ae

Larry C. Andrews |
USED
 Larry C. Andrews
University of Central Florida

‘Scott, Foresman and Company

~ Glenview, Illinois.
Oakland, N.J. Palo Alto, Cal. Tucker, Ga. London, England
Dallas, Tex.
Library of Congress Cataloging in Publication Data

Andrews, Larry C.
Ordinary differential equations, with applications.

Includes bibliographical references and index.

1. Differential equations. I. Title.
QA372.A594 515.3'52 81-2127]
ISBN 0-673-15800-4 AACR2

ISBN: 0-673-15800-4

Copyright © 1982 Scott, Foresman and Company

All rights reserved.
Printed in the United States of America.

1234567 8-RRC-88 87 86 85 84 83 82 81
Preface, ix

Basic Concepts, 1

1.1 Introduction, 2
1.2 Classification of DEs, 2
1.2.1 Origin and Application of DEs, 4
1.3 Solutions of DEs, 6
1.3.1 General Solutions, 9
1.4 Initial and Boundary Value Problems, 14
References, 17

Solution Techniques for First-Order Equations, 18

2.1 Introduction, 19
2.2 Separation of Variables, 20
2.3 Miscellaneous Techniques, 25
2.3.1 Exact Equations, 25
2.3.2 Integrating Factors, 28
2.3.3 Linear Equations, 30
2.3.4 Homogeneous Equations, 31
*2.4 The Theory of Linear First-Order Equations, 36
2.4.1 The Homogeneous Equation, 37
*2.5 The Nonhomogeneous Linear Equation, 41
2.5.1 Physical Interpretations, 45
2.5.2 Existence-Uniqueness Theorem, 47
*2.6 Bernoulli’s Equation, 52
References, 54

*Sections marked with an asterisk may easily be omitted for a shorter course.
vi CONTENTS

Applications of First-Order Equations, 55

3.1 Introduction, 56
3.2 General Applications, 56
3.2.1 Orthogonal Trajectories, 56
3.2.2 Free-Falling Bodies, 59
3.2.3 Flow of Water Through an Orifice, 62
3.2.4 Curves of Linear Pursuit, 63
3.3. Applications Involving Linear Equations, 69
3.3.1 Growth and Decay Problems, 70
3.3.2 Motion of a Particle and Simple Electric Circuits, 73
3.3.3 Cooling and Mixing Problems, 76
References, 81

Linear Equations of Higher Order, 82

4.1 Introduction, 83
4.2 Linear Dependence and Independence, 84
4.2.1 Wronskians, 86
4.3 Constructing a Second Solution from a Known Solution, 91
>» 4.4 Homogeneous Second-Order Equations with Constant Coefficients, 94
4.4.1 Differential Operators, 98
4.5 Higher-Order Linear Equations, 101
4.5.1 Linear Independence, 102
4.5.2 Homogeneous Constant-Coefficient Equations, 104
4.6 The Nonhomogeneous Equation, 108
4.6.1 The Method of Undetermined Coefficients, 109
4.7 Variation of Parameters, 115
4.8 The Cauchy-Euler Equation, 121
References, 124

Applications Involving Initial Value Problems, 126

5.1 Introduction, 127

5.2 Small Motions of a Spring-Mass System, 128
5.2.1 Undamped Free Motion, 130
*5.2.2 The Pendulum Problem, 133
5.2.3 Damped Free Motion, 134
5.3. Forced Motions, 139
5.3.1 Undamped Forced Motion, 139
5.3.2 Damped Forced Motion, 141
5.4 Simple Electric Circuits, 146
*5.5 The Method of Green’s Functions, 150
5.5.1 The One-Sided Green’s Function, 152
 CONTENTS vii

5.5.2 A Table of Some One-Sided Green’s Functions, 155

*5.6 Impulse Functions, 159
5.6.1 A More General Definition of the Green’s Function, 161
*5.7 Elementary Oscillation Theory, 164
References, 170

6 The Laplace Transform, 171

6.1 Introduction, 172

6.2 The Laplace Transform of Some Elementary Functions, 172
6.3 Operational Properties, 178
6.3.1 The Laplace Transform of Derivatives and Integrals, 180
6.3.2 Derivatives and Integrals of Laplace Transforms, 182
6.4 Inverse Laplace Transforms, 187
6.4.1 Partial Fractions, 190
6.5 Solution of Initial Value Problems, 194
6.5.1 Spring-Mass Systems, 197
6.6 Discontinuous Functions, 201
*6.6.1 Impulse Functions, 207
6.7 The Convolution Theorem, 210
*6.7.1 The One-Sided Green’s Function, 213
6.8 Table of Some Laplace Transforms, 216
References, 218

7 Systems of Equations, 220

7.1 Introduction, 221

7.2 The Operator Method, 222
7.3. The Method of Laplace Transforms, 227
7.3.1 Coupled Spring-Mass Systems, 228
7.3.2 Electrical Networks, 230
7.4 First-Order Linear Systems, 234
7.4.1 Homogeneous Linear Systems with Constant Coefficients, a8)
7.4.2 Nonhomogeneous Linear Systems, 244
*7.5 Matrix Methods, 248
7.5.1 Homogeneous Equations, 249
7.5.2 Linear Equations with Constant Coefficients, 250
7.5.3 Fundamental Matrices, 252
7.5.4 Nonhomogeneous Equations, 254
*7.6 Nonlinear Systems and Stability, 258
7.6.1. The Phase Plane and Critical Points, 258
7.6.2 Stability of Linear Systems, 263
7.6.3 Local Stability of Nonlinear Systems, 267
References, 273
viii CONTENTS

Numerical Methods, 274

8.1 Introduction, 275

8.2 Numerical Methods for First-Order Equations, D5 fe)
8.2.1 The Euler Method, 276
8.2.2 The Improved Euler Method, 279
8.2.3 The Runge-Kutta Method, 281
8.3 Systems of Equations, 285
8.3.1 Higher-Order Equations, 287
References, 289

The Power Series Method, 290

9.1 Introduction, 291

9.2 The General Method, 292
9.2.1 Ordinary and Singular Points, 294
9.2.2 Solutions Near an Ordinary Point, 296
9.3 Solutions Near a Regular Singular Point, 305
9.3.1 The Method of Frobenius, 307
9.3.2 Roots Differing by a Noninteger, 309
9.3.3 Equal Roots, 311
9.3.4 Roots Differing by a Nonzero Integer, 314
References, 332

Answers to Odd-Numbered Problems, 323

index, 335
This book is intended as a one-semester, one-quarter, or two-quarter introductory
course in ordinary differential equations and their applications. It is designed for
students in mathematics, engineering, or science who have already successfully
completed a basic course in calculus.
Applications in the physical sciences as well as some in the social and life
sciences are prominent throughout the text, but kept at a fairly elementary level so
that little background in the various sciences is required to understand them. And
although applications are discussed throughout the text, I have attempted to main-
tain a close relationship between mathematical theories and applications. In this
regard, I have tried to avoid the temptation of introducing a multitude of applica-
tions in many diverse areas of application. My experience is that exposing the
student to too many applications proves distracting, at least to the average student.
(More of something good is not always better.)
For the most part, the text contains the standard material that is usually found
in introductory one-term courses. There are, however, some distinctions in the text
that are perhaps noteworthy. For example, in addition to the integrating factor
technique for solving first-order linear equations, a separate discussion of the
general theory of these differential equations is presented (in Chapter 2) from the
same point of view that is used in Chapter 4 in developing the theory of higher-order
linear equations. Thus, not only is a more unified approach to the theory of all linear
equations possible, but the simpler theory for the first-order equation can be used
for motivation in developing the corresponding theory for higher-order equations.
Another feature of the text is the introduction of Green’s functions (Chapter 5)
for handling nonhomogeneous equations in a systematic and physically meaningful
fashion. I believe an early exposure to Green’s functions involving ordinary differ-
ential equations enhances the student’s understanding and use of these functions
when required in more advanced courses involving partial differential equations.
A brief discussion of the qualitative methods used in oscillation theory is pre-
sented in the chapter on initial value problems (Chapter 5). Because many students
take differential equations prior to a course in linear algebra or matrix theory, an
elementary (nonmatrix) treatment of linear systems of differential equations is
presented in the early sections of Chapter 7. Hence, using techniques already
familiar to the student, a natural transition from higher-order linear equations to
ix
PREFACE

linear systems of equations can be attained. A separate section on the use of matrix
methods in solving linear systems of differential equations is also included for those
students who are already versed in matrix techniques.
The text contains more than enough material for a one-semester course, and most
of the chapters after Chapter 4 are sufficiently independent of each other so that
various arrangements of topics can be made to suit individual needs. Also, sections
that can easily be omitted for a shorter course are marked [O]. Hence, the text
should be flexible enough to lend itself to several different types of courses. For
example, a course may consist of selected material from Chapters 1-5, 7, and 9,
or from Chapters 1-6, 8, and (possibly) 9. If applications are not stressed, a course
could be based upon Chapters 1, 2, 4, 6—9, and so forth.
With respect to the physical layout of the text, I have numbered sections and
subsections in decimal form. The subsections do not contain separate exercise sets.
Various equations, theorems, figures, and tables are numbered consecutively in
each chapter, but not according to section. For instance, Theorem 4.5 is the fifth
theorem in Chapter 4 without regards to any particular section. Since worked
examples usually play an important role in learning new material, I have included
a large number of them (over 170) of varying difficulty throughout the text. Each
example is generally indicative of typical problems to be found in the exercise sets
at the end of each section (which consist of more than 1100 problems). Further-
more, all examples are set apart from the textual discussion by the use of horizontal
lines in the hopes of making them easy for the student to find. The exercise sets
usually contain a blend of drill-like problems, some more difficult, and some that
extend the theory and applications beyond that discussed in the exposition. Prob-
lems that are considered to be more challenging and those used to extend the theory
are generally separated from the more routine or drill-like problems by placing them
either at the end of the exercise sets, or identifying them by a star (*), or both.
References are listed at the end of each chapter.
I am grateful to Gerald Bradley, William Fitzgibbons, John Gregory, Samuel
Rankin, Burton Rodin, and Chester Scott, who served as reviewers, and to Leslie
Rochlin for suggesting Scott, Foresman and Company to me. Also, I would like to
record my appreciation to the editorial and production staff of Scott, Foresman and
Company for their helpfulness and cooperation with this project. And finally, I wish
to thank Jack Pritchard, editorial vice-president, who became closely acquainted
with this project near its end, but who was very helpful in working out the final
details and seeing it through to completion.

L. C. Andrews
Differential equations play a fundamental role in engineering, mathematical sciences,
and the life sciences because they can be used in the formulation of many physical
laws and relations. The development of the theory of differential equations is closely
interlaced with the development of mathematics in general, and it is indeed difficult
to separate the two. In fact, most of the famous mathematicians from the time of
Newton and Leibniz had some part in the cultivation of this fascinating subject.
In developing the theory of differential equations systematically, it is helpful to
Classify the various types of differential equations, since equations in a particular
Class can often be solved by the same method. Therefore, in Section 1.2 we discuss
some of the various classification schemes, emphasizing order and linearity.
In Section 1.3 we carefully define what we mean by a solution of a differential
equation. Here we first discover that differential equations are peculiar in that they
generally possess many different solutions, and for this reason we usually seek a
specific function, called a general solution, with the property that all solutions can be
obtained from it.
We introduce the notions of initial value problems and boundary value problems
in the last section of the chapter. These are the names attached to those problems
occurring in applications wherein the solution of a differential equation must satisfy
certain additional auxiliary conditions. The study of these problems is so important in
applications that we have devoted a significant portion of the text to developing the
theory associated with them.
1.1 INTRODUCTION

The theory of differential equations (DEs) has played an important role in science
and engineering since the introduction of the calculus by Newton* and Leibniz.t
Problems in the physical sciences have long since been investigated primarily by
formulating them as DEs. Differential equations were first used in the early eigh-
teenth century to solve problems in mechanics, but more recently the theory of DEs
has found its way into the social and life sciences.

1.2 CLASSIFICATION OF DEs

nite number of it ‘ivatives.t One of the simplest

examples that occurs early in the calculus is to find all functions y = (x) for
which

y’ =f). (1)
For instance, if f(x) = x’, the unknown function y is obtained through a simple
integration to yield
x?

Ruy Hove (2)

where C is a constant of the integration which can assume any value. Other
F)
examp or VES
ir

LINENe. (3)
oe: ky Sis:

*Sir ISAAC NEWTON (1642-1727) was born on Christmas Day. Although mathematics began as a
recreation for Newton, he was soon known as a great mathematician after his invention of the calculus,
discovery of the law of universal gravitation, and experimental proof that white light is composed of
light of all colors. He accomplished all this before the age of 24! A controversy erupted between
Newton and Leibniz over the invention of the calculus, brought on primarily by friends of Newton who
accused Leibniz of plagiarism. Eventually both Newton and Leibniz themselves became occupied with
the controversy, even though both were convinced that they had reached their results independently.
Newton spent the last two years of his life in constant pain, and he finally died in his sleep on March
XO), WPT,
+GOTTFRIED WILHELM VON LEIBNIZ (1646-1716) is considered one of the great thinkers of modern
times. Known as both a philosopher and mathematician, he is credited with building a remarkable
system of modern philosophy and was greatly responsible for the development of the calculus. The last
years of his life were saddened by ill health, controversy, and neglect. A bitter controversy had been
started by friends of Newton over the legitimacy of Leibniz’s discovery of the calculus. Newton had
also discovered the calculus a few years earlier than Leibniz but failed to publish his results until after
Leibniz had published his. Leibniz died alone on November 14, 1716, and only his faithful servant
attended the funeral. Many members of the scientific community were not even aware of his death until
almost a year later.
{Derivative identities such as d/dx sinx = cosx are not included in our basic definition of DEs.

2
 SEC. 1.2 / CLASSIFICATION OF DEs 3

(1 — x?)y” — 2xy’ + 6y =0, NEAR. (4)

y" + Ksiny =0, Non - Lin. (5)
+ xy" + y? Non - Lil, (6)
(y')? + 3xy = Non - Lin, (7)
Gd ee, ks, (8)
un, = 0,
(9)
Wea el ayy 1 ayy 0: (10)

Remark. Various notations for derivatives are commonly employed, depending

upon which issa omen at the time. SaStange. it is onary to make ee
association eo =Peli! ve = qd’vides

When the ankeowa Fangtion pean upon moreethan onei neeencent vari-
able, the derivatives will be partial derivatives and the equation is called a partial
differential equation. Examples of ordinary DEs are given by (3) through (7) above,
while (8) through (10) are partial DEs.
Another classification is according to the order of the DE.

Definition 1.1 |
CHAP. 1 / BASIC CONCEPTS

cannot be put down in the fo ven by | ) is nonline Thus we see

that Equations (3) and (4) are linear, and (5), (6), and (7) are nonlinear.

eran (11)
Or

M(x, y)dx + N(x, y)dy = 0. (12)

For example

1.2.1. Origin and Application of DEs

As mentioned in the introduction, DEs originated out of a study of certain kinds of

problems in mechanics. Today, however, their use is far more widespread. They
occur in various branches of engineering and the sciences, and are used in

the study of particle motion,

the analysis of electric circuits and servomechanisms,
continuum and quantum mechanics,
the theory of diffusion processes and heat flow,
electromagnetic theory, and
ae
SL
ee
ee
ed the theory of vibrations and sounds.

Disciplines such as economics and the biological sciences are also now making use
of DEs to investigate problems in

7. interest rates,
8. population growth, and
9. the ecological balance of systems,

among other types of problems.

The mathematical formulation of problems like those listed above gives rise to
a DE. This occurs because the various scientific laws employed in the formulation
of these problems involve certain rates of change of one or more quantities with
respect to other quantities, and such rates of change are expressed mathematically
by derivatives. Hence, the ensuing mathematical equations involve derivatives
in
the unknown quantities, and this is what we mean by a differential equation.
In formulating the DE, one must normally make certain simplifying assumpti
ons
so that the resulting DE is tractable. Determining just what assumptions
are reason-
SEC. 1.2 / CLASSIFICATION OF DEs 5

able is often the most critical part of a problem. Sometimes certain aspects of a
problem seem relatively unimportant and can be modified by assuming an approxi-
mate situation; sometimes an aspect of a problem may even be entirely eliminated.
The DE resulting from any such assumptions will actually be that of an idealized
situation.
Even after making a number of simplifying assumptions, the mathematical
formulation of a problem can still lead to a DE that is troublesome to solve. This
is particularly true when the resulting DE is nonlinear, since they are generally
difficult or impossible to solve exactly. Linear DEs are much easier to handle in
many ways, mostly because various properties of their solutions can be character-
ized in a general sort of way and standard solution techniques have been developed
for solving many linear DEs. For this reason, linear equations occupy a more
prominent place in the theory and applications of DEs.
Because of the difficulties encountered in solving most nonlinear DEs, they are
often approximated by linear DEs when the resulting linear equations can describe
certain fundamental characteristics of the nonlinear systems. For instance, the angle
6 that an oscillating pendulum of length 6 makes with the vertical direction
(Figure 1.1) is governed by

Ci)

then for small angular displacements 6, it may be reasonable to set sin = @, and
thus (13) can be replaced by the linear equation
4
a 0 (14)
(fe ae
In this particular example the linear equation (14) gives fairly accurate information
about the behavior of the pendulum when @ is small. There are a number of
applications, however, for which this approach is fruitless, since the phenomenon

*The constant g is the gravitational constant.

 CHAP. 1 / BASIC CONCEPTS

being studied does not lend itself to any reasonable linear approximation. In such
cases we try to solve the nonlinear DE itself. Techniques for handling such DEs will
be discussed somewhat in Chapters 2 and 7.

EXERCISES 1.2

For each of the following DEs, state whether the equation is linear or nonlinear, and
give its order.
es Nhs |
SeNC 1.) —35+3=—+y=0
Sera a s 2. y' +x’ =0
wohpetbe rT eles : sei
3. (x7? + y’) dx = 2xydy rh 4. y' t+a@y =f(x)
[ IP
ee. View)
y ets Oy ea Vineex 6. saancs — bP?

A
Ss 4y x . \ f|
SN . Te mt
i 1) 8.F y ho .eS ||

= “\ 9. 12 + Ri = EG) of0/ hP- 29)* + y =0

as axl di ; ge '

a a 1 xy Ba aay = Oe 12. x? dy +(y — xy — xe*) dx =0

. oo wi ns
. N13/ Hg. Fay = 0 14. me +y=0
Ins x
<15. yy" ye = y lox 0 16." Vi 4 ky ee Giney ee
“17. y’ + 1 = &’) — sinx 18. NV Dy ye ee
Rint 19) (x? =Q7)y’ + 3xy =0 20.. x7y" + ay’ —y = 3 cos*x

Pee) 2 ls OOLUTIONSIORIDES

ous requ

Definition 1.3

EXAMPLE 1 Verify that y = e “is a solution of y’ + y = 0, and state the interval of validity.

Solution The function y = e * is continuous and has derivatives of all orders for all x.
Furthermore,
 SEC. 1.3 / SOLUTIONS OF DEs 7

for all values of x.

It should be pointed out that a given DE will usually possess many solutions
example, it is easily verified that)y== Cen rene: ihe DE i in Example | for any
value of the constant C. Thus we obtain the family of solutions shown in Figure 1.2.
Even the trivial solution y = 0, obtained by setting C = 0, is a solution of
y +y=0.

c=4 ‘
C=3
C=2
ae Cai
oo ee
;

ae

ee

C=-2
C=-3
C=-4
Figure 1.2 Solutions of y’ + y = 0.

EXAMPLE 2 Verify that y; = C, cos 2x and y) = C,sin 2x, where C, and C) are any constants,
are both solutions of
y" + 4y =0.

Solution For y, = C,cos 2x, we find

yy — Or sin De. yi = —4C, cos OKs

so that
y! + Ay, = —4C,cos2x + 4C,cos2x = 0.

Similarly, it follows that

y! + 4y) = —4C,sin2x + 4C,sin2x = 0.
We might observe here that the functions
y = C,cos2x + Cysin 2x
 CHAP. 1 / BASIC CONCEPTS

and
y = C;sinxcos x
are also solutions of the DE, the verification of which is left to the reader.

EXAMPLE 3 Verify that g(x, y) = 0 is a solution of y’ = x/y, where g(x, 7y) =x? —y?— 1.

Solution n of g(x, y) = O, we obtain

d d
—(x*)
ot — ee
—(y?) =0 or 2x — 2yy’
vA =0;
and solving for y’ yields

(y' = -1
clearly does not have a (real) solution. The question concerning the existence of a
solution is very important in applications where the DE is our mathematical model
of some physical situation. If our model does 1not HO aUE it is probaaly a
poor model, since meaningful physical problems should lution:
Another question Anat solutions concerns uniqueness. In the Saree we have
discussed, the solutions are not unique. Nonetheless, certain additional require-
ments can be imposed in most instances so that a unique solution can be singled out
for those problems having solutions. More will be said about this situation later.
The existence and uniqueness questions are generally difficult to answer. Usu-
ally it is preferable to discuss these questions with respect to rather narrow
classifications of equations, and this we do to some extent as we go along. A more
practical question that concerns us is, How do we find solutions? We will devote
most of our efforts in succeeding chapters to answering this question. Observe,
however, that once a solution has been found, we have answered the question of
existence. But if no solution is found, it is not clear whether this is because a
solution doesn’t exist or simply because we are unable to find it. It is embarrassing
 SEC. 1.3 / SOLUTIONS OF DEs

at the least to use, say, a computer to produce a numerical approximation to the

solution of some problem only to find out later that the problem has no solution.
i em a

EXAMPLE 4 Determine the values of m such that y = e” is a solution of Yost) = 12 yO:

Solution The direct substitution of y = e” into the DE yields

y" —y' — 12y =m*e™ — me™ — 12e™

= (m?—m — 12)e™,
which equals zero only if (e”" # 0)

m* —m — 12 =(m = 0.
— 4)(m + 3)
Thus, m = 4 or m = —3, and we find the two solutions

Dee. pean Wes ee WPy te

y/ De LN /

The reader should verify that y = C,e** + C,e~* is also a solution of the DE.

EXAMPLE 5 Find a first-order DE involving both y’ and y for which y = x? — 1 is a solution.

Solution Computing
p os y’ = 2x, it becomes clear that xy’
) — 2y) = 2. Also, we observe that
‘(y')? — y = 1. Hence, we conclude that given a particular solution function, we can
find several different DEs that it will solve.

1.3.1 General Solutions

From the examples considered thus far, it is apparent that DEs generally have more
than one solution—in fact, infinitely many solutions. For this reason we often seek

ee fev
Thati
is to say, be reduced to a fe

Tht oy, CARO the foll wing amples

EXAMPLE 6 Bothy = C,cos2x + C;sin2x and y = C;sinxcosx + C,sin2x are two-parameter

sulutions of y” + 4y = 0. Are both general solutions?

Solution The first solution, y = C,cos2x + C,sin2x, is a general solution, since no algebraic
manipulation will allow us to combine the two constants into a single constant. On the
other hand, we find that
10 CHAP. 1 / BASIC CONCEPTS

y = C\sinxcosx + C,sin2x

= 501sin2x + C,sin2x

= (50 a c,)
sin 2x
Z
=< C3 sin 2x

so that this solution cannot be a general solution.

EXAMPLE7 Both y = Cje* + Coe* and y = C,coshx + C,sinhx satisfy the DE y” — y = 0.

Are both general solutions?

Solution In this case both functions do represent general solutions, since neither can be reduced
to a single-parameter solution. This shows that general solutions can assume different
forms for the same equation.
Using the definition of hyperbolic functions, we also note that
y = C,coshx + C,sinhx

ae Cle
l
He, Gt = ) Tek 31 Cle ¥, ren)
=<

= 51 (Ci + Ce" + (C1

1 — Che #3
= C3e* aR Geen

illustrating that we can obtain one general solution from the other.

nt s : r DEs; they de not arise in

elvine linear DEs site Asner restrictions. Eee instance, we can easily verify
that
« —CyY+y7?—1=0 (15)
is a solution of the DE
; 1
Coe by
(16)
for any value of the constant C. Moreover, it can likewise be shown that y = 1 and
y = —1 are also solutions of (16) but not members of the family (15), since no
value can be assigned to C to obtain these solutions. They are singular solutions (see
Figure 1.3).

Remark. Because of the possibility of singular solutions, some authors do not

consider (15) to be a general solution, since it does not contain all solutions of (16).
 SEC. 1.3 / SOLUTIONS OF DEs 11

na

=] Sa

Figure 1.3 Solutions of (y’)? + 1 = 1/y?.

Since this situation arises only in connection with nonlinear DEs, and since most
of our analysis concerns linear DEs, such a distinction need not worry us for the
most part.

Another significant difference between linear and nonlinear DEs is that a linear
combination of two or more solutions is also a solution in the case of many linear
equations (specifically those for which F(x) = 0 in Definition 1.2). This property,
however, is generally not true for nonlinear equations, as illustrated in the follow-
ing example. This is one of the reasons why nonlinear DEs are usually much harder
to deal with than linear DEs.

Remark. ‘O or more s ; to ew S
the superpc n principle This principle, which HE an aancnan role
throughout our
o Frectision of linear DEs, is first introduced in Chapter 4.

EXAMPLE 8 Show that although y, = e * and y, = e* are solutions of both the linear equation
y" os y' i 2y =0

and the nonlinear equation

We a= 0,
the expression y = Cje * + Cye” is a solution of only the linear equation for
arbitrary nonzero values of C; and C).

Solution By direct substitution of y = Cje * + C,e~* into the linear DE, we see that

Wi eye yen 1 40.e2) (Cre, 20,6") — 2(Cie" + Cye>)

= (C, + C, — 2C,)e* + (4C, — 2C, = 2C,)e*
== ten til eo t==n()s
whereas the substitution of this expression into the nonlinear equation leads to
yy" jt Gy) = (Ciena: 4b Corer Crew aE 4C,e*) ~ (—C\e™* di NOTA

= (C? — C?)e~* + (5C,C> + 4C,C,Je* + (4CZ — 4C3)e*

= 9C Coe",
CHAP. 1 / BASIC CONCEPTS

which is clearly not identically zero for arbitrary values of C, and C). In fact, it is zero
only when either C, or C; itself is zero.

Throughout the calculus we have become accustomed to the idea that the
“solution” of a problem is an exact, closed-form, analytical expression in terms of
elementary functions from which numerical values can be generated when needed.
In practice, however, this is rarely the case, especially when solving a DE. That
is to say, comparatively few DEs occur in applications for which simple exact
solutions can be found by known methods. Hence, in practice the “solution” often
consists of some type of approximation function or a set of numerical values
approximating the true numerical values of the solution over some interval. Several
such approximation methods are available for this purpose, and they are becoming
increasingly more common due to the complexities of the DEs arising in modern
science and engineering problems, and also due to the widespread availability of
modern high-speed computers.
In spite of what we have just said, most of the problems encountered in this text
will lead to exact solutions in much the same way as we have come to expect in
the calculus. The reason is that studying problems with exact simple solutions aids
the student in understanding the general theory and solution methods available; for
example, they provide an answer that can often be easily checked. Some approxi-
mation methods are briefly discussed in this text, but these techniques should be
taken up in more detail, such as in a course in numerical analysis.

EXERCISES 1.3
In problems 1-14, verify that the given function is a solution of the specified DE.
Assume C, and C, denote constants.

< Il = < ll * th fo)

Pa
+
o

(3.) ayy = xs yy =x log x, xi 0

as + (1+ x?) dy = 0;
(1 + yo)de —1 = 0
xy ta ty
J Byd= xy, x =G
= Cys

6. Oy ==, Vy
= Ve +3, 2 >0
f—», aP ae®
N( 7.) —
dt = aP — bP?, P =~
(1 + be”) (a, b constants)

' LA s
Bs Mey ote) eae” y= pret + x #0
xX

@.) y" + 2y’ — 3y = 0; -y = Cye* + Cre™

10. y" c= 6y' +. Oy = 0; y = (C, + Coxe"
 (sec. 1.3 / SOLUTIONS OF Des) 13

Py " — 2y’ + Sy
= 0; y = Cie*cos2x
+ Cre*sin2x
12. y"+y’ — 2y = 2x — 40cos2x:

y = Cie* + Cre * — :=x + OCS 241-2 Sin 2x

ni 13) xy" Sxy’ + 4y,= 07. y = Cx 4+ Cox logx, x > 0

—* 14. Vim oxy =: y=e™ [ie ear +e”

2|
%
0

Vv >

Each of problems 15-20 has a solution of the form y = e™ for some value of the
constant m. Determine for what values of m the following linear DEs have such a
solution.
& ne = ae
45) y'-2z=0 © 16. y"-y =0
Y)\ oe 18. y” + 2y’ =0

ah Pet oh py, =) () ; * é yo" 1 Oy 1 ily’ + 6y = 0

In problems 21-25, not all constants that appear in the given expression are
essential. Rewrite the expression using only essential constants.

Y\ 21) y=Ce"** = Pe’ 22, y = C\cos(x + a) + Crcos(x — a)

~ +} XK
23) yi oe s ya y=, Gini; log (ax’) *
oe Cis anwe) AG
v\ (25) y=<C, sinx + Cosin3x + C3sin?x Can
26. Show that g(x, y)=x°>+3xy?-—1=0 is an _ implicit solution of
2xyy' + x? + y* = 0 on the interval O <x <1.
Show that g(x, y) = 5x’y*— 2x°y?-1=0 is an implicit solution of
27.)
~ xy’ + y = xy? on the interval 0 < x < 3.

In problems 28-35, find a first-order DE involving both y’ and y for which the given
function is a solution. sy! Rie
aA pe 1
j 28. 1p = 69" ate = A C9)oy =x-4 30. y= zxe"

~@1) Re eel 32. byi=ncos es (33) y = sinh 3x

9A. & — 1% +y?-1=0 KSB) y? =x? +y
*36. Verify that y = Cx — C’ is a solution of
OD Saya 0
_for all values of the constant C. Determine a value of K such that y = Kx? is a singular
/ solution of the DE.
37) Verify that y = (jx? + C)? is a solution of y’ = xy”. Also find a singular solution
by inspection. . ; ’

38. The function y = C,e***® is a solution of the DE \. ¥ ¥

vf vf ‘
wan K
| f
14 CHAP. 1 / BASIC CONCEPTS

Ye ee ee
> for any choice of the constants C, and Cp. Explain why this is not a general solution.
&39.) Show that the piecewise-defined function
lee ieee O
tas Ms je = 0)

is a solution of xy’ = 4y as is y = Cx*. Can the solution defined piecewise be

obtained from y = Cx* by an appropriate choice of C? Explain. |
*40. The functions y = V1 — x? andy = —V 1 — x? are both solutions of yy’ + x = 0
on the interval —1 <x < 1. Is the function

EN leet Sa 0
: VI-x, O<x<1
also a solution? Explain.
Verify that y; = 1 and y. = x° satisfy each of the DEs
xy” 3 y’ = 0 and 2yy" es (y')? as 0,

but that y = C; + C2x’ only satisfies the first equation for arbitrary values of C, and
C>. Explain.
42. Verify that y,; = 1 and y. = x'” satisfy each of the DEs

2xy"+y'=0 and 8x*(y")? — yy’ =0,

but that y = C, + C2x'” only satisfies the first equation for arbitrary values of C, and
C2. Explain.

1.4 INITIAL AND BOUNDARY VALUE PROBLEMS

In most applications the unknown function y must satisfy certain restraints, or

auxiliary conditions, in addition to satisfying the DE. These conditions imposed
upon the problem determine which of an infinite collection of solutions are peculiar
to the given problem, and the number of these conditions is usually equal to the
order of the differential equation. ~ + .
, ¥ ee in Q) CS
Naw y! ind \/

EXAMPLE 9 Solve y’—y=0, ms

y(0) = 2, given that y = Ce* is a general solution.

Solution The general solution y = Ce* is a family of curves in the xy-plane (Figure 1.4). By
specifying y(O) = 2, we are seeking that one particular curve which passes through the
point (0, 2). Thus, by substituting x = 0 in the general solution, we obtain 2
= Ce°®,
or C = 2. The solution we seek is therefore

y = 2e’.
 SEC. 1.4 / INITIAL AND BOUNDARY VALUE PROBLEMS 15

Figure 1.4 '' uv

EXAMPLE10 Solvey”+y=0, y(0)=2, y’'(0) =0, given that y = C,cosx + C)sinx is

a general solution.

; g~ The auxiliary conditions require that the solution we seek pass through the point (0, 2)
of the xy-plane with zero slope at that point. Imposing these conditions on the general
-m>
SS) solution leads to _

¥ % y(0) = C,cosO + C;sin0 = 2,

d x y'(0) = —C,sin0
+ C,cosO = 0,

which, upon simplifying, yields.C; = 2, C, = 0. Hence, our solution becomes

y = 2cosx.

EXAMPLE 11 Solve y”+y=0, y(0)=0, y(m/2) = 1, given that y = C,cosx + C,sinxisa

general solution. ) , £
ce UU ae ery
ox
§Solution Applying the first condition y(0) = 0 to the general solution leads to hee
é He RRO
y(0) = C,cosO + C,sin0 = 0,

& y from which we deduce C, = 0. The second condition yields

a :
f ; ™ \ N
ee A »(3) = C,sin= = 1, )
SS Sau ye
or C, = 1. The solution we seek is therefore

y = sinx.

When the auxiliary conditions are all specified at a single value of x, such as in
Examples 9 and 10, we refer to the problem as an initial value problem. Although
we could choose the single value of x as any value, it is customary in practice to
use x = O. If the auxiliary conditions are specified at more than one point on the
interval of interest, the resulting problem is called a boundary value problem (see
16 CHAP. 1 / BASIC CONCEPTS

Example 11). The boundary conditions are usually specified at only two points of
the interval, and we refer to these problems as two-point boundary value problems.
Such problems involve DEs that are at least second-order, since first-order DEs
have but a single auxiliary condition and as such are classified as initial value
problems.
In general, initial value problems are well behaved in that they almost always
lead to unique solutions. Unfortunately, the same is not true of boundary value
problems, and so the general theory of these problems is more complicated. The
following examples give some hint of the difficulties that arise in solving boundary
value problems.

EXAMPLE 12 Given the general solution y = C;cosx + C,sinx, solve the boundary value problem
y’+y =0, yO) =0, yo) = 2.

Solution Imposing the prescribed boundary conditions, we find

y Oe G a0
and
y(m) = Chsinaw = 2.

But since sin 7 = 0, we conclude heh no member in the tebe family of Seats
satisfies both boundary conditions. “Since the equation is li e know t the DE
praseeisemive any SireLAP SOLUTIONS: SO we must accept the factthatrans pou value
problem has no solution.

EXAMPLE 13 Solve the boundary value problem in Example 12 if the boundary conditions are
changed to y(0) = 0, y(7) = 0.

Solution This time the boundary conditions demand that

y(0) =C, =0
and

y(ar) = Ci sin7 = 0,

which are satisfied for C; = 0 and any value assigned to C). Hence, we obtain the
family of solutions

y = Cysinx,
where C, is an arbitrary constant.

EXERCISES 1.4

Solve the initial value problems 1-10, for which the general solution is specified.

Q.) y’-2y =0, yO)=1; y =Ce”

 SEC. 1.4 / INITIAL AND BOUNDARY VALUE PROBLEMS ili
‘

@ xy’ +2y=0, y=4 yao bm

-_ ; : iT

ioe 8 pO) = 0 poet Cie

4. y"+y=0, yO=H1, y'0)=-1; y =C\cosx + Cosinx
(re
9.) y"=1,
”
yO)=0, y’0)=0;
,
y= 5x + Ci + Cox
l 2

‘6. yo=e. yO)=2,.. (0) =—-l; y=e + Cy + Cx

Y.) eee exyee st tye Oy (l= 0, ey (1) 1s ye = Cine Cox
8." — y =0, >y(0) =0, y'(0)=0; y =Cyet + Cre
9) y"— 2y’ + 2y =0, yQ =T1, y'0)=0:° y = e*(C,cosx + Cosinx)
Omhc y=a7 ep 0 = 3, y'1) = —12
yl) -y = Civ + Coxloex

Find solutions (if they exist) for each of the boundary value problems 11-15. The
general solution is provided.

11) y"=0, yO=0, yU=l; y=C.

+ Cox
12f y"=0, yO)=0, ya)=0; y=Ci
+ Cox
Ww
i y’ =“ty =0) y'0O)=0, y'(m=0; y =Cicosx
+ Crsinx
14 yo my =0, yO)=1, y(l)=0; y = Cicos mx + C2 sin mx

By 0;
y =Cre
y"-y =0, yO) =0, y')= +Cre
46. Verify that y = C,; + Cox? is a solution of the linear DE
xy Sy = Oe

/ Can constants C, and C; be found such that y(O) = 0 and y'(0) = 1?

*47. For which values of k (if any) does the boundary value problem

y"+k’y =0, y(0)=0, y(m =0,

have nonzero solutions? The general solution is y = C,coskx + C2 sinkx.

REFERENCES

Bell, E. T. 1937. Men of mathematics. New York: Simon & Schuster.

Boyce, W. E., and DiPrima, R. C. 1977. Elementary differential equations. 3rd ed. New
York: Wiley.

Eves, H. 1964. An introduction to the history of mathematics. 3rd ed. New York: Holt,
Rinehart & Winston.

Ince, E. L. 1956. Ordinary differential equations. New York: Dover.

Ross, S. L. 1980. Introduction to ordinary differential equations. 3rd ed. New York: Wiley.

Zill, D. G. 1979. A first course in differential equations with applications. Boston: Prindle,
Weber & Schmidt.
In this chapter we consider certain basic types of first-order DEs for which exact:
solutions may be obtained by clearly designated techniques. Proficiency in solving
such equations rests heavily on the ability to recognize the various types of DEs and
apply the corresponding method of solution, which often consists of certain
“devices” or “tricks.” This is particularly true when solving nonlinear DEs, since no
clear underlying theory can be applied to them.
In Section 2.2 we discuss equations that can be solved by the method of sepa-
ration of variables. This is the easiest technique to apply and comes up most often in
practice. Other equations that permit exact solutions in closed form are discussed in
Section 2.3, although some of them occur infrequently in applications. These include
exact equations, equations solvable by the use of integrating factors, linear equa-
tions, and homogeneous equations which are solved by reduction to variables
separable by a change of variable.
In Sections 2.4 and 2.5 we again discuss linear DEs of the first order, but in the
context of the general theory of linear equations of any order so as to provide a
more comprehensive approach. Thus we introduce the notions of a homogeneous
solution and a particular solution of a DE, the sum of which forms a general solution.
A physical interpretation of a general solution is given, and the basic existence-
uniqueness theorem concerning linear initial value problems of the first order is dis-
cussed.
A special nonlinear equation, called Bernoulli's equation, is solved in the final sec-
tion of the chapter by first reducing it to a linear equation through an appropriate
change of variable.
For a shorter course, Sections 2.4 through 2.6 can be omitted.

18
 2.1 INTRODUCTION

First-order DEs arise in a variety of problems, including the determination of the

velocity of free-falling bodies subject to a resistive force; finding curves of popu-
lation growth, radioactive decay, and the pursuit of a predator tracking its prey; and
finding the current or charge in an electrical circuit. The types of DEs involved in
these applications fall into several classifications, each of which demands a differ-
ent method of solution. We will discuss the various solution techniques in the pres-
ent chapter and take up the applications in Chapter 3.
The DEs to be studied maybe expressed either in the derivative form
‘y' = FQ, y) (1)
or in the differential
form
y)dx + N(x, y)dy = 0,
MC, Q)
depending upon the type of DE and the solution treatment. The equivalence of these
two forms can be seen in writing (2) as

d wa M(x, y) a

In applications the solutions of these DEs are usually required to satisfy an auxiliary
condition of the form

y (Xo) = Yo, (3)

which specifies that the solution function pass through the point (%, yo) of the
xy- plane.
Nonlinear DEs do not lend themselves to the development of general solution
formulas, even in the case of first-order equations. Except for the few special cases
treated in this chapter, finding the solution of a nonlinear DE is almost always very
difficult or impossible by known techniques. Therefore, approximation and quali-
tative methods (see Chapters 7 and 8) often play a marked role in the analysis of
a nonlinear DE, and for that reason it is important to be able to answer the existence
and uniqueness questions in advance. We state the following relevant existence-
uniqueness theorem without proof.

Theorem 2.1 If F and dF/dy are both continuous functions in some domain of the xy- plane
containing the point (xo, yo), then there exists a unique solution of the initial value
problem
y' = F(x, y), yo) = yo,

defined on some interval |x — x9| < h, where h is “sufficiently small.”*

*For a proof of Theorem 2.1, the reader should consult E. I. Ince, Ordinary Differential Equations
(New York: Dover, 1956).
19
20 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

In order to understand the significance of the continuity requirement in Theorem

2.1, consider the initial value problem
y'=y"3, (0) =0.
It is easily verified that
2x \ 3/2

n= (3) ’ ge)

is a solution, but it is also obvious from inspection that

yz =0
is another solution (called a singular solution). A unique solution does not exist
because
aF _ —a ( uy
ite 2/3
=-y
‘ay Oy 3
is not a continuous function in any region containing the x-axis (y = 0). If we
prescribe the initial condition at any other point (xo, yo) not on the x-axis, a unique
solution can then be found.

SEPARATION OF VARIABLES

(4)

or

(6)

important method that we will discuss for i ae ee DEs. -

 SEC. 2.2 / SEPARATION OF VARIABLES 21

EXAMPLE 1 Solve (1 — x)dy + ydx = 0. |

Solution Division by (1 — x)y leads to { | x)Oy d/ ) f

dy) _\ ax
igh y hing See Ov4 Rte
Thus, integrating both sides, we get ~~ \i) X

logy = log(1—x)+Ci, . \Imy =H bi -ylte

which can also be expressed in the form
y=e290 =x)

Since C, is arbitrary, so is e“!, and we can redefine it as C so that the solution

reads y = C(1 — x).

In Example | we are using the symbol logx to denote the natural logarithm, also
commonly denoted by Inx. Since integrals leading to logarithmic terms are quite
prevalent when employing this solution technique, some care should be exercised
in evaluating these integrals. Recall from the calculus that
\
die
[= = gle +c, ago)
u
where the absolute value is usually retained unless we know in advance that u > 0.
As a general rule, however, we will retain the absolute value only in those cases
for which a logarithmic term remains in the final solution form of the DE.

EXAMPLE 2 Solve y= s3yer 1)/x7, (x0, y > -4. ius

Solution Separating the variables gives (du 1%

dy (a a '
a?2 |
eet
x
and integrating yields
ieee | e
i + Z loge ,/\C >0.*
A |3 los Gy + |)= a

3y fags wat aut | ow (

ds to
Simplifying the algebra finally lea
aaa Oe is
form of the term
*Writing the arbitrary constant as tlogC is a simplifying device suggested by the
is arbitrary, it can assume any form that seems
involving y in the general solution. Since the constant
convenient for the problem.
22 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

3y +1 = Ce
or

eiaGGs
= 1}

Although we technically have the general solution of the DE once the integration
has been completed, it is usually advantageous to simplify the algebra in this
solution when possible. The simplicity of the final solution in the next example
clearly illustrates the gain resulting from a little algebraic maneuvering.

EXAMPLE 3_ Solve the initial value problem

(Gear Dh ae ge a TOR MO) ih

Solution Rearranging terms, we have

dy dx
ea ieee
Integrating this equation gives the general solution
arctany = —arctanx + C
or
arctany + arctanx = C.
From the trigonometric identity
tanA + tanB
tan(A + B) = —————_——_.,
ll = (enn Alien hs
we find

ih AP Se
tan (arctan y + arctanx) = i
_ xy ;

and consequently our general solution takes the form

WP Se BS
= tanC.
Leaety
Applying the initial condition y(0) = 1 reveals that tanC = 1, and hence

WS Vs
fx
Or
 SEC. 2.2 / SEPARATION OF VARIABLES 23

which is the intended solution.

EXAMPLE 4 Solve (x — 4)y*dx — x3(y? — 3)dy = 0.

Solution We can separate the variables by dividing by x*y*, finding

x-4 3
" 3 Pax - Pay =0
or
(x* — 4x") dx — (y~? — 3y dy = 0.
Upon integrating, we get

ae, y y
In separating the variables, it was necessary to assume that x # 0 and y # 0.
Note, however, that y = 0 is also a solution of the DE, which cannot be obtained
from the general solution by selecting an appropriate value of C. It is a singular
solution of the problem.

This last example illustrates one of the frustrations encountered in solving

nonlinear DEs, namely, the possibility of singular solutions. These solutions may
be lost in the solution process as in the above example, and yet in some instances
they may be just the solution we seek. In practice, therefore, the conditions of
Theorem 2.1 should be carefully checked with each problem to determine whether
the problem has a unique solution.

EXERCISES 2.2

In problems 1-25, obtain the general solution by separating the variables.

(N 2ydx = 3xdy 2 ye (3) y=--

. y
pm atia Va we dl _
—~

efsinx sinydx + cosxcosydy = 0

ve) sec*xdy + cscydx = 0

8. x71. ty*)dx + v7 + x*)dy =0 Y\

24 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

dV V oh aN
10. —S> = -—— =
4, a)a7 ae N = Nt
e
Lace

12... e*(y — 1)dx + 2(e* + 4)dy =0

t \GeISoy (yitix) dram ty? + x2 tay? deol dy = 0
14. xcos*ydx + tanydy = 0

! 15. (1 + logx)dx + (1 + logy)dy = 0 16. x*yy" = e”

[RoNS (17.) 2yy’ — 1 = 2x 187 yee

ah ~19, “(y + 1)dy — ylogxdx =0 6, dy =yV1—-y* dk

BA, (x? — ly’ = 2xylogy 22. (y'P =y?—y?

"23, x(y')? = y "24. y(y'P =1 25.0 yO) ee

In problems 26-30, obtain the particular solution satisfying the prescribed auxil-
iary condition.

~ d
26. oa =g, Uv(to) = vo 27) y’ ' =—2xy, y(0)=yo

(28) xy’ =1+y’, y(2)=3

9
29. {sin¥e
4 my
dx =
—ydy =
=0, y@O)=1
=
30. y ‘= ie
2x
y(2)
p) = O

*31. Giveny’ = y? — 4:
(a) Derive the general solution

heed Us Cer)
die hay he ehegs
(b) Show that the initial condition y(0) = 2 cannot be satisfied by the solution in (a).
(Can you explain?)
(c) By inspection, obtain two singular solutions and verify that one of these satisfies
the initial condition in (b).
*32. Given the initial value problem y’ = xy'?, (0) = 0:
(a) Find two particular solutions of the DE satisfying the prescribed initial condition.
Hint: One is a singular solution.
(b) Verify that the piecewise-defined function
O; #2 <i
y= AND
ad x = 1
/ 16
42 vA is also a solution. Are there any others?
Y\ (*38.) Solve xy’ — y* + 1 = 0 subject to
a) yO CD) y (0) ade (c) y(O) = 2.
*34. Solve y’ — x*(y? + 1) = 0 when
@tk a =i (b) k = -1.
*35. Find all solutions of (vi) + yy? sata a ee:
 2.3 MISCELLANEOUS TECHNIQUES

Although most of the first-order DEs that permit exact solutions are of the separable
variables type, several other types occasionally arise for which exact solutions can
be obtained.

2.3.1 Exact Equations

Recall from the calculus that if f(x, y) is a function of two variables, then its total
differential is given by the expression

df=eeade a+ Ol:
ay. (7)

EXAMPLE 5 Find the total differential of f(x, y) = x?y?.

Solution We have df/dx = 2xy? and df/day = 3x*y?, so that

df= Jxy?dx. + 3x-y" dy.

In this case it follows that the solution of (9) is given by the family of curves
Haar) Ce (10)
where C can assume any constant value.
For example, if
2Qxy3dx + 3x*y2dy =0, (11)
then we recognize the left-hand side of this expression as being the total differential
of the function f(x, y) = xy? (see Example 5). Hence, (11) has the solution
yr =C (12)
for any constant C.
Equations like (8) for which the left-hand side is an exact differential are called
exact differential equations. The difficulty in solving such equations is that it is not
always easy to determine by inspection whether a given differential expression is
exact, as we determined in the above example. Therefore, the following theorem
provides us with a test for deciding whether the expression M(x, y)dx + N(x, y) dy
is an exact differential, and the proof provides us with a scheme for finding the
function f(x, y) in those cases for which it is an exact differential.
25
26 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

Theorem 2.2

Proof: To prove the necessity part of the theorem, we merely observe that if the
expression is exact, then
0
M(x, y)dx + Mx, y)dy= Las + od

and hence

Ofaas of
ee M, 7 N

Taking mixed partials of f, we obtain

Bol
ey CM ein ey
Oydx
0 <ey > oxoy mmow
from which we deduce 0M/dy = dN/dx.
To prove the sufficiency part, let us assume that 0M/dy = dN/0x and

ie i)
eee
If we formally integrate this latter expression with respect to x, we find

fx, y) = iM(x, y)dx + g(y)

where the function g is the “constant” of integration (constant with respect to x).
Differentiating this last expression now with respect to y yields

Ze
0
jp[Mes de + 8G) =Nw»).
The function g is then a solution of the first-order DE

gO) =NG9)— jy| Mew y) dx,

the right-hand side of which is independent of x. To see this, observe that —

Oar = ON eee oe ONG. Oia

2(n # |mar) Ox 22 |mar) Ox dy e
and the theorem is proved. L_]
 SEC. 2.3 / MISCELLANEOUS TECHNIQUES 27

Remark. In the sufficiency proof of Theorem 2.2, we could just as easily have
started with df/dy = N(x, y) and integrated to find

f(x,y) = iN(x, y)dy + h(x).

Then A(x) is determined by

6)
h"(x) = Mx, y) — = |N(x, y)dy.

EXAMPLE 6 _ Show that the DE (4x — 2y + 5)dx — (2x — 2y)dy = 01s exact and find a gen-
eral solution. | ;
ia

Solution We identify M = 4x — 2y + 5andN = — (2x - 2y), from which we calculate

Yu a a
=) | dy Ax
Hence the equation is exact. By Theorem 2.2 a function f(x, y) exists such that

=of
ay =4,-2y
a y +5 and
an —of
ay =2y 'y — 2x.
Be if

fC) 2x ie oxy ox. F 2 (9).

“ 3 1 Crp dae
g it with respect to y and setting the result equal to N, we find a Nab iy
cy
EN Tees g'(y)= 2y 2x.
of = = ae U — D, SSG f

4 a\'=>— JU
It now follows that

g'(y) =2y of Oia

where a constant of integration is not needed since one appears in the final solution.
Therefore,
fG, y) = 2x? = 2xy + x tee
and our solution is
Dy ayuy xy5x= C..
28 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

Remark. Observe in Example 6 that the solution is not f(x, y) = 2x? + y*

— 2xy + 5x, but f(x, y) = C where C is any constant.

EXAMPLE 7 Solve (y* — 1)dx + (2xy — siny)dy = 0.

Solution We first test for exactness and find

0M oN
oy Bs
Thus,

OE a
== l
of
ax ey a daay xy sin Ms

For the sake of variety, let us integrate the second of these expressions this time to
get
f(x, y) = xy? + cosy + h(x),
from which we obtain

of
Neues ee = I
Hence, h'(x) = —1 or h(x) = —x, and our solution is

Aye het COS a= Ge

—————
a Ee

2.3.2 Integrating Factors

Occasionally we find that while the DE of interest

M(x, y)dx + N(x, y)dy = 0 (14)
is not exact, it

(15)

alsolution We i calition; however, that ae use of aire ne fear may

theloss or ee of solutions toee original equation. Furthermore , except _
a vneneatheadinheanatieremenrcnt ; on, this
thod o alate can be very v difficult.

EXAMPLE 8 Solve (3x + 2y)dx + xdy = 0.

Solution Testing for exactness reveals that

 SEC. 2.3 / MISCELLANEOUS TECHNIQUES 29

and thus the equation is not exact. However, if we multiply the DE by x, the new
equation

(3x7\+ 2xy) dx + x?dy = 0

is exact, i.e., M/dy = 2x = dN/dx. Now solving by the method of Section 2.3.1,
we have that f(x, y) = x* + x*y, and hence a general solution is
2 a Niaws acatia 8 Oye

Mostly we determine integrating factors by recognizing certain groups as differ-

entials of known expressions. The following formulas may be helpful in this regard:

es x
y dx z nS a(®) (17)
y y
ydx + xdy = d(xy) (18)

ie OE (tos) (19)
xy y
pein
= alas
eyy= a{tun'*) \ (20)
Any y \

eer TAO = aflog (x? + y]

enya
ar ZY .\
21)
For instance, from (16) it appears that w(x, y) = 1/x? is an integrating factor of

xdy — yava()dx =10:

There are no general rules for finding integrating factors. The procedure is one
primarily of trial and error and educated guesses.

EXAMPLE 9 Solve y*dx — x(xdy — ydx) = 0.

Solution The combination xdy — ydx suggests division by either y* or x*. Here division by
y*, accompanied by a division by x, leads to the expression

a ~ EES) = 0 or d(log|x|) + a(®)30)

XG ie y

which is an exact differential. Direct integration now gives

CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

fs.
log|x}g |x| ;
+-=C

as a general solution. In this case the integrating factor is u(x, y) = liye

2.3.3 Linear Equations

This section can be omitted if Sections 2.4 and 2.5 are discussed.
In Chapter 1 we defined the general form of a linear DE. For first-order equa-
tions, this form reduces to

Ai(x)y’ + Ao@)y = FQ). (22)

It is customary to divide (22) by A(x) to get the more useful form

y' + ays FO). (23)

To solve (23), we wish to find an integrating factor (x) such that the left-hand
member of (23) can be expressed as the derivative of a single function p(x)y. That -
is, we want

d
MOXOLy’ + ao(x)y] = pyleooyl = LOY + feey (24)

This condition is satisfied for some p(x) such that

M(x)ao(x)y = mw'(x)y,
or

b'(x)
a)
es)o(X) (25)
25

If we assume that p(x) > 0, integrating (25) gives

log u(x) = |ao(x) dx,

and finally, upon exponentiation,

Mx) = exo|
|do(x) ax}, (26)

Of course, s(x) is not unique, since any constant multiple of it also does the job.
Returning now to (23) multiplied by (x), we write

d
Hoy] = p(x)f (x).
Therefore, it follows that

Moy |w(x) (x) dx + Cy,

 SEC. 2.3 / MISCELLANEOUS TECHNIQUES
31

or

es
see - |M(x) Mix)f(x) dx
dx C Ce (27)
The function y given by (27) represents a general solution of Equation (23).

Remark. This solution technique has the disadvantage of not being adaptable to
higher-order linear DEs. A more encompassing technique is discussed in Section
BX

EXAMPLE 10 Solve y’ + 2xy =x.

Solution The integrating factor is

p(x) = exp [2/40] ert

Therefore, from (27) we have

2.3.4 Homogeneous Equations

Let us now consider a class of DEs that can be reduced to the separable variables
type by a change of variable.
If the DE
M(x, y)dx + Mx, y)dy = 0 (28)

has the property that

M(tx, ty) = t"M(x, y),
(29)
N(tx, ty) = t"N(, y),
then we say the functions M and N are homogeneous functions of degree n, and (28)
is called a homogeneous DE.* Because of the homogeneous nature of the
coefficients M and N, we can always express them as

*The term homogeneous as used here should not be confused with the meaning of this term as used
in subsequent sections and chapters.
32 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

M(x, y) pa= x"M ltJ, ;

N(x, y) = vn(1, AE
x
and hence (28) can be written in the form
dy _ _ MG, y) bil _ x"M(l, y/x)
dx N(x, y) x INCISED)

or, equivalently,

dy am A yCal
rae 30)
dx * (*) (
This last form of the DE suggests either the substitution y = vx or x = vy.
Setting y = vx, we get

through application of the chain rule. Equation (30) now becomes

d
ee +v = F(v),

which is the same as

EXAMPLE 11 Solve (x* + y*)dx + 2x \\y/= 0.

Solution We note first that M and N are both homogeneous functions of degree 4. If we try
the substitution y =vx), there follows
(XX + x4v4) dx + 2x*v(vdx + xdv) = 0,
or, if x #0,
(1 + 20? + v*)dx + 2xvudv = 0.
Separating variables, we find
‘dx
eee
20 Be =
x Neeacehely ee
which yields the solution
 SEC. 2.3 / MISCELLANEOUS TECHNIQUES 33

log
|x| =o" sh 1p t= Cc.
Replacing v with y/x now gives

Vere C
1 thatof dy, it 1s somewha
_x = vy rather than y = wx. Making this substitution, we obtain
dv
= ee ea
Il Se “Be y

which leads to
arctan v + log|y| = C
or
3
arctan (*)+ logly|=C, y #0.

EXERCISES 2.3

In problems 1-10, test for exactness and solve the equation.

1. Gx? — 6xy)dx — (3x? + 2y)dy = 0

2.) (2xy — cosx)dx + (x? — 1)dy = 0
> Qxtdx.y (x? — 1)dt =0 <
4) (yx — 3)dx + Qyx? + 4)dy =0
5. (cos @sin@ — 6r?)d@9 + r(1 — 6*)dr =0

6. (1+ logx + *)ae~ (1 — logx)dy =0

7. (Qu — e*”)du — 3(ue*” — cos3v)dv = 0
8. 3x(xy — 2)dx + (x* + 2y)dy =0
9. (tanx — sinxsiny)dx + cosxcosydy = 0
GLO me ee yO rye) (ext lye sae x(x? + y”) '|]dy =0

In problems 11-15, use the suggested integrating factor to solve the equation.
/
f

eet yy de hey 1) dy" 0, x Jy — AY JAK +5 ali

bet . Ty) pe
? el ds A\

125 x + y)ide ce mlogxdy-= 0; p ale

34 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

| y(x +y + Idx + (x + 2y)dy = 0; jth ABs

* 2, bilge. 2 ps 2 = {): =
1 :
142 Go + oxy = y*) de (ys 2x — % dy =0 Gane

1
15. (+ 2 ores
y°—x)dx —
—ydy == 0;1) pb =
Ona,

In problems 16-22, find an integrating factor by inspection and solve the equation.

a xdx + ydy = y*(x? + y”)dy

N ydx —xdy = x? dx
1 (x + 3y) dx — (y= 3x)dy =
“d (x* — y)dx + (x?y? + x)dy =0
*20. (x*y + 3x? + y)dx + (Qxy* + 6y° + xdy = 0
*21. (3x°y* + 4y)de + (2x°y*? + 3x)dy =0 Hint: Multiply by x*y?.
22. (x + y)dx + (y —x)dy
=0

In problems 23-32, verify that the given DE is linear and find its solution. ACheck
for linearity in both x and y.)

a y a 2y = 0 24. y' + (cosx)y

=0
ef Axy* dy + dx =0 26. (x*
+ 9)dy + xydx =0
sd y' + Ixy =xe™ 28. y' + (cosajy =e" _
29. xdy + ydx = (xsinx)dx 30. y*dx + xdy = Sdy ae
31. (y* + 1)dx + xydy = dy “32. (x? — ly’ + 2xy = Cosx

In problems 33—40, show that the equation has homogeneous coefficients and solve
the equation.
a Tae tee : aD >
+ @? — xy)dy=0
'338) Ge? + y*)de
~pi 3erga N34) xy’ =y He
eae
(u+v)du + (v —u)dv =0 36. xdx + (y — 2x)dy =0
xydx + (x? + y*)dy =0
*38. (s —1(4s + Dds + s(5s — thdt =0
‘39. ( — ylogy + ylogx)dx + x(logy — logx)dy = 0

“40. E — y Arctan (2)dx + x Arctan (:)dy =0

x

Le Maes 41-52, solve the equation by any method.

