!

іс

Series Sn I

Mathematical

hvdcs.
NUNC COCNOSCO EX PARTE

TRENT UNIVERSITY
LIBRARY
 Digitized by the Internet Archive
in 2019 with funding from
Kahle/Austin Foundation

https://archive.org/details/partialdifferent0001tikh
PARTIAL DIFFERENTIAL EQUATIONS
OF MATHEMATICAL PHYSICS

Volume I
 HOLDEN-DAY SERIES IN MATHEMATICAL PHYSICS

Julius J. Brandstatter, Editor

S. G. Lekhnitskii, Theory of Elasticity of an Anisotropic Elastic Body

D. A. Pogorelov, Fundamentals of Orbital Mechanics

A. N. Tychonov and A. A. Samarski, Partial Differential Equations of

Mathematical Physics

M. M. Vainberg, Variational Methods for the Study of Nonlinear Operators

 A. N. Tychonov and A. A. Samarski

PARTIAL DIFFERENTIAL EQUATIONS

OF MATHEMATICAL PHYSICS

Volume I

Translated by S. Radding

HOLDEN-DAY, INC.
San Francisco, London, Amsterdam
1964
© Copyright 1964 by Holden-Day, Inc., 728 Montgomery Street,
San Francisco, California. All rights reserved. No part of this
book may be reproduced in any form, by mimeograph or any
other means, without permission in writing from the publisher.

Library of Congress Catalog Card Number: 64-15462

Printed in the United States of America

ONULP
 EDITOR’S PREFACE

This text reflects the authors’ unique approach to the study of the basic
types of partial differential equations of mathematical physics. The system¬
atic presentation of the material offers the reader a natural entrde to the
subject. Each of the basic types of equations which are to be studied is
motivated by its physical origins. The derivation of an equation from the
physics to its final mathematical structure is very instructive to the student.
The authors have gone to great length to make clear the meaning of a
solution to an initial value or boundary-value problem. Various methods of
solving such problems are treated in great detail, as are the questions of
existence and uniqueness of solutions. Thus, the student gains an apprecia¬
tion of the theoretical foundations of the subject and simultaneously acquires
the manipulative skills for solving such problems.
The exercises which accompany each chapter have been selected to test
the student’s ability both to formulate the correct mathematical statement
of the problem and to apply the appropriate method for its solution. The
applications treated by the authors are non-trivial and are completely worked
out in detail.
The present volume covers the two dimensional class of partial differential
equations of mathematical physics and is well suited as a basic text for both
the undergraduate and graduate level at the university. The second volume
will cover the three dimensional counterparts of the present volume and
contain an additional chapter on the special functions which arise in mathe¬
matical physics.
In view of the dearth of texts having the scope and depth of the present
one, Holden-Day felt the need for a careful and articulate translation of
Tychonov and Samarski’s valuable contribution to the literature in the field.
The translator and editor exercised great care in making the English edition
read smoothly and correctly. Several misprints and other flaws have been
corrected, and we are pleased to make this excellent text available to the
English-reading scientific community.

Julius J. Brandstatter
Stanford Research Institute
 FROM THE PREFACE TO THE RUSSIAN EDITION

The class of problems of mathematical physics are closely bound to dif¬

ferent physical processes. To these belong the phenomena which are treated
in hydrodynamics, in the theory of elasticity, in electrodynamics, etc. The
mathematical problems which are associated with these phenomena contain
many common features and form the subject matter of mathematical physics.
For this purpose, a knowledge of the characteristic methods of investiga¬
tion is in its nature mathematical; however, the representation of the prob¬
lems of mathematical physics proceeds in certain directions. Thus, for
example, the initial and final states of a qualitative process are of different
character and therefore require different mathematical methods for their
description.
We have attempted to choose the subject matter following conventional
physical viewpoints; these also correspond to the arrangement of the basic
types of differential equations.
Each chapter begins with the simplest physical problems which lead to
equations of the type considered therein, and each chapter contains exercises
of which several are also of physical interest.
Since the simplest problems treated in the text do not accurately rep¬
resent the multiplicity of the problems and their role in mathematical
physics, each chapter concludes with physical applications in which the
methods first presented in the main body of the chapter are used for the
solution. We have also included examples which go beyond the scope of the
problems treated in the text.
The text contains only a part of the material which comprises the
methods of mathematical physics; thus we have not treated integral equa¬
tions in detail nor have we discussed variational methods. Approximate
methods of solution are only briefly mentioned.
The present text is an outgrowth of lectures given by A. Tychonov
before the Physics Faculty of the Moscow Lomonossov University. We wish
to thank our students and our colleagues, V.B. Glasko, V.A. Iljin, A.V.
Lukianov,O.I. Panuch,B.L. Roshdestvenski, A.B. Vassiliewa, A.G. Svechnikov,
and D.N. Chetaev, without whose help we could not have compiled this book
in a reasonable time, and to J.L. Rabinovich who read the manuscript and
gave us valuable suggestions. We wish to thank both V.I. Smirnov for
many valuable comments and A.G. Svechnikov for his aid in the prepara¬
tion of the second edition.

A. Tychonov
A. Samarski
 PREFACE TO THE GERMAN EDITION

Just as geometric representation is necessary for the study of differential

and integral calculus, so physical representation is required in the study of
partial differential equations. In this book, we have placed most emphasis
on the formulation of the basic problems of mathematical physics and on the
physical interpretation of the solutions of boundary value problems of partial
differential equations. We hope that our book will prove useful as an intro¬
duction to mathematical physics.
We wish to take this opportunity to give thanks to Dipl. Math. Gerhard
Tesch for his careful translation, to Dr. Herbert Goering and Dr. Wolfgang
Schmidt for the scientific editing and to VEB Deutscher Verlag der Wissen-
schaften for the publication of this book.

Moscow, January 1959 A. Tychonov

A. Samarski
 TABLE OF CONTENTS

1. CLASSIFICATIONS OF PARTIAL DIFFERENTIAL EQUATIONS OF

THE SECOND ORDER
1-1. Differential equations of the second order with two
independent variables 1
1-2. Differential equations of the second order with several
independent variables 7
1- 3. The canonical forms of linear equations with constant coefficients 9

2. HYPERBOLIC DIFFERENTIAL EQUATIONS

2- 1. Simple problems which lead to hyperbolic differential
equations and boundary-value problems 12
2-2. Wave-propagation method 36
2-3. Separation of variables 66
2-4. Problems with auxiliary conditions on the characteristics 101
2-5. Solutions of general linear hyperbolic differential equations 107
2- 6. Applications to Chapter 2 117

3. PARABOLIC DIFFERENTIAL EQUATIONS

3- 1. Simple problems which lead to parabolic differential equations 153
3-2. The method of separation of variables 171
3-3. Problems for the infinite straight line 188
3- 4. Problems without initial conditions 210
3.5. Applications to Chapter 3 215

4. ELLIPTIC DIFFERENTIAL EQUATIONS

4- 1. Problems which lead to Laplace’s differential equation 241
4-2. General properties of harmonic functions 252
4-3. Solutions of the boundary-value problems for simple
regions by separation of variables 270
4-4. Green’s function (source function) 279
4-5. Potential theory 288
4-6. The difference method 325
4-7. Applications to Chapter 4 332

APPENDIX
Tables of error integrals and some cylindrical functions 363
Literature references of the editor 371
 1

CLASSIFICATION OF PARTIAL DIFFERENTIAL EQUATIONS

OF THE SECOND ORDER

Many problems of mathematical physics lead to partial differential equa¬

tions. Differential equations of the second order occur most frequently; in this
chapter we shall consider their classification.

1-1. DIFFERENTIAL EQUATIONS OF THE SECOND ORDER WITH

TWO INDEPENDENT VARIABLES

A relation between an unknown function u(x,y) and its partial derivatives1

up to and including the second order derivatives is designated as a partial
differential equation of the second order in the two independent variables x
and y:

*p(x, У, My U X у Му у Uxx у Мху у Муу) 0 .

The equation has an analogous form for more than two independent variables.
A partial differential equation of the second order is called linear with
respect to the highest derivative, if it has the form

auHxx + 2a^uxy + a22uyy + F(x, у, и, их,му) = 0 , (1-1.1)

where the coefficients an, «12, and a22 are functions of x and y.
If the coefficients are not only functions of x and y, but also F is a function
of x, у, M, Mx, and My у then it is called a quasilinear differential equation.
The equation is called linear if it is linear in the higher derivatives
Mxx, Mxy, Myy, also in the function u(x, y), and in the first derivatives ux, му\

auMxx T 2a.[2Mxy T a22uyy + biUx + b2My + cu + f — 0 , (1-1.2)

where an, a12, a22, blt b2, c, and / are functions which depend only on x and
y. If the coefficients are independent of x and y, then Eq. (1-1.2) is a linear
differential equation with constant coefficients. Equation (1-1.2) is called
homogeneous if f(x, y) = 0.
With the aid of a unique inverse transformation

1 For the derivatives we use the symbols

ди ди дги d2u d2u
Ux — - , Uy = , Uxx — a 2 » Uxy — л Д у иУУ ~ 3 2 •
ox ay ox oxoy oy
2 PARTIAL DIFFERENTIAL EQUATIONS OF THE SECOND ORDER

f = <p(x, y), 7} = ф(х, у)

we obtain, under certain assumptions on <p and ф, a new differential equation
which is equivalent to the original equation. There now arises the question
of how the variables £ and tj are to be selected so that the transformed dif¬
ferential equation assumes as simple a form as possible. This question will
be answered now for a linear equation of the form (1-1.1) with two independent
variables x and y.
If we transform the derivatives to the new variables, we obtain

Ux UtJ^x “Ь Uxi7Jx

Uy - U£$y у

Uxx = Utffx + 2 UfrZxTlx + Ur,v71x + щ£хх + Щ7ІХХ (1-1.3)

Uxy — U££$x$y "Ь U%x,(£ x:7]у ~Ь $y7Jx) Ux,r,7] x7] у “Ь U^xy “b UrfQxy

‘2
Uyy - U££^ у + 2u&t£y7)y ~b UrjXqTJ у ~b U^yy + UxqTjyy .

If these expressions are inserted into (1-1.1) an equation results of the form

anu& + 2а.хгЩх) + al2iir,r, + F = 0 , (IT-4)

where

йп = flnfi + 2a12£x$y + Un^y

a 12 = an£xT)x + аіг(£х7]у + v^y) + «22?^,
fl22 = ЙЦ?! + 2ац7]х7]у + аггѴу

and F is independent of the partial derivatives of the second order of w(£, >?)
with respect to f and rj. If the initial equation is linear, that is,

F(x, y, u, ux, uv) = biUx + b-iiiy + cu + f{x, y) ,

then F has the form

F(f, ri, и, щ, Ur,) = щ + 02ІІУ, + к и + <J(£, rj) ;

that is, the transformed differential equation is likewise linear.2

We now want to choose the transformation such that the coefficient an
vanishes. To this end, we examine a partial differential equation of the first
order of the form

flu z\ + 2a,\iZxZy + fl222y — 0 (1-1.5)

Let z = <p(x, y) be an arbitrary particular solution of this equation. If we set

£ = <p(x,y), then the coefficient an is obviously equal to zero. In this manner
the above-mentioned problem of the selection of the new independent variables
£ and rj is linked with the solution of Eq. (1-1.5).
First we shall prove the following lemmas.
Lemma 1. If z = <p(x,y) is a particular solution of the equation

2 If the transformation of the variables is linear, then F = F, since the second deri¬
vatives of f and 7] vanish in formula (1-1.3), and in the expression for F none of the
transformations of the second derivatives appears in the preceding sums.
 1-1. DIFFERENTIAL EQUATIONS WITH TWO INDEPENDENT VARIABLES 3

Йп2* + 2йі2ZxZy + «22Z2y — 0 ,

then (p{x, у) — C is a general integral of the ordinary differential equation

andy2 — 2al2dxdy + a22dx2 = 0 . (1-1-6)

Lemma 2. Conversely, if <p(x, у) = C is a general integral of the ordinary

differential equation

ЙЦdy2 — 2andxdy + a22.dx? = 0 , (M.6)

then z = (p{x,y) satisfies Eq. (1-1.5).

Proof of the first lemma. Since the function z = <p(x,y) satisfies Eq. (1-1.5),
the equation

aj&Y- 2aJ-^) + й22 = 0 (1-1.7)

\ <Py/ \ 4>y)

is valid for all x, у of the region in which z = <p(x, y) is defined and <py(x, у) Ф 0.
The relation <p(x,y) = C is a general integral of Eq. (1-1.6) if the function у
defined implicitly by <p(x, у) = C satisfies Eq. (1-1.6). If, namely, y=f(x,C)
is this function, then it satisfies

so that у = f(x, C) satisfies Eq. (1-1.6). The expression in the brackets vanishes,
not only for у = f(x, C), but for all values of x, y.
Proof of the second lemma. Let <p(x,y) = C be a general integral of Eq. (1-1.6).
We can show that for each point (x, y)

+ 2al2<Px<Py + а22<р2у — 0 (1-1.7)

is valid. Let (x0,y0) be any given point. If it can be shown that Eq. (1-1.7)
is satisfied at this point, it follows that Eq. (1-1.7) is valid at all points, since
(х0,Уо) is arbitrarily chosen. The function <p(x,y) then represents a solution
of Eq. (1-1.7). We now construct through (x0, y0) an integral curve of Eq.
(1-1.6) in which we set <p(x0,yo) = C0 and consider the curve y=f(x,C0).
Obviously, y0 = f(x0,Co). For all points of this curve the following equation
is valid,

(dy У dy Г / <pxV ( <px\

Hill , ) — 2«12~У h ^22 — «111 — 1 — 2«12І — ) + «22 = 0
V dx) dx L \ 4>y) \ 4>y) JJ ?=/<*. c0>

If we set x = x0 in this equation, we obtain

4 PARTIAL DIFFERENTIAL EQUATIONS OF THE SECOND ORDER

dn<p\{xо, Уо) + 2аіг(рх(х0, y0)<py(x0, y0) + a22<p\{x0, y0) = 0 ,

which was to be proved.3

Equation (1-1.6) is called the characteristic equation of the differential Eq.
(1-1.1); its integrals are called characteristics.
If we set f = <p{x, y), where <p(x, y) = const, is a general integral of Eq.
(1-1.6), then we find that the coefficient of и^ vanishes. Likewise the coefficient
of Ur,r, equals zero if ф(х,у) = const, is an additional general integral of (1-1.6)
independent of <p(x, у) [see footnote 5] and if we set rj = ф(х,у).
Equation (1-1.6) yields two equations

dv _ d\2 + л/a\2 — flnfl2г

(1-1.9)
dx йп

dy CL 12 і/ Й12 ЙЦЙ22 ^ -j^ IQ-j

t/x CL л

The sign of the expression under the root determines the type of the dif¬
ferential equation
anuxx + 2al2uxy + a22uyy + F = 0 . (1-1.1)

At the point M we shall say that it is4

of the hyperbolic type if at this point an — ana22 > 0 ,

of the elliptic type, if at this point a-n — апаг2 < 0,
of the parabolic type, if at this point a\2 — ana22 = 0 .

We can easily show the validity of the expression

Й12 ЙПЙ22 = (йі2 йцвгг) (f* f]y £УУх) ,

from which the invariance of the type of equation follows under a trans¬
formation of the variables. At different points of the region of definition,
the equation can be of changing type.
For the following considerations we take as basic a region G, at each
point of which Eq. (1-1.1) is of one and the same type. Through each point
of the region G two characteristics arise which are real and distinct for a
differential equation of the hyperbolic type, complex and distinct for a dif¬
ferential equation of the elliptic type, and real and equal for a differential
equation of the parabolic type. We shall investigate each of these cases
separately.
1. For an equation of the hyperbolic type a\2 — ana22 > 0, the right sides
of the differential Eqs. (9) and (10) are real and distinct. The general

3 This relationship between Eqs. (1-1.5) and (1-1.6) is the equivalent of the well-known
relation between a linear partial differential equation of the first order and a system of
ordinary differential equations. This can be shown if the left side of Eq. (1-1.5) is
represented as the product of two linear differential expressions.
[See V. I. Smirnov: Course in Higher Mathematics, Part II, 2d ed., Berlin, 1958, p. 62,
and V. V. Stepanov: Textbook of Differential Equations, Berlin, 1956, p. 328 (Translated
from Russian).
4 This terminology is taken from the theory of curves of the second order.
1-1. DIFFERENTIAL EQUATIONS WITH TWO INDEPENDENT VARIABLES 5

integrals (p(x, у) — C and ф(х, у) — C of these equations determine a real set

of characteristics. We shall set

£ = <p(x, у) г] = ф{х,у) (1-1.11)

and reduce Eq. (1-1.4) after division by the coefficient of иcv to the form

F
U(i! = Ф(£, у], и, U(, Ur)) with Ф =— ЙГ2 Ф 0
2 a.

This is the so-called canonical form for an equation of the hyperbolic type.5
Frequently, a second canonical form is used. If we set

£ = a + ft f] = a — /3
i.e.,

a =
£ + у n _ £—у
2 ‘ 2
where a and /3 are the new variables, then

«£ — " r) (^aJ + «в) Uy) — 2 (^® Wq) Wfrj — ^ (Маи «Зб) j

whereby Eq. (1-1.4) finally assumes the form

uaa — ирв = Фі Фі = 4Ф

2. For an equation of the parabolic type, an — апагг = 0. Consequently

Eqs. (1-1.9) and (1-1.10) coincide, and we obtain only a single general integral
of Eq. (1-1.6): <p{x, y) = const. In this case, we set

£ = <p(x, y) and t) = r](x, y) ,

where rj(x, y) is an arbitrary function independent of <p. By this choice of
the variables we find

«11 «11 £x + 2«і2 £x £y T «22 £ у — ("і/ «11 £x + «22 £>) 0 ,

5 The introduction of the new variables f and rj through the functions <p and ф is only
possible when these functions are independent of each other. Thus it is sufficient that
the corresponding functional determinant obtained from these functions be distinct from
zero. This is the case here, since if
<Px фх

V>y фу
at any point M were zero, then for this point the columns of the determinant would be
proportional to each other; hence

Vx _ фх
<Py фу

but since

<Px «12 + ~\/a.i2 — ante , фх «і2 — I' аГг — апягг 2 . a

- =-and -X— = —- ai2 — апягг > 0 ,
<py an фу an

this is impossible (without loss of generality we assume an Ф 0). Thus, the independence
of functions <p and ф is demonstrated.
6 PARTIAL DIFFERENTIAL EQUATIONS OF THE SECOND ORDER

since a12 = («п)1/2(а22)1/г; from this it follows that

«12 = an$xrjx + an{£xr]y + £y 7]x) + cintyVy

= (l/cnfx + Ѵаіг£у) {Vaur]x + ѴагггіУ) =0 .
After dividing Eq. (1-1.4) by the coefficient of и„, the canonical form
f
Urn = Ф(£, fj, U, Tj f, Ur,) with Ф =-r— «22 =£ 0
«22

results for an equation of the parabolic type. If, in particular, щ does not
appear in this equation, then it is an ordinary differential equation with £ as
a parameter.
3. For an equation of the elliptic type, fli! - йцй22 < 0, and the right
sides of Eqs. (1-1.9) and (1-1.10) are complex conjugates of each other. Thus,
if
<f(x, у) = C
is a complex integral of the differential Eq. (1-1.9), then

<p*(x, у) = C ,
where <p* is a complex function conjugate to <p, a general integral of Eq.
(1-1.10), and a complex conjugate to (1-1.9). We introduce now complex
variables by setting
? = 4>(X, У) v = 9*(x, У) .

In this way an equation of the elliptic type, as in the case of the hyperbolic
type, is converted to another form.
In order to avoid calculations with complex variables we introduce new
real variables a and p, through
1 * *
a
9 + 9 9-9
(8 =
2 2i
such that
£ = a + ij3 rj — a — i[3 .
Thus we obtain

«11 fx T + «22 ^у
= (fluff! + 2al2axay + a22ay) — (anfi + 2anpxfiy + a22fil)
T 2і(&ц (Xx fix “b «12{oLx fly -}“ OCyflx) T «22 CCу fiy) — 0 ,

from which it follows that

«и — «22 and «12 — 0 .
After dividing by the coefficient of uaa, Eq. (1-1.4) takes the form6

6 Such a transformation is valid only if the coefficients of Eq. (1-1.1) are analytic
functions. Namely, if aL — ana22 < 0, then Eqs. (1-1.9) and (1-1.10) are complex; con¬
sequently, the function у takes on complex values. We can only speak of the solutions
of such equations when the coefficients of aik(x, у) are defined for complex values of y.
For the conversion of the differential equation of the elliptic type to canonical form we
shall limit ourselves to equations with analytic coefficients.
1-2. DIFFERENTIAL EQUATIONS WITH MANY INDEPENDENT VARIABLES 7

uaa + Щs = Ф(а, /3, u, ua, ue) with Ф = #22 0

CL 22
Depending on the sign of the discriminant а\2 — апа2г, the following
canonical forms of Eq. (1-1.1) result:

hyperbolic type: Lixx Uyy — Ф

elliptic type: lixx "h Uyy — ф
parabolic type: Uxx = Ф

1-2. DIFFERENTIAL EQUATIONS OF THE SECOND ORDER WITH

SEVERAL INDEPENDENT VARIABLES

We shall consider now the linear differential equation with real coefficients.

2 2 a%jUxixj T 2 Mi; cu f—0 Uij — an , (1-2.1)

j=1 i=l <= 1

where a, b, c, and / are functions of Xl, x2, • • •, xn. We introduce a new

variable £* by

?4 = Ы*1, *2 , • * •, Xn) k = 1, • • •, n .
Then

Uxj — 42= 1 Wft

n n
Ui{Xj — 4=11=1
2 2 f г оспе an + 42= 1 иt/с)хіх j >

where for brevity аік = д£к/дхі is introduced.

If now we substitute the expressions for the partial derivatives into the
initial equation, we obtain

2 2 akiUtn, + 2 bkU(k + cu + f — 0
4= 1 i = l 4=1

with
n «П
aki = 2 2 ацсцкап
t=l i =l

_ n n n
bk 2 biajk T 2 2 aij($k)xiXi »
»=1 i=l J = 1 J

We now consider the quadratic form

2 2 аЬуіУі, (1-2.2)
<=іj=i

whose coefficients coincide with the coefficient of ац of the initial equation

at a point M0(xi, •••jxl). Under a linear transformation

У>= 2 aikrjk
4=1
8 PARTIAL DIFFERENTIAL EQUATIONS OF THE SECOND ORDER

we obtain a new expression for the quadratic form

n n
1 1 alirjkr]i
k=1 1 = 1

where

flkl — ^ dijOCikOCjl
i= 1 j=1

The coefficients of the principal parts of the equation transform like the
coefficients of a quadratic form under a linear transformation. As is well
known, with the help of a suitable linear transformation the coefficient matrix
(a°j) of a quadratic form can be transformed into a diagonal matrix in which

\a°u \ — 1 or 0 and a\j = 0 i Ф j; i = 1, 2, • • •, n .

This is called the transformation of the quadratic form to the normal form.
According to a theorem of Sylvester, the so-called inertia rule for quadratic
forms, the number of coefficients of a°u distinct from zero is equal to the
rank of the coefficient matrix, and the number of negative coefficients is
invariant.
We call Eq. (1-2.1) at the point M0 of the elliptic type if all n coefficients
of a°i are different from zero and have equal signs; of the hyperbolic (or of
normal-hyperbolic) type, if, likewise, all а°ц Ф 0, n — 1 coefficients of а°ц have
the same sign, and one coefficient is different from the other in the signs;
of the ultrahyperbolic type if m coefficients of a°u are equal and n — m have
opposite signs (m > 1, n — m > 1); and of the parabolic type if at least one
of the coefficients a°u vanishes.
We choose the new independent variables so that at the point M0

_ д£к о
ОСік ^ ОС-ік у
ot\

where aik is the coefficient of the transformation which converts the quadratic
form (1-2.2) to the canonical form, in which we can set = 'Z.a'iV xt. Then
Eq. (1-2.1) at the point M0 can be transformed to one of the following
canonical forms:

elliptic type: u*i*i + MV2 + • • • + u*n*n + Ф - 0

hyperbolic type: Ux.x. = I UHx: + ф

t=2

m n
ultrahyperbolic type: I Ux{Xi = I uHH + Ф
i=l i = m-\-l
n—m
parabolic type: I (± uHH) + 0 = 0 m > 0
i= 1

The further classification of equations of the parabolic type into equations

of the elliptical-parabolic type, hyperbolic-parabolic type, etc., will not be
discussed here.
1-3. LINEAR EQUATIONS WITH MANY INDEPENDENT VARIABLES 9

If Eq. (1-2.1) at a given point M belongs to a definite type, it can be

transformed to the corresponding canonical form.
We now investigate further whether an equation in a definite neighbor¬
hood of a point M can be transformed into the corresponding canonical form,
if at all points of this neighborhood it belongs to one and the same type.
If Eq. (1-2.1) can be transformed to the simplest form in a region in which
the elements of the coefficient matrix off the principal diagonal vanish, then
the functions

£i(Xl i %2 t ***i %n) f — 1 2, * •,

> • Yl

must satisfy the relation аы = 0 for кФ 1. The number of these relations

is equal to n(n — l)/2, and hence for n > 3 it is larger than the number n
of the functions to be determined. For n = 3 the nondiagonal elements of
the coefficient matrix (an) usually can be made to vanish; then, however,
the elements of the principal diagonal can be distinct from each other.
Consequently, for n ^ 3 it is impossible in a neighborhood of the point
M to transform the differential Eq. (1-2.1) to canonical form. For n — 2,
it can happen that the single nondiagonal coefficient of the second-order
matrix vanishes, and the two coefficients on the principal diagonal are equal
to each other as outlined earlier in this section.
If the coefficients of Eq. (1-2.1) are constant, then after a transformation
of (1-2.1) to canonical form at a point M we obtain an equation which has
the same canonical form in the entire region of definition.

1-3. THE CANONICAL FORMS OF LINEAR EQUATIONS WITH CON¬

STANT COEFFICIENTS

In the case of two independent variables, a linear equation of the second

order with constant coefficients has the form

anuxx + 2ai2uxy + a22uyy + b1ux + b2uy + cu + f(x,y) = 0 . (1-3.1)

A characteristic equation with constant coefficients corresponds to it.

Consequently, the characteristics in this case are the straight lines

lyi + a\2
V1 & 11 #2 «12 + Ѵа\г - йп О-гг
У У = C2

After a corresponding transformation of the variables, (1-3.1) assumes one of

the following simple forms:

elliptic type: U(l + U-q-q + Ь\ Mf T 62 Uq ~b CU + f — 0 (1-3.2)

U(q — bi U( + b2 Uq + cu + / = 0 or
hyperbolic type: (1-3.3)
Mff - Uqq + by U{ + bl Uq T CU + f — 0
parabolic type: Mff -)- by и^ T bi Uq cu T f — 0 (1-3.4)

For further simplification we introduce

и = v ,
 10 PARTIAL DIFFERENTIAL EQUATIONS OF THE SECOND ORDER

which yields a new function of v where X and у are still undetermined

constants. Then

Mf = ex^'(vz + Xv)
Ur, = /f+M(f„ + yv)
Mff - eXf+M(fff + 2,XV( + X2v)
Wfl? = eXf+M(ff, + Xvr, + yv^ + Xyv)
tirjrj = e*+M(vr,r, + 2 yv r, + yv)

If we substitute these expressions for the derivatives in Eq. (1-3.2), after

division by we obtain

fee + vVt) + (b, + 2X) ve + (b2 + 2у) Vr, + (i2 + у + b,X + b2y + c) v + f, = 0 .

If in this equation the parameters X and у are so chosen that there are two
coefficients, say in which both of the first derivatives are made to vanish,
that is, X = —(bj2) and у — —(b2l2), then we obtain

fee + Vvv + yv + /i = 0 ,

where у is a constant defined by c, b,, and b2, and f, — /•e_(Xf+,1’’). In the

same manner we can derive the equations corresponding to (1-3.3) and (1-3.4).
Thus, we are led to the following canonical forms for differential equations
with constant coefficients:

elliptic type: fff + Ѵг)Г) + yv + /i = 0

ff>) + yv + /i = 0
hyperbolic type:
fff - Vm + yv + fy = 0
parabolic type: ffe + b2 Vt) + f i = 0

We have already noted (1-2) that a differential equation with constant co¬
efficients in the case of several independent variables,

I I ац инч + I Ь{ uXi + cu + / = 0 ,
t—1 j = 1 i=l

under a suitable linear transformation of the variables, can be transformed

to a canonical form which is the same for all points in the region of defini¬
tion. If now we set
n
_
и = e,=1 f ,

a new function of v is introduced, and if Xi is selected appropriately, the

transformed equation can be further simplified so that in the case n — 2, a
corresponding canonical form obtains.

Problems

1. Determine the region in which

uxx “b уМуу — 0

is hyperbolic, elliptic, or parabolic, and transform the differential equation,

1-3. LINEAR EQUATIONS WITH MANY INDEPENDENT VARIABLES 11

in the region in which it is hyperbolic, to canonical form.

2. Transform the following differential equations to canonical form:
<i) uxx T хущу 0 .
b) yuxx — xuyy + ux + yuy = 0 .
c) в uxx T 2,6 yuXy T 6 Уиуу — 0 .
d) uxx + (1 + y)2uyy = 0 .
e) xuIX + 2л/~ху uxy T yUyy — ux = 0 .
f) (x — y)uxz + {xy — y2 — x + y)uxy = 0 .
g) угихх — e2xuyy + ux - 0 .
h) sin2 yuxx — e2xuyy + 3ux — 5u — 0 .
i) uxx + 2uxy + \iiyy T 2ux T 3tiy = 0 .
3. Transform the following differential equation to canonical form and sim¬
plify it as much as possible:
duxx T 2o.uXy T ciUyy T bux T cUy T и —0 .
4. Simplify the following equations with constant coefficients by introducing
the function v = ueKz+>Ly and by a suitable selection of the parameters X and
V-
a) uxx + Uyy + aux + [іи, -4- yu = 0 .
b) uxx — —2 Wy T оси + pux ,
a

c) uxx-\uyy — aux + fiUy + yu .

a
d) uxy — aux + Puy .
 2

HYPERBOLIC DIFFERENTIAL EQUATIONS

Partial differential equations of the second order of the hyperbolic type

occur principally in physical problems connected with vibration processes.
The simplest hyperbolic differential equation

Uxx Uyy - 0
is usually called the differential equation of the vibrating string. In this
and the following chapters we shall restrict ourselves to the treatment of
linear differential equations.

2-1. SIMPLE PROBLEMS WHICH LEAD TO HYPERBOLIC DIFFERENTIAL

EQUATIONS AND BOUNDARY-VALUE PROBLEMS

1. The differential equation of small transverse vibrations of a

string
Each point of a string of length l can be characterized by the value of
its abscissa x. The vibration of a string can be described by the position
of the points of the string at different times. In order to characterize
the position of the string at time t, it is sufficient to know the components
ux{x, t), u2(x, t), u3(x, t) of the displacement vector u at time t at the point
X.
We shall consider now the simplest problem related to the vibrating
string; we shall assume the displacement of the string takes place in a plane,
for example, the x, и plane, and the displacement vector и always lies per¬
pendicular to the x axis. Then the vibration process can be described by a
single function u(x, t) which characterizes the vertical displacement of the
string. We consider the string to be a flexible-elastic filament. The flexi¬
bility is expressed mathematically by the assumption that the tension in the
string is always in the direction of the tangent to the existing profile of the
string (Figure 1). This requirement states that the string offers no resistance
to bending.
The magnitude of the tension, which arises in the string because of the
elasticity, can be calculated using Hooke’s law. We shall consider only small
vibrations and can therefore neglect the square of ux, since this quantity
is small compared with unity. In accordance with these requirements, we
can calculate the elongation which a segment (xi, x2) of the string undergoes.
2-1. SIMPLE PROBLEMS AND BOUNDARY-VALUE PROBLEMS 13

FIG. 1.

The length of arc belonging to this segment is equal to

S'=[ l/l + {UxY dx ~ x2 — Xi = S .

Thus no elongation of a single segment of the string occurs within the scope
of our required limits of accuracy, whereas according to Hooke’s law it
follows that at each point, the tension T is independent of x, i.e.,

T — T0 = const.
For the projection of the tension on the x and и axes, indicated by Tx and
Tu, we find

Tx(x) = T(x) cos a = -г--1'-- - ~ T(x)

VI + (ux)*

Tu(x) = T(x) sin a « T(x) tg a = T(x)ux

where a is the angle between the tangent to the curve u(x, t) and the x axis.
A tensile force, an external force, and an inertial force act on the seg¬
ment (xi, Хг). The sum of the projections of all these factors on the x axis
must be equal to zero (since we consider only transverse vibrations). Since
the inertial force and the external force act along the и axis according to
hypothesis, there results

Тх(Хг) — Tx(xi) = 0 or T(xt) = Т(хг). (2-1.1)

But since Xi and x2 were selected arbitrarily, it follows that the tension is
not dependent on x; i.e., for all x and t values,

T = To. (2-1.2)

To derive the equation of a transverse vibrating string, we use Newton’s

second law. The total momentum of an element (xx, x2) in the direction of
the и axis is, first of all, equal to

( w«(f, t)p(£)d£ ,
J

where p denotes the linear density of the string. We now set the change
of momentum during the time Jt = t2 — ,
14 HYPERBOLIC DIFFERENTIAL EQUATIONS

/0(6)[И«(£, ti) — Mt(f, ty)]d£ ,

*l

equal to the impulse of the acting force which arises from the tension T0uz
at the points Xi and x2, as well as from the external force. The latter we
assume as continuously distributed with the density p (the load) and denote it
(per unit of length) by f(x, t). Thus we obtain the equation of a transverse
vibrating string in integral form

Ot(f, tt) — ut(£, /i)]|o(f)rff

(2-1.3)
rxz
T0[ux(x2, r) — ux(Xi, t)]dz + \ /(?, T)dZdT
Jxj

In order to go from this integral equation to a differential equation, we assume

the existence and continuity of the second derivative of u(x, t).1 Equation
(2-1.3) is then transformed, after a twice-repeated application of the mean-
value theorem, to

uu(t, t*)p{£*)MAx = {T0[m„(£**, t**)] + /(?***, t***)}JtJx ,

where
£*, ?**, £*** e (Xi, xt) , t*, t**, t*** e (tt, U).
If we now divide by ДxAt and carry out the limit as x2—*Xi, then
we obtain the differential equation of a transverse vibrating string:

T0uxz = putt — fix, t). (2-1.4)

In the case of a constant density p, this equation is usually written in the

form

utt = a2uxx + Fix, t) (^a—-)/~-) . (2-1.5)

where

Fix, t) = —f(x,t) (2-1.6)

P
is the force density per unit of mass. If no external force acts, we obtain
the homogeneous equation

utt — a2uxx or uxx — uyy — 0 , у = at,

which describes the free vibrations of a string. This equation represents
the simplest example of a hyperbolic differential equation.
If a concentrated force f0it) acts at the point x0 where Xi < x0 < хг
(Figure 2), then Eq. (2-1.3) takes the form

1 By the assumption that the function u(x, t) is twice continuously differentiable, we

have practically asserted that we shall consider only those functions that possess this
property. This does not mean, however, that no functions exist which satisfy the vibra¬
tion equation in integral form and have no second derivatives. Such functions exist and
are, in practice, of extraordinary interest. More details will be discussed later (2-7).
2-1. SIMPLE PROBLEMS AND BOUNDARY-VALUE PROBLEMS 15

(u) -ut£, - ГТ/(?,T)dMr

J*1 JxjJq
C<2 r *2
= l Tо[их(хг, z) — ux(x 1, z)]dz + l f0(z)dz .
J«1
Since the velocity of a point of a string is bounded, the integral on the left
side of this equation tends towards zero as xt —> x0 and x2 —* x0 and (2-1.3)
assumes the form
S<2

T0[ux(x0 + 0, r) — ux(x0 — 0, z)]dz = — fo{r)dz . (2-1.7)

«1
By using the mean-value theorem after
dividing both sides of the equation by
At and taking the limit as t2 —> 1t we
obtain
*0+0

ux(x, t)

Hence, we see that the first derivative

has a discontinuity at the point at
which a concentrated force acts. The
differential equation loses its meaning for x= x0. At this point, both conditions

u(xо + 0, t) = u(x0 — 0, t)
(2-1.8)
ux{xо + 0, t) — ux(xо 0 ,t) =~±fo(t)
1 0

must be fulfilled. The first expresses the continuity of the string, while the
second determines the magnitude of the jump of the derivative at x0 in terms
of f0(t) and the tension T0.

2. Differential equations of longitudinally vibrating rods and

strings
Longitudinal vibrations of rods, strings, and springs lead to equations of
the same type. We shall consider a rod which lies in the interval (0, l) of
the x axis. Then the longitudinal vibration can be described by a single
function u(x, t) which represents at time t the displacement of those points
which had abscissa x in the equilibrium state.2 In longitudinal vibrations

2 The geometric variable x chosen here is called the Lagrange coordinate. In the
Lagrange coordinates, each physical point of a rod in the course of an entire process is
characterized by one and the same geometric coordinate x. A physical point, which at
the initial time (in the equilibrium state) is at the point x, is found after an arbitrary
time t at the point with the coordinate X = x+u(x, t). If we choose any geometric point
A with the coordinate X, then at different times different physical points (with different
Lagrange coordinates x) would be found at this point. Frequently we also use the
Eulerian variables X, t, where X is the geometric coordinate. If U(X, t) denotes the
displacement of the point with the Eulerian coordinate X, then the Lagrange coordinate
is x = X — ЩХ, t). An example of the use of Eulerian coordinates is found in § 2-6.
16 HYPERBOLIC DIFFERENTIAL EQUATIONS

the displacement is along the rod. To derive the vibration equation, we

shall assume that the tension which causes the vibration is given by Hooke’s
law.
First we shall calculate the relative elongation of the elements (x, x + Ax)
at time t. The end points of these elements at time t have the coordinate
values

x + u(x, t) , x + Ax + u(x + Ax, t),

so that the relative elongation is equal to

[Ax + u(x + Ax, t) - u(x, f)] - Ax_ = + dJx ^ 0 ^ Ѳ й 1.

Ax
Taking the limit as Ax-^0, it follows that the relative elongation at the
point x is defined by ux(x, t). According to Hooke’s law, the tension T(x, t)
satisfies the equation

T(x, t) = k(x)ux(x, t) , (2-1.9)

where k(x) denotes Young’s modulus at the point x.

By applying the law for the change of momentum we then arrive at the
vibration equation in integral form:

f [ut($, t2) - ut(£, tt)]p(£)d£

= ( [kix2)ux[x2, t) — k{Xi)ux{xi, r)]dr + [ i f(£,z)d£dT (2-1.10)

where f(x,t) is the density of the external force per unit of length.
If now the second derivative of the function u(x, t) exists and is con¬
tinuous, then by use of the mean-value theorem and after taking the limits*
as Ax = хг — Xi —»0 and At = t2 — tt-> 0 we obtain

[k{x)ux]x = puH — fix, t) (2-1.11)

as the equation of a longitudinal vibrating rod.4

If the rod is homogeneous, then this equation becomes

utt = a2uxx -F F(x, t) , (2-1.12)

where

Fix,t) = f(x’{)- (2-1.13)

P

represents the density of force per unit of mass.

3 In the following we shall waive the discussion of such details for a transverse
vibrating string.
4 The requirement that the vibrations be sufficiently small depends in the present
case only on the limits of applicability of Hooke’s law. Generally, T = k{x, ux)ux; we
then arrive at the quasilinear equation [k(x, ux)ux]x = putt — fix, t).
2-1. SIMPLE PROBLEMS AND BOUNDARY-VALUE PROBLEMS 17

3. Energy of vibration of a string

We now seek an expression for the energy E = К + U of a transverse

vibrating string where К is the kinetic and U is the potential energy. An
element dx of the string moving with velocity ut possesses the amount of
kinetic energy given by

ymv2 = -^-p{x)dx(ut)2,

so that the kinetic energy К of the entire string is given by

K = y\1 p(x)[ut(x,t)]*dx. (2-1.14)

The potential energy of a transverse vibrating string which has the form
u(x, t0) = u0(x) at time t = t0 is equal to the work done in transforming the
string from the equilibrium state to the state u0(x). The profile of the string
at time t is given by the function u{x, t). Thus

u(x, 0) = 0 , u(x, t0) = u0(x).

The element dx under the influence of the resultant tension force is related
to the displacement by
j, du _ j, du
Tuxxdx ,
dx x+dx dx

whereas the time dt is related by means of ut(x, t)dt. The work done is
equal to

|| T0uxxutdx^dt — jT0uxut |o — J T0uxuxtdx^dt

J_ d_
T0{uxfdx + T0uxut I
2 dt
By integration over t from 0 to t we obtain

—^-| T0(ux)2dx |o° + I ToUxut\odt=—j| T0[ux(x, t0)fdx + | T0uxut\[dt.

The meaning of the latter terms of this equation is easy to see; indeed,
ToUx U=0 is the magnitude of the tension at the end point x = 0; ut(0, t) is
the displacement of the end point while the integral

j ToUxUt\x^dt (2-1.15)

represents the work which must be expended by the displacement of the

end point x = 0. The meaning of the terms corresponding to the case x = l
is analogous. If the end points of the string are fixed, then the work done
at these points is zero; here u(0, t) — 0 and ut(0, t) = 0. Consequently, by a
displacement of a string with fixed end points from the equilibrium state
и = 0 to a state u0(x) the work done does not depend on the manner in which
the string is brought into this new state; indeed the work equals
18 HYPERBOLIC DIFFERENTIAL EQUATIONS

T,[u'o(x)Ux, (2-1.16)

i.e., it equals the potential energy of the string at time t = t0 but with op¬
posite sign. Therefore we have

E = y\‘[To{Ux)2 + pWutfWx (2-1.17)

as the total energy of the string. We similarly obtain the potential energy
of a longitudinal vibrating rod. Moreover, we arrive at the potential energy
of the rod starting from the formula

From this it follows directly that

U= k(ux)2dx.

4. Derivation of the equation of electrical vibrations in con¬

ductors

The passage of an electric current through a conductor with distributed

parameters can be characterized by the current strength i and the voltage v,
which are functions of the position x and the time t. From Ohm’s law for
an element of the conductor of length Ax it follows that the decrease in
voltage in the element of the conductor Ax is equal to the sum of the electro¬
motive forces:
— vxAx = iRAx + itLAx , (2-1.18)

where R is the resistance and L is the coefficient of self-induction (both are

expressed per unit of length).
The amount of electricity flowing through the conducting element Ax
during the time At is given by

[i(x, t) — i(x + Ax, t)]At —— ixAxAt (2-1.19)

and is equal to the sum

C[v(x, t + At) — v(x, t)]Ax + GAx • vAt = (Cvt + Gv)AxAt, (2-1.20)

which is the amount of electricity necessary for the charging of the element
Ax plus the amount which is lost due to insufficient insulation, where C is
the capacity and G is the loss coefficient (both are expressed per unit of
length). We assume that the magnitude of the loss is proportional to the
voltage at those points of the conductor under consideration.
From (2-1.18), (2-1.19), and (2-1.20) we then obtain the so-called system of
telegraphic equations5

5 These equations within the structure of the theory of electromagnetic fields have
only approximate validity, since the electromagnetic vibrations in the material-filled
space are neglected.
2-1. SIMPLE PROBLEMS AND BOUNDARY-VALUE PROBLEMS 19

ix Cvt Gv — 0
(2-1.21)
vx 4" Lit T Ri = 0 .
In order to obtain only a single equation defining the function i, we dif¬
ferentiate the first of the two equations (2-1.21) with respect to x and the
second with respect to t after we have multiplied these by C. By subtrac¬
tion of the equations thus obtained we then find, under the assumption that
the coefficients are constant,

ixx T Gi>x — CLitt — CRit — 0 .

If we insert for vx the value given by the second equation of (2-1.21), we
obtain the differential equation for the current strength in the conductor

ixx — CLitt + (CR + GL)it + GRi. (2-1.22)

Analogously, the equation for the voltage reads

vXx = CLvH T (CR + GL)vt + GRv . (2-1.23)

The differential Eq. (2-1.22) or (2-1.23) is called the telegraphic equation.

If the insulation loss can be neglected and the resistance is very small
(G ~« 0), we obtain from (2-1.23), the well-known vibration equation

vtt = a'Vxx , a =j/. (2-1.24)

5. Transverse vibrations of a membrane

By а membrane we mean a sufficiently thin elastic film whose boundary

is stretched firmly into a closed-plane curve C and offers no resistance to
stretching and distortion. Of interest are the transverse vibrations of the
membrane by which the stretching occurs perpendicular to the plane of the
membrane.
Let ds be the element of arc length of an arbitrary closed curve lying on
the membrane surface through the point M(x, y). A tension of magnitude
Tds acts on this element. Based on the above hypothesis about the mem¬
brane, the vector T lies in the tangent plane of the instantaneous membrane
surface and is perpendicular to the element of arc ds. On the same basis
the magnitude of the tension is also independent of the direction of the
element ds. The stress vector T = T(x, y, t) is therefore a function of x, y,
and t alone. Obviously this is the mathematical expression for the absence
of resistance to stretching or distortion.
In the following, we will investigate the small vibrations of a membrane
in which the square of the first derivatives ux and uy can be neglected.
From this assumption it then follows directly that the projection Th(x,y,t)
of the tension on the x, у plane is equal to the absolute magnitude of the
tension. For arbitrary orientation of the arc element ds, namely, the angle
y' between the vector T and the x, t plane is not larger than the angle у
which is formed by the normal to the membrane surface at the point (x, y)
and the z axis. Thus,
20 HYPERBOLIC DIFFERENTIAL EQUATIONS

COS y' Si COS у

V 1 -\~ иx u)
and

Tk(x, y, z, t) = T cos y' « T(x, y, z, t). (2-1.25)

The vertically acting tension is obviously equal to Tu = T(du/dn). Now we

choose an element on the membrane surface whose projection on the x, у
plane forms a rectangle ABCD, whose sides are parallel to the coordinate
axes (Figure 3). On this element acts a tension force of magnitude defined by

Г* = <f Tds. (2-1.26)

JABCD

Since along the x and у axes no displace¬

D c ment occurs, the projections of T* on
these axes equal zero; thus

x T(x2, y, T(xx ,y, t)dy

a в A

I J
X = Xy {T(x2, y, t) — T(xi, y, t)}dy = 0
yi
FIG. 3.

or

T* =[ {T(x,y2,t) — T(x,yl,t)}dx = 0 .
J
Hence according to the mean-value theorem, since the surface ABCD is arbi¬
trarily chosen, it follows that

T(x, уi, t) = T(x, y2, t)

T(xlty, t) = T(x2, y, t) ,

i.e., the tension T does not vary with x and у and therefore can only be a
function of t alone.
The surface area of a membrane element at time t in the sense of our
approximation is equal to

= \\Vl+u\ + u\ixd, = j\dxdy . (2-1.28)

Consequently, no extension occurs during the vibration process, whereby

according to Hooke’s law the independence of the tension on time follows.
Therefore, the independence of the tension on all three variables x, y, and
t is proved; i.e.,
T{x, y, t) = const. = T0. (2-1.29)

We shall now derive the equation of a vibrating membrane. For this

purpose we proceed from the law for the change of momentum. Let Si
be the projection of any one membrane element on the x, у plane, and let
2-1. SIMPLE PROBLEMS AND BOUNDARY-VALUE PROBLEMS 21

Ci be the boundary of Si. We set the change of momentum equal to the

impulse of the vertical components of the tension and the externally acting
force whose density is f(x,y,t). Then we obtain the equation of the vibrat¬
ing membrane in the integral form

j У, t2) - ut[x, y, t,)]pix, y)dxdy

= Г( rj^-dsdt + f* f (fdxdydt, (2-1.30)

JqJcq °n Iqls,.)
where pix, y) is the surface density of the membrane and f(x, y, t) is the
density of the external force.
In order to go from this equation to a differential equation we shall
assume the function u(x, y, t) to be twice continuously differentiable. First
we transform the line integral by means of Green’s theorem6 into a surface
integral

\ ^ndS = i \"Uxx + u^dxdy •

Then the above vibration equation is transformed from the integral form into

\ \ {putt — T0(uxx + uyy) — f{x, у, t)}dxdydt = 0 .

ilJSj J

If we now apply the mean-value theorem and take into consideration that
both Si and the time interval (G, t2) are arbitrary, then we see that the ex¬
pression in the brackets must vanish identically. In this manner we arrive
at the differential equation of a vibrating membrane:

pUa Toft xx “f” Uyy) T fix, у, t). (2-1.31)

For a homogeneous membrane the vibration equation can also be written in

the form

utt — a2iuxx + Uyy) + Fix, y, t), a2 — (2-1.32)

P

where Fix, y, t) is the force density per unit of mass of the membrane.

6. Basic equations of hydrodynamics and acoustics

In order to describe the motion of а fluid one uses three functions vdx,y, z, t),
v2ix,y,z,t), and v3ix, у, z,t), which are the components of the velocity vector v
at the point ix,v,z) at time t. Further quantities for the characterization of
a fluid motion are the density p(x,y,z,t), the pressure pix,y,z,t), and the
density of the external force given in units of mass Fix,y,z,t), in case such
forces are present.
We consider a fixed portion of a fluid occupying a region of space T, and
calculate the force acting on it. Moreover we shall neglect the friction forces

6 See V. I. Smirnov, Textbook of Higher Mathematics, 2d ed., Part II, Berlin, 1958,
p. 175; in the literature it is also called Gauss’ integral theorem.
22 HYPERBOLIC DIFFERENTIAL EQUATIONS

caused by the viscosity, i.e., we consider an ideal fluid. For the resultant
of the pressure forces we obtain the following expression in the form of a
surface integral

-JJ/>ndS, (2-1.33)

where 5 is the surface of the region T and n is the unit vector in the direc¬
tion of the outward directed normal. The formula of Green then yields

— I ^pndS = — j grad pdr . (2-1.34)

For the calculation of the acceleration of any fluid element, naturally the
displacement of the points themselves are to be considered. If x = x(t), у —
y(t), and 2 = z(t) are the paths of the points of this element, then the derivative
of the velocity with respect to time is

dv dv dv dv dv dv dv dv dv dv
—— = —- + -т-x + ——y + z= Vi + ——v2 + ——va = —- + (vp)v ,
dt dt dx dy dz dt dx dy dz dt
where

This derivative with respect to time, which takes into account the motion of
a single element of the medium (the substance), is called the substantive
derivative. The equation of motion of a fluid which describes the connection
between the acceleration of the element and the forces acting upon it is
given by

SlS4-*4lSgrad*Hi>№- (2-1'3S

The last integral represents the resultant of the external forces which act on
the region of space T. But since T was chosen arbitrarily we obtain the
equation of motion of an ideal fluid in Eulerian form:

vt + (vp)v =—-grad£ + F. (2-1.36)

P

To derive the needed continuity equation, we shall assume that at all points
of the streaming fluid region T no additional fluid is introduced, i.e., we limit
ourselves to the study of source-free streams, in which we regard a sink as
a negative source. Then the change in the amount of fluid contained in T
per unit of time is equal to the flux through the boundary S of the region

d_
^ \^pdz=— J \^pvndS. (2-1.37)
dt
The transformation of the surface integral to a volume integral yields

+ div pv \dv = 0 .
2-1. SIMPLE PROBLEMS AND BOUNDARY-VALUE PROBLEMS 23

Since this relation is valid for an arbitrarily small region, there follows the
continuity equation

-уу + div (pv) = 0

at
or

+ v grad p 4- p div v = 0 . (2-1.38)

at

To Eqs. (2-1.36) and (2-1.38) there is still to be added the thermodynamic

equation of state which we shall write here in the form

P = Ap)-

Thus we obtain a system of five equations for the five unknown functions
vx, vy, v,, p, and p. If the equation of state also includes the temperature,
then a heat-conduction equation must be added. The system of equations

-fy- + (vp)v = F —— grad p

ot p

-£r + div (pv) = 0 (2-1.39)

ot
P=f(p)
thus represents the complete system of equations for the motion of an ideal
fluid.
To use the hydrodynamic equations for the propagation of sound in a gas,
we make the following assumptions: (1) no external forces are present; (2)
the process of sound propagation in a gas proceeds adiabatically, so that
Poisson’s adiabatic equation

can be taken as the equation of state (p0 designates the initial density and p0
is the initial pressure, cp and cv are the specific heats at a fixed pressure and
a fixed volume); and (3) the vibrations of the gas are sufficiently small so
that the higher powers of the velocity, the gradients of the velocity, and the
gradients of the density can be neglected.
For the condensation of the gas we designate the relative change in density
by the quantity

P
s(x,y, z, t) (2-1.40)
po

Then the condensation is defined by

p Pod + s) (2-1.41)

With the above assumptions the hydrodynamic equations take the form
24 HYPERBOLIC DIFFERENTIAL EQUATIONS

Vt——- grad p
P0
pt + po div I? = 0 (2-1.42)

P = A>(1 + s)y ~ Po( 1 + rs)

since

— grad p = —(1 — s + • • •) grad p =— grad /> + •••

/0 |O0 Po

div pv = v grad p + p div v = p0 div !> + •••

where the dots denote the terms which are of second and higher order of
smallness. With the notation a2 = rPolpo we rewrite the system (2-1.42) in
the form

Vt = a grad s (2-1 42')

st + div v = 0 .
If we now apply the divergence operator to the first equation of (2-1.42') and
interchange the order of differentiation, we obtain

div v — — a2 div (grad s) — — a2p2s — — ci As ,

at
where

2 d2 d2
V
dy2 dz2
is the Laplace operator. Then the second expression of (2-1.42') can be used
to yield the vibration equation

(2-1.43)

or
a (sxx T Syy T Szz) — s*t •
From this and from (2-1.40) we obtain the density equation

a (pxx T Pyy T pzz) — ptt •

We now introduce the velocity potential and show that it satisfies the
same vibration Eq. (2-1.43), as does the condensation. From the expression

Vt — — a2 grad s
follows

v(x, y, z, t) = v(x, y, z, 0) — a2 grad (2-1.44)

where v(x, y, z, 0) is the initial velocity distribution. If the velocity field at

the initial time possesses the potential

»lt=o =- grad f(x, y, z), (2-1.45)

2-1. SIMPLE PROBLEMS AND BOUNDARY-VALUE PROBLEMS 25

we can obtain the expression

v = — grad|y(x, y, z) + a2^ = — grad t/. (2-1.46)

This means that a velocity potential U(x,y,z,t) exists. The knowledge of

the velocity potential yields for the determination of the complete process
of motion the following:

v — — grad U
(2-1.47)

By inserting these quantities into the continuity equation

st + div v = 0 ,

we obtain the vibration equation for the potential:

a2(Uxx -4- Uyy + Uzz) — Utt

or

Utt = cfAU. (2-1.48)

Equations for the pressure p and the velocity v similar to Eq. (2-1.48) can
be derived. These are usually called the equations of acoustics.
For the solution of two- and one-dimensional problems the Laplace opera¬
tor is replaced in Eq. (2-1.48) by дг/дх* + д2/ду2 or дгІдх2.
For the vibrations of a gas in a bounded region of space, specific bound¬
ary conditions must be fulfilled; that is, certain restrictions are imposed which
are to be applied directly on the sought functions at the boundary of the
gas-filled region of space. The simplest example is for the case of a gas
moving in a vessel whose walls are fixed. Since the stream velocity must
always be directed tangentially on these walls, the normal component of the
velocity must be equal to zero. This leads to the condition

or (2-1.49)
дп л-
The constant

possesses the dimension of velocity and, as will be shown later, represents

the velocity of propagation of sound.
We shall now calculate the velocity of sound in air at normal atmospheric
pressure. In this case, we have 7- = 7/5, p0 = 1.293 • 1СГ3 gem-3, and p0 =
1.033 kg cm-2; consequently,
26 HYPERBOLIC DIFFERENTIAL EQUATIONS

7. Boundary and initial conditions

For the mathematical description of a physical process, a problem must

first of all be defined; that is, the conditions must be formulated which are
sufficient for the unique determination of the process.
Both ordinary and naturally partial differential equations possess, in
general, infinitely many solutions. Therefore, in those cases in which the
physical problem to be considered leads to a partial differential equation, for
a unique characterization of the process it is necessary to add further con¬
ditions on the differential equation.
For an ordinary differential equation of the second order the solution
can be determined from the initial conditions, i.e., from the initial value of
the function itself and its first derivative at the initial value of the argu¬
ment. Also possible are other forms of the conditions of state; for example,
the function can be prescribed at two distinct points (problem of the catenary).
For partial differential equations there are, likewise, different forms of the
conditions of state.
We shall consider, first, the simple problem of a transverse vibrating
string fixed at the ends. Here u(x, t) denotes the displacement of the string
from its equilibrium position (x axis). If the ends of the string 0 ^ x ^ l
are held fixed, the boundary conditions which must be satisfied are

u(0, t) = 0 , u(l, t) = 0 . (2-1.50)

On the other hand, there are initial conditions to be satisfied, i.e. the dis¬
placement and velocity of the string at the initial time t0, say

и (x, t0) = <p{x)

(2-1.51)
ut(x, t0) = ф{х) .
The conditions of state therefore consist of boundary and initial conditions.
Later we shall show that these conditions completely determine the solution
of the vibration equation

utt = a2uxx. (2-1.52)

If the end points of the string move in a prescribed way, then the boundary
conditions assume another form:

u{0, t) = frit) ,
(2-1.50')
U(l, t) = fjt2(t) ,
where (i^t) and fi2(t) are prescribed functions of the time /. The statement
of problems for longitudinal vibrating rods or springs reads analogously.
Boundary conditions occur also in other forms. We shall consider, for
example, a longitudinal vibrating spring which is fastened at one end (sus¬
pended point) while the other is free to move. The motion of the free end
is not known and is thus the function to be determined. At the point of
suspension x = 0 there can be no displacement,

u(0,t) = 0;
2-1. SIMPLE PROBLEMS AND BOUNDARY-VALUE PROBLEMS 27

while at the free-end point x = l the spring tension is given by

T(l, t)= кр¬ (2-1.53)

ах x—l

and equals zero (no external force), so that the mathematical formulation of the
conditions for the free end takes the form

ux(l, t) = 0 .
If the fixed end x = 0 moves according to a definite law //(/), whereas at x= l
the prescribed force v(t) acts, then we write

u(0, t) = fj(t), ux(l, t) = v(t) with v(t) — —-Ht).

k
Typical also is a condition of an elastic constraint, say at x = /:

kux(l, t) = —au(l, t)
or

ux(l, t) = — hu(l, t), h=— , (2-1.54)

k
about which the end point x = l can be displaced, while the elastic force of
the constraint at this point gives rise to a tension which causes the displaced
point to return to its earlier position. The rigidity of the constraint is
characterized by the coefficient a.
If at a point of a system at which an elastic constraint acts, the dis¬
placement and its deviation from the initial position is given by the func¬
tion 0(t), then the boundary conditions read

ux(l, t)——h[u(l, t) — 6(t)] with h=—. (2-1.55)

k
Note that in the case of a rigid constraint (large a), even when small
displacements cause large tensions, the last boundary condition is transformed
into the first for (i(t) = ѲЦ). For a weak constraint (small a), in which large
displacements produce only small tensions, the condition for a free-end point
occurs in place of the last noted boundary condition.
In the following, we shall speak of three principal types of boundary
conditions:
1. The first type 2<(0, t) — /jt(t), a prescribed motion of the end point x = 0;
2. The second type ux(0, t) = v{t), a prescribed force;
3. The third type ux{0, t) = h[u{0, t) — ѲЩ], an elastic constraint.
In an analogous way we formulate the boundary conditions for the second
end point x = l. If the functions on the right sides— цЩ, v{t), or 0(t) —equal
zero, then one speaks of homogeneous boundary conditions.
Combinations of these different types of boundary conditions result in
six types of simple boundary-value problems. Complicated boundary condi¬
tions occur, for example, for an elastic constraint which does not satisfy
Hooke’s law—that is, if the tension at the end point is a non-linear function
28 HYPERBOLIC DIFFERENTIAL EQUATIONS

of the displacement u(l, t), namely,

ux(l, t) = —F[u(l, t)]. (2-1.56)

k
This boundary condition is, in contrast with the law stated above, non-linear.
In addition, the relations between the displacements and the tensions are
possible at different points of the system; e.g., the boundary conditions for
problems of the vibrations of a ring, where x = 0 and x = / designate one
and the same point, read:

u{l, t) = u(0, t) , ux(0, t) — ux(l, t) (2-1.57)

and amount to the requirement that и and ux are continuous. In the boundary
conditions the derivative with respect to t can also occur. If the end of a spring
undergoes a resistance from the outside, which is proportional to its velocity
(perhaps by a disc fastened to the end of the spring, and for which the plane
of the disc is perpendicular to the spring axis), the boundary conditions read:

kux(l, t) = — aut(l, t). (2-1.58)

If a mass m is suspended at the end of the spring x — l, then the condition

mutt(l, t) = —kux(l, t) + mg (2-1.59)

must be fulfilled at x = l.
In the following discussion we shall limit ourselves to the consideration
of the three simplest types of boundary conditions, directing attention to an
example of boundary conditions of the first type, and indicate only incidentally
the peculiarities which occur in connection with the second and third types.
To formulate the first boundary-value problem for Eq. (2-1.52), let us seek
a function u(x, t) which satisfies the equation

Uu — a2uxx for 0 < jc < 1, t > 0

as well as the boundary and initial conditions

и (0, t) = g^t)
и (/, t) = fJ2(t)
(2-1.60)
и (x, 0) = (p(x)
ut(x,0) - </>(x)

We speak of the second or third boundary-value problem if boundary

conditions of the second or third type are imposed at both ends. If the
boundary conditions for x = 0 and x — l are of different types, then we mean
mixed boundary-value problems without classifying them more precisely.
We now turn to the limiting cases of these problems. The influence of
the boundary conditions at a point M0 which is sufficiently far from the
boundary, first enters the expression at a sufficiently large interval of time.
Of interest to us is the behavior during a small time interval in which
the influence of the boundary conditions is still negligible; then instead of
treating the complete problem, we can consider the problem with initial con-
2-1. SIMPLE PROBLEMS AND BOUNDARY-VALUE PROBLEMS 29

ditions for an unbounded region. Thus we seek a solution of the equation

Uu = cfuxx + f{x, t) for — oo < * < oo , t > 0,

with the initial displacement

[;u(x, /)],.„ = <p(x) for — oo < x < oo , (2-1.61)

and the initial velocity distribution

[ut{x, /)](=о = фіх) for — oo < x < oo .

This problem of the infinite string is ordinarily known as Cauchy’s problem.

If, however, we wish to study the behavior in the neighborhood of the
boundary, and if the influence of the boundary condition of one boundary on
the other in the course of a prescribed time interval remains insignificant,
we are led to a problem for a semi-infinite line 0 5S x < oo, where apart from
the differential equation itself we also have the conditions

и (0, t) = fi(t) , t ^0
и ix,0) = <pix) , о ^ x <QO (2-1.62)
■Utix, 0) = фіх) .
The character of the process for a time interval which is sufficiently
long from the initial time t = 0, is completely defined by the boundary values
since the influence of the initial conditions, because of the friction which
occurs in every real system, vanishes with increasing f.7 Such cases occur
mostly when the system under consideration is acted on by a periodic bound¬
ary influence which persists for an indefinitely long time. The formulation
of such problems without initial conditions is as follows:
Find a solution of the considered equation for 0 й x^l and t > — oo with
the boundary conditions

w(0, t) (i\it) (2-163)

uil, t) = fl2it) .

This is analogous to the problem without initial conditions for the semi¬
infinite line.
Besides the fundamental boundary-value problems, we shall also consider
in what follows the limiting cases:
1. Problems for an infinite region when one or both boundaries lie at infinity.
2. Problems without initial conditions when the solutions considered are
defined in the course of an infinitely large time interval.

8. Reduction of the general problem

For the solution of complicated problems one endeavors to trace them

back to the solution of simpler problems. To this end we shall represent

7 The vibration equation, taking into consideration the influence of friction which is
proportional to the velocity, has the form utt = a2uxx — aut, where a > 0. For details
of the above problem without initial conditions for a = 0, see § 3-7.
30 HYPERBOLIC DIFFERENTIAL EQUATIONS

the solution of the general boundary-value problem as a superposition of the

solution of specific boundary-value problems.
The functions Ui(x,t), where i= 1,2,•••,«, are required to satisfy the
equations

3 - а2 du* + f\x, t) for 0 < x < l, t > 0 (2-1.64)

at ox
and the conditions

Ui(0, t) = fA(t)
Ui(l, t) = /4(0
Ui(x, 0) = <p'(x) (2-1.65)

^(x,0 ) = ф\х).
at

Obviously, the solutions can be superposed in such a way that the function

U(0)(X, 0 = 1 Ui{x, t) (2-1.66)

i— 1

satisfies the analogous equation with

fm(x, 0=1 fix, 0 (2-1.67)

i—1

and the conditions with right sides given by

i /4(0 , k = 1, 2
i= 1

<Pw(x) = І 9\x) (2-1.68)

i—1

Фт(х) = i ф\х).
i—1

We see that this superposition principle is valid not only for the aforemen¬
tioned problem, but also for every linear equation with linear auxiliary con¬
ditions. We shall use this property repeatedly in the following.
The solution of the general boundary-value problem

utt = a2uxx + f(x,t) , (0 < x < l, t > 0)

и (0, t) = fitf)
и (/, 0 = /г2(0 (2-1.69)
и (x, 0) = 9(x)
ut(x,0) = ф(х)

can be represented as a sum

u(x, 0 = Mi(*. 0 + u2(x, 0 + u3(x, 0 + ut(x, 0 , (2-1.70)

where uy, u2, u3, ut are solutions of the particular boundary-value problems
2-1. SIMPLE PROBLEMS AND BOUNDARY-VALUE PROBLEMS 31

2 о u>%
■32

= a * = 1,2,3 ,
dt2 dx2
(2-1.71)
d2ut
= a2^- + f{x, t) , 0 < x < l, t > 0 ,
dt2 ox

Mi(0, t) = 0 Ui(l, t) — 0 иi(x, 0) = cp(x) иit(x, 0) = <p{x)

иг(0, t) = Ці{t) и2(/, 0 = 0 u2(x, 0) = 0 u2t(x, 0) = 0
щ(0, t) = 0 щ(1, 0 = a*i(0 щ(х, 0) = 0 u»t{x, 0) = 0
ut(0, 0 = 0 м4(/, 0 = 0 ut{x, 0) = 0 uu(x, 0) = 0 .

We shall limit ourselves for this formal reduction to the characterization

of those special boundary-value problems which are important intermediate
ones for the solution of the general problem. An analogous reduction can
also be obtained for the limiting cases of the general problem.

9. Formulation of boundary-value problems for several variables

We have considered above only the formulation of boundary-value problems

for the case of one independent geometric variable я and the time t. If the
number n of the geometric variables is larger than one (for example, n — 3),
then the first boundary-value problem reads:
We seek a function u(M,t) = u(x,y,z,t) which is defined in the interior
of a prescribed region T with boundary I; in the interior of T the function
must satisfy the equation

utt = a2Au + f{x, y, z, t), M(x,y,z)eT, t> 0 (2-1.72)

and on the boundary 2 must satisfy the boundary condition

u\i —/л(х, y, z, t) , M(x,y,z)£Z, t> 0; (2-1.73)

it also must satisfy the initial conditions

и (x, y, z, 0) = <p(x, y, z) , M(x,y,z)eT, (2-1.74)

ut(x, y, z, 0) = <p(x, y, z)
where u{x, y, z, t) is a function defined on 2.
The reduction of the general boundary-value problem to a series of simpler
problems is analogous to that in §2-8. It is to be noted that here also the
limiting case for an unbounded region, semi-infinite space, etc., can be con¬
sidered.

10. Uniqueness theorem

In the solution of boundary-value problems both of the following ques¬

tions arise.
1. Are auxiliary conditions sufficient for the determination of a unique solu¬
tion?
2. Will the problem be overdetermined by the auxiliary conditions, i.e., are
these conditions incompatible?
32 HYPERBOLIC DIFFERENTIAL EQUATIONS

The first question is answered by the uniqueness theorem, the second by

the existence theorem. The proof for the existence of solutions depends
strictly on the method that one uses for its determination. In the following
we shall prove the uniqueness theorem.
The differential equation

p^r = 0 + F(x, t), 0<x<l,t>0 (2-1.75)

possesses only one solution which satisfies the initial and boundary condi¬
tions

и (x, 0) = <p (x)

ut(x, 0) = <p(x)
(2-1.76)
и (0, t) = nAt)
U (l, t) — fi2{t)

Here it is assumed that the function u{x, t) and its first and second deri¬
vatives are continuous in the interval 0 x ^ l for t ^ 0, and p(x) > 0, k{x) > 0
are continuous functions.
Suppose there are two solutions

Ui(x, t) and u2{x, t)

of the problem under consideration. Their difference

v(x, t) = Ui(x, t) — u2(x, t)

obviously then satisfies the homogeneous equation

a2 v_ _ д Л dv \
_
(2-1.77)
dt2 dx \ dx )

and the homogeneous auxiliary conditions

v {x, 0) = 0
vt(x, 0) = 0
(2-1.78)
v (0, 0 = 0
v(l, 0 = 0.

Then we shall prove that the function v(x, t) defined by the conditions (2-1.78)
is identically zero.
To this end, we consider the function

E(t) — {k(vx)2 + p(vt)2)dx (2-1.79)

and prove that it is independent of t. Physically the function E(t) represents

the total energy of the string at time t. We differentiate E(t) with respect
to t\ since the second derivatives are continuous we can differentiate under
the integral sign. This gives
2-1. SIMPLE PROBLEMS AND BOUNDARY-VALUE PROBLEMS 33

dE(t)
(kvxvxt + pvtvtt)dx.
dt

An integration by parts of the first term of the sum on the right side gives

kvxVxtdx = [kvxvdh - 1 vt(kvx)xdx (2-1.80)

On the basis of the boundary conditions the expression in brackets vanishes;

vt(0, 0 = 0 follows from v(0, t) = 0, and similarly for x = l. Thus

dE(t)
vt{kvx)x vt[pvtt — (kvx)x\dx = 0 ;
dt

this means E(t) = const. By taking into consideration the initial conditions
we find further that

E(t) = const. = E(0) - ~ [k(Vx)z + p(vt)2]t^0dx = 0 (2-1.81)

LJ 0

because

v(x, 0) = 0 and vt(x, 0) = 0 .

Since by hypothesis p(x) and k(x) are always positive, it follows from (2-1.79)
and (2-1.81) that

vx(x, t) == 0 and vt(x, t) = 0

and hence the identity

v{x, t) = const. = C0. (2-1.82)

From the corresponding initial conditions we then find

v(x, 0) = C0 = 0 ,
and therefore we prove that

v(x, t)= 0 . (2-1.83)

If, therefore, ity{x, t) and u2(x, t) are two functions which satisfy all the con¬
ditions of our theorem, then Uy(x, t) = u2(x, t).
For the second boundary-valife problem the function v = иУ — иг satisfies
the boundary conditions

vx(0, /) = 0, Vxd, 0 = 0, (2-1.84)

so that the brackets in (2-1.80) vanish again. The rest of the proof remains
unchanged.
With regard to the third boundary-value problem, the proof of the unique¬
ness theorem requires several changes. If we again consider two solutions
иi and и2, then for their difference, v(x, t) = uy — u2, the differential Eq.
(2-1.77) results, while the boundary conditions read:

vx(0, t) - hM0, 0 = 0, hy ^ 0
(2-1.85)
Vx{l, 0 + h2vd, t) = 0 , ht ^ 0 .
34 HYPERBOLIC DIFFERENTIAL EQUATIONS

Next we shall write the expression in the brackets in (2-1.80)

[kvxvt\о = — -гг 4r[h2v2(l, t) + hiV2(0, /)].

Z ot

If we integrate dEjdt from 0 to t, we find

t ri
E(t) - E(0) = Vt[pvtt — (kvx)x]dxdt

- f {h2[v\l, t) - v\l, 0)] + h\v\0, t) - v\0, 0)]}

from which, with respect to the equation and the initial conditions, it follows
that

E(t) = —\-[кгѵ\1, t) + h,v\0, /)] ^ 0 . (2-1.86)

Li

However, since the integrand is non-negative, then E(t) 2: 0, from which

necessarily follow
E{t) = 0 (2-1.87)
and
v(x,t) = 0. (2-1.88)

Thus the uniqueness theorem is proved also for the third boundary-value
problem.
The method of proof used here, which was based on the use of the expres¬
sion for the total energy, is often used for the proof of uniqueness theorems
in different branches of mathematical physics (for example, in electrodynamics,
elasticity theory, and hydrodynamics).
The proof of existence for these and other boundary-value problems will
be discussed later when we consider the corresponding questions.

Problems

1. Prove that the differential equation for small torsional vibrations of a rod
has the form

Ѳн = а2Ѳхх,

where Ѳ is the angle of torsion of the cross section of the rod at the abscissa
x, G is the modulus of torsion, / is the polar moment of inertia of the cross
section, and k is the moment of inertia per unit of length of the rod. Give
a physical interpretation of the first, second, and third boundary conditions
for this problem.
2. Let an absolutely flexible homogeneous cable be fastened at one end.
Under the influence of its weight, it is aligned then with the vertical axis.
Find the differential equation for small vibrations of the cable.
Solution'.
d2u 2 3 V.du~\ ... 2
IF = a-s7Vl ~ х)~эІ \ w‘tha=!''
2-1. SIMPLE PROBLEMS AND BOUNDARY-VALUE PROBLEMS 35

where u(x, t) is the displacement of the points, l is the length of the cable,
and g is the acceleration due to gravity.
3. A heavy homogeneous cable of length /, which is fastened at its upper
end (x = 0) of a vertical axis, rotates about this axis with a uniform angular
velocity w. Derive the differential equation for small vibrations in the neigh¬
borhood of the vertical state of equilibrium.
Solution:

d2u 2 d V. du 1 . 2 ... 2
= a — (/ - x,)— + ш u with a = g.

4. Find the equation of a transverse vibrating string which exists in a medium

whose resistence is proportional to the velocity.
Solution'.

Vtt = a2vxx — h2vt.

5. Formulate the boundary conditions for the differential equation of a longi¬
tudinal vibrating elastic spring where the upper end of the spring is rigidly
fixed and, in contrast, a weight P hangs at the lower end when
(a) The rod in the equilibrium state under the action of an immovable
weight P, which is fixed at its lower end, exists in a state of elongation
(static deformation).
(b) The rod in the state of equilibrium is unstretched. This arises, for
example, when at the initial time under the weight, the support on which
the weight was previously resting, has moved; at first, the weight begins
to stretch the rod.
6. Determine the differential equation and the auxiliary conditions for the
torsional vibrations of a rod, at both ends of which discs are fastened.
Solution: For x = 0 and x = l the boundary conditions

в«(0, t) = а\Ѳх{0, t) , Ѳн{1, t) =— аІѲх(1, t)

must be satisfied.
7. At an arbitrary point x = x0 of a string (0 й x ^ l) let there be placed a
weight of mass M. Find the conditions determining the state of the vibration.
8. A weight of mass M is at the end x = l of an elastic rod which is elastically
fixed at the point x = 0. Determine the differential equation and the condi¬
tions for the longitudinal vibrations of the rod under the assumption that it
is acted on by an external force. Hence there are two cases to be considered:
(a) The force is distributed along the rod with the density f(x, t).
(b) The force is concentrated at a point x = x0 and equals F0(t).
9. Consider small vibrations of an ideal gas in a cylindrical tube. Derive
first the fundamental differential equations of hydrodynamics and, next, under
the assumption that the process proceeds adiabatically, [equations] for (a) the
density p, (b) the pressure p, (c) the velocity potential у of the gas particles,
(d) the velocity v, and (e) the displacement of particles u. In addition, con¬
struct an example which realizes the boundary conditions of the first, second,
and third types for these differential equations.
36 HYPERBOLIC DIFFERENTIAL EQUATIONS

10. What similarities exist between the phenomena of mechanical, accoustical,

and electrical vibrations?
11. Construct an example of the boundary conditions of the first, second,
and third types for telegraphic equations.
12. Consider the longitudinal vibrations in an inhomogeneous rod (k =
for x < x0, k —for x > x0), and derive the conditions which must be satisfied
at the boundary between segments of inhomogeneous rod (for x = x0).
13. Give a physical interpretation of the boundary conditions
aux(0, t) 4 /Зм((0, t) — 0 .
-

14. Give an example of a mechanical model which can be described by the

equation

uH = a2uxx + but + cu .

2-2. WAVE-PROPAGATION METHOD

1. The D’Alembert method

The subject of the following investigations is methods for the construc¬
tion of boundary-value problems for hyperbolic differential equations. We
begin by considering the problem of an infinite string with the initial condi¬
tions
Пн a uxx — 0 , (2-2.1)

u(x, 0) = <p(x),
(2-2.2)
ut(x, 0) = ф{х).

First we transform this equation to the canonical form in which the mixed
derivative is obtained. The characteristic equation

dx2 — a2dt2 = 0

reduces to two equations

dx — adt = 0 , dx + adt — 0 .

Their integrals are the straight lines

x — at = Ci , x + at = C2.
As in the previous chapter, we introduce the new variables
£ = x + at , 7) = x — at.
If we then calculate the derivatives
ux — U( -f- Utj , uxx = Mff T 2U(y) T Uryr),

Ut — a(U( — Ur]), Utt — — 2U(t) T ,

we see that the equation of a vibrating string can be transformed to the form
Me, = 0 . (2-2.3)
2-2. WAVE-PROPAGATION METHOD 37

Obviously for each solution of Eq. (2-2.3)

v) = f*(v)
is valid, where f*(rj) is a function of rj alone. By integration of this equation
with respect to rj for fixed £ we get

u($, rj) = j f*(v)dV + Ш = Ш + Ш - (2-2.4)

where depends only on £ and A only on rj. Conversely, any arbitrary dif¬
ferentiable functions fi and /2 which define the function w(£, rj) through (2-2.4)
represent solutions of Eq. (2-2.3). Since every solution of Eq. (2-2.3) for a
suitable choice of /, and /2 can be represented in the form (2-2.4), then (2-2.4)
yields the general solution of this equation. Consequently,

u(x, t) = fdx + at) + f2(x — at) (2-2.5)

is the general solution of the differential equation (2-2.1).

Now we assume that a solution of the problem under consideration exists.
This is then given by (2-2.5). The functions A and /2 are determined so that
the initial conditions
u(x, 0) = AM + f2(x) = (p(x) (2-2.6)
ut(x, 0) = af!(x) — afl(x) = ф(х) (2-2.7)

are fulfilled. By integration of the second equation we obtain

fiix) — Ш = — f </>(a)da + C ,
a J*o

where x0 and C are constants. From the equations

AM + AM = 4>M

AM - AM = —{ <p{a)da + C

we then find

AM = -^vM
(2-2.8)

AM = y^M

In this manner, A and /2 are defined by the given functions (p and ф,

where Eq. (2-2.8) must be satisfied for arbitrary values of the argument.8
By substitution of the arguments of A and A as in (2-2.5) we obtain
1 л

j
s rx+at rx—at

u(x, I) = £l£±SSl± I**.-“4 +. ф(а^а — фіа^а^

8 In formula (2-2.5), /i and /2 are not uniquely defined. That is, if we subtract
from f\ any fixed number ci and add this to /2, и remains unchanged. In 2-2.8, how¬
ever, the constant C is not determined by <p and ф. It can be omitted or replaced by
another without changing the value of w. When and /2 are added, the sum differs
by ± Cl2.
38 HYPERBOLIC DIFFERENTIAL EQUATIONS

or
I (•!+at
(p(x + at) + <p(x — at)
u{x, t) + —\ <f>(a)da . (2-2.9)
2 j x—at

Formula (2-2.9), the so-called D’Alembert formula, was derived under the
assumption that a solution of the given problem exists. This formula proves
the uniqueness of the solution. If there were to exist another solution of
the differential Eq. (2-2.1) with initial conditions (2-2.2) then it would have
to be a solution of the form (2-2.9) and hence would coincide with the first
solution. It can be shown that the function u(x,t), defined by (2-2.9) under the
assumption that <p be twice differentiable and that ф is only once differenti¬
able, satisfies both Eq. (2-2.1) and also the initial conditions (2-2.2), so that
the D’Alembert method besides proving uniqueness also proves the existence
of the solution of the given problem.

2. Physical interpretation
The function u(x, t) which appears as a solution of the wave equation
with initial conditions, is the sum of two functions

<p(x + at) +' <p(x — at)

ифх, t)
2
(2-2.10)
*x+at

u2{x, t) — ф(а)(1а.
2a J x-at
The first sum ифх, t) represents the path of propagation of the initial dis¬
placement without the initial velocity, ф(х) — 0; the second sum u2{x, t) con¬
tains the initial velocity (the initial impulse) for vanishing initial displace¬
ment. The function u{x, t) can be interpreted geometrically as a surface in
u, x, t space (Fig. 4a). The intersection of this surface with the plane

FIG. 4.
2-2. WAVE-PROPAGATION METHOD 39

t = t0 is analytically given by и = u(x, t0) and yields the profile of the string
at time t0. On the other hand, the intersection of the surface u(x,t) with
the plane x = x0 gives и = u(x0, t), which is the path of motion of the points
Xo-
The function u(x, t) given by и — f(x — at) is described in physics as a
propagating wave.* * * 9 The displaced profile defined by this function at different
times t can be easily illustrated in the following manner: We shall assume
that an observer moves parallel to the x axis with the velocity a (Figure 4b).
If the observer then is found at the initial time t = 0 at a position x = 0,
then up to time t he has moved along the path toward the right. If we
now introduce a new coordinate system through

x' = x — at and t' — t

which moves with the observer, then и — f(x — at) is defined in the new
coordinate system by
u{x', t') =/(*');
i.e., the observer during the entire time t sees one and the same profile
fix'), which coincides with the profile f(x) at the initial time t = 0.
Therefore f(x — at) represents a fixed profile f(x') moving toward the
right with the velocity a (propagating wave).
If we consider the x, t phase plane, then the function и = f(x — at) re¬
mains constant on the straight line

x — at = const.
The surface и = f(x — at) is also a cylindrical surface whose generators are
parallel to the straight lines x = at. Thus the form of the cylindrical surface
is determined by the profile of the initial displacement.
In the interval (xltx%) let the function f(x) now be different from zero
and outside of this interval equal to zero. The straight lines x — at = x2
and x — at — Xi represent the forward surface and the rear surface of the
propagating wave f(x — at). These lines divide the x, t plane into three
regions I, II, and III.
The regions I and III consist of the
points {x, t) which correspond at the time
t considered to the point x of the string
that lies behind the forward propagating
wave. On the other hand, the points (x, t)
of region II correspond to those points x
through which, at time t, the forward pro¬
pagating wave travels (Figure 5).
X
It is obvious that f(x + at) represents
a wave propagating toward the left at a
FIG. 5. velocity a. For this wave a similar inter-

9 In physics one uses the form и

40 HYPERBOLIC DIFFERENTIAL EQUATIONS

pretation can be given. The function ux(x, t) which describes the propagation
of the initial displacement <p(x) with vanishing initial velocity, ф(х) = 0, is
given by formula (2-2.10) as the sum of two waves propagating to the right
and left at a velocity a. Thus the initial form of both waves is characterized
by the function <p(x)/2, which is equal to one half of the original displacement.
As a first simple example we shall consider the propagation of an original
displacement which has the form of an equilateral triangle. The string
preserves this form if it lengthens in the middle of the interval (xi, x2) while
the points Xi and x2 remain fixed. Figure
6 shows the successive behavior of the
string after a time interval of amount

x2 — Xi
At =
8a
If one wishes to exhibit the behavior of the
string in the course of a sufficiently small
time interval, then one can, so to speak,
group together snapshots of the propaga¬
tion of the original length.
If we place the characteristics in the
x, t phase plane through the end points of
the intervals P{xГі,0), Q(xz, 0) (Fig. 7), the
plane is divided into six regions—regions
I and V at the time t considered in which
lengthening has yet to occur, region III in
which maximum lengthening has already
occurred, and regions II, IV, and VI in
FIG. 6. which lengthening has just occurred.
As a second example we shall investi-
gate the case in which there is no initial lengthening, but in which an
initial velocity exists. Let it be different from zero only in the interval
(xi, x2) and possess there the constant value ф0. Then we can write

u(x, 0) = <p(x) = 0
Ut(x, 0) = (p(x) = фо for Xi S x й x2,
= 0 for X > X2 or X < Xi .

Formula (2-2.9) then takes the form

u(x, t) = u2(x, t) = -J—f ф(а)с1а = [W{x + at) — W{x — at)} .

£Cl Jx-at

The function u(x, t) is given also in this case as the sum of two waves.
Thus W{x) is the integral of ф(а) and represents the profile of a wave pro¬
pagating to the left:

V{x) - j ф{а)сіа .
0
2-2. WAVE-PROPAGATION METHOD 41

We choose x0 = Xi appropriately. The auxiliary function ¥(x) so obtained is

represented in Figure 8. Hence,

¥(x) = 0 for X 5; Xi ,

= T^-(X - Хі)ф0 for X,. й X ^ X2 ,

Хі)фо for x^ x2.

In order to find u(x, t) we must form the

difference of the left and right waves defined by
¥(x). Figure 9 shows the position of these waves
and their difference after a time interval At =
(x2 — x^/Sa. For t > (x2 — Xi)/2a the profile of
the displacement is given by a trapezoid which
expands uniformly with the time. If <p(x) is not
a constant, the problem under consideration re¬
mains essentially unchanged. By using the phase
plane (Figure 7) we can easily show in which
region the lengthening has not yet occurred
(I and V), where the lengthening has already at¬
tained its maximum value (III), and where this
is not yet the case (II, IV, and VI).
Consequently, the lengthening of the string
occurs not only at one but at many places; thus
we obtain the form of a propagating wave in
which one adds the lengthenings corresponding
to the influence of single moving points. The
two examples consequently illustrate the pro¬
pagation of waves also in the general case.
As a third example we shall consider the
vibration of a string under the influence of a
concentrated acting impulse. By striking the
string at the point (x, x + Ax) with any object
(for example, with a hammer), we produce an
FIG. 9. impulse / at this position which is equal to the
42 HYPERBOLIC DIFFERENTIAL EQUATIONS

change of the momentum of the object struck during the time of the stroke.
Let the change in velocity of the point in the interval Ax be equal to v,
where v is the initial velocity. Under the assumption that the initial velocity
v is constant in Ax, then we obtain the change of momentum,
pvAx = I,

where p is the linear density of the string. Consequently, we must solve

the wave equation with the initial velocity,

^int — v — —t— x, x + Ax ,
pAx
</>ext = 0 X, X + Ax ,

for the initial displacement.

The lengthening obtained by the action of the impulse can be described
by a trapezoid whose lower base equals (2at) + Ax and whose upper base
equals (2at) — Ax, for t > Ax/2a. Obviously, the quantity I/Ax = I0 can be
interpreted as the impulse density. As Ax—>0 the following results for the
form of the displacement: the lengthening is equal to zero everywhere out¬
side the interval (x — at, x + at), and inside it is equal to l/2a • 1/p. Loosely
speaking, one can say that the displacement is produced by the point im¬
pulse I.
We consider now the x, t phase plane
n (Figure 10) and place the two characteristics
through (x0, t0):

x — at = x0 — at0
x + at = x0 + at0 .
They determine two angles oa and a2,
the so-called upper and lower characteristic
angles at the point (x0, to) •
The action of a point impulse at the
fig. io. point (x0,t0) produces a lengthening which
in the interior of the above characteristic
angles equals 1/2a • 1/p and outside the interval equals zero.
Of interest to us now is the region in which the solution is uniquely
defined by the initial conditions when these are prescribed in a given interval
PQ of the lines t = 0.
Formula (2-2.9) shows that it suffices for the determination of the func¬
tion и at any point M(x, t) of the x, t phase plane (Figure 7) when the initial
conditions in the interval PQ are known. Thus, P, Q are the points of the
x axis with the coordinates x — at and x + at. The segments MP and MQ
of the characteristics passing through the point M and the segment PQ of
the x axis form a triangle MPQ called the characteristic triangle of the
point M.
If the initial conditions are not given on the entire line — oo < x < oo
but are given only in a fixed interval PQ, then these initial conditions define
2-2. WAVE-PROPAGATION METHOD 43

the solution uniquely within the characteristic triangle which has the interval
PQ as a base.

3. Stability of the solution

The solution of Eq. (2-2.1) is uniquely determined by the initial condi¬

tions (2-2.2). We shall prove that between this solution and the initial condi¬
tions exists a continuous dependence and, in fact, we have the theorem:
For each time interval 0 ^ t ^ t0 and for arbitrary e there exists a number
<5(e, t0) such that two solutions ut(x, t) and u2(x, t) of Eq. (2-2.1) differ from
each other by an amount less than e:

I Ui(x, t) — u2(x, t) I < s , 0 ^t0,

provided that the initial values

Ui(x, 0) = <pi(x) u2(x, 0) = <p2(x)

, and a
-Jt-(*, 0) = ффх) ~ЧГ(Х’ °) = &(*)
at at
differ from each other by an amount less than d:

I <Pi(x) — <p2(x) I < S , I <f>i(x) — 4>2{x) I < d. (2-2.11)

The proof of this theorem is surprisingly simple. The functions ифх, t) and
u2{x, t) are linked to the initial values by formula (2-2.9), so that

I щ(х, t) - u2(x, t) I ^ 1 + at) + at) I

I rx+at
I <Рі(х — at) — <p2(x—at) I
у-\ \фі(ос) — </)2(a)\da,
+ 2 + 2 a J x—at

whereas on the basis of the inequality (2-2.11), there follows

Ui(x, t) — u2(x, t) I ^ 2at

2 Za • =

Hence our assertion is proved if we take

l + ^o
Every physically defined process must be capable of description through
functions which depend continuously on those initial conditions determining
the process. If the solution of a boundary-value problem depends continuous¬
ly on the initial conditions, then one also says that the boundary-value problem
is well set or the solution is stable.
If this continuous dependence did not exist, there could be two essentially
different processes corresponding to practically the same set of initial con¬
ditions (whose difference lies within the limits of the accuracy of measure¬
ment); that is, the solution would not be stable. It cannot be asserted that
such processes are determined by the initial conditions (in a physical sense).
From the above theorem, it follows that the vibrations of a string are deter-
44 HYPERBOLIC DIFFERENTIAL EQUATIONS

mined not only mathematically but also physically by the initial conditions.
We shall now consider such a problem in which the solution is not
stable. The functions resulting as solutions of the Laplace equation

uxx + uyy — 0
are defined uniquely by its initial conditions10

u(x, 0) = <p{x) , uy(x, 0) = ф(х).

