrary constant.
The general solution of the given PDE is
<I> (2y - z,y + 2(z - x _ y)1/2), (2.22)
where <I> is an arbitrary function of the class C l (R2). However, note that
z=x+y
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.
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:
28 CHAPTER 2. FIRST ORDER PDEs
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
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.
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)'
2.1. QUASI-LINEAR PDEs 29
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 ).
(y + vr=?) dx (2Z + x - ~) dy _ 2y dz
G'(y) - ~G(y)
y
=~
2y2
- ay2.
30 CHAPTER 2. FIRST ORDER PDEs
This linear ODE by the unknown function G = G(y) has the solution
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
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)
where <I> and III are arbitrary functions of the class Cl(R). Hence
and
1 1
v(x,y) = 4x 2 + 2'<I>(x 2 - y2) _ 2'1lI(x 2 + y2).
and
v(x,2x) = 6x 2 = 4x 2 + ~ (<P(-3x 2) - 1lI(5x 2)).
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 .
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
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).
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.
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,
where C is an arbitrary constant.
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
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
and thus
<p(y) = ~ +~ (C E R).
C
x
So we obtain -Z2 - -y + -2y -- 0 , or
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.
f) 2(y+z)dx-(x+z)dy+(2y-x+z)dz=0;
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
F(x,y,z,p,q) = 0, (2.32)
veX, y, z, a, b) = O. (2.33)
36 CHAPTER 2. FIRST ORDER PDEs
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,
<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 :
Clearly, the functions cp and ' from (2.36) should satisfy the equality
(2.37)
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 <I> 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 <I> , 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 <I> are in involu-
tion. By definition, this means that it holds
[F,<I>] = O.
where
of of of of of of
-+p- -+q-
op ox oz oq oy oz
[F, <I>]
0<I> 0<I> o <I> + 0<I> o <I> 0<I> (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
of of of of
op ox oq oy
(F, <I 0<I> o <I> + 0<I> 0<I> (2.40)
op ox oq oy
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.
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
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.
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.
(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
which after the elimination of the constants a and b gives the PDE
(2.44 )
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;
40 CHAPTER 2. FIRST ORDER PDEs
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 .
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 ).
2.3. NONLINEAR FIRST ORDER PDEs 41
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 < b < a, and R > 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.
42 CHAPTER 2. FIRST ORDER PDEs
b) By supposition we have
(R J a 2 - b2 =f J a 2 _ z2) 2 = x 2 + y2.
z = px + qy + pq
and pass through the parabola x = 0, z = y2.
Solution. One easily finds the complete solution:
t2 = bt + abo (2.48)
Example 2.32 Find the complete, general and singular solution of the PDE
(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) 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.
44 CHAPTER 2. FIRST ORDER PDEs
Answers.
yzp2 = q.
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
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
= o.
Using this equality, we get a linear first order PDE in q>, whose solution is
easily found to be
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.
4Z + p 2 + Q2 _ x 2 _ y2 - 2(xP + yQ) = 0,
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.,
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.
p2 + q2 + pq _ qx - py - 2z + xy = o.
Answer.
x + a = p, and thus WI =p - x.
q = --
a
2
+ J 2z + ay - (x + a)2 + -.
(12
4
Thus we obtain Pfaff's equation
48 CHAPTER 2. FIRST ORDER PDEs
or
a
dz+-dy-(x+a)dx
2 = O.
J2Z + ay - (x + a)2 + ~
Hence the complete solution is
pq = x+y+ z.
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.
z = pq-x-y
(x y)2
= -
4
+a(x-y)-(x+y)+1-a 2 .
Chapter 3
(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
49
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
and
<I>
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: :
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
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 .
y = 3x - xx + G1 , y = 3x + zx + G2 ,
54 CHAPTER 3. CLASSIFICATION OF THE SECOND ORDER PDEs
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
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,
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.
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
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.
(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
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'
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;
Exercise 3.4 Prove that using suitable changes of variables the given PDEs can be
reduced to the given canonical forms.
a)
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
which implies
a (av)
a~ a1] = 0,
hence for some function G j from CI(R) it holds
av
a1] = G 1 ( 1] ).
J
1]
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
where F and G are functions of the class C 2(R), while m = -BIA is the double
solution of (3.22).
Answer. Since ml = m2 = 1, it follows from Example 3.6 that the general solution
IS
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.
60 CHAPTER 3. CLASSIFICATION OF THE SECOND ORDER PDEs
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
dy 1
dx = -),1 = -2" hence x + 2y = C},
and
dy 1 2y3
-dx = -),2
2y2'
hence x - -
= -
3
= C2
J g(r) dr.
3 x+2y
u(x,y) = f (x - 2~ ) + ~
x-2y 3/3
62 CHAPTER 3. CLASSIFICATION OF THE SECOND ORDER PDEs
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(
whose general solution is
e- x / 2 F(2x) + G(-2x) = 0
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
(3.30)
1
2B'(x) = cp~(x) + CPl(X), hence
which gives us
Therefore the solution of the problem (3.28), (3.29) has 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 <I> 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
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:
02u 02u OU ou ou
c) - - - - -
oxoy oxoz
+ -ox + -oy - -OZ = 0, U = u(x,y,z);
Solutions.
A = [
1 1 -1
1 2 0
1
-1 0 6
[ ~l=B[:l=[ y~Xzl'
( z x--+-
Z 2
the equation transforms to
-2 0
o -1 1
-1 0
Since
3.2. n INDEPENDENT VARIABLES 67
( = 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:
a)
b)
c)
Solutions.
k
(k = L Xf, k = 1,2, ... , n.
=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
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
b) Show that in the case n = 2 the Lorentz transformation can be represented in the
following from:
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:
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.
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)
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)
70 CHAPTER 3. CLASSIFICATION OF THE SECOND ORDER PDEs
which specifies the value of U OIl aQ, or the Neumann boundary condition:
aul
an 8Q
=f
'
(3.40)
(3.41 )
i.e., the ends of the wire are being kept on the zero temperature.
Chapter 4
Hyperbolic Equations
fPu fPu
ot 2 - a 2 ox 2 = F(x, t), (4.1 )
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
(x E R, t > 0),
Solution. Let us change the variables v = x + at, W =x- at, and denote by
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) 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.
1 1 lx+at
u(x, t) = -2 (J(x + at) + f(x - at)) + - ~ _~
g(Od( ( 4.4)
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.