Kb
*

\(8 .@ — y)dx + 3x + yydy =0, yQ)=—-

42. 2x(y + l)dx — ydy = 0
43. y(2xy* — 3)dx + (3x*y? — 3x + 4y)dy =0
SEC. 2.3 / MISCELLANEOUS TECHNIQUES
35

44//x = 2y)dx + Ay — dy =0
“AS. xdy + 3ydx = (x — 2)dx (Solve two ways.)
46. (y — x)dx + dy =0 (Solve two ways.)
47. (xy? + x — 2y + 3)dx + x’ydy = 2x + y)dy, y(1) = 1
48./ x*y’ — y — xy = 0 (Solve two ways.)
Ue en ree ydy = y*(x? — y”)ady
50/ 2ny'=1+y%, yQ)=3
*31. 3ydx + 2xdy = xy*(xdy + ydx)
52. Sear es+ (y? — x*)dy =0
Prove that
p Pevaxyrrby
ad — bc #0
cx + dy”
is an exact DE if and only if b + c = 0, and find a general solution in this case.
*54. Prove that 4 = u(x, y) is an integrating factor for Mdx + Ndy = O if and only if u
satisfies the relation

it — yt 4 (2M _ aN)
dy Ox dy Ox
55. Using the result of problem 54, show that if u(x, y) = x’y’ is an integrating factor
of Mdx + Ndy = 0, then p and q must satisfy the relation
OM aN
pyN — qxM = »(Sy -S)
Ox
56. Use the result of problem 55 to obtain an integrating factor of the form w(x, y) = x”y4
and solve the equation.
Gey Cyaan idx cdye—() (OO) Cae) decd yaa)
(c) 2ydx + 3xdy = 3x~'dy
p57: Any DE of the form
y= xy fly)
is called an equation of Clairaut.
(a) Show that a general solution of this equation is the family of straight lines
y = mx + f(m)
where m is an arbitrary constant.
(b) Show that this DE also has the singular solution given by the parametric equa-
tions
Dale bet) ae eet Ata).
Hint: Differentiate both sides of the Clairaut equation with respect to x to obtain
Passa)ya 0:
58. Referring to problem 57, solve the following Clairaut equations:

@ y=n'+,0F% ® yan'tOP © v'-y =e"

[O] 2.4 THE THEORY OF LINEAR FIRST-ORDER EQUATIONS

The general theory of linear DEs is discussed in Chapter 4. However, one can
achieve a great amount of insight into this theory by first developing the corre-
sponding theory for first-order linear equations. Although we have presented a
solution technique for this type of DE in Section 2.3.3, it cannot be adapted to
higher-order equations and therefore cannot provide the kind of insight into the
general theory that we are seeking.
A first-order DE is said to be linear if it can be arranged in the form

A\(x)y’ + Ao(x)y = FO). (32)

The functions Ao(x), A;(x), and F(x) are known functions generally assumed to be
continuous on some specified interval, and in most cases it is also assumed that
A,(x) # 0 on this same interval. Applications involving (32) usually require the
solution function y to satisfy the additional condition
y(Xo) = ko, (33)
where kp is a known value. Although it doesn’t have to be, we frequently think of
the number x as the initial point in the interval of interest, and thus (33) is called
an initial condition.
We say that (32) is homogeneous * when F(x) = 0; similarly, when ky = 0, the
initial condition (33) is said to be homogeneous. Any other specialization of either
(32) or (33) is called nonhomogeneous.
If we divide each member of (32) by the function A;(x), we get

y' + aloy =f@, BD

where do(x) = Ao(x)/A,(x) and f(x) = F(x)/A;(x). We will henceforth refer to (34)
as the normal form of the linear equation.
In additioi to” writing the equation in normal form, notation is simplified by
5 Bile enor ae
introducing the concept of a ayjerential operator i certain situations. For instance,
the symbol D is an example of a meeatal operator that is defined by the rule

Dy =y’. (35)
A more general example of a differential operator is
M =D + a(x), (36)
where a,(x) is any function. We interpret this operator such that

[D + apx)]y = y' + avy, (37)

and thus the operator M can be used to abbreviate (34) as simply M[y]=
= f(x). We
refer to M as a normal linear operator on any interval for which ao(x) is con-

*The use of the term homogeneous as given here and in succeeding chapters is not the same as
implied
in Section 2.3.4.

36
SEC. 2.4 / THE THEORY OF LINEAR FIRST-ORDER EQUATIONS 37

tinuous.* (When using operator notation, it is customary to use brackets to separate

the operator M from the function y.)

2.4.1 The Homogeneous Equation

If we set f(x) = 0 in Equation (34), the resulting expression

- ec SOLUTION | J rt) (38)
VoRMDL oR pes
is called the associated homogeneous equation. One feature of a homogeneous
linear DE (common to all homogeneous linear DEs of any order) is that y = 0 is
a solution, called the trivial solution. Of course, what concerns us most in practice
is the possibility of nontrivial solutions.
For the special case when a(x) is a constant, i.e., do(x) = a, then (38) becomes
y' + ay = Oor, equivalently,

Vann (Ly: (39)

The solution of (39) must therefore be a function whose derivative is a constant
multiple of itself. Clearly, y = e ™ is a function with this property, as is
y =Cie™ (40)
for any value of the constant C;. Hence (40) is a general solution of (39).
In the more general case when ap(x) is not a constant, we solve (38) by separating
terms involving the variables x and y (i.e., separation of variables). Hence

=dM4 = -aja)dr, y #0, (41)

direct integration of which yields the solution function

log |y| = = |ala) a eM OR, (42)

where C is a constant of integration and log|y| denotes the natural logarithm.

Exponentiation of this expression leads to the explicit solution function

(43)

where exp (x) = e* and C, = e°. Notation is simplified by writing

yi(x) = exp(- Ag(x) ax), (44)

and then our general solution (43) becomes simply

y = Ciyi(%). (45)

*In general, a linear operator M is one for which M [Cif (x) + Cog (x) = Ci M[f(®)] + C.M[g (x)] (see
problem 27).
38 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

The function y, represents the only nontrivial solution (to within a multiplicative
constant) of Equation (38). Thus (45) is truly a general solution ,since it contains
all solutions of (38), including the trivial solution y = 0 obtained by setting
C, = 0. Also observe that (45) reduces to (40) when a(x) = a.
In Section 2.5 it will be necessary to distinguish the solution function given by
(45) from another part of the general solution required in solving nonhomogeneous
equations. Therefore, the symbol y, will be used in that section to identify (45),
while here we will continue to use the symbol y alone since it is the only solution
that presently concerns us.

Remark. In obtaining the general solution (45), it was necessary in (41) to

assume y # 0. Now that we have solved (38), however, it is clear that this
restriction can be removed.

EXAMPLE 13 Solve y’ +(Qxy = 0.

Solution Separating the variables x and y, we find

d = -2xdx,
SY
integration of which yields
log |y| =-x°+C

or, equivalently, at 3
sp C cee nC =e oN
Of course, we could have used (43) as a solution formula, which requires only
that we identify ao(x) = 2x. Completing the integration in (43) then leads to the
same result.

EXAMPLE 14 Solve the initial value problem

xy’ +y =0, y(1) =3.

v= (pee
Solution Assuming x # 0, we divide the DE by x to get the normal form .

Thus, do(x) = 1/x, and substituting this expression for ao(x) into (43) gives us

a C
y = Cjexp(-[©) =e es 0.
XxX x

By imposing the initial condition y(1) = 3 on this solution function, we obtain the
- value
 SEC. 2.4 / THE THEORY OF LINEAR FIRST-ORDER EQUATIONS 39

VU Os),
Hence, the solution we seek is

which is valid on any interval not containing x = 0.

If we wish, we can use the general solution (45) to produce an explicit solution
formula for the initial value problem
y'’ ta(x~y =0, yx) = ko. (46)

That is, imposing the initial condition in (46) upon the general solution
y = C\y\(x), we find

y(Xo) = Ciyi(%) = ko.

Assuming y;(%9) # 0, we obtain the value

k
C1. =—, (47)
yi(Xo)
and therefore the solution of (46) can be put in the form

k
y = OD) vil) # 0. (48)
yi(Xo)
If ao(x) is continuous throughout an interval containing the point x = Xo, then it
cannot happen that y,(%9) = 0 (see problem 26). In such a case, (48) is the unique
solution of (46). For those cases where ao(x) is not continuous, the initial value
problem (46) may still have a unique solution, but the problem may also have no
solution or more than one solution. Consider the following example.

EXAMPLE 15 _ Solve the initial value problem xy’ = 4y, y(O) = 0. a KR = O

Solution Dividing the equation by x, we see that a(x) = —4/x has a discontinuity atx = 0.
Nonetheless, we can still produce the general solution
VSS Cie

which is a well-behaved function for all x. Imposing the prescribed initial condition
on y, however, does not require C; to assume any particular value, and hence the
given problem has infinitely many solutions.
We should observe that if a nonzero value was prescribed for y(0), the problem
would have no solution.
40 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

EXERCISES 2.4

1.) Show that

(a). ene = eae (c) e~tsleex! = legsx|

(b) e (1/2) log x ae Vx, x>0 (d) e707 +2 logx) = >, x >0
x

In problems 2—12, find a general solution.

2. y'’ +2y =0 3.) ae + by =0 (a, b constants)

a dy dy y
4. y' f + (tanx)y = =0 5! a
—_—_— = xy 6. ae
—>
= >

7. xy ye 0 8. 4x°ydx + dy =0 *9. (logx)ydx + dy =0

10. tds + s(2t + l)dt =0
11. CSS yo ya="0 12. @* + 9)y’ + xy = 0

In problems 13-24, find a solution of the given initial value problem and state the
interyal for which the solution is valid.

13. y' + (cosx)y = 0, y(O) =2

14. y’ + Ginx)y = 0, (4) =2)) 15. “ = 5x*w, w(0) = —-7 NY

16. eee —(3sinx)y, y(Q) = L 7 y'’+ey=0, yO)=-1

dx 2 7
18. z'+4x°z =0, z(1) =2 49) y' + (6x? +2ny =0, y(D=1
20. y’ +ky =0, y(0O)=yo (k,yo constants)
74 Ne xy’
-y =0, y(1)=1 22. x*y’
Py =0) Fe) =3 ’
Sao
‘23. (la x)y ety =O. S(O) — 2 24. ‘y= (cotx)y = 0, (3) =j ‘
L=(O1
*25. Given the DE y’ + ao(x)y = 0, it can be shown that if ao(x) has a point of discon-
tinuity, the solutions may or may not be discontinuous at this point. In problems
21-24, the function ao(x) has certain points of discontinuity. Discuss whether the
solutions of these equations are continuous or discontinuous at these points.
*26. If ao(x) is continuous on some interval /, prove that the solution function

ya) = exp( — |(x) ax)

cannot vanish anywhere on /.

27. Given that f and g are differentiable functions and C; and C> are any constants,
demonstrate that the linear operator M = D + ao(x) has the property
M[Cif(x) + Cog(x)] = CiM[f()] + CoM[ go].
 [O] 2.5 THE NONHOMOGENEOUS LINEAR EQUATION

We now consider the nonhomogeneous linear equation

M abe,=y'+ a) = f(x). (49)

corresponds i in niyieal PAENA toSRE applied are or

SETA Thus
it is often called the forcing function.

(50)
ge tion 49). To see this, we will assume Y is any solution of (49)
and ieaney=ZY — yp. The direct substitution of y = Y — yp into (49) gives

My] = (Y — yp)’ + ao(x) (Y — yp)

= Y' + ao(x)Y — Lye + ao(x)ye]
= f(x) — f(x) = 0.

Hence, the function y = Y — yp is a solution of the homogeneous equation

M[y] = 0 and as such must be contained in yy, i.e.,
y = Y — yp = Cy).

It now follows that

Ye= ori)
and therefore belongs to the general family of solutions described by (50). Since
Y could be any solution of (49), we conclude that (50) is a general solution.
Summarizing, we have the following theorem.

Theorem 2.3 | If yp is any particular solution of the nonhomogeneous equation

‘EXAMPLE 16 Meaty. that yp = —2x — | and zp = e** — 2x — | are both particular solutions of
y’ — 2y= 4x, and find a general solution in each case.

Solution For yp, we have

ee ype er AX 2 = 4x,
41
42 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

and for zp it follows that

zp — Izp = 2e* —2 — 2e% 4+ dxet 2 = 4x,
so that both functions are indeed particular solutions.
The associated homogeneous equation is y’ — 2y = 0 with solution yy =
C,e**. Thus the two general solutions obtained from this result are

w= yaya —2x —1+ Cres

and

Vi= Zp ot yee eee Let Gea

However, in the second case we note that

y = —2x —1+(1 + Cie* = —2x — 1 + Cre”,

and so the two general solutions are actually equivalent.

that the pa
/any pa

Remark. The difference of two particular solutions of a linear nonhomogeneous

DE is always a solution of the associated homogeneous DE, as we have already
demonstrated.

Since the homogeneous solution y, can be found by the method discussed in

Section 2.4, we now concentrate on finding a particular solution of (49). In the
method of constructing yp that we employ, called variation of parameters, we
assume the form

yp = Uu(x)y\ (x) (S1)

for some function u(x) to be determined, and where y,(x) is defined by (44) in
Section 2.4. The technique derives its name from the fact that the arbitrary constant
in the homogeneous solution y = C,y\(x) is replaced by the unknown function
u(x).
The operator M applied to yp leads to

M[yp] = (uy1)' + ao(x)uy,

=u'y, + uly + ao(x)yi],
FO
Zero

which reduces to M[ yp] = u'y, since y, satisfies the homogeneous equation. Now
if also M[ yp] = f(x), then clearly the function u(x) must be chosen such that

u'(x)y(x) = f(x),
or
 SEC. 2.5 / THE NONHOMOGENEOUS LINEAR EQUATION 43

u(x) =
F(x)
|——dx + C. 4
yi) Ce
Technically any constant C can appear in (52), and the resulting function will lead
to a particular solution, so it is customary to set C = 0 and write the particular
solution as

(53)
Combining yp with the homogeneous solution function y, leads to the formula

yY=yp + yy = vila) |EG) dat Cry) (54)

yy (x)
as a general solution of (49).

Remark. Motivation for the technique used in constructing yp is difficult to give.

In fact, many of the techniques commonly used in solving DEs are the result of
someone having tried various “schemes,” “tricks,” or “devices” until something
worked.

EXAMPLE17 Find a general solution of xy’ + (1 — xy = xe*.

ye(ipen sea 2
Solution We first rewrite the equation in normal form,
I g
ote | en: # 0, }

and hence
1 e*
yi) = exp |
—[( ~~ 1)ar|Sti

From (53) we obtain the particular solution

e* |e e l
yp : |e
=—_— —
dx =
=. x d dx
— =i
5 xe
— om

be
which, added to the homogeneous solution yy = C\e*/x, gives

l C\e*
y= ae = aa x #0.

EXAMPLE 18 Solve the initial value problem

dye (sin 20—sytanx) dx, y(0) =".

Solution Rearranging terms, we have

vat (tan x)y "sin 2x,
44 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

from which we immediately deduce

y\(x) = exp(— {tanxdr) = cOsx.

Therefore, the general solution is

sin 2x
y = cosx dx + C,cosx
cosx

= 2eosx |sinxdx + C,cosx

= —2cos*x + C,cosx,
where we have made use of the identity sin2x = 2sinxcosx. According to the
initial condition, we find

9) = C, = =),

or C; = 1. The solution of the initial value problem is then

Y. = COSH 2 COS.x,

When checking the linearity of a given DE, checking for linearity in just one
variable is not always enough, as our next example illustrates.

EXAMPLE 19 Find the solution of dy/dx = 1/(x + y’), which passes through the point (—3, 0)
of the xy-plane.

Solution The DE is not linear in y due to the presence of the term y*. However, if we invert ©
the equation, we get

y = x de Vw Ola, dx
2) eee Vie
dy
which is linear in x. Thus, interchanging the roles of x and y in (54), we first
calculate

x\(y) = exp (|a) =e,

and hence

x(y) = e” |ye dy + Cie? = —y? — 2y —2 + Cre’.

To satisfy the prescribed condition x(0) = —3, we find that C; = —1, and so

x(y) = -y? — 2y -2 -e?’.

SEC. 2.5 / THE NONHOMOGENEOUS LINEAR EQUATION 45

2.5.1. Physical Interpretations

When an initial condition is prescribed, the problem is characterized by

My] =f), yo) = ko, (55)

where M = D + ao(x). As before, we assume the solution can be expressed in the
form y = yp + yy, where yy = C,y,(x). In order to derive a general solution
formula, and also to anticipate physical interpretations, it is advantageous to select
a “special” particular solution function yp. For example, if we choose to write yp
in the form

ye-= yi) ibs

fO 4
sat (56)
then we obtain the important property that when x = Xo,
t
p(X)= yi (Xo)
a)
‘0 f) de 0)

Hence we write

y= Yeok Ya

= y\ (x) lai + Ciyi(),

and imposing the prescribed auxiliary condition, we find

y(Xo) = 0 + Cry: Qo) = Ko

from which we deduce C, = ko/y;(xo). Our solution formula for the initial value
problem (55) can therefore be represented in the form

3 FO 4,4 ku)
y =yi0) | SO ey >?
provided y,(%o) # 0.

Remark. It may be interesting to note that the solution formula (57) for the initial
value problem described by (55) depends only upon the solution function y,(x) of
the associated homogeneous DE and the input parameters ky and f(x). This is a
fundamental characteristic of all linear nonhomogeneous DEs and is the primary
reason why so much effort in succeeding chapters on linear DEs is directed at
finding solutions of homogeneous equations.

Not only have we derived the solution formula (57) for the initial value problem
(55), but our choice of yp and yy now leads to important physical interpretations.
For instance, since yp(x9)= 0, we can think of yp [defined by (56)|as the solution
of the initial value problem
or both limits
*It is customary to introduce a dummy variable of integration such as in (56) when one
of integration are variable.
46 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

M[y] =if(X)s y(%) = 0, (58)

wherein the initial condition is homogeneous. The physical implication of yp, then,
is that it represents the response of a system at rest which at some time x = Xp is
subjected to the external disturbance f(x). Moreover, since yy must now satisfy the
auxiliary condition y;(xo) = ko, it is a solution of the initial value problem

Mly] = 0, y(%o) = ko, (59)

which physically describes the behavior of a system due entirely to the initial
condition, without any external disturbance.
We find, not only here but also in studying higher-order linear DEs, that it is
often convenient (but not necessary) to think of nonhomogeneous problems as
composed of two simpler problems—one having a nonhomogeneous DE and ho-
mogeneous auxiliary conditions [such as (58) ] and the other having a homogeneous
DE but nonhomogeneous auxiliary conditions [like (59)]. In this way, each solu-
tion function will describe important physical characteristics of the system being
studied that may not be as discernible once the two solutions are superimposed.

EXAMPLE 20 Solve the initial value problem

y'’ —2y = 4x, y(0)=S.

Solution Since ao(x) = —2, we readily calculate

y
(2far)=
yi) = exp (—fasco ax)= exp
Therefore the homogeneous solution is yy = C,e**. If we utilize the solution
formula (59), then we can apply the initial condition directly to this solution
function without first finding the general solution of the nonhomogeneous DE. This
action leads to
Mi Se ae

[We should take note of the fact that if we were using (54) to solve this problem,
we could not impose the initial condition at this step of the solution process.|
Substituting y, into (56), we find

yp = fal 4te-*dt = —2x —1+ e*,

0

and thus the solution we seek is

y = yp + yy = —2x — 1 + Ge.
pe ee

One of the difficulties that arises from time to time in solving initial value
problems is that the integrals that need to be evaluated are not elementary. Consider
the next example.
 SEC. 2.5 / THE NONHOMOGENEOUS LINEAR EQUATION 47

eee ee ee ee ee ee ee Se Se

~ EXAMPLE 21 Solve the initial value problem

yo —2xy = 1, yi0)= 3.

Solution Direct integration of —ao(x) = 2x yields

y\(x) = exp (2{xar)=e.

Substituting this function into (57), we find

y= eaiee 2 ci + 3e,

but the integral to be evaluated is no longer an elementary integral. When this

happens, we generally leave the solution in integral form. If numerical values of
y are required, they can be obtained by numerical integration techniques, such as
Simpson’s rule. In this instance, however, the integral is related to a function
known as the error function , which has been extensively tabulated. This function
is defined by
OY te
erf (x) = Ve ; Ce at.

so that we can express our solution of the initial value problem as

y= [sVeeton + Jer.

This last example illustrates that even simple-looking DEs may lead to non-
elementary solution functions such as the error function. A host of other functions
of this type falls under the category of special functions. Many of these other special
functions also arise in the process of solving a DE that does not possess solutions
expressible as elementary functions. As long as these new functions are tabulated,
we can treat them essentially as known functions. In computing numerical values
of functions, there is little difference between consulting a table to evaluate an
exponential or trigonometric function and consulting a different table to evaluate,
say, the error function.
.
2.5.2 Existence-Uniqueness Theorem

To begin, we prove the following lemma.

Lemma 2.1 The homogeneous initial value problem

y’ + ax)y =90, y(%) = 0
has only the trivial solution y = 0 on any interval containing x = Xo for which a(x)
is a continuous function.
 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

Proof: The general solution of the DE has been shown to be

y = Cry;(x)
where

yi(x) = exp |-[axe ax|.

(

Since ao(x) is assumed to be continuous, its integral is also continuous, and there-
fore y; is always positive. Imposing the initial condition now yields

y(xo) = Ciyi%) = 0,

from which we deduce that C; = 0. Hence the only solution of the initial value
problem is y = 0.

We are now prepared to state and prove the following important existence-
uniqueness theorem for linear initial value problems of the first order.

Theorem 2.4 If M =D + a(x) is a normal differential operator on the interval x, <x < x
containing the point x = xo, and f(x) is continuous on this same interval, then the
initial value problem

My] =y' + a(~y =f), yo) = ko

has exactly one solution on this interval, given by

“£0 4, koi) \
Yen de xo Yi(d) yi(Xo) ”

where

yi(x) = exp |— [aon ax},

Proof: To verify that y is a solution, we observe that

iz Te I xf koyi (x)
eRe ano aty sien
and so it follows that

y' + ax)y = f(x) + Ly + a0) yi |pau

oe diate yee a0) es
a a a) —————— =
Zero Zero

= f().
 SEC. 2.5 / THE NONHOMOGENEOUS LINEAR EQUATION 49

To show that y is a unique solution, let us assume that uw, and uw are both solutions
of the initial value problem and define z = u; — u2. The function z then satisfies
the homogeneous equation

M[z] = M[u] — M[uw2] = f@%) — f@) =0

and the homogeneous initial condition

Z(X0) = Ui(X%o) — U2(Xo) = ko — kyo = O.

Based upon Lemma 2.1, z is the trivial solution and uw, = wp for all x in the interval
of interest. [J

Remark. Observe that the existence theorem in Section 2.1 (Theorem 2.1) also
applies to the linear equation discussed here. The advantage of Theorem 2.4 over
Theorem 2.1 for linear DEs is that it not only informs us of the existence of a
solution but also provides us with a solution formula. ~

Some comments about Theorem 2.4 should be made at this point. First, the
theorem not only guarantees the existence of a solution but also states that there is
only one solution, i.e., that the solution is unique. Furthermore, the solution is
differentiable and has a continuous derivative on the specified interval. This can be
observed by writing the equation in the form

y’ = —ax)y + f@),
where the right-hand side is composed of continuous functions. However, the
condition that ao(x) and f(x) be continuous on the interval of interest is not a
necessary condition, but a sufficient condition of Theorem 2.4. For instance, in
Example 17, the solution y = 5xe* + C,e*/x is not valid at x = 0. Yet if we
impose the initial condition y(0) = 0, then we can select C; = 0, and the solution
reduces to the well-behaved function y = 4xe* for all values of x. In other words,
the situation here is similar to that discussed in Section 2.4. That is, if ao(x) is not
continuous, then a unique solution may still exist, but it is also possible that either
no solution exists or more than one solution exist.
When the function f(x) is not continuous, there is also no guarantee that a unique
solution of the initial value problem exists. And in certain applications the input
function f(x) commonly exhibits a discontinuity at some point due to, say, a switch
being (instantaneously) turned on or off.
Pere er

EXAMPLE 22 Solve the initial value problem

We oe Se Ua ae
where

FO et OF Os n= al
50 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

Solution Here the input function has a jump discontinuity at x = 1, so we solve the problem
in two parts,
y ty =0, yO=H1, Osx a1
and
We yyoas lees
The solution of the first problem is found to be
y =e) Ose reel:
For x > 1, we have y = 1 + Cye*, and hence we write

i. Ca? ty Ome Kaeeel

Swamial Pe etyisr ibaaSs Ft:
In order that our solution be continuous for all x, we want to select the constant C;
such that

lim(1 + Cye™*) = lime ~.

xt x17

Thus, C; = 1 — e, and our solution takes the form

Soe) eee
ys Vile (1 eres
Although this solution is continuous, it is not differentiable at x = 1 (see Figure
2a

EXERCISES 2.5

In problems 1-20, find a general solution.

eeUD oa
ekey 2c Yui
ye= Sy
 SEC. 2.5 / THE NONHOMOGENEOUS LINEAR EQUATION 51

xy'+ y =xsinx 4. xy’ —y =x’sinx

xy’ = 4y + x%e* 6. (x° + 3y)dx — xdy =0

z'’ =x — 4xz 8. “ = cscx + ycotx

dy
er = cscx — ycotx “10. udt + Bt — tu + 2)du =0

11. + 9)y' +xy =x

(x7 12. 31 woe + ex =y
dy
13. y'’ — my =e™ 14. y'—
my =e" (k #m)
15. fy Oxy = xe = 1G. ye (cosx)y =e)
17. xy’ — ay = bx 60 > 0,> a4) 18. xy’ — ky = bx* (x > 0)
19. toy vn *20. (x7 — l)y’ + 2xy =cosx

In problems 21—26, solve the given initial value problem.

di
“ah L7. + Ri=E, i(0)=0 (L,R, E constants)

di
22. L—+Ri=Asi
FF 1 sinwt, El i(0) == 0

23. (2x + 3)y’— y = @x +3)'*, y(-1)=0

24. (Do i\yetexe vy(0) =" 1

25) xy' + 2y = sinx, (4) = 1

26: y'’ =x? - 2x,

/
yl)=1
Zi: Solve the initial value problem
| Ute pei
y ‘'+y=
y =f(x), ; 0) =0,
y(O) wh
where f(x) a tkaS es

28. Solve the initial value problem

we Oe <1
y' + 2xy =f, yO) =2, where f(x) =4>OO), dee
7dl.
29. Solve the initial value problem

(1 + x7)y' + 2xy =f(x), y(O)=0, where f(x) = _

*30. Given the initial value problem

y' +k =Asinot, y0O)=y (wo #4),
(a) find its solution.
(b) The part of the solution in (a) that persists as t+ is called the steady-state
solution. Find the steady-state solution.
(c) What is the steady-state solution when w = k?
31. If y = (x) is a solution of
52 CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

y' + ao(x)y = f(x)

and y = (x) is a solution of
y’ + aox)y = g(a),
show that y = (x) + w(x) satisfies

y! + aoa)y = f(x) + g@).

32. Use the result of problem 31 to solve
y + p= 2sinx + 3e*.
*33. Show that the substitution z = logy transforms the DE
y’ + ado(x)y = f(x)y logy
into a linear DE in the variable z.

*34. The nonlinear DE

y' = P(x)y* + Oy + R@)
» is called a Riccati equation. If y = yp is a particular solution of the equation, show
that y = yp + 1/z is a general solution where z is a general solution of the first-order
linear DE

Zit [2ypPO) 7.0G) z= SP).

*35. Referring to problem 34, find a general solution of
y= yee = x)y +x?7-—x +1,
given that yp = x is a particular solution.
*36. When P(x) = —1, the Riccati equation (problem 34) becomes
y’ = -y? + O@)y + RO. =
Show that the substitution y = u'/u reduces this equation to the second-order
linear DE
u' = OG)u" — R@)u = 0:

[O] 2.6 BERNOULLI’S EQUATION

The nonlinear DE

DE eke bi wide ape dec 10 A (60)

is called Bernoulli’s equation after the Swiss mathematician Jakob Bernoulli.*'It
is assumed that n is any real number except as noted. The values.n = 0 andn = |
are excluded from the discussion, since a linear equation results in each case.

*JAKOB BERNOULLI (1654-1705) was a member of the famous Bernoulli family, which produced eight
mathematicians in three generations. Besides studying the equation named in his honor, Jakob’s
research included probability theory, the integral and differential calculus, and the calculus of vari-
ations wherein he solved the isoperimetric problem and, along with several other prominent mathe-
maticians, the famous brachistochrone problem proposed by his younger brother Johann.
 SEC. 2.6 / BERNOULLI'S EQUATION 53

Bernoulli’s equation is an example of a nonlinear equation that can be reduced

to a linear equation by means of a suitable change of variable. For example, if we
make the substitution

N
II
<

then

ea) yey
and (60) can be reduced to the linear DE in z

oS ee (le nate = Cl’ = 1)y (x), (61)

which is solvable by our previous technique.

EXAMPLE 23 Solve dy + 2xydx = xe™* y> dx.

Solution Dividing by dx, we find

y! + 2xy = xery3,
which is a Bernoulli equation with n = 3. Therefore we set
we yi se y?,

from which we obtain the linear DE

z' — 4xz = —2xe.
Here we calculate

z= ee ia rca + Cye*

=se*
it ee
3 + Ce?noo) I

or, changing back to the dependent variable y,

By 2 =e + Ce.

EXAMPLE 24 Solve the initial value problem

—dPdt = aP — bP, P(0)

=Po (* #2).
where a and b are fixed constants.

Solution Rearranging terms, we have

dP i
— —aP = —bP’,
dt 4
CHAP. 2 / SOLUTION TECHNIQUES FOR FIRST-ORDER EQUATIONS

which is recognized as a Bernoulli equation with n = 2. Making the substitution

z = P' leads to the linear DE in z and initial condition
dz 1
et Gee aes MDS

Here the general solution is

z(t) = en |bem at ge ery aes

— B ae Ges
a
Imposing the initial condition (0) = 1 /Po on z(t), we find C; = 1/Po — b/a, and
hence

DPot (Gee Obeen =

3) = aP>

Since z(t) and P(t) are reciprocals, the solution we seek is

CeaP»

OO bP + (a — bPaje*
We should also observe that P = a/b is a singular solution of the Bernoulli
equation, but it does not satisfy the prescribed initial condition.

EXERCISES 2.6

Solve the given Bernoulli equation.

Oa 2B) Spey 2. y'=ay — by® (a, b constants)

oxy ot Quy y° 4. xy’ = (0 Pay = ay

"5. 6y dx — x2x° + y)dy =0 "6. y' +y = Gy) ¥O)= 5

REFERENCES

Boyce, W. E., and DiPrima, R. C. 1977. Elementary differential equations. 3rd ed. New York:
Wiley.

Coddington, E. A. 1961. An introduction to ordinary differential equations. Englewood Cliffs,

N. J.: Prentice-Hall.

Finney, R. L., and Ostberg, D. R. 1976. Elementary differential equations with linear algebra,
Reading, Mass.: Addison-Wesley.

Ince, E. L. 1956. Ordinary differential equations. New York: Dover.

Reiss, E. L., Callegari, A. J., and Ahluwalia, D. S. 1976. Ordinary differential equations with
applications. New York: Holt, Rinehart & Winston.
 First-order DEs can be applied in many diverse areas of the physical and life
sciences. In spite of this, we often find the same DE evolving out of our
mathematical formulations of problems, regardless of the field of application. That
is, the DE together with an initial condition is simply a mathematical model that
someone has developed of a problem under study, and the same model can often
be used to describe different phenomena, such as the motion of a free-falling body
and_the growth rate of a certain type of bacteria. Once the problem has been
_ properly formulated mathematically, the origin of the problem makes little difference
until we need to interpret the solution.
In Section 3.2 we discuss some general elementary applications wherein the
governing DE may be either linear or nonlinear. These applications include finding
orthogonal trajectories of a given family of curves, finding the velocity of free-falling
bodies, applying Torricelli’s law to the flow of water through an orifice, and finding
curves of pursuit that describe the path of a pursuer tracking its prey.
We restrict the applications in Section 3.3 to those involving linear equations.
Here we discuss problems of growth and decay, free-falling bodies, electric circuits,
cooling bodies, and the mixing of two solutions. Again we point out that all of these
areas of application have basically the same governing DE.

oo

55
 3.1 INTRODUCTION

Differential equations were first applied in science problems, viz., problems in

mechanics. Most of these problems involve quantities that change in a continuous
manner, such as distance, velocity, acceleration, and force. This is not the case,
however, in problems in the life sciences where the quantity of interest may be a
particular population size. Clearly, the total population of a community changes by
discrete amounts rather than continuously; for that reason we might not expect to
describe such changes by derivatives, or DEs, since these concepts are meaningful
only for continuous variables. This is indeed the case for small populations, but if
a population size is sufficiently large, it can often be modeled as a continuous
system. That is to say, the continuous system may describe the general character-
istics of the problem being studied and even predict certain results that can be
verified experimentally. The justification for using such a model then depends
simply on whether it works! Because continuous models have proven effective in
a number of different problems in the life sciences, the use of DEs in this area has
grown significantly in recent years.
In the application of DEs, we are concerned with more than just solving a
particular initial value or boundary value problem. The complete solution process
consists of the following three steps.

1. Construction of a mathematical model. The variables involved must be

carefully defined and the governing physical laws identified. The mathe-
matical model is then some equation(s) representing an idealization of the
physical laws, taking into account some simplifying assumptions in order to
make the model tractable.
2. Solution of the mathematical equation(s). When permitted, exact solutions
are usually desired, but in many cases one must rely on approximate solu-
tions; in this case it is reassuring to be able to establish the existence and
uniqueness of a solution of the model.
3. Interpretation of the results. The solutions obtained should be consistent with
physical intuition and physical evidence. If a good model has been con-
structed, the solution should describe many of the essential characteristics of
the system under study.

3.2 GENERAL APPLICATIONS

The examples in this section are typical of applications that lead to first-order DEs.

3.2.1 Orthogonal Trajectories

It is well known in analytic geometry that the slopes of perpendicular lines are
negative reciprocals; i.e., m, = —1/mp. In a more general setting, we say that two
intersecting curves are orthogonal if and only if their tangent lines are perpendicular
at the point of intersection.
56
 SEC. 3.2 / GENERAL APPLICATIONS 57

Suppose we have the one-parameter family of curves defined by

FX ye) = 0; (1)
where each member of the family corresponds to a particular value of the parameter
c. In certain applications it is important to be able to obtain a second family of
curves given by

g(x, y,.k) = 0 (2)

with the property that all intersections of the two families are orthogonal.* The two
families are then said to be orthogonal trajectories of each other. This means that
their slopes at the points of intersection are negative reciprocals (see Figure 3.1).

Figure 3.1 Curves and their orthogonal trajectories.

Suppose we imagine (1) to be the general solution of a DE having the form

Vil Tey): (3)

It then follows that the DE whose general solution is (2) must be

4
=p
F(x, y).
(4)
The procedure is therefore to find a DE (3) for which family (1) is a general
solution, and then obtain the orthogonal trajectories as solutions of (4).

EXAMPLE 1 Find the orthogonal trajectories of the family of circles

xe+y?=c’.

Solution The constant in the equation x? + y* = c’ can be eliminated by implicit differ-

entiation, leading to

l curves (i.e.,
*For instance, in an electric field the lines of force are orthogonal to the equipotentia
curves of constant potential).
58 CHAP. 3 / APPLICATIONS OF FIRST-ORDER EQUATIONS

aE
eoie aa iby Set
te) 0,

or
2x + 2yy' = 0.
Solving for y’, we find
: x
a ss ae

and thus the DE for the family of orthogonal trajectories is

vi y a)
ee ul

with general solution

y = kx.
Hence, the orthogonal trajectories are straight lines passing through the origin (see
Figure 3.2).

> Xx

Figure 3.2

EXAMPLE 2 Find the orthogonal trajectories of the family of parabolas y? = 4px.

Solution If we rewrite the equation as.y*/x = 4p, then differentiation leads to the DE

yas2a
The orthogonal trajectories must therefore satisfy the DE

pees y ’

leading to the family of ellipses given by 2x? + y? = k? (Figure 3.3).

SEC. 3.2 / GENERAL APPLICATIONS 59

Figure 3.3

3.2.2 Free-Falling Bodies

The basic principle of mechanics used in studying particle motion is Newton’s

second law of motion, which states that “the time rate change of momentum of a
body is equal to the resultant force acting on the body.” The momentum of a body
is defined by the product mv, where m denotes the mass of the body and v its
velocity, so that in symbols Newton’s law reads
d
=7 (mv) =F.

If the mass is considered constant, this equation becomes

dv
a F
dt
Since a = dv/dt is the acceleration of the body, Newton’s second law is often
formulated
F = ma.

Consider the problem of a free-falling body that is acted upon only by the force
of gravity. If the body is close to the surface of the earth, the weight of the body
is essentially the constant mg, where g is the gravitational constant. It has been
determined experimentally that at sea level g = 32 ft/s* (English system) or
g = 980 cm/s? (metric system). Newton’s law in this case is simply
oD
ae = mg,

where our sign convention is such that the positive direction is downward.
If the body encounters air resistance as it falls, then the weight mg must be offset
somewhat by a resistive force Fr. In such a case the governing equation is
dv
me = mg — Fr. (5)
60 CHAP. 3 / APPLICATIONS OF FIRST-ORDER EQUATIONS

The amount of air resistance depends upon the velocity of the body, but a general
law expressing this dependency is not known. We often make the assumption that
Fr = cv or Fr = cv”, where c is some positive constant.

EXAMPLE 3 A body of mass m is dropped from a height of 5000 feet. If the air resistance is
described by Fr = mv’/40, find the velocity of the body for any time t.

Solution At time t = 0, the velocity of the body is assumed to be v = O. At any later time,
the velocity is described by the solution of the DE
dv mv ~ Moye
FeBerge, :
Separating the variables, we find
dv
5)
Ate
40g —v*
Ae
\40
! =
which leads to

1 " (ae) =a
fiemaG:
AV 10g : dV 10g —"6 40
Imposing the auxiliary condition v = 0 when t = 0, we see that C = 0. Hence we
rewrite this last expression as
2V 10g ata) = 9 Velo:

PASAY gb
and, solving for v, we get the desired result
Veilor —

e Ve/l0r + |

=e, i =p
1 —e Vegi0r

1 +e" Veil0
Observe that as t>, we find v2 V 10g. This is called the terminal velocity,
or limiting velocity, of the body.

At large radial distances r from the center of the earth (distances beyond sea
level), the weight of a body differs from the weight at sea level, since the acceler-
ation a is not equal to the constant g. According to Newton’s law of gravitation,
the acceleration of a body is inversely proportional to the square of the distance
from the center of the earth. In symbols we write

ae: ier.
 SEC. 3.2 / GENERAL APPLICATIONS
61

where k is a proportionality constant. If the body is falling to earth, the constant kis
positive, since the velocity is increasing; if the body is leaving the earth, k is
negative. Suppose we denote the radius of the earth by R and consider a body
projected upward (Figure 3.4). Then a = —g whenr = R, and thus —g =k/R’,
or :
k = —gR°.

Earth’s
surface

Figure 3.4

Invoking Newton’s second law, we get the equation of motion

dv mgR?
dt aE i (6)
For solution purposes we wish to express the velocity v in terms of the distance
variable r. To do so, we observe that

dv _ dvdr dv
dt drdt dr’
and thus (6) becomes

ee ee (7)

If initially the body left the earth’s surface with velocity uv, then we have the
auxiliary condition v(R) = vo. Hence, a particle projected radially outward from
the earth‘s surface with velocity vp will have the velocity given by
2

Pe a oR. r>R (8)

r

when the body is r units from the center of the earth.

To calculate whether the velocity vo of the body is sufficient for the body to
escape the earth’s gravity, we note that the velocity v must remain positive, for
62 CHAP. 3 / APPLICATIONS OF FIRST-ORDER EQUATIONS

otherwise the body will stop and then fall back to earth with negative velocity. The
critical initial velocityvp is such that v6 — 2gR = 0, or vy = V2gR. The minimum
value of v, is

Ve = V22gR, (9)
called the velocity of escape.

Remark. Observe that air resistance was neglected in obtaining (9).

EXAMPLE 4 Given that R = 3960 miles and g = 32 ft/s’, determine the escape velocity of the
earth.

Solution Converting units, we have that g = 32 ft/s* corresponds roughly tog = 6 X 1073
mi/s*. Therefore,
Ve = V(2)(6 X 10-*)(3960) = 6.9 mi/s.
Of course, the escape velocity for other heavenly bodies, such as the moon or
Mars, will be different, since both g and R are different for these bodies.

3.2.3 Flow of Water Through an Orifice

Consider a tank filled with water that is pouring out near the bottom of the tank
through an orifice. Torricelli’s law* states that the velocity with which water issues
from an orifice is

v = 0.6V 2gh,
where g is the gravitational constant and /h is the instantaneous height of water
above the orifice (Figure 3.5). Hence, if V is the instantaneous volume of water in
the tank, it follows that

dV
a = Av = —(06A \V 2eh, (10)

where A is the cross-sectional area of the orifice.

Figure 3.5

*Named in honor of the Italian physicist EVANGELISTA TORRICELLI

(1608-1647).
 SEC. 3.2 / GENERAL APPLICATIONS
63

For solution purposes it may be advantageous to express V in terms of the height

h and use the chain rule
dV _ dVdh
dt dhdt’
If B(h) denotes the cross-sectional area of the tank at height h, then dV/dh = B, and
(10) becomes

dl
oe = —0.6AV2¢h. (11)

oe Sag ee ee ee ee ee ee, SS ee) ee ee ee

EXAMPLE 5 How long will it take to empty a cylindrical tank of radius } foot and vertical axis
2 feet if the tank is initially full of water and the orifice is a }-inch hole in the bottom
of the tank?

Solution The volume of the water in the tank at a height of h units is V = 7 (4)7h. Therefore

wv
dh 4”
and (11) becomes

1edh Bats
—a— =) 675 75 V 2(32)h.

Separating variables, we obtain the general solution

2h fe end G.
ay oe aa
770°

The initial condition is h(0) = 2, so thatC = 2V2. Solving fort when h = 0 then
leads to the result
t = 270(2V2) = 764 seconds,
or approximately 12 minutes and 44 seconds to empty the tank.

=
NJ) /
iS j 3.2.4 Curves of Linear Pursuit

_A curve of pursuit is a path generated by a point P that is always moving in the

direction of a second point Q constrained to move along a prescribed path. In other
words, it is the path of a pursuer tracking its prey. The problem of finding such a
curve seems to have originated with Leonardo da Vinci in the fifteenth century, but
its curious difficulties still intrigue modern mathematicians.
The general problem of determining pursuit curves is very difficult, but certain
special cases lend themselves to solution methods that we have already discussed.
For simplicity, let us assume the “prey” is located at Q and constrained to move
64 CHAP. 3 / APPLICATIONS OF FIRST-ORDER EQUATIONS

P(x, y)

0 (a, 0)

Figure 3.6 Pursuit curve.

along the y-axis (see Figure 3.6). The pursuer is assumed to be at the point (a, 0)
when Q is at the origin. It is further assumed that the speeds of the pursuer and prey
are always in the same constant ratio.
To illustrate, let us imagine a large field in which a fox is located at the point
(a, 0). He spots a rabbit at (0, 0) running along the y-axis in the positive direction
with constant speed v. The fox immediately runs toward the rabbit with speed w,
and after t seconds the rabbit is now at Q(0, vt) and the fox at P(x, y). The line PO
is always tangent to the path of the fox and therefore has the slope
y= tt
Yet ie
: x (12)
In order to solve (12), we need to eliminate the parameter t. To do so, we observe
that the length of the path traveled by the fox at speed w can be computed by the
arc length formula

wt = [VI \G) dx. (13)

Solving (12) and (13) for ¢ and equating the resulting expressions, we find

Seek allV1 + (y'P ae.

= ' 1 a

(14)
Although (14) at first appears to be formidable, let us differentiate it with respect
to x to get (after simplification)

xy" = “<VitGy =kV1+ (y"),

w
where & is the ratio of the two speeds. Now setting p = y’ leads to

xp’ =kV1 +p’,

which can be separated according to
dp k
ek
Vila ey
Integrating both sides of this expression, we have

log (pFaNe | ¥p") = klogx — C,. (15)

SEC. 3.2 / GENERAL APPLICATIONS
65

At time t = 0, the slope of the pursuit curve is zero, so that we write p = 0 when
= a; hence, C; = kloga. Using properties of logarithms and some algebra, we
can rewrite (15) in the form

r=v=al(e) -@)}
If we assume that the fox runs faster than the rabbit, i.e., k < 1, the integral of
this last expression yields the general solution

y= tafe ea),
1 x/

lee %
EEK

Pas
lek

Since y = 0 when x = a, we deduce that C, = ak/(1 — k’), and thus

leita), ala); ak
‘ ES t—-k | 1-# a
=> ———-—- — + ‘

The fox will catch the rabbit when x = 0, and this happens when y = ak/(1 — k’).
Those cases for which the two speeds are equal or the rabbit runs faster than the
fox are taken up in the exercises.

EXERCISES 3.2

In problems I-12, find the orthogonal trajectories of the given family of curves.

1) xy =e 2 Eee eke @.) yaxte

4x2 c Oot aie auc 6506-62 = ¢
8. Circles through the 9.) y? = 4x°(1 — cx)
ety
origin with centers
on the y-axis
sates 3xale ice —3x 1/3 iso c [ex
105. sy Tivexeseey 120 ey, = ewes
*13. In the calculus it is shown that the angle w measured positive in the counterclockwise
direction from the radius vector to the tangent line at a point satisfies the relation
(Figure a)

. tanw =r—,

where r and @ are polar coordinates. If two curves in polar coordinates are orthogonal,
show that (Figure b)

Problem 13(a) Problem 13(b)

 CHAP. 3 / APPLICATIONS OF FIRST-ORDER EQUATIONS

1
tan WN = ~ tanya’

In problems 14-18, find the orthogonal trajectories of the following families of

curves using the result of problem 13.

14. r =c(1 + cos@) 15. r=c(l1 — sin@) 16. r = 2ccosé

17. r =ccos’?6 18. r?=csin26
“19. A family of curves that intersect another family of curves at a constant angle a # 90°
are called isogonal trajectories. If y' = F(x, y) is the DE of the given family of
curves, show that
ee Can) eetanigay
ewes ce y)tan @
is the DE of the isogonal family.
*20. Referring to problem 19, find the isogonal families of the following families of
curves.

(a) y(xt+ec)=1; a = 45°

(b) y=cx; a = 45°
(c) y=cx; a = 30°

(ny e— a Gaetan (4)

A given family of curves is said to be self-orthogonal if it has the property that its
family of orthogonal trajectories is the same as that of the given family. Verify that
the family of parabolas y* = 2cx + c? is self-orthogonal.
Solve Example 3 when the resistive force is given by Fg = mv/40. Find the limiting
velocity of the body in this case.
Solve Example 3 when the resistive force is cv. What is the limiting velocity?
Suppose a parachutist falls from rest toward the earth. Assume the chutist weighs 160
pounds and his speed is 30 ft/s at the instant the chute opens (t = 0). If the air
resistance is Fr = cv?
(a) determine the Pca v at any later time t.
(b) What is the skydiver’s limiting velocity as t>%?
(c) Calculate the limiting velocity if the air resistance is cv instead of cv?.

A parachutist weighing 160 pounds falls from rest toward the earth. Before the
parachute opens, the air resistance is equal to $v. The chute opens 5 s later, ane the
air resistance changes to 5v7/8. Find the velocity of the skydiver
(a) before the parachute opens.
(b) after the parachute opens.
(c) If the parachute never opened, what would be the limiting velocity of the
skydiver? Compare this value with the value obtained after the chute opens.

*26. A man and his parachute together weigh 192 pounds. Assume a safe landing
velocity
 SEC. 3.2 / GENERAL APPLICATIONS 67

is 16 ft/s and that the air resistance is known to be proportional to the square of the
velocity, equaling 5 pound for each square foot of cross-sectional area of the parachute
when it is moving 20 ft/s. What is the cross-sectional area of the parachute necessary
for the chutist to make a safe landing?
27. If it takes time T for a ball thrown upward to reach its highest point (Fr = 0), show
that the return time is also 7. What is the velocity of the ball upon return if the initial
velocity is Uo?
28. Determine the escape velocity from the moon given that the moon’s radius is roughly
1080 miles and the acceleration of gravity is 0.165g, where g is the acceleration of
gravity on the earth’s surface.
29. Determine the escape velocity from Mars given that its radius is 2100 miles and the
acceleration of gravity is 0.38g.
30. Given that the force of gravity on Venus is about 85% of the earth’s gravity and the
radius of Venus is roughly 3800 miles, determine the escape velocity.
31. At 200 miles above the earth’s surface, the atmosphere offers almost no resistance.
What velocity should a rocket have at this altitude in order to reach a height of 4000
miles if all its fuel is exhausted at this point?
*32. If a body is shot straight up from the earth’s surface with an initial velocity vp and no
air resistance is assumed, show that the rising time t as a function of the distance r
of the body from the center of the earth is given by

a Ca 7VAT pn sper Arcsin (24),

where A = 2gR*, B = 2gR — vo > 0, and C is a constant to be determined by the

initial condition t = 0,r =R.
Hint: Show that dr/dt = 1/r Ar — Br’.
33. How long does it take to empty a cylindrical tank of height 20 feet and radius 10 feet
initially full of water if the cross-sectional area of the orifice in the bottom is 1 ft”?
34. A conical tank of circular cross section standing on its apex whose angle is 60° has
an outlet of cross-sectional area 0.5 cm”. The tank is initially full of water, and at time
t = 0 the outlet is opened and the water flows out. Assuming the height of the tank
is 1 meter, how long will it take for the tank to empty?
35. A water tank in the shape of a paraboloid of revolution measures 6 feet in diameter
at the top and contains water 3 feet deep. When will the tank empty if a hole at the
bottom has a 1-inch diameter?
*36. A large hemispherical cistern with a 25-foot radius is filled with water. A circular hole
of radius 1 foot is cut into the bottom of the bowl. How long will it take for the cistern
to empty?
37. In the fox-rabbit problem discussed in Section 3.2.4, find the path of the fox when
k = 1; i.e., when v = w. Will the fox ever catch up with the rabbit?
38. In problem 37, show that the fox will never get even as close as 4a to the rabbit.

39. Referring to the fox-rabbit problem in Section 3.2.4, find the fox’s path when v > w,
or k > 1. Compute the distance between the fox and the rabbit in terms of the vari-
able x.
 CHAP. 3 / APPLICATIONS OF FIRST-ORDER EQUATIONS

*40. A man standing at O holds a rope of length a to which a weight is attached, initially
at the point (a, 0). The man walks along the positive y-axis, dragging the weight after
him (see figure).

(a, 0)
Problem 40

(a) Show that the slope of the path along which the weight moves is

Hint: Find tan @ directly from the figure.

(b) Solve the DE in (a) to find the path of the weight. This particular curve is called
a tractrix.
(c) Show that the solution in (b) can be expressed both in terms of a logarithm and |
in terms of the inverse hyperbolic secant.
41. Two skaters are located on the x-axis, Q at the origin and P at the point
(36, 0).
Suppose that Q skates along the positive y-axis and that P skates directly toward
Q
at all times. If P skates twice as fast as Q, how far will Q travel before being caught
by P? Answer the question if P skates 3 times as fast as Q.
*42. A pilot always keeps the nose of his plane pointed toward a city C due west of
his
starting point at (a, 0). If the plane’s speed is v mi/h, and a wind is blowing from
the
south at the rate of w mi/h, show that the equation of the plane’s path is

where k = w/v.
medida |
43. In problem 42, if the wind speed and plane speed are equal, show that the
path is that
of a parabola. Will the pilot ever reach city C?
44. Find the equation of a curve that passes through the point (4,
1) and has slope
—y/(x — 3) at any point (x, y) on the curve.
45. Find the shape of a curved mirror such that light from a distant source
will be reflected
to the origin.
Hint: The slope of such a curve in the xy-plane is given by

y=
) SN EEE RE aie
y
 SEC. 3.3 / APPLICATIONS INVOLVING LINEAR EQUATIONS 69

*46. Ona winter day it began snowing early in the morning, and the snow continued falling
at a constant rate. The speed at which a snowplow can clear a road is inversely
proportional to the height of the accumulated snow. The snowplow started at 11:00
A.M. and had cleared 4 miles of road by 2 P.M. Another 2 miles was cleared by 5 P.M.
At what time did the snow begin?
“47. Acable of constant density hanging from two pegs (such as a telephone line) assumes
a shape determined by the DE
ytkVl+o'y¥=0
where k is a constant. Show that the cable assumes the shape of a hyperbolic cosine,
called a catenary.
*48. (Brachistochrone problem) One of the most famous problems in mechanics is called
the brachistochrone problem. The problem is to determine the curve along which a
particle will slide (without friction) from point O to point P in the shortest time where
gravity is the only acting force. Point P is below O, but not directly beneath it. The
curve that solves the problem is a solution of the nonlinear DE (derived through
principles of the calculus of variations)
[1 + (y')]y = 2a,
where a is a constant.
(a) Solve the DE for y’ using the negative square root. (Why use the negative square
root?)
(b) Introduce a new variable ¢ through the relation
y = —2asin’t
and solve the resulting DE.
(c) By writing 0 = 2t, show that the solution of the original DE can be expressed
in the parametric form
ve alee sin 0),; y = — a(1 = cos @);

which satisfies the auxiliary condition y(0) = 0. These last equations are the
well-known parametric equations of a cycloid.

Jeg

Problem 48

3.3. APPLICATIONS INV@LVING LINEAR EQUATIONS

First order linear DEs occur in a wide variety of applications, some of which are
discussed in the following subsections.
70 CHAP. 3 / APPLICATIONS OF FIRST-ORDER EQUATIONS

3.3.1 Growth and Decay Problems

The simple linear DE

eles (17)
arises in numerous physical theories concerning either growth or decay of some
entity. For instance, Equation (17) might describe the rate at which a radioactive
substance decomposes or the rate at which temperature changes in a cooling body.
This same equation might be used to predict the population growth of certain small
animals over short intervals of time or to describe the growth rate of certain
bacteria.
The graphs of solutions of (17) are shown in Figure 3.7 for cases when k > 0
and k < 0. Since dy/dt represents the slope of y, the sign of k gives an indication
of whether the function y is increasing (k > 0) or decreasing (k < 0).

Figure 3.7 Solutions of dy/dt = ky

EXAMPLE 6 (Radioactive decay problem) Experimental evidence indicates that a radioactive

substance decays at a rate directly proportional to the amount present. Starting at
time t = 0 with yo grams of undecayed matter, find the amount present at some later
time.

Solution If y denotes the amount of undecayed matter at any time f, then the experimental
evidence is mathematically described by the initial value problem

dy
It
hh ky, > y(0)=y,
= -) k>O0
=

where the negative sign indicates that y is a decreasing function. This is a homoge-
neous equation with general solution
y a Ges

and application of the initial condition determines C, = yo. Thus

 SEC. 3.3 / APPLICATIONS INVOLVING LINEAR EQUATIONS
71

a
y(t) = woe ™.
Se a mre ce ee

In physics, the stability of a radioactive substance is measured in terms of its

half-life, i.e., the time it takes for one-half of the atoms in an initial sample of the
substance to decompose. The longer the half-life is, the more stable the substance.
For instance, radium that decomposes quite rapidly has a half-life of approximately
1700 years, whereas the isotope uranium 238 has a half-life of about 4.5 x 10°
years.
An interesting application involving radioactive decay is in approximating the
age of fossils. It is known that cosmic radiation interacting with nitrogen produces
the isotope carbon 14 in the atmosphere. Curiously, the ratio of carbon 14 to
ordinary carbon in the atmosphere is roughly constant. Thus, the same proportion
of carbon 14 is found in the bodies of all breathing creatures as in the atmosphere.
By comparing the amount of carbon 14 present in a fossil with the constant ratio
found in the atmosphere, the age of the fossil (i.e., the time when the organism
stopped breathing) can be reasonably approximated. The method, which is based
upon knowledge of the half-life of carbon 14 (see problem 3 in this section), was
devised by Willard Libby and won him the Nobel Prize for chemistry in 1960.

EXAMPLE 7 (Population growth problem) The population of a particular community is observed

to increase at a rate proportional to the number of people present at any one time.
In the last 5 years the population has doubled. How many years will it take for the
population to triple?

Solution The governing equation is

dP_ le.
——i—
dt

where P(t) denotes the population at any time and a is a positive constant (since P
is increasing). Let us assume the population is Po at time t = O, although we don’t
know its numerical value. The solution of this initial value problem is therefore

P(t) = Poe”.
Now when t = 5 years, it is known that P(5) = 2P), which leads to

2Po = Poe’ or EEN IN

and solving for a yields the value

a = =log2 = 0.1386.

Thus
72 CHAP. 3 / APPLICATIONS OF FIRST-ORDER EQUATIONS

P@ = Pye? B®
To find the time at which the population has tripled, we solve
3P) = Poe? 186
for t. Taking logarithms of each side of this expression, it follows that
0.1386r = log3
or
=] .9 yeats:
pete ie. Ot ee ee eee ee ee eee ee

The simple equation

dP
me aP (18)

used in Example 7 to describe population growth is open to criticism, since pro-

longed exponential growth is unrealistic for most populations. Thus, Equation (18)
is generally assumed valid only in the initial stages of growth but not over a long
period of time.
To offset the rapid growth predicted by (18), an inhibitive factor, which is
sometimes referred to as the “death rate,” is usually introduced into the model. One
of the earliest mathematical formulations of the problem employing such a term was
suggested in 1844 by the Belgian mathematician P. F. Verhulst. He assumed that
~ the inhibiting factor was proportional to —P?, resulting in the DE

dP ‘
i aPe—iDPe, (19)

where both a and b are positive constants. This equation is now known as the
logistic equation, and its solutions are referred to as logistic curves.
Although (19) still generally does not provide a very accurate model for human
population growth, it has proven quite effective in predicting the growth patterns
(in a limited space) of fruit flies, for example, and certain types of bacteria.
The solution of (19) subject to the initial condition P(0) = Pp was given in
Example 24 in Section 2.6,7 viz.,

aPo
re
OT FRG bP jee ey)

*This solution can also be expressed in the equivalent form P(t) = Po2°, which more clearly reveals
that P(t) doubles every 5 years.
+Equation (19) is also solvable by separation of variables.
SEC. 3.3 / APPLICATIONS INVOLVING LINEAR EQUATIONS
73

Figure 3.8 The logistic curve.

The general shape of (20) is illustrated in Figure 3.8 for the cases when
0 < Po < a/2b, a/2b < Py < alb, and Py > a/b. In the first two cases, the value
P = a/2b cuts the graph at a point of inflection, as can be verified by examining
the second derivative P”(t). Regardless of the value of Py, we observe that

lim P(t) = ; (saturation level).

When the initial population Py exceeds the value a/b, then P(t) approaches this
value asymptotically from above rather than from below. Finally, if the initial value
of the population is a/b, it has this population value for all t.

3.3.2 Motion of a Particle and Simple Electric Circuits

Let us imagine a particle of mass m moving in a straight line. Newton’s second law
of motion states that the total force acting on the particle is proportional to the
acceleration a of the particle (see also Section 3.2.2). In symbols, we write

F = ma, (21)

where the mass m is the constant of proportionality. But since acceleration is the
rate of change of velocity, Equation (21) can also be expressed as a first-order DE,

dv
Le rept 22
(22)

In those problems for which the force is due entirely to gravity, then F = mg, and
(6) becomes simply

m— = meg, (23)

where g is the gravitational constant (approximately 32 ft/s*, or 980 cm/s”). Some-

times the particle may also encounter a resistive force Fr due to the medium in
which the motion takes place (see Figure 3.9). For example, the air resistance
encountered by a parachutist is such a force, as is the resistance offered by a viscous
74 CHAP. 3 / APPLICATIONS OF FIRST-ORDER EQUATIONS

Fp

Mass

mg
Figure 3.9 Free-falling body.

liquid into which a ball is dropped. Such a resistive force is frequently proportional
to the velocity of the moving mass, i.e., Fr = cv, so that the governing equation
is then modified to

m aS OTC eee a
dt 2 ; ‘
or
dvaee
ae (24)
dimeaetit 2

The value of the positive constant c is determined by the nature of the resistive
force.

EXAMPLE 8 (Particle motion problem) A parachutist weighing 150 pounds opens his chute when
_ his downward velocity is 100 ft/s.
(a) If the force of air resistance is 25v, find the velocity of the parachutist at any later
time (prior to hitting the ground).
(b) What is the limiting velocity of the parachutist?
(c) If distance s is related to the velocity by ds/dt = —v and the parachutist opens
his parachute at 2000 feet, how close to the ground will he be after 5 minutes?

Solution The governing equation is

d
m au +250 = mg, v(0) = 100.
dt
Since g = 32 ft/s* and the mass is m = 150/32 slugs,* the equation reduces
to

— + —p = 32, (0) = 100.

(a) The solution of this initial value problem is readily found to be

v(t) = 6 + 94e~ 163

*A “slug” is a unit of mass equal to | Ib/(ft/s’).

 SEC. 3.3 / APPLICATIONS INVOLVING LINEAR EQUATIONS
75

which gives the velocity at all later times.

(b) The limiting velocity* is found by taking the limit of v as to, i.e.,

Vx. = 6 ft/s.
(c) Since the position s of the parachutist is determined from the initial value
problem

7 = 2,t (0)
Cpe: 0) =
= 2000 20 ,

a single integration yields the general solution

SOS 561-1 7.65 Gi

Setting t = 0, we see that C; = 1982.4, and since 5 minutes correspond to 300
seconds, the position of the parachutist at this time is

s(300) = —1800 + 1982.4 = 182.4 feet

above the ground.

Figure 3.10 RC circuit.

Consider the single-loop electric circuit containing a resistor R and capacitor C

connected in series with a voltage source (battery) E(t), as shown in Figure 3.10.
Kirchhoff’s second law states that the sum of instantaneous-voltage drops across
each part of the circuit is equal to E(t). From experimental observations, we get
voltage drop across a resistor = Ri,
: aed
voltage drop across a capacitor = Cc

where qg denotes the electric charge on the capacitor and is related to the current i
by i = dq/dt. Thus, the governing equation is

*From a practical point of view, the velocity approaches the limiting velocity in a relatively short
interval of time and remains at this value from that point on.
+Named in honor of the German physicist, GUSTAV R. KIRCHHOFF (1824-1887).
76 CHAP. 3 / APPLICATIONS OF FIRST-ORDER EQUATIONS

q
Ri i+ C—=E(p),
(t)

or, in terms of charge q, we find

dag wast
—+--—qc4 = E(t).
Ro (t) (25)

For the special case when E(t) = Eo, a constant, we find that (25) has the general
solution

q(t) = et |pence BEN OfY mel RES) TORN EYEE

Se LO (26)

Also, since i(t) = dq/dt, it follows that the current is

: Car, e de
i(tt)= - ae eI 1g Coe (27)

where ip is the initial current in the circuit.