The functions

u{'\x, y) = 0 and um(x, y) =— sin Xx • cosh Xy

satisfy the Laplace equation. In uw(x, у), X plays the role of a parameter.
The initial conditions

u"\x, 0) = 0 , un\x, 0) = <p(x) = — sin Xx

и[у1){х, 0) = 0 , и\г\х, 0) = ф(х) = 0

differ arbitrarily little from each other for sufficiently large X. On the other
hand the solution uw(x, y) for a fixed value of у can become arbitrarily large.
Therefore, the problem with initial conditions for the Laplace equation is
not well set.
We note the following. Obviously, a function u(x, t) defined by
x+at

U(X, t) = 9ІХ + at) + *-— + -—■ \ <p(a)da

can only be a solution of Eq. (2-2.1) provided <p{x) is once and <p(x) is twice
differentiable. Hence, it follows that the functions represented in Figures

u.

x, xz хг

FIG. 11. FIG. 12.

11 and 12 cannot be solutions of Eq. (2-2.1) because they are not twice dif¬
ferentiable throughout. Beyond this, the assertion holds in that a solution of
the wave equation does not exist which satisfies the initial conditions (2-2.2)
when (p{x) and ф(х) do not possess the required derivatives. By repeating

10 These conditions define uniquely the solution of the Laplace equation mathemati¬
cally. The origin of the function uy(x, 0), is, of course, equivalent to the origin of the
function vx(x, 0), where v(x, y) is the harmonic function conjugate to u(x, y). Hence that
analytic function of which the function u(x, у) is the real part is defined uniquely up to
an arbitrary constant. (See §4-1, 4-5.)
2-2. WAVE-PROPAGATION METHOD 45

the reasonings which led us to formula (2-2.9) we can show that from the
existence of a solution of the wave equation its representation follows ac¬
cording to (2-2.9). If, however, cp and ф are not a sufficient number of times
differentiable, then (2-2.9) defines a function which does not satisfy Eq. (2-2.1),
i.e., no solution to the problem exists.
If, however, we change the initial conditions a little—replace them by
differentiable functions cp{x) and ф{х)—then these new initial conditions cor¬
respond to a solution of Eq. (2-2.1). Moreover, it is still to be noted that
according to the proof of the last theorem we have in fact proved the con¬
tinuous dependence of the functions (p and ф defined by formula (2-2.9)—
independent of whether these are or are not differentiable. If, therefore,
certain functions <p, ф do not correspond to a solution of the wave equation
which satisfies the conditions (2-2.2), then the functions defined by (2-2.9) are
boundary values of the solutions of the wave equation with somewhat smoother
initial conditions.

4. Semi-infinite line and the method of continuation

In the following we shall concern ourselves with the propagation of a

wave along the semi-infinite line x ^ 0. This problem plays an essential role
in the investigation of the reflection of a wave at one end.
Statement of the Problem: Find a solution of the wave equation

агихх = Uu for 0<x<oo, t > 0,

which satisfies the boundary conditions

u(0, t) = pit) or ux(0, t) = v(t)

and the initial conditions

u{x, 0) = <p{x) , ut(x, 0) = ф(х).

Our first concern is the homogeneous boundary conditions

u(0, t) — 0 or ux{0, t) = 0 ,
i.e., the propagation of the initial displacement of a string with a fixed end
point x = 0 (or a free end point).
For the solutions of the wave equation which are defined for the infinite
straight line, the following two lemmas are valid.
1. If the functions <p(x) and ф(х) occurring in the initial conditions with
respect to any point x0 are odd, the corresponding solution at this point
is equal to zero.
2. If the functions <p(x) and ф(х) occurring in the initial conditions are even
with respect to any point x0, then the derivative of the corresponding
solution with respect to * at this point equals zero.
Proof of the first lemma. We select x0 as the origin of coordinates, i.e.,
x0 = 0. The conditions on the function in question are odd; therefore we have

<p(x) =— <p{— x); ф{х) = —ф{—х).

46 HYPERBOLIC DIFFERENTIAL EQUATIONS

Consequently, for x — 0, the function u{x, t) defined by (2-2.9) equals

«(0, t) = ^й + ѴІ-а*!. + J_r‘ фШа = 0 ;

2 2a J —at

then the first summand vanishes, since <p(x) is odd, whereas the second, the
integral of an odd function within the limits shown, is symmetrical with
respect to the origin and likewise is equal to zero.
Proof of the second lemma. The conditions on the functions in question are
even; therefore we have

<p{x) = <p{— x) ; ф(х) — <j>(— x).

Now the derivative of an even function is odd, i.e.,

<p'(x) =— <p'(— x).

Therefore, it follows from (2-2.9) that

u,(0, t) = y,(gf) +/(~ at) + -±-[</>(at) - ф(- at)] = 0 ;

2 2a

then the first term of the sum vanishes since <p'(x) is odd and the second
vanishes, since ф(х) is even.11
With the help of these two lemmas the following problems can be solved:
Find a solution of Eq. (2-2.1) which satisfies the inidal conditions

и (x, 0) = <p(x)
(2-2.2)
ut(x, 0) = ф(х),
and the boundary condition

u(0, 0 = 0
(first boundary-value problem).
The functions Ф(х) and ¥(x) defined by the relations

<p(x) for x > 0

Ф(х) =
- <p{- x) for x < 0

W{x) = ^ for x > 0

— ф{— x) for x < 0
are the odd continuations of <p(x) and ф(х).
The function

u(x, t) = 0{x + at) + Ф(х ~ at)■ + 4-[+at W{a)da

2* bQ, J x—at

defined with their help is defined for all x and t > 0. According to the first

11 These two lemmas are consequences of the fact that for even (or odd) initial con¬
ditions the function u(x, t), given by the formula of D’Alembert, for t > 0 is likewise
even (or odd); we leave the proof of this to the reader. Geometrically, one sees im¬
mediately that an odd continuous function, as well as the derivative of an even dif¬
ferentiable function, vanishes at x = 0.
2-2. WAVE-PROPAGATION METHOD 47

lemma, there results

u(0, t) = 0 .
Moreover, u(x, t) for t — 0 satisfies the initial conditions

и (x, 0) = Ф{х) = <p (x)

x > 0.
ut(x, 0) = ¥(x) = ф(х) ,
Consequently, the function u(x, t), which we should consider only for x > 0,
/ > 0, satisfies all the requirements of the given problem.
With regard to our original functions, we have
x + at
_ (p(x + at) + <p(x — at) + 1
u(x, t) <p(a)da for t <— , x > 0
2a ( a
x+at
(2-2.12)
_ <p(x + at) — (p(at — x) 1
cb{a)da for t >— , x > 0 .
2a r a
In the region t < x\a the influence of the boundary conditions does not enter
into the expression. For these values of t, the expression u(x, t) coincides
with the solution of (2-2.9) for the infinite straight line.
We proceed in a corresponding way when a free end exists at the point
x = 0,

«*(0, 0 = 0,
and, in fact, we take the even continuations of <p(x), and <p(x)

<p(x) for x > 0

Ф(х) =
<p(— x) for x < 0
ф{х) for x > 0
V(x) =
ф(— x) for x < 0 .

As a solution of the wave equation we then obtain

Ф(х + at) + Ф{х — at) 1 rx+at

u(x, t) = —1 ¥{a)da
J x-at
or

u(x, t) = 9{Х + at) + (P {x—+ fr^ for t<^

b “O' J x-at a
_ <p(x + at) + <p(at — x)
2
z' лх+at rat— x x

+ ф(а)<і(х + j •/’(a)daj for t>^

2a a

This solution in the region x ^ 0 satisfies the initial conditions (2-2.2)

and the boundary condition

ux(0, 0 —0.
In the following we shall frequently have occasion to use the above ap-
48 HYPERBOLIC DIFFERENTIAL EQUATIONS

plication of the method of continuation, even if the initial conditions are

defined only for a finite subregion. We shall reiterate the results obtained
in the two following rules:
1. For the solution of a problem of a semi-infinite line with the boundary
conditions u{0, t) = 0, the initial conditions are continued oddly along
the entire axis.
2. For the solution of a problem of a semi-infinite line with the boundary
condition ux(0, t) = 0, the initial conditions are continued evenly along the
entire axis.
We shall consider two examples. Let the initial conditions be given on
the semi-infinite line x 2: 0 bounded by x = 0 and different from 0 only in
the interval 0 < a < x < b. In this interval an initial displacement given by
the function <p(x) occurs, which is represented by an isosceles triangle, whereas
<p(x) — 0. We arrive at the solution of this problem if we continue the initial
conditions oddly along the entire straight line. Figure 13 shows the course
of the wave propagation. First the propagation proceeds as though both sides
of the straight line were unbounded. The initial displacement is distributed
on two waves which progress at a constant velocity with respect to the

FIG. 13.
2-2. WAVE-PROPAGATION METHOD 49

different sides. This continues as long as the half-wave propagating toward

the left has not reached the point x — 0 (Figure 13). When the half-wave
reaches the point x — 0, where the corresponding process is taking place, a
wave with opposite phase arises. Accordingly a reflection of two half-waves
occurs at the fixed ends. Figure 13 shows the reflection process in its in¬
dividual stages. The resulting profile of the string is shortened, the dis¬
placement vanishes, after which a displacement (with a negative phase) begins
again, and finally the reflected half-wave moves toward the right following
the half-wave which likewise is propagating toward the right. Consequently,
the phase of the wave due to reflection of the wave changes its sign at the
boundary point.
For the investigation of the second example we shall assume that no¬
where on the semi-infinite line x ig 0 bounded by x = 0 does an initial dis¬
placement occur, and further, that the initial velocity ф(х) is different from
0 only in an interval 0 < xt < x < x2. Hence ф(х) — const. For the solution
of this problem the initial conditions are continued oddly. Then the displace¬
ment is split into each of the intervals (x;, x2) and ( — x{, —x2) which resemble
the displacements represented in Figure 14. As seen from these figures, the

process proceeds for x > 0 as if it were initially on the infinite straight line.
Accordingly, the reflection occurs at the fixed end, and finally a wave moves,
whose profile in this case is an isosceles trapezoid with constant velocity
toward the right.
The investigation of the reflection at a free end proceeds analogously only
if the initial conditions are continued evenly, so that the reflection of the
50 HYPERBOLIC DIFFERENTIAL EQUATIONS

wave at the free end does not proceed with a changing phase but with the
same phase.
Finally, we shall consider problems with homogeneous boundary conditions

u(0, t) = /jt(t) = 0
or
ux{0, t) = v(t) = 0 .
In the general case of nonhomogeneous boundary conditions the solution
can be written as a sum, each of whose terms satisfies only one of the stated
conditions (either the boundary or the initial condition).
We turn now to the solutions of differential equations with homogeneous
initial conditions and prescribed boundary conditions. Let

й(х, 0) = 0 , ut(x, 0) = 0 , U(Q, t) = f*(t) , t > 0.

Obviously such a boundary condition produces a wave which moves away
from the string toward the right with the velocity a. This wave has the
analytical form

u(x, t) = f(x — at) .

The function / is defined by the boundary condition

й(0, t) = /(— at) = fi(t).

Then

and therefore

This function is defined, however, only for the region x — at -й 0 since fi(t)
is defined only for t ^ 0. In Figure 15 this region is shown as the shaded
part of the phase plane. Now in order to determine u(x, t) for all the values

FIG. 15.
2-2. WAVE-PROPAGATION METHOD 51

of the argument, we define fj.{t) also for negative values of t setting = 0

for t < 0. Then

u{x, t) = fJL^t ——^

is defined for all values of the argument and satisfies the homogeneous initial
conditions.
The sum of these and function (2-2.12) defined earlier represent the solu¬
tion of the first boundary-value problem for the homogeneous wave equation:

^ _ <p(x T at) + <p(x — at)

, 1 f*+0<
t) _---+¥)j (p(a)da for t <—
a

= filt
<p(x + at) — ip(at — x) 1 r+ttt
+ + <p(a)da for t >— .
2a a
(2-2.13)
Analogously one can construct the solution of the second boundary-value
problem. For the third boundary-value problem see page 61. We shall
limit ourselves here to the solution of the boundary-value problem for the
homogeneous wave equation. For the solution of the nonhomogeneous wave
equation see page 61.

5. Problems for a bounded interval

We shall take now a bounded interval (0, l) as a basis and begin our
investigation with the search for the solution of

Hit Cl tlxX t

which satisfies the boundary conditions

u(o, t) = ^(t) t > 0

U{1, 2
t) = fJt (t) ,

and the initial conditions

и {x, 0) = <p(x)
О й x ^ / .
ut(x, 0) = ф{х) ,
Moreover, we shall consider introducing the case of homogeneous bound¬
ary conditions
u(0, t) = u(l, t) = 0 .

For these we shall seek the solution by means of the method of continua¬
tion. Hence we construct the expression (Ansatz)

u(x, t)
Ф(х + at) -f Ф(х — at)
Sx+at

¥(a)da ,
2 x—at

where the functions Ф and ¥ are still to be defined appropriately. First Ф

and ¥ are determined only for the interval (0, /) by the boundary conditions

и (x, 0) = Ф(х) = (p(x)

ut(x, 0) = ¥(x) = ф(х) ,
52 HYPERBOLIC DIFFERENTIAL EQUATIONS

In order that now Ф{х) and ¥(x) satisfy the homogeneous boundary con¬
ditions, we require that Ф(х) and ¥(x) be odd with respect to the point x = 0,
x = /, i.e.,
Ф(х) = - Ф(- x) , Ф(х) = - Ф{21 - x)
¥(x)=-¥(-x), ¥{x) = - V(2l - x) .

From these equations it follows that

Ф(х') = Ф(х' + 21), {x1 =— x).

A corresponding relation holds for ¥(x), i.e., Ф and ¥ are periodic functions
with a period of 21.
Obviously the continuations of Ф(х) and ¥(x) are such that these func¬
tions will be odd and, moreover, periodic with respect to the origin of co¬
ordinates, and are defined on the entire straight line — oo < * < oo. By in¬
troducing these continuations into formula (2-2.9) we obtain the solution of
the problem.
Figure 16 combines the л:, t phase plane and the x, и plane, in which the
initial displacement and its continuations are included. In the phase plane
the shaded strips represent the regions in which the displacement is different
from zero (see Figure 7). The plus or minus signs in the strips indicate the

t, /-21
' a

STT І yk

jp 4-X
тЖр
“41

/l\
■y^v4 1

ІХ X"FXi4 1 N
+ІХ7=Х
1—
i
i
1 1
1
x -01 \X- ti i ! 1 1
i 1 1 1
1 1 1
i 1 Ui , 1 I ! 1
i I i 1 1
1 ' - i i 1 ' ! 1 1
i 1 1
i i i 1 1
1 1 i i 1 i i i
i i 1 t
1 t i i i i i 1 1
i 1
1 i '
i 1 ! 1 ' 1 1
i 1 1 1
1 1 1
i i i 1 1
i
1
1
1
1
1
1 1 1 1
I 1
! !
1 ■ 1 1i >
У'Х ! i 1/NJ

x=0 x=l ^*=21

FIG. 16.

signs (the phase) of the displacement (in the form of an isosceles triangle).
With the aid of this figure one can easily illustrate the profile of the string
at any arbitrary time t. Thus one can recognize at time t = 2l/a a displace¬
ment which coincides with the original displacement. The function u(x, t)
is therefore a periodic function of t with period T = 2l/a.
We shall now consider the propagation of the boundary effects. For this
purpose we shall seek the solution of the equation
_ 2
Utt ^ Mxx

with the homogeneous initial conditions

2-2. WAVE-PROPAGATION METHOD 53

u{x, 0) = <p(x) = 0 , ut(x, 0) = ф(х) = 0

and the boundary conditions

u(0, t) = fi{t) t > 0

u{l, t) = 0 .
From the results of Section 4 it was shown that for t < l/a the function

u{x, t) = /l(t - —^ with &(t) = 1 > ^

is a solution. This function, however, does not satisfy the boundary condition

u(l, t) = 0 for t >— .

The reflected wave which propagates to the left and at x = / has a displace¬
ment of magnitude Д(і — l/a) is represented analytically by the equation

_l_ l — x I . 21 x
W = H[t-+ —
a a a a
The difference between the two waves, i.e.,

is then obviously a solution of the equation for t < 2l/a.

By repeating this process one obtains a solution in the form of the series

2 nl ” _/ 2nl
u(x, t) = 1 v(t - 1M + — (2-2.14)
n=0 a
This contains only a finite number of terms distinct from zero since the
argument with each new reflection about 2l/a is decreased whereas ДЦ) = 0
for / < 0. That the boundary conditions are satisfied we prove directly when
we set x = 0 in (2-2.14). The summand for n = 0 of the first sum is then
equal to whereas the remaining terms of the first and second sums are
cancelled pairwise for equal n values. Thus u(0, t) = цЦ).
If we now replace n by n — 1 and vary accordingly the summation limits,
then the first sum reads

2 nl 21 — x
I
n= a a

If we now set x — l, it can be seen directly that the summands of the first
and second sums mutually cancel each other.12
Formula (2-2.14) has a simple physical significance. First

12 The initial conditions can likewise be proven directly, since the arguments of all
functions for t = 0 are negative.
54 HYPERBOLIC DIFFERENTIAL EQUATIONS

represents a wave caused by the effect of the boundary at x — 0, independent

of the effect at the point x — /, as though it were an infinitely long string
(0 < x < oo). The remaining summands represent the successive reflections
at the point x = l (the second sum) and at the point x = 0 (the first sum).
Correspondingly,

(2 n + 1)/ x_ t _ (2n + 1)/ x_

u{x, t) = -in
n—0 a a n=1 a a

is the solution of the homogeneous equation with homogeneous initial con¬

ditions u(x, 0) = 0, ut(x, 0) = 0 and the boundary conditions u(0, t) = 0, u(l, t)
= Here we shall not go further into the uniqueness proof and the proof
of the continuous dependence of the solution on the initial and boundary
conditions.

6. Wave dispersion

We have seen that the equation

litt - & Мхх

of propagating waves has solutions of arbitrary form. On the basis of the
class of partial differential equations, we ask which wave solutions are of
arbitrary form. We shall limit ourselves, therefore, to the consideration of
linear differential equations of the second order with constant coefficients,

йцИи -4- 2al2uxt T a.22Utt biUx + b2Ut + cu = 0 . (2-2.15)

Our problem thus consists of constructing relationships between the coefficients

which guarantee that the differential equation will be solved by functions of
the form

u(x, t) = f(x — at), (2-2.16)

where / represents an arbitrary function and a is a constant.

By inserting (2-2.16) into (2-2.15) we obtain the linear differential equation

f"(x — at)[au — 2al2a + a22a2] + f'(x — at)[bl — b2a] + cf(x — at) — 0 ,

that the wave profile must satisfy. For arbitrary / it is obviously solvable
when all the coefficients are equal to zero:

ал 2^12^ T a22a — 0
bt — b2a = 0 (2-2.17)
c=0 .

If the differential equation for waves has solutions of arbitrary form then
we speak of a lack of wave dispersion. For this, it is necessary and sufficient
that the conditions (2-2.17) be satisfied.
From the first relationship follows the wave velocity:

a — fli2 dr a12 2

a22
2-2. WAVE-PROPAGATION METHOD 55

Two velocities of wave propagation exist for hyperbolic differential

equations (a]2 — ana22 > 0); the requirement for the fulfillment of all three
relations for both values of a gives

bt = b2 = c = 0.

Consequently, a solution in the form of a propagating wave with two possible

velocities exists only for an equation of the form

#n#xx + 2#i2^x( “b u22Wtt —- 0 . (2-2.18)

If агг Ф 0 then (2-2.18) represents a wave equation in a moving coordinate

system; that is, if we set

$ = x - yt , у=t,
we obtain the equation

(#11 2«12 7" T #22 T )#f£ T (2#i2 2,(Z22y)U{ri “b #22 — 0 ,

which for у = #i2/#22 coincides with the equation of the vibrating string.
In this case we have
2
Ut) tj — CL

with

q2 = #12—~ #11 #22 >

#22

For elliptic differential equations (a?2 — #n#22 < 0), waves with real velo¬
cities as solutions are not possible. For parabolic differential equations
(#*2 — #n #22 = 0), solutions in the form of waves with real velocities likewise
are rejected. There is an exception in the case of Eq. (2-2.15), which de¬
generates into an ordinary differential equation.
By taking into consideration that al2 = l/#ii • l/#22, # = #іг/#22 = l/#n/#22,

and c = 0, as well as the relation a = bjb2 — V/anla22, we obtain

b2_
byux + b2ut (l/ #11 и* +1/ a22 #i) •
Л/ #22
The resulting equation can be written in the form

и = 0

and is reduced to the ordinary differential equation

d2u b2
+ b^- = 0 , b = -^=
de #? л/ #22

when we introduce the new variables

x — Va, t ~Va22^ + rj.

There also exists the relation
56 HYPERBOLIC DIFFERENTIAL EQUATIONS

d_ /— Э /— d
— V an~z + V a22——
ds dx at
Consequently, in this case every arbitrary function of the variables

Ѵогг _ . _ _£
V = t
T/ou a
is a solution of the equation. Solutions in the form of propagating waves
are thus possible for the simplest wave equations in a moving or nonmoving
coordinate system.
In physics the concept of wave dispersion is usually introduced some¬
what differently.
Thus, consider a harmonic wave of the form

u(x, t) = eUa,t~kx) , (*)

where w is the frequency, k = 2к/к is the wave number, and к is the wave
length.
The velocity with which the phase of the wave moves in space,

a = at — kx ,
is called the phase velocity and is obviously equal to

One speaks of wave dispersion if the phase velocity of the harmonic wave
is dependent on the frequency.
An impulse or a signal of an arbitrary form can be represented by a
superposition of harmonic waves of the form (*), that is, by a Fourier in¬
tegral. If the phase velocity depends on the frequency, the harmonic signals
are displaced relative to each other so that a distorted signal appears. In
this case the wave dispersion takes place in the sense of the definition on
page 54.
If the solution of the equation under consideration, which yields a wave
of arbitrary form, can be inverted, the phase velocity can be determined from
the first equation of (2-2.17) and thus does not depend on the frequency.
The concept of wave dispersion in the sense of our definition therefore
coincides with the property that the phase velocity depends on the frequency.
We shall now determine the class of Eq. (2-2.15) which permit solutions
in the form of damped waves

u(x, t) = fi{t) f{x — at),

where fi(t) is a function of t.
We shall substitute this expression into Eq. (2.2.15) to obtain

f"li(t)(an — 2 al2a + a22a2) + f'[{bx — Ь2а)ц + 2 (an — a22a)^]

+ + Ь2 fl' + «22 p") = 0 .
Since / is arbitrary, the coefficients of f", /', and / must be equal to zero.
2-2. WAVE-PROPAGATION METHOD 57

The function /u(t) thus satisfies an ordinary differential equation with con¬
stant coefficients and has the form
—kt
(t — e
If we set the coefficients of the last equation equal to zero, then we obtain
for the determination of a and k the relations

йи — 2 ana + a22a2 = 0
(bx — b2a) — 2k(an — a22a) = 0 (2-2.19)
a22k2 — b2k + c = 0 .
By elimination of a and k from (2-2.19) we obtain a condition for the compat¬
ibility of these three relations. The first equation shows that only the
hyperbolic differential equation allows damped waves as solutions. The
damping coefficient k is obtained from the second relation. Accordingly if
we insert k into the third equation, we arrive at the following relation be¬
tween the coefficients:

A(an — ana22)c + (anbl — 2ax2b]b2 + a22b\) = 0 . (2-2.20)

If these are satisfied, solutions of the equation exist in the form of damped
waves.
Example: The “telegraphic” equation

uxx = CLiitt T (CR T LG)ut T GRu (2-2.21)

does not permit a propagating wave as a solution if G or R are different

from zero. We shall investigate whether it has solutions in the form of
damped waves. The velocity of the damped wave is given by the first equa¬
tion of (2-2.19), namely,

1 - a2CL = 0 .

From the second equation, we obtain the damping coefficient

, CR + LG
k= ZCL -■
The condition for the validity of Eq. (2-2.19) reads:

ACLGR - (CR + LG)2 =-(CR- LG)2 = 0

or
CR = LG.
If these are satisfied, the damped wave

u(x,t) = e~ktf(x - at) , = a = ^~^L’

where / is an arbitrary function, is a solution of Eq. (2-2.15).

That no damping of the wave occurs with propagation in a cable is of
special importance in telephonic propagation over great distances. If the signal
can be propagated undamped, then by a corresponding reinforcement an un-
58 HYPERBOLIC DIFFERENTIAL EQUATIONS

distorted reproduction of the acoustical effect can be obtained. The wave

dispersion which occurs encroaches on the purity of reception independent
of the quality of the telephonic apparatus. The dispersion phenomenon has
a corresponding significance for telegraphy over great distances.

7. The integral equation for waves

As the starting point for the derivation of the differential equation of

waves, Eq. (2-1.3) was used as the basis for the conservation of momentum.
In order to go from this integral equation to a differential equation we
assume the function u(x, t) to be twice differentiable. Every limitation on
the class of functions under consideration naturally requires us to reject those
functions which do not satisfy these requirements. Thus in going from the
integral to the differential equation we exclude from consideration all of those
functions which are not twice differentiable.
To prove that a theory in the class of piecewise continuous differentiable
functions can be developed, we start with the integral form of the wave
equation

ЯЩ - (£).>* - - (*£),>+\Хтіт (2-2-221

and cast it into the form

j (p-Jfdx + + j ^Fdxdt = 0 , (2-2.23)

where G is the region of the x, t plane bounded by the piecewise smooth

curve C. For a homogeneous medium this relation assumes the form

( (X^dx + a2~dt) + i ( fdxdt = 0 , / =— . (2-2.23')

Jo\ dt ox J ))0 p

If C is the contour of a rectangle whose sides are parallel to the coordinate

axes, then formulas (2-2.22) and (2-2.23) coincide. If C consists only of seg¬
ments parallel to the coordinate axes, then G can be written as the sum of
rectangles. If we sum the boundary integral which corresponds to individual
summands, we see that the summands belonging to the interior boundaries
mutually cancel each other, since the sense of integration in each case is
taken in the opposite sense. The remaining summands yield formula (2-2.23)
directly. Further, C now contains an arc C which is not parallel to the axes
on which the integrand has no point of discontinuity. Then in the plane
we construct a system of squares with parallel axes and consider the set G*
of squares of the net which have a point in common with G. Let C* be
the boundary of G*. In G*, formula (2-2.23) is applicable. Finally, if we
make an unlimited refinement of the net, then formula (2-2.23) holds exactly
for the boundary curve C.
If we apply formula (2-2.23) to G* then the first summand consists of
summands of the form
2-2. WAVE-PROPAGATION METHOD 59

\ Ф(х, t)dx or l Ф(х, t)dt,

where Ф(х, t) is a continuous function and C„ is the arc of the boundary C*,
which approximates C (Figure 17).
Let t = tn(x) be the equation of Cn and t = t{x) be the equation of C.
Obviously then tn{x) converges uniformly towards t(x) and we can write
ь
lim \ Ф[х, tn(x)]dx Ф[х, t{x)]dx ,
™—“ Jo

whereby passage to the limit is proven.13

If there exists an arc C on which functions to be integrated have dis¬
continuities, then formula (2-2.23) remains valid when we take for the value
of the integrand its limit value as we approach the curve C from the interior
of the region G. With this observation, the representation of (2-2.22) in the
form of (2-2.23) also has been completely proven for this case.

FIG. 17.

We shall consider now the following problems:

Find a piecewise smooth function u(x, t) defined on — oo < x < oo and which
satisfies the equation

j jj f(x,t)dxdt = 0 (2-2.23')

and the initial conditions

и (x, 0) = <p(x)
ut(x, 0) = ф{х).
Solution. Let <p(x) be piecewise smooth and ф(х) and f(x,t) be piecewise con¬
tinuous. C is an arbitrary piecewise smooth curve which lies in the region
/ ^ 0. We shall prove that this problem has a single solution which can
also be determined by the D’Alembert formula.
Let us assume that the function u{x, t) is a solution of our problem.
Then we shall consider the triangle ABM, whose base lies on the x axis,

13 Since dx — 0 at the vertical parts of the polygonal arc Cn> t — tn(x) represents
in this formula the equation of a horizontal part of the polygonal arc Cn.
60 HYPERBOLIC DIFFERENTIAL EQUATIONS

whose upper vertex is the point M(x, t), and whose remaining sides have been
formed from the corresponding segments of the characteristics x — at — const,
and x + at = const. (Figure 18). Within this triangle we apply formula
(2-2.23'). Since along AM the relation dx/dt = a is valid, it follows that

dt dx \dt dx J
Along MB, by contrast, dx/dt ——a is valid; we obtain, therefore,

du , . 2 du ( du ,, . du , \ ,
-—dx + a -—dt = —a -—dt + -^—dx — —adu .
dt dx \dt dx J
Consequently, the integrand along the characteristics is a total differential,
and by integration along BM and MA we immediately obtain

\M(^dx + a2^-dt] ——a\u{M) - u{B)\

\ dt
) b dx J

[A(—dx + a2^-dt) = a[u{A) - u{M)\.

)M\ dt dx )

Formula (2-2.23') thus reads

u(B) + u{A) B du , 1
u(M) = +
iri ——dx + ——
л dt 2a ABM
f dx dt

or
j rx+at (*x+a(t-t)
<p{x + at) + cp(x — at)
u{x, t) + Ф(№ + _ dv\ /(£, :№•
2a L-ai^
x-at J x—a U-'

(2-2.24)

Consequently, the existence of а solution to our problem proves it is uniquely

determined by the initial conditions. In the case of a homogeneous equation
(/= 0), Eq. (2-2.24) coincides with the D’Alembert formula.
As one easily confirms, every function of the form

u(x, t) = ft(x + at) + f2(x — at) + f dr[ /3(f, r)df ,

Jo Jx-aU-r)

where /i,/2 are piecewise smooth and /3 is piecewise continuous, satisfies Eq.
(2-2.22) and therefore also Eq. (2-2.23). The solutions of the problems con¬
sidered as examples in § 2-4 are piecewise
smooth functions and thus are accounted for
by the above theory.
We shall now turn to the first boundary-
value problem for the semi-infinite line and
seek a solution of Eq. (2-2.23) at a point
M(x, t) for t > x/a, since in the region t < x/a
(outside the characteristic x = at) the influ¬
ence of the boundary conditions still does not
enter into the phenomenon and the solution FIG. 19.
2-2. WAVE-PROPAGATION METHOD 61

there is defined by formula (2-2.24). First we apply formula (2-2.23) to the

quadrangle MAA'B. Here MA, MB, and A A' are segments of the charac¬
teristics (Figure 19). By integration along MA, MB, and AA' we obtain

2au{M) — 2au(A) + au(B) — au(A') + f ^—dx + f f f dx dt.

Ja' dt Jj maa'b

If we insert the coordinates of M, A, B, and A' there finally results

u{x + at, 0) — u(at — x, 0)

u(x, t) = и +
2
x + a( t—r)
+ /(£, Adz
I* — a (( — т) I

or

*<*, t) = „( t - f) + y(* + af) - ~ *> +

д:+а(і —r)

+ ” f(S,r)dS, t > — (2-2.25)

ljQ’ Jo J Ix—ait — r)| Cl

For /=0, this formula, as is easily seen, coincides with formula (2-2.13).
Similarly, one obtains the solution of the second boundary-value problem,
as well as the solution of problems for a bounded interval.
For the investigation of the first boundary-value problem we saw pre¬
scribing both the initial conditions

u(x, 0) = <p(x) , ut(x, 0) = <p(x)

and the boundary condition

u(0, t) = /j(t)

was sufficient for the complete determination of the solution. Hence it follows
that a relation must exist which connects the functions <p, ф, p, v with v(t) —
ux(0, t). If we differentiate (2-2.25) with respect to x and set x — 0, we get

v(t) = —{$&(<*/) ~.[p(t) - acp'(at)}} , (2-2.26)

a
where for simplicity we have set / = 0. With the relation (2-2.26), for ex¬
ample, the third boundary-value problem can be reduced to the first.

8. Distribution of the points of discontinuity along the charac¬

teristics

In the following, we will be concerned with the points of discontinuity

of the first kind in the derivatives of the solution of Eq. (2-2.23). We shall
prove that the points of discontinuity of a function u(x, t), which is a solution
of (2-2.23), can only lie along the characteristics

x — at = const. x + at = const.
62 HYPERBOLIC DIFFERENTIAL EQUATIONS

Let us assume that a differentiable curve

X = x(t)

exists, on which the derivative of the con¬

tinuous piecewise smooth function u(x, t)
possesses a point of discontinuity. For con¬
venience of calculation we shall assume x(t)
to be a monotone increasing function. Then
we apply formula (2-2.23') to the quadrangle
ABCD (Figure 20):

du , . 2 du du , . 2 du ,,
——ax + a -—at + —ax + a ——at = 0
at ox DO + OB dt dx

and also along the curvilinear triangle Л = BAD and d2 = BDC:

du , . 2 du j, du , . 2 du \ j, л
-—ax + a ——dt
BA+AD dt dx dt ox /1

du du t . 2 du \ i. л
dx + a —л: + a —— ) dt = 0 ,
DO+OB dt dt dx

where the parentheses ( )t and ( )2 will indicate that the corresponding limit
values are taken from the interior of Л or zf2 respectively. By subtraction
of the first equation from the sum of the last two equations we obtain

f l(±Lx' + a‘^L) _ (ІѴ + а'Щ 1

jdsIV dt dt /1 \ dt dx J2/

or, since DB was chosen arbitrarily,

(2-2.27)

The square brackets indicate here, as is usual, the magnitude of the jump
of the function, i.e.,

[/]=/.-/1.
We now form the derivative of the function u(x, t) with respect to t

du du
—u(x(t), t) = x' + £ — 1,2
dt dx dt

Therefore, we can choose both the value of the derivatives and the limit
values corresponding to and J2 (taken from the interior). The difference
of the right sides for i = 1 and i = 2 is

= 0.
2-2. WAVE-PROPAGATION METHOD 63

If we equate this equation with (2-2.27) and assume at least one of the jumps
[du/dt], [du/dx] to be different from zero, we see that both equations are simul¬
taneously satisfied if the determinant of the system vanishes:

r'
XU n 2 / /\2 2 a
, = (x ) — a = 0

or
x — +at + const.

Consequently the points of discontinuity of the derivatives of a function

u(x, t) which is a solution of the wave equation lie along the characteristics.

Problems

1. Construct a profile of a string at different times for the following cases:

(a) The unbounded string
(1) The initial velocity is equal to zero [ф{х)
— 0], while the initial profile of the string is
given as in Figure 21.
(2) The initial displacement is equal to zero
while the initial velocity of an element {xy, x2)
of the string possesses a constant value ut{x, 0)
FIG. 21. and outside of the latter is equal to zero.
(3) The initial conditions have the form

(p(x) = 0 ; ф(х) = 0 for x < c

-—гх{2 c — x) for c < x <2c

2c

= 0 for x > 2c .

(b) The one-sided bounded string

(4) The initial velocity is equal to zero [ф{х) = 0] while the initial dis¬
placement has the form of the triangle given in Figure 21. One end
of the string is assumed to be fixed.
(5) As in (4), but for a free end x — 0.
(6) The initial conditions read

<p(x) = 0 , ф(х) = 0 for 0 < x < c

= фо = const. for c < x <2c
- 0 for x > 2c

and the end x = 0 is assumed to be fixed.

(7) As in (6), but again for a free end x = 0. The string profile is in¬
dicated for all the problems (1) through (7) at the times

t0 — 0, k = l,2,---,8.
64 HYPERBOLIC DIFFERENTIAL EQUATIONS

They are in the x, t phase plane, which is interpreted for the zones
corresponding to the different states.
2. Find the solution of 1(1) for all values of x and t (thus the formula
representing u{x, t) which differs for the different zones of the phase plane).
3. Determine the displacement at a point (x0, t0) by using the x, t phase plane
and the x, и plane in which (Figure 21) an initial displacement (<p = 0) is given
for an unbounded string and also for a semi-infinite string with a fixed (or
a free) end.
4. At one end of a long cylindrical tube filled with gas there is a piston
which moves according to an arbitrary law x — f(t) with the velocity v =
/'(/) < a. The initial displacement and the initial velocity of the gas particles
are equal to zero. Find the displacement of the gas at the plane with co¬
ordinate x. In this case assume that the piston moves with a constant velocity
c < a. What can be said about the solution of the problem if the piston at
a certain time has a velocity v > a?
5. A wave u(x, t) — f(x — at) propagates along an unbounded string. Choose
the condition of the string at the time t = 0 as the initial condition and then
solve the wave equation with the corresponding initial conditions; compare
with problem 1(1).
6. By joining two elastic rods with the characteristics
Uehal

о
V

kl, for
II

ki , Рг i for x > 0
Г p2

at point x = 0, an elastic infinitely long rod is obtained,

(a) In the region x < 0 a wave

u{x, t)

is prescribed where / is the given function. Find the reflection and trans¬
mission coefficients that are due to the passage of the wave at the boundary
(x — 0). Investigate the conditions under which no reflection of the wave
occurs.
(b) Solve the corresponding problem when the local initial displacement

u(x, 0) = 0 for x < Xi

= <p(x) for Xi < x < хг < 0
= 0 for X > Хг

is given and the initial velocity is equal to zero.

7. At any point x = x0 of a string hangs a weight of mass M, and in the
region x < 0 the wave

u(x, t) — f I

is prescribed. Determine the transmission and the reflection of the wave.

2-2. WAVE-PROPAGATION METHOD 65

8. At one end (x — 0) of a semi-infinite tube (x > 0) filled with an ideal gas,

there is a freely moving piston of mass M. At time t = 0 the piston acquires
an initial velocity v0 by an impulse. Find the path of the wave propagation
in the gas when it is known that the initial displacement and the initial
velocity of the gas particles are equal to zero. Consider the solution of the
wave equation for x > 0. Also use the boundary condition

Mutt(0, t) = Syp0ux{0, t),

where p0 is the initial gas pressure, 5 is the area of the cross section of
the tube, у — cp/cv, and the initial conditions at the boundary are u{0, 0) =
0, ut(0, 0) = 0.
Solution:

u(x, t) = ^Mh-[ 1 _ е(№о«/*а*»<*-•*)] for д: - at < 0

rPoS
u{x, t) = 0 for x — at > 0 .

9. An infinite string, which at point x = 0 has a concentrated mass M, is

in a state of equilibrium. At the initial moment t = 0, the center of gravity
of the mass M acquires by an impulse an initial velocity v0. Prove that the
string at time t > 0 has the form shown in Figure 22 where их(х, t) and u2{x,t)
are defined by

FIG. 22.

Mavp
Ui(x, t) [1 for x — at < 0
direct wave: 27
= 0 for x — at > 0
Mav о [l_e-<*r/*.u(*+.«>]
fl , A wave:
reflected
,
Ui{x’l)
,ч
= ~OlT
^
for x — at < 0

= 0 for x — at > 0

Use the conditions

t) - 0, t) = 7-^(0, t) = 7^(0, t).

at at ox ox
10. Solve the problem of the propagation of electrical vibrations in an un¬
bounded conductor with the condition

G_ R_
C - L '
66 HYPERBOLIC DIFFERENTIAL EQUATIONS

and with arbitrary initial conditions.

Solution:

v{x, t) = e ^R/L)t[(p{x — at) + ф{х + at)}

i Ce~[R/L)t[(p(x — at) — ф(х + at)] .

І-/

If

v(x, 0) = f{x)

then the following holds:

/(*) + F{x) _ fix) - Fix)

2 ~ 2

11. Find the solution of the integral wave equation for a semi-infinite string
when the boundary conditions prescribed are of the third kind.
12. A membrane is fastened at the end x = 0 of a semi-infinite spring. It
is undergoing resistance due to the longitudinal vibrations of the spring and
is proportional to the velocity uti0, t). Investigate the course of vibrations
if the initial displacement is known and ut(x, 0) = fix) = 0.

2-3. SEPARATION OF VARIABLES

1. The free vibrations of a string

The method of separation of variables, also called the Fourier method,

is one of the best-known methods for the solution of partial differential equa¬
tions. We shall investigate this method using as an example the vibrations
of a string fastened at both ends. We shall discuss the problem in detail
and then refer to the investigations which follow without repeating the in¬
dividual proofs.
First, we seek a solution of the equation

Utt - @ Mxx f (2-3.1)

which satisfies the homogeneous boundary conditions

ui0, t) = 0 , uil, t) = 0 , (2-3.2)

and the initial conditions

uix, 0) = <pix) , Utix, 0) = фіх) . (2-3.3)

The differential Eq. (2-3.1) is linear and homogeneous. Hence the sum of
individual solutions again is a solution of this equation. Accordingly, when
we have a sufficient number of individual solutions, the desired solution is
obtained by summing the individual solutions when each solution is multi¬
plied by a suitable coefficient.
For this purpose we shall consider the following fundamental lemma:
Find a solution of (2-3.1), which does not vanish identically, which satisfies
2-3. SEPARATION OF VARIABLES 67

the homogeneous boundary conditions (2-3.2), and which is represented in the

form of a product

u(x, t) = X(x)T(t), (2-3.4)

where X(x) is a function of x alone and T(t) is a function of t alone.

By inserting expression (2-3.4) into Eq. (2-3.1) we obtain

X"T \t"x (2-3.5)

a
or, after dividing by XT,

X"(x) _ 1 T"(t)
(2-3.6)
X(x) a2 T{t)
Thus, if the function defined by (2-3.4) is a solution of (2-3.1), Eqs. (2-3.5)
and (2-3.6) must be identical, i.e., for all values of x, such that 0 < x < l, t > 0.
The right side of (2-3.6) is a function of t alone, while the left side is de¬
pendent only on x. If, for example, x is fixed and t changes (or conversely),
we see that the right and the left sides of (2-3.6), for changes in their argu¬
ments, remain constant:

X"(x) _ 1 T"(t)
(2-3.7)
X(x) a2 T(t)
where X is a constant which for the following investigations is appropriately
marked with a minus sign. Nothing will be assumed about the sign of X.
From the relation (2-3.7) we then obtain for the determination of X(x)
and T(t) the ordinary differential equations

X"(x) + XX(x) = 0 (2-3.8)

T"(t) + a2XT(t) = 0 . (2-3.9)

The boundary conditions (2-3.2) yield

u(0, t) = X(0)T(t) = 0
u(l, t) = X(l)T(t) = 0 .
Hence it follows that X(x) must satisfy the auxiliary conditions

X(0) = X(l) = 0 (2-3.10)

while in particular if

T{t) = 0 and u(x, t) = 0

we could still find a nontrivial solution. For T(t), no auxiliary conditions
are prescribed.
Consequently we are led to the relation for the determination of the
function X(x) for the simplest eigenvalue problem:
Determine the values of X for which a nontrivial solution of the problem

X" + XX = 0, X(0) = X(l) = 0 (2-3.11)

exists and find the corresponding solutions. The X value for which a solution
68 HYPERBOLIC DIFFERENTIAL EQUATIONS

exists is called an eigenvalue, and the corresponding solutions are called

eigenfunctions. The problem just formulated is a special case of a Sturm-
Liouville eigenvalue problem.
In the following considerations we will distinguish between three cases:
1. For X < 0, the problem possesses no nontrivial solution. The general
solution of this equation has the form

X(x) = Cie'/~K* + C2e~'r=*x,

while the boundary conditions are

X(0) = C, + C2 = 0 ,

X(l) = C1e* + C2e~* = 0 , a = iV^l,

i.e.,

Ci =— C2 and Ci(ea — e~a) = 0 .

Since now a for X < 0 is real and positive, we have ea — e~a Ф 0. Thus

Cy = 0 , C2 = 0

and

X(x) = 0.

2. For X = 0 likewise, no nontrivial solutions exist, since here the general

solution is

X(x) = ax + b ,
while the boundary conditions are

X(0) = [ax + b]x=о ~ b = 0 ,

X(l) = al = 0 ,
i.e., a = 0 and b = 0; thus

X(x) = 0.
3. For X > 0, the general solution possesses imaginary exponents and
therefore can be represented in the form

X(x) = Dy cos i/Ix + D2 sin i/Xx .

The boundary conditions are

X(0) = A = 0
X(/) = D2 sin \/ll = 0 .
If X(x) does not vanish identically, then D2 Ф 0 so that

sin V~Xl = 0 (2-3.12)

or
2-3. SEPARATION OF VARIABLES 69

where n is an arbitrary integer. A nontrivial solution is therefore possible

only for the values

These eigenvalues correspond to the eigenfunctions

Xn(x) = Dns'm~-x.

Hence for those values of X which equal

(2-3.13)

there exist only the nontrivial solutions

-\r / \ TtYl
•
Xn(x) = sin —j~x > (2.3.14)

which, except for arbitrary factors which we have set equal to unity in (2-3.14)
throughout, are determined uniquely. The solutions of Eq. (2-3.9) corre¬
sponding to these X values are

Tn(t) = An cos —-at + Bn sin —af > (2-3.15)

where An and Bn are coefficients yet to be defined.

With regard to the three problems above, we know that the functions

л 7T71 . r, . Tfyi
I \ 7m I
un{x, t) = Xnix)Tn(t) = Ancos—j-at + BnSm-j-at (sin (2-3.16)

are specific solutions of Eq. (2-3.1) which satisfy the boundary conditions
(2-3.2). They can be described by (2-3.4) as the product of two functions, of
which one is dependent only on x and the other only on t. These solutions
must satisfy the initial conditions (2-3.3) of our original problem only for
prescribed functions <p(x) and ф(х).
We turn now to the solution of the problem in the general case. Because
of the linearity and homogeneity of (2-3.1) the sum of the particular solutions

u(x,t)= J.un(x,t)= У (An cos-^y-at + Bn sin -^-at] sin-^p-x , (2-3.17)

n= 1 71=1 \ I l ) l

if it converges and is twice differentiable with respect to x and t, term by

term, likewise satisfies this equation and the boundary conditions (2-3.2).
We shall treat this question in greater detail later (see Section 2-3 § 3). The
initial conditions yield

u(x, 0) = <p(x) = 2 Unix, 0) = 2 A„sin —j-x

n 1 /
(2-3.18)
Utix, 0) - ф(х) = i -%-(*, 0) = I ^f-aBnsm^-x .
n=1 Ol n~1 L l
70 HYPERBOLIC DIFFERENTIAL EQUATIONS

From the theory of Fourier series it is now known that an arbitrary piece-
wise continuous and piecewise differentiable function f(x), which is defined
in an interval 0 й x sS /, can be developed in a Fourier series

fix) = S bn sin ^-x (2-3.19)

n=1 /

where14

bn = —— f1 f(£) sin . (2-3.20)

I J0 *
If <p(x) and ф{х) can be developed in a Fourier series, we obtain

<p(x) = І <pn sin -^j-x ,

»=i /
(pn -4 <p(£) ~[
/Jo
sin
/
, (2-3.21)

Ф(Х) - 1 фп sin 22-x , Фп = 4 f 0(0 - sin ^ . (2-3.22)

»=i / /Jo /
Comparison of these series with Eq. (2-3.18) shows that in order to satisfy
the initial equations we must set

An = <pn , Bn — ——фп . (2-3.23)

7rna

Hence the function (2-3.17) which gives the solution to the problem under
consideration is defined completely.
The solution can be determined in the form of an infinite series (2-3.17).
If, however, series (2-3.17) diverges or the function determined by this series
is not differentiable, then, of course, it is not the solution of our differential
equation.
We shall limit ourselves at this point to the formal construction of the
solution. The conditions under which the series (2-3.17) converges and rep¬
resents a solution will be investigated in Section (2-3.3).

14 Usually one considers the periodic functions of period 21

F(x) = — + X cos —x + bn sin ~j~xj ,

an = yj' F(e)cos-y-edf, bn =yj|_ F(f)sin-y-fde .

If F{x) is odd then <z„ = 0; thus

F{x) = ХЪп sin^A-x ,

71 = 1 i

bn = у £ F(f) Sin -y-£d£ =y[W) sin -y-£d£.

If F(x) is defined only in an interval (0,/) then F(x) can be continued oddly and developed
in the interval from -l to +/. This then leads to formulas (2-3.19) and (2-3.20).
2-3. SEPARATION OF VARIABLES 71

2. Interpretation of the solution

We shall turn now to the interpretation of the solution obtained. The

function un(x, t) can be written in the form

, ( л ttn , n . лп \ . лп
un(x, t) = ( zUcos ——at + Bn sm—j-fltfjsin-y-

— a„cos——a(t + On) sin —-—x (2-3.24)

i L

with

a» = V.Al + Bl , ^а5п = ” arc tg ~ . (2-3.25)

Each point x0 of the string describes a harmonic vibration

Ttfl . 7ГН
un(xо, t) = an cos —-a(t + dn) sin ~j~x0

with the amplitude

7Г n
an sin —j~x° •

Each such motion of a string is designated as a standing wave. The point

x = miUn), where m — 1, 2, • • •, n — 1, in which sin innll)x = 0, remains fixed
in the course of the entire process and is called a node of the standing wave.
By contrast the point x = (2m + l)/2w, where m = 0, 1, • • *, n — 1, in which
sin іппЦ)х = ± 1, vibrates with the maximum amplitude an. One designates
this as the maximum of the standing wave.
The profile of the standing wave is described for arbitrary t by

Unix, t) = Cn(t) sin —j~x .

Thus

Cnit) = a cos Wnit + <5„) , (on = ~Y~a •

When cosa>n(/ + <5„) = ±1 at the time t, the displacement reaches its maxi¬
mum value when the velocity equals zero. By contrast the displacement for
those t values for which cos wnit + dn) = 0 is valid equals zero but the
velocity is maximum. The frequencies of vibration of all points of the string
coincide and have the values

ЛП
(On —-—CL (2-3.26)

The frequencies oon are called the eigenfrequencies of the vibrating string.
For a transverse vibrating string a2 = Tip, and accordingly

лп /Т
GO (2-3.27)
n_ / V ~p '
72 HYPERBOLIC DIFFERENTIAL EQUATIONS

The energy of the nth standing wave (the nth harmonic) for a transverse
vibrating string equals

F —
=Ц.Ш)+ Tif)]dx

= j \fw“ s^n2 Wn^ ^ sin2T- ^ cos2 Wn^ ^cosi! ~f~x^x

= ~^~2 sin2 wn{t + dn) + COS2 wn{t +5„) (2-3.28)

>]
and

• 2 КМ i I 2 7ГП 7 l
ь
sin ——xdx = \ cos ——xdx — — .
/ l
By using the expressions for am and wn as well as the relation T = a2p, we
obtain
2 2 /I 2 I D2
77 pOCn^n / 2 -^T-n I Dn
bn — 1—t—I — 0)nM--- , (2-3.29)
4 4

with M = Ip, the mass of the string.

The vibrations of a string can usually be observed acoustically. Without
going into the process of wave propagation in air and the observation of
sound vibrations by ear we can say that the acoustical effect of a string is
composed of simple tones. The splitting into simple tones is not only a
mathematical operation; it can also be observed with the help of resonators.
The pitch of a tone depends on the frequency of the vibration corre¬
sponding to this tone. The strength of the tone, by contrast, is dependent
on its energy, i.e. on its amplitude. The lowest tone which the string pro¬
duces is determined by the lowest natural frequency, au = (n/l)(\/Tlp), and is
called the basic tone of the string. The rest of the tones whose frequencies
are multiples of oij are called overtones of the string. The tone color, finally,
is obtained from the presence of the overtones in addition to the basic tone
and the distribution of the energy of the individual overtones.
The lowest tone of a string and its tone color are dependent on the ex¬
citation of the string. The type of excitation is defined namely by the initial
conditions

u(x, 0) = <p(x) , ut(x, 0) = <p(x), (2-3.3)

and these determine in turn the coefficients An and Bn ■ If = Bl = 0, the

lowest tone is that one which corresponds to the frequency wn where n is
the smallest number for which An or Bn is different from zero.
Usually a string produces one and the same tone. In order to under¬
stand this, we start a string vibrating by striking it on one side, and neglect
the initial velocity. Then

ut(x, 0) = 0 , u(x, 0) = <p(x) > 0

and
2-3. SEPARATION OF VARIABLES 73

<p(£) sin > 0

о

because

• 7Г - л
sin — £ > 0.

The resulting coefficients are generally significantly smaller than At,

since the function sin (nn/l)t; for n 2; 2 changes its sign. If y>(x) is in parti¬
cular symmetrical with respect to the center of the interval, then A2 = 0.
Consequently the lowest tone of a string which is set to vibrating by
one-sided impulses, cp(x) > 0, is just that tone whose energy is in general larger
than the energies of the other harmonics.
A string can also execute other types of vibration. For example, if the
function <p{x), occurring in the initial conditions with respect to the center
of the interval, is odd, then

Аг= 0,
and the lowest tone corresponding to the frequency

is obtained. If a tuning string is contacted exactly in the middle, then it

vibrates an octave higher than its natural tone with changing tone color.
This type of tone conversion is often used in playing violins, guitars, and
other string instruments. The tones thus produced are called flageolet tones.
From the standpoint of the theory of a vibrating string this phenomenon is
completely understandable. At the moment in which the string touches ex¬
actly in the middle, those standing waves having displacement at this point
are extinguished, and only those waves remain which possess a node at the
point of contact. Thus, only the even harmonics remain, so that the lowest
frequency is

If the string (0, l) is touched at the point x — 1/3, it increases the pitch of
the basic tones threefold, since here only those harmonics whose nodes lie
at the point x = Z/3 remain. The formula

or (2-3.30)

for the frequency or the period of the basic vibration explains the following
rules which were first found experimentally. (1) For strings of uniform
density and uniform tension, the vibration period is proportional to the
length of the string. (2) In a prescribed length of string, the period is
inversely proportional to the square root of the tension. (3) For prescribed
lengths and tension, the period is proportional to the square root of the linear
74 HYPERBOLIC DIFFERENTIAL EQUATIONS

density of the string. These three can be easily demonstrated in a monocord.

In this section the existence of standing waves by the vibrations of a
string with fixed ends is demonstrated. It must still be mentioned that the
existence of solutions of the form

u(x, t) = X(x)T(t)
is equivalent to the existence of standing waves, since the profiles of such
a solution at different times are proportional to each other.

3. Description of arbitrary vibrations by superposition of stand¬

ing waves

Earlier, in Section 2-3 § 1, we treated the free vibrations of a string fixed

at the ends. There we proved the existence of special solutions in the form
of standing waves and presented a formal scheme for the representation of
an arbitrary vibration as an infinite sum of the standing waves. We shall
prove in this section that such a representation is possible. First of all, a
generalization of the superposition principle, known for finite sums, will be
applied to infinite series.
Let L(u) be a linear differential operator. This means that L(u), that is,
L applied to a function u, is equal to the sum of the corresponding deriva¬
tives of the function и with coefficients which are independent of u. There¬
fore, both ordinary as well as partial derivatives are admissible.

Lemma (generalized superposition principal): If the functions Ui (i = 1,2, • • •)

are separately solutions of a linear and homogeneous differential equation

L(u) = 0, then the series и = J CiUi is likewise a solution of the differential

t=i

equation, provided the derivatives of и appearing in L(u) can be differentiated

termwise.
Thus, if the derivatives of и occurring in L(u) = 0 can be differentiated
termwise, we have
дпи — у r d„Mj
dxmdtn~m ~i~'iidxmdtn m ’
and because of the linearity of the equation we can write

L(u) = L( 2 CiUi) 1 CiL(iti) = 0

i 1

since a convergent series can be added termwise. Hence it is shown that и

satisfies the differential equation. As a sufficient condition for the termwise
differentiability, we shall use the uniform convergence of the series15
oo
dniij
ICi (2-3.31)
dxmdtn~m '
We return now to our boundary-value problem. First we have to prove
the continuity of the function

15 See V. I. Smirnov, Textbook of Higher Mathematics, 2d ed., Part II, Berlin, 1958.
2-3. SEPARATION OF VARIABLES 75

u(x, t) = J. un(x, t) = 2 (An cos at + Bn sin ^~at]sin ^—x . (2-3.32)

n=l n=l\ / l J l

From this it then follows that u{x, t) depends continuously on the correspond¬
ing initial and boundary conditions. Here it is sufficient to prove the uniform
convergence of the series which represents the function u(x, t), since the
general term of the series is a continuous function, and a uniformly conver¬
gent series of continuous functions represents a continuous function. There¬
fore we can proceed as follows: From the well-known inequality

\un(x,t)\ й I An \ + \Bn\
we conclude that the series

2 (I An I + I Bn I) (2-3.33)

is a majorant of the series (2-3.32). If the series (2-3.33) converges, then the
series (2-3.32) converges uniformly, and u(x, t) is continuous.
Furthermore, in order to see that ut(x, t) depends continuously on the
initial conditions, the continuity of this function must be demonstrated; there¬
fore, it is sufficient to prove the uniform convergence of the series

£ du„ ПП 7ІЯ ПП
ut(x, t) ^ TT
— 1 a~T~( — AnSm-^B^at + Bn cos -^-at jsin-^f1* (2-3.34)
' n 1 dt l l l l

or the simple convergence of the majorant series

I n(\ An \ + Bn\). (2-3.35)

L 71 — 1

Finally, in order to show that the function u(x, t) is an integral of the

differential equation, it is sufficient to show by the use of the generalized
superposition principle that the series representing u(x, t) is twice differenti¬
able termwise, which suffices for the proof of the uniform convergence of
the series

/ у d2un n x- 21 j-,
Л • nn j \ .nn . J. ГЬ HfL
2. n ( An cos ——at + Bn sin ——at sin —— x
'Л dx2 l ) n—\ l l l
■ИЛ nn Л . nn
2
Z d2un nn
un n ( An cos—j-at + B.sin — at \ sm-j-x
n=i dt2

This corresponds within constant factors to the common majorant

I n2{\ An\ + \Bn\ (2-3.36)

Л=1

Since

An - фи J Bn -фп ,
nna
where
2 f! . . . 7in , . 2 [l .... nn ,
<рп = —l <p(x) sin ——xdx , = —j ФМ sm~J~xdx •
76 HYPERBOLIC DIFFERENTIAL EQUATIONS

our problem is solved, as soon as the convergence of the series

f nk 1 <pn 1 , k = 0,1,2
n=1
(2-3.37)

Іпк\фп\ , k =- 1,0,1
n —1

is demonstrated.
To demonstrate the convergence of (2-3.37) we use some well-known pro¬
perties of Fourier series.16
If a periodic function F(x) with period 2/ possesses continuous derivatives
up to and including the kth order, while the (k + l)th derivative is only
piecewise continuous, then the series

2n*(|e»| + |6.|) (2-3.38)

n -1

converges, where an and b„ are the Fourier coefficients. If we are dealing

with development of a function f(x), which is given only in the interval
(0,/), in a series with respect to the functions in sin (nn/l)x, we require that
the conditions stated above hold for the odd continuation F(x) of the func¬
tion f(x) in order to guarantee the convergence of this series. Further for
the continuity of F(x) it is necessary that /(0) = 0, since otherwise a dis¬
continuity in the odd continuation would appear at the point x = 0; accord¬
ingly, /(/) must also equal zero at the point x = l, since the continued func¬
tion is continuous and has a period equal to 21. The continuity of the first
derivative at x = 0, x = l, follows automatically for the odd continuation.
In general, we must require that the even derivatives of the continued func¬
tion satisfy
/‘*’(0) = f{k)(l) = 0 , k = 0,2, 4, • • • ,2n , (2-3.39)

while for the continuity of the odd derivatives no auxiliary conditions are
imposed.
For the convergence of the series

I nk 1 <pn I , k = 0,1,2
n=1

it is sufficient to require that the initial displacement <p(x) satisfy the follow¬
ing requirements:
Condition 1. The derivatives of <p(x) are continuous up to and including
the second order, the third derivative is piecewise continuous, and moreover

y>(0) = <p{l) = 0 , ip"(0) = <p"(l) = 0 . (2-3.40)

For the convergence of the series

Іпк\фп\, k =- 1,0,1
n= l

we require that the initial velocity ф(х) satisfy:

19 Ibid,., cf. footnote 14.

2-3. SEPARATION OF VARIABLES 77

Condition 2. The function ф(х) is continuously differentiable and possesses

a piecewise continuous second derivative, and

Ф(0) = ф(1) = 0 . (2-3.41)

In summary, we have demonstrated that an arbitrary wave u(x, t) can be

represented by the superposition of standing waves if the functions <p(x) and
<p(x) occurring in the initial conditions satisfy Conditions 1 and 2. Conditions
1 and 2 are sufficient for the method of proof used here.
An analogous problem was solved in Section 2-2 §5 with the help of the
method of wave propagation, and the solution was given by

Ф(х — at) + Ф(х + at)

x+at
1
u(x, t) ¥(a)da , (2-3.42)
2 + ~2d x_ at

where Ф and ¥ are the odd continuations with respect to 0 and l of the
prescribed initial functions ф(х) and <[>(x) defined in the interval (0, l). The
functions Ф and ¥ are, as shown, periodic functions of period 21 and there¬
fore can be represented by the series

л/ \ £ , • 7ГИ
Ф(х) = Z <pn sin——*,
nri \
¥(x)=Z<pn sin——x,
n 1 i n I /

where y>„ and фп are respectively the Fourier coefficients of the functions
<p{x) and ф(х). By insertion of this series in (2-3.42) we obtain, with the aid
of the corresponding addition theorem for the trigonometric functions,

u(x, t) = Z (<Pn cos —-—at H-фп sin ——at sin ——x , (2-3.43)
n i\ l nna l ) t

which is exactly the representation given by the method of separation of

variables.
Formula (2-3.43) therefore is valid under the same assumptions as (2-3.42)
(see Section 2-3§l), which was derived under the conditions that Ф(х) is twice
and ¥(x) is once continuously differentiable.
With regard to the functions <p(x) and ф(х), besides the differentiability
conditions we still require that the following be satisfied:

<p{0) = <pd) = 0 , ф(0) = ф(1) = 0 , <p"(0) = <p"(l) = 0 . (2-3.44)

Thus Conditions 1 and 2 from the method of proof, while sufficient for
the exact foundation of the method of separation of variables, depend on and
contain additional conditions, in comparison with conditions which guarantee
the existence of the solution.
For the foundation of the representation of the solution by a superposi¬
tion of standing waves, we used the first method for the proof of the con¬
vergence of the series, since it did not depend on the special form of solu¬
tion (2-3.42), which is applicable only for the solution of the simplest vibra¬
tion problems. Furthermore, this method is carried over without difficulty
to a series of other problems, although more stringent requirements are
imposed on the initial functions.
78 HYPERBOLIC DIFFERENTIAL EQUATIONS

4. Inhomogeneous equations

We shall now consider the inhomogeneous wave equation

Utt — d Uxx ”1“ f ІХ , t) t 0 < x < l, (2-3.45)

with the initial conditions

и (x, 0) = <p(x)
(2-3.46)
Mt( x, 0) = <p(x)
and the homogeneous boundary conditions

u(0, /) = 0
t > 0. (2-3.47)
u(l, t) — 0

We shall represent its solution as a Fourier series with respect to x:

u(x, t) = 2 un{t) sin ^fLx , (2-3.48)

n= l I

in which t is considered as a parameter. For the determination of u(x, t) we

must determine un(t). The function f(x,t) and the initial conditions likewise
can be written as a Fourier series:

fix, t)=f fnit) sin 2*-x fnit) = -f-f /(?. t) sin ,

n=i / l J0 /

<pix) =i<Pn Sin Щ-Х Vn = 4- ('1 №) sin , (2-3.49)

n=l l l Jo /

Фіх) = 1 Фп sin dff-x фп = ^-[ ф(£) sin-^L£d$ .

»=l / /Jo /

If we insert (2-3.48) into the original Eq. (2-3.45) we obtain

f sin-y-xj— unit) - Unit) + fnit)I = 0 .

This relation is satisfied if all the coefficients of the series development

vanish, i.e., if

unit) +(-y-)*(fu'nit) = Ш . (2-3.50)

Therefore, we have obtained for the determination of u„it) an ordinary dif¬

ferential equation with constant coefficients. The initial conditions read

uix, 0) - <p(x) — 2 uni0) sin f—x = f p, sin fy-x t

n—1 / n =1 /

Utix, 0) = фіх) = 2 M„(0) sin -ff-x = 2 фп sin dff-X

71 = 1 L 71 = 1 L

Hence it follows that

2-3. SEPARATION OF VARIABLES 79

Unity - (fn
(2-3.51)
Unity = фп •

These auxiliary conditions completely define the solution of Eq. (2-3.50). Now
un{t) can be represented in the form

Unit) — Un \t) + ui">it),

where

ul!\t) = -1— f sin ^-a(t - r) • fnWv (2-3.52)

тіпа Jo l
represents a solution of the inhomogeneous equation with homogeneous initial
conditions,17 and
(II)/ .. ТіП I . t ft ТіП I (Л n I— r\ ,
un it) = co„cos——at -1-фп sin ——at (2-3.53)
/ тіпа l
is a solution of the homogeneous equation with the prescribed initial condi¬
tions. Consequently, the solution sought has the form

uix, t) = 2 —— ( sin —r~ait — r) sin ^-x • fnitydr

n=i xna Jo I I

ПП Д ЛП ,0 о C/U
+ I U, cos ——at H-фп sin ——at ) sin ——x .
, / nn t I I I • .
(2-3.54)
n=i\ / тіпа 11 l
The second sum solves the problem of a freely vibrating string with pre¬
scribed initial conditions and was completely investigated earlier. The first
sum by contrast represents the forced vibrations of the string under the
influence of an external force with homogeneous initial conditions. With
the use of (2-3.49) for fn(t) we find

ua>ix,t)=\ [ Л—— sin -^j-ait — t) sinsin -^p-?l/(f, т)(1Ыт

Jo Jo l I n=1 тіпа l It)

= j* j!G(x, f, t - г)/(£, T)dtdv (5-3.55)

with

Gix, £,t — z) — j[ — sin —j-ait — z) sin -^-x sin -^-£ . (2-3.56)

na »=i n l It
To determine the physical significance of this solution, we first assume
that the function fi$,z), in a sufficiently small neighborhood

£o ^ ^ £o + , т0 й т.й t0 + At

of a point Mo(£o,r0), is different from zero and vanishes outside this neigh¬
borhood. The function pfi£, r) represents the force density of the acting
force; the force developed in the interval (f0.fo + ^£) is therefore equal to

№ flrff •
fo

17 See the comments at the end of this section.

80 HYPERBOLIC DIFFERENTIAL EQUATIONS

Then
Stq+Jt

F(T)dz = p\
Г то+Лт +
f(Z,z)d$dz
tq J r0

is the impulse of the force during the time dr. By applying the mean-value
theorem to the expression
pZ pZ p то+Лт p

u(x,t)=\ \ G(x,€,t— z)f(£,z)d£dz = \ 1 G(x, t — z)f(Z,z)d£;dz

J0 J0 J r0 J f0
we obtain
STQ + dr p ^0 + ^^
\ /(£, z)d^dz (2-3.57)
^0 J f0
with
fo й I ^ £o + d£ , z0 ^ f ^ г0 + dr .
If in formula (2-3.57) we take the limit as >0 and dz —> 0, we obtain
the function

u(x, t) = G(x, 6o ,t — r0)— , (2-3.58)

P,

which can be regarded as the effect of an instantaneously concentrated im¬

pulse I.
If the function (1 /p)G(x, $,t — z), which represents the effect of a concen¬
trated impulse, is known, then it is immediately clear that the effect of a
continuously distributed force f(x, t) can be represented by

u(x, t) = Г' Г G(x, 6, t - r) / (£, :)d№ .

J0 J0
(2-3.59)

This representation coincides with the representation (2-3.55) above.

The function describing the action of concentrated impulses was investi¬
gated for the case of the infinite straight line in the preceding paragraphs.
We recall (see Figure 10) that it is piecewise constant, and within the upper
angle for the point (£, r) this function is equal to (1/2a)(I/p)\ everywhere out¬
side of this angle the function is equal to zero.
From this the function describing the action of
concentrated impulses for the bounded string
J_ I
2a p (0, l) can be found by the odd continuation with
respect to the points x — 0 and x = l.
Let t be so near to г that the influence of
the reflection at x = 0 and x — l still does not
appear. At this time the influence function takes
the form shown in Figure 23. If we develop it
£-o(/ - TV £ * f + oU- t)
in a Fourier series with respect to sin (ш/1)х,
fig. 23. in which we set / = p, its Fourier coefficients
are then equal to
•e+a(t- T)
7Г П
G(a, t — z) sin-^j-ada = -( sin ada
^d J f—a{t — z l
2-3. SEPARATION OF VARIABLES 81

= —— (cos -^-[£ — a(t — r)] — cos + a(t — r)]l

ann [l l )
2 nn ^ ■ nn .. ,
=-sin—— f sin —— a(t — r) .
ann l l
From this we obtain the formula

G(at, f, t — t) = - 2 — sin ——a(t — r) sin ——x • sin —— £ . (2-3.60)

па n=m l l l
Thus it agrees with the formula (2-3.56) found by the method of separation
of variables.
For values t ^ z, where the influence of the fixed ends has already ap¬
peared, there are difficulties; in the construction of the influence function by
means of characteristics its description by Fourier series remains to be ob¬
tained, even in this case.
We shall limit ourselves here to a formal scheme of solution, without
going more precisely into the conditions for the applicability of the formulas
obtained.
We shall now consider an ordinary inhomogeneous linear differential equa¬
tion with constant coefficients
(Гн
L(u) — u[n) + piii(" 11 + • • • + pn-iUil) + pnu —f (t) , a (»> (1*)
df
with the initial conditions

M(i,(0) = 0 , * = 0,1, 1. (2*)

Its solution is given by

u(t)=\ U(t-T)f(z)dT, (3*)

where U(t) is the solution of the homogeneous equation

L(U) = 0,
with the initial conditions

U{i)(0) = 0, г = 0,1, — 2
{/'"-■’(О) = 1 . (4*)

If we calculate the derivatives of u(t) by differentiation of the right side

of (3*) with respect to t, we find

um(t) * U{l)(t-z)f(z)dz + U(0)f(t) , 17(0) = 0

0

um(t) ' Ui2)(t - r)f(z)dz + Un\0)f(t) , U\0) =0

0

.(5*)

* U(n~l)(t-z)f(z)dz + t/u"2,(0)/(() , Uin-2)(0) = 0

0

u'n'(t) * U{n)(t - z)f(z)dz + U{n-'\0)f(t) , Uin-"(0) = 1 .

82 HYPERBOLIC DIFFERENTIAL EQUATIONS

If we substitute (5*) into (1*), we find

L(u) = 1 L[U(t-T)]f{T)dT + f(t) =/(/);

i.e., the differential equation is satisfied. Obviously, u(t) also satisfies the
initial conditions (2*).
For the function U(t) and formula (3*) we can easily find a clear physical
interpretation. Usually u(t) designates the displacement of a given system
and f(t) is the force acting on this system. Our system is in a state of rest
for t < 0. Let the displacement of the system be given by the non-negative
function let fe(t) be different from 0 only at the time 0 < t < e. We
denote the impulse of this force by

1= f / (z)dr .
J0
Further let ue(t) be the function corresponding to /e(/). Therefore we can
consider e as a parameter and set 1=1. Obviously then, as e—>0, lim ue(t)
£—►0

exists independent of the selection fe(t). Also one easily recognizes that this
limit value is equal to the function U(t) defined above; namely

U(t) = lim ue(t) ,

e-*o

if (/(/) = 0 for t < 0. U(t) is called the influence function of the instantaneous
impulses.
If we apply the mean-value theorem to (3*) we obtain

ue(t) = U(t — T*) J f{z)dr = U (t — r*) , 0 ^ t* e < t.

By passage to the limit as e—>0 we then obtain the limit value

lim ue(t) = lim U(t — r?) = U(t) ,

e-*o £ —»o

which therefore proves our assertion.

We now want to represent the solution of the inhomogeneous equation
by (7(0. the influence function of the instantaneous impulses. For this purpose
we divide the interval (0, t) by points и into equal intervals

and write f(t) in the form

m

/(» = i m, t=i

where

fi(t) = 0 for t < ті and t ^ Ti+i

= f(t) for Ті й t < Ti+l .
2-3. SEPARATION OF VARIABLES 83

Then

U{t) = I Ui(t) ,
І—1

where Ui(t) is the solution of the equation L(ui) = f{ with homogeneous initial
conditions.
If m is sufficiently large, Ui(t) can be regarded as the influence function
of the instantaneous impulse of intensity

I = fi(Ti)JT — f (гі)Тг ,
so that for u(t) we have

u(t) — 1 U(t — Z{) f{zi)Ar —» f U (t — z)f{z)dz

t=l Jr-0 J0

and arrive at the formula

u(t) — f U (t — z) f (z)dz .
J 0

This shows that the influence of a continuously distributed force can be

represented by the superposition of the influences caused by the instantaneous
impulses.
In the case considered above, m»1 satisfies Eq. (2-3.50) and the conditions
UnifA) — un{0)- For Uit) we have

17+(-y-)Vf/= 0, 17(0) =0, 17(0) = 1,

so that

U it) =-sin ——at.

nna l

From this and from (3*) there results

u"\t) = [*Uit- z)fniz) = —

Jo япа
Г sin 22-ait - z)fniz)dz .
Jo I

The integral representation (3*) derived above for the solution of the
ordinary differential Eq. (1*) has the same physical significance as formula
(2-3.59), which gave the integral representation of the solution of the homo¬
geneous wave equation.

5. The general first boundary-value problem

The general first boundary-value problem for the wave equation reads:
Find a solution of the differential equation

Uu — a2uxx + fix, t), 0 < x < l, t > 0 (2-3.45)

with the auxiliary conditions

и (x, 0) = <pix)
(2-3.46)
Utix, 0) = фіх)
84 HYPERBOLIC DIFFERENTIAL EQUATIONS

w(0, t) = /^(f)
(2-3.47)
u(l, t) = fi2{t) .
First, by means of

u(x, t) = U (x, t) + v(x, t)

we introduce a new function v(x, t). This then signifies the difference of the
function u(x, t) from a function U(x, t) still to be determined.
The function v(x, t) can be determined as a solution of the wave equation

vtt = a2vxx + /(x, t) , f {x, t) = f{x, t) — [Uit — a2Uxx\

with the auxiliary conditions

v(x, 0) = ф{х) , vt(x, 0) = ф(х)

v(0, t) = ThU) , v (/, t) = M2(t)
ip {x) - (p{x) — U (x, 0) , ф{х) = <p(x) — Ui(x, 0)
AhU) = Vi{t) — U(0, t) , P-г(t) = Mt) - и (l, t) .

Now we choose U {x, t) such that

= 0 and /u2(t) = 0 .
Thus it suffices to set

U(x, t) = Mt) + f Ыі) - ftM .

Hence the general boundary-value problem for u(x, t) has been reduced
to a boundary-value problem for v(x, t) with homogeneous boundary condi¬
tions which we have already solved (see Section 2-3 §4).

6. Boundary-value problems with stationary inhomogeneities

A very important class of problems is formed by the boundary-value

problems with stationary inhomogeneities. In these problems the boundary
conditions and the right side of the equation are independent of the time
t:

Zltt — Wxx T fto(x) (2-3.45')

и (x, 0) = <p(x)
(2-3.46)
ut(x, 0) = ф(х)
и (0, t) = и,,
(2-3.47')
и (l, t) — иг .
In this case we seek the solution of the problem in the form

u(x, t) = U(x) + v{x, t)

where u(x), the stationary state (the static deflection) of the string is deter¬
mined by the conditions
2-3. SEPARATION OF VARIABLES 85

агй"(х) + fa(x) = 0
m(0) = ut
u(l) = u2
while v(x, t) designates the displacement from the stationary state. It is
easily seen that

u(x) = Ul + (ut - +
* ( Jo Jo a Jo jo &
If, in particular, /„ = const., then

u(x) = uy + (и2 — Uy)~ + -£\(lx — x2).

L Z«

The function v(x, t) obviously satisfies the homogeneous equation

Vu = a2vxx
with the homogeneous boundary conditions

v(0, t) = 0
v(l, t) = 0
and the initial conditions

v(x, 0) = ф(х) , <p(x) = <p(x) — й(х) , vt(x, 0) = (p{x) .

Hence v is the solution of the simplest boundary value problem treated

in Section 2-3 § 1.
In the derivation of the equation of the vibrating string and for a series
of other cases the influence of the force of gravity was not considered. From
the above it follows that it is sufficient to assume the displacement from
the stationary state instead of the direct influences of the force of gravity
(and in general for time-independent forces).
We shall now give the solution of the simplest problem of this type with
homogeneous initial conditions:

utt = a2uxx + f0(x) (2-3.45")

u(x, 0) = 0 , ut(x, 0) = 0 (2-3.46")
u(0, t) = Uy , u{l, t) = u2 . (2-3.47")

For v(x, t) the problem then reads

vtt = a2vxx v(x, 0) = <p(x) =— U{x) , v(0, t) — 0

vt(x, 0) = 0 , v(l, t)= 0 .

We can easily see that for the solution of this problem it is not necessary
to know the exact analytic expression for й(х).
According to formula (2-3.17), v(x, t) has the form

v(x, t) = f (An cos ал/Тпі + Bn sin a\/Tnt)Xn(x)

86 HYPERBOLIC DIFFERENTIAL EQUATIONS

where

X„(x) = sin v%x , i/T„ = ~~J~~

are the eigenfunctions of the boundary value problem defined by

X" + IX=0 (2-3.8)

X(0) = 0 , X(l) = 0 . (2-3.10)

From the initial conditions it follows that

Bn — 0
and
2 f! _
An= —— 1 u(x)Xn(x)dx .
I JO
The following method is appropriate for the calculation of these integrals.
From (2-2.8) we find

Xn(x) =-±Xi'(x).
An

We introduce this expression into the formula for An and integrate twice
by parts:

An -u'Xn + [ U"Xn(x)dx\ ,
0 J0 J
from which, by taking into consideration the differential equation and the
boundary conditions for й(х), there follows
_2_
An —

ttn [ u2Xn(l) — UiXn(O) —

or
21 fo(x)
An -[«*(- 1)" — Ml] — 2 Xn{x)dx .
nn Tin2 ) о
In particular we find for the homogeneous equation (f0(x) = 0)

An = -|=[M2(- 1)” - Щ\ =-[«*(- 1)” - uA .

Ivin ГСП

With the aid of this method it is possible to calculate the Fourier coefficients
also for the boundary conditions of the second and third type and for the
boundary-value problems of the inhomogeneous string

+ *p[x)X = 0 ’
if the eigenfunctions and the eigenvalues are known.

7. Problems without initial conditions

As shown above, the general first boundary-value problem for the equa-
2-3. SEPARATION OF VARIABLES 87

tion of the vibrating string can be reduced to the solution of an inhomo¬

geneous equation with homogeneous boundary conditions. However, it often
happens that the solution of a problem by this method is more difficult
than by the direct solution.
Thus it is important in the investigation of the influences of boundary
conditions to find some specific solution (of the homogeneous equation) which
satisfies the prescribed boundary conditions, since the calculation of the
necessary corrections owing to the initial conditions leads to the solution
of the same equation with homogeneous boundary conditions.
A very important class of problems for the study of boundary influences
is composed of the so-called problems without initial conditions.
That is, if the boundary conditions have acted sufficiently long, the in¬
fluence of the initial conditions vanishes owing to the friction which occurs
in every real physical system. We arrive quite naturally at the following
problem (1) without initial conditions:
Find the solution of the equation

Utt = a2uxx — ocUt , a > 0 (2-3.61)

with the prescribed boundary conditions

u{0, t) = nS) , u(l, t) = fi2(t).

The term aut on the right side corresponds to a frictional force proportional
to the velocity.
We shall first investigate the consequences of periodic boundary influences:
u(l, t) = A cos wt or u(l, t) = В sin wt , (2-3.62)
u(0, t) =0. (2-3.63)

For the following, the complex representation of the boundary conditions

is found to be advantageous:
u(l, t) = Аеш . (2-3.64)

If
u(x, t) = un>(x, t) + iu(2){x, t)

satisfies Eq. (2-3.61) with the boundary conditions (2-3.62) and (2-3.63), then
un)(x, t) and uw{x, t), the real and imaginary parts of u(x, t), separately satisfy
the same equation (since it is linear); the condition (2-3.63) and the boundary
conditions at x = l lead to
un)(l, t) = A cos wt , uw(l, t) = A sin wt .
We therefore seek the solution of the problem

Utt = Cl2Uxx — OLUt

u(0, t) =0 (2-3.65)
«(/, t) = Аеш .
With the expression
88 HYPERBOLIC DIFFERENTIAL EQUATIONS

u(x, t) = X(x)eiu,t
we obtain the following problem for X(x):

X" + k2X = 0, кг = ^г- iaAt (2-3.66)

а аг
W(0) = 0 (2-3.67)

X(l) = A . (2-3.68)

From Eq. (2-3.66) and the boundary condition (2-3.67) we find

X(x) = C sin kx .
The condition at x = l gives

A
C (2-3.69)
sin kl
Thus

X(x) = A^Z = XAx) + iXt(x), (2-3.70)

sin kl

where Xi{x) and X2(x) are the real and imaginary parts of X(x).
We can then represent the sought solution in the form

u(x, t) = [Хі{х) + іХ2(х)]егші = u"\x, t) + u'2)(x, t)

with

un\x, t) = Wdx) cos ctit — W2(x)sin oft

u{2\x, t) = Xi(x) sin wt + X2{x) cos wt .
By passage to the limit as a—>0 we obtain

k = lim k = — (2-3.71)
a 0 a

and correspondingly

U[l)(x, t) = lim um(x, t) = ASm{o,la)x COS wt , (2-3.72)

a 0 Sin (wla)l

j/ .4 _ л sin (w/a)x
и \x, t) — lim и
12
(x, t) = A— sin wt . (2-3.73)
a -o sin (w/a)l

The functions и (x, t) and и (x, t) are obviously solutions of the equation
litt — CL lixx

with the boundary conditions (2)

iim(0,t) = 0 йП)(0, t) = 0

й(1)(/, t) = A cos wt й<2>(/, t) — A sin wt.

A solution for a = 0 does not always exist. Thus if the frequency w of the
forced vibrations coincides with a characteristic frequency wn of the vibrat-
2-3. SEPARATION OF VARIABLES 89

ing string with fixed ends, that is,

then the denominator vanishes in the equation for ма> and uw and no solu¬
tion exists for the problem considered here.
This fact has a simple physical significance, namely: for w = wn there
occurs a resonance. At a given time t = t0 the amplitude grows without
bound. With the existence of friction (а Ф 0) a vibration process is possible
for each w.
If f(t) is a periodic function which can be represented by the series

f(t) = + 2 (An cos amt + Bn sin w nt), (2-3.74)

Z 71 — 1

where w is the lowest frequency and An,Bn are the Fourier coefficients,
then the solution of the equation for a = 0 has the form

u(x, t) = f (An cos a>nt + Bn sin cont)Sjn

Zl n=1 si п(а)п/а)1

provided none of the frequencies wn coincides with the characteristic fre¬

quencies of the bounded string.
If f(t) is a non-periodic function, we represent it by a Fourier integral
and the solution is obtained in integral form in a corresponding manner.
We note that the solution of the problem without initial conditions for
a = 0 is not uniquely defined if no further additional conditions are furnished.
Thus if one adds to any solution of this problem even an arbitrary linear
combination of standing waves

_ / ,, Tin . . D . nn A . nn
2 ( An cos -j-at + Bn sin —j~at J sin ——x ,

where An and Bn are arbitrary constants, then the sum so obtained likewise
satisfies the equation and the boundary conditions.
In order that problem (1) for a = 0 be uniquely solvable we introduce
the additional condition on the vanishing friction:
We say a solution of the problem (2) has vanishing friction if it is the
limit value of a solution of problem (1) as a—>0. The problem for а fixed
end x — l and with a prescribed boundary condition u(0, t) = /x(t) at x = 0 is
solved in a similar manner.
The solution of the general problem without initial conditions

m(0, t) = fiM) , u(l, t) = Hi(t)

is calculated as the sum of two summands, each of which satisfies an inhomo¬

geneous boundary condition.
We shall now prove the uniqueness of a bounded solution of the problem
without initial conditions for Eq. (2-3.61). For this purpose we shall assume
that the solution and its derivatives up to and including the second order are
90 HYPERBOLIC DIFFERENTIAL EQUATIONS

continuous in the region 0 й x й l, —co<t<t0 when the boundary values

u(0, t) = fxi(t) , u(l, t) = fi2(t)

are defined for — oo < t < t0.

Let Hi(x, t) and u2(x, t) be two bounded solutions of problem (1) under
consideration,

I иг \ < M , I u21 < M,

where M > 0 is a fixed number.

The difference

v(x, t) = Ui(x, t) —u2(x, t)

of these functions is likewise bounded (\v \ < 2M) and satisfies Eq. (2-3.61)
as well as the homogeneous boundary conditions

v(0, t) = 0 , v(l, t) = 0 .
The Fourier coefficients for v

vn(t) = -y-f v(x, t) sin -^-xdx

L Jо I
obviously satisfy the equation

Vn “l- OCVn “r ^ j ^ f ( )

since the second derivatives of v(x, t) for 0 й x ^ l are continuous.

The general solution of (*) reads
(I), (2).

vn(t) = Ane "» + Bne (**)

where

a) a , / a2 2 (2) a /a2
+ a,n> Qn -y--и», « > 0

are the roots of the characteristic equation. There are now two possible cases:
1. The roots are real and negative.
2. The roots are complex and possess a negative real part. From this
it follows that every solution (**) of Eq. (*) either is identically zero or is
such that its absolute value as t—*—oo becomes arbitrarily large. Now,
however, | vn \ < 4M is valid for all n so that v„ = 0 for all n.
Hence we must have

v(x, t) = 0 and Ui(x, t) = u2{x, t) ,

from which the unique solvability of the problem (I) is proved under the
given assumptions.

8. Action of a concentrated force

A string is set to vibrating under the action of a concentrated force which

acts at the point x = x0. If the force is distributed in a given interval
2-3. SEPARATION OF VARIABLES 91

(x0 — e, x0 + s) then the solution is found from formula (2-3.55). By passage

to the limit as e—>0 we obtain also the solution for a concentrated force.
On the other hand we have seen in the derivation of the wave equation
that at the point x0 at which the concentrated force acts, a point of discon¬
tinuity of the first derivative occurs, whereas the function itself remains con¬
tinuous. The solution u(x, t) can be represented for such a displacement of
the string by two different functions:

u(x, t) — Wj(x, t) for x0

(2-3.75)
u(x, t) = u2(x, t) for x0 Ш x S I.
These functions must satisfy the equation

tin = a uxx, (2-3.76)

the boundary and initial conditions

«ДО, t) = 0 и (x, 0) = <p(x)

(2-3.77)
u2(l, t) = 0 ut(x, 0) = ф(х) ,
the continuity condition at the point x = x0

MlUo , t) = u2(x0, t) , (2-3.78)

and the condition which determines the magnitude of the jump in the first
derivative at the point x0 at which the concentrated force f(t) acts:

дм/0+0 du2 dux fit)

Uo, t) — (x0 ,t) - (2-3.79)
dx Xq—o dx dx k
If the initial conditions are satisfied we need not investigate further. Thus
if we find a particular solution of Eq. (2-3.76) which satisfies both the bound¬
ary conditions of (2-3.77) and the relations (2-3.78) and (2-3.79), the prescribed
initial conditions can also be satisfied only when the corresponding solution
of the homogeneous equation is added to this solution.
We shall now seek a solution of the equation for the special case

fit) = A cos wt — oo < t < + oo .

Hence only the boundary conditions are to be fulfilled in which we assume

that the force has been acting from t—— oo on, i.e., it is a problem without
initial conditions. For the solution we set

ux{x, t) = Xi{x) cos wt for 0 ^ x ^ x0

u2(x, t) = X2{x) cos wt for Xo ^ X й I •

Then there follows from (2-3.76)

Xi" + (y)2^ = 0 for 0 ^ x g x0

(2-3.80)
Xf + (—Xx,
\ a /
=o for x0 ^ x fk I •

The functions Xx and X2 on the other hand must satisfy the boundary con¬
ditions
92 HYPERBOLIC DIFFERENTIAL EQUATIONS

*,(0) = 0 , Xt(l) = 0 (2-3.81)

which results from (2-3.77) and also satisfy the continuity and jump conditions

XAxo) = X2(x0) , X((x0) - X'i(x0) = 4 . (2-3.82)

k
which are derived from (2-3.78) and (2-3.79).
From Eq. (2-3.80) and the condition (2-3.81) we find

Xi(x) — C sin— x , X2(x) = D sin — (/ — x).

a a
The connection conditions (2-3.82) give

C sin—x0 — D sin — (/ — x0) = 0 ,

a a

^о) w , со cd ., .
гл A
C— COS —X0 + D— COS —(l — X0) =— .
a a a a k

Therefore, when the coefficients C and D are determined, we obtain

, ,, Aa sin (w/a)(l — x0) ■ oj , c A ,

u(x,t) = uL~---■——-——sin—xcoscut for 0 5S x -й x0
kcd sin {at I a) l a
Aa sin (а>/я)*о • cd ,, \ , r ^ ^ ,
= и2 = —-:-sin—(l — x) cos wt for x0 ^ x й I ■
kw sin {w/a)l a

The solution for f(t) = A smwt is written similarly.

Thus the problem for the case f(t) = A cos cut and F(f) = Asinwf is
solved. Now if f(t) is a periodic function, say

f{t) — -77 + 2 (a, cos wnt + j}n sin wnt),

where w is the lowest frequency, then obviously we have18

, ^ 1 [а0х[л x0\ , ^ й sin (wn/a)(l — x0) wnx

U(x, t) = u 1 = -tT — 1 ~—r ) + I -——r,-sin-
k [ 2 V / / n=i aw sin (wnla)l a
x (a„ cos аші + pn sin wnt) , 0 5S x ^ x0
_ J f а0х0 Л _ _£_\ у д sin (wnla)x0 .
wn(l — x) (--3.83)
г<2 2 \ / / n i wn sm (wn/a)l S1U a
X (a, cos «и/ + Bn sin cow() , x0 ^ x ^ l .