J g(~)d~,
1 x+at
u(x, t) = ~ (J(x + at) + f(x - at)) + 2a
x-at
a2 a
= 2" (J"(x + at) + f"(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).
74 CHAPTER 4. HYPERBOLIC EQUATIONS
Example 4.3 Let the following Cauchy problem on the set {( 3;, t) I x E R, /, > o} bc
gzven:
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
x-at
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
4.1. ONE-DIMENSIONAL CAUCHY PROBLEM 75
Solutions.
(4.5)
U(x,O) = F(x), aU(x,t) = G(x) (-00 < x < (0),
at
has the solution given by
x-at
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
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
g x
()d
X (x> 0, t < ~),
x-at
ou(O, t) = o.
Ox
4.1. ONE-DIMENSIONAL CAUCHY PROBLEM 77
Solutions. Since the initial conditions aTe homogeneous the solution of the consid-
ered problem is
J J
t X+a(t-T)
JdT J
t a(t-T)
8 t)
8u(O,
x
1
=~
2a
J(J(a(t - T),7) - f(-a(t - T),T))d7 =
t
O.
o
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
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
J J
t x+a(t-r)
+21a f(Z,T)dzdT
t-x/a x-a(t-r)
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)
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 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
. 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.,
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 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
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.
aatu -.6.u =
2
F(x, t), ( 4.9)
2
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
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.
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, 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
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
84 CHAPTER 4. HYPERBOLIC EQUATIONS
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)
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.
4.2. N-DIMENSIONAL CAUCHY PROBLEM 85
!....
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
Mu(x, r, t) =~
an
J u(x + rv, t) dSv (4.16)
BB(a,I)
(4.18)
86 CHAPTER 4. HYPERBOLIC EQUATIONS
[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)
Jg(s)
rH
vCr, t) =~ ds,
2r
r-t
where
a
g(s) = at(sv(s,t))lt=o = sf(s).
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
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) ,
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)].
88 CHAPTER 4. HYPERBOLIC EQUATIONS
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
Example 4.20 (Poisson formula) Prove that the Cauchy problem for two-dimen-
sional wave equation
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
1
47r
J g(x + tv) -dS v
t
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.
is called the Fourier series of function f, if the coefficients An, n = 0, 1, ... , En, n =
1,2, ... , arc given by
The coefficients An, n = 0,1, ... , En, n = 1,2, ... , are called Fourier coefficients.
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
(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.
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
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 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
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
~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
where ). is a separation constant. Thus our task is to solve these two ordinary
differential equations.
4.3. THE FOURIER METHOD OF SEPARATION VARIABLES 93
(4.25)
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).
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.
1
cos a . cos 13 = 2" (cos( a - 13) + cos( a + 13)) ;
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
4.3. THE FOURIER METHOD OF SEPARATION VARIABLES 95
2 (-It+!
f(x) = - L
00
7r n=l n
sin
mfX
-0-'
{.
(4.31)
For = 7r we have
(_l)n+l
=2L
00
X"(X) = _A T"(t)
X(x) , a 2 T(t) = -A,
or
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.
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.
X(O) = C1 = 0,
(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.
4.3. THE FOURIER METHOD OF SEPARATION VARIABLES 97
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 )
Using the superposition principle on the set of solutions we take the solution of the
given problem as an infinite series
The series (4.43) converges uniformly if the series L(IAnl + IBn!), where An and
n=l
En are Fourier coefficients, converges.
00
for k = 0,1,2 converges, if the function has continuous second derivative and the
third derivative is piecewise continuous and it holds
for k = -1,0,1, converges if the function tP has continuous first derivative and the
second derivative is piecewise continuous and it holds
n?rx
E Ansin-e-;
00
u(x,O) = f(x) =
n=l
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
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
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
-:21 ~ E
L..., n
(
cos
na7r(x
C
+ at) - cos
na7r(x - at))
C .
n=1
100 CHAPTER 4. HYPERBOLIC EQUATIONS
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
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
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
n7rX
=L An sin -e-
00
x(e - x) (0 < x < e),
n=1
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
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(x,O)=o, 8u(x, 0) =
8t
(0 < x < C),
8v(C,t)
v(O, t) = 0, --=0 (t > 0), (4.48)
8x
P
S(x) = EX.
The method of separation variables leads the problem (4.4S) Lo the systems of ODEs
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'
8 zu 8 zu +2 -3x
(0 < x < 1, t > 0),
8xz = 8t Z e
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
v ( x, ) = -92 ( 1 - e-
3) x, ov(O, t) __
ot (0<x<1),
IS
L
00
we obtain
4
Cn = - - (1 - e- 3 )
Jl x sin(n7rx)dx = 4(-1t
- - ( 1 - e- 3 ) (n EN),
9 9n7r
o
u(x,t) = ~(e-3X_1)+~(1_e-3)x
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)
Show that the solution of this problem is not unique on the given rectangular.
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 ....
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.
106 CHAPTER 4. HYPERBOLIC EQUATIONS
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
J
b
C n= IicPnll- 2 r(x)f(x)cPn(x)dx n E N, (4.54)
a
where
J
b
(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
-=
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
A=n(n+1)
-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
and the corresponding eigenfunctions are the Legendre functions P:;' given by
Specially for m = 0 the ball functions are axial symmetric, i.e., independent of cpo
The Bessel equation is given by
4.4. THE STURM-LIOUVILLE PROBLEM 109
+ 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),
X" + AX = 0,
110 CHAPTER 4. HYPERBOLIC EQUATIONS
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
L
00
Jr(x, t)Xn(x)
l
Denoting by
J
l
where An and En are arbitrary constants for n E N. The solution of the given
equation with boundary conditions is
L AnXn(x),
00
u(x,O) = f(x) =
n=l
wherefrom we get
An = IIXnll-2l f(x)Xn(x)dx,
cPu 8 2u
8x 2 = a- 2 8t 2 - R(x)sinwt (0 < x < e, t > 0),
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
8~p =wC(x)coswt-wD(x)sinwt,
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
=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
4.4. THE STURM-LIOUVILLE PROBLEM 113
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
u(x, t)
u(x,O) = 0 = L
00
An sin (n7rx)
-C- ,
n=l
114. CHAPTER 4. HYPERBOLIC EQUATIONS
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, ....
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
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, ....