EXAMPLE 9 (Electric circuit problem) If a resistance of 2000 ohms and a capacitance of

5 x 10~° farad are connected in series with a voltage source of 100 volts as shown
in Figure 3.10, what is the current at any time, given that i = 10 * ampere at the
time the switch is closed (t = 0). Also compute the initial charge on the capacitor.

Solution . Substituting R = 200, C = 5 Xx 10~°, Ey = 100, and iy) = 10? into (27) we get

it) = 10° "ee

To obtain the initial charge q(0), we observe that the constant C; appearing in (26)
is related to the initial current ip by [see (27)|
C,; = —RCiy = — 10%,
and therefore
q(t) = 5 X 10-4 — 10~4e
=100",
which yields the initial value g(0) = 4 x 10~‘ volt.

3.3.3 Cooling and Mixing Problems

Newton’s law of cooling states that “the rate of change of temperature u in a cooling
body is proportional to the difference between u and the temperature Ty of the
surrounding medium.” In symbols, this law reads
d
$ =e k(us— 15); (28)

where k is the proportionality constant.

 SEC. 3.3 / APPLICATIONS INVOLVING LINEAR EQUATIONS 7

EXAMPLE 10 (Cooling problem) A metal ball is heated to a temperature of 100°C and then
immersed in water at temperature 30°C at time t = 0. After 3 minutes the tem-
perature of the ball is reduced to 70°C. Find its temperature at all later times.

Solution The governing equation is (28) with Ty) = 30°C. (It is assumed that the water can
be maintained at 30°C even with the metal ball immersed).* The solution of (28)
satisfying the initial condition u(0) = 100 is
u(t) = 30 + 70e™.
The additional information u(3) = 70 allows us to determine a value for k, 1.e.,

70 = 30 + 70e *,

or

1 t/
k = 3 log (7)
— | = 0.1865.

Using this value of k, we see that subsequent temperatures are given by

u(t) = 30 + 70e"°18*,
Observe that if we wait long enough (>), the temperature of the metal ball will
eventually approach 30°C, the temperature of the water. From a practical point of
view, this will happen in approximately 15 minutes (see Figure 3.11).

Time (minutes)

Figure 3.11

The mixing of two solutions can also give rise to a first-order linear DE. In the
next example we will consider the mixing of pure water with a salt brine solution.

EXAMPLE 11 (Mixing problem) A certain tank contains 100 gallons of a solution of dissolved salt
and water, the mixture being kept uniform by stirring. If pure water is now allowed

large compared with the size of the metal ball.

*This will be true if the volume of water is relatively
78 CHAP. 3 / APPLICATIONS OF FIRST-ORDER EQUATIONS

to flow into the tank at the rate of 4 gal/min, and the mixture flows out at the rate
of 3 gal/min (Figure 3.12), how much salt will remain in the tank after ¢ minutes
if 15 pounds of salt is initially in the mixture?

4 gal/min

3 gal/min

Figure 3.12

Solution Let us denote the amount of salt present in the tank at any one time by x(t). The net
rate at which x(¢) changes is given by
dx ;
¥ = rate of salt in — rate of salt out.

Since pure water is coming in, the rate of salt entering the tank is zero. The rate
at which salt is leaving the tank is the product of the amount of salt per gallon and
the number of gallons per minute leaving the tank, i.e.,

rate of salt out = (3 galiin §Ibgal),

The governing equation is therefore

dx
i x/V, x(0) = 15

However, the volume V of the mixture is not constant but is determined by

dV
—, = rate of liquid in — rate of liquid out = 4 — 3 = 1,
dt
or
VO =t+C,.
Because at time t = 0 the volume is known to be 100, we find that C; = 100. Thus

: te = =0, x(0) = 15,

the solution of which is readily found to be
TiN 15a 10°
(2-681 00)2
ee ee eee
SEC. 3.3 / APPLICATIONS INVOLVING LINEAR EQUATIONS 79

EXERCISES 3.3

c) The half-life of a certain radioactive substance is 1620 years. If 10 grams are initially
present in a given sample, how much will be left after 162 years?
2: A certain breeder reactor converts uranium 238 into the isotope plutonium 239. After
15 years, 0.043% of the initial amount of plutonium has decayed. What is the half-life
of this isotope?
3) By comparing the amount of carbon 14 found in a fossil with the constant ratio found
in the atmosphere, the age of the fossil can be estimated. Assuming the half-life of
carbon 14 is 5600 years, determine the approximate age of a fossil that is found to
contain 0.1% of the original amount of carbon 14.
A certain chemical is converted into another substance at a rate proportional to the
square of the amount of unconverted chemical. Starting at time ¢ = 0 with an amount
yo of unconverted chemical, determine the amount of unconverted chemical for all
P= 0.
The population of a certain country was | million in the year 1950, and the instanta-
neous growth rate since that time is observed to be 3% of the current population.
Assuming this trend continues, what population is expected by the year 2000?
Suppose that a certain population has grown to 10,000 after 3 years and that after 4
years the population will be double the original amount. What was the original
population, and what will be the population after 10 years if it continues to grow at
a rate proportional to the number of people present at any time?
Tf If we allow a population to change by either immigration into or emigration out of,
then the governing equation is modified to

—dP
LC =aP + mattef(0,
where f(t) is the rate that members of the population are being added or subtracted
from outside the system. Determine the population growth if a = —3 and the immi-
gration is governed by the periodic function
f(t) = 100001 + 5b sin 2).
Assume the initial population is Po, and discuss separately the cases |b| <1 and
|b| > 1; i.e., which case represents immigration and which emigration? Finally,
determine the steady-state population by considering the limit of P(t) as t>~.
The infusion of glucose into the bloodstream is an important medical technique. Let
us imagine glucose is infused into the bloodstream at the constant rate of b grams per
minute. At the same time, the glucose is converted and removed from the bloodstream
at a rate proportional to the amount of glucose present.
(a) Show that the amount of glucose G(t) present at any time is governed by the DE

ass =b—kG (k constant).

dt
(b) What concentration of glucose is attained after a sufficiently long period of time
(720) f
Suppose a sum of money M,p is deposited into a bank that pays 6% interest.
(a) Show that the value M(?) of the investment at the end of t years is given by the
expression
80 CHAP. 3 / APPLICATIONS OF FIRST-ORDER EQUATIONS

0.06\“
M(t) = Mo| 1+ pF ;

where k is the frequency (number of times each year) at which the interest is ©
compounded.
(b) Determine the amount of money in the bank after 10 years ifk = 1, k = 4, and
k = 365.
(c) If the interest is compounded continuously, write the DE for M (?) illustrating this
growth of investment, and show that its solution is the same as obtained by
allowing k—> in the formula given in (a). Determine the amount of money in
the bank at the end of 10 years as predicted by this model.
In problem 9, how long will it take to double the original investment if the interest
is compounded continuously? If k = 1?
A stone weighing 4 pound falls from rest from a tall building. If the air resistance is
known to be v/160, where v is the velocity of the stone, determine v at any time.
If the building in problem 11 is 200 feet high, what is the velocity of the stone when
it hits the ground? (Approximate your answer.)
An object is thrown upward with an initial velocity vo, and the air resistance is cv.
(a) Find the time required for the object to reach its maximum height.
(b) Find the maximum height if c = 7, mg = 4, and vo = 8.
In the RC circuit shown in Figure 3.10, how long will it take the current i(f) to
decrease to one-half its original value if the voltage source is the constant Eo?
Find the steady-state current in an RC circuit when the voltage source is
E(t) = Eosin wt. Hint: Let t—~© in the solution.
“16. A variable resistance R = (5 + t)”' ohm and a capacitance of 5 x 10~° farad are
connected in series with a voltage source of 100 volts. If the initial charge g is zero,
what is the charge on the capacitor after 60 seconds?
17: The current i(¢) in a circuit containing only a resistance R and an inductance L in series
with a voltage source E(t) is governed by
di . ;
tes + Ri =E(), iO) = io,

where L and R are known constants (see figure). Solve this initial value problem when
(a) E(t)= Eo (constant).

000

Problem 17
SEC . 3.3 / APPLICATIONS INVOLVING LINEAR EQUATIONS 81

(b) E(t) = Eosinoat.

(c) R =1 ohm, L = 10 henrys, io = 6 amperes, E(t) = 6 volts forO<t < 10,
E(t) = 0 for t > 10.
A thermometer reading 20°F is brought into a room kept at 72°F. Two minutes later
the thermometer reads 46°F. Find the temperature reading for any time. What is the
temperature reading of the thermometer after 6 minutes?
19, A thermometer is taken from an inside room to the outside, where the temperature is
7°F. After 1 minute the thermometer reads 50°F, and after 4 minutes the reading is
32°F. What is the temperature of the inside room?
20. A pie is removed from a 350°F baking oven. If the room temperature is 75°F, how
long will it take the pie to cool to 100°F if it cooled 150° in the first 4 minutes? How
long will it take the pie to reach 76°? How long to reach room temperature?
21. Fifty pounds of salt are initially dissolved in a tank holding 300 gallons of water. A
brine solution is pumped into the tank at the rate of 2 gal/min, and the (well-stirred)
solution is allowed to flow out of the tank at the same rate. If the salt concentration
entering the tank is 2 lb/gal, determine the amount of salt in the tank at any time. How
much salt is present after 60 minutes? How much salt will remain after a long period
of time?
22. Solve problem 21 if pure water instead of a brine mixture is pumped into the tank at
the rate of 2 gal/min.
23. Solve problem 21 if the mixture is allowed to flow out of the tank
(a) at the slower rate of 1 gal/min.
(b) at the faster rate of 3 gal/min.
24. A particular tank contains 200 liters of a dye solution with a concentration of 1 g/l.
Fresh water is entering the tank at the rate of 2 I/min, and the (well-stirred) solution
is flowing out at the same rate. Find how much time will elapse before the dye
concentration in the tank reaches 1% of its original value.

REFERENCES

Boyce, W. E., and DiPrima, R. C. 1977. Elementary differential equations. 3rd ed. New
York: Wiley.
Coddington, E. A. 1961. An introduction to ordinary differential equations. Englewood
Cliffs, N.J.: Prentice-Hall.

Derrick, W. R., and Grossman, S. I. 1981. Elementary differential equations with applica-
tions. 2nd ed. Reading, Mass.: Addison-Wesley.

Rainville, E. D., and Bedient, P. E. 1980. Elementary differential equations. 6th ed. New
York: Macmillan.

Ross, S. L. 1980. Introduction to ordinary differential equations. 3rd ed. New York: Wiley.
The study of linear DEs is of both theoretical and practical importance. In Chapter 2
we investigated the general theory concerning first-order linear equations and
presented various applications of these equations in Chapter 3. Here we wish to
build upon that theory and in some instances use it for motivational purposes for
developing the higher-order linear theory. Applications of higher-order linear DEs will
be taken up in subsequent chapters.
The notion of linear independence, which is so critical in the development of a
general solution, is introduced in Section 4.2 for second-order equations. The ques-
tion of linear independence is later reduced to evaluating a special determinant
called the Wronskian, the nonvanishing of which is established as both a necessary.
and sufficient condition for linear independence of a set of solutions of a homo-
geneous linear DE. In Section 4.3 a method is discussed for producing a second
linearly independent solution of a second-order equation given that one solution is
known. Although this technique is often considered to be mostly of theoretical
importance, it is used in practice from time to time.
The solution of second-order, homogeneous, constant-coefficient equations is dis-
cussed in Section 4.4. The significant feature here is that such equations can be
solved entirely by algebraic methods. In Section 4.5 we generalize the theory and
solution techniques to equations of order greater than 2.
We turn our attention to nonhomogeneous DEs in Sections 4.6 and 4.7. The
method of undetermined coefficients, which requires us to guess at the structural
form of the particular solution, is introduced first. Although this method is fairly easy
to apply, its application is restricted to a narrow class of DEs. A more general
method called variation of parameters is then introduced, which theoretically is
applicable to all linear equations.
In the last section we extend the solution techniques for constant-coefficient
equations to a special variable-coefficient equation called the Cauchy-Euler
equation.

82
 41 INTRODUCTION

Most nonlinear equations of order greater than | are very difficult to solve. This is
because there is no general theory concerning the solution of these equations, and
even the techniques discussed in Chapter 2 are not generally applicable. We will
therefore restrict our attention in the remainder of the text almost exclusively to
linear equations.
Linear DEs of order greater than | can be applied in numerous areas. Of these
equations, the most prominent are those of the second order, and so we will direct
most of our theory development and applications at them. For example, the forced
and free oscillations of a spring-mass system are governed by the second-order
linear DE

my" + cy’ + ky = F(), (1)

where m, c, and k are system parameters and F(t) is a prescribed function. A similar
equation arises in electric circuit problems where the circuit components may
include resistors, coils, and capacitors. The static displacements of a string or wire
supporting a distributed load are described by a particular second-order linear DE,
as are the steady-state temperatures in a rod. Problems involving cylindrical sym-
metry often lead to Bessel’s equation
xy + xy + (a — )y = 0, (2)
whereas Legendre’s equation
(1 — x*)y” — 2xy'’ + n(n + ly =0 (3)
frequently occurs in physical situations featuring spherical geometry.
In the chapters that follow, we will discuss all of the above phenomena in terms
of their mathematical models, and we will also discuss applications that sometimes
require a DE of order greater than 2. The purpose of the present chapter, however,
is to discuss appropriate solution techniques for some of these equations and to
develop the general underlying theory of linear DEs.
Recall from Chapter | that a linear DE of order n has the form
ARG)y A O)y ce tel (A) y et Ag(x)y: = FU) (4)
where the coefficients of y and its derivatives are functions of x alone. When
F(x) # 0, we refer to (4) as being nonhomogeneous, and when F(x) vanishes
identically on some interval, the equation reduces to the associated homogeneous
equation

Ay ase Ane)ee tae A(X) A(x)y.= 0. (5)

An important property of homogeneous linear DEs is contained in the following
theorem, sometimes called the linearity property or the superposition principle.

Theorem 4.1 If y = y,(x) is a solution of the homogeneous DE (5), then so is y = Cy,(x) for
any constant C;. More generally, if yi, y2, . - -» Ye are all solutions of (5) on some
specified interval, then
83
 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

y = Ciy(%) + Coyo(x) + ++ + + Cine)

is also a solution on this interval for arbitrary constants C,, C2, . . ., Cy.

Proof: For notational simplicity we will prove only the case when n = k = 2.
Let y, and y, be two solutions of

Ar(x)y" + Ai(x)y’ + Ao)y = 0.

The substitution of y = C,y,(x) + Cry2(x) into this DE gives

Aa(x) (Ciyt + Crya') + Arle) (Ciyt + Cayz) + Ao(x) (Cin. + Crya)

= C,[Ar(x)yi' + Ai(x)yl + Aod)yi] + C2[A2~dy2 + Ai)yz + Ao) yr]
CA eee a eee
Zero Zero

= Ci: 0 37C.20
= Ole)

In the development of the theory of linear DEs that follows, we will generally
present the theory for second-order equations first and later generalize the results
for equations of order n. Unfortunately, even for second-order linear DEs no
general solution techniques are available, unlike the case for first-order linear
equations. Only for certain narrow classes of equations can formulas for general
solutions be found.
In order to SENCLED S
some oo the theory for second-order DEs, it is helpful to put

(6)

(7)
A,(x)/A2(X), do(x) =
AsG)Aa(x), and FQ) = FOD/As(,

4.2 LINEAR DEPENDENCE AND INDEPENDENCE

dn Sa 1

(9)
 SEC. 4.2 / LINEAR DEPENDENCE AND INDEPENDENCE

EXAMPLE 1 Show oe yi = x and y)= 5x are linearly dependent on any interval while y, and
y3= x? are linearly independent on any interval.

Solution In the first case, we merely observe that

y2(X) = 5 for all x;

yi(x)
hence they are linearly dependent. Likewise,

y3(x)
=x # constant,
yi (x)
which implies that y, and y; are linearly independent.

EXAMPLE 2 Discuss the linear independence of y,; = x and y, = |x| over the intervals
Q=x <™~ and —~ < x < & (see Figure 4.1).

Vi 25 V2 = |x|

Figure 4.1

Solution Here we see that

yo(x) _ Ute OrSant ne

y(x) SUN SES) 255 oa Oh

Thus, since the ratio y2/y, remains constant over the interval 0 = x < %, these
functions are clearly linearly dependent on this interval. However, over the larger
interval —0%© < x < ©, the ratio y2/y, changes values (from —1 to +1), which
implies linear independence on this interval.

The significance of linear independence is apparent from the following theorem,

which we state without proof.

Theorem 4.2 | If y, and y, are linearly independent ons of the homogeneous linear equation ©
86 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

where C, andC) are any Furthermore, a set of linearly independent

solutions always exists.*

EXAMPLE 3 _ Bothy, = e* and y, = e ~ are solutions of

Wet yay
What is the general solution?

Solution Note that y, and y2 are linearly independent on any interval, since

ya(x) =
e ** # constant,
yi (x)
and hence the general solution is
y = Cie* + Coe”,
where C, and C, are arbitrary constants.

4.2.1 Wronskians

When more than two functions are involved, our definitions of linear dependence
and independence do not apply. Thus we seek a more general definition of these
terms which can be applied to situations involving more than two functions.
Observe that if constants C; and C, (not both zero) can be found such that

Ciyi(x) + Cry2(x) = 0 (10)

for all x in some interval J, then y, and y2 are linearly dependent on /. That is to
say, if (10) is true, then (C, # 0)

C
yi) C220):

which shows that y, and y2 are proportional and hence linearly dependent.
Now suppose we consider the expression

Ciyi(x) + Coy2(x) = 0 (11a)

and its derivative

Ciy1 (x) oF Cry (x) = 0. (11b)

*For a proof of Theorem 4.2, see E. L. Ince, Ordinary Differential Equations (New York: Dover,
1956).
 SEC. 4.2 / LINEAR DEPENDENCE AND INDEPENDENCE 87

If we think of C, and C, as the unknowns in (11a) and (11b), then using Cramer’s
rule we see that nonzero values for C, and C; are possible only when the coefficient
determinant of these two equations is zero, i.e., when

yitx) ya(x)
yi(x) y2(x) Vash//s-/
I STK CICC &.
(12)
This coefficient determinant is called the Wronskian , named after the Polish math-
ematician J. Wronski* and denoted by

yix) yo(x)
| W(y,
Yis 2 y)@)= ae
|", :
yi(x) = yilx)y,(x)y2(x)
y2 (x) — yi
yi(x)y2(x).
(x)a(x) 13
(13)
‘a on \Y\>

et
( Theorem 4.3 If y, and y possess first derivatives on some interval J, and
8
“——— (a) if W(y1, y2) (x) # 0 for at least one point in /, then y, and y> are linearly
independent.
(b) ify, and y, are linearly dependent on /, then W(y,, y2) (x) = 0 for all x in J.

EXAMPLE 4 Show that y, = cosx and y,= sinx are linearly independent.

Solution Computing the Wronskian of y, and y2, we find

W(cosx, sinx) = |ibe ae =1#0

=—Sin.x* \CcOSx
for every value of x. Hence cosx and sinx are linearly independent.

EXAMPLE 5 Show that y, = e”" and y. = e”™ are linearly independent if m, # mp.

Solution Here we find the Wronskian to be

Cae
Wem", em = a= (m, a2 me (tm + 0
me™* me™

for all x. Thus the functions are linearly independent.

We should be aware that the conditions listed in corte =oaare only ene
{Dhara That iiS to) Say,
not may Ca! yen when —

in
*JozeF M. H. WRONSKI (1778-1853) studied mathematics in Germany but lived most of his life
France. His sole lasting contribution to mathematics appears to be the Wronskian determinant.
88 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

EXAMPLE 6 Show that the functions y, = x? and y, = |x|’ are linearly independent on the
interval —%© < x < ©, but that the Wronskian of y, and yp is identically zero (see
Figure 4.2).

Solution By setting x = 1 and x = —1 in the expression

Gx G0)
we find
C,+ C, = 0,
=Cr + C, = 0;
from which we deduce C; = C, = 0. Hence the functions are linearly independent.
On the other hand, if x => 0, then

peo eS
W(y1, y2) (x) i Le 3x2 <a 0,

and if x < 0, then

3 ahs
W(y1, y>) (x) et ooo) Soe! — 0.

This shows that the converse of Theorem 4.3 is false.

Because our interest is confined to only those sets of functions that are solutions
of linear DEs, an additional restriction can be imposed on any given set of func-
tions, which, together with the nonvanishing of their Wronskian, imply linear
independence. The only requirement is that the functions involved be solutions of
a homogeneous linear DE in order that the nonvanishing of the Wronskian becomes
both necessary and sufficient for the functions to be linearly independent. The proof
of this result relies on the following important lemma.
 SEC. 4.2 / LINEAR DEPENDENCE AND INDEPENDENCE 89

Lemma 4.1 |

for an appropriate value of the constant C.

Proof: Taking the derivative of the Wronskian leads to

£
oy Worn)
Visey2) ie=iyLonrys
Yi¥2 —— Wiyi )ya) == Viy2
yh —— Yiyfyr,
and since ty = —a,(x)y’ — ao(x)y, we can rewrite this expression in the form

£ Won, yy) = yil-axyz — ao(x) 2] = yal —ay(x)y1 7. ao(x)y:]

—ay(x) (yiy2 — yry2)

—ay(x)W(y, 2).

Thus the Wronskian satisfies the first-order linear DE

w'+ acowe =n wy
with general solution (5 fuction af,Horo agn)

Wo, y2) (2) = Cexp i-[ao tx).oO

EXAMPLE 7 Find the Wronskian of the solutions of

By ey ary a0)
to within a multiplicative constant.

Solution We first rewrite the DE in normal form to get

y' +/-y_
” 1 '
+y = 0.
x
Using Abel’s formula, we then have

*NeILS H. ABEL (1802-1829) was one of six children born into a poor Norwegian family. One of his
early accomplishments was proving that the general fifth-degree algebraic equation has no radical
solution. Stricken with tuberculosis in 1827, he died two years later at the age of 26.
 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

dx C
W(y1, Ya) (x) = cexp(-[©) Sane

Theorem 4.4 If y, and y, are solutions of

y" + a(x)y’ + a(x)y = 0

on some interval J where a,(x) and ao(x) are continuous, then y, and y, are linearly
independent on / if and only if W(y,, y2) (x) # O for every x in /.

Proof: First, if W(y1, y2) (x) # 0 for every x in J, it follows from Theorem 4.3
that y, and y2 are linearly independent. Conversely, if y, and y2 are known to be
linearly independent, then from Lemma 4.1 their Wronskian is given by

W(y1, Y2) (%) = Cexp (-ia(x) ax),

which is clearly nonzero for C # 0, and the theorem is proved. L]

From Theorem 4.4 it is clear that when y, and y, are solutions of a second-order
homogeneous linear DE on some interval J, either their Wronskian is identically
zero or is never zero on J. Moreover, if the Wronskian is identically zero on J, it
now follows that the functions y, and y, are linearly dependent.

EXERCISES 4.2

In problems 1-5 , determine whether the given functions are linearly dependent or
independent .
\fi) Vumale ype 2. yp =x", yo =x"!

® yi = logx, y2 = logx? 4. y,=sin2x, y2 = sinxcosx

y= ex as e°, ie e* ome

6. Show that the functions y, = x* and y2 = x|x| are linearly dependent on any interval
for which either x > 0 or x < 0, but are linearly independent on —2%2 < x < ~,
&@) Verify that y; = sin3x and y2 = cos 3x are solutions of
y eye= 0
and show that their Wronskian satisfies Abel’s formula (Lemma 4.1).

In problems 8-11 , use Abel’s formula (Lemma 4.1) to determine the Wronskian to
within a multiplicative constant of the solutions of the given DEs.

8. y"—4y'
4y+=0 @) y"-3y' +2y =0
10... Cl = x*)y" = 2xy" + nin Dy 0 41) XH" + xy’ + (x? — ny =0
 SEC. 4.3 / CONSTRUCTING A SECOND SOLUTION FROM A KNOWN SOLUTION
91

“12. Prove Theorem 4.3.

\“"13. Show that if y, and y2 are linearly independent functions, then
Pe a vee BEd y= Vata 74
ae are linearly independent.
A44. Prove that if y, and y2 vanish at the same point in the intervala <x <b, they cannot
form a set of linearly independent solutions of a second-order equation.
Given the DE
y" — 6y' + 9y = 0,
verify that y; = e™ and y2 = xe** are linearly independent solutions and write the
general solution.
16. Given the DE
xy)
Pea)
ay = 4ye—=(,
verify that y; = x * and y2 = x are linearly independent solutions and write the
general solution.

4.3. CONSTRUCTING A SECOND SOLUTION FROM A KNOWN SOLUTION

: . eT ee _— Ce ae
It is a curious fact that given one nontrivial sol f a d-order me
linear DE,
a second linearly inc

Theorem 4.5 | If y = y,(x) is a nontrivial solution of the homogeneous second-order DE

y’ + ay’ + aay = 0,
then

exp (—[asada]
yo = yi) av err

is a second linearly independent solution.

Proof: If y, and y, are linearly independent solutions of the DE, then

yiX)y2 — yi@)y2 = Wy, y2) >)

where the right-hand side is nonzero. We wish to interpret this equation as a
nonhomogeneous first-order linear DE in y). Thus, after dividing the equation by
y,(x) to put it in normal form, we recall that a solution is given by
W(y1, Y2) @) dx.
pe yi) |
yi(x)
92 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

From Lemma 4.1 it now follows that

exp (—[au ax)

y= (| oes

Regardless of the value of C, this last expression is a solution of our second-order

DE in the theorem, and so by letting C = 1 we obtain the desired result. LA

EXAMPLE 8 The function y, = e” is a solution of the DE

y” — 4y’ + 4y = 0.
Find a second linearly independent solution.

Solution Using the result of Theorem 4.5, we obtain

ae ipa, 4 Ria
f dx) S ee1)Sa
re
© (y-e™
=O, e%4Ca pom (Ca Cee ao*
EXAMPLE 9 _ Given that y, = x“! is a solution of
Boy A Bx i ya Oe
find a general solution valid for x > 0.

Solution Here it is necessary to put the DE in normal form first:

y
tee) res.
sey ae 20 = 0.

Thus a;(x) = 3/x, and a second linearly independent solution is given by

fr

Therefore, we can write

yx. (Ci +.Glogx emer 0

as a general solution.
sa eee eee
 SEC. 4.3 / CONSTRUCTING A SECOND SOLUTION FROM A KNOWN SOLUTION
93

The utility of Theorem 4.5 is clearly limited to those situations wherein one
solution of the DE is known. In most situations it is just as difficult to produce one
solution of a DE as it is to produce both, and so the theorem is of little help.
However, occasionally one solution of a DE is obtained by “inspection,” a series
method, or some other means, and the theorem can then be very useful in the
construction of a general solution.

EXERCISES 4.3

In problems 1-15, use Theorem 4.5 to construct a second linearly independent

solution of the given DE.

y" + 2y'=0; yw =1 N ( 2.) Vout 2y) = 3y = 0:9 y) = xe"

A y"(—-(@' + 9 =0; y, =e” AL oy" =4y' + 4y =0; y, = xe?
~ oe ye ® y' +y =0; yi = sinx 6. y* =y =0; y= sinhx
AS : pay 2y a Sy = Or yi = e* cos2x 8. Vxry ie xy’ —"6y = OF yp =
ate xy Oy = 0: y= x° 100 x2 = 20y = 05 y=
. 4x*y"+y=0; y, =x" logx
12) x7y" xy 2y = 0: yy = xsin (logx)
pI) - xy"
— 2y' = 0; ys = 1
14 ye xy ty 02 y=

"IB. x*y" + xy’ + (x — )y =0; y =x '*sinx

16. Verify that y,; = e* is a solution of

xy = ny tony =0;
where n is a nonnegative integer.
(a) Find a second linearly independent solution for the case when n = 1.
(b) Repeat (a) for the case when n = 2.
(c) orn a positive integer, verify the interesting result
4 x? isx”

ig wae 8ne ap Sa Pie

which is simply the first n + 1 terms of the Maclaurin series for e*.
“17. By assuming y2 = u(x)yi(x) is a second solution of
y" + alx)y’ + ao(x)y = 0,
given that y, is a known solution,
(a) show that the function u satisfies
(x
te ae |)(C8) oF 2H | =0
yi(x)
(b) Let v = w’ and solve the resulting first-order DE in v to obtain the result
 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

v(x) = age
Ae |-
-{ ao ar|.

(c) From (b), obtain an expression for u(x) and verify that y2 = u(x)yi(@) is the same
solution as given in Theorem 4.5.

4.4 HOMOGENEOUS SECOND-ORDER EQUATIONS WITH CONSTANT

COEFFICIENTS

Differential equations with constant coefficients are the easiest class of linear
equations to solve by any general technique. This particular fact coupled with the
fact that they occur in such a wide variety of physical applications accounts for the
special place that these equations occupy in the general theory of linear equations.
We have previously found that
thefirst-orconstant-coefficient
der linear DE
y! + ay = 0)
has the exponential solution y = Cje~“ valid on the interval —-* <x < &. Be-
cause of the special property associated with derivatives of the exponential func-
tion, it may seem natural to inquire as to whether higher-order linear DEs with
constant coefficients also exhibit exponential solutions.
For example, suppose we consider the second-order equation
ay" + by' + cy = 0, (14)

where a, b, and c are constants. Let us assume that (14) has an exponential solution
of the form

ier ere
for some value or values of the parameter m. Direct substitution of this function into
the left-hand side of (14) leads to

Gyn > by = ey s=" ane" 5 bme pices

= (am? + bm + c)e™
which can vanish if and only if (since e’“ # 0)
am? + bm +c =0. (15) J
This quadratic equation in m is called the auxiliary equation of (14) and has the
roots
—b + Vb*
— 4ac —b — Vb?
— 4ac
sd erry Sime GL Cs (om
Tu (16)
a
Clearly now, there are three separate cases to consider: whether m, and m, are real
and distinct roots (b* — 4ac > 0), real but equal roots (b?= 4ac), or complex
conjugate roots (b? — 4ac < 0).
 SEC. 4.4 / HOMOGENEOUS SECOND-ORDER EQUATIONS WITH CONSTANT COEFFICIENTS 95

We have already shown these functions to be uly) scent in Example 5,

and thus it follows that a genera Y
(17)

EXAMPLE 10 Find a general solution of 2y" + 5y’ — 3y = 0.

Solution The auxiliary equation is 2m? + 5m — 3 = Oor

(2m — 1)(m + 3) =
Thus the roots are m, = } and m, = —3, giving us the general solution
a Cyere aia Ges

Find a general solution of y” — 4y’ + Ay = 0.

Solution This time the auxiliary equation is m? — 4m + 4 = 0 with the double root m = 2.
Hence,

yee (Cet Coxe = tie + (yer

 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

In this case the general solution is”

y = CyePriax + Cre?

= eP(Cye™ + Cre), (19)

For physical problems, real solutions are generally preferred to (19). In order to
accomplish this, we invoke the famous Euler formulas
e® = cos@ + isind, e =cosé — isin#, (20)
to rewrite (19) as
y =e([C,(cosgx + isingx) + C,(cosgx — isingx)]
= e[(C, + C,)cosgx + i(C, — C2)singx]
Lf = e” (Cscosqx + C,singx),
*

where C; and C, are any constants. We usually relabel the constants as C; and C,
once again and write
y =e (C,cosqxoog + Cysingx) (21)
as a general solution. That y; = e” cos qx and y, = e”* sin qx are linearly indepen-_
dent solutions of (14) will be verified in the exercises.

EXAMPLE 12 Find a general solution of y” — 4y’ + 13y = 0.

Solution The auxiliary equation is m? — 4m + 13 = 0, with complex roots m,; = 2 + 3i

and m, = 2 — 3i. Hence we write the general solution as
= e*(C, cos 3x + C,sin 3x).

EXAMPLE 13 Solve the initial value problem

OA Say BIO eh a Gyice Fb

Solution The auxiliary equation m? — 2m + 1 = Ohas the double root m = 1. The general
solution is therefore
y = (C, + CyxDe’.
By differentiating, we obtain

y' = (C, + Coxe* + Cre’.

From these expressions for y and y’ and from the initial conditions, it follows that
yO =C,=3, yOSeq+rQ=i.
Hence, C, = 3 and C, = —2, leading to the desired solution

y=i(3i— 2x)e™
 SEC. 4.4 / HOMOGENEOUS SECOND-ORDER EQUATIONS WITH CONSTANT COEFFICIENTS
97
'

oe SS eg ee a
EXAMPLE 14 Find general solutions of
y"+k’y =0 and y"—k’y =0.

Solution The first DE has the auxiliary equation m? + k? = 0 with pure imaginary roots
m, = ik and m, = —ik. Thus the general solution is

y = C,coskx + Cysinkx.
In the second case the auxiliary equation is m? — k? = 0, with m, = k and
m2 = —k the roots. Its general solution is therefore
Ye Cer air Gea.

However, there is another representation of the general solution that is more

convenient in particular applications. Moreover, it more closely resembles the
solution of the first DE. Since the hyperbolic functions are defined by the linear
combinations
1
cosh kx = iC +e"). “sinh kx’ = x(e" —e*),

and these are linearly independent functions (see problem 5, Exercises 4.2), it
follows that ;
y = C3coshkx + C,sinhkx

is also a general solution of y” — k*y = 0 (see Example 7 in Section 1.3.1).

EXAMPLE 15 _ Find a general solution of y” + y’ — y = 0.

Solution Here we get the auxiliary equation m? + m — 1 = 0, which cannot be factored.

Thus we use the quadratic formula, finding that

m, = 5(-1 + V5) and =m) = s(-1 — V5).

Since these roots are distinct and real, the general solution is

Vo= Cyexp Le + V3)s| + Cyexp (fet Vays]

However, using a device similar to that used in Case III for obtaining a general
solution in terms of trigonometric functions, we can write the general solution of
this problem in the alternate form
1
ines eat? E cosh (55) + Cysinh (5v5:)].

the verification of which is left to the reader. (See problem 20 in this section.)
98 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

4.4.1. Differential Operators

ae ee examples of such transformations are integration and differ-

entiation. Since transformations of one type or another are so common in many
areas of application, operator methods are playing an ever increasing role in
mathematics.

< In Chapter 2 we introduced the notion of an operator for the first time, the
simplest one of which is D, where Dy = y’. We now consider the second-order
differential operator
P(D) = aD? + bD +c (22)
associated with the constant-coefficient equation ay” + by’ + cy = 0. The inter-
esting feature of the operator P(D) is that it can be manipulated according to the
basic rules of algebra applied to any polynomial. To see why, let us apply the
operator (22) to the exponential function y = e”™, which leads to the relation
P(D)[e™] = (am? + bm + c)e™ = P(m)e™
Hence P(D) and P(m) have the same polynomial form and therefore factor exactly
the same.
The above remarks suggest that the constant-coefficient DE
(2D? —"D —3)y =0
can also be written in either of the equivalent forms
oeet ied — 3)y =0
or a Pk c= 51)

QDs 3)(D =-4) yy,=.0:

Thus simple inspection of the AUTRES operator reveals that the roots of the auxil-
lary equation are m, = —1 and m, = 3. A linear differential operator of any order
with constant coefficients can also be Galtored’ in the same manner. Factorization
of operators with variable coefficients, however, does not satisfy the basic laws of
algebra, so one must exercise more caution when dealing with operators of this
type.
ee ee

EXAMPLE 16 _ Verify that

(D + 1)(2D — 3)y = (2D — 3)(D + 1)y

whereas

(xD + 1)(D — 2)y\# (D — 2)(xD + 1)y.

 SEC. 4.4 / HOMOGENEOUS SECOND-ORDER EQUATIONS WITH CONSTANT COEFFICIENTS 99

Solution In the first case, we have

ID GB — 3)y-\(D + I2y* = 3y)

= (2D? — D — 3)y
and

CO) Del)y De 3 yay)

fy Oye aay
= (2D? — D — 3)y,
which shows the equivalence of the two expressions. On the other hand,

GD + 1)(D — 2)y = GD # 1)(y’ — 2y)

PA (le 2X)y oy
= [xD? + (1 — 2x)D — 2]y
and

(D — 2)@D + l)y =D — 2)xy' + y)

Ayia aa) yen oy
= [xD? + 211 — »D — 2p,
which are clearly not the same. The distinction, of course, is that in the second case
the operators have variable coefficients.

EXERCISES 4.4

For problems 1-6 , show that the auxiliary equation has distinct real roots and find
the general solution.
@ 3y"—y' =0 2). y" — 4y =0

@y y” + 2y' — 3y =0 4. 2y" — S5y’ — 3y =0

\53 y” — 10y' + 17y = 0 6. 3y” — 10y’ + 4y =0

For problems 7-12 , show that the auxiliary equation has a repeated root and find
the general solution.

@ yr oye ty = 0 Gay = 10y 2 25y = 0

©) 9" +o’ +y=0 WWE 2g eae

41) Oy” — 12y’ + 4y =0 12. y"—2V2y’
+ 2 =0
 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

For problems 13-18, show that the auxiliary equation has complex roots and find
the general solution.

43) ory yee G 14. =0

y” + 25y
“45. y” — 6y' + 25y =0 16. ey 4y E13 y= 0

y= Vie ye 18. 2y” — 3y’ + 10y =0

Given the two functions
y, =e” cosqx and y2 = e”*singx,
(a) show that they are linearly independent.
(b) Verify that y = Ciy,;(x) + Cry2(x) satisfies the DE
(Dom pyas ‘gsly.=.0.
20. Given the two functions
y, = e”*coshqx and y2 = e”*sinhqx,
(a) show that they are linearly independent.
(b) Verify that y = Ciyi(x) + Cry2(x) satisfies the DE
[Oppo Wie 20.
*21. Show that the general solution of y” + y = 0 can be expressed in the form
y = Acos(x — @)
or
y = Bsin(x — 8),
where A, ¢, B, and @ are arbitrary constants.
*22. Show that the solution of the initial value problem
Vor Oy. tel jye=O) ay ( Omi wey aC) ieee
can be expressed in the form
y = Vi3e*sin(4x — 6),
where the angle @ is defined by the equations

5)
2 8
cos
9 = —= and sin@d
= ——.
V 13
Hint: See problem 21.

In/problems 23-30, perform the indicated multiplication.

3) (D - 54D +3) 24. 02D +72D-7 G5) (DW -1)~D2+D +1)

26. 0+2nD-1?7 OF O-nD+n 2. O+HNO-»
(xD + 3)(xD — 2) 30. (xD — 2)(xD + 3)

In problems 31-35, find the general solution. Assume that x is the independent
variable.

31.) ( — 5\(4D + 3)y =0 32. (D? — 8D + 16)y =0

 SEC. 4.5 / HIGHER-ORDER LINEAR EQUATIONS 101

GD Gp? - 14D — 5)y =0 34. (4D? - 12D + 5)y =0

*35. (D + 1)(5D — 3)(D + 4)y =0
In problems 36-40, solve the DE subject to the prescribed conditions.

G6) y"-4y'
+13y =0, yO=-1, y'@=2
(37) y"- 2y'—3y=0, y)=0, y'@=-
(38) y"+ ky =0, yO =yo, y'O=
*39,, y"— 3y'+ 2y=0, y(1I)=0, yQ)=
\40. (0?7-D-6y
=0, yO=0, yo) =e?

4.5 HIGHER-ORDER LINEAR EQUATIONS

Although second-order linear DEs are far more prevalent in practice than equations
of higher order, there are applications wherein the mathematical model demands a
DE of order greater than 2. For example, in studying the small deflections of a beam
supporting a distributed load, we find that the governing DE is fourth order, as is
the DE describing the buckling modes of a long, slender column under an axial
compressive force.
In this section we wish to extend the theory of second-order linear equations
developed in previous sections to higher-order linear equations. Much of the theory
is simply a natural generalization of that developed for second-order DEs, so we
will mostly state appropriate theorems without providing their proofs. In discussing
the results for these higher-order equations, we find it convenient once again to put
the DE in normal form, but this time we will also introduce the concept of a
differential operator. Assuming A,,(x) does not vanish anywhere on the interval of
interest J, let us divide the equation

AMY et Ay Fi * TAY F Any = FX) (23)

by A,,(x) to get the normal form —
© + a, soy" - (24)
102 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

4.5.1 Linear Independence

Extending our notions of linear dependence and independence from Section 4.2.1,
we get the following definition.

Definition 4.1

of en can be expressed as a linear combination of the others. For example,

suppose (28) is true where at least one of the C’s, say C,, is different from zero.
This being the case, we can solve for y,, getting the result

ioe ys (29)

EXAMPLE 17 Show that y; = 3x* — 8x, y. = x”, and y; = 4x are linearly dependent on any
interval.

Solution We see that y, = 3y, — 2y; for all x, and thus it follows that the functions are
linearly dependent.

Theorem 4.6 | If M is a normal differential operator of order n on an interval J, and if y,, y»,.
-» Yn Constitute a set of linearly independent solutions of MLy] = 0 on J, then
the general solution of this homogeneous equation on J is

y= CyyiG) a Cayce ~ 2 Gi yG).

where C;, C,, . . ., C, are any constants. Moreover, a set of linearly independe
nt
solutions always exists.

Remark. The lepende eran vie:++ Yn ISoften re-

ferred
to as a Pp pepersisset ofsolutions of ML y] =
 SEC. 4.5 / HIGHER-ORDER LINEAR EQUATIONS
103

Next, generalizing the notion of a Wronskian to a set of n functions, we have

yi(x) y2(x) +++ Yn(x)

yi) —-y3(x) s+ Yn (X)

Vie(x)ayy)ox) eee amy oer)

Theorem 4.7 | If the set of functions yi, Y2, - - » » Yn possesses at least n — 1 derivatives on some
interval J, and

EXAMPLE 18 Discuss the linear dependence of y; = 3x? — 8x, y. = x, and y; = 4x, using the
Wronskian.

Solution The Wronskian gives us

WO, y25 y3)(x) = 6x — 8 2x 4 = (0).

0
Nonetheless, ¥

2 = i I ( ident.
Example 17, however, we did establish that these functions are indeed linearly
dependent. )

Abel’s formula (Lemma 4.1) can also be extended to nth-order equations, which
in turn can be used to prove Theorem 4.8 below.

Lemma 4.2 | (Abel’s formula) If y,, y2, . . . , y, are linearly independent solutions of M[y] = 0
on the interval J where M is the normal linear operator
; ra D'+ ano tee e + a,(x)D as a(x),

the associated Wronskian is given by

Wy,-- > Yn)(X) = Cexp E|ay) ds

for an appropriate value of the constant C.
104 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

Theorem 4.8 . » +Yn are solutions of M[y]= 0 on the interval J where M is normal
a
TE TTOGERATTE of order n, then the set of solutions is linearly independent
on J if
and only if.
Wor.) y)Q) + 0
| for
everyx in J.

4.5.2 Homogeneous Constant-Coefficient Equations

If we wish to solve the linear DE of order n,

G,Dit sD chs voteaD ray i—a0. (31)

where the coefficients do, a;, . . ., a, are all constants, it is necessary to find the
roots of the nth-degree po lynonial
a,m" + a,\m" '+-++ + am + a =0, (32)

which is the auxiliary equation associated with (31). We obtain (32) by putting
y = e”™ into (31) and simplifying. Because of the many possible combinations of
roots of (32) depending upon 2, it is difficult to identify all cases when n > 2.
However, the following generalizations are fairly easy to establish:

1. If m, is a real distinct root of the auxiliary equation, there corresponds the

single solution
yy — (ues

2. If m, is a real root of multiplicity k, there corresponds k linearly independent

solutions given by
yy, = er ty, = xe", ey — ee
3. If p + ig are distinct complex roots of the auxiliary equation, there corre-
spond the linearly independent solutions
yi = e”*cosqx, y2 = e” singx.
4. Ifp + ig are complex roots of multiplicity k, there correspond the 2k linearly
independent solutions
y= "e”" COs x; yz = xe COS GX, «say ye = xe! COS
Yer = eP* sin gx, Yaro = xe" sin gx, . . ., yx = x*-'e”* sin gx.
We leave it to the reader to verify that in each case the given functions are indeed
solutions of the DE and, moreover, linearly independent solutions.

eee
se ea eee

EXAMPLE 19 Find the general solution of the sixth-order DE

(D =1)7 O27 6D" = 2) y= 0:

 SEC. 4.5 / HIGHER-ORDER LINEAR EQUATIONS
105

Solution The auxiliary equation is obviously given by

(m — 1)°(m + 2)°(3m — 2) = 0,
with roots m = 1, 1, 1, —2, —2, 3, obtained by inspection. Hence the six solutions
corresponding to these values of m are
MigesGes aye = Ae sys =e, =e ys = xe“. yem en
providing us with the general solution
y = (Ci + Cox + Cax*)e* + (Cy + Cexle~™ + Cre?*?,
acre e eee 1 i eg th ee ae
EXAMPLE 20 Find the general solution of the eighth-order DE

DD? — 2D + 5)*y =0.

Solution Here we find the auxiliary equation to be

— 2m + 5)? =0
m*(m?
with roots m = 0, 0, 0, 0, 1 + 2i, 1 + 2i, leading to the solutions

Viegas Aesth = Xi, Ysa e1C0S 2x,

Ys = xe*COS2x, y, = e*sin2x, ys, = xe* sin 2x.
Hence,

De Cy a0 1 Cox. 4 Cax e?[ (Cs Cox) cos 2x24 (C, + Cex) sin 2x].

When the DE does not have the operator in factored form (which is the normal
situation in practice) as in Examples 19 and 20, finding the roots of the auxiliary
equation can be the most difficult part of solving the DE. According to the general
theory of polynomials, if (32) has a rational real root of the form m, = p/q, where
p and q are integers, then p must be a factor of dp and q a factor of a,. Of course,
once we have found one root m,, we can divide (32) (either directly or by using
synthetic division) by the factor (m — mj) to obtain a polynomial of one less degree
from which to determine the remaining roots.

EXAMPLE 21 Solve (4D? — 3D + l)y =0. + |

Solution The auxiliary equation is
4m?
— 3m + 1=0.
If there are any real rational roots, they are among the possibilities m = +1, + 5,
+4. Checking, we find that m, = —1 is such a root. By division, we then obtain
4m3 — 3m + 1 =(m + 1)(4m? — 4m + 1) = 0,
106 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

which shows that the remaining roots are m, = m3 = 5. Thus, the general solution
is :

Ne oe Cyrene oP (C, ar C3x)e*.

EXAMPLE 22 Solve (4D‘* — 15D? + 5D + 6)y = 0.

Solution For the auxiliary equation

4m* — 15m?
+ 5m + 6 = 0,
the rational root possibilities are
mice 1), 222 te beth aed ees eee
By checking, we find that m,; = 1 is a root, so we can write

4m* — 15m? + 5m + 6 = (m — 1)(4m? + 4m? — 11m — 6).

The remaining roots must then satisfy the reduced equation
4m?’ + 4m? — 11m — 6 = 0.
This time we obtain m, = 3 and write
4m* — 15m? + 5m + 6 = (m — 1)(m — 3)(4m? + 10m + 4) = 0,
and so the remaining two roots are readily found to be m; = —} and m, = —2. The
general solution is therefore given by
Eye Grier ate Gere + C3e eee Cy4e aoe,

EXERCISES 4.5

In problems I—5, determine whether the given functions are linearly dependent or
independent.

Vro=ok, Yo = 3x25, ys =x? = 9x

Yr=x%, yo= 3x", ys =e — Tx + 1
Yi =e", yo=xe*, ys = x7e*
Vie) yr =6: ueVar rcosho:
yi =cos’x, y2=sec?x, ys =sin?x, yg = tan2x
Given the DE
yl = yk Sy Pal 2yaee
verify that yi =e *, y2 = e™, and y3 = e* are linearly independent solutions and
write the general solution.
*7. Prove Lemma 4.2 for the case n = 3.
*8. Prove Theorem 4.8.
 SEC. 4.5 / HIGHER-ORDER LINEAR EQUATIONS 107

In problems 9-29, find the general solution. Assume that x is the independent
variable.

(D — 1)*y =0 10. D(2D + 3)y =0

(D — 4)(D + 2)*y =0 12. (D? + 1)?y =0
(D* + 3D? + 3D + 1)y =0 14. (D*- l)y =0
(D?+ l)y =0 16. D(D’? + 3D — 4)y =0
17. D*(D? + 3D + 1l)y =0 18D 9 (Dp 11)7y4= 0
19) (4D? + 4D? + D)y =0 20. (D*+'D? —2)y =0
21. (3D? — 19D? + 36D — 20)y =0
22: (D* — 5D? + 6D? + 4D — 8)y =0
(4D* — 4D? — 23D? +. 12D + 36)y =0
24. (D*—4D? + D? + 6D)y = 0
25)
P=—\

(4D* — 8D? — 7D? + 11D + 6)y =0

26. (D*° — D*)y = 0
a D*(D? + 9)*(D? — 2D + 1)*°y =0
*28. D* — 4D? + 10D? — 20D + 25)y =0
29.) (D + 2D? + D)y =0
In problems 30-34, solve the DE subject to the prescribed conditions.

(D? —- 3D —2)y =0, yO)=0, y’0)=9, y"0)=0

Vie ey yea) lea ey(2) 05 thin y"—10)
y” + 5y” + 17y’ + 13y = 0, y(O)=0, y’O)=1, y"(0) =6
y®=0, yO)=2, y'0)=3, y"0)=4, yO) =5
(D* + 6D? + 9D”)y = 0, y(0)=0, y'(0)=0, y"(0)=6, limy’=1
The roots of the auxiliary equation of some tenth-order DE are known to be 3, 3, 3,
3, 1 + 2i, 1 + 2i, 5, and —1. Write the general solution of the DE.
The roots of the auxiliary equation of a third-order DE are known to be m, = —4 and
m2 = m3 = 4. What is the corresponding DE?
Repeat problem 36 when m, = 3, m2 = 2 + i, and m3 = 2 — 1.
Given that y; = sinx is one solution of
yO 2p" by” + 2y" +5y = "0,
find the general solution.
*39.) Given that y; = e *cos 2x is one solution of
(D* + 4D? + 14D? + 20D + 25)y =0,
find the general solution.
40. Find four linearly independent solutions of
y® — ay =0
108 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

for the case when

(a) A=0.
(bn A = .0.
Hint: Set X = k*.
“41. Solve problem 40 subject to the boundary conditions
yO) =.0,- y(0) = 09 y= iO a:

4.6 THE NONHOMOGENEOUS EQUATION

In Section 2.5 we stated that every solution of the first-order linear DE

y' + aox)y = f(x)

is of the form y = yp + yy, where yp is any particular solution of the non-
homogeneous equation and y;,* is a general solution of the associated homogeneous
equation
y’ + a(x)y = 0.

This situation can be generalized to linear equations of any order n as the next
theorem states.

Theorem 4.9 If yp is any particular solution of the nonhomogeneous equation

yi + a, Gy?) +--+ + aay’ + ally = FQ),

and if yy = Ciy\(x) + - - + + C,y,(x) is a general solution of the associated homo-
geneous equation resulting from the deletion of f(x), then y = yp + yy is a general
solution of the nonhomogeneous equation.

Proof: For simplicity, we will present the proof only for the case when n = 2.
Suppose that yp is a particular solution of the second-order equation

y" + ay’ + a(~y =f,

and let Y be any solution of the same equation. The quantity y = Y — yp then
satisfies

y" + ax)y' + a(x)y = ¥Y" + a(xX)¥' + ag(x)¥ — yp — a(x)yp — ao(x)yp

=f) ea)ly ac.
Thus, since Y — yp satisfies the associated homogeneous DE, it must be express-
ible in the form

*The function yx is frequently called the complementary solution and is also denoted by
the symbol
yc in many texts.
SEC. 4.6 / THE NONHOMOGENEOUS EQUATION 109

Y= yp = Cy) 'Coy2@):
Transposing, we obtain

Y = yp + Ciyi(x) + Coya(x) = yp + yu,

which shows that every solution of the nonhomogeneous equation is contained in
the sum yp + yy. [_]

4.6.1 The Method of Undetermined Coefficients

To illustrate the gist of the method, suppose we wish to find a particular solution
of
y" + y = 3e*. (33)

Since differentiation of an exponential function merely reproduces the function

again with at most a multiplicative constant, it seems natural to “guess” that a
particular solution exists of the form
Pia NOX

where A is an “undetermined coefficient.” We substitute yp into (33), getting

Ae* + Ae* = 3e* or 2Ae* = 3e",

which reduces to an identity if A = 3. Thus

is a particular solution of (33).

Now suppose the nonhomogeneous term is changed to 3x? so that the equation
we wish to solve is
y" + y = 3x’. (34)

Proceeding as before, we might guess

110 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

This time the substitution of yp into (33) yields

QA + Axe = 3x7,
which cannot be satisfied for any choice of the constant A. The problem is that the
derivatives of Ax* produce new functions that are linearly independent of it. To get
an idea of what to do, we observe that (34) can be transformed into a homogeneous
DE by taking three derivatives of each side. That is, (34) becomes

yO + "= 0 (35)
since d°/dx?(3x*) = 0. Now the auxiliary equation associated with (35) is
m> + m? = 0 with roots m = 0, 0, 0, +i, so that its general solution can be
expressed as
y =C, + Cox + Cax? + a
Qe ee
Cycosx + Cssinx. (36)
yp Yu
It can be argued that every solution of (34) is also a solution of (35), and since

yy = Cycosx + Cs sinx
is the homogeneous solution of (34), it follows that the particular solution of (34)
gives
yp = Ax? + Bx +C

for some choice of the constants A, B, and C. The substitution of this yp into (34)
gives

DAS TrAL@
IBY + 1Ce—sKs
or
(QAs-2 OC) Bx Ax base
By equating like coefficients in this last identity, we have

2A +C=0

which gives the simultaneous solution

A=3, B=0, C=-6.
Thus a particular solution this time is

yp = 3x? — 6,
The genera] rule illustrated here is that yp should have the basic structure of the
nonhomogeneous term f(x), plus all linearly independent derivatives of f.
Another difficulty in the method arises when f(x) is composed of a function
that
occurs in the homogeneous solution. For example, suppose we wish to solve
SEC. 4.6 /THE NONHOMOGENEOUS EQUATION 111

y" + y = 3cosx. (37)

Assuming
yp = Acosx + Bsinx,

we find that substituting this expression into (37) leads to

—Acosx — Bsinx + Acosx + Bsinx = 0 = 3cosx,

which is absurd. To correct for this situation, we must assume yp to be a function

linearly independent of any functions in the homogeneous solution. Therefore we
write (since yy = C,cosx + C,sinx)

yp = x(Acosx + Bsinx),

from which it follows that

yp = Acosx + Bsinx + x(—Asinx + Bcosx),
yp = —2Asinx + 2Bcosx — x(Acosx + Bsinx),
and, when these expressions for yp, yp and yp are substituted into the DE, leads to

—2A sinx + 2Bcosx = 3cosx.

Equating like coefficients, we have
—2A =0, 2B =3,

Yp Xx .

The method of solution is summarized in Table 4.1. Using the table consists of
the following four steps.

Step 1: If the DE is of the form (not necessarily normal form)

My] = fi@) + fh) +---> +f),

where each f(x), i = 1,2, . . ., r, is a different type of function occurring in one
of the six categories in Table 4.1, then we replace the DE with the equivalent
system of equations
MLy] aa fi (x);

My] = fix),

My] = f-().

Step 2: Identify each function 7,(x)nti— 1, 25... 57) as belonging to one of

the major categories in the table.
112 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

Step 3: For each f;(x), assume the form of yp given in conditions (a) or (b),
depending upon whether the specified value of m is a root of the auxiliary equation.

Step 4: Add all yp’s found in Step 3 together to form the proper yp for the
original DE.
Table 4.1 Method of Undetermined Coefficients

DEE 4any OAs Opa oat yeaa ye=)

j Auxiliary equation: a,m" + a,-ym" | ++++ +am +

f@) = oxi +--+ +dx+b, j=

a. m#0 = 0, k times

yp = Ayx’ ++ +> + Ay + Ao yr = x*(Ayx! +--+ + Ax + Ao)

FQ) = be™
b. m =c,k times

ye = Ax*e™

Mi fG)= Oa = bx bole. | = 0) 1a eee

a m#c b. m=c,k times

= (Ajx’ ++ + + Aix + Anve® ye = x*(Ajxi ++ + + Arle™

IV. f(x) = acosqx + bsingx

m # +iq = +iq, k times

yp = Acosqx + Bsinqgx ye = x“(Acos qx + Bsingx)

V. f@) = (bx' +--+ + dix + bo)cosgx + (xi +--+ + Co) Sin gx

a. m # tig b. m = ig, k times

pos A Aas = teAG) Cos gx

+ (B,x’ + +--+ + By) singx, “Multiply yp in (a) by x‘.

f= max (27)

f(x) = ae” cos qx + be” sin qx

a. mF#p
= ig b. m =p + ig, k times

p = Ae”* cos gx + Be” singx Multiply yp in (a) by x*.

ee ee eee

EXAMPLE 23 Determine the form of yp fory"+ 2y’ + y = 3x2e>.

 SEC. 4.6 / THE NONHOMOGENEOUS EQUATION 113

Solution The roots of the auxiliary equation are m = —1, —1. In order to determine yp, we
first note that if m = —1 were nota root of the auxiliary equation, we would select
yp = (Ax? + Bx + Che™
as suggested in Category III in Table 4.1. But since m = —1 is a double root of
the auxiliary equation, the proper form to assume for yp is
yp = x*(Ax? + Bx + C)e™ = (Ax* + Bx? + C*x%)e™*

EXAMPLE 24 Solve D*(D — 1)y = 2sinx — Se’. /

)

Solution By inspection, we see that the auxiliary equation has roots m = 0, 0, 1. The
homogeneous solution is therefore
a ae ge Cote Gg. Cae.

Following Step 1 as outlined above, we reexpress the DE as the system of

equations
D*D — 1)y = 2sinx,
D*(D.—1)y = —S5e*,
The term 2 sinx in the first equation occurs in Category IV of Table 4.1, and since
m # +1, we assume

yp = Acosx + Bsinx.
Corresponding to the term —Se*, however, we need to write

Vian (Ce),
since m = 1 is a root of the auxiliary equation (see Category II). Thus, we write
the complete particular solution as
yp = Acosx 2 Bsinx + Cxe*.

Computing derivatives, we have

Ve = —Asinx. + Bicosx = C(xe> a €*),
Vie A COSY. = Bisinx 1, C(xe a 2°),
Vp Asing = -Bcosxe C (xem je),

and substituting these expressions into the DE yields (after simplification)

(Agar p Vsittxes (Au BD) COS uC e sa SIN Xe Oe

Comparing like coefficients, we get,

ee pet Ant bo= C= 5.

Therefore, A = B = 1, C = —5S, and our solution is

114 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

Vo yest Ya
=cosx + sinx — Sxe* + C, + Cox + Ce".

EXAMPLE 25 Solve y” + y = (x — 1)cosx.

Solution The roots of the auxiliary equation are m = +i, and thus

Yo = C,cosx + Cysinx.

Since m = +i are roots (one time), the proper form to assume for our particular
solution is
yp = x[(Ax + B)cosx + (Cx + D)sinx]
= (Ax? + Bx)cosx + (Cx? + Dx) sinx.
The substitution of yp into the DE yields
yp + yp = —(Ax? + Bx)cosx — (44x + 2B)sinx + 2Acosx
— (Cx? + Dx)sinx + (4Cx + 2D)cosx + 2Csinx
+ (Ax? + Bx)cosx + (Cx? + Dx) sinx
= 4Cxcosx + 2(A + D)cosx — 4Axsinx + 2(C — B)sinx
= xcOSx — COSx.
Equating like coefficients gives
4C =1
24 +2D = -1
—4A =0
2G.— 2B = 0;
from which we deduce A = 0, B = j, C = 4, D = —4. The general solution is
therefore

YS yp tye
=a 1
= 4x’sinx + 4xcosx — 4xsinx + C;cosx + C) sin x.
. . .

EXERCISES 4.6

In problems 1-20, obtain a general solution by using the method of undetermined

coefficients.

\ A) y" + y’ = —cosx. 2. y" +9y = 18

Y\ (3,)) VY, Oy +t Oy se" 4. y"+ 3y' + 2y = 6x3
\S3 y” + 8y = 2e*
+ 5x 6. y"—y =3e*
% y"
+ 9y = xsin3x 8. y"
+ 4y = 3sinx + 4cosx — 8
 SEC. 4.7 / VARIATION OF PARAMETERS 115

y” + 4y = 2sinxcosx
Q)
— 7 10. y" —3y’ — 4y = 30e*

Ap)! y" — y = 8xe* 12. y" — y =coshx

13. y"+y' =@ + 1)? 14. yy! = Dy" + Sy = e* sinx
15. y eyes y = 2sin*x 16. y” + y" — 4y' — 4y = 3e™* — 4x - 6

0 219)
y" + ay" + 4y' = xe™ “18. 2y” — 3y” — 3y’ + 2y = 4cosh?x
l6y — y = 6e*? 20. y® — 4y" = 8e-* + 3e7 -— x +8

In problems 21-24, set up the correct form for yp but do not solve for the
coefficients.