If by contrast f(t) is a nonperiodic function then in a corresponding

18 The first summand of these sums corresponds to the stationary deflection which is

determined by the force f(t) = ao/2 = const, through the functions

и = ui(x, t) = Wi(x) = -j£- ж(і — 4) f°r 0 <; x й Xo

= Мг(х, f) = u2(x) = — xo^l —j ) for Xo й x й I

2-3. SEPARATION OF VARIABLES 93

manner, in which we represent f(t) by a Fourier integral, we obtain the

solution in integral form.
The resonance phenomena which can occur in the process described by
the function (2-3.83) are easy to overlook. Resonance occurs when a denomi¬
nator in these functions is zero:

ami A 7:m
sin-= 0 , wn = ——a - a)m ,
a l
i.e., if the frequency spectrum of the driving force contains one of the
characteristic frequencies of the vibration.
If one of the nodes of the standing wave is found at the point x0 of
action of the force which corresponds to the free vibration with frequency
a>„, then

sin x0 — 0 , sin -^-(Z — x0) = 0 .

a a

Here the numerators of the corresponding summands for и are equal to zero,
and no resonance occurs. However, if a displacement of the corresponding
standing wave of frequency a>m is found at the point of action of the force
which acts with the frequency a>m, then

sin-x0 — 1 ,
a

and the resonance phenomena is pronounced.

From the above calculations, we can formulate the following rule: In
order to stimulate to resonance a string which remains under the influence of
a concentrated acting force, it is necessary that the frequency a> be equal to
one of the characteristic frequencies of the string and that the point of action
of the force coincide with one of the maxima of the corresponding standing
waves.

9. A general scheme for the method of separation of variables

The method of separation of variables can be used not only for the wave
equation of a homogeneous string, utt = a2uxx, but also for the wave equation
of the inhomogeneous string,

L[u] = - Q(x)u = p(x)-, (2-3.84)

where k, q, and p depend on x and are positive (k > 0, p > 0, q > 0).'9
For example, consider the first boundary-value problem for Eq. 2-3.84:

u{0, t) = 0 , u(l, t) = 0 (2-3.85)

u(x, 0) = (p(x) , ut(x, 0) = ф{х) . (2-3.86)

We shall seek the solution of this problem by the method of separation of

19 The case that k{x) vanishes at some point must be considered separately.
94 HYPERBOLIC DIFFERENTIAL EQUATIONS

variables. Therefore, in a search for specific solutions as we did earlier, we

shall first turn to the following auxiliary problems.
Find a nontrivial solution of Eq. (2-3.84) which satisfies the boundary
conditions

u(0, t) = 0 , u(l, t) — 0
and is representable as a product

u(x, t) = X(x)T(t).
If we substitute this product expression into the partial differential equa¬
tion, we obtain, with consideration of the boundary conditions for X(x) and
T(t), the two ordinary differential equations

4~\тЩ
ax L ax J
-qx + ipx = о
T" + ЛТ = 0 .
For the determination of the function X(x) we are led to the following
eigenvalue problem:
Find those values of the parameter X for which the boundary-value problem

L[X] + XPX = 0 (2-3.87)

X(0) = 0 , X(l> = 0 (2-3.88)

possesses nontrivial solutions, the so-called eigenvalues of the boundary value

problem, as well as the corresponding solutions—the so-called eigenfunctions
of the boundary-value problem.20
Next we shall formulate the basic properties of the eigenfunctions and
the eigenvalues of the boundary-value problems (2-3.87) and (2-3.88) which are
necessary for the following investigations.
1. There exists a denumerably infinite number of eigenvalues < X2 <
• • • < Xn < • • •, which correspond to the nontrivial solutions of the
boundary-value problem, the so-called eigenfunctions Xx{x), X2(x),- • •,
Xn(x)
2. For all the eigenvalues Xn are positive.
3. The eigenfunctions in the interval 0 ^ x ^ l are orthogonal to each
other with respect to the density function p(x):

^ Xm{x)Xn{x)p(x)dx = 0 , m Ф n . (2-3.89)

4. (Development theorem of V. A. Steklov.) An arbitrary function F(x)

20 For p = p0 = const., к = ко = const, we obtain the boundary-value problem for the

characteristic vibrations of a string bounded on both sides:

X" + pX = 0 , p=-^X
fc 0
X(0) = 0 , X(l) = 0,
which we have already investigated.
2-3. SEPARATION OF VARIABLES 95

which is twice continuously differentiable and satisfies the boundary condi¬

tions F(0) = F(l) = 0 can be developed in a uniformly and absolutely conver¬
gent series with respect to the eigenfunctions Xn(x):

F(x) = f FnXn(x),
»=>
Fn = -i- Y F(x)Xn(x)p(x)dx
iVnJo

Nn = j Xl(x)p(x)dx . (2-3.90)

The proof for propositions 1 and 4 are usually given in the theory of
integral equations. We shall limit ourselves to the proof of propositions 2
and 3.
Before we prove these properties we shall first derive Green’s formula.
Let u(x) and v(x) be two functions which are twice differentiable in the inter¬
val a < x < b and possess continuous first derivatives in a S x й b. Next we
consider the expression

uL[v] — vL(u) — ufkv')' — v(ku')' = [k(uv' — vu')\ .

If we integrate this equation with respect to x from a to b, we obtain Green’s

formula
b b

[uL[v] — vL[u])dx — k(uv' — vu') (2-3.91)

a

Proof of the third proposition. Let Xm(x) and Xn(x) be two eigenfunctions
with the corresponding eigenvalues 2m and Xn. Then if и = Xn{x), v = Xn(x)
are inserted in the formula (2-3.91) we obtain, by consideration of the bound¬
ary conditions (2-3.88),21

{XmL[Xn) —XnL[Xm]}dx = 0 , a = 0, b =1,

from which because of Eq. (2-3.87) we obtain

(^n Xm{x)Xn{x)p{x)dx — 0 .

Therefore provided that 2„ Ф Xm, we obtain the following relation

11 Xm(x)Xn(x)p(x)dx = 0 , (2-3.92)

i.e., the eigenfunctions Xm(x) and Xn(x) are orthogonal to each other with
respect to the density function p(x). We shall now prove that every eigen-

21 The derivatives X^(x) and X„(x) are continuous throughout 0 ^ x й l, including

the points x — 0 and x = l. Eq. (2-3.87) gives

k{x)Xn(x) — j \q — Xmp)Xmdx + C .

From this follows in particular the existence of the derivative X^(x) for x = 0 and x =1.
96 HYPERBOLIC DIFFERENTIAL EQUATIONS

value to within a constant factor corresponds to only one eigenfunction.22

Thus it follows that each eigenfunction is uniquely defined as a solution of
a differential equation of the second order by the value of the function itself
and by its first derivative at x = 0. Let us assume the existence of two
functions X and X which correspond to one and the same X and vanish at
x = 0, and consider the function
X*(x) = Ш^-Х{х).
X'(0)
Then we see that this function satisfies the same equation of the second order
(2-3.87) and the same initial conditions as the function X(x):

X*(0) = 0
_X'(0)

"40) =i^-F(0) = z'(0).

dx X (0)
Therefore, we have shown that X*{x) — X{x) and that

X(x) = AX(x), А=ЦМ-

x'(0)
is valid.
Let it be noted that for the proof, the condition X'(0) ф 0 was used.
This, however, is not necessarily satisfied, since the solution of the linear
equation (2-3.87) determined by the initial conditions

Z(0) = 0 , Z'(0) = 0
is identically zero and, therefore, no eigenfunction can exist.
If Xn(x) is an eigenfunction corresponding to the eigenvalue Xn, then
AnXn(x), where An is an arbitrary constant, because of the linearity and the
homogeneity of the equation and the boundary conditions, is likewise an
eigenfunction corresponding to Xn.
We proved earlier that the class of eigenfunctions has been completely
exhausted. The eigenfunctions which differ from each other only by con¬
stant factors will appear as not essentially different. However, in order to
avoid ambiguity in the selection of factors the eigenfunctions will be normal¬
ized by

Nn = j Xl(x)p(x)dx = 1 .

If, at the outset, an arbitrary eigenfunction Xn(x) does not satisfy this nor¬
malizing condition, then it can be normalized by multiplication by a suitable
coefficient An:

22 This property of the first boundary-value problem therefore depends on the fact
that two linearly independent solutions of a differential equation of the second order
cannot be at one and the same point equal to 0. This assertion is based on the bound¬
ary-value problem with homogeneous boundary conditions. With other boundary con¬
ditions (for example, with X(0) = X(l), X'(0) = X'(l)) two different characteristic func¬
tions can be given which correspond to one and the same characteristic value.
 2-3. SEPARATION OF VARIABLES 97

AnXn(x) = Xn(x) , An
l a
Xn{x) p(x)dx

Consequently, the eigenfunctions Xn(x) of our boundary-value problems

(2-3.87), (2-3.88), form a normalized orthogonal system (orthonormal system)
•i
m Ф n
I Xm(x)X„(x)p(x)dx = ’
m = n

Proof of the second proposition. Let Xn(x) be the eigenfunction corresponding

to the eigenvalue Xn. Then,

L[Xn] = — inp(x)Xn{x).
Multiplication of both sides of this equation with Xn{x) and integration with
respect to x from 0 to l yields

Xl{x)p{x)dx = — j Xn(x)L[Xn\dx

or

Jn = — ( Xn-^7-\k(x)^^-~\dx + i q{x)Xl{x)dx ,
Jo dx L dx J Jo

since the function Xn(x) was assumed to be normalized. We then obtain by

partial integration and by use of the boundary conditions (2-3.88)

1 Г! Г1
^n — nk^Cn + \ k(x)[Xn(x)fdx + j q(x)Xl(x)dx

k(x)[X'n(x)fdx + 1 q(x)Xl(x)dx , (2-3.93)

from which it follows that

in > 0
since by assumption k(x) > 0 and q{x) ^ 0.
We soon arrive at the calculation of the development coefficients of
F(x), but not at the proof of the development theorem itself, which we
shall use in this case. Obviously,

( p(x)F(x)Xn(x)dx
Fn = -^гт-. (2-3.94)
j p(x)Xn(x)dx

Now if we multiply both sides of the equation

F(x) =f n~ 1
FnXn(x)

by p(x)X„(x), integrate this expression with respect to x from 0 to /, and

bear in mind the orthogonality of the eigenfunctions, we obtain directly the
expression shown above for Fn (the Fourier coefficients).
98 HYPERBOLIC DIFFERENTIAL EQUATIONS

We turn again to the partial differential Eq. (2-3.84). For T(t) we obtain
the equation

T" + XnT = 0 (2-3.95)

without any auxiliary conditions. Since Xn is positive, its solution has the
form

Tn{t) = An cos i/Hnt + Bn sin \/AJ ,

where An and Bn are arbitrary coefficients. Our auxiliary problem therefore
has infinitely many solutions of the form

Unix, t) = Tn(t)Xn(x) = [AnCosV^nt + BnSin V*nt)Xn{x).

The solution of the problem with prescribed initial conditions, from which
we started at the beginning of this section shall be introduced in the form

u{x, t) — 2 (AncosVTnt + Bn sin \ZTj)Xn{x) . (2-3.96)

71 — 1

The formal scheme for satisfying the initial conditions (2-3.86) depends on
the Steklov theorem (4) above. It is completely analogous for the homo¬
geneous string, and we find from the equations

u(x, 0) = <p(x) = f AnXn{x), ut(x, 0) = <p(x) = 2 BnV2„Xn(x)

n—1 тг — 1

the relation

An = <pn, Bn=^ . (2-3.97)

Hence (pn and фп are Fourier coefficients of the functions (p{x) and ф(х) which
occur in the development of these functions with respect to the orthogonal
system of functions [Xn{x)} with the weight factor p(x).
Since we have limited ourselves here to the general scheme of separation
of variables, the conditions for the applicability of this method, with regard
to both the coefficient of the equation and the initial functions, have not been
discussed.
The basic ideas of this method are due to V. A. Steklov.23

Problems

1. Determine the function u(x, t) which describes the vibrations of a string

(0, l) fixed at the ends, if the string is excited in such a way that at the
point x = c it has been displaced to a magnitude equal to h from the original
position (see Figure 24), i.e., let u(c, 0) = h.
2. A string fixed at both ends has been displaced at the point x = c by a
force F0. Determine the string vibrations if the force at the initial moment
ceases to act.

23 Report of the Kharkov Mathematical Society, second series, Vol. 5, nos. 1 and 2
(1896); also Fundamental Problems of Mathematical Physics, Vol. 1 (1922).
 2-3. SEPARATION OF VARIABLES 99

3. Determine the function u(x, t) which

describes the vibrations of a string (0, /)
fixed at both ends if the string is excited
by an impulse К distributed in the interval
(с — д, c + d) for the case when the dis¬
x=0 tribution is (a) uniform, (b) subject to the
law v0 cos (x — c)/2S.
FIG. 24. 4. Determine the function u(x, t) which
describes the vibrations of a string fixed
at both ends if the excitation results from impulse К acting at point
x — c.
5. Prove the additivity of the energies of the individual harmonic vibrations
for a vibration process with the boundary conditions и = 0, ux = 0. Consider
also the case of the boundary condition ux + hu = 0 (all of the series occurring
are assumed to be uniformly convergent). Calculate the energies of the in¬
dividual harmonic vibrations in Problems 1, 2, 3, and 4.
6. A spring which is fastened on one side at the point x — 0 is stretched
by a weight of mass M fastened at the point x = l. Determine the spring
vibration if the weight is removed at time t = 0 and subsequently no force
occurs again at x = l.
7. Let a rod be fastened at one end; at the other end acts a force F0. What
vibrations does the rod perform if the force ceases to act at the initial
moment?
8. What vibrations are performed by a spring which is fastened at one end
while at the other end a weight of mass M hangs at the initial moment?
Let the initial conditions be homogeneous.
9. A mass M is fastened at the point x — c on a homogeneous string with
fixed ends x — 0 and x = l. Determine the displacement u(x, t) of the string
if (a) the string at the initial moment moves from its equilibrium state at
x — c to a value equal to h without losing its initial velocity; and (b) the
initial displacement and the initial velocity are equal to zero.
10. Determine the course of vibration of a spring with free ends with uni¬
form initial expansion (give a model of this problem).
11. Describe the vibrations of a spring which is fastened elastically at both
ends as in Problem 10, if the initial conditions are arbitrary. The solution
should be investigated for small h (“soft” attachment) and for a larger h
(“rigid” attachment), and the corresponding correction to the eigenvalues of
the spring with fixed and free ends should be calculated.
12 . Find the displacement u(x, t) of a string whose ends are rigidly fixed if
the vibrations occur in a medium in which the resistance is proportional to
the velocity. Let the initial conditions be arbitrary.
Solution:

utt = a2uxx — 2vut , v > 0

u(x, t) = e v< 2 (fln cos wnt + bn sin w„t) sin y7Xnx ,

n=l
100 HYPERBOLIC DIFFERENTIAL EQUATIONS

bn — v^--\--p—
<t>n
( <p(x) s\n~[/InXdx
Wn Jo

a>„ = л/аг?.п — v .

.
13 Let an isolated electrical conductor of length / with characteristic co¬
efficients L, R, C, and G = 0 be charged to a fixed potential v0. One end of
the conductor is grounded at the initial moment while the other remains
isolated during the entire time of the process. Determine the voltage dis¬
tribution in the conductor.
Solution.-.

v(x, t) — e iR/U)t f an sin 2?го-|—■ xx • sin (wnt P <pn) ,

71—0 Zl

(2 n -P l)n CR42
°>n ~ Я1
2lVLC^ Ln (2n + If
4y0 . о L
anП г. I 1 \ • > фп 2<J)n .
n(2n + 1) sin (pn R
14 . A string which is rigidly fastened at the ends vibrates under the in¬
fluence of a harmonic force which is distributed with density f(x, t) — Ф(х)
sin wt. Determine the displacement u{x, t) of the string with arbitrary initial
conditions. When is resonance possible and what solution results in the case
of resonance?
15 . Solve Problem 14 under the assumption that the vibrations occur in a
medium in which the resistance is proportional to the velocity. Find the
vibrations which comprise the principal part of the solution for / —» °o.
16 . An elastic rod of length l is in a vertical position and at its upper end
a free-falling elevator is fastened, which at the moment it is stopped, has
attained a velocity v0. What vibrations are performed by the rod if the
other end is freely movable?
17. Solve the equation

Uu = a uXz -p b и -p A
with homogeneous initial conditions and the boundary conditions

u(0, t) — 0 u(l, t) = В ,
where b, A, and В are constant.
.
18 Solve the differential equation

utt = cfuxx -P A sinh x

with homogeneous initial conditions and the boundary conditions

u(0, t) — В u(l, t) — C ,
where A, B, and C are constant.
.
19 On a homogeneous string with rigidly fixed ends x = 0 and x = l acts a
harmonic force
2-4. PROBLEMS WITH AUXILIARY CONDITIONS ON THE CHARACTERISTICS 101

F{t) = P0 sin cut,

which starts acting at t — 0 at the point x = c (0 < c < l). Determine the
displacement u(x, t) of the string with homogeneous initial conditions.
20. Describe the vibrations of an inhomogeneous rod of length / with rigidly
fixed ends, which consists of two homogeneous rods connected at the point
x — c (0 < c < l), when the initial displacement has the form

u(x, 0) - —X for 0 5S x ^ c ,
c

= ~~—(l — x) for C-^XSl

l—c
and the initial velocity is equal to zero.
21. Determine the vibrations of a spring which is fixed at one end while at
the other acts a force given by

F(t) = A sin coj + В sin w2t .

22. Describe the vibrations of an inhomogeneous rod which consists of two
homogeneous rods connected at the point x — c when one end of the rod is
fixed and the other moves according to the law

u{l, t) = A sin wt .

2-4. PROBLEMS WITH AUXILIARY CONDITIONS ON THE CHARAC¬

TERISTICS

1. Statement of the problem

In the following discussion, we shall consider a series of problems which

are extensions of the first boundary-value problem for the equation of a
vibrating string. For simplicity we shall investigate phenomena only in the
neighborhood of boundary points where the other boundary point moves
toward infinity, i.e., we proceed from the problem for the semi-infinite line.
The equation of the vibrating string ut'f — a2uxx is symmetrical with
respect to л: and t if a2 = 1, i.e., if the unit of time is changed by the in¬
troduction of the variable t — at'. Of course, the auxiliary conditions in¬
troduce an asymmetry in the mathematical interpretation of x and t; in the
initial conditions (for t — 0) two functions u(x, 0) and ut(x, 0) are prescribed,
while in the boundary conditions (at x = 0) only one function u(0, t) is pre¬
scribed.
As noted in 2-2§7, between the derivatives of the displacement function
at t = 0 and x = 0 the relation

ut(0, z) + ux{0, z) = ut(z, 0) + ux(z, 0) , a2 = 1 ,

exists for an arbitrary z. Hence it follows that these functions at x = 0 and
t = 0 cannot be prescribed independently of each other; only three relations
can be prescribed arbitrarily, which proves that the auxiliary conditions can¬
not be arranged symmetrically.
102 HYPERBOLIC DIFFERENTIAL EQUATIONS

The auxiliary conditions can be prescribed either on the straight lines

x = 0, / = 0 (as in the problems considered hitherto), or on known curves in
the phase plane. For example, if the boundary values are given on curve
Ci (x = /i(/)), then this curve, in order to guarantee the solvability of the
problem considered, must still satisfy the determined auxiliary conditions in
addition to certain continuity and differentiability conditions.
We shall consider the vibration process of a gas in a tube in which a
piston moves. The velocity of the piston whose law of motion is given by
x = fi(t) cannot be prescribed arbitrarily: it must not exceed the velocity of
sound a, [dfi(t)/dt < a]. Geometrically it follows that the curve Ct (x —
from the straight line t — 0, which supports the initial value, must be separated
from the characteristic x — at (Figure 25). Also if only one point of Ci were
to lie below the characteristic x = at,
then the value of the function u(x,t)
would be completely determined by
the initial value and could not be
arbitrarily prescribed. The physical
significance of these relations is as
follows: if the motion of a gas has
* a velocity which is larger than that
of the velocity of sound, the equa¬
tions of acoustics lose their sense
and must be replaced by the non¬
linear equations of gas dynamics.24
The initial conditions can be
given on the straight line t = 0 and
also on a curve C2 (t = /2(x)), which
satisfies the inequality | /2(x) | < 1/й. such problems can be easily solved with
the aid of the integral vibration equation (see Section 2-2§7).
Without going into a complete review of all the possible boundary-value
problems, we shall now deal more precisely with the problem in which the
auxiliary conditions are prescribed on the characteristics. This problem is
known as the Goursat problem. Such problems are of extraordinary interest
from the standpoint of the physical applications. They arise in problems of
adsorption and absorption of gases, for example, (see Section 2-6 §5) with
evaporation processes (see Problem 1, page 106) and many other problems.

2. The method of successive approximation

We shall first consider the simplest of the problems in which the auxiliary
conditions are prescribed on the characteristics. It reads as follows:

uxy = f(x, y)
u(x, 0) = <pi{x) (2-4.1)
w(0, y) = <p2(y) .

u See Application, Section 2-6, 4.

2-4. PROBLEMS WITH AUXILIARY CONDITIONS ON THE CHARACTERISTICS 103

The auxiliary conditions here are given on the straight lines x = 0 and v = 0,
which represent the characteristics of Eq. (2-4.1). We assume the functions
<Pi(x) and (p2(y) to be differentiable and related to each other by the relation¬
ship <р2(0) = <p2(0). If we integrate Eq. (2-4.1) with respect to x and then with
respect to y, we obtain

uy(x, y) = uy(0, y) + J /(£, y)dZ

u(x, у) = u(x, 0) + u(0, у) - u(0, 0) + I j /(f, rj)dZ

or

u{x, y) = <Pi(x) + (p2(y) — Pl(0) + f [ f(Z, r;)d£dr]. (2-4.2)

J0 J0

Therefore, the solution in this case, in which the differential equation contains
neither the first derivatives nor the sought function itself, can be represented
explicitly by the analytic expression (2-4.2). From (2-4.2) the uniqueness
and the existence of the solution of our problem follows directly.
We shall now seek the solution of the linear hyperbolic differential equation

uxy = a(x, y)ux + b(x, y) uy + c(x, y)u + f(x, y), (2-4.3)

with the auxiliary conditions given on the characteristics x — 0, у = 0

u{x, 0) = <pi(x) , u(0, y) = <p2(y),

where <р2(х) and y>2{y) satisfy the above cited conditions, differentiability and
y>,(0) = (p2{0). Let the coefficients a, b, and c be continuous functions of x and y.
From the differential Eq. (2-4.3) we know that the function u(x,y) satis¬
fies the integro-differential equation

u(x,y) - j I И6,7])U£ + bit, Г])иѵ + c(Z, T])u\d£dr]

+ yM + <p2(y) — ^i(0) +[ [ f(Z, rj)dZdrj . (2-4.4)

Jo JO

We shall determine its solution with the help of the method of successive
approximations. For the zero-th approximation we choose the function

U0(x, y) — 0 .
Eq. (2-4.4) then gives for the successive approximations the expression

иi(x, y) = <fi(x) = <p2{y) - <pi(0) + f \ f(Z, r;)dsdri

J0 J0

ltn(x, y) = Ui(x, y) (2-4.5)

diin-i
+ a{Z, v)~ + b(Z, + c(Z
dZ 3ry

Incidentally we note the validity of the expressions

104 HYPERBOLIC DIFFERENTIAL EQUATIONS

dun dlli
= + [ \a{x, rj)—in -1- + b(x, г})дНп 1 + c(x, r])un-Adr]
dx M JoL OX Of] J
(2-4.6)
ou
У) "-1 + c(£
ОУ Оу JoL Oi; dy

In order to demonstrate the uniform convergence of the following sequences

{Un{x,y)} , {ІМ, {^>}.

we shall consider the differences

Zn(x, y) = un+1(x, y) — Unix, y)

v) dzZ.~ dtdr; ,
drj + c(t> r])zn-i(^> 7)1
dznjx, y) _ dun+lix, y) dun(x, y)
dx dx dx

a(x,v) dZgx± + b(x’ 7) 1 + c{x, rjjzn-iix, rji^dr) ,

dzn(x, y) _ dun+i{x,y) _ dunix, y)

dy dy dy
dzn-i
a{£, y) b{£, y) dz*~l + c($, y)zn-1($, y)J
d$
Let M be an upper bound for the absolute values of the coefficients a{x,y),
b(x, y), c(x, y), and H an upper bound for the absolute values of the function
z0(x, y) and its partial derivatives
dz0 dz0
I 20 I < H , <H, <H,
dx dy
where л: and у vary in a square (0 ^ x 5S L, 0 ^ у ^ L). We shall now esti¬
mate the functions zn, dzjdx, dzjdy according to the above. First, the in¬
equalities are valid;
■(x +у)2
I 2,1 < 3HMxy < ЗЯМ-
21
dzi
< ЪНМу < 3HM(x + y)
dx
dzi
< ЪНМх < 3HM(x + y) ,
dy
and the recursion relations

2n I < 3HMnKn~l {x + y-)n+1

in + 1)!
dzn
dx
< 3 нмпк п~1 (x n\
+ -v)n

dzn
< 3HMnKn~l (x + y)n-
dy n\
2-4. PROBLEMS WITH AUXILIARY CONDITIONS ON THE CHARACTERISTICS 105

where К > 0 is a constant number whose value will be specified later. With
these estimates and the formula for the (n + l)th approximation we obtain
after a series of simplifications

n+ijfn-i [x_ + yY^_( x + у

Zn+i I < тмп+1к +2 <
(n + 2)! \n + 3

< 3нмп+1кп (x + -уГ+2 <r 3H VKLM^

(я + 2)! Я2М (я + 2)!
X + у
n+1
dZn+i
< 3HMn+1Kn~l{x + y) +2 <
dx in + 1)! \n + 2

< 3нмп+1кп {x + У)П+1 < зя (2KLM^n+l

(n + 1)! К (n+1)!
n+1
dzn+l (x + y) X + у
К
n+1 rsn-1
< 3HM +2 <
dy (n + 1)! \n + 2
и+1„п (x + ѵГ1 , ЗЯ (2KLM)n+1
< 3HMn+1K <
in + 1)! К in + 1)!
where

К = 2L + 2 .
On the right sides of these inequalities we have, to within constant factors,
the general term in the series development of e2KLM. Thus the above estimates
show that the sequences of functions

Un = Uo + Z\ + • • • + Zn-1

dun __ du„ dzi__ dzn-i

dx dx dx dx
dun du0 dzi dzn-i
dy ~ dy + dy dy

converge uniformly. We shall denote their limit functions by

u(x, y) = lim un{x, y) ,

П-’Оо

v(x, y) = lim -^-(x, y) ,

n -oo OX

tv(x, y) = lim Щ^-{х,у) .

n OO Оу

By passage to the limit under the integral signs in (2-4.5) and (2-4.6) we then
find

u(x, y) = udx, y) + [ [ [a(f, rj)v + 6(£, rj)iv + c(f, rj)u]d£dy ,

J0 J0

v(x, v) = ^-(x, y) + [ [a{x, rj)v + b(x, jj)w + c(x, rj)u)drj , (2-4.7)

ox Jo

w{x, y) = ~~ix, y) + [ [c(f, y)v + b(S, y)w + c(£, у)иЩ .

dy ' Jo
106 HYPERBOLIC DIFFERENTIAL EQUATIONS

Hence the following equations

v = ux w — uy
establish the fact that u(x, y) satisfies the integrodifferential equation

u(x, у) = (pM + <p2(y) — <pi(0) + J J j (6, r))dtdv

+ [ [ [a(f, >i)u( + b(£, rj)ur, + c(£, r])u]dtdr) (2-4.4)

J0 J0

and also the original differential Eq. (2-4.3). We verify this directly by dif¬
ferentiation of (2-4.4) with respect to x and y. Moreover, as one easily sees,
u(x, y) satisfies the auxiliary conditions.
The uniqueness of the solution of the stated problem will now be de¬
monstrated. If we assume the existence of two solutions и{(х, у) and u2(x, y),
we obtain for their difference

U(x, y) - uAx, y) — u2{x, y)

the homogeneous integrodifferential equation

U{x,y)—\ [ (aUx + bUy + cU)dl;dy .

J0 J0

Now if Hi is a common upper bound of the magnitudes \U\, \UX \ and | Uy\
for 0 S x S L and 0 ^ у ^ L, then by a repetition of the estimates which led
to those on the function zn(x, y), the correctness of the inequality

(x + ѵГ2 3 Hi (2 KLM)n+*
I U\ < 3tf,Mn+1/r <
(n + 2)! K2M (n + 2)!
for an arbitrary value of n follows easily. Hence it follows that

U(x,y) = 0 or Ui(x, y) = u2(x, y),

whereby the uniqueness of our solution is shown.
If the coefficients a, b, and c are constant, then Eq. (2-4.3) with the help
of the substitution

и = ve^y,
can be brought to the form

vxy + CiV = j . (2-4.8)

For C| = 0 we obtain the simple Eq. (2-4.1) whose solution was given by
formula 2-4.2.
By contrast, if Ci Ф 0 then we obtain for the solution of (2-4.8) an explicit
analytic representation by the method given in Section 2-5.

Problems

1. Air with a velocity v passes through a tube (x > 0) which is filled with a
moist fluid. Let v(x, t) be the concentration of the moisture in the saturated
2-5. SOLUTION OF GENERAL LINEAR HYPERBOLIC DIFFERENTIAL EQUATIONS 107

substance and u(x, t) be the concentration of the free vapor. What equations
are satisfied by the functions u(x, t) and v(x, t) describing the drying process
when (a) the process proceeds isothermally; and (b) the dry isotherm has the
form и = у ■ v where у is the isothermal constant (see also Section 2-6, Applica¬
tion 5)?
2. Hot water with velocity v flows through a tube (x > 0). Let и be the
temperature of the water in the tube, v the temperature of the walls of the
tube, and u0 the temperature of the environment. Derive the equations for
и and v by neglecting the temperature distribution in the individual cross
sections of the tube and the walls. Assume further that a temperature
gradient exists at the boundaries of the water-tube wall and the wall environ¬
ment, and that the flow of heat follows Newton’s law (see Chapter 3-1).

2-5. SOLUTION OF GENERAL LINEAR HYPERBOLIC DIFFERENTIAL

EQUATIONS

1. Adjoint differential operators

In order to represent the solutions of boundary-value problems in integral

form we need some auxiliary formulas. Let

L[u\ = uxx — uyy + a{x, y)ux + b(x, y)uy -I- c(x, y)u , (2-5.1)

where a(x, y), b(x, y), c(x, y) are differentiable functions, be a linear differential
operator corresponding to a linear hyperbolic differential equation. We
multiply L[u\ by a function v and write the individual summands in the form

vuxx — (vux)x — (vxu)x + uvxx

VUyy = (VUy)y — (VyU)y + UVyy

vaux — (avu)x — u(av)x

vbuy = (bvu)y — u(bv)y
vcu = ucv .
By summation of the individual summands we obtain

rr i wr i 3K (2-5.2)
vL[u] — uM[v] = —— + —
dx oy

where

M[v] — vxx — vyy — (av)x — (bv)y + cv (2-5.3)

H = vux — vxu + avu = (vu)x — (2vx — av)u (2-5.4)

= — (vu)x + (2 ux + au)v (2-5.4')

К =— vuy + vu + bvu =— (vu)y + (2vy + bv)u (2-5.5)

= (uv)y — (2uy — bu)v . (2-5.5')

Two differential operators L and M are said to be adjoint to each other

108 HYPERBOLIC DIFFERENTIAL EQUATIONS

if the difference
vL[u] — uM[v]
is equal to the sum of the partial derivatives of the expressions H and К
with respect to x and y.
In particular, if L[u\ = M\u\, then the operator L is called self-adjoint.
For the double integral, the difference vL[u\ — uM[v\, where и and v are
twice differentiable functions throughout a region G which is bounded by a
piecewise smooth curve C, we have

[ \{vL[u\ - uM[v])d£drj = [ (Hdrj - Kd£) (2-5.6)

Jo J Jc
(Green’s formula).

2. Integral form of the solution

We shall apply Formula (2-5.6) for the solution of the following problem.
Find a solution of a linear hyperbolic differential equation

L[u] = uxx — Uyy + a(x, y)ux + b(x, y)uy + c{x, y)u — — f{x, y) , (2-5.7)

which satisfies the initial conditions

u\a = <p{x) , un\c - <p{x)

on a curve C (un is the derivative in the direction of the normal to the curve C).
Obviously the operators L and M are adjoint to each other.
Ffence let C be given by
У = fix) ,
where fix) is a differentiable function. On
the curve C we impose the requirement that
every pair of characteristics у — x = const,
and у + x = const, intersects curve C at
most once (therefore it is necessary that
I f\x) \ 5S 1). Formula (2-5.6) then yields for
the curvilinear triangle MPQ which is
bounded by the arc PQ of C and the seg¬
ments of the characteristics MP and MQ
(Figure 26).

1 {vL[u] — uM\v])d£dr)
M J PQ

= Г(Hdv - Kd£) + \P(Hdrj - Ш) + \Q(Hdr) - Ш).

JQ Jjf JP

We perform the first two integrals which are taken along the characteristics
MQ and MP, and by taking into consideration the relations
ds
dt; =— dr] on the characteristic QM
l/2
2-5. SOLUTION OF GENERAL LINEAR HYPERBOLIC DIFFERENTIAL EQUATIONS 109

ds
d£ = drj on the characteristic MP
l/2

(where ds is the element of the arc along QM and MP), and the formulas
(2-5.4) and (2-5.5), we obtain

dv a +b \ j
(Hdrj — Kd$) = — \ d(uv) -f
ds 17Tvrs

-—{uv)M + (uv)q + j (2-^22- — ^-—-vjuds

and correspondingly

j (Hdrj — Kd£) = — (uv)M + {uv)P + j ^2-^- — vjuds .

From this and from (2-5.6) it follows that

{uv)p + (uv)q (l dv b —a \ j , [M( dv a + b \ ,

(uv)M = + u
br wr+Ц1Г-wr
-

+ -Lf ill dr- — Kdt)—Lf f (vL[u] — uM\v])d£d*j. (2-5.8)

£ )P £ J m) pq

This relation is an identity which is correct for arbitrary but sufficiently

smooth functions и and v.
Now let и be a solution of the problem with above stated initial condi¬
tions, while v depends on the point M as a parameter and fulfills the require¬
ments

M[v] = Ѵн — Vr,r, — (av) f — (bv)v + cv = 0 (2-5.9)

in the interior of dMPQ
and
dv _ b — a
on the characteristic MP
ds ~ 2V2V
dv b + a (2-5.9')
on the characteristic MQ y
IT 2г/2
v(M) = 1 .
From the conditions which must be fulfilled on the characteristics and the
last condition we then find

(Ь-а)/(»ѵТ) is

v = e *° on MP
C (6 + o)/(2V2) ds

v = e s° on MQ

where s0 denotes the value of s at the point M. As we saw in Section 2-4,

Eq. (2-5.9) and the value of v on the characteristics MP and MG define the
function v completely in the region MPQ. The function v is ordinarily called
the Riemann function.
Consequently formula (2-5.8) for a function и which satisfies Eq. (2-5.7)
по HYPERBOLIC DIFFERENTIAL EQUATIONS

takes the following definitive form:

u(M) = (uv)>‘ + (uv)t}

2
1
+ — 1 [viu^drj + Ur,d£) — u(v(dr] + Vr,d£) + uv(adrj — bd c)]

+M( v(M, M')f(M')daM’ , doM. = dtdv . (2-5.10)

^ J JMPQ

The stated problem is solved by this formula since the expression which
stands under the integral extended along PQ contains only functions which
are known along C. The function v has already been defined above while
the functions

и \c = <p{x)
ip'(x) - <P(x)f'(x)V\ + fn(x)
ux — us cos (x, s) + un cos (x, n) =
1 +/'*(*)
cp,(x)f'(x) + ф(х)Ѵ l+f'\x)
Uy — Us cos (y, S) + Un cos (y, n) =
1 +f'\x)
can be calculated from the initial conditions.
If the initial conditions along the arc PQ are known, the function и in
the triangle PMQ can be completely determined according to formula (2-5.10)
when f{x, y) is known in this region.25
Formula (2-5.10), obtained under the assumption of the existence of a
solution, defines the solution from the initial conditions and the right side of
Eq. (2-5.7). Most important, it also proves the uniqueness of the solution
(compare it with the D’Alembert formula, Section 2-2).
It can also be shown that the function и defined by (2-5.10) satisfies the
conditions of the problems.

3. The physical interpretation of the Riemann function

In order to clarify the physical meaning of the Riemann function v(M, M')
we shall first determine the solution of the inhomogeneous equation

L[u\ ——2fy, /=2/,

with homogeneous initial conditions on a curve C. We see from (2-5.10) that

the desired solution has the form

u{M)=\[ v(M,M')UM')doU' (2-5.11)

J Jifee

If we now assume that except in a small neighborhood S? of the point

Mi, is everywhere equal to zero and satisfies the normalization condition

25 If the characteristics intersect the curve C at the two points P and Mi (Figure 26),
the value u(M\) cannot be arbitrarily given. On the contrary, it is determined accord¬
ing to formula (2-5.10) by the initial value on the arc PQi and the value of f{x,y) in
APMiQi.
2-5. SOLUTION OF GENERAL LINEAR HYPERBOLIC DIFFERENTIAL EQUATIONS 111

J j/i(MW = l. (2-5.12)

Then the above formula for ue(M) assumes the form

ue(M) = j |y(M, M')f\{M')doM' . (2-5.13)

By use of the mean-value theorem we can write

u<>(M) = v{M, M*) j j UM')daM■ = v(M, Mi*) ,

where M* is a determined point in 5г.

If now the е-neighborhood of the point M{ shrinks (e—>0) then we obtain
u(M) = lim Ut(M) = v(M, Md . (2-5.14)
e -0

The function fx, as we have already seen in a series of examples, represents

a force density whereas у denotes the time. The expression

j j MM')dou> = U Ш, V)dZdrj (2-5.15)

then denotes the impulse of the force. Hence, because of (2-5.11) we conclude
that v(M, Mi) is the influence function of the unit impulse acting at Mx.
The function v(M, Mx) = v(x, y, £,rj) can be defined as a function of the
parameter M{x,y), which with respect to the coordinates of the point
Mi satisfies the equation
M,e.„[y] = 0 (2-5.16)
with the auxiliary conditions (2-5.9').
We shall now consider the function
и = u(M, Mi) ,
which depends on the parameter M,(f, rj) and with respect to the coordinates
x, у of the point M satisfies the equation

L и. у)\м\ — 0 (2-5.17)
with the auxiliary conditions

du b — a
и on the characteristic MxQx ,
8s 2i/2
(2-5.18)
^ on the characteristic МХРХ,
ds 2\/ 2
u(Mi, Mi) = 1 .

From these conditions there results

' (* (Ь-о)/(2^2 ) Ія
e "u on MiQi
u(M, Mi) =
\ ' (b + a)/(2Vs') ds (2-5.19)
\e on MiPi
u(Mi, Mi) = 1
112 HYPERBOLIC DIFFERENTIAL EQUATIONS

In the quadrangle MP1M1Q1, which

is bounded by the intersection of the
characteristics MPX, MQi, MiPx and
MiQi (Figure 27), и is completely deter¬
mined by Eq. (2-5.17) and the conditions
(2-5.18).
If we apply formula (2-5.6) to the
quadrangle MPiMiQi we obtain

{vL[u\ — uL[v])dt;dy

=
pi
{Hdrj - Kd£) +
I .if Je,
+[Qi+ p
J Mi JPi
= 0

If we use formulas (2-5.4) and (2-5.5) for К and H and the conditions (2-5.9'),
the first two integrals on the right side can be easily calculated

{Hdrj — Kd£) =— (uv)M + (uv)Pl ,

(.Hdrj — Kd£) = — (uv)M + (uv)0l ,

just as in the derivation of formula (2-5.10).

In a similar manner we find with the use of equations (2-5.4') and (2-5.5')
as well as conditions (2-5.19)

l“l Гя!
{Hdrj — Kd£) = \ [— (vu)(drj — {uv)vd£\ + j v[{2 u(dr] + 2 uvd£) + (audrj — bud£)]

= J ld(uv) + ^ - ~~u)jvds = (uv)Ml - (uv)Pl ,

ds
{Hdrj — Kd$) = {uv)Ml — (иѵ)ол , d£ = dr] = —y=
ЛГ, V b

By addition of these equations, we obtain

2 (uv)M = 2 (uv)Ml

or

u(M, Mt) = v(M, , (2-5.20)

since

{V)M, = {V)M = 1 •

The influence function of the unit impulse acting at the point Mi can also
be defined as the solution of the equation

L ix, y)[v{M, Mi)\ = 0 ,

with auxiliary conditions (2-5.18).
2-5. SOLUTION OF GENERAL LINEAR HYPERBOLIC DIFFERENTIAL EQUATIONS 113

4. Differential equations with constant coefficients

As the first example of application of formula (2-5.10) we shall consider

the following initial-value problem for the equation of the vibrating string:

f
uyy = uxx + fy{x, t) , у = at, fy = »
a
u{x, 0) = (p(x) ,

Uy - фу{х) , фі=— .
a

As in (2-5.10) a strip of the axis у — 0 occurs as the arc PQ. The operator

L - uxx Uyy

because of
M — L — uxx — Uy,

is self-adjoint.
Because a — 0 and b = 0, v equals one on the characteristics MP and MQ.
Hence it follows

v(M, M') = 1

for every point M' which lies interior to the triangle PMQ.
Now in our case,

dr] = 0 for PQ .

Thus we obtain

u(P) + u(Q)
u(M) — —\ Uvds + -7Г /(6, ѵШѵ
PMQ

If we bear in mind that P = P(x — y, 0), Q = Q{x + y, 0), where x and у

are the coordinates of the point M= M(x,y), we obtain by use of the initial
conditions
x+y У rx+ly—ri)
_ u(x — y) + u(x + y) 1
u(x,y) <!>№dZ + -~ /i(6, rj)d£dr).
> x-y “ J 0 Jx- (y-V)

For the variables x and t, therefore, we arrive at the formula

x + at t Az+a(t-T)
u(x — at) + u(x + at) 1
u(x, t) — + Ф(№ + /(f, r)d£dr,
2a 0 J x—a (t—t )

which we recognize from Section 2-3 §7.

As а second example we shall consider the following initial-value problem
for differential equations with constant coefficients:

uxx — Uyy + aux + buy + cm = 0 (a, b, c are constants) (2-5.21)

m |>=0 = <p{x), (2-5.22)
Uy\y=0 - ф(х). (2-5.23)
The substitution
114 HYPERBOLIC DIFFERENTIAL EQUATIONS

U = ueKl^y (2-5.24)

transforms Eq. (2-5.21) into the simpler form

U„ -U„ + clU = 0 , cl = i-(4c -a2 - b2) (2-5.25)

with the auxiliary conditions

U\y=0 = <p(x)e,a/2u = <pM (2-5.22')

Uy U = (ф(х) - I-ср(х)уа/ш = ФМ (2-5.23')

only when the parameters A and и are suitably selected; that is

i=f, v=-y- (2'5>26)

The determination of the function

U(x,y) from the initial conditions and
Eq. (2-5.25) leads to the construction
of the Riemann function v(x, y; 6, >?).
The function v must satisfy the
conditions (Figure 28)

vxx — vyy + CiV = 0 (2-5.27)

v = 1 on the characteristic MP ,
(2-5.28)
v — 1 on the characteristic MQ .

We chose v in the form

v = v(z) (2-5.29)

with

2 = V (X — f)2 — (y — 7]f

or (2-5.30)
z2 = (x — £)2 — (y — r])2 .
On the characteristics MP and MQ, z is equal to zero so that v(0) = 1.
The left side of (2-5.27) can be brought to the form

Vxx — Vyy + CyV = v"(z){z\ — Z2y) + v'(z)(zxx — Zyy) + CiV = 0 .

By differentiating twice the expression for z (with respect to x and y), we

obtain

zzx = x — $
ZZy = — (y - v)
zzxx + z\ — 1
I 2 1
ZZyy I Zy — -L •

From this and from (2-5.30) it follows

2-5. SOLUTION OF GENERAL LINEAR HYPERBOLIC DIFFERENTIAL EQUATIONS 115

Consequently, the equation for v can be written in the form

v" +— v1 + ctv = 0
z
with the condition v(0) = 1. A solution of this equation is the Bessel func¬
tion of the zero order (see Appendix)

v(z) = Мл/Тіг)
or

V(X, y\ f, rj) = J0{VCy [U — I)2 — (у — j?)2]) (2-5.31)

For the determination of U(x,y) we shall now use formula (2-5.10) which in
our case reads

U(P) + U(Q)
U(M) =
2
+ -ІГ
2
(vUvdZ - Uvr,d$) , drj = 0 . (2-5.32)
' 2 )i

First we calculate the integral along PQ (rj = 0):

^{vUr, - Uvv)d$ = p'j/,0/Ci[(x - f)2 - y2])Ur,($, 0)

me, ъ)ѴсхуПѴсУ(х - z? - /) (2-5.33)

Vcdb-ef-f]
On the basis of the initial conditions (2-5.22'), (2-5.23') we find

U(x,y) = 9l{x ~У) + Уі(* + У) +-1рѴо(Ѵ/^ V{x - ?)2 - у2)фЛ№

+ J-л/~С\ у Г М^Т/ІХ-П'-ЛЫШ. (2-5.34)

2 ' )*-, Ѵ(х - f)2 - у2
from which because of (2-5.24), (2-5.22') and (2-5.23') there results the formula

. <p(x - у)е-{іа-Ь)/г)у + <p(x + у)е(Ы+Ь)/г)у

u(x, y)

- Vix -f)! - Л

. 'T~ — >') ] -I./3IU-

+ VC,y V(x - (>■ - / Г

+ V(x - ()• - (2-5.35)

^ J x-y

which represents the solution of our problem.

We shall consider the special case a = 0, b = 0, i.e., the equation

uxx — uyy + cu = 0.
116 HYPERBOLIC DIFFERENTIAL EQUATIONS

From (2-5.35) directly follows

«<*, y) = JO +ЛІ1 + » +

/.(V^c. l/U - g)2 ~ /) (2-5.36)

V(x - f )2 - y2
If we put Ci = 0 and у = at, we arrive at the D’Alembert formula

/ y>(-£ — fl/) + y>(# -f- at) 1 f +11 ic.\j£

u(x, t) = JL-—^-+ — \ ф(£)</£ . (2-5.37)
^ ^ Jx-ai

This solves the equation of the vibrating string

1
UXx 2 H
a
with the initial conditions
и (x, 0) = <p(x)
ut{x, 0) = ф(х)
ф(х) — аф(х) = auy{x, 0) .

Problems

1. Solve Problem 1 from Section 2-4 under the assumption that at the initial
moment the concentration of the liquid in the entire tube is constant and dry
air streams in through an opening.
2. Solve Problem 2 from Section 2-4 under the assumption that the initial
temperature of the system equals u0 while the temperature at the ends of the
tube during the entire time maintains a constant value v0 > u0.
3. Find a solution of the telegraphic system of equations (see Eq. (2-1.21)),

ix T Cvt T Gv — 0
vx T Lit T Ri — 0

for the infinite straight line with the initial conditions

i(x, 0) = <p(x),
v{x, 0) = ф(х).

Hint: First reduce the equation system to an equation of the second order
for one of the functions i(x, t) or v(x, t), then find, for example,

ixx — CLitt T (CR + GL)it + GRi

with the initial conditions

i(x, 0) = <p(x),
di
+ -J-i) = - \ф\х) - j-<p(x) = фо(х)
at (=0 L L, /1=о L* L,

Then use formula (2-5.35). Investigate the solution of the telegraphic equation
2-6. APPLICATIONS TO CHAPTER 2 117

which is obtained for small G and R, formula (2-5.35). Consider the limiting
case G—>0, /?—>0 and obtain from (2-5.35) the D’Alembert formula for the
solution of the equation of the vibrating string.

2-6. APPLICATIONS TO CHAPTER 2

1. The vibration of strings of musical instruments

A vibrating string produces vibrations in the air which the human ear
perceives as the tone emitted from the string. The tone strength is deter¬
mined by the energy or the amplitude of the vibrations, the tone pitch by
the vibration period, and the tone color by the energy relation between the
lowest tone and the overtones.26
We shall not go into the physiological process of sound perception and
the propagation of sound in air; on the contrary the acoustic effect produced
by a string is characterized by the energy, the period, and the distribution
of the energy in the overtones.
Ordinarily in musical instruments transverse vibrations of the strings
are generated. We can distinguish three types of string instruments, depend¬
ing on whether the strings are stroked, plucked, or struck. Strings which
are struck (for example, in a piano) are thus given a fixed initial velocity,
but undergo no initial displacement. With the plucked instruments (for ex¬
ample, a harp or guitar) the strings are made to vibrate from a fixed initial
displacement without initial velocity.
The free vibrations of a string excited in an arbitrary manner can be
represented in the form (see Section 2-3)

u(x, t) = 2 (ancoswnt + 6nSino)„D sin^x , wn = ~a .

n=1 l l

Problem 1 given in Section 2-3 as an exercise underlies the very simple

theory of vibration of the stroked instrument. The solution of this problem
shows that the above coefficients are given by

2 hi2 • лпс
sin —j— ,
, A
bn — 0 (2-6.1)
an —
n n c(l — c)

provided that the initial displacement has the form of a triangle with the
height h at the point x — c (Figure 29). For the energy of the nth harmonic
vibration we find

En = ~Plwl al = Mh2 2 /***—-ysin2^-, M = pi (2-6.2)

4 7:nc(l—c) l

which is inversely proportional to n .

In Problem 4, Section 2-3, the simple theory of the struck string was
investigated; that is, the string was excited by a concentrated blow with an
impulse К at a point c. The solution of this problem has the form

26 Rayleigh Theory of Sound, 2d ed., Vol. I, Ch. VI, London, 1894.

118 HYPERBOLIC DIFFERENTIAL EQUATIONS

, .. 2 К ^ 1 . ППС . nn . , ПП r o\
u(x, t) =- 2—sin—-—• sin — x • sin wnt , wn = —a (Z-b.3)
nap n=i nil l

En = ~7~r sin —— . (2-6.4)

M l

Consequently, for a string which is excited by a concentrated blow in a

small interval of length 8, (for which <5 is small in comparison with the
diistance between the individual nodes) the energies of the different harmonics
can differ only slightly, and the tone of a
string thus excited produces overtones to
a strong degree. This conclusion can be
easily demonstrated experimentally as
follows: If a stretched string (as in a
monochord) is struck with the edge of a
knife, then it rings very clearly, i.e., the
overtones are very noticeable. In the
piano, the strings are struck by hammers
which are padded with felt or leather. Such an excitation of the string can
be represented by the following scheme:
(1) Let the string under consideration be excited so that it obtains a
constant initial velocity v0 in the interval (c — 8, c + 8). This case corresponds
to a flat rigid hammer which has a width of 28 and strikes at the point c.
Here the vibrations are described by the function

4v0l 1 • nnc . nn ~ . nn . .
u{x, t) = 2 —5- sin - sin — 0 sin — X sin Wnt
n a n=1 n2 l l l

(see Section 2-3, Problem 3) while the energy of the individual harmonic
vibrations is given by

AMvl . 2nnc . 2 nn8

E =
■L^n sin —-— sin
П2 7Г2 ~r

(2) The string is excited by an initial velocity of the form

n
V0 COS for I x — с I <5
0) 0 ~2
at
0 for I x — с I >8 .

This case corresponds to a rigid convex hammer of “width” 28. Such a

hammer gives the string at the center of the interval 28 the largest initial
velocity described schematically by the last-mentioned function. The vibra¬
tions so produced have the form

nn - • nn
. „ cos — 8 • sin — c
u(x, t) —
8 v08 j J_ /
2 - sin у л; • sinaij
n2a *-1 n 1
I

{see Section 2-3, Problem 3). The energy of the nth harmonic vibration is
2-6. APPLICATIONS TO CHAPTER 2 119

2 югд . 2 nnc
cos —-— sm
l

(3) A hammer which strikes a string is not ideally rigid. Then the
vibrations are no longer determined by the initial velocity but by a time-
varying force. Thus we arrive at an inhomogeneous equation with the
function

F(x, t) =
(T-, X— С
Г о COS—-— •
0
7Г . nt
— sin —
Z т
for I x — с I < d, 0 ^ t йт ,

0 for I x — с 1 > <5, t > T

in the right-hand member. The solution of this equation for t > r reads

^xno 0)nT . лпс

„ cos—— cos-£- sm-
, .. 16F0n5 ” 1 l 2 I 7Г П
U(X, t) = -3- I sin -XSin 0)J t — —
лъ pa
~я['-т)['-т\
The above example shows that the width of the interval on which the
blow acts as well as the duration of the blow has a substantial influence on
the energies of the higher overtones. Moreover the appearance of the factor
sin (nnll)c shows that when the center of the blow lies at a node of the wth
harmonic, the energy of the harmonics are equal to zero.
The appearance of higher overtones (from the seventh on) disturbs the
harmonics of the sound and produces a dissonance phenomenon.27 Conversely,
the presence of lower overtones produces a greater sound volume. In order
to reduce the energies, the striking point in the piano is chosen in the
neighborhood of the point of attachment, between the nodes of the seventh
and eighth overtones. By a suitable selection of the width of the hammers
and the cushioning, an increase of the energy of the lower (the third and
fourth) overtones is again obtained. With older types of pianos, the use of
narrower and harder hammers produces a shriller and even a clinking tone.

2. Vibrations of rods

Differential equations of the second order occupy an important place in

textbooks on methods of mathematical physics, although many vibration
problems—for example, the vibration of rods, plates, etc.—lead to equations
of higher order.
To exemplify an equation of the fourth order, we shall consider the
natural vibrations of a tuning fork. This problem is equivalent to the
problem of the vibration of a thin rectangular rod which is permanently

27 When, for example, the basic frequency (the first harmonic) corresponds to 440
vibrations per second —A in the center octave—then at the sevenfold frequency the
A, G of the fourth octave is produced. The interval A-G, the so-called minor seventh,
is unpleasant to the ear and is felt as a dissonance.
120 HYPERBOLIC DIFFERENTIAL EQUATIONS

fixed at its ends as in a vise. The determination of the form of the vibra¬
tions of a tuning fork and its frequencies leads to the equation of a transverse
vibrating rod,

g2v d*y
+ a‘ (2-6.5)
dt2 ox

From this equation we are led to several problems in vibrations of rods, the
calculation of the stability of rolling waves, and investigations of the vibra¬
tion of vessels.28
To give an elementary derivation of (2-6.5) we shall consider a rectan¬
gular rod of length l (0 ^ x S l) of height h and of width b (Figure 30). Let

dx be the length of an element. The boundary surfaces of the chosen rod

element, which is assumed as planar, after the deformation forms the angle
d<p as shown in the figure. If the deformation is small and the length of
the rod axis does not change under the deformation (dl = dx) then

dy dy
dcp dx
dx dx x+dx

is valid. The section of the rod which is at a distance rj from the rod axis
у = 0 changes its length by an amount rjdy> (Figure 31). Therefore, according
to Hooke’s law the tension acting along the section is equal to

dN= E ■ bdrj .2^° = -E • bpzrjdr) ,

dx dx

where E is the modulus of elasticity of the corresponding section. The

complete moment of flexure of the force acting on the slice at x is

(2-6.6)

where

28 See for example, the monograph of A. N. Krylov, The Vibrations of Vessels,

Moscow, [and] Leningrad, 1948.
2-6. APPLICATIONS TO CHAPTER 2 121

is the moment of inertia of the rectangular layer

i-(M + dM)
with respect to its horizontal axis. We denote by
M(x) the moment acting on the slice at x on the
right side of the rod. Then obviously (M + dM)
is the moment of the force corresponding to the
slice at x + dx.
FIG. 31. The moment dM coincides with the moment
of the tangential force

dM = Fdx .

Hence, because of (2-6.6), we obtain

(2-6.7)

for the magnitude of the tangential force

If we set the resulting force on the element under consideration,

equal to the product of the mass of the element by the acceleration

where p is the density of the rod and 5 is the area of the cross section
(here we neglect the rotary motion arising from the distortion), we obtain
the equation of a transverse vibrating rod:

(2-6.5)

The boundary conditions at the fixed end x = 0 are the immovability of

the rod and the horizontal position of the tangent:

(2-6.8)

At the free end the moment of flexure (2-6.6) and the tangential force
(2-6.7) must be equal to zero. This yields

In order to completely define the motion of the rod even the initial
conditions (the initial displacement and initial velocity) must be given in the
form

(2-6.10)
122 HYPERBOLIC DIFFERENTIAL EQUATIONS

The problem considered therefore leads to the solution of equation (2-6.5)

with boundary conditions (2-6.8), (2-6.9), and the initial conditions (2-6.10).
We shall solve this problem by separation of variables with the aid of
the expression

у = Y(x)T (t) . (2-6.11)

By putting (2-6.11) into (2-6.5) we obtain

T"(t) Yu\x)
a2T(t) ~ Y(x) ~ Л ’

from which the eigenvalue problem

Y^-XY = 0 (2-6.12)

dY d2Y d3Y
Y = 0 , = 0 , = o,
x=0 dx x=0 dx2 x=l dx3

results for Y(x). The general solution of Eq. (2-6.12) reads

Y(x) = A cosh Vx x + В sinh fyX x + C cos y/X x + D sin y/X x .

From the conditions F(0) = 0, Y'(0) = 0, we find C — — A and D=—B.

Hence it follows that

Y(x) = A (cosh y/Xx — cos yJJx )+ В (sinh y/J x — sin y/X x) .

From Y"(l) = 0 and Y"'(l) = 0, we find

H(cosh у/XI + cos y/X l) + В (sinh yfxi + sin у/XI) = 0 ,

H(sinh у/XI — sin y/X l) + B(cosh y/X l + cos у/XI) = 0 .

This system of homogeneous equations possesses a nontrivial solution A and

B, provided its determinant is equal to zero. If we set this determinant
equal to zero then we obtain the eigenvalues from the transcendental equation

sinh2 y/ll — sin2 у/XI = cosh2 y/J 1 + 2 cosh y/X l cos fyX l + cos2 y/11.

Because cosh2x — sinh2A: = 1, we can write

cosh (j. ■ cos fi = —1 , Ц— у/XI. (2-6.14)

The roots of Eq. (2-6.14) can be calculated without difficulty29, as, for example,
graphically we find

A*! = 1.875 ,
[iz - 4.694 ,
Иг = 7.854 ,

fjn ~ ~7г (2я — 1) for n > 3 .

Cj

*» For the calculation of the roots of Eq. (2-6.14) see Rayleigh, op. cit., fn. 26.
2-6. APPLICATIONS TO CHAPTER 2 123

The last formula accurately gives the value of pn for n = 3 to the third
decimal place and accurately for n = 7 up to the sixth place.
Now we shall consider the frequency of vibration of a tuning fork. The
equation

T" + a2i.nT = 0

is satisfied by the trigonometric functions

Tn(t) = an cos 2л vnt + bn sin 2itvnt

with the frequencies

_ aVln _ VXn /EJ _ pi jEJ

2л 2п У pS 2лl2 У pS

The frequencies pn of the eigenvibrations therefore behave as the square

of fin. Because

4 = 6.267 , 4 = 17-548
Pi Pi

the second eigentone is more than two and one-half octaves higher than the
base tone, i.e., higher than the sixth harmonic of the string with equal base
tones, while the third eigenvibration is more than four octaves higher than
the base tone. If the tuning fork, for example, has as a basic frequency of
440 vibrations per second (usually one takes for A' the small A of the
first octave) then the following eigenfrequency of the tuning fork equals
2757.5 vibrations per second (between C"" = 2637.3 and = 2794.0, thus
between the E and F of the fourth octave of uniformly tempered scale) while
the third eigenfrequency of 7721.1 vibrations per second already is higher
than those frequencies used in music.
If the tuning fork is set to vibrating by a blow, higher frequencies ap¬
pear in addition to the first, which explains the initial metallic sound. The
higher harmonics are nevertheless quickly damped so that the tuning fork
soon rings out with the pure basic tone.

3. Vibrations of a string loaded by masses

1. Statement of the problem. In the following discussion, we shall limit

ourselves to vibrations of a string (0, l) fixed at the ends and loaded at fixed
points x = Xi (i = 1, 2, • • •, n) by concentrated masses Mi.
The conditions at point Xi can be of two different types. If a concentrated
force Fi(t) acts at point Xi (i — 1,2, •••, n), the conditions

u(Xi — 0, t) = u(Xi + 0, t) (2-6.15)

kux I ^ = -Fi (2-6.16)

must be satisfied. In the present case, Fi signifies an inertial force. If we

substitute into (2-6.16) the following

Fi = -MiUuiXi, t) ,
124 HYPERBOLIC DIFFERENTIAL EQUATIONS

then we obtain

MiUtt(Xi, t) = kux I • (2-6.17)

The condition (2-6.17) can be derived by still another method. First we

distribute the mass Mi in the interval (Xi — e, Xi + e) with uniform density
8i and then use the vibration equation for the inhomogeneous string

(p + di) uH =Yx(Kklhc) Xi~ £ < % < Xi + e , (2-6.18)

where p is the density of the string. Let ue(x, t) be a solution of this equation.
By integration of Eq. (2-6.18) with respect to x from xx — e to Xi + £ and
passage to the limit as s—>0, the condition (2-6.17) results for the function
u(x, t) = lim Ue(x, t). We shall not go into a more precise investigation of the
8—0
passage to the limit.
The complete formulation of our problem then reads:
Find a solution of the equation

d2u _ _d_ (,fu\

(2-6.19)
P dt2 dx \ dx)

which satisfies the boundary conditions

u{0, t) = 0,1
(2-6.20)
u(l, t) = 0, J

the transition conditions

u(Xi — 0, t) = u{Xi + 0, t)
i (2-6.21)
MiUtt{Xi, t) = kux I

at the point x = xit and the initial conditions

u(xt 0) (p{x) ) (^6 ^^)

ut{x, 0) = ф(х) )

where <p(x) and <p(x) are prescribed functions.

2. Eigenvibrations of a string loaded by masses. We shall first consider
the eigenfrequencies and the profile of a standing wave for a loaded string.
Therefore, we shall seek the solution of this problem in the form

u(x, t) = X(x) T(t) . (2-6.23)

If we substitute (2-6.23) into (2-6.19) and use the boundary conditions, we

obtain by separation of the variables

T" + IT = 0 (2-6.24)

and

4-(kX') + ЛРХ= 0,
dx (2-6.25)
X(0) - 0 X(l) = 0 .
2-6. APPLICATIONS TO CHAPTER 2 125

The transition conditions yield

X (Xi — 0) = X (хі + 0)
МіХШТ" = kX' \:\t°0T.
By consideration of (2-6.24) we can write this relation in the form

kX' I xx\t°o = -ЖіХ^Хі) .

In this manner we arrive at the following eigenvalue problem for X(x):

4~(kX') + XpX = 0 , X(0) = 0 , X(l) = 0 (2-6.25)

ax

X(Xi - 0) = Х(хі + 0) , i = 1,2, • • -, n (2-6.26)

kX'(Xi + 0) - kX'(xt - 0) + ЖіХ(хі) = 0 . (2-6.27)

The peculiarity of the boundary-value problem considered, which differs

from the problems treated previously, lies in the occurrence of the parameter
к not only in the equation but also in an auxiliary condition.
It was shown in Section 2-3 that the eigenfunctions

Xi(x) , X2(x), • • •

of the problem

4-(kX') + ipX = 0, X(0)=0, X(l) = 0

dx
are orthogonal in the interval (0, /) with respect to the density function p(x):

j Xm(x)X„(x)p(x)dx = 0 , m Ф n . (2-6.28)

We now distribute each mass Mi with uniform density <5 in the interval
X, — e < x < Xi + e, where e > 0 is an arbitrarily small number. Hence we
arrive at a corresponding vibration problem for the inhomogeneous string
with density pe(x). Let and Xn(x) be the eigenvalues and eigenfunctions
respectively of this problem. The eigenfunctions must satisfy the orthogo¬
nality relation,

j Xem{x)Xen(x)pe(x)dx = 0 . (2-6.29)

If we break up the integral in (2-6.29) over (0, /) into ranges from

(0, Xi — e) plus the integral first from (Xi — e, Xi + e) about the portion of
the interval, then from (x, + e, /), and take the limit as e—>0, we obtain the
relation

( Xm(x)Xn(x)p(x)dx + І' MtXm(Xi)Xn(Xi) = 0 m Ф n , (2-6.30)

Jo i=l

the condition for the so-called weighted orthogonality.30

30 R. Courant and D. Hilbert, Methods of Mathematical Physics, Vol. I, Ch. 6,

Berlin, 1931.
126 HYPERBOLIC DIFFERENTIAL EQUATIONS

We shall not go into the investigation of the question as to whether such

passage to the limit is permissible.
The orthogonality condition (2-6.30) can also be derived by a purely
formal method from the differential equation and the conditions (2-6.25)
through (2-6.27). Let Xm(x) and X„(x) be two eigenfunctions of the problem.
Let Лт and Xn be the eigenvalues corresponding to it. These satisfy the
conditions

Tx (k^t) XmpXn + = ° and + x*PX' = 0 '

We multiply the first relation with Xn(x), the second with Xm(x), and
subtract the resulting equations from each other. If we integrate this dif¬
ference over (0, Xi), {Xi, хг), • ■ • (хк, l) and add the results, we find

Цп - Xn) [lXJx)Xn(x)p(x)dx - І \H+14-[XmkX'n - XnkX'm]dx = 0 , (2-6.31)

Jo i=° Jxj dx

where we have set x0 = 0, Хк+1 = l. If the integration is performed in each

summand of the sum, and if the terms which correspond to the boundaries
x = Xi — 0 and x = Xi + 0 are collected, we obtain

Ai — (XmkX'n — XnkX m)x=Xi- о — (XmkX n — XnkX m)x=xj-f

Here the expression evaluated at x = 0 and x = l on the basis of the boundary
conditions equals zero.
For the calculation of Ai we use the transition conditions

Xj(Xi — 0) — Xj(Xi + 0) . /о с. оn
, , i — m, n . (l-b.A7
kX j(Xi + 0) — kX j(Xi — 0) = —MiXjXj(Xi)

If we then write Л, in the form

Ai = Xm(Xi) [kX'n(Xi — 0) — kX'n(Xi + 0)] — Xn(Xi) [kX'm(Xi — 0) — kX'm(Xi + 0)]

and take (2-6.27) into consideration, we obtain

А{ = Xm(Xi)MiXnXn(Xi) — Хп(Хі)МіЛтХт{Хі) = MtXm(Xi)Xn(Xi)(Xn — Zm) .

Equation (2-6.31) can now be written in the form

(Лт — Л«) jj Xm(x)Xn(x)p(x)dx + 2 MiXm(Xi)Xn(Xi) I=0.

For }.m Ф 2n the orthogonality condition (2-6.30) follows immediately.
The norm of the eigenfunction Xn(x) is defined by

Nn = [ Xl{x)p(x)dx + І МіХІ(хі) . (2-6.32)

Jo »=1

We shall not at this point go into the proof of the existence of infinitely
many eigenvalues and eigenfunctions, the positiveness of the eigenvalues,
and the corresponding development theorem. The boundary-value problem
considered above, as also the problem investigated in Section 2-3, leads to an
2-6. APPLICATIONS TO CHAPTER 2 127

integral equation, in the present case to a weighted integral equation,31 which

is equivalent to an integral equation with Stieltjes integrals.
For the development of a function fix) in a series

fix) = І fnXnix)
71 = 1

the following formula for the coefficients holds:

\ f (x)Xn(x)p(x)dx + I Mif{Xi)Xn{Xi)
fn = ^-. (2-6.33)
Nn

The initial-value problem described in Section 2-6 § 1 is solved according

to the above method of separation of variables. We can treat analogously
the vibrations of a rod (or a beam) which is loaded with a concentrated mass.
In physics and in technology vibration problems of a string loaded with
a concentrated mass often occur. Poisson had already solved the problem of
the motion of a weight hanging from an elastic string. Then A. N. Krylov^
showed that the theory is applicable to the indicators of damping machines,
the torsional vibrations to balance wheels, to valves of different types, etc.,
and that they lead back to this problem. Also of great significance for the
theory of various measuring instruments is the study of the torsional vibra¬
tions of fibers with a mass fastened to one end (for example, a mirror).
A specific instance occurs with a similar problem in investigating the
stability of vibrations of the air foils of airplanes. For the solution of this pro¬
blem one must calculate the eigenfrequencies of the air foils (which can be
considered as a beam of variable cross section) which are loaded by masses (say,
the motor). Another example is the calculation of the eigenvibrations of
antennas with lumped capacity and self-inductance.
We shall not concern ourselves here with the approximation methods for
the determination of the eigenvalues and the eigenfunctions of these problems,
since they are analogous to the approximation methods for the determination
of the corresponding quantities for the inhomogeneous string.
3. The vibrations of a string with a loaded end. The vibration of a
homogeneous string which is fastened at the end (x = 0) while at the other
end (x = l) hangs a weight of mass M is, in practice, of special interest.
In this case the condition at x — l assumes the form

Muuil, t) = —kuxfl, t) .

For the amplitudes of the standing waves we obtain the equation

X'r! + XnXn = 0

with the boundary conditions

31 See A. Kneser, “Integral Equations and the Representation of Arbitrary

Functions of Two Variables,” Rend. Palermo 27, 117-147, 1908.
32 A. N. Krylov, Differential Equations of Mathematical Physics, Ch. VII, Academy
of Science of USSR, 1932.
128 HYPERBOLIC DIFFERENTIAL EQUATIONS

Xn(0) = 0, Xtil)= — *nXn(t).

P
Hence we find

sing/ lnx
Xn(x) =
sin VYnl
where is fixed by

ctg VU = —ѴК ■ (2-6.34)

The orthogonality condition for the functions Xn(x) assumes the form

Xn(x)Xn(x)p(x)dx + MXn(l)Xm(l) = 0 .
)

For the norms we have

Nn = j Xl(x)p(x)dx + MX 1(1) .

From (2-6.34) we therefore obtain

N =— + — + ^—21

The initial-value problem is solved here in the usual manner.

4. Corrections on the eigenvalues. We shall calculate the corrections to
the eigenfrequencies for the case of smaller and larger loads M. For the
sake of simplicity we shall consider the case in which the weight hangs at
the end of the string. Both of the following limiting cases are possible:
(1) M = 0. The end x = l is free. For the eigenvalues, then, we have

i /ТіТГ 2n + 1 n
V An — ry
2 l

(2) M = со. The end x — I is rigidly fixed: u(l, t) = 0. Here the eigen¬
values can be calculated from

Of interest to us now are the eigenvalues for small M {M —>0) and large
M (M—> oo).
(a) M is small. In order to find the corrections to the eigenvalues
we set

VJn = VW + eM , (2-6.35)

where e is a fixed number. If we substitute (2-6.35) into Eq. (2-6.34) and

neglect Мг and the higher powers of M, we obtain

К = a (2-6.36)
2-6. APPLICATIONS TO CHAPTER 2 129

i.e., the eigenvalues of the weighted string increase as M—*■ 0 and approaches
the eigenvalues of the string with a free end.
(b) M is large. Now we set

л/~Ап = л/.Af' + £— •

Then (2-6.34) yields

Vfff ’
where the terms which contain 1/M2 and higher powers of 1/M can be
neglected. Thus, there results

Vln = VAn' + ycjjlij m ’ = + М/ ’ (2-6.37)

i.e., the eigenfrequencies with increasing loads becomes smaller and approaches
uniformly the eigenvalues of the string with rigidly fastened ends.

4. Equations of gas dynamics and the theory of shock waves

1. Equations of gas dynamics. The conservation of energy expression.

The equations of acoustics (see Section 2-6 §1) were derived under the as¬
sumption that the stream velocity of the gases and the pressure changes are
small. Under these assumptions the equations of hydrodynamics assume a
linear form.
On the other hand, one deals with hydrodynamic processes in problems
which result from the investigation of rocket flights, the velocity of airplanes,
ballistics, with detonation waves, etc., for which high velocities and large
pressure differences are characteristic. In these cases, the linear approxima¬
tion of acoustics is useless and instead we must use the nonlinear equations
of hydrodynamics.
Since in practice such motions occur predominantly in gases, the hydro¬
dynamics of high velocities are usually called gas dynamics.
The equations of gas dynamics have, for one-dimensional motion of the
gas (in the direction of the л: axis), the form
a?|-?

о
+

continuity equation: (2-6.38)

II

dv . dv dp
momentum equation: (2-6.39)
pTt pVTx - ~x
+

equation of state: P =f(p, T) . (2-6.40)

Therefore, the equations of gas dynamics represent the equations of

motion of an ideal compressible fluid in the absence of external forces.
We shall now derive the conservation of energy law. The energy per
unit of volume is equal to
130 HYPERBOLIC DIFFERENTIAL EQUATIONS

(2-6.41)

where the first member signifies the kinetic energy and the second, the
internal energy. Obviously, here e is the internal energy per unit of mass.
For an ideal gas, e = cvT where cv is the specific heat at constant volume
and T is the temperature. For the energy change in a unit of time we
therefore have

_a_
(2-6.42)
dt

If we carry out the differentiation in the first summand, then on the basis of
Eq. (2-6.38) and (2-6.39) we obtain

d / pv2 \ 2 dp dv v2 d . .
~~ + pv (2-6.43)
dt \ 2 ) dt dt YTX {"v)
To transform the derivative (d/dt){pe) we apply the first law of thermodynamics
which expresses the law of conservation of energy:

dQ = de + pdz . (2-6.44)

Thus dQ is the amount of heat which the system obtains from the outside
or gives to the outside, and pdz is the work which must be performed in
order to change the volume by an amount dz (z = 1/p is the specific volume).
If the process proceeds adiabatically (i.e., no heat exchange occurs with
the surrounding medium) then

dQ = 0 ,

and

de = -pd— =Adp . (2-6.45)

P P
With the help of this expression we find

d(pe) = edp + pde = edp + — dp = wdp (2-6.46)

P

Jr (pe) = (2-6.47)
at at

where

w = e +— (2-6.48)
P
is the heat function (enthalpy), or the heat content per unit of mass.
The derivative діѵ/dx on the basis of the expressions (2-6.46) and (2-6.48)
satisfies the equation
2-6. APPLICATIONS TO CHAPTER 2 131

By consideration of Eqs. (2-6.39), (2-6.42), (2-6.43), (2-6.47) and (2-6.49), we

obtain the conservation of energy relation in the differential form

d_
(2-6.50)
dt

To find the physical significance of this equation we integrate it over an

interval (xt, x2):

d_ *2

'2/£t + P£)dx=-Pv(T w
dt 11

On the left side is the energy change per unit of time in the interval (x,_,
x2)\ on the right is the amount of energy which in a unit of time flows out
from the considered volume.
If the heat conduction can not be ignored, the conservation of energy
equation assumes the form

d_
(2-6.51)
dt

where к is the coefficient of thermal conductivity.

2. Shock waves: the Hugoniot conditions. With high velocities motions
are possible, which on certain surfaces can propagate discontinuities of the
hydrodynamic quantities (pressure, velocity, density, etc.) in space. These
surfaces of discontinuity are usually called shock waves.
On the surface supporting the discontinuity (the front of the shock wave),
the conditions for the continuity of the streaming of the fluid (or of the gas),
the energy, and the momentum (the Hugoniot conditions) must be satisfied.
To derive these conditions, we first transform Eq. (2-6.38) to a form
suitable for our purpose. For this purpose we multiply (2-6.38) by v and add
the result to (2-6.38). This gives

£w = -£<p + *»'>• (2-6.39')

Further we write the continuity equation, the equation of motion, and the
conservation of energy relation in the following form,

(2-6.38')
!=->

(2-6.39')

\ d r■ fv‘ , Vi
/ “ dx[ fv(j + w) J (2-6.50)

In the x, t plane we consider the “spur” x = a{t) of the shock front. Let
AC be any arc of x = a(t). The points A and C must therefore have the
coordinates x,, tx, or x2 = Xi + Ax, t2 = tl + dt. Further, let the points В
and D be situated so that the sides of the rectangle ABCD run parallel to
the coordinate axes.
132 HYPERBOLIC DIFFERENTIAL EQUATIONS

We write the law of the conservation of mass in the integral form

( 2[(p)tz — (p)t1\dx - — [ г[{рѵ)Іг - (pv)Xl]dt . (2-6.52)

Here the left side is the change of mass in the interval (xt, x2) during the
period of time (t,, t2); on the right side is the amount of fluid (the gas)
which flows out from (x,, x2) during the time (1,, t2). If the functions p and
pv are continuous throughout and differentiable in ABCD, Eqs. (2-6.52) and
(2-6.37') are equivalent. In the present case, however, this does not occur.
We apply the mean-value theorem to each individual summand, with the
result

[{p)t=t2 - (p)t=t1] j- = -(/»).=,, + (pv)x=Xl .

Here x*, ***, t*, t** signify the corresponding mean values of x and t.
If we pass to the limit as dx —»0 (x2 —»■ xj and At —>0 (t2—> td and denote
the value of the function above the curve x = a(t), behind the front of the
shock wave, by the index 1 and the value of the function below this curve,
before the front, by the index 2, we obtain

{p2 -Pi)U= -(pv), + (pv)2 , (2-6.53)

where

TT da ,. Ax
U — — = lim —
at jt-oAt
is the velocity of the shock wave.
In the coordinate system moving with the shock wave we denote by

и, — U — v,, u2 = U — v2
the velocity of the elements before and behind the front. Therefore (2-6.53)
can be written in the form

p{U 1 — p2^2 • (2-6.53')

This equation signifies the continuity of the matter flowing across the fronts
of the shock wave.
The law of conservation of momentum in integral form reads
x2 Г *2
[ipv)t2 — (pv)tjdx = — 1 [(p + рѵ2)Хг — (p + pv2)Xl\dt ,
xi Jq
where the right side represents the sum of the acting forces produced by the
impulses (pressure forces) and the flow of the momentum. As Ax—>0 and
At -> 0 we obtain
U[(pv)2 — (pv),] = —(p + pv2), + (p + pv2)2

or

pi T p,u\ — p2 + p2u\ , (2-6.54)

2-6. APPLICATIONS TO CHAPTER 2 133

the conservation law for the flow of the momentum across a front; and for
the conservation of energy on the front we obtain the equation

pv
£%- + P*) U~ + joe ) U — — piVA — + W ) + p2Vz(— + W

which after a single simple transformation assumes the form

PiUi P2U2

or according to (2-6.53)

(2-6.55)

Thus, on the front of a shock wave, the following equations (the Hugoniot
conditions) must be satisfied:

PiUi — pzWz (2-6.53')

pi + PiU\ = pi + piUi (2-6.54)

. u\ . u\
Wi + ~2 — wi 3—ту • (2-6.55)

From both of the first Eqs. (2-6.53') and (2-6.54) we can express щ and uz
through p and p:

.fi — Pz Pi ~ Pi ni — Pi Pi~ Pi
Pi P1 pz Pz pi P2

from which it follows that

u\ — u\ — —p' +-p- (pi — pi) .

P1P2
If we then insert this expression into (2-6.55), we find between the energy
values on both sides of the wave front the relations

Wi — Wi — (pi + pz)(pi — Pz)

2piPi

and

1
£i — e2 (pi — Pi)(pi + Pz) •
2 pipi
For an ideal gas we have

Cp — У t P
p = RpT , s = cvT , w = cPT =
cP — cv у — 1 p

that is,

(2-6.56)
134 HYPERBOLIC DIFFERENTIAL EQUATIONS

After a simple transformation we obtain from (2-6.56) the so-called Hugoniot

adiabatic equation

Pz_ _ (r + 1 )рг + (r — l)/>i (2-6.57)

Pi (r — Dpi + (r + Dpi

Pi _ (t Pp2 (t 1)|Qi /2.0 53)

pi (r + Dpi — (r — Dpi
From this, the quantities ti, pit t2, p2 can be calculated if the remaining three
are known.
A shock wave can move continuously with respect to the gas from
positions of high pressure to positions of lower pressure: p2~+pi (law of
Zemplen). Hence it follows that the density of gas behind the front is larger
than that before the front.
Formula (2-6.57) expresses the dependence between p2 and p2 for prescribed
pi and pi. For a given pi and piy p2 = p2{p2) is a monotone increasing function
which for p2lpi —> oo (shock wave of larger amplitude) tends towards a finite
limit value

Pi _ Г + 1 (2-6.59)
Pi Г ~ 1 ’

This formula gives the maximum jump in the density which can occur on
the front of the shock wave. For diatomic gases у — 7/5, and the maximum
density jump is then equal to 6:

Ti = 6
Pi
From Eqs. (2-6.53'), (2-6.54), and (2-6.57), we find for pi = 0,

Ml = m2 = (r- D2 Pi
2(r + 1) Pi
If the motion of the shock wave with respect to the gas is considered as
static (Vi = 0), the velocity of propagation of the shock wave equals

U = Jl±l.h,
2 pi

i.e., it increases as the square root of p2.

We shall now treat a quite simple problem in the theory of shock waves,
whose solution can be given analytically. A cylindrical tube x > 0, which is
unbounded on one side and enclosed by a piston at the other (x = 0) contains
a static gas with constant density pi and constant pressure px. At the initial
moment t — 0, the piston begins to move with a constant velocity “v” in the
direction of the positive x axis. In front of the piston a shock wave forms,
which at the initial moment is in the same position as the piston and then
moves away from this position with a constant velocity U > v. Between the
2-6. APPLICATIONS TO CHAPTER 2 135

piston and the front of the shock wave a region 2 forms, in which the gas
moves with the velocity of the piston. Before the front (region 1) the gas
is found in the original state: p = plt p = pi (v = 0).
With the conditions (2-6.53'), (2-6.54), and (2-6.55) for the front, we can
easily determine its velocity, the magnitude of the jump, the density, and
the pressure.
To this end we introduce the nondimensional quantities

o, = U=— v =— Р=Щ (2-6.60)

Pi Ci Ci PiCi

where cx = т/rPilpi is the velocity of sound before the front (in the non-
excited region 1). Then the equation of state can be written in the form

a,U= U-v or 0=—^, (2-6.61)

1 — 0)

p = 1 + jUv or p = 1 + r~r^— * (2-6.62)

pw = 1 + (r - l)(^Uv - J- v2^j . (2-6.63)

Then, by elimination of p and 0, we obtain for the determination of «>

the quadratic equation

2a? - 44 + (Г + 1) 0] + [2 + (r - 1) v2) = 0 . (2-6.64)

Since, obviously, a, < 1, (p2 > p{), the smallest root is given by

o,2 = 1 + (r + D v2-vJ 1 + {Г . (2-6.65)

4 У 16 V

From (2-6.61) and (2-6.65) there results then

U = lL±iU + J\ + {.L±il!y2 (2-6.66)

4 У 16

P = l + № + 1). ѵг + rj) J\ + \LJL±L6*

(Г + 1) (2-6.67)
4 r 16

In the old quantities, therefore,

рг — Pi
1+(xTi-?)+x/^CT
16 C (2-6.68)

1 +
(r - 1)^
2c!

u = L±±v + cji+<L0v>, (2-6.69)

(2-6.70)
136 HYPERBOLIC DIFFERENTIAL EQUATIONS

Since the velocity of the shock wave is constant there, we find for the
position of the front at time t

x = «(/) = „ + c, /1 } t . (2-6.71)

In the limiting case v/ci » 1 (shock wave of greater intensity) we find from
the formulas (2-6.68) through (2-6.70) the limiting relations

r + 1 7 + 1 rir + 1)
Рг — p 1 U— v , Pi — Pi
r- 1 ’ 2 2 c\
obtained earlier.
If v/ci € 1 (shock wave of smaller intensity) then the term ѵг!с\ can be
neglected:

рг — Pi

U = Ci + (r + 1)
4

3. Weak discontinuities. We have hitherto considered the motion of a

shock wave, on whose front the quantities p, p, v, and others change their
values by jumps. Such discontinuities are called strong discontinuities.
However, there are also possible motions for which only the first deriva¬
tives of p, p, v, etc., are discontinuous on certain surfaces, while these
quantities themselves are continuous. We call these weak discontinuities.
In Section 2-2 § 8, we investigated the motions of such discontinuities
and determined that these discontinuities propagated along the characteristics.
We shall therefore proceed from the equations of acoustics. Corresponding
results can be obtained also from the nonlinear problems of gas dynamics.
We can easily see that a surface, on which weak discontinuities lie with
respect to the gas, propagates with a velocity which is equal to the local
velocity of sound. Therefore, if we consider a small neighborhood of the
surface of discontinuity and take the average value of the hydrodynamic
quantities in this neighborhood, then the weak discontinuities can obviously
be considered as small disturbances which satisfy the equations of acoustics
and propagate with the local velocity of sound.
As an example, we shall consider the flow of a gas in a vacuum. Let
the gas which is enclosed in a semi-infinite body x > 0 be found in the rest
state at time t = 0, and have in the entire region x > 0 the constant density
p and constant pressure p0. For t = 0 the external pressure which acts on
the plane x = 0 vanishes, and the gas begins to move; hence, a weak dis¬
continuity (the attenuation wave) arises, which propagates with the velocity
of sound c0 in the direction of the positive xaxis. The density and the
pressure for t = 0 have discontinuities on the interior side of the front
2-6. APPLICATIONS TO CHAPTER 2 137

x = x,(t) of the gas.These discontinuities vanish however, immediately after

the beginning of the motion.
From the conditions for the continuity of the flow of gases and the
momentum for x = x,(t),

0 = pT(vi — vl) = pt(Vi — vt) ,

pi + pZ(vx — vlf = pt + pt(vi — Ff)z ,
where pi, pi, vГ are the left-hand and pt, pt, vt are the right-hand values
at the point x{(t), we obtain

pt = 0 and pt = 0
because pi = pi = vt = 0 .
For an adiabatic process the equation of state of an ideal gas reads

P = Po(fJ ‘ (2'6'72)
For the solution of the problem we assume that

p = p(£) , P = />(f) , v = v(£) ,

where

If we then calculate the derivatives

dj_= ]_ df
dt t ^ d£ ’ dx t d£
where f — p, v, or p, and substitute the found expression in (2-6.38) and (2-6.39)
then we obtain
, g..dp dv
(v - £)-£ = -P-Ji
dt; d£
(2-6.73)
. dv dp
lv-()Trd7-
Now if we multiply the first equation by (v — £) and add the product to
the second, we obtain
2 dp
(v-$)
d$
or
dp
= (v- $)2 .
dp
Hence

°-!=ЧТг±С'
where r is the velocity of sound for the adiabatic process.
138 HYPERBOLIC DIFFERENTIAL EQUATIONS

Since the motion of the weak discontinuity was considered to be in the

direction of the positive x axis, then in the last formula we take the minus
sign, i.e., we have

v — $ = —c . (2-6.74)

If we put this solution into (2-6.73),

dv_c_
(2-6.75)
dp p
or

dv_1_
dp pc

With the help of the equation of state (2-6.72) we find

cл = y—
P
P
and after integration of Eq. (2-6.75)

(2-6.76)

By use of this formula p can be expressed through v:

Г- 1
P — (Ool 1 + (2-6.77)
2 c0

Here we denote by

c0
Г Pa

the velocity of sound for v = 0 (in the static gas). For (2-6.76) we can also
write

v = --— (c — c0) . (2-6.78)

Г - 1

By introduction of the expression (2-6.77) for p in the equation of state

(2-6.72) we obtain

Y- 1 ^,T/IHI
P — poll + (2-6.79)
2 Co

From (2-6.78) and (2-6.73) follows the formula

x
--Co (2-6.80)
r +i Vt
which gives the dependence of the velocity n on т and t. Accordingly, if
we insert expression (2-6.80) for v into (2-6.77) and (2-6.79), we obtain the
explicit dependence of the quantities p and p on x and t. Hence, all the
2-6. APPLICATIONS TO CHAPTER 2 139

quantities are dependent only on x/t. Therefore, if the distance is measured

in units which are proportional to t, the form of motion does not change.
We seek now the velocity of the forward front Vi(t). If we set p — 0 in
Eq. (2-6.79), we obtain

Vi =-- c0 . (2-6.81)
T —1
From this it follows that the flow velocity of the gas in vacuum is finite.
For diatomic gases у — 7/5 and, therefore,

Vi = —5c0 •

We can also obtain the expression (2-6.81) for the velocity of the forward
front x = xx(t) from the relation
r*z
l pdx ~ poXz — poCot , (2-6.82)
J xi
which is the equation of conservation of mass. If, namely, we introduce

f = -
C t ’
we obtain

1 pd£ — p0c0 ■
J'1
Here we substitute the expression for p from (2-6.77). With

1 +
T + 1 co

it follows that

Гл 2 r-i
= (2-6.83)
'лі Г + 1

with

У — 1 _
li = 1 + +-І- ■ -—^ ,
V\ — C0
Ъ = 1
r + 1 Co

By calculating the integral (2-й.83) we obtain

2 ^ (Y+l)/(y-i> _

i.e.,

Л = 0 ,

where

2c0
Vi =
r- 1
The problem of the flow of a gas in a vacuum is therefore solved.
140 HYPERBOLIC DIFFERENTIAL EQUATIONS

In the above considerations we have limited ourselves to the simplest

problem of gas dynamics. For a detailed study of the questions occurring
here we refer to the list of the corresponding special literature.33

5. Dynamics of gas absorption

1. Equation of gas absorption. We shall consider the problem of the

absorption of a gas.34 Let a gas-air mixture be passed through a tube (whose
axis is directed along the x axis of our coordinate system) be filled with an
absorbing substance. Further, we shall indicate by a(x, t) the amount of gas
which is absorbed per unit of volume of absorbent and by u(x, t) the con¬
centration of the gas in the pores of the absorbent in the layer x.
We first derive the condition for the conservation of mass assuming that
the velocity v is sufficiently large and that the diffusion for the passage of
the gas plays no essential role. For the layer of absorption from xi to x2 in
the time from Л to t2 there obviously holds the relation

[vu I *x — vu I SAt = [(« + и) I (2 — (a + и) 1 (J SAx . (2-6.84)

If we divide this by AxAt, as Ax —> 0 and At-+0 we obtain

-ѵдЛ = ±{а + и) , (2-6.85)

ox ot
The left side of this equation represents the amount of gas absorbed during
the passage with respect to the volume and per unit of time, while the right
side gives directly the amount of gas consumed in raising the concentration
of the gases absorbed and found in the pores. To this equation one must
still add the equation of the kinetic absorption

~ = p{u - y) , (2-6.86)
ot
where ft is the so-called kinetic coefficient and у is the concentration of the
gas which exists in equilibrium with the amount of gas absorbed.
The quantities a and у are linked to each other by a relation

a = f(y) , (2-6.87)

the so-called characteristic of the absorbent.