V(x, t) = (1 + t)x,
116 CHAPTER 4. HYPERBOLIC EQUATIONS
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
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
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
The solution of the considered problem is
cPu 82u
-=--1 (0 < x < I, t > 0),
8x 2 8t 2
t2
u(O, t) = 2' u (1, t) = - cos t (t > 0),
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
x3 - X _8v....:(,-x,--'0...:..) = 0
v(x , 0) = ---
6' (O<x<l).
8t
118 CHAPTER 4. HYPERBOLIC EQUATIONS
The solution of this problem is v(x, t) = Vh(X, t) + vp(x, t). The homogeneous
part of the solution is
00
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
The solution of the considered problem is
t 2 (1 - x)
u(x, t) = 2 + x cos t
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),
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
Til + a2 n :
4 4
T =0 with condition T' (0) = 0,
IS
n 2 7f 2 at
T(t) = A cos - f -
2 -'
=L
7fX 00 n7fX
xsmT Ansm--,
n=l
120 CHAPTER 4. HYPERBOLIC EQUATIONS
[)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),
X" + AX = 0, X(O) = 0,
(s < 0 does not produce a new solution). We have here a continuum of eigenvalues
generated by ). = s2. The solutions of the problem
where A(s) and B(s) are arbitrary functions. Using the superposition principle,
from (4.70) and (4.71) we obtain
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).
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.
122 CHAPTER 4. HYPERBOLIC EQUATIONS
Answer. This is a special case of the previous example. The solution of this
problem is
00
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
Figure 4.3 u(x, y, 0) = sin(nx) sin(my) for (n, m) = (1,4), (1,3), (2, 2), (2,5).
124 CHAPTER 4. HYPERBOLIC EQUATIONS
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 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
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
(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
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 )
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
(4.82)
From
au aU or - aU atp aU - sin tp aU
ox = -or. -+
-
ox atp. -ox =costp-
or ----,
r atp
4.4. THE STURM-LIOUVILLE PROBLEM 127
aU ar aU a<p
-au = - . aU cos <p aU
. - + - . - = SlfI<p- + ----,
ay ar ay a<p ay ar' '1' a<p
it follows
( ) =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
~T"X - T6.X =0
c2
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)
128 CHAPTER 4. HYPERBOLIC EQUATIONS
We shall solve the problem (4.87), (4.88) in the polar coordinate system (r, cp) and
using again the Fourier method. We take
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
u(r, cp, t)
k=Oj=l
+ (Ckj coscxkjt + d kj sin CXkjt) sinkcp)Jk(rxkj),
fPu 1
L(u) = - - + -u = 0 ((x, y) E R2), (4.91 )
8x8y 2
Figure 4.5
8v J8v J(8v
u(M).v(M)=u(B).v(B)-_u8xdY+_U8ydx+_
J 8u) ,
u8xdx+v8ydy
AM BM AB
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
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
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
Xo Xo
Note that all the quantities on the right-hand side are knowIl.
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
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
Examrle 4.44. Prove that the following Cauchy's problem on the set
{(x,y) Ixl<l, O<y<l}:
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
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
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
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.
t a real parameter,
and
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
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!
(4.103)
Solution.
(
[;;!
n Yi
::; n (t]
=1 0
ID;J(XI, ... , Xi-I, Si, Yi+I,, YnW dSi) .
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
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
and
x x
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
Hints.
a) Taking (the Schwartz derivative)
plll(X) 3 pll\X)
F(x) = pl(X) - 2' p/2(x)
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
p "() 1smp,
z = ab .
,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
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.
140 CHAPTER 4. HYPERBOLIC EQUATIONS
Example 4.53
L (_l)mk m b(ao(ml(x))
00
[Pu a2u
-2--2=0,u(x,0)=slnx,u x,-
. (X) =x
& ~ 4
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.
p(x)+q(x)=smx,p . (5X)
4 +q (3X)
4 =x.
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
(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
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
au
L(u) = F OIl Q, on + a = f (x E oQ).
where a E C(oQ).
flu=O (xEQ),
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
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 ),
X"(x) Y"(y)
X(x) = -A, Y(y) = A.
5.1. DIRICHLET PROBLEM 145
(n EN), (5.4)
(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
where
Bn =
2
e
h n7r 1
Jf(x) . -e-
l
n7rX dx
sm (n EN).
o .
{;sm -e- 0
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),
X n () . n1rX
X = sln-- (n EN).
(n EN),
where An and En are arbitrary constants. From the boundary condition Y(d = 0,
we obtain
n1rl
n 1ry cosh -- . n 1ry
An cosh -- - An . smh --
sinh n1r 1
00 1 ,n1rx. n1r(l - y)
u(x,y) = LAn sm --' smh '
n=l sinh n1r 1
5.1. DIRICHLET PROBLEM 147
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
fPu fPu
ox 2 + oy2 = 0 (0 < x < 1, 0 < y < 2),
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
1 1 4 (-It
+ 2" I:
(Xl
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
u(x,y) =];00 (
Ancosh y1 +
/ n7ry
-e- . /
+ B n smh y 1 +
n7r Y). n7rX
-f- sm-C-
and
g(x)
~( AncoshVl~
=~ . . ~. n7rX
+ T-f- +smh y1 + T-f-) sm-f- (0 < x < e).
fJ2u a2 U
aX 2 + ay2 - ku =0 (0 < x < i, 0 < y < J\),
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
Let us first solve the problem (5.6). Taking v(x,y) = X(x) Y(y), we obtain the
problem
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).
152 CHAPTER 5. ELLIPTIC EQUATIONS
IS
(n EN),
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
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
with solutions
y.n () . n7ry
y =sln-e;-.
(n EN).
5.1. DIRICHLET PROBLEM 153
w(x,y) = L00 (
Ancosh ~7rX
k+ -C-+Bnsinh ~7rx. n7ry
k+ -C- .sm-C-
n=l 1 1 1
= 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
Jcp(y)
l,
An cosh v'k + Bn sinh v'k = : sin n;y dy,
1 0 1
and
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)'
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
wherefrom it follows that
n = 1,3,5, ...
0, n=2,4,6 ....
5.1. DIRICHLET PROBLEM 155
. ((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
Example 5.8 Determine the solution of the problem given in previous Example 5.7
by using double Fourier series.
Using
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['),
(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'
R " + -R
1 ,
H"
1'2 r = -- = A
R H'
wherefrom we obtain the problem
H" + AH = 0, (5.11 )
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).