-@) y" — 2y'’ ty = 5x? — Tx + 4x2" . Be re

22. Woe eye oxen tx, )
N23. y” — 4y” + 4y' = 5x°e* —x + 10

e4) y” + 4y = 3 + le*cosx

In problems 25-28, solve the given initial value problem.

25. y”"+y=sinx, y(0)=-—1, y'(O=1

y = Sy = x= 2, yO) = 9, y'(0) =2
y” + y
27. = 8cos2x — 4sinx, (3) = l, y'(¥) =9

28. y — sy 4y Ser, yQ)=3." 7 C= 0

In problems 29-34, solve the given boundary value problem.

29) y"+2y’+y=x, yO=-3, y=-1

” 1
30.

31) y” + y = 2cosx, yO)=0, y(a)= 0

32. yoy sinx, y(0) =1, »(2) = -|]

33. viceAyes e 7 =y(Q)=~ 1, “yi 2)3

34. Vai = (0) 0, ay(1).= 0
35. Show that when the nonhomogeneous term is a constant, the resulting DE
ay. ot by! + cy = K_ (K constant)
has the particular solution yp = K/c.

4.7 VARIATION OF PARAMETERS

When f(x) is not of the form assumed in Section 4.6.1, or when the DE has variable
coefficients, a more general method of constructing yp is required. This more
+
116 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

general method, called variation of parameters, is analogous to the method used

in Section 2.5 for solving first-order linear DEs. For simplicity we will restrict our
development to second-order equations in normal form
y" 12 a(x)y' + a, (x)y = f(x), (38)

although the technique can be extended to higher-order equations.

Suppose the general solution of the associated homogenous DE
y” + a(x)y’ + anx)y = 0 (39)

is known to be

Yo = Ciyi(x) + Cry2(x), (40)

where y, and y, are linearly independent on an interval a = x = b. We then seek
a particular solution yp that is of the form

ye = u(x)y(x) + v(x)y2(x), (41)

where we try to determine u(x) and v(x) in such a way that (41) satisfies (38).
Notice that (41) has the form of (40) with the arbitrary constants C, and C, replaced
by the functions u(x) and v(x).
To determine u and v, we need two relations that they satisfy. Only one such
relation is obtained by requiring (41) to satisfy (38), and so we must come up with
a second relation that will lead to an easy determination of these functions.
Specifically, we will force u and v to satisfy two first-order DEs, since such
equations are fairly straightforward to solve.
From (41), we get

yp = u(x)yi(x) + v()y2(x) + u'@)yiQ) + v'(a~)y2(x).

Let us eliminate the terms involving derivatives of u and v by equating them to zero:

u'(x)y(x) + v'(x)y2(x) = 0.
Thus,

yp = u(x)yiQx) + v(X)y2Q), (42)

and, differentiating again, we find

yp = ux)yi + ox)y2 + u'(@yi(x) + v'(~)y3(a). (43)

Substituting (41), (42), and (43) for yp, yp, and ys into (38) leads to

yp + ay(x)yp + ag(x)yp = u(x)[yt + ai(x)yi + ao(x)y1]

+ v@)[y2 + ay2 + ao(x)ye]

Zero
+ u'(x)yi(x) + v(x) y3(X)
 SEC. 4.7 / VARIATION OF PARAMETERS 117

= u(x):0 + v(x):0 + u'(~yi(x) + v'@)y2(x)

= f(x).
Therefore,

u'(x)yi(x) + v'(x)ys(x) = f),

and so we find that the two functions uw’ and v’ satisfy the simultaneous equations

u'(x)yi(x) + v'(x)y2(x) = 0, (44)

u'(x)yi(x) + v'(x)y2(x) = f(x). (45)
Nontrivial solutions of (44) and (45) exist provided that the coefficient
determinant

yilx) yo)
yi) y2(x)
does not vanish. But since this determinant is precisely the Wronskian of the
linearly independent solutions y, and y2, it can never be zero on the interval
a =x <b, and so the simultaneous solution of (44) and (45) yields

~m ya(x)f (x) eu = vif) (46)

ui (x)=
W(y1, Y25)() W(y15 Y2)Q)
Taking any integral of these expressions provides suitable functions for u and v, and
we conclude that our particular scenoee is ‘given bysa

yO ney) | 47
-10) |Fo, pent * 1 Wor ynen ee
In certain instances these integrals cannot be evaluated explicitly, and so we must
leave yp in the integral form suggested by (47).

EXAMPLE 26 Solve 2y” + 18y. = csc 3x.

Solution We first rewrite the DE in normal form

1 Pee
y” + 9y = 5 esc 3x. fe aon e

The solution of the homogeneous equation y” + 9y = 0 is

4” = C,cos3x + C,sin 3x,

where we identify y; = cos3x and y, = sin3x. Computing the Wronskian, we

have
cos3x sin 3x
W(y1, Y2)(X) =—
SSO Mi esiCOS. 3X ar 3
118 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

With f(x) = 3 csc 3x, we find from (46)

1
u'(x) = -: sin 3x csc 3x = We

and
ere toe _ COS 3%
v'(x) = 6 cos 3x csc 3x = Aaa

and thus u(x) = —x/6 and v(x) = 7 log|sin 3x|. Hence

yp = u(x) cos 3x + v(x) sin 3x

cy ih
—
6 608 3x
a +
ie log |sin 3x| sin 3x,
s

and the general solution y = yp + yy becomes

1
y= (c,2 ;)cos 3x + (c.bos log |sin 2) sin 3x.

EXAMPLE 27 Solve y" — 3y’ + 2y =1/(1+e%).

Solution The associated homogeneous DE has the solution

Vie C\e* ae Coe"

with Wronskian W(y,, y2)(x) = e*. Hence,

u'() ae ex macs e-

e*(1 +e) [see

and

c.. E

2a e*(1 + e*) alee

Integrating the above functions, we find
‘Cie

u(x) ee
Fearee
hr =
log(1 + e~*)ail =x

and
an e a e

0) = [aaa = |(6 ae
or

u(t) =" est log(1 iee,):

SEC. 4.7 / VARIATION OF PARAMETERS 119

Then, from (47)

yp = e*log(1 + e*) — e* + e*log(1 +e),

which leads to the general solution
y = C3e* + Cre + (e* + e*) log(1 +e”),
where C, = C7 — 1.

EXERCISES 4.7

In problems 1-20, use variation of parameters to find a general solution.

Ue 25 yl ay =x
3. y" + 9y = sin3x 4. y"+y =cscx
Bx y" + y =secx 6. y"+y =sec*x
\7/ y" +y = sec*x 8. y"+y =tanx
(9) y+ y = cotx 10. y"’+y =tan’x
14"y" + y =secxescex 12. y"+y =cscxcotx
yaa 3x
f ) iD , a2 x * 1B fs ’ =
13, y”" — 4y' + 4y = + De 14 ry Bye Fey. crraee
"3 2x
19.) by 2y y= e 7 logx,, x20 *16. Ve area

47.) y” —y- =e *sin(e*) 18. y" — 3y' + 2y = sin(e*)

19. y” +y =csc*xcotx *20. 4y" — 4y’ #y = (1 — x?) e*”

Use variation of parameters to solve the initial value problems 21-24.

21) y"-y=xe*, yO)=2, y'0)=0

22. y" + 2y’ — 8y =2e*—e%, yO=1, y'0)=0
‘ i at
23) y" + y = 2cscxcotx, »(2) =a (z)Tt

2y”"+y'’-y=x+1, yO)=1,. y’0)=0

24.

In problems 25-30, obtain a general solution where the homogeneous solution is

specified.
—.

95.) (1 —x)y” tx’ —y =2e — We; ye = Cx + Goo"

26. x?y" — x(x + 2)y' + + 2)y = x75 yr = Cix + Coxe*
97) x Dy" Se Dy 2 — Dy = 3x*x — 2)°e%s: ye = Cx? + Coe”
28. xy" — (1 + 2x?)y' = x5"; yw = Ci + Core™
120 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

Vax
*29,. (1 —x?)y” — 2xy’ = 2x; yw = Ci + Clog she

*30. x2y" + xy’ + (x? - i)y hay? sin hw.

yy = Cx? cosx + Cox? sinx

31. If y = $(X) is a solution of

y" + ax)y’ + aox)y = f@)

and y = w(x) is a solution of

y" + ai(x)y’ + aolx)y = g@),

show that y = d(x) + W(X) satisfies

y" + aix)y’ + aolx)y = f(x) + g(@).

32. Use the result of problem 31 to solve
y" + 8y = 2e * + Sx.
*33. Use the variation of parameter method to show that

yp = i,sin(x — s)f(s)
ds

is a particular solution of the ee Sconmne DE

ye be
34. Use the result of problem 33 to solve the DE
yy Seow
*35. Use the variation of parameter method to show that

yp = iksinh(x — s)f(s)
ds

is a particular solution of the nonhomogeneous DE

Wi ye FG):
*36. Extend the variation of parameter technique to DEs of the third order,
yl wa(x)y” + bG)y” + ca)y = fG@),
by starting with
ye = U(x)yix) + v@)ya(x) + w(a)ys(x),
where yj, y2, and y; are linearly independent solutions of the associated homogeneous
DE.
*37. Referring to problem 36, solve the third-order DE
ye Dy em We uD yi eo
*38. Find a general solution of the DE
(1 + x*)y" — 4xy' + 6y = 3(1 + x?)
given that y; = 1 — 3x? is a solution of the associated homogeneous equation.
 4.8 THE CAUCHY-EULER EQUATION wy.

Thus far we have solved only constant-coefficient equations although most of the
theory developed in the early sections of this chapter is applicable to general linear
equations with variable coefficients. Unfortunately, we usually cannot solve these
general linear equations as easily as constant-coefficient equations. That is, to solve
variable- coefficient equations one must ass resort to some sort of

of this eS can be enamine into constant- Berheitnt equations by means of a

ens aabseil Ee

and

2 4 (dy\ _@ (ay\ a _ fay

yea low di de)di be Nae ear).
which leads to (now using the notation D = d/dt) |

ay = Dy, xky p= sD (Desai)ye KY

Under this transformation, our original DE (48) becomes the constant-ccoeticent
equation

(49)

Generalizations to higher-order DEs are treated in the exercises.

EXAMPLE 28 Solve x’y" TA, + 2y =0, x2 0.

=

Solution Using the fe eeron x =e', we get a ene

d’y dy + 2y =0. | ne amt

Fm er Fala n
The general solution of this equation is
Vi Ce, Ce,

121
122 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

and transforming back to the original variable x leads to

y(x) = Cyx + Cox’.

EXAMPLE29 Solve x*y” — xy’ + 5y =0,x >0. Rags. 38

| -+

Solution The related constant-coefficient equation is given by

with 1 + 2i as the roots of the auxiliary polynomial. Thus

y(t) = e(C, cos 2t + C, sin 2t)
or
y(x) = x[C; cos (2 logx) + C,sin (2 log x)].
eer ee ae EF I ee ee ee ee ee ee
EXAMPLE 30 Solve x*y" — 3xy’ + 3y = 2x‘e”, x > 0.

Solution This time the DE is nonhomogeneous, and so we must find a particular solution yp
(by variation of parameters) as well as the homogeneous solution y,. The homoge-
neous DE x*y”" — 3xy’ + 3y = 0 transforms into

d*y
Se dy
ae - =

dt? dt eae
by making the change of variable x = e'. Thus the roots of the auxiliary equation
are m = 1, 3, giving us

yu(t) = Cye' + Cre*

or
yy (x) = Cyx + Cox?.
Before using variation of parameters, we must put the DE into normal form by
dividing both sides by x’. This action leads to

y" - *y" a =) = 2x7e",

from which we identify f(x) = 2x7e".

Now it follows that

W Gx =

and thus
SEC. 4.8 / THE CAUCHY-EULER EQUATION 123

: x3 2x2e% 2x? x

Ui) == Ss : == xe; es (x): = a = os

Integration of these expressions leads to

u(x) = —x*e* + 2xe* — 2e*, v(x) = e’,
and hence
yp = u(x)x + v(x)x? = 2x7e* — 2xe*.
Finally, we obtain the general solution
Y= yer yy Sox — er Cx + Ox

EXERCISES 4.8

In problems 1-7, find the general solution of each homogeneous DE valid for
hee Ue

V2 x7y” = S5xy’ + Sy = 0 2. xy" — 4xy’ + by = 0

3) x7y"= xy’ ty =0 Al 4x7)" +'y = 0
HS). 3xy”
+ 2y' = 0 6. .x*y" + xy’ ty =0
Ty =x’y” — Sxy' + 25y = 0 .

In problems 8-10, find the general solution of each nonhomogeneous DE by the

method of undetermined coefficients. (Transform the entire DE by letting x = e’,
find the general solution of this transformed nonhomogeneous DE, and then
transform back to the variable x.)

8. x2y" — 2xy’ + 2y = x? Ae Cee Sia =a

10. x?y" + xy’ + 4y = 2xlogx

Solve problems 11-15 by any method (assume x > 0).

11. — 4xy' + 6y = 4x — 6
x?y" 12) xy ty ty = 4xlosx
130 x2y xy > y= x 14. — 2xy’ + 2y = x°e*
x2y"”
(15 x2y" — 2xy’ + 2y = x? logx?

Solve problems 16-20, subject to the prescribed initial conditions.

16. x2y"—2xy’- l0y=0, y)=5, y'() =4

17. xy" —4xy'’ + 6y =0, y@)=0, y'@Q)=4
18) 4x2y” + 8xy’ ty =0, y(lI)=1, y'()=0
124 CHAP. 4 / LINEAR EQUATIONS OF HIGHER ORDER

(19.) x?y" — 3xy' + y = x3, y(1)=1, y'(1) =0

20. x2y" + xy’ + 4y = sin(logx), y(l)=1, y'(1I)=0

In problems 21-23, solve the equation by assuming a solution of the form y = x",
where m must be determined.

Gr) 2x*y"”
+ 3xy'’ -— y =0 22; x2y"
— 2xy' + 2y =0
23. x2y" + Txy’ + Sy =0
*24. Show that under the transformtion x = e’,
a) meDD SAND ey,
xty = D(D: = 1KD-— 2)D —-3)y:
where D = d/dt, and deduce that in general
x"y™ = D(D — 1(D —2)---(D-—n + Dy.

In problems 25-30, use the result of problem 24 to find the general solution of each
DE Vor. 0:

(25) ay oxy. tay =O 26. x*y” + x*y” — 2xy’ + 2y = 0

2h. xy Or yet xy” = y = 0
28s xy" = 3x7y" = bxy’ — 6y'= 3+ logxt
*29. x°y" — x7y" + 2xy' — 2y = x3
*30. x*y + 6x°y" + 15x2y” + Oxy’ + l6y =0

Solve problems 31 and 32 by first making an appropriate change of independent

variable to reduce the given DE to a Cauchy-Euler equation.

“31. @ + 5)*y" = @ +.5)y' = 3y =0

"32. (3x = 2)*y" — 6(3x — 2)y’ + 12y = 0

REFERENCES
Boyce, W. E., and DiPrima, R. C. 1977. Elementary differential equations .3rd ed. New
York: Wiley.
Coddington, E. A. 1961. An introduction to ordinary differential equations .Englewood
Cliffs, N.J.: Prentice-Hall.
Ince, E. L. 1956. Ordinary differential equations. New York: Dover.

Rabenstein, A. L. 1972. Introduction to ordinary differential equations .2nd ed. New York:
Academic Press.
Rainville, E. D., and Bedient, P. E. 1974. Elementary differential equations .5th ed.
New
York: Macmillan.
REFERENCES
125

Ross, S. L. 1980. Introduction to ordinary differential equations. 3rd ed. New York:
Wiley.
Zill, D. G. 1979. A first course in differential equations with applications .Boston, Mass.:
Prindle, Weber & Schmidt.
In the present chapter we consider applications of DEs to problems involving me-
chanical vibrations and electric circuits. Such applications naturally lead to initial
value problems, since time is normally the independent variable and the auxiliary
conditions are all prescribed at a particular instant of time.
Newton’s law of motion is used in Section 5.2 to derive the DE governing the
small vertical movements of a spring-mass system. Free motions are discussed here
for cases of both undamped and damped systems. The latter type of motion results
when frictional or other types of resistive forces in the system are taken into account.
The three cases of overdamped, underdamped, and critically damped motions are
carefully distinguished, and typical solution curves are illustrated for each case. In
Section 5.3 we examine certain kinds of forced motions resulting from an external
stimulus applied to the system. In particular, we consider the case of an impressed
periodic force whose frequency is at or near that of the natural frequency of the sys-
tem, leading to a state of resonance. Analogous systems involving simple electric
circuits are briefly discussed in Section 5.4.
In Section 5.5 we introduce the notion of a Green’s function, which permits us to
develop general solution formulas for initial value problems similar to the formula de-
rived in Section 2.5.1 in connection with first-order equations. The concept of an im-
pulse function, which is so useful in circuit analysis, is presented in Section 5.6, and
this in turn helps to describe a physical interpretation of the Green's function.
In Section 5.7 we look at equations with variable coefficients, which arise in appli-
cations where some of the system parameters vary over time. Since solution tech-
niques for such DEs have not yet been developed (see Chapter 9), we limit our
study to the qualitative behavior of the solutions. We are particularly interested in the
oscillatory characteristics of the solutions, which are discussed in terms of the fa-
mous Sturm separation and comparison theorems.
Fora shorter course, Sections 5.5, 5.6, and 5.7 can be omitted.

126
 5.1 INTRODUCTION

Initial value problems arise in the study of particle motion, population dynamics,
and electric circuits, as well as several other areas of application. The general
problem is to solve the linear equation

where M is the nth-order linear differential operator (D = d/dt)

MES Di aD. pert ai). da), (2)
subject to the n auxiliary conditions

yo) =k, yi) =k, ..., ye) = ket, (3)

all specified at a single point. Although the value is usually chosen as zero, the
general theory does not require this choice. Equation (1) is said to be in normal
form, and the operator M is said to be a normal differential operator on any interval
for which the coefficients in (2) are continuous.

Remark. Since the independent variable in initial value problems is usually time,
it is customary to use f rather than x.

An interesting and important property of initial value problems is that they

always possess unique solutions when the coefficients and forcing function are
continuous. Without proof, we state the following fundamental existence-
uniqueness theorem.*

Theorem 5.1 If f(t) is continuous on the interval t = f and M is a normal differential operator
of order n on this same interval, then for any choice of the constants ko, ki, . . . ,
k,-1, there exists a unique solution of the initial value problem

y(t) =ko, yi(to) =k... 5 y(t) = ky-1-

As an immediate consequence of Theorem 5.1, we have the following corollary.

Corollary 5.1 If M is anormal differential operator of order n on the interval t > f, then the initial
value problem
M[y]=0, t>h,

¥q) HV (o)=H=°-o = y(t) = 0,

*For a proof of Theorem 5.1, see E. A. Coddington, An Introduction to Ordinary Differential

Equations (Englewood Cliffs, N.J.: Prentice-Hall, 1961).

127
128 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

has only the trivial solution y = 0.

Proof: By inspection we see that y = 0 is a solution of the initial value problem,

and by Theorem 5.1 it is the only solution.

EXAMPLE 1_ Show that y = 0 is the only solution of

ye by 10s) yO) 20, yO) 0:

Solution The general solution of the DE is

y =C,cost + C)sint.
Imposing the prescribed initial conditions, we find
y(0) = C, + C20 = 0,
y'(0) = -—C;;0 + C, = 0,7
from which we deduce that C; = C, = 0. Clearly, y = 0 is the only solution.

5.2 SMALL MOTIONS OF A SPRING-MASS SYSTEM

When different weights are attached to an elastic spring suspended from a fixed
support, the spring will stretch by an amount that varies with the weight. Hooke’s
law* states that the spring will exert an upward restoring force proportional to the
amount of stretch s (within reason); i.e., F = ks. The constant of proportionality,
denoted by k, depends upon the “stiffness” of the spring and thus is different for
each spring. For example, if a 10-pound weight stretches a spring 6 inches (3 foot),
then 10 = k(3), or k = 20 lb/ft, whereas if the weight only stretches the spring 2
inches, we find k = 60 lb/ft.
The two most common systems of units and their abbreviations are given in
Table 5.1.
Suppose the natural length of a spring is b units and a weight W = mg is attached
to the spring. The weight, which is also referred to as the “mass,” will then attain
a position of equilibrium at y = 0, which is s units from the equilibrium position
of the spring itself (see Figure 5.1). The upward restoring force is ks, which is offset
by the weight mg. If the system is now subjected to an external force (downward)
of magnitude F(1), the weight will move in the vertical direction. Let us assume
such motions are “small” so that Hooke’s law will remain valid.
In addition to an external force, there frequently exists a retarding force caused
by resistance of the medium in which the motion takes place or possibly by friction.
For example, the weight could be suspended in a viscous medium, connected to a

*Named in honor of the English physicist ROBERT HOOKE (1635-1703).

SEC. 5.2 / SMALL MOTIONS OF A SPRING-MASS SYSTEM 129

(a) (b)

Figure 5.1 (a) Unstretched spring. (b) Equilibrium position with

mass. (c) Spring-mass system with a forcing function and damping.

Table 5.1 Common Systems of Units and Their Abbreviations

System of Units Force Length Mass Time

International newton (N) meter (m) kilogram (kg) second (s)

English pound (1b) foot (ft) slug second (s)

Note: To convert from one system to another, we use the following relations:
1 N = 1 kg-m/s? = 0.2248 Ib 1 kg = 0.0685 slug
1 m = 3.2808 ft 1 Ib = 1 slug-ft/s? = 4.4482 N

dashpot damping device, etc. In practice, many such retarding forces are approxi-
mately proportional to the velocity y’. Hence, we will assume the resistive force
is cy’, where c is a positive constant, and this force acts in a direction opposing the
motion. Now, taking into account all forces acting on the system, we deduce
im REGFY Nes a WS— am is a a 9) [ ;

— ONIN S aKa Gye elt (t))

Zero

which is a consequence of Newton’s second law of motion (F = ma). Simplifying

this equation, we have |

| my" + cy’ + ky = FQ. | (4)

}

Remark. Note that the terms on the left-hand side of (4) represent system forces
such as restoring and damping forces, while the function F (¢) on the right-hand side
represents an external force to the system. In this sense, the function F (ft) is referred
to as a forcing function or input function, which gives rise to additional motions of
130 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

the system superimposed onto the free motion that would result in the absence
of F(t).

Equation (4) describes the general motions of a spring-mass system. Observe

that it is a nonhomogeneous, second-order, linear DE with constant coefficients.
The motion is said to be undamped when c = 0 and damped when c # 0. The
motion is further classified as free when F(t) is absent and forced when F(t) is
present.
If the mass is initially displaced a distance yo from the equilibrium position and
released from that point with velocity vo, then we prescribe the initial conditions

y(0) = yo, y'(O) = t%. (5)

In order to investigate the solutions of (4) subject to the initial conditions (5), it is
best to consider several special cases.

EXAMPLE 2 Derive the initial value problem for a 2-kg mass suspended by a spring with spring
constant k = 32 N/m. A force of 0.5 sin 3t is applied to the mass, and a dashpot
damping mechanism is such that c = 5 kg/s. The mass is released from rest 3 cm
below the equilibrium position.

Solution The relevant constants are m = 2, c = 5, and k = 32, leading to the DE

2y" + Sy’ + 32y = 0.5 sin 3¢.

The initial position is y(0) = 3 cm = 0.03 m, and the initial velocity is y'(0) = 0
since the mass is released from rest.
eee

5.2.1 Undamped Free Motion

For the case when c is sufficiently small compared with mk and the time span is
short, it may be acceptable to neglect the damping term cy’. (All systems have a
certain amount of damping, no matter how small the motions or how short the
period of time.) If this is done and if no external force acts on the mass, then the
initial value problem whose solution describes the motion is
my + ky= 0,=-t> 0, yO) =x, y'(0)i= wp. (6)
The general solution of the DE above is
y = C\ COs wot + Cr sin wot,
where we write w) = Vk/m for convenience. If we subject this solution function
to the prescribed initial conditions, we see that

y(0) = C,cosO + C,sin0 = yo,

y'(0) = w(—C,sin0 + C,cos0) = vo,

which identifies the constants C,; = yo and C, = Up/Wo. The motion of the mass
is
therefore described by
SEC. 5.2 / SMALL MOTIONS OF A SPRING-MASS SYSTEM 131

Uo.
Y = YoCOS @pt + — sin Wot. (7)
Wo

Making use of the trigonometric identity

cos(a — b) = cosacosb + sinasinb,

we can rewrite (7) in the more compact form

y = ACOs (aot — d). (8)
The number A, defined by

A = 4/98 + ey (9)
Wo

is called the amplitude of the motion. It gives the maximum (positive) displacement
of the mass from its-equilibrium position. The angle ¢ is referred to as the phase
angle and is chosen in such a way that

cos d =7
tang = te ; (10)
sing =—~ 4
WA
Any motion described by a single sinusoidal function as in (8) is called simple
harmonic motion. Such motion is clearly periodic, since the mass will oscillate
between y = —A and y = A. The time between successive maxims, or the length
of time required to complete one cycle of the motion, is the period of the motion
and is given by

aWo 2mk (11)

The reciprocal of the period, or the number of cycles per second*, is called the
natural frequency of the system, which we denote by fo = wo/27r. The value
Wy = Vk/m is known as the angular frequency.
Regardless of the values of the input parameters yo and vo, the graph of (8) is
simply that of a cosine curve (see Figure 5.2). By changing either yo or vo, the

20
Wo

Yo A
——_> t

Figure 5.2 Simple harmonic motion.

*The unit cycles per second (cps) is now called hertz (Hz).
132 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

cosine curve will appear to shift a certain amount along the t-axis and possibly
change amplitude.

EXAMPLE 3 Find the natural period of a spring-mass system for which the spring is stretched 4
in. by a 6-lb weight.

Solution Since 4 in. corresponds to } ft, Hooke’s law leads to

6=k--,|
3
or k = 18 lb/ft. The mass m = W/g= % slug, and hence
m\ 1/2 6 ve = aV6
T= 2n() = an aim Fie ch

EXAMPLE 4 Suppose a 16-lb weight is attached to the spring in Example 3 and released 3 in.
above the equilibrium point of the spring-mass system with an initial velocity of 2
ft/sec directed upward.* Describe the subsequent motion of the mass.

Solution In Example 3 we found that k = 18 lb/ft, and the 16-lb weight corresponds to a
mass of 3 slug. Thus the motion of the mass is described by the solution of the initial
value problem

jy mely a0: es aia: yiOVE

] "” 1 Uy

Solving, we get

1 1
y=- cos 66 — 3 Sin 6r.

Putting this solution in the form of Equation (8), we find the amplitude of motion
given by

whereas the phase angle satisfies cos@ = —3, sind = —!, or d = 4.07 rad
(233°). Hence,

2
Va
D cos (6¢ — 4.07)

for which the period of motion is T = 2a(m/k)'*= 7/3 (see Figure

5.3). The
frequency is fo = 3/m Hz.

*The positive y-axis has been chosen downward so that, for example, y(0)
= —Yo if the mass is pushed
yo units above the equilibrium position. The same is true of the velocity.
 SEC. 5.2 / SMALL MOTIONS OF A SPRING-MASS SYSTEM 133

Sometimes it is useful to know the values of time for which the graph of y(t)
crosses the positive f-axis. This corresponds physically to the mass passing through
its equilibrium position. Writing the solution in the form of Equation (8) is very
helpful for such calculations. For instance, using the solution in Example 4, we
observe that cos (6t — 4.07) = 0 when

(27271) as
6t — 4.07 = ee
Z
where n is an integer. The first positive value of ¢that satisfies this relation is found
to be tf; = 0.42 rad (n = 0), whereas the next value is t, = 0.94 rad (n = 1), and
so on (see Figure 5.3).

Figure 5.3

[O] 5.2.2 The Pendulum Problem

A mass m is suspended from the end of a rod of constant length b whose weight
is negligible (see Figure 5.4). We wish to determine the equation of motion of the
mass in terms of the angle of displacement @.
Summing forces makes it clear that the weight component mg cos 6, acting in the
normal direction to the path, is offset by the force of restraint in the rod. Therefore,

Figure 5.4
134 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

the only weight component contributing to the motion is mg sin 6, which acts in the
direction of the tangent to the path. If we denote the arc length of the path by s,
then Newton’s second law of motion reads

Uke
us kon meg sin 0 ;
dt? °
where the minus sign illustrates that the tangential force component opposes the
motion for s increasing. Now the arc length s of a circle of radius b is related to
the central angle 6 through the formula s = b@, and so the equation of motion (after
simplification) transforms to
2
d‘0
bt 8 sing = 0. (125

The equation of motion (12) is nonlinear due to the term sin 0. To solve this
equation exactly would necessitate introducing a special function called the Ja-
cobian elliptic function ,* since the equation has no solution that can be expressed
in terms of elementary functions. However, if we restrict the motion so that 6 is
always a “small angle,” then we might use the approximation sin 9 = @ and replace
(12) with the linear DE
d*6
b—>; + 2g =0. 13

In this form we recognize (13) as being equivalent in structure to the equation of

motion of the undamped spring-mass system. Hence the pendulum problem for
small motions is mathematically equivalent to the spring-mass problem, which
once again illustrates the fact that the same DE can be used to describe contrasting
physical phenomena.

5.2.3 Damped Free Motion

When damping effects are taken into account, the free motions of the spring-mass
system are described by solutions of

IY ty ee ky =O, = (14)
The auxiliary equation from which the solutions of (14) are determin
ed is
mA* + ct +k = 0,
with roots

—c + (c? — 4mk)!”
XA, a0) = a
(13)

The solution obviously takes on three different forms, depend

ing upon the mag-
nitude of the damping term. The three cases are:

*See, for example, T. C. Bradbury, Theoretical Mechani

cs (New York: Wiley, 1968).
tWe use the parameter A here since m denotes mass.
SEC. 5.2 / SMALL MOTIONS OF A SPRING-MASS SYSTEM 135

Case I—Overdamping: c? > 4mk (Xj, 2 distinct and real)

Case II—Critical damping: c? = 4mk (x:=),= -<)

2m
Case I]]—Underdamping: c? < 4mk (\;, \2 complex conjugates)

Let us discuss each case separately.

Case I—Overdamping: The damping is large compared with the spring con-
stant. Both roots of the auxiliary equation are real and distinct, leading to the
solution formula
: t
y= corny exp Gc ~ ami)? |+ C,exp |-« _ amy? || (16)

This equation represents a smooth, nonoscillatory type of motion. Typical graphs

of this motion are illustrated in Figure 5.5.

Case II—Critical damping: For this case the roots of the auxiliary equation are
equal, so the solution takes the form
wasyen UC hrr.Cof): (17)
The motions here are similar to those of the overdamped case, as shown in Figure
oO:

Vo = initial velocity

yo Vv) > O

> <x
ae

Figure 5.5 Overdamped motion. Figure 5.6 Critically damped motion.

Case III—Underdamping: This case is the most interesting of all three cases.
The roots of the auxiliary equation are complex, and so the solution can be
expressed as
y a3 (CAN OF cos pt + C sin Lt),
(18)

where = (4mk — c’)'?/2m. Regardless of the initial conditions prescribed, the

mass will oscillate back and forth across the equilibrium position with the amplitude
of motion steadily decreasing in time (see Figure 5.7).

In all three cases, the solution of the homogeneous DE (14) contains the multi-
plicative factor e~’””", which tends to zero after a sufficiently long period of time.
136 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

Taw

Figure 5.7 Underdamped motion.

Hence, regardless of the prescribed initial conditions, the solution of (14) must
approach zero as time goes on. This situation confirms what we intuitively expect—
without damping (friction), the motion of the system continues indefinitely; but
with damping, the motion eventually dies out.
a es eee eee

EXAMPLE 5 A spring-mass system involves a mass of 4 kg, a spring with k = 64 N/m, anda
dashpot with c = 32 kg/s. The mass is lowered 1 m from its equilibrium position
and released from rest. Determine the subsequent motion.

Solution The initial value problem describing the motion of the Spring-mass system is
Ay" + 32y + 64y = 0, yO) =1, yO) = 0.
Dividing the equation by 4, we find the auxiliary equation A2 + 8A + 16 = 0 with
solutions A = —4, —4. Thus the general solution of the DE is

y= (Ce Collen:
and imposing the initial conditions leads to C; = 1 and C; = 4, resulting in
y=(1 + 4pe™.
The system is therefore critically damped. The mass will never pass through
its
equilibrium position y = 0, but will slowly approach it as time goes on.

Although damped motion is not truly periodic, it is sometimes

convenient to
introduce the notion of a quasiperiod, defined as the time between
successive
maxima of the displacement. That is, we write

i 2 —1/2
aot nan ~ | : (19)
which can also be expressed in terms of T [given by (11)]
b)

ae ce \ate oe
TFS (1 ets | ein ee
W, ( a r(1 ) f Cu
SEC. 5.2 / SMALL MOTIONS OF A SPRING-MASS SYSTEM 137

where wo = (k/m)'*. Here we see that when damping is small, i.e., when
c7/4km<<1, the quasiperiod T is approximately equal to T.

EXERCISES 5.2

A Find the frequency of oscillation of a spring-mass system if the mass is 4 kg and the
spring constant is 100 N/m. What is its natural period?
re Calculate the time necessary for a 0.03-kg mass hanging from a spring with spring
constant 0.5 N/m to undergo one full oscillation. What is the natural frequency of the
system?
A spring-mass system is stretched 6 in. by a 12-lb weight. If the weight attached to
the spring is pulled downward 4 in. below the equilibrium position and started upward
with a velocity of 2 ft/s, show that the subsequent motion is described by the function
y = 4cos 8r — fsin 81.
Show that the solution in problem 3 can be expressed in the form
y = Acos(8t — d),
and find the first two positive values of time for which y = 0.
A 10-lb weight stretches a steel spring 2 in.
(a) Determine the natural period of the spring-mass system.
(b) If the spring is stretched an additional 2 in. and then released, describe the
_ subsequent motion of the mass.
4 ew aus RANKKOAM ;

The period ‘of free oscillations of a mass on a string is 7/2 s. What is the numerical
value of the length of the string in feet? Hint: See Equation (13).
A clock has a pendulum | m long. The clock ticks once for each time the pendulum
makes a complete swing, returning to its original position. How many ticks will the
clock make in | min?
Hint: Use Equation (13).
A 24-lb weight stretches a spring 4 in. Determine the equation of motion if the weight
is released from a point 3 in. above the equilibrium position with a downward velocity
of 2 ft/s.
The period of free, undamped oscillations of a spring-mass system is observed to be
m/4 s. If the spring constant is given by 16 lb/ft, what is the numerical value of the
weight in pounds?
10. Find the solution of the DE
my" + ky =0
in the form A cos (wot — ) when the initial conditions are prescribed by

(a) y()=yo, y'(0)=0.

(by yO) = 0. y(0) = vo,
11. A mass of 1 slug is attached to a spring with k = 9 lb/ft. The mass initially starts
moving from a point | ft above the equilibirum position with velocity V3 ft/s directed
downward. Find the first positive value of time for which the mass is moving down-
ward with a velocity of 3 ft/s.
138 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

12. Prove that the maximum value of the speed of a mass undergoing simple harmonic
motion occurs when y = 0.
13. Determine the natural period of oscillation of the pendulum in problem 7 if the
pendulum is 2 m long.
14. Solve Equation (13) of the pendulum problem when the weight is 8 Ib and the rod is
1 ft long. Assume the weight is released from an angle of 3 rad with a positive velocity
of 3 rad/s.
15. At what time does the pendulum in problem 14 first pass through the angle 6 = 0?
What is its velocity at this time?
16. An 8-lb weight, attached to the end of a vertical spring, is pulled yo ft below its
equilibrium position and released at time t = 0 with a downward velocity of 3 ft/s.
Determine the spring constant k and the initial displacement yo if the amplitude of the
resulting motion is known to be V5 and the period is 77/2.
mae Show that underdamped free motion has the following characteristics:
(a) The characteristic angular frequency pw is independent of the initial conditions
but decreases as c increases.
(b) The natural logarithm of the ratio of two consecutive maximum amplitudes is the
constant 6 = mc/mp. The number 6 is called the logarithmic decrement of the
oscillation.
(c) Find 6in the case y = e ‘cost, and determine which values of t correspond to
maximum and minimum displacements.
18. Show that overdamped free motion has the following characteristics:
(a) The mass cannot pass through y = 0 more than once.
(b) If the initial conditions are such that the constants C; and C>2 in the general
solution have the same sign, the mass never passes through y = 0.
19. Under what conditions on yo and vo, where yo = y(O) and vo = y'(0), will critically
damped free motion have a maximum or a minimum for t > 0?
20. A spring is stretched 6 in. by a 3-lb weight, which is started from its equilibrium
position with an upward velocity of 12 ft/s. If a retarding force equal in magnitude
to 0.03v exists, find the resulting motion.
A certain straight line motion is described by the initial value problem \

y" + 2by' + 169y = 0, y(0)=0, y’0)=8, b>O0.

(a) Find the value of b that leads to critical damping.
(b) Find the solution for that value of b in (a).
(c) At what time does the motion momentarily stop, if at all?
22. A 4-lb weight is attached to a spring with spring constant 2 lb/ft. If the weight is
released from | ft above the equilibrium position with a downward velocity of 8 ft/s,
determine the time that the weight passes through the equilibrium position, assuming
the retarding force is equal in magnitude to the instantaneous velocity. Find the time
and position of the weight at its maximum displacement after passing through equi-
librium.
23. Determine the maximum displacement of the free motion of a spring-mass system
governed by the initial value problem
 SEC. 5.3 / FORCED MOTIONS 139

y" + Sy’ +4y =0, yO)=1, y’O)=1.

Does the graph of y ever cross the f-axis?
24. Consider a spring-mass system experiencing a viscous damping term.*
(a) Ifthe mass m is given an upward initial velocity of 50 m/s from the equilibrium
position, find the motion given that m = 4 kg, k = 64 N/m, and c = 40 kg/s.
(b) Determine the time between consecutive maximum
displacements of the mass
when m = 30 kg, k = 2000 N/m, and c = 300 kg/s.

5.3 FORCED MOTIONS

If an external force F(t) is also present, the initial value problem describing the
possible forced motions of a spring-mass system reads

my + cy +ky =F), yO)=yo, y'(0) = vo. (21)

To discuss such motions we will again take cases.

5.3.1 Undamped Forced Motion

For undamped motion we set c = 0 in (21) to get

my’ + ky = F(). (22)

In order to investigate undamped motions due entirely to the input function F(#),
it is convenient to assume the system is initially at rest. Thus we prescribe the
conditions

OPS AOL oD) ah (23)

Let us suppose the external force is a constant described by F(t) = P. Assuming
a particular solution of (22) of the form y, = A, we find that A = P/k, giving the
general solution
y = C,cos wot + CzSiN Wot + P/k,

where w) = Vk/m is the natural (angular) frequency of the system. Applying the
initial conditions (23), we deduce

y= (1 — COS Wot). (24)

In this case the mass oscillates at its natural frequency between the points y = 0 and
y = 2P/k with a period of 277/a s.
If a sinusoidal force F(t) = P cos wt is applied, we assume a particular solution
exists of the form (see Section 4.6)

*A viscous damping term arises if the spring-mass system is suspended in a fluid like oil or water, or
if air resistance cannot reasonably be neglected.
140 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

yp = Acoswt + Bsinat,

where the constants A and B are to be determined. The substitution of y, into the
DE leads to
(k — mw’)(Acos wt + Bsin wt) = Pcos at.
Hence, equating like coefficients of coswt and sinwt, we obtain A =
P/(k — mw) = P/m(ws — w*) and B = 0, so our general solution is

y = C, COS @ot + CzSin wot + COS wt.

m(we — w’)
If we now impose the initial conditions (23), we see that C; = —P/m (we — w’) and
C, = 0, and therefore
P

Va= isa LC OSIOR RCOS One) (25)

m(@§ — w’)
provided w # wo. This time the motion consists of two modes of vibration—the
natural mode at frequency wp and the forced mode at frequency w.
An interesting phenomenon occurs when the forcing frequency in (25) is close
to the natural frequency, i.e., when |@ — | is small. By setting wt = a + band
Wot = a — b, we can use the trigonometric identities

cos(a + b) = cosacosb + sinasinb

to rewrite (25) in the form

Pdee ‘
y= ree aya (o - 05| sin (o25 oy}. (26)
m(@? — w

Since |@ — @o| is small, the period of the sine wave sin[(w — wp) t/2] is large
compared with the period of sin[(@ + @p) t/2]. The motion described by (26) can
then be visualized as a rapid oscillation with angular frequency (@ + wp)/2, but
with a slowly varying sinusoidal amplitude (see Figure 5.8). Motion of this type,
possessing a periodic variation of amplitude, exhibits what is called a beat. The

al 9

27
i | or oy
Figure 5.8 Phenomenon of beats.
SEC. 5.3 / FORCED MOTIONS 141

phenomenon of beats can most easily be demonstrated with acoustic waves—for

example, when two tuning forks of nearly the same frequency are sounded at the
same time.
In the special case when the system is excited at its natural frequency (w = wp),
the response of the system becomes [from (25)|
P
y = lim (COs wt — cos ont |;
So Ooo)
which, evaluated through use of L’H6pital’s rule, leads to

y= i Sin Wot. (27)

2M@o

It is clear that the amplitude of motion will become unbounded in (27) as t>~,
and thus we have the phenomenon of resonance (see Figure 5.9). Of course, we
recognize that in a physical problem the amplitude cannot become unbounded. A
certain amount of damping, however small, is always present, and this has the
effect of limiting the amplitude. Also, if the amplitude should become large
enough, the system is likely to fail. This situation has actually caused certain
bridges to collapse, such as the Tacoma Narrows bridge at Puget Sound in the state
of Washington. On November 7, 1940, only four months after its grand opening,
a huge portion of the bridge collapsed into the water below. From the very begin-
ning the bridge had experienced large undulations, which were later attributed to
the wind blowing across the superstructure. Therefore, in designing such struc-
tures, it is very important to make the natural frequency of the structure different
(if possible) from the frequency of any probable forcing function.

cals

Figure 5.9 Phenomenon of resonance.

Remark. It should be noted that the reason soldiers do not march in step across
bridges is to avoid any possibility of resonance occurring between the natural
frequency of the bridge and the frequency of the uniformly stomping feet.

5.3.2 Damped Forced Motion

When damping effects are important, we can think of the solution of (21) as
composed of two parts, i.e., y = yp + yy. Since it contains the multiplicative
 os

142 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

factor e~/2™, the solution function y, contributes only initial effects to the motion
and it is called the transient solution. The function yp dominates the response of the
system after initial effects diminish and is therefore referred to as the steady-state
solution of the system.*
Let us assume the forcing function is F(t) = Pcos wt. Then it seems natural to
assume that the steady-state solution has the general form
yp = Acoswt + Bsinat, (28)
where A and B must be determined. Observe that we are assuming that the fre-
quency of the steady-state motion is the same as the frequency of the forcing term
producing the motion. When (28) is substituted into the DE, we obtain
[((k — mw*)A + wcB|cos wt + [(k — mw?)B — wcA]|sin wt = P cos at.
Equating like coefficients of cos wt and sin wt results in the simultaneous solution
of A and B
P(k — mw’) Pwc
Ae Ge
(k — mw’) + wc”
re Bg ee
(k — mw’) + wc?
The particular solution, or steady-state solution, is then
P(k — mw’) Wwe :
ee = Goma ame ¢ Sa
kona.
ot: (29)

Since the amplitude R of the oscillation is of some importance, let us rewrite (29)
in the equivalent form
yp = Rcos(at — ), (30)
where (@) = Vk/m)

P
Kes [m%(we — w2)? + wc?” (31)

and the phase angle ¢@ is such that

tangd =——~
k— mw
a
m(w — w*)
(32)
Unlike the case of undamped motion, here we see that the maximum value of R
does not occur when w = wo. The maximum amplitude of the motion can be found
by setting dR/dw = 0, which occurs when (see problem 19 in this section)

Spee (33)

However, for sufficiently large damping such that c? > 2m?w4, there is no value
of w satisfying (33), and hence no maximum amplitude. For damping coefficients

*By steady state, we mean only that portion of yp that does not go to 0 as t>,
 SEC. 5.3 / FORCED MOTIONS 143

satisying c* < 2m*w$, the maximum amplitude is found to be (see problem 20 in

this section)
2Pm
R max
~ ¢[4m we — c Py: (34)

The amplitude R given by (31) is plotted in Figure 5.10 as a function of the input
frequency w. It thus becomes clear that large amplitudes due to resonance can be
avoided by a sufficient amount of damping.

c= Mwy
p Ee
mw
C= MW

c= 2m = c< V2 may
(critical damping) —] TSS - Seinen a
-=<—

Figure 5.10 The amplitude R as a function of o.

EXAMPLE 6 Interpret and solve the initial value problem

1
3)"Dy elOy = Scos2t, y(0)= =" y'(0) = 0.

Solution We can interpret the problem as representing a spring-mass system consisting of a

mass of 1/2 unit, spring constant equal to 10 units, a damping term equal to twice
the instantaneous velocity, and a periodic forcing term equal to 5 cos 2t. The mass
is lowered 1/2 unit from the equilibrium position and released with zero velocity.
To solve this problem, we find it convenient to first rewrite the DE in the form
y” + 4y’ + 20y = 10 cos 2z.
The auxiliary equation of the associated homogeneous DE is m* + 4m + 20 = 0
with roots m = —2 + 4i. Hence,

Ya = e *(C, cos4t + C,sin4t).

CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

Using the method of undetermined coefficients, we set

yp = Acos2t + Bsin2t.

Upon differentiation and substitution into the nonhomogeneous DE, we find that
the coefficients A and B must satisfy the equations
—8A + 168 =0, 16A + 8B = 10.
Calculating, we have A = 3 and B = j, so that
1 Lay
yp = gos ott qoin2t.

Combining y, and yp leads to the general solution

1 Le
y =e *(C,cos4t + Cysin4t) + 5 008 21 + qsin2e.

Applying the initial conditions, we see that

el
+-=-
o rh

Gn 1 AG 2 Se2 eye’

and hence it follows that C, = 0, C, = —%. Therefore,

Peer sine
&
y= ge a 5I 008 2t a 4Lisesin2t.
(ee
transient solution steady-state solution

After a short period of time (¢ > 3), the dominant part of the solution is the
steady-state term, which represents simple harmonic motion with maximum am-
plitude

Oia:
Ne Tye
k= = Wert = feree Os

and period 7.

The undetermined-coefficient method of producing the steady-state solution is

effective only when the forcing function is of a simple nature like Example 6. Of
particular importance in many applications, however, is the case when the forcing
function is a periodic function other than a simple sinusoid. When this situation
occurs, the method of Fourier series becomes a useful device, and when the forcing
function is of a more general nature, the method of Green’s function (Section 5.5)
can be quite effective.
SEC. 5.3 / FORCED MOTIONS 145

EXERCISES 5.3

A 2-lb weight stretches a spring 6 in. An impressed force 16 sin 87 is acting upon the
spring, and the weight is pulled down 3 in. below the equilibrium position and
released. Determine the equation of motion.
If an impressed force }cos 41 is imposed upon the system in problem 3 of Exercises
5.2, determine the subsequent motion.
*3. A spring with spring constant k = 0.75 lb/ft has a weight of 6 Ib attached, which is
at rest in the equilibrium position. A 1}-lb force is applied to the weight in the
downward direction for 4 s and then removed. Discuss the subsequent motion.
A spring stretches 6 in. when a 4-lb weight is attached. If the weight is started from
the equilibrium position with an upward velocity of 4 ft/s and has an impressed force
of 5cos 8t acting on the weight, determine the position of the weight for all time. What
is the position when ft = 2 s?
A 2-kg mass is attached to a spring with k = 32 N/m. A force of 0.1 sin 4t is applied
to the mass, which is at rest. Neglecting damping, calculate the time required for
failure to occur if the spring breaks when the amplitude of oscillation exceeds 0.5 m.
Show that the solution of
y’ + 25y = 10cos7t, y(O)=0, y'(0) =0,
is given by y = 2sin¢sin6t. How many seconds are there between beats?
A 20-N weight is suspended by a frictionless spring for which k = 98 N/m. An
external force of 2cos7t acts on the weight. Find the frequency of the beat, and
determine the maximum amplitude of the motion that starts from rest.
How many seconds are there between the beats in problem 7?
Verify Equation (27) by solving directly the initial value problem

my" + ky = Pcosaot, y(0)=y'(0)=0, wo = Vkim.

10. Assuming c* — 4mk <0, determine the complete solution of the DE
my" + cy’ + ky = Pcosat,
satisfying the following sets of initial conditions:
(a) y(0)=yo, y'(0) = 0,
(b) y(0)=0, yO) = to,
(c) y(0)=yo, y'(0) = to.
ne Show that the motion of a body rising with drag proportional to velocity is given by
my" + cy’ + mg = 0.
12. For problem 11, assume the initial velocity of the body is 100 m/s upward, c = 0.4
kg/s, and m = 2 kg. How high will the body rise?
13. For the body of problem 12, determine the time required for the body to reach its
maximum height, and compare this with the time it takes for the body to fall back to
its original position.
14. Verify Equation (31).
15. Determine both the transient and steady-state solutions of the initial value problem
146 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

y' + 2y’ + 2y = 4cost + 2sint, y(0) = 0, y'(0) =3.

16. A mass of 3 slug is attached to a spring with spring constant k = 6 Ib/ft. A damping
force numerically equal to twice the instantaneous velocity acts on the system.
(a) Find the steady-state response of the system due to an external driving force
F(t) = 40sin2t.
(b) Will Riax occur?
(c) What is the amplitude in this case?

17. Find the steady-state response of the system in problem 16 if the driving force is
constant; i.e., F(t) = P.
“18. The ratio of successive maximum amplitudes of a particular underdamped spring-
mass system is found to be 1.25 when the system undergoes free motion. If k = 100
N/m, m = 4 kg, and a driving force of F(t) = 10cos 4t is imposed on the system,
determine the amplitude of the steady-state motion.
19. Show that dR/dw = 0, where R is defined by (31), occurs when the frequency w
satisfies w? = we — c7/2m?.
20. Derive Equation (34).

5.4 SIMPLE ELECTRIC CIRCUITS

Let us consider an electric circuit composed of a resistance R, capacitance C, and

inductance L, connected in series with a voltage source E(t), as shown in Figure
5.11. When the switch is closed at time t = 0, a current i = i(t) will flow in the
loop. The network featured here is called an RLC circuit. The current is determined
at each point in the network by solving appropriate differential equations that result
from applying Kirchhoff’s laws:

R E(t)

Figure 5.11 RLC circuit.

1. The sum of the currents into (or away from) any point is zero.
SEC. 5.4 / SIMPLE ELECTRICAL CIRCUITS 147

2. The sum of the instantaneous voltage drops in a specified direction is zero

around any closed path.

. The first Kirchhoff law indicates that the current is the same throughout the
circuit. To apply the second law, we must know the voltage drop across each of the
idealized elements in the RLC circuit. From experimental observations, we have
voltage drop across a resistor = Ri, f
di
voltage drop across an inductor = Le
t

voltage drop across a capacitor = a

where q denotes the electric charge on the capacitor and is related to the current i
by i = dq/dt. The impressed electromotive force E(t) contributes to a voltage gain.
Applying the second Kirchhoff law to the circuit shown leads to the differential
equation
z
ts +Ri t+ Cg =E(). (35)
In terms of the charge q, we find
] La’ +Rq' + Cg =EQ@, t>0/ (36)
We recognize that (36) is the same equation in form as that which governs the
damped motions of a spring-mass system. In fact, we notice the following analogies
between mechanical and electrical systems.

Charge q corresponds to position y.

Current i corresponds to velocity y’.
Inductance L corresponds to mass m.
. Resistance R corresponds to damping constant c.
. Inverse capacitance C~' corresponds to spring constant k.
AKRWN
. Electromotive force E(t) corresponds to input function F(t).

Such analogies between mechanical and electrical systems prove very useful in
practice. For example, in studying a certain mechanical system that is either too
complicated or too expensive to build, the electrical counterpart is often constructed
instead for the purpose of analysis. Interestingly, the phenomenon of resonance
also occurs in electrical systems, but it does not have the undesired side effects of
mechanical resonance. Quite the contrary, it is primarily because of electrical
resonance that we can tune a radio to the frequency of the transmitting radio station
in order to obtain reception.
The units most commonly used are listed in Table 5.2
148 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

Table 5.2 Common System of Units and Their Abbreviations

emf or voltage (E) volt (V)

charge (q) coulomb
current (i) ampere (amp)
resistance (R) ohm (Q2)
capacitance (C) farad (F)
inductance (L) henry (H)

ee ee eee ee ee

EXAMPLE 7 The circuit shown in Figure 5.11 contains the components L = 1H, R = 10000,
andC = 4 x 10°°F. Attimet = 0, both current and charge are zero, and a battery
supplying a constant voltage of 24 V is instantaneously switched on. Find the
charge q(t) on the capacitor and the current i(t) for any later time.

Solution The DE to be solved is given by

q’ + 1000q’ + :x 10°q = 24.

The associated homogeneous equation has the general solution

qu(t) = (Cy + Cyt)e™.

To find a particular solution gp, we substitute gp = A into the nonhomogeneous
DE, getting
A =9.6 X10°.
Therefore, since g = gp + qu, we find
Git pO Oexal OS Gre Cote.
and differentiating gives
git) =Hi(e=o S00C e+ G0 5001
)e
The prescribed initial conditions q(0) = 0 and i(0) = 0 lead to

0= 9.6 X 10° + C;, 0 = —500C;+ Co.

Hence, C; = — 9.6 X 10° and C; = — 4.8 X 10-7, from which we deduce
q(t) = 9.6 X 10°71 — e7) — 4.8 X 10-7te
> Goulombs
and
i(t) = 24te °°” amp.
SEC. 5.4 / SIMPLE ELECTRICAL CIRCUITS 149

EXERCISES 5.4

If the resistance R is not included in the RLC circuit of Figure 5.11, we have what is
called an LC circuit. Show that the current i(f) of the LC circuit satisfies the relation
5 E(O) (0)
es a: LC # 0,
where q(Q) is the charge in the capacitor at time t = 0.
Using problem 1, find the current in the LC circuit when i(0) = 0, g(0) = 0, and
(a) L =1H,C =0.25 F, E(t) = 30sint V.
(b) L=10H,C =0.1 F, E() = 10t V.
A steady-state current in the RLC circuit results after a sufficient length of time
(t—). Find the steady-state current where
(a) R=40,L =1H,C =2~x 10%F,
E(@) = 220 V.
(b) R=100,L =2H,C =0.5 F, E(t) = 10.9cos2r V.
Show that the current i(f) in the RLC circuit satisfies
IN eeE(O) _ ed
R q(0)
ee
i'(O) L Oo) LC” LC
# 0,

‘ where q(0) is the charge in the capacitor at time t = 0.

Find the current in the RLC circuit assuming i(0) = q(0) = 0, and
(a) R=60,L =1H,C =0.04 F, E(t) = 24cos
St V.
(b) R=800, L = 20H, C = 0.01 F, E(t)= 100 V.
What conditions on the circuit parameters R, C, and L must be satisfied for the charge
variation to be (see Section 5.2.3)
(a) underdamped?
(b) overdamped?
(c) critically damped?
A series RLC circuit has components L = 4 H, R = 10 0, C = 10° F, and
E(t) = 150 V. Determine the instantaneous charge on the capacitor for t > 0 if
initially g(0) = 1 and i(0) = 0. What charge persists after a long period of time?
Anelectrical circuit has components L = 10°*H, C = 2 x 10°° F, and a resistor R.
Determine the critical resistance necessary to lead to an oscillatory current if the
elements are connected in series.
a9: The amplitudes of two successive maximum currents in a series circuit with L = 1052
H and C = 10° F are measured to be 0.2 amp and 0.01 amp. Determine the
resistance R.
10. A particular RLC circuit connected in series has an electromotive force given by
E(t) = Eosinat.
(a) Show that the steady-state current (as t—>) is
150 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

i(t) = Eo aor = SE
vA i
where X = Lw — 1/Cwand Z = VX? + R’. The quantityX is called the reac-
tance of the circuit, and Z is the impedance of the circuit.
(b) Show that when
1
o-=— =,
VEC
the amplitude of the steady-state current is a maximum. Electrical resonance is
said to occur for this value of w.

[O] 5.5 THE METHOD OF GREEN’S FUNCTIONS

We now wish to develop a general solution technique applicable to initial value

problems involving higher-order DEs. The method of attack is attributed to George
Green* and is based upon the construction of a particular function known as the
Green’s function.
Since most of the DEs that commonly occur in practice are of the second order,
we will develop the theory for these equations and generalize the results to order
n in the exercises. Thus we will confine our attention to the initial value problem

My] =f, t >t, y(t) =k, y'(t) =k, (37)

where M is the normal second-order differential operator
M =D?*+ aD + a(t). (38)
For solution purposes it is convenient to separate (37) into two simpler problems:

Mly]=0, yt) = ko, y'(to) = ki, (39)

and

Mly]=f@), yo) =0, ym) = 0. (40)

The solution of (39), which we denote by y,,, physically represents the free response
of the system described by (37) entirely due to the initial conditions. Therefore, the
solution of (40) can be interpreted as the response of the system which is at rest until
time t = %, when it is subjected to the external disturbance f(t). This latter
solution function is denoted by yp, and the sum y = yp + yy then constitutes the
solution of (37) (see problems 46 and 47 in this section).
Let the general solution of M[y] = 0 in (39) be denoted by

Yo = Ciy\(t) + Cryr(t). (41)

*GEORGE GREEN (1793-1841) gained recognition for his important works concerning the reflection and
refraction of sound and light waves. He also extended the work of Poisson in the theory of electricity
and magnetism.
 SEC. 5.5 / THE METHOD OF GREEN'S FUNCTION 151

Imposing the prescribed initial conditions on (41), we get

Ciyi(to) + Cry2(to) = ko,

Ciyi(to) + Cry2(to) = ki,
and using Cramer’s rule leads to the unique determination
ko ya(to)
ky y2(to)
yilto) alto) W(y1, y2) (to)
Yi(to) y2(to)
(42)
yilto) ko
yilto) ky kyyi(to) — Koy i(to)
Cc = ————- =
; Yillo) yo(to) W()1, y2) (to)

yilto) y3(to)

Remark. The explicit representation for the constants C, and C; as given by (42)
is mostly of theoretical importance. In practice, it is usually just as convenient to
solve for these constants directly, as illustrated in Example 8 below.
Notice that when ky = k,; = 0, we get C; = C, = 0, so that necessarily yy = 0.
The physical implication of this result is that a system which is initially at rest and
not subject to any external disturbance must remain at rest.

Theorem 5.2 | The homogeneous initial value problem

Mly] ca 0, y(t) = 0, y (to) = 0,

has only the trivial solution yy = 0.*

ee ee 00ST
eg

EXAMPLE 8 Solvey’+y=0, y(O)=1, y(O)=—l.

Solution The general solution of the DE is

yy = C,cost + C)sint.

Imposing the initial conditions, we find

(ONC Cy 0 = Ce
y'(0) = —C,- 0+ C, = Co,
and deduce that C; = 1 and C, = —1. Thus, our solution is
yy = cost — sint.
a
Oe ee ee

*Theorem 5.2 is a repeat of Corollary 5.1 for the case of second-order DEs.
152 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

5.5.1. The One-Sided Green’s Function

In order to solve the problem described by (40), we will start by assuming yp has
the form
yp = u(t)y\(t) + v(y2(t) (43)
for some functions u(t) and v(t). Using the method of variation of parameters as
discussed in Section 4.7, we find that

yf (0) yilOf(0)
u(t) = —
TENG ee lao
+. 7 ; = pe LEN TRA OFM aS

where W(y;, Y2) = yiy3 — yiy2 is the Wronskian function. If we choose u(t) and
v(t) as any indefinite integrals as indicated above, the resulting solution function
yp is unlikely to satisfy the prescribed homogeneous initial conditions. A proper
choice turns out to be

t y(T)f(7) t y(T)f (7)

=—| ————dr, v() = | =~ dz, 44
uM) to W(y1, Y2)(T) el to W(y1, Ya(T) y of
and substituting these expressions back into (43) leads to

t ya T)f(7) t y(t
)f(7)
Meet
=-
Cah a
ORR eo kicckse
Te
olaile
+
Aa
which we choose to write as

ye =| gilt, f(a) dr, (45)

where

yi(t) y2(T)
yi(t)ya(t) — yidy2(7) y(t) y(t)
Oa 46
cee W(y1, y2)(7) y(t) y(7) “
yi(t) ya(t)
To show that (45) satisfies the homogeneous initial conditions in (40), first
observe that when t = fo, we have

Yetta) = |"gulto, F(z) dr = 0.

to

Next, using the Leibniz formula (from calculus)

ang t OF
al
aes F
Gonliai
b
igs6 =
[ es5p m2 t) ae oe AC ae

we find

19 02
yp (to) = Ba (to, TIF(T) dt + gi(to, tof (to),
t9

which also is zero since g)(t, to) = 0 by definition.