The curve a = f(y) is called the absorption isotherm. If

33 See N. E. Kochin, J. A. Kibel, and N. W. Rose, Theoretical Hydromechanics,

Part II, Ch. I (trans. from Russian), Berlin, 1954; L. Landau and E. Lifschitz, Mechanics
of Deformable Media, Ch. VII, Moscow, 1944; Ya. B. Zeldovich, Theory of Shock
Waves and Introduction to Gas Dynamics, Moscow, 1946; L. I. Sedov, “Propagation of
Strong Detonation Waves,” Applied Mathematics and Mechanics, 10: 2, 1946; R. Sauer,
Introduction to Theoretical Gas Dynamics, 2d ed., Berlin, 1951; K. Oswatitsch, Gas
Dynamics, Vienna, 1952.
34 A. N. Tychonoff, A. A. Zhukhovitskii, and J. L. Zabezhinskii, “Absorption of
Gases from an Airstream by the Layers of a Granulated Substance,” J. Phys. Chem.,
U.S.S.R. 20: 10, 1946.
2-6. APPLICATIONS TO CHAPTER 2 141

f(y) =
u0 + py ’
then we speak of a Langmuir isotherm. A simpler form of the function /
corresponds to the so-called Henry isotherm, which is accurate in regions of
low concentration:
1
(2-6.88)

where l/p is the Henry coefficient.

In this case the following problem arises: From the equations

du da
(2-6.85)
= Hi+ Tt
~ = P(u - ya) (2-6.89)
at
determine the functions u(x, t) and a(x, t) which satisfy the auxiliary condi¬
tions
a(x, 0) = 0
(2-6.90)
u(x, 0) = 0
u(0, t) = Uo (2-6.91)

where u0 is the concentration of the gas at the entrance of the tube.

By neglecting the derivative du/dt (the amount of gas necessary to increase
the free concentration in the pores of the absorbent) compared with the
derivative da/dt (the amount of gas necessary for the increase of the amount
of absorbed gases) we obtain35
du da
(2-6.85')
V dx dt

= p{u- у a) (2-6.89)

a(x, 0) = 0 , u(0, t) — u0 .
If we eliminate the function a(x, t) by introducing the second equation into
the first equation and differentiating with respect to t, then

—vuxt - Put — Prat - Put + Pvyux

or
О

uxt H-u% + Pyux = 0

If we set t = 0 in (2-6.85'), we obtain the initial value of u\

—vux(x, 0) = pu(x, 0) , u(0, 0) = m0 .

From this we find
u(x, 0) = u0e~{fi/v)x .

35 For the system, Eqs. (2-6.85') and (2-6.89) satisfy a single condition, since the axis
t = 0 here is characteristic.
142 HYPERBOLIC DIFFERENTIAL EQUATIONS

The determination of u(x, t) thus leads to the problem of the integration of

the equation

uxt — — ut + = 0 (2-6.92)
v

with the auxiliary conditions

u(x, 0) = м0е_(Р/ѵ)І (2-6.93)
u(0, t) — u0 . (2-6.91)
The characteristics of this equation are the straight lines
x = const. , t — const.
The auxiliary conditions for this problem thus prescribe the values of the
sought function u{x, t) on the characteristics. Analogously, the problem for
a{x, t) reads:

axt = —at + Pra* — 0 (2-6.94)

V

a(x, 0) = 0 (2-6.90)

a(0, t) = ^(1 - e~Mt) . (2-6.95)

Г
Note that a similar problem occurs for a whole series of other questions
(for example, with the process of drying in a stream of air, with the heating
of a tube by flowing water, etc.).36

36 In the transition to Eq. (2-6.85') we have neglected ut . Moreover, we can still

show that we arrive at the same equation if we introduce the variables

T — t — -
V
t = T + — e = x

(Figure 32) in which the time at the point x is mea¬

sured from the moment t° = x/v at which the stream
of the gas-air mixture has reached this point. We
obtain
du du 1 du
dx v dr

d_ d_
dt ' дт
so that Eq. (2-6.85) has the form
du _ da
(2-6.85")
d£ dx

~ = P(u - ra) ■ (2-6.89) FIG. 32.

ОТ

The initial conditions (2-6.90) and Eqs. (2-6.85) and (2-6.89) read

£: Sj: о.
The problem is therefore to determine the function и in the region between the straight
<2-6-90'>
lines t — 0 and the f axis corresponding to the initial conditions (2-6.90') (Cauchy pro¬
blem). Obviously, in this region u(x, t) = 0 but also a = 0). Further, from (2-6.85')
and (2-6.89) it is evident that the function u{x, t) for г = 0 is discontinuous, while a(x, t)
remains continuous. Therefore, и for r = 0, as was shown above, is defined by Eq.
(2-6.85') for a{x, 0) = 0. As shown on pages 141-142 (see formulas (2-6.93) and (2-6.95)),
when we define the value u(x, 0) and a(x, 0) we obtain for u(x, t) and a(x, t), problems
with auxiliary conditions on the characteristics.
2-6. APPLICATIONS TO CHAPTER 2 143

A solution of Eq. (2-6.92) can be obtained explicitly by the method

investigated in Section 2-5, and we obtain

u(Xl, tj = u0e~Xl [V4 (2V'xltl) + ^ VTAl/0(2y7)rfr] , (2-6.96)

where xx — (fix/v), L = (fitly) are dimensionless variables and I0 is the Bessel

function of the first kind of zero-th order with imaginary argument.
If an asymptotic formula is used, for 70, an asymptotic description of
the solution for large values of the argument can easily be obtained.
2. Asymptotic solution. In the above we investigated the absorption
process of a gas for the case of a Henry isotherm. This relates the amount
a of the absorbed gases with the equilibrium concentration by the linear
expression

We now consider a general absorption isotherm

a— f(y) .
If we introduce the dimensionless variables

„ _xfi
X\ — -
_tfi
t1 — и - z = у v =■
г Щ M0 МоГ

then the system of Eqs. (2-6.85), (2-6.89), (2-6.90), and (2-6.91) becomes
eg |

CO
II

dxi dti
(2-6.97)
dv .
—— = (u — z)
dti

v = fi(z) = —f(zuo) (2-6.98)

UoT
with the auxiliary conditions

m(0, t) = 1 (2-6.99)
v(x, 0) = 0 . (2-6.100)

We are interested in the asymptotic forms of

the solutions of (2-6.97).
To this end, with regard to the function
fi(v), we impose the following conditions:
1. Let fi(z) be an increasing function and let
/i(0) = 0.
2. Let fi(z) possess, for all z with 0 ±5 z ^ 1,
a continuous derivative.
3. Let the line leading from the origin of the coordinates to the point
(1, fi (1)) lie for 0 ^ z ^ 1 below the curve fi(z) (Figure 33). This holds in
particular for convex isotherms.
144 HYPERBOLIC DIFFERENTIAL EQUATIONS

For the inverse function corresponding to /1(2) we introduce the relation

z =/л\ѵ) = F(v)

and seek the asymptotic solutions in the form of propagating waves

й = ф(£) v = <p(£) £ = x — at , (2-6.101)

where г is the velocity of propagation of the waves still to be determined.

This means that for large distances (as x —> 00) or for large times (as
t —> 00)

v{x, t) — v = <p(x — at) , й(х, t) — й — ф{х — at) .

The concentration U and a for x = 00 or t = 00 must satisfy the equilibrium

condition

v = /і(м) or и = F(v) .
From (2-6.99) it follows

й|*=о = Ф( — °o) = 1 , <p( — 00) = гф=0 =Л(1) (2-6.102)

and from (2-6.100)

v\x='x, = ^(-foo) = 0, ^( + 00) = mL=o= = F( 0) = 0 . (2-6.103)

It=0 It=0

The condition (2-6.102) states that as i-+oo (f—>—00) saturation must occur
throughout.
If we insert (2-6.101) into (2-6.97), we obtain

ф' = oip' = 0 , (2-6.104)

—ay' — ф — F((p) . (2-6.105)

From (2-6.104) and (2-6.105) we derive the relation

Ф(£) — o<p(£) = 0 . (2-6.106)

Hence it follows from (2-6.102) that

1
(2-6.107)
P(f) ^=-00
Ml)
or (in the dimensionless quantities)

a = Г- (2-6.107')
do
From (2-6.105) and (2-6.106) we find

(2-6.108)
o<p — t(<p)

and hence after integration

<0(9) = € — £0 i (2-6.109)
2-6. APPLICATIONS TO CHAPTER 2 145

where w(p) is an integral of the left side and f0 is the integration constant.
The sought function p(£) is defined by this relation to within the unknown
constant f0 :

P = - fo) (2-6.110)

Ф = оаГ\£ - f0) • (2-6.111)

We want now to determine whether the function oTl is defined at all and
whether the functions p and ф as £ -> oo and f —> — oo for the inverse function
satisfy the required conditions. Therefore we show that

da>
—a < 0 (2-6.112)
dp ^9 — f i \p)
i.e. £ — £o = «(у)

is a monotone decreasing function of <p. For the denominator in (2-6.112) the

following relation holds,

op — fAp) P — f i \p) •

The first summand represents the abscissa of points on the line corresponding
to the ordinate p which runs from the origin of the coordinates to the point
(1, /i(l)), (Figure 33). Since by hypothesis the curve p = /1(2) lies above these
lines, we have

ГАр) < о < p <Ml)

and consequently

op — /1 \p) > 0 .

Moreover,

op — f 11(p) = 0 for p = 0 and p — fi( 1) .

From this it follows that

^ — fo = o)(p) — 00 f°r P — 0
? — f 0 = o){p) = — 00 for P = /i(l) •

For the inverse function we therefore obtain

p= (0 ‘(f — $0) =/,(1) for f = —OO

p = aT‘(£ — <f0) = 0 for f = со .

Further because of (2-6.112) we also find

ф — op 9 = 1 for f = — 00
/i(l)
1
ф = op = p = 0 for f = 00 .
Л(і)
 146 HYPERBOLIC DIFFERENTIAL EQUATIONS

The conditions (2-6.102) and (2-6.103) are therefore satisfied. Besides this it is
shown that our system of equations permits a solution in the form of a pro¬
pagating wave which still contains an arbitrary constant ?0-
In order to determine £0 we integrate the first equation of (2-6.97) from
0 to t0 and 0 to x0:

u(x0, r)dr — j m(0, r)rfrj + |^ v(x, t0)dx — j v{x, 0)<ілГ| = 0 . (2-6.113)

This relation expresses the law of conservation of mass. By passage to the

limit as Xq —> oo we find by using the initial condition for й and v that
Soo C <0
v(x, t0)dx = 1 m(0, r)dr = to .

We now assume that the solution of our problem for large values of t
approximates the functions й and v which we have found above as propagat¬
ing waves.
If then we determine ?0 according to the condition

j v(x, t0)dx — t0—>0 (t0—> oo) (2-6.114)

then this is just the f0-value which corresponds to the functions й(х, t) and
v(x, t).
We form our integral in the following manner:

I v(x, to)dx = j (p(x — ato)dx = j W~'(x — ato — fo)dx

=Г
J—<TtQ—tQ
<o~\№ = Г аГ\Ж J£j

C=% ot0 ?0 Cl == <7^0 ?0 •

We denote by <p* the value of a>_1(C) for £ = 0:

®_1(0) = <p* .
Then it is easy to see that

«r\Qdt: + w_1(C№
fi Jfl
“_1(^ p
Г-Сіа»"‘(С.) + a>(ip)d<p + (2-6.115)

is valid when (p = a> ‘(£) which is the inverse function corresponding to £ = w(tp)
(Figure 34). Hence it follows that instead of the conditions (2-6.114)

( <o~\OdC - to0 =
— -|(<rfo + 6o )<p(—oto — ?o) +
54>{-<Tto
-<r<0—fo)
u)((p)d(p J* — /о —> 0
о
<
J -<rtQ-fo

(2-6.115')

can also be written as t0 —* °°.

2-6. APPLICATIONS TO CHAPTER 2 147

As t —* oo,

<j(p(—ot0 — £o) —<► <np ( — oo) = afi( 1) = 1 .

(2-6.115")

In order to calculate the limit values of

the expression

ot0<p( ato fo) — t0

we use Eq. (2-6.108). By developing

/ \<p) = F(<p) in a series in the neigh-
borhood of the point <p0 = /i(l) we obtain

ocp — F(ip) = a((p — <p0) + 1 — F{(p)

= a(ip — <po) — \F{(p) — F(<p0)\

= [o — E'^V’o)] W — <Po) + • • •,
where
_(Up_
= d£ . (2-6.116)
[<J — F'(<p0)] (<p — <fo) +

The dots here denote the terms of higher order with respect to (<p — cp0).
From the requirement 3 for /, it follows that

1
F'(<p0) > <j
/,(1) '

From (2-6.116), the order of magnitude of <p as £-» — oo results:

ip = AeH + (pB , (2-6.117)

where A and k > 0 are constant.

The expression (2-6.117) gives

lim t0[o<p(-ot0 — c„) — 1] = lim toAoe~k"Tta^) = 0 . (2-6.115"')

tQ—oo Iq .oo

If now in (2-6.115') we still let t0~> oo and bearing in mind (2-6.115") and
(2-6.115'"), we obtain

£o = (2-6.118)

Therefore the profile of the wave (и, v} is completely determined.

Of particular interest is the case of the Langmuir isotherm. For this
we seek now an asymptotic solution of the absorption process.
Eq. (2-6.108) here assumes the form

_(Up_
= dt (2-6.119)
ay — <p/( 1 - pip)
where a — l//i(l) = 1 + p is the wave velocity. From (2-6.119) we find

£ — £o = o>(<p)
with
148 HYPERBOLIC DIFFERENTIAL EQUATIONS

w{(f) = a \ —--^ )d(p- + A = —- Г— ln(ff - 1 - pap) - In y>l 4- A .

J (p — <799(1 — pep) a — \ \_ a J

If (p varies from 0 to /i(l), then obviously w(<p) varies from —00 to + 00.
We choose A so that

i.e.,
* 1 1 1
w(<p ) = 0 for
'—2Л<1,“2 l + p

The conditions for it are

A =-
a
and

w((p) = —1n 2( 1 — op) — In 2(1 + p) .

The value of £0 can be calculated from

1 r/i<n
f°=-7777-1 o)(ip)d<p = -(In 2 — 1)
J\\4 Jo

and does not depend on p = u0ly, the prescribed concentration.

The sought asymptotic solution then has the form

v(x, t) = a)~1(x — at — £0)

(2-6.120)
й(х, t) — aw~1(x — at — £0)

where аГ‘(£) is the inverse function of a»(p).

12 16 20 24 2Ѳ
FIG. 35.
2-6. APPLICATIONS TO CHAPTER 2 149

Figure 35 shows the results of the numerical integration of Eq. (2-6.97)

for the Langmuir isotherm (by means of difference equations). We considered
here the range of values 0 < / ^ L = 10. For t = ty the results of the numerical
integration agree with the asymptotic solution to within 1 per cent. For
t > t{ we can use the asymptotic formula.

6. Physical analogies

In the investigations of phenomena in different branches of physics one

often establishes a common characteristic. This then leads in the mathe¬
matical formulation of the corresponding problems to one and the same
equation which describes simultaneously different physical phenomena. The
following equation serves as a simple example:

d2x , л
a + bx — 0 ,

which describes different vibration processes of a simple system: vibrations

of a mathematical pendulum, a weight under the influence of an elastic
spring, electrical vibrations in a simple conductor with inductance and capa¬
city, etc. The fact that the different physical processes can be described by
the same mathematical equation permits us, on the basis of the investigation
of one of these processes, to reach conclusions about the properties of the
other (less accurately investigated) processes.
The propagation of electric vibrations in systems with distributed con¬
stants can be given by the telegraphic equations

dl _ c ЗУ
+ GV
dx dt
(2-6.121)

— = L — + RI
dx dt
where C, G, L, R are coefficients of capacity, the loss, the induction, and the
resistance of the system. If we can neglect R and G, the voltage V and the
current density / satisfy the ordinary wave equations

d2V _ d2V
=о
dx2 dt2
ЭѴ _ Trll = 0
dx2 dt2

while Eqs. (2-6.121) become

dl =cdV
dx dt (2-6.122)
dV = L dl
dx dt
On the other hand, in the investigation of the propagation of sound in a
unique direction, we arrive—for example, for the motion of air in a tube—
150 HYPERBOLIC DIFFERENTIAL EQUATIONS

at the equations
dp _ dv
dx P dt
(2-6.123)
dv _ 1 dp
dx t dt
where v is the velocity of the vibrating particles, p is the density, p is the
pressure, and г = p0y is the elastic coefficient of air.
The similarity of Eqs. (2-6.122) and (2-6.123) point to the analogy between
the acoustic and the electrical quantities. The potential difference corresponds
to the pressure, the current density to the displacement velocity of the
particles. Further, the induction of the electrical system corresponds to the
density which determines the inertia properties of the gas, and the capacity
corresponds to the quantity 1/r, i.e., the reciprocal of the elastic coefficient.
This analogy can also be determined from the expressions for the kinetic
and potential energy of the electrical and acoustical systems.
By a glance at Eqs. (2-6.121) we can, by analogy with the corresponding
electrical quantities, introduce an acoustic resistance and loss. The magnitude
of the acoustic resistance is then to be considered if, in the motion, the
friction of the gas on the side of the vessel plays an essential role. By
analogy with electrical resistance, which is defined as the ratio of the voltage
to the current density, we define the acoustic resistance as the ratio of the
pressure to the current in the medium which is proportional to the displace¬
ment velocity of the gas particles:

uv
In those cases in which the motion of a gas in a porous medium is considered,
one has to introduce a quantity which corresponds to the loss in an electrical
system. This quantity (which we designate by P) is called the porosity and
is defined per unit volume for those materials which are filled with air.
The mechanical analogy of telegraphic equations are the equations of a
longitudinal vibrating rod. These can be written, similar to Eqs. (2-6.122), as

dv 1 dT
dx k dt
dT dv
dx

where T is the tension of the rod, v is the velocity of the vibrating points,
p is the density, and k is the elastic coefficient of the rod.
By comparison of these equations with Eqs. (2-6.122) we can define further
analogies, this time between mechanical and electrical quantities. If the
electrical voltage can correspond to the tension of the rod and the current
density to the motion velocity of the particles, then the reciprocal of the
elastic coefficient is known to correspond to the capacity and the density to
the induction.
2-6. APPLICATIONS TO CHAPTER 2 151

The consideration of similar dynamic problems leads thus to an analogy

between a series of electrical, acoustic, and mechanical quantities. This is
illustrated in the following table.’7

Electric systems Acoustic systems Mechanical systems

Variables
Voltage V Pressure, p Tension, T
Current density, / Particle velocity, v Particle velocity, x
Charge, e Displacement, и Displacement, x
Parameter
Inductance, L Inertia (density), p Mass density, pm
Capacity, C Acoustic capacity, Rigidity, elastic modulus
CA = 1/r CM = l/k
Resistance, R Acoustic resistance, RA Mechanical resistance, RM

The above table is based on the results of acoustic problems about the
character of the phenomena and gives an insight for the solution of the
problem.
Thus the problem of motion of air in a porous material for simple
harmonic waves leads to the equations38

— i(Dpmu + ru = —grad/)

Ap + (r — iwpm)p = 0
pc

where и is the spatial velocity of air through the pores, p is the pressure,
p is the density, pm is the effective density of air in the pores (pm can be
larger than p, since the material particles and the air can also vibrate in the
pores), P is the porosity, c is the velocity, w is the frequency of sound, and
r is the flow resistance. The latter can be characterized by prescribing the
pressure in the material. If we put r—RA, pm = LA, yPIpc2 = CA, then the
above equations take the form

La— + Rau = -grad p

ot

CalA + CarJ£- = Ap .
ot ot

These equations are completely analogous to the equations for the propaga¬
tion of electrical vibrations in a conductor. Therefore, by analogy with the
wave impedance in conductors,

Z=
/ R T iu)L
G + iwC
G = 0

37 See, for example, H. F. Olson, Dynamic Analogies, New York, Toronto, London,
1944.
39 W. Furduev, Electroacoustics, Moscow, 1948.
152 HYPERBOLIC DIFFERENTIAL EQUATIONS

we can write the expression

Z=

for the so-called characteristic impedence of porous materials. The charac¬

teristic impedence causes a dampening of the waves propagating in a porous
material.
With the help of these analogies between electrical and acoustic pheno¬
mena, the investigation of many acoustic problems can be replaced by the
consideration of equivalent electrical systems. These methods of analogy
have in recent times found many applications with analog computers, in
which an equivalent electrical circuit is constructed for the solution of an
equation which corresponds to a physical process.
 3

PARABOLIC DIFFERENTIAL EQUATIONS

Partial differential equations of the second order of the parabolic type

occur principally in problems connected with heat conduction and diffusion.
The simplest parabolic differential equation

uxx — uy = 0
is usually denoted as the heat-conduction equation.

3-1. SIMPLE PROBLEMS WHICH LEAD TO PARABOLIC DIFFERENTIAL

EQUATIONS

1. The linear problem of heat propagation

We shall consider a homogeneous rod of length l which can be heat
insulated and is sufficiently thin so that at an arbitrary time the temperature
at all points of the cross section can be regarded as equal. If the ends of
the rod are held at the constant temperatures иx and щ, then, as is known,
along the rod a linear temperature distribution occurs (Figure 36):

U(x) = Ul + X . (3-1.1)

Here the heat flows from the warmer

to the colder end of the rod, that is, in
the direction in which the temperature
decreases. The amount of heat which,
in a unit of time, flows through a cross
section of area S is given by the experi¬
mentally determined formula

The coefficient k, the so-called thermal conductivity, depends on the material

of the rod. The magnitude of the heat flow is taken to be positive if the
heat flows in the positive x direction.
The course of the temperature distribution in a rod can be described by
a function u(x, t), which gives the temperature in the cross section at x at
time t. What equation must u(x, t) now satisfy? For the derivation of this
154 PARABOLIC DIFFERENTIAL EQUATIONS

equation we formulate first the physical laws which govern heat conduction.
1. The Fourier law If the temperature of a body is distributed non-
uniformly, then there arises in the body a heat flow in the direction of
temperature decrease.
The amount of heat which flows through the cross section at x during
an interval of time (t, t+dt) is equal to

dQ — qSdt , (3-1.3)
where

q=- k(x) (3-1.4)

ox

is the density of the heat flow. This is equal to the amount of heat flowing
through an area of 1 cm2 per unit of time. This law represents a generali¬
zation of formula (3-1.2). It can also be written in the integral form

Г12 ди
Q=-S\ k—(x,t)dt. (3-1.5)
Jq OX

Then Q is the amount of heat flowing in the time interval (tit t2) through
the cross section at x. If the rod is inhomogeneous, k is a function of x.
2. The amount of heat which must be added to a homogeneous body
in order to raise its temperature by an amount Au, is equal to

Q — cm Au — cpV Au , (3-1.6)

where c is the specific heat capacity, m is the mass of the body, p is its
density, and V is the volume of the body.
If the temperature change differs at different places of the rod or the
rod can be treated as inhomogeneous, then

*2

Q = cpSAu(x)dx . (3-1.7)
q
3. Within a rod, heat can arise or vanish (for example, by the flow of
an electric current, because of chemical reaction, etc.). The occurrence of
heat can be completely described by a function F(x, t), which is a measure
of the heat source at the point x at time f.39 The effect of these sources
on an element of the rod (x, x + dx) during an interval of time (t, t + dt)
induces an amount of heat given by

dQ = SF(x, t)dxdt . (3-1.8)

By integration we find

Q = S f f F{x, t)dxdt , (3-1.9)

J q J*!

39 For example, if heat is produced by an electric current of strength / in a rod,

whose resistance per unit length is equal to R, then F — 0.24- Iz- R.
3-1. SIMPLE PROBLEMS WHICH LEAD TO PARABOLIC DIFFERENTIAL EQUATIONS 155

which is the amount of heat which exists in the interval (xlt x2) of the rod
in the time interval (t%, t2).
We obtain the equation of heat conduction from the law of conservation
of energy for an interval (xt, x2) in a time interval (tt, t2) and with the use
of formulas (3-1.5), (3-1.7), and (3-1.9) for the energy balance, we can write

,du.
-[ k —dudx {x,,
I ,
r)
,
— k—(x, t)
dx >+П“
F($, t) d£ dr
(3-1.10)
= 1 cp[u(£, t2) - «(f, ti)]d£ .

This is the equation of heat conduction in integral form.

In order to derive the differential form we shall assume that the func¬
tion u(x, t) possesses the continuous derivatives uxx and ut.*°
Then from the mean-value theorem of integral calculus we obtain

t) -k^(x, r)| At + F(xt, tt)AxAt

[kfxiX’ ox xijr=(. (3-1.11)
= {cp[u(Z, t2) - M(f, tA]}(=xJx

and hence with the help of the mean-value theorem of differential calculus

Г k ~~ (x, 0 ~| AtAx + F(x4, t4)AxAt = [cp—(x, t)l AxAt , (3-1.12)

dx Ах=хь L dt _]*=,.

where t3, tt, tb and x3, x,, xb are suitable intermediate values in the intervals
(tу, t2) and (xi, x2).
If we divide the last equation by AxAt, we find

d_ du
+ F(x, t) (3-1.13)
dx cpm

All these considerations hold for arbitrary intervals {xi, x2) and (A, t2). Ac¬
cordingly if Xi, x2 tend towards x and , t2 tend towards t, there results

£(* ^
the so-called equation of heat conduction.
We shall next consider a special case.
1. If the rod is homogeneous then k, c, p can be taken as constant, and
the heat-conduction equation takes the form

ut = агихх + f(x, t)

a‘ = ±, /(*,,)= «£-'>■
cp cp

40 Through these restrictions on the function u(x, t), in general, we will lose a class
of solutions, namely those which satisfy the integral equation but not the differential
equation. In the case of the heat conduction equation, however, these requirements do
not exclude any possible solutions. That is, we can show that a function which satisfies
Equation (3-1.10) must be differentiable.
156 PARABOLIC DIFFERENTIAL EQUATIONS

where a2 is the coefficient of temperature conductivity. If there are no

sources present, i.e., if F(x, t) = 0, then the equation simplifies:

ut = a2uxx. (3-1.14')

2. The density of the heat sources are dependent on the temperature.

If heat exchange with the surroundings exists which obeys Newton’s law,
the amount of heat emanating from the rod per unit of length and time
equals

F0 = h{u — Ѳ) ,

where Ѳ{х, t) is the temperature of the surrounding medium and h is the

heat exchange coefficient.41 Therefore the density of the heat sources at the
point x at time t is equal to

F -- F^x, t) — h(u — Ѳ) ,

where Fx{x, t) is the density of other heat sources.

If we are dealing with an inhomogeneous rod, then the heat conduction
equation considering heat exchange with the surroundings takes the form

ut = cfuxx — au + f(x, t) , (3-1.15)

where a = h/cp and f(x, t) = аѲ{х, t) + F^x, t)[c,p is a known function.

3. The coefficients k and c as a rule are slowly varying functions of
temperature. Therefore, from the above cited assumption, these coefficients
can now be assumed as constant, which is valid for small temperature fluc¬
tuations. The consideration of the course of temperature for large tempera¬
ture fluctuations leads to a quasilinear heat-conduction equation, which for an
inhomogeneous medium can be written in the form

Yx(b(u’ x^Yx) + F('X’ ^ = C('U’ X'>p('u’ ^ lit

(see Application 3).

2. Diffusion equation

If а medium is filled nonuniformly with а gas, then a diffusion occurs

from the places of higher concentration to places of lower concentration.
This phenomenon also occurs in solutions when the concentration of the dis¬
solved material is not everywhere equal.
We shall now consider the diffusion in the interior of a hollow tube or
in a tube which is filled with a porous substance, under the assumption that
at any arbitrary moment the concentration of the gas (the solution) in any
cross section of the tube is constant. Then the diffusion process can be de-

41 Since with our approximation the temperature distribution inside an individual cross
section was not considered, the influence of the surface sources is equivalent to the in¬
fluence of the heat sources in the interior.
3-1. SIMPLE PROBLEMS WHICH LEAD TO PARABOLIC DIFFERENTIAL EQUATIONS 157

scribed by a function u{x, t), which gives the concentration in the cross sec¬
tion at x at time t.
According to the Nernst law, the mass of the gas passing through the
cross section at x during the time interval (t, t + At) equals

dQ = - D^(x,t)Sdt , (3-1.16)
ox
where D is the diffusion coefficient and 5 is the area of the tube cross sec¬
tion.
From the definition of concentration,

Q = uV
results for the amount of gas found in the volume V. Here we know that
the change of mass of the gas in the interval (xi, x2) of the tube equals

c{x) Au-Sdx
*1

when the concentration changes by an amount An. Here c(x) denotes the
coefficient of porosity.42
The condition for the conservation of mass in the interval (xi, x2) during
the time interval (ti, t2) reads

S p[£>(*2)|^(*2, t) - D(Xl)yx (Xl, r)Jt/r = 5 U) - M(f, tA]dt.

Hence, as in Section 1 we arrive at the equation

д_ ( и ди\ _ c du
(3-1.17)
йж\ dx) dt ’

the so-called diffusion equation. It is completely analogous to the heat-con¬

duction equation. For its derivation we assume that in the tube no material
sources exist, and no diffusion takes place through the tube walls. If we
take into consideration also the sources, then we arrive at equations which
correspond to Eqs. (3-1.14) and (3-1.15).
If the diffusion coefficient is constant, the diffusion equation assumes the
form

ut — auxx
with

If с = 1 and the diffusion coefficient is constant, the diffusion equation reads

Ul — Dllxx •

42 The porosity coefficient we understand to be the relation between the volume of

the pores and the total volume Vo, which in this case is equal to Sdx.
158 PARABOLIC DIFFERENTIAL EQUATIONS

3. Spatial heat propagation

Spatial heat propagation can be characterized by the temperature u(x, y, z, t)

as a function of the point (x, y, z) and the time t.
If the temperature is not constant, a heat flow occurs which again proceeds
from places of higher temperature to places of lower temperature. The
amount of heat flowing through the surface element da at the point (x, y, z)
during the time interval (t, t + At) is given, according to the Fourier law,
by the formula

dQ — — k — u(x, у, z, t)dadt .
on
Here k is the heat conductivity of the body43 and n is the normal to the
surface element da in the direction of the heat flow. As is known,

ди ди , ч , du , \ , ди , , , лт
— = — cos (n, x) + — cos (и, у) + — cos (n, z) = grad u-N ,
on ox oy oz
so that we can write

dQ — — k grad u-Ndadt

where N is the exterior normal. Hence it follows that the heat flow per unit
of time and area is equal to

qn = q-N (3-1.18)

where q = — k grad и is the vector density of the heat flow.

For the amount of heat which flows through a surface 5 in the time
interval (tlt t2), we have

Qi = - k -~-dt da

or

= —[ f grad u-Ndadt = [ [ \q-Ndadt . (3-1.19)

Jq-taj JqJsJ

Further, the amount of heat necessary to raise the temperature of a point

of the body by

Au(x, y, z) = u{x, y, z, t2) — u(x, y, z, tx)

is equal to

Qi = \^cp[u{£, TJ, c, t2) — w(£, V, C, ty)]dV . (3-1.20)

Finally, if we denote the density of the heat sources by F(x, y, z, t), then
for the amount of heat freed in the volume V in the time interval (t{, t2)
we obtain

43 In this case, a homogeneous isotropic body.

3-1. SIMPLE PROBLEMS WHICH LEAD TO PARABOLIC DIFFERENTIAL EQUATIONS 159

E(f, r), C, t)dVdt . (3-1.21)

We shall now formulate the law of conservation of energy for a volume V

whose surface will be denoted by 5. Obviously, we must have

Q2 — Q3 — (Qi . (3-1.22)

Under the assumption that the function u, in the region considered, with
respect to the variables x, y, z, is twice continuously differentiable, and with
respect to t, is once continuously differentiable, we now form the relations
(3-1.19), (3-1.20), and (3-1.21). By applying Green’s theorem44

s Jr

and the mean-value theorem of integral and differential calculus for functions
of several variables, we obtain

Qi = div qfa, yi, Zi., U) VAt

du
Q г = cp[u(x2, у ,2 Z2, t2) - m(x2 , у ,
2 Z2, fi)] V = cp — (x2, Уг, 22, ti) VAt
ot
Q3 = F(x 3, y3, zs, tb) VAt
or, after division by VAt follows

Cp -І7 {x2, Уг, г2, ti) = - div q(Xi, Уі, zlt t3) + F(x3 ,y3,z3, t3) . (3-1.23)
at

Here all the values of the argument lie within the region considered, i.e.,
they are the coordinates of certain interior points of V for a certain value
of time in the interval (ti, t2).
Equation (3-1.23) holds for any volume V within the body. If we shrink
the volume to the point with coordinates x, y, z, and carry out the passage
to the limit tlf t2-+t, we obtain45

cp — {x, y, z, t) = — div q(x, y, z, t) + F{x, y, z, t) (3-1.24)

at
because of continuity of the derivatives.
If now q is replaced according to (3-1.18), we obtain the heat-conduction
equation

cpiit = div (k grad u) + F

or

(3-1.25)

If the body is homogeneous, we usually write this in the form

44 In this formula as in (3-1.19) the exterior normal is assumed.

45 F(x, y, z, t) is assumed to be continuous in the region considered.
160 PARABOLIC DIFFERENTIAL EQUATIONS

lit - О (lixx T Uyy ~f” Uzz) “Ь , (3-1.26)

cp

where a2 = klcp is the temperature conductivity, or also

ut = агДи + f, f—— , (3-1.26')

cp

where

4 = jL.u JL + A
dx2 з/ a*2
is the Laplace operator.

4. Formulation of boundary-value problems

In order to determine the solution of the heat-conduction equation in a

unique manner in each case we must still consider the initial and boundary
conditions along with the equation itself.
In contrast to the differential equations of the hyperbolic type, only the
initial conditions arise here in the prescription of the values of the function
u(x, t) at the initial time t0-
The boundary conditions can assume different forms, according to the
temperature conditions considered on the boundary. We thus distinguish
three principal types of boundary conditions.
1. At the end of a rod {x = 0 or x = l) the temperature is prescribed,
e.g.,
z*(0, t) = p(t) , (3-1.27)

where p(t) is a function defined in the interval f0 = ^ = T. Here T charac¬

terizes the time interval in which the process is considered.
2. At one end, the value of the derivative is prescribed, e.g.,

= (3-1.28)
ax
We arrive at this condition when the heat flow Q{1, t) occurring at the end
of the rod is given by

Q(l, = t) .
ox
From this there results {du{l, t)dx) = v(t), where v(t) is a known function; then
we obtain

Q(L t)
v{t) =
k
3. At one end a linear relation exists between the derivative and the
function given by
3-1. SIMPLE PROBLEMS WHICH LEAD TO PARABOLIC DIFFERENTIAL EQUATIONS 161

This boundary condition corresponds to Newton’s heat exchange of the surface

of the body with the surroundings whose temperature Ѳ is known. There¬
fore, if for the flow of heat which flows through the cross section at x = l
we use the two expressions
Q - h{u — Ѳ)
and

we obtain the mathematical formulation of the third boundary condition in

the form

-fl-(l,t)=-t[u(l,t)-0(t)\, (3-1.29)
ox

where Л = h/k is the heat-exchange coefficient and 0(t) is a prescribed func¬

tion. For the cross section at x = 0 of the rod (0, l) the third boundary con¬
dition reads

■I-«),*) = + t[u(0,t) - ѲЦ)) . (3-1.29')

ox

Naturally, the boundary conditions for x = 0 and x = l can be different

so that the number of possible boundary conditions is large.
If the system considered is inhomogeneous and the coefficients of the
differential equation are discontinuous functions, we divide up, in a suitable
manner, the interval (0, /) in which the solution is sought, by the points of
discontinuity, and the coefficients into several subintervals, in such a way
that the function и within these subintervals satisfies the heat-conduction
equation and at the points of discontinuity satisfies the corresponding transi¬
tion conditions.
In the simplest case these conditions are the continuity of the tempera¬
ture and the continuity of the heat flow,

u(Xi — 0, t) = u{Xi + 0, t)

k(Xi - оАхі - 0, t) = k(Xi + + 0, t)

ox ox

where x, are the points of discontinuity of the coefficients.

Besides the problems discussed here, limiting cases also arise. We shall
consider the heat conductivity in a very long rod. In the course of a suffi¬
ciently small time interval, then, the influence of the temperature conditions
prescribed at the end points on the middle portions of the rod is very small,
so that the temperature of these parts alone are determined by the initial
temperature distribution. In this case the exact consideration of the rod
length is of no significance, since a change in the rod length has no essential
influence on the temperature of the portions of the rod of interest to us; for
such problems we usually assume the rod to be of infinite length. Thus we
162 PARABOLIC DIFFERENTIAL EQUATIONS

have an initial-value problem (the Cauchy problem) for the temperature dis¬
tribution in an infinite straight line, i.e.:
Find a solution of the heat-conduction equation in the region — oo < x < oo,
t ^ t0, which satisfies the equation

u(x, to) = <p(x), — oo < x < + oo ,

where <p(x) is a prescribed function.
Correspondingly, the temperature in an element of the rod which lies
nearer to one end and far from the other end is determined in practice by
the temperature condition of the near end and the initial conditions. In such
cases, we at least assume the rod to be bounded on one side and the coordi¬
nates from the bounded end to lie within the region defined by O^r^co,
As an example we shall formulate the first boundary-value problem for a
one-sided bounded rod:
Find a solution of the heat-conduction equation in the region 0 < x < oo,
t0 й t which satisfies the conditions
u{x, to) = <p(x) , 0 < X < CO
(3-1.30)
U(0, t) = fi(t) , t ^ t.

where <p(x) and fi(t) are prescribed functions.

The above formulated problems represent limiting cases (degenerate) of
the fundamental boundary-value problems. A different limiting case of the
fundamental boundary-value problem occurs when the exact initial conditions
are not taken into consideration. Obviously, the influence of the initial con¬
ditions on the temperature propagation along a rod weakens in the course of
time. If the time point of interest is sufficiently long from the initial time,
the temperature of the rod is determined primarily by the boundary condi¬
tions, since a change of the initial conditions shows no change in the tem¬
perature condition of the rod (within the limits of accuracy of observation).
In this case we must also assume that the process continues indefinitely and
the effect of the initial conditions has ceased.
In this manner we arrive at boundary-value problems without initial con¬
ditions: Find a solution of the heat-conduction equation for 0 й x й I and
t > — oo which satisfies the conditions

u{0, t) = fiAt)
(3-1.31)
u(l, t) = //,(/) .
According to the nature of the boundary conditions, other types of problems
without initial conditions are also possible. Of great importance is the prob¬
lem without initial conditions for a one-sided bounded rod (l = oo): Find a
solution of the heat-conduction equation for 0<x<°o,t>—oo which satisfies
the condition

u(0, t) = tft) , (3-1.27)

where fi(t) is a prescribed function.

Often one encounters problems without initial conditions but with periodic
3-1. SIMPLE PROBLEMS WHICH LEAD TO PARABOLIC DIFFERENTIAL EQUATIONS 163

boundary conditions,

pit) = A cos wt (3-1.32)

(see Section 3-5 § 1).

Naturally we presume that the temperature of the rod varies after a
sufficiently long time and, similarly, periodically with the same frequency.
However, if the influence of the initial conditions were taken into considera¬
tion exactly, we could never obtain a periodic solution; since the influence
of the initial conditions is continuously decaying, it completely vanishes;
indeed the consideration of such cases, because of unavoidable errors of ob¬
servations, are senseless. The investigation of a periodic solution is analog¬
ous, therefore, to a neglect of the initial conditions.
Finally, the above formulation of the boundary-value problem does not
refer only to equations with constant coefficients. Under the “heat-conduc¬
tion equation” we can subsume each of the equations of the preceding sec¬
tions.
In addition to the linear boundary-value problems cited above we also
have to investigate problems with nonlinear boundary conditions, for example,
of the form

k ^ (0, t) = a[u\0, t) - Ѳ*(0, t)] . (3-1.33)

ox
This boundary condition corresponds to the Stefan-Boltzmann law underlying
heat radiation from the point x = 0 in a medium with the temperature 0(t).
We shall now consider the formulation of boundary value problems in
some detail and begin with the first boundary-value problem for a bounded
region.
A function u(x, t) is called a solution of the first boundary-value problem
if it has the following properties:
1. It is defined and continuous in the closed region

0 йх<>1, to£t£T.

2. It satisfies the heat conduction equation in the region

0 < x < l, t0 < t < T .

3. It satisfies the prescribed initial and boundary conditions, i.e.,

U(X, to) = <p(x) , U(0, t) = , u(l, t) = f*2(t) ,

where <p(x), fii(t) and fi2(t) are continuous functions which satisfy the transi¬
tion conditions

<p(0) = [h{t0) [= w(0, *0)]

and
<р(1) — ^г(^о) [— u(l, t0)]

which are necessary for the continuity of u(x, t) in the closed region.
We consider now the x, t phase plane (Figure 37). In our problem we
164 PARABOLIC DIFFERENTIAL EQUATIONS

seek a function u{x, t), which is defined in the interior of a rectangle ABCD.
This region is already determined by the statement of the problem, since
the course of the heat propagation in the rod 0 й x ^ / during the time in¬
terval t й t = T, in which the heat behavior
of the boundary is known, was already in¬
vestigated. Let t0 = 0; we assume that u(x, t)
D C
satisfies the heat-conduction equation only
for 0 < x < l, 0 < / S Г, i.e., not for t = 0
(the side AB) or for x = 0, x = / (the sides
AD and BC). For / = 0, as well as x = 0
В ..x and x = l, the value of this function is given
directly by the initial and boundary condi-
fig. 37. tions. To require that the heat-conduction
equation, for example, be satisfied also for
/ = 0 would imply that the derivative <p" = uxx(x, 0) in this equation exists.
Therefore, the generality of the physical phenomena to be investigated is
limited, and thus the basic functions which do not satisfy this requirement
are eliminated from consideration. The condition (3-1.3) loses its meaning
when it is not required that u(x, t) in the region 0 ^ x ^ l, 0 ^ t ^ T (i.e., in
the closed rectangle ABCD) be continuous or this requirement must be replaced
by another appropriate assumption.46 To understand the significance of this
requirement we consider the function v(x, t) defined by the following condi¬
tions:
v(x, t) = C , 0 < x < l, 0 < t й T
v(x, 0) = cp(x) , 0 й x й /
v(0, t) = цМ) , v(l, t) = [і,Ц) , 0 йТ
where C is an arbitrary constant. The function v obviously satisfies both
condition (3-1.2) and the boundary conditions. However, this function in no
case describes the course of the heat distribution in the rod with an initial
temperature <p(x) Ф C and boundary temperatures ^,(/) Ф C and f*2(t) Ф C, since
it is discontinuous for t = 0, x = 0, x = /.
The continuity of u{x, t) for 0<x<l,0<t<T directly follows in that
u(x, t) satisfies the differential equation. Therefore, the requirement that
u(x, t) be continuous in 0 x й /, 0 ^ / й T, is based essentially only on those
points at which the boundary and the initial values are prescribed. In the
following, by a solution of the equation which satisfies the boundary condi¬
tions, we shall always mean a function which satisfies the requirements (3-1.1),
(3-1.2), and (3-1.3) and hence not repeat these each time, unless there are
special conditions.
Correspondingly, this is the case for other boundary-value problems, in
particular for problems of an infinite rod and problems without initial condi¬
tions.

46 Later, boundary-value problems with discontinuous boundary and initial conditions

will be considered. For these, the problems will be properly defined so that the boundary
conditions are fulfilled.
3-1. SIMPLE PROBLEMS WHICH LEAD TO PARABOLIC DIFFERENTIAL EQUATIONS 165

For problems with several independent geometric variables the above

statements remain valid. In these problems, an initial temperature and
boundary conditions determined on the surface of the body are prescribed
for / = 0. We can also investigate problems for infinite domains.
With regard to all the problems discussed, the following problems exist* :
1. Are the solutions of the problems discussed uniquely determined?
2. Does a solution exist?
3. Do the solutions depend continuously on the auxiliary conditions?
If a problem admits of many solutions, then we naturally cannot speak
of “the solution of the problem,” and we must first prove the uniqueness.
In practice, the second question above is the most important, since generally
in proving the existence of a solution, we simultaneously find methods for
its calculation.
As noted earlier (see Section 2-2 §3) we speak of a physically determined
process when a small change in the initial or boundary conditions causes a
small change in the solution. In the following, it will be shown that heat
propagation is determined physically by the initial and boundary conditions,
i.e., a small change in the initial or boundary conditions implies a small change
in the solution.

5. The principle of the maximum

In the following we shall investigate differential equations with constant

coefficients,

vt = агѵхх + flvx + yv . (3-1.34)

As already shown, these equations, by the substitution of

v = e*x+Ktu with Ц = — , X — у —j—f

Za Aa

can be brought to the form

ut = a2uIX . (3-1.35)

The solutions of this equation have the following properties which will
be denoted as the principle of the maximum.
A function u(x, t) defined and continuous in the closed region 0 ь / T,
0 x й I and satisfying the heat-conduct ion equation
Ut = a2uxx (3-1.35)

in the region 0<t<T,0<x<l assumes its maximum or minimum at the

initial moment t = 0 or at the boundary points x = 0 or x = /.
Before we prove this, note that the function u(x, 1) - const, obviously
satisfies the heat-conduction equation and assumes a maximum (minimum) at
each point. However, this does not contradict our assertion, because it means
only that when a maximum (minimum) is assumed in the interior of the region
it is also (but not only) assumed for t = 0 or for x = 0 or .v /.

47 Cf. Section 2-2.

166 PARABOLIC DIFFERENTIAL EQUATIONS

The physical significance of this statement is immediately clear: if the

temperature on the boundary and at the initial moment does not exceed a
value M, then in the interior of the body no temperature higher than M can
be attained. We shall limit ourselves to the proof of the statement of the
maximum and give an indirect proof. We shall designate by M the maxi¬
mum value of u(x, t) for t = 0 (0 ^ x S l) or for r = 0 or т = / (0 ^ 1 g Г) and
assume that the function u(x, t) assumes its maximum at an interior point
(x0, t0), (0 < x0 < l, 0 < to ^ T):*s
u{xо , to) = M + £ .
We now compare the signs in Eq. (3-1.35) at the point (x0, t0). Since the
function at (x0, t0) assumes its maximum,49 then necessarily

zr(x0,ta) = 0 and pL(Xoft0)gO. (3-1.36)

ox ox
Also, since u(x0, t) for t = t0 has a maximum,50 then

^(Xo,to)^0. (3-1.37)
ot
By comparison of the signs on the left and right sides of (3-1.35) it follows
that both sides can be different. These considerations, however, still do not
prove the correctness of our theorem; since the right and the left sides can
simultaneously equal zero, it would signify no contradiction. We bring forth
this consideration simply to emphasize the fundamental concepts of our proof.
For the completion of the proof we shall seek more than one point (xi, ti)
at which d2u/dx2 ^ 0 and du/dt > 0. Therefore, we consider the auxiliary
function
v(x, t) = u(x, t) + k{to — t) , (3-1.38)
where k is a constant. Obviously then

v(x0 , to) = u[xо , to) = M + £

and
k(t0 — t) kT .
48 If the continuity of u(x, t) were assumed in the bounded region
then the function u(x, t) could not exceed its maximum, and further considerations would
be contradictory. On the basis of the theorem that every continuous function in a bounded
region attains its maximum, then (a) the function u(x, t) attains a maximum within or
on the boundaries which will be denoted by M; (b) if u{x, t) also were to exceed M only
at a point, then a point (xo, U) would exist at which the function u(x, t) assumes a maxi¬
mum which is larger than M: u(x0, to) = M + e (e > 0), where 0 < x0 < l, 0 < to ^ T.
49 As is known from analysis, for the existence of a relative minimum of a function
f(x) at an interior point x0 of an interval (0, l), the conditions

df dV
= 0, > о
dx dx2 x=x0
are sufficient. If, therefore, at the point xo the function /(x) has a maximum value, then
(a) /'(xo) = 0, and (6) /"(xo) > 0 cannot hold; therefore /"(xo) = 0.
50 Obviously, du/dt = 0, in case to < T, whereas for to = T, then du/dt — 0 must hold.
3-1. SIMPLE PROBLEMS WHICH LEAD TO PARABOLIC DIFFERENTIAL EQUATIONS 167

We now select k > 0 so that kT < e/2, i.e., let k < e/2T; then the maximum
of v(x, t) for 1 = 0 or for x = 0, x = / does not exceed the value M+e/2,
i.e.,

v(x, t) й M + ~ for / = 0 or x = 0, x = / , (3-1.39)

since for this argument the first summand of (3-1.38) is not larger than M,
and the second is not larger than e/2.
Now, v(x, t) is a continuous function. Thus a point (jq, ti) exists at which
it assumes its maximum. Then we have

v(Xi, /1) S; v{x0, t0) = M + s .

Therefore, tx > 0 and 0 < xt < l, since for t = 0 or x = 0, x = l the inequality
(3-1.39) is valid. It follows that

vxx(Xi, ti) = uxx(xi, /i) ^ 0

and
Vt(xi, ti) = ut{Xi, ti) — k ^ 0 or ut(Xi, ^i) ^ k > 0 .

By comparison of the signs on the right and the left sides in (3-1.35) at the
point (xi, ti) we conclude that Eq. (3-1.35) at the point (xlt ti) cannot be satis¬
fied, since the quantities on the right and left sides have different signs.
Therefore, the first part of our proposition is proved. The statement for the
minimum can be proved analogously, and it is sufficient to apply the first
part to Mi = — u.

6. The uniqueness theorem

We turn now to a series of consequences of the principle of the maxi¬

mum. First, we prove the uniqueness theorem for the first boundary-value
problem. If the functions Ui(x, t) and иг(х, t), which are defined and continu¬
ous in a region 0^x^l,0^t^T, and which satisfy the heat-conduction
equation

ut = агиІХ + /(*, t) for 0 < x < l, t > 0 (3-1.35')

as well as the same initial and boundary conditions

Ui(x, 0) = u2(x, 0) = <p(x)

Ui(0, t) - m2(0, t) = Hi{t)
Ui(l, t) = u2(l, t) = fl2{t) ,

then necessarily51

Ui(X, t) = иг(х, t) .
For the proof of this theorem we consider the function

51 Previously this theorem was refined and the continuity requirement at t = 0 was
dropped.
168 PARABOLIC DIFFERENTIAL EQUATIONS

v(x, t) = U2(X, t) — Ui(X, t) .

Since Ui(x, t) and u2(x, t) for

О йх^І, О^ійТ

are continuous, their difference v(x, t) in the same region is continuous. Further,
v(x, t) as the difference of two solutions of the heat-conduction equation for
0<x<l,t>0 is similarly a solution of the heat-conduction equation in that
region. Consequently, the principle of the maximum can also be applied to
this function, and the maximum and the minimum of v(x, t) for t — 0 or x = 0
or x = l is assumed. According to the hypothesis we obtain

v(x, 0) = 0 , v(0, t) — 0 , v(l, t) = 0 .

Therefore, also

v(x, t) = 0 ,
i.e.
Ui{x, t) = u2(x, t) ,

from which the uniqueness of the solution of the first boundary-value problem
follows.
We shall now prove a series of direct conclusions from the principle of
the maximum. In the following discussion We shall refer to “the solution
of the heat-conduction equation,” instead of enumerating the properties of
the function in detail which also satisfy the initial and boundary conditions.
1. If two solutions Ui(x, t) and u2{x, t) of the heat-conduction equation
satisfy the conditions

Uy{x, 0) 5S U2(x, 0) , Ui{0, t) ^ щ{0, t) , Mi(/, t) u2(l, t) ,

then

uy{x, t) 5S u2{x, t)

for all 0£x£l,0£t£T.

The difference v(x, t) — u2(x, t) — u^x, t) satisfies the conditions on which
the principal of the maximum is based; also

v(x, 0)^0 v(0, t) ^ 0 v(l, t) ^ 0 .

Therefore

v(x, t) ^ 0 for 0 < x < l, 0<t£T,

since v(x, t) in the region

0<x<l, 0<tST

would otherwise have a negative value.

2. If three solutions

u(x, t) , u(x, t) , u(x, t)

of the heat-conduction equation satisfy the conditions
3-1. SIMPLE PROBLEMS WHICH LEAD TO PARABOLIC DIFFERENTIAL EQUATIONS 169

u{x, t) < u{x, t) s' u(x, t) for f — 0 , X — 0 , X= l ,

then this inequality is fulfilled for all x in 0 g x ^ / and all f in 0 ^ t 52 T.
This assertion represents an application of conclusion (1) to the func¬
tions

u{x, t) , U(x, t) and u{x, t) , u{x, t) .

3. If, for two solutions Ui{x, t) and u2(x, t) of the heat conduction equa¬
tion, the inequality

I Ui(x, t) — u2(x, t) I S £ , for t =0 , x=0 , x — l

is valid, then

I Ui(x, t) — u2(x, t) I -й e
for all x, t in

Ъйх-йі, ОйійТ
is satisfied.
This assertion results from conclusion (2), when we apply the heat-con¬
duction equation to the solutions

u(x, t) = — e
u{x, t) = Hi(x, t) — u2(x, t)
й(х, t) = e .

The question regarding the continuous dependence of the solution of the

first boundary-value problem on the initial and boundary conditions is an¬
swered completely by conclusion (3). To understand this, we consider a solu¬
tion u{x, t) which satisfies other initial and boundary conditions, instead of
the solution of the heat-conduction equation which corresponds to the initial
and boundary conditions

u(x, 0) = <p(x) , u(0, t) = ЦіЦ) , u(l, t) = fi2(t) .

Let these be given by functions <p*(x), and /it{t) which differ by less than
e from the functions <p{x), Ці(/), and pi2(t):

I <p(x) - <p*(x) \йе, I цт - /it it) I ^ , I Mt) - I ^ e .

However, the function Ui(x, t) according to conclusion (3) differs by less than
e from the function u(x, t):

I u(x, t) — Mi(x, t) I g £ .

Here the principle of the physical determination of a problem arises directly.

We have investigated in detail the question of the uniqueness and the
physical determination of a problem in the case of the first boundary-value
problem for a bounded interval. The uniqueness theorem for the first boundary-
value problem for a two- or three-dimensional bounded region can be proven
by a verbatim repetition of these deliberations.
Similar questions arise in the investigation of other problems, an entire
170 PARABOLIC DIFFERENTIAL EQUATIONS

series of which was discussed in the preceding paragraphs. These problems,

however, require certain changes in the method of proof.
The solution of the problem for an unbounded region (see Section 7) or
a problem without initial conditions is uniquely determined only if the sought
functions are still subjected to certain auxiliary conditions.

7. The uniqueness theorem for the infinite straight line

For the solution of а problem on the infinite straight line it is essential

to require the boundedness of the sought function in the entire region, i.e.,
there exists an M such that | u{x, t) \ < M holds for all — oo < x < + °o and
f = 0.
If Ui(x, t) and u2(x, t) are two continuous and bounded functions for all
values of the variables x and t considered, and if they satisfy the heat-con¬
duction equation

ut = a2uxx , — со <. x < со, t > 0 (3-1.35)

and the conditions

Ui(X, 0) = u2{x, 0) , — oo < x < со

then
Ui(x, t) = u2{x, t) , — оо<дг<оо, / 2; 0 .

We consider, as before, the difference

v(x, t) = Ui(X, t) — u2(x, t) .

The function v(x, t) is continuous, satisfies the heat-conduction equation, is
bounded in the entire region

I v(x, /)|^| Ui{x, t) I + I U2(x, t) I < 2M , — co<X<oo,teZ0,

and satisfies the condition
v(x, 0) = 0 .
The principle of the maximum, which was used for the proof of unique¬
ness for the case of a bounded interval, is not directly applicable here, since
the function v(x, t) in an unbounded region need not assume its maximum
at any point. In order to use this principle we consider a region

\ x\ й L ,
where L is a parameter which is permitted to increase unboundedly, and a
function

v(x, *) = 7т-(у + л)> (3-1-40)

which is continuous, satisfies the heat-conduction equation (which we can

easily verify by differentiation), and also satisfies the inequalities

V{x, 0) ^ I v(x, 0) I = 0
V(± L, t)^2M^ y(± L, t) .
3-2. THE METHOD OF SEPARATION OF VARIABLES 171

If we apply the principle of maximum to the region | x | ^ L, we obtain

m (x2
+ «** ) = v(x> *) =
L2 v2
+ a2t (3-1.41)

If we now consider any fixed pair of values (x, t) and let L increase un¬
boundedly, we obtain

v(x, 0 = 0,
and thus our theorem is proved.

3-2. THE METHOD OF SEPARATION OF VARIABLES

1. The homogeneous boundary-value problem

We turn now to the solution of the first boundary-value problem for the
heat-conduction equation

lit — a2uxx + f(x, t) (3-2.1)

with the initial conditions

u(x, 0) = ip(x) (3-2.2)

and the boundary conditions

u(0, t) = fit(t) , u(l, 0 = ц2Ц) . (3-2.3)

For the investigation of the general first boundary-value problem we begin

with the solution for the simplest case.
Problem 1. Find the solution of the homogeneous differential equation

ut = a2uxx , (3-2.4)

which satisfies the initial conditions

u{x, 0) = <p(x) (3-2.2)

and the homogeneous boundary conditions

u(0, /) = 0, u(l, 0 = 0. (3-2.5)

For the solution of this problem we shall first consider, as is customary

in the method of separation of variables, the following general auxiliary
problem:
Find a nonidentically vanishing solution of the differential equation

lit — CL Uxx у

which satisfies the homogeneous boundary conditions

u{0, 0 = 0, u(l, 0 = 0 (3-2.5)

and is representable in the form

u(x, 0 = X(x)T(t) , (3-2.6)

where X(x) is a function dependent only on x and T(t) depends only on t.

172 PARABOLIC DIFFERENTIAL EQUATIONS

If we put (3-2.6) into (3-2.4) and divide both sides of the equation so ob¬
tained by a2XT, then we obtain

±Г X^_
Л , (3-2.7)
a2 T X
where X = const., since the left side of the equation depends only upon t and
the right only on x.
Hence, we obtain the equations

X" + XX=0 (3-2.8)

T' + a2XT = 0 , (3-2.8')

whereas the boundary conditions (3-2.5) become

X(0) = 0 , X(l) = 0 . (3-2.9)

For the determination of X(x) we obtain the eigenvalue problem

X" + XX=0, X(0) = 0, X(/) = 0, (3-2.10)

which has already been investigated for the solution of the wave equation
(Section 2-3 § 1), where it was shown that only for the value of the parameters

Xn = (jj , n = 1, 2, 3, • • • (3-2.11)

do nontrivial solutions of Eq. (3-2.8) exist which are then equal to

Xn(x) = sin — x . (3-2.12)

These values Xn correspond for Eq. (3-2.8') to the solutions

Tn(t) = Cne-^ , (3-2.13)

where C„ are still arbitrary constants.

For the auxiliary problem, therefore, the functions

un(x, t) = Xn(x)Tnit) = Cne~a2л*( sin yi (3-2.14)

are particular integrals of Eq. (3-2.4) which satisfy the homogeneous boundary
conditions (3-2.5).
To solve Problem 1, we formally construct the series

u(x, t) = f Cne-^'M sin^x . (3-2.15)

The function u{x, t) satisfies the boundary conditions since these are satisfied
by every member of the series. If the initial conditions are to be satisfied
also, then we must have

<p(x) = u(x, 0) = J Cn sin x , (3-2.16)

n=1 l

i.e., Cn are the Fourier coefficients of the function <p{x) when these are de-
3-2. THE METHOD OF SERARATION OF VARIABLES 173

veloped in the interval (0, l) in a sine series:

Cn = <pn = W <p{£) sin . (3-2.17)

i Jo l

The coefficients Cn of the series (3-2.15) must now have the value defined
by (3-2.17). To show that this series then satisfies all the conditions of
Problem 1, we must prove that the function defined by (3-2.15) is differenti¬
able, satisfies Eq. (3-2.4) in the region 0 < x < l, t > 0, and at the boundary
points of this region (for t = 0, x = 0 and x = l) is continuous.
Because of the linearity of Eq. (3-2.4), a series consisting of particular
integrals, according to the superposition principle, is also a solution if this
series converges and is termwise twice differentiable with respect to x and
once with respect to t (see the auxiliary theorem in Section 2-3 § 3). We
show next that the series arising by termwise differentiation

У dun У d2un
and
»=i dt »=i dx2

for f ^ f >0 (1 isa fixed number) converges uniformly. Therefore, we con¬

sider the expression

diin KYI '

7Г \ 2 -- 2 (.„/п2я2» • _
p—\1Zn/ ll a t
Cn( —j- 1 a'n'e < |C„|(y ) -a n e~isn/l) a t
dt

In the following we shall formulate the additional conditions which the func¬
tion <p(x) must satisfy. First, let <p(x) be bounded | <p{x) \ < M\ then

Cn I = rt£) sin^e# < 2 M,

from which

dtin
< 2M(-j) for t^t ,
dt

and correspondingly there follows

d Un
< 2M(-j) пге~^п/1)2аЧ for t^t .
dx2

In general, we find
2 k+l
dk+lun < 2 M(y) пгк+Іагке-<.пп/1) 2e2t for t si t
dtkdxl

The convergence of the majorant San with

an = Nnqe-Un,l)2a2i (3-2.15')

results from the D’Alembert criterion, since

(n 4- 1 \q o~Стг/i) za2(nz+2n+l) t
&n+1 — lim ( Ц-) e~{*/l) a {2n+l)t = 0 .
lim lim iHJZ-LL-X.- = 1І
lim
П—*oo 71—oo Yfi g-(jT/I) 2a2n27
n
174 PARABOLIC DIFFERENTIAL EQUATIONS

Hence, it follows that the series (3-2.15) for t ^ t > 0 is arbitrarily often term-
wise differentiable. Further, we conclude from the superposition principle
that the function defined by this series satisfies Eq. (3-2.4). Since / is arbi¬
trary it holds for all t > 0. Consequently, the series (3-2.15) for t > 0 represents
a function which is arbitrarily often differentiable and satisfies Eq. (3-2.4).52
If the function <p(x) is continuous, possesses piecewise continuous deriva¬
tives, and satisfies the conditions $o(0) = 0 and <p{l) = 0, then the series

u(x, t) = I Спе~|ягп/!| sin ^ x (3-2.15)

defines a continuous function for t 0.

From the inequality

I un(x, t) I < I Cn I for t Ш 0, 0 ^ x й I

follows directly the uniform convergence of the series (3-2.15) for / ^ 0, 0 ^ x fs /.

Since for a continuous and piecewise smooth function tp(x) the series of the
absolute values of the Fourier coefficients converges uniformly in case
y>(0) = cp(l) = 0, we have proved the proposition.53 We have, therefore, completely
solved the first boundary-value problem for the homogeneous equation with
homogeneous boundary conditions with continuous, piecewise smooth initial
conditions.

2. Green’s function

We turn to Eq. (3-2.15) obtained above in which we introduce the corre¬

sponding values for C„:

u(x, t) = 5 Cne~Un,l) 2“2‘ sin x

n=1 l

= i r^tW^in^r?
» = l|_ l Jo
/
dp[-e^«n/Ms-m™x
J /

= (T4
JоL i
ii r("/1,!*21 sin^x-sin^fly>(f)J?
/ / J
.

Summation and integration can be interchanged for t > 0, since the series
for t > 0 in the brackets converges uniformly with respect to f.54
We now set

G(x,£, 0=45 е-(и/,,!Л sin^x-sin^. (3-2.18)

I n=l l l

With the use of this function G(x, t), u{x, t) can be written in the form

52 In proving that the series (3-2.15) satisfies the equation щ — a2uzx for t > 0, only
the boundedness of the Fourier coefficients Cn was used. This, however, is the case for
bounded <p(x).
53 See Section 2-3 § 3.
54 The series Ian, where an is taken according to formula (3-2.15'), represents a
majorant corresponding to the series, standing for q = 0, in the brackets.
3-2. THE METHOD OF SEPARATION OF VARIABLES 175

u(x, t) = j'GOt, f, t)<p(Z)dt . (3-2.19)

The function G(x, f, /) is known as Green’s function (source function). We

shall investigate its significance.
We shall show that G(x, f, f), considered as a function of x, represents
the temperature distribution in the rod 0 ^ x 5S l at time t, provided (1) the
temperature at the initial moment t = 0 is equal to 0, and (2) at this time
at the point x = $ a certain amount of heat (whose magnitude we shall deter¬
mine later) is free while at the boundary (x = 0, x = / of the rod) in the course
of the total process the temperature remains equal to 0.
By the term “amount of heat Q free at the point £” we shall mean as
usual that an amount of heat exists which is freed in a sufficiently small
interval about the point $ considered. The temperature change <pe(£) which
is produced by the appearance of the amount of heat about the point £ outside
of the interval (f — e, f + e) is equal to zero; by contrast <pe(£) inside of this
interval can be regarded as a positive, continuous, and differentiable function
and is such that

cp[+tM$)de = Q , (3-2.20)
Jf-e

since the left side of this equation precisely represents the amount of heat
producing the temperature change of amount ^8(f). The temperature distri¬
bution in this case is given by formula (3-2.19)

ut(x, f) = ( G(x, $, t)(ps($)df . (3-2.21)

Now we let e tend toward 0. From the continuity of G(x, f, t) for t > 0
as well as Eq. (3-2.20) we obtain next by use of the mean-value theorem for
fixed x, t

«.(*, t) = Г+tG(x, $, t)<pe(Z)dt = G(x, e, t)

J f—* Jf-e
= G(x, f*, o— ,
Cp

(3-2.21')

where is a determined point in the interval (f — e, f + e) and the integral

of <pt(£) exists as e-»0 and is equal to Q/cp. Because of the continuity of
G(x, f, t) with respect to £ for t > 0 there then results

lim u,(x, t) = Q-G(x, ?,/)=—• 4 І sin^x-sin^f . (3-2.22)

e—0 Cp Cp l n=l I I

Hence, it follows that G(x, f, t) represents the influence of the temperature

of a heat pole of intensity Q — cp occurring instantaneously, which is found
at time t — 0 at the point £ in the interval (0, /).
We shall now prove the following property of the function G(x, f, t): For
arbitrary x, f and t > 0, G(x, f, t) ^ 0. For the proof we consider the initially
given function (pe(x) with the above stated properties and the corresponding
solution (3-2.21). Since the initial conditions and the boundary conditions are
176 PARABOLIC DIFFERENTIAL EQUATIONS

nonnegative, it follows from the maximum principle that

ме(x, t) ^ 0

for all 0 iS x g l and t > 0. Hence, it follows by consideration of (3-2.21)

Ug(x, t) = G(x, f*, t)Q- ^ о for t> о (З-2.21")

ср
as e —» 0, and from (3-2.21') we obtain

G{x, ?, t) ^ 0 for O^x, f iS /, t > 0 .

This result has a simple physical significance which, however, can be
recognized directly only from (3-2.19), since G(x, f, t) can be represented there
by an alternating series.

3. Boundary-value problems with discontinuous initial conditions

The theory treated above was based on solutions of the heat-conduction

equation which are continuous in a bounded region This
requirement is a significant restriction. We consider, for example, the simple
problem of the cooling of a uniformly heated rod on whose boundaries the
temperature is equal to 0. Here the auxiliary conditions read

u(x, 0) = u0 , u(0, t) = u{l, t) = 0 .

If u0 Ф 0 then the solution of this problem at the points (0, 0) and (0, l) must
be discontinuous. This example shows that the continuity of the initial con¬
dition required above and its compatibility as defined above are caused by
the boundary conditions—a fact which, in practice, excludes very important
cases from consideration. However, formula (3-2.19) also yields the solution
of the boundary-value problem in this case.
If we wish to use the results of the above theory without neglecting its
domain of applicability, we must concern ourselves with an extension of this
theory that also encompasses the fundamental problems. There are numer¬
ous formulas used in applications outside their domains of validity in addition
to those which are, in general, stated according to the conditions for their
applicability. The consistent basis of all formulas would be too time con¬
suming and would deviate from the quantitative and qualitative aspects of
those processes which are characteristic for these physical methods.
On the other hand, we retain what is necessary, at least with respect to
the simplest examples, to give a basis of the mathematical apparatus suffi¬
cient for the solution of the fundamental problem.
We shall consider boundary-value problems with piecewise continuous
initial conditions without assuming that the initial function in the above-defined
sense is compatible with the boundary conditions. This class of auxiliary
conditions is, in practice, sufficiently general and for the explanation of the
theory sufficiently simple. Our goal, therefore, is to show that formula (3-2.19)
still gives the solution of the problem described. The necessary investigations
will be carried out in single steps. First, we shall prove the theorem:
3-2. THE METHOD OF SEPARATION OF VARIABLES 177

The solution of the heat-conduction equation

ut = a2uxx , 0 < x < l, t > 0 , (3-2.4)

which is continuous in the closed region 0 S t й T and satisfies the

conditions

u{0, t) = u{l, t) = 0 (3-2.5)

uix, 0) = <p(x) (3-2.2)

where <p{x) is an arbitrary continuous function vanishing at x = 0 and x = l,

is determined uniquely and is represented by

u{x, t) = ^ G(x, £, t)<p{£)d£ . (3-2.19)

We have already proved this theorem under the assumption that y>(f)
possesses a piecewise continuous derivative.
We shall now prove the theorem without this assumption. Therefore
we consider a sequence of piecewise continuous differentiable functions <pn(x),
<pn{0) = <pn(l) = 0, which converges uniformly towards <pix), since (pn(x), for
example, can be chosen as the function which represents the step function
which coincides with <p(x) at the points l-k/n, k = 0, 1, 2, • • •, n. The func¬
tions un{x, t) defined by (3-2.19) through <pn(x) then satisfy all the assumptions
of the theorem, since the <pn{x) are piecewise differentiable. The functions
ujx, t) converge uniformly towards a continuous limit function uix, t). To
each e —> 0, therefore, we can find an nie) such that

I ?п^х) — <Pn2(x) I < e , о^Xй l

when Пі, n2 ^ п(е), since these functions, by hypothesis converge uniformly.

From this, on the basis of the principle of the maximum it also follows that

I Uniix, t) - Un2ix, t) I < e , O^x^l, 0

when Wi, w2 ^ nie). Therefore, the uniform convergence of a sequence of
functions Unix, t) toward a continuous limit function uix, t) is demonstrated.
If now for fixed x and t we pass to the limit under the integral sign, it
follows that the function

uix, t) = lim unix, t) - lim ( Gix, ?, t)(pni£)d£ = [ Gix, 6, t)<pi£)d£

n—*°° n-»oo Jo Jo

exists in the closed region 0^x^l,0^t^T, is continuous, and satisfies

the conditions (3-2.2). Referring to footnote 52 we see that this function also
satisfies Eq. (3-2.4) and hence the theorem is proved.
As the following deliberations will show, the function uix, t) defined by
(3-2.19) is a uniquely determined continuous solution of our problem.
We turn now to the proof of the uniqueness theorem for the case of a
piecewise continuous initial function <p(x) without assuming that this function
is compatible with the boundary conditions and we prove:
A function which is continuous in the region t > 0 and for 0 < л; < /, t > 0
178 PARABOLIC DIFFERENTIAL EQUATIONS

satisfies the heat-conduction equation

lit - ^ Hxx f (3-2.4)

the homogeneous boundary conditions

о
(3-2.5)

II

II
and the initial condition
u(x, 0) = <p(x) (3-2.2)

is uniquely determined when

1. It is continuous at the points of continuity of the function <p(x)]
2. It is bounded in the closed region 0 ^ x < l, 0 ^ t й t0 (t0 is an arbitrary
positive number).
We assume that such a function exists, and on the basis of the preceding
theorem for t > t this can be represented by

u(x, t) = j G(x, S, t - t)<p-t(?)de , t>t> 0 (3-2.19')

for arbitrary t (0 < t й t, <p~t(x) = u(x, t)).

In (3-2.19') we now carry out the passage to the limit t—> 0 where x and
t are fixed. We shall show55 that the passage to the limit under the integral
sign is possible, and therefore the function u(x, t) is represented uniquely by

u(x, t) = JW £, , ?>(£) = u($, 0) . (3-2.19)

Let Xi, x2, • • •, xn be the points of discontinuity of the function <p(x).

Then, if we set x0 = 0 and xn+l = l (Figure 38) and consider the closed intervals

X
"V L>
хк~д
4J К
V*5 V.
fv
(У
,_<5
1 J
1 1
у - 7

FIG. 38.

Ik (Xk + S ^ x ^ Xk+i — 5), k = 0, 1, ■ • •, n, where 5 is an arbitrarily small posi¬

tive number, then we know that the integrand in (3-2.19') in each of Д, k =

55 The theorem proven in the following is a special case of a theorem of Lebesgue,

which states that the passage to the limit under the integral sign is possible in case
the sequence of functions Fn(x) converges almost everywhere toward a summable limit
function F(x) and if this sequence is bounded by a summable function. This proof can
be carried out without the use of the notions of measure theory. If measure theory is
used, one can prove completely analogously that a solution of the heat-conduction equa¬
tion u(x, t) which satisfies the homogeneous boundary conditions is defined uniquely when
the following conditions hold: (1) u(x, t) ;S F(x), where F(x) is a summable function; (2)
if, almost everywhere

lim u(x, t) = <p(x) ,

i—о

where <p(x) is a prescribed summable initial function.

3-2. THE METHOD OF SEPARATION OF VARIABLES 179

0, 1, 2, • ■ •, n, converges uniformly towards the integrand in (3-2.19). In the

,
intervals /* (xk — d ^ x ^ Xk + 8), k = 1 2, • • ■, n, on the other hand, the in¬
tegrands of (3-2.19) and (3-2.19') are bounded by a fixed number M for each
t (0 й t 10), since u(x, t) was assumed to be bounded and G(x, f, t), for 0 g
f g /, t > 0, is continuous. If we split up the difference of the integrals (3-2.19)
and (3-2.19') into the 2n + 3 integrals which correspond to the intervals Ik,
k = 0, 1, • • •, n, and h, k = 0, 1, • • •, n + 1, then we see that this difference
can be made smaller than an arbitrary prescribed number e if

° = 2n + 3 AN
so that

_ [G(x, t - t)<pi(£) - G(x, f, t)<p(?)]d£ <

ik 2n + 3 ’
and if t is sufficiently small so that

I G(x, e, t - t)cpm - G(X, tw) I

= 7T^T for *= t in Ik , k = 0,1, • • •,«.
I 2n + 3
Therefore,

G(x, f, / - tfatf) ~ G(x, f, f)p(f) I d$

Hk
< for t ^ t , k — 0, 1, • • •, n .
2n + 3
Hence follows the inequality

l[G(x, 5, t - - G(x, f, t)(p($)}d£ < e for tS t .

Therefore, the passage to the limit under the integral signs is permissible,
and if a function u(x, t) exists which satisfies the assumptions of our theorem,
it can be represented by the formula (3-2.19), from which also the unique de¬
termination of such a function is proved.
We shall now show that formula (3-2.19) represents a bounded solution
of Eq. (3-2.4) which satisfies the conditions (3-2.2) for an arbitrary piecewise
continuous function <p(x) and is continuous everywhere that <p(x) is continuous.
We shall prove this theorem in two steps. First, to show that it is true
in case <p(x) is a linear function

<p(x) = cx , (3-2.2')

we consider the sequence of auxiliary functions (Figure 39)

<pn(x) = cx for 0^x йі

- a(l — x) for l ^x йі.

180 PARABOLIC DIFFERENTIAL EQUATIONS

The number a is determined so that <pn(x) at the point x = l( 1 — 1/и) shall

be continuous:

, n—1 l
cl-= a— i.e., a = (n — l)c .
n n

The functions u„(x, t), which for the

<pn{x) are defined by formula (3-2.19),
then are known continuous solutions
of the heat-conduction equation with
homogeneous boundary conditions and
the initial conditions

Un(X, 0) — фп{х) •

<Pn(x) £ <Pn+l{x) , OSXfkl,

then on the basis of the principle of the maximum

un(x, t) ^ wn+1(x, t) .
The function U0(x) = cx is a continuous solution of the heat-conduction equa¬
tion. It follows from the principle of the maximum that necessarily

Un(x, t) ^ U0(x) ,
since this inequality is valid for x = 0, x = l, and / = 0. The sequence {un(x, f)}
is therefore monotonic, nondecreasing, and is bounded above by the function
U0(x)', consequently, it converges. We can now easily recognize the validity
of the relations

u{x, t) = lim Unix, t) = lim [ G{x, $, t)(pn($) d£ = \ G{x ?, t)<p{$)d£ ^ U0(x)

П—юо n-*oo JQ J0

and pass to the limit under the integral sign. On the basis of footnote 52
this function satisfies Eq. (3-2.4) and the homogeneous boundary conditions
(3-2.5) for t > 0. Moreover, it is continuous at t — 0 and 0 ^ x < 1, as we
now prove. Let x0 < l. We choose n such that x0 < 1(1 — l/и). In this case,
(pn(xo) = Uo(x0). If we bear in mind that

Unix, t) 5S u(x, t) ^ U0(x)

and

lim un(x, t) = lim U0(x) = <p(x0) ,

then we can conclude that the double limit exists, namely

lim u(x, t) = cp(x:0) ,

which is independent of the order of passage to the limit x —> x0 and t —* 0.

However, this implies the continuity of u(x, t) at the point (x0, 0). This func-
3-2. THE METHOD OF SEPARATION OF VARIABLES 181

tion is also bounded since it does not exceed U0(x). Therefore, the theorem
for (p(x) — cx is demonstrated.
If x is replaced by l — x, we see that the theorem holds also for

<p(x) = b(l - x) . (3-2.2")

Hence, each function of the form

<p(x) = В + Ax
is valid, since such a function can be obtained by addition of (3-2.2') and
(3-2.2"). Further, it follows that the theorem is also true for an arbitrary
continuous function without the assumption ^(0) = <p(l) — 0. Every function
of this type can therefore be represented in the form

<p(x) = [,(0) + -y-(?>(/) — y>(0))^j + <p(x) ,

where the sum in the brackets represents a linear function and <p(x) is a
continuous function which vanishes at the ends of the interval: <p(0) = <p(l) = 0.
However, we have already seen that the theorem is true for both summands;
therefore, it is also true for <p(x).
We turn now to the proof of the theorem for an arbitrary piecewise con¬
tinuous function <p{x). In this case, formula (3-2.19) also determines a solution
which satisfies Eq. (3-2.4) and the homogeneous boundary conditions (3-2.5).
Let Xo denote any point of continuity of the function <p(x). To every
positive s, a d{e) can be found such that | u(x, t) — cp(x0) | < e holds when
I x — x0 I < d(e) and t < 8(e). In order to understand this we first note that
because of the continuity of the function <p(x) at the point x0, an ^(e) exists
such that

I (p(x) - <p(xо) I ^ ~ for I * - *0 I < y(e) ,

from which results

<p(xo) — Y = 9(x) й <p(xo) + -J for I X — Xo I < y(e) . (3-2.23)

We now construct the following continuous differentiable functions ф(х) and

<p(x):

<p(x) = <p(x o) + 4- for 1 X - *0 I < V(e)

it n

(a)
1 > Ф)
21

IIV

for 1 ^ - Xo
A

<p(x) = <p(x o)- — for

O)

1 * - Xo 1
(b)
'2.

2.
3
ІІЛ

for 1 *0 1
O)

X -

In the interval | x — x01 > y(e), Ф and <p must satisfy only the requirements
(a) and (b), but are otherwise arbitrary. On the basis of the inequality (3-2.23)
we already have
182 PARABOLIC DIFFERENTIAL EQUATIONS

<p(x) g <p(x) й <p(x) . (3-2.24)

We consider now the functions

u(x, t) = j G(x, f, t)<p(£)d£

u(x, t) = j G(x, f, t)<p(£)d£ .

Because of the continuity of ф(х) and <p(x), й(х, t) and u(x, t) are also continuous
at the point x0, i.e., a 5(e) exists such that

I u(x, t) - <p{x) I ^ -|-

for I X — Xo I < 8(e) , t < 8(e)
I u(x, t) - tp(x) I ^ Y

and it then follows that

U(x, t) й ф(х) + Y = ¥>(*o) + £

for I x — Xo I < 5(e) , / < 5(e)
u(x, t) ^ </>(x)-= p(*0) — £ .

Since G(x, f, f) is nonnegative, formula (3-2.24) gives the relation

u(x, t) ^ и(лг, /) ^ й(лг, t) , (3-2.25)

and hence follows the inequality

<p(x0) — e g u(x, t) ^ p(*„) + £ for I X — x0 I < 5(e), £ < 5(e) ,

i.e.,
I u(x, t) — <p(xо) I < e for I ДГ — Xo I < 5(e), t < 5(e) ,

which was to be proved. The boundedness of the function | u(x, t) | follows

from (3-2.25) and from the boundedness of the functions u(x, t) and u(x, t).
Thus the theorem is proved.

4. The inhomogeneous heat-conduction equation

We shall consider the inhomogeneous heat-conduction equation

ut = агихх + f(x, t) (3-2.1)

with the initial condition

u(x, 0) = 0 (3-2.26)

and the boundary conditions

o

о о
s'

II

(3-2.5)
a

II

The solution u(x, t) of this problem is sought in the form of a Fourier series
in terms of the functions sin (nn/l)x:
3-2. THE METHOD OF SEPARATION OF VARIABLES 183

uix, t) = 2 Unit) sin^x . (3-2.27)

Thus, we consider i as a parameter. In order to find u(x, t) we must deter¬

mine the functions un{t). To this end we use for f(x, t) the series represen¬
tation

fix, t)= f fnit) sin^x ,

71 = 1 /

with

Ш = fie,t)sm?jZdS . (3-2.28)
/Jo /

If we substitute this solution expression in the initial Eq. (3-2.1), we obtain

j^sinyx (funit) + Unit) — fnit)I = 0 .

This equation will be satisfied if all the development coefficients are equal
to 0, i.e., if

Unit) = - а2 (уУu^t) +fnit) . (3-2.29)

From the initial condition for uix, t)

uix, 0) = 2 u„i0) sin-y-x - - 0 ,

n=l l

we find the initial condition for Unit)

uni0) = 0 . (3-2.30)

If we solve the ordinary differential Eq. (3-2.29) with the homogeneous initial
conditions (3-2.30)56 then we find

Unit) = )fniT)dr . (3-2.31)

If we introduce this expression for unit) into (3-2.27) we obtain a solution of

our problem in the form

uix, t)= І IT
*=iLJo
g-(,rn/!>2a2((-r>y(r) I .
Jsln
nn
l
X . (3-2.32)

With the expression (3-2.28) for /„(r), (3-2.32) is transformed into

uix, t) = j ^ jy І е-(«/и1о*(*-г> sinyx-siny £ j/(f, r)d£ dv

(3-2.33)
= \ \ G(x, £, t — t)/(£, T)d£dr

where

56 See footnote 17.

184 PARABOLIC DIFFERENTIAL EQUATIONS

G(x, = e~Un/l)2“2t‘-f> sin^x-sin^f (3-2.34)

I n=1 l l

coincides with Green’s function given by formula (3-2.18).

To determine the physical significance of the solution

u(x, t) = f j G(x, f, t — t)/(f, z)d£dz (3-2.33)

we assume that /(f, z) differs only from 0 in a sufficiently small neighborhood

fo ^ f ^ fo + ^f , T0 g г ^ r0 + Az0
of a point M„(f0 > r0). The function

F(f, z) = Cjo/(f, z)
denotes the density of the heat source. The total amount of heat which is
set free in the interval (0, l) during the time of the action of the source (i.e.,
during Az) is then equal to
Sr0+ar r
cpf(Z,z)dtdz. (3-2.35)
Г0 Jfo
If we use the mean-value theorem, we arrive at the expression

u(x, t) = G(x, ?, t — r)/(f, z)d£dz

fr0-Mrf f0-Mf _ Г)
= \ \ G(x, f, t — z)f($, z)d£dz = G(x, f, t — z)—- = u{x, t)
J ro J fo CP
where

fo < f < fo + A£ , z0 < Z < z0 + Az .

Hence, as df—*■0 and Az—>0 we obtain the function

u(x, t) = lim u(x, t) = — G(x, f0, t zq) . (3-2.36)

Jf-0 cp
Jz-0 1
Consequently, this can be interpreted as an influence function of an instantane¬
ous heat source which appears at the point f0 at time r0.
When the function (Q/cp)G(x, f, t — z) representing the action of an in¬
stantaneous unit source at a point is known, the action of a continuously
distributed source of density F(x, t) = cpf(x, t) can be represented by formula
(3-2.33). This follows directly from the physical significance of the function
G(x, f, t — z).
Therefore, the influence of the temperature of the heat source occurring
in the region (f„, f0 + Af), (r0, r0 + Az) can be represented by the expression

G(x, f, t - r)/(f, r) Jf Az , Q- = /(£, z)A{Az .

cp

For the case of a continuous distribution of heat sources in the entire region
0 ^ £ й 1,0 ^z S t we obtain
3-2. THE METHOD OF SEPARATION OF VARIABLES 185

u(x, t) = j ^ G(x, £, t - Г)/(£, т)(іЫт

by a passage to the limit as —>0 and Jr-*0.

Therefore, proceeding from the physical significance of Green’s function
G(x, f, 0 one can immediately write the expression (3-2.33) for the solution
of the inhomogeneous equation.
Since we know the form in which the solution of our problem must be
represented, we can investigate the conditions which the function /(£, z) must
satisfy for this formula to be applicable. We shall not go into this now.
We have considered the inhomogeneous equation with homogeneous initial
conditions. If, however, the prescribed initial condition is inhomogeneous,
the solution obtained with the homogeneous initial condition must be added
to the solution of the homogeneous equation with initial condition u(x, 0) = <p(x)
which we found in Section 3-1.

5. The general first boundary-value problem

For the heat-conduction equation, the first boundary-value problem reads

as follows: Find the solution of the equation

Ut a Hxx + ./"(-r, t) (3-2.1)

with the auxiliary conditions
u{x, 0) = <p(x) (3-2.2)
u{0, t) = fixit)
(3-2.3)
u{l, t) = t*i{f) .
In order to find the solution we introduce a new function v(x, t) by

u(x, t) — U(x, t) + v{x, t) , (3-2.37)

which represents the difference between u{x, t) and a known function U(x, t).
We define v(x, t) as the solution of the equation

vt - a2vxx = f(x, t)
f(x, t) =f(x, t) — [Ut — агихх\
with the auxiliary conditions

v(x, 0) = <p(x) , Ф(х) = <p(x) — U(x, 0)

v(0, t) = (t) , mi(t) = Mi{t) — U(0, t)
v(l, t) = Д2(і) , Д2(і) = ц2(і) — U(l, t) .
Further we choose the auxiliary function U{x, t) such that

Mi(t) — 0 and Mi(t) — 0 .

Obviously for this purpose it is sufficient to set57

U(x, t) = fii(t) + у IMt) — Мі(0] •

57 See Chapter 2, Section 3 § 5.

186 PARABOLIC DIFFERENTIAL EQUATIONS

Consequently, the determination of the function u(x, t) as a solution of the

general boundary-value problem is based on the determination of a function
v(x, t), which represents the solution of the above boundary-value problem
with homogeneous boundary conditions. A method of determination of v(x, t)
was given in Section 3-2 §4. To be sure, the “solution scheme” described
there is not always suitable for the representation of the function u{x, t) when
inhomogeneities exist in the equation itself and in the boundary conditions.
The degree of difficulty in determining the auxiliary function v(x, t) depends
on the function U(x, t). In particular, for a problem with stationary inhomo¬
geneities it is appropriate to distinguish a stationary solution and to find the
difference between it and u(x, t).iS
As an example, consider the following problem for a bounded rod (0, l)
whose ends are held at the constant temperatures m0 and uv:

lit — d i'ixx

u(x, 0) = <p(x)
u(0, t) = u0
u(l, t) = Ui .

We establish the solution in the form

u(x, t) = U(x) + v(x, t),

where й(х) is the stationary temperature and v(x, t) is the deviation of the
function u(x, t) from й(х).
Then the conditions
й"=0, Vt - a2 vxx
й(0) = щ , v(x, 0) = cp(x) — U{x) - (pi{x)
й(1) = Ui , v{0, t) = 0 , v(l, t) = 0
hold for й(х) and v(x, t).
Hence we find
— X
u{x) = u0 + — (Ui — Mo) .

The function v(x, t), determined by the initial condition and the homoge¬
neous boundary conditions, is then easily found by separation of variables.

Problems

1. Derive the equation for the heating of a homogeneous thin wire due to
a constant electrical current when heat exchange with the environment takes
place on the surface of the wire.
Solution:

Ut = a2uxx — hu -f q ,
where h and q are fixed constants.

58 See Chapter 2, Section 3 § 6.

3-2. THE METHOD OF SEPARATION OF VARIABLES 187

2. Derive the diffusion equation in a medium which moves in the direction

of the x axis uniformly with a velocity w. Consider the case of one inde¬
pendent variable.
Solution:

ut = Duxx — wux ,

(where D is the diffusion coefficient).

3. Proceeding from Maxwell’s equations under the assumptions EX = EZ = 0,
Нг = 0 and disregarding the displacement current, prove that in a homogene¬
ous conducting medium the resulting electromagnetic field Ey satisfies the
differential equation

дгЕу _ Ana dEy

dz2 ~ ~ ~дГ ’

where a is the conductivity of the medium and c is the velocity of light.

What equation does Hx satisfy?
4. Give a physical interpretation of the following boundary conditions in
problems of the theory of heat conduction and diffusion:

(a) u(0, t) — 0 (b) ux(0, t) — 0 (c) ux(0, t) — hu(0, t) = 0

ux(l, t) + hu(l, t) = 0 h > 0 .