5.1. DIRICHLET PROBLEM 157
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
_1_ sin(ncp)) dO
Rg71" -1T
1 J1f
R or.
n I( 0)( cos( n( 0 - cp) )dO,
-1T
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).
a 2u 1 au 1 au
-+--+--=0
ar2 r ar r2 a<p
(0 < a < r < b, -7r < <p < 7r),
U(r,<p)
(5.17)
00
a) After applying the boundary conditions we obtain the following system of equa-
5.1. DIRICHLET PROBLEM 159
tion
Ao + Bo In a =;:
1 J" J(cp)dcp,
-"
JJ(cp) cos(ncp)dcp
1r
An an + Bna-n = ~
-1r
(5.18)
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
In ::.
U(r) = Tl + (T2 - Td--t
In -
a
160 CHAPTER 5. ELLIPTIC EQUATIONS
(-oo<x<oo,o<y<oo),
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.
J
00
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
5.1. DIRICHLET PROBLEM 161
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
Solution. Applying the Fourier method of separation of variables first in the form
u(r,cp,z) = V(r,z) <I>(cp)
and then
V(r,z) = R(r) Z(z),
where <I> : R -+ R, R: (0,00) -+ R, Z: R -+ R and <I> 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<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
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
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.
on the square
s= {(x,y) i 0 ~ x,y ~ 7r},
but it attends its maximum inside of S at the point (~, ~).
164 CHAPTER 5. ELLIPTIC EQUATIONS
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).
~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
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
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
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
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
Solution. The functions u = 1 and u = I~I are solutions of the given problem.
GlaQ = O.
168 CHAPTER. 5. ELLIPTIC EQUATIONS
1
c) 0< G(x, y) < 47rX-y
/ / (x E Q,x =I- V)
Solution.
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,
.
Figure 5.1
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
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
170 CHAPTER 5. ELLIPTIC EQUATIONS
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
+ 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
5.3. THE GREEN FUNCTION 171
(
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)
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
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.
1
u(x) = 47rR J R2Ix _-lxl2
B(O,R) y
f(y)dS 13 y (x E B(O, R,
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
u(x)
1
= -47rr
J r2
I
-lxl 2
1 u(y) dSy
x- y3
(x E B(O,r)).
BB(O,r)
Ql = {x I x E Q,u(x) = yE8Q
maxu(y)},
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.
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
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.
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)
J au(y)
an
aE(O,r)
dS y = O.
u(O) = _1_1-
rn - O-n
J u(y) dS y (5.29)
aE(O,r)
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
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
176 CHAPTER 5. ELLIPTIC EQUATIONS
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.
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
Since
8
d(B(xO,8/4),8Q5/2):::: 4
and for every multi-index a we have
(5.32)
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.
v(x,y) = J (_au~X'Y)dx+au~~Y))dy,
L((xO,yO),(x,y)) Y
5.4. THE HARMONIC FUNCTIONS 179
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.
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
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
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
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.
is harmonic on B \ {OJ.
5.4. THE HARMONIC FUNCTIONS 181
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.
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.
182 CHAPTER 5. ELLIPTIC EQUATIONS
Exercise 5.47 Prove that the solution of the Dirichlet problem in R2, R3
R2
where y' = Y1Yf2 is obtained by inversion.
where p is the density of a mass at (~,."e) and k is a positive constant. The function
u satisfies the Laplace equation
Parabolic Equations
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
+
j j (47raF(e,r)
t
(t_r))n/2 exp
2
(
o Rn
183
u E C(Q),
ou 02U
and
au -
ot
- t:.u
-< O.
Then
ma~ u(x, t) = max u(x, t).
(x,t)EQ (x,t)E&'Q
u(x, t) =
1
c;
lco fey) exp ( - (x -
2
y)2) dy, (6.4)
2ay7rt 4a t
-co
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 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 '
u(x,O):=
u(x, t) =
T
r=;
Jd exp ((x_
-
y )2)
dy. 2
2av-rrt c
4a t
u(x, t) =T vk J (x-d)/a-/2i
(x-c)/a-/2i
e-e/ 2 d(
J e-
x
<I>(x):= t2 / 2dt,
-00
u(x,O)=f(x) (xER),
ou(O, t)
b) ax = 0, if the function f is even;
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).
au 202U
- =a - 2 (x E R, t > 0),
at ox
U(x,O) = w(x) (x E R).
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
188 CHAPTER 6. PARABOLIC EQUATIONS
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);
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,
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
Answers.
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)
190 CHAPTER 6. PARABOLIC EQUATIONS
au
at = a .6.u + F
2 (
x E Rn t > ,
)
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
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) 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
6.1. CAUCHY PROBLEM 191
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
a2 lj
acp2 - lj = 0,
u E C(Q),
au
- - ~u on Q
at
and of u on the boundary a'Q (see Preliminaries).
Therefore u = on Q.
192 CHAPTER 6. PARABOLIC EQUATIONS
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
va(x, i) ~ e + u(x, i)
where
v(x,t) = lim va(x,t) = (
a ..... 00 41ft
\ n /2 J exp (_IS - x12) f(s)ds.
4t
Rn
v(x,O) = f(x).
6.2. MIXED TYPE PROBLEM 193
8h _ 8 2h _ 0 V
8t 8x2 - on,
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.
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).
194 CHAPTER 6. PARABOLIC EQUATIONS
where
Solution. Let us suppose that the solution of given equation has the form
(n E N) (6.10)
(n EN),
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
en = p.2 Jf(x)sm-l-dx,
l
. n7rX
(n EN).
o
ou 202U
at = a ox2 (0 < x < 7r, t > 0),
(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).
L: An sin(nx)exp( _a
00
u.(x, t) = 2 n 2 t).
n=l
u(x,O) = To = L: Ansin(nx),
n=l
4To
{
n = 1,3,5, ...
n7f
0 n = 2,4,6, ... ,
u ( x, t ) = -4To ~
L.J
sin((2n - l)x exp ( -a 2(2 n - 1)2) t ) .
7f n=l 2n - 1
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.
6.2. MIXED TYPE PROBLEM 197
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(X,0)=5+2X (O<x<7r).
(-It+!
u(x, t) = 5 + 4 L + 4)t).
00
sin(nx) exp( _(n 2
n=I n
S"(x) = 0,
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
where
.".
en = ~J(x(11"-X)-Tl-(T2-Td~)sinnxdx
11" 11"
o
(n EN).