 SEC. 5.5 / THE METHOD OF GREEN'S FUNCTION 153

The function g,(t, 7) is called the one-sided Green’s function for the initial value
problem described by (37). Its construction depends only upon knowledge of the
homogeneous solutions y(t) and y2(f); i.e., it is independent of to and the prescribed
initial conditions and is completely determined by the operator M = D? +
a,(t)D ar a(t).

Remark. It is important to observe that the particular solution (45) was derived
under the assumption that the DE was in normal form. Strict adherence to this form
is necessary for proper identification of the forcing function f(t).

The original initial value problem

M[y]=f(t), y(to) =ko, y'(to.) = ki,

has the solution y = yp + yy, which can now be expressed as

y = [ sie. WG) dr + Cy. + Corl,

t0
(47)
where the constants C; and C, are defined by (42).

Remark. Although (47) represents a general solution formula for the given initial
value problem, it does not always represent the simplest approach to finding the
solution. The Green’s function method is important for developing and under-
standing some of the general theory, and it can be useful in those situations where
the same DE must be solved a number of times with various input functions. It is
the responsibility of the practitioner to determine those occasions for which such
general formulas are useful.

EXAMPLE 9 Determine the one-sided Green’s function for the operator M = D? + 1.

Solution Linearly independent solutions of M[y] = 0 are simply y\(t) = cos ¢ and y2(t) =
sin t. The Wronskian is
COS.t ssinft)P
W(y1, y2)(t) = |
—sint cost
?

and hence, from (46), it follows that

Gapl= COS 7, SINLF = sintcos7 — costsin7 = sin (¢t — 7).
i’; cost sint
en er

With g,(t, 7) determined for the operator M =D?’ + 1 as given in Example 9,

we are now in position to solve all DEs of the form y" + y = f(t), a fact that clearly
illustrates the power and economy of using the Green’s function.
oe a eee ee ari ee

EXAMPLE10_ Using the one-sided Green’s function, solve the initial value problem
154 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

yay isin’ty) «y(0) = 1p (ON aa I:

Solution From Example 9, we have g,(t,7) = sin (t — 7). Thus,

yp i)sin (4 — 7) sin 7 dt
0

t t A
li sin | cos T sin T dt — cos ¢ | sin? t dt
0 0

= Lets ir =it COS).

2,
From Example 8, it has been determined that y, = cos ft — sin ft, and hence,
Lee :
Nae ty SI Teg COS) CO tel

= eee bot
2 2 :

EXAMPLE 11 Solve the initial value problem

t?y” — 3ty’ + 3y = 2t*e’, y(1)=0, y'(1) =2.
Solution This is a Cauchy-Euler equation. Using the method of Section 4.8, we find the
general solution of the associated homogeneous DE to be
») Hapa Gite Cate

Subjecting this solution to the initial conditions, we find

Get Gss—0;
Cy 3G, =;
from which it follows that C, = —C,; = 1. Hence,

yy = 0? —-t.
The Wronskian of y, = ft and y) = f° is
raat
W(t, ?) = = 273,
338
and therefore the one-sided Green’s function is given by
3

gilt, T) =
ihe a Mee earn is
Noms Pot 27° ;
Dividing the DE by the leading coefficient r* puts the equation in normal form, from
which we identify f(t) = 2r’e'. Therefore,

yp = |ets \2s%e)das
1

*Note that the lower limit of integration is | since the initial conditions are prescribed there.
SEC. 5.5 / THE METHOD OF GREEN'S FUNCTION 155

ty t

| et dr ~ 1| te’ dt
1 1

(t — Pe + 2t(t — Ne’.
Finally, combining solutions, we have

y =yp + yy =t(e — 1) + (1 — e) + 2tt — Net.

It should be observed that one of the distinct features of the Green’s function
technique is that the nonhomogeneous initial conditions are imposed only upon the
solution of the associated homogeneous DE. This is in sharp contrast with the
methods employed in Chapter 4, wherein it was first necessary to find a general
solution of the nonhomogeneous DE before applying the prescribed initial condi-
tions. The reason for this situation, of course, is that we are selecting a particular
solution yp that always satisfies homogeneous initial conditions.

5.5.2 A Table of Some One-Sided Green’s Functions

A listing of one-sided Green’s functions for some of the more common differential
operators is provided in Table 5.3 for easy reference.

Table 5.3. Table of One-Sided Green’s Functions

Le 1B Peg

(t — 7)"|
Ga = i

1 5
2
8. Ds 2
tp p sin b(t — 7)

4. D? — b? +sinh b(t — 7)

I alm). b(t—7)
5: (D
=
—a)(D
=
—b),a#b eh se e

6. (D — a) (<7) cone

n = (OS tT) a(t—7)

ds (Da) ane Ie 2 3ae se Tere

8. D? — 2aD + a? +b? en sin b(t — 7)

D*— 2aD + a* — b* ren? sinh b(t — 7)

?D?+tD —b?
156 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

EXERCISES 5.5

In problems 1-8, find the solution satisfying the prescribed initial conditions.

yoy = 0, yO) S07 ry. C=

y= y= 0,7 7 0) = 0, ay) al
ye Ayles syursc0) ay (O)=en Le ry(0) eas
yy aye On y(0) 2am wy(0) seg ae
y cat jy ee On 6V(O) 9250), (0) nay) aed
Py" + 4ty' +2y =0, y=1, y'()=2
[yu sty) -4y=0, yl) = yey (y= |
Oe Py et ty
ee
re
SOS ay Oe ey(Di= 2, ly(8
In problems 9-18, determine the one-sided Green’sfunction for the given operator.

9. M =D? 10. M=D?+5

Wo M=pD*=5 12. M=D*+4D+4
13. M=4D’?-8D+5 14. M=D’?-D-2
15. M=2?D?+
WD — 16 16. M=?D*?-1tD*+1
"7. M =D(i—¢*)D “18. M = 1D? — (1 + 2t7)D

In problems 19-24, use the one-sided Green’s function to solve the given initial
value problem.

19. y"=1, y(@)=0, y'(0)=0

20. y"=e', yoO)=2, y(0)=-1
21. y"-y=1, yO)=0, y’O=1
22 ye. 3. YUH On y= 0
\(28) y” — 3y' — 4y Sy. Vis y (2 0

24. y" +y = 2 csc tcot t, (3) = ily. y'(3) |

25. Show that the one-sided Green’s function associated with the undamped spring-mass
system described by
my" + ky = F(t), y(O) =yo, y'(O) = vo,

feo kK
gi(t, T) =— sin[wol(t — 7)], wo= ve
Wo m

26. Using the one-sided Green’s function given in problem 25, find the response of the
spring-mass system described there when the system is initially at rest and subject to
the forcing function
(a) F(t) =P (constant).
 SEC. 5.5 / THE METHOD OF GREEN'S FUNCTION
157

(b) F(t)=Pcos
wt, w # a.
(c) F(t)
=P cos wot.
wai Show that the one-sided Green’s function associated with the damped spring-mass
system described by

my" + cy’ + ky = F(t), yO) =yo, y'() = to,

for each of the three cases of damping is as follows:

(a) Underdamped: g\(t, T) = ~ ~~ 72m sin [u(t — 7)], where

KM = (dk
=62)am :

(b) Critically damped: g\(t, T) = (t — r)e “0.

(c) Overdamped: g\(t, T) = +e~e— 92m sinh [a(t — 7)], where

ae (c? — 4mk)'?
2m ;
*28. Using the one-sided Green’s function given in problem 27(a), (b), and (c), find the
steady-state response of the system in all three cases when the driving force is

(a) F(t) =P (constant). (b) F(t) = P cos at.

eat) elseer <<a7 aa) _ | Sir, OS <a
© FO= {9 py @Fo={ OMS a.
In problems 29-32, solve the Cauchy-Euler initial value problems using the method
of Green’s function.

29. iy tye Sy i. ay (0. ay <1) =O)

30. t?y” — Sty’ + 8y = 2t7, y(-—2)=1, y’(-2)=7

31. t’y"—6y
”
=logt, y)=—2,l y()=
'
areI
32. t?y’ + ty’ + 4y =sin(logt), y(1)=1, y'(1)=0
33. Show that for all 7 = fo,
(a) gi(t, 7) = 0.
(b) gi(t, 7), along with its first and second derivatives with respect to ft, is con-
tinuous.
34. Show that for a fixed value of 7, the function @(t) = g,(t, 7) is a solution of the initial
value problem
y’ + a(t)y’ + a(t)y = 0, (7) =0, O'(7) = 1.
Hint: Use the result of problem 33.

*35. Consider the third-order initial value problem

y" + anlt)y" + a(t)y’ + ao(t)y = f(t),
y(to) = ko, y' (to) =ki, y"(to) = ke.
158 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

(a) Show that the one-sided Green’s function is defined by

yi(t) y(t) ys(7)

yi(t) y2(t) y3(7)
yi(t) y2(t) s(t)
GNC ee
ne ee Ee
W(y1, y2, y3)(7)

where y:, y2, and y3 are linearly independent solutions of the associated
homogeneous DE.
(b) Establish the solution formula

y= irgilt, Df (adr + Ciyi(t) + Cryr(t) + Csys(O),

and derive determinant expressions for C,, C2, and C3 similar to those in (42).
(c) Generalize the results in (a) and (b) to nth-order problems.

In problems 36-41, find the one-sided Green’s function for the given operator
using the result of problem 35.

36. M =D", n=2,3,4,... 37. M =D*(D2—-1)

38. M = D(D? + 4) *39. M =D? +3D?-3
*40. M=D*-1 *A1. M =D?—-
6D’ + 11D -6
*42. Show that the one-sided Green’s function associated with an nth-order normal linear
operator and its first n — 2 derivatives with respect to t vanish when t = 7, but that
the n — 1 derivative equals unity when ¢t = 7.
*43. Show that for a fixed value of 7, the function #(t) = gi(t, 7) is a solution of the initial
value problem
M[¢]=0, t>7r
(7).
= 0,6 b'7)
= 0,8 = ie Ps
(7) = OO eG)ale
where M is an nth-order normal linear differential operator.
Hint: Use the result of problem 42.

Solve the initial value problems 44 and 45 using the one-sided Green’ sfunction (see
problem 35).

*44, y”+y m
=te’, yd)=0, y')=0, y")=1
*45. yr ye tay oe dye 1 meyO) c= Ons ye Okan Umm (Olea
46. Prove that M = D? + a,(t)D + ao(t) is a linear operator; 1.e., that
M[C\y; + C2y2] = CiM[yi] + CoM[y2]
for any constants C,; and C).

47. If yp satisfies the initial value problem

M[y]=f(t), y(t) =90, y'(to) =0
 SEC. 5.6 / IMPULSE FUNCTIONS
159

and yy satisfies the initial value problem

Mly]=0, y(t) = ko, y'(to) =k,

show that the sum y = yp + yy satisfies

Mly]= f(t), y(to) =k, y'(to) = hk.

[O] 5.6 IMPULSE FUNCTIONS

In certain applications it is convenient to introduce the concept of an impulse

function, which is the result of a sudden excitation administered to a system, such
as a sharp blow or a voltage surge. Let us imagine that the sudden excitation, which
we will denote by d,(t), has a nonzero value over the short interval of time
a—eé<t<a+e, but is otherwise zero. The total impulse (force times du-
ration) imparted to the system is thus defined by

Le i FCawili i. d,() dt (€ > 0). (48)

The value of J is a measure of the strength of the sudden excitation.
In order to provide a mathematical model of the function d,(t), it is convenient
to think of it as having a constant value over the intervala — e <t <a + €(see
Figure 5.12). Furthermore, we wish to choose this constant value in such a way that
the total impulse given by (48) is unity. Hence, we write
1
aE
i <a t €
d,(t)= 3 2e’
0, otherwise.
(49)

Figure 5.12 Impulse function.

Now let us idealize the function d,(t) by requiring it to act over shorter and shorter
intervals of time by allowing e—0. Although the interval about ¢ = a is shrinking
to zero, we still want J = 1; thus it follows that

limi =lim [ d,(t) dt = 1. (50)

e>0 e>0 J-*
160 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

We can use the results of this limit process to define an idealized unit impulse
function, 6(t — a), which has the property of imparting a unit impulse to the
system at time ¢ = a but being zero for all other values of t. The defining properties
of this function are therefore
Of — a) =O tas

ie S(t — a)dt = 1. (51)

By a similar kind of limit process, it is possible to define the integral of a product

of the unit impulse function and any continuous and bounded function f1.e3

ie 8(t — a(t) dt = f(a). (52)

The verification of this result, which makes use of the mean value theorem for
integrals, is left to the exercises (see problem 8).
Obviously the “function” 5(t — a), also known as the Dirac delta function,* is
not a function in the usual sense of the word. It has significance only as part of an
integrand. It is an example of what are commonly called generalized functions .In
dealing with these functions, it is usually best to avoid the idea of assigning
“functional values” and instead refer to its integral property (52), even though it has
no meaning as an ordinary integral. Following more rigorous lines, generalized
functions can be defined as a limit of an infinite sequence of well-behaved functions
(see problems 14 and 15 in this section).

EXAMPLE 12 . Solve the initial value problem

y"+y =6( —7), t >0, y(O)=0, y'(@) =O.

Solution Because of the homogeneous initial conditions, we only need consider the particu-
lar solution; i.e., y = yp. The one-sided Green’s function for the operator
M = D* + I has previously been shown to be (see Example 9)

Oi sae slll (ba 27),

and thus it follows that

y= [sin ~ 1) 8(r~ mar =|Sint

0
Oy PS op
("ane t= a7.

where we have made use of (52).

We can interpret this solution as the response of some system that remains at rest
until time ¢ = a, when it is subjected to a unit impulse. After time t = 7, the
response of the system follows that of a simple sinusoid with frequency equal to that
of the natural frequency of the system (see Figure 5.13).

*Named after PAUL A. M. DirAc (1902-), who was awarded the Nobel Prize (with E. SCHRODINGER)
in 1933 for his work in quantum mechanics.
SEC. 5.6 / IMPULSE FUNCTIONS
161

Figure 5.13

a ee

5.6.1 A More General Definition of the Green’s Function

The unit impulse function is also useful for providing us with a physical inter-
pretation of the one-sided Green’s function. To see this, let us consider the initial
value problem

My] = 6 — a), y(t) =0, y'@) =0. (53)

The solution of (53) is formally given by
0, iy) So SE
y = git, 1) (r~a) dr =|
10 git, a), t = 0), oy

where g(t, 7) is the one-sided Green’s function associated with the operator M.
Thus the function g(t, a) must represent the response of the system described by
(53) for t > a, which was formerly at rest and then subjected to a unit disturbance
(impulse) at time ¢ = a.
Based on the above interpretation of g)(t, 7), let us introduce the more general
function

0, i St Se
co Una he
gi(t, as 7 Sy Se,

or equivalently,

si.) hes={ Ci(P Aete otetotoeoToa

att
(55)
so that we can express the response of the system to a general input function f as

y= i g(t, Tif (t) dr, (56)

where the integration now takes place over all 7 = f. The function g(t, 7) can then
be interpreted as the response of the system for all time ft due to a unit impulse
delivered at time t = 7. We will refer to g(t, 7) as simply the Green’s function,
or influence function as it is sometimes called.*

*The influence function in some textbooks is defined as a constant multiple of g(t, tT) rather than g(t, +)
itself.
162 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

For a fixed value of r, we see that g(t, 7) must necessarily satisfy the differential
equation M[g] = 5(t — 7), where M is the differential operator associated with the
construction of g\(t, 7). The Green’s function must also satisfy the homogeneous
initial conditions g(t, T) = (dg/dt)(t, T) = O for any % less than or equal to the
fixed value of 7. It is a continuous function for all values of t since g\(7, T) = 0
and g\(t, 7) is itself a continuous function of t by definition. However, there is a
jump discontinuity in the first derivative of g(t, 7) att = Tof unit magnitude. That
is, for t < 7, the derivative is clearly zero, while

0g _ yilt)y2(7) = yiryal7) _ ,
Ot |r=r+ W(y1,¥2)(T)
The jump discontinuity in the derivative turns out to be an essential feature of the
Green’s function.
In summary, we have the following definition.

Definition 5.1 The Green’s function g(t, T) associated with the nonhomogeneous initial value
problem
M[yl=f(t), t >t, y(t) =k, y'(o) =k,
is a function satisfying the following conditions:

(a) M[g]= 6(¢ — 7) (rfixed),

(6) eto, 7) = B(wp, 7) = 0,
(c) g(t*, tT) = g(t, T) (continuous function),
0g t=rt
(d) Fe = 1.

EXERCISES 5.6

In problems 1-5, find the solution of the given initial value problem.

yy) ly OF — yey (ON oe eC) ea

y"+y = 6(t — 7) - 6(t — 27), y(O)=0, y'(0)=1
y"+y =6(t-—7m)+Acosat, y0)=0, y'0)=0 (w# 1)
y y= 260 —1), yO)
= 1) 40) ==1

y"+y= Aa( = 5)sint, y(0)=0, y'(0)=2

ee
OR A particular spring-mass system has spring constant k = 72 N/m, m = 2 kg, and
c = 40kg/s. If the system is at rest and at time t = 0 given a sharp blow of magnitude
100 N, what is the subsequent motion?
Hint: Set F(t) = 100 6(t).
 SEC. 5.6 / IMPULSE FUNCTIONS 163

Te Solve the initial value problem

y"+y=e'+ 6¢-1), yO=1, y'O)=1, yO) =2.
Show that [- 8(t — a)f(t) dt = f(a).
Hint: The mean value theorem of the integral calculus states that

[ fe) dx = Ob ~ a)
b

for some € in the interval a < & < b. Thus,

a [dingo dt eet
lim = lim 52 is ery
f(t) dt ke= lim
ye 5- l+
east f(€) >2¢,
whereas
€1— 6 <daiet €.

Show formally that dh(t)/dt = 6(t), where h(t) is the Heaviside unit function defined
by
Ona)
nia)= | be,
Hint: Use integration by parts to show that

carat
a CFO) dt = FO).
10. Show that 6(¢ — x) = 6( — f).

ah Show formally that f(t) 6(¢ — a) = f(a) 6(t — a), and use this result to deduce that
t d(t) = 0.

Hint: Show that thy e(t)if(t) 6(t — a)] dt = iL e(t)if(a) d(t — a)] dt.

12. If a> 0, show that 2 S(at)f(t) dt = a~'f (0).

13. Show formally (using integration by parts) that

@) fo spi dr= -7'O.

(-1)"f(0).
(b) ig 6(t)f (t) dt =

*14, Consider the sequence of rectangle functions defined by

n l

Wp (t) =
ppg ea 1

jor tu =sls 2. eee

(a) Show that for each n the area enclosed by the rectangle is unity, and deduce that

lim | (dt = 1.
164 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

V,

n
2
t
Leo
n =e

Problem 14

(b) More generally, if f(t) is any function continuous at t = 0 and bounded, show
that

lim ib Un (tf (t)dt = f(0).

Hint: See problem 8.
15. Any sequence of continuous and differentiable functions (1), wo(t), . . -,Un(t), . - -
satisfying the condition of problem 14(a) is called a delta sequence. Show that the
following sequences are delta sequences:

(a) Y(t)= ee
LD PF Waal eu
aw(1 + n°t?)’

(b) Un) = ae" i ie ache

2 72

Hint: ip edt = Vr.

—oo

[O] 5.7 ELEMENTARY OSCILLATION THEORY

We now turn our attention to equations more general than the constant-coefficient
DEs arising in most spring-mass systems or electric circuits. In particular, we want
to discuss the DE
A2(t)y” + Ai(t)y’ + A,(t)y = 0, (57)
in which certain parameters of the system vary over time. Although in general we
cannot obtain explicit solutions of (57), we can study the qualitative behavior of
these solutions by directly analyzing the equation itself. For example, information
concerning the existence and relative positions of the zero-crossings can be ob-
tained without formal solutions. In the theorems that follow, we will treat the
coefficients of (57) as continuous functions on the interval of interest and assume
A,(t) # 0 on this interval.

Theorem 5.3 (Sturm separation theorem) If y, and y, are linearly independent solutions of
the second-order equation

Ax(t)y" + Ady!’ + ApOy = 0

SEC. 5.7 / ELEMENTARY OSCILLATION THEORY 165

on some interval I, then y, has precisely one zero between any two consecutive
zeros of y2 on the interval /.

Proof : Lett = aandt = b be consecutive zeros of y) (Figure 5.14). The linear

independence of y, and y, implies that the Wronskian W(y,, y2)(t) does not vanish
on the interval a = t = b. And since y,(a) = y(b) = 0, it follows that

W(yi, y2)(@) = yila)yz(a) # 0,

W(y1, y2)(b) = y\(b)y5(b) # 0.

Figure 5.14

The Wronskian is clearly continuous on a St S b, and since it does not vanish,

it must have the same sign at all points in the interval, and in particular, the same
sign at t = a and t = b. On the other hand, since a and b are consecutive zeros
of y>, then y; must have opposite signs at t = a and t = b. In order to prevent the
Wronskian from changing signs, y(a) and y,(b) must have opposite signs, and the
continuity of y, implies that it must assume a zero value at least once between t = a
and t = b.
If we reverse the roles of y, and y in the preceding argument, we conclude that
y2 must have at least one zero between consecutive zeros of y,. Hence, y,; cannot
vanish more than once between t = a and t = b. LJ

Put another way, Theorem 5.3 states that the zeros of y, and y, occur alternately,
and that on any finite interval the number of zeros of y, and y, can differ by at most
one. For example, the solutions cost and sint of y” + y = 0 have zeros that
alternate on the entire real axis. A somewhat less obvious consequence is that any
two functions of the form
y, = C,cost + Cysint, y, = C3cost + C7 sin t

have alternating zeros whenever C\C, # C>C;, as all such pairs of functions are
also linearly independent.
The Sturm separation theorem does not imply the existence of any zeros of the
solutions y, and y). Furthermore, it does not allow for a comparison of the number
of zeros (rate of oscillation) of solutions of two distinct DEs. To pursue these
matters it becomes convenient to put the differential equation in what we call the
standard form.
166 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

If we divide (57) by the leading coefficient A,(t), the resulting equation is

Var at) pore) yess (58)

which is in normal form. Next we put y = u(f)v(t) so that y’ = uv’ + u'v and
y" = uv" + 2u'v' + u"v; when these expressions are substituted into (58), we get
vu" + (2v' + avju' + (v" + av’ + bu)yu = 0. OF
The coefficient of u' can be made to vanish by selecting

u(t) = exp ISiat) (60)

and hence (59) reduces to

u"+ O(pu = 0, (61)

where

01 = by — Flat — 5a’. (62)

We will refer to (61) as the standard form.
Because the function v cannot vanish, the transformation y = uv above can have
no effect on the zeros of y, and hence leaves unaltered the oscillation phenomena
we are investigating.

EXAMPLE 13 Find the standard form of Bessel’s equation

eceaeha cid ead 235, II S

Solution We first rewrite the equation as

0 ETON (ipa Yaa

By. 4 t Ve SINe

which identifies a(t) = 1/t and b(t) = 1 — v?/t?. Hence

ONC M te Perel 1
sen ea Stent loz 4y2
=
and Bessel’s equation reduces to

t+ (14259) y= 08

We can now show that if Q(t) is negative on a given interval, the solutions of
(61) do not oscillate on that interval. To better understand this situation, consider
the solutions of y” — y = 0.

*Although we don’t need it, we see that v(t) = ¢~"/? in this case.
 SEC. 5.7 / ELEMENTARY OSCILLATION THEORY
167

Theorem 5.4 If Q(t) < 0 on the interval J and uw is a nontrivial solution of u” + Q(t)u = O, then
u has at most one zero on the interval /.

Proof: Let t =a be a point on the interval such that u(a) = 0. Since u is a

nontrivial solution, it follows from Theorem 5.2 that u'(a) # 0. Let us assume that
u'(a) > 0 so that u is positive as rt increases from t = a. Since Q(t) < 0, then
u” = —Q(t)u must be positive as t increases. And since u” is the rate at which the
slope of the function u changes, we conclude that the slope is increasing, and hence
u cannot have a zero for t > a. A similar argument can show that u has no zero for
t <a, and the proof is also the same for u'(a) < 0.

Theorem 5.5 (Sturm comparison theorem) Let u, and u2 represent nontrivial solutions of

u"+ QO(tu =0, u” + Q(hu = 0,

respectively, and suppose that Q,(t) > Q,(t) everywhere on the interval of interest.~
Then there exists at least one zero of u, between every two consecutive zeros of up.

Proof. Let t = aandt = b be consecutive zeros of uz, and assume that u, does
not vanish on the interval a < t < b. We will further assume that both u, and uw,
are positive on the interval a < t < b. Since the zeros of u, are the same as —u,
this assumption is justified. Following the argument presented in the proof of
Theorem 5.3, we have

W(u1, ur)(a) = uj(ajus(a), W(u;, u2)(b) = u,(b)uz(d).

However,

d d ; '
qu u2)(t)] = Gy i — Uj Up)

= WU — UjUy
—UjQouz + U,Q {Uy
u,U(Q; — Qo),

and since Q,(t) > Q(t), we conclude that the Wronskian is an increasing function
on a <t <b. Also, since we have assumed that u, is positive on this interval, it
follows that u5(a) = O and w3(b) S 0, and hence

W(uy, u2)(a) 20, W(u;, u2)(b) S 0.

But this condition implies that the Wronskian cannot be an increasing function.
Thus, contrary to our assumption, we must conclude that u; vanishes at least once
on the intervala <t <b. LJ

An immediate consequence of the Sturm comparison theorem is that the larger

the coefficient of uin u” + QO(fu = 0, the more rapidly the solution oscillates. For
example, let us examine the solutions sin? and sin 2t of the two equations
168 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

y"+y=0, y'+4y =0.

The solution (sin 22) of the second equation oscillates more rapidly than the solution
(sin?) of the first equation. That is, the zeros of sin 2f are located at t = 0, 7/2)
a, 37/2, . . . , whereas the zeros of sint are located att = 0, 7, 27, . . . (see
Figure 5.15).

>

\ y=sint
\
N\
—IF

Figure 5.15 Graph of y = sint andy = sin2r.

EXAMPLE 14 Show that every solution of y” + t*y = 0 has infinitely many zeros on the interval
seca

Solution The equation y” + y = O has one solution sin with infinitely many zeros at tr= 0
and t =n, n = 1, 2, 3,.... Since t? > 1, it follows from the Sturm com-
parison theorem that one nontrivial solution of y" + t?y = 0 has at least one zero
between nar and (n + 1) for n = 1, 2,3, ... , and hence has infinitely many
zeros. By comparison with cost, it can be established that a second linearly
independent solution of y" + ¢*y = 0 has infinitely many zeros between the zeros
of cos ft. Since all other solutions of this DE are linear combinations of these two
linearly independent solutions, we have our intended result.

The zeros of Bessel’s equation are of both practical and theoretical importance,
and are found to satisfy the conditions of the following theorem by comparison with
the zeros of solutions of y” + y = 0.

Theorem 5.6 (Zeros of Bessel’s equation) Every nontrivial solution of Bessel’s equation, which
has the standard form

eee ( eee = 0,
4t-
has infinitely many zeros on the positive t-axis. Moreover, the distance between
suCCesSive Zeros is
SEC. 5.7 / ELEMENTARY OSCILLATION THEORY 169

(a) less than 7 forO Sv < ?

(b) equal to 7 for vy = 3, an ee

(c) greater than a for v > 5.

Remark. Interestingly, the zeros of the solutions of Bessel’s equation coincide

with the zeros of the sinusoidal functions when v = 3. The reason for this situation
is that the solutions of Bessel’s equation are directly related to the sinusoidal
functions for this particular value of v.

EXERCISES 5.7

Prove that the zeros of any solution of the second-order DE

A2(t)y" + Aifi)y'’ + Ao(t)y = 0
are simple; i.e., if y(to) = 0, then y'(to) # 0.
Hint: Use Theorem 5.2.
Prove that if y,; and y2 are linearly independent solutions of
A2(t)y” + Ai(t)y’ + Ao(t)y = 0,
they cannot vanish at the same point.
Hint: Assume y\(a) = y2(a) = 0 at some point t = a, and show that this leads to a
contradiction that the Wronskian cannot vanish anywhere on the interval of interest.
Prove that between every pair of consecutive zeros of sint there is one zero of
sint + cost.
Show that the zeros of y, = sin(log?#) and y2 = cos (log #) alternate.
Hint: Find a DE for which y, and y2 are both solutions.
Explain how the Sturm separation theorem applies to y" — y = 0, if indeed it does.
Show that the substitution y = ¢ ‘uv transforms Bessel’s equation to standard form.
Show that every nontrivial solution of y” + (sinht)y = 0 has at most one zero in the
interval t < 0. What can be said about the interval t > 0?
How many zeros does every solution of Airy’s equation y” + ty = 0 have on the
interval t > 0?
“Gh Show that every nontrivial solution of y” — 2ty’ + ky = 0 has
(a) at most one zero when k S —1.

(b) only finitely many zeros when k > 0.

10. Which of the equations
=0 and
y"+(1+2?)y + 2ty = 0
y"
10?
has the most rapidly oscillating solution in the interval 0 = t <
y” + ty = Ois less
11. Show that the distance between consecutive zeros of a solution of
than 7 for at least t > 7.
of y” + t’y =0
12. Show that the distance between consecutive zeros of a solution
approaches zero as [>.
170 CHAP. 5 / APPLICATIONS INVOLVING INITIAL VALUE PROBLEMS

13. Putthe DE my” + cy’ + ky = 0 instandard form, and show that it does not possess
oscillatory solutions unless c* — 4mk <0.
14. Prove Theorem 5.6.

REFERENCES
Birkhoff, G., and Rota, G.-C. 1978. Ordinary differential equations. 3rd ed. New
York: Wiley.
Boyce, W. E., and DiPrima, R. C. 1977. Elementary differential equations and bound-
ary value problems. 3rd ed. New York: Wiley.
Bradbury, T. C. 1968. Theoretical mechanics. New York: Wiley.

Coddington, E. A. 1961. An introduction to ordinary differential equations. Englewood

Cliffs, N.J.: Prentice-Hall.

Finney, R. L., and Ostberg, D. R. 1976. Elementary differential equations with linear
algebra, Reading, Mass.: Addison-Wesley.
Potter, M. C. 1978. Mathematical methods in the physical sciences. Englewood Cliffs,
N.J.: Prentice-Hall.
Ross, S. L. 1980. Introduction to ordinary differential equations. 3rd ed. New York:
Wiley.
 rm isan e ficient method for s near, stant-coefficient -
ns with scribed auxiliary conditions, usually in the form of initial conditions.
It offers the advantage of solving the problem directly without first producing the
general solution or even having to solve separately for the homogeneous and
particular solutions.
The transform method is formally equivalent to the operational calculus devised
by Oliver Heaviside (1850-1925) for the solution of transient problems in physics
and electrical engineering. It is an especially useful tool for solving problems in
circuit analysis, where the nonhomogeneous terms are frequently of a discontinuous
or impulse nature. More conventional methods tend to be clumsy when piecing
together solutions valid in different intervals, whereas the Laplace transform has the
effect of “smoothing” the problem in the transform domain and making it more
tractable.
In Section 6.2 we calculate the transforms of some elementary functions directly
from the integral definition. We follow this in Section 6.3 with some of the operational
properties of the transform, which permit methods of calculating certain transforms
other than straight evaluations of the defining integral.
We discuss methods of computing inverse Laplace transforms in Section 6.4,
which rely mostly on operational properties and sometimes partial fraction
expansions.
in Section
Solutions of initial value problems by the transform method are studied
g some of the
6.5 and the vibrating spring-mass system is revisited, illustratin
benefits of a transform analysis.
impulse func-
The use of the Laplace transform in dealing with discontinuous and
d in Section 6.6, and in Section 6.7 the convoluti on theorem and its
tions is discusse
last section contains a
relation to the one-sided Green’s function are examined. The
short table of Laplace transforms.

171
 6.1 INTRODUCTION

The idea of transforming one function into another is commonplace in mathematics;

for example, the operation of differentiation transforms the function f into the
function f’. Another transformation that is prominent in the calculus is that of
integration.
An integral transform is a relation of the form

F(s) = [xo t)f(t) dt (1)

such that a given function ie() is

i SES into seme Eu pee
tf eu means

K(s, 0)
a ia

(2)

6.2 THE LAPLACE TRANSFORM OF SOME ELEMENTARY FUNCTIONS

The Laplace transform of many elementary functions can be obtained through

routine integration of the defining integral. Consider the following examples.

*Named in honor of the French mathematician PIERRE SIMON DE LAPLACE (1749-1827). Laplace made
use of this particular integral transform in his work in probability theory, although it is believed that
the integral was really discovered by Euler.

172
 SEC. 6.2 / THE LAPLACE TRANSFORM OF SOME ELEMENTARY FUNCTIONS 173

EXAMPLE 1 Let f(r) = 1, t = 0. Then

—st|
#1} = [edt = 0
e
eats
=+
S

provided that s > 0 so that e “— 0 as ¢ tends to infinity.*

Although the variable s may be real or complex, in the computation of Laplace

transforms we normally assume s is real and greater than some constant, as in
Example 1. Such a restriction on s, however, has little effect in applications.

EXAMPLE 2 Let f(s) =e”, 1 > 0. Then pera

L{e*} = [ ¢werat = [(eee Se
0 0 ie
In the case s S a, the integral diverges, while for s > a, we get
=(s= apt 4
e 1
LHe*}) = ———_|_ =/——_; ad
P(Oa-s CO) | eee Soa

EXAMPLE 3 Let f(t) = sint, t = 0. Then integration by parts yields

\
oO \>

Lif} = I e “sint dt
0
ey

=
= st
CmaCOSt — s| en costdt
0 0

ae CT COSdia Ss 1 0:
0

A second integration by parts now gives

FFM} =1— fe "sin eat

0

PS KSCAN ADE
and solving for S{f(t)}, we find

Lt fit} = stl s>0.

174 CHAP. 6 / THE LAPLACE TRANSFORM ag
_

Not all functions of t have a Laplace canst even if they are continuous,
since the defining infegral is1 imprOPee The basic requirement for the existence of
i
transform is that f be of «
the 1

Definition 6.1

a:see Figure 6.1). For instance, ihe functions 1, e”, and sin tareall Bf eteoucn
order, whereas the function f(t)= e” is not since its graph grows faster than any
linear power of e for t > fo (see Figure 6.2).

} S Me“

>t

Figure 6.1 A function of exponential order. Figure 6.2 f(t) =e".

Although most of the functions we encounter in practice are continuous, they do

not have to be in order to have a Laplace transform.

Definition 6.2 | A function f(t) is said to be piecewise continuous in a given interval provided
(a) f(t) is defined and continuous at all but a finite number of points in the
interval, and
(b) the left-hand and right-hand limits exist at each point in the interval.

Remark. The left-hand and right-hand limits are defined, respectively, by

lim, f(t — €) = f(t) and lim, f(t +e) = f(t*). Furthermore,
when tf is a point of
continuity, f(t) = f(t*) = fd).

A piecewise continuous function f need not be defined at every point in the

interval of interest. In ithe ee it is often not defined at a point of een

*This relation is also denoted by f(t) = O(e").

 SEC. 6.2 /THE LAPLACE TRANSFORM OF SOME ELEMENTARY FUNCTIONS 175

and even when it is, the functional value assigned at these points really doesn’t
matter. Also, the interval of interest may be open or closed, or open at one end and
closed at the other (see Figure 6.3).

Figure 6.3 A piecewise continuous function.

Theorem 6.1 | ecewise continuous and of exponential order, then it has a Laplace ~

Proof: Consider
+| era
[e“feoa = [Pepa
‘CO

to

The first integral on the right exists, since f is assumed piecewise continuous. Since
fis also of exponential order, the second integral on the right satisfies the inequality

i* ef(t)dt < [ ero

t0
dt < Apart
to
0

Hence, by direct integration, we get

Me —(s—c)tg
[ero dt <=
>
Me eae

For s > c, this last expression vanishes in the limit as fy so that we say the
integral is absolutely convergent. [a]

Most functions met in practice satisfy the conditions of Theorem 6.1. However,
these conditions are sufficient rather than necessary to ensure that a function has a
Laplace transform. For example, both t-'! and t'? have infinite discontinuities at
t = 0, and while the integral

Aft} = [etetas
0

diverges, we can show that

is 1/2
it Ne [ efi at — @ (5)
0 5
176 CHAP. 6 / THE LAPLACE TRANSFORM

Hence t~'’? has a Laplace transform, but t~' does not. Finally, we remark that if
the transform of a-function exists, it is unique, since the definite integral of a
function is uniquely determined.

EXAMPLE 4 Find the Laplace transform of the piecewise continuous function f(t) where

tes t AO eee
FOES Sh nip ele

Solution From the definition, we have

9) foe)

LEED} = [ e~“tdt + [ e-"5 dt.

0 2

Using integration by parts on the first integral above, we deduce that

bitten “4 : id
20

ALF (O} Se ek St ier

S s o
ai |P See
X) 2

1 1 =
Se 6 te Oa Ore ee
5 RY S RY

which simplifies to

Hitt (Dip
{elabirencore aa5e Te
Figen

Remark. Observe that the original function f(t) in Example 4 had a discontinuity
at t = 2, whereas the transformed function F(s) is continuous for all s > 0. This
phenomenon is known as the “smoothing effect” of the transform.

Finding the transform of discontinuous functions by the method used in Example

4 is especially awkward when the function has several discontinuities, because of
the many integrals to be evaluated. A more appealing method for transforming such
functions will be discussed in Section 6.6.

EXERCISES 6.2

In problems 1-12, evaluate the Laplace transform of each given function directly
from the defining integral.

1) f@=t 20 FG)
On ue ue ee 4. f(t) = sinkt
 SEC. 6.2 / THE LAPLACE TRANSFORM OF SOME ELEMENTARY FUNCTIONS 177

e f(t)= coskt 6. i) =2e" =e *

Pah e w—~e 8. f(t) = sinhkt
9. f(t)= coshkt 10. f(t) = cos(at + b)
TL. FO]2? — 34:45 Ven fiiiee
. tbe)

In problems 13-16, evaluate the Laplace transform of each function, recalling the
identities coshx = (e* + e *)/2 and sinhx = (e* — e*)/2.
a
{aw
\43. f() = cosh?kt 14. f(t) = sinh*kt
15. f() = e“coshkt 16. f(t) = e“sinhkt

In problems 17-20, evaluate the Laplace transform of each function using integra-
tion by parts.

) Jit. Soe 18. f(t) = tsinkt

19. f (i = t? coskt 20. f(t)
= te ‘cost
Sat. Using the relation cos*x = 3(1 + cos 2x), find L{cos? kt}.
22. Using sin?x = 1 — cos*x and the results of problem 21, find L{sin? kt}.
4 “23° \ Show that if lime ~“F(t)= 0, then f(t) is of exponential order.
24. / Using the criterion in problem 23, determine whether the following functions are of
exponential order.
@) f@=r' Ghornran + (c) f@) = te
(@) f() = 3e"™* (e) f@ = Ssin(e") MO ==
Fx, Vi 25. Sketch the graph of each of the following functions and evaluate its Laplace trans-
form.

(om ies een

Sn, OSe< a Ud unt ZH
(b) 70) = | paire= a
fh O<%< il
(Oi =oo et. | ata 2
Ls ip ep
*26. The gamma function, denoted by I(x), is defined by the integral
re = |eete db,
1)

which converges for all x > 0.

(a) Show that T(x + 1) = xI(x).
(b) Show that when x = n,n = 0, 1, 2 then Lin 1) = nh

*27. Using the results of problem 26 and the special value (3) = V7, verify the follow-
ing identities.
178 CHAP. 6 / THE LAPLACE TRANSFORM

/2

(fy Pee (2) esa 0.

(b) Pir} = +(2) s>0.

1/2

(c) #07} = 8(2)

s
ats 0!
IN@e se ib)
(d) L(t} = gerd ? s>o0.

*28. Referring to problem 26, show that

logs
L{logt} = {rd = S ,

where

I’@) = [e-og Nt odio

0

*29. Show that [() = Vz.

Hint: Establish that [(¢) = 2ffe~*dx = 2ffe dy.
Then multiply these integrals and evaluate the resulting double integral by changing
to polar coordinates.

6.3 OPERATIONAL PROPERTIES

Operational properties often offer an attractive alternative for evaluating the trans-
form of a function to tedious and often cumbersome evaluations of the integral that
defines the transform. For example, it becomes fairly easy to evaluate the trans-
forms L{e~* cos 2t}, L{t? sin St}, and L{r°e*} with these operational properties
once we know the transforms of L{cos 27}, f{sin 5t}, and L{t°}, respectively. First
among such properties is the linearity property, which is a simple consequence of
integrals.

Theorem 6.2 (Linearity) If L{f(p} and L{g(H} are the Laplace transforms of f(t) and g(#),
respectively, then

K ACif() + Cig} = CLF} + CoL{g (O}

for any constants C, and C).

Proof. From the definition,

LC f(t) + Cog (0} = [“e(Cif() + Cog (Dl dt

= Cie“ penat + Co |"eg dt
0 0

= CAFO} + CAz(o}. O
 SEC. 6.3 / OPERATIONAL PROPERTIES 179

EXAMPLE 5 Evaluate {7 — 3e* + Ssinz¢}.

Solution From the linearity property, we have

SALT — 3e% + Ssint} = 7A{1} — 3Lf{er4 + 5L{sint}

indSoe verte
St
Sin
es eal
where we have used the results of Examples 1, 2, and 3.

EXAMPLE 6 Find &f{sin ks} and L{cos kth.

Solution While both of these transforms can readily be found through routine integration
methods, another approach using Theorem 6.2 is available to us. Setting a = ik in
L{e"} = 1/(s — a), we get
1 S
Tier deed teege
From Euler’s formula, e’® = cos @ + isin 6, and Theorem 6.2, it follows that

ew sicos it} 1Asin kt}:

Matching up real and imaginary parts, we deduce
S
Li{cos kt} = ee S{sin kt} =
s? Cee

Functions multiplied by exponentials are easily handled because of the ex-

ponential function occurring in the defining integral of the Laplace transform.

Theorem 6.3 | (Shifting) If F(s) = LYf(}, then

K [HetfOh= Fo a}
Wards :
Proof: By definition,

Le"f (Ot = i‘eLemp(p]dt = [Wee (di,

from which our result follows. [_]

EXAMPLE 7. Given that £{r*} = 4!/s°, evaluate L{t*e*}.

Solution Direct application of Theorem 6.3 leads to

180 CHAP. 6 / THE LAPLACE TRANSFORM

Wes < aN Un OMe

Litre oe 553° GUE

EXAMPLE 8 Evaluate L{e~~cos 34}.

Solution From Example 6, we know that S{cos 37} = s(s? + 9)~', and hence, through the
shifting property, it follows that
SAEZ
Die =2i
ta cos3ty =
G +2
ee
+9"
are Se

6.3.1 The Laplace Transform of Derivatives and Integrals

The real merit of the Laplace transform is revealed by its effect on derivatives.
Suppose that f(1) is continuous with a piecewise continuous derivative f’(1). Then
by definition

LE f'()} = i“en f'(0) dt

Tee To) alae [* e-F(1) dt,
0 0

where we have performed an integration by parts. If the function f(‘) has a Laplace
transform F(s), then lime “f(t) = 0, so that

LL f'(O} = sF(s) — f (0). (6)

Remark. If f(#) is continuous for t > 0 except for a finite jump discontinuity at
t = a, then it can be shown that (6) takes the form

AS (t)pi= SH (S)Paf(0) = 1) a") mel(aene

EXAMPLE 9 Use the result L{sin kt} = k/(s? + k”) and Equation (6) to evaluate L{cos kt}.

Solution Let f(t) = sinkt and then f'(t) = kcoskt. From (6) it follows that

L{k cos kt} = sL{sinkt} — 0

PLES
s?+ k?’
and hence we deduce that (as before)

L{cos kt} a eae

5 2 :
 SEC. 6.3 / OPERATIONAL PROPERTIES 181

Similarly, if f(¢) and f'(t) are continuous and f(t) is piecewise continuous, and
all three have transforms, it follows that

Af} = sLAF'O} — f'O)

s7F(simist(O) aah), (7)
Generalizing these results, we state the following theorem.

Theorem 6.4 | (Differentiation) If f(t), f'(),..., f" ?@ are continuous for t = 0 and of
exponential order, and if f(t) is piecewise continuous and of exponential order,
then

EXAMPLE 10 Evaluate f{r"} forn = 1,2,3,....

Solution Let us first take the case when n = 1 and define f(t) = t. Then f’(‘) = 1 and from
previous results we know L{1} = 1/s. Substituting these expressions into (6), we
get
1
we Sty 0) = sit}
or

Dike S

The function f(t) = ¢” and all its derivatives are continuous and of exponential
order. Also we observe that

{OQ Sf OH 222 Sf 7D =0,, £90 = n!.

Substituting these results into Theorem 6.4 leads to
ns
eee ei Ol ah) eae i),

RY

and therefore

A different application of Theorem 6.4 involves the Laplace transform of inte-

grals.
182 CHAP. 6 / THE LAPLACE TRANSFORM

Theorem 6.5 If f(¢) is piecewise continuous and F(s) = L{f(H}, then

Proof: Let us define

a(t) = [fw) du,

which is continuous (why?). Furthermore, g(0) = 0 and g(t) = f(t). Hence,

FFD} = 5} [Fey au}=

or

4 [709 du|= 7 eal

0 5

Remark. Theorem 6.5 is a special case of the convolution theorem, discussed in

Section 6.7.

In these last properties, we found that differentiation and integration of a given

function in the t-domain correspond roughly to multiplication and division, re-
spectively, in the s-domain. In this fashion, the Laplace transform has the effect of
relacing the operations of calculus in the t-domain with algebraic operations in the
s-domain. It is primarily for this reason that the Laplace transform is so useful in
applications.

6.3.2 Derivatives and Integrals of Laplace Transforms

Sometimes we need to evaluate the transform of a function that is expressed as
t"f(t), when the transform of f(1) is either known or readily obtained. In order to
derive the needed property, let us start with the relation

F(s) = i“e-F(t) dt (8)

0

and formally differentiate both sides with respect to s. This action leads to

F's) = [ Cofe “ar,

0

and thus it follows that

Lif (O} = —F'(s) (9)

 SEC. 6.3 / OPERATIONAL PROPERTIES 183

EXAMPLE 11 Evaluate L{rsin r}.

Solution From the known relation £{sint} = (s* + 1)~', simple differentiation with respect

ee tee a
to s yields
d 2s
{t sin t} ras [(s Dal G+ De

EXAMPLE 12 Evaluate L{te~*

cos r}.

Solution Starting with L{cos rt} = s(s? + 1)~', we first evaluate

d s s*— 1
L{tcos t} ae Ae - ) i. (s? + 1) |

and then apply the shifting property to obtain

(ste
2) al
L{te cost} = nea ay :

If we differentiate (8) n times, we get

F2X(s) = [“(1 (De dt,

0

from which we deduce the following theorem.

(0) =(DFM).
Theorem 6.6 If F(s) = L{f(}, then forn = 1, 2,3,...

The integration of both sides of (8) produces yet another interesting result.

af
heorem 6.7 If F(s) = L{f(o}, then

ee = [ Fwdu,

provided that f(#)/t has a transform.

Proof. Integrating both sides of

F(s) = i“ef(0)dt
184 CHAP. 6 / THE LAPLACE TRANSFORM

with respect to s from s to , and interchanging the order of integration on the right,
we find

[F(u)du = [ =| f()dt
ioe) 2]! e -—st |]

s Fh tage! aah

= ea Ka dt,

which is our intended result. [_]

EXAMPLE 13 Evaluate £{(sint)/t}.

Solution Since f{sint} = (s* +1)7!, we have

int)\ je d 1
SA aia! = i 5 i—= Arctan— = Arccots.
LWA bawaPensa Ss

EXAMPLE 14 _ Using properties of Laplace transforms, evaluate the integral

| te ~costdt.
0

Solution In Example 12, we have shown that

co Zs jj
L{tcost} = I te" costdt = — sO
0 (5? +i 1)2

Letting s = 2, we find
% 3
I e te ~costdt
cos = —.
5

EXERCISES 6.3

In problems 1-12, evaluate the Laplace transform of the given function using
known results and any of the operational properties.

Ley i) =a3tes 2) fe =te *

La) fa = sink 4. f(t) = 2e~'sin3z
C5. f() = cosh kt cos kt 6. f@) =e (t? — 2t +7)

7.) f(t) = 3e- “(cos4t — tsin 41) 8 f= Sah

=

“igs fin
=2 —< 10. f= [waa + en")du
0

Ba al f(@) =e ‘cos*t 12. f(t) = 5te* sin7t

 SEC. 6.3 / OPERATIONAL PROPERTIES 185

‘ep eae and f(t), where f(t) = t + 1,0 <t S2, and f(t) = 3, t > 2.
(a) Find PL}.
(b) Find S{f'(t} in two ways.
14. If L{f()} = F(s), show that L{f(at} = (1/a)F(s/a), a > 0.
15. Using problem 14 and L{cos t} = s(s* + 1)", find L{cos 44}.
| 46. If n denotes a positive integer, obtain L{t"e“} from L{e“} = 1/(s — a) using The-
orem 6.6.
17. Using Theorems 6.5 and 6.7, determine the Laplace transform of the sine integral

sity = [du
a
18, Using transforms, evaluate the following integrals.
og Se St? = ea CE!
(a) | ee ea (b) | Saeed
10) 0

(c) ee at (d) [Pe-*sineat.

0 t 0

Hint: Set s = 0 iD) eee a for part (a), etc.

0

*19. Show that | [eau = [ea —e “) du.

a su Sto ew
*20. If f(x) is a periodic function such that f(t + 7) = f(d,

(a) show that F{f(9} = ¥ |Oe =O nT

Gy

(b) Using the relation 5’ e “= (1 —e ~*~ verify that the Laplace transform of
n=0

a periodic function with period T is obtained from

T —st

(poy = LOE
2
*21. Use the result of problem 20 to establish the following Laplace transform relations:

(a) L{f}= “tanh(Fs). where f(t + 2c) =f(t) and

= i, OS
rp Se
FO = ors ee.

L c 2c 3c 4c 5c
=

Problem 21(a)
186 CHAP. 6 / THE LAPLACE TRANSFORM

(b) Lf} = tanh (es), where f(t + 2c) = f(d) and

OSS
fO = Ne Sie SiS We

Problem 21(b)

(c) {| sint|} = (s? + 1)~'coth (3s).

Pa
|
2a
At |
Problem 21(c)

e'd
n

*22. The Laguerre polynomials are defined by L,(t) = nian! é),. m=O; 172° ee
Show that

Lait} _1fs—1\n
= 4(— 4)",
Hint: First find L{e 'L, (b}.
*23. The error function (erf) and complementary error function (erfc) are defined, re-
spectively, by

Die e gre Es
erf ) =| ear, erfe(1) = =e) ede
Veal (=a) eta
(a) Show that erf() = 1.
Hint: Use the gamma function relation TG) _ Vox, given in problem 29 of Exercises
(9),7%,
(b) Show that erfc(t) = 1 — erf(a).
*24. Using problem 23, establish the following Laplace transform relations.
(a) L{e~V") = Ware” erfe(s).
Hint: Write jt? + st = (4t + s)? — s? and make the change of variable y = 4¢ + s.
 SEC. 6.4 / INVERSE LAPLACE TRANSFORMS 187

(b) Lferf(t)} = ste” erfe (4s).

Hint. Change the order of integration.

(c) ‘Heriot )} = te S
co > eras qr?

Hint: | ee aS Pia a
0 2a

(4) From (c), show that fferf(¢-!7)} = 21 — 2%).

*25. Expand sin r'” in an infinite series and
(a) show that Lf{sint’?} = (7/28? %)e"'*.
(b) From (a), determine £{t~'sinr'’*}. Hint: Use problem 23.
*26. Starting with the integral relation for the Bessel function
D) a/2 3
Jo(t) = =) cos (tsin 0) dé,
TT Jo

(a) show that

Qs pa csc?6
L{Jo(o} = =f (2+ 1) + aeER
(b) Make the change of variables x = cot @ in (a), evaluate the resulting integral,
and establish the transform relation
1D Oe Saat) eae

6.4 INVERSE LAPLACE TRANSFORMS

In applications the use of the Laplace transform is effective only if, given some
function F(s), we can find the inverse transform f(t). In symbols, we write

fO = L NFO}. (10)
We might well wonder if an explicit representation for the inverse Laplace trans-
form exists analogous to the integral representation of the transform itself. Such a
representation does exist, but it requires integrations performed in the complex
plane that involve techniques of complex variables. Therefore, we will rely on other
methods for constructing the inverse transform.
In the process of finding the inverse transform f(t) of a function F(s), there are
three basic questions of concern:

1. Does the inverse transform exist?

2. When the inverse transform exists, is it unique?
3. How do we find inverse transforms?

The answer to question | is not necessarily. In order for F(s) to be the transform
of some function, it must possess certain continuity requirements and behave
suitably as s—>00, as stated in the following theorem.
188 CHAP. 6 / THE LAPLACE TRANSFORM

Theorem 6.8 If f(t) is piecewise continuous and of exponential order, then

lim F(s) = 0,
Ker hs)

where F(s) = L{f(d}.

The proof of this theorem follows that of Theorem 6. | and is left to the exercises.
The real significance of the theorem is that if F(s) is any function such that lim so

F(s) # 0, then it does not represent the Laplace transform of any regular function
f(t). This condition immediately rules out many functions as possible Laplace
transforms, such as polynomials in s, e*, cos ks, and so forth.
To respond to question 2, we observe that a discontinuous function can be
transformed into a continuous function by use of the Laplace transform. If f(f) and
g(t) are two identical functions except for a finite number of points, they will have
the same transform, say F(s). Hence, either f(#) or g(t) can be considered the inverse
transform of F(s). That is to say, the inverse transform of a given function is
uniquely determined only up to an additive null function (Lerche’s theorem). Since
null functions are normally of little consequence in applications, the apparent
difficulty of finding unique inverse transforms is of no practical concern.

Remark. A null function n(f) is one satisfying [; n(u) du = 0 for all t¢.
0

As an illustration of the above remarks concerning uniqueness, recall from

Section 6.2 that £{1} = 1/s. Also, we note that the function defined by
he UIT Syeee
g(t) = 42, t=5
bo ss Ss

&(t)

Figure 6.4

(Figure 6.4) has the transform

LF{e(t)} = [ e“ewar = [eat + [eva

0 5

1
yp
 SEC. 6.4 / INVERSE LAPLACE TRANSFORMS 189

Hence, either | or g(t) can be considered the inverse transform of F(s) = 1/s. When
a choice can be made between a continuous function f(f) and a discontinuous
function g(f) to represent the inverse Laplace transform of F(s), we always choose
the continuous function. There can only be one continuous inverse transform in any
situation.
In answer to question 3, we find that for many routine problems the inverse
Laplace transform can be obtained directly from existing tables of transforms. (A
short table is provided in the last section of this chapter.) Also, many of the
operational properties used in finding the transform itself can likewise be used in
constructing the inverse transform. For instance, the linearity property for inverse
transforms reads

L£'NC FG) + C,G(s)} = CL 4F(s)} + C24 {G(s}. (11)

Hence, optimum use of tables coupled with certain operational properties will
generally produce the required inverse transform in most elementary problems.

EXAMPLE15 Find £-'{(3s + 7)/(s* + 5)}.

Solution Using the linearity property (11), we have

et 3s + 7 = =1 KY FE
(have Site 1 V5
a oe a oe gee |e
ee
= 3cos V5t + Gesin V51.

This last result is obtained from the tables.

EXAMPLE 16 Find Zo) Sar :

Solution Completing the square in the denominator, we get
ee a Sao Woes
se +6s +13 (s +3 +4 (s +3440
g the shifting property (Theorem 6.3)inthe form
PINE CE HO)SALA
we obtain

oe ape reeeeies —s1cp—|

Cae, S38
Ss
- 8 a a
LEN bok K) ie oF 2
— a7 ‘| feds! 49. ie=a}|

e~*(cos 2t — 4sin2r).
a
eee
190 CHAP. 6 / THE LAPLACE TRANSFORM

6.4.1 Partial Fractions

In many cases of practical importance we wish to find the inverse transform of a

rational function, i.e., a function having the form

Ets) a),
Q(s)
where P(s) and Q(s) are polynomials in s. The inverse transform in such a case can
most easily be effected by representing F(s) in partial fractions. The partial fraction
representation is the same as that found in the calculus, for example, as a means
of integrating certain rational functions. It is assumed that P(s) and Q(s) have no
common factors and that the degree of P(s) is lower than that of Q(s). Let us
illustrate the technique with some examples.

EXAMPLE 17 Find the inverse transform of

OR eas |
ENS) a
SitaramS:

Solution Using partial fraction expansions, we have

DSce T1552
SS ee
Suamel A B —
re:
si—s ePID
Ka Ie ey ee AP eh
or, upon clearing fractions,

2s? + 5s —1=A(s — 1)(s + 1) + Bs(s + 1) + Cs(s — 1).

By setting s= 0,s = 1, ands = —1, respectively, we deduce thatA = 1,B = 3,

offs oof ae(t} a)

and C = —2. Hence,
2 ms

Gg = 6% Ky Gg = |! Suatael
=O ee.) Cae

EXAMPLE 18 Find a(S Da '(s2 tah ye

Solution To determine the inverse transform, we first assume an expansion of the form

1 eA BsiaC
(s +1)? +1) st+1 57417
Clearing fractions gives
1=A(s?+1)+ (Bs + C\(s + 1).
Setting s = —1, we find A = 3, and equating the coefficients of s* and s° yields the
equations
 SEC. 6.4 / INVERSE LAPLACE TRANSFORMS 191

QO=A+B, 1=A+C,

from which we deduce B = —} and C =}. Thus

PHM eet keer

a2 s+1
te 2,
les+] el ol pt)
2 +1
1
= oe “2 COSIA7 Sill):

: =f Ee
EXAMPLE 19 Find & ee
e 5}

Solution Let us write

onl, A B
SSS = 7 =F c Sacite ee Lpvee
ss) 202) 9) Sai st Ssee Gist) Gey?
Clearing fractions, we obtain
s+1=As(s + 2)? + Bis + 2)? + Cs%(s + 2)? + Ds*(s + 2) + Es’.
Solving for the constants yields C = —A = %, B =4, D = 0, and E = —%.
Hence
m
ge! s == || 1
eae Vera 1 21 wi) =1 =1 @e=i 1
= - a (6s {2|i 37 ‘=|y Ts t;- |

| | +

The method of partial fractions can be greatly systematized by use of the

Heaviside expansion theorems.* These theorems provide an approach to deter-
mining the unknown constants in partial fraction expansions that is more so-
phisticated than clearing fractions and matching like terms. The technique involves
finding all linear factors (both real and complex) of the polynomial in the denom-
inator, producing a partial fraction expansion in terms of these linear (and possibly
repeated) factors, and then evaluating the unknown constants one by one. In other
words, it is not necessary to solve simultaneous equations as we did here. Such an
elaborate procedure, however, is not required for the elementary problems consid-
ered in this chapter.

York: Wiley,
*See, for example, E. Kreyszig, Advanced Engineering Mathematics, 4th ed. (New
1979).
192 CHAP. 6 / THE LAPLACE TRANSFORM

EXERCISES 6.4 _

In problems 1-12, determine the inverse Laplace transform of each function using
Table 6.1 on p. 217 and various operational properties.

h) F(s))=4 2. FS) = 55

Pt FO - Go 4 =a
5) Fs) =a bh ilinerer rie
7) Wore TR 2 Oe ae

(9 Fo = Gea) Ke soxyaae
( 11; F(s) = (ess 12. F(s) = >

In problems 13-24, evaluate the inverse Laplace transform by the method of partial
fractions.

ons (Ge ak Sy” 14. F(s)=

(Saal) (Sate? (Smtr)

15) F(s) ae 16. F(s) = ee

17. esos ke 18. F(s)

= (st — 1)7!

Ming a
“23. Pe ge?
25.\ Given that F(s) = L{f(}, show for constants a, b, and k, that

(a) LMF (ks)} = (3). k>0.

(b)) Seu Figs tb) = te-mer(*), Gea

26. Using the results of problem 25, evaluate the inverse Laplace transform of each of the
following functions.
3 ¥/

(a) F(s) = (2) (OMG Oh a 4s*

+ 45 + 5
eee
SEC. 6.4 / INVERSE LAPLACE TRANSFORMS 193

1
(ers) =
Oso (508 OS oo]
ane If it is known that Lo '{s~'/2e—
15} = (at)7
"7 cos (2t'”), find
Sia hype eaAsal ts ee WY
Hint: Use problem 25.

28. Show that £-'{(s + a)"} = —_

n!
e 5 1
29. Show that £ ff = BoA cos bt — asin bt).

a .
*30. Show that £ tg(: ‘)|= es
y= t
Hint: Write
lo § se 1 =p als:
Bs aT AGE Sy in
and use
l+x el axe
os(+ =")=2 rere ea
Hest
Find the inverse transform of the series and sum the result.