5. Solve the problem of the cooling of a uniformly heated homogeneous rod

at whose ends the temperature is equal to 0, under the assumption that no
heat loss occurs on the lateral surface.
Solution:

4f70 °° e~ia2{2k-l)2*2/l2)t (2k - 1 )n

u{x, t) — sin --;-x u(x, 0) = U0.
n »?i (2k - 1)

6. Let the initial temperature of a rod be given by u(x, 0) = u0 = const, for

0 < x < l. The temperature of the ends are held constant: u(0, t) = ult u(l, t)
= и2 for 0 < t < oo. Find the temperature of the rod when no heat exchange
occurs on the lateral surface. Determine the stationary temperature.
7. Solve Problem 6 under the boundary conditions that one end has a con¬
stant temperature and that the other is heat-insulated.
8. Solve the problem of the heating of a thin homogeneous conductor due
to a constant electrical current when the initial temperature, the temperature
of the boundary, and the temperature of the environment are equal to 0.
9. A cylinder of length / is filled with air and has the same pressure and
temperature as the exterior environment. At the initial moment, the cylinder
is opened, and a diffusion of gases into the cylinder commences from the
surrounding atmosphere in which the concentration of a known gas is equal
to u0. Find the amount of gas diffused into the cylinder during the time t
when the initial concentration of the gases in the cylinder is equal to 0.
10. Solve Problem 9 under the assumption that the left end of the cylinder
is closed by a semipermeable membrane.
188 PARABOLIC DIFFERENTIAL EQUATIONS

11. Solve the problem of the cooling of a homogeneous rod whose lateral
surface is heat-insulated; its initial temperature is u{x, 0) = <p(x), and at the
ends a heat exchange occurs with the environment whose temperature is 0.
Consider the special case <p(x) = u0.
12. Solve Problem 11 under the assumption that the temperature of the
environment is equal to u0.
13. Solve Problem 11 under the assumption that on the lateral surface a
temperature exchange occurs with the environment whose temperature (a)
equals 0, (b) is constant and equals .
14. What temperature does a rod assume when one end is heat-insulated
and through the other end passes a heat current changing harmonically with
time, if the heat exchange on the lateral surface is neglected?
15. Solve Problem 11 now assuming that the temperature at one end of the
rod is 0 while the temperature at the second end of the rod changes harmoni¬
cally with time.
16. Let a rod (0, l) be composed of two homogeneous pieces of equal cross
section which touch at the point x = x0 and are characterized by , ki and
a2, k2 respectively.What temperature is assumed in such a rod when the
temperature at one end of the rod, x = 0, tends continuously toward the value
0 while the temperature at the other end changes sinusoidally with time?
17. The left end of the combined rods in Problem 16 has a constant tem¬
perature of 0; the right end, by contrast, has the temperature u(l, t) = uy.
Let the initial temperature of the rod be 0. Determine the temperature of
the rod (taking into consideration only the first term in the series develop¬
ment).
18. Find the temperature u(x, t) of a rod whose initial temperature is equal
to 0 when the boundary conditions have the form

u(0, t) = Ae~at , u(l, t) = В .

A, B, and a > 0 are constant.

3-3. PROBLEMS FOR THE INFINITE STRAIGHT LINE

1. Green’s function for the unbounded straight line

In the preceding paragraphs for the Green’s function of a bounded interval

(0, l), we obtained the expression59

Gt(x, f, t) = 4
t
I
n=l
e Un'l)at sin-if-*, sin ^4 (3-3.1)

If the heat pole possesses the intensity Q, the temperature distribution is

described by the function

59 Here we introduce the symbol l to the function Gi(x, $, t) in order to distinguish

it from Green’s function G(x, $, t) for the unbounded region to which we shall limit our¬
selves in these paragraphs.
3-3. PROBLEMS FOR THE INFINITE STRAIGHT LINE 189

u(x, t) = —Gi(x, £, t) . (3-3.2)

Cp

From our observations in the formulation of the heat-conduction problem for

an infinite straight line, it follows that the sought function G(x, £, t) is con¬
sidered as the limit value of the corresponding function (3-3.1) for a finite
interval when both ends tend towards infinity. In order to calculate this
limit value we transform (3-3.1) so that the ends of the given intervals con¬
tain the coordinates —1/2, Ц2. This is accomplished by the introduction of
new coordinates x' and £'

X=X~Y' f=f-T-
The Green’s function of an instantaneous point-forming source of intensity
Q = cp which is found at the point f' of the interval (—//2, l\2) then has the
form

2 1 \ ■ КП (. I
Gl(x', t) (3-3.1')
/ »=i l 2 )-Sm—(f +T
We transform the product of both sine functions. If n is even, i.e., n = 2m,
then

2кт ( / . I 2кт (
(. / \ . 2лm /
2ю . 2-кт
sin ——I x + — X • Sin ;—$ .
l V ' 2)
sin
l (f +Y) = sm-
If n is odd, then n = 2m + 1, and

(2m + V)n (2m + 1)k

sin sin
l (' + *)
(2m + 1 )tt , (2m + l)n
cos---——x • cos-1--——£ .

Consequently,

Gi(x', £', t) = 4- I" е~Ып/1)2Л sin -^y-x' sm-^-f'

I n=0 l l

2 -ыпп>2<А nn , xn
+fr« ° ' cos —-—x cos —-—£' (3-3.1")
/ n=1 L l
where 2" ranges over the even and 2; over the odd n.
We shall next determine the limit value of the first sum as /—»oo. It
can be written in the form

4-
I
f " <Гх"а2( sin /U' • sin
n=0
' = —
71
f " fVInW
u=0
(3-3.3)

with
2 2 О-г
j , пк
/І(Л) = g-* “ e sin Ля' sin Л£', ЛЛ =—— and An = —j— .

The sum (3-3.3) suggests the corresponding sums for the integral of the func-
190 PARABOLIC DIFFERENTIAL EQUATIONS

tion /і(Л) over the interval 0 ^ Д < oo. As l—>oo then AX —>0. By passage to
the limit we obtain for the integral60

lim — =— WW = — [VxV‘sin Д*' • sin X^'dX . (3-3.4)

J\—0 7Г n=0 7Г Jo 7Г Jo

Analogously we can write the second sum in the form

4 f cos Л У cos XJ' =— I' UXn)AX (3-3.5)

L п=1 7Г n = 1

with

/2(Л) = e x 0 f cos Дх' cos Л?' d/1 = -y- and Д„ = -y- .

60 In formula (3-3.4) we obtain an improper integral on the left side as the limit value
of the integral sum is extended throughout the infinite interval (0, oo). For the justification
of this passage to the limit, one must show that it does not contradict the usual definition

(°7(Д)ЙД = lim [Lf(X)dX

Jo i—oojo
of an improper integral. Therefore, we shall prove that if for 0 fg Д < °° we consider
a continuous function /(Д) such that for any subdivision of the integral (0, °o) into equal
subintervals (a 5S AXi ^ /3) the integral sum

£ f(Xi)(Xi Ді-і)і Xi—i ^ Xi ^Xi

1=1
converges for an arbitrary choice of X*, then the integral

{>№
exists.
We show that the limit value
C^k
lim \ f(X)dX
oo J 0

exists, independent of the manner in which ^4 approaches infinity. First, we shall con¬
sider the case цк = Xk- Then

(*) k+hAx)dx - f(x)dx = k+hf(x)dx ^ z f(Wi+1 - Xi)

Jo Jo JAk i=k

and correspondingly,
j. i k-\-h 1
(**) k+hf(X)dX ^ V f(Xi)(Xi+i - до,
JAjt i=k

where /(Xi) and /(Xi) denote the largest and smallest values of /(Д) in the interval (Xi, Ді+і).
From the convergence of the integral sums with respect to the chosen sequence of Xk,
it then follows that for every e an N(e) exists such that for к > N(e),

I 2 f(Xi)(Xi — Дг —l) I < £

i=k

holds. From this and from the expressions (*) and (**) we obtain

'k + h
f(X)dX < e for к > N(e).

(Next page)
3-3. PROBLEMS FOR THE INFINITE STRAIGHT LINE 191

As AX—>0, we obtain the limit value of the second sum

lim— f'/*('InW = — \™fzWX =— (VxV‘cos te'costfUl. (3-3.6)

J\~0 n n=l n Jo 7Г J 0

The integrals for fx and /2 are formed by splitting the interval (0, l) into
equal subintervals of length AX = 2n/I. In this way it can be seen that the
intermediate point is the right end point of the interval in the first sum and
the middle point in the second sum.
Summarizing the results, we obtain

(Cont’d footnote 60)

From the convergence of each integral sum with respect to a given subdivision it follows
that f(X) approaches 0 for X -* 0, and, therefore, that

'^k+l !'Ki-
lf{X)dX - \ k+hf(X)dX < e for к > N(e) ,

where Xk and Xk+h are those points of the subdivision that lie nearest to the points
and fik+i.
This proves the existence of the improper integral

f(X)dX .
s;
We show now the existence of a limit value

lira 2 f(X*)AXi = f f(X)dX .

Jk;-0 i = o Jo
For the difference there holds

\ f(X)dX - Zf(X*)AX
Jo i=о

< \Xkf(X)dX - Zf(X*)/IXi + \~f(x)dx\ + \i:f(x*)dx

Jo 0 I I \k
and by a suitable selection of AXi and Xk each of the sums on the right side can be made
less than e/3. Therefore, our proposition is valid and our lemma is proved. (However,
the definition of the improper integral arising from the lemma is not suitable, since an
example can be found of a function which satisfies the ordinary definition but not the
definition given here.)
In our case

f(X) = e~x2a'it cos Xx cos X£ or f{x) = sin Xx sin (X .

For the integral sum arising from the subdivision by equidistant points Xi — Xi-i = AX,
it follows that

H f{X*)AX ^ Ef(ii-1)AX = E e~xi-ia2‘AX <AXE е~^г^г^

i=0 i=0 i=0 t=0

The convergence of this sum is easily shown by using the D’Alembert criterion in which
we investigate the ratio of the fth term of the (г—l)th term

(2г—1) (J\)2azt
g —(i —1)гЛЛ2а 2f
192 PARABOLIC DIFFERENTIAL EQUATIONS

G(X, f, t) = lim Gi(X, £, t)

1 — 00

v 4_£ .
- a a t
sin Лх sin Л&/Л f — \ e
1 Г" x*e*< cos Лх COS Л&//1
=-r
JT Jo tt Jo
’•-"•cos X(x-S)dX .
7Г о

Green’s function for the infinite straight line therefore has the form

G (x, £,« = —(
7Г JoVxV‘ cos X(x - $W . (3-3.7)

We now evaluate the integral

/ = j Vх*- cos XfklX , a > 0, (3-3.8)

which depends on both parameters a and ft. To this end we fix a and denote
the integral by /(/3). Here, obviously, in order to calculate the derivative
we must differentiate under the integral sign:

dl
dp -І e~K aX sin xpdX .

By partial integration it follows that

dl . 1 _\2«
JL[
2« Jo
c * cos XpdX = ——2^—f(P)
a

In this manner we obtain for I(P) a differential equation with separated

variables

Zl = _ jL
I 2a ’
whose solution is

I(p) = Ce~p /ia .

The value of the constant is obtained from p 0, and indeed is
о 1 о
C oo f °o
1 \/к
C = 7(0) = e~k *dX - -4=1 e~*dz
Jo V a Jo [/a

because

l/ n
e z dz
s0 ^

Hence we obtain

Kp) = (V^cos;^ = p2/4e (3-3.9)

Jo 2 Va

Now, if we substitute (3-3.9) into Eq. (3-3.7), we arrive at the expression

 3-3. PROBLEMS FOR THE INFINITE STRAIGHT LINE 193

G(x, £, t) = —^==е~и-()Z/"'h (3-3.10)

2Vna2t
for the Green’s function for the infinite straight line. This function is usually
called the fundamental solution of the heat-conduction equation. We can
verify directly that the function

G(x, £, t - t0) _===== v»* (3-3.10')

ср2л/ na2[t — t0)
represents the temperature at the point x at time t, when at the initial
moment t = t0 and at the point £ the amount of heat (J cp is set free:
1. The function G(x,$,t — to) satisfies the heat-conduction equation with
respect to the variables x and t which is proved directly by differentiation.
In fact,
1 x—£
Gx =
2 i/n 2[a\t - /„)]3/2
1 1
Gxx 1* - g)1 1 c
2\/n [ 2 [«2(* - /0)]3/2 + 4[a2(/
At - /„)j5/: J
аг(х - £)2 I ,, -OI/*a*U-t0)
Gt =
2\/k [ 2|a2(/ - G)]3/2 1 4At\t - to)A2 Iе
i.e.,
Gt = a2Gxx.
2. The amount of heat found on the x axis at time / > t0 is equal to

І „-(х-()г/4а2и-10) dx
c/>\ G(x, £, t — t0)dx = —(4=
V* j 2Va\t- U)
Q
Г c-\la Q = cp
Vx
since

Г e~*da =
J -
(a =
V 2Va\t
e
- U)
, da
dx
2 Va\t ■ t.) /
Therefore, the amount of heat on the straight line does not change with time.
The function G(x, £, t — t0) depends only on the time through the argument
0 = a2(t — to), and therefore we can write

r— 1 1 ,, C 07*n (3-3.10)
U “ W7 VO
Figure 40 shows the graphic representation of G with respect to x for
different 0 values. Almost the entire area which is bounded by this curve
lies in the interval
(£ — s, £ + s),
where e is an arbitrarily small number when 0 dl{t. t„) is sufficiently small.
The magnitude of this area when multiplied by cp coincides with the amount
194 PARABOLIC DIFFERENTIAL EQUATIONS

of heat present at the initial moment. Therefore, for small values of t — t0

> 0 almost the entire amount of heat is concentrated in a small neighbour¬
hood of the point £. From what has been said it also follows that the entire
amount of heat at time t0 is found at the point ?.

If we consider the temperature change at a fixed point x = £ + h in the

course of the time for h = 0, i.e., for x = $, we obtain

G x=$
1 _ 1
2\/к -[/f '

The temperature, at this point in which the heat source is found, therefore
becomes unbounded for sufficiently small Ѳ.
If x Ф i.e., x = t; + h with h Ф 0, then G is represented by the product
of two factors:

GA

G-*' ~ [W7 Vt]

The second factor is smaller than 1: for large
Ѳ it is « 1; for small Ѳ, by contrast, it is «0.
Hence, it follows that Gx& « Gx=( for large Ѳ and
Gx^f € Gx=f for small Ѳ. The smaller h is—that is,
the closer x lies to f — the larger is the second
factor. Figure 41 shows the course of Gx=( and
FIG. 41. G*#f for < hi. It is easy to see that
3-3. PROBLEMS FOR THE INFINITE STRAIGHT LINE 195

lim GWf - 0.
0-0

By application of de l’Hopital’s rule we find,

_1_1_1 1 - \Ѳ~Ь/2
lim -кг/*Ѳ
■=lim
Ѵѳ
= 0 .
e-o [ 2г/тГ j 2V 7Г 0-0 (/г2/40Ѵ2/4й

From formula (3-3.10') it follows that at each point x the temperature

produced by an instantaneous point-forming source acting at the initial time
t = 0 is different from 0 for any small interval of time. This phenomenon
can be considered the result of an infinitely fast temperature propagation
(propagation with an unbounded velocity). This, however, contradicts the
molecular-kinetic concept of the nature of heat. This contradiction is con¬
nected with the concept of heat flow used in the derivation of the heat conduc¬
tion equation and with the neglect of the inertia of molecular motion.
2. Heat Conduction in the Infinite Straight Line. The problem of heat
conduction in this case reads as follows: Find a bounded function u(x, t),
where — oo < д: < oo, f2:0, which satisfies the heat-conduction equation

ut = a2uxx, —co<x<oo, t > 0 (3-3.11)

and the initial condition

u(x, 0) = <p(x). (3-3.12)

As previously stated, since the initial condition is satisfied, the function

u(x, t) for t — 0 not only satisfies the condition but also is continuous.61
In order to derive an analytical representation of the sought solution, we
introduce an auxiliary function Ф(х), which we set equal to 0 everywhere
outside of a small interval (f„ — 5, f0 + 8), and inside it coincides with <p(x).
In order to increase the initial temperature from 0 to the value <p(x), an initial
amount of heat is introduced into the interval (£0 — 5, f0 + 8) which is given by
fe0 + 5
Q — cp\ (p(x)dx « cp(p(£)d$ , = 2d .
Je0-5

The temperature at the point x at time t is then equal to

—G(x, f, t) - G(x, £, t)y>(€)J€ , (3-3.13)

cp
where I is a suitable average value in (f0 — 8, $ + d).
By splitting up the entire straight line into small intervals we can represent
u(x, t) on the basis of the superposition principle as the sum of summands
of the form (3-3.13). More precisely, we are concerned with an integral sum
which as 5 — 0 is transformed into

u(x, t) = G(x, f, t)<p(S)d£ (3-3.14)

61 As we saw in Section 3-1 § 7, the solution of the heat-conduction equation is deter¬

mined uniquely by the initial conditions if it is bounded. Therefore, the requirement
that u(x, t) be bounded enters into the formulation of the theorem.
196 PARABOLIC DIFFERENTIAL EQUATIONS

or

(3'ЗЛ4,)

Obviously, the first function represents the solution of our problem.

These assertions do not constitute a proof; consequently we seek condi¬
tions for the applicability of this formula.
We shall show that in the case of a bounded function ^(f), | (p(£) \ < M for
t > 0, the so-called Poisson’s integral

=27гГ.77Ге"“'',Ѵ“'Ѵ(?№
represents a bounded solution of the heat-conduction equation which for t = 0
at all points of continuity of this function u(x, t) is continuously connected
with <p(x).
Therefore, we shall prove the following lemma (generalized superposition
principle). If the function U(x, t, a) for each fixed value of the parameter a,
and with respect to the variables x and t satisfies a linear differential equation

L(U) — 0,
then

u{x, t) = j U(x, t, a)<p(a) da

is also a solution of the differential equation L(U) = 0, provided the derivatives

occurring in the linear differential operator L can be obtained by differentia¬
tion under the integral sign.
The proof of this lemma is very simple. A linear differential operator
L(U) represents a linear combination of derivatives of the function U where
the coefficients depend on x and t. Now by hypothesis the differentiation of
и can be carried out under the integral sign. The coefficients obviously can
also be brought under the integral sign. From this it follows that

L(u) = jL(U(x, t, a))(p(a)da = 0 ,

i.e., the function u(x, t) satisfies the differential equation L(u) = 0.

Finally, let us consider the sufficient conditions for differentiability under
the integral sign when the integral occurring depends on a parameter. A
function

in which a and b are finite limits of integration, can be differentiated under

the integral signs when df(x, a)/dx, in the closed region of x and a, is a
continuous function of these variables.62

62 See V. I. Smirnov, Textbook of Higher Mathematics, 2d. ed. Part II, Ch. Ill, §3,
no. 80, Berlin, 1958.
 3-3. PROBLEMS FOR THE INFINITE STRAIGHT LINE 197

We also easily see that

Fi(x) = J f{x, a)<p(a)da ,

in which a and b once more are finite, similarly can be differentiated under the
integral sign if fix, a) satisfies the above stated conditions and tpia) is a
bounded (or absolutely integrable) function. If the limits of integration are
infinite then in this case we must require the uniform convergence of those
integrals which are obtained by differentiation of the integrands with respect
to the parameter. This observation holds also for multiple integrals which
depend on parameters.
For linear differential equations Liu) = 0 the superposition principle holds;
that is, the function

uix, t) = 2 CiUfx, t),

i=1

which is a linear combination of several particular solutions, again represents

a solution. If the functions uix, t, a) are solutions which depend on a para¬
meter, then the integral sum

1 uix, t, an)Cn, Cn = <p(an)da (3-3.15)

is also a solution of Liu) = 0. The proven lemma, similar to the one above,
gives the conditions under which the limit value of the sum (3-3.15), i.e., here

uix, t) = \ Uix, t, a)(pia)da ,

is also a solution of equation Liu) = 0. Therefore, it is natural from the

viewpoint of the proven lemma, as with the first, to designate it as a gene¬
ralized superposition principle.
We shall now investigate the integral in (3-3.14'), first by showing that
the definite integral in (3-3.14') converges and represents a bounded function
provided <p(x) is bounded, | <pix) | < M. In this case

If” 1 -U-e)2/4a2<
uix, t) I < M—7=1 -y—
2ук J-oo Va t

= M-4=\
1
e~a2da = M , a
6—x
Vic Waft ’
since

e~a da =у/яГ .

Further, we shall prove that the integral in (3-3.14') for t > 0 satisfies
the heat-conduction equation. Therefore, it is sufficient to show that the
derivatives of this integral for t > 0 can be calculated by differentiation under
the integral sign.
In the case of finite integration limits, this is certainly permissible since
the derivatives of the function
198 PARABOLIC DIFFERENTIAL EQUATIONS

1 g-(x-f)2/4a2t

2V7 n a2t

for t > 0 are continuous functions. To prove the possibility of differentiation

under the integral signs for infinite integration limits it is sufficient to show
the uniform convergence of the integral arising by differentiation under the
integral sign. We carry out this proof, for example, for the first derivative
with respect to x.
The differentiability of function (3-3.14) with respect to x and the validity
of the equation
du
~ t))<p(Z)d$
ox — ox

are proven when we have shown that the integral standing on the right side
converges uniformly; in particular, for the proof of the differentiability at
the point (x0, to) it is sufficient to show the uniform convergence of the integral
in a fixed region

ti й t0 й U , I x — x01 й x

of that point.
Sufficient for the uniform convergence of an integral (an analogous crite¬
rion holds for the uniform convergence of a series) is the existence of a
positive function F(£) which does not depend on the parameters x and t and
which majorizes the function

-~G(x, $, t)<p(£) £ — x0>x, £ — x0 < — x, (3-3.15')

such that

< °o, p F(t)d$ < oo . (3-3.15")

Therefore, let x^ be a number which satisfies the inequality (3-3.15').

First of all we seek an upper bound for the magnitude of the integrands
in the formula for du/dx. It is

-j~G(x, $, t) • I рЮ 1_ lg-*l g-U-e)2/4a2(

M£)l
dx 2і/тг 2[аЧ]ъ/г
< I £ — Xp I + x g-(|f-xo|-T)2/«a2(2 — p(g) (3-3.16)
= 2і/тг 2[a2t i]v*
for f — x0 < x, \x — Xo x, and Л g t ^ t2. For this function F(£), however,
(3-3.15") is valid, and

•- 1 1 g — JCo I + X.
~F(€)de = g-(|{-l0l-J,2/«“2<2
d$
x, 2Vк 2[аги}ъ/г
“ 1 _ + 2x _
*j-7 2\/я 2[аН$/ів

£l = I £ — X0 I — X,
3-3. PROBLEMS FOR THE INFINITE STRAIGHT LINE 199

and this integral converges, since a factor of the form (a£ + b)e cj2 appears
under the integral sign. Therefore

ax
= Г
J-oo ox
• (3-3.17)

By a completely analogous method, we can also prove that all the above
derivatives can be calculated by differentiation under the integral sign.
Therefore, we have proved that the function (3-3.14') satisfies the heat-con¬
duction equation.
We turn now to an important property of the integral (3-3.14), and indeed
we shall show that at all the points of continuity x0 of the function (p(x),
the following limit

u(x, t) —> (p(x0) for t —> 0 and x —> x0

is valid.
Let <p(x) be continuous at x0. Then we have to show that

lim u(x, t) = <p(x0) .

t~ о
l-x0
For each e > 0 we must therefore find a 5(e) such that
I u(x, t) — (p(xо) I < £

holds when

\x — x0\ < 5(e) and 11 \ < 5(e) .

On the basis of the assumed continuity of <p(x) at x0, an rj(e) exists with

I <p(x) — <p(xо) I < -|- (3-3.18)

for

I X — Xo I < У] •

We now split up the integration interval and represent u(x, t) as the sum of
three summands:

-и-еі2/4Л
<p(WZ + l_p..*+l {-...*
U[X,t) 2v/*L~-l/^ 2і/л- Jxj 2l/TZ Jx2
= иг(х, t) + uz(x, t) + u3(x, t), (3-3.19)

where

Xi = x0 — v and x2 — x0 + v • (3-3.20)

For the second summand we can then write

(pjx oj_r2
u2(x, t)
2i/?r )XlVaH

1 _Г2 — " {х-()'іЛ[<р(£) - <p(x0)Ш = /, + h •

+ 2\/n JХ]л/аН
200 PARABOLIC DIFFERENTIAL EQUATIONS

The integral Л can be calculated directly, and we therefore obtain

2
(x -*l/*Va 2(
Г _ 9(*o) f9(x0)f
_e~a2 da
I_ 21/Т]Ж1 г/л г/*)
with

g — x
da = (3-3.21)
2\/a2t 2\/a2t
Now if I x — x01 < y, then the upper limit is positive and the lower, negative,
and as / —»0 the upper limit tends towards + oo and the lower limit toward
— oo. It then follows, however, that

lim Ii = <p(x0).
t—0

Consequently a <5, can be prescribed such that

h - y>(x0) I < -g (3-3.22)

when

x — x01 < and 111 < di .

We shall now prove that the remaining integrals, that is, I2, их and щ, can
be made arbitrarily small. First we estimate I2 as follows:

ЛІ ^ 1 Г2 i_ g-lx-t)2/*a2t j ^ _ ^o) I #
2р//7Г Jxj
VaH
From (3-3.20) it is seen that for

Xi < f < x2
the inequality

I f - x01 < r)

is valid. If we use the inequality (3-3.18) and the relation

1
e~*2da < e~a2da = 1 ,
г/я

we obtain for arbitrarily chosen x' and x'

e 1 f u2-x)/2il*t -
:0-(*-£>г/4в«^£ =
/21 ^ 4-4= -e^da<4-,
6 2т/л- )XlVa2t u 1/ x J u1-jr)/2Va2£ b
(3-3.23)
where the new variable a is defined by (3-3.21). Further,

1 1===e U |l/4a Af
«з(х, t) I = < e a2da—*0
2у/ ,VaH і/тг

for x —>x0 and t —> 0 (3-3.24)

and analogously
3-3. PROBLEMS FOR THE INFINITE STRAIGHT LINE 201

Uj—x)/2 v^a2«
U{(X, t) I = e a da—* 0
2і/ 7Г

for x-+ x0 and t —> 0 . (3-3.25)

As *—►*<>, then дг2 — x > 0 and x, — x < 0. As / —>0, the lower limit in the
last member of (3-3.24) tends toward +oo and the upper limit in (3-3.25)
tends toward — oo. Consequently, there exists a <S2 such that

\u3(x, t)\ <—■ and \ui(x, t) I (3-3.26)

holds when

I x — x01 < <\ and 111 < <52 .

With the use of the estimates (3-3.24) and (3-3.25) obtained above, we find

I u(x, t) — (p(xо) I ^ I Ml I + I /, — <p{xо) I + I Л I + I Мз I

<T + T + T + T = £’ p-3-27*
provided

Ix — x01 < <5 and \t I < о ,

where 8 denotes the smaller of the numbers 8, and <52.
Therefore we have shown that the function

u(x, t) = -i-f* e-{x-()2/ta\(t)d$ (3-3.14')

21/Л-

is bounded and satisfies the heat-conduction equation as well as the initial

condition.
If the initial value is given not for t = 0 but for t = t0, then the expres¬
sion for u(x, t) assumes the form

1 p 1 е~и-()г/* a*lt-t0)
u{x, t) <p(Z)dt . (3-3.14")
2i/?Tj - Va\t - to)

The uniqueness of this solution for a continuous function <p(x) results from
the theorem proven in 3-2§3. If the initial function <p(x) possesses a finite
number of points of discontinuity, then (3-3.14') represents a bounded solution
of Eq. (3-3.1), which is continuous everywhere with the exception of the
points of discontinuity of <p(x).63
We shall consider as an example the following problem: Find a solution
of the heat-conduction equation when the initial temperatures (at t = t0 — 0)
for x > 0 and x < 0 are constant but have different values, i.e.,

u(x, 0) = <p(x) = Ti for x > 0

= T2 for x < 0 .

63 With the aid of the method described in 3-3 §3, we know that the function u(x, t)
is uniquely defined by the enumerated properties.
202 PARABOLIC DIFFERENTIAL EQUATIONS

For the solution, formula (3-3.14') yields

r° x-()‘/4a2t_d{__ _T\_f _
-U-f>‘/4a2*

l/jrj — 2y/a2t т/тг Jo " 2\/ a11

Тг
i/2Va 2t
Г Tx
e~a da + Ц"
l/ff J- і/яЛ J— x/2^a%t

7\ + Г2 , Тг-ТгГ^
+ і/ 7Г
e a da (3-3.28)

since

—7=1
1 f-
e da =—7=4 e
1 f° -«S 1 fl-»2J 1 1 f*-.2,
da-7=1 e da— — — —17=1Г e da
V n J_» V К )-*, V К Jo 2 Vx Jo
1/7Г Ji
and
* .2
x
e da, 2 =
г/jTj-* 2VaH *

In particular,

2 =■
M(*’/} " 2 (X + 1/* Le " ^) ’ 2i/a2/

holds for

T2 = 0 , Ti = 1.

The temperature profile at time t is given by the curve

1 1 r* 2

/w-2+tt

Here 2 is the abscissa of the point at which the temperature is considered

when the value 2i/a2t dependent on t is chosen as the unit of length. The
construction of this curve is not difficult, since the integral

2 f1 _e*
Ф(г) - —7=1 e da ,
vn Jo

the so-called error integral, occurs often and has been tabulated many times
in probability calculations.64
For arbitrary Тг and T2, formula (3-3.28) can also be written in the form

u(x,t)= T't-T±+ Tl~-T^/ X (3-3.29)

Ф 2VaH ) ‘

64 See, for example, A. A. Markov, Text Book of Probability Calculations, where

this integral is tabulated to six places. At the end of this book, likewise, are given
values of this integral.
3-3. PROBLEMS FOR THE INFINITE STRAIGHT LINE 203

Here we recognize that the temperature at the point x = 0 during the entire
time is constant, and therefore is equal to half of the sum from the right
and left sides of the initial value, since Ф(0) = 0.
Finally, the solution of the inhomogeneous equation

ut = a2uxx + f(x, t) , —co<x<oo,t>0

with the homogeneous initial condition

u(x, 0) = 0
obviously can be expressed by the formula

u(x, t) = G(x, £, t - t)/(£, т)с1Ыт . (3-3.29')

This results from the meaning of the Green’s function G(x, f, t) (see Section
3-2 §4). At this point, however, we shall not investigate this formula and
the conditions imposed on f(x,t) for its applicability.

3. Boundary-value problems for the semi-infinite line

As we have already noted in 3-1 §4, when we are interested in the dis¬
tribution of heat in the neighborhood of one end of a rod, while the influence
at the other end of the rod is unessential, we make the assumption the latter
end lies at infinity. This then leads to the problem of determining those
solutions of the heat-conduction equation

ut — a2uxx (3-3.11)

on the semi-infinite line x > 0 for t > 0 which satisfies the initial condition

u(x, 0) = <p(x) , x > 0

and a prescribed boundary condition. This depends on the character of the
boundary influence and can be given in one of the following forms:

first boundary-value problem u(0, t) = fi(t) ;

du
second boundary-value problem -(0, t) — v(t) ;
dx
or

third boundary-value problem -^-(0, t) — Л[м(0, t) — ѲЩ] .

ox
In the following investigations we shall consider only the first boundary-value
problem, i.e., the determination of a solution of the heat-conduction equation
with the auxiliary conditions

u(x, 0) = cp{x) , m(0, t) = n(t). (3-3.30)

Therefore, if the problem is to have a unique solution certain conditions

must still be met at infinity. Thus we require that the solution u(x, t) is
bounded everywhere:
204 PARABOLIC DIFFERENTIAL EQUATIONS

I u{x, t) \ < M for 0 < x < oo and t ^ 0 ,

where M is a constant. From this it follows that the initial function <p(x)
must also satisfy the condition

I <p(x) I < M .

The solution of this problem can be represented in the form

u(x, t) = uAx, t) + u2(x, t),

where их(х, t) refers to the influence of the initial condition and u2{x, t) to the
influence of the boundary condition. These functions can be defined as those
solutions of Eq. (3-3.11) which satisfy the conditions

0) = cp(x) , мА0, t) = 0 (3-3.30')

or
u2(x, 0) = 0 , u2(0, t) = pt(t). (3-3.30”)

The sum of these functions then satisfy conditions (3-3.30).

We shall next prove two lemmas for the function defined by Poisson’s
integral

= (3'3-31»

1. If ф(х) is an odd function, i.e.,

Ф(х) =- ф{— x),

then the function

u(x, t) =
1 1
■ /-^\ —7==- e--U-f)2/4a2(
2\/k J_oo з/аЧ
Фта
vanishes for x — 0:

u{0, t) = 0 .

Here, it must be assumed that the integral defining u(x, t) converges; this is
the case when ф(х) is bounded. The integrand in

1 _f~ 1
u(0, t) -- е-иг,іаг"ф{!;Ш
2-[/л j_oo i/ a21

is odd with respect to $ since it is the product of an odd function and an

even function. Now, however, the integral of an odd function between the
limits which lie symmetrical to the origin of the coordinates, is equal to 0;

consequently
u(0, t) = 0 ,
which proves our lemma.
2. If ф(х) is an even function, i.e.,

ф(х) = ф(— x),

the derivative of the function u(x, t) from (3-3.31) for x = 0 and all t > 0 is
equal to 0:
 3-3. PROBLEMS FOR THE INFINITE STRAIGHT LINE 205

-f^(0, /) = 0 .

That is,
1 f+-U-g) g-U-f)2/4a2t
= 0,
2\/n J_»2(a2t)3/2 x=0

since the integrand at x = 0 represents an odd function when ф(х) is even.

The determination of the function u(x, t) now presents no further difficulty.
First, we introduce an auxiliary function U(x,t), which is defined on the
entire straight line — oo < * < oo, and which satisfies Eq. (3-3.11) and the
conditions
U(0, 0 = 0,
U(x, 0) = <p(x) for x > 0 .

This function can be defined by use of the first lemma and an initial func¬
tion ¥ (x), which for x > 0 coincides with <p(x) and for x < 0 represents the
odd continuation of <p(x), i.e.,

¥(x) = <p(x) for x > 0

— <p( — x) for x < 0
so that
1 f°° 1_
U(x, t) = e-lx'*)2/ta4W($)d$ .
гт/тГ'-со 1/аН

In the region of interest, x ^ 0, we then obtain

u(x, t) = U(x, t) for x ^ 0 .

Introduction of the function ¥(x) results in

1 Г° 1 -<*-«) 1 1i -«.-«*/*0*1 щме

U(x, t) = i/4e2‘TO# +
2і/тГ J_oo Vc
VaH 2]/л Jo л/a11
1 -U+f )2/4a2( 1 1 „-U-f>2/4a2(
<p(£)d£ + о /— <p{£)d£
2і/7Г Jo 1/a21 У 2~\/n Jo \/a2t

where the substitution $'= — f is introduced in the first integral and the
relation
¥(£) = -&-£) = -?(£')

is also used. By combining both integrals we arrive at the sought function

,-U-{)2/4a2t _ -U + f)2/4a2«
{e }<p($)d$ (3-3.32)
Ui(x’t} ~ гѵАо i/m

in a form which no longer contains auxiliary functions. We also note that

for x = 0 the expression in braces vanishes and zq(0, t) — 0.
With the help of the second lemma one can see that the solution of the
heat-conduction equation with the homogeneous boundary condition of the
second type (d/lfiO, t)/dx) = 0 and an initial condition U^x, 0) = <p(x) is represent¬
able in the form
206 PARABOLIC DIFFERENTIAL EQUATIONS

1 f~ 1_
Мі(дг, /) (3-3.32')
2г/^Г]о 1/аН

We now apply the formula so obtained to the problem of cooling a uni¬

formly heated rod whose boundaries are held at a constant temperature. Let
this temperature equal zero. The problem leads to the determination of those
solutions Vi of the heat-conduction equation for which the conditions

Vi(x, t0) = T , Vi(fi, t) = 0

are fulfilled.
Since the initial condition is prescribed not for f = 0 but for t = t0, in
place of (3-3.32) we obtain

T_f“ ,-<x-f)2/4 a2U-«0> -(x+e)2/4a2(t-«0) d?

Vi(x, t) = {e (3-3.33)
2л/ж Jo Va2{t — t0) '

If we now split up the integral into two summands and introduce the variables

$ — X £+X
a = ai =
2Va\t-U) ’ 2Va\t - to)
we obtain

Vl(x,t) = -^=ГГ _e~*2da- j"_е~*ЧаЛ

V ft L J-x/2Va2»—tg) J х/гѴа2и — <0) J
х/гѴа2((-(п) о
T г
rx/2Va2(t-40)

_e~a2da = T-4=\ «г*2 da

Т/гг J. x/VazU-t0) Jo

or

Уі(дг, f) = ТФі _ * ^ (3-3.33')

2Va\t -t0) J ’
where

0(z) = —7= ( e 0,2 da

Vft Jo

is the error integral.

We turn now to the determination of u2(x, t)—that is, the second part of
the solution of the first boundary-value problem.
Let

fi(t) = Цо — const.
The function

К 2T/4-J (M-34)

then represents those solutions of the heat-conduction equation which corre¬

spond to the conditions

v{x, to) = fio , v(0, t) = 0 .

3-3. PROBLEMS FOR THE INFINITE STRAIGHT LINE 207

From this it follows that

**■ « *'- m *> ""t1 “ K'

= - Want -J] (3'3'35)

is the sought function, since it satisfies the same equation and the conditions

v(x, t0) = 0 , x > 0 and y(0, t) = fi0 , t > t0.

For v(x, t) we write

v(x, t) = fi0U{x, t),

where

U(x, t) = 1 — Ф _e-*2da (3-3.36)

2Va\t - to) ) lA і/гѴа2(е-(0)

is the solution of the same problem in the case of fi0 = 1.

By definition, the function U(x, t) is defined at first only for t ^ t0. We
now extend the region of definition in which we set

U(x, t) = 0 for t < t0.

Obviously this definition is compatible with the function values of U(x,t) for
t — 0. Also, the function so defined satisfies the heat-conduction equation for
all t and x > 0. The boundary value of this function (for x = 0) forms a step
function which for t < t0 is equal to 0 and which for t > t0 is equal to 1.
The function U(x,t) occurs frequently in applications and is an aid in the
determination of u2{x, t).
We consider now a second auxiliary problem, namely to solve the heat-
conduction equation with the initial and boundary conditions

v(x, to) = 0, v(0, t) - fi(t) = lOT 1° < tl

(0 for t > ti .
We can verify directly that

v(x, t) = th[U(x, t — to) — U(x,t — fi)] .

By similar reasoning we find that:
If the boundary function p{t) is given as a step function

fl{t) = Ho for to < t ^ ti

= A<1 for ti < t й t2

- f^n—l for tn— 1 t ^ tn r

then with the help of the function /n(t) the solution of the second boundary-
value problem can be represented in the form

u(x, t) = I fa[U(x,t — ti) — U(x, t — ti+i)] + Hn-iU(x,t — tn-i). (3-3.37)

i=0

The mean-value theorem then yields

208 PARABOLIC DIFFERENTIAL EQUATIONS

/ V dU(x,t — z)
U{X, t) = I Щ-K-r-- Az + fin-iU(x, t — tn-i) with ti й т< ^ ti+i.
t=0 ot
(3-3.38)

We shall now determine the solution u{x, t) of the heat-conduction equa¬

tion with homogeneous initial conditions and the boundary condition

u(0, t) = pit) , t > 0,

where fi{t) is a piecewise continuous function. An approximate solution is
easily obtained in the form (3-3.37) when we replace fj(t) by a piecewise con¬
stant function. Now when we make arbitrarily small the interval in which
the auxiliary functions are constant, we recognize that the limit value of the
sum (3-3.38) is equal to

j -^rU, t — z)fi{z)dt ,

since for x > 0

lim /Лп-іЩх, t — tn-i) - 0 .

Obviously, the sought solution u2(x, t) of the second problem must satisfy

иг{х, t) = [ ^-(x, t — z)fi(z)dz . (3-3.39)

Jo dt

We shall not consider here whether a passage to the limit is permissible, or

the requirements that must be placed on the function /n(z) in order that this
passage to the limit be applicable.
As we easily see, we have

ax ?—*2/4a2t
3/2
2i/V [a2t]
1
■2a'^-(x, 0,0 = 2 G = -Д=
dx d£ f=0 2f/ я т/azt
Consequently the sought solution in the case of an arbitrary function fi(t)
can be represented in the form65

a2 f *_x_
u2(x, t) = g-x2/4a2( t-r)^T)dz
2т/гг Jt0 [a\t — r)]3/2
or

u2(x, t) = 2а2 Г Щг(х, 0, t - z)n(z)dz . (3-3.40)

J‘o
We note further that in the derivation of formula (3-3.40), besides the
linearity no further special properties of the heat conduction equation were

65 This representation of the solution of the first boundary-value problem with homo¬
geneous initial conditions was mentioned here so that it can be compared to an alterna¬
tive solution in Chapter 5, Section 4.
3-3. PROBLEMS FOR THE INFINITE STRAIGHT LINE 209

used. Further, we never used the analytical expression of the function U(x, t)
but only the property that it satisfies the boundary and initial conditions

U(0, t) = 1 for / > 0

U(x, 0) = 0 for x > 0
or

for t > 0
for t < 0

If a linear differential equation exists with a boundary condition

u(0, t) = fi{t) , t > 0,

homogeneous initial conditions, and auxiliary homogeneous boundary condi¬
tions (when such are present, for example, where x = /), then the solution of
this problem can be expressed in the form

u(x, t) -
s ‘dU_
0 dt
(x, t — т)ц(т)сіт ,

where U(x, t) is the solution of the corresponding boundary-value problem for

(3-3.41)

U(0, o=i.
The principle formulated here, the so-called Duhamel principle, shows
that the constant boundary value causes the principal difficulty in the solution
of boundary-value problems. If a boundary-value problem for a constant
boundary value has already been solved, then one can immediately determine
the solution of the same problem with variable boundary conditions by use
of formula (3-3.41). This principle is often used for the solution of several
boundary-value problems in which the solution is determined only for a con¬
stant boundary value, and it is assumed that the solution for a variable
boundary condition fi(t) then can be represented by formula (3-3.41).
The sum

Ui(X, t) + u2(x, t)
gives the solution of the first boundary-value problem for the semi-infinite
line in the case of the homogeneous equation.
By the use of formula (3-3.29) and the method of odd continuation, it
follows easily that the solution of the inhomogeneous equation

lit — auxx + f{x, t) , 0<x<oo,/>0,

when the initial and the boundary conditions are homogeneous (u(0, t) = 0),
can be represented by the formula

иъ{х ,t) = -^Л=ІТ / 2,! -e-'x+t)2'ia2u-T)}f(Z,r)dZdr.

2Vn Jo Jo V a2(t — r) 1 j
(3-3.42)

The sum
210 PARABOLIC DIFFERENTIAL EQUATIONS

uAx, t) + U2{X, t) + U3(X, t) — ll(X, t) (3-3.43)

is then the solution of the first boundary-value problem

ut = a2uxx + f(x, t) , u(0, t) = ii{t) , u(x, 0) = <p(x).

3-4. PROBLEMS WITHOUT INITIAL CONDITIONS

If one investigates the process of heat conduction at a time which is

sufficiently long from the initial moment, the influence of the initial conditions
on the temperature distribution at the time of the observation is practically
negligible. In this case, then, there arises the problem of determining a
solution of the heat-conduction equation which satisfies one of the three types
of boundary conditions for all t > — oo. If the rod is bounded on two sides
then the boundary conditions are prescribed on both ends of the rod. By
contrast, for a one-sided bounded rod one uses only a single boundary con¬
dition. We shall consider the first boundary-value problem for a one-sided
bounded rod:
Find a bounded solution of the heat-conduction equation in the region
x > 0 which satisfies the boundary condition

u(0, t) = fi{t), (3-4.1)

where /i{t) is a prescribed function. We assume that the functions u{x, t)

and fi(t) are bounded everywhere, i.e.,

I u(x, t) \ < M

I MO I < M.

It will be shown later that the function u(x, t) is uniquely defined. Often
we deal with the boundary condition

fi(t) = A cos wt. (3-4.2)

Fourier has treated this problem, and it was applied to determine the tem¬
perature variation of the earth.66
We write the boundary condition in the form

MO = Аеш . (3-4.2')

From the linearity of the heat-conduction equation it follows that the real
and imaginary parts of a complex solution of the heat-conduction equation
separately satisfy this equation.
If one also finds a solution of the heat-conduction equation which satisfies
the condition (3-4.2'), then its real part satisfies condition (3-4.2) and its imagi¬
nary part satisfies the condition

u(0, t) = fix(t) = A sin cot.

We consider therefore the problem

66 See Section 3-5 § 1.

3-4. PROBLEMS WITHOUT INITIAL CONDITIONS 211

Ut — d tiXx у
Awt (3-4.3)
u{0, t) = Ae'

We seek its solution in the form

u{x, t) = Aeax+pt (3-4.4)

where a and /3 are constants to be determined.

If we insert (3-4.4) into Eq. (3-4.3) and into the boundary conditions we
find

a = —t /3 , /3 = iw ,
a
from which there results

a —
CO (1 + i)

o>2 л/ 2
Hence we have
u(x, t) — Ae±у,(ш/2а2)д:+<(±',(ш/2“2))с+“() (3-4.5)

The real part of this solution, namely

u(x,t) = yle±V(“/2“2>)tcos ^+-j/ ~^~2X + , (3-4.6)

satisfies the heat-conduction equation as well as the boundary condition (3-4.2).

Formula (3-4.6) first of all defines not only one but two functions, since there
are both plus and minus signs in the exponents before the roots. If we
observe that only the functions corresponding to the minus sign are bounded
then we obtain as a solution of our problem:

(о
u(x, t) — Ле~Ѵ(ш/2а2)Іcos cot (3-4.7)
2 a2

The problem without initial conditions for a bounded interval is treated

similarly:
lit - d lixx

u(0, t) — Л cos ad (3-4.8)

u(l, t) = 0 .

If we write the boundary conditions in the form

й(0, t) = Ае~іш1 , й(1, t) = 0

and seek the solution in the form

й(х, t) = X(x)e~iwt (3-4.9)

then we obtain from (3-4.8) for X(x) the differential equation

rf / , id)
X’ -X = 0 or x" + fx = 0 ,
(3-4.10)

’• = /¥=/ik(1+i)
212 PARABOLIC DIFFERENTIAL EQUATIONS

and the auxiliary conditions

X(0) = A , X(l) = 0 . (3-4-11)

Consequently, for X(x) we find

X(x) = A Sln r.(l ~ x) = ХМ + iX,(x), (3-4.12)

sin yl

where Xi and Хг are the real and imaginary parts of X{x). Therefore, for
й(х, t) we obtain the representation

й(х, t) = A Sm r.(l ~ х)-е~іы*. (3-4.13)

sin у l

By separating out the real parts of й(х, t), we finally find the solution of the
original problem without initial conditions in the form

u(x, t) = Xi(x) cos wt + X2(x) sin wt. (3-4.14)

From the above, the explicit determination of Xi and Xz gives no difficulty.

If the boundary function is a combination of harmonic vibrations of dif¬
ferent frequencies, the solution of such a problem can be represented by a
superposition of the solutions corresponding to the individual harmonics.
We shall now prove the unique solvability of problems without initial
conditions for the semi-infinite line. We proceed from the formula

+ -Д=Г 7 J {e-U-f)V4a*(t-«o> _ g-(x+f.*/4„*Ct-t0,}M(£ to)d$

2Vn Jo V a2(t — t0)
= Ii + h , t ^ to (3-4.15)
which represents every bounded solution of the heat-conduction equation
through its initial function u(x, t0) and its boundary function u(0, t) — yi(t) in
the region x ^ 0, t t0.
We show next that the limit

lim /,(*, t) = 0 (3-4.16)

exists if the following holds for all t:

I u(x, t)\ < M .

In fact, we have

ЛI < e 0,1 dai e azdaz

V* IJ -x/2 Va2(i-(.)

fyj Г*/г Va*t*-*0>

e~a da ,
- 1/^T2 0
where
3-4. PROBLEMS WITHOUT INITIAL CONDITIONS 213

£ X i ^ X
«1 - ——7===- and a2 = ——. =- .
2l/aHt -t0) 2i/a2(t - t0)
From this (3-4.16) follows, since as t0-*co both x and t remain fixed. If in
(3-4.15) the variables * and t remain fixed and t„ tends towards — oo, then
u(x, t) is equal to the limit value of the first summand. We therefore obtain
_x_
u(x, t) = _«!_[ (x2/4a2(f—r)
p(x)dx, (3-4.17)
2V7) [a2(t-r)f2
which shows that there cannot be two different solutions of our problem.
We can also prove that formula (3-4.17), for every bounded piecewise con¬
tinuous function represents the solution of the corresponding problems.
We can proceed analogously for problems without initial conditions for a
finite interval 0 ^ x ■й /. Here if the boundedness of the solution is not re¬
quired these problems would not be uniquely solvable, since the function

un(x, t) = Ce (*n/1) a ‘ sin —j~x

for arbitrary n produces a solution of the problem in question with homo¬

geneous boundary conditions. These solutions as > —со become unbounded.
The uniqueness of a bounded solution can be shown immediately.

Problems

1. Determine the Green’s function for (a) a one-sided bounded rod for bound¬
ary conditions of the first and second type when no heat exchange results
on the lateral surface; (b) a two-sided unbounded rod when there is heat
exchange on the lateral surface; (c) a one-sided bounded rod in the presence
of a heat exchange on the lateral surface and with boundary conditions of
the first or second type.
2. Find the Green’s function for a one-sided bounded rod with heat insulated
lateral surface for the third boundary value problem, where the boundary
conditions have the form

-f--hu(0,t) = f(t).
dx

Solution:
_1_ _|_ g-(x+()2/ta2lt-T)
G(x, £,t — t)
2Vruil{t
g-hz-(a+t+z)2/ia2 (

}•
3. Solve the heat-conduction equation for (a), (b), and (c) of Problem 1 when
(1) a heat source Q = Q(t) develops at the point x = £0 (Q = Q0 = const.); (2)
an initial temperature distribution u(x, 0) = <p(x) is prescribed, where
for 0 < x < l
outside of the interval (0,/) ;
214 PARABOLIC DIFFERENTIAL EQUATIONS

(3) the heat sources are distributed throughout the entire rod with the density
f(x, t) while the initial temperature is equal to zero, where f(x, t) = q0 = const,
(stationary sources).
4. A one-sided bounded rod with heat-insulated lateral surface is uniformly
heated up to a temperature

u(x, 0) = n0 — const., x > 0.

Let the ends of the rod be held at the time t = 0 at the constant temperature
zero

«(0, 0 = 0, t > 0.
Determine the temperature u(x, t) of the rod and use the tables for the error
integral
2 f1 2
<P(z) = —7=1 e a da
V к Jo

to construct the graphic representation of u{x, t) with respect to x in the

interval

0 йх ^ l for
16a2 2 a2

Hint: In a suitable manner introduce the dimensionless variables

5. The end of a one-sided bounded cylinder is opened at the initial moment

/ = 0. The surrounding atmosphere contains a gas whose concentration is u0.
Find the concentration u(x, t) of the gas in the cylinder for t > 0 and
x > 0 when the initial concentration in the cylinder is equal to zero. L^se
the tables of the error integral to determine how much time must pass until
the concentration of the gas in the cross section at the distance / from the
end of the cylinder has reached 95 per cent of the exterior concentration.
Further, determine the law of motion of the layers of constant concentration.
6. Let a heat flow kux(0, t) — q(t) be directed toward the end of a one-sided
bounded rod whose initial temperature is equal to zero. What temperature
u(x, t) does the rod attain when (a) the lateral surface of the rod is heat in¬
sulated; (b) on the lateral surface the rod undergoes a heat exchange (accord¬
ing to Newton’s law) with the environment whose temperature is equal to
zero? Consider the special case q — q0 = const.
7. The end of a one-sided bounded rod is maintained at the constant tem¬
perature u0; on the lateral area of the rod there occurs a heat exchange with
the surroundings which is at the constant temperature /q. The initial tem¬
perature of the rod is zero. Find the rod temperature u(x, t).
8. Solve problems 6(a) and (b) under the assumption that u{x, 0) equals ua
— const.
9. Determine the temperature distribution in a one-sided bounded rod with
heat-insulated lateral areas at whose ends (а) а temperature k(0, t) — A coswt
3-5. APPLICATIONS TO CHAPTER 3 215

exists; (b) a heat flow Q{t) = В sin wt occurs; (c) Newtonian heat exchange
with the environment exists, whose temperature changes according to the
law v(t) = C sin cot.
10. By using the method of reflections construct the Green’s function for a
two-sided bounded rod with heat-insulated lateral surfaces for boundary con¬
ditions of the first and second types.
11. Let a two-sided unbounded rod be comprised of two homogeneous rods
which are brought together at a point x = 0 and are characterized by ax, kx
and a2, kz respectively. Let the initial temperature be

for x < 0
u(x, 0) = <p(x) = j^1
for x > 0 .

Determine the temperature u(x, t) in the rod when the lateral surface is heat-
insulated.

3-5. APPLICATIONS TO CHAPTER 3

1. Temperature waves

The treatment of the problems of propagation of temperature waves in

the earth is one of the first examples of the application of the mathematical
theory of heat conduction developed by Fourier for the investigation of natural
phenomena.
As is known, the temperature of the earth’s surface changes very dis¬
tinctly in a daily (day-night) and a yearly (summer-winter) period. We shall
treat the propagation of the periodic temperature distribution in the earth
which we assume to be a homogeneous half-space 0 ^ x < °° (from the surface
to the interior). This problem is a characteristic example of problems with¬
out initial conditions. After several temperature variations on the earth’s
surface, therefore, the influence of the initial temperature is small in com¬
parison with the influence of other factors which we likewise disregard (for
example, the inhomogeneity of the earth). In this way we then arrive at
the following problem:67
Determine a bounded solution of the heat-conduction equation
| a”
^ **

ato

*-4.
ІІЛ

(3-5.1)
H

J3

8
II

1
Cu

which satisfies the condition

u(0, t) = A cos wt. (3-5.2)

We have already considered this problem earlier in this chapter. Its

solution has the form (see Section 3-4§7)

u{x, t) — Ae' 1ш/2а")І cosf — -^2 x — ooA . (3-5.3)

67 H. S. Carslaw, Mathematical Theory of the Conduction of Heat in Solids, London,

1921.
216 PARABOLIC DIFFERENTIAL EQUATIONS

On the basis of this solution the temperature waves in the earth can be
described in the following manner: When the temperature of the surface
changes periodically over a long period of time, then a temperature fluctuation
with the same period is developed in the earth. Therefore:
1. The amplitude of the distribution decreases exponentially with the
depth: A(x) = Ae~v^iu2^, i.e., an arithmetic increase in the depth corre¬
sponds to a geometric decrease of the amplitude (Fourier’s first law).
2. The temperature distribution in the earth takes place with a phase
displacement. The time <5 between the occurrence of the temperature maximum
(minimum) in the earth and the corresponding time point on the earth surface
is proportional to the depth:

5=/ 2^X
(Fourier’s second law).
3. The depth of penetration of the temperature in the earth depends on
the amplitude of the temperature distribution on the surface. The relative
change of the temperature amplitude is equal to

A{x) _ _ ''(ш/2аг) *

A
This formula shows that the depth of penetration of the temperature is
smaller when the period is smaller. For two temperature distributions with
periods Ti or T2, the corresponding depths xt and x2 in which the relative
temperature changes coincide, are connected by the relation

(Fourier’s third law). For example, it shows the comparison between the
daily and the yearly variations (T2 = 3657\):

x2 = і/365жі = 19.1 Xi
that is, the depth of penetration of the yearly distributions with equal am¬
plitude on the surface is 19.1 times as great as the depth of penetration of
the daily distribution.
For example we shall give the results of observations of the yearly tem¬
perature distributions of the station Gosch (Amur):68

Depth (m) Amplitude (°C)

1 11.5
2 6.8
3 4.2
4 2.6
These values show that the amplitude of the yearly distributions at 4 m

68 M. I. Sumgin, S. P. Kachurin, N. I. Topspikhin, V. F. Tumel, General Ground

Frost Science, Moscow, 1940, Ch. V.
3-5. APPLICATIONS TO CHAPTER 3 217

depth has diminished to 13.3 per cent of its value on the earth’s surface,
which is equal to 19.5 degrees.
On the basis of these values one can determine the thermal conductivity
of the earth:

1n Mx) ПГ .» _ (ox2
A ~ V2tf ’ 2 Inг(А(х)1А) ■
From this we find the thermal conductivity,

а2 = 4-l(T3 —.
sec

The maximum temperature at 4 m depth four months later would therefore

reach that of earth’s surface.
However it should be noted that this theory holds only for heat conduc¬
tion in dry soil or rocky terrain. With moist soils the temperature pheno¬
mena are more complicated; with frozen ground a latent amount of heat would
be freed which this theory does not consider.
The thermal conductivity of a body is one of the characteristic quantities
which is of importance in the investigation of its physical properties and also
for different technical calculations. The investigation of the propagation of
temperature waves in rods depends on one of the laboratory methods for the
determination of the thermal conductivity.69
Let the end of a sufficiently long rod be heated and cooled periodically
as If we develop this function in a Fourier series,
oo
2ixn Clp 2j:n
+ 2 (a„cos
n=l \ T
t + bn sin
~2
2 An cosi
n=1 T

An — Va\ + b\ , d°n = -2~_(гг + arctg-^ .

(T is the period) and determine the temperature waves corresponding to the

individual summands, then we find that the temperature u{x, t) for an arbitrary
x is a periodic function of time; hence for its nth harmonic vibration we have
2k n 2n n .
un{x, t) — an(x) cos t + bn{x) sin
~T~

КП
= Ane~'/unlTai)x cos x 2m-t + A
Taz T
or

l/ an(x i) + bnjx i) = g-'/(irn/ru2Hi1—x2)

Val{xi) + blix2)
This formula shows the following: If one measures the temperature
changes at any two points Xi and хг during an entire period, then by deter-

69 Special Physical Practices, vol. I, Prob. 35, Moscow, 1945; V. I. Iveronova, Physical
Practice, Prob. 23, p. 117, Berlin, 1957.
218 PARABOLIC DIFFERENTIAL EQUATIONS

mination of the coefficients an{xft, bn{Xi), an{xf), bn{xf) by means of harmonic

analysis, one can determine the thermal conductivity a2 of the rod.
The periodic temperature variations (temperature waves) in a rod, for
example, can be generated in the following manner. One end of the rod is
placed in an electric furnace and a repeated pulsed current passed through.
Because of this periodic warming in the rod, a periodic temperature fluctua¬
tion can result after a certain time. Now by means of a thermostat, we can
measure the temperatures u(xi, t) and u(x2, t) at any two points Xi and x2
during a complete period of change of the boundary influence and thus obtain
U\ and u2 as described above; in this manner the thermal conductivity a2
of the material of the rod can be determined. There are two conditions
necessary for application of this theory. First, the rod must be heat-
insulated on the lateral surface; and second, the temperature at the other end
of the rod must be controlled, if we are to be justified in using here the
theory of temperature waves in a one-sided bounded rod. Therefore, we have
to prove that the temperature at the free end of the rod is constant, and
this is accomplished with the help of another thermostat.

2. The influence of radioactive decay on the temperature of the

earth's crust

For the estimation of the state of temperature in the earth’s interior some
hint can be obtained from observations on the earth’s surface. The impor¬
tance of this lies in the fact that the daily and the yearly temperature varia¬
tions take place only in a relatively thin layer of the surface (approximately
10—20 m for the yearly variations), while the temperature below this layer
changes very slowly in the course of time.
It has been observed in ravines and caves which lie 2—3 km beneath the
earth’s surface that the temperature increases with increasing depth on the
average of 3°C per 100 m.
The first test, carried out at the end of the 1800’s, to give a theoretical
explanation for the observed geometric gradient, met with insurmountable
difficulties.70 It was concluded that earlier the earth must have been radiat¬
ing heat and that it cooled gradually. The initial temperature characteriz¬
ing this cooling process must have been of the order of magnitude of T0 —
1200°C (the melting temperature of rock); the surface temperature is of the
order of magnitude of 0°. Further, the surface temperature may not have
changed essentially (not more than 100°) from the period when vegetable or
animal organisms first occurred on the earth. The cooling process, considered
in this sense as purely quantitative, leads then to the solution of the heat-
conduction equation
du _ 2 d2u
dt dz

70 H. S. Carslaw, Mathematical Theory of the Conduction of Heat in Solids, London,

1921, Ch. 3.
3-5. APPLICATIONS TO CHAPTER 3 219

in the half space 0 < z < со with the initial and boundary conditions

u(z, 0) = T0
u(0, t) — 0 .
The solution of this problem was treated in Section 3-3; we obtained it
in the form

2[г'г
u(z, t) — Tf da .
°і/іГ)о
The gradient of this function at z — 0 is given by

du_T0 ^-(z2/4a2t) Ta _
dz z=o л/л Vazt z=o Vл i/a2t ’
If we substitute in this expression the value of the geothermal gradient у =
3 • КГ4 degrees/cm, T0 — 1200°C, and the value a* = 0.006 cm2/sec (the average
value of the experimentally found coefficients of the thermal conductivity
of granite and basalt), then for the time duration of the cooling process we
obtain the value / = 0.85 • 1015 sec = 27 million years. Such an estimate of
the age of the earth is, however, incompatible with the geological facts. The
approximate nature of the theory being considered—that is, disregarding the
earth’s curvature, the variability of the thermal conductivity, and the in¬
accurate value of To—naturally cannot strongly influence the order of mag¬
nitude of the value found for the age of the earth, which according to modern
investigation has been estimated at around 2 • 109 years.
The physical scheme of the course of the temperature in the earth can
be represented in a different manner, using radioactive decay. The radioactive
elements distributed in the earth’s crust generate by their decay a certain
amount of heat which naturally contributes to the warming of the earth.
Therefore, the corresponding heat-conduction equation must have the form

du 2 d2u r
~dt a +
where A represents the volume density of the heat source. On the basis of
numerous measurements of the radioactivity of rocks and the amount of heat
generated by their radioactive decay we usually choose the value

A = 1.3 • 10"12—^— .
cm sec

This value is considered to be the heat developed by uranium, thorium, and

potassium, including their decay products.
We shall now assume that the density of the radioactive source in the
interior of the earth sphere is constant and has the value A on the upper
layer of the earth’s crust. Under this assumption we obtain for the amount
of heat generated in the entire sphere during a unit of time

Q = -|-7ГІ?3Л .
220 PARABOLIC DIFFERENTIAL EQUATIONS

We assume further that the earth is not warmed by the radioactive-generated

heat. In this case, for the amount of heat flowing per unit of surface area
we obtain
du Q
q — k
dz 4 kR2

where k is the coefficient of heat conductivity and du/dz\z=0 is the geothermal

gradient at the earth’s surface.
From this we find

du > AR ^ g 3 iq-2 degrees

dz *=о = 3k cm

where R = 6. 3 • 103 km is the earth’s radius and k = 0.004 is the average heat
conductivity of the soil rocks.
Accordingly, under the assumption that the distribution of the radioactive
elements are constant and the earth is not warmed by the radioactive decay,
the calculated geothermal coefficient exceeds the observed value

у = 3 • 1СГ4 degree/cm
by two orders of magnitude.
We now drop the hypothesis that the radioactive elements are distributed
uniformly and assume instead that they lie in a layer of thickness H of the
earth’s crust. Without considering the curvature of the earth we then obtain
for the determination of the stationary temperature the equation

d2u _ A
for 0 ^ г ^ H
dz2 k
= 0 for z > H

with the conditions

u(0) = 0 , = 0.
The solution of this problem is obviously given by

0 йг^Н
-у) ■
_ A H2
z^H
~ k 2 '

since this function including its first derivative at z — H is continuous and

the conditions of the problem are satisfied.
If now we calculate the value of the gradient of this function at z = 0

du _ AH
dz г=о k ’

and compare it with the observed value

у = 3 • 1СГ4 degree/cm ,
3-5. APPLICATIONS TO CHAPTER 3 221

then we find

H— « 106 cm = 10 km .
A
We shall now estimate how the assumption that the temperature is station¬
ary affects the size of the geothermal coefficient. To this end we consider
the solution of the heat-conduction equation

dw
+f
~dt

f=— , 0 йгйН
cp

= 0, z > H

with the homogeneous initial and boundary conditions

w(z, 0) = 0
w(0, /) = 0 .

As we saw in Section 3-3 the solution of this problem is represented by the

integral

w(z, t) = I I G(z, C, t - z)f(C)drdC ■

In this case G is the Green’s function for the one-sided bounded straight line:

1 ~(z—g) 2/4a2(t — г) (z+£) 2/4a2(t — r) j

G{z, C, t - r) =
2\/л і/a2(t — t)

For the value of the gradient of the function w(z, t) for z = 0, by considera¬
tion of the value of / there results therefore

dw
e-ii/^t-T)d^dz
dz Cp2\/ к Jo Jol/[tf2(/ — t)]
A ft 1 ГЙ2/ 4a2(t-r)
——7=1 —. e-«dadz
cpv л JoVa\t — z) Jo

[1 - e~a'l"4\de , 0 = t —z.
cpVn Jol/«2^
Therefore

dw A Г 2 Vt H f „-СТ2 da \
Л \ ^ 2 f f
dz z=0 cp\/n 4 « a )o0 0 J
where

a=
H
(T0 —
я da a2 dd
2yVtf ’ 2\/aH ’ a2 ~~ H VaM
Now
222 PARABOLIC DIFFERENTIAL EQUATIONS

о
e^'da e a da ,
Oo ao
so that

dw
—Y-t [i - <r"2/<‘2'] + нЛЛШ _e~^da\ . (3-5.4)
dz cpa l. \/n V К Jff/Fa2! J
Further,
dw
lim-
(dz

then cpa2 = k, the limit value of the first summand in the curved brackets
vanishes and the limit value of the second summand is equal to H.
We now calculate the difference between the limit value of dw/dz obtained
above and the value of dw/dz for

t = 2 • 109 years = 6 • 1016 sec.

Then <r0 is small:

H 10e 1
<T0 : 0.025.
2УаЧ 2г/6 • ІО-3 • 6 • ІО16 2 • 19

By expanding in series the functions appearing in formula (3-5.4) we obtain

А ц _ dtv 1,2. , , 2 1 A H ■ 0.04,

-(ffl + •) +
k dz -i" l/ тта0 И7Г “J ~ k

i.e., dw/dz |z=0 differs by about 4 per cent from its limit value as / —»oo.
We can calculate the function w{z, t) for z > 0 and prove that for z Ш H,
when one introduces the age of the earth for t, the function still deviates
considerably from its limit value71 (although, as we see at the surface, the
gradient is practically equal to its limit value).
These considerations are all only estimates; nevertheless we must conclude
from the fact that the rate of the radioactive decay does not change with
the temperatures and pressures obtainable, that the concentration of the radio¬
active elements with increasing depth decreases rapidly when the value of
A for the upper layers of the earth’s crust, as it has been determined by
numerous measurements, is taken as a basis. A physical theory which would
allow us to derive a law for the decrease of the concentration of radioactive
elements with increasing depth has not yet been developed.

3. The method of analogy in the theory of heat conduction

The method of analogy has been shown to be very useful for the solu¬
tion of a series of problems of heat conduction. As an example, therefore,
we shall consider the following two problems.

71 A. N. Tychonoff, “Concerning the Influence of Radioactive Decay on the Tempera¬

ture of the Earth’s Crust,” Investigations of the Academy of Science of the USSR,
Division of Mathematics and Natural Sciences, pp. 431-459, 1937.
3-5. APPLICATIONS TO CHAPTER 3 223

1. The Green's function for the infinite straight line. The heat-conduction
equation obviously remains unchanged by the substitution

x' — kx , t' = k2t. (3-5.5)

That is, when the linear scale of the rod changes &-fold, and the time scale
changes &2-fold.
We seek next the solution of the heat-conduction equation

ut — a2uxx (3-5.6)

with the initial condition

for x > 0
u( x, (3-5.7)
for x < 0 .

Also, the initial condition (3-5.7) is not changed by the above scale changes.
Therefore, for the function u(x, t), the equation

u(x, t) = u(kx, k2t) (3-5.8)

holds for all x, t, and k.

Now if we set

(3-5.9)

then we obtain

u(x’,) = {wr- т)=щ/(Фг)- (3-5.10)

The function и therefore depends only on the argument

x
(3-5.11)
2 yT •

For the derivatives there results from (3-5.10)

d2u _ d2 f 1 du _ x ■ Up d f _ _ z_ d f
dx2 U° dz2 4/ ’ dt 4ti/2 dz U° 21 dz

If we introduce this expression into Eq. (3-5.6) and divide it by u0l4/, then it
follows that

>d2f (3-5.12)
— 2
dz2 dz
with the auxiliary conditions

/( —oo) = 0 , /(o°) = l, (3-5.13)

which correspond to the initial condition for the function u(x, t).
By integration of Eq. (3-5.12) we obtain
Ct „ fz/a
a2^r=-2z, f = Ce-^2 , f=c\ e~i la2d£ = Ci \ e--?2JC-
J —oo J —oo
224 PARABOLIC DIFFERENTIAL EQUATIONS

Therefore, the lower limit is so chosen that the first condition of (3-5.13) is
fulfilled. The second will similarly be fulfilled if we set

Therefore

u(x, t) — e-**de = _££оГ 1 + Ф (3-5.14)

V*)- 2L 2 VaH ]•
where

Ф(г) = -%=[
VK Jo
is the error integral. If the initial value has the general form

uо for X > X
u{x, 0) (3-5.15)
0 for X < X

then

x — X
u(x, t) = у j^l + ф(- (3-5.16)
2\/a4 )]
We turn now to the solution of the second auxiliary problem for which
the initial conditions have the form

u(x, 0) = 0 for X2 < X

= Uo for ДГі < X < Xz (3-5.17)

= 0 for X < Xi .

In this case

u(x, t) = Y

The initial temperature u0 corresponds to an amount of heat

Q — cp{x2 — Xi)u0.
If Q = cp, then

u(x, t) =- (3-5.18)
x2 — Xi
The Green’s function then obviously represents the limit value of the func¬
tion u(x, t) as (хг — ДГі)—>0 in the passage to a point source.
The corresponding passage to the limit in formula (3-5.18) gives

u(x, t) — — (3-5.19)

since on the right side of (3-5.18) the difference quotient arises whose limit
value is the derivative in (3-5.19).
Now if we perform the differentiation we find
 3-5. APPLICATIONS TO CHAPTER 3 225

и^‘)=-^77-Ѵше~{-ч"''л • (3-5'20)

1. e., u(x, t) = G(x, Xi, t) is the Green’s function.

2. Boundary-value problems for the nonlinear heat-conduction equation. We

consider the equation

д Г,. . ди 1 du
ITJ = cpir (3-5.21)

and seek the solution which satisfies the boundary condition

u(0, t) = Mi (3-5.22)

and the initial condition

u{x, 0) = m2 . (3-5.23)

Also in this case a substitution of the form (3-5.5) changes neither the Eq.
(3-5.21) nor the auxiliary conditions (3-5.22) and (3-5.23).
From this it follows that

(3-5.24)
"fe') = /(w) = /(2)’ 2 = w-
By using this expression we obtain for / the equation

(3-5.25)

with the auxiliary conditions

/(0) - Mi, /(oo) = m2. (3-5.26)

The function f(z) can be calculated by numerical integration in those cases

in which it cannot be found in closed form.
Equation (3-5.25), under very general assumptions on k and cp, also pos¬
sesses a uniquely determined solution which satisfies the conditions (3-5.26).
As an example of (3-5.21) we consider the case in which k(u) = k0 + kxu
is a linear function in м and cp is a constant. By a suitable change of the
time scale and a scaling of the values of м we then obtain the equation

3 Гп I i ди 1 du_
1*11 + ІІГ (3-5.27)

with the initial and boundary conditions

u(x, 0) = 0 , m(0, t) ! . (3-5.28)

Now if we set

x
u(x, t) = f(z) ,
2ѴГ '

then f(z) must satisfy the condition

 226 PARABOLIC DIFFERENTIAL EQUATIONS

/(0) = 1 , /(oo) = 0. (3-5.30)

Figure 42 shows the results of the numerical integration of (3-5.29) for dif¬
ferent values of a.

FIG. 42.

4. The solidification problem

If the temperature of a body changes, then its physical (state) condition
changes. If, in particular, the temperature change includes the melting point,
then the body is transformed from the fluid to the solid state (or conversely).
Moreover, the temperature on the transition surface during the entire time
remains constant. Owing to the motion of the transition surface, a latent
heat of fusion is freed (melting). In the following we shall formulate the
additional conditions which must be fulfilled by the solidifying surface.72
To this end we shall consider the problem in which the transition surface
is described by the plane x = £((). During the time interval (t, t + At) the
boundary x = £ can be displaced from the point £ = x% to the point £ = x2 =
Xi + J£. In so doing, the mass pJ£ solidifies (or melts in case J£ < 0) and
an amount of heat XpA£ is released.
To satisfy the energy conservation law, this amount of heat must be
equal to the difference of the amount of heat passing through the boundaries
£ = Xi and £ = хг. Therefore, the condition

72 Ph. Frank and R. v. Mises, Differential and Integral Equations of Mechanics

and Physics. Braunschweig. 1930/1935; Ch. 13.
3-5. APPLICATIONS TO CHAPTER 3 227

must also be fulfilled where kx and k2 are the heat conductivity coefficients
of the first and the second states while X is the heat of fusion.
If now in this expression we let Jt tend to zero, then the auxiliary con¬
dition on the transition surface takes the form

(3-5.31)

This condition is valid for the solidification process (for > 0 and d$/dt > 0)
and also for the melting process (for < 0 and dtldt < 0); the direction in
which the process proceeds can be determined from the signs on the left side.
Now we shall consider the freezing process of water where the melting
temperature is zero. Let the half space x ^ 0 be filled with water. This
amount of water then is bounded on one side by the plane x = 0. At the
initial moment t = 0, the water has the constant temperature c > 0. If the
surface x = 0 steadily maintains the constant temperature ci < 0, then the
transition surface x — f penetrates, with time, into the fluid.
According to the above, the problem of the temperature distribution in
the freezing water and the determination of the velocity of propagation of
the transition surface leads to the equations

dui d2ul
for 0 < x < £
~df dx2
(3.5.32)
ди 2 32m2
for £ < X < OO
dt dx2
with the auxiliary conditions

Ml = Ci for X = 0
(3-5.33)
u2 — c for t = 0 .

Moreover, on the transition surface the conditions

Mi = m2 = 0 for x = £ (3-5.34)

Зм i , Зм 2 dt,
ki (3-5.31)
dx i*={ dx ^if

must be fulfilled where ki, a\ or k2, al are the coefficients of the heat and
thermal conductivity of the solid and liquid states.
For the solution of the problem we make the Ansatz

Mi = Ai + ВіФ\
2міі/1

— л‘+

where At, , A2, and B2 are constants to be determined while Ф is the error
integral
2
228 PARABOLIC DIFFERENTIAL EQUATIONS

If we consider the conditions (3-5.33) and (3-5.34) then we obtain from (3-5.33)

Ai = Ci, A% + B2 — c
and from (3-5.34)

А' + В'І2^7Г) = °

A° + В’Ч^7Г) = 0 •
These conditions must hold for arbitrary values of t. This is possible only
when

f = aVT (3-5.35)

is valid, where a is a fixed constant. The expression (3-5.35) determines the

law of motion of the transition surface between the fluid and the solid states.
For the constants Alt Bu A2, B2, and a the following expressions result:

Al Cl ’ Bl ФМЩ
(3-5.36)
л _ сФ(а/2а2) B _ c
1 — Ф(а/2а2) ’ 2 '1— Ф(а/2а2) ’
For the determination of a the requirement (3-5.31) leads to:

^1clg~0,2/4a ■ k2ce~a2/*a 1 _ .
(3-5.37)
ахФ{а!2аі) a2[l — Ф(а/2а2)] ^ 2

The solution of this transcendental equation gives directly the value of a.

That at least one solution of cx < 0, c > 0 exists follows from the fact that
the left side of this equation varies from — oo to +oo when a passes through
all values from 0 to oo,,s while the right side for such a assumes only the
values between 0 and — oo. In the case when c is equal to the melting tem¬
perature (c = 0), expressions (3-5.36) and (3-5.37) for the determination of the
coefficients reduce to a simpler form:

Ci
A — B2 — 0 ,
2 A i — Ci f Bi (3-5.36')
Ф{а/2аі)
and

kicxe a /iai _ _ . т/гг

(3-5.37')
аіФ(а/2аі) ^ 2
If we set <x\2a.i = /3, then (3-5.37') can also be written in the form

1 е~Р2

where D is determined by

73 For an asymptotic representation of the function 1 — Ф(г) as z -> oo see the Ap¬
pendix.
3-5. APPLICATIONS TO CHAPTER 3 229

D= < 0 .
kiCi

By use of the graphic representation of the function <p(/3) = Ф(і3) given

in Figure 43, the value of a can be determined.
The solution of the solidification prob¬
lem can also be found with the help of the
analogy method as discussed in Section
3-5§3. To a certain degree the solidifica¬
tion problem is a limiting case of a non¬
linear boundary-value problem similar to
that in the previous section. The coef¬
ficients of heat conductivity and heat
capacity for solidification problems are
piecewise constant functions. Moreover,
the heat capacity for u(x, t) = 0 is infinite.
This case can also be obtained as a limiting
case as e—>-0 if one assumes that the latent
heat is not freed instantaneously but is
released during a fixed interval (—e, +e).
Therefore, we naturally must have
FIG. 43.
c(u)du = X.

One can easily prove that all the conditions of the problem remain un¬
changed when the length scale changes &-fold and the time scale &2-fold.
Therefore the solution depends only on the argument xj\/T, i.e.,

u(x-t)=fwij-
Hence, in particular, it follows that the motion of the zero isotherm can be
described by the equation

f = aq/1 ,
when a is the value of the argument, for which

f(a) = 0
is valid.
For the determination of the of the function f the following conditions
result:
dfj_
a i d2fi 2z for 0 < z < a
dz2 dz
-«d'A dft
^2 2
i
2z for a < z < oo
dz2 dz
/i(0) = Ci , /2(00 )-c , fi(a) = /2(a) = 0

kifl(a) — кгf'2(a) = Xp~ .

230 PARABOLIC DIFFERENTIAL EQUATIONS

Therefore,

f(z) =
Ш
fi{z) — A i + ВіФ\
for0
< z < a

fi{z) — A2 + 52Ф for a < z < oo .

For the determination of the constants Al, Bi, Az, Д,, the requirements
(3-5.33) and (3-5.34) must again be used, from which (3-5.36) results. For the
determination of a the condition (3-5.37) is used. Therefore the respective
analytical parts of both methods of solution coincide.
These considerations show that the solidification problem can also be
solved in the cases in which the latent heat is freed not for a fixed tempe¬
rature but in a certain temperature interval. In a similar manner the solution
is possible when not one but many critical temperatures are given. Such
cases can occur for changes of state in the transition from one crystalline
structure to another (for example, for the recrystallization of steel).

5. The Einstein-Kolmogoroff equation

Microscopic particles which are free to move in a medium are in a con¬

stant, permanent state of unordered motion (Brownian molecular motion). The
probability therefore that a particle found at time t0 at the point M0 is found
at time t in a small neighborhood AV of the point M can be described by
the function

W(M, t\ Mo, t0) • AV . (3-5.38)

In this case by “probability” we mean the following:

If a sufficiently large number N of particles moves from the point M0
(we shall disregard the opposing influences) during a small time interval
t0 + At, W(N, t\ M0t0) is equal to the concentration of these particles as At-+0
at the point M at the time t, if the total mass of the particles issuing from
the point M0 is chosen as the unit of mass of the particles.
A similar phenomenon occurs also in the diffusion of gases in, for ex¬
ample, a gaseous medium.
The function W(M, t; M0, t0) represents the Green’s function correspond¬
ing to the unit mass. Obviously, we must have the condition

j W(M, t; Ma, t0)dV,M — 1 , t > to • (3-5.39)

If the initial concentration of the particles at a fixed time t0 is equal to <p(M),

the concentration w(M, t) of these particles at time t > t0 must be equal to

u(M, t) = J W(M, t; P, to)<p(P)dVP , (3-5.40)

where the integral is extended throughout all space.

3-5. APPLICATIONS TO CHAPTER 3 231

From this relation results the Einstein-Kolmogoroff equation:74

W(M, t; Mo, t0) = I W(M, t\ P, Ѳ) W(P, Ѳ\ Mo, t0)dVP , U < Ѳ < t. (3-5.41)

We shall now show that, under certain conditions which the function
W(M, t\ Mo, to) must satisfy, the solution of the Einstein-Kolmogoroff equation
satisfies a parabolic differential equation. Therefore, we shall consider the
case in which the position of the point M is described by a single coordinate
x and assume that the function W{x,t; x0, t0) satisfies the following conditions:

(a) lim —-— = lim — [(# — ?) W{x, t + r; £, t)df = A(x, t). (3-5.42)
r-0 г T J

If the particles during the time z are displaced from the position £ to
the position x, then (x — £/r) is the average velocity of the particles. Con¬
sequently, requirement (a) means that the ordered motion of the particles
takes place with a finite velocity.

(b) lim (x ~ ^)Z- = lim — ( (x - 6)2 W(x, t + r; £, t)d£ = 2B(x, t). (3-5.43)
r-o r : ]

The quantity (x — f)2 does not depend on the direction of the displace¬
ment of the point x with respect to the point £. The average value of the
square of the distance during the time r,

{x — f)2 = j(* — f)2 W(x, t + z;£, t)d$ ,

is ordinarily regarded as the measure of the unordered motion during this

time interval. The requirement (b) says, therefore, that the average square
for sufficiently small z shall be linearly dependent on the time.

(c) lim = lim— • W(x, t + r; £, t)d$ = 0. (3-5.44)

r-0 Z T J

The Green’s function W(x, t + r; f, t) for small z values shall decrease

rapidly when \ x — £|—>oo and increase when | x — £ | is small.
Now in order to derive the Einstein-Kolmogoroff differential equation we
multiply both sides of Eq. (3-5.41) with an arbitrary function ф(х), which,
including its derivatives, vanishes at the boundary of the region of integra¬
tion, and integrate over the entire region:

J W(x, t + r; x0, t0)<p(x)dx = J W{£, t\ x0, t0)df j W(x, t + z; $, t)<p(x)dx .

Then we develop ф(х) on the right side according to Taylor’s formula,

Ф(х) = ф(£) + ф'Шх — f) + ^kx - О2 - f)3,

74 M.A. Leontovich, Statistical Physics, Moscow, 1944, Ch. 6; A. N. Kolmogoroff,

“Analytical Methods in the Calculus of Probability,” Uspechi mat. nauk. 5, 1938.
232 PARABOLIC DIFFERENTIAL EQUATIONS

where f* is a value lying between x and f. After some simple transforma¬

tions, division by t gives

W(x, t + r„ Xp, t0) — ITU, f; x0, f0) ^

= | WG, t; x0,t0)^'(S)^^-+ ф"(£){х ~J)2 УZ

+ ^ ^ W(f. *; *», 4) Щт, t + r; £, .

If we assume ф"'(х) to be bounded, i.e.,

\[ф"'(х) \ <A,
and take into consideration

I TT(f, t; x0, t0)d£ = 1,

we obtain

1
'"(£*)(* - ?)3 W(Z, t; Xo, /0) W(X, t + r,$, tmdx
T
^ — fI * — f |3 W(x, t + r; $, t)dx = AAl—£11 .
T J Г

From condition (c) it follows that this expression approaches zero as

r—>0. Therefore, with the use of conditions (a) and (b) and as r—>0, we find

dW(x, t; x0, t0)

Шх) dx
dt

= j w(?, t-, xo, toWitmt, t) + 0"(f)5(f, tm .

Now if we integrate the right side by parts and note that the function ф(х),
including its derivatives, vanishes at the boundary of the region of integra¬
tion, then we find

d(AW) d\BW)
dx dx2 ]
dx — 0 .

Since this relation must hold for an arbitrary function ф(х), the Einstein-
Kolmogoroff differential equation

dW d(AW) d\BW)
dt dx dx2

results for the function W{x, t; x0, t0).

This equation, similar to the heat-conduction equation, is a parabolic dif¬
ferential equation and can be written in the form

Wt=-H-(BWx) + aWx + pW, (3-5.46)

dx
3-5. APPLICATIONS TO CHAPTER 3 233

where

a = — A + Bx
P = — Ax + Bxx = ax.
From Eq. (3-5.46) we recognize that the quantity В has the physical signi¬
ficance of a diffusion coefficient. If the process considered is homogeneous
in time and space, i.e., the function W depends only on the difference f =
x — x0 and Ѳ — t — t0, then the coefficients A and В are not dependent on x
and t and are constant. Eq. (3-5.45) is then a differential equation with con¬
stant coefficients:

dW л dW , „ d2W
—T — — /І-г -О -5“ . (3-5.47)
dt dx dx
If W depends only on \x — f|, i.e., the probability for a right-and left-sided
displacement at the same distance from the point f, coincide, then obviously
A must be equal to zero. Analytically this follows from the formula (3-5.42),
since the integrand is an odd function.
In this case, Eq. (3-5.45) is transformed into the simple heat-conduction
equation

dW „d2W
(3-5.48)
dt dx2 '

6. The d function

1. Definition of the d function. In addition to the continuously distributed

quantities (mass, charge, heat sources, impulse, etc.), point-form quantities
(point mass, point charge, point-form heat source, point-form impulse, etc.)
also often occur. We emphasize that this idea represents a “limiting case,”
although it is used by physicists principally as an independent concept—by
omission of the corresponding passage to the limit.
With regard to the physical significance of the d function, we shall first
consider the following relations. Let a unit of mass exist inside a fixed region
of space T in the neighborhood of a point M0. At some other point M of
the space, the mass then produces a fixed potential (see Chapter 4, Section 5).
We now select a sequence of functions {pn} (pn > 0), each of which is equal
to zero outside of a sphere S of radius e„ about the center point M0, whereby
e„ for n—>oo approaches zero. From a fixed n on, we always have

рп(Р)йтр = \\\ pn(P)dTP = 1 (3-5.49)

Then we consider the sequence of functions

This obviously represents the potential of the masses which are distributed
with the density pn. For n—>°o we obtain
234 PARABOLIC DIFFERENTIAL EQUATIONS

lim m„ = —-— . (3-5.50)

П-.00 ТмйМ

These results obviously do not depend on the choice of the sequence {,»„},
and although the sequence {un} now approaches 1/r, the sequence {p*} pos¬
sesses no limit value in the class of the considered piecewise continuous func¬
tions. The “limit form’’ corresponding to the sequence {/>„} is called the
Dirac <5 function and is designated by 5(M, M0).
The basic property which defines the 8 function is the formal relation

SSS/<M-M,/WWr"={'o(W £ m. IT <M-51)
where f(M) is an arbitrary continuous function of the point M. From the
fact that the functions pn, for n—> oo in each region which does not contain
the point M0, converge uniformly to zero and become unbounded in the neigh¬
borhoods of 5 of the point M0, one often defines the 8 function formally by the
relations

8(M, M0) = 0 for M Ф M0

(3-5.52)
8(M, M0) = oo for M = M0

and

for M0 e T
8(M, M0)dr (3-5.53)
-f. for Mo £ T .

Eq. (3-5.53) is then a trivial consequence of (3-5.51) for / = 1.

In order to consider a sequence of functions in different problems we
must introduce some new concepts of convergence:
We say that a sequence of functions

{un(x)} = Ui(x) , щ(х), •••, Unix), ■ • • (3-5.54)

converges in an interval (a, b) uniformly if an N exists for every e > 0 such

that the condition

I Unix) — ип(х) I < e for n,m > N

is satisfied for n,m> N and an arbitrary x in (a, b).
A sequence (3-5.54) is said to be convergent in the mean in an interval
(a, b) if an N exists for every e > 0, such that the relation
(b
I Unix) — Unix) 12dx < e

holds for all n, m > M.

A sequence (3-5.54) is said to be weakly convergent in an interval (a, b)
if the limit value

lim f fix)un(x)dx
n-ooja

exists for each continuous function /.

3-5. APPLICATIONS TO CHAPTER 3 235

In the treatment of convergent sequences we usually introduce the notion

of limit elements of a sequence. Let us consider the class of continuous
functions in the interval (a, b). In the case of uniform convergence the limit
value already belongs to the same class of functions, but this is not neces¬
sarily true for mean and weak convergence.
If the limit element does not belong to the considered class of functions,
then this class can be extended by addition of the limit elements. In this
case, by an extension will be understood all the original functions (perhaps
by equivalent classes) and the limit element. Such an extension is already
known from the theory of real numbers. There the irrational numbers are
introduced as the limit elements which are defined by classes of equivalent
sequences of rational numbers.
With respect to the limit element for weak convergence we shall say
that two sequences {«„} and {vn} possess one and the same limit element
if they are equivalent, i.e., if the sequence {un — vn) converges weakly to¬
wards zero:

lim f f(x)[un(x:) — vn(x)]dx = 0 .

Ja
We shall now call a sequence {<?„} of nonnegative functions a locally normal¬
ized sequence of a point x0 if Sn is equal to zero outside the interval (x0 —
en, x0 + e„), where en for w—>oo approaches zero, while

Obviously then, {<5„} is a weakly convergent sequence. The limit element

of the sequence {£„} will be called the 8 function of the point x0.
If in the case of weak convergence the limit element и is formed by a
sequence {un} obtained from the class of functions un, the integral of the
product of a function f(x) with the element и is defined by

[ f(x)udx = lim \ f(x)un(x)dx .

Jo n-~Jo

Obviously, the equation

( f(x)8(x0, x)dx = f(xo)

holds for the 8 function of the point x0.