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
iPu -2 i}u
-=a - (0 < X < 1, t > 0),
Ox 2 01,
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
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
cPu -2 iJu
iJx2 =a iJt (O<x<I!, t>O)
iJu(O, t) =0 iJu(,t)
--=0 (t > 0),
iJx ' iJx
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, ...
_ ~
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- ~
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
av(O, t) = 0
ax ' vel, t) = 0,
202 CHAPTER 6. PARABOLIC EQUATIONS
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
Exercise 6.27 Solve the following problem
202u ou
a-=- (0 < x < , t > 0),
ox 2 at
ou(O, t)
ox = qo, u(, t) = 110, (t > 0),
u(x, t) = qox
2n + 1
+ Uo + ~ An sin (-U~xr.) exp (-t
00
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.
6.2. MIXED TYPE PROBLEM 203
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),
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
(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
L
00
Jf(x) cos(Anx)dx.
l
Cn =
o
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
(n EN).
GU(O, t)
ax - hu(O, t) = u(, t) = 0 (t > 0),
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
h tan A = -A.
6.2. MIXED TYPE PROBLEM 205
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)
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.,
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
206 CHAPTER 6. PARABOLIC EQUATIONS
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]
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,
IS
S(x) = ~ (1 -e 3X ) - ~ (1 - e3 ) x.
The solution of second part of the considered problem, i.e., of the problem
IS
v(x, t) = 2 2 ).
n=1
Cn = -4 (1 - e3 ) x
Jx sin(n7rx)dx = -'---~--'-
1/2
4(1-e )(-lt 3
(n EN).
9 9n7r
o
(_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
Solution. Let us suppose that the solution of the problem (6.23) has the form of
generalized Fourier series
00
(6.24)
n=1
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
J
f
n=l
~U
= :LTn(t)X:(x) = - :L ASn(t)Xn(x).
00 00
ox2
n=1 n=1
(6.25)
i
with the solution
n=1
J
f
Jrn( T)
t
Tn(t) = in exp( -Ana2t) + exp( -Ana2(t - T))dT.
o
6.2. MIXED TYPE PROBLEM 209
-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).
x OCJ
where
Tn(t) =
.
Ilsmn7rxll
_;1 -cosismn7rxdx
x. (_l)n
= --cost.
2
o
2 n7r
Let us consider the solution in form
u(x, t) =
n=1
whose solution is
From u(x, 0) = 0, it follows that en = 0 and the solution of the considered problem
IS
210 CHAPTER 6. PARABOLIC EQUATIONS
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),
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)),
av 2 a2 v aV
at - a ax 2 = r(x, t) - 8t (0 < x < , t > 0),
(6.29)
u(f!, t) = bo + blt (t> 0),
u(x,O)=o (O<x<f!).
av
- - a -
2a 2V = - ( a1 + -X) (b 1 - aj)
at ax 2 f! '
with initial condition
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 -
(6.30)
with the initial conditions
X) n?TX
- ( ao + e(bo - ao) = L In sin -f-'
00
n=l
In = . n?TX 2
-II sm-f-W J
l
(aoX . n?TX
+ e(bo - ao) ) sm- d
f- X,
o
212 CHAPTER 6. PARABOLIC EQUATIONS
+ 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) ) )
+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)
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
6.2. MIXED TYPE PROBLEM 213
L
00
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
J(2 + (1 -
1
- ~ (cos t + 2 (1 -
n7f
( -1 n) , n = 1,2,3, ....
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
[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),
a 2 w _ aw 2
ax2 - at + (0 < x < 1, t > 0),
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.
We still have to find the function v satisfying the equation
where
An = 2 J 1
X
2
sin(mrx) dx = - ( _l)n+l.
n7r
o
The solution of the considered problem is
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
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),
(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
n7rX . n7ry
2:= 2:= cm,n sm -f..- sm -f..-'
00 00
f(x, y) = (6.39)
n=lm=1 1 2
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
'
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.
6.2. MIXED TYPE PROBLEM 217
ou(O,y,t)
-a} Ox + f3lU (O,y,t ) = 0 (0 < y < 2, t > 0),
ou(x,O,t)
-a3 oy + f33 U(X, 0, t ) = 0 (O<X<l, t>O),
ou(O, t)
-a} ox + /31 u(O, t) = 0 (t > 0),
+ f32U(1l t) =
OU(l, t)
a2 ax (t > 0),
au(O, t)
-a3-[}y- + f33U(O, t) = (t > 0),
aU(2, t)
a4 oy + f34U(2,t) = 0 (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
U(x,y,O)=xy (0<X<1,0<y<2).
D..W+.\W=O, T'+.\T = 0
by using the separation variables method
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
au
Further on, let us suppose that u and ax are bounded when x --+ 00.
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
J
00
JB(s) sin(sx)ds
00
B(s) = -;
2
Jf(O sin(sOd(
00
o
(6.41)
220 CHAPTER 6. PARABOLIC EQUATIONS
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
)
200
u(x,t) = ;[f(O (00
[(coss(x-0-coss(x+O)exp(-a2s2t)ds ) d~
J
00
u(x,t) = exp( _a 2s 2t). (A(s) cos(sx) + B(s) sin(sx))ds. (6.42)
o
where
-00
-00
6.2. MIXED TYPE PROBLEM 221
u(x,O) =e x2
(-00 < x < (0),
u and ax
au arc bounded when x ---+ 00.
-00
1 e- X2 . 1
00 s 00 _x2 S
- - - sm(sx) - - ; e cos sxdx =--A(s).
7r 2 -00 27r 2
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
Exercise 6.49 Solve the mixed problem for r > 0 and t > 0
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
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
X2 t2
Q3 = sJ J F(x,t)dtdx. (6.48)
or
X2
= Cp J(U(~,t2) -U(~,tl))d~
XI
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'
au 2a 2u
at =a ax 2 + lex, t),
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
kp p
t1
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
(6.50)
or
226 CHAPTER 6. PARABOLIC EQUATIONS
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
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
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'
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,)'
Xi - h S; i S; Xi + h, YJ - k <
-
y"Z- <
-
Xt - k,
229
satisfying
8U(Xi,yj) D 8U(Xi,Yj) G ( )
+Ci,j 8x + i,j 8y + i,jll Xi, Yj
(7.3)
(7.4)
Taking
230 CHAPTER 7. NUMEIUCAL METHODS
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),
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
o 1
Figure 7.2
Using the boundary conditions we can evaluate the terms on the right-hand side
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:
These are the approximate solutions of the considered problem at the corresponding
points.