EL Show that (see problem 30)

a iy x l=@~
CRP ee {ioe(1+ “) a

ae G =i * en = et

(by acy {te—7 “) eae

*32. Solve problem 30 by using Theorem 6.6 in the form (n= 1)

1
fo = - GE ME (Oh

*33. Solve problem 31 by using Theorem 6.6 (see problem 32).

= sin tx
*34. Letfo = | 0
nbc. if S> 0)

(a) Show that by taking Laplace transforms of both sides, we formally obtain

F(s) = [@ + 52)"dx
(b) Evaluate the integral in (a), and by taking the inverse transform of the result,
deduce the value of f(t).
*35. Using the technique of problem 34, show that
a 1/2

(a) Ka
he po eae!
ed 7
sae Pm on pe As
194 CHAP. 6 / THE LAPLACE TRANSFORM

= a \12
(c) [ xo * sintxdx = (z) pam ios 0;
0 ; 2t
*36. Prove that lim sF(s) = lim, f(o.

Hint: Set f() = >) cnt”.

n=0

*37. Prove Theorem 6.8.

6.5 SOLUTION OF INITIAL VALUE PROBLEMS

The Laplace transform is a powerful tool for solving linear DEs with constant
coefficients—in particular, initial value problems. Let us illustrate the method with
a simple example.

EXAMPLE 20 Solve y’ = 3e', y(O) = 7.

Solution We first apply the Laplace transform to both sides of the DE:

Ly'(O} = Le}.
From the tables we find that £{3e'} = 3/(s — 1), and making use of Theorem 6.4,
we get

Ay'(O} = sLHyO} — yO) = s¥(s) — 7.,

Thus, the differential system is reduced to the algebraic equation

sY¥(s) —-7 = es
. suit
and solving for Y(s) yields 7
3 7
Y(s) = +—.
(s) SS abd) os
Expressing Y(s) in terms of its partial fractions, we get

Yi) = ——
Sua el
+2,as
and hence the solution of the original problem is symbolically ‘given by

YQ) = LOXY(s)} = eres + we

Se S
or

y(t) = 3e’ + 4.
ee
EE ee eee
 SEC. 6.5 / SOLUTION OF INITIAL VALUE PROBLEMS 195

Although this example was rather trivial and could be solved more easily by
another method, it does illustrate the basic features of the transform method. The
usefulness of the Laplace transform rests primarily on the fact that the transform of
the DE together with the prescribed initial conditions reduces the differential
system to an algebraic equation in Y(s). Such an algebraic equation is readily
solved, and the inverse transform of its solution then yields the solution of the initial
value problem; i.e., y(t) = L'{Y(s)}. Furthermore, the solution of the DE satis-
fying certain initial conditions is found directly without first producing a general
solution and then solving for the arbitrary constants.

EXAMPLE 21 Solve y" + 2y' + 5y =0, y(0)=2, y'(O) = —4.

Solution The transform of the DE leads to

[ys on a iss) 21 ors) — 0;
or
(s* + 2s + 5)¥(s) = 2s.

Solving for ¥(s) gives

2s
SD 8 Os SR TAS

Making use of the shifting property, we can write

2s = 2s CSTE
Soest 5: a(S Ped (sl ler

and thus

y(t) = ages°+4
= |

# (285 z i}z af :i)

Nee COS chasis),
eS eee ee
;7
£510 Btre

EXAMPLE 22 Solve
y" — 6y + Sy = Te~, 2 ot
OPS Tes

Solution Taking the transform of the DE and simplifying gives us

D
(s? EaLOS ai 9)Y(s) = INS ra; 3) ey nay

Hence,
196 CHAP. 6 / THE LAPLACE TRANSFORM

and taking inverse transforms, we find

Biber figist payer GR!
viet ae eae |
— Shae tee
|
2e te4 3t

eee ee er, Pee Se SS SS eS eee

It may be of interest to see exactly where each input parameter ends up in the
transform domain by considering the general second-order DE

ay” + by’ + cy =f (9), (12)

where a, b, and c are known constants.
If we apply the Laplace transform to each term in (12), we get

als?¥(s) — sy(0) — y'(0)] + B[s¥(s) — y(0)] + c¥(s) = Fis),

or
(as? + bs + c)Y(s) — (as + b)y(O) — ay'(0) = F(s),
where F(s) = L{f(t)}. This algebraic equation can be rearranged in the form

cs (as + b)yy(O) + ay'(0) a ee ed

F(s)
setd 13
Ay as*+bs +c as*>+bs +c ds
Now, taking the inverse transform leads to
2 |(as: + &)y(O) + ay? (0) > F(s)
= Ly Se pl) 14
10s =? as*+ bs +c as*+bs +e Cg)
a
Yu (t) yp(t)

Here it is interesting to observe that the solution (14) has naturally split into two
parts—the function y,, which is a solution of the initial value problem

ayisibye + ocy = 0:en yO) hoya) eka, (15)

where ky and k; are prescribed values, and yp, which satisfies

ay" + by'+cy =f), y(O)=0, y'(0) =0. (16)

We see, therefore, that the transform method splits the problem into two separate
problems, much like we did in Section 5.5 because of physical significance. This
makes the Laplace transform an effective tool for analyzing the basic characteristics
of a system in response to each of the input parameters ky and k,, or f(#).
Finally, we should note that the coefficient as* + bs + c of the function ¥(s) in
the transform domain is precisely the polynomial that appears in the auxiliary
equation of the DE. Thus, the use of the Laplace transform does not avoid the
necessity of factoring this polynomial. Once again, this factorization may represent
one of the most difficult aspects of the problem when the DE is of higher order.
As a last example, let us solve a third-order DE.
 SEC. 6.5 / SOLUTION OF INITIAL VALUE PROBLEMS 197

a ee eee ee ee ees ee ee ee Ve
EXAMPLE 23 Solve y” + y"”—y'’—y =9e", y0)=2, y'(0)=4, y"(0) =3.

Solution Applying the transform, we have

[s*¥(s) — 2s? —4s-— 3] +'[s7¥(s) — 2s — 4] — [s¥(s) - 2] - Ys) = ay.

and solving for Y(s) gives

DNs ollOS Os) 2

Y(s) =
GG 4.1) (s ~2)6 —Dw@+41y
Mase AOS StS )s = 2) 4-9
(s — 2)(s — 1)(s + 1)?
Using partial fractions, we write
A B C D
Y(s) = —— + ——_ +.———+ —
RV ar ae ey gean
from which it is found that A = B = C = 1 andD = 0. Hence

Whj=— 6. 1 6 + te!

6.5.1 Spring-Mass Systems

The equation of motion governing small movements of a damped spring-mass

system is known to be (see Section 5.2)
my" + cy’ +ky =f@, t>0, (17)
where m denotes the mass, k is the spring constant, and f(¢)* is an externally applied
stimulus. The term cy’, where c is a positive constant, represents a retarding force
due to damping effects. Let us assume that the initial position and velocity of the
mass are given by
y(0) = yo, y'(O) = . (18)
To discuss some of the possible motions of the system described by (17) and
(18), let us first set f(#) = 0. We will find it convenient to also introduce the new
parameters

= mee 2 =
2m’ m
so that the governing equation now takes the form
y" + 2ay' + b*y = 0, y0)=yo, y'(O) = vo. (19)

Because of the nature of the parameters involved, it is clear that a > O7, and we
may choose b > 0.
*We use a lowercase letter here for f(t) in order to be consistent with our Laplace transform notation.
ta = 0 only in the absence of the damping term, i.e., when c = 0.
198 CHAP. 6 / THE LAPLACE TRANSFORM

If we apply the Laplace transform to (19), we find

s°Y(s) — syo — Uo + 2a[s¥(s) — yo] + b°Y(s) = 0
with solution
3 SYo aE Uo SF 2ayo

xe) s? + 2as + b?’

which we choose to write as

= yo(s + a) + (Uo + ayo) 20

oo RP SSEDet pelome io
Hence,

ip Krac?)
nie ee Pe Se 21)
It is here where the benefits of using the Laplace transform become evident. In
order to invert the expression in (21), we need to know whether the denominator
s’.+ b* — a? has distinct real factors, has equal factors, or is the sum of two
squares. These three cases correspond to the following:

Case I—Overdamping: b <a

Case II—Critical damping: b = a
Case II1I—Underdamping: b > a

Case I: When b < a, then also b* — a* < 0, and we write

b?-—a*= —- a’.
By taking the inverse transform in (21), we get the result

YoS + (Vo + ayo)

y(t) = erro eae

(vo + ayo)
=e “a(ncoshat + sinh x : (22)

which describes overdamped motion.

Case II: Critically damped motion results when b = a, and under this condition

y(t) = oe ae (Vo ae

S Sig

e~“Lyo + (vo + ayo)t]. (23)

Case III. Oscillatory motion can occur only when b > a, which means that the
system is underdamped. Here we find it convenient to put
be--q@e= (en:
 SEC. 6.5 / SOLUTION OF INITIAL VALUE PROBLEMS 199

and making the appropriate parameter change in (21), we obtain

y(t) = e "ZL"! yos + (vo + ayo)

57+ ine

2 Uo +
=e *(sncosLiss. a URL ut) (24)
uv
The reader should compare all three solutions with those obtained in Section 5.2.3
for each of the three cases of damping.
When the forcing function f() is not identically zero, the Laplace transform
method leads to the solution form

y(t) = ete! es + (v9 + 20 +g F(s) | oo

s?°+b*- a’ (s tay +b? - a’?
where F(s) = L{f(H}. Let us illustrate this case with an example.

EXAMPLE 24 Determine the response of a spring-mass system that is initially at rest and then at
time t = 0 subjected to the sinusoidal forcing function f(t) = P cos wt. Neglect
damping effects.

Solution The spring-mass system is characterized by

y" + wy = ~cos wt, yO)=0, y'(0)=0,

where w) = Vk/m. The transform of this DE leads to

Ps/m
(s? + w6)¥(s) =
se + w?

with solution
Ps/m
Y(s) =
(So a) srk)
No resonance: If w # wo, a partial fraction expansion yields
S As Baw Cs ete)
(tosis ee on sk | sw.
from which we find A = —C = 1/(w3 — w?) and B = D = 0. Hence
yy) = LY(s)}
P
Se ae ay (COS WF a COS Wot).
m(wo = )

whose fre-
This solution represents the superposition of two harmonic motions
frequen cy w of the forcing term and the natural
quencies correspond to the angular
angular frequency w» of the system.
200 CHAP. 6 / THE LAPLACE TRANSFORM

Resonance: When w = @o, we get

Pie s
He Be \2 - ai
je ;
= t Sin Wot,
2m Wo

this last result being obtained from the tables. The solution this time corresponds
to resonance in the system, since y>® as 1 (see also Section 533-2).

EXERCISES 6.5

In problems 1-20, solve the initial value problem using the Laplace transform
method.

| | |
nN & < ® =S ~— | _—

Ary= Oy Oe Lary (Oe

i eataet) yO)—0 yO
6. y"+y=1, y)=0, y'(0)=0
Ud.) y” ty = 2e', yO) =0, y'O)=0
8. y”+y'—2y =-—4, yO =2, y’O)=3
9.) y” + 2y' + 2y =sin2t — 2cos2t, y()=0, y'(0)=0
10. “y"+ 2y' ty =3re™, yO) =4," y'0)=2
11) y"+4y'
+6 =1+e%, yO=H1, y'O=-4
12. y" + loy =cos4t, y(0)=0, y'(0)=1
SAB.) y"— 4y' + 4y =t, yO=1, y'0)=0
14. y"—y'=e'cost, yO)=0, y'(0)=0
15.5 sy" 1 6y" lly’ + 6y = 0, yO) =2, yO) =1, y0) = —1
16. y"—y" + 4y’—4y =1t, yO)=0, y'(0)=0, y"(0)=1
V7. 2p" + Sy" By" — dy =e yO) =-0,"ay, (0) = Orary, (0) er
18." yO =y =0, yO=T, y'O)=0, y"0)=—1, »"O),=0
19. y®-y=t2, y(0)=0, y'0)=0, y"0)=0, y"0) =0
"20. y® — 4y® + 6y" — 4y'+y =0, y(0)=0, y'O)=1, yO) =0,
y"0) = 1
In problems 21-31, use the Laplace transform to solve the problem specified.

\21., Problem 3, Exercises 5.2 22. Problem 5, Exercises 5.2

 SEC. 6.6 / DISCONTINUOUS FUNCTIONS 201

Problem 8, Exercises 5.2 24. Problem 20, Exercises 5.2

Problem 21, Exercises 5.2 26. Example 6, Section 5.3
Problem 4, Exercises 5.3 28. Problem 15, Exercises 5.3
Problem 5, Exercises 5.4 30. Problem 7, Exercises 5.4
Problem 8, Exercises 5.4

Show that the Laplace transform of the differential system

eye 2 Ly y= Oy (0) 10 yi(O) = 4,
leads to the first-order DE in the transform domain

(s? + 2) + (45 + 4)Y =0.

Solve this first-order DE to obtain Y(s), and invert it to find the solution y(t). (This
problem is one of the few variable-coefficient DEs for which the Laplace transform
method proves fruitful.)
*33. Apply the method of problem 32 to Bessel’s equation of order zero,
ty” + y! + ty = 0.

(a) Show that Y(s) satisfies the first-order DE

(di + so + s¥(s) = 0.

(b) Show that the general solution of the DE in (a) is

Vey CULE ase)o,
where C is an arbitrary constant.
(c) Express the term (1 + s*)~'” in a binomial series valid for s > 1. Assuming it
is permissible to take the inverse transform termwise, deduce that

y =y SE ae qe (n 1)? 0 an >

where Jo is the Bessel function of the first kind of order zero. Although a second
solution of Bessel’s equation can be found by other methods, it is not bounded
at t = 0 and so does not arise in the transform method. Because it is not
bounded at t = 0, it is discarded anyway in most applications.

6.6 DISCONTINUOUS FUNCTIONS

The Laplace transform has the effect of “smoothing” discontinuous functions in the
transform domain. Thus, one of the most interesting and useful applications of the
transform method is in solving linear DEs with discontinuous or impulse forcing
functions. Equations of this type are commonplace in circuit analysis problems as
well as in some problems involving mechanical systems.
In order to effectively deal with functions having finite jump discontinuities, it
is helpful to introduce the Heaviside unit function, or unit step function as it is often
called. We denote this function by the symbol h(t — a) and define it by
202 CHAP. 6 / THE LAPLACE TRANSFORM

On) =a
mea) ={ (26)
len fA Ose
where a = 0 (see Figure 6.5).

Figure 6.5 Heaviside unit function.

EXAMPLE 25_ Sketch the graph of f(t) = A(t — a) — h(t — b), a <b.

Solution From the definition of h, we find

0-0=0, 0<1<a
fO®=41-0=1, a<t<b
PSs eie eS ih
The graph of this function is shown in Figure 6.6.

Figure 6.6

Suppose we have a function f that has nonzero values only on the interval
Gai <b (ricure 6.7)... that is,

Hove ee Wareb
0, otherwise. (27)
In terms of the Heaviside unit function, we can write

FO = fAOlhG — a) — he — b)] (28)

rather than expressing f piecewise as in (27). More generally, if
Tig,
Om ta
TAD) esl) ae ett (29)
TOR tee
 SEC. 6.6 / DISCONTINUOUS FUNCTIONS 203

(Figure 6.8), then we can write

FO = ACAD — h(t — a)] + ALAC — a) — Act — by] + fh(t — b)

= fi) + LAM — AOMIAC — a) + LAW — AMIAC — 4) (30)
fort = 0.

Figure 6.7 Figure 6.8

EXAMPLE 26 Express the function

tp eysae
th) =

FO ie ft.
(Figure 6.9) in terms of the Heaviside unit function.

Figure 6.9

Solution Following (30), we write

Tipe) 2 ae = 2) She 2) 4
RG ne 2), 120.
ec
NSE ES ee i

The Heaviside unit function is also useful in translating a function to the right
a distance of a units. For example, the function y = g(t) is shown in Figure 6.10,
while the function y = g(t — a)A(t — a) shown in Figure 6.11 represents a trans-
lation of g by a distance a in the positive f direction.
Calculating the Laplace transform of h(t — a), we find directly
204 CHAP. 6 / THE LAPLACE TRANSFORM

Figure 6.10 y = g(t). Figure 6.11 y = g(t — ajh(t — a).

{h(t -— a} = i“e-h(t — a) dt

[ Ge at
a

or

SAR i — a) eet Soe Se. (31)

Observe that when a = 0, L{A(H} = L{1} = Is.
The next theorem, which relates the Laplace transform of the function
g(t — a)h(t — a) to that of g(t), is most important in our present discussion.

Theorem 6.9 (Translation) If L{g(t)} = G(s), then

LAg(t:— a)h(t — a)} = e “G(s).

Proof: From definition,

LF{e(t — ah(t — a)} = [eee — a)h(t — a)dt

0

= [eee ade

Introducing the new variable

:
x = t — a, we have
t-a.> %
S{e(t — a)h(t — a)} = (Leone dx

‘CO

i eal e * 9(x) dx

e *G(s),

and the theorem is proved. [_]

EXAMPLE 27 Find the Laplace transform of

ft) =
Hahaee
IM fe bi 2s
 SEC. 6.6 / DISCONTINUOUS FUNCTIONS 205

Solution From Example 26, we have

Tit Ga tnt = 2),
In order to use Theorem 6.9, we must rewrite 3 — 1? as a function of t — 2.
That is,
3-2=3-(-—2)-4t+4
=3—(f —2)—4¢ — 2) +4—'8
= —1 — 4(t — 2) — (t — 2).

Hence,
SG] inns oA t*}
2, pal | 4
=4-e(+5+4),

a) Sree Oe ese
ek a

6.9 in
Another way of solving Example 27 is to express the result of Theorem
the form
Fig Ones — a} =e “L{g(t + ads,
3)
x/+ \=> $44 (3 bp(4elt
and then it follows that
L{F(O} = L{t7} + eB =-(t + 24
= Ge +e eae ie 4 ateae ae

yh
Sy
aan os eta 3)
(1+44 eee
first writing 3 — t? asa
Using this latter technique, we avoid the necessity of
function of t — 2.

si ne
es a
EXAMPLE 28 Find the inverse transform of
1 — 3e>8
F(s) => ager as

Solution Writing Theorem 6.9 as

P'e-*G(s)} = at — a)htt — a),
we find

Bere ecet S|
] ¢ —5s
= 8% —3s

Il pte ie ae
206 CHAP. 6 / THE LAPLACE TRANSFORM

or

ve fo TOSS ie
Oi 1S aie
EXAMPLE 29 Solve the initial value problem

Yes) = (1) sanOe Oey dO ea

where

fd =

Solution We can interpret this problem as a spring-mass system (no damping) that is at rest
until time ¢ = 0 and then subject to the sinusoidal forcing function sint until time
t = q/2, after which the forcing function is removed.
Writing f() = sin[1 — h(t = 7/2)] and finding

Lio Lea 7 sin + 5)

set] 2

= I ame 2 Ft 1COS7
set] { }
1 se 7/2

s+ 1s? +1’
it now follows that

Liy"} + Ly} = Af oh,

or
1 se 7/2
Me eae 0G) es ae eee
(s) (s) se+1 s?+]
with solution
1 se m2

MAO) (s? + 1) al (s? + 1)?’

Thus,
me & 1 e % ge m2

erat fe + ‘al 4 le + a
car LAN ME take awa 7 TT. \e 1
7 (sine tcos f) sH(t 5)(+ ~ $)
sin(s - 2)

or, equivalently,
 SEC. 6.6 / DISCONTINUOUS FUNCTIONS 207

Line 7
=(Sinit
tat? COS) a Olt
— —
- 2
MO
ae Daa poe
Zz 4 : ok
Observe that the amplitude of vibration becomes steady as soon as the external
force is removed. Otherwise resonance would take place, since the forcing function
has the natural frequency of the system.

[O] 6.6.1 Impulse Functions

Closely associated with the Heaviside unit function is the impulse function
5(t — a), also called the Dirac delta function (see Section 5.6). For t # a, the
derivative of A(t — a) is clearly zero, and at t = a this function doesn’t have a
derivative in the usual sense because of the discontinuity. However, in a gener-
alized sense the derivative can be defined in terms of the impulse function. Consider
the following argument.
Since L{h(t — a)} = e~%/s, it follows from Theorem 6.4 that
dh
Fe = a|= sf{h(t —at=e™. (32)

Also, using integral properties of the impulse function, we have

£5(0 — a} = [e 5 — addr =e. (33)

0

The two results (32) and (33) suggest that

h (34)
tC =10)=10(i—aG)s

The interpretation of (34) is that the unit step function has a jump discontinuity
of unit magnitude at t = a. This is not too surprising since we defined it that way!
The significance of the result, however, is that it allows us to generalize the concept
of differentiation to include functions with finite discontinuities. Wherever a dis-
continuity occurs in the function, a delta function occurs in the derivative of this
function, and moreover, the magnitude of the jump discontinuity appears as a
multiplicative constant of the delta function.

Remark. Observe that £{5(1)} = 1, which follows from (33) by setting a = 0.

Although lim F(s) # 0 for F(s) = 1, we are not in violation of Theorem 6.8, since
6(t) is not a regular function.

EXAMPLE 30 Solve the initial value problem

y"+y =d¢-—7), 0)=0, y'O=)0.
208 CHAP. 6 / THE LAPLACE TRANSFORM

Solution We previously solved this problem in Section 5.6 (Example 12) by using the
method of Green’s function. Now, taking the Laplace transform of each term in the
equation, we get
s7¥(s) +: ¥(s) = e0™
with solution
y( = Cae

oe s? +17
Using the result £-'{1/(s* + 1)} = sint, we have

y(t) = ee | = sin(t — mAh(t — 7),

or (see Figure 6.12)

7 3m Sr Tt ee

Figure 6.12

OF) Ones tae

Vee.
Sin(fs— anja f= ae
We might interpret this problem as representing an electric circuit to which an
instantaneous unit voltage impulse is applied at time t = 7. Hence, there is no
response until time t = a (due to the homogeneous initial conditions), but the
impulse voltage produces a response that lasts indefinitely (there is no resistor for
damping effects). Interestingly, the solution is a continuous function for all t > 0
in spite of the singular nature of the impulse function. There is, however, a jump
discontinuity in y’ at the point t = 7. (See also the discussion in Section 5.6.1
concerning the Green’s function.)
re
NO ee a ee ee eee

EXERCISES 6.6

In problems 1-6, sketch graphs for t = 0.

vA , ;
AL.) f(t) = t?h(t) + (5 — t?)A(t — 3)
2. = h(t —
f() 1) + 2h(t = 3) = (ag — 2h
3. fit) = h(t) — h(t — m) + sinth(t — 27)
SEC. 6.6 / DISCONTINUOUS FUNCTIONS 209

4. f() = g(t — 2)A(t — 2), where g() = 0?

5.) f() = t? —(t — 2)*h(t — 2)
6. f() = t? — t7*h(t — 2)

In problems 7-14, express each function in terms of the Heaviside unit function and
find F(s).

2 Saal Pome One ee3

8. =
Ls fio = {? fee Fo 18, ee
sin 3t, O<t<t
WS to - sint, O<t<a7 10. f= ett
‘ Os Sty
; 2
Onn O17<< 3 a, WS S23
12. =
Ee fo= |e P33 Io i 4 1>3
fe ORs ORsie al
Kay fw = ee iA OQ, bere
ORE 4 ta OS
©, ¢=3

In problems 15-22, find the inverse Laplace transform.

re

f6,) Fs) = ~~ 16. F(s) = Si)

ps

17. F(s) — peas” 18. LOSS ae

—35)

19.
é
Be, = Gn 0 ; WS;
ee
s°+4
é so = s(1 a e =) he (Stee Chae
21 ehs) = any ee Os yar erer

Solve the initial value problems 23-34.

0, 0O<t<il
f
Ae ai GeO) — Ine wherenel (ice tf ji = tk,
(ye ASE |
y@)=0, where f=
24. y'+ty=f@, if > lle
Me Ome< il
‘96. y'+4y=f@, yO =1, yO- Qumwheros) (a= ‘i
peal

Ae Oise
t <2
Poti fy 0) iO (0) = 02 awhere f(D ( + 2, ipSD
210 CHAP. 6 / THE LAPLACE TRANSFORM

cos4t, Ost<7
27. y"+4y =f), y(@)=0, y’(0)=1, where f(#) = |
Os 6 Sar
28. y" + 4y =sint — h(t — 2m)sin(t — 2m), y(0)=0, y'(0)=0
29. y" — 5y'+ 6y =A -— 1), yO)=0, y'(0)=1
30. y"+2y’+2y =d(t-—m, yO)=0, y'0)=0
31. y”+2y'+2y =6@¢ -—7m, yO)=1, y'0)=0
32. y" + 4y = 6 — m — 5(t —2m), y()=0, y’0)=0
33. y" +2y’+y =&) + h(t —27), yO)=0, y'0)=1
34. y"+y =6(t — mcost, y(0)=0, y’0)=1
35. A spring-mass system at rest until time t = 0 is then subjected to the sinusoidal force
f() = Psinwt. At time t = 7 seconds, the mass is given a sharp blow from below
that instantaneously imparts an upward impulse of 5 units to the system. Neglecting
damping effects, describe the motion of the system.

6.7 THE CONVOLUTION THEOREM

In applications we must often find the inverse transform of a function that is the
simple product of two other transforms. Unfortunately, it happens that

L-NF(s)G(s)} # LF (s}L-Y{G(s)}.
We will presently show that the inverse transform of this product is equal instead
to the convolution of two functions f(t) and g(t).
The Laplace convolution of two functions f(t) and g(t) is defined by

(Fea) = [F( ~ wetw)du. (35)

There are several immediate consequences of this definition. First, we note that
making the change of variables v = t — u leads to

(Feed = —[Foe — ydo = ['florgtt — abo,

from which we conclude

(f*g)(t) = (g*f)(t) (commutative law). (36)

Sa a es eee
EXAMPLE 31 Calculate both (f*g)(A)and (g *f)(t), where f(t) = t and g(t) = sinkt.

Solution From (35) we have

(f*g)(t) = [« — u)sinkudu

t|sin ku du = [usin kad

0 0
SEC. 6.7 / THE CONVOLUTION THEOREM 211

t 1 t
= ratil = cos kt) ae
in ki Eepoos kt
SR

k jsinkt),

whereas

(g*f)O I, usin[k(t — u)|du

0

= sinkt |wcoskudu = cos kt |usin ku du

0 0

1 fia. Lage
= yasin kt cos kt + sin’ kt = point = 7300sktsin kt + 7.0087kt

l Ih. 2
= i(:= isint).

Other properties that readily follow from the definition are the following, the
proofs of which are left to the exercises:
f¥(kg) = (kf)*g = k(f*g), & constant (37)
f#(g +h) = f*g + f*xh (distributive law) (38)
f*(g *h) = (f*g) *h (associative law) ©?)

The important result we need is that concerning the Laplace transform of the
convolution. Let us consider

A(faso} = [eo |F — wa dudr

which we can write as a double integral,

LA fegy(O} = |‘ |,ef (t— u)g(u) du dt.

over the region
We can interpret the integrals on the right as an iterated integral
ange the order of
0 <u <t,0<t <~®, as shown in Figure 6.13. If we interch
<*,0 <u <, and
integration, we find that the region is characterized byu <t
thus

A(feg yo} = [| ef — weward

0 Ju

The change of variables x = 1 — u leads to the expression

than the other, which should be
* Although f*g = g*f, one of the integrals is often easier to evaluate
taken into account when using the convolution integral.
212 CHAP. 6 / THE LAPLACE TRANSFORM

Figure 6.13

FA(feg)(O} = I- i“en + F(x) 9(u) dx du

= |“eof(x)dx: i“e-9(u) du
or

Li(feg(} = F(s)G(s). (40)

In summary, we have the following important theorem.

Theorem 6.10 | (Convolution) If F(s) = L{f(H} and G(s) = L{g(t)}, then

£-{F(s)G(s)} = [f(t — wg(u)du = (feg)(0).

EXAMPLE 32 Find £-{s-(s? + k*)|}.

Solution Let us select F(s) = 1/s” and G(s) = (s? + k*)~'. From the tables, we find
1
oe = f(t) =t.

LNs? kh?)| = o(t) = “sinkr

Thus we write
ny i
x ee rk ifmM Ae
= [« 2 te
0 k
 SEC. 6.7 / THE CONVOLUTION THEOREM 213

[O] 6.7.1 The One-Sided Green’s Function

The Laplace transform also turns out to be a useful tool for constructing the
one-sided Green’s function (see Section 5.5) for constant-coefficient DEs.
Consider the initial value problem

y" + ay' + by =f), yO) =0, =0,

y'(0) (41)
where a and b are constants. The Laplace transform of this DE leads to the algebraic
solution
a F(s)
AG Facgs wre? +b
Thus, if we define the function
1
— 2 5 42
ae ea cot
the convolution formula yields the solution

yi)= [ar — 2)fla. (43)

Comparison of (43) with the solution obtained through use of the one-sided Green’s
function,

y(t) = I‘gilt, DAD dr, (44)

identifies the function k(t — 7) appearing in (43) as the one-sided Green’s function;
moe
KG ee 20 4kt, a): (45)

EXAMPLE 33 Find the one-sided Green’s function associated with the differential operator M =
Da 2D irs:

Solution We first determine

1 1
leks eae a taescaa |
= it Pai
eee ee =) ee ee

which leads to
1
k(t) = on On

Hence,
I
k(t — 7) = gi(t,7) = Bo se 77).
214 CHAP. 6 / THE LAPLACE TRANSFORM

Comparison of the convolution theorem of the Laplace transform and the

Green’s function method show that they are equivalent for constant-coefficient
equations. In fact, it now follows that the one-sided Green’s function for a constant-
coefficient DE can always be expressed as a function of the variable t — 7. The
method of Green’s function is more general than that of Laplace transforms,
however, since it can also be applied to variable-coefficient equations (at least in
theory) and is more readily adapted to problems when the initial data is prescribed
at a point other than ¢ = 0.

EXERCISES 6.7

In problems 1-4, find the Laplace transform of each convolution integral.

1. [« — u)sin2udu 2. [« — u)’e du
av 0

\3/ e*| et cosudu 4. [cost — u)sinudu

0 10)

In problems 5-10, find the inverse transform of each function using the convolution
theorem.

-5. ¥(s) = Guin 6. Y(s) = (s? + 1)?

ne ae! q :
Cuiel is a) SD ia Ing
Ga e
9. Ms) =
(ais 4) 10: 19 = gr
GS ee s
ashe a
“11. Establish the properties (37), (38), and (39) of the convolution integral.
12. Show that (f*1)() # f(A) in general.
Hint: Find a counterexample.
13. Show that (f*f)(1) is not necessarily nonnegative by letting f(t) = sint.
14. Given ¥(s) = (s + 1)/(s? + 4), explain why we cannot set F(s) = s + 1 and
G(s) = 1/(s* + 4) in the convolution theorem to evaluate y(t); 1.e., explain why

y(t) # [fe — u)g(u) du,

10)

where f(t) = L£'{F(s)} and g(t) = £7 '{G(s)}.

15. Given that £{f(p} = (s? — a*)~', evaluate [fe — u)f(u) du.
10)

16. Show that[Jo(1 — u)Jo(u) du = sint, where Jo is the Bessel function of the first kind
_ of order zero.
Hint ETD ls eye
17. Solve the integral equation* for y(t),

*The equation receives its name from the fact that the unknown function y appears under an integral.
SEC. 6.7 / THE CONVOLUTION THEOREM 215

y(t) = 4t - 3 yap sin( = Pam.

18. Solve the integral equation for y(t),

[ —u) *y(u)du = t"?.

0

19. Solve the integrodifferential equation* for y(t),

y') = [ ypeos —u)du, y(0)=1.

*20. Starting with f() = [Wo — Wo duxny > 0,
0

(a) use the convolution theorem to show

Fis) = LO) x+y

ise ’

where I’ denotes the gamma function (see problem 26, Exercises 6.2).
(b) Take the inverse transform of F(s) and establish the formula
Tol
; et — uy ws 'du = aS
bes .

21. Show that

15
lee bake ett 51

m — 1)!nte"""

22. Using the convolution theorem, show that the solution of the spring-mass system
my ky — J), (0) = Oey (0) = 0;
can be expressed in the form (@o = V k/m)

y = (ma)' Ihsin[wo(t — u)|f(u) du.

0

23. Using the result of problem 22, determine the response of the spring-mass system
when the forcing function is given by
(a) f(t) = P (constant).
(b) f(t) = Pcosat, w F Wo.
(c) f(t) = Pcos wot.
Determine the current i(t) in a single-loop RLC circuit when L = OHA Rs— 20s
24.
C = 1073 F, (0) = 0, and the impressed voltage 1s
DOA OSS
A Oo Ged

The system is characterized by

under both an
*The equation receives its name from the fact that the unknown function y appears
integral and a derivative.
216 CHAP. 6 / THE LAPLACE TRANSFORM

014 + 201 + 10° [cw au = E(t), i(0) = 0.

"25. Abel (1802-1829) studied a particular Volterra integral equation that has many im-
portant applications. In particular, suppose a particle of mass m is constrained to move
(without friction) along a certain path in a vertical plane under the influence of gravity
alone. Given the time T required for the particle to descend the curve, we wish to
determine the equation of the curve. This problem reduces to finding the solution f(y)
of the Volterra integral equation of the first kind
y f(u)du
T= Ky) =
0 V2e(y — w)
where g is the gravitational constant and f(y) is the length of the path.
(a) Take the Laplace transform of the integral equation and deduce that
K(s) = (a/2gs)'*F(s).
(b) Solve for F(s), and by taking inverse Laplace transforms, show that
9) 1/2

fO)= se [o — u) '?k'(u)
du.
7 0

Hint: Write F(s) = (2g)'(as)~?sK(s) and use Theorem 6.4.

26. Determine the one-sided Green’s function for the following differential operators
using the Laplace transform method.
(a) M =D?
(b) M =(D - ay
(c) M=D?+2D +5
(4d) M=(D -a\(D —-b), a#b
*27. Show that the one-sided Green’s function for an nth-order linear DE
V2 dnd > aly. agy =
is given by k(t — 7) = £~'{1/p(s)}, where
p(s) = 5 Fa, 2(s" * + + ays oh ae
is the auxiliary polynomial in the transform domain.
28. Use the result of problem 27 to determine the one-sided Green’s function associated
with the following differential operators:
(ale Mae D". n= 2: 304:
Ae
(b) M =D*(D? — 1)
(c) M=D*-1
(d) M=D?*-
6D’? + 11D —6

6.8 TABLE OF SOME LAPLACE TRANSFORMS

Listed on pages 217-218 is a short table of Laplace transforms and their inverses
that commonly occur in applications.
SEC. 6.8 / TABLE OF SOME LAPLACE TRANSFORMS 217

Table 6.1 Table of Laplace Transforms

F(s) = L{f(O} f() = £ '{F(s)}

Se aa)
(sod (Sue) u
S
——_—_————— (a # b)
(s)(s5 0)
1
— sinkt
k

cos kt

1
— sinh kt
k

cosh kt

1
—e sin bt
(s — ay + b* b

S Beene
ae il eee e” cos bt
(See a)’\+ b*
218 CHAP. 6 / THE LAPLACE TRANSFORM

Table 6.1 Table of Laplace Transforms (cont.)

F(s) = £{f(O} Fi) = LF}

1 1
19. e+) pl — cos kt)

1 1 :
20. x97 £ ptt — sinkt)

1 geen
PAN G+ aie — ktcos kt)

5 ks” nt on
225 (2 + ee pe ee <a pee Dian kt

Sa 1
D3 ware ye + ktcos kt)

S
(a* # b*) 5(cos at — cos bt)
b?— a

1
ae (sinktcosh kt — cos kt sinh kt)

1
| 26. ——— aye in Mt sinh kt

1
ee (sinh kt — sin kt)

1
1 (coshkt — cos kt)

(a > 0) h(t — a)

é; G(s) (a = 0) g(t — a)h(t — a)

31. F(s —a) e“f(t)
I ffx
3. =F(2) (a > 0) f (at)

33 2 Sal (S) 18 f (O) i eae af (()) °°

BAe Ne (5) (=f)
SS nC aan (geal) 6(t — a)

F(s)G(s) [re — u)g(u) du

REFERENCES

Churchill, R. V. 1958. Operational mathematics. 2nd ed. New York: McGraw-Hill.

Finney, R. C., and Ostberg, D. R. 1976. Elementary differential equations with linear
algebra. Reading, Mass.: Addison-Wesley.
REFERENCES 219

Kreyszig, E. R. 1979. Advanced engineering mathematics. 4th ed. New York: Wiley.
LePage, W. R. 1980. Complex variables and the Laplace transform for engineers. New
York: Dover.
Miles, J. W. 1971. Integral transforms in applied mathematics. London: Cambridge Uni-
versity Press.
Sneddon, I. H. 1972. The use of integral transforms. New York: McGraw-Hill.
Spiegel, M. R. 1965. Laplace transforms. New York: Schaum.
The first five sections of this chapter are devoted to solving linear systems of differ-
ential equations with constant coefficients. Several methods can be employed to :
solve such equations. :
In Section 7.2 we introduce a method that changes the system of equations to a f
single DE by successive elimination of the unknowns. The technique can be greatly i
systematized with the aid of Cramer’s rule. We use the Laplace transform in Section
7.3 to reduce the differential system to an algebraic system solvable by elementary
techniques and then apply the inverse transform to produce the desired solution set.
Another method, presented in Section 7.4, consists of assuming a particular trial
solution form that, when substituted into the system, reduces it to a linear system of
algebraic equations. Solving the algebraic system identifies the parameters occur-
ring in the trial solution set. Most of the general theory concerning linear differential
systems is also discussed in this section. In Section 7.5 we briefly discuss the use of
matrix techniques that permit the system to be expressed as a single-matrix DE ina |
form similar to that of a first-order linear equation. J
The qualitative aspects of certain kinds of nonlinear systems are discussed in Sec-
tion 7.6 without any explicit, formal solution functions. The general questions of con-
cern are mainly associated with the idea of stability of a solution. In particular, it is
important to know whether small changes in the input data (initial conditions) pro-
duce only small changes (stability) or large changes (instability) in the output (solu-
tion). When a nonlinear system is approximated by a linear system, it is equally im-
portant to know whether such an approximation is “reasonable.” That is to say, some
nonlinear systems do not iend themselves to reasonable linear approximations.

220
Wal INTRODUCTION

In applied mathematics, many applications involve not one but several dependent
variables, each of which is a function of a single independent variable, usually
time. The formulation of such problems leads to a system of differential equations
rather than a single equation, as we have studied thus far.
For example, the motion of a single particle in space governed by Newton’s
second law of motion leads to the vector DE

—_ =F, (1)
which is equivalent to the system of scalar DEs

mx" = Py,

my is, (2)
mz" = F3,
where m is the mass of the particle, atl he y, Zz) is the position vector of the
particle, and F = (F\, F, F3) is the external force acting on the particle. Systems
of equations are also used to describe the motions of coupled spring-mass systems
and RLC circuits connected in parallel. Examples of such systems will be given
later.
Systems of equations arise in the classical ecological problem of the prey and the
predator, which involves the struggle for survival among different species of
animals living in the same environment. One kind of animal eats the other as a
means of survival, while the other develops methods of evasion in order to avoid
being eaten. For example, suppose x(t) denotes the population of rabbits at any
time on an isolated island, and y(t) the number of foxes at any time on this same
island. The foxes eat the rabbits, and the rabbits eat clover, which is in ample
supply. When the rabbits are plentiful, so too are the foxes, and their population
grows. When the foxes become too numerous and eat too many rabbits, they enter
a period of famine and their population begins to diminish. The rabbits eventually
become relatively safe again, and their population increases, which also initiates
new increases in the number of foxes. Under the proper ecological balance, these
cycles of population increases and decreases can be repeated incessantly. On the
other hand, if the balance of nature is disturbed in the right way, both species could
die out.
Problems of this nature were independently modeled by A. J. Lotka (1880—
1949) in 1925 and by V. Volterra (1860-1940) in 1926. They suggested that the
instantaneous populations x(t) and y(t) of both species are solutions of the system
of equations
Xx
,
D682 16 Bes 3)

Vos ,
Gynt OXY.

where a, b, c, and dare all positive constants. The constants a and c represent the
growth rate of the prey (rabbits) and death rate of the predator (foxes), respectively,
221
222 CHAP. 7 / SYSTEMS OF EQUATIONS

whereas b and d are measures of the effect of the interaction between the species.
These equations are now widely known as the Lotka-Volterra equations.
Finally, another way in which a system of equations can arise is to convert an
nth-order DE to a system of first-order equations. That is, given the nth-order DE
(not necessarily linear)

VON R [tSey oe ete ele (4)

let us introduce the variables x,, x2, . . . , x, defined by

Lp = J t= Yo, ts SS ie a es
Hence (4) is equivalent to the first-order system of equations

X| = X2,

X2 = X3,
Be Pa ee (5)

Site Ane
Xe UL a ee re Ie

7.2 THE OPERATOR METHOD

Before proceeding with this section, it may be helpful to review the operator
notation introduced in Section 4.4.1.
Let us consider the simple system of equations
Key =. OF
y' + 4x =0, oI
or, equivalently (using operator notation D = d/dr),
Dx —y =0,

4x + Dy = 0. Om
Suppose we operate on the first equation by D to get
D*x — Dy = 0,
4x + Dy = 0.
Adding the two equations now eliminates y from the system. Hence,

(D* + 4)x =0
with general solution
X(t) ="Cncos 21+ Gorsini ce (8)
In a similar manner we multiply the first DE in (7) by 4, operate on the second
DE by D, and then subtract the results to eliminate x from the system. This action
leads to
SEC. 7.2 / THE OPERATOR METHOD 223

(D? + 4)y = 0,
from which we get
Wit) =1G a COSe2? a Ga sin. Zt: (9)

Although (8) and (9) represent solutions of the system (7), they are not solutions
for every choice of the constants C,, C2, C3, and C4. Therefore, it is necessary to
find a proper relationship between these constants. To do this, we substitute (8) and
(9) into the first equation in (7) to obtain

—2C, Sin.2i + 205 cost 2h — Cs Cost i— Cy sin 2b = 0;

or
(2Ce Gs) cos 2b = (2G Cy) Sin 2 = 0;

from which we deduce C3 = 2C> and Cy = —2C,. Hence, the solution of (7) is
given by
x (f) = Ci. Cos: 26.4 Cousin: 2h,
(10)
y(t) = 2C, cos 2t — 2C, sin 2t.

In order to systematize the procedure we just used, let us now consider the
general system of equations

L{x] + Ly] =f,

(11)
Lx] + Lay] =A0,
where L;, L2, L3, and Ly are any linear differential operators with constant
coefficients. Operating on the first equation by L,4 and the second equation by Lp,
and subtracting the results, we eliminate y from the system and are left with

(LiL, — Inks) x = 9\(), (12)

where g}(t) = Lal f,()] — L.{fx(d]. In a similar fashion, the elimination of x gives

(ils — I4L,) y = end), (13)

where g2(t) = LiLfa(d] — Ls[fi(O]. Equations (12) and (13) can now be solved
independently.
The above procedure can be systematized even further by formulating it in
determinant notation. Observe that if we write
L, I, fi@® 14
Ee ? (14)
fo) Ls

which is precisely what Cramer’s rule would yield for an algebraic system, then the
expansion of the determinants on each side reduces (14) to (12). However, it is
important to keep in mind that the determinant on the right-hand side has the proper
meaning only if the operators L, and L, operate on f(t) and f(t), respectively. That
is, the expansion of this determinant must produce the function g,(t) defined above.
Likewise, we can express (13) as
224 CHAP. 7 / SYSTEMS OF EQUATIONS

LT, Ly Ly fit)
(15)
Learns L3 fxd)
The use of Cramer’s rule to solve systems of equations can be applied to systems
of any number n of equations, regardless of the order of the operators L;, I2,. . .,
L,, . . . , Lon. From a practical point of view, it works best on systems of two or
three equations with either first- or second-order operators, since the amount of
computation becomes unwieldly for larger systems. Also, we might note that the
auxiliary polynomials for x and y are always the same, since the determinants on
the left-hand sides are identical [see (14) and (15)].Therefore, with the exception
of the arbitrary constants, the homogeneous solutions (complementary functions)
of x and y are identical. The number of independent arbitrary constants in the final
solutions should be equal to the order of operator determinant on the left-hand
sides. For a 2 X 2 system, for example, this is the order of L,\L4 — L5L3.

Remark. If L,L; — L,L3 = 0, the system either has infinitely many solutions or
no solutions, depending upon whether the determinants on the right-hand sides of
(14) and (15) vanish. Thus the situation here is similar to that of a system of
algebraic equations.

EXAMPLE 1 Solve the system of equations

xs Bp ay Paes
Ve = Oxe ly.

Solution We first express the system in operator notation

(DSN)
2y 0:

orem (De 2) ys 08

and then formulate as two determinant equations

Det eee aie =®
SP ole My vay Sl

Dele © hs) Sawer

Sie. Dr, 5 is eee
Hence, we obtain the two homogeneous DEs
(D?=33D)—4 4) =;
(Di= 3D .— Ayyi=.0,
with general solutions

x(t) = Cye' + Coe*,

y(t) = C3e' + Cye*.
 SEC. 7.2 / THE OPERATOR METHOD 225

Substituting these solutions into the first DE of the system, we have

Cer aCe = (Cr eC.
je rt (Catt 2Ci)e":
The constants must therefore satisfy the equations
C, + C; = 0,
3C, — 2C, = 0,
or C; = —C, and C, = 3C). Thus
Wii Cien 4 Ce

yi Cies 5Cre"

We might point out here that instead of solving both homogeneous DEs forx and
y, we could have solved only for, say, x, and then used the first DE of the system
in the form y = }(x’ — x) to find y. Such observations can be helpful in some
problems.

EXAMPLE 2 Solve the system of equations

X= DN a Vals

eye c

Solution In operator notation, the system becomes

Oat 2) ey es
—x +(D -2)y =?’,
which leads to the determinantal formulation

D—-2 = || eee — |
tery" AK, =) 4
D—2 =| Die eart
Shewd es es awe;?
After expansion of the determinants, we find
(D?— 4D +3)x =2?—2t + 1,
(D2? — 4D + 3)y = —2t? + 3¢.
The homogeneous solution for x is
Xy(t) = Ge ap Ge

while a particular solution is found to be

 ‘

226 CHAP. 7 / SYSTEMS OF EQUATIONS

Se ee
AU Pea yom
Hence,

(CS ip (Eee)
1 2 11
= Cie
Cie’ + Cre
Ce* + 3!
or + =t
9! + —.
7

Rather than solve the nonhomogeneous DE above for y, we merely observe that
y =x' — 2x — t, obtained from the first DE of the system. Therefore,

he yal BE
2
ae fp
g
at pence!
16
y(t) Cient Cee 3 9 7
——————

EXERCISES 7.2

In problems 1-10, find the general solution of the system.

1. x= 2x —y 2. x’ = 4x — 3y
y' = 3x — 2y y' = 5x — 4y
3. x' = 4x — 3y 4. x' = 3x — 18y
y' = 8x — 6y y' = 2x — 9y
5. x’ =x— 4y 6. x’ = 3x — 2y
y’=xty Yo Seleoy
7. x’ = 6x — 5y 8. x’ = 3x + 2y
y t= erry yas Ix ay
9, x’ =x —y + 42 "10. x’ =-x+2
A eh Ga Pay Soars y'=-ytz
PgNec DBE* ers 2 =x ey
11. Find the general solution of
x' = ax + by,
y' = bx + ay,
given that b # 0.

In problems 12-28, solve the given system of equations. Make sure the proper
number of constants appears in the general solution.

12. x’ — 2x + 5y = —sin 2, x(0) =0 13. x=

2 y= 7
Vix
a 2y = ry (0) Vorcie eye aot
Vie 29S 22x ay ie’, xO) Eu 15.44 yt + 2y = 0
yi 4x + y =e", yO) =—1 Xe 3X 2y 0
16, =0
x5 +4 5% —-2y
—24 ty" + 2ye= 0
 SEC. 7.3 / THE METHOD OF LAPLACE TRANSFORMS 227

Via) Soe Ky by eyed”)

xo 2x + 2y"— dy = 0
18. (D* 3D +2)x (1D — lly =0
(i 20 te (Daly = 0)
19, (D* —4D. + 4)x + (D?2.+.2D)y =0
(D? — 2D)x + (D?+ 4D + dy =0
20. x' +x hy = y = 2 22 OX oy eS
Sc + 2y = 1 2x’ + 3x + 2y' + By =2
225 Xk — 2x ey —4y =e" 23. 4% +7 + 2y = sint
ty hy e-* ee
rye ye)
24. (2 -1)x+(D
—- Dy = 1 25. (D?—4D + 4)x + 3Dy =1
(D2) aan (Deel) Ve (D2)
(Da 2)ye— 0
p20, Mle ee Vt
7 ote =) t2i. x —6y= 0
oltre 2ieeeay ee ot Oz.=0 Keaa Veit 70)
oe Nee Vale ee () Dy tee)
(20, e HF oe=res
= x toy bz’ =0
Sot 2Y ize aes

In problems 29-31, show that each system is degenerate in that it either has no
solution or infinitely many solutions.

29. x TV Sy —e- 300) xk St oy Sy = Sint

y aey tt = & 8 x ex + y = 2e’

ie, tO eka, lO
x exe
yy =-0

7.3. THE METHOD OF LAPLACE TRANSFORMS

If initial conditions are prescribed along with a system of differential equations, the
Laplace transform method is generally more convenient to use than the operator
method of the last section.
The Laplace transform applied to a differential system reduces it to an algebraic
system in the transformed functions. Standard solution techniques can then be
applied to solve the algebraic system, and the inverse transform of the algebraic
solution functions will produce the solution functions of the original differential
system.
nn nn UIE IIEEEEEEIUIEEStnSa a

EXAMPLE 3 Solve the differential system

228 CHAP. 7 / SYSTEMS OF EQUATIONS

subject to the initial conditions

x(0)=0, x'(0)=0, y(0)=0, y'(0)=0.
Solution If we let X(s) = L{x(n} and Y(s) = L{y(H}, then application of the Laplace trans-
form to each equation results in the pair of algebraic equations

s*X(s) — X(s) + 5s¥(s) = 5,

2
s°Y(s) — 4Y(s) — 2sX(s) = are

Upon simplification, we have

(s* — 1)X(s) + 5sY(s) =

Sl >

~25X(s) + (0?= 4K) = ==,

the simultaneous solution of which is given by
lls? —4
yest ee ee
(s) ss? + Is? + 4)’
HONE —25? US + 4 2.

s(s? + 1)(s? + 4)"

In order to invert these expressions, we first expand the right-hand members in
_ terms of their partial fractions

XG)(s) =e§e.1 5
gg? eT
—4
st Ae

Y(s) =. = = ohcan
S ge se Il Ger agai
Now,

x 1
x(t) = -Y pees
i {| si SL Be1 {=1 Fan = I 2
‘} 2 {=

Gland
Sine fi oasin were
and, similarly, we find

2r. y(t) = | — 2 cost + cos

eiee ee

7.3.1 Coupled Spring-Mass Systems

One elementary application leading to a system of DEs involves

the coupling of
spring-mass systems. Suppose two masses m, and m are connected
to two springs
 SEC. 7.3 /THE METHOD OF LAPLACE TRANSFORMS 229

Figure 7.1 Coupled spring-mass system.

having spring constants k, and k,, respectively (see Figure 7.1). Let y; and y2 denote
the vertical displacements of the masses from their equilibrium positions. When the
system is in motion, the stretching of the lower spring is y. — y), exerting a force
of k,(y2 — y,) on the upper mass m, (Hooke’s law). Also, the upper spring exerts
the force —k,y, on this mass, and so, invoking Newton’s second law of motion, we
have

my = —ky, + kly2 — yi) (16)

describing the small motions of the upper mass m,. In a similar manner, it follows
that the equation of motion for the lower mass my is
M2 = "KA¥o = i): (17)

Hence, the coupled spring-mass system in the absence of damping effects and
external forces is governed by the system of second-order DEs

mMy\ (+ (k,
(ky + ky)y,2)Y1 — kyr
22 = 0,7 (18)

my, — key, + kay2 = 0.

ct i a ea ns ne Snr EEE

EXAMPLE 4 Determine the free motions of a double spring-mass system composed of two unit
masses and springs with constants k, = 6 and k, = 4. Assume that both masses
unit initial
start from their equilibrium positions, but that m, has a downward
velocity and m, an upward unit initial velocity. Neglect damping.

proper
Solution The system here is described by Equations (18) above, and when the
parameter values are written in, we obtain
Vt LOV gee: Ayo 0,

yz — 4y, + 4y2. = 0,
230 CHAP. 7 / SYSTEMS OF EQUATIONS

with initial conditions

y(0)=0, yi()=1, y(0)=0, y2(0) = -1.

Using the notation Y\(s) = L{y,(H} and Y2(s) = L{y2(1)}, the transform of the
differential system reduces to
s*Y¥\(s) — 1 + 10Y,(s) — 4Y,(s) = 0,
s*Y,(s) + 1 — 4Y,(s) + 4Y,(s) = 0,
or
(s? + 10)Y,(s) — 4Y3(s) = 1,
—4Y,(s) + (s? + 4)Y.(s) = —1.
Solving these equations simultaneously, we find
2
Visa
Grader pay
. s* +6
ISG)
(S2+ti9 Wis? 125
The partial fraction expansion of these functions yields
Sie 6/5
OES s?+2. 52+ 12’
DYES 3/5
YS
Oy eee tea ot
and taking the inverse transform of each expression gives

y(t) = oe sin V2t + V3 sin 2V3t,

y(t) = ea WO n= V3 sin IN Si.

Hence, each mass is governed by the superposition of two harmonic motions.
I,2 Uy ee ee

7.3.2 Electrical Networks

An electrical network composed of components connected in parallel also gives

rise
to simultaneous DEs. For example, the RLC circuit in Figure 7.2
has two loops.
By Kirchhoff’s voltage law (Section 5.4), we obtain

Li{ + Ri, = E(2),

Cau “# Ris — 0, (19)

where i), i2, and i; denote the current in each part of the
circuit. However, by
Kirchhoff’s current law,
 SEC. 7.3 / THE METHOD OF LAPLACE TRANSFORMS 231

Figure 7.2 RLC circuit.

1 = 1» ap 13,

and thus we can eliminate i; by writing i; = i; — i, and then (19) becomes

Li; + Ri, = E(t),

(20)
ie iNoe s
in + pa lh — i) = 0.

EXAMPLE 5 LetR = 1009, L =4H,C = 10% F, and E(t) = 100 V in the network of
Figure 7.2. If initially the currents i, and i, are both zero, find the currents at all
later times.

Solution Substituting the appropriate values of the parameters into (20), we have
4i; + 100i, = 100,
i, + 100 — i) = 0,
with i,(0) = i,(0) = 0. The Laplace transform applied to each equation gives (after
simplification)
25
sI,(s) + 251,(s) = rae

—100/,(s) + (s + 100)/,(s) = 0,

where J,(s) = L{i,(t)} and 1,(s) = L{i.(D}, and solving these equations, we find

IY Se age) ERO Gee

2500 1 50 1
LIM iatiie( 50% s 7% +50" 5s +507

Hence,
HG) ea SOR Oe
pusala Clem ea,
232 CHAP. 7 / SYSTEMS OF EQUATIONS

and

10) = 1(f) et) = ote ee

Observe that both 7, and i, approach a constant unit value as t—™, but that i
tends to zero.

EXERCISES 7.3

In problem 1-14, solve the system of equations by use of the Laplace transform.

1. X=
— ey (0) 0 2. x’=y, x(0)=
y =2x, yOu I y'’ =x, y(0)=0
3. c= yx)= —I 4. 25 4r 2ysex 0)= 2
V part
oe ¥en WU) ae Y= ori 2y, yO)
= 2
59) AP ea Na era | 39(8) Pe
y = 2x +33y +e") yO) =.0
x’ — 44 +2y =e", x(0)=
Va Ox
ly it (0)
=0
2X VSy= ty x(0)=
ictal oemmy (\e—10)
Dea Via re= seer (0) =O
AC heya 13y =92 ay (Oy 0
Xe!xe yu= Ox (0) =0.- x'(0) =
yoriye«= 0, .y(0)=:0, Fy (0), =
x) ay + y" ="6 sin) x0) 0
Lia 2X — 0% Sy(O)e=nOey (0)
= Onn oyOn=1G
Lee Dey
fat 6 ey (en ame (0) nO
A ay) Gye ae (0)
Kiet Oy tery (1), ox(O)=0 er (0) 0
yo Ay = 2x!’ = 0, —y(O) = 0, y'(0)=0

oO, OS %7<—2
where fo = {0 Aas

13. Solve the coupled spring-mass system described by (18) when m, = m = 1,

k,;= 3, kz = 2, and the initial conditions are prescribed by y,(0)= 0, yi(0) = 1,
y2(0)= 1, y2(0)= 0.
14. Solve the coupled spring-mass system described by (18) when m, = 4, m. = 2,
k; = 8, ko = 4, and the initial conditions are yi(O) = 0, yi(O) = 0, yO) = 0,
yx(0) = —2. What are the natural frequencies of the system?
15. Solve problem 14 if a force f(t)= 40 sin 3¢ is suddenly applied to m; while the system
is in equilibrium.
16. Solve problem 15 if the force is applied to mz instead of my.
SEC. 7.3 / THE METHOD OF LAPLACE TRANSFORMS 233

17. Show that the spring-mass system in the accompanying figure is governed by the
system of equations
yi + 2ky; — kyo = 0,
y2 + Qky2 — ky, = 0.

Problem 17

18. Solve the system of equations in problem 17 if k = 3 and the initial conditions are
yi(O) = 1, yi(0) = 3, y2(0) = 1, y2(0) = —3.
19. Find the currents i;, i2, and i3 in the network in Figure 7.2 when R = 1000, L = 1
H, C = 10°* F, and E(t) = 100 V. Assume that i, and i are initially zero.

20. Find the currents i;, i2, and i; in the network in Figure 7.2 when R = 100 Q,L =4
H, C = 10°* F, i(0) = i,(0) = 0, and
(ay) EG = 100er Ov: (b) E(t) = 100sin607tV.
21. Show that the network illustrated in the accompanying figure is governed by the
system of equations

i,(t)

L,

R,

Problem 21
234 CHAP. 7 / SYSTEMS OF EQUATIONS

Ry, = Lis ar Roiz _ E(t),

Ri, + Lois = E(t).

Eliminate i; and rewrite the system of equations in terms of i; and i2 alone.

22. Solve the system of equations in problem 21 when R; = 6 0, R2=5 0, Li =
L, = 1H, E(t) = SOsint V, and i, and iz are initially zero.
*23. A double pendulum oscillates in a vertical plane under the influence of gravity alone.
For small displacements, it can be shown that the equations of motion are given by
(m, + m2)bi0! + m2b1b205 + (m: + m2)big6, = 0,
m2b303 + m2bb261 + m2b2g0, = 0.
Find the natural frequencies of the system when (assume g = 32)
(a) m=m=1, b =b=1.
(b) m=2, m=1, b=h=
(c) m=1, m=2, b=h=
(4d) m=m=1, b=1, bh
II = ll | slrede
(e) m nN Sy i) > nN tee
eS

Problem 23

*24. Solve the system of equations in problem 23 when the initial conditions are
|
9(0) = 0, 60) =0, 6,(0) = 5 (0) = 0,
and m, = 3, m2 = 1, and b, = bo = 16.

7.4 FIRST-ORDER LINEAR SYSTEMS

In this section we wish to discuss some of the general theory associated with linear
systems involving first-order DEs. Although the results can be generalized to
 SEC. 7.4 / FIRST-ORDER LINEAR SYSTEMS 235

systems of n equations, we will find it convenient to develop the theory for systems
of two equations. Therefore we will be interested in the general nonhomogeneous
system
x’ =ay()x + any + fd,
| (21)
y= an()x + an(ty + fr).
By the associated homogeneous system we mean

x' =ay()x + anldy,

(22)
y' = ay (tx + andy,
which is obtained from (21) by setting f,(t) = f,(t) = 0. The point of view presented
here will emphasize the similarities between such systems and the linear second-
order DEs discussed in Chapter 4. That such similarities exist is a consequence of
the fact that all second-order linear DEs can be expressed as a system like (21).

EXAMPLE 6 Reduce the second-order DE my" + cy’ + ky = f(t) to a system of first-order

linear DEs.

Solution Write the DE as

"
ic Vi
k
pas + ff)
t '

m m m
and then let y = x,, y’ = x2. Thus, since xj = y’ = %2, we have

x} = X2,

c k fd)
SO hay ier Noel 5) meme

Most linear systems that are not first-order systems can also be rewritten as
first-order systems but will generally involve more than two equations. To accom-
plish this it is usually best to solve for the highest-order derivative in each of the
unknowns before introducing new variables. Let us illustrate with an example.
een ey ele wenn Ae ae es ee

EXAMPLE 7 Reduce the system

= Vie,
Yk
—Ox ity: II ~~ S

to a system of first-order DEs.