This relation is often used also in the definition of the 8 function.
2. Development of the 8 function in a Fourier series. The 8 function can also
be defined as the limit element of another series when this, in the sense of
a weak convergence of the above given sequence {<5„(x)}, is equivalent to the
locally normalized functions of the point x0.
We consider the sequence of functions defined in the interval (—1,1) by
mn , . mn гп
. mn \
<5„(x0 ’ *) = T7
ZL
+T І (cos
l m=l\ L
cos ——x + sin —— x0 sin
H
1 1 " mn. ,
= TT7 T
Z/
~~T
l
I COS —
m=l l
(X - X0) (3-5.55)
236 PARABOLIC DIFFERENTIAL EQUATIONS

or, in complex representation

ітЫ/l) (x—xq)
8n(x, x0) I e (3-5.55')
21 —n

Obviously then, for each function g(x) which can be developed in a

Fourier series, the relation

lim [ 8n(x0, x)g{x)dx = g(x0) (3-5.56)

*-“> J-i
holds, which shows that in this class of functions {g(x)} the above-introduced
sequence {Sn(xB, x)} in the sense of weak convergence is equivalent to the
sequence {8n(x0, x)}. Further,

S(xо, x) I cos -j-(x0 — x), (3-5.57)

21 + / m = 1 /

when this expression is understood in the sense of the above explained weak
convergence; and in the same sense the relation

<5(x0, x) = f <pn(x)cpn(xo) (3-5.58)

n—1

holds, where {(pn{x)} is a complete orthonormal system in the interval (a, by,
similarly

a(x0,x)=-^-[“ eiku°~x)dk = — (°°cos£(Xo — x)dk . (3-5.59)

2n J_„ 7Г Jo

We shall now show that for the calculation of the integrals which con¬
tain the d function, the series (3-5.57) can be used and can be integrated
term by term. To this end we consider a function g(x) which can be developed
in a Fourier series, and the integral

j g(x)d(x0, x)dx .

Here if we introduce, in place of o(x0, x), the expression from (3-5.57)

and integrate the series term by term, we obtain

/ \ Oo , /- я-m . = • 7Гm \
g{x) = -7г + I \Qm cos ——X + gm sin - x (3-5.59')
2 m=i\ I I J
with
If!
Qo 9(Xo)dx0

gm = 4* \ g(x0) cos ^^x0dx0 (3-5.60)

l J-i !
__ 1 • TZYYl
gm-—\ g(xo) sm —j-x0dx0.

By comparison of (3-5.59) with

3-5. APPLICATIONS TO CHAPTER 3 237

l
d(x, x0)g{x)dx = g{x0) , — / < Xo < I
-i

we recognize that the above-executed term-by-term integration of the series

for the <3 function leads to a correct result.
Therefore, in the class of functions developable in a Fourier series, the
sequence of partial sums
1 ^
_ k gi(xnll)(x—xf)
21 n=—k

is equivalent to the locally normalized sequence {(?„}.

Other representations of the <5 function are based on the use of a certain
sequence of functions which in the sense of weak convergence is equivalent
to the sequence {5„}.

3. Use of the 8 function in the construction of the Green's function. We con¬

sider the following problem:

ut = a2uxx (3-5.61)
u{x, 0) = <p(x) (3-5.62)
u(0, t) = u(l, t) = 0 . (3-5.63)
Here a prescribed function cp(x) corresponds to a uniquely determined solution

u{x, t) = L[(p{x)]
of the problem. We assume that the operator L can be represented in the
form

u{x, t) = L[y{x)] = j G(x, ?, t)<p($)d£ (3-5.64)

where G(x, f, t) is the kernel of the operator L.

For the determination of G(x,£,t) we set
<p(x) = <5(x — x0). (3-5.62')
If in formula (3-5.64) the initial function (p{x) is replaced by the о function,
the result is
u(x, t) = G(x, x0, t), (3-5.65)
i.e., G(x,x0, t) is the solution of (3-5.61) with the initial condition (3-5.62).
We now write the 8 function in the form of a Fourier series:
s, . ^ 2 . m . ш
8(x — x0) = 2. —sin —--*sin——дг0.
n=l It t

The kernel G must then obviously be of the form

G(x, Xo, t) = f An(t) sin 2ff— t (3-5.66)

n=1 /

where each summand must satisfy the heat-conduction equation. Hence there
238 PARABOLIC DIFFERENTIAL EQUATIONS

follows
An{t) = Bne.
From the initial condition there also follows
n 2 . m
B, = TS'n—ДГ..

Therefore we arrive formally at the following expression for the kernel G,

G(x, x0, t) = -7- 2 е~Ып/1> ° ‘ sin -^-x sin —~Xq . (3-5.67)

This agrees with the representation of the Green’s function given in Section
5-3. The solution of the problem (3-5.61)-(3-5.63) is given by the formula
(3-5.64), where G(x,x0, t) is the function defined by (3-5.67).
In a similar manner, we can arrive at an expression for the Green’s func¬
tion for an infinite straight line. The function G is defined here by the
conditions
Ut — a2uxx = 0, — со < x < 00 (3-5.68)
u(x, 0) = <p(x) = d(x — x0) . (3-5.69)
Since here the 8 function should be defined by a Fourier integral
1
8(x — x0) = —\ cos X(x — x0)dX ,
n Jo

we make for G(x, x0, t) the Ansatz

G(x, x0, t) = — { Ax(/) cos X(x — .t0№ . (3-5.70)

X Jo

From Eq. (3-5.68) we then find

Лх(« = Al0,g-aV‘. (3-5.71)

If we set t — 0, then we obtain
АГ = 1

by comparison of formulas (3-5.69) and (3-5.71).

Consequently,

G(x, Xg, t) = - ( e~a л ‘ cos Л(х — Xt,)dX .

X Jo

The calculation of these integrals as carried out in Chapter 3, Section 3, gave

Hence it follows that the solution of the problem of the propagation of an

initial temperature on an infinite straight line must be representable by the
formula
 3-5. APPLICATIONS TO CHAPTER 3 239

u(x, t) = у G(x, $, t)<p($)d£ (3-5.72)

Under which conditions the formulas derived by use of the <5 function are
valid, is a question that must be investigated separately.
As an example, we shall now treat the inhomogeneous equation

tit = a2uXI + . (3-5.73)

cP
In this case F(x, t) is the density distribution of the heat sources. If at the
point x — 6, at time t = t0, a momentary heat source of intensity G0 exists,
then

F(x, t) — Q0o(x — £)8(t — t0). (3-5.74)

Therefore, we must find the solution of the inhomogeneous equation

lit = агиХІ + -Q^-d(x — £)8(t — t0) , t0 > 0 (3-5.75)

cp

with the homogeneous initial condition

u{x, 0) = 0 .
Thus, by consideration of the integral representation

8{x — f) = — ( cos fШ
7Г Jo

for the function и(х, f), we make the Ansatz

u(x, t) = — \ ii\{t) cos 2(x — £)dk .

n Jo

If we put this expression into Eq. (3-5.75), we obtain for u\(t) the equation

th(t) + a2fux(t) = -Q-d(t - to)

cp

with the initial condition

uk{ 0) = 0 .
Now, as is known, the solution of the inhomogeneous equation

й + аги = f(t) , u(0) = 0

has the form

u(t) = ‘ g-a>2(t-T) f(T)dr . (3-5.76)

0

In our case, however,

for t < to
Uk(t) = в-,х*““т)3(т - /о№ = (°
Jo j _Оо_с-а2кги-10) (3-5.77)
for t > to •

V Cp
240 PARABOLIC DIFFERENTIAL EQUATIONS

Therefore,

”g-a2A2U-t0> cos ;(д. _ W = Oo_Q{X) ^ , _ f0) ,

u(x, t) = —
ср 7Г o CjO

where

_1_ -(*-«>*/4a2(t-l0>
G(x, £, t — t0)
2\/na2(t - to)

represents the Green’s function. Similar methods for the construction of the
Green’s function have often been used in theoretical physics.75

75 A detailed investigation of the theory of the 8 function and a number of examples

of application of the 8 function are found in D. D. Ivanenko and A. A. Sokolov, Classical
Field Theory, (trans. from Russian), Berlin, 1953, Ch. I.
 4

ELLIPTIC DIFFERENTIAL EQUATIONS

The investigation of different stationary physical processes (vibrations,

heat conduction, diffusion, etc.) usually leads to elliptic equations. Very often
Laplace’s differential equation occurs:

du = 0 .
A function и is said to be harmonic in a region T if in this region the
function and its derivatives up to the second order are continuous and satisfy
Laplace’s equation.
For the investigation of the properties of harmonic functions, different
mathematical methods have been constructed which can be successfully ap¬
plied to hyperbolic and parabolic differential equations.

4-1. PROBLEMS WHICH LEAD TO LAPLACE’S DIFFERENTIAL EQUA¬

TION

1. Formulation of the boundary-value problem for stationary

heat fields
It was shown in Chapter 3 that the temperature of a nonstationary heat
field satisfies the differential equation

1
ut = a du ,
, 2
a =— .
k
cp

If a stationary process occurs, then a stationary heat field also occurs, so

that the temperature distribution is constant in time; thus it is only a func¬
tion of position u{x, y, z). Consequently, in this case it satisfies Laplace’s
equation

du = 0 . (4-1.1)

If heat sources are present, however, the equation

du = f, /=£ (4-1.2)
k

results, where F is the density of the heat sources and k is the coefficient
of heat conductivity. The inhomogeneous Laplace Eq. (4-1.2) is usually called
the Poisson differential equation.
 242 ELLIPTIC DIFFERENTIAL EQUATIONS

We consider now a region T of space which is bounded by a surface I.

The problem of the stationary temperature distribution u(x, y, z) in the interior
of the body T reads as follows:
Determine a function u(x, y, z) which in the interior of T satisfies the
differential equation

Au = — f(x, y, z) (4-1.2)

and one of the following boundary conditions:

first boundary-value problem: и = /х on E

second boundary-value problem: 4r~ = /2 on E

on
• du
third boundary-value problem: —— + h(u —/3) = 0 on E .
on
Here /1, /2, /3, h are given functions and du/dn is the derivative in the direc¬
tion of the exterior normal to the surface T.76
The physical significance of these boundary conditions is clear (see Sec¬
tion 3-1). The first boundary-value problem for Laplace’s equation is usually
called the Dirichlet problem; the second is called the Neumann problem. If
we are dealing with the solution of a problem in a region T0 which lies out¬
side of the surface E, then one speaks of an exterior boundary-value problem.

2. Irrotational fluid motion (potential flow); the potential of a

stationary flow and of an electrostatic field

As а second example, we now treat the potential flow of a source-free

fluid. In the interior of a region of space T with the boundary I a stationary
flow of an incompressible fluid (with density p = const.) moves with the velo¬
city v(x, y, z). If the fluid motion is irrotational then the velocity v has a
potential, i.e., a scalar function ip exists such that

v = grad (p (4-1.3)

or in terms of components

dip dip dip

vx = vv = Vz =
dx ~dy ’ dz
In a curl-free fluid motion, therefore, the three velocity components can be
derived from a single function <p(x, y, z), which is called the velocity potential
of the motion in question. If the flow is source-free throughout, then

div v = 0 . (4-1.4)

76 Obviously a stationary temperature distribution can occur only when the total heat
flow through the boundary of the region is equal to zero. Consequently, the function
/2 must still satisfy the auxiliary requirement
4-1. PROBLEMS WHICH LEAD TO LAPLACE S DIFFERENTIAL EQUATION 243

If we introduce the expression (4-1.3) for v then we obtain

div grad <p = 0

or

J<p = 0 , (4-1.5)

i.e., the velocity potential satisfies Laplace’s differential equation.

In a homogeneous conducting medium, let a stationary electrical current
flow whose spatial density is j(x, y, z). If there are no spatial current sources
in the medium, then

div j = 0 . (4-1.6)

The electrical field strength E is determined by the current density, that is,
according to Ohm’s law

E = ^j, (4-1.7)

where Л is the conductivity of the medium. Since now the process is stationary,
the curl-free electrical field must be a potential field,77 i.e., a scalar function
<p(x, y, z) exists such that
E = — grad <p . (4-1.8)

From this we obtain by consideration of (4-1.6) and (4-1.7) the relation

A<p — 0 , (4-1.9)

i.e., the potential of the electric field of a stationary flow satisfies Laplace’s
differential equation.
We consider further an electrical field generated by stationary charges.
Because of the time independence,

rot E = 0 (4-1.10)

also follows here, i.e., the field is a potential field, hence

E- — grad<p . (4-1.8)

Now let p(x, y, z) be a spatial charge density in a medium whose dielectric

constant e is equal to one. Then, according to a known law of electrodynamics,

^ dS = 4л- J ві = 4л:^ j ^pdz , (4-1.11)

where T is the region of space considered, 5 is the boundary and is the

sum of all the charges in the interior of T. By Green’s theorem

div Edz (4-1.12)

we obtain

77 From the second Maxwell equation (plc)H = — rot E, it follows that rot E — 0.
244 ELLIPTIC DIFFERENTIAL EQUATIONS

div E = 4Ttp .

If we substitute the expression (4-1.8) for E, it follows that

J<p = — 47rp , (4-1.13)

i.e., the electrostatic potential <p satisfies the Poisson differential equation.
If, however, no spatial charges exist (p = 0), then <p satisfies the Laplace equa¬
tion
J<p =0.
The basic boundary-value problems for the processes considered correspond
to the three types given above. We shall not consider here other boundary-
value problems which also are important for the description of definite physi¬
cal processes. One of these problems is discussed in Section 4-7.

3. Orthogonal transformation of the Laplace differential expres¬

sion in curvilinear coordinates

In our further investigations, we shall consider a new orthogonal curvi¬

linear coordinate system defined by the unique reversible and continuous dif¬
ferentiable transformation

* = (pMi. <72, q3), У = Ч>г(Ц\, <h , <h), z = <p3(qit q2, q3) , (4-1.14)
rather than use the rectilinear coordinates x, y, z. The new orthogonal curvi¬
linear coordinate system is denoted by

<7i = fi(x, y, z), q2 = f2(x, y, z), q3 = f3{x, y, z) . (4-1.15)

If we set <7i = C\, q2 = C2, q3 = C3, where Ci,C2,C3 are constants, we

obtain the three families of coordinate surfaces

fi(x, y, z) = Ci , f3{x, y, z) = C2
and
A(x, y, z) = C3 . (4-1.16)

Now we consider an element of volume in

the new coordinates qi,q3,q3, which are
bounded by three pairs of these coordinate
surfaces (Figure 44). Hence qz = const., q3 =
const, along AB, qy = const., q2 = const, along
AD, and q{ = const., q3 = const, along AC. The
direction cosines of the tangents at the edges
AB; AD, and AC are then proportional to
d(px d<p2 d<p3
dq, ’ dq, ’ dqi
d<Pi dcp2 d<f3
л У
dq2 ’ °q3 dq2
d<Pi dcp2 d(p3
dq3 ’ dq3 ’ dq3
 4-1. PROBLEMS WHICH LEAD TO LAPLACE'S DIFFERENTIAL EQUATION 245

Therefore, the orthogonality conditions for the edges of the elements of the
volume can be written in the form

dip i dip, + dip2 dip2 + dip3 dtp3

i Ф k . (4-1.17)
dqi dqk dq, dqk dqi dqk

The line element of a curve in the new coordinates is given by

dip i dip! , 42
ds2 = dx2 + dy2 + dz2 = ( dqx + dq2 + -P-
dqi dq 2 dq3
/ 5^2 Э^2_
+ ( dqi + dq2 + dq3
\ dqi dq2 dq3

I 4- +
dip
dq3 (4-1.18)
V dq3

If we expand the parentheses and use the orthogonality conditions (4-1.17),

we obtain

ds2 = Hi dqi + Hi dqi + Hi dqi , (4-1.19)

where

(4-1.20)

Only one coordinate changes along each edge of the element of volume. The
lengths of these edges are therefore given according to (4-1.19)

dsi — Hidqi, ds2 = H2dq2, ds3 = H3dq3. (4-1.21)

so that a volume element equals

dv = dsids2ds3 — HXH2H3 dq{dq2dq3 . (4-1.22)

Given a vector field A(x, y, z), then by a known formula of vector analysis

\AndS
div A — lim —- , (4-1.23)
uM-o vM

where S is the boundary of a spatial region vM which contains the point M,

and An is the component of the vector A in the direction of the exterior
normal. When we apply this formula to the element of volume dv represented
in Figure 44, we can calculate the divergence of the field vector A.
Next, the difference of the flux through two opposite boundary surfaces,
for example, through the left and right, by use of the mean-value theorem,
can be written in the form

Qi — A\ds2ds3 Aids2ds31?1 .
246 ELLIPTIC DIFFERENTIAL EQUATIONS

Now from (4-1.21) we obtain

Qi — [H2H3A\ I^1+<ii71 H2H3Ai \gi\dq2dq3 — — - (H2H3Ai)dq\dq2dq3 .

In a similar way, we find for the other pairs of bounding surfaces the net
flux:

Qi — —— (H3H\A2) dQi dq2 dq3 (4-1.25)

oq 2

Q3 = (H1H2A3)dqldq2dq3 . (4-1.26)
dq3
Now if we substitute in formula (4-1.23) the value

AndS — Qi + Q 2 + Q3

and use (4-1.22), we obtain for the divergence, in the orthogonal curvilinear
coordinates, the expression

д {НгНМ+4-іНгНгА»
1
div A (H2H3Ai (4-1.27)
НуНгНг Г—
=
Uqi dq dq3 ]•
If we assume that A has a potential, i.e.,
A — grad и , (4-1.28)

then

du _ J_ ди — J_ ^u 1 du
(4-1.29)
dSi Hi dqi ~ H2 dq2 ’ H3 dq3

If these expressions for Alt A2, and A3 are substituted in (4-1.27), then
Laplace’s differential equation becomes

Ли = div grad и

= ттіЪтЬг
HiH2H3ldqi \ Н1 dqi ) + Т-{пг¥)
(Щг-Т-) dq2 \ Н2 dq2 J + аЧтг
dq3 \ H3 I1)!
dq3 )J • I4'1-30»

Laplace’s differential equation Ли — 0 therefore has the form

j _ 1 f д /Н2НЪ ди \ d /H3Hi du \_d (HjH2 du \| _ q

НіН2Нз\ dqi \ Hi dqi ) dq2 \ H2 dq2 ) dq3 \ H3 dq3 Jj
(4-1.31)

in the orthogonal curvilinear coordinates qit q2, q3.

We shall now consider two special cases.
1. Spherical coordinates. In this case q3 = r, q2 = Ѳ, and q3 = <p. The trans¬
formation formula (4-1.15) therefore assumes the form

X = r sin Ѳ COS <p , у — r sin Ѳ sin <p , z — r cos Ѳ .

The line element is defined by
4-1. PROBLEMS WHICH LEAD TO LAPLACE’S DIFFERENTIAL EQUATION 247

ds2 = (sin d cos (p dr + r cos Ѳ cos ipdd — r sin <9 sin <p d<p)2
+ (sin d sin (pdr + rcosdsinipdd + rsin<9cosy> d<p)2 + (cos ddr — rs'mddd)2
where, after simplification

ds2 = dr2 + r2de2 + r2 sin2 d dip2 ,

i.e.,

Hi = 1 , H2 = r , Нъ — r sin d .
By substituting these values for Hlt H2, H3 in (4-1.31) we arrive at Laplace’s
differential equation in spherical coordinates:

_1_ da
sini? sin d + = 0
r2 sin d dd
or finally

i-JL 1 du 1 d и
—( sin Ѳ + = 0 . (4-1.32)
r2 dr r sin Ѳ dd dd r sin2 d dip2

2. Cylindrical coordinates. Here <?! = p, q2 = <p, <7з = z;

x — p cos ip , у = p sin ip , z — z ;
thus
Hi = 1 , H2 = p , Яз = 1
Therefore Laplace’s differential equation in cylindrical coordinates has the
form

——( \_1 d2и d2и (4-1.33)

p dp \ dp ) p2 dip2 dz2

In particular, if the sought function и does not depend on 2 as in plane

potential flow, then Eq. (4-1.33) simplifies to

— — ( дм \ J_ d2u
(4-1.34)
p dp\P dp ) p2 dip2

4. Some particular solutions of Laplace’s differential equation

Of great interest are the spherically and cylindrically symmetric solutions

of Laplace’s differential equation, that is, those which depend only on the
variables r and p, respectively.
The spherically symmetric solutions и = U(r) of Laplace’s equation are
defined by the ordinary differential equation

A/О dU\ = 0,
dr\ dr J
whose general integral is given by

U=^- + C2,
r
248 ELLIPTIC DIFFERENTIAL EQUATIONS

where Ci and C2 are arbitrary constants. For example, if we set Ci = 1,

C2 = 0, then we find the function

U, = — , (4-1.35)
r

which is often denoted as the fundamental solution of the three-dimensional

Laplace differential equation.
In a similar way, by setting

и - U(p)
and by use of (4-1.33) or (4-1.34), we find the general form of the cylindrically
or circularly symmetric solutions (in the case of two-dimensional problems):

U(p) = Ci In p + C2 .
For Ci = —1 and Cz = 0 we obtain the function

U0 = In— , (4-1.36)
P
which is also often called the fundamental solution of the two-dimensional
Laplace differential equation.
The function U0 = 1 lr satisfies the equation Ли = 0 throughout with the
exception of the point r — 0, where it becomes infinite. To within a constant
factor it coincides with the potential due to a charge e placed at the origin
of the coordinates; the potential of this field is actually

e
и —— .
r
The function In 1 Ip satisfies Laplace’s differential equation throughout with
the exception of the point p = 0, at which it becomes logarithmically infinite.
To within a factor it agrees with the potential of an infinitely long wire (see
Section 4-5 § 2), whose potential is given by

и — 2ex In —
P
where ex is the charge density per unit of length. This function plays an
important role in the theory of harmonic functions.

5. Harmonic and analytic functions of a complex variable

A very general method for the solution of two-dimensional problems for

Laplace’s differential equation results if one employs the theory of functions
of a complex variable.
Let

w = f(z) = u(x, у) + iv(x, у)

be a function of the complex variable z = x + iy\ let и and v be real functions
of the variables x and y. Of particular interest are the so-called analytic
4-1. PROBLEMS WHICH LEAD TO LAPLACE'S DIFFERENTIAL EQUATION 249

functions for which the derivative

і^=1іт^=1іт^4±І£НѴЫ
dz 4z-o Az jz-o Az

exists. The increments Az — Ax + iAy can approach zero in an arbitrary man¬

ner. In general, each of the paths by which Az approaches zero corresponds
to a definite limit value. If w — f(z) is an analytic function, the limit
value lim Af/Az = f'(z) is independent of the choice of path. In this case, the
dz—*0
limit value is unique.
A function w = f(z) is analytic if the partial derivatives ux, uy,vx,vy
exist and are continuous, and provided, moreover, that the real part u(x, у)
and the imaginary part v(x, y) satisfy the Cauchy-Riemann differential equa¬
tions

ux = vy
(4-1.37)
liy — Vx •

These conditions are necessary and sufficient. We can derive them as fol¬
lows:
Let w = и + iv — f(z) be an analytic function. If we calculate the deriva¬
tives

dw(z) dw
wx = ux + ivx
dz dz
dw(z) _ . dw_
wy = uy + ivy
dz Zy dz
and require that the value of dw/dz coincide with both expressions, then it
follows that

dw
ux + ivx = vy — iuy
dz ’
from which the Cauchy-Riemann differential equations result directly. We
shall not prove here that this is also sufficient.
In the theory of functions it is shown that every analytic function defined
in a region G of the z plane has derivatives of all orders and can be developed
in a power series. In particular, for each such function, the real part u(x, y)
and the imaginary part v(x, y) are twice continuously differentiable with respect
to x and y.
By differentiation of the first expression of (4-1.37) with respect to x and
the second with respect to у we obtain

uxx + uyy = 0 or A2u = 0 .

Correspondingly, by reversing the scheme of differentiation we obtain

vxx + vyy — 0 or Агѵ = 0 .

The real and imaginary parts of an analytic function therefore satisfy
Laplace’s differential equation. Two functions и and v which satisfy the
250 ELLIPTIC DIFFERENTIAL EQUATIONS

Cauchy-Riemann differential equations are said to be harmonic conjugate func¬

tions of each other.
We consider now the transformation

x x(u, v) и u(x, у) 1 (4 1 38)

у = y(u, v) v=v(x,y).f

This maps a region G of the x, у plane onto a region G' of the u, v plane
such that each point of G uniquely corresponds to a point of G', and con¬
versely.
Now let

U = U(x, y)
be a real, twice continuously differentiable function which is defined in the
interior of the region G.
We shall investigate how the Laplace operator behaves under the trans¬
formation U = U[x(u, v), y{u, v)] = U(u, v). To this end we first calculate the
derivatives

Ux = Uuux + Uvvx, Uy = UuUy + Uvvy

UXX - UuuUx T UyyVX T 2/UuvMXVx T Uy,UXX “1“ UyVxX
Uyy - GJixuUy I GJyy'Vу I 2 GJuvUyVy I GJy.Uyy I GJyVyy
from which we obtain

GJxx + GJyy = GIuu{Ux + uy) + Uvv{vl + Vy)

T 2/Uuv(uxvx T UyVy) T Uu{UxX T Uyy) T Uy{vxx d- ^yy) • (4-1.39)

Now if и and v are harmonic functions conjugate to each other, then the
transformation (4-1.38) is equivalent to those transformations which are brought
about by the analytic function

w = f(z) = и + iv , z — x + iy . (4-1.40)

In this case, then, because of the Cauchy-Riemann differential equations

for и and v, the expression
2, 2 2. 2 2. 2 I W, i ,!
Ux + Uy = Ux d- vx = Vy + vx - I / (г) I
uxVx d- UyVy = 0

must be satisfied. Therefore from (4-1.39) we obtain

Uxx + Uyy = (Uuu+ U„) I f(z) I2 (4-1.41)

or

Ли,ѵи=—±—2Лх,уи. (4-1.41')
I / (2) I

By the transformation (4-1.40), the harmonic function U(x, y) in the region G

is transformed into the harmonic function U = U{u, v) in the region G'
provided I f'(z) |2 Ф 0.
4-1. PROBLEMS WHICH LEAD TO LAPLACE’S DIFFERENTIAL EQUATION 251

6. Transformation by reciprocal radii

In the investigation of the properties of harmonic functions one often

uses the transformation of reciprocal radii. By a transformation of reciprocal
radii of a sphere of radius a we mean a transformation with the following
properties:
1. To each point M there corresponds a point M' which lies on the same
straight line drawn from the origin of the coordinates through the point M.
2. The radius vector r' of the point M' is related to the radius vector
r of the point M by the expression

/ 2
rr— a or (4-1.42)

In the following, we shall assume a = 1, which is always possible by a suit¬

able choice of the unit of length.
We shall now show that a harmonic function u(p, <p) of two independent
variables is transformed by a transformation of reciprocal radii into a harmonic
function

v(p', <p) = u{p, (p) with P =— • (4-1.43)

P
The function u(p, <p) and, therefore, the function v(l/p, <p), considered as
functions of the variables p and <p, satisfy the differential equation

du d2u
= j+^ =°

and

2 . d dv d2v
p Др,<рѴ = p P— ) + = 0.
dp \ dp / d<p
From this, by passage to the variables p and <p, it follows that

dv _ dv dp' _ / dv
P 77 ,
^ dp P dp' dp ‘ dp
i.e., v(p', <p) satisfies the equation t<pv = 0, since
i2 л r d / I dv > d2 v
P A?'.<pV = P 77 P 77 + = 0
dp \ dp dtp2

We proceed now to functions of three variables and show that

v(r', Ѳ, <p) = m(r, Ѳ, <p)(^r = (4-1.44)

satisfies Laplace’s differential equation dr',e,vv = 0. If u(r, Ѳ, tp) is a harmonic

function of the variables г, Ѳ, tp, then dr,e,<pii = 0 is valid.
The transformation (4-1.44) is called the Kelvin transformation.
Next, we can show by differentiation that the first summand of Laplace’s
differential expression assumes the form
252 ELLIPTIC DIFFERENTIAL EQUATIONS

J_ A дги 2 ди _ 1д2(ги)
Г
(4-1.45)
Эт г dr г Эг2
T“2 ^ -л 2
r dr -2

Consequently,

or

Now, however,

дѵ___дѵ_ dr' _ /2 dt;

3r dr' dr dr'
so that v satisfies Laplace’s differential equation АГ’ ,в,ѵѵ = 0 since

or
r'4Ar',e,<pV = 0 .

4-2. GENERAL PROPERTIES OF HARMONIC FUNCTIONS

In the present discussion we give a description of the integral of harmonic

functions which will prove to be an important aid in the investigation of its
general properties. One of the important consequences of this integral repre¬
sentation is the principle of the maximum which we shall use many times—
for example, for the proof of uniqueness, as well as for the solution of boundary-
value problems. Further in the subsequent paragraphs the interior and ex¬
terior boundary-value problems for Laplace’s differential equation will be
formulated. Finally we shall prove the uniqueness theorem and the continu¬
ous dependence of the solutions on the auxiliary conditions (the stability).

1. Green’s integral formulas; integral representation of the

solution

In the treatment of elliptic differential equations we shall often use Green’s

formula, which can be shown to be a direct consequence of the formula of
Ostrogradski.
In the simplest case this formula has the form

(4-2.1)

where T is a definite region of space bounded by a sufficiently smooth surface

I; furthermore, R(x, y, z) is an arbitrary function which is continuous in
T + S and which in T possesses continuous partial derivatives; у is the angle
4-2. GENERAL PROPERTIES OF HARMONIC FUNCTIONS 253

between the z axis and the exterior normal to the surface I. The correct¬
ness of this formula is obtained by integration with respect to z. Ordinarily
we write the Ostrogradski formula1 in the form

ii \(^ + l^ + ir)dT=\ ^PcoSa + ^ cos ^ + Rcos^ da ’ (4-2.2)

/\ /\ /\
where dr = dxdydz is the element of volume, and a — (nx), /3 = (ny), у = (nz)
are the angles between the exterior normal n to the surface I and the
coordinate axes; P, Q, and R are arbitrary differentiable functions.78
If we denote P, Q, R as components of a vector A = Pi + Qj + Rk, then
the Ostrogradski formula can also be written as follows

^div^4dr=j \^Anda , (4-2.2')

where

dP dQ 3R
div A =
dx dy dz
and

An — P cos a + Q cos /3 + R cos у

denotes the component of the vector A in the direction of the exterior normal.
We proceed now to the derivation of Green’s formulas.
Let и = u(x, y, z) and v = v(x, y, z) be two functions which including their
first derivatives are continuous in T + I and which in T possess continuous
second derivatives.
If we set

„ dv ^ dv D dv
P = u— , Q = u— , R — u— ,
ox dy dz

then by the use of the Ostrogradski formula we obtain the so-called first
Green’s formula

ГГ Г/ди dv du dv du_ dv \
SS,\иШт=\Утпл° J J tJ dx dy dy^ dz dz )
(4-2.3)

where A = dz\dxz + d2/dy2 + d2/dz2 is the Laplace operator, and

d d 0d , d
— = cos a— + cos /9— + cos y —
dn dx dy dz

is the derivative in the direction of the exterior normal.

By use of the relation

78 In the following, it is always assumed that the Ostrogradski formula is applicable

in the region considered.
t Editor’s note: In the English literature the formulas of Ostrogradski and Green
are identical.
254 ELLIPTIC DIFFERENTIAL EQUATIONS

, , du dv du dv du dv
grad и grad v = Fu ■ rv = — — + — — + -—— ,
dx ox ay ay dz dz

Green’s formula can also be represented in the form

dv ,
JLH*=-fLSr“-r”*+Lf и —— da
dn
(4-2.3')

By interchanging the roles of и and v there follows

^vAudz = — j j ^pv-pudz + I ^v-^-da . (4-2.4)

If now we subtract the identity (4-2.4) from (4-2.3'), we obtain the second
Green’s formula

^(uAv — vAu)dz — j —V’lht'S)d(7 ' (4-2.5)

The region T can be bounded by several surfaces. Green’s formulas also

remain applicable in this case in which the surface integral is extended over
all surfaces which bound the region T.
As we have already seen (Section 4-1 §4), the function U0(M) = 1 lr, where
r — л/{x — x0)2 + (y — y0)2 + (z — z0)2 is the distance between the points M(x, y, z)
and M0(x0, Vo, z0), satisfies Laplace’s differential equation.
Now let u(M) be a harmonic function which including its first derivatives
is continuous in Г + І and which in T possesses a second derivative. Fur¬
ther, let us consider the function v =
1/Ѵ«лг0, where M0 is any interior point
of T. Since this function in T possesses a
point of discontinuity at M0(x0, y0, z0) the
second Green’s formula in T cannot be
applied directly to the functions и and v.
However, v = 1 /гмм0 is bounded in
T — K* with the boundary I + , where
Kz is a sphere of radius e, with center
at M0 and surface Z<. (Figure 45).
Now if we apply the second Green’s
formula (4-2.5) to the functions и and v
in T — Кг then we obtain

S$L.(“J7-T *)*
(4-2.6)
du
da .
dn
On the right side of this relation the last two integrals depend only on s.
For the derivative in the direction of the exterior normal to Ze we find

d_ d_ 1
dn dr r=e ~ £2
4-2. GENERAL PROPERTIES OF HARMONIC FUNCTIONS 255

•Therefore,

I ^ и^do = *4- j \udo = \ Ane'u* = Anu* , (4-2.7)

where u* is a suitable average value of u(M) on Іг. Further, the third in¬
tegral can be transformed as follows

l
(4-2.8)
r on e JjJ on e
where (du/dn)* is a suitable average value of the normal derivative du/dn on
If If we substitute the expression (4-2.7) and (4-2.8) into (4-2.6) and note
that d(llr) = 0 in T — Kt, then

1 du du
—^-\du dr = do + Anи — Ane
T— К p \ r) dn r dn dn
(4-2.9)
Now if we let the radius e approach zero, we obtain:
1. lim u* = u(M0), since u{M) is a continuous function and и* is an average
8—"0

value on the sphere of radius e with center point M0;

2. lim 4ne{du/dn)* = 0, since from the continuity of the first derivatives
e-0

of u(M) in T, the boundedness of the normal derivative

du du , du „ , du
— = -^— cos a -f —— cosj9 + —- cos?-
dn dx dy dz

in the neighborhood of M0 follows;

3. according to the definition of an improper integral,

'Л? i i ,L,(-T J“) * = i i j ( - T J“) * •

The passage to the limit e—>0 therefore yields the fundamental formula of
the theory of harmonic functions:

1
Anu(MQ) = — (4-2.10)
Г MqP

If we apply formula (4-2.10) to a harmonic function u(M), (du = 0), then

we obtain

u(M) = (4-2.11)

Therefore, the value of a harmonic function at an arbitrary interior point

M can be expressed in terms of its values and its normal derivative on the
boundary of the region. Here we assume the continuity of the function и
and its first derivatives in the closed region. We shall prove directly that
each of the integrals

and t tr~(—\(P)doP,
JjJ dnp \ r up )
(4-2.12)
256 ELLIPTIC DIFFERENTIAL EQUATIONS

where fi and v are continuous functions, represents a harmonic function. That

is, since the integrands including their derivatives are continuous throughout
with the exception of the boundary S, the derivatives of arbitrary order of
(4-2.12) can be formed by differentiation under the integral signs. Moreover,
since the functions

1
and -L (J-') |(|)cos«, +^(!)cosA- +!(t)cos»
r MP dnP \Гмр )
satisfy Laplace’s differential equation with respect to the variable points
M(x, y, z), the functions (4-2.12) by the generalized superposition principle
(see the lemma on page 63) similarly satisfy Laplace’s differential equation
with respect to x, y, z.
Hence we arrive at an important conclusion: Every harmonic function,
in the interior of the region in which it is harmonic, possesses derivatives
of all orders.79 Further we note that a harmonic function in a region T at
each point M0 in T is analytic (i.e., it can be developed in a power series).
This assertion also results from the integral representation (4-2.11).
Corresponding formulas hold also for harmonic functions of two independent
variables. Let S be any region in the x, у plane and C its boundary curve.
Further let n be the direction of the exterior normal (with respect to 5) to
this curve. Then if v = ln(l/гм0р) is introduced where Гм0р = л/{х— х0)г + {у— y0)2
is the distance from a point M0 in the interior of S, then, by similar consid¬
erations as above, instead of (4-2.10) we arrive at the expression

2ku(M0) = - ( Гм — fin — ^ - In— —]dsP - ( (in — AudS , (4-2.12')

JcL on \ Гм0рI Гм0р on J JsJ r

where M0 is an arbitrary fixed point in the region S.

For a harmonic function u(M) it follows that

du
u(M0) = (4-2.12")
2k dn

2. Some fundamental properties of harmonic functions

Harmonic functions possess the following important properties:

1. If v(M) is a harmonic function in a region T which is bounded by
the surface I, then

(4-2.13)

where 5 is an arbitrary closed surface which lies entirely in the region T.

If in the first Green’s formula (4-2.3) an arbitrary harmonic function v (Jv=0)
is introduced and и = 1 then formula (4-2.13) follows. From (4-2.13) we con-

79 If for a function и which is harmonic in T the condition that it and its first de¬
rivatives are continuous on the boundary E is not fulfilled, the theorem still remains
valid. By this we mean that each point M is enclosed by a region, including its boundary,
lying in the interior of T.
4-2. GENERAL PROPERTIES OF HARMONIC FUNCTIONS 257

elude that the second boundary-value problem (Au = 0 in T, du\dn = f\x) pos¬
sesses a solution only if

Ц'л-°-
This property of a harmonic function can be interpreted as a condition that
no sources exist in the region T.
2. If u(M) is a harmonic function in a region T, and M0 is a point lying
in the interior of T, then

u(M0) = ——г- ( (и do , (4-2.14)

JxJ

where Ea is a spherical surface with radius a and center point M0 which lies
entirely in the region T. This property is stated by the average-value theorem,
which reads:
Theorem. The value of a harmonic function at any point M0 is equal to
the average value of the function on an arbitrary spherical surface Ea with
center at M0, provided Ea lies entirely in the region in which u(M) is harmonic.
For the proof of this proposition we apply formula (4-2.11) to a sphere
Ka with center M0 and surface Ia:

4ku(M0) =

If we bear in mind that

—=— on T0 and ( \^-do — 0

r a jxoJ dn

as well as

d_ d_ 1
2
dn dr a

(the direction of the exterior normal to la coincides with the direction of

increasing radius) then we obtain

1
u{M0) и do ,
4k a2

which was to be proved.80

For two independent variables the analogous theorem is valid:

80 For the proof of this theorem we have used Eq. (4-2.13). This, however, assumes
that the derivatives also exist on the spherical surface. If a function u(M), which is
continuous in the closed region T + I, satisfies the equation Au — 0 only for interior
points of T, then this conclusion would not be correct for a sphere Zai), which borders
the boundary T. However, if the theorem for each a < ao is true, then as a -* do we
obtain

u(Mo) = —2\ \u(M)do.

47T<X0 J Xao J
258 ELLIPTIC DIFFERENTIAL EQUATIONS

u(M0) — —— ( uds , (4-2.15)

ЗтГЙ Jca

where Ca is а circle of radius a about the center M0.

3. If a function u(M) is continuous and defined in a closed region T + Z
and satisfies Laplace’s differential equation Ди = 0 in the interior of T, then
it assumes its maximum and its minimum on the boundary Z (principle of
the maximum).
If the function u(M) were to assume its maximum at an interior point
M0 in T, then M0 = u(M0) S u(M) for each M in T.
Now enclose the point M0 with a sphere of radius p whose surface Zp
lies entirely within T. Since by hypothesis u(M0) is the maximum of the
function u(M) in T + Z, then и U ^ u(M0). Therefore, by use of the average-
value formula (4-2.14), provided that everywhere under the integral signs we
replace u(M) by u(M0), we obtain

u(M0) -- j \u{M) da и T I ^u(M0)da = u(M0) . (4-2.16)

Now if we assume that at least one point M exists on Zp such that the
inequality u(M) < u{M0) is valid, then obviously in the last formula the in¬
equality sign must hold, which in turn implies a contradiction; consequently
on the entire surface Zp we must have u(M) = u(M0).
If po1 is the minimal distance of the point M0 from the boundary Zp,
then u(M) = u(M0) is also valid for points belonging to TP“. Hence, because
of continuity, it also follows that at those points M* which belong to the
intersection of TP™ and Z the relation u(M*) = u(M0) is valid. Therefore our
theorem is proved; and the last conclusion shows that the maximum u(M0)
is assumed at least at one point on the boundary.
It is easily seen that u(M) = u(M0) must be valid in the entire region
when the region T is connected and if at least at one interior point M0 the
maximum is assumed.
To demonstrate the above, we select another arbitrary point Mi0) in T
and connect it with M0 by the polygonal line L (Figure 46), whose length is
designated by l. Let My be the last current
point of L through the spherical surface
TP™. At this point, then, и(Мг) = u(M0) is
still valid. Now we enclose this point by
the spherical surface Zp™, where pГ is the
minimal distance of the point Mi from the
boundary. We obtain another such point
M2 as the last current point of L through
the spherical surface Tpj". By this procedure,
we arrive after a finite number of steps
(the number p of necessary steps is certainly
not larger than 1 lpim' if p[m) denotes the
minimum distance between L and T) at a
4-2. GENERAL PROPERTIES OF HARMONIC FUNCTIONS 259

spherical surface which contains the point M(0). From this it follows that
u(M{0)) = u{M0). Because of the arbitrary choice of the point M(0) and the
continuity of u(M) in the closed region T + 2 we conclude that u(M) = u{M0)
holds everywhere (including the boundary). Of all harmonic functions, there¬
fore, only the constant functions can assume their maximum at an interior
point.
The corresponding statement is also true for the minimum.
Conclusion 1:
If и and U are continuous in T + 2 and are harmonic functions in T
for which

и tk U on 21

then also и 5S U holds at all points in the interior of T.

The function U — и is therefore continuous in T — 2, harmonic in T; hence

U — и 'Sz 0 on I .

Consequently, according to the principle of the maximum we must have

U-u^O
at all points in the interior of T— precisely our assertion.
Conclusion 2:
If и and U are continuous in the region T + 2 and are harmonic functions
in T for which

I и I 5= U on I

then also

I и I 5І U

at all points in the interior of T.

From the above assumptions it follows that the three harmonic functions
— U, u, and U satisfy the relation

— U -й и -й U onJ.

However, it follows by a twice-repeated application of Conclusion 1 that

-ийийU
at all points in the interior of T, or

\ u\ й U
in the interior of T.

3. Uniqueness and stability of the solution of the first boundary-

value problem

Let us consider а region T which is bounded by the surface 2. Then

the first boundary-value problem for Laplace’s differential equation in the
region T reads as follows;
260 ELLIPTIC DIFFERENTIAL EQUATIONS

Determine a function и which

(1) satisfies the equation z/m = 0 in the interior of the region T;
(2) is defined and continuous in the closed region T + I, i.e., both in the
interior as well as on the boundary.
(3) on the boundary T assumes a prescribed value.
Condition (1) requires that the sought function is harmonic in the interior
of T. To require that the function also be harmonic on the boundary would
result in additional limitations for the boundary values. This requirement
is superfluous. On the other hand, continuity in the closed region (or any
other corresponding condition) for the uniqueness of the solution is necessary.
If we waive this requirement, then each such function, which in T is equal
to a constant C and coincides on I with the prescribed function /, could be
considered as a solution of the problem, since the conditions (1) and (2) would
also be fulfilled.
We now prove the uniqueness theorem:

Theorem. The first interior boundary-value problem for Laplace’s differential

equation is uniquely solvable.

We assume that two different solutions ux and m2 exist which are continu¬
ous in the closed region T + I, satisfy Laplace’s differential equation in the
interior of T, and on the boundary I assume one and the same value /.
Then the difference и = Ui — m2 has the following properties:
1. Au = 0 in the interior of T.
2. и is continuous in the closed region T + S.
3. и U = 0.
Consequently, u(M) represents a continuous and harmonic function in
the interior of T which on the boundary is equal to zero. As is known, a
continuous function assumes its maximum in a closed region. Therefore we
must have и = 0. That is, if и Ф 0 and for at least one point и > 0, then и
must assume its maximum at an interior point of the region, which is impos¬
sible. In exactly the same way we show that и is never negative in T.
Consequently,

и =0 .

We turn now to the proof of the continuous dependence of the solution

of the first boundary-value problem on the boundary conditions. As stated
previously, a problem is said to be determined physically if a small change
in the conditions which determine the solution of the problem—in our case
the boundary conditions—implies a small change in the solution itself, that
is, “the solution is stable.”
Let Mi and m2 be two functions which are continuous in T + I and harmonic
in the interior of T, for which the magnitude of the difference of the boundary
values does not exceed s > 0. Then throughout the region under considera¬
tion,
4-2. GENERAL PROPERTIES OF HARMONIC FUNCTIONS 261

This assertion follows directly from Conclusion 2 in the previous section,

since U = e is a harmonic function.
Therefore we have proved the continuous dependence of the solution on
the boundary conditions and the unique solvability of the first interior boundary-
value problem.

4. Problems with discontinuous boundary conditions

Often one has to deal with a first boundary-value problem with discon¬
tinuous boundary conditions. Naturally, a continuous function in the closed
region cannot be a solution to this problem, and we must therefore modify
the formulation of the first boundary-value problem accordingly.
Let a piecewise continuous function f(P) be defined on a curve C which
bounds a bounded region S. We seek a function u(M) with the following
properties:
1. u(M) is harmonic in the interior of S.
2. u(M) continuously approaches the boundary values at all points where
these are continuous.
3. u(M) is bounded in the closed region 5 + C.
We note that the additional requirement of boundedness is essential only
in the neighborhood of the points of discontinuity of f(P). Therefore, the
following theorem holds:
Theorem. The solution of the first boundary-value problem with piecewise
continuous boundary conditions is uniquely determined.
Let Ui and u2 be two solutions of the stated problem. Then we have for
the difference v — — u2:
1. v is a harmonic function in the interior of 5;
2. v continuously approaches the value zero on the boundary with the excep¬
tion of the points of discontinuity of f{P) at which v can be discontinuous;
3. v is bounded in S + C; | v \ < A.
We now construct the harmonic function

U(M) = e І In— .
i= l Гі

Here e denotes any positive number, D is the set theoretic diameter of the
region 5, and r, is the distance of the zth point of discontinuity Pi under
consideration. The function U(M) is positive since all the summands are
positive.
Further we enclose each point of discontinuity Pi with a circle Ki of
radius 8 and let 8 be sufficiently small so that each summand

i D
£ ІП-
Гі

of the corresponding circle Ci is larger than A, i.e., let £ In D/8 > A. Then,
in the closed region S — U v is continuous.
t=l
According to the principle of the maximum, U in this region is a majorant
of v:
 262 ELLIPTIC DIFFERENTIAL EQUATIONS

I v(M) I ^ U(M) .
For a fixed point M in S, as e —> 0 we obtain
lim U(M) = 0 .
8—0

Therefore,

v(M) = 0 ,
since v is independent of £; or
ti\ — U-2 у

which was to be proved.

5. Isolated singularities

We shall now investigate the singularities of a harmonic function. If P

is an isolated singular point in a region in which the function и is harmonic
then two cases are possible:
1. The harmonic function is bounded in the neighborhood of the point P.
2. The harmonic function is unbounded in the neighborhood of the point P.
We have already dealt with singularities of the second type (for example
In Hr). The following theorem shows that singularities of the first type can¬
not occur:
Theorem. If a bounded function u(M) is harmonic in the interior of a region
5 with the exception of a point P, then u(M) at the point P can be defined
such that the function u{M) is harmonic throughout 5.
For the proof of this theorem we choose a circle Ka of radius a about
the center point P, which lies entirely in S, and consider a harmonic func¬
tion v in the circle which coincides with the function и on the circumference
Ca of the circle Ka.sl
The difference

w=и—v
then has the following properties:
1. With the exception of the point P, at which w is not defined at all, it is
harmonic everywhere in Ka.
2. It continuously approaches the boundary value zero on Ca.
3. It is bounded in the closed region Ka + Ca, (I v \ < A).
As in the proof of the previous theorems (Section 4-2 §4), we again con¬
struct the non-negative harmonic function

U(M) = e In — .
r
Here £ denotes a positive number, a is the radius of the circle Ka, and r is
81 The existence of such a function will be shown in Section 4-3. Its construction
therefore does not depend on the theorem proved here.
4-2. GENERAL PROPERTIES OF HARMONIC FUNCTIONS 263

the distance of the point M from the singularity P. Further we choose a

circle Ks with center point P, whose radius d is such that on its circumference
the value of U is larger than A. Then we consider the region Ka — Ks.
The function w in the closed region 5 -й r ^ a is continuous and on its boundary
\w\tkU. On the basis of the principle of the maximum, therefore, the posi¬
tive function U is a majorant of w:

\ w \ S U(M) for дй r^a.

Now if we fix any point M on Ka, which does not coincide with P, and let
e approach zero, we obtain
lim U(M) = 0 ,
8— 0

thus
w =0 .
Therefore, the function u, with the exception of the point P, coincides with
the function v everywhere in the region S. Now if we set u(P) = v(P) then
we arrive at the function и = v, which is harmonic everywhere in S. There¬
fore our theorem is proved.
Similarly, we proceed to the proof of the theorem for the three-dimen¬
sional case in which the function U(M) = 2(l/r — 1/a) can be used as the
majorant function.
For the proof of this theorem we should assume that the function и
remains bounded in the neighborhood of the point P. The proof also remains
valid if instead of assuming the boundedness we assume only that the func¬
tion u, in the neighborhood of the point P, satisfies the inequality

I u(M) I < e(r) log —— , (4-2.17)

Грм
where e(r) is an arbitrary function which as r —> 0 also approaches zero. This
requirement indicates that u(M) in the neighborhood of P increases slower
than log l/грм ■
Therefore if the function u(M) is harmonic with the exception of a point
P in a region S, where as M—> P it increases slower than log 1 ІГрм, then
in the neighborhood of P, u(M) is bounded, and the value u(P) can be defined
so that и is harmonic in the entire region 5.
This is also the case for three independent variables. If a harmonic
function u(M) in the neighborhood of an isolated singular point P increases
slower than 1 /r:

\u(M)\<e(r)—— , e(r)->0, r-> 0, (4-2.18)

Tup

then it is bounded in the neighborhood of this point, and the value u(P) can
be so defined that u(M) is also harmonic at the point P.

6. Regularity of a harmonic function at infinity

A harmonic function u(x, y, z) is said to be regular at infinity if for r ^ r0

264 ELLIPTIC DIFFERENTIAL EQUATIONS

the following conditions are fulfilled:

du A du A du A
и 1 < — , ^ 2 f < . < 2
(4-2.19)
r (be r dy dz Г

The following theorem is valid: If a function u, outside of a closed

surface I, is harmonic and as r—> со tends uniformly towards zero, then it
is regular at infinity.
The assertion that и tends uniformly towards zero as r—> oo requires
also that a function e*(r) exist such that

I u(M) I < e*(r) , e*{r) —* 0 , t —* oo , (4-2.20)

where r denotes the radius vector of M.

For the proof we use the Kelvin transformation

v{r', Ѳ, <p) = ru(r, Ѳ, <p) with r' = — .

r
The function v so obtained is then harmonic everywhere in the interior of
the surface I', which arises by the transformation of reciprocal radii from
I. An exception is the origin of coordinates at which v has an isolated
singularity.
From the condition (4-2.20) it follows that in the neighborhood of the
origin of coordinates the function v satisfies the inequality

\ѵ{г',Ѳ, ip) I й*[у^у = £{г')^

with

e(r') = 7^-^-0 for r' —* 0 .

Therefore, v(r', Ѳ, <p), on the basis of the last theorem in Section 4-2 §5,
is bounded and harmonic for r' = r0':

I v(r', Ѳ, <p) \ S A for r' S r'o, (4-2.21)

from which it follows that

■ , a , , 1 v(r', Ѳ, <p) I A r 1
I u(r, Ѳ, cp) I = -r - ^ — for r ^ r0 = —
r0
Since now for r' =0, v is harmonic, we can write:

ЩьЛ = АЛ1. у, у.
x . 1 Г дѵ дх' дѵ ду' , дѵ dz' "1 .. 0
= -— ‘V + — —— + —7’*+-*-, (4-2. 22)
r r \_dx дх ду ox dz dx J
where

/ X / / V/ / Z f
x — —r , у ——r , z — —r
r r r
4-2. GENERAL PROPERTIES OF HARMONIC FUNCTIONS 265

If we now calculate the derivatives дх'/дх, ду'/дх, dz'/дх and consider the
boundedness of the first derivatives of v in the neighborhood of the point
r' - 0, we find
du
for r OO .
dx
Corresponding estimates are obtained for du/ду and du/dz.

7. Exterior boundary-value problems; the uniqueness of the

solution for two- and three- dimensional problems

In the formulation of the exterior boundary-value problems we have to

distinguish between two and three independent variables.
We consider first the three-dimensional case. Let T be an unbounded
region which can be bounded by a closed bounded surface I. The exterior
first boundary-value problem then reads as follows:
Determine a function u(x, y, z) which satisfies the following conditions:
1. Au — 0 in the unbounded region T.
2. и is everywhere continuous, including on the surface 1.
3. u\i = /(x, y, z), where / is a prescribed function on I.
4. u(M) tends uniformly to zero as M-* оо.
A simple example will show that the last condition is essential for the
uniqueness of the solution. To this end we consider the following problem:
Determine a solution of the exterior boundary-value problem for a sphere
5д of radius R with a constant boundary condition

и |sB = const. = /о .
If Condition 4 is dropped, then both the functions ut — f0 and u2 — foR/r as
well as the functions

и = aih + ft u2 with a + P = 1
are admissible solutions of the problem. On the other hand we have the
following theorem:
Theorem. The exterior first boundary-value problem for the three-dimensional
Laplace differential equation has a uniquely determined solution.
If Ui and m2 are two solutions which satisfy Conditions 1 to 4, then their
difference и — Ui — u2 represents a solution which satisfies the corresponding
homogeneous boundary condition. Since Condition 4 for the function и is
also fulfilled then for every e > 0, an R* can be found such that

I u(M) I < e for r^R* .

If the point M lies in a region T' (Figure 47), which lies between the surface
I and a sphere Sz (r 2; R*), then obviously u(M) < e. This follows from the
principle of the maximum applied to the region T'. Since e can be chosen
arbitrarily we conclude that u, in T' as well as in the entire region T, vanishes
identically. Therefore we have demonstrated the unique solvability of the
266 ELLIPTIC DIFFERENTIAL EQUATIONS

exterior first boundary-value problem in the three-dimensional case.

The exterior first boundary-value problem in the two-dimensional case
reads as follows:
Determine a function u(x, y) which satisfies the following conditions:
1. Ди = 0 in the unbounded region I under consideration whose boundary
forms a closed curve C.
2. и is continuous everywhere and also on C.
3. и \c — f{x, y), where / is a prescribed function on C.
4. u{M) is bounded at infinity, i.e., a number N exists such that \ u{M) \ ^ N.
Here the requirement that the solution is equal to
zero at infinity would also be sufficient to prove that
there could not be two distinct solutions. However, this
would be too strong a requirement and in general the
problem would not be solvable.
The following theorem is valid:
Theorem. The exterior first boundary-value problem for
fig. 47. the two-dimensional Laplace differential equation has a
uniquely determined solution.
From the existence of two distinct solutions Ui and иг it follows that
their difference и = Ui — u2 represents a solution of the first boundary-value
problem with homogeneous boundary conditions. Further, because of Con¬
dition 4 we have
I и I ^ N = N + N2,
where Ni and N2 are such that | Mi | й Ni, | иг | ^ Nt • We denote by 11 the
complement of T, so that I + represents the entire plane. Further we
select a point M0 on , and also a circle of radius R with center point
M0, which lies entirely in (Figure 48). Then
the harmonic function In 1/гМм0 has no singularity
in the region I\ the function In rMMJR is positive
in all of I and on C.
Further, let CRy be a circumference with radius
CRi
and center point M0 which entirely contains the
curve C, and let I' be the region bounded by C
and CRl. The function uRl defined by

— лг In r/R (4-2.23)
Rl N In RJR

is then a harmonic function which is equal to N on the circumference CRl

and is positive on C. From the principle of the maximum it follows that
uRl is a majorant of the absolute value of u(M) in the region I:
I u(M) I < uRl(M) .

We fix M and let R{ increase unboundedly. Obviously, then, as Rx—>oo, uRl{M)

approaches zero, so that u(M) = 0. However, since M can be chosen arbi¬
trarily, the unique solvability of the problem considered is demonstrated.
4-2. GENERAL PROPERTIES OF HARMONIC FUNCTIONS 267

We can also prove the unique solvability of this problem by using the
transformation of reciprocal radii. With this transformation the curve C in
this case is transformed into the curve C', so that the region lying in the
exterior of C is transformed into the region lying in the interior of C'. Thus,
the point at infinity transforms into an isolated singular point in whose neigh¬
borhood the function v remains bounded. From the theorem of the harmonic
functions in Section 4-2 § 5, it then follows that v is harmonic at the origin
of coordinates and the solution is unique.
From these considerations it follows that a harmonic function which is
bounded at infinity tends towards a definite limit value as M—> oo.
The difference in the formulation of the exterior first boundary-value
problem for two and three variables can be clarified by the following physical
example. Given a sphere of radius R whose surface is held at a constant
temperature u0, find the stationary temperature distribution of the surrounding
space. The function и = u0R/r represents the solution of this problem which
vanishes at infinity.
We consider now the corresponding two-dimensional problem. Thus, let
a circle of radius R be given, on whose circumference a constant temperature

u\s = fo = const.
prevails. In this case и = f0 is the uniquely determined bounded solution of
the problem and no further solution exists which would be equal to zero at
infinity. Earlier we emphasized the essential difference in the relation of
harmonic functions at infinity for two and three independent variables (for
example, the relation of 1/r and In 1/r at infinity).
Also, for spatial and plane unbounded regions the principle of the maxi¬
mum remains valid. This we know from the considerations which parallel
the arguments in the proof of the uniqueness theorem. From this follows
again the continuous dependence of the solution on the boundary conditions.

8. The second boundary-value problem, regularity at infinity,

and the uniqueness theorem

A function и is said to be the solution of the second boundary-value

problem if it and du/dn are continuous in a region T + I, and if и satisfies
Laplace’s differential equation in T and on the boundary I satisfies the pres¬
cribed condition

The solution of the second boundary-value problem is uniquely determined

to within an arbitrary constant.
We proceed to the proof with the additional assumption that the function
u(M) has continuous first derivatives in the region T + T.82
82 In order to simplify the proof, we assume the continuity of the first derivatives
in T + I. Under more general assumptions, the uniqueness theorem was proved by
M. V. Keldysh and M. A. Lavrentev (Doklady A. N. SSSR 16, 1937). See also V. I.
Smirnov, op. cit. footnote 15, Part IV.
268 ELLIPTIC DIFFERENTIAL EQUATIONS

Let us assume that wx and иг are two distinct continuously differentiable

functions in T + T, each of which satisfies the equation Au = 0 in T and the
condition du/dn |x = f(x, y, z) on The difference и — и^ — uz then satisfies

du
= 0 .
дп г
Now if we set v — и into the first Green’s formula (4-2.3) and take into con¬
sideration that An — 0 and du/dn |x = 0, we obtain

dr — 0 .

However, because of the continuity of the function и and its first derivatives,
it follows that
du du _ du
dx dy dz
i.e.,
и = const. ,
which was to be proved.
The method of proof used here can also be applied to the case of an un¬
bounded region when the function under consideration is regular at infinity.
First of all, we prove the following theorem:
Theorem. In the case of an unbounded region, Green’s formula (4-2.3) is ap¬
plicable to functions which are regular at infinity.
Therefore, we consider an unbounded region T whose boundary we denote
by 1. Further we choose a sphere of radius
R and denote by TR that part of our region
which lies entirely in this sphere. The
region TR is then bounded by that part of
the spherical surface IR which belongs to
T and by the surface which forms a
portion of the surface I (Figure 49). Now
if we apply Green’s formula in the region
TR to the two functions и and v which are
regular at infinity, we obtain

ГI пн
nr du dv du
uAvdr
rjl_ dx dx + dy dy
+ dz dz J dz

(4-2.23')
dv ,
+ Lj'“irA' + u——do .
dn
By using the regularity of the functions и and v we estimate the integral
over SR:

dv ,
и-do u(vx cos a Ar vy cos + vt cos y) do
dn
3A , ЗЛ2 12тгЛ2
< =S^r4 тгД2
i?da R3 R
4-2. GENERAL PROPERTIES OF HARMONIC FUNCTIONS 269

Hence, there follows

lim f f u-^—do = 0 .
dn

The integral over TK on the right side as R—>■ oo tends toward the integral
extended throughout the entire region T. This integral exists since the in¬
tegrand, because of the regularity of и and v at infinity, approaches zero as
l/R*. If, however, the limit value of the integral over Ti, which is equal
to the integral over I, exists, then the limit value of the right side of (4-2.23')
also exists and is given by

lim uAvdv — uAvdz .

R—oo

Therefore we arrive at the formula

ff ГГ du_dv_ du dv du dv ,
uAvdz = — и-da . (4-2.24)
JJt-JLsx dx dy dy dz dn
Hence, the applicability of the first and consequently also of the second Green’s
formula in the unbounded region, under the assumption that the functions
which occur are regular at infinity, is proved.
We are now in a position to prove the following uniqueness theorem:
Theorem. The second boundary-value problem for an unbounded region pos¬
sesses a uniquely determined solution which is regular at infinity.

If in formula (4-2.24) we set v = и = ux — m2 and bear in mind that Au = 0

and du/dn |x = 0, then we obtain

(u2x + iiy + uz)dr = 0 .

Hence, because of the continuity of the derivatives of u, it follows that

ux — 0 , uy — 0 , uz = 0 , and и = const .
Now, however, at infinity и — 0, so that

и = 0 ,
i.e.
Hi — ІІ2 у

which was to be proved.

There now remains the question of whether or not the uniqueness of the
solution of the first boundary-value problem can also be proved by this method.
Let Mi and m2 be distinct solutions of the first (interior) boundary-value
problem. We then apply formula (4-2.3) to the function м = Mi — u2 and v = и
in the region T. Let I be the bounded boundary surface of T. Then we
obtain

uAudz = — (ux + uy + ut)dz + \ \ u-^-da ,

270 ELLIPTIC DIFFERENTIAL EQUATIONS

(ul + Uy + Uz)dz = 0

results. Consequently we find

ux = uy — uz = 0 and и = const .
The function и on the surface 2 is equal to zero. We can therefore conclude
that

и = 0 and Mi = m2 .

This proof is incorrect, however, since in the method of proof the existence
of the derivatives of the sought function on the surface 2 was assumed. The
proof of uniqueness which is based on the principle of the maximum does
not contain this weakness.

4-3. SOLUTION OF THE BOUNDARY-VALUE PROBLEMS FOR SIMPLE

REGIONS BY SEPARATION OF VARIABLES

For simple regions, the solution of the corresponding boundary-value

problem can be treated by separation of variables. Accordingly, certain
auxiliary equations occur which make it necessary to use special classes of
functions. This discussion will be limited, however, to those problems which
can be solved by means of trigonometric functions alone. Later, in the in¬
vestigation of certain special functions, we shall consider the solution of other
problems.

1. The first boundary-value problem for the circle

We begin with the solution of the first boundary-value problem for the
circle.
Determine the function и which satisfies the equation

Ju = 0 (4-3.1)

inside a circle and the boundary condition

u —f (4-3.2)

on the periphery, where / is a prescribed function.

Let us first assume / to be continuous and differentiable; later we shall
drop the differentiability requirement and also the continuity of / (see Sec¬
tion 4-2 §4). We shall consider both the interior and the exterior boundary-
value problems.
We first introduce the polar coordinates p, <p whose origin lies at the
4-3. SOLUTION OF THE BOUNDARY-VALUE PROBLEMS 271

center point of the considered circle of the radius a. Eq. (4-3.1), as is known,
has the form

1 dzu
ли
4
=-
1 3
+ 2 о 2
= 0 (4-2.3)
P dP P °<P

in polar coordinates (see formula (4-1.34)). We now investigate the separation

of variables, i.e., we seek the solution in the form

u(p, <p) = R(p)0(<p) .

By substitution into Eq. (4-3.3), we obtain

d/dp-(p(dRldp)) _ Ф" _ ;
Rip Ф

where X = const. Hence, we obtain for Ф and R the two differential equations

ф" + ХФ = 0 (4-3.4)

and

(4-3-5»
From Eq. (4-3.4) there results

Ф(<р) = A cosi/ly> + В sin \/l(p .

Moreover, we note that the single-valued function u(p, cp) under a change of
the angle <p by an amount of 2ir is assumed to equal its original value

u{p, (p + 2л-) = u{p, <p) .

Consequently,

Ф{(р + 2л-) = Ф(<р) ,

i.e., Ф{(р) is a periodic function of the angle <p with the period 2л-. This,
however, is possible only when l/J = n is an integer and

Фп{<р) = An cos mp + Bn sin n<p .

We assume the function R{p) to be of the form

R(p) = / .
With this and (4-3.5),

nz = p or pt = ± n , n > 0
results after replacing R{p) by /Л Consequently,

R(p) = Cpn + Dp~n ,

where C and D are arbitrary constants.
For the solution of the interior problem we have R(p) = Cpn (p = + n),
i.e., D = 0. For, if D Ф 0, then u(p, <p) = R{p№{<p) for p = 0 would increase
unboundedly and would not represent a harmonic function in the circle. By
272 ELLIPTIC DIFFERENTIAL EQUATIONS

contrast, R(p) = Dp~n (p = — n) must be assumed for the solution of the exterior
problem, since this solution must be bounded at infinity.
Therefore we have found the following particular solutions of our prob¬
lem:88

un(p, <p) = pn(An cos nip + Bn sin n<p) for p ^ a ,

un(p, <p) = \ (An cos n(p + Bn sin nip) for p^a .

9

The sums of these solutions,

u(p, ip) = f pn(An cos n<p + Bnsmwp) for the interior problem
71 = 0

and

°° 1
u{p, <p) = J, — (An cos nip + BnSinncp) for the exterior problem,
71 — 0 p

are also harmonic functions, provided they converge.

Now in order to determine the coefficients An and Bn we use the boundary
condition

u(a, <p) = ^ an(An cos rup + Bn sin n<p) = f. (4-3.6)

71 = 0

We assume that / can be written as a function of the angle <p. Then the
Fourier series for / has the form

(X
f{<P) = -7Г + 1 {an COS nip + /3* sin nip) (4-3.7)
Z 71=1

with

«о = — \ f(<P) dip , an= — \ f((p) cos nip dip

n J-* 7Г

f(ip) sin пф dip

7Г

By comparison of series (4-3.6) and (4-3.7) we then obtain

83 The Laplace differential expression in polar coordinates (4-3.3) loses its meaning
for p = 0. Therefore, we still have to prove that Aun = 0 also for p — 0. For this proof
we shall use not polar but rectangular coordinates; the particular solutions

pn cos nip and Pn sin nip

are then the real or imaginary parts of the function

pngincp = (реі(р)п = (x + iy)n :

it is a polynomial in x and y. Now, however, it is clear that a polynomial which satis¬

fies the equation Au = 0 for p > 0 also is satisfied for p = 0, because of the continuity
of the second derivatives of this equation.
4-3. SOLUTION OF THE BOUNDARY-VALUE PROBLEMS 273

A - h- A — —
■n'n
Ao~ 2 ’
—
a ’

for the interior problem and

A0 = An — OCn& , Bn = P nan

for the exterior problem. In this manner, we have found a solution of the
interior first boundary-value problem for the circle in the form

u{{o,(p) = ~-+ I (—) (ocn cos rup + Pn sin rup) (4-3.8)

and the solution of the exterior problem in the form

u(p, Ф) = («» cos n<P + Pn sin rup) . (4-3.9)

2 n=i \ p /

In order to prove that these functions represent the sought solutions,

we also have to prove the applicability of the superposition principle. For
this purpose we show the convergence of the series, the termwise differen¬
tiability, and finally the continuity of the functions represented by it on the
circumference of the circle. Both series can be subsumed in the homogene¬
ous formula

u(p, <p) = 2 tn(ancos,rup + pn sin tup) + ,

71 - 1 Z

where

U <1 for f) S a (interior problem)

a

’ *-<l for p 2: a (exterior problem)

P
and an, pn are the Fourier coefficients of f((p).
To prove that series (4-3.8) and (4-3.9) for t < 1 are arbitrarily often term-
wise differentiable, we set

Un = tn(an COS rup + Pn sin n<p)

and calculate the &th derivative of the function un with respect to <p\

= tnnk cos ^ rup + k + pn sin ^ rup + k •

Hence we obtain the estimate

дкип
й tnnk2M,
d<pk

where M is the maximum of the absolute values of the Fourier coefficients

an and pn, i.e.,

I an I < M , I pn\ < M . (4-3.10)

274 ELLIPTIC DIFFERENTIAL EQUATIONS

Now we choose a fixed value p0 < a (for the interior problem) or oy — a2/p0 > a
(for the exterior problem) where t0 = p0/a < 1. Then,

2 tnnk(\an\ + \pn\)^2MItnonk , Ій t0,

n=1 n=l

from which because of the convergence of the series on the right we infer
the uniform convergence on the left for t й to < 1 and arbitrary k. Therefore
series (4-3.8) and (4-3.9) at each point lying inside (outside) the circle can be
differentiated arbitrarily often with respect to <p. Correspondingly, we show
that series (4-3.8) and (4-3.9) inside (outside) a circle of radius p0 < a can be
differentiated arbitrarily often with respect to p.
Now po was chosen arbitrarily under the given restrictions. Therefore,
series (4-3.8) and (4-3.9) at each interior (exterior) point of the circle are term-
wise differentiable. Consequently, the superposition principle is applicable
and the functions (4-3.8) and (4-3.9) satisfy the equation Au — O.84
For the proof we only used the fact that the Fourier coefficients of /(y>)
were bounded (formula (4-3.10)). However, this is valid for every bounded
function (indeed for every absolutely integrable function). Therefore, provided
f(<p) is a bounded function, the series (4-3.8) and (4-3.9) define functions which
satisfy the equation

Au = 0 for t < 1 .
We shall use this fact later for the generalization of the results found
in this section.
We turn now to the proof of the continuity of the solutions in the closed
region t й 1. It is easy to see that we must go into the properties of the
function f(<p) more precisely.
Because of the assumed continuity and differentiability of f(p), it follows
that f(<p) can be developed in a Fourier series and that the series

І (I а» I + I I) (4-3.11)
n—1

converges. On the other hand, we have

I tnan cos n<p I й I 0Ln I

I tnpn sin n<p I g I pn I

so that series (4-3.8) and (4-3.9) converge uniformly for t й 1. The functions
represented by them are therefore continuous on the circumference of the
circle. Finally, formula (4-3.11) shows that the function (4-3.9), as the solution
of the exterior problem, is bounded at infinity. Therefore, we have proved

84 The Laplace differential equation is also satisfied for p = 0; if, therefore, the de¬
rivatives with respect to the polar coordinates are expressed through the derivatives in
terms of the cartesian coordinates, we can easily see that the functions (4-3.8) and (4-3.9)
are differentiable for t й to arbitrarily often with respect to x and y. On the basis of
footnote 83, it follows that

Au — 0 for p =0 .
4-3. SOLUTION OF THE BOUNDARY-VALUE PROBLEMS 275

that series (4-3.8) and (4-3.9) completely satisfy the conditions of the problem.

2. Poisson’s integral

We shall first transform formulas (4-3.8) and (4-3.9) to a simpler form.

For this purpose we consider, for example, the interior problem. Correspond¬
ing results hold also for the exterior problem.
By introduction of the expression for the Fourier coefficients into (4-3.8)
and by an interchange of the order of summation and integration we obtain
first of all

u(p, <p)
_ J_
7Г
Г m\[-І-
l 2
+ f (p. y,
n=i \ a )
I пф cos n<p + sin пф sin tup) > йф

Г
(4-3.12)
J_
71
m\
l 2
[i-
+ I (^Y'cos n(<p
»=i \ a /
— Ф) |# •

Now we use the following identity:

1 oo
+1 tn cos n{<p — Ф)
} ”
— I
2 n—1 = y + ' 2 n=i
4-
2 [
/l + n=2i [{teu<e~^)n + (fir^^’nlJ
teUv~^
-ii+— +
2 \ ' 1 — }
1 1 - t2
t = -P-< 1.
2 1 — 21 cos (<p — ф) + t2 ’ a

If we introduce this result into (4-3.12) we obtain

1 - P2la2 йф
u(p, <p) = j' № (p2la2) — 2(p/a) cos (<p — ф) + 1

or

a — p
u{p, <p) = 7Г , №
27Г J —; 2 — lap cos {<p — ф) + a2^
(4-3.13)

This formula for the solution of the first boundary-value problem for the
interior of a circle is the Poisson integral, and the expression under the in¬
tegral sign
„2 2
a — p
K{p, <p, а, ф) =
p — lap cos (<p — ф) + a

is called the Poisson kernel. For p < a, we have K(p, <p, а, ф) > 0, since
lap < a2 + p2 for p Ф a.
The Poisson integral was derived under the assumption that p > a\ for
p — a, expression (4-3.8) loses its meaning. However,

lim u(p, <p) = f(<p0) ,

p—a
<P-<Pq
276 ELLIPTIC DIFFERENTIAL EQUATIONS

because the series from which we obtained the Poisson integral is a continu¬
ous function in a closed region.
Consequently the function defined by

for p < a ,
u{p, (f) = I 2k L,mpl - 2ap cos (<f-ф) + atd<P (4-3.13')

{/(?) for p — a

is a solution of the equation Au = 0 for p < a, and in the closed region, in¬
cluding the boundary p = a, it is continuous.
The solution of the exterior problem has the form

1 p — a
7Г №p ap cos (<p — ф) + a'
АФ for p > a ,
u(p, (p) — 2k J_ — 2 (4-3.14)

l/M for p = a .
Under our initial assumption that the function f(<p) is continuous and
differentiable, we proved that the solution can be represented by a series.
Finally, by means of the above-cited identity, we arrive at the Poisson in¬
tegral.
Now we shall prove that the Poisson integral represents the solution of
the first boundary-value problem when the function f(cp) is continuous only.
The Poisson integral represents the solution of the Laplace differential
equation in the region p < a (t < 1) for an arbitrary bounded function f(<p).
Therefore for p < a (t < 1) the Poisson integral is identical with the series
(4-3.8) and the equation Au — 0 is satisfied for every bounded function f(<p).
It still remains to be proved that the function u(p, <p) in our case continu¬
ously approaches the boundary values. For this purpose we choose any sequence
of continuous and differentiable functions

M<p), ■■■, fk{<p).

which converges uniformly towards the function /(p):85

lim/*(£>) = f(<p) .
fc —oo

The sequence of boundary functions then corresponds to a sequence of harmonic

functions Uk(p, <p), which are defined by (4-3.13) or (4-3.8).
From the uniform convergence of the sequence {/4(9?)} it follows that
for every e > 0, a k0(e) > 0 exists such that

I fk(<p) — fk+i(<p) \ < £ for k > k0(e) , / > 0 .

For the functions Uk{r, <p) which are solutions of the first boundary-value
problem it follows from the principle of the maximum that

I ик{р, (p) — ик+і(р, <p) I < £ for p й po , k > k0(e) , e > 0 .

Consequently the sequence {uk} converges uniformly towards a definite

85 We shall not concern ourselves how it arises; such a sequence can be chosen in
many ways.
4-3. SOLUTION OF THE BOUNDARY-VALUE PROBLEMS 277

function и = lim Uk(p, <p). The limit function u(p, <p) is continuous in the closed
к— °o
region, since all functions Uk(p, <p) which are represented by the integral

Uk(p, <p) = J_r _ МФ) dtp

2* W-22ap cos {<p — <p) -f a2
are continuous in the closed region. Finally we have
2 2
1 a — p
f(<p)d<p for p < a
u(p, <p) = lim uk(p, <p) = 2л J-іб — 2ap cos {<p — tp) + p2
k — oo
f(<p) for p - a

since the sequence {fk(<p)} converges uniformly towards f(tp) and therefore
the passage to the limit can be carried out under the integral sign.
The function

If11 a2 — o’
u(p, V) = -f 2 9 / . № dtp
2л ) -np —2ap cos (tp — <p) + a

is thus, for every continuous function f(<p), a solution of the Laplace differen¬
tial equation which can assume the continuous boundary values given on the
circumference of a circle.

3. Discontinuous boundary values

We now show that formulas (4-3.13') and (4-3.14) for every piecewise
continuous function f(tp) represent the solution of the boundary-value problem,
i.e., that this solution in the entire region is bounded and tends continuously
to the boundary value at the points of continuity of the function f(<p).
Therefore also it is a unique solution (see Section 4-2 § 4). If <p0 denotes
any point of continuity of the function f(<p), we have to prove that a d exists
for every e > 0 such that

I U{p, tp) — f(tpo) I < £ ,

provided
I p — a I < d(e) and | tp — <p0 \ < <5(e) .

Because of the continuity of f{<p), a d0(e) can be found such that

І/Ы-/ЫІ <y

holds when
I ip — <Po I < SQ(s) .

We consider now two continuous differentiable auxiliary functions f(<p) and

f(<p) which satisfy the conditions

f(<p) = f(<po) + y for 1 <p — <Po 1 < <5o(e)

IIV

for 1 <P ~ <Po 1 > do(e)

278 ELLIPTIC DIFFERENTIAL EQUATIONS

f(<p) = A<po) --j for 1 9 — 9o 1 < <50(e)

f(<p) =; f(<p) for \9 — 9o \ > ^o(e)

but are otherwise arbitrary and for which formula (4-3.13) is applicable.
Then if we define the functions U(p, <p) and u(p, <p) by the formula (4-3.13)
for / and /, then these are harmonic functions which have the functions f(<p)
and f(<p) as boundary values.
Since the Poisson kernel is positive, then

U(p, (f) ^ u(p, <p) S u(p, <p)

because

f(<p) й f(<p) Sf{<p) .

Further, from the continuity of u(p, (p) and u(p, <p) on the boundary for 9 = <p0,
it follows that a 5i(e) exists such that

I u(p, <p) —f(<p0)\

for

I p — a I < 5i(e) , \<p — 9o \ < di(e) .

and

I u{p, <p) — f((p0) I S~-

for

I p — a I < <Ы£) , I <p — <Po I < 5i(e) .

From these inequalities we find

u(p, <p) ^/((po) Л-— - f(<p0) + £

f(<p0) - £ =f(<Po) —= %(P> <°)

for

I p — a I < 5(e), I ip — <p01 < S(e) d = min (50, 5,) .

If now we collect these inequalities, we obtain

f(<Po) — £ ^ u(p, <p) ^ u(p, <p) -й. й(р, <p) й f(<p) + £

or

I u{p, (p) — f((po) I < £ for I a — p I < 5(e), | <p — <p0 \ < 5(e)

and hence the continuity of u{p, <p) at the point (a, (pQ).
We shall now show the boundedness of u(p, <p). Since the Poisson kernel
is positive, then
4-4. GREEN’S FUNCTION 279

2*
a
u(p, <p) < M^- йф = M,
jo a2 + p2 — 2ap cos (<p — <p)

provided \f{<p)\ < M. Now, however, we have

_1_ (a2 - p2)d<p

= 1 ,
2л- Jo p2 — 2ap cos {<p — ф) + a‘

since the left side, as was proved earlier, represents a harmonic function
which tends continuously toward the boundary value / = 1. Such a func¬
tion, however, is identically equal to 1. Correspondingly, we [could] show
that u(p, (p) > Mi for f > Mi and therefore prove the boundedness of the
absolute value of u(p, <p).

4-4. GREEN’S FUNCTION (SOURCE FUNCTION)

The use of Green’s function is a convenient aid in the analytical treat¬

ment of boundary-value problems. In this section we give the definition
and basic properties of the Green’s function for the Laplace differential
equation as well as the construction of the Green’s function for a class of
simple regions (circle, sphere, semi-infinite body). The construction follows
by using the method of images in electrostatics.

1. Green’s function for the equation Ди = 0 and its basic

properties

Let u(M) be а function which with its first derivatives in a closed region
T bounded by a sufficiently smooth finite surface I, is continuous; further,
let u(M) in the interior of T possess continuous second derivatives. Then,
as was proved in Section 4-2 § 1, the integral representation

u(M0) = —
гг i ди Ди
dru (4-4.1)
4л 1J L rPMo dn ?MM 0
is valid. If u{M) is a harmonic function, the volume integral equals zero;
if u(M) satisfies the Poisson equation, the volume integral represents a
known function.
Further let v(M) be a harmonic function which nowhere possesses singu¬
larities. The second Green’s formula

then yields

0=i j (“s " “!)*' - J У vjudt ■ (44-2)

By addition of (4-4.2) and (4-4.1) we obtain

u(M0) = Ди • Gdr (4-4.3)

280 ELLIPTIC DIFFERENTIAL EQUATIONS

where

G(M, AT.) = —-— + v (4-4.3')

4 лТлшо

is a function of two points M0(x,y,z) and M(£,rj,0- The point M0 is held

fixed so that the variables x, y, z play the role of parameters.
In the interior of the region T the function G(M, M0), except at point
M = M0 where it possesses a singularity of the form 1/4лт, everywhere
satisfies the equation

AG = 0 .

We now choose v(M) so that

G |r = 0 ,

i.e.

The function G defined in this manner is called the Green’s function of the
first boundary-value problem for the Laplace differential equation Au — 0 (or
also, the function of an instantaneous point source of the first boundary-value
problem for the differential equation An = 0). It permits us to give a direct
representation of the solution of the first boundary-value problem for the
equation Au = 0. From (4-4.3) follows

u(M0) = -l\u^dff = -\y£da, f=u\2. (4-4.4)

It must also be noted that formula (4-4.4) was derived by the use of
Green’s formula, where certain conditions concerning the functions u(M) and
G(M, M0) as well as the boundary 2 were assumed. Moreover, the expres¬
sion dG/dn occurs, whose existence on 2 is not immediately seen from the
definition of G.
In the derivation of (4-4.4) we proceed from the existence of a harmonic
function u(M) which assumes the value / on I. For the region in which
the Green’s function exists and for which the Green’s formula is applicable,
formula (4-4.4) thus yields an explicit representation of only such solutions
u(M) of the first boundary-value problem as satisfy the conditions for the
applicability of the Green’s formula. (This formula also proves the uni¬
queness of this class of solutions of the first boundary-value problem.)
A detailed investigation by A. M. Liapunov showed that under very
general assumptions formula (4-4.4) represents the solution of the first
boundary-value problem for a wide class of boundaries—the so-called Liapunov
surfaces (see Section 4-5).
We turn now once more to the definition of the Green’s function G(M, M0)
which was introduced by means of the function v(M)—itself a solution of the
first boundary-value problem for the equation Av = 0 with the boundary-values
4-4. GREEN’S FUNCTION 281

v\i = — 1/4лт. This gives the impression that here this one problem (the
determination of the solution и of the first boundary-value problem) would
lead back to another of equal difficulty (the determination of v as the solution
of the same problem). This is not the case, however, since the knowledge
of the Green’s function allows the boundary-value problem to be solved for
an arbitrary boundary condition {u\i — /), while for the determination of the
function G itself only the boundary-value problem with the particular
boundary condition {v\z = —1/4лr) is to be solved. This problem is consider¬
ably simpler, as we shall see from a series of examples.
In electrostatics, the Green’s function

G(M, M0) -- —-h v

47zr

represents the potential86 at a point M of a point charge which is found

inside of a grounded conducting surface I at the point M0.
The first summand 1/4яr is obviously the potential of the point charge
in free space whereas the second summand v represents the potential of the
fields which arises from the charges induced on the conducting surface I.
The construction of the Green’s function therefore follows from the deter¬
mination of the induced fields.
In the following, we shall investigate several properties of the Green’s
function. We shall assume that the region considered is such that the Green’s
function G(M, M0) exists for it and possesses a continuous normal derivative
dGIdn. Moreover, we assume that Green’s formula is applicable on I.
1. The Green’s function is positive everywhere in the interior of T.
G is equal to zero on the boundary I of the region and is positive on the
surface of a sufficiently small sphere about the pole. Therefore G must be
positive according to the principle of the maximum in the entire region.
Further we note that {dGIdn) | z ^ 0, which follows directly from the positivity
and the condition G \ s = 0.
2. The Green’s function is symmetric with respect to its arguments
M0{x, y, z) and M(f, г], C):

G(M, M0) = G(M0,M) .

86 It is well known that the potential v of a point charge of magnitude e in a

medium with dielectric constant s in the cgs system is given by v = e/er. Hence the
Green’s function corresponds to a charge of «/4K in absolute electrostatic units. If, on
the other hand, we use the so-called Giorgi system, in which Coulomb’s law has the
form / = ee'l^jzsr, then in vacuum (e = 1) the Green’s function G(M, Mo) corresponds
to a unit charge.
In heat-conduction theory, the stationary temperature of a point-forming heat source
of intensity q is determined by the expression q/Ankr, where к is the coefficient of
thermal conductivity. Therefore, G{M, Mo) is the temperature at the point M, provided
the temperature of the surface of the body is equal to zero and at the point Mo a heat
source exists of intensity q = k.
If the units of length are so chosen that к = 1, then the function G{M, Mo) cor¬
responds to a source of unit intensity.
282 ELLIPTIC DIFFERENTIAL EQUATIONS

For the proof, we consider two spheres Ti and T2 of radius e with Mi and
Mo as center points; thus let Mi and Mi' be two arbitrary fixed points in
T (Figure 50). By 7\ we shall designate the region which results from T by
the removal of the regions enclosed by and I2. Now if we set
и = G(M, M'o) , v = G(M, M'o')

and apply the Green’s formula

(:uAv — vAu)dv = (4-4.5)

to Te, we obtain

[ \\G(M, Mo) dG(M’ -G(M, M'o')dG(M’ M'o)~\d<js

JxJL dn dn J

+ J j j~G(M, M'o)- G(M, M'o')dG{M’ M'o]

dn
= 0

since because AG = 0, the left side of (4-4.5) is equal to 0, while the integral
over I vanishes because of the boundary conditions.
Now, if we let e approach zero and consider the singularity of the
Green’s function, we obtain the expression:87

G(M'0, Mo) = G{M'0', M'o)

or
G{M, M0) = G(M0, M) .

The above-proven symmetry of the Green’s function is a mathematical

expression of the principle of reciprocity in physics: A
source situated at the point M0 has the same effect at a
point M as the same source situated at point M has at
point M0. The principle of reciprocity is of a very
general character and arises in different physical fields
(electromagnetics, elastic, etc.).
In the two-dimensional case the Green’s function
G(M, Mo) is defined by the following conditions:
1. AG = 0 everywhere in the region 5 under con¬
sideration, except at the point M = M0 .
2. At M = M0, G(M, M0) has a singularity of the
form

51~ln-.
i 1
Ал Тммо

3. G\o — 0 if C is the boundary of S. In this case the Green’s function

has the form

G(M, Mo) = In—!— + v(M, Mo) ,

Ал Ymmo

87 Liapunov derived this theorem in the application to the class of so-called Liapunov
surfaces.
4-4. GREEN'S FUNCTION 283

where v is any harmonic function everywhere continuous and on the boundary

satisfies the condition

. 1 , 1
v\a = — 7Г In-•
Умм0

The solution of the first boundary-value problem for Ли = 0 is then given by

u(M0) = — \ }^~ds , f — u\o .

J 0 on

2. The method of images in electrostatics (the electrostatic

image) and the Green’s function for the sphere

A method often used for the construction of the Green’s function is the
method of images in electrostatics. Its fundamental idea in the construction
of the Green’s function

G(M, M0) = —-— + v

47Г У MMQ

arises from the fact that the induced field v can be represented by the field
of the charges which are distributed on the exterior of the surface I and
are so chosen that the condition

is satisfied. These charges are called the electrostatic image points of the
unit charge found at the point M0, which, if I were not present, would
produce the potential 1/4ят. In many cases the choice of such charges causes
no difficulty. We shall give three examples for the construction of the
Green’s function by means of the method of electrostatic images. From the
representations of the Green’s function, which will be given for the three
examples, the continuity of the first derivatives of G on S will be obvious.
As the first example we consider the Green’s function for the sphere:
Given a sphere with radius R whose center lies at the origin of coordinates
O, find the corresponding Green’s function.
For the determination of this function we place a unit charge at the
point M0 and choose the segment OMi on the radial line passing through M0
such that

PoPi — R , (4-4.6)

where p0 = OM0 and px = OM, (Figure 51).

Figure 51 shows the correspondence of the point M0 with a definite
point Mi, Ъу а transformation of reciprocal radii. The point M{ is called
the point conjugate to M0. This construction is reversible, that is, the point
M0 can be regarded as the point conjugate to M,.
We prove now that the distances of all the points P lying on the spherical
surface are proportional to M0 and Mt respectively. Therefore we consider
284 ELLIPTIC DIFFERENTIAL EQUATIONS

the triangles OPM0 and OPM\ (Figure 51); they are similar since the angle
at О is the same for both triangles and the adjacent sides are proportional:

Po R OMo R
or
R pi R OMi •

However, from the similarity of the triangles

Го _ Po_ _ R_ (4-4.7)
Гі R pi

follows, where r0= IM0P |, rx = \MiP\. From

(4-4.7) we obtain

for all points of the spherical surface. There¬

fore, the harmonic function v = — (R р0)ОіГі)
FIG. 51. on the spherical surface assumes the same
values as the function 1>0. It obviously
represents the potential of the charge found at Mi of magnitude —R/pa.
Therefore the function

(4-4.8)

is the sought Green’s function for the sphere; thus it is a harmonic function
which at M0 has the singularity (1 4—)(1 r„) and is equal to zero on the spherical
surface.
The solution of the first boundary-value problem can then be obtained
from formula (4-4.4).
We now calculate the normal derivative

(4-4.9)

where n is the exterior normal and ri = MXM (in general, M does not lie on
the spherical surface).
For the derivatives of l/r0 and l/гі along the normal n we have

(4-4.10)

since

dr
4^- = cos (r0, n) , = cos (iq , n) . (4-4.11)
on on

For cos(r0, n) and cos(r!, n) we easily find

4-4. GREEN’S FUNCTION 285

_ R2 + rl - pi
COS (r0, n) (4-4.11')
2Rr0
r>2 I 2 2
R + ri — p,
COS (Гі , It) (4-4.11")
2 Rrt

With the help of (4-4.7) we obtain

p2 ,
Л -Г 2 Го 2 2,2 т-,2
Po Po _po 4" Го — /t
COS(r1; Л) I г =
г,г> R 2роГо
ZR—r
Po

since according to the definition of the points Mi the relation

R2
Pi = —
po

holds, and

Гі = — Го on I .
po

By using (4-4.10), (4-4.9), (4-4.11'), and (4-4.11"), we find

2 - 2
r>2-
dG 1 [ 1 R2 + rl-pl pl R po + Уo —
R 1 R2-Pl
dn x 4я[ Го 2 Rr0 R2rl po 2роГо 1 4 kR rl

Consequently u{M0), with the use of (4-4.4), equals

и(Mo) = ( [f(P) ——dap . (4-4.12)

47r/?JXJ Го

We now introduce a spherical coordinate system whose origin lies at the center
of the sphere. Now let R, Ѳ, <p be the coordinates of the point P and p0,
e0, <po the coordinates of the point M0; let у be the angle between the radius
vectors OP and OM0. Formula (4-4.12) can then be written in the form
p л2й|-я p2 _ 2
u(po, 00, <Po) = T-\ 1 fi.0,<p)-775.-np , 2\3/2-^ітіѲdPdtp (4-4.12')
4л: Jo Jo {R — ZRp0 COS у + Po) '

where88

cos у = cos Ѳ cos Ѳо + sin Ѳ sin 0o cos (<p — <p0) . (4-4.13)

This formula is called the Poisson integral for the spherical surface.
By the same method the Green’s function can also be constructed for
the region lying exterior to the sphere:
—У —^

88 The direction cosines of the vectors OP and OMo are therefore equal to
(sin Ѳ cos <p, sin Ѳ sin <p, cos Ѳ) or (sin Ѳо cos <p0, sin Oo sin <p0, cos do) .