232 CHAPTER 7. NUMERICAL METHODS
Solution.
'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).
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
234 CHAPTER 7. NUMERICAL METHODS
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)
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 ,
1 1 1
4U11 -
, , = -16'
2U1 2 - 4U2 1 -
. 2U1 1 -
, . = -16'
U2 2 - 4U2 2 -
, , = - -
4U1 2
16.
[Pu [Pu
ox2 + oy2 =0 (0 < x < I, 0 < y < I),
1L1,2 = U3,2,
we obtain
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
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)
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)
(7.16)
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
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,
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+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)
(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
1 1
a) mesh h = -, and k = -;
5 50
1
b) mesh h = k = -
5'
Solution.
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)
Example 7.10 Determine the numerical solution for the following problem
au fPu
(t > 0, 0 < x < 1),
at ox2
U(x,O) = x (O<x<l),
Solution. If h =I I I
-5' k = -,
50
then c> = - and
2
therefore we have
(7.25 )
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
Example 7.11 Determine the numerical solution for the following problem
Solution. The given PDE is the two dimensional heat equation. Let us take
Xi = ih, Yj - kj, h = k,
241
'(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
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,
1) 1 1
U2,3 ( 64 =4 (U3,3(0) + U1,3(0) + U3,2(0)) = 8;
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;
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),
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),
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),
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,
244 CHAPTER 7. NUMERICAL METHODS
ah
IU"l- ull < -2M2 .
I, I, -
(7.29)
where
(O<x<l, t>O)
i = 0,1, ... , 10
Ui,j+1 = Ui-l,j + U;,j - Ui,j-b i=I,2, ... ,10, j=1,2, ... 10.
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).
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.
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
-,
246 CHAPTER 7. NUMERICAL METHODS
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
-.
(t>O,O<x<I),
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
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
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
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
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.
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
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
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
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
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
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
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
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.
1
f(x) = -
Ixla
is integrable on B(O, 1) = {x Ilxl < 1 }?
Answer. a < n.
JJ
00 00
there exist the iterated integrals but not the double 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,
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
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).
F(t) = J
A
f(x,t)dx
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
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
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.
Ix-YI<h Ix-YI<h
< ~ j If(x+v)-f(xWdv
Ivl<h
~ j dvjlf(x+v)-f(xWdv.
Ivl<h Q
Using now the continuity of the integral we obtain the desired conclusion.
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
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. 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
i=l
Then by the fact that 'Pi = 1 in some neighbourhood of the set K; we obtain the
desired equality.
260 CHAPTER 8. LEBESGUE'S INTEGRAL, FOURIER TRANSFORM
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
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
is a Banach space.
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
where
el = (1,0, ... ,0),e2 = (O,l,O, ... ,O), ... ,en = (0,0, ... ,1)
and
(<I>(i)t 1 = ((<I>(i)t1h, ... ,((<I>(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
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
Exercise 8.24 Let Q eRn be a locally quadratic bounded region. Prove that GQ is
a locally quadratic part of surface.
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
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}.
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},
264 CHAPTER 8. LEBESGUE'S INTEGRAL, FOURIER TRANSFORM
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)
el e2 e3
cos cp sin !P 0
-r sin cp r cos !P 0
8.3. THE SURFACE INTEGRALS 265
eJ e2 C3
-Rsincp Rcoscp
001
By (8.3) we have
= (1
f
R
) It
a -.
~
4
3n-
(cpd)(rcoscp,rsincp,O)r dcp)dr
+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'
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.
R 27r
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
2" H
+R 1(1 f(Rcosrp,Rsinrp,z)dz)drp. (8.8)
o 0
Hence
R 21r 2" H
1(x +
av
2 y2) dS,
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).
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).
n
where z . x =L ZiXi
;=1
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
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).
always has a solution u(x, t) E cm(Rn+1) for every f E q(Rn) for enough big s.
8.4. THE FOURIER TRANSFORM 269
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.
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
270 CHAPTER 8. LEBESGUE'S INTEGRAL, FOURIER TRANSFORM
Applying now the inverse Fourier transform ;::--1 on both sides we obtain
u(x, t)
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.
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)
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
u(x,t) = (27ft'
n 1. Rn
e ozy
sin(lylt) g(y)
Iyl
. dy.
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
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.
Remark 8.38. 1. The opposite statement is also true. Therefore the following
theorem is true.
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
belongs to S iJ J E 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)
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)
1:
problem (8.14), then the following is true
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.
-
If(y)1 ~ Ck,o
1 00
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
b) Since we have
u(x,t) = - 1
27r
1 00
-00
f(y)exp(-zty2
- -zxy) dy,
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
1:
Adding the last two equalities we obtain by (8.14)
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
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
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 )
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
u(x) = - -
1 foo F-I(f)(z)
e-'xz dz.
,j2; -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
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
if and only if
F- 1 (f)(z)
F-l(k)(z) E L2(R).
()
u x =
~2
VZ7r
1"" F-F-l(k)(z)
-00
1
(f)(z)
e
-,.:z d
z.
Chapter 9
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.
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
Solution. For an arbitrary but fixed <p E CJ (R) there exists R > 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.
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
}
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.
where
/h(x) = k f(y)8 h ([x - y[) dy.
Moreover, for a compact subset J( of Q we have
Solution. We have
By this theorem and the property 4) (section 8.2) of the function 6h 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
282 CHAPTER 9. GENERALIZED DERIVATIVE AND SOBOLEV SPACES
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 )
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
~ =
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).
284 CHAPTER 9. GENERALIZED DERIVATIVE AND SOBOLEV SPACES
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
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
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 9.7 (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.
286 CHAPTER 9. GENERALIZED DERIVATIVE AND SOBOLEV SPACES
L n2(a~ + b~)
n=O
converges, where
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.,
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
Prove that
1 x
a) (Dcxf>.)(x) = ~(DcxJ)(-:\), which implies f>. E Wk(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)
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).
9.2. SOBOLEV SPACES 289
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
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$.'"
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
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
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
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
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
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. 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
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
9.2. SOBOLEV SPACES 293
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
ok
Hence f EW (Q').
294 CHAPTER 9. GENERALIZED DERIVATIVE AND SOBOLEV SPACES
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);
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)
ok
c) Since W (Q) is a Hilbert space we obtain c) by Parseval's identity.