Solution Let us rewrite the system as

236 CHAP. 7 / SYSTEMS OF EQUATIONS

x" — x! te y! Gs en

y=
6x + 4,
and introduce u = x, v = y, w = x’. Thus, uw’ = w and

w’ = 2w + v! — e%,
vo' = 6u +f,
or
u'=w,

v' = 6u +4,
w' = 6u + 2w +t — e”.

One of the reasons we are so interested in first-order systems is that most

computational algorithms in numerical techniques are established for first-order
equations, and in most practical applications the problem ultimately requires some
numerical calculations. A clear understanding of the general theory is therefore
desired to help facilitate the use of these numerical methods.
By a solution of the system (21), we mean simply a set of functions {x(), y(O}
with continuous derivatives that simultaneously satisfy both equations of the system
identically on some intervala St <b.

Theorem 7.1 | If the functions a;;(0), ai2(t), an,(0), an(t), fi(), and f,(1) are continuous in some
interval a S t < b containing the point fo, then there exists exactly one solution
pair {x(t), y(t)} of the system (21) such that x(t) = xo, y(to) = yo.

Theorem 7.1 is the basic existence-uniqueness theorem for the system (21),
which is comparable to Theorem 5.1 for a single DE.
The superposition principle, which is so important in the study of linear ho-
mogeneous DEs, also applies to systems of DEs as stated in the next theorem. The
proof is left as an exercise (problem 33).

Theorem 7.2 | /f {x,(t), yi(} and {x2(2), y2(t)} are both solution sets of the homogeneous system
(22), then the pair

X(t) = Cixi) its Goxst)y

y(t) = Cy) + Cry,(0),

is also a solution for any constants C, and C).
 SEC. 7.4 / FIRST-ORDER LINEAR SYSTEMS 237

EXAMPLE 8 The function sets {e~', —e‘} and {e*, }e“} are each solutions of
yh Aa
yous On ar 2s
Verify that x(t) = Cye~! + Cre” and y(t) = —Cye ' + $C e™ are also solutions.

Solution Direct substitution of x(f) and y(t) into the first DE gives
x — x = 2y — —Gier ot 4G se* — Ce * = Coe 4°2C
ie | — 3Che™
= 0.
and, similarly, for the second DE,

pray = Fire OC Ce = 3036 Cie 3 C2e*

= 0.
e
rere en ee eg S

Definition 7.1 | The solution sets {x,(t), y\(t)} and {x(t), yo} are said to be linearly dependent
on some interval a <t < b if and only if there exist constants C, and C2, not both
zero, such that

C,x,(t) + Cox2(t) = 0,
Ciyi) + Cry2(t) = 9,
Cy = 0, we
for every t in the interval .* If these relations are true only for C, =
say the solution sets are linearly independent.
ek e ee ee
e
se
dependent on every
EXAMPLE 9 Show that the sets {e”, 2e’} and {—4e*, —8e'} are linearly
interval, while {e*, 2e"} and {—4e”, e’} are linearly indepen dent on every interval.

Solution In the first case, we find

Cie™ = 4C,e* =U,

2C,e' mF 8Ce' = 0,

C; = 4C. Hence the first

which is satisfied for every choice of C; and C, such that
ns, the conditions are
pairs are linearly dependent. For the second set of functio
Ce" a 4C,e* — 0,

2C,e' + Coe’ = :

the sets are linearly indepen-

These conditions imply that C; = C, = 0, and thus
dent.
ee eee neice, eee Se

are linearly dependent if one set is a constant

*In the case of only two solution sets, we can also say they
multiple of the other.
238 CHAP. 7 / SYSTEMS OF EQUATIONS

As acriterion for linear independence concerning a single DE, we introduced the

concept of a Wronskian. Here once again we have a similar criterion for systems
of equations, which we state in the following theorem without proof.

Theorem 7.3 If {x,(2), y,(t)} and {x2(t), y2(t)} are solution sets of the homogeneous system (22)
on some interval a <t <b, a necessary and sufficient condition that these
solution sets be linearly independent is that the Wronskian

Wi =
X(t) x(t)
Oe pate
never vanishes ona St Sb.

Remark. As was the case in Chapter 4, the Wronskian defined in Theorem 7.3
can be shown to be either identically zero on the interval a < t < b or never
vanishing there. Also notice that our present definition of the Wronskian does not
involve differentiation.

Based on the above results, if {x,(4), yi()} and {x2(), y2(} are linearly indepen-
dent solution pairs of the homogeneous system (22) on some interval a <t <b,
then we define a general solution of this system to be the solution pair {xj (t), yy (O},
where

yD) = (Gan) + Cra),

(23)
yu(t) = Ciyi@) + C2y2(2),
and where C; and C; are arbitrary constants.

EXAMPLE 10 Verify by the Wronskian test that the solution sets {e~', —e~} and {e*, 4e* given
in Example 8 are linearly independent, and write the general solution of the system

Xe tx ae Dy
y = 3x0 2y.

Solution The Wronskian of {e', —e~‘} and {e*, 3¢4% is

Cones 5
W(t)(2) = mea Sei eft = —e*
We ?

which is never zero. Hence, the solution sets are linearly independent, and from
Equation (23) the general solution is

Xu(t)= Cie + Cre“,

yu(t) = -—Cye'+ 5Cre
ge
 SEC. 7.4 / FIRST-ORDER LINEAR SYSTEMS 239

Finally, with respect to the nonhomogeneous system (21), we state the following
theorem, which is analogous to Theorem 4.9. The proof also parallels that of
Theorem 4.9 and is left to the exercises (problem 34).

Theorem 7.4 If {xp(t), yp()} is any solution pair of the nonhomogeneous system (21) and {xy (0),
yx (t)} is the general solution of the associated homogeneous system (22), then the
general solution of (21) is given by {x(t), y(t}, where
x(t) = Xy(t) + xp(t) = Cixi(t) + Crx2(t) + xe,

y(t) = yu(t) + ye(t) = Ciyi(t) + Crya(t) + yet).

7.4.1. Homogeneous Linear Systems with Constant Coefficients

In this section we wish to confine our attention to the homogeneous system

U
Ke Cha Gi),
4
VS Oat ays 2)
where the coefficients are constant. In order to solve this system, we will steal an
idea from the solution technique used in solving a single DE with constant
coefficients. That is, let us assume that a solution of (24) exists of the form
x(t) = Ae, y(t) = Be™, where A, B, and A are constants yet to be determined. The
direct substitution of these trial solutions into (24) yields
Ade = a,,Ae™ = apbe.

Bhe™ = ay,Ae™ + ayBe™.

After dividing by the nonzero factor e™ and rearranging the remaining terms, we
are left with the algebraic system
Gi NA apB = 0,
(25)
aA + (ax — AB = 0.

We wish to find values of A such that the system (25) has a nontrivial solution
is possible if and
for A and B, i.e., where A and B are not both zero. This situation
only if the coefficient determinant

Aca) = |" a
=A. a\2
(26)
fig —

vanishes. Setting A = 0, we are led at once to the quadratic equation

+ (Ax — 2421) = 0. (27)
dN? — (ayy + am)A
or the characteristic
We refer to (27) as the auxiliary equation of the system (24),
equation, as it is sometimes called.
roots. If x(t) = Ae™,
The roots A; and A, of (27) are called the characteristic
A must be one of these roots.
y(t) = Be”™ are to be solutions of the system (24), then
240 CHAP. 7 / SYSTEMS OF EQUATIONS

For A = A,, we substitute this value back into the algebraic system (25) to obtain
the nontrivial solutions A, and B,. Hence, with these values of A, A, and B, we have
one solution set {x,(1), y,(4)}, where

x(t) = Aje*",

yi() = Bye. Ca2

Obtaining a second linearly independent solution set corresponding to A = A, will
vary depending upon whether the characteristic roots A, and A; are real and distinct,
real and equal, or complex conjugates.

Case I—Distinct real roots: When the roots A, and A, of the auxiliary equation
(27) are real and distinct, there are two linearly independent solution sets corre-
sponding to {A,e*", B,e*"’} and {A,e*”, B,e'} (see problem 25). The constants A,
and B, are found in a similar way as A, and B,, but this time the value A = A; is
substituted into (25). The general solution in this case is therefore
x(t) — GAler ota C,A,e*,

y(t) = C\Bye*" + C,Bye*. Ce

EXAMPLE 11 Solve the system

=e +82 ys,

y = 3x 4+ 2y.

Solution Evaluating A, we have

AQ) =e [=U -ae- 1-6-0,

3 2}
or
A*—3A —4=(A — 4)(A + 1) = 0.
Hence, A, = —1 and A, = 4. For A, = —1, the system of equations (25) yields

2A, + 2B, = 0,
3A, + 3B, = 0,
which shows that B, = —A,. For A; = l, a possible solution pair is {e~', —e~4.
Likewise, with A, = 4, we obtain from (25)

—3A, + 2B, = 0,
3A, — 2B, = 0,
3
or By = 3A). Thus, {e”, 3e*} is another solution pair, and our general solution
4 3 4 . . . . .
is
x(t) aa Gem q= Cre.
 SEC. 7.4 / FIRST-ORDER LINEAR SYSTEMS 241

y(t) = —C,e"' oF >Cre"

Case II—Equal roots: If the two roots A, and A, are real and equal, then setting
A = A, = Ay produces only one solution set of the system. By analogy with
second-order DEs, we might expect to find a second linearly independent solution
set of the form
x(t) — Aste”,

y2(t) = Byte™.
However, this is not the case. Without providing the details, it turns out that we
must seek a second solution set of the slightly more general form

AD) = (Ast + Aje™,

yo(t) = (Bot + B3)e™. oo

This being the case, our general solution this time becomes
x(t) = rages = Cy(Aot =F Aaje”,

y(t) = Cpe + C2(Bot te Bae: OY)

Let us illustrate this situation with an example.

eese Se

EXAMPLE 12 Solve the system

Ke A aay
yx a 2y.

Solution Here we have

Ao = =I
= =e\ 24+ 6\)+2.9/— 0,
a) l pg
in (25), we see that
which has the double root A; = Az = —3. Setting A = —3
—A, a B, = 0,

A se B, = 0.

A possible nontrivial solution is A, = —B, = 1, from which we get the solution

n set
pair {e*, —e~*\. We now substitute the trial solutio
xt) = (Ast + Azle,

y(t) = (Bat + Bes

to
into the original system of DEs. Such action leads
+ Bse~*,
eA, — 3A3 — 3Act) = —4(Aot + Aj)e~* — (Bot
242 CHAP. 7 / SYSTEMS OF EQUATIONS

e-*(B, — 3B; — 3B2t) = (Ant + A3)e~* — 2(Bot + B3)e~*.

Equating the coefficients of like terms, we obtain
A, — 3A; = —4A; — B;,
—3A, = —4A, — B,
B, — 3B; = A; — 2B3,
—3B, = A, — 2B).
One possible solution of these equations is Ay = A; = 1, B) = —1, and B; = —2.
Thus a second solution pair is {(t + l)e~*, —(t + 2)e~*}. It is very easy to verify
that the solution pairs given here are linearly independent since |W(t)| = 1, and
therefore a general solution set is
x(t) = Cie + Ct + De,
y(t) = —Cye~* — Cyt + 2)e*.
———————————
e —— ae ee

Remark. It should be observed that the special case when aj); = a and
2 = a2, = O leads to the uncoupled systemx’ = a,x, y’ = ayy. Here {A,e™, 0}
and {0, B,e”} are linearly independent solution pairs. However, our general pro-
cedure outlined above is for coupled systems.

Case I1I—Complex conjugate roots: When the characteristic roots are complex
conjugates, A, = p + iq and A, = p — ig, we obtain two distinct solution pairs

x(t) = Aye?) x(t) = Ae?

yi(t) = Bye?*™, y(t) = Bye", (32)
where A, and B, denote the complex conjugates of A, and B,. (The verification that
the coefficients are indeed complex conjugates is left to the exercises.) Moreover,
these solution pairs are linearly independent, although they are complex. To obtain
real solutions, we proceed in a fashion similar to that in Section 4.4, where
we
previously obtained real solutions from complex solutions. Let A, = a, + ia, and
B, = b, + ib, so that A, = a, — ia, and By = b, — ib; Using Euler’s formulas

e = cos6 + isin@, e~ = cos@ — isin@,

and the superposition principle, we can write

1 l
3 ba) + x,(t)] = 5 (a1 + iaz)e”(cos qt + isingt)

+ Pa — ia)e’"(cos qt — isin gt)

= e?(a, cos qt — az sin gf)

 SEC. 7.4 / FIRST-ORDER LINEAR SYSTEMS 243

and
Le Aare :
sbi) + y,(t)] = e?(b, cos gt — b2 singt).

These sums represent another solution pair of the system involving only real
functions. In a similar fashion, we see that

= shu = x(t) | == 5(a + iay)(cos gt + isin qt)

+ 5a — ia)(cos qt — isin qt)

e”'(ay COs qt — a, Sin qt)

and

a sb. — y,(t)] = e?(b, cos gt + b, sin qt),

which is also a solution pair. The Wronskian of these solution pairs is found to be
W(t) = e?'(a,b> = ayb;).

For complex A, the constant B; cannot be a real multiple of A, and thus a,b, —
asb, # 0. These last solution pairs are therefore linearly independent, and the
general solution for this case can now be written as
cosgt — azsingt) + C(a,cos gt + a,singt)],
x(t) = e?[C,(a,
cos gt — bzsingt) + C(b2.cos qt + b, sin qt)].
y(t) = e? [C\(b, ee)

EXAMPLE 13 Solve the system

R= O85,
y’ = 5x + 4y.

Solution The auxiliary equation is

637 Nex Hl
|=27— 108 + 29 =0,
3) aA

with complex roots A; = 5 + Qi and A, = 5 — 2i. For A; = 5 + 2i, we must solve

Ct 21)A, 28 i.—.9,

then
Therefore, if we choose A; = a, + ia) = 1 and By =o; 4+ 10) = M27,
a, = 1, @ = 0, b; = 1, and bp = —2. The general solution then becomes
x(t) = e*(C, cos 2t + C, sin 21),
y(t) = eriC ie 205) COS 20 (2G) + Cs) sin 2].
244 CHAP. 7 / SYSTEMS OF EQUATIONS
§
Notice that it was not necessary to substitute A, into (25) for this case. (Can you
explain why?)

The results of this section can be generalized to systems of equations larger than
two. Let us illustrate with an example.

EXAMPLE 14 Solve the system

x II Xmas

y rz
Peace ie ee

Solution To obtain the characteristic equation, we can expand the determinant

ESE 0 1
0 tuaeN. 2 | =A? — 7A? + 6A =A(A — LIA — 6).
1 2 ae 7,"
Hence, we find A, = 0, A, = 1, and A; = 6. The substitution of A; = 0 into the
algebraic system
GloAACH= 0),
Cli HBr
2C* =/0,
At QBS iG AC =0,
yields

A, + C, =0,
B, + 2C, = 0,
A, + 2B, + 5C, = 0,
from which we obtain A; = 1, B; = 2, and C, = —1. Similarly, the substitution
of A, = | and A; = 6, respectively, into the algebraic system leads to A> = 2,
By= 1, Cyi=0; \Ag= 1) By= 2;eand Gy'= 5eiThe general solution of this
system is therefore
x(t) = C, + 2Cr,e' + Ce",
y(t) = 2G} = Cre! ote 266-3

z(t) = = = SGse a

7.4.2 Nonhomogeneous Linear Systems

We now turn our attention to the nonhomogeneous system of DEs
 SEC. 7.4 / FIRST-ORDER LINEAR SYSTEMS 245

'
x = ax + apy + f,(0),
(34)
y' = ayx + any + f(t).
For the solution technique to be discussed, the coefficients need not be constants,
although that is the only case we will solve. The general solution of this system
(Theorem 7.4) is the pair {xy(t) + xp(t), yu (0) + ye(O}, where {xy (t), yu (D} denotes
the solution set of the associated homogeneous system and {xp(t), yp(} is a
particular solution set yet to be determined.
A general method for finding a particular solution set is the variation of param-
eter method, which is similar to the technique previously given this name in Section
4.7. We seek a particular solution of the form
Xp(t) = uj(O)x\() + un(t)x2(0),
(35)
ye(t) = u(dyi() + ur(Dy2(d),
where we have replaced the arbitrary constants in the homogeneous solution by
unknown functions. The substitution of (35) into (34) gives
ae U5X2 = A U,X| se Q,\U2X2 55 Q\2uUy,y\ ae Q\2U2y2 hi
UX} == uyX) te UX)

uy, + uty, + Uayy + Uby2 = Ay UyX, + AyiMyxX2 + Ayu)

' / , amar
F Ar2U2y2 Hale:

Rearranging terms, we find

, 0
X\ X2

_—— ————————

UX} pa a UrX2 + 1X2. = uy(ayyx, + ayy) + Up(Ay 1X2 + a\2Y2) erty is

uy, + uy, + Uys + Wyy2 = Uy(arX1 + Axyy\) + U(X. + Ayy2) + fr»

——ESS a
Uy ,
yi y2

OT

Xu, + XW = fi,
Pe a

yu + yoy = fr.
Solving for uj and uw, we obtain
fi) x(t) xt) fi

RG) BU, —In® AO (36)

fe ey (e ee Wa
an defined in The-
where W(t) = xi(fy2(t) — x2(Dyi(/ is the nonvanishing Wronski
from which a particular
orem 7.3. Hence, a single integration yields u,(t) and u(t),
solution set can be constructed.

a
oe th ee ee e
ee

EXAMPLE 15_ Solve the system

1
x' = —4x + 2y Hp
246 CHAP. 7 / SYSTEMS OF EQUATIONS

' 2

Solution We first solve the associated homogeneous system. Thus we set

—fa—%) 2
, aes =)? + 5A = 0,

which gives A, = 0 and A, = —5. Setting A, = 0 in the algebraic system (25)

yields
—4A, + 2B, = 0,
2A, — B, = 0,
or A; = | and B, = 2. Also, the value A, = —5 leads to A, = 2 and B, = —1.
Therefore,

Nyt) = Cy + 20re57,
ViDi= 2G — Cree,
Assuming a particular solution of the form
Xp (0) = u(t) + 2u,(He—*,
ye(t) = 2u,(t) — u(de™,
we find

l
- Je
t

ete
u(t) = i = 8 + 1

1 e202 ae Mn
2 -e*

ae
t

Deas : ;
S(t = dash LF HoT«
walt) = Tos ca.
2 -e*
Therefore,

8
u(t) = sé eA Oe

i) =.=
4 St
U( ) 5 ’

giving us the particular solution set

 SEC. 7.4 / FIRST-ORDER LINEAR SYSTEMS 247

8 8
xp(t)
Xp(t) = <t
5 + logt -—55°

16
yp(t) = cz. Sle Peet ~.

Finally,
8 8
a)(0) S.C 1 2G 2€ bet
5 logt -—55°

- 16 4
y= 2Cp— Cre + au + 21097 a ar
e
a a S

EXERCISES 7.4

In problems 1-7, rewrite the DEs as first-order systems of DEs.

1. y"+k’y = Psinoat 2) 4y + ye

3. y®—-kty =f 4. ty"+ty'—y =?’ logt

ihe bode phy pie a Sh Ake Bes Se a
xy =f —410 x'—y'=e™
We myi te (ky + ka) y1 ae Kay =0

moy2 — ky: + kay2 = 0

. In problems 8-20, solve by the method of this section.

Oa 4s Sy. 10. x’ =4x — 3y

Se ed ce i aeMe
y'’ = 3x — 2y y’ = 5x — 4y y' = 8x — 6y

eX ey AD ey 13.0 x 3x Ly
41
va = het y Vir 2s Sy
y Oxy
14. x' = 6x — 5y 15.0 + -2y
x! 33x 162 x) = 3x = I8y
Veoh oy yi = Se aetery by aaa OY

AZo 2 Sy 182 tay ez 19% Hay

View 2k Vie Vian Xa Vem ig
y' = 3x + 4y
z' = —3x + 2y + 42 yo! = ay ap 2

20 ee = ee
ia oii),
Ft ea ee

of functions is linearly de-

In problems 21-24, determine whether the given set
pendent or independent using the Wronskian test.

21. {e%, 2e*} and {—3e%, —e"} 29. {e*, 2e*} and {—3e*, —6e"}

23 ieee + Sie, andl {=e 0c — 8re'}

248 CHAP. 7 / SYSTEMS OF EQUATIONS

*24. {1+ t.—2 4 27,4 + 211, —20. 4}, and {3 4 °2t, Gi ar 2 rar|
25. Show that the solution (29) is composed of linearly independent solution sets.
*26. Show that two of the characteristic roots of
Gd ie had A mH
Vesey es,
AE nD Wacay he we
are the same but that three linearly independent solution sets can be found which are
all of the form {Ae*’, Be”, Ce*}.

In problems 27-32, solve the nonhomogeneous system of equations.

27. X= 2 ye— Sy = sin 2s 28. x'=2x-yt+t

Vix Dy ft y =3% —2y + 2
29. ge ORS Fen ap! 30. x’ =4x —2y +e’
y= yee y =sx bly —t
31. x’ =x + 2y 32. x’ = 2x — S5y + csct
y= 5x + y + e’tant
Va k sary er SCer
*33. Prove Theorem 7.2. *34. Prove Theorem 7.4.
*35. Show that the complex solutions (32) do indeed have coefficients A, and B,, and A 1
and B,, all of which are complex conjugates.
*36. Show that the Wronskian associated with the homogeneous system
Uy
ae SS ay (t)x + a)2(t)y,
i
y = an (t)x + an(ty,
is given by Abel’s formula

W(t) = Cexp {[leno + att) Jar},

where C is a constant (see also Lemma 4.1).

[O] 7.5 MATRIX METHODS

In this section we assume that the reader is somewhat familiar with matrices and
their basic operations, as taught in a beginning course in matrix methods or in linear
algebra. Those readers lacking this background can skip this section without loss
of continuity.
The theory and solution techniques discussed in the last section can all
be
formulated in terms of standard matrix operations. For example, the system itself

x' =ay()x + andy + fd,

, (37)
y = ay (Dx + andy + fd),
SEC. 7.5 / MATRIX METHODS 249

can be expressed as a single matrix equation

Y'=A(Y + Fd, (38)

where

A(t) = ay(t) ay2(t)

© (ee a ce
is the coefficient matrix and

x(t) fild)
Y(t)
@
=
2) a
d F(j) =
&3 fy
are both vectors.* [The form of Equation (38) suggests that the notation for a matrix
equation for a system of n equations is as simple as that for a system of two
equations. ]The use of vectors and matrices not only is notationally expedient but
also facilitates calculations and emphasizes the similarity between systems of
equations and first-order linear (scalar) equations. We say that the vector Y is a
solution of (38) provided it is differentiable and its components satisfy the system
of equations (37).

7.5.1 Homogeneous Equations

equation
When F (1) = O in Equation (38), we obtain the associated homogeneous
V = Aipy. (41)
Suppose we know that
42

oe
d Yy = 4

aes ae)
= a8

SE
(1)

ey (
e, it follows that
are both solutions of (41). Then, from the superposition principl
the linear combination
Y = CY + GY (43)
is also a solution of (41) for any constants C, and C).
Let us now introduce the matrix
x,(t) x(t) 44
=
ie aay
eu co)
say that Y and Yo
whose columns are the solution vectors Y“) and ¥. We then
7.1 in Section 7.4). How-
are linearly independent if det (X) # Of (cf. Definition
in Theorem 7.3, 1.e.,
ever, this determinant is also the Wronskian defined
det (X) = W(t). (45)

ngeably in the literature on linear algebra.

*The terms vector and column matrix are used intercha
+det (X) means the determinant of X.
250 CHAP. 7 / SYSTEMS OF EQUATIONS

Thus, if we have that Y“’ and Y™ are indeed linearly independent vectors on some
interval / and are also solutions of (41), then a general solution of the homogeneous
equation (41) on J is given by (43).

7.5.2 Linear Equations with Constant Coefficients

We now restrict our attention to the linear system

Y’ = AY, (46)
where A is a2 X 2 matrix with constant elements. By analogy with our solution
treatment of second-order constant-coefficient equations in Section 4.4, we seek
solutions of (46) that are of the form

ME ad a (47)
where A and the constant vector E are to be determined. The substitution of (47)
into (46) leads to

Ane 4 AKeo
which can be rearranged in the form (after canceling the common factor e™’)

(A — ADE = 0, (48)
where

I = (as)
hw)
(49)
49

is the identity matrix. Now from the general theory of matrix equations, we know
that (48) can have a nontrivial solution if and only if

det(A — AI) = A(A) = 0. (50)

Thus we are led once again to Equation (27) in Section 7.4.1 for finding the
characteristic roots A, and A).
In the matrix notation given here, we recognize that finding the characteristic
roots of (50) is simply finding the eigenvalues of the constant-coefficient matrix A,
and the eigenvectors of A are then used to construct the nontrivial solutions
obtained from (47). Let us formalize these statements in the following theorem.

Theorem 7.5 For each eigenvalue A; of the constant-coefficient matrix A and each eigenvector
E” belonging to A,, the function
Y = Ee%!
is a solution of the homogeneous matrix equation

Y’ = AY.
 251
SEC. 7.5 / MATRIX METHODS

e
eee re eee ee S

EXAMPLE 16 Solve the system

by matrix methods.

Solution (We previously solved this system in Example 11 of Section 7.4.1 by conventional
methods.) Here we assume Y = Ee”, which leads to the system of equations
ae 2D C1a\ ee 0

3 2 eA €7 0 ‘

Nontrivial solutions of this equation are possible only if

La" x 2
J neat Se
ase ee
Shae ee nA
This equation has roots A; = —1 and A; = 4, which are the eigenvalues of A. For
A, = —1, the above system of equations reduces to the single equation (after
simplification)
e\ aie = 0.

can be
Thus e, = —e,, and the eigenvector corresponding to the eigenvalue A,
represented by (any constant multiple will also suffice)

e®=(1)).
= 1

ctor
In the same way, corresponding to A, = 4 we find the eigenve

()
E =
2 2

system of equations
Based upon these results, the corresponding solutions of the
are
yo = BE Me, yy = Eee)

which lead to the general solution

1 2.
Y=C (Gi Oem (3)eae

where C, and C, are any constants.

i.e., if it is equal to its

Remark. If any n X n matrix A is real and symmetrical,
rs of A have important and useful
transpose, then the eigenvalues and eigenvecto Moreover,
eigenvecto rs will be real.
properties. In particular, all eigenvalues and in the case of
rs always exists , even
4 full set of linearly independent eigenvecto
252 CHAP. 7 / SYSTEMS OF EQUATIONS

repeated eigenvalues. There is no guarantee of these properties when A is not

symmetrical. For example, complex eigenvalues and eigenvectors can arise as in
Case III, discussed in Section 7.4.1.

The matrix formulation permits us to generalize the solution formula for a single
first-order (scalar) DE with a constant coefficient. For instance, the equation

yi ay (51)
has the general solution (see Section 2.4.1)
y = Ce”, (52)
where C is any constant. Likewise, the matrix equation

Y’ = AY, (53)
where A is a constant matrix, has a solution that can be represented by (where C
is any constant vector)
Y =e'C. (54)
Of course, we must now define the exponential matrix function e“’ so that (54) is
meaningful (see problem 11). This can be done, but the use of this matrix function
as a practical means of solving systems of DEs would necessitate a fairly thorough
knowledge of matrix theory. Such knowledge is considered beyond the scope of
this text.

7.5.3 Fundamental Matrices

_ The theory of systems of equations can be further enhanced by introducing the

notion of a fundamental matrix. Suppose that Y“” and Y are linearly independent
solutions of
Y' = A(OY, (55)
where A(t) is not necessarily a constant matrix. We then define the fundamental
matrix

Wr) = (ie z)
yi(t) y(t)
(56)
whose columns are the vectors Y‘’) and Y. Observe that W is nonsingular, i.e.,
det (W) # 0, since this determinant is also the nonvanishing Wronskian W(t). In
terms of the fundamental matrix, the solution of (55) is simply

Nea) C. (57)
where C is an arbitrary column vector.
In the case of an initial value problem, we seek a solution of (55) such that

Y(0) = Yo, (58)

 SEC. 7.5 / MATRIX METHODS 253

where Yo is a prescribed vector. Imposing this condition on the solution (57), we

find
W(O)C = Yo, (59)

which has the formal solution

C = W'0)Yo, (60)
where W>! denotes a matrix inverse.* Hence the solution of the initial value
problem described by (55) and (58) is formally given by
Y = WNW '(O)Yo. (61)

—
ee
a

EXAMPLE 17 Solve the initial value problem

eit
Y eee
=(3 sib vo = (0).

Solution Referring to Example 16, we know that

Yo = |
] ee. Y” = z Jes
(
=a 3

are linearly independent solutions of the equation. Hence, the fundamental matrix
is
Caer le.
Wi) < G =I 2) i

Next, we set tf = 0 to get

1
wor = (_1 ah det (W(0)) = 5,
from which we compute
a peli =a)

The solution we seek is therefore

*Recall from matrix theory that the inverse of a 2 x 2 matrix

A — ay,
(int a ‘|
a2, 422

is given by
— 1 ie nee d A) # 0
=1

« det(A)\—ax ai] ’ ny) ;

254 CHAP. 7 / SYSTEMS OF EQUATIONS

ea erier:
-( —e! ae
z Ben Qe
~ \—3e7 + 3e%] ”

7.5.4 Nonhomogeneous Equations

When the system of equations is nonhomogeneous, i.e., when

Y'’=A(QY + Fd, (62)

then a general solution has the form
Y = Y, + Yp, (63)
where Y,, is a general solution of the associated homogeneous equation and Yp is
any particular solution of (62).
To find Yp, we again call upon the method of variation of parameters. Thus we
assume
Yp = W(fu(d), (64)
where u(f) is a vector to be found. The substitution of (64) into (62) leads to

W'adu() + WOu'() = A(DW(Nu(D + Fd. (65)

However, since W is a fundamental matrix, it follows that

Wi) = AOW(D,
and therefore (65) reduces to simply
Wnu'(t) = F(d. (66)
Hence,

is69) (67)
and, upon integration, we deduce that

u(t) = |Wo (EF (d) dt. (68)

Putting this result back into (64) gives

Yp = Wr) |WF (1)dt, (69)

and so a general solution of (62) takes the form

Y = WC + Wa iW>'()E(2)dt, (70)
 SEC. 7.5 / MATRIX METHODS 255

where C is any constant vector.

Finally, if an initial condition
Y(O) = Yo (71)
is also prescribed, the complete solution of the initial value problem becomes (see
problem 21)

Y = W()¥'(O)Y, + WO) |‘WF(a)dr. (72)

0

EXAMPLE 18 _ Use matrix methods to solve the nonhomogeneous system

1

Solution (We previously solved this system in Example 15 of Section 7.4.2 by conventional
methods.) The coefficient matrix A has eigenvalues given by solutions of
det(A — AI) = A? + 5A =0.

Hence A, = 0 and A, = —5, and the corresponding eigenvectors are found to be

Be) ee)
@b) es 1 O) D

From these results we construct the fundamental matrix

[eileen
Wit) = (; es ;

The matrix inverse of W is

I 2
5 5)
Wa) = '
2 St I St
=e —-—e
5 5

so from (69) we obtain

yp =W) iWo '()F (1) dt

1 2 1
ot
= ¢ ze | ; 2 i dt
Rae St eet. Ot 4 46 =

5° ae
256 CHAP. 7 / SYSTEMS OF EQUATIONS

Sted
lecee | plead g ns
TA ee Ene 5
5

( ze spurns
=a t

Wy \oe=c ied
25©
8 8
5! t loge 75

16 4
et
es
clogt +
75
—

Therefore, from (70) the general solution is

& Me ipzenea YG, ar | 7

as +7 log f= a
Eran B25
BEE ENG? LSpdastygpheee ate .
5 SMS
or, equivalently,

EXERCISES 7.5

In problems 1-7, use matrix methods to solve the linear system.

1tgY, ,_-(3
(2 -1
ah 2aFN, ,_-(§
(4 -3
aly WY ,_ (4 3
(3 iy

In problems 8-10, find a unique solution of the system of equations satisfying the
prescribed initial condition.
SEC. 7.5 / MATRIX METHODS 257

8. Y
. — ¢cha ake
@| ¥(0) = (i)
1
9. Y Utpe &=3 tll
|
¥(0) =
(*)
9)

Diener ware 1
pLOgmevee ten Ogee |Ye Y (Oe =| 2.0
Lisaglee a
“11. By analogy with the Maclaurin series
>
: x2 eae
Oye ee Ne er ec
PA BN
we define the matrix exponential function

eee At ANIA i
GAGMe ge

where I is the identity matrix and A* = AA, A* = A(A’), and so on.

(a) Use the approximation
2
At — at
e“=I+At+A T

C4)
to compute the matrix exponential function when ¢ = | and

(b) Compute e“’ exactly for the matrix A given in part (a) when t = | and compare
the result with that for (a).
(c) Compute e*’ exactly for any t when

Use the results of problem 11 to solve the linear systems

a-(55):
*42.
pany Ont e/a 0
(a) Y -(i ae (b) Y =il, ake

sys-
In problems 13-18, use matrix methods to solve the nonhomogeneous linear
tems.

Re ps
(Ay oe ke(5 eA)
—sin2t nm =I t
13. Y -(i ie ; ‘= ats

ik WEIR Lyte h = Dae

=a ok ae! e' ye 4 2 e'

1
0
We: Se) Sapa 1 sel beater 18. vie Uh Ble
2
problem
19. Find a unique solution of the nonhomogeneous initial value
 oe

258 CHAP. 7 / SYSTEMS OF EQUATIONS

aS 10e~* oa
‘= + , YO)= :
uf ies “)¥ ec) ©) (2
20. Find a unique solution of the nonhomogeneous initial value problem

Y ee- (|
cree 3e!
bees ¥ (0) ay ih
©

*21. Verify that Equation (72) satisfies the initial value problem described by Equations
(62) and (71).

[O] 7.6 NONLINEAR SYSTEMS AND STABILITY

A DE provides the basic model from which a mathematical analyst pursues in-
vestigations into the existence, characterization, and construction of solutions to
physical systems. For a large class of problems, either the oscillatory nature of the
system under study or its stability is of greatest interest. But because finding the
solution of many DEs in a convenient and useful explicit form is either difficult or
impossible, we often resort to qualitative methods for much of the analysis, as we
did in Section 5.7 in studying the oscillatory characteristics of certain linear
systems. We now wish to use qualitative methods once again in studying certain
nonlinear systems, but with respect primarily to their stability.

7.6.1 The Phase Plane and Critical Points

Suppose we consider the initial value problem (primes denote derivatives with
_ respect to 7)

Kie= (Cenex) = ane (Ore (73)

where F is assumed to have continuous first partial derivatives with respect to x and
x'. If we put y = x’, then we can replace (73) with the equivalent system of
equations

tea yee Olea.

y' = F(x, y), y(0) = B. (74)
By a solution of (74), we mean a set of differentiable functions {x(t), y(t)}, which
on some interval containing t = 0 reduces (74) to a set of identities.
We can think of the functions x = x(t) and y = y(t) as parametric equations of
an arc in the xy-plane that passes through the point (a, 8). From this point of view,
it is customary to refer to the xy-plane as the phase plane of the system (74) and
the arc defined by its solution as a trajectory, path, or orbit. The direction of
increasing f is considered the positive direction along a given trajectory.
The trajectories of a system can be found by eliminating the parameter t between
the equations x = x(‘) and y = y(t). In other cases where a solution cannot be
found, the trajectories can be determined by forming the ratio y'/x', which leads to
 SEC. 7.6 / NONLINEAR SYSTEMS AND STABILITY 259

dy _ F(x,y)
(75)
dx yo oF
and then solving this first-order DE.

EXAMPLE 19 Determine the trajectories associated with the simple harmonic oscillator whose
governing equation is
mx takx
= 0.

Solution We first reexpress the DE as the first-order system

x'=y,

y
es aE

and then form the ratio y'/x’ to get

Li a
dx my
The general solution of this DE can be found by separating the variables, which
leads to kxdx + mydy = 0, or
kx? + my? = C,

where C is an arbitrary constant. By allowing C to assume various values, we find

the trajectories to be the one-parameter family of ellipses shown in Figure 7.3. Each
trajectory represents a possible motion of the system (depending upon the pre-
scribed initial conditions), and each point on a given trajectory represents an
instantaneous state of the system. The direction of the arrows in Figure 7.3 suggests
is so
that the representative point moves clockwise along a given trajectory. This
x(7) is increasing , and y < O implies that x(7)
since y = x’, and y > 0 implies that
is decreasing.

y
A

(Ww y(t))

NZ, ‘

Figure 7.3 Trajectories of a harmonic oscillator.

260 CHAP. 7 / SYSTEMS OF EQUATIONS

EXAMPLE 20 Find the trajectories of the nonlinear system

| x = xi ely C087,
y= —y — siny.

Solution Dividing the second equation by the first, we obtain

dy neg tS ie sin y
av ly, ex cos ye
or
(y + siny)dx + (x + 2y + xcosy)dy = 0.
This DE happens to be exact and can be easily solved by the method of Section
2.3.1. Its solution is
Ky oh yee ex sin ye.
where C is any constant. Typical configurations of the trajectories are shown in
Figure 7.4 for various values of C.

> x

Figure 7.4

Remark. Since the parameterization of a curve is not unique, the terms solution
and trajectory are not synonymous. For example, x = e', y = e” and x =1,
y = t’ are both parameterizations of the parabola y = x’, but in the first case only
the right half (x > 0) of the parabola is defined by the parametric equations.

Example 20 is characteristic of nonlinear systems described by

 261
SEC. 7.6 / NONLINEAR SYSTEMS AND STABILITY

x' = P(x, y),

(76)
y = OG,)).
Since the independent variable ¢ does not explicitly appear in (76), the system is
said to be autonomous. This means that the physical parameters of the system are
not time dependent, which frequently is the case in practice and is the only kind of
system we will discuss in the remainder of this chapter. Forming the quotient of the
two equations in (76), we obtain

dy _ Qi y) (77)
dvi PO.)

the integration of which provides the integral curves or trajectories except at points
If both
for which P(x, y) vanishes. (If P = 0, we simply consider dx/dy = P/Q.)
the phase plane, that point is said to be a critical
P and Q vanish at a point (xo, yo) of
point, for then a unique slope is not defined. However, in reference to the original
um point
system (76), the critical point (xo, yo) is best described as an equilibri
there. That is to say, once we reach this point,
because both x and y are stationary
a critical point is one for which both velocity
we can never leave it. In mechanics,
and acceleration vanish, and hence the motion stops.
ee ee ae
ee
A
EXAMPLE 21 Determine the critical points of the nonlinear system
ee y
ary
»
!
ey!

Solution Solving simultaneously,

— ? ae y = 0,

x —-y’=0,
le):
we see that there are two critical points, (0, Oywand

a DE is altered if either the

It is usually of interest to know how the solution of
is slightly changed. If small
initial conditions or the input function of the equation
small changes in the solution
changes in these input parameters produce only
ely, we have the following
function, we say the solution is stable. More precis
definition.

of the system
Definition 7.2 |Let {x(t), y(O} denote the solution set
x' = P(x,y), 0) = a,
y’ = Q,y), y0)=B8,
critical point is said to be
which has a critical point at (xo, yo). The
262 CHAP. 7 / SYSTEMS OF EQUATIONS

(a) stable iffor every € > 0 there exists some number 6 > 0 such that

VS Gn 2)? Gy = yy He ate
for all t = 0 whenever

[Gg On) eto 8) 0:

(b) asymptotically stable if it is stable and there exists a positive number A such
that

lim ig = 4) (Yo — y)]'"? =i)

whenever

[Og ay yo By = A and
(c) unstable otherwise.

a Se a ee ee oe ee ee
EXAMPLE 22 Discuss the stability of the critical point (0, 0) of the linear system

eee) == 28
y = —2y, y(0) = 1.

Solution Because the system is linear with constant coefficients, we readily find the solution
em wen) =Thuse according to Definition 7.2, given € > 0, we wish to find a 6
such that whenever

(4 1) 5 <6,
we have

(454 nes4) <e

for all t > 0. Here we see that by choosing 6 = e, it follows that
(den? tem MA a5\/ Se<06 eer ee 0)
Hence the critical point (0, 0) is stable, and since x(t)—0 and y(t)—0 as t—>-~,
the
critical point is also asymptotically stable.
IEOS ee ee eee

The following statements can be made about the trajectories of autonomous

systems, although we will not present any proofs.

1. There exists at most one trajectory passing through any point of the
phase
plane that is not a critical point.
2. A particle starting at a point other than a critical point cannot reach the
critical
point (if indeed it reaches it at all) in a finite amount of time. If the solution
is asymptotically stable, it will approach the critical point as t> (see Figure
SEC. 7.6 / NONLINEAR SYSTEMS AND STABILITY 263

7.5). If the solution is unstable, no matter how close to the critical point the
particle starts, there are solutions that move away from it.

Trajectory
(a, B)
[8 Yo)

(Xo, Yo) (Xo, Yo)

Stable Asymptotically stable Unstable

Figure 7.5 Stability trajectories.

3. If a trajectory crosses itself at a point of the phase plane that is not a critical
point, that trajectory is a closed path and corresponds to a periodic solution
of the system (see Figure 7.6).

Figure 7.6 Periodic motion.

hes a critical
These statements suggest that a particle (solution) either approac
hes a closed path, or goes off
point as t>*, moves along a closed path, approac
of critical points is most important
to infinity for increasing time. Thus the study
in the analysis of autonomous systems.
n set of the system
In order to test for stability by using Definition 7.2, the solutio
do not have a known solution set,
must be known. Since nonlinear systems often
y without first explici tly finding the
it is desirable to be able to test for stabilit
let us begin by examin ing more
solution of the system. To see how this is done,
extend our ideas to similar nonlinear
closely linear autonomous systems and then
systems.

7.6.2 Stability of Linear Systems

Consider the linear system

264 CHAP. 7 / SYSTEMS OF EQUATIONS

x'
ff
= ax + by, aa
y =cx
t+dy,
Ul

where a, b, c, and d, are constants such that ad — bc # 0. This system has one
critical point at the origin of the phase plane.

Remark. The requirement that ad — bc # 0 is to ensure that only isolated

critical points occur. If ad = bc, all points on the line ax + by = 0 (or the line
cx + dy = 0) are nonisolated critical points. Also note that the most general linear
system is x’ = ax + by + c,, y' = cx + dy + cp, where c, and c, are constants.
However, if the critical point is (xo, yo), then the simple translation X = x — x,
Y = y — yo reduces this linear system to the form (78) with (0, 0) as the critical
point. Hence, there is no loss of generality in treating the case (78).

The auxiliary equation associated with (78) is (see Section 7.4)

dX? —(a + dd + (ad — bc) = 0, (79)

with roots A; and A. Observe that A = 0 cannot be a root since ad — bc #0.
From our definition of stability, it is clear that if either or both A, and A, have a
positive real part, the critical point is unstable, since x(t) and y(t) would both
become infinite as >. However, if both A, and A; have negative real parts, then
the trajectory described by every nontrivial solution of (78) approaches (0, 0) as
to. In summary, we state the following theorem.

Theorem 7.6 The critical point (0, 0) of the linear system

x’ =ax
+ by, ad—bc#0
{?
y =cx + dy,
is stable if and only if both roots of the auxiliary equation

A?
— (a + dd + (ad — bc) = 0
have nonpositive real parts. The critical point has asymptotic stability if
both roots
have negative real parts.

Some of the possible trajectories for the linear system (78) are shown
in Figures
7.7 through 7.11. The critical point (0, 0) is referred to as a node
in Figures 77
and 7.8, a focus in Figure 7.9, a center in Figure 7.10, and a saddle
point in Figure
7.11. These figures make clear that small changes in the initial
conditions produce
only small changes in the solution for stable systems but can lead
to large changes
in the solution for unstable systems.
SEC. 7.6 / NONLINEAR SYSTEMS AND STABILITY 265

y
A A

: Xx |

2 < A, < 0 (stable) d, > Ay > O (unstable)

Figure 7.7 Real distinct roots leading to nodes.

y y
rN r

: by:

\, = A. < 0 (stable) \, = A, > 0 (unstable)

Figure 7.8 Equal roots leading to nodes.

oy J
A
A

‘ © ‘

p <0 (stable) p > 0 (unstable)

Figure 7.9 Complex roots = p + ig leading to foci.

266 CHAP. 7 / SYSTEMS OF EQUATIONS

><

(stable)

Figure 7.10 Pure imaginary roots A = +iq leading to a center.

d\, > 0> A, (unstable)

Figure 7.11 Real distinct roots leading to a saddle point.

EXAMPLE 23 Discuss the stability of the system

Le Xe roy,
'

yay.
Solution The auxiliary equation is A* + 2 = 0 with roots A= + V2i. Thus the origin is
a center and the solution is stable.

EXAMPLE 24 Discuss the stability of the system

x' = —-4x —y,

Vee vem sve

Solution
The auxiliary equation is A* + 6A + 9 = 0 with double root A = —3. Hence the
origin is a node, which is asymptotically stable.
CUO ae
 SEC. 7.6 / NONLINEAR SYSTEMS AND STABILITY 267

EXAMPLE 25 Discuss the stability of the system

x= 4y-— 7,
ye Ox — Sy.

Solution Here we find A2 — A — 6 = O with roots A; = 3 and A, = — 2. The origin is thus

a saddle point, which is unstable. ee
a ee
a a a SN

Since periodic solutions are of special interest in many problems, it should be

noted that they arise in the case of a linear system when and only when the roots
of the auxiliary equation (79) are pure imaginary. In such a case, the trajectories
are closed curves (ellipses), and the critical point is a center. In other words, closed
trajectories imply periodic motion.

7.6.3 Local Stability of Nonlinear Systems

Here now we wish to examine certain types of nonlinear systems

xl = "PO, y);
(80)
y’ = OG, y),
is a critical
where for simplicity we assume P(0, 0) = Q(0, 0) = Oso that the origin
that P(x, y) and Q(x, y) possess at least third partial
point. We further assume
derivatives so that they can be expanded in Taylor series,
2

P(x, y) = P(O, 0) + P,(O, Ox + Py(0, Oy + Pxx(0, 05

Ne

+ PO, Oxy + PyO,OF + °°

Zz

Q(x, y) = Q0, 0) + 2,0, 0)x + O,(0, Oy + Qx(, n>

2

+ Qn (0, Oxy + Qy(0, Oe ++

can write
Then, using the fact that P(O, 0) = Q(0, 0) = 0, we
P(x, y) = ax + by + pt, Y),
Q(x, y) = cx + dy + qx, y),
derivatives and are small
where p(x, y) and q(x, y) have continuous first partial
enough near the origin in the sense that
oe) eae nyoenn ) ee
(x? + y?)!?
G2 + yy)?
 CHAP. 7 / SYSTEMS OF EQUATIONS

Our nonlinear system (80) now has the form of a perturbed linear system (nearly
linear system)

ax tiby + p(X, y),

/
x

y
t/
Ch GV AUN y). (81)
In most cases, stability behavior of the perturbed linear system in the neigh-
borhood of a critical point (0, 0) is closely related to the stability behavior of the
corresponding linear system arising when p(x, y) = q(x, y) = 0. In particular, we
have the following theorem, which we state without proof.

Theorem 7.7 Let (0, 0) be a critical point of the perturbed linear system
,
X= aXe DY +,DUX..y),
Uy
y ay rag):

If the critical point (0, 0) of the associated linear system

x = ax by dd = be 0,
y' =cx + dy,

(a) asymptotically stable, then the nonlinear system is asymptotically stable

near this point;
(b) unstable, then the same critical point of the nonlinear system is unstable; and
(c) stable, but not asymptotically stable, then the same critical point of the
nonlinear system may be asymptotically stable, stable, or unstable (i.e., no
conclusion).

Even though the stability of a perturbed nonlinear system may be the same as
that of the associated linear system, the trajectories of the nonlinear system may
differ greatly from those of the linear system. Also, if we move sufficiently far from
a stable critical point of the nonlinear system, it may no longer be a point of stability
of the system. This is in sharp contrast with linear systems, where stability is not
localized.

EEE

EXAMPLE 26 Discuss the stability of the simple pendulum whose equation of motion is

6” + k? sin@ = 0.

Solution Setting x = 0 and y = 6’, we obtain the nonlinear system

x'=y,
y' = —k?*sinx.
'
SEC. 7.6 / NONLINEAR SYSTEMS AND STABILITY 269

This system has an infinite number of critical points (0, 0) and (nz, 0), where
n= +1, +2,.... The associated linear system near the critical point (0, 0) is
(sinx =x as x0)
'
x M5

eed coe
which
Here we find that the roots of the auxiliary equation are pure imaginary,
7.7, however,
means that (0, 0) is a center of the linear system. Based on Theorem
(0, 0).
we cannot conclude anything about the stability of the nonlinear system near
about the
At the critical point (77, 0), the situation is different. Expanding sinx
point x = 77 yields
i
Sue Tee) Gay Tek) ote

so that the associated linear system this time becomes

aie
!

y =k yo):

— k* = Oare
Letting z = x — 77, we see that the roots of the auxiliary equation A?
system.
A, = kand A, = —k, so that (77, 0) is an unstable critical point of the linear
ar system.
Therefore, it is an unstable critical point of the nonline
At the point x = 77, the pendulum is pointing vertica lly upward. Clearly, such
cause large motions of the
a position is unstable, since a small displacement will
pendulum hanging vertically
system. Also, the point x = 0 corresponds to the
downward. It is equally clear that such a positio n is stable based on physical
considerations in spite of the fact that we could not derive this conclusion from
points of the form (n7r, 0) will
Theorem 7.7. In fact, it is evident that all critical
of n and of the type (77, 0) for odd
be of the same type as (0, 0) for even values
stable (but not asymptotically
values of n. Thus the critical points are alternately
stable) centers and unstable saddle points.
7.12. Close to the stable
The trajectories of this system are shown in Figure

Figure 7.12 Trajectories of a pendulum.

270 CHAP. 7 / SYSTEMS OF EQUATIONS

critical points are the closed paths suggesting periodic motion. The trajectories that
cross at the unstable critical points are called separatrices. Finally, the wavy paths
outside the separatrices correspond to whirling motions of the pendulum.

For purposes of contrast with Example 26, it may be interesting to consider the
effect of frictional forces acting on the pendulum. Assuming such a force is
proportional to the angular velocity 6’, the equation of motion becomes
6" + co’ +ksind =0, c>O0, (82)
or, equivalently,
/
Am kame) 5

y' = —k*sinx — cy,

id (83)
where x = 6. Again we observe that the origin is a critical point, but it now
becomes asymptotically stable. To see this, let us examine the associated linear
system
,
am)

Va=
,
ke cy, (84)
with auxiliary equation A? + cA + k? = 0. The roots are
Or ti(C = a4)
di, Az » 5)

and hence the origin is a stable node if c > 2k, or a stable focus if c < 2k.
According to Theorem 7.7, in all cases the origin of the nonlinear system is also
asymptotically stable.
A more direct method than presented here is also available for analyzing the
stability of nonlinear systems. This method, which rests upon the construction of
a suitable auxiliary function, is a more powerful method in that it provides more
global type of information. This method is a generalization of the physical prin-
ciples associated with a conservative system and is due to A. M. Liapunov
(1857-1918). The Liapunov theory, however, would take us too far afield for our
purposes, so we refer the interested reader to the references.

EXERCISES 7.6

In problems 1-4, sketch the trajectory corresponding to the solution Satisfying the
prescribed initial conditions, and indicate the positive direction of motion.

Less Oe Ss PEE ae Sa
y' =x, y(0)=0 y' =x, yO) =0
3.nitesty,~ Tx(0)) Sted 4. x'=2x+ 4y, x0) =4
y= Xe). 0 y' =—2x
+ by, 3(0).=0
SEC. 7.6 / NONLINEAR SYSTEMS AND STABILITY 271

In problems 5—10, determine the critical points of the system.

5. x'=y Gi x=
Ye ax yea xy
Lk = Soy Bat, Skee
y’ =x + 3y yi pax 1 2y ite!
oO. x=A xy 10, x0 =e — x7 xy

y' =y + 2xy ole

eel elk
ree!
ao aed 4

In problems 11-17, find the one-parameter family of trajectories by solving Equa-

tion (77).

11. x’ =y 12a al a 2y
pix | Vo teDyan)
13.7% = 3x 14, x’ = —2x’y
y' =x + 3y y’ =x*+y*
15. 2% =2 —\y* 16. — siny
x' =2xy

yi ny y'=1-y?
TJ = 3x yx ty
y' = y(3 — 2xy’)
18. Consider the linear autonomous system
yo S54
t

yo =x ey-
conditions x(0) = 1,
(a) Find the solution of this system that satisfies the initial
y(0) = 3.
(b) Repeat (a) for x(4) = e, y(4) = 4e.
both represent the same
(c) Show that the two different solutions in (a) and (b)
trajectory.

point (0, 0) and tell whether

In problems 19-25, determine the nature of the critical
it is stable, asymptotically stable, or unstable.

ly 20. + 4y
x' =2x
19 sex = 2h
BY y pr 2x + Oy
yl = xr
+ 3y 222k aay
21. x’ =x
y'=3x ty Vespa = ge)
ey 24 ety
23 eee =
y' =x +5y
y' = 4x + 5y
25x = xray
 to"

272 CHAP. 7 / SYSTEMS OF EQUATIONS

*26. Given the system

x. = aK. — ¥,
yl =x + ay,
show that the trajectories are spirals when a # 0. Discuss the stability of the critical
point (0, 0) for both cases a > 0 and a < 0. What are the trajectories when a = 0?
Hint: Transform to polar coordinates to show spiral trajectories.

In problems 27-32, determine the nature of the critical point (0, 0) by analyzing
the related linear system.

2/7. x =x + 2y + xcosy,; 28. x'=y

y = y= siny SMe hai)
29. x = y 30. x’ =3x
+ 4y + x?
ah ae y' = 4x — 3y — 2xy
312. x) =x + 2yi+ 2siny 32. x’ =e *~'” —cosx
SS i hee y’ = sin(x
— 3y)

Determine the critical points in problems 33-35, and discuss their nature and
stability.

33. x’ = 8x -y? 34. x’ =2y

+ x?
y' = —6y + 6x? yx ay
35. x’ =1-xy
eet
*36. The Lotka-Volterra equations for the predator-prey problem are given by (see Section
ie)

X30. DN
'

Na Se wsban
Lf

(a) Show that the change of variables x = cX/d, y = aY/b leads to the system
X' = a(X — XY),
YaA= =) aX):

(b) Show that the trajectories are given by

(e*/X)° = K(Y/e’)* (K constant).
(c) The point (0, 0) is obviously a critical point of the system in part (a). Find a
second critical point, and discuss the stability of the system at each critical point
when a = 2 andc = 1.
(d) Ifx and y represent two competing biological species and they start initially with
“small” populations, do you expect the populations to become extinct, or can
they continue to exist without the threat of extinction? Explain your answers.
REFERENCES

Birkhoff, G., and Rota, G.-C. 1978. Ordinary differential equations. 3rd ed. New
York: Wiley.
Boyce, W. E. and DiPrima, R. C. 1977. Elementary differential equations and bound-
ary value problems. 3rd ed. New York: Wiley.

Campbell, H. G. 1971. Linear algebra with applications. New York: Appleton-

Century-Crofts.

Cullen, C. G. 1979. Linear algebra and differential equations. Boston, Mass.: Prindle,
Weber & Schmidt.

Finney, R. L., and Ostberg, D. R. 1976. Elementary differential equations with linear
algebra. Reading, Mass.: Addison-Wesley.
Ross. S. L. 1980. Introduction to ordinary differential equations. 3rd ed. New York:
Wiley.

273
Numerical methods are becoming more and more important in applications, partly
because of the difficulties encountered in obtaining exact analytical solutions but
also, more recently, because of the ease with which numerical techniques can be
used in conjunction with today’s high-speed automatic computers.
Several numerical procedures for solving initial value problems involving
first-order DEs are discussed briefly in Section 8.2, all of which are based upon
Taylor series approximations. They include Euler’s method, the improved Euler's
method, and the Runge-Kutta method. Of these, the Runge-Kutta method is the most
widely used because it is far more accurate.
These same numerical techniques are generalized in Section 8.3 to include
systems of first-order equations and higher-order equations, which are solved by
reducing them to a system of first-order DEs.
Although error analysis is an important part of any numerical procedure, we have
limited our discussion primarily to the use of the procedure itself. The theory of errors
is sometimes fairly complex and goes beyond the intended scope of this chapter.
The interested reader should consult a text on numerical analysis.

274

6
A
are,
8.1 INTRODUCTION

In applications we must often solve an initial value or boundary value problem that
is either difficult or impossible to solve exactly by analytical methods. For this
reason it becomes either convenient or necessary to employ some method that
yields accurate numerical estimates of the true behavior of the system. However,
the method itself can involve a considerable amount of analysis concerning the
errors involved in using the method as well as errors due to rounding off in the
computations. We do not intend to discuss these matters deeply, as is done in
courses on numerical analysis; instead, we will simply present some techniques that
provide quantitative information about certain systems we wish to study.

8.2 NUMERICAL METHODS FOR FIRST-ORDER EQUATIONS

Let us consider the initial value problem

F(t, y), y (to) = Yo; (1)

y’ a

where the function F is suitably “well behaved.” We will assume the problem has
a unique solution in some interval containing the point fo (see Theorem 2.1).
The methods to be discussed here are step-by-step procedures, wherein each
calculated value makes use of the previously calculated value. We start at the initial
point on the solution curve (fo, Yo) and increase t) by the fixed (positive) number
h to get t; = tf + h, and then compute a value y, that approximates the solution
value y(t,). In the second step of the procedure, we find an approximate value y2
for y(t), where tf = t; + h = f + 2h, and continue in this fashion.
The calculations are all done by the same formula at each step of the process.
These formulas are suggested by the Taylor series

yee +h) = yO + hy" + FV" too 2)

where from (1) we find y'(t) = F(t, y) and

—
d
eee i aL
dy
pee
”

with respect to
and so forth. The subscript variables indicate partial differentiation
e, we set ¢ = % and
the designated variable. In the first step of the procedur
y(t.) = yo to calculate
fer
yo) cor at ee)
Vite 0 a hF (to, Yo) “12 5h As

and in the second step we calculate

ie a
yo == yy + AFC, y1) + —h?—(t
A (ti, yi)t-+°:,
, yi)

275
 hi
276 CHAP. 8 / NUMERICAL METHODS

whereas, in general,

Yntt = Yn + AF (tus Yn) + ae (tns Yn) Hct. (3) 9

For computational purposes, the series in (3) must be truncated after a certain
number of terms. This leads to an error, which is appropriately referred to as a
truncation error. If h is picked to be sufficiently small, terms involving h’, h°, and
higher powers of h can often be neglected in (3), making a first-order approxi-
mation. The truncation error per step is then of the order h*. When higher-order
terms of (3) are required for greater accuracy, we must compute the derivatives of
F. Since these derivatives must be done by hand, using (3) directly is often not
feasible when F is of a complicated nature. In such instances it is usually preferable
to replace the derivatives by certain numerical equivalences so that the formulas
lend themselves to computer calculations. That is precisely the approach used in the
improved Euler method and Runge-Kutta methods to be discussed.

8.2.1. The Euler Method

In the Euler method we make our calculations using the formula

Yn+1 7 Yn 4 hF (ty, Vas (4)

obtained from (3) by truncating all terms involving powers of h higher than 1.
Geometrically, this technique approximates the true solution curve with a polygon
whose first side is tangent to the solution curve at (f, yo) (see Figure 8.1). Hence
it is also called the method of tangent lines.

S
True solution

(to, Yo)

Figure 8.1
oe
eee

EXAMPLE 1 Use Euler’s method to approximate the solution of the initial value problem

y = = 22, yO) =a,

on the interval 0 S$ t < 0.5, using h = 0.1 and also h = 0.05.

We identify F(t,, y,) = —2t,y? so that (4) becomes

V5 dy meee cil,Vane =
For h = 0.1, we set n = O and calculate
 277
SEC. 8.2 / NUMERICAL METHODS FOR FIRST-ORDER EQUATIONS

n=0: y= Yo — 2(0.1)toy
1 — 2(0.1)(0)(1)?
= |,

which is our estimate to the value y(0.1). Continuing, we find

n=l: y. = y, — 20.Dnyi
1 — 2(0.1)(0.1)(1)"

= 0.98,
Tiel f2 y3 = y2 — 2(0.1)tay3
kes = 2(0.1)(0.2)(0.98)
= (0).

= (0)Sil).

well as the exact values,

and so forth. These, plus a few additional calculations as
s, the exact
are summarized in Tables 8.1 and 8.2. [By separating the variable
solution is found to be y = 1/(t? + 1).]

Table 8.1 Euler’s method for y’ = —2ty”, y(0) = 1, withh = 0.1

1.0000
1.0000
0.9800
0.9416
0.8884
0.8253

1.0000
1.0000
0.9950
0.9851
0.9705
0.9517
0.9291
0.9032
0.8746
0.8440
0.8119
278 CHAP. 8 / NUMERICAL METHODS

Since we had the exact solution in Example 1, we could calculate the amount
of error incurred in the method at each step of the procedure. For instance, over the
interval in which we made the calculations, the maximum percentage error in the
first case is found to be
|error|
elles socal ee _ 0.0263
= ye = 3.05%,
exact value oY 0.8621 Bt : 4

while in the second case we have

O0125
0.862] We
x 100 ceil 1.45% !