Thus
cos у = cos Ѳ • cos Ѳо + sin Ѳ sin Ѳо (cos <p cos <p0 + sin <p sin <рй)
= cos Ѳ cos Ѳо + sin Ѳ sin Ѳо cos (<p — <p0) .
286 ELLIPTIC DIFFERENTIAL EQUATIONS

(4-4.14)
4 л: Vn pi r0
Here ri = MMi is the distance of a fixed point Mx lying outside of the sphere,
r0 = MM0 is the distance of the point M0 conjugate to Mi, pz is the distance
of the point Mi from the origin of coordinates, and R is the radius of the
sphere.
If we bear in mind the distinction between the direction of the normals
for the interior and the exterior problem, then we obtain
л 2it p it

Pi R2
u{pi, Ѳі, (pi) = „— ,
4л: J о J, о го* 9 VnCOSc.I
[R — 2piR у + pi\ ' Ode d<p ,

where cos у is given by formula (4-4.13) (the subscript zero is there replaced
by 1).

3. The Green’s function for the circle

The Green’s function for the circle can be obtained in the same manner
as the Green’s function for the sphere. In this case the Green’s function
assumes the form

G=^-ln— + v. (4-4.15)
2л: r
If we repeat the arguments of the previous
section from formula (4-4.6) to formula
(4-4.8), we obtain for G(P, M0) the repre¬
sentation

G(P, м0) =-!-Гіп— -In——1 , (4-4.16)

2л:|_ r0 p о r ij
where p0 = OM0, r0 = M0P, ri = MXP, and R — OP is the radius of the circle
(Figure 52). We see immediately that the harmonic function so defined on
the boundary C is equal to zero,
G\o = 0 .

For the solution of the first boundary-value problem the normal derivative
dG/dn on the circumference C must be calculated. The calculation proceeds
as for the sphere and yields
8G 1 R2-pl
дп о 2лR rl
Now let p, Ѳ be the polar coordinates of the points P lying on the circum¬
ference and p0, Ѳ0 the coordinates of M0. Then, first of all, we have

To = R + po — 2Rpo cos (Ѳ — Ѳо) •

If we substitute this expression for r0 into the formula
4-4. GREEN'S FUNCTION 287

and note that

u(P) I о — f(0) and ds = Rdd

then we obtain for the function u(M0) the expression

1 Г
2
R2 po
u(p0, Ѳ0) = — \ /«?) dff , (4-4.17)
Lll Jo R + (Oo — 2i?p0 cos (0 — 0O)
the so-called Poisson integral for the interior of a circle. Except for the
signs this formula also gives the solution of the exterior problem.

4. The Green's function for the half space

The notion of the Green’s function and formula (4) are also valid for an
unbounded region of space, provided the functions considered are regular at
infinity (see Section 4-2 § 6).
To determine the Green’s function for the half space z > 0, we place at
the point M0(xo, y0, z0) a unit charge which produces in the unbounded region
of space a field whose potential is defined by the function

TIT—> (rMQM =V(x — x0Y + {y — :y0)2 + (z - z0)2) •

Г MqM

We easily see that the induced field v is the field of a negative unit charge
placed at the point Mi(x0, Уо, —z0), which is the mirror image of the point
M0 in the plane z — 0 (Figure 53). The function G(M, M0) is equal to

1 1
G(M, M0)
4лг0 4лгі
where

Го = MoM = V{x — XoY + (у - Уо)2 + (2 — ZoY

ry — МуМ — q/(x — x0Y + (у — Уо)2 + (z + zo)2.

G(M, Mo) is equal to zero for z = 0 and has the required singularity at the
point Mo ■

We now calculate

3G dG
dn dz

Obviously,

dG Zp Z + Z0
=i.r.
dz 4 7Г [_ Го r\ ]■
For z = 0, we thus find

dG _ dG Zp

dn z=o dz 2nrl '

FIG. 53 The solution of the first boundary-

288 ELLIPTIC DIFFERENTIAL EQUATIONS

value problem is therefore given by

_£o_
u(Mo) = — ' f{P)dop ,
Jx, .3
M0P

where T0 is the surface 2 = 0 and f(P) = и |*=0. Thus we can also write
OO Г со

u(xо , y0 , Zo) =
Zo
■f(x,y)dxdy. (4-4.18)
2n J-coJ-,» {{x — Xof + (у — Уо)2 + .31З/2
Zo]

4-5. POTENTIAL THEORY

The function
2_1_
r _ V(X - I)2 + (y - >y)2 + (2 - C)2 ’

which represents the potential of the field of a unit mass (charge) placed at
the point M0(f, у, 0, is a solution of Laplace’s equation. Here £, 77, and C
are interpreted as parameters. The integral of this function with respect to
the parameters is called the potential. It has a significant meaning both in
physics and also in the development of the methods of solution for boundary
value problems.

1. Spatial potential (Newtonian potential)

Let а mass m0 exist at M0(f, rj, £). According to the law of gravitation,
the attracting force on a mass m placed at the point M(x, y, 2) is

F=_rmmor
(4-5.1)

Here r = M0M and 7- is the gravitational constant. By a suitable choice of

units we can make у = 1. Further, if we set m = 1, we obtain

F=

The projections of this force on the coordinate axes are

X = F cos a = - ™°(x-£)
r

Y = F cos /3 = — -y (y — y) (4-5.2)
r

Z = F cos у = -^т(г-С)
г
where a, [3, and у are the angles between the vector F and the coordinate
axes.
Through

F = grad и
or
4-5. POTENTIAL THEORY 289

v _ du Y _du du
dx dy dz

we introduce the so-called potential u(x, y, z) of the force field.89 In our case

we have

By superposition of the force fields, the potential field of n mass points can
be expressed by the formula

V' Y' ''H

u = 2 Ui = 2 — •
t=1 i=l r

We treat now the case of a continuous distribution of mass and consider

a bounded region of space T with a density distribution p(£, rj, Q. In order
to determine the potential of this spatial region at a point M{x, y, z), we
divide T into sufficiently small volume elements At and assume that the in¬
fluence of a spatial element At is equivalent to the influence of a mass con¬
centrated at an interior point (£, rj, 0 of dr.90
For the X component of the force acting on the point M we then obtain

AX= -£$■(*-£) ,
r
with

r2 - (x - 6)2 + (y- v? + (z- C)2 -

The X component of the potential of the total region T at M is obtained

by summation of the individual potentials over all spatial elements, i.e., by
integration over T:

x=~\l\p£vldT- m-3)
Therefore we obtain for the potential of a spatial region T at the point M
the so-called Newtonian potential

«(Af) = jj ^р-уЛт . (4-5.4)

If M lies outside of T, then this follows directly by differentiation under the

89 The potential introduced here should not be confused with the energy potential
of a force field. Potential here is understood in the sense of the force function in
mechanics.
90 More precisely stated, it is assumed that the influence of a body T of mass m
on a point lying outside a convex region T containing this body can be replaced by the
influence of a certain concentrated force with the same mass m which lies in the interior
of T.
290 ELLIPTIC DIFFERENTIAL EQUATIONS

integral sign.91 We calculate the higher derivatives in a similar way. For

each point M lying in the exterior of T, u{M) satisfies the Laplace differential
equation.
In the following we shall use the above-mentioned properties of the
potential and formulate a series of theorems under the condition that p(£, rj, £)
is a bounded (and also integrable) function. We do not intend to develop
the theory under the most possible general assumptions.
If the point M lies in the interior of T, then we cannot directly conclude
that X = du/dx. We shall go into the necessary additional investigation later.

2. The planar problem (logarithmic potential)

We now investigate a special mass distribution which depends only on

the two coordinates л: and y. Obviously then the potentials on each of the
planes v = const, assume one and the same value.
Therefore it is sufficient to consider the potential
at the points (x, у) which lie in the plane 2 = 0.

First we shall determine the potential of a

homogeneous infinite straight line L. For this we
choose the 2 axis in the direction of this straight
line. Let the linear density (i.e. the mass per
У unit of length) be equal to p. For an element Az
fig. 54. on a point P(x, 0) (Figure 54), we obtain the attract¬
ing force whose component with respect to the x
axis is

/iAz _ pAz
1?~ _ (X2 + 22)

AX = AF cos a = — uAz-■■ .
r V{x2 + 22)3
Hence it follows that

dz 2 1
й =-
2P
X= */2
cos a da — ,
fX (X2 + 22)3/2 “ ^ X3 -*/2 x
z
— = tg
* a .
r
If P{x, y) is an arbitrary point, then the attracting force acting on a point
—»
at О obviously is in the direction of OP and its magnitude is equal to
91 A sufficient condition for differentiation under the integral sign, with respect to
a parameter, for an integral of the form

f(M) = \tF(M, P)<p(P)drP

is the continuity of the derivative of F(M, P) with respect to the parameter and the
absolute integrability of <p{P). Ordinarily this theorem is formulated only for <p{P) — 1,
although the proof for this and the general case are not different.
4-5. POTENTIAL THEORY 291

where

p = ~[/x2 + yz .

The potential of this force is called the logarithmic potential; it is equal

to

1
V = 2pln (4-5.5)

which we recognize directly by differentiation.

The logarithmic potential represents a solution of the two-dimensional
Laplace differential equation which is circularly symmetric with respect to
the point p = 0 and at this point becomes
logarithmically infinite. Thus the potential
of a homogeneous straight line represented
by formula (4-5.5) produces a planar field.
The representation of the potential in the
form of an integral was derived only for a
bounded region.92
We note that in contrast to the spatial
potential, the logarithmic potential at infinity fig. 55.
does not equal zero but has a logarithmic
singularity there.
For the components of the attracting force at point P (Figure 55) we find

X — F cos a = —2p — cos a = —

P P

Y = Fsina = —2p-~ , У
sin a = —
P

If several points (an infinite straight line with

distributed mass) exist, then because of the super¬
position of the force fields the potentials at the
point P can be added (straight line).
In the case of a region 5 with continuous density distribution (Figure 56),

92 In the calculation of the potential of an infinitely long straight line, the potential
of the individual elements cannot be integrated directly since an improper integral
would result. The potential of an element Az is therefore equal to

Az
Ди — p-
*/ p2 + z2
so that formal integration would yield the improper integral

dz
и
F p2 + z2
292 ELLIPTIC DIFFERENTIAL EQUATIONS

the components of the attractive force on the point P can be expressed by

the double integrals

X=
x-e ■d$drj
v)
(x - o2 + (y- v) (4-5.6)

y— ~2 fJsjUe,9)7(y-£=-?—
— y)+(x — t)
-d£ dr]

and for the potential we find

y) = 2 Ц*. ,) d£ dr] , (4-5.7)

which we can prove by differentiation if the point lies outside the region S.
If P lies inside of S, then an additional investigation is necessary in the
proof.

3. Improper integrals

Potentials and components of gravitational forces can be represented by

integrals whose integrands become infinite when we consider their value at
the points at which the attracting masses are placed.
If the integrand at any point of the region of integration is infinite, the
integral as is known cannot be defined as the limit value of the integral
sum. In this case, therefore, the integral sum has no limit value, since the
summand corresponding to the element of volume containing the singularity
can arbitrarily change the magnitude of the sum, depending on how one
selects the corresponding intermediate points. The integrals of such functions
are defined as improper integrals.
F(x, y, z) be a function which becomes infinite at an arbitrary point
Let
M0(x0,yo,z0) in a region T. We then consider the integral over a region
T — Ke, where Kz denotes an arbitrary neighborhood of the point M0 whose
diameter does not exceed e.
If now the sequence of regions Ks shrinks in an arbitrary manner about
the point M0 and if the sequence of integrals

/. = [[ \Fdv en^0
JJ T— K^nJ

has a limit value independent of the choice of Ken, then the limit value is
called an improper integral of the function F(x, y, z) over the region T and
is usually denoted by

If at least one sequence of neighborhoods KSfl exists such that as en —>■ 0 a

limit value / exists, whereas for another sequence {/Gn} this limit value has
another value or does not exist, then the limit value I is called a conditionally
convergent improper integral. Naturally, for the consideration of a condi-
4-5. POTENTIAL THEORY 293

tionally convergent integral, a sequence of neighborhoods {A%re} must exist

through which it is defined.
We limit ourselves here to the consideration of those cases in which the
integrand has an isolated singularity, such as integrals of the form

(4-5.8)

where C and a > 0 are constants and

r = rMД-0 = l/Uo — £)2 + (Уо - ѵ)г + Uo — О2

and M0 denotes a point in T. Without loss of generality we can assume T

to be a sphere of radiusR about M0. Further, we choose as neighborhoods
Ktn spheres of radii en about M0. For the calculation of the sought limit
value of the corresponding sequence of integrals we first obtain

==2к - 2CL” £
47гС Г ——~—F a~\ for аФ3
Ы — a
4л:C [In r]f„, for a = 3 .

Now if we let en—>0, then it can be shown that the limit value exists for
a < 3 but not for a ^ 3.
Now let F(x, y, z) be a nonnegative function and let the following limit
value exist

/ = lim ((( F dr ,
sn-° J J Jr-Ken

where Ks„ represents spheres of radii e„ about

M. Further let {Aen} be an arbitrary sequence
of regions, which shrink to the point M. Then,
also, for this sequence of neighborhoods the
limit value of the corresponding sequence of
integrals exists and the value of this limit
value is independent of the form of the KSn.
For the proof we proceed from the fact that
each of the regions KSn can be enclosed by
two spheres АгПі and Kirl^, whose radii enj and е*2 together with en approach
zero (Figure 57). Since the integrand is always nonnegative, we obtain

F dr F dv S: Fdv
T-K„ T-Kp
»i c»2

It follows, however, that

F dr = lim F dz = I,
J J JT-Ke
1ітШ T-Kp

since the limit value of the outer integral exists and coincides with it.
294 ELLIPTIC DIFFERENTIAL EQUATIONS

Thus we have the following results:

In the case of three independent variables, the improper integral

4dr (4-5.8)
tJ У

exists, provided a < 3, and does not exist if a ^ 3.

In general, if the critical value a which determines the convergence of

the integral of the form (4-5.8) is equal to the number of independent
variables, then in the two-dimensional case the integral

converges for a < 2 and is divergent for а Ш 2.

We shall now establish the following criterion for the convergence of
improper integrals:
To guarantee the convergence of an improper integral

F(x, y, z)dx dy dz (4-5.9)

it is sufficient that a nonnegative function F(x, y, z) exist for which the

improper integral over the region T converges and such that the inequality

I F(x, y, z) I ^ F{x, y, z) (4-5.10)

is valid.
For the proof, we consider any sequence of regions K<.n which entirely
contains the singular point M0. Because of the convergence of the sequence
{/„} of the integrals of F(x, y, z), an N(e) exists for each e > 0 such that

Fdr < e
-K-tn
n\

is valid when nx, nz > N(e). Since F(x, y, z) is a majorant of \F(x, y, z) |, we

can write

Inx In2 Fdr < IFI dr < Fdr < e

ffi
JJ -K> /Tg
nl
—iiC g
n2
К2 _ —Kg
n1 tn2
(4-5.10')

when tii, пг > N(e) holds. On the basis of the Cauchy convergence criterion,
the condition (4-5.10') is, however, sufficient for the convergence of the
sequence

/» = Fdr
-K?n
toward a limit value

I= lim I„ = (( [Fdr .
JJrJ
4-5. POTENTIAL THEORY 295

We easily see that this limit value is not dependent on the choice of the K!n.
Therefore, the existence of the improper integral (4-5.9) is completely proved.
If, on the other hand, a nonnegative function F(x, y, z) can be found
such that F(x, y, z) 2; F(x, y, z), and if the improper integral of F diverges
over the region T, then the integral of F also diverges.
Conclusion:
If for a function F(M, P) which becomes unbounded at P = M the ine¬
quality

I F(M, P) I < , a< 3,

У mp

is satisfied, then the improper integral

F(M, P)drP

converges over the region T which contains the point M.

As is known from the theory of integrals containing a parameter, the
continuity of the integrand with respect to the parameter and the independent
variables is sufficient for the continuity of the integral with respect to the
parameter. For the improper integrals considered here, however, the in¬
tegrands are not continuous functions and therefore the above criterion is
not applicable.
To find a criterion for the continuity of improper integrals which depend
on a parameter, we shall consider the improper integral

V(M) = jrF(P, M)f(P)dtP , (4.5-11)

for which F(P, M) at P — M is unbounded but with respect to M is a con¬

tinuous function, whereas f(P) is a bounded function.
An integral (4-5.11) is said to be uniformly convergent at the point M0
if for every s > 0 a 8(e) exists such that the inequality

Vtm(M)\ = F(P, M)f(P)drP < e

TS да

is valid for each point M whose distance from M0 is less than 8(e), and for
every region TS{e) which contains the point M0 and whose diameter d ^ 8(e).
We prove now that at the point M0 a uniformly convergent integral

T(M) = j F(P, M)f(P) drP

represents a continuous function. Thus we have to show that for every

e > 0, a 8(e) can be found such that
I V(M0) - V(M) I < e
is valid when

\MM0 \ < S(e) .

296 ELLIPTIC DIFFERENTIAL EQUATIONS

First, inside the region T we select a subregion Ti which contains the

point M0 (Figure 58), and split the integral into two summands

V = Fi + V2,

where the integral Vx is taken over 7\ and the integral V2 over T2 = G — 7\.
We shall consider later the magnitude over 7\. The following estimate is
valid:

I V(M0) - V(M) I ^ I VM) - Ѵг№ I + I Vx(Mo) I + I Vi(M) I .

Each summand standing on the right can be made smaller than e/3 for
sufficiently small \M0M\. Now if we choose Ti in the interior of a sphere

TltTz also is determined.

the selection
Since Mi lies in the exterior of T2, V2 is a continuous function at this
point. From this follows the existence of a d"(e/3) such that

I VM) - Vt(M) I ^ у ,

provided

Now if we set

8(e) = min [<5'(e), <5/;(e)] ,

then we obtain

I V(M) - V(M0) I ^ e for I M0M \^5 ,

whereby the continuity of a uniformly convergent integral is proved.

These results are valid not only for spatial integrals but for surface and
line integrals as well, a fact that we shall utilize in the following calculations.
We consider now a potential

V(M) (4-5.12)
rMp

and the components of the corresponding attracting force

4-5. POTENTIAL THEORY 297

Х(М) = -[\[^Р-(х-Штр
JJrJ TMP

Y(M) = \^P-(y~v)dTP - (4-5.13)

JJrJ Г.МР

Z(M)=-\\\^(z-OdTP ,
JJrJ Tup
at the points which lie in the interior of the attracting body. The improper
integrals (4-5.12) and (4-5.13) converge when the density p(M) is bounded,
p\(M)\ < C. For the potential V(M) this is obvious since

For the components of the attractive force the assertion follows from the
inequality

Ы \x - £| C_
„2 „ ^ A a=2 < 3

since I я — ? I < r.
As an illustration of the uniform convergence of an improper integral
we show that the integrals (4-5.12) and (4-5.13) are continuous functions.
Therefore, we have to prove that the integrals (4-5.12) and (4-5.13) con¬
verge uniformly at each point M0.
First,

dr p
йc
T.MP Tmp

is valid,93 where Af° is a sphere of radius о about M0 which contains the

region Td. For the calculation of the integral on the right over the sphere
К*0 we introduce suitable spherical coordinates whose origin is at the point
M. Then obviously
dzP
C Ѣc - 8Сло2
Tmp

where is the sphere of radius 2o about M. Now if s > 0 is prescribed,

then we have only to choose

о^

Thus the condition for the uniform convergence of the integral V is satisfied.
By a corresponding consideration for the integral

X(M)= \p(P)^Z±dTP,
JItJ T sip

93 The integral (4-5.12) results from integral (4-5.11) for F(M, P) = 1 /гмр, AP) = p(P)-
298 ELLIPTIC DIFFERENTIAL EQUATIONS

we obtain

dr i
p(P)^AdzP й c г» 0
dzf
<c = 8n5C < e
Ts J rip ГІР 25 rip
when

d Y d(e)
8nC ‘

Therefore the potential V and the components X(M), Y(M), and Z(M) of
the attracting force in the entire space are shown to be continuous functions.94

4. The first derivatives of spatial potentials

The integrands of the integral

X(M) = - jjj p(P)^-^dTP , Y(M) , Z(M)

are the derivatives of the integrands of

W) = ([ \№-drp .
JJrJ Гup

If the differentiation under the integral signs is carried out for the function
V(M), then
dV _ dV dV
X= (4-5.14)
dx ’ dy ’ dz
i.e., V is the potential of a field with components X, Y, and Z.
If the pointM lies outside the region T, then
a: — <?_—(x — g)_3_1_
rip [(x — f)2 + {у — T])2 + (2 С)Г2
— дх r HP
is a continuous function with respect to the two arguments M(x, y, z) and
P(£, 7], Q. Consequently, in this case differentiation under the integral sign
can be easily carried out.
The higher derivatives can be calculated by differentiation under the
integral sign at all points exterior to T. Hence it follows on the basis of
the lemma from Chapter 3, Section 2 that the potential outside the attracting
mass satisfies the Laplace differential equation

JV = 0 .
We shall now show that the derivatives of the first order of the potential
V also can be obtained by differentiation under the integral sign when the
point M considered lies inside of the region T.
94 The uniform convergence of the integrals V(M) and X(M) was shown under the
assumption that the density p is bounded (| p \ < C). Consequently, these integrals also
are continuous at the points of discontinuity of the function p, in particular on the
boundary of the region covered with a continuous distribution of mass.
4-5. POTENTIAL THEORY 299

For the proof we use the boundedness but not the continuity of the
function p(x, y, z) (I p(x, y, z) I < C). It follows that V(x, y, z) also is continuous
at the boundary points of T. These points can be considered as points of
discontinuity of the function p(P) which vanishes outside of T.
We shall now prove that for every e > 0 a 5(e) exists such that

V(x + Ax, y, z) — V(x, y, z)

X < e
Ax
when
I Ax I < 5(e) .
To this end we enclose the point M0(x, y, z) by a sufficiently small sphere
Ky°, whose magnitude will be fixed later, and split V into two summands
V — Vi + V2 ,
where V2 corresponds to integration over 7\ = Ky°, and V2 to integration
over T2 — T — Kf.\ Then
V(x + Ax, y, z) — V(x, y, z)
Ax
_ Vi(x + Ax, y, z) - VAx, y, z) + V2(x -f Ax, y, z) — V2(x, y, z)
Ax Ax
For an arbitrarily fixed 7І, we have

lim V‘ix + ?■£ ~ ѴЛх’ У- г) = X, = j jP(f, 0j-x(j)dr ,

since M0 lies outside of T2.

Now the following estimate is valid:

% _ V(x + Ax, y, z) — V(x, y, z) < x _ V2(x + Ax, y, z) - V2(x, y, z)

+ \X2\
Ax = Ax
Vi(x + Ax, y, z) — Ti(x, y, z)
+ Ax
Each of the summands occurring here can be made smaller than e/3. First
we find

Гі T r
• g / л 2it p it 2
x f j
< С sin MM-VAL = 4nC5' < ~ , (4-5.15)
l*il = P——
r dr
Jo Jo Jo r 3

since \(x — Z)jr\ ^ 1 and \p \ < C. Further, we have for the third summand

Vi(x + Ax, y, z) - V,(x, y, z)

SI =
Ax

with
4-(( Ui—i-r
Ax JJrJ \Гі
dz
m r — r1 ,
p-dz
ГГ 1

Гі = V[(x + Ax) — f]2 + (у — г])2 + (z — C)2

r = V(x- ^)2 + (y- V)2 + (2 - C)2
300 ELLIPTIC DIFFERENTIAL EQUATIONS

The sides of the triangle М0ММ, are equal to r, rir and \Ax\. Hence it
follows that

I r — Гі I ^ I Ax I ,
so that

SI ^ C

since for arbitrary numbers

Ш-ЖЗНШ'
)r,J rr1
a and b
1
cib 'S — (a2 + b2)
Li

Therefore,

= 4лд' ,

and

where Кщ' is a sphere of radius 28' about Mi.

By a suitable choice of 81 the inequality

ISI < 12*8' = 6kCo' < ~ (4-5.16)

Lj О

can be satisfied. If we choose o' according to the condition (4-5.16), then the
inequality (4-5.15) also is satisfied. Now we set 7\ = К*0 and therefore also
T2 = T — Ti.
The first equation of (4-5.14) applied to the region T2 so selected means
that for every e > 0, ад" exists such that
V2(x + Ax, y, z) - Ѵг{х, у, z) £
T2
Ax T

when \Ax\< 8".

Finally, if we choose 8 = min [5', 8"}, we obtain
V(x + Ax, y, z) — V(x, y, z)
X < £ for Jx I <8
Ax
Therefore, it is proved that

8V
X (4-5.17)
dx
The expressions

dV
and = Z
dz
can be proved similarly.
4-5. POTENTIAL THEORY 301

Therefore we have proved that by differentiation under the integral

sign the components X, Y, Z of the force field are the components of grad
V.

5. The second derivatives of spatial potentials

The improper integral

dr (4-5.18)

does not converge absolutely at an interior point P of the bounded region

T. Here the majorant of the magnitude of the integrands has the form

with a=3 .

In the following, a formula will be given which permits the calculation

of the second derivatives of the potential V(M) in the region T when the
continuity and the continuous differentiability of the density p(x, y, z) are
assumed in the neighborhood of point P. Above all, the investigation is not
applicable at a boundary point at which the density as a rule has a point of
discontinuity.
We again write the potential V(M) in the form
V — Vi + vz,
where as before the summands refer to the two regions 7\ and T2. Here
Ti — K‘s° denotes a sphere of radius S about M0 in which the density p(x, у, z)
is differentiable.
The second derivatives of V2 can be obtained by differentiation under

т-м
the integral sign since the point M0 lies outside of T2, e.g.,

дгѴг
dx2 dx
For the first derivative of Vi with respect to x we find

(4-5.19)

since

8_
dx
By applying the Green’s formula, the integral (4-5.19) is transformed
into
302 ELLIPTIC DIFFERENTIAL EQUATIONS

where Tf° is the spherical surface bounding 7\ and a denotes the angle
between the exterior normal to the surface l'*0 and the x axis. The first
summand is a differentiable function at the point M0, since M0 lies outside
of Tf°. Similarly, if the second summand is differentiable in the neighborhood
of the point M0, then p is differentiable in 7\. From this it follows that at
the point Mo the second derivative of the function Vi exists. Consequently,

cos ado + ДА'ІЙ*

dx dt
For the second summand at M0 the following estimate holds:

(4.5-20)
dx\r JdZ
provided

dp_
< Cx .
dt
If we apply the mean-value theorem to the surface integral, we obtain

-k*»S4(-7)cosaA’ = ■

Here p* is a suitable intermediate value of the density at a point of Tf0.

Further,

d_ x—6 1
3 — — cos a
dx r
and

= yj^oj4(cos2« + COS2/3 + cos2r)do = ^-K .

By passage to the limit as d—>0, we obtain

&Н^4(т)С08Н = • <4’5-21)

,
The equation

d2V d2V dzV2

dx2 ~ dx2 + dx2
is true for every 5. Since, moreover, its left side does not depend on 8, then
d2V (d2Vl , d2V2
= lim ■ + ■ _|P(M) + lim \p~r(y)dr . (4-5.22)
dx2 Ѵ-оД dx2 ' dx2

The existence of the limit value

«JLj'w)*=іігЩт)* (4-5'23)
follows from the existence of the second derivative d2V/dx2 proved above.
4-5. POTENTIAL THEORY 303

This integral was obtained by a special choice of passage to the limit, and
indeed the sphere was shrunk to the point M0.95 This is indicated in formula
(4-5.23) by the bar above the integral sign. In general, a change in the shape
of this region implies a change of the limit value; the integral (4-5.23) is
conditionally convergent (in the sense of the above definition). Accordingly,

=ЛИИт)* ■-т‘oim ■ (4'5-24)

From this we see that the calculation of the second derivatives of the
potential by formal differentiation under the integral sign would lead to an
incorrect result.
Analogous expressions hold for the derivatives дгѴ1дуг and dzVldz2. By
substitution of the three derivatives into the expression for the Laplace
operator we find

дгѴ d2V d2V

AV =
dx
2 + , 2 + dz1

dv — 4 лр(М0) (4-5.25)

— —4л-p(M0)
since 1 lr is a harmonic function.96
Therefore the spatial potential satisfies Poisson’s equation

AV = -Акр
inside the body, and the Laplace equation

AV = 0

outside the body.

The inhomogeneous equation

Au = — f (4-5.25')

possesses the particular solution

provided / is differentiable in a fixed region T.

From this it is evident that the solution of a boundary-value problem
for the inhomogeneous Eq. (4-5.25') can lead back to the solution of the cor¬
responding boundary-value problem for the Laplace equation Av = 0, in which
the sought function is put in the form и = u0 + v.
95 The limit value (4-5.23) can be regarded as the principal value of the integral.
96 Formula (4-5.25) was derived under the assumption that the function p(M) was
differentiable. This condition can be replaced by weaker conditions. However, the
continuity of p(M) would be insufficient for the validity of (4-5.25), since continuous
functions p(M) exist for which the spatial potential possesses no second derivatives.
304 ELLIPTIC DIFFERENTIAL EQUATIONS

6. Surface potentials

As the Green’s formula shows:

U(M)
= l4
4л JjJL rMp dn on \r J J

(see Section 4-2), every harmonic function can be represented by integrals

which are surface potentials.
We will now determine the potential of the field of a surface T loaded
with a mass. For the definition of the surface density ц(Р) at the point P
of I, we proceed in the following manner. Let {Tv} be a sequence of sur¬
face elements which shrink down to a fixed point P on I, i.e., the relation
lim (Tv = 0

holds provided we understand crv to be its surface area. Each Iv corresponds

to a mass layer m, which is considered as a function of ov, i.e., it can be
written in the form m = m(ov). Thus we denote the limit value

МП = lim «<£!>. =
<T)/ '0 (fy dop

as the surface density of the surface I at the point P.97 The potential of
this mass load can then be represented by the surface integral

V(M) = f {-^-dop , (4-5.26)

J^J Tup

the so-called potential of a single layer.

Another type of a surface potential represents the potential layer whose
definition we shall now develop.
We consider a dipole which is
formed by two masses — m and m
situated at the points Px and P2 and
at a distance M (Figure 59) from each
other. The product mJl = N is called
the moment of the dipole. For the
potential at an arbitrary point
M(x, y, z) we then have
m m_ / \_
= N- 1 /1
_
Гг Гі V Гг d/ \ T2

where ту and r2 are the distances of the points M from 7\ and T2, respectively.

97 If the mass with spatial density p is distributed into layers of thickness h on the
surface I and the field is considered at the points whose distances from the surface
are large in comparison with h, (h/r •€ 1)> then the consideration of the thickness
of the layer in general is unimportant. It is suitable however to consider, instead of
the spatial potential with density p, the corresponding surface potential with a surface
density /( = ph.
 4-5. POTENTIAL THEORY 305

If Al is small in comparison with the distance to the point M (АІІГі € 1),

then by use of the mean-value theorem we can write

r=

Here the vector l is along the direction from the repelling mass to the at¬
tracting mass, and r is the distance of the point M(x, y, z) from a fixed inter¬
mediate point P($, 7], Q of the interval Al.
For the derivative in the l direction we obtain

where the vector r is directed along the dipole to a fixed point M, whereas
ip is the angle between the vectors l and r. Hence, the potential of the
dipole is

(4-5.27)

where N is the moment of the dipole.

Now let two surfaces E and S' (Figure 60) which are parallel and at a
distance d from each other be loaded
with a mass in such a way that the
mass of each of the elements of E' M(xyz)
accordingly is equal and opposite to / уУ
the magnitude and the signs of the ^ JJ
corresponding element of E. By n, we \\ V' \
denote the normal common to the \\ da
surfaces I and E', which are directed \ \z У
from the repelling to the attracting ~ ]_}/ fig. 60.
masses. Now if we let 8 approach zero,
then we obtain the double layer as the totality of two infinitesimally close
v is the surface density of the
layers whose densities have opposite signs. If
moment, then the moment of the surface element dap equals
dN = vdop .
For the potential of the element da at the point M(x,y,z) we then have

—>

where <p is the angle between n and PM.

The integral

(4-5.28)

is called the potential of a double layer. This definition obviously cor¬

responds to the case in which the outer side of the surface acts so as to
repel and the inner side acts so as to attract.
306 ELLIPTIC DIFFERENTIAL EQUATIONS

If (p\ is the angle between the interior normal and the direction PM,
then

W= ( \^^-v{P)d(Xp .
JjJ r HP

If the surface is not closed then we have to regard it as two-sided, since

the potential of the double layer is defined only for such surfaces.
The potential of the single and double layers assumes, in the two-
dimensional case, the form

V = [ ii{P)\n—^—ds , (4-5.29)
Jo Tup

W=-[ *P\-T—(In—V5 = ( ■^^Ё^£Lv(P)^/s (4-5.30)

Jo drip \ Гмр / Jo Гагр

where C is an arbitrary curve, ц is the linear density of the single layer, v

is the moment density of the double layer on the curve, and is the angle
between the interior normal to the curve and the direction to the test point.
If the test point M(x,y,z) lies exterior to the surface (i.e., outside of the
attracting mass), then the integrands in the equations

do p

including their derivatives of arbitrary order are continuous functions of

x, y, z. Therefore the derivatives of the surface potential at the points which
lie outside of the surface I can be calculated by differentiation under the
y integral sign. Hence it follows on the basis
of the superposition principle that the sur¬
face potential everywhere outside the at-
с. tracting mass satisfies the Laplace differential
\ equation. The functions (4-5.29) and (4-5.30)
v Уу obviously satisfy the two-dimensional Laplace
ууУ differential equation.
p At the points on I, the surface potential
fig. 61. can be represented by an improper integral.
If the surface possesses a continuous curva¬
ture, then we can prove directly that the potential of a double layer exists
at those points of the surface where the curvature is continuous. To give
the proof for the case of two independent variables

cosy
W vds ,
-L-
we consider the curve C in the x, у plane and select as the origin of
coordinates the point P. The x axis coincides with the tangent to the curve,
the у axis with the normals at this point (Figure 61). In a fixed neighbor-
4-5. POTENTIAL THEORY 307

hood of the point P, the equation of the curve can be written in the form

У = У(х) .
Now C, by assumption, has a continuous curvature, i.e., y(x) possesses a
continuous second derivative. Consequently,

У(х) = y(0) + xy\0) + jy"($x) , 0 < 3 < 1 ,

where, because of the special choice of the coordinate axes, it follows that

y(x) = —хгу"{Ях)

Therefore, we have

/*■ + ** = У1 + ,■ ,

cosy У xу" {Эх)

2/1 + /[jqMj
and

cos у _y"(9z)_
2{1 + ^J}
Further, for the curvature,

y"( 0) =K(P)
results from the expression

кy
(1 + y'2?'* ’

Therefore,

,. cos (p 1 jr. n.
lim — = —K(P)
MP—0 r 6

This expression shows the continuity of (cos <p)lr along the segments of the
curve and therefore also the existence of the potentials of the double layer
at the points of the curve C, provided v is bounded.
In the three-dimensional case, the potential of the double layer also exists
at the points of the surface considered, since the function

cosy

has an integrable singularity of order 1/r. The existence of the potential of

a single layer presents no difficulties.
308 ELLIPTIC DIFFERENTIAL EQUATIONS

7. Liapunov surfaces and curves

The requirement that the surface 2 considered possess a finite curvature,

is considered superfluous for the existence of the corresponding surface
potential.
As is known, the potential of the single and the double layer are im¬
proper integrals at the points of I. We shall show that these integrals for
a definite class of surfaces, the so-called Liapunov surfaces, converge, if the
density of the layer is bounded, that is, if \v{P) \ < C,C = const.
A surface 2 is called a Liapunov surface if the following conditions are
satisfied:
1. 2 a well-defined normal (tangent plane) exists.
At each point of
2. A numberd > 0 exists such that the elements 2'P of the surface 2
parallel to the normals at any point P of 2 which lie within a sphere of
radius d about Pdo not intersect 2 more than once. These surface elements
2'p are called a Liapunov neighborhood.
3. The angle y{P, P’) = {nP, nP') formed by the normals at two points
P and P' satisfies the condition
r(P, P') < Ars , (4-5.31)

where r is the distance between the points P and P', A is a fixed constant,
and 0 < d ^ 1.
Now let P0 be an arbitrary point of 2. We then choose a rectangular
coordinate system whose origin is at P0 and whose z axis is in the direction
of the exterior normal. The x, у plane then coincides with the tangent
plane at this point. Because of condition (4-5.2), a p„ exists such that the
equation of the surface 2 can be written in the form98
2 = f(x, y) , (4-5.32)

provided

p — л/хг + у2 < Po. (4-5.33)

Now let 2Po be a neighborhood of the point P0 on the surface 2 deter¬

mined by the conditions (4-5.32) and (4-5.33).
From the existence of the normals at each point on the surface (Con¬
dition 1) the differentiability of f(x, y) follows. The direction cosines of the
exterior normals, therefore, are given by the formula

cos a = cos /3 = cos t =

Vl + z\ + z. V7! + z9 + vT + 2" + 2?
Because of the special choice of our coordinate system, however, zx{P0) = 0,
and Zy(P0) = 0. We assume the surface 2Po to be so small (i.e., p0 is sufficiently
small), that

98 If the function f(x, y) [in the neighborhood of the point P0 has a continuous
second derivative, then the surface satisfies the Liapunov conditions. Consequently,
surfaces with continuous curvature belong to the class of Liapunov surfaces.
4-5. POTENTIAL THEORY 309

1
1 > COS у= > ѴГ (4-5.34)
V\ + zl + zl 2

Further, we denote by n'P the projection of the vectors nP on the x, у plane

and by a , /3' the angles formed by the vectors n'P with the x axis and the
у axis, respectively. Obviously then,

cos a — sin у cos a' cos/3 = sinf sin a'.

Condition ( 3 ) now asserts that

sinr < у < ArPPo ,

hence

I cos a I < A/PPq , I cos /31 < ArPPo, (4-5.35)

and since

cos a cos В
zx --, zy —-— with < 2
cos?- cos у cos у

then

I zx I < 2ArPPo, I zy I < 2ArPPo .

If we now apply Taylor’s formula to 2 = f(x, y) in the neighborhood of
the point P0(0, 0), we obtain

z(x, y) = 2(0, 0) + xzx{x, y) + yzy(x, y) ,

where

0^ x^x, 0 ^у^у.
From this follows the estimate

I z(x, у) I < 4ArPp0 . (4-5.36)

The estimates (4-5.34) and (4-5.36) now permit the proof of the proposition
that the potential of the double layer

W(M)'=[ f(p)dap (4-5.28)

J Z J r MP
at points of the surface I represents a convergent improper integral, pro¬
vided T is a Liapunov surface. Now let M = P0 be a point on the surface
I. If we choose our coordinate system again in the above-described manner,
we can write the equation of the surface I in the neighborhood of the point
P0 in the form

2 = f(x, y) ,
where the function f{x, y) satisfies the inequalities (4-5.34) and (4-5.36).
Now if <p is the angle between the direction of the interior normal at
the point P($, 1}, 0 and the direction PP0, then we can easily obtain the
relation
310 ELLIPTIC DIFFERENTIAL EQUATIONS

£ « Г \r\
I COS (p —cos a + —cos/3 H-cos у 5S I COS a I + I COS /3 I + —
r r r r
= + Лг/>р0 T 4Лг/>/-0 = 6 Ar]>p0

and

cosy
< 6Л- 2-S о<г<l (4-5.37)

Further, we can write the integral W{M) as the sum of two integrals:
W — Wi + Wi .
Here Wi denotes the integral over the surface S'Fq containing the singular
point P0, and W2 is the integral over the remaining part of the surface,
that is, it is extended over I — Z'p0. For proof of the convergence of W it
is sufficient to show the convergence of W{, since the integrand of W2 re¬
mains finite everywhere. In polar coordinates p — i/£2 + Vz and Ѳ in the
x, у plane,
— d£dr] _ pdpdd
cos у cos у

holds, so that by a transformation of the variables we obtain

On the basis of the estimates (4-5.34), (4-5.36), and (4-5.37) and because p < r
we have for the integrands

cosy 1 12 AC
v(P) ^ F=
r2 cos у i°2~S
This form of the majorant function F, however, guarantees, in the case
of two independent variables, the convergence of the improper integrals (see
Section 4-3).
It can easily be shown that for a Liapunov surface the potential of a
single layer

V(M) = f f—p{P)daP (4-5.26)

J^J У up

also converges at points of this surface. Let it also be noted that the
potential for a more comprehensive class of surfaces converges.
In the two-dimensional case the potential of the single and double layers
converges at points of the curve considered [see formulas (4-5.29) and (4-5.30)],
provided this potential exists on a Liapunov curve. The conditions for these
curves are analogous to Conditions 1-3 for the Liapunov surfaces.

8. Discontinuities of the potential of a double layer

We shall now prove that the potential of a double layer at a point P0

4-5. POTENTIAL THEORY 311

on the surface I is a discontinuous function, for which the relations

WAP') = W(P„) + 2kv(P0)
Wa(Po) = W(PQ) - 2ttv(P„)
are valid. Here Wi{PQ) denotes the limit value of the potential of the
double layer for an approach to the point P0 from the inner side, and Wx(P)
the limit value for an approach from the exterior side of the surface."
In the case of two independent variables, the corresponding formulas
have the form
Wt (P) = W(P0) + rcv(P0) (4 5
Wa(Po) = W(P0) - яѵ(Р0) .
For simplicity, we shall limit ourselves to the proof of these formulas for
two independent variables.
The potential of a double layer is represented in this case by the integral

W(M) = -^^^(P)Jsp .
JO fMP

We begin with an arbitrary element of arc ds, whose end points are P
and Pi. Through the point P we construct a circular arc of radius MP about
M. This cuts the ray MP at the point Q. Then we can write to within
terms of higher order

ds cos <p = da , = da) (4-5.40)

r

(Figure 62), where ds = PP and da — PQ, while da> denotes the angle by which
one observes the arc ds from M. The sign of da) coincides with the sign of
cosy>. Consequently, day > 0, provided y> (the angle between the interior
normal at the point P and the vector PM) is less than ?r/2, and day < 0, pro¬
vided <p > 7t/2. If day > 0, i.e., <p < к/2, then the point M is seen from the
“interior” side, and if day < 0 (y> > тг/2), by contrast, it is seen from the
“exterior” side of the curve C. From this it follows that the angle by which
one sees an arc PyP2, from the point M equals the angle PiMP2, which de¬
scribes the ray MP, when the point traverses the arc PP2 •
We now consider the potential W° of a double layer with constant
density v = v0 = const. The ray MP describes the angle
[ 2л: providing the point M lies in the interior of curve C
& = j 7Г providing the point M lies on the curve C
1 0 providing the point M lies in the exterior of curve C
when the point P traverses the entire curve C. From this we find for the
potential W°:

99 If I is an unclosed surface, we have to decide which normal shall be designated

at the point Po as “interior” and which as “exterior”; this choice is completely arbi¬
trary. We have only to observe that for unclosed surfaces the potential of the double
layer is defined only for two-sided surfaces.
312 ELLIPTIC DIFFERENTIAL EQUATIONS

( 2kv0 providing the point M lies in the interior of curve C

W° = Qv 0= nv0 providing the point M lies on the curve C
1 0 providing the point M lies in the exterior of C.

Therefore, the potential with constant density is a piecewise constant func¬

tion and is given by
W° = W°0 + itva
(4-5.41)
W°A = Wno-тгѵ„
when W°, W'c, W°A are the values of potentials in the interior, on the curve,
and on the exterior of the curve C, respectively.
Correspondingly, for three independent variables:

da cos <p , ,, r
-5—— = dw , (4-5.42)
r
where dw is the solid angle subtended by the element
da of the surface 2. Now let da' be an element of
the spherical surface which we obtain when we let
a cone whose vertex is at point M and whose base
is the surface element da intersect the sphere of
radius MP about M. We then obtain for the spheri¬
cal surface element
da' = da cos <p .
From this formula (4-5.42) also follows. The above
observations with regard to the signs of dw also remain valid here. This
leads to the relations.
[ 4л-ц, provided M lies inside of 2
W° = vaQ — j 2nv0 provided M lies on S
1 0 provided M lies outside of 2

by which the piecewise constant function W° is determined. Further, we

arrive at the formulas

Wj = Ws + 2nv0 and W°A = W°s - 2nv0 (4-5.41')

in which W°, W°A denote the values of the potential W° in the interior and
in the exterior of the surface 2, while W\ is the value of the potential on 2.
We proceed now to the consideration of the potential of a double layer
with a variable density and show that at the points of continuity of the
density, relations exist which correspond to formulas (4-5.41) and (4-5.41').
Let P0 be a point of the surface 2 at which the function v(P) is con¬
tinuous. Further, let W° be the potential of the double layer with constant
density v0 = v(P0). Then the function

KM) = W(M) - W\M) = \ [v(P) - Vo]^LdoP ,

J^J T HP

as we can prove immediately, is continuous at the point P0. Therefore, it

is sufficient to prove the uniform convergence of the integrals I(M) at the
4-5. POTENTIAL THEORY 313

point P0. Let us prescribe an arbitrary number e > 0. From the continuity
of v(P) at the point P0 it follows that for an arbitrary prescribed r] > 0 a
neighborhood Ex exists of P0 on the surface I, such that

I v{P) — v(P0) I < Г]

when P is in Ei. Now we write the integral I(M) in the form
/=/, + /,,
where Іл is extended over Ei and I2 over E2 = E — EFrom the definition
of Ei it then follows that

I h I < f]Bs ,

where B* is a constant which for all M satisfies the inequality

(4-5.43)

but which does not depend on the choice of the surface Ex. More details
will be given later about this constant.
If we now select у =
e/Bs, we know that for every e > 0 there actually
exists a Ei containing a P0 such that for each point M we have

I h(M) \<e.
From this, however, follows the uniform convergence of the integral I(M)
at the point P0 and therefore also the continuity at this point.
If W,(P0) and Wa{P0) are the limit values of the potentials Ws as M—
on T+ and E~, respectively (E+ and E~ denote the interior and the exterior
sides of E), then

Therefore, the validity of (4-5.43) is proved.

The above proof holds for surfaces which satisfy the boundedness con¬
dition (4-5.43). For convex surfaces in which each ray from the point M
cuts the surface twice at most, we find Si ^ 8л-; for surfaces composed of
a finite number of convex pieces, Вs likewise is bounded. Therefore the
proof given here holds for a very wide class of surfaces.
These calculations also remain valid for functions of two independent
variables, except that here formula (4-5.41) has the form

Wz{Po) = W(P0) + л v(P0)

wA(Pa) = w(P0) - ыРо).

9. Properties of the potential of a single layer

In contrast to the potential of the double layer, the potential of a single

layer
314 ELLIPTIC DIFFERENTIAL EQUATIONS

V(M) = ( [-^—{i{P)daP (4-5.26)

JlJ f/np
is continuous at points of the surface 2. For the proof it is sufficient to
show the uniform convergence of the integrals V(M) at points of the surface
V

If P0 is an arbitrary point of the surface 2, we can write the potential

V{M) in the form

V= f f—p(P)daP + ( \—ii{P)daP = Vi + V2 ,
Гнр JX2J r»IP

where 2t is a sufficiently small part of 2 which is contained in a sphere of

radius 8 about the point P. We shall consider the magnitude of 8 later.
We consider now a coordinate system whose origin is at the point P0
and whose z axis is in the direction of the exterior normal at P0. Let M(x, y, z)
be an arbitrary point whose distance from P0(0, 0, 0) is equal to MP0 < 8.
Further, let 2[ be the projection of 2\ on the x, у plane and K2s a circle of
radius 28 about M'(x, y, z) which lies entirely in 2[. If we assume the
boundedness of the density p(P)

I /л{Р) I < A
and note that

da — — d£drj
cos у cos у
and

r =T/{x — + {y — r]Y + (z — O2 ^ V(x — f)2 + (y — rjY = p ,

we obtain the estimate

VAM) < A <h_=A[ t 1

ГМР Jx; J V(x - £)2 + (y - 1])* + (z- C)2 cosr

< 2A • didy <2a[ [ dr}

) V{x — £)2 + (y — j?)2 ~~ Jk^'j VTx — f)2 + (y— rjY

when 8 is so small that cos;- > 1/2.

Now we introduce the polar coordinates p and <p in the x, у plane with
origin at M'. Then we can write

_ dz dr; _
a„s .
VAM) < 2A
V(x — f)2 + (y — j?)2 0 p
For 8 = е/8л-A,
VAM) < e ,
when MPa < 8. Consequently, at each point P0 on 2, V(M) converges uniform¬
ly and therefore is a continuous function.
On the other hand, however, the normal derivatives of a single layer
have discontinuities on 2 of the same kind as the potential of a double layer.
4-5. POTENTIAL THEORY 315

The exterior and interior normal derivatives of V, that is, dV/dnA and
dV/dni, are defined as follows. Let P0 be an arbitrary point on 2. Through
Po we draw the z axis, which then can have the direction of either the ex¬
terior or interior normal.
We now consider the derivative dV/dz at a
point on the z axis. We denote by (dV/dz)r and
(dV/dz)A the limit values of the derivative dV/dz
as the point M approaches the point P0 from 2+
or 2~, respectively. If the z axis has the direction
of the exterior (interior) normal, then this value
is called the inner (outer) limit value of the deriva¬
tive with respect to the exterior (interior) normal
at the point P0.100 FIG. 63.

Now we investigate the discontinuity of the

2. The deriva¬
inner normal derivative of the potential of the single layer on
tive dV/dz at a point M of the z axis directed along the interior normal is
equal to
dV(M)
(4-5.44)
dz
where ф is the angle between the 2 axis and the vector MP. Now we draw
through P (Figure 63) the normal PQ and the straight line PN parallel to
the 2 axis (the normals at P0). We denote by Ѳ the angle NPQ. This coin¬
cides with the angle between the normals at the points P and P0*101
The expression for the potential of the double layer W(M) contains the
factor cosyi/r2 where <p = < MPQ. Since the angle MPN equals л — ф,

cos(7r — ф) = cosy> cos<9 + sin<p sin# cosf2 = — cos^ ,

where Q is an angle defined by the surface and PQ.102 From this it follows
that
dV(M) f f. cos<p , f f . „ n siny> ,
—г-= — l Uucostf)-ir—do — \ \/*sin#cosf2-r~da
dz JjJ r JxJ r
= -Wl-I(M) (4-5.45)
where W^M) is the potential of the double layer with density jtq = ficos#.
100 The limit value of the difference quotient (V(M) — V(Po))IMPo as M -* Po is equal
to the existing limit value of the derivative with respect to the exterior normal ap¬
proached along the exterior normal, or equal to the limit value of the derivative with
respect to the interior normal approached along the interior normal, according to the
manner in which M approaches the point P, when these derivatives are continuous
along the normal and on 2.
101 Obviously, as P-> Po, Ѳ and sin Ѳ approach 0. If the surface in the neighborhood
of Po has finite curvature, i.e., if its equation can be written in the form z — f{x, y),
where /(x, у) is twice differentiable, then sin# is a function differentiable with re¬
spect to x and y; consequently, sin# < Ar (for Liapunov surfaces sin<9 < Ars).
102 If the PQ direction is chosen as the axis of a new polar coordinate system,
then this formula coincides with formula (4-5.13). For the geometrical significance of
this angle, see footnote 88.
316 ELLIPTIC DIFFERENTIAL EQUATIONS

Wi{M) has discontinuities on I. If the integral I(M) is a continuous func¬

tion at P0, then 7(M) converges uniformly at this point (see footnote 101).
By consideration of formula (4-5.45) we obtain at the points of discon¬
tinuity of Ц

Wy{P0) — 2тг/*і(Р0) — I(Po)

Wi{PQ) + 2kMPo) — I(Po) • (4-5.46)

Now we set

- - wm - KPo)
= £— J ^(/z cosѲ) C0S2^-~do — I j^sinffcos#

ГР0Р

where <p0 is the angle between the z axis and the vector P0P. If we note
that MPo) = t*(Po), then
fdV
— 2tt/z(P0)
\drii

+ ^Kfi(Po) (4-5.47)
\ОПі Jo
results, since by convention the 2 axis has the direction of the interior
normal. If the z axis is in the direction of the exterior normal, the sign of
cos<p changes and we obtain

2 лyz(P0 (4-5.48)
\dnAJA \dnAJ0
In the two-dimensional cases there are corresponding formulas; we have
only to replace 2л: by л.

10. Application of the surface potential to the solution of

boundary-value problems

The surface potential is a convenient analytical tool in the solution of

boundary-value problems. We consider the interior boundary-value problem
for a curve C.
Determine the function u, which is harmonic in the region T, bounded by
the curve C, and on C satisfies one of the boundary conditions
first boundary-value problem: и Ic =f
du
second boundary-value problem: =f
dn c
4-5. POTENTIAL THEORY 317

The exterior boundary-value problem reads correspondingly.103

We write the solution of the first interior boundary-value problem as
the potential of a double layer

W(M) = { -^v(P)dsP = - ( -4—(In——•

Jo r\tp Jodnp\ Гцp )

For an arbitrary choice of v(P) the function W(M) satisfies the Laplace
differential equation interior to C. On C, W{M) is discontinuous. There¬
fore if the boundary condition is to be fulfilled, we obviously must have

W,(Po) = f(Po) •
By using formula (4-5.41) we obtain the equation

nv(P0) + f S2^Lv(P)dsP = f(P0) (4-5.49)

Jo гРйР
for the determination of v(P). If s0 and s are the values of the length of
the arcs on C corresponding to the points P0 and P, then Eq. (4-5.49) can
also be written in the form

mv(So) + j K(s0, s)v(s)ds = f(s0) , (4-5.50)

where L is the length of the curve C and

K(s0 ,s) =-/Ціи-і-) = (4-5.51)

drip \ rppj rpp0

is the kernel of the integral equation. Here we are dealing with a Fredholm
integral equation of the second kind.104 For the exterior problem we obtain
the corresponding equation

- rrv(So) + J K(s0, s)v(s)ds = /(s0) . (4-5.52)

For the second boundary-value problem we find the equations

interior boundary-value problem: — nfi(s0) + J Ki(s0, s)/u(s)ds — f(s0) (4-5.53)

exterior boundary-value problem: Kfi(So) -F l Kt(So, s)fji(s)ds — /(So) (4-5.54)

I 0

юз por the formulation of the second boundary-value problem, it makes no difference

whether in treating the interior or the exterior problem, the inner normal is always
selected as the normal in the boundary condition.
104 Linear integral equations with fixed limits of integration are called Fredholm
equations:
f*
first type: \ K(x, s)<p{8)ds = f(x)
Ja
rb
second type: <p(x) + \ K{x, s)<p{8)ds = /(x) .
318 ELLIPTIC DIFFERENTIAL EQUATIONS

where

COS фо
к list, s) ln-M = (4-5.55)
dn, r pp0) Гррп

when its solution is written as the potential of a simple layer105

1
u(M) In-/Lt(P)dsi
-I Г UP

The questions which relate to the solvability of these equations will be

P(s) discussed in Section 4-5 § 11.
In the following we shall treat the boundary-value
problem for a simply connected region for which the cor¬
responding integral equations are easily solvable.
1. The first boundary-value problem for the circle. If the
curve C is a circle of radius R, then the interior normal
at the point P is in the direction of the diameter and is
FIG. 64.
cosy 1
2R

since (p is the angle of P0PP' (Figure 64). The integral equation for the
function v(s) therefore assumes the form

y(s0) + — f -^5-v(s) ds = —/(s0) . (4-5.56)

it J о £R л

The solution of this equation is the function

1
v{s) - —f(s) + A, (4-5.57)
7Г

where A is a constant yet to be determined. If we put this expression for

the solution in (4-5.56), then

-/(so) + A + -f -^(-f(s) + A)ds=-f(s0) ,

from which we obtain for A the expression

A = -ih\fs)ds-
Therefore

v(s) = -f(s) -
7Г
— 4 —

R Jo
f f(s)ds (4-5.58)

is the solution of the integral equation.

For the corresponding potential of the double layer we obtain

юг we easily see that K(so, s) = Ki(sf so). Such kernels are called adjoint and the
corresponding equations, adjoint integral equations.
4-5. POTENTIAL THEORY 319

wm = (^*мрхі*г = (-^Гі/м f(s)ds>^J ds

J 0 r MP Jo r \_n 4-k R
If M lies interior to the curve C, then the right side of the last formula is
transformed as follows

1
W = -[ ———^—f(s)ds — f(s)ds) ^ds
Л Jo r 4 k2R /Jo r
1
■f(s)ds f(s)ds) • 2л (4-5.59)
л J0
=1( r 4 n2R

-7І(*1*-к>*
Now we see from AOPM (Figure 65) that
COS (p _l_27? cos <p — r _ 2Rr cos <p — r
К
r 2R ~ 2 Rr ~ 2 Rr2
r>2 2
R — pa (4-5.60)
2R[R2 + pi - 2R/>0cos(0 - Ѳ0)] '

Then

pi = R2 + r* — 2Rr cos<p .
Finally, if we substitute expression (4-5.60) for К into (4-5.59), we obtain
Poisson’s integral

1 (R2 — pl)f(0)d0
U = W(p0,0a) = (4-5.61)
2л R + po — 2Rp0 cos {0 — 0 o)
as the solution of the first boundary-value problem for the circle.
These considerations show that for every continuous p
function f(0), a harmonic function is defined by formula
(4-5.61) which continuously approaches the boundary value
№.
If / is only piecewise continuous, then the function
W, because of the properties of the potentials of the
double layer, also is everywhere continuous where / is
continuous. Further, from the boundedness of /,

I/I <C,

follows the boundedness of the function W from (4-5.61):

Г
2
1 /?2
W(p0, Oo) I < C;
(>n
■do = C ;
2л-Jo R + f>o — 2Rp0 cos(0 — 0o)
then1
r>2 2
R — po ■dO = 1 (4-5.62)
2it)o R + pi — 2Rp0cos{0 — 0o)
106 Equation (4-5.62) results from the fact that the left side represents the solution
of the first boundary-value problem for /=1 (see page 281).
320 ELLIPTIC DIFFERENTIAL EQUATIONS

2. The first boundary-value problem for the half space. Let us determine a
harmonic function which is everywhere continuous in the half space z is 0,
assumes a prescribed value f(x, v) on the boundary 2 = 0, and at infinity is
equal to zero.
We write the solution of this boundary-value problem as the potential of
a double layer:

W{x, y, 2) = j ^ y)d$ dr], гг — (x — £)2 + (y — 1?)2 + 22 .

In the present case,

cosy
Г2 Г3 [(x — f)2 + (у — 7])г + 22]3/z
and for the kernel of the integral equation we find

1 ( cosy \ _ 0 .
2Д r / 2= 0

Therefore, the density of the potential of the double layer is given by

v(P) = ±f(P) ,

and the function sought is equal to

2)=2iXli[(,-f)- + (,T^ + /]■//«• vWd4 .

It can now easily be shown that u(x, y, 2) tends to zero uniformly as r —
-[/x2 + у2 + 22 —> 0 when / has this property also.

11. Boundary-value problems and their equivalent integral

equations

In the solution of boundary-value problems for the Laplace differential

equation with the help of the potentials of the single and double layer we
arrived at Fredholm integral equations of the second type (4-5.50).
The conditions for solvability of the Fredholm integral equation of the
second kind with continuous kernel and bounded (integrable) right side are
similar to the solvability conditions for systems of linear algebraic equations
(to which the integral equation leads if the integral is replaced by a correspond¬
ing integral sum).
The first Fredholm theorem reads as follows:
Theorem. An inhomogeneous Fredholm integral equation of the second kind
has one and only one solution if the corresponding homogeneous equation
has only the zero solution.
The Fredholm theorem is directly applicable to curves of bounded curva¬
ture, since here the kernel of the integral Eq. (4-5.50) is continuous.
The Fredholm theory, however, is also applicable if one of the iterated
kernels
4-5. POTENTIAL THEORY 321

Кы+і\Р,,Рг) = j M) Kin)(M, Р2) da и

Ка)(Р, М) = К(Р, М)
is continuous.

Theorem. We now prove the theorem: If I is a Liapunov surface, then

the iterated kernels of the corresponding integral equation are continuous
from a certain index on. As we have seen earlier, for a Liapunov surface
we have

The iterated kernels can be written in the form

Ku г(Рі , Рг) = ^ ^К^МШМ, P2)doM .

If

I Ki I < (n = PiM, ai>0, i = l,2),

then
c
\Kl>2\ < г-Лі~аг for tt! + a2 < 2 , r = P,P2 .

Obviously, it is sufficient to prove this estimate for the case in which the
point P2 lies in the Liapunov neighborhood 1 „ of the point Px. Therefore,
instead of the integrals over , we can consider the integral over the pro¬
jection So of this neighborhood on the tangent plane at the point P!, since

^ В> 0
r(P,M)
(where p(P, M) is the distance between the pro¬
jections of P and M on the tangent plane, В is a
certain constant), and since the relation do = dSI cosy
exists between the surface element do and its pro¬
jection dS; then according to formula (4-5.34),
cos у > 1/2.
The following lemma holds for a plane region:
If
Ci
Ki I <
r]~ai
then
C
ІЛ ( \Kl(P1,M)Kl(M,Pt)dxdy ^ r2-“l-«2 *
Js0J

Let R be the diameter of the region S0. Then we split the integral I
into the integral Л, over a circle Gi of radius 2r about P,, and /2, over the
322 ELLIPTIC DIFFERENTIAL EQUATIONS

remaining region G2 (Figure 66). Now for a point M lying in G2 we find

2^S:J- {г,йгг + г^2гг, г2йг, + г йг1+^ = ^-^ .

There results for /2 the estimate

C,
2ir ГR 2-Л1-Л2 * аі + а2 < 2
1
I h I < 4СіС2 Oby—Cb 2
ridr{d(p < •
0 J 2r A
С /^л1+а>2-2 «і + а2 > 2 .

Further, if we introduce in P the substitution

x = rx'
у — ry'
then we obtain

1 ( C]C2
/il < -®l-*2 rn-a, rn-^dx'dy'

In the last integral, extended over the circle G[ of radius 2r, r[ is the dis¬
tance from the center and A is the distance from the bisection point of
the radius. Consequently, this integral converges, i.e., it does not depend
on r.
From this it follows that
C,
l/il < yi-a\-<t2

Now if we set C3 + C4 = C, then we obtain the sought inequalities:

C
2-«і-оіг «1 + o-i < 2
/I <
C 2 «1 + «2 > 2 .

Accordingly, from a certain index on, the integral, which can be represented
by the iterated kernel, is bounded and converges uniformly, i.e., they are
continuous functions of their arguments.
With the use of the first Fredholm theorem we shall now prove that
the integral Eq. (4-5.50) has exactly one solution.
We limit ourselves here to the consideration of convex curves which
contain no straight line elements. In this case, the kernel of Eq. (4-5.50) is
not negative, since
K(P0, P)dsP — dco ,
where dw is the angle formed by the arc dsP as seen from the point P0.
We consider the first interior boundary-value problem. The homogene¬
ous integral equation corresponding to Eq. (4-5.50) has the form
4-5. POTENTIAL THEORY 323

As we have already seen (Section 4-5 §8), we have the relation

j K(s0, s) ds = n .

With the help of this expression we can write the homogeneous Eq. (4-5.63)
in the form

J Ms0) + v(s)] K{s0, s) ds — 0 . (4-5.64)

Now let P*(s*) be a point of the curve C at which the function | v(s) |
assumes its maximum. Then the sum v(s*) + v(s) has a constant sign. If
now in (4-5.64) we set s0 = s* and note that K(s0, s) = 0, we obtain

v(s*) + v(s) = 0 or v{s) = — v(s*) .

This result, however, contradicts the continuity at the point s*, if only
v(s*) Ф 0.
Consequently, the homogeneous Eq. (4-5.63) possesses only the zero solution
so that the inhomogeneous equation for an arbitrary /(5) has exactly one solu¬
tion.107
The exterior second boundary-value problem leads, as we have already
seen (Section 4-5 § 10), to the integral equation

^(s0)+I Ki(s0, s)/n(s)ds = /(s0) , (4-5.54)

whose kernel Ky (s0, s) is adjoint to the kernel К (s0, s), i.e., (s0, s) = K(s, s0)
now holds.
We shall now consider the second Fredholm theorem, which reads:
Theorem. The number of linearly independent solutions of a homogeneous
Fredholm integral equation is equal to the number of linearly independent
solutions of the adjoint equation.
According to this theorem, the solution of Eq. (4-5.54) is uniquely defined.
The exterior first boundary-value problem corresponds to the integral equation

—nv{s0) + [ K(s0, s)v(s)ds = /(s0) . (4-5.52)

Jo
The homogeneous equation (/ = 0) can, on the basis of the above considerations,
be reduced to the form

[i4s0) — v(s)] К(s0, s) ds = 0 . (4-5.65)

0

If So* is a point at which |v(s)| assumes its maximum, then

v(s*) = v(s)

results from (4-5.61). Therefore

107 If straight line boundary elements are present, then the considerations are some¬
what complicated but present no special difficulties.
324 ELLIPTIC DIFFERENTIAL EQUATIONS

v(s) = const. = V0

is the single solution of the homogeneous equation. On the basis of the second
Fredholm theorem, therefore, the adjoint homogeneous equation has exactly one
solution.
The third Fredholm theorem reads:
Theorem. If a homogeneous integral equation

<p(x) = J K(x, s)<p(s)ds

possesses k linearly independent solutions, <pi(x) (i — 1,2, • • •, k) , then the in¬

homogeneous adjoint equation

Ф(х) = J K(s, x)<p (s) ds + f (x)

has solutions when

j f(x)<pi{x) dx = 0 .

If we apply the third Fredholm theorem to Eq. (4-5.53), which corresponds

to the interior second boundary-value problem, then we obtain

^ f(s)ds = 0 (4-5.66)

as a solvability condition for this problem. We have already encountered this

condition in Section 4-5 § 1.
The solvability condition for the exterior first boundary-value problem has
the form

^ f(s)h(s) ds = 0 , (4-5.67)

where h(s) is the solution of the homogeneous problem corresponding to (4-5.53).

We shall discuss briefly the physical significance of this function.
Let a cylindrical conductor, with cross section S, be charged up to a fixed
potential Vo. The total charge on the conductor is found on its surface; h (s) is
the density of the surface charges. The potential produced by the surface
charges is then the potential of a simple layer with the density h (s) and can be
represented by formula (4-5.29). The normal derivative for an approach from
inside the conductor is equal to 0, since there V = const. Therefore h (s) satis¬
fies the homogeneous Eq. (4-5.53) and is proportional to the above-introduced
function h(s) , whereby the physical significance of this function is made
clear.
In conclusion, we can state that the integral equations to which the bounda¬
ry-value problems considered here lead are always solvable for the interior first
and exterior second boundary-value problems, whereas the interior second and
exterior first boundary-value problems must satisfy the conditions (4-5.66) and
4-6. THE DIFFERENCE METHOD 325

(4-5.67). We shall not investigate the questions regarding the solvability

of the remaining boundary-value problems.108

4-6. THE DIFFERENCE METHOD

1. The difference method for the Laplace differential equation

In cases in which no analytical expression for the solution can be given,

then besides considering the numerical methods for the solution of the corre¬
sponding integral equations, we can also go back to difference equations which
arise from the differential equation.
For a function of a single variable у = f(x), instead of the second derivative
we can use the difference quotient

£f
h2 4[ A+f
h
A~fl
h \
[f(x+h)-f(x)\-[f(x)-f{x-h)]
h2
f{x + h) +f{x—h)—2f(x)
(4-6.1)
h2

Here A+f = f(x+h) — f(x) is the first right-sided and A~f — f(x)—f(x—h) is the
first left-sided difference. In the case of two independent variables, the second
differences read

AIzu = u{x + h, у) + u(x—h , y) — 2u{x, y)

A2yyu = u{x, y+h) + u(x, y—h) — 2u(x, y) .

The difference quotient for the Laplace differential expression obviously

then has the form

Ahu u(x+h , y) + u(x,y+h) + u(x—h , y) + u(x, y—h) — 4u(x, y)

(4-6.2)
h h2

The transition from a differential equation to a difference equation corre¬

sponds to the transition from a continuous argument to a discrete argument.
We take now an arbitrary positive num¬
ber h and construct a network in the x, у plane
consisting of two systems of straight lines at
right angles to each other, the lines of which
are at a distance h from each other (Figure 67),
and consider only the values of the function
at the net points.
In the following we shall treat the first
boundary-value problem for the Laplace dif¬
ferential equation in a region S, which is FIG. 67.

bounded by the curve C. On C let a bounded

continuous function f be prescribed.

108 For further details, see, for example, I. G. Petrovski, Lectures on Partial Dif¬
ferential Equations, Interscience, 1954.
326 ELLIPTIC DIFFERENTIAL EQUATIONS

For the difference method, the given region S is replaced by a network

region Sh which approximates S. We assume that the region Sk consists of
those squares of our network which lie entirely in S (one could also have
chosen as the approximating region Sh the totality of all the squares of the
network which have at least one point in common with S). Let Ch be the
straight line segments bounding the region Sa. Obviously the distance of
each corner point of Ch from the curve C does not exceed h~\/~2 .
Now at the corner points of Ch we define a boundary function fh, which we
take at each of these corner points to be equal to the value of the function / at
the nearest point of the boundary C (or at one or more of the closest neighboring
points if there are several).109
The first boundary-value problem for the difference equation corresponding
to the Laplace differential equation can then be formulated as follows:
Determine a function which at the net points Mik(Xi,yk) interior to the
region Sh satisfies the Laplace difference equation
./HI _i_ ,/H) i ./fc> i ,,(Н) л„/Н> a
Ui-и, fc+ Щ, i+i+ М(-1, k + Mi, k-\—4мі,* — U , (4-6.3)
where
Ui]k = u{Xi , yk)

and assumes the value fh on Ch.

For the solution of this problem we must determine the value of the net
function Uihl at the interior net points Mik of the region Sh, and at each interior
net point Mik the difference Eq. (4-6.3) must be satisfied. Thus we obtain for
the determination of the net function a system of algebraic equations of the
first order whose number equals the number of the unknowns.
The system of difference equations possesses exactly one solution.
To prove this proposition it is sufficient to show that the corresponding
homogeneous system possesses only the trivial solution.
The system of difference equations is a system of inhomogeneous equations,
since the values Ui,k at the corner points of Ch are prescribed and are equal to the
values of fh. The transition to the homogeneous system is therefore equivalent
to the boundary function being everywhere equal to zero, fh = 0 . We shall
show that in this case the solution of the difference system at all the net points
of the region Sh is equal to zero.
Suppose a value Ui,k Ф 0 exists; without loss of generality we can assume

Mq.jq > 0 .

Further, let uio, *0 be the maximum of our net function, so that

Mj, k ^ MtQ, 4q

at all points of Mik in S*. However, the equation

_ Mi0_i, ic0 + M<0.*0 1 + Mi0+1, t0 + Mi0, *0-m

MlQ, Icq - л

109 The degree of arbitrariness in the selection can, because of the continuity of /,
be made arbitrary small, when h is chosen sufficiently small.
4-6. THE DIFFERENCE METHOD 327

can hold only when

Mi0-1. k0 Ui0, k0-1 ^t'o+1. *о = «<0. fco+1 — «<o. l0 •

Proceding in this way for иіо+1,ко, uio+2,k0, respectively, we finally arrive at a

boundary point and thus obtain a contradiction since, by assumption, the bounda¬
ry values are equal to zero. Also, the assumption wio, k0 < 0 leads to a contra¬
diction.
From this it follows that

и{, к = 0

at all interior points of the region S, i.e., the homogeneous difference system has
only the trivial solution. Simultaneously, the uniqueness of the solution of
the first boundary-value problem for the difference equations corresponding to
the Laplace differential equation is proved.
If we solve the difference equation, the net function results as an approxi¬
mate solution of the original problem for the Laplace differential equation.

2. The method of successive approximation for the solution of

difference equations

The difference method in the present case consists of finding, instead of the
solution of a boundary-value problem for the Laplace differential equation Au —
0 , the solution of the boundary-value problem for the corresponding difference
equation. For the basis of this method one has to prove that for sufficiently
small h, the function Uh differs arbitrarily little from u, the exact solution of the
equation Au = 0. We will not go into the proof of this here.110
Instead, we shall consider in more detail the methods for the solution of dif¬
ference equations. The solution of the boundary-value problem by means of
the difference method leads to the solution of a system of algebraic equations
with many unknowns, of which there can be hundreds and even thousands.
The solution of such systems using the methods of determinant theory gives rise
to severe technical computational difficulties. On the other hand, the method
of successive approximations is essentially more suitable.
For systems of linear algebraic equations the method of successive approxi¬
mations proceeds as follows.
We write the existing system of equations in the form

Mi — f\ — {al2u2 H-
u2 — f2 — (auUi -Valnun) (4-6.4)

Mn — jn \U\~\~ * * * Tйв, n— lMn—l) .

First we choose numbers u\,u\,--%, unn as the zero-th approximation and sub¬
stitute these in the right side of the Eq. (4-6.4); then we obtain the first approx¬
imation Ui1,-• •, u(nl). We continue this process. The (k + l)-th approxi-

110 See I. G. Petrovski, op. cit., fn. 109.

328 ELLIPTIC DIFFERENTIAL EQUATIONS

mation then can be calculated from the formulas

«'ifc+1) = A — (апи'гк) H-\-a,nu[nk))

Ma+D = f2— (anu[k) H-

mu+d — fn — (anlu[k) H-|-o„, ! и;_і) .

Now if the successive approximations {u[nk)} always converge towards a limit

value
lim u{nk) = un,
k-ao

then these limit values are the sought solutions of the systems (4-6.4).
The method of successive approximations is not applicable to every system
of equations. However, it can be used to solve the first boundary-value problem
for the system of difference equations Jhu — 0. The system of algebraic
equations which correspond to the equations Jhu = 0, has the form
(A) , .. (A) , (A) , (A)
(A)_Ui +1 к ■+ Ui k + 1 + Ui-i, к "Г Ui, к-1
,
Ui,k-- ,
4

where each equation is solved for the corresponding unknowns.

We begin the successive approximation on the boundary on which the
boundary values are prescribed. By С^> we denote the totality of the net points
of the region Sh, which are at a distance h from the corner points of the bounda¬
ry Ch. To carry out the successive approximations at the points on CJ,1’, the
values of fh given on Ch will be used directly. Further, we denote by C*2) the
totality of the net points of Sh which are at a distance h from CS,l). The succes¬
sive approximations of 1 are carried out using the values u[hl on Corre¬
spondingly, we define the “zones” С'Д C;4’,---. Then the last of each of the
net points in Sh belongs to one of the zones Ci{), (i = 1,2,* • *, N) . N is the
number of the last zone СУТ
Now let Ui, к be the exact solution of the system of difference equations and
«<!* the nth approximation of this system. Then the difference

on Ch is equal to zero, whereas in the interior of Sh the equations

(я—i) I (n—1) , (n—1) , (n — 1)
_ Vi + x.k + Vi, fc+1 + Pi-1, к + Vi, к-1
ѵі, к — -:- (4-6.5)

hold.
We prove that the difference sequence {Vi*l} approaches zero:

lim vl*I = 0 .

Thus, we set max v\*l = An and estimate the (n + l)-th approximation, Obvi-
ously,

(»+l) о .
Vi,k Si -r An on Cll)
4
4-6. THE DIFFERENCE METHOD 329

holds, since in this case at least one of the summands in (4-6.5) is equal to zero.
From this it follows that

v\T] Ѣ (l on Ci2\

and, in general,

i4n*+1)^ (l-^r)^n on Ci‘\

For the last zone Ci‘v> we thus obtain

on Oh •

Therefore,

An+i = ex Ant
a ^ 4v ’

i.e., lim An — 0 .
n—*oo

If now we set min v^l = Bn, we can similarly obtain

lim Bn = 0 .
71—*°°

Consequently,

lim ѵ[Ук = 0 or lim и(*1 = Ui, к

n—* OO 71 —*°°

for all points Mi,k(Xi, yk) in Sk.

The convergence proved here is valid for an arbitrary choice of the zero-th
approximation. However, the degree of convergence (speed of convergence,
etc.) depends strongly on the choice of the zero-th approximation.

3. Electronic integrators

In recent times several mathematical machines have been used for the
solution of systems of difference equations.
M2
The construction of these machines is based
on the analogies that exist between different Ѵг

physical processes which are described by

one and the same differential equation.
For the solution of the Laplace differen¬ Mn
R
tial equation (and also certain complicated мх -лМЛЛЛ- -АЛАДѴ-

equations) electronic integrators are often

used. To investigate one of the simplest
electrical systems for the solution of the
Laplace differential equation (which was con¬ FIG. 68.

sidered by Gerschgorin), we consider a net¬

work of equal ohmic resistances. One of its components is shown in Figure 68.
Let Vi be the potential at the point Mit and the current in М0М<.
 330 ELLIPTIC DIFFERENTIAL EQUATIONS

From Ohm’s law

i= 1,2,3,4

and Kirchhoff’s law

П + ji + ja + jt - 0
there results
F.+ T2 + Т3 + V<
4

Therefore the potential at an arbitrary net point of a compound electrical

system of resistances is equal to the arithmetic mean of the potential at the four
neighboring points. This relation, which corresponds to Eq. (4-6.5) of the dif¬
ference method described above, is of fundamental significance for the electro¬
nic integration of the Laplace differential equation.
The simplest electronic integrator consists of a sheet in which the elements
are suitably arranged. Between the elements are found equal resistances. For
example, let us consider the first boundary value problem for a region 5 of the
x,y plane. Let C be the boundary of 5.
We then choose, in the x,y plane, a network with distance h and construct
in the above-described manner the region Sh with boundary Сл. At the net
points of the boundary Ca, by means of a special voltage distribution we apply
voltages which correspond to the boundary conditions of the first boundary-
value problem. The voltage distribution then obtained yields an approximate
solution of the problem.
In certain integrators the resistance between the individual net points can
be varied. In this manner we find that equations with variable coefficients of
the form

with

ki = kl (x, y) , k2= kz (x, y)

can be solved. Such electronic integrators were constructed by L. Ya. Guten-

macher. Also, for complicated regions of the x,y plane these integrators make
it possible to obtain a rapid solution of boundary-value problems.
Still other methods exist for the machine integration of the Laplace dif¬
ferential equation—for example, the method of electrolytic tanks.

Problems

1. Find the function и which is harmonic in a circle of radius a and on the cir¬
cumference C assumes the value

(a) и \c = A cos <p

(b) и \c = A + В sin (p .
4-6. THE DIFFERENCE METHOD 331

2. Solve the Laplace differential equation Ди = 0 in a rectangle 0

0 = У = b with the boundary conditions

^ lz=0 f l(y) t M |^=o fM \x=a — 0 » ^ \y=b — 0 •

Show that the formulas derived here define the solution of the problem for every
piecewise continuous function which is prescribed on the boundary. Further,
solve the problem for the special case

Ш = Ay (b-y) , f2( x) = Be os Л = Л = o.

3. Solve the equation Ди — 1 for a circle of radius a with the boundary con¬
dition и |p=0 = 0 .
4. Solve the equation Ди = Axy for a circle of radius a with center at (0,0) for
the boundary condition и |p=0 = 0 .
5. Determine the solution of the differential equation Ди — A + B(x2 — y2) in a
circular ring a ^ p ^ b if

и = Alt = 0 .
p=a P=b

The origin of coordinates is at the center of the ring.

6. Construct the Green’s function of the Laplace differential equation (first
boundary-value problem) for (a) a half circle; (b) a ring; (c) a layer (0 ^ z ^ /) .
7. Determine a harmonic function in a ring where a ^ p b , which satisfies
the boundary conditions

U |p=o — f i(y>) , U |р=ь — /г(<р) •

8. Determine the solution of the Laplace differential equation Ди — 0 in the

half plane у ^ 0 with the boundary conditions

u(x, 0) =
Г for
for
x < 0
x > 0 .

9. Find a function u(p, cp) which is harmonic in a spherical sector p 5S a, 0 ^<p

fg (p0, and satisfies the boundary conditions

(a) и |v=0 = , и |p=<p0 = <7i, и |p=tt = q2 (<7i and <?2 are constants)
(b) U Itp= о — U |ір=(р0 = 0 , U |p=a = f\ф) •

10. By the difference method solve the first boundary-value problem for Ди = 0
in a rectangle 0-йх-йа,0-й.у-йЬ, with each side of the rectangle divided into
eight equal parts. The boundary conditions read:

^ Ix^o — L{f , I I ^ l>=6 sin X , U |x=a — ^ l>=0 — •

b \ b ) a a

Compare the results with the analytical solution.

11. Calculate the spatial potential of a sphere of constant density p = p0. Hint:
Solve Ди = 0 in the exterior of the sphere and Ди = Акра in the interior of the
sphere and match the solutions on the sphere.
12. Determine the potential of a simple layer distributed on a sphere with
332 ELLIPTIC DIFFERENTIAL EQUATIONS

density v = vQ. Hint: Solve Ди — 0 in the exterior and in the interior of a sphere,
and for the matching of the solutions on the spherical surface use the condition
that the potential of a simple layer is discontinuous across the surface.
13. Solve the first boundary-value problem for a bounded circular cylinder
(p g a, 0 g z si /):
(a) On the end surfaces of the cylinder let the boundary value (first or second
type) be equal to zero, while on the lateral surface it must satisfy

U |p=a = f(z) .

(b) On the lateral surface and on one of the end surfaces let the boundary value
(first or second type) be equal to zero, while on the other end surface of the
cylinder the condition is

и = f(p) ,

for example, f(p) = Ap( 1 — (p/a)).

14. Solve the inhomogeneous equation

Ди — — /

in an unbounded cylindrical region with homogeneous boundary conditions (first

or second type) and construct the Green’s function.
15. Find a harmonic function in the interior of a sphere which is equal to иi on
one half of the spherical surface and u2 on the other half.
.
16 Calculate for the density the series development using spherical functions
for the charges induced on the surface of a conducting sphere due to a point
charge.
17. Solve the problem of the polarization of a dielectric sphere in a field due to
a point charge.
18. Determine the gravitation potential of a plane disc. Compare the solution
with the asymptotic representation of the gravitation potential at large dis¬
tances.
19. Find the magnetic potential of a circular current.
20. Solve the problem of the excitation of a plane-parallel electrical field for an
ideal conducting sphere. Solve the same problem for a completely nonconduct¬
ing sphere.

4-7. APPLICATIONS TO CHAPTER 4

1. Asymptotic representation of the spatial potentials

For the investigation of the spatial potential

V(M) = jj j №']ddrp , d = rMP (4-7.1)

at large distances from a body T, we usually take the value of the potential as
equal to m/R, where m is the total mass of T and R is the distance of the center
of gravity from the exterior point M. We shall now derive an exact asymptotic
4-7. APPLICATIONS TO CHAPTER 4 333

representation for V.U1

Let I be a sphere about the origin which entirely contains the body P. Its
center is at the origin of the coordinates. Outside this sphere the potential is a
harmonic function.
The distance of the test point M(x,y,z) from the variable source points
My(xy ,yi,2i) in T (Figure 69), over which we integrate, is given by

d — VV2 + — r\ 2rry cos Ѳ , r = OM, ry = О My, (4-7.2)

from which

1
d
1 1
r l/l + a2 — 2ap ’ a
^ n
r
= cos Ѳ . (4-7.3)

Гу < r means a < 1; we therefore have the development

— І ccnPn{[i) , (4-7.4)
d Г n=0

where Pn(/d is the Legendre polynomial of wth order.* If we substitute this

expression into formula (4-7.1) and note that 1/r does not depend on the varia¬
bles of integration, we obtain

V(M) = у] П pI anPn(v)dT = Vy+ V2+ V3 + '

pryPy{(j.)dr + j pryP2{p)dr +

(4-7.5)

The first member here is equal to m/r, where m is the total mass of the body.
This yields a first approximation for the calculation of the potential for large r.
We now calculate the following terms in (4-7.5). The integrand of the second
member is

рРу(р)гу = ppry = joricos Ѳ = ~ - x

r
Pzz> .
The quantities x, у, z, and r do not depend on the variables of integration and
can therefore be placed in front of the integral sign. Then the second term
takes the form

11 i Prip№dr = -p- (MyX + M2y + M3z) = M(xx + yy + zz) ,

where

My = jj ^ pXydr = Mx , M2 = ^ pyydr = My

M3= jj I pzydz = Mz

111 V. I. Smirnov, Part III2, op. cit. fn. 15.

* See Whitaker and Watson, also Bateman in List of References
334 ELLIPTIC DIFFERENTIAL EQUATIONS

is the moment of the first order and x, у, z are the coordinates of the center of
gravity. Therefore, the second term is as small as 1 lr2. If the origin of the
coordinates is at the center of gravity (x = 0,y = 0,z = 0), then V2 = 0 .
We consider now the third term of the development. For the integrand,
we have

...in,,) » 3/г2 — 1 3(xx, + yy, + гг,)2 - r\r2

Pr —pr,--— — pr\-2Ууг-

= [3(дтхлг + у,у + z,z)2 — г\гг] .

With the notations

Мы = j j j pXiXkdz (x = xt , у =x2 , z = x3)

we obtain for V3 the expression

F3= у jy pr\P2(p)dz

= {x2[3Mu — (Mi, + M22 + M,»)] + y2[3M22 — (M„ + M22 + M33)]

+ z2[3M33 — {M,i T M22 + M33)] + 2-ЗхуМіг + 2-3xzM13 + 2-3yzM2%) .

The polynomial inside the braces is harmonic, since it can be written in the
form

V3=~{(x2- у2) [Мп - M22] + (z2 -x2) [Mn - M33]

2r
+ (y2 — z~) [M22 — M33] + 6[xyMt2 + xzMis + _yzM23]}

in which each summand satisfies the Laplace differential equation. The co¬
efficients standing in the square brackets can be expressed in terms of the
moment of inertia with respect to the coordinate axes. The moment of inertia
of a body T with respect to the x axis, as is known, is given by

A — 11 ^ p(yi + Zi)dv — M22 + M33.

Correspondingly, the moments of inertia with respect to the у and z axes are,
respectively,
В = M22 + Мц , C = Мц T M22.

From this it follows, however, that

Ми- M22 = В - A , Mn — M33 — C — A , м22 - Мзз = C - В.

Therefore we arrive at the following asymptotic representation of the

spatial potential:

V ~ — + ™ (xx + yy + zz) + -V {(x2 - y2)(B -A)+ (y2 - z2)(C -B)

r r 2r
+ (z2 — x2) (A — C) + 6 (хуМи + yzM23 + zxMu)} (4-7.6)
4-7. APPLICATIONS TO CHAPTER 4 335

which is exact up to terms of the order of 1/r6.

The representation (4-7.6) can be simplified to

m 1
V + {(x2 - у2) (B-A) + (/-z2) (C-B) + (z2 - x2) {A — C)} , (4-7.7)
2r
provided the origin of the coordinates
coincides with the center of gravity and
the coordinate axes have the direction of
the principal axes of inertia.
This asymptotic representation of the
spatial potential permits us to give the
answers to a series of questions on the
inverse problem of potential theory. This
consists of determining the characteristic
quantities of a body from its potential (or
from any of the derivatives of its potentials).
That is, from the coefficients of the develop¬
ment (4-7.6), we can calculate the mass,
the coordinates of the center of gravity, and the moment of inertia.

2. Problems of electrostatics

In electrostatics the solution of Maxwell’s equations leads to the determi¬

nation of a scalar function, the potential (p. Between tp and the electrical field
strength E there exists the relation

E = —grad (p .

By using Maxwell’s equation

div E = 4rtp

we obtain

A(p = —4np .
Therefore, the potential at points of space at which electrical charges are found
satisfies the Poisson differential equation, and at points where no charges are
found it satisfies the Laplace differential equation.
1. The basic problem of electrostatics is to determine the field which is
produced by a system of charges on a given conductor. Two different formu¬
lations of this problem are possible.
(a) Given the potential of the conductor, determine the field exterior to
the conductor and the charge density on the conductor. The mathematical
formulation here reads as follows:
Determine the function <p which satisfies the Laplace differential equation
J(p = 0 everywhere exterior to the given conducting system, vanishes at in¬
finity, and on the conducting surface assumes the prescribed value <pp.

9 = 9i •
 336 ELLIPTIC DIFFERENTIAL EQUATIONS

In this case, therefore, we arrive at the first boundary-value problem for the
Laplace differential equation. Its unique solvability results from the general
theory discussed above.
(b) The inverse statement of the problem is also possible, i.e., given the
total charges on the conductor, determine its potential, the charge distribution
on its surface, and the field in the interior of the conductor. The solution of
this problem leads to the following problem:
Determine a function <p which satisfies the Laplace differential equation
A(p — 0
exterior to the given conducting system, vanishes at infinity, and on the surface
of the conductor assumes a certain constant value

<P I Si = const.

and also satisfies the condition

<j>* da = —Апві.
J s{ dn

Here ві is the total charge on the fth conductor.