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.
J J
b b
J1'(z) dz.
x
f(x) - f(a) =
a
J1'(z) dz l)2
x
< 2
<
a
<
a
(b - a)lf(a)12 ::; 2
a a
with sides
JIfl2 ~ ~l J
BP P
If(xW dx + C2E t,=1 JIDJ(xW dx,
P
Solution.
a) Using the same procedure as In Exercise 9.26 only starting now from the
equality
J
Xi
J
bi
~ 21f(xW + 2(Xi - ai) IDJ(Xl, ... , Xi-I' Zi, Xi+l, ., XnW dzi
ai
< E(b
,
~ a) PJIf(xW dx + E(bi - a;) J
P
IDJ(xW dx
JIfl2 ~ ~l J
BP P
If(xW dx + C2E t,=1 JIDJ(xW dx.
P
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
The preceding inequality (9.6) can be written also in the following form
1
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
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)
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.
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;
9.2. SOBOLEV SPACES 299
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.
If(Xl, x'W :::; J IDd(zl, x')1 dZl J dYl :::; (Xl - ad J Ddh, x') dzl
a] al al
Therefore
XI
b2 bn bl
o 1
b) Since Ca (P) is dense in the space HI (P) there exists a sequence {lj LEN from
0 1
J Ih(xW dx :::; C 2 J
P
t
3=1
IDjfk(xW dx.
1II(xW dx = 1II(xW dx 1t
P Q
and
P J=1
IDjl(xW dx = 1t
Q J=1
IDd(xW dx,
Since trivially
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
1J'(x) dx.
b
I(b) - I(a) =
a
Exercise 9.33 Does there exist any lunction I E C[a, b] such that I (j. W 1 ([a, b])?
9.2. SOBOLEV SPACES 301
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])
o 0
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
Therefore we have
( 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)
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).
(k E N U {O})
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
(x E D(T)).
303
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'
l)xlci)ci
ieI
converges to x, where in the sum there are not more than countably many nonzero
members.
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
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
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
{(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~).
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
lim Xij
J~OO
= 0 (i EN), lim Xij
t-OO
= 0 (j EN), lim Xii
t~OO
= O.
Moreover, the elements of the set I can be ordered in a increasing sequence {Pi}iEN
such that 00 00
(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
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
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
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
<
j=1
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
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;
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
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
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
J1 OxO
Ik(x, yW dxdy < 00
Tf(x) = 10 k(x,y)f(y) dy
is a compact operator on L2(0).
Hint. Show that an unbounded set A can not be compact, constructing a sequence
in A which has no Cauchy subsequence.
T(x) = 2]2:tijxi)bj,
j=1 i=1
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
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.
= (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)
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.
(>..J-T)x=O (10.6)
either has the only trivial solution x = 0, and then the equation
(1 - T)x =y
can be written in the form:
(1 - T*)x* = y*
Hints. Use Examples 10.15 and 10.16. See the solution of Example 10.20.
(I - T)x =0 (x E H)
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)
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
n
xj = z; + Ltj;xi, j > n, (10.17)
i=l
316 CHAPTER 10. SOME ELEMENTS FROM FUNCTIONAL ANALYSIS
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
n n
< 2 L IYjl2 + 2 L(L It;jI2) L ly;J2
00 00
n
L Ihl12 ::; (ccd 21IYI12,
j=l
and therefore
n
IIhl12 L Ihj l2 + L
00
(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
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).
(J - T)x =Y (10.21 )
n 00
L(Yj + L tijYi)xj = O.
j=l i=n+1
It follows
n (X) n n 00
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,
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:
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
Therefore
The above inequality implies that the sequence TX1, ... , Tx n , ... has not a Cauchy
subsequence. A contradiction.
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
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
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
lim
n~oo
< X",X~ >=< X",XI >
A Banach space X is reflexive if X = X" ( in the sense ofthe canonical map x f-7 x").
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.
10.3. NORMED VECTOR SPACES 323
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
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
Hence by the continuity of the functionals < T(x), > and < x, >
< T(x),y' >=< x,z' > (x EX).
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
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
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 10.29 (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'.
326 CHAPTER 10. SOME ELEMENTS FROM FUNCTIONAL ANALYSIS
(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.
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
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.
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
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 ~~)}
(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
L
00
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
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
329
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)).
L(u) =L aO'(x)DO'u, k ~ 1,
1<>I~k
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.
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
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
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
(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
332 CHAPTER 11. FUNCTIONAL ANALYSIS METHODS IN PDEs
L(u) = L a",(x)Dau, k ~ 1,
lol$k
Therefore, taking
L m~IDaL(u)1 <
lal$s-k xEQ
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).
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)
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)
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)
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)
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
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)
11.1. GENERALIZED DIRICHLET PROBLEM 335
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.
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
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)
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 11.15 !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
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,
338 CHAPTER 11. FUNCTIONAL ANALYSIS METHODS IN PDEs
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
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 that
Solution.
= 1aQ
Ov au
P( U-;:;- - V-;:;-) dS.
un un
(11.5)
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)
au
- = --u
a
for f3(x) > 0 (x E aQ)
an 13
and
u=O for f3(x) =0 (x E aQ),
340 CHAPTER 11. FUNCTIONAL ANALYSIS METHODS IN PDEs
we obtain by (11.6)
(u E U). (11.9)
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
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
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)
Solution.
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
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
(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
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).
-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)
11.1. GENERALIZED DIRICHLET PROBLEM 345
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.
a} that the classical solution u E CI(Q) of the problem {11.15} is also a general-
ized solution;
Example 11.27 Which of the following differential operators are uniform elliptic
Solution.
a) The equality
n n
- L Zi(-I)Zj = LZ?,
;=1 ;=1
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
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
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
-ttli=lj=1 Q
a;j(D;u)(Djzp) dx + 1
Q
fZP dx = 1 Q
Fzp dx.
Ilull w.1
(Q)
~ CIIFIIL2(Q).
Solution.
a) It is easy to check that
o 1
(J,g E W (Q))
11.1. GENERALIZED DIRICHLET PROBLEM 347
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
~ C 10 ~ IDj/12 dx -
1 10 cl/1 2dx ~ C 10 ~ IDj/12 dx.
1 (11.18)
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
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
(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
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
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.
350 CHAPTER 11. FUNCTIONAL ANALYSIS METHODS IN PDEs
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
(11.24)
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)
11.1. GENERALIZED DIRICHLET PROBLEM 351
where
sn sn
M-L~~-~-~
]=1 (bi - aD", (b~ - a~)
Hence {Jrm} mEN is a Cauchy sequence in the space L2 (Q) and therefore also
a convergent sequence.
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
u-AT(u)=h (11.28)
01 01 01
Jar hEW (Q), where T W (Q) --+ W (Q) lS a compact selJadjoint positive
operator.
-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
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)).
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 )
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.
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 = 0,
354 CHAPTER 11. FUNCTIONAL ANALYSIS METHODS IN PDEs
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) (<p E W (Q)).
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).
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.,
Solution.
(11.39)
By the definition of the trace of the function Jon S (see Chapter 9.)
fk L2(S)
~
fl
s (11.40)
356 CHAPTER 11. FUNCTIONAL ANALYSIS METHODS IN PDEs
t
We have (see Chapter 9.)
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
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
On the other side, since {In. hEN is a convergent sequence in the space L 2 ( Q)
there exists ko E N such that
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}.
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
vlaQ = o. (11.45 )
Using the Fourier method of separation of variables construct the classical solution
(supposing that there exists)
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.
(L(u))(x, t) (11.51)
;=1 i=1
Applying the operator D; on (11.50) we obtain
00
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,
00 00
where we have taken the expansions of the functions f and 9 with respect to the
eigenfunctions {Vi hEN. Hence
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
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
11.2. THE GENERALIZED MIXED PROBLEMS 359
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
T.ik W2(O,T)
~
T.i as k ......; 00
(.l E N) . (11.63)
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.
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)
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
'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
By (11.68) we obtain
d) Follows by c).
u= LT;Vi
;=1
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;
(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
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)
364 CHAPTER 11. FUNCTIONAL ANALYSIS METHODS IN PDEs
L Ib;(Vil v i)I/212.
00
(fl!) =
i=l
(flf) (11.76)
i=1
L[Tj V iI 7i v ;] (11.78)
i=1
11.2. THE GENERALIZED MIXED PROBLEMS 365
(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
and
IT:(t)12 < 3 (lb;J2),i + Ic;J2 + I faT ai(7) cos( [\;(t - 7)) dTI)
L[TiViITiVi]
i=l
b) The inequality (1l.71) follows by the last inequality in a), the Parseval equality
and the inequality
00
A(u) = f. (1l.80)
Figure 11.1
(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.
(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.
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)
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)))
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.
(11.82)
(u E Bz,n EN).
11.4 Miscellaneous
11.4.1 Preliminaries
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.,
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
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.
(i) S(O) = Ii
a} closed set;
370 CHAPTER 11. FUNCTIONAL ANALYSIS METHODS IN PDEs
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
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
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.
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 11.45 For the mixed type problem for heat equation
au a u 2
< x < 7r,t > 0),
at - ax z (0
=L
00
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
(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.
373
(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)
-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
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
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
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
a) { (1
7r
J
+ J2x
2)} jEN
j
Solutions.
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
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) .
j /2 if Ixl ~ ~, if Ixl ~ ~,
h(x) = { J and J
o otherwise, otherwise.
If cp E V'(R), then, using the mean value theorem for definite integrals, we obtain
J
IJj
(gj, cp) = j2 cp(x) dx. (12.9)
-I/j
(
x
-1 ) _
,cp -
;00 cp(x) - cp(o) H(1 - x) d
x (cpEV), (12.10)
x
-00
I if x > 0,
H(x) ={
o if x :S O.
X-I. X = 1, (12.11)
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
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
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
Jr.p(x)dx = (1,r.p).
00
=
-00
Example 12.4 Show that the following two distributional products exist:
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).
x 0 = xl x=o 0= 0.
380 CHAPTER 12. DISTRIBUTIONS IN THE THEORY OF PDEs
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
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.
Solutions.
a) For cp E 'D(R) 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
o= (0, cp).
12.1. BASIC PROPERTIES 381
for j > p,
for J = p,
and therefore we have from (12.17)
p(x) . T =0 (12.18)
382 CHAPTER 12. DISTRIBUTIONS IN THE THEORY OF PDEs
(12.19)
Moreover, the solutions of (12.19) (hence alsQ of (12.18)) are of the form
m
T = 'I:.,Cj oai'
j=l
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
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
((x - a) T, 'l/J(x)).
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
12.1. BASIC PROPERTIES 383
(Oikl, (x - a) cp(x))
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
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)
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
(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) :
= -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
12.1. BASIC PROPERTIES 385
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)
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
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
386 CHAPTER 12. DISTRIBUTIONS IN THE THEORY OF PDEs
Example 12.9
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
J x(x) dx =
+00
1.
-00
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
J cp(x)dx=(C,cp)
+00
(y,cp)=C
-00
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).
388 CHAPTER 12. DISTRIBUTIONS IN THE THEORY OF PDEs
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) 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 < b < f), a plane heat source of constant strength q. If both ends
(x = and x = e) of the rod are kept at the temperature OOC, find the temperature
u = u(x,t) in the rod fort > 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(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
Solution.
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
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
+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)
/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.
392 CHAPTER 12. DISTRIBUTIONS IN THE THEORY OF PDEs
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
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
88(0,e)
J (E--r.p-
or.p OE)
on on dS.
12.2. FUNDAMENTAL SOLUTIONS 393
Therefore we have
~ (_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
We shall prove that the last limit is equal <p(0), i.e., < ~, <p > . Using the polar
coordinates we have that dS. = c: n - l dSl and so
=
r=.
Jc: ~~ dSl + (n - 2)r=.J<p(c:, 0 1 , ... , On-d dSl. (12.39)
Since
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
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 the function <P was chosen as in Example 12.22. Since then on the domain of
integration it holds jPN(OI 2: C (1 + 1~12)N, the last integral converges for N > 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)
Rn-J
J J
!S(n)=~(')
Frp( -0 d~n d~
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
[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.
[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.
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.
[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.
[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.
[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) .
[31) Taylor M.E., Partial Differential Eql1,ations I,II, III, Springer-Verlag, 1996.
[33) 'rreves, F., Basic Linear Partial Differential Equations, Academic Press, 1975.
[37] Zeidler E., Nonlinear Functional Analysis and its Applications, Vol. IV: Ap-
plications to Mathematical Physics, Springer-Verlag, New York, 1988.
Index
401
402 INDEX
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