2. The unique solvability of the secondary-value problem does not result
from the general theory; however, it can easily be proved.
Let us assume that there are two distinct solutions <pt and <p2 of Problem (b)
above. Then their difference

<p' = фі— Ѵг
satisfies the equation
/Up' = 0

and the conditions

<p' U, = const. , ^—da — 0, y>'|~ = 0.

J bn
We imagine the given conductor to be completely enclosed by a sphere IR of
sufficiently large radius R and apply to <p' the first Green’s formula in the
region TR bounded by the surface of the sphere IR and the conducting surfaces
Si;

(P(pr) dz
_ . dcp'
(p -^7—ida + Z
" ,d<p' ,
Zr dn <=1
From this, because of the above conditions,112 we obtain

lim (
R-oo J Tr (Pcp'fdz = 0 .

112 As a consequence of the condition <p' !« = 0, the function <p' is regular at in¬
finity (see page 270); then

dv'
——do 0 for R -> OO .
on
4-7. APPLICATIONS TO CHAPTER 4 337

Since the integrand is positive it follows that

r<p' = 0
or
<p' = const.
everywhere in the region under consideration. By considering the condition

<P' |oo = 0

we finally obtain

<p' = 0 ,
and hence the uniqueness is proved.
3. From the uniqueness of the solution of the boundary-value problem for
the Laplace differential equation it follows that the potential <p of an individual
conductor is directly proportional to its total charge e, i.e.,

el<p - C .
Therefore, if the charges e and e' — me lie entirely on a single conductor, then
the corresponding potentials and <p' must satisfy the equations

A<p — 0 , A<p' — 0
and the boundary conditions

d<p j 1 Г d<p' ,
-т-da — e t Ф
—4л — me
4л J 5 dn

Consequently, <p' — m<p = 0 , i.e., <p'l<p = e'le .

On the surface of a single conductor we therefore have

e'l<p' = e/<p = C = const.

The constant C is called the capacity of the single conductor. It does not depend
on the charge on the conductor but is determined by its shape and size. There¬
fore, the relation

e — C<p
exists for a single conductor. The capacity of an individual conductor is
numerically equal to the charge when the potential of the conductor is equal to
unity. If, in addition to the given conductor, still another conductor is present,
its potential depends essentially on the shape and distribution of charges on the
other conductor; indeed, for a system of conductors we have

ei = Cn<pi + Cn(<P2 — <Pi) 4- h Cln(<pn — <pi)

02 = Сц(<Рі — <рг) + Сгг<Рг + • • • + С2и(у>п — У>г)

en ~~ <pn) T Cnii^Pz *pn) * * T Сппфп

where and <pi are the charge and the potential, respectively, of the г th
conductor. The coefficient C<* is the “partial capacity” of the fth conductor
338 ELLIPTIC DIFFERENTIAL EQUATIONS

with respect to the &th conductor. Ca can be defined numerically as the

charge which must reside on the z’th conductor in order for the &th conductor
to maintain the potential when all the remaining potentials become zero.
4. The matrix of coefficients C« is symmetric:

Ca = Cki •
For simplicity we shall consider the case of two conductors; for a larger
number of conductors the proof proceeds analogously.
Given two conductors a and b, the determination of the coefficients Саь
and Cba leads to the determination of the solutions u{1) and и<2) of the equa¬
tions Aua>=0 and Auw — 0, which satisfy the boundary conditions
,(Di ,.(1> 1 _
= 0, u ISb — 1 , иШ loo

= Cab
-i:L bnd°
,(2) I ua)\Isb -— o,
= 1, u M(2,U
_J_ ( r dum,
= вьг> = Cba
47Г „ L. ьпіа

Now let SR be a sphere of sufficiently large radius R which encloses

both conductors a and b. Then we apply the Green’s formula to the functions
un) and ul2) in the region TR lying between the surface of sphere SK and the
conductor surfaces Sa and Sb:

dum 2 dun)
( )
(ull)Aum -uwAuil))dr \ —u \da .
Tr J ZR+Sa-\- Sf, dn dn

The integral on the left side of this equation vanishes. By using the
boundary conditions and the conditions at infinity we therefore obtain

duw dun)
da — da = 0
dn dn

ОГ Cab — Cba r

which was to be proved.

5. As a concrete example we shall consider the field of a charged sphere.
On the surface of a charged sphere of radius a let the potential <p0 be given.
The field and the charge density (see Problem 1 above) on the sphere can
be represented by

w = —a and a=-^~
r 4na
If the total charge e0 on the sphere is given instead of the potential <p0, then

во во во
<P о = a — (r > a) .
а
<P
4ла r
Here the sphere has the capacity

C= a ,
4-7. APPLICATIONS TO CHAPTER 4 339

i.e., in absolute units the capacity of a sphere is numerically equal to its

radius.
As another example, we shall investigate the spherical condenser (a
system of two concentric conducting spherical surfaces).
The inner sphere has the radius and the potential V0; the outer sphere
has a radius r2 and is grounded. Then the determination of the field inside
of the condenser amounts to the determination of the function <p which
satisfies the equation

A(p = 0
and the conditions

<P lrt = Vo, <p 11*2 = 0

As one can easily see, in this case

ГіГг
V Vo —-—
r2 — r 1 \Г Гг
so that the spherical condenser has the capacity

ГіГг
C =
Гг —Гг
A complicated problem is represented by the calculation of the potential
of a sphere when the second sphere is not concentric to the first. This
problem is solved with the help of the method of images. Since the analy¬
tical solution is quite extensive, we shall not go into it here.113
The first sphere can be mapped onto a plane by means of an inversion.114
We shall now show that by means of a second inversion the determination
of the potential of a plane surface and of a sphere leads to the calculation
of the potential of a system of two concentric conducting spheres.
For our purposes it is sufficient to consider, instead of a plane surface
and a sphere, a straight line E and a circle К with the center 0 and radius

-H

113 See Ph. Frank and R. v. Mises, Differential and Integral Equations of Mechanics
and Physics, Vol. II, Braunschweig, 1937, p. 713.
114 Ibid.
340 ELLIPTIC DIFFERENTIAL EQUATIONS

P (Figure 70). From the point 0 we drop the perpendicular OF on the straight
line E (let l be the length of OF) and from the point F draw the tangent
FD to the circle K. With FD as a radius we let the point F trace a circle
H, which cuts the extension of the perpendicular line OF at a point S. We
choose the point S as the center of the inversion.
By this inversion, the straight line L = SO, which cuts the circle К at
the point A, and the circle H are transformed into the straight line L and H
perpendicular to each other, while the straight line E and the circle К are
transformed into the circles Ё and K.
Since orthogonality is preserved under an inversion, the circles E and К
must be orthogonal to the straight lines L and H. This is possible, how¬
ever, only when the centers of the circles E and К coincide with the inter¬
section point В of the straight lines L and H.
Consequently, the given inversion transforms the straight line and the
circle into a pair of concentric circles.
By some simple calculations we find the radii of the circles К and Ё:

VF^-d-p) _ 1
2^/12— p2 id — p) + ~[/12 — p2] У'Л/12 — p2
Thus

2 in V7/2 - P 2 + (/- p) '

Vl2-P2~{l-P)

6. As an example of another two-dimensional problem, we shall in¬

vestigate a cylindrical condenser which is formed of two infinitely long co¬
axial cylinders. Let a uniform electrical charge be distributed on one of these
cylinders. Obviously, the potentials in all planes which are parallel to the
normal cross sections of the cylinders are equal. The problem can therefore
be treated as a plane problem and, instead of using the total charge, only
the charge к per unit length need be given.
If the exterior cylinder of radius r2 is grounded, whereas on the inner
cylinder of radius rx the charge к is given, then the field potential in the
condenser is

tp — 2k In ,
Гі

and for the capacity per unit of length for the cylindrical condenser we
obtain

2 In (rjri) '

The above-considered example allows us to solve a somewhat difficult

problem, namely, to determine the capacity of a wire which lies above a con¬
ducting surface. An infinitely long wire of radius p is placed above an infinite
plane and at a distance l from it. Let a charge with density к (charge per unit
4-7. APPLICATIONS TO CHAPTER 4 341

length) be distributed on this wire. This problem can also be solved as a

two-dimensional problem.

3. The basic problem of electrical exploration

Several electrical methods are used in the investigation of the inhomo¬

geneities of the earth’s crust for the purpose of assaying the soil. The
basic idea of electrical exploration using direct current consists of the fol¬
lowing: By means of grounded electrodes a current is passed through the
earth from a battery. On the surface of the earth we measure the strength
of the field so produced. From the results of the surface observations con¬
clusions can be drawn about the structure of the interior of the earth. The
methods used here are based on the mathematical solution of the correspond¬
ing problem.
The potential of the field of a direct current in a homogeneous medium
satisfies the Laplace differential equation

/77 = 0, 2 > 0 (4-7.8)

with the auxiliary condition

dV
= 0 , (4-7.9)
dz z= 0

which asserts that the vertical component of the current density on the surface
2 = 0 is equal to zero. This is the case since the half space 2 < 0 is non¬
conducting (air).
We now consider a point-forming electrode on the boundary of the half
space at the point A. Obviously, the field potential satisfies

ІР (4-7.10)
V=
2xR ’

where R is the distance from the source, p is the specific resistance of the
medium, and I is the current strength.
This function differs from the Green’s function in an unbounded space on
the basis of condition (4-7.9) by a factor of 2.
Now if we measure, by means of a resistance bridge, the potential dif¬
ference between two points M and N which lie on a straight line through
A, we obtain

V(M) - V(N) = Щг&г,

dr
where Ar is the distance between M and N.
Under the assumption that the points M and N lie sufficiently close to
each other, we can write

V(M) - V(N) dV Ip
Ar dr 2 л: r2

Here r is the distance of the point 0 (the center point of the segment MN)
342 ELLIPTIC DIFFERENTIAL EQUATIONS

from the current-conducting electrode. The current strength /, as is known,

must be considered since it can be measured during the observation. From
this we obtain, for the resistance of the homogeneous half space,

2кг2 dV
P = ~T~ (4-7.11)
dr
If the medium is inhomogeneous then we call the quantity p determined by
(4-7.11) the apparent resistance. We denote it by pk\ pk is not constant.
We shall now consider the problem of vertical electrical probing. Let
the layers of the earth’s crust be horizontal, and let their resistance depend
only on the depth: p = p(z).
In this case, the apparent resistance is a function of the distance r = AO.
The problem raised by the vertical electrical probing thus consists of deter¬
mining the function p(z), which yields the “electric layer’’ of the medium
with respect to a known value pk(r).
We turn next to the problem of a two-layered medium. A homogeneous
layer of density l and resistance p0 is in contact with a homogeneous medium
of resistance pt.

Obviously for distances r < l the apparent resistance pk is equal to p0,

since the influence of the lower medium here is very weak. For large dis¬
tances (R » l), by contrast, pk becomes equal to pi.
Our problem therefore leads to the determination of the solution of the
Laplace differential equation which in the layer 0 < z < l is equal to V0 and
in the half space z > l is equal to L. At z = l the potential must satisfy
the continuity condition

= FxU. (4-7.12)
Also the normal components of the current density must be continuous:

1 5F0 1 dVi
(4-7.13)
Pо dz Z— I
p1 dz
At z — 0 the potential V0 must satisfy the condition (4-7.9), while at the
point A, which we chose as the origin of a cylindrical coordinate system
(<p,r,z), it must possess a singularity of the form (4-7.10):
pj 1
L0 + Vo , (4-7.14)
2k 1/z2 + r2

where v0 is a bounded function.

The potential V’i must be bounded at infinity. The functions V0 and Vi
satisfy Eq. (4-7.8), which, because of the cylindrical symmetry of the problem,
assumes the form

d2V J_ dV_ dzV

dr2 + r dr ^ dz2
4-7. APPLICATIONS TO CHAPTER 4 343

By separation of variables two types of solutions for V can result, which are
bounded at r = 0:

е±ХгШг).
Here /о is the Bessel function of the zero-th order (see Appendix) and X is
the separation parameter. We write the solution in the form

Vo(r, z) = &L.
2n
_L_ + Jo\°°(A0e~Xz + B0eXz)JQ(Xr)dX
V r1 + z2

ѴЛг, z) = [“(Arf-** + BieXz)J0(Xr)dX

where A0, B0, Alt Bx are constants. The condition (4-7.9) relates A0 and B0.
We calculate the derivative

= ■- ^r- ' 7 2 2'чз/Т + Г(- ЛА0<ГХг + ХВоеХг)ШгШ .

02 Z7T (2 + Г ) Jo

Condition (4-7.9) then assumes the form

^ (B0 — A0)J0(Xr)XdX = 0 .

This relation must hold for arbitrary r, therefore,

B0'= A0.
From the boundedness of Vi as z—>oo there follows
Bi = 0.
Therefore,

ѴЛг, z) = [“А^ЫШХ ,

and

Voir, 2) = + A0(e~Kz + e^mtrW .

Here we have used the formula

1
Jo(Xr)e XzdX (4-7.15)
Ѵгг + 22 0

(see Appendix) and р0І/2л- = q .

We have yet to determine the constants A0 and from the conditions
(4-7.12) and (4-7.13) at 2 = /, which leads to the system of algebraic equations

А0(е~ш + 1) - A,e~M =- qe~M

— А0(е~ш - 1) -J-А,<ГШ =- ^е~ш .

Po Pi Po
From this we find
344 ELLIPTIC DIFFERENTIAL EQUATIONS

(Pi - Ро)е-ш
A0
(joi + p0) — (^>i — ро)е~ш

so the solution V0 for the upper layer is given by

(4-7.16)
F”(r- -££1/“+ + ■
Here we have set

Pi Pa _ fo
Pi +"i°o
We shall now transform expression (4-7.16). Since \ k\ <1, we can write
со
кё~ш
I F • в_Ш*
1 - ке~ш
and
Г oo
о X'
oo oo r oo

V0(r, л) = ІР
1 j^n g—\(2nl+z) Ipo n —\ { 2nl —z)
2i7T n = о J q
Ja{Xr)dX + -I
2л n = 1
kne JoUr)ctt . (4-7.16')

Thus, by use of formula (4-7.15) we obtain

Ipo/Z,n 1
V0(r, z) — I kn + I kn- (4-7.17)
2л \и=і \/ г2 + (z — 2nl)2 -\/r2 + (z + 2nl)2 ) '

This expression for solution (4-7.16) can be written immediately when

the problem is solved by the method of images. If 2 = 0, then for the poten¬
tial on the earth’s surface we obtain

Vo(r, 0) — ІР° Г J_ _i_ 2 у ^ 1 (4-7.18)

2л- 1 r n=o-\/r2 + {2nl)2j ’
from which follows

, о f knr
dr 2л \_гг n=i [г2 + (2и/)2]3/2

whereas for рк, according to formula (4-7.11), we find

knr3
+ 21
1
n— 1 [r2 + (2w/)2]3/2 ]
F(f/2)3
1 + 2 І (4-7.19)
n=. i [(f/2)2 + n2]3/2

where f = r/l and /(f) are the expressions in the square brackets. For r € l,

Pk ~ Po •

In order to estimate the behavior of pk for large r, use formula (4-7.19) and
let r—>■ °o (f —>oo). The limit value of this nth term of the sum equals kn;
hence

lim pt = p,(l + 2**) = ft(l +YZj)

1 + к pl + po + (Pl — Po)
= Pol1- = pi.
— kh = P°'^~L
Pl +- 7-
Po — (Pl — Po)^
4-7. APPLICATIONS TO CHAPTER 4 345

By a comparison of the experimentally determined data with the curve deter¬

mined by formula (4-7.19), we can determine p0 from the values of pk for small
r and pi from the values of pk for large r. The thickness l of the upper
conducting layer is determined according to the selection principle. It is
equal to that value of l for which the empirical curve as a function of p(£)
= p(r/l) coincides best with the curve calculated by formula (4-7.19). Here
we shall not dwell on the technicalities of the selection which can be carried
out by means of bilogarithmic measurements.115
In the case of multiple layers, the curves for pk can be calculated analo¬
gously. The character of the electric layers of the medium under considera¬
tion is determined by the selection of those theoretical curves that coincide
best with the empirical curves. By increasing the number of layers, the
technicalities of interpretation become complicated, since the number of theo¬
retical curves entering into consideration is greatly increased.
We have proved, therefore, that for different electrical layers pfz) Ф p2(z)
the corresponding apparent resistances also are different:

Pk\r) Ф p?\r).

Consequently, the problem—to determine the electric layer from the apparent
resistances—appears from the mathematical side to have a uniquely deter¬
mined solution.116
Problems analogous to those of electrical exploration considered above
occur in different areas of physics and technology. Electrostatic problems
of this type arise in the construction of electronic equipment, whereas heat-
theory and hydrodynamic problems arise in different areas of technology (heat
loss of buildings, filtration of water under a retaining dam, etc.).117 The
problem of the determination of a magnetic field in an inhomogeneous medium
occurs, for example, in magnetic testing of materials. To determine an
ultimate failure in a sample—for example, an air bubble below the surface—
the metallic sample is placed between the poles of a magnet, and the magnetic
field on the surface is measured. From the perturbations of the magnetic
field, the presence of the defects is observed, and if possible their extent,
the depth at which they occurred, etc., are also determined.
For the solution of such problems several methods of analogy can be
applied. Therefore the analogy between the potential fields of different
physical processes is used.118
We consider now the potential field in an inhomogeneous medium (for
example, a stationary temperature field, a magnetic field in an inhomogeneous

115 See, for example, the excellent book by A. I. Zaborovskii, Electrical Explora¬
tion, 1943.
116 A. N. Tychonoff, “On the uniqueness of solutions of problems of electrical ex¬
ploration,’’ Doklady, 69, (6): 797 (1949).
117 N. N. Pavlovskii, Theory of Motion of Ground Waters for Hydro-technical In¬
stallations and its Fundamental Applications, 1922, Ch. XIV.
118 A. V. Lukyanov, “On the electrolytic simulation of spatial problems,” Doklady, 75
(5): (1950).
346 ELLIPTIC DIFFERENTIAL EQUATIONS

medium, an electrostatic field, the velocity field of a liquid during filtering).

The potential functions u(x, y, z) of these fields in each homogeneous sub-
region satisfy the Laplace differential equation Ли — 0. At the boundary
between two regions Gi and G2 with different heat-conduction coefficients,
magnetic permeabilities, etc., the condition

, du^l _ U2duw
1 dn dn
holds in which ki and k2 are the corresponding physical constants.
On the boundaries of similar geometric regions the corresponding numeri¬
cal values of the potential or the normal derivatives of different physical
fields are prescribed. We shall assume that the physical inhomogeneities of
this region are not different geometrically and are distributed similarly. The
ratios of the physical constants (heat conductivity, magnetic permeability,
etc.) of an arbitrary pair of corresponding inhomogeneities are also equal.
Then the values of the potential of this field at the corresponding points
coincide numerically, since the potential solutions are one and the same
mathematical problem and this problem permits only a single solution.
The direct measurement of a temperature, in magnetic or other fields,
is significantly more difficult than the measurement of a current field in
an electrolytic tank. We replace it therefore in a suitable manner by meas¬
urements in such a tank. We can also choose the units in a suitable way.

4. The determination of vector fields

In addition to scalar problems, the problem often arises in electrodynamics

and hydrodynamics of determining a vector field from the given rotation and
divergence of these fields.
We shall prove that a vector field A interior to a region G, whose bound¬
ary S is bounded, is uniquely defined when the rotation and the divergence
of A are known in G, that is,
rot A= В (4-7.20)

div A =C (4-7.21)

and on the boundary S the normal component of A is prescribed:

An |s = /(M). (4-7.22)

The functions В, C, and / are not arbitrarily prescribed. Often the re¬
lations

div В =0 (4-7.23)

jj f{M)dS= jjj Cdr (4-7.24)

must be satisfied. Now we assume / to be continuous on the surface S and

the functions В and C, including their derivatives, to be continuous in G;
further, let the surface S be so constituted that the second interior boundary-
4-7. APPLICATIONS TO CHAPTER 4 347

value problem with continuous boundary values is solvable for it. We shall
solve our problem in three steps. First, we shall seek a vector Ax which
satisfies the conditions

rot Ai = 0 (4-7.25)

div Ax =C . (4-7.26)

From (4-7.25) there follows

Ai — grad (p . (4-7.27)

If we assume <p in the form

C(Q)
<P(P) dzQ (4-7.28)
в Г PQ

then we satisfy (4-7.26). Now we define a vector A2 such that

rot A2 — В (4-7.29)

div Л2 = 0 . (4-7.30)

If we set

A2 = rotcp, (4-7.31)

then the condition (4-7.30) is satisfied.

If we substitute (4-7.31) into (4-7.29), we obtain

grad div ф — Аф = В . (4-7.32)

Now we require

div ф=0. (4-7.33)

Then Eq. (4-7.32) for ф assumes the form

Аф = — В. (4-7.34)

We consider a region Gi which entirely contains G and is bounded by

a surface Si. We continue В into the region Gx — G, where the following
conditions are to be fulfilled:
(a) The normal component Bn of В on S is continuous (В itself in general
is discontinuous), Bn{ = ВПа.
(b) Bn = 0 on Si. (4-7.35)

(c) In Gi — G, div В = 0. (4-7.23')

We show how this continuation of В into Gx — G can be realized and

therefore set

В = grad x in Gi — G .
The condition div В — 0 yields
AX = 0 in Gi — G . (4-7.36)
ГИЯ ELLIPTIC DIFFERENTIAL EQUATIONS

The boundary condition, according to (a) and (b), has the form

= BH on S (4-7.36')
dn

= о on S, (4-7.36")
an
where B„. is the limit value of Bn from the interior side of 5. For the func¬
tion we obtain the second boundary-value problem (4-7.36)—(4-7.36").
The necessary and sufficient condition for the solvability of this problem
is that

dS B„dS = 0
S I S dn
be satisfied, because the relation

div Bdz = 0
a
is satisfied.
If we set

f IHQ) dvQ
Ф(Р) (4-7.34)
Jff, r,-Q
with P P(x, y, z), Q = Q (£, rj, 0, then (4-7.34) is obviously satisfied; the con¬
dition (4-7.33) also is satisfied.
Therefore, we calculate the derivatives

dфx дфу дфг

дх ’ ду ’ dz '
If we represent the integral over Gi as the sum of integrals over G and
Gi — G and bear in mind the relation

then by partial integration

dx
d
dx i S \.Л*=i S U-Л v*+S L(a,-^s - J
follows, where cos a = cos (n, x) |$, cos at = cos (n, x) |s , and n is the direction
of the exterior normal to the surface.
For Ьфх/dx we find

1 .-/г I 1 Oleosa Bx^^dS.

dx 4л: d£ г 4л: JJ r 47tJJ5i r
Analogous expressions result for dфy\dy and dфI|dz. Hence, on the basis of
4-7. APPLICATIONS TO CHAPTER 4 349

(4-7.23) and the continuity of the normal components of В on S (ВПа — BH),

it follows that

div ф= f —dS + {Bn-z_^iiids = о.

)al Г An
4 r s Г

The vector Аг defined by (4-7.31) satisfies (4-7.29) when the vector ф satis¬
fies conditions (4-7.33) and (4-7.34).
Therefore, the sought vector A satisfies the boundary condition (4-7.22).
Now a vector Az is to be determined which satisfies the following conditions:
rot A3 = 0 (4-7.37)
inside of G
div A3 —0 (4-7.38)

An\s = f{M) — Л,„|5 — Агп U = f*(M) on S. (4-7.39)

It is clear that the function f*(M) is uniquely defined, From (4-7.37)

there follows
A3 = grad 0 .
If we substitute A3 into (4-7.38), in the interior of G we obtain
АѲ — 0), (4-7.40)
which yields
dO
(4.7.41)
dn s
i.e., for the determination of Ѳ we have to solve the second boundary-value
problem. A is therefore uniquely determined.
The problem therefore possesses exactly one solution,

A = Ai + Ai + A3.

5. Conformal mapping in electrostatics

1. For the solution of two-dimensional electrostatic problems the theory

of functions is often used. We shall consider the following problem as an
example:
Determine the electric field of several charged conductors whose potentials
are equal to щ , иг, • • •.
This problem, as is known, leads to the equation

Au = 0 (4-7.42)

with the boundary conditions

и = Ui, (4-7.43)

where S; designates the surface of the z'th conductor. If the field can be
regarded as a planar field which does not change along the z axis, then Eq.
(4-7.42) and the boundary conditions assume the form
350 ELLIPTIC DIFFERENTIAL EQUATIONS

u\a< = Ui (4-7.45)

where Ci is the contour of the bounded region Si.

We write the potential и as the imaginary part of an analytic function

/(2) = v(x, y) + iu(x, y) , z = x + iy . (4-7.46)

Therefore the Cauchy-Riemann differential equations must follow:

vx = uy , vy = — ux (4-7.47)

and

vxvy + uxuy = 0 . (4-7.48)

From the boundary condition (4-7.45) it follows that the imaginary part of
/(2) on the contour C, is constant.
From the conditions (4-7.47) we know that

v(x, y) = const. (4-7.49)

represents the equation of the lines of force,119 and the lines of equipotential
can be described by the equation

u(x, y) = const. (4-7.50)

because of (4-7.48).
To solve the problem under consideration it is sufficient, therefore, to
find the conformal mapping of the 2 plane (2 = x + iy) on the w plane (w =
и + iv) for which the boundaries of the conductor are mapped into the straight
lines

и = const.

or

Im w = const.

If such a mapping function w = /(2) is known, then the sought potential

is determined from

и = u(x, y) = Im/(2).

The knowledge of these potentials permits the calculation of the electrical

field strength

du du
Ey=- (4-7.51)
dx 7 dy

and the calculation of the density of the surface charges along the 2 axis:

1 ✓bJTIF; 1 / du
+
/ du у
dx \dy) '

119 The equation of the lines of force results from dxlux = dy/uy. If one replaces ux
and uy according to (4-7.47) by — vy and vx, respectively, then we obtain vxdx + vydy =
dv = 0, or v(x, y) — const.
4-7. APPLICATIONS TO CHAPTER 4 351

Also, on the basis of the Cauchy-Riemann differential equations we obtain

(4-7.52)

2. We shall now determine the field of a

one-sided bounded plate condenser, using as an
example a condenser which is formed from two
infinitely thin metal plates у — — <Ц2 and у = сЦ2.
The plates lie in the regions x < 0. We avoid
the construction of the conformal mapping func¬
tion which maps the region represented in Figure
71 onto the region | Im w | ;£ к, and, instead, apply FIG. 71.

this mapping directly for the solution of the

problem.120
The mapping defined by

IV = <p + Іф (4-7.53)
2= 1t(W + eW) ’
carries the lines z =± d/2, x < 0, which cuts the
z plane (z = x + iy) into the region | ф | ^ к of the 7T

w plane (w — <p + іф) (Figure 72). For the complex

potential we choose the function

-Jr2"», (4-7.54)

where щ is the potential difference between the fig. 72.

condenser plates. The potential of the electric

field is then represented by the function

u{x, y) = Цр-ф . (4-7.55)

Here ф is connected with x and у by the relations

x = <p + ev cos ф)
Лк
(4-7.56)
у - -ту-іф + e9 sin ф) .
Zk

Figure 73 shows the equipotential and the lines of force of a one-sided

bounded plate condenser. We now investigate the field in the neighborhood
of the boundary of the condenser.
We see from formula (4-7.56) that as <p-+— 00

x и У (4-7.57)

120 See Ph. Frank and R. v. Mises, Ch. XV, §5, op. cit. fn. 113.
352 ELLIPTIC DIFFERENTIAL EQUATIONS

i.e., inside the condenser at large distances from the boundary the field is
planar; but as <p—>co

p — т/*2 + У2 ~ > (9 = arc tg-^- « ф , (4-7.58)

In X

i.e., outside the condenser, at large distances from the boundary, the equi-
potential lines are circles.
If instead of w the complex potential / = (u0l2n)w is introduced so that
w = {2n/ut)f, then the relationship between z and f(z) is given by

= d(-£- + -Ц?2л//“Л .
V Mo An )
From this it follows that
dz_d_
1 + e2*//u°^
df M0
on the other hand, for f = (u0/2n)[<p ± ni) we obtain

dz Mo
—(1 - ev) or f\z)
df M0 d( l-e*) '

If we assume m0 = 1, then for the density of the charge a, according to

formula (4-7.52), we obtain the value

1
a = (4.7.59)
An And\l-ev\ ‘

OO

i
a
And ’

and as <p OO

1
And# ’

i.e., in this case, the density of the charge

on the outside of the plate decreases as
1ІР-
fig. 73. Formula (4-7.59) shows that for <p = 0
(on the boundary of the condenser) a = oo.
The boundary of a plane plate possesses an infinite curvature; therefore, in
order to charge the plate to a definite potential, an infinitely large amount of
charge must be placed on it.
A wide range of problems can be solved with the help of conformal
mapping. With this method, for example, the question regarding the in¬
fluence of the boundary of thick-walled plates of a plane condenser can be
answered successfully, as well as a series of questions which are related to
the influence of deformation in condensers. The method of conformal mapping
can also be used for the treatment of dynamic problems. Its disadvantage
4-7. APPLICATIONS TO CHAPTER 4 353

is that it can only be used in plane problems which lead to the two-dimen¬
sional Laplace differential equation Аги = 0.

6. Conformal mapping in hydrodynamics

1. For the solution of problems of the motion of а solid body in а liquid,

the boundary conditions on the surface of the body play an important role.
In the case of an ideal fluid, the boundary condition is that the normal
component of the fluid velocity

Vn(x, y, z) = vx(x, y, z) cos (n, x) + Vy(x, y, z) cos {n, y) + vz(x, y, z) cos (n, z)

for all points on the surface of the body must be equal to the normal com¬
ponent of the velocity of motion of the body. If the body does not move,
then the boundary condition on the surface of the body assumes the simplest
form

Vn - o .

If the motion under consideration possesses a potential, i.e., a function <p

exists such that
v = grad <p ,

then the boundary conditions read

dtp
= 0 in case the body does not move;
dn S

dtp
— un in case the body moves with the velocity u.
dn s

From hydrodynamics it is known that the velocity potential for an in¬

compressible fluid satisfies the equation

A<p = 0 .

Therefore the determination of the potential for the streaming about a solid
body in an incompressible ideal fluid is reduced to the solution of the Laplace
differential equation
A<p = 0,

where on the boundary of the streaming body the boundary condition

dtp
Un
dn s

must hold, i.e., the second boundary-value problem for the Laplace differential
equation must be solved.
If the motion is planar, then the solution of the problem can be obtained
by using function theory.
For planar motion of an incompressible fluid the continuity equation is:

dvx _ d(— vy)

(4-7.60)
dx dy
354 ELLIPTIC DIFFERENTIAL EQUATIONS

We describe the equation of the stream lines

dx _ dy
vx Vy

in the form

vxdy — Vydx = 0 (4-7.61)

an introduce a new function ф by the relations

дф дф
Vx = ~ , Vy = - .
ay ox

From Eq. (4-7.60) it then follows that the left side of expression (4-7.61) is
the total differential of the function ф\

vxdy — Vydx = dф .

The one-parameter family of curves

ф(х, у) = С

therefore represents the stream lines of the incompressible fluid motion.

If a velocity potential exists, then the equation rot v = 0 is equivalent to

Аф = 0.

From the relations for vx and vy it then follows that

dip _ дф dip _ дф
dx dy ’ dy dx

i.e., the functions <p and ф satisfy the Cauchy-Riemann differential equations.
Consequently, the complex function

w{z) = ip(x, y) + іф(х, у)

is an analytic function.
Therefore, every irrotational (rot v = 0) planar motion of a fluid has a
corresponding definite complex analytic function, and conversely every analytic
function corresponds to a definite form of fluid motion, i.e., two motions
correspond to an analytic function, since the roles of the functions <p and ф
can be interchanged.
2. In conclusion, we shall treat a concrete example of the application of
the theory of functions to the solution of problems for the streaming of a
body—here, a circular cylinder—in a planar fluid.
On a circular cylinder at rest of radius r = a, let a planar fluid stream
impinge which at infinity has the constant velocity u. In the case of a
static fluid, we would consider the motion of the cylinder moving with the
constant velocity и with respect to the stationary fluid.
As a reference system, we choose a rectangular coordinate system (x,y,z)
attached to the cylinder in which the x axis is chosen parallel to the direc¬
tion of the velocity of the cylinder.
On the surface of this moving body, the boundary condition
4-7. APPLICATIONS TO CHAPTER 4 355

дФ dy
— u~r~
ds ds

must be satisfied, where ds is the element of arc on the boundary of the

body.
In the case of a motion propagating with the velocity и this condition
can be integrated over the surface of the body. We then obtain on the
surface of the body

ф — uy + C .

Therefore, our problem amounts to the solution of the equation

Дф = 0

with the following boundary conditions:

on the surface of the cylinder ф — uy + C

at infinity and approach zero.

dx dy

The last condition means that

div _ dф + ■ dф
vx — ivy
dz dy dx

exterior to the circle C is a single-valued analytic function, which vanishes

at infinity. The function w can therefore be represented in the form

w = Ci In z —

Now we set

Ck — Лк + iBk

and determine the constants Ak and Bk from the boundary condition

ф = ua sin Ѳ + C ,

which are transformed to polar coordinates by z = аеіѲ.

For the constants then we obtain

Ai=0, A2 = ua2 , B2 = 0 , Ая = B3 = 0 , Zfi = —.

Zn

From this there results

1 . a
IV = ——Г In z — и—
2лг z
/ л п а
T~i 2

<р — -^—0 — и cos 0—

2л г
, Г . , . . аг
ф =-— lnr + MSinw — .
2л г

The first term in the expression for w denotes the circulation around the
356 ELLIPTIC DIFFERENTIAL EQUATIONS

cylinder with the intensity Г. Where there is no circulation, we obtain

a2
iv = — и— .
z

For the flow about a static cylinder, which at infinity has the velocity u, the
complex potential has the form

. ua . I
w = uz A-V ——r In z .
2 T~t
.
z Zki

3. With the results for a streaming about a circular cylinder we can

solve the problem of the streaming about an arbitrary contour by using the
method of conformal mapping. We consider its application to the problem
of the flow about a rectangular plate.
On the x axis let us place an infinitely long plate of width 2a (Figure 74).

On the plate a constant planar flow occurs which at infinity has a velocity
c with components и and v. By means of the analytic function

2=І(с+т)=/ю
we obtain a reversible single-valued mapping of the region exterior to the
plate in the z plane onto the region exterior to the unit circle in the C plane.
Here the point z — oo corresponds to the point С = °°» and

We shall now investigate the nature of the mapping of the condition at in¬
finity. For the complex potential

w(z) = <p + іф

we have

(dw\ .
—Г- = и — IV = Voo ,
\ (l Z J z oo

i.e., the conjugate value of the complex velocity.

We determine now the value of the complex velocity of the fictitious
flow in the C plane:
4-7. APPLICATIONS TO CHAPTER 4 357

r r, m dw dw dz
MO = w[f{z)) , .
aC az a£

From this we obtain

/ dw\
kVo k= —
V dZ Л=с * 2

The fictitious flow therefore represents the flow about a cylinder of unit
radius and at infinity has the complex velocity kvFor such a motion, the
complex potential has the form

w(0 — kv„Q 4 -b -r-т In c

С 2ттг
From z — /(0 there now follows:

^ _ г + т/z2 — a2 l/z2

By use of this expression, we obtain, for the complex potential of a fluid

streaming about a plate, the expression

Г 2 + t/z2 — Й2
w(z) — uz — г/t/z2 — й2 in
27гг

If no circulation occurs, we have instead the following relation

w(z) = uz — г/t/z2 — a2.

From these relations it is seen that the velocity at the ends of the plate
becomes infinitely large. Under real conditions, of course, this is not the case.
Our conclusions can be explained on the basis that we have assumed an ideal
fluid. With the use of the Bernoulli equation an expression can be found
for the forces which the fluid exerts on the streaming body.
The investigation of those forces with which air acts on a wing of an
airplane constitutes the subject of aerodynamics. Toward the development
of this theory Russian scholars have made outstanding contributions, fore¬
most of which were Joukowski and S. A. Tschaplygin. A paradox arises in
the simplest case of a circulation-free flow about a cylinder by a planar fluid
in that the flow exerts no force on a cylinder. If, for the propagating flow,
a circulation of the velocity about the cylinder exists, then a force appears
which acts on the cylinder perpendicular to the fluid velocity at infinity.
The theory of functions can be applied only to planar motions. In the
three-dimensional case we must turn to other methods in order to treat the
problem of streaming about a solid body by a fluid. The solution of this
problem in complete generality is very difficult. Let us consider, for example,
the motion of a sphere in a static fluid. Let velocity of the sphere be con¬
stant. The problem then consists of solving the equation

d<p = 0

with the boundary conditions

358 ELLIPTIC DIFFERENTIAL EQUATIONS

—>— — u cos Ѳ
072 |r=a

on the surface of the sphere and

dip _ d(p _ dip

dx dy dz

at infinity. The solution has the form

<p = Ли grad— .
r

By using the boundary condition, we obtain

a3ur

as a solution of the problem.

An ideal fluid was assumed in all the cases considered. For viscous fluids
the boundary conditions take another form, and on the surface of the body
the so-called adhering condition must be satisfied. This condition states that
directly on the surface of the body the velocity of the fluid must coincide in
both magnitude and direction with the velocity of the corresponding boundary
points of the body.
The problem of flow about a body by a viscous fluid leads to great mathe¬
matical difficulties. In the development of this part of hydrodynamics, bound¬
ary-layer theory plays a significant role.

7. Biharmonic equation

In Section 2-6 §1 we found

(4-7.62)

as the equation of a transverse vibrating rod. The vibrations of an unloaded

thin plate whose boundaries are fixed can be described by the analogous
equation121

or + a2JJu = 0. (4-7.63)

Here the boundary conditions

и — 0 and (4-7.64)
dn
must be satisfied.
Moreover, the function и must satisfy the initial conditions

u(x, y, 0) = <p(x, y), y, 0) = <p{x, y) . (4-7.65)

or

121 V. I. Smirnov, Part III2, op. cit. fn. 15.

4-7. APPLICATIONS TO CHAPTER 4 359

If external forces act on the plate which are distributed with the density
f(x,y), then the deformation of a plate whose boundaries are fixed can be
described by the equation

ДДи = /, (4-7.66)

with the boundary conditions

и — 0 and = 0 . (4-7.64)
on

The equation

ДДи = 0 (4-7.66')

is called the biharmonic equation and its solutions biharmonic functions. These
functions must possess continuous derivatives up to and including the fourth
order.
The fundamental boundary-value problem for the biharmonic equation
reads:
Determine a function u{x, y), which including its first derivatives in the
closed region 5 4- C is continuous, in 5 possesses continuous derivatives up
to the fourth order, in S satisfies Eq. (4-7.66) or (4-7.66'), and on the boundary
C satisfies the boundary conditions

du
и Ip = g(s), = h(s) , (4-7.67)
dn о

where g{s) and h(s) are continuous functions of the arc length s on C.
For the solution of the above-formulated initial value problem (4-7.63)—
(4-7.65) by means of separation of variables, we usually set

u(x, y, t) = v(x, y)T(t). (4-7.68)

If we introduce this expression into (4-7.63) and separate the variables, then
for the determination of the eigenvalues we arrive at the equation

ДДѵ — Лѵ — 0 (4-7.69)

with the boundary conditions

v = 0 , —=0 on C. (4-7.70)
on

1. Uniqueness of the Solution. We shall prove that the biharmonic equation

ДДи = 0

with the boundary conditions

u\c = g(s), = h(s) (4-7.64')

о

has at most one solution.

Let us assume that two distinct solutions Ui and uz exist. We consider
their difference
360 ELLIPTIC DIFFERENTIAL EQUATIONS

V = Ui — U2 ,

which obviously also satisfies the biharmonic Eq. (4-7.66) and the homogeneous
boundary conditions

By application of Green’s formula

j {фДу — <pA(p)dS = J — <P^')ds

to the functions <p = v and ф — Дѵ, we obtain

( {AvfdS = 0
Ja

and hence

Дѵ = 0.

Now, however, v \c = 0. Consequently,

v = 0 and Ui = u2,

1. e., the biharmonic functions are uniquely determined by the boundary con¬
ditions (4-7.64).
2. Representation of a Biharmonic Function by Harmonic Functions. We first
prove the theorem:
Theorem. If ux and u2 are two harmonic functions in a region G, then и =
xu,. + М2 is a biharmonic function in G.
For the proof we use the identity

Д(ірф) = (рДф + фДу + -J^ + -J^- -J^-) . (4-7.71)

With

<p = X , ф = Ui

it follows that

Д(хиД = 2——4 . (4-7.72)

ox

By repeated application of the d-operators and by consideration of ДДи2 = 0,

we obtain

JJ(xui + m2) = 0 .

If the region G is so chosen that each straight line parallel to the x axis
cuts the boundary of G at two points at most, then the converse of the
above-proven theorem also is valid.
For every biharmonic function и in the region G, two harmonic functions
Mi and m2 can be found such that
4-7. APPLICATIONS TO CHAPTER 4 361

u = xu, + иг.
For the proof of this statement it is sufficient obviously to consider a
function и I which satisfies the two conditions
Au, = 0 (4-7.73)
and
A(u — xui) = 0 . (4-7.74)
From condition (4-7.74) and formula (4-7.72) we find

Au = A(xui) = # (4-7.75)
bx
Eq. (4-7.75) is satisfied by the function
Г 1
u,(x, y) = \ —Au(s, y)dZ .
Jx0

Since

-z—Au, = A-r—iil = -j-AAu = 0 ,

ox ox 2
Aui depends only on y:
Айі = v(y).
We now define a function uAy) so that

Au, = —r = — v(y),
л t

by
holds, and set
u, = u, + u, .
This function then obviously satisfies the conditions (4-7.73) and (4-7.74), which
was to be proved.
We shall now give another representation of a biharmonic function. Thus
we assume that the origin of coordinates lies inside of the region G and each
line from the origin touches the boundary of G at only one point. Under
these assumptions the following is valid: Every biharmonic function и in G
can be represented by two harmonic functions u, and иг in the form
и = (гг — r\)u, + иг. (4-7.76)
Here гг = хг + уг, and r0 is a given constant.
The proof here proceeds analogously to the preceding one by means of
the identity (4-7.71) and the relations
bu, au, bx . bu, by
Ar2 = 4 , — —- —+ -T—
or bx or by or
3. Solution of the Biharmrmic Equation for the Circle. Given a circle of radius
r0 about the origin of coordinates, we seek a biharmonic function which
362 ELLIPTIC DIFFERENTIAL EQUATIONS

satisfies the boundary conditions (4-7.67) for r = r0. As was shown above,
the sought function can be written in the form
и — (r2 — rl)ui + u2, (4-7.76)
where и, and u2 are harmonic functions. From the boundary conditions we
find
u2\ r=r0 = g. (4-7.77)
From this we recognize that u2 is a solution of the first boundary problem
for the Laplace differential equation. Therefore, u2 can be represented by
the Poisson integral

_ 1 f2* (rl-r2)gda
(4-7.78)
2л Jo г2 + Го — 2rr0 cos(« — Ѳ)
From the second boundary condition we obtain

2r0Ui + 4^- = h . (4-7.79)

dr r=r0
By differentiation we find that the function

9 ru (4-7.80)
r0 dr
also satisfies the Laplace differential equation and therefore can be expressed
by the Poisson integral

г ди2 __ 1 f2*_(rl — r2) hda_

(4-7.81)
r0 dr 2л- Jo r2 + Го — 2rr0 cos (a — Ѳ)
If we differentiate (4-7.78) with respect to r and substitute the value obtained
for dUi/dr in (4.7.81), then we obtain zq. Finally, if we substitute in (4-7.76)
the expression found for и, and u2, then we obtain the sought function:

и 1 (r« _ M 1 Г - hda
2лг0 L
2 Jo rz + rl — 2rr0 cos (a — 0)
f2* g[r0 — r cos (a — 0)]da ~|
Jo [r2 + rl — 2 rr0 cos (a — 0)f J '
 APPENDIX

TABLES OF ERROR INTEGRALS AND SOME CYLINDRICAL FUNCTIONS

In the following pages we list tables of some special functions used for
the solution of boundary-value problems of mathematical physics. These tables
consist of a compilation of the simple properties of the functions considered.

The error integral

The error integral is defined by

Ф(г)
2 f -.2
da
"v^ Jo*
For small values of z, it has the series development
3 5
z z
m = v~Jz 1!3 + 2!5

An asymptotic representation for large values of Z

If we introduce а new variable p in the integral

1 - 0(Z) = Д» е~*гda
V n )z
by means of a = z + (/9/2z), then we obtain
1. e~z
e-*2 f°° g-p-tp2/4*2, dp .
1 - Ф(г) =
T/tt Z Jo
Consequently
,. 1 - 0(Z) ,
lim-r——r
z-*oo 1 £, z
= i;
~[/ n z
ье.,
1 е~*г
1 -Ф(2)~4=— .
Vл z
If we develop e p2/4z2 in the series

в-*/.* = у lull! £1
»=o (2z)2n n\ ’

multiply this series with e~p and integrate from zero to infinity, then we
obtain the divergent series
“ (-1)" (2м)! = “ (-1Г(м + 1)(m + 2)---2m
nt0(2z)2n n\ nt0 (2z)2"
364 APPENDIX

This divergent series represents the asymptotic development of the function

1 - Ф(2):

1 <r*2 Л 1 3-4 4-5-6

1 - 0(2)
\/к z \ 2z2 ^ (2z)* (2г)6

CYLINDRICAL FUNCTIONS

Series Asymptotic representation

1. Bessel functions

T(xs-i ШЛх№ /„W = v/|cos(x-i)+.”

Jo(x) - 1 1! + (2! )2
= X_ Г1 _ (x
(хІ2)г
/,(*)
2 L 1 2 /iw = /Jes1"(x_t)+-
(ХІ2)*
+ 1-2-2-3
v 1
AW = /|cos(x-|-.-i)+...
' 2 ) Г(у + 1)

2. Neumann functions

№) = у Jo(x) (in у + c) N0(x) = n/A sin (x — -^Л H-

гл-* \ 4 /

7Г

№) =-—
nx jv'w = -/Icos(^t)+--’

+ у /і(х) (in у + с) + • • •

ІѴп(х) = - у (y)"(n -!)!+•••

(n > 1)
(C — 0.577215... is the Euler constant.)

3. Hankel functions

H^\x) = Jt,(x) + iNv(x) Hll\x) =Дг'Н*-т)+-

У nx

Hl2)(x) = Mx) - ІЩх) НІг)(х) = JLe-i'-Tv“t) + • • •

¥ nx
CYLINDRICAL FUNCTIONS 365

4. Functions with imaginary arguments

I Ax) = J0(ix) +■
ш = /У7
= i , (x/2f (xffl
1! + (2!)2
I Ax) = — iJAix) №) = /2Хе-
_ x Г1 (*/2)2 (ХІ2У A
' 2 L Г 1-2 1-2-2-3 J
Iy(x) = (-ifMix) +■
ад = /гБ
1
2 / Г(ѵ + 1)

K0(x) = -J ІНИ(іх) /вд + ■

= /г/'

— — (In ~2 + CJIM + y~2 ) +'

К Ax) = Hll)(ix) = — + №) = 1/iLe“I+'

2 x г 2х

км = ^леп/г)лпіНИ(іх) Я»(*) = л/£-е-х +■

5. Functions of one-half order

/ О
/ілМ -л — sin л:; J-і/Ах) = — cos x ;
г хх t 7ГХ

г , ч
/з /Ах) =
НЕ( sin*
— (-cos X /_,дМ = 1/І(-^-5іп.Л
jtt\ л: Г 7ГХ V X /

6. Recursion formulas

d \Mx) 1 _ _ At i(x)
-y-[*7v(*)] - x7v-i(*) and
ax rfx L xv J x
2v
Jy-Ax) + Jy+Ax) = — Jy(x) ,
X

Jo(x) = — JAx) , jAx)dx = 1 - JAx) ,

-J-[xJAx)] = xjo(x) , J xj0(x)dx = xJAx) .

ax

Analogous formulas are also valid for other cylindrical functions with
real arguments.
For functions with an imaginary argument we have:
366 APPENDIX

/ѵ-iW - /ѵ+і(ж) = — Л(*) , Кѵ-і(х) - Кѵ+ М) --КМ) ,

X
Іо(х) = /і(х) , К'М) = - КМ) .

7. The Wronski determinant for the cylindrical function

Mx)Nl(x) - Ату(х)Л(х) = — ;
nx
In particular,
2_
Jo(x)NM) - N0{x)JM =
nx

8. Integral formulas

1 f*
Jn(x) = — \ e~ix sin ѵ*іпіе dip
2л J —*

(-»)" f it eix cos <p+in<p dy = л-i

(— {У1 fit
exxCOS(p cos n<pd<p ,
2 л- ] — n ^ J0
— * cosh £—tn£ yjt •

In particular,

K0(x) =

9. Integrals which contain Bessel functions

j %_х7.(^) dX = v -г (г > 0) ,
+ 2‘

oo -</KZ-ki |,| ikVp2 + г2 ikr

wp) ; ,--XdX = > ..... -
0 ^ l//!2 - k2 Vp1 + Z2

10. Differential equations which lead to Bessel's differential

equation

/' + ±y-(l + JL), = 0; у - aim + в km) ;

y" + xy = 0 ; у = VxZui(^ i/*3) ;

y" + =0 ; У — л/xZl/M+2)(m 2~l//x m+2^ .

Here Z denotes а solution of the Bessel differential equation.
CYLINDRICAL FUNCTIONS 367

TABLE 1. The Error Integral

<p(z) = -|=Г
Vn Jo
da 0 ^ z ^ 2.8

z m z 0(z) z 0(z) z Ф(2)

0.00 0.0000 0.40 0.4284 0.80 0.7421 1.20 0.9103

0.01 0.0113 0.41 0.4380 0.81 0.7480 1.21 0.9130
0.02 0.0226 0.42 0.4475 0.82 0.7538 1.22 0.9155
0.03 0.0338 0.43 0.4569 0.83 0.7595 1.23 0.9181
0.04 0.0451 0.44 0.4662 0.84 0.7651 1.24 0.9205

0.05 0.0564 0.45 0.4755 0.85 0.7707 1.25 0.9229

0.06 0.0676 0.46 0.4847 0.86 0.7761 1.26 0.9252
0.07 0.0789 0.47 0.4937 0.87 0.7814 1.27 0.9275
0.08 0.0901 0.48 0.5027 0.88 0.7867 1.28 0.9297
0.09 0.1013 0.49 0.5117 0.89 0.7918 1.29 0.9319

0.10 0.1125 0.50 0.5205 0.90 0.7969 1.30 0.9340

0.11 0.1236 0.51 0.5292 0.91 0.8019 1.31 0.9361
0.12 0.1348 0.52 0.5379 0.92 0.8068 1.32 0.9381
0.13 0.1459 0.53 0.5465 0.93 0.8116 1.33 0.9400
0.14 0.1569 0.54 0.5549 0.94 0.8163 1.34 0.9419

0.15 0.1680 0.55 0.5633 0.95 0.8209 1.35 0.9438

0.16 0.1790 0.56 0.5716 0.96 0.8254 1.36 0.9456
0.17 0.1900 0.57 0.5798 0.97 0.8299 1.37 0.9473
0.18 0.2009 0.58 0.5879 0.98 0.8342 1.38 0.9490
0.19 0.2118 0.59 0.5959 0.99 0.8385 1.39 0.9507

0.20 0.2227 0.60 0.6039 1.00 0.8427 1.40 0.9523

0.21 0.2335 0.61 0.6117 1.01 0.8468 1.41 0.9539
0.22 0.2443 0.62 0.6194 1.02 0.8508 1.42 0.9554
0.23 0.2550 0.63 0.6270 1.03 0.8548 1.43 0.9569
0.24 0.2657 0.64 0.6346 1.04 0.8586 1.44 0.9583

0.25 0.2763 0.65 0.6420 1.05 0.8624 1.45 0.9597

0.26 0.2869 0.66 0.6494 1.06 0.8661 1.46 0.9611
0.27 0.2974 0.67 0.6566 1.07 0.8698 1.47 0.9624
0.28 0.3079 0.68 0.6633 1.08 0.8733 1.48 0.9637
0.29 0.3183 0.69 0.6708 1.09 0.8768 1.49 0.9649

0.30 0.3286 0.70 0.6778 1.10 0.8802 1.50 0.9661

0.31 0.3389 0.71 0.6847 1.11 0.8835 1.51 0.9671
0.32 0.3491 0.72 0.6914 1.12 0.8868 1.6 0.9763
0.33 0.3593 0.73 0.6981 1.13 0.8900 1.7 0.9838
0.34 0.3694 0.74 0.7047 1.14 0.8931 1.8 0.9891

0.35 0.3794 0.75 0.7112 1.15 0.8961 1.9 0.9928

0.36 0.3893 0.76 0.7175 1.16 0.8991 2.0 0.9953
0.37 0.3992 0.77 0.7238 1.17 0.9020 2.1 0.9970
0.38 0.4090 0.78 0.7300 1.18 0.9048 2.2 0.9981
0.39 0.4187 0.79 0.7361 1.19 0.9076 2.3 0.9989

2.4 0.9993
2.5 0.9996
2.6 0.9998
2.7 0.9999
2.8 0.9999
368 APPENDIX

TABLE 2. Values of Bessel Functions of the Zero and

First Order from x = 0 to x = 12.00

X Jo(x) Hx) X Jo(x) Mx) X Mx) Mx)

0.00 + 1.000 + 0.000 4.00 -0.397 -0.066 8.00 + 0.172 + 0.235

0.10 + 0.997 + 0.050 4.10 -0.389 -0.103 8.10 +0.148 + 0.248
0.20 + 0.990 + 0.099 4.20 -0.377 -0.139 8.20 + 0.122 + 0.258
0.30 + 0.977 + 0.148 4.30 -0.361 -0.172 8.30 + 0.096 + 0.266
0.40 + 0.960 + 0.196 4.40 -0.342 -0.203 8.40 + 0.069 + 0.271

0.50 + 0.938 + 0.242 4.50 -0.321 -0.231 8.50 + 0.042 + 0.273

0.60 + 0.912 + 0.288 4.60 -0.296 -0.257 8.60 + 0.015 + 0.273
0.70 + 0.881 + 0.329 4.70 -0.269 -0.279 8.70 -0.013 + 0.270
0.80 + 0.846 + 0.369 4.80 -0.240 -0.298 8.80 -0.039 + 0.264
0.90 + 0.808 + 0.406 4.90 -0.210 -0.315 8.90 -0.065 + 0.256

1.00 + 0.765 + 0.440 5.00 -0.178 -0.328 9.00 -0.090 + 0.245

1.10 + 0.720 + 0.471 5.10 -0.144 -0.337 9.10 -0.114 + 0.232
1.20 + 0.671 + 0.498 5.20 -0.110 -0.343 9-20 -0.137 + 0.217
1.30 + 0.620 + 0.522 5.30 -0.076 -0.346 9.30 -0.158 + 0.200
1.40 + 0.567 + 0.542 5.40 -0.041 -0.345 9.40 -0.177 + 0.182

1.50 + 0.512 + 0.558 5.50 -0.007 -0.341 9.50 -0.194 + 0.161

1.60 + 0.455 + 0.570 5.60 + 0.027 -0.334 9.60 -0.209 + 0.140
1.70 + 0.398 + 0.578 5.70 + 0.060 -0.324 9.70 -0.222 + 0.117
1.80 + 0.340 + 0.582 5.80 + 0.092 ^0.311 9.80 -0.232 + 0.093
1.90 + 0.282 + 0.581 5.90 + 0.122 -0.295 9.90 -0.240 + 0.068

2.00 + 0.224 + 0.577 6.00 + 0.151 -0.277 10.00 VO.246 + 0.043

2.10 + 0.167 + 0.568 6.10 + 0.177 -0.256 10.10 -0.249 + 0.018
2.20 + 0.110 + 0.556 6.20 + 0.202 -0.233 10.20 -0.250 -0.007
2.30 + 0.056 + 0.540 6.30 + 0.224 -0.208 10.30 -0.248 -0.031
2.40 + 0.002 + 0.520 6.40 + 0.243 -0.182 10.40 -0.243 -0.055

2.50 -0.048 + 0.497 6.50 + 0.260 -0.154 10.50 -0.237 -0.079

2.60 -0.097 + 0.471 6.60 + 0.274 -0.125 10.60 -0.228 -0.101
2.70 -0.142 + 0.442 6.70 + 0.285 -0.095 10.70 -0.216 -0.122
2.80 -0.185 + 0.410 6.80 + 0.293 -0.065 10.80 -0.203 -0.142
2.90 -0.224 + 0.375 6.90 + 0.298 -0.035 10.90 -0.188 -0.160

3.00 -0.260 + 0.339 7.00 +0.300 -0.005 11.00 -0.171 -0.177

3.10 -0.292 + 0.301 7.10 + 0.299 + 0.025 11.10 -0.153 -0.191
3.20 -0.320 + 0.261 7.20 + 0.295 + 0.054 11.20 -0.133 -0.204
3.30 -0.344 + 0.221 7.30 + 0.288 + 0.083 11.30 -0.112 -0.214
3.40 -0.364 + 0.179 7.40 + 0.279 + 0.110 11.40 -0.090 -0.222

3.50 -0.380 + 0.137 7.50 + 0.266 + 0.135 11.50 -0.068 -0.228

3.60 -0.392 + 0.095 7.60 + 0.252 + 0.159 11.60 -0.045 -0.232
3.70 -0.399 + 0.054 7.70 + 0.235 + 0.181 11.70 -0.021 -0.233
3.80 -0.403 + 0.013 7.80 + 0.215 + 0.201 11.80 + 0.002 -0.232
3.90 -0.402 -0.027 7.90 + 0.194 + 0.219 11.90 +0.025 -0.229

12.00 + 0.048 -0.223

CYLINDRICAL FUNCTIONS 369

FIG. 75. Graphical Representation of the Bessel Functions or

Cylindrical Functions of the First Type J0(x) and Ji(x).

FIG. 76. Graphical Representation of the Neumann Functions or

Cylindrical Functions of the second type No(x) and Ni(x).

TABLE 3

Roots of the Equation /0(/O = 0 and the corresponding values of |

n l^n Jl(//n) n JlOn)

1 2.4048 0.5191 6 18.0711 0.1877

2 5.5201 0.3403 7 21.2116 0.1733
3 8.6537 0.2715 8 24.3525 0.1617
4 11.7915 0.2325 9 27.4935 0.1522
5 14.9309 0.2065 10 30.6346 0.1442
370 APPENDIX

TABLE 4. Values of the Functions K0{x) and Ki(x)

X Ko(x) Щх) X Ko(x) Щх)

0.1 2.4271 9.8538 3.1 0.0310 0.0356

0.2 1.7527 4.7760 3.2 0.0276 0.0316
0.3 1.3725 3.0560 3.3 0.0246 0.0281
0.4 1.1145 2.1844 3.4 0.0220 0.0250
0.5 0.9244 1.6564 3.5 0.0196 0.0222

0.6 0.7775 1.3028 3.6 0.0175 0.0198

0.7 0.6605 1.0503 3.7 0.0156 0.0176
0.8 0.5653 0.8618 3.8 0.0140 0.0157
0.9 0.4867 0.7165 3.9 0.0125 0.0140
1.0 0.4210 0.6019 4.0 0.0112 0.0125

1.1 0.3656 0.5098 4.1 0.0098 0.0111

1.2 0.3185 0.4346 4.2 0.0089 0.0099
1.3 0.2782 0.3725 4.3 0.0080 0.0089
1.4 0.2437 0.3208 4.4 0.0071 0.0079
1.5 0.2138 0.2774 4.5 0.0064 0.0071

1.6 0.1880 0.2406 4.6 0.0057 0.0063

1.7 0.1655 0.2094 4.7 0.0051 0.0056
1.8 0.1459 0.1826 4.8 0.0046 0.0051
1.9 0.1288 0.1597 4.9 0.0041 0.0045
2.0 0.1139 0.1399 5.0 0.0037 0.0040

2.1 0.1008 0.1227 5.1 0.0033 0.0036

2.2 0.0893 0.1079 5.2 0.0030 0.0032
2.3 0.0791 0.0950 5.3 0.0027 0.0029
2.4 0.0702 0.0837 5.4 0.0024 0.0026
2.5 0.0623 0.0739 5.5 0.0021 0.0023

2.6 0.0554 0.0653 5.6 0.0019 0.0021

2.7 0.0492 0.0577 5.7 0.0017 0.0019
2.8 0.0438 0.0511 5.8 0.0015 0.0017
2.9 0.0390 0.0453 5.9 0.0014 0.0015
3.0 0.0347 0.0402 6.0 0.0012 0.0013
 LITERATURE REFERENCES OF THE EDITOR

In the following, there is no claim of completeness. Textbooks, monographs, and

collected reports, but no journal articles, are given. Special works from journals are
given in part in the footnotes of this book. As a rule, only literature after 1900 has
been used.

LITERATURE ON THE THEORY OF PARTIAL DIFFERENTIAL EQUATIONS

OF HIGHER ORDER

Albrecht, R., and Hochmuth, H. Exercises in Higher Mathematics. Part III. R. Olden-
bourg: Munich, 1956.
Bateman, H. Partial Differential Equations of Mathematical Physics. Dover Publica¬
tions: New York, 1944.
Baule, B. “Mathematics for Scientists and Engineers,” in Partial Differential Equations.
Volume VI, 5th edition. S. Hirzel: Leipzig, 1955.
Bergmann, S., and Schiffer, M. Kernel Functions and Differential Equations in Mathe¬
matical Physics. Academic Press, Inc.: New York, 1953.
Bernstein, D. L. Existence Theorems in Partial Differential Equations. 2nd printing.
Princeton University Press: Princeton, 1951.
Bieberbach. L. Introduction to the Theory of Differential Equations in the Real Domain.
Springer Verlag: Berlin, 1956.
-. Theory of Differential Equations. 3rd printing. Springer Verlag: Berlin, 1930.
Borgnis, F. E. and Papas, С. H. Boundary-value Problems in Microwave Physics-
Springer Verlag: Berlin, 1955.
Budak, В. M., Samarskii, A. A., and Tychonoff, A. N. A Collection of Problems for
Mathematical Physics. Gostechisdat: Moscow, 1956.
Collatz, L. Numerical Treatment of Differential Equations. 2nd printing. Springer
Verlag: Berlin, 1955.
-. Eigenvalue Problems with Technical Applications. Academic Verlagsgesells-
chaft: Leipzig, 1949.
-. “Numerical and Graphical Methods,” in Handbook of Physics. Volume II,
Mathematical Methods, Part II. Springer Verlag: Berlin, 1955.
First Conference on Equations with Partial Derivatives. Louvain, 1953 (CBRM). George
Thone: Liege, 1954.
Second Conference on Equations with Partial Derivatives. Bruxelles, 1954 (CBRM).
George Thone: Liege, 1955.
Contributions to the Theory of Partial Differential Equations. Princeton University
Press: Princeton, 1954.
International Conference on Linear Equations with Partial Derivatives. Trieste, 1954;
Edizioni Cremonese, Rome, 1955.
Courant, R., and Hilbert, D. Methods of Mathematical Physics. Volume I, 2nd printing;
Volume II, 1st printing. Springer, Verlag: Berlin, 1931 and 1937, respectively.
Courant, R., and Schiffer, M. Dirichlet's Principle: Conformal Mapping and Minimal
Surfaces. Interscience Publishers, Inc.: New York, 1950.
Duff, G. F. D. Partial Differential Equations. University of Toronto Press: Toronto,
1956.
Duschek, A. Lectures on Higher Mathematics. Volume III. Springer Verlag: Vienna, 1953.
372 LITERATURE REFERENCES OF THE EDITOR

Frank, Ph., and R. v. Mises. Differential and Integral Equations of Mechanics and
Physics. Part I, 2nd printing; Part II, 2nd printing. F. Vieweg: Braunschweig,
1930 and 1935, respectively.
Ford, L. R. Differential Equations. 2nd edition. McGraw-Hill Book Co., Inc.: New York,
1955.
Forsyth, A. R., and Jacobsthal, W. Textbook of Differential Equations. (Translated
from the English.) 2nd edition. F. Vieweg: Braunschweig, 1912.
Freda, H. The Method of Characteristics for the Integration of Linear Hyperbolic
Partial Differential Equations. Mem. sci. math., fasc 84. Gauthier-Villars: Paris,
1937.
Friedman, B. Principles and Techniques of Applied Mathematics. John Wiley & Sons,
Inc.: New York, 1956.
Geronimus, J. L., and Steklov, V. A. Integration of the Differential Equations of
Mathematical Physics. (Translated from the Russian.) Verlag Technik: Berlin,
1954.
Godeaux, L. Mathematical Analysis. Volume III. Sciences and Letters: Liege, 1947.
Gosse, R. The Method of Darboux for the Equation s = f(x, y, z, p, q). Mem. sci.
math., fasc 12. Gauthier-Villars: Paris, 1926.
Goursat, Ed. Lessons on the Integration of Equations with Partial Derivatives of the
Second Order in Two Independent Variables. 2 vols. A. Hermann: Paris, 1896
and 1898, respectively.
Guldberg, A. “Partial and Total Differential Equations,” in E. Pascal, Repertory of
Higher Mathematics. Volume I, Part II, 2nd printing. B. G. Teubner: Leipzig,
1927.
Gunter, N. M., and Kusmin, R. O. Collection of problems in Higher Mathematics. Volume
II. (Translated from the Russian.) VEB Deutscher Verlag der Wissenschaften:
Berlin, 1957.
Hadamard, J. Lectures on Cauchy's Problem in Linear Partial Differential Equations.
Reprinting of the 1st edition of 1923. Dover Publications: New York, 1952.
-. Cauchy's Problem and Linear Hyperbolic Partial Differential Equations.
(Revised translation of the preceding title.) Hermann et Cie: Paris, 1932.
Hahn, H., Lichtenstein, L., and Lense, J. “The theory of Integral Equations and Func¬
tions of Infinitely Many Variables and Their Application to Boundary-value
Problems for Ordinary and Partial Differential Equations,” in E. Pascal, Repertory
of Higher Mathematics. Volume I, Part III, 2nd edition. B. G. Teubner: Leipzig,
1929.
Heilbronn, G. Integration of Equations with Partial Derivatives of the Second Order
by the Method of Drach. Mem. sci. math., fasc. 129. Gauthier-Villars: Paris,
1955.
Hildebrand, F. B. Methods of Applied Mathematics. Prentice-Hall, Inc.: New York,
1952.
Hoheisel, G. Collection of Problems in Ordinary and Partial Differential Equations.
2nd edition. W. de Gruyter: Berlin, 1952.
-. Partial Differential Equations. 3rd edition. W. de Gruyter: Berlin, 1953.
Horn, J. Partial Differential Equations. 4th edition. W. de Gruyter: Berlin, 1949.
Hort, W., and Thoma, A. Differential Equations of Technology and Physics. 6th edi¬
tion. J. A. Barth: Leipzig, 1954.
Janet, M. Lessons on Systems of Equations with Partial Derivatives. Gauthier-Villars:
Paris, 1929.
-. Systems of Equations with Partial Derivatives. Mem. sci. math., fasc. 21.
Gauthier-Villars: Paris, 1927.
LITERATURE ON THE THEORY OF PARTIAL DIFFERENTIAL EQUATIONS OF HIGHER ORDER 373

John, F. Plane Waves and Spherical Means Applied to Partial Differential Equations.
Interscience Publishers: New York, 1955.
Julia, G. Exercises in Analysis. Volume IV. Gauthier-Villars: Paris, 1935.
Катке, E. Differential Equations of Real Functions. 3rd edition. Akademische Verlags-
gesellschaft: Leipzig, 1956.
Kantorowitsch, L. W., and Krylov, W. I. Approximate Methods of Higher Analysis.
(Translated from the Russian.) VEB Deutscher Verlag der Wissenschaften:
Berlin, 1956.
Krylov, A. N. On Certain Differential Equations of Mathematical Physics. 5th edition.
Gostechisdat: Moscow, 1950.
Kupradse, W. D. Boundary-value Problems of Vibration Theory and Integral Equa¬
tions. (Translated from the Russian.) VEB Deutscher Verlag der Wissenschaften:
Berlin, 1956.
Ladyzhenskaya, 0. A. Mixed Problems of the Theory of Hyperbolic Partial Differential
Equations. Gostechisdat: Moscow, 1953.
Lense, J. “Partial Differential Equations,” in Handbook of Physics. Volume I, Mathe¬
matical Methods, Part I. Springer Verlag: Berlin, 1956.
Leray, J. Hyperbolic Differential Equations. Institute for Advanced Study: Princeton,
1955.
Lewin, W. I., and Grosberg, J. I. Differential Equations of Mathematical Physics.
(Translated from the Russian.) Verlag Technik: Berlin, 1952.
Lichtenstein, L. “New Developments in the Theory of Partial Differential Equations of
the Second Order of the Elliptic Type,” in Encyclopedia of Mathematical
Sciences. Volume II, Part III, Second half. B. G. Teubner: Leipzig, 1924.
Love, A. E. H. Textbook of Elasticity. (Translated from the English.) B. G. Teubner:
Leipzig, 1907.
Madelung, E. Mathematical Aids for Physicists. 6th edition. Springer Verlag: Berlin,
1954.
Malkin, I. G. The Methods of Liapunov and Poincare in the Theory of Non-Linear
Vibrations. Gostechisdat: Moscow, 1949.
Margenau, H., and Murphy, G. M. The Mathematics of Physics and Chemistry. 2nd
edition. D. van Nostrand Co., Inc.: New York, 1956.
Miller, F. H. Partial Differential Equations. 6th reprint. John Wiley & Sons, Inc.:
New York, 1953.
Miller, K. S. Partial Differential Equations in Engineering Problems. Prentice-Hall,
Inc.: New York, 1953.
Miller, N. A First Course of Differential Equations. University Press: Oxford, 1935.
Milne, W. E. Numerical Solutions of Differential Equations. John Wiley & Sons, Inc.:
New York, 1953.
Miranda, C. Equations with Partial Derivatives of the Elliptic Type. Springer Verlag:
Berlin, 1955.
Morse, P. M., and Feshbach, H. Methods of Theoretical Physics. 2 vols. McGraw-Hill
Book Co., Inc.: New York, 1953.
Muskhelishvili, N. I. Some Fundamental Problems of the Theory of Elasticity. 4th edi¬
tion. Akademie-Verlag: Moscow, 1954. (This work contains a complete list of
references which primarily give an insight into the Soviet literature in this field.)
Page, С. H. Physical Mathematics. D. van Nostrand Co., Inc.: New York, 1955.
Panov D. J. A Collection of Formulas for the Numerical Treatment of Partial Dif¬
ferential Equations by Difference Methods. (Translated from the Russian.)
Akademie-Verlag: Berlin, 1955.
374 LITERATURE REFERENCES OF THE EDITOR

Pertowski, I. G. Lectures on Partial Differential Equations■ (Translated from the Rus¬

sian.) B. G. Teubner: Leipzig, 1955.
Picard, E. Lessons on Certain Types of Simple Equations with Partial Derivatives
with Some Application to Mathematical Physics. Nouveau tirage. Gauthier-
Villars: Paris, 1950.
Proceedings of the Conference on Differential Equations (Dedicated to A. Weinstein.)
University of Maryland Book Store: College Park, Md., 1956.
Purday, H. F. P. Linear Equations in Applied Mechanics. Interscience Publishers,
Inc.: New York, 1954.
Rashevski, P. K. Geometrical Theory of Partial Differential Equations. Gostechisdat:
Moscow, 1947.
Reed, M. B., and Reed, G. B. Mathematical Methods in Electrical Engineering. Harper
& Bros.: New York, 1951.
Riquier, C. The Method of Majorant Functions and Systems of Equations with Partial
Derivatives. Mem. sci. math., fasc. 32. Gauthier-Villars: Paris, 1928.
-. Systems of Equations with Partial Derivatives. Gauthier-Villars, Paris, 1910.
Rothe, R. Higher Mathematics. Part III, 8th edition; Part IV, No. 5/6, 7th edition. B. G.
Teubner: Leipzig, 1954 and 1955, respectively. Part VI, I. Szabo and B. G.
Teubner: Stuttgart, 1953.
Runge, C. and Willers, F. A. Numerical and Graphical Quadrature and the Integra¬
tion of Ordinary and Partial Differential Equations. B. G. Teubner: Leipzig,
1915.
Sauer, R. Initial Value Problems of Partial Differential Equations. Springer Verlag:
Berlin, 1952.
Sauter, F. Differential Equations of Physics. 2nd edition. W. de Gruyter: Berlin, 1950.
Schlogl, F. “Boundary-value Problems,” in Handbook of Physics. Volume I, Mathe¬
matical Methods, Part I. Springer Verlag: Berlin, 1956.
Schwank, F. Boundary-value Problems. B. G. Teubner: Leipzig, 1951.
Smirnov, M. M. Problems in the Partial Differential Equations of Mathematical Physics.
(Translated from the Russian.) VEB Deutscher Verlag der Wissenschaften:
Berlin, 1955.
Smirnov, V. I. Textbook of Higher Mathematics. (Translated from the Russian.) Part
II (Chapter VII), 2nd edition; Part IV (This book is included in the text (Chapter
III) in the valuable references from the original work.) 1st edition. VEB Deutscher
Verlag der Wissenschaften: Berlin, 1958.
Sneddon, I. N. Elements of Partial Differential Equations. McGraw-Hill Book Co.,
Inc.: New York, 1957.
Sobolev, S. L. Equations of Mathematical Physics. 3rd edition. Gostechisdat: Moscow,
1954.
Sommerfeld, A. “Boundary-value Problems in the Theory of Partial Differential Equa¬
tions,” in Encyclopedia of Mathematical Sciences. Volume II, Part I, first half.
B. G. Teubner: Leipzig, 1900.
-. “Partial Differential Equations of Physics,” in Lectures on Theoretical Physics-
Volume VI, 4th edition. Akademische Verlagsgesellschaft: Leipzig, 1957.
Stepanov, V. V. Textbook of Differential Equations. (Translated from the Russian.)
VEB Deutscher Verlag der Wissenschaften: Berlin, 1956.
Sternberg, W. “The Theory of Boundary-value Problems in the Domain of Partial Dif¬
ferential Equations,” in E. Pascal, Repertory of Higher Mathematics. Volume
I, Part III, 2nd edition. B. G. Teubner: Leipzig, 1929.
Transactions of the Symposium on Partial Differential Equations. University of Cali-
LITERATURE FOR THE POTENTIAL THEORY (CHAPTER 4, § 5 OF THIS BOOK): 375

fornia, Berkeley, 1955. Interscience Publishers, Inc.: New York, 1955.

Tricomi, F. G. Lessons on Equations with Partial Derivatives. Gherardi: Torino, 1954.
Vekua, I. N. New Methods for the Solution of Elliptic Differential Equations.
Gostechisdat: Moscow, 1948.
Weber, E. von “Partial Differential Equations,” in Encyclopedia of Mathematical Sciences.
Volume II, Part I, first Half. B. G. Teubner: Leipzig, 1900.
Webster, A. G. and Szego, G., Partial Differential Equations of Mathematical Physics.
(Translated from the English.) B. G. Teubner: Leipzig, 1930.

LITERATURE FOR THE POTENTIAL THEORY (CHAPTER 4, § 5 OF THIS

BOOK)

In addition to the books already enumerated, the theory is explained in the following
works.
Burkhardt, H., and Meyer, F., “Potential Theory,” in Encyclopedia of Mathematical
Sciences. Volume II, Part I, first Half. B. G. Teubner: Leipzig, 1900.
Gunter, N. M. Potential Theory and its Application to the Basic Problems of Mathe¬
matical Physics. (Translated from the Russian.) B. G. Teubner: Leipzig, 1957.
Kellogg, O. D. Foundations of Potential Theory. Springer, Verlag: Berlin, 1929.
Korn, A. Five Treatises on the Potential Theory. F. Dummler: Berlin, 1902.
Liapunov, A. M. Studies in Potential Theory. Gostechisdat: Moscow, 1949.
Lichtenstein, L. “New Developments in Potential Theory,” in Encyclopedia of Mathe¬
matical Sciences. Volume II, Part III, first Half. B. G. Teubner: Leipzig, 1909.
Neumann, E. R. Studies of the Methods of C. Neumann and G. Robin for the Solution
of the Two Boundary-value Problems of Potential Theory. B. G. Teubner:
Leipzig, 1905.
Plemelj, I. Potential Theoretic Investigations. B. G. Teubner: Leipzig, 1911.
Rothe, R. Higher Mathematics. Part VII by W. Schmeidler (also treats potential theory.)
B. G. Teubner: Stuttgart, 1956.
Sternberg, W. Potential Theory. 2 vols. W. de Gruyter & Co.: Berlin, 1926.
Sternberg, W., and Smith, Turner L. The Theory of Potential and Spherical Harmonics.
3rd edition. University of Toronto Press: Toronto, 1952.

LITERATURE ON THE THEORY OF SPECIAL FUNCTIONS (APPENDIX

OF THIS BOOK)

Besides the references given above, special functions are also discussed in the fol¬
lowing works. Tables and Tabular work will not be included in the reference list. We
have used L. M. Ryshik, and I. S. Gradstein: Tables (Translated from the Russian),
VEB Deutscher Verlag der Wissenschaften, Berlin, 1957, as a reference for special values
of these functions. This book contains a list of references which gives further references
to tables.
Angelesco, A. On Some Polynomials Generalizing the Polynomials of Legendre and
Hermite and the Approximate Calculation of Multiple Integrals. Gauthier-
Villars: Paris, 1916.
Appell, P., and de Feriet, J. Kampe. Hypergeometric and Hyperspherical Functions'.
Hermite Polynomials. Gauthier-Villars: Paris, 1926.
Bateman, H. Higher Transcendental Functions. 2 vols. McGraw-Hill Book Co., Inc.:
New York, 1953 and 1955.
376 LITERATURE REFERENCES OF THE EDITOR

Bowman, F. Introduction to Bessel Functions. Longmans, Green and Co.: New York,
1938.
Goudet, G. Bessel Functions and Their Application in Physics. 2nd edition. Masson
et Cie.: Paris, 1954.
Graf, J. H., and Gubler, H. Introduction to the Theory of Bessel Functions. 2 Parts,
2nd edition. K. J. Wyss: Bern, 1900.
Gray, A., Mathews, G. B., and MacRobert, T. M. A Treatise on Bessel Functions and
Their Applications to Physics. 2nd edition. Macmillan: London, 1922.
Hilb, E. “Spherical Functions, Bessel and Related Functions,” in E. Pascal, Repertory
of Higher Mathematics. Volume III, Part III, 2nd edition. B. G. Teubner: Leipzig,
1929.
Lagrange, R. Polynomials and Legendre Functions. Mem. sci. math., fasc. 97. Gauthier -
Villars: Paris, 1939.
Lebedev, N. N. Special Functions and Their Applications. Gostechisdat: Moscow,
1953.
Lense, J. Spherical Functions. Akademische Verlagsgesellschaft: Leipzig, 1950.
Magnus, W., and Oberhettinger, F. Formulas and Theorems for the Special Functions
of Mathematical Physics. 2nd edition. Springer Verlag: Berlin, 1948.
McLachlan, N. W. Bessel Functions for Engineers. 2nd edition. Clarendon Press:
Oxford, 1955.
Meixner, J. “Special Functions of Mathematical Physics,” in Handbook of Physics.
Volume I, Mathematical Methods, Part I. Springer Verlag: Berlin, 1956.
Nielsen, N. Handbook of the Theory of Cylindrical Functions. B. G. Teubner: Leipzig,
1904.
Petiau, G. The Theory of Bessel Functions. CNRS, Masson et Cie: Paris, 1955.
Prasad, G. A Treatise on Spherical Harmonics and the Functions of Bessel and Lame.
2 vols. Mahamandal Press: Benares, 1930 and 1932.
Relton. F. E. Applied Bessel Functions. Blackie & Son, Ltd.: London, 1946.
Schafheitlin, P. The Theory of Bessel Functions. B. G. Teubner, Leipzig, 1908.
Smirnov, V. I. Textbook of Higher Mathematics (Translated from the Russian.) Part III,
Section 2, Chapter VI. VEB Deutscher Verlag der Wissenschaften: Berlin, 1955.
Sonin, N. J. Investigations of Cylindrical Functions and Special Polynomials. Gostechis¬
dat: Moscow, 1954.
Wangerin, A. “Theory of Spherical Functions and Related Functions, in Particular the
Lame and Bessel Functions,” in Encyclopedia of Mathematical Sciences. Volume
II, Part I, Second Half. B. G. Teubner: Leipzig, 1904.
Watson, G. N. A Treatise on the Theory of Bessel Functions. University Press: Cam¬
bridge, 1922. (This book contains an enlarged literature list in which all publi¬
cations on the Bessel functions prior to 1922 are given.)
Weyrich, R. Cylindrical Functions and Their Applications. B. G. Teubner: Leipzig,
1937.
Whittaker, E. T. and Watson, G. N. A Course of Modern Analysis. 4th edition. Uni¬
versity Press: Cambridge, 1952.
 INDEX

Absorption, 140 weak, 234

Absorption isotherm, 141, 143 Coordinates, Euler, 15
Adiabatic equation, Hugoniot, 134 Coordinates, Lagrange, 15
Analogy method, 222 Current
Approximation, successive, 102, 327 irrotational, 242
Average-value theorem, 257 source-free, 242
Curve, Liapunov, 308
Basic problem of electrostatics, 335 Cylindrical coordinates, 247
Basic tone of a string, 61
Bessel function, 115, 364, 368 d’Alembert formula, 38, 60, 116
Boundary conditions, 26, 160 d’Alembert method, 36
of the first type, 27 <5 function, Dirac, 234
homogenous, 27 Derivative, substantive, 22
of the second type, 27 Determinant, Wronski, 366
of the third type, 27 Development thorem of V. A. Steklov, 94
Boundary-value problem, exterior, 242 Difference equation, Laplace, 326
exterior first, 265, 266 Differential equation, Cauchy-Riemann, 249
first, 28, 31, 46, 90, 162, 205, 242, 270 Einstein-Kolmogoroff, 231
general first, 83, 185 of the elliptic type, 4, 8
homogeneous, 171 of the first order, particular solution of,
mixed, 28 2
second, 28, 242 fundamental solution of Laplace’s, 248
solution of first, 163, 174, 319 homogeneous, 1
solution of first interior, 317 of the hyperbolic type, 4, 8
solution of general, 29 Laplace’s, 44, 241
solution of second, 257, 267, 317 linear, 1, 7
solution of special, 31 of the parabolic type, 4, 8
stable solution of 43 partial, 1
with stationary inhomogenenities, 84 Poisson, 241
third, 28, 242 quasilinear, 1
well set, 43 solution of Laplace’s, 44, 276, 288, 291
solution of the linear hyperbolic, 103,
Capacity, 337 108
Cauchy problem, 29, 162 of the ultrahyperbolic type, 8
Cauchy-Reimann differential equations, 249 of the vibrating string, 12, 113
Coefficient, Henry’s, 141 Differential operator, adjoint, 108
Coefficient of temperature conductivity, 156 self-adjoint, 108
Concentration, 157 Differential expression, Laplace, 325
Condensation of gases, 23 Diffusion equation, 157
Constant, Euler, 364 Dirac Л-function, 234
Conditions, Hugoniot, 131 Dirichlet’s problem, 242
Continuity condition, 23 Discontinuity, strong, 136
Continuity equation, 129 Discontinuity, weak, 136
Convergence Duhamel’s principal, 209
in the mean, 234
uniform, 234 Eigenfrequencies, 71
378 INDEX

Eigenfunctions, 68 third, 324

Eigenvalues, 68 Function
Eigenvalue problem, Sturm-Liouville, 68 analytic, 248
Einstein-Kolmogoroff differential equation, Bessel, 115, 364, 368
232 biharmonic, 359
Einstein-Kolmogoroff equation, 231 Dirac, 234
Elasticity coefficient, 150 Green’s, 175, 280, 286
Electronic, integrators, 329 Neumann’s, 364
Energy, internal, 130 Riemann’s, 109
Energy, kinetic, 130 Functions, conjugate harmonic, 250
Energy conservation law, 130 Fundamental formula for harmonic func¬
Equation tions, 255
biharmonic, 359
characteristic, 4 Gaussian integral theorem, 21
Einstein-Kolmogoroff, 231 Gerschgorin, 329
of longitudinal vibrating rods, 150 Goursat’s problem, 102
of state, 129 Gradient, geothermal, 220
of state of ideal gases, 137 Green’s formula, 22, 108
of state, thermodynamic, 23 first 254
Equations second, 254
of acoustics, 25 Green’s function, 175, 280, 286
of gas dynamics, 129 of infinite straight lines, 193
hydrodynamic, 23 Green’s law, 22
Error integral, 202 Gutenmacher, L. Ya., 330
Euler’s constant, 364
Euler coordinates, 15 Heat-conduction equation, 153, 155, 233
Existence thorem, 32 fundamental solution, 193
inhomogeneous, 182
Flageolet tones, 73 solution of, 162, 168, 195, 201, 203, 208,
Formula 210, 223
d’Alembert, 38, 60, 116 Heat content per unit of mass, 130
first Green’s, 253 Heat exchange coefficient, 156, 161
Green’s, 108 Heat function, 130
of Ostrogradski, 253 Henry’s coefficient, 141
second Green’s, 254 Henry’s isotherm, 141
Fourier, 210, 215 Hugoniot’s adiabatic equation, 134
Fourier coefficient, 76 Hugoniot’s conditions, 131
Fourier series, 70
Fourier’s integral, 89 Image, electrostatic, 283
Fourier’s law, 154 Impedance, characteristic, 152
first, 216 Inertia, moment of, 334
second, 216 Inertia law of a quadratic form, 8
third, 216 Initial conditions, 26
Fourier’s method, 66 Initial-value problem for temperature dis¬
Fredholm’s integral equation tribution, 162
of the first type, 317 Integral
of the second type, 317 Fourier’s, 89
Fredholm’s theorem improper, 292
first, 320 Poisson, 196, 276, 285, 287
second, 323 uniformly convergent, 295
INDEX 379

Integral equation Momentum equation, 129

adjoint, 318 Motion of an ideal fluid, 23, 129
of the first type, Fredholm, 317
of the second type, Fredholm, 317 Neighborhood, Liapunov, 308
Integral formula, 366 Neumann’s function, 364
Integral theorem, Gauss, 21 Neumann’s problem, 242
Integro-differential equation, 103 Newtonian potential, 289
Isotherm Nodes of standing waves, 71
Henry, 141 Norm of an eigenfunction, 126
Langmiur, 141, 147 Normalizing an eigenfunction, 96
Normalizing condition, 96
Joukowski, N. E., 357
Operator, Laplace, 24
Kelvin transformation, 251 Orthogonal system, normalized, 96
Kernel Orthogonality, weighted, 125
adjoint, 318 Orthogonality relation, 125
iterated, 321 Orthonormal system, 96
Poisson, 275 Overtones of a string, 72
Krylov, A. N.. 127
Phase planes, 39
Lagrange coordinates, 15 Phase velocity, 56
Langmuir isotherm, 141, 147 Point, conjugate, 283
Laplace’s difference equation, 325 Poisson differential equation, 241
Laplace’s differential equation, 44, 241 Poisson integral, 196, 275, 285, 287
fundamental solution of, 248 Poisson’s kernel, 275
solution of 44, 276, 288, 291 Porosity coefficient, 157
Laplace’s differential expression, 325 Potential, 218
Laplace operator, 24 logarithmic, 290
Law, Fourier’s 154 Newtonian, 289
first, 216 Potential
second, 216 of a double layer, 305
third, 216 of a force field, 289
Lebesgue, theorem of, 178 of a single layer, 304
Liapunov, A. M., 280, 282 Potential field, 243
Liapunov curve, 310 Potential flow of a source-free fluid, 242
Liapunov neighborhood, 308 Principal value of an integral, 303
Liapunov surface, 308 Principle, Duhamel’s, 209
Limit element, 235 Principle of the maximum, 165, 258
Probability, 230
Mass conservation law, 132 Problem
Maximum of a standing wave, 71 Cauchy’s, 29, 162
Method Dirichlet, 242
d’Alembert, 36 Goursat, 102
of electrostatic images 283 Neumann, 242
Fourier, 66 Problems
of successive approximation, 103 , 327 with initial conditions, 28
Modulus, Young’s, 16 without initial conditions, 29, 87, 162
Moment
of a dipole, 304 Reciprocity, principle of, 282
of inertia, 334 Recursion formula, 345
380 INDEX

Regularity at infinity, 263 of Sylvester, 8

Resonance, 93 third Fredholm, 324
Riemann function, 109 Tone, simple, 72
Transformation, Kelvin, 251
Separation of variables, 66, 93 Transformation by reciprocal radii, 251
Sequence, locally normalized, 235 Triangle, characteristics, 42
Series, Fourier, 70 Tschaplygin, S. A., 357
Shock waves, 131
Solidification problem, 226 Uniqueness theorem, 31, 167, 260, 265, 266,
solution of, 229 267, 269
Source functions, 175
Spherical coordinates, 246 Velocity potential, 24, 242
Steklov, V. A., 98 Vibration equation, 24
Steklov, development theorem of, 94 homogeneous, 45
Sturm-Liouville eigenvalue problem, 68 inhomogeneous, 78
Superposition principal, generalized, 74, solution of the homogeneous, 45
196 solution of the inhomogeneous, 78
Surface, Liapunov, 308
Sylvester, theorem of, 8 Wave
System of difference equations, 326 propagating, 39
standing, 71
Telegraphic equation, 18, 57, 149 Wave dispersion, 54, 56
Temperature conductivity, 160 lack of, 54
Temperature waves, 218 Wave equation, 149
Theorem Wronski determinant, 366
first Fredholm, 320
Green’s, 22 Young’s modulus, 16
of Lebesgue, 178
of Zemplen, 134 Zemplen, theorem of, 134
second Fredholm, 323
 Date Due
AUG -5

«rm
ЛШ'З' 1976

MflPg i«y
t
L
btB A

w V- >»

PRINTED IN U. S. A, NO. 23233

(«у CAT'
 QA401 .T513 V.1
Tiknonov A. N. (Andrei N 010101 000
Partial differential equations

63 0 27475 3
TRENT UNIVERSITY

QA401 .T513 v.l

иТікѢоШ.ѵ>- -

40337
QA Tikhonov, A N
401 Partial differential
T513 equations of mathematical
v.l physics

Trent
University