We might conclude that the Euler method works quite well, since the errors are
acceptable in many applications. However, this kind of accuracy is usually not
realized by Euler’s method in practice (see Example 2).

a ee Ped ee hee, ee
EXAMPLE 2. Use Euler’s method to solve the initial value problem

Vasa
last titty my O)eals
on the interval O S¢ <0.5 withh = 0.1.

Solution Here we find F(t,, y,) = 1 — t, + 4y, and thus

Yuet = Ya + (0.1) 7, 4y,):

Our first calculations yield
n=0: Vie Vo ee (O.
2) CPt tg eeygn
bet (Onl Ce 0st 4)
= 1.5,
n=1: Yo = viet (OT) tty

Table 8.3 Euler’s method fory’ = 1—1+ 4y, y(0) = 1, withh =0.1

Exact
Value
SEC. 8.2 / NUMERICAL METHODS FOR FIRST-ORDER EQUATIONS 279

oe.1) Gt 016)
Beh.
Additional values are provided in Table 8.3 along with the exact values for com-
parison.

The maximum percentage error in this last example is 27.4%, significantly

greater than in our previous example. We recognize, of course, that the error is
controlled to some extent by the size of the increment h. By reducing the size of
h we can reduce the error, but the number of calculations then increases. Also, the
error normally builds up as we move farther away from the initial point on the
solution curve so that it eventually exceeds what is deemed acceptable, no matter
how small we choose the value of h. Therefore Euler’s method is seldom used in
practice even with the high-speed digital computers available today. It is introduced
here primarily because it is simple and may be helpful in understanding the basic
procedure used in other numerical techniques.

8.2.2 The Improved Euler Method

Euler’s method can be made more accurate for a fixed value of h by first computing
the auxiliary value
Vie =I Yn ats hF (th, Yn) (5)

and then the new value

1
LF Un, Yn) a5 Fein yee]. (6)
Yn+1 = Yn ig 5

This technique, called the improved Euler method (or Heun’s method), is an
step we
example of what is called a predictor-corrector method. That is, at each
predict a value by (5) and then correct it by (6).
true
The geometric interpretation of this new method is that we approximate the
(tf, y,) with
solution y in the interval from f, tot, + sh by the straight line through
slope F(t,+1, Ye+1) UP to tr+1 (see
slope F(t,, Yn), and then along a new line with
sum Fh, Va) + Fass y*.1)] as
Figure 8.2). Therefore, we might interpret the

y True solution
280 CHAP. 8 / NUMERICAL METHODS

some average slope over the interval t, < t < t,+4,. It can be shown that (6) is
equivalent to a Taylor series through the term containing h?, but it has the advantage
that dF/dt does not have to be calculated.
ee
ee eee
EXAMPLE 3_ Use the improved Euler formula to obtain an approximate solution of
Vee ye ty(Oars

Solution We will illustrate the technique for h = 0.1 on the interval 0 < t < 0.5. We first
calculate

na 0; yi = yo — 2htoys
= 1 — 2(0.1)(0)(1)’
— ¢

and then

1
y= Yo, 5 hl —2toyo eI 2n(yi)"]

1 — (0.)[)C)* + O.1)d)]
EO:
The remaining calculations are provided in Table 8.4 along with those obtained by
Euler’s method and the exact values.

Table 8.4 Approximate solutions of y’ = —2ry*, y(0) = 1, using

the Euler method and the improved Euler method with h = 0.1.

Improved
Euler

Example 3 shows that the results using the improved Euler method with
h = 0.1
are better than those obtained with the Euler method even for h = 0.05. In
general
this is the case, and although a few more calculations are required at each
step using
the improved Euler’s method, the greater accuracy is worth the extra effort.
SEC. 8.2 / NUMERICAL METHODS FOR FIRST-ORDER EQUATIONS 281

8.2.3 The Runge-Kutta Method

Perhaps the most commonly used as well as most accurate technique is the Runge-
Kutta method .* At each step we must first compute four auxiliary quantities
ky = F(t, Vas

1 1
k, = F(t ar ah, yer sh),

: (7)
= F(t + ah, Vice ss),

kg = Filta Yn XG hk3),

and then calculate the new value

h (8)
Vee yee 60k + 2k. + 2k3 + ka).

y of a Taylor
The purpose of the Runge-Kutta method is to achieve the accurac
to calculat e higher- order derivati ves. For example,
series expansion without having
(8) was derived by obtaini ng appropr iate constants
the algorithm given in (7) and
A, B, C, and D so that
Ynt1 = Yn + Ak, + Bky + Ck; + Dk,
term of the series (see
agrees with the Taylor expansion out to h4, or the fifth
-order Runge-Kutta
problem 26). Thus the technique is often called the fourth
+ k4)/6 appear ing in (8) can be interpreted as
method. The term (k; + 2k, + 2k3
an average slope over the interval t, = t S t+1-
our previous formulas,
Although the Runge-Kutta formula is more complex than
even the improved Euler
it yields results that are many times more accurate than
es this accura cy with larger comparative in-
formula. And it generally achiev
computer, the calculations
crements of h. Furthermore, with the use of a modern
in a few second s for most problems.
are routine and can be performed
on F does not explici tly depend on y, then
Finally, we note that if the functi
1
ky = F(t), ke = k3 = F(t +5h), ky = F(t, + h),

and (8) can be written in the form

1
i,
PES ates | FeoF aF(«ote 5h)Teh (Ce

Simpson’s rule of integrationt on the interval

which is recognized as
ee yk

(1856-1927) and WILHELM KUTTA

the German mathematicians CARL RUNGE
*Named after
(1867-1944).
).
+Named after THOMAS SIMPSON (1710-1761
282 CHAP. 8 / NUMERICAL METHODS

EXAMPLE 4 Find an approximate solution to the initial value problem

yee yas (0) a

on the interval 0 < ¢ S | using the Runge-Kutta method with h = 0.2.

Solution With F(t, y) = —2ry’, we first set up the expressions

ky = ive

ky = —2(t, + 0.1)(y, + 0.1k,)’,

kx = —2(t, + 0.1)(y, + 0.1k,)*,
Kg 2 Go) Vat Oka)ae
For n = 0, our calculations yield

n = 0: ki =0, k= -0.2, ko = —0.192, k, = —0.37,

and

02
y=1- “6 (2(0.2) + 2(0.192) + 0.37] = 0.9615,

The remaining calculations are given in Table 8.5 along with similar results ob-
tained by the Euler methods. The improved Euler method with h = 0.1 is also
included for comparison.

Table 8.5 Approximate solutions of y’ = —2ry*, y(0) = 1, using

the Euler methods and the Runge-Kutta method with h = 0.2.

Improved Runge- Improved

Euler Kutta Euler Exact
yr(h = 0.1) Values

EXAMPLES5 Find an approximate solution to the initial value problem

yale tetudve y (0) a8

on the interval 0 <7 < | using the Runge-Kutta method with h = 0.1.

Solution Here F(t, y)=1—t + 4y, which leads to (n = 0)

SEC. 8.2 / NUMERICAL METHODS FOR FIRST-ORDER EQUATIONS 283

k, = F(0,1) = 5,
ky = F(O + 0.05, 1 + 0.25) = 5.95,
ke = (040.05, 140.2975) =.6.14,
k, = F(0.1, 1 + 0.614) = 7.356.

Thus,

| 0.1
me =1+ a. + 2(5.95) + 2(6.14) + 7.356]

1.6089.
A tabulation of the remaining values is given in Table 8.6 as well as the results
obtained from the Euler methods. The superiority of the Runge-Kutta method is
clearly demonstrated by this example.

Table 8.6 Approximate values of y’ = 1 — 1 + 4y, y(0) = 1, using

the Euler methods and the Runge-Kutta method with h = 0.1.

Improved Runge-
Euler Kutta Exact
Values.

requires
Remark. The Runge-Kutta method has certain drawbacks in that it
calculations of the function F(t, y) at successive steps of the
time-consuming
expensive. These calcu-
procedure. In some applications this technique may be too
rrector methods, such
lations can be greatly reduced by use of certain predictor-co
However, these methods create some inconve-
as the Adams-Moulton method.
change in step size as the
niences of their own in many cases by requiring a
choose will greatly depend
calculations proceed. The proper numerical method to
the necessary computer
upon the application and the budget allowed for making
calculations.
284 CHAP. 8 / NUMERICAL METHODS

EXERCISES 8.2

In problems 1-10, use the Euler formula to obtain an approximation to four places
after the decimal, to the indicated value of y with h = 0.1.
1. y’ =2y, yA) =1, ys) =7
y =1 + y yO) =U, yOs) = 7
ye tery Sl)s yO) = 20 yOs) a4
y ‘= pty, yO)
= 1) [email protected]) =?
y '=??+y’, yO)=1, yO.5)=?
y '=ty+Vy, yO)=1, y(0.5)=?

< PoGh aos ssOS th. sil Me |

Vou= sit (tay eeyy (0) = Oey OSes
9. y' =e”, yO) =0, yO5)=2
10. y'’=y-y’, y0) = 3, y(0.5)
=?

11. Repeat the calculations in problems 1-10 using the improved Euler formula.
12. Repeat the calculations in problems 1-10 using the Runge-Kutta method.
13. Using Euler’s formula with h = 0.2, find an approximate value for y(1) where y is
the particular solution of the initial value problem
y 2 yw O= 1
(Your answer approximates the value of e. Can you explain why?)
14. Solve problem 13 using the improved Euler’s formula.
15. Solve problem 13 using the Runge-Kutta method.
16. Using Euler’s formula with h = 0.2, find an approximate value for y(1) where y is
the particular solution of the initial value problem
y' =(? +1)", yO) =0.
Use your answer to approximate the value of 77.
Hint: The exact solution is y = Arctan (tf).
17. Solve problem 16 using the improved Euler’s formula.
18. Solve problem 16 using the Runge-Kutta method.
“19. Derive Euler’s formula (4) by integrating both sides of the DE y’ = F(t, y) from x,
to x,+, and then approximating the integral by

[feoax = (b — a)f(a).
Give some justification for using the above approximation for the integral.
20. Using the step size h = 0.2, approximate the value y(1.4), where y is a particular
solution of the initial value problem
y=r?ty’, y)=1,
 SEC. 8.3 / SYSTEMS OF EQUATIONS 285

(a) by Euler’s formula.

(b) by the improved Euler’s formula.
(c) by the Runge-Kutta method.
*21. Show that the Runge-Kutta method reduces to Simpson’s rule of integration on the
interval x, <x <x,+4, when the DE is of the form y’ = F(x).
*22. Using the first three terms of the Taylor series (2), show that it leads to the approx-
imation formula

Yost = Yn + hyn + Shyh,

where y’ = F(t, y) and y” = F,(t, y) + F(t, y)y’. Such a formula is called a three-
term Taylor formula.
23. Use the three-term Taylor formula in problem 22 to approximate the value y (0.5) for
the solution of the initial value problem
Wei sey) et,
using h = 0.1, and compare your answer with those listed in Table 8.4.
*24. Show that when F(t, y) is linear in ¢ and y, the three-term Taylor formula in problem
22 reduces to the improved Euler’s formula.
*25. Derive a four-term Taylor formula similar to the result given in problem 22, and use
it to show that y(0.4) = 1.5152, where y is a particular solution of
y= t ye y(ONe Ll.
Use the step size h = 0.1.
*26. Prove that the fourth-order Runge-Kutta formula (8) agrees with the Taylor expansion
out to h*; i.e., show that (8) and (2) are equivalent through the fifth term of (2).

8.3. SYSTEMS OF EQUATIONS

The methods discussed in the preceding section for solving initial value problems
featuring first-order DEs can be extended to a system of first-order equations. For
instance, let us consider a system of two equations
x' = F(t, x, y), x(t) = 0,
(9)
Ve ae G(t, x, y), y (to) = die

As before, we will assume that F and G are suitably behaved so that (9) has a unique
solution (for linear systems, see Theorem 7.1).
For illustration purposes we will generalize the Euler method, since it is the
simplest to apply. Thus, we first calculate
X, = Xp + hF (to, Xo, Yo);

y1 = Yo + AG(to, Xo, Yo),

for some prechosen positive increment h, whereas in general our calculations

involve
286 CHAP. 8 / NUMERICAL METHODS

(10)
x, oi See DE Gn)

Yn+1 = Yn Fe hG(t, ; Xn» Vals

where ¢, = t + nh.
ee ee ee ee eee ee Se

EXAMPLE 6 Approximate the solutions of the system of equations

f° =x —y +e, xO) =i,

ym ts yad eu env (Oye. 0,

at the points t = 0.1 and t = 0.2 using the Euler method.

Solution Choosing h = 0.1, we first calculate

Xi = Xo ar h(xo eae yO) SF e'0)

1+ 0.Dd -—0+ 1)
= 1.2
and
V1 = yo + h(2x% + 3yo + e ”)
= 0 + (0.1)[2(1) + 3(0) + 1]
= 0.3.
Similarly,
ApS Ae es!)
= 172) (O81) Ce 2e— Ona en)
= 1.4005
and
Vi VA ap h(2x, ap 3y, ar e ")

= 0.3 + (0.1)[2(1.2) + 3(0.3) + e*"]

= 0.7205.

For comparison purposes, the exact solution is

ft) =
1
—
1
x(t) 10° 2t Qi cost =— 13isin'
. ey
tire te a
10° .

l
y(t) = go (A cost +17 sing) eo— ae"

from which we calculate

 SEC. 8.3 / SYSTEMS OF EQUATIONS 287

x(0.1) = 1.1980, (0.1) = 0.3665,

x(0.2) = 1.3818, y(0.2) = 0.8957.

8.3.1 Higher-Order Equations

In Chapter 7 we found that higher-order DEs can always be reduced to a system of

first-order DEs. For example, the initial value problem

SE ee 50) Xo, AO) = Vo, (11)

is equivalent to the system of equations

4 =7);, X(0) = 2)
, (12)
PS Ns) 2 (0) = yo,
and thus can be solved by methods already discussed.*

EXAMPLE 7 Approximate the solution of the initial value problem

ates kee peer (OQ) al ex (0) = 2,
at the points t = 0.1 and t = 0.2 using the Euler method with h = 0.1.

Solution The equivalent system of equations is

x'=y, x(0)=1,
y’ =t—3x-—t’y, y(0) =2.
From (10) we have
xX; = X + hyo
= 1+ (0.1)(2)
=al22.
yi = yo + (to — 3x0 — fYo)

So (OUD SO o ew)

I,

and

*Some numerical analysts believe that greater accuracy is achieved by applying a numerical procedure
directly to the higher-order DE rather than reducing the DE to a system of first-order equations and then
applying a numerical procedure. See Peter Henrici, Discrete Variable Methods in Ordinary Differential
Equations (New York: Wiley, 1962).
288 CHAP. 8 / NUMERICAL METHODS

D9) a 6) ar hy,

= 1.2 + (0.1)(.7)

= 1.37,

yo = yi + A(t, — 3m — fi)
= 167-42 (O5L) (0: Deas Gle2 (0.1)?(1.7)
ll 1.3483.

Hence, x(0.1) = 1.2 and x(0.2) = 1.37.

ee ee ee

EXERCISES 8.3

In problems 1-5, use the Euler method with h = 0.1 to determine approximate
values of the solution at t = 0.1 and t = 0.2.

1. x’ =x-—4, x0)=1 2. x =x+¥, x0) =—1

yi ayn), Oi O y =x —y: 2yO) =e
Seb oe ety,
eS x (Oa 4.0%) =xor yaat, x OHA
Way ey (Oy at y'=4x -2y, yO)=0
Bee ih ye Oat
eS
6. The generalizations of the Runge-Kutta method to the system of equations
x' = F(t, x, y), x(to) = Xo,
y' = Gt, x, y), y(to) = yo,
leads to the formulas

Xn+1 2s 2K, + 2K, + 2K. + K,).

Yat = Ya oh FL I) Lo wate 2Las 1 naps

where
K, = I Cg Xny Male

Ly, = G(tn, Xn5 Valk

K>
oe
— F(tn
1
qe 5h, Xn
A
ate 7 hKi, Yn a
u5 Li),

[ReeG n 2) 9 An D hk l> Yn ee 2 l}>

end aCe ey net

n D 9» An D) 29 Sree isi
Yn 5) 2)>

eet
ce ik y Ne, n 5) 9 An D 29 ay,osSing
Yn D 2)>
SEC. 8.3 / SYSTEMS OF EQUATIONS 289

Ka = F(t, + h, Xn + hK3, yn + AL),

La = Gat Rite TAKsy Vn ce La):
Use these formulas to solve problem | at t = 0.1.
7. Use the Runge-Kutta formulas in problem 6 to solve problem 2 at t = 0.1.
8. Use the Runge-Kutta formulas in problem 6 to solve problem 3 at t = 0.1.
9. Use the Runge-Kutta formulas in problem 6 to solve problem 4 at ¢ = 0.1.
10. Use the Runge-Kutta formulas in problem 6 to solve problem 5 at t = 0.1.

Change the initial value problems 11 and 12 to a system offirst-order DEs, and use
the Euler method with h = 0.1 to determine approximate values of the exact
solution at t = 0.1 and t = 0.2.

Ailes ty exe — Oe (0)

— tee (0) R92
12 on tt 3 Ste 0) HO) 1
13. Use the Runge-Kutta formulas in problem 6 to solve problem 11 at ¢ = 0.1.
14. Use the Runge-Kutta formulas in problem 6 to solve problem 12 at ¢ = 0.1.

REFERENCES

Acton, F. S. 1970. Numerical methods that work. New York: Harper & Row.
Burden, R. L., Faires, J. D., and Reynolds, A. C. 1978. Numerical analysis. Boston,
Mass.: Prindle, Weber & Schmidt.

Henrici, P. 1964. Elements of numerical analysis. New York: Wiley.

Hildebrand, F. B. 1974. Introduction to numerical analysis. 2nd ed. New York:
McGraw-Hill.
Up to this point we have been able to solve homogeneous linear DEs of order
greater than 1 entirely by algebraic methods when the coefficients were constant. In
fact, we were even successful in extending the technique to a special variable-
coefficient DE called the Cauchy-Euler equation. Unfortunately, this method cannot
(in general) be extended to other variable-coefficient DEs because their solutions
typically involve nonelementary functions. Variable-coefficient equations like Bessel’s,
Legenare’s, and the hypergeometric equation, along with many others, arise in nu-
merous important engineering applications, and so other solution techniques must
be found. The method to be discussed produces solutions in the form of power peer
eee
Rt

series, and for that reason the procedure is referred to as the power series method.
Because of computers, this method is no longer as useful as it once was. Never-
theless, the theory is still very important since it can be used to determine regions
where the solutions are analytic, and this information is essential even in the applica-
tion of numerical techniques.
The general method is discussed in Section 9.2 for the case of power series
expansions about ordinary points. For this case we always find two linearly indepen- nha

dent solutions of second-order equations.

In Section 9.3 we first distinguish between regular singular points and irregular
singular points. The Frobenius method for finding a solution of a DE about a regular
singular point is discussed at length in terms of three separate cases depending
upon the nature of the roots of the indicial equation. The three cases correspond to
the roots differing by a noninteger, equal roots, and roots differing by a nonzero in-
teger. Finding solutions about irregular singular points is not considered at all, since
there is no general theory for this case.

290
0
 9.1 INTRODUCTION

A power series is an infinite series of the form

€o + ex(x — x0) + cox — Xo)? ++ ++ = Dene — 0)", (1)

n=0

where Co, C1, -- - » Cn» - - - are called the coefficients of the series and xo is the
center of the series. The series has the sum co when x = Xo, but generally we are
interested in whether the series also has a sum for other values of x.

Definition 9.1 A power series == cn (x — Xo)" is said to converge for a particular value of x if
N

limDen(x — Xo)"
2
exists. Otherwise the series is said to diverge.

The values of x for which the series (1) converges is called the interval of
convergence. That is, to each series corresponds a number R, called the radius of
convergence, with the property that the series converges if |x — x9| <R and
diverges if |x — x| > R. The radius of convergence of many power series can be
found by means of the ratio test.

Ratio Test If
Coico) a |= |x — xo|lim
Cnt+l
L = lim )
roa
Cyl FxG)" now
Cn
> 1.
then the series > Ca(X — Xo)" converges when L < | and diverges when L

The ratio test administered to the series

este anne came canes Re cae

n=0
le
for example, shows that it converges for |x| < 1 and diverges for lelee-a
defines a
A power series in x — Xo with a positive radius of convergence R
function f in the interval of converg ence by the rule

(Cn Xeao) (2)

n=0

us and possesses deriv-

where c, = f”(xo)/n!. This function is necessarily continuo
convergence. Moreover,
atives of all orders everywhere in the original interval of
e differen tiation of (2). Familiar exam-
these derivatives can be found by termwis
ples of functions with power series expansi ons are
291
292 CHAP. 9 / THE POWER SERIES METHOD

t= Sx",
MiaBh alent
s1<x <1, 3)
foe] x"

e Sri DN
2 met
nt 1GOOl. OX
x Pome, (4)
(oe) antl (5)

in
sinx 2 ((eee
1) On # DI’ Oran (

xn
cosx =ae ere yp OO xy <0, (6)

Many of the series of interest have centers at x = 0 as in (3) through (6), but
not all power series do; for example,
co Hy!

logx = >) (uty CeO ey exaee (7)

n=1

Observe, however, that by making the change of variable X = x — 1 in (7), we

obtain a power series with center at X = 0,
— a

log(1 + X) = Deora
— 2 paki Xe. (8)

For this reason much of the pe esetin subsequent sections will be confined to
power series for which x9 = O without loss of generality.

9.2 THE GENERAL METHOD

From the calculus we know that a differentiable function can be expanded in a

power series. Since the solution of a DE must satisfy certain differentiability
requirements, we assume it too can be expressed in the form

y= » Cue (9)
n=0

It follows that power series for y’, y”, . . . can be obtained from (9) by termwise
differentiation. That is,

y= nga iy ae ie Meese Pooks (10)

n=0 n=0

The method now is somewhat like the method of undetermined coefficients in that
all that remains is to determine the constants c, appearing in (9) in order to have
the solution.
Let us illustrate the basic procedure by applying it to the simple equation

egaly, =i0; (11)

The general solution of this DE is known to be y = Ce*, but now we wish to find
it by the method of power series. Substituting the series for y and y’ into (11), we
have
SEC. 9.2 / THE GENERAL METHOD 293

foo} foo}

SON Ya SONY Fol ty

n=0 n=0
SS ——
n—n-1

To combine these two series, we must have the same exponent for x in each
corresponding term, and both summation indices must start at the same value. Let
us begin by replacing the index n by (n — 1) in the second sum to get
foe] oo

Seca ae n= x" “hs 0,

n=0 n=

and then we can write

foa} foe}

Oe a> eric te Dy Cea nO

n=1 n=1

or, adding the two series termwise,

OMe ag eas:
ICH acral (12)
n=1

Now, since x can take on various values, the coefficient of x”! must vanish in order
to have (12) identically zero; 1.e.,
NC Cp ON Na yo,ees (13)

while cy can remain arbitrary since the term in which it appears is already zero. The
relationship (13) is called a recurrence formula for the unknown coefficients. Since
n # O, we can rewrite (13) as

Cp =e Na Ooi
2)e (14)
n

Successively substituting n = 1, 2,3,... into (14) yields

Cee CO,

C2
Peli
he. 5 0

Lae ac oape 60
Eas 2003 amet
a ee COME CO
BON TSE TTD ate
tle Mee etier eto, wiememie§ es ene) leuce,

Hence, from (9) we are able to write

ye Cyt Cx Hk? Cax? Cah

Co + cox + 5% ‘ae parime:

2 3 4
Xx Xx
Stet tion ait gyt 1a
294 CHAP. 9 / THE POWER SERIES METHOD

or

00 n
x
fy a= co» Hid
n=0

In this case we recognize the infinite series as that of e* and can therefore write the
general solution as
y = coe’, (15)
in agreement with our previous result.
Although the above example was quite elementary, it illustrates the basic manip-
ulations required to solve a DE by the power series method. Even when the
coefficients of the DE are polynomials in x rather than constants, the calculations
required differ little from those used in this simple example.
The power series method does not provide a general solution to all variable-
coefficient DEs, even for the special case when the coefficients are polynomials in
x. Before discussing the method any further, however, let us first clarify the two
rules of manipulation that were used in the above example and will be used in all
future examples.
Rule 1: When making an index change, always make all exponents on x equal
to the smallest one occurring in the various series.

Rule 2: When combining series under one summation sign, start all series with
the largest of all the beginning values. Terms preceding the new first value of the
summations must then be added outside the summation sign.

9.2.1 Ordinary and Singular Points

Consider the second-order DE

A2(x)y" + Ai(x)y’ + Ao(x)y = 0, (16)

where the coefficients Ao(x), A,(x), and A2(x) are any polynomials with no common
factors. The method of solution and the behavior of the solutions of (16) at a point
X = Xo depends upon whether this point is an ordinary point or a singular point of
the equation.

Definition 9.2 A point x = Xo is called an ordinary point of the DE

Ag(x)y" + Ai(x)y" + Ao(x)y = 0

provided A,(xo) # 0. If A2(%) = 0, we say that x = Xo is a singular point of the
equation.

Remark. When classifying points as singular or ordinary, complex values must

be considered as well. Also, infinite values can be included, although we will limit
our discussion to finite values.
 SEC. 9.2 / THE GENERAL METHOD 295

ee
ee ee ee ey Oe ee ee

EXAMPLE 1 Identify all singular points of the DEs

(a) (L—w4)y" = 2xy + 2y =.0}
(Dy sexy 4: Vo oxy, = Oand
(c) (&? + 4)y" + xy’ — 2y = 0.

Solution In (a), we set 1 — x* = Oand find x = +1 as the singular points. All other points,
real and complex, are ordinary points.
The only singular point in (b) is x = 0, and thus all others are ordinary points.
The DE in (c) has singular points at the solutions of x? + 4 = 0, or x = +2i.
Again, all other points are ordinary.

More generally, if the coefficients in (16) are not polynomials, we first rewrite
the equation in normal form

y era(ny, 4 0(x)y =10, (17)

and then define ordinary point as one for which both a(x) and b(x) are analytic. If
one or both of these functions fails to be analytic at the point x = xo, we say that
point is a singular point.

Remark. A function analytic at x = x) has a power series expansion about

xX = Xo with a positive radius of convergence.

EXAMPLE 2 Determine whether the DE 2xy” + (sinx)y = 0 has any singular points.

Solution Writing the equation in normal form

we see that all points are ordinary points, including x = 0, since (sin x)/2x has the
power series expansion

pine
2x Zz
etn asthe
St a |
which converges for all x and therefore must be an analytic function.
BA EE ——————E——————e—e———— en

In the remaining sections of this chapter, we will confine our discussion to only
those DEs that have polynomial coefficients. Furthermore, we will present the
theory only for second-order DEs and make no attempt at generalizing to higher-
order equations.
296 CHAP. 9 / THE POWER SERIES METHOD

9.2.2 Solutions Near an Ordinary Point

For most problems we are mainly concerned with whether the point x = 0 is an
ordinary point or a singular point of the DE, since this point is the easiest point at
which to apply the power series method. If x = 0 is an ordinary point the following
theorem is applicable. We state it without proof.

Theorem 9.1

n=0

Moreover, these series will converge at least in the interval |x| < R, where R is the
distance from 0 to the nearest singular point.

Remark. If the nearest singular point is complex, then we define the modulus of
this point as the distance from the origin. For example, the modulus of 1 — 2i is
[1 — 2i] = V5.
Theorem 9.1 is easily generalized to any ordinary point x = Xo. The series
solution in such a case will then be of the form

Y= Gea ook
n=0

EXAMPLE 3 Solve (1 — x7)y" — 2xy’ + 2y ='0.* \

Solution Since x = 0 is an ordinary point, we assume

from which we formally obtain

y= Dine", y" = Sint — 1)c,x"

n=0 n=0

Replacing y, y’, and y” in the DE with their series, we have

a - SDSS n(n — 1)c,x"-? — 2x5) ne,x" | + 2S c,.x" = 0,

n=0 n=0 n=0

*This DE is a special case of Legendre’s equation.

SEC. 9.2 / THE GENERAL METHOD 297

Daten Cy ete lo,c > iaric,x et >, 2C,t" = 0.

ate (a dheac arteries hed ler, ay n=0
n—n—2 n—>n-2 n—n—2

Now let us replace n by (n — 2) in the last three sums to get

co

Ds n(n — 1)c,x"? — S (n — 2)(n — 3)Cy-2x"? — 3 (n — 2)cy-2x"*

n=0 n=2 n=2

te S|Nip e =
n=2

or

O-cox 2 + O-cyx7! + >) [eel Caer ie— 2) (a 3) Cy a2 Nae 2)C=

n=2

+ 2c,-2|x""? = 0.

Clearly, co and c, are arbitrary, but the remaining constants must satisfy
n(n — 1)c, — ([(n — 2) — 3) + 2(n — 2) — 2]e,-2 = 0.

This last expression can be simplified to U1

C, = (AF) ens Fe a ne ie

which is ourrecurrence formula.

Setting n = 2, 3, 4, . . . into the recurrence formula yields
C9 = —C9,

Oy = Wy

C4
ye
3° 3 60

Z
Coa— ie = 0,

C6
ae
a
ee5 Or
4
C7 6a: = 0,

> 1
Oy = mice aw

Thus we get
y = Cy + C1x + gx? + C3x? + cax* + Cx
5
+--
 CHAP. 9 / THE POWER SERIES METHOD

Co 4 Co 5 Co 7
=c t+ cx — Ox? — —xt- =x x
Bue : 3 5 7

= oleae? ye tes Vek ee eee

J 3 5 7 ;
which has the form

y = coyi(x) + cy y2(x).

If the pattern in each series is clear, it is sometimes useful to write the solutions
in summation notation. For instance, in the above example we can write
xn

yxy) =1- > Happs

n=1

whereas y2(x) = x is already in this form. The ratio test can readily be applied to
solutions written in this form to determine the radius of convergence. In the present
example it is easy to verify that the series converges for |x| < 1.

Remark. In some cases the interval of convergence extends beyond the closest
singularity.

EXAMPLE 4 Solve y” + xy’ + y =0, y(0)=3, y’() = —7.

Solution Again we see that x = 0 is an ordinary point. Thus we have

Mesne pl cane ees een ee cet 03

n=0 n=0 n=0
co fo2} co

Sa a eee: ») NC,x" + »; Cree

n=0 n=0 n=0 '
n—n-2 n—n—2

Reindexing to obtain equal powers of x, we get

[e-0)

>) n(n — 1)c,x"? + ss(nh — 2)e,-9x"7 + See = 0

n=0 n=2 n=2

or

O-cox? + O-cyx 1 + y, [n(e = ey ch Dyce Oe eee

n=2

We now see that co and c, are arbitrary, and the recurrence formula is

nA — 1) Ce hs (1 —52) Cen, Ge Os
or, upon simplification,
SEC. 9.2 / THE GENERAL METHOD 299

Ceee deen CeSoe JUS ea | 55

n
Therefore,

i2 7 60

joe
3 30

] C

pa
5 5
heBLE
3 2: ’

ee
: 6x 2-4-6’
eee
c7 = ae eae!
aR IEG

Hence the power series for y is

Virco ix t Cx + Oye cax h cst oh cox? hi
Fie oe x sey ate? | EAE ees Ne pepiXe a ca
ees,
xe Lata aetes
= ee eS
col 4 Pell ewe ax A Ge GG ),
or

y = Coy(x) + cyy2(x).

Although it is not obvious, the first series can be written in terms of an elemen-
tary function. That is,

~-L4+—-
3g
+
LD) sia) Sa Sire aPC
N Ww

ll | | ab | te
NI Sake epi NI
+

Se bedi
fe
—

aS Se ae
a

or

series.
This is not the case for the second series, so it must be left as an infinite
Hence we write

erste
3
X
Bex agraPahl nx 5) 7
+e).
ae + a(x Be Caer
300 CHAP. 9 / THE POWER SERIES METHOD

Imposing the first initial condition y(0) = 3 on this general solution, we imme-
diately find that
Co 5.

In order to impose the second initial condition y’(0) = —7, we first calculate
4 6
7) x Xx
y’= —coxe r+ a1 7 fot fees ,

from which it follows that

C— =).

Thus the solution of the initial value problem is (for all x)

3 ) 7
-2/2 X X X
=3e 7%? — —-—+——- +e),
Pe i(: 53-5 3-529

If the initial values of y and y’ in Example 4 were prescribed at some point other
than x = 0, then the solution technique would be altered. For example, suppose we
wish to solve the initial value problem

yore Ayn te yo 0 aeyich t= 03Wey 1) rl. (18)

Since the initial conditions are prescribed at x = 1, it would be best to seek
solutions in powers of (x — 1) rather than in powers of x; that is, seek solutions
in the form

ig Os eee a (19)
n=0

To do this, we set X = x — 1 in (18), which leads to the equivalent initial value

problem

yor Xk Dy ty =O yO) =33, yy Oj 7, (20) :

where the primes now denote differentiation with respect to X. Setting X = x — 1
also in (19), we can now proceed as before.
Also, in some cases the substitution X = x — x9 is used to simplify the algebraic
structure of the DE. Consider the next example.

EXAMPLE 5 Find a power series solution of

Venki) y 4 1) yf
about the point x = 1.

We are seeking a power series solution expressed in powers of (x — 1). To simplify

matters, let us make the translation of axes X = x — 1, which changes the DE to

y" + X*y — 4xXy =0,

SEC. 9.2 / THE GENERAL METHOD 301

where the primes now denote differentiation with respect to X.

Assuming

we find that

SG DG et Ancacae acA = 0,
n=0 =0 n=0
nn 3 in 3

and with a shift in index, this expression becomes

O-coX7?+ O-¢,X~! + 2e.X° + Si [n(n — lea + (H — 3)Cn-3

n=3

Be ye 2a
Once more Cp and c, are arbitrary, but we must set cz = 0. The remaining constants
are then determined from the recurrence formula
n(n — 1)c, + (n — T)ep-3 = 0

or
lumen) =} oe) ee
CcPoe nh 3 1) Gaeasmlt
n—39

Calculating the first few constants, we see that a

ye RS) eae -
C3 3.2 Co 3-2 0>

SAAC, ON
Oe MUU nae g
—2)
Ck = 54 c, = 0,7

DE (ean RIT A
Coin ae 6154 ee 69523220 4
Gia ’

C3 = 0,

ee 24
OOS eg O 78 9655:6 0.
3
= — = 0
C10 10:9.” >

Cie 0,
302 CHAP. 9 / THE POWER SERIES METHOD

and so on. Thus,

Y = Co + OX + 2X? + cgX? + cg" +=
ei ah es 4 6 = 2-4 X? 9 +--+ ol(
X L +-XeE
;)
cot + 35x eer inacewy) 4
or, since X =x — Il,

ya all + she D+ (x — 1)° pee regs 1° +---

3-2 6-5-3-2 958: 6155972, 4s |
+a 6—1)+ aC — v'.

Because the original DE has no singular point in the finite plane, this solution is
valid for all finite values of x.

In some cases the recurrence formula involves more than two terms in the
unknown coefficients. Consider the next example.

EXAMPLE 6 Solve y” + (1 + x)y =0.

Solution Making the appropriate substitutions for y, y’, and y”, we find after simplification
foc) io) ie)

SSA = TE CoGD Get a> ea Sa),

n=0 n=0 n=0
n—n-2 n->n-3

Ssn(n — l)c,x"? + S Chg k!? + s)Cr-3x" 2 = 0,

n=0 n=2 n=3

O:cova cs O-cin ett (2cy ricotacta >) — Vera |x ames

iCrenict Gyaa
n=3

Thus, co and c, are arbitrary, and also

Co
Oana?

Cheats
¢, = oo an =3,4,5,....
nie)
To simplify matters here and still obtain two linearly independent solutions, let
us first set c; = O and let co remain arbitrary. Therefore
1
Se A Sei
2
 SEC. 9.2 / THE GENERAL METHOD 303

A
: A aera
Cc BRR C2 SRC ie WO

4-3 4!’
C3 + C2 4co
Cia => —

eC ee Cee eee Oe a ne

cep eee MES

Da ier Oe | =r COy Xs

Similarly, if we set co = 0 and leave c, arbitrary, we find

c, = 0,

pet aa
: 3-2 B17
pee Sone ed
4 4-3 At;

S).6 ese leas! Pele mien ra: nee, e)ne

giving the second solution

eee Oe
y=c |x ———+—+--+] =c y2(x).

The general solution is therefore (for all x)

y = cCoyi(x) + Cyy2(x).
epee ee es

EXERCISES 9.2

to rewrite the series as

In problems 1-4, make an appropriate index change
directed.

o~
&
oo
ts ee) ee spa
ere as a series in x”.
\A. > Dues 2
n=2 n

ES (Hece | ee
2. y a ea as a series in x”.
n=0
n

SW SCA) eee reorrpbe ts series in x°”'.

 CHAP. 9 / THE POWER SERIES METHOD

41 Le as a series in x”*'.
> @yay-- Gay
n=1

In problems 5-8, list all of the singular points in the finite plane.

‘5.
\

Giek ly" — Gry eye 0 6. xt xy) (Bik) yee cere

io ; 4
8/ y" —-———y' + ene
1
Lm Sy" + 6xy' - x*y =0 x(1 + 4x?) x71 + 4x2)

In problems 9-12, find a power series solution about x = 0 for the given first-order
DE.

(l+x)y’+y=0 10. y’+2xy =0, yO) =3

“11. y' + (sinx)y =0
Hint: Expand sinx in a power series about x = 0 and find only the first three
nonzero terms in the general solution.
12. Se P83
In problems 13-22, find two linearly independent power series solutions around the
origin and state the region of validity as predicted by Theorem 9.1.

at 3)
y" Hay = 05, 14, y"—4y =0
45) at 4x2)y" — 8y = 0 16.5 yy" Felxy = 5y = 0
17. (1 — x2)y" — 2xy' + 12y =0 18, (x? + 4)y" + 2xy — 12y = 9
19. y" + xy’ + (x? + 2y =0 202, Hat ixry tray = 0
21. «@-—Dy"’+y'
=0 22. @ +3)y"
+ Ge 2)yas y= 0

In the initial value problems 23-26, find the first four nonzero terms in the power
series expansion about x = 0 of the solution.
we

laa. Yi nay tye Oy (0) al ys (0) 2250

24. Weert) poe dy aesvey (O)e= Ome y (0) aaa
25. (xP |v Sxy ot xy me 0) 5y(0)i="4 Sy (Oho
(
*26. DY are) ata (x) el) Ya ey (Oe 00 yh(0) es)
In problems 27-29, find the power series expansions about x = 1 of the DE.
¢
1
yo (x — ly =O
xy te yr texy oe Omi eee Osage Clie)
(x? — 2x + 2)y” — 4@ — Ly’ + 6y =0
Find the first four nonzero terms in the power series solution of
y" — 4xy' — 4y =e".
Hint: Expand e* in a power series about x = 0.
 SEC. 9.3 / SOLUTIONS NEAR A REGULAR SINGULAR POINT 305

In problems 31 and 32, find the power series expansion about x = 0 of the general
solution.
*31. ve cox 3xy' =’ y = 0 339% ye ate Noy aL Sx! ae 3y = 0

"33. The DE
Voy, reeny 10,8 yu;

is called Hermite ’s equation .* Obtain two linearly independent solutions for the cases
when
(a) n=l.
(b) n= 4.
(c) Show that when n is any nonnegative integer, one solution of Hermite’s DE is
always a polynomial of degree n.
*34. The DE
(1 — x*)y"— xy’ + n’y = 0, n 20,
is called Chebyshév’s equation .+ Obtain two linearly independent solutions for the
cases when
(a) ore — al
(b) n=4.
(c) Show that when n is any nonnegative integer, one solution of Chebyshév’s DE
is always a polynomial of degree n.

9.3 SOLUTIONS NEAR A REGULAR SINGULAR POINT

some
The solution of a DE in the neighborhood of a singular point usually exhibits
type of peculiar behavior, and for this reason the behavior of the physical system
singular
governed by such an equation is frequently most interesting around this
technique ,
point. Therefore, rather than avoiding the singular points in our solution
we need to investigate precisely these points in many situations.
be possible
Unfortunately, when x = 0 is a singular point of the DE, it may not
to find a power series solution of the form

(1822-1901). Hermite showed in 1858 that

*Named after the French mathematician CHARLES HERMITE
and proved in 1873 that e is a
the general fifth-degree equation can be solved by elliptic functions
He made many contributions to the fields of
transcendental number (i.e., a type of irrational number).
number theory and analysis.
CHEBYSHEV (1821-1894). Much of his
+Named after the Russian mathematician PAFNUTI LVOVICH
work in probability, the theory of numbers,
work involved prime numbers, but he is also known for his
mation of functions . Other spellings of his name include Tchebysheff and Tchebycheff,
and the approxi
 CHAP. 9 / THE POWER SERIES METHOD

Under proper conditions, however, we may be able to find a solution of the more
general form

VS LD) Cadi eae

n=0

where s is an unknown parameter to be determined.

In general, a singular point of the equation

Vide GUY ae Dy (22)

has been defined as one for which either (or both) a(x) or b(x) fails to be analytic.
Singular points are further classified as regular or irregular according to the
following definition.

Definition 9.3 If x = Xo is a singular point of (22), it is classified as a regular singular point

(R.S.P.) if both
(x — x)a(x) and (x — Xx)*b(x)
are analytic functions at x = Xo; otherwise, we say that x = Xo is an irregular
singular point (I.S.P.).

Remark. In the special case when a(x) and b(x) are rational functions reduced to
lowest terms, an R.S.P. is a singular point for which the factor (x — x9) appears
at most to the first power in the denominator of a(x) and at most to the second power
in the denominator of b(x).

EXAMPLE 7 Classify the singular points of

XY ny ye =

Solution Clearly, x = 0 is the only singular point, and dividing the DE by x shows that
|
a(x) = 2 and b(x) = 1.

Thus, since both xa(x) = 1 and x*b(x) = x’ are analytic atx = 0, we conclude that
x = Ois anR.S.P.

EXAMPLE 8 Classify the singular points of

X= 1)? ES) y Sey BF (Ox + ay 0,

Solution The singular points are x = 0, 1, —3. Putting the DE in normal form identifies
oer
EET RG LLY
 SEC. 9.3 / SOLUTIONS NEAR A REGULAR SINGULAR POINT 307

and

2x2 th)
eax 423)"
Therefore, since x and (x + 3) appear at most to the first power in the denominators
of a(x) and b(x), it is clear that x = 0 and x = —3 are both R.S.P.’s. Likewise,
since the factor (x — 1) appears to the second power in the denominator of a(x),
x = 11s anI.S.P.

9.3.1 The Method of Frobenius

In 1873, a method was published by the German mathematician G. Frobenius* for

finding a solution of a DE about a regular singular point that is based upon the
following theorem.

Theorem 9.2 if — 0 iS an R.S.e- o1 the DE

Vee aA) Very 0) ye 0;

then at least one solution of the form
co fo2]

ya cyt Crue ere 0,

n=0 n=0

always exists with a positive radius of convergence.

In finding a solution of the form suggested in Theorem 9.2, it is necessary to

determine values of the parameter s in addition to the cocthicicntSico, Ci. Co. ee
The restriction x > 0 is necessary to prevent complex solutions that might arise for
certain values of s. If we need a solution that is valid for x < 0, we can make the
simple change of variable X = —x and solve the resulting DE for X > 0.
If we formally differentiate the series

we find

and analysis. In addition to

*GEORG F. FROBENIUS (1849-1917) is known for his research in algebra
singular point, he made great contributions to the theory
his infinite series solution of a DE at a regular
of groups.
308 CHAP. 9 / THE POWER SERIES METHOD

and

y"=SYatsnts— Dox eS
n=0

Using these expressions for y, y’ and y”, we can rewrite

y" + alx)y’ + b@)y = 0

as

DS Gites tes W)cey ee te OO)Dytea Sct a cet ee

n=0 n=0 n=0

or, more compactly, as

Sy) (Gris) eis 1) ea 0res osye x OO) ee 0. (23)

n=0 ;

Assuming x = 0 is anR.S.P., both xa(x) and x7b(x) have power series expansions;
1263)
xa(x) = do + ax + @x?+---,
DG) = boa Dyan Dyk ae
If we substitute these expressions into (23), the smallest exponent of x occurring
in (23) is (s — 2), corresponding to n = 0. Let us separate this term from the rest
and write

[s(s — 1) + aos + bolcoxs? + Ss[(n + s\n +5 — 1)

n=1

h(a Fae Pin eS) Eb iD eee) oe een

As before, this equation can be satisfied only if the coefficients of all powers of x
vanish independently. This yields a system of equations in the unknown constants
C,. In particular, we must set

[s(s — f) + aos + Doleg = 0.

But in order to obtain at least one solution of the DE, we must be assured of at least
one arbitrary constant. We select this arbitrary constant to be co and therefore set
its coefficient to zero, giving us
5° =F (ao me l)s aie bo = 0. (24)

This important quadratic equation is called the indicial equation of the DE (21).
There are two roots of the indicial equation, s, and s;, leading to three different
procedures for generating a second linearly independent solution, depending on the
nature of the roots:
 SEC. 9.3 / SOLUTIONS NEAR A REGULAR SINGULAR POINT 309

Case I—Roots differing by a noninteger

Case II—Equal roots
Case III—Roots differing by a nonzero integer

We will consider the three cases separately under the assumption that s; and s, are
real solutions of the indicial equation (24). Although these roots can be complex,
we will not discuss this case.

9.3.2 Roots Differing by a Noninteger

This case always leads to two linearly independent solutions of the form

Vi = XY) cni)xs. Gols) = 1, (25)

n=0

and

2 = X25 Ca(2)x", Co(S2) = 1. Pees)

n=0

constant for
These solutions are clearly linearly independent, since y)/y2 cannot be
are obtained
our choice of s, and s). The two sets of coefficients c, (s,;) and c,,(s)
recurrence
independently by replacing s by s; and s by 52, respectively, in the
formula.
Ee Le ee ae
ee
EXAMPLE 9 Solve 2xy” + (1 — 2ny’ — y = 0.
the infinite series
Solution We first observe that x = 0 is an R.S.P. Next, substituting
expressions for y, y’, and y”, we find

> 2nd se cries (2 SOK) Yon aii

n=0 n=0
co
Loo}

Setar Glees 10s ee Ul

n=0 n=0

n—>n-1 n—>n-1

combining all terms under one

Making the appropriate change of indices and
summation, we are led to

(Os(st) Les icone Ss[2n + s)\(n +s — Ve,

n=1

+(n +s)¢, — 2(n +8 — \en-1 — Cele = 0.

the indicial equation
The coefficient of co equated to zero gives
310 CHAP. 9 / THE POWER SERIES METHOD

2s? — 5 = 0*
with roots s; = 0 and 5) = 5. The recurrence formula is
(n + s)[2(n +s) — Je, = [2(n + 5) — 1en-1,
or, since 2(n + s) — 1 # 0 for either choice of s,

(a + se, = Een SH 27 ee

Now putting s = 0 in this recurrence formula and setting co = 1 -for con-

venience, we find c, = C,—\/n, which gives

ey = els

Cy = co = 1,

St
2 5 J?

Cae fl

2 mipane:
5 eecee ak
ult 4e eate

leading to the first solution

2 3 4
yi 0
=a(1te+ eater
b+ heuaei
S4 apie
te Eh) Jee.
BYbees

Similarly, for s = 5 we get c, = 2c,-\/(2n + 1), ice.,

Co = 1

aenee
| 3 Be

e GMs
Daeees a5
é ee 2:

AES PDE TL

and thus

9x 2x? Dee
yo(x) = x7( 1 + — +
Ne ( 3 S. 3u5et
The general solution is therefore

*Observe that we could find the indicial equation directly from (24) by noting that xa(x) = 3 — x and
x?b(x) = —x/2, and hence ap = 4 and by = 0.
 SEC. 9.3 / SOLUTIONS NEAR A REGULAR SINGULAR POINT 311

y = Ayi(x) + By2(x)
Dane S400
= Asspsl tenet ec ee eal}
where A and B are arbitrary constants.

9.3.3 Equal Roots

When s, = 52, the procedure used in the last case will yield only one solution.
Nonetheless, two linearly independent solutions can be shown to exist in this case
corresponding to

yitx) = y, 51) = ny Gh ps Saree) Saale (27)

n=0
and

Pe ae S=S|

= y,(x) logx + I CASi)t aes (28)

where -

Wise oik. eh 3 Cs, (29)

n=1

Remark. By (29), we simply mean the general function obtained for arbitrary s,
which becomes y; upon setting s = 5).

Another way of obtaining a second linearly independent solution in this case is

to use the method of Section 4.3. That is, a “second” linearly independent solution
of y” + a(x)y’ + b(x)y = 0 is given by the expression
exp[—JS a(x) dx] Hes 30
yx) = yi) | yi(x) ee

where y(x) = y(x, Si).

aren
nen
ee
EXAMPLE 10 Solve xy" + y’ + xy = 0.

Solution The assumption y = 2 c,x"** leads to

[s(s — 1) + s]eox’ |+ (ese Liew ot S [in + s\(n'+ s — le,

n=2

pe oes ee all
312 CHAP. 9 / THE POWER SERIES METHOD

Setting the coefficient of co to zero, we see that the indicial equation reduces to
s? = 0 with double root s,; = s, = O. For this choice of s, the coefficient of c; does
not vanish and so we are forced to set c; = 0. For values of n greater than 1, we
have the recurrence formula
(n? + S)'Ce 4 Cee 0, Bh = 2, 3, Al ea
or

In this last form of the recurrence formula we are emphasizing the dependence of
these constants on the parameter s. Since the solution formula (28) requires deriv-
atives of the c’s with respect to s, we will not substitute the value s = O into the
recurrence formula until later. From the recurrence formula, there follows (with
Co = | once again for convenience)

C(s) = ee Oem 2 ee
Y GE) Gey *
c3(S)
teeee
(s aE 3) 0,

c4(s) = eel ea ee
(s+ 4)? (s + 2)°Xs + 4)’
c5(S) = 0,

Cc (s) = ee hart = — Be ee

; (GEG) Soa)

while in general we deduce that c, = O forn = 1,3,5,... , and for even values
of n we write

us alle
Ch
lig 3 m
Semen
= EP
(ten
EP WL
een?
ir a

where m = 1, 2, 3, . . . . To obtain the first solution, we now set s = 0, finding

Eq (0) = ee
—j| m =j| m

(2Y(4) - + (m?) 2?" !)?’

and therefore
a (ea ns 00 (- Tye
=a] ce
x 2m
y(x) 2 2m mij a) —
YY (m!)? (5
a ) ;
m=0

Next, using logarithmic differentiation, we deduce that

Com (S) = i Oh et
eer _S Com(S),
SSP 0
 SEC. 9.3 / SOLUTIONS NEAR A REGULAR SINGULAR POINT 313

which for s = 0 leads to

; (eat : 1 l 1
+ototer-s +}.
Com) =Damm (1 ny ve 42]
It is common to introduce the notation

1 ae | 1
| |Htkcactl MG ensseigen iat ag chiidadtm
iS) m

for the partial sum of the harmonic series. Thus, from (28) we obtain the second
solution

y(x)
; = yi) logx + ec
(7m) ae Jane
The general solution, valid for x > 0, is

y = Ay,(x) + By2(x)
(—1)"(x/ 2)?” ro) (-1)" 'An x 2m

= (A + Blog Se ene 3

ce ta lt Sl lage See ae I a 2 a

An alternate solution technique for finding y2(x) in Example 10 rests upon the use
of (30), which leads to
dx
Yo 0) | ag:

Performing some lengthy calculations, we have

Pye 1
yiix) [1 — (2/4) + (464) — (9/2304) + ig

igre eaie Sane al eee te ae

Wit 2) Ox 7/32) = (6x9/5716) a
Xe SK ee free,
= a
ae a3 aesI6
this last step resulting from long division. Thus,
Tae we gee
yx) ~=iG) |(7seat Reape
eR eer. oe
ae hs

a ‘as a sao k ae Ee
See 1) OE Ai fie 198 13456
By x2 «5x* = 23x°® 1B G6
eea ee e | ee |.
=
yi) logx + ( 4 64 MG * 128 ” 3456
or
314 CHAP. 9 / THE POWER SERIES METHOD

e be ee ‘i ise
yalx) = vila) logx + 7 — 58 + F3e54 (31)
We have therefore obtained the first few terms of the second solution by this
alternate technique. Although it is not immediately obvious, this is the same
solution found in Example 10 for y2, the verification of which is left to the reader.
The solution function y, in Example 10 defines a particular function called the
Bessel function of the first kind of order zero and is denoted by the special symbol

yi = Jo(x). (32)
In most applications involving the Bessel function, it is customary to choose a
certain linear combination of Jp and y, and to take this combination as the “second”
solution of the DE rather than simply to take y). This special combination is called
the Bessel function of the second kind of order zero and is defined by

Yo(x) = <[yx(a) + (y — log 2)Jo(x)] (33)

where y is called Euler’s constant given by

y a, in ( 1
eSSee 1 ee 1
z og) = (0. S772 :
noo

Using the Bessel functions, we can then express the general solution of
xy" + y' + xy = 0; (34)

known as Bessel’s equation of order zero, in the form

y = AJo(x) + BYo(x), (35)

where A and B are any constants.

9.3.4 Roots Differing by a Nonzero Integer

Let us suppose s; — s, = N, where N is a positive integer. Thus s, > s>, and in this
case there exist two linearly independent solutions of the form

LED Se BBO) = SsCa(Si)e"

"1, ColS i ale (36)

and 5

yale) Z = - S = 8)

= Ky\(x) logx + 3 CACSa yes (37)

n=0
where K is a constant that in some instances is zero, and

yx, 8) = vy CaS) oy Cols) 45 ss (38)

n=0
 SEC. 9.3 / SOLUTIONS NEAR A REGULAR SINGULAR POINT 315

Therefore, we are always assured of one solution of the form (36) corresponding
to the larger root s, of the indicial equation. Using the smaller root, however, we
can always produce two solutions of the DE, although there are two subcases to
consider here. In most situations we can produce the second linearly independent
solution by a procedure similar to that used in Case II in Section 9.3.3 which leads
to a logarithmic type of solution as suggested by (37). An essential point of
dissimilarity between the two cases is the choice co(s) = 8 — 52 in constructing
y(x, s), rather than setting co(s) = 1 as before. By constructing the function y(x, 5)
in this manner, it can be shown that

yitx) = y,, $2) (39)

is equivalent to calculating y, by use of the larger root of the indicial equation.
Hence, by using (39) instead of (36), fewer calculations are necessary. In some
special cases the use of the smaller root of the indicial equation will lead to two
linearly independent solutions, neither of which involves a logarithmic term. Typ-
ically this will happen when 5s, is a positive integer and both co and c,, turn out to
be arbitrary. Otherwise c,, will be impossible to calculate (with the assumption
Cy) ¥ 0), and the general solution will then contain a logarithm. It is usually best to
first try the smaller root s> of the indicial equation in the hopes that it will produce
two solutions of the DE not involving logarithms. Let us illustrate these cases with
some examples.

a SSS aetna eee

ee

EXAMPLE 11 (Nonlog case) Solve xy" — (4 + Rye teeye— Ue 0,

Solution The substitution of y = os> c,x"*’ into the DE leads to

> (n+ s(n 3s — Dex = SY) 4(n + jet

n=0 n=0

a » Noe as 0,
Bes > (n ae Genie

n=0 n=0
> ——_——_——_
a,

n—>n-1 n—n-1

which simplifies to

+ s)cn
Be = ir Aslegx® 5 + 5 [7 + s\n + 5 — Ic, — 4
n=1

—(n +s — Wer-1 + Dome ie a3

s; = 5 and 52 = 0 as roots. If
Therefore, the indicial equation s? — 55 = 0 gives
formula reduces to
we try the smaller root s,; = 0, the recurrence
n(n — 5)cn = (n — 3)en-1,. 1 = ee Oe, ole
5, it is best to write out
Since division by (n — 5) is not permitted when n =
separate relations for the c’s through n = 5. Thus,
316 CHAP. 9 / THE POWER SERIES METHOD

n=1 —4c, = —2co,

n=2: —6c, = -c),

n = 3: —6c3 = —0°c2,

n=4: —4c, = 63,

n=5: O-cs = 2¢4,

from which we deduce

cy = ue C2 re oe Con. 3 =c,=0,

and since 0-c; = 0 is satisfied for any value cs, we see that cs is arbitrary. For
n > 5, it follows that

=
(n — 3)Cn-1 Sag)
a7 tae ens
ie Ce) Lae
Proceeding now as usual, we find

cau
6 6:1 5>

4 3-4
a ane OO ee
Polat, |e eenase)
RE per
@: 19" a) (eo) 6; Ue: 6:8 6)iyo) 0) ome 0), (ee) fal. of rer ..0-<e:

Collecting terms, we have the general solution

+ cox? + eax? + cgx* + ogxi +++ -),
y = x%(eo + eyx
or finally

y =o ol(1+ *
=x + =x?
|

3 3-4 3-4-5
+ ¢5x5( 1)+is—ex + 7 r r-
CRESS om
es ih eR,

EXAMPLE 12 (Nonlog case) Solve x*y" + xy’ + (x? -—)y =0,x>0.

Solution Proceeding as before, we have

Ms(n + s\n tos = Dex is (n + s)e.x"*

n i] 0 n=0
 SEC. 9.3 / SOLUTIONS NEAR A REGULAR SINGULAR POINT 317

ee) co 1
Fs42 se
+ See = rine ‘= 0,
n=0 n=0
n—n-2

. |
ri, C + s(n +s — Mey + (nm + S)ep + Cn-2 — mas
4 20:
n=2

We find that s; = 3 and s; = —} are roots of the indicial equation s? — ; = 0.

Using the smaller root s; = — 4 leads to both cy and c, arbitrary in this instance, and
the remaining constants are then determined by

Cn-2
Ch = n = 2, 3, 4,
‘ nin — |)
We obtain successively

Co Co
——
Oe 2!
C\ Ci
QS SS SS=
caper 320 owae 73!
C= -
C2 =
Co
oS ae

E ey al
C3 Ci
a, pg
5-4 me Nem
5!

and thus
Viera (Cot top On tsk te Cv te)
= = 1/2
Kien 2
cA gs2
kgpice?4 x6
SSeS ew. VP =
CE2 3 ea 5
44 22 es
S7 Ee 6 6 ‘

co 2 Al 6h )+s as 305!
which can also be expressed in the form
cos x sinx
Ween ONE. x >.

EXAMPLE 13 (Log case) Solve xy" + 3y’— y = 0, x > 0.

Solution Ify = Zenx"™, then

318 CHAP. 9 / THE POWER SERIES METHOD

> (n + s\n +5 — le,x"s! + >) 3(n + s)x"™* = a c,x"** = 0,

=0 n=0
n=0
n->n-1 n—>n-\|

[s(s — 1) + 3s]eox*! + S [((ietes\in des eal )c,

n=1

+ 3(n + s)O, — G1" >= 0.

Here we find that the coefficient of cy set to zero gives s* + 2s = 0, or s; = O and
5) = —2. Making the substitution s = —2 in the recurrence formula
ig =F ST) it Salen 2 Cee aC)

yields
Ni —=2)C
= Cay eat Slee
ee

Again we do not divide by the factor (n — 2) until n > 2. Thus,

n=l: —C; =C,

n = 2: O-co
= C1,

and

Cn-1
Bl ees eo Rear! Fake Pete te
: n(n = 2)

These relations can be satisfied only for c} = co = 0, which contradicts our as-
sumption that cp # 0. Therefore we have failed to produce two solutions this time
that do not involve logarithmic terms. (It may at first seem that the way to proceed
is to set co = c; = O and leave c, arbitrary. Doing so will only produce one
solution, which will be reobtained automatically when solving the DE by the
method discussed here.)
Let us write the recurrence formula above in the form

mstnta?
GF "Ps
Cei(S)
CAS a=

where we now choose c¢o(s) = s + 2. Therefore,

iC) epee Oa as
(s + I)(s +3) (s + 1X(s + 3)’

(5) = ao a a
(Set 2) ire 4) (Se 1) (8 Sse a)

€3(S) = (A Om Ss = Xe Seah Ore

(Sits3)(SeFi Se (sal) (5 3) (says)

c4(S) = =
(s + 4)\(s + 6)
SEC. 9.3 / SOLUTIONS NEAR A REGULAR SINGULAR POINT

Z 1
~ (s + Is + 3)%X(s + 4)°(s + 5)(s + 6)’
and so on. We have

y(x, s) = ey C.CS)x"
n=0

(s ak ace x5t2
(See ei
(Sta 1CSa 3) sea) Gra 3) t- 4)
xst3 +

bi(s + 1)(s + 3)*(s + 4)(s + 5)

xst4
+

Gaines Sac
from which we can generate y, by setting s = —2; i.e.,

yi) = ys S2)
= 2 Spee 1Oe eee
% ee Xeae oe x?
on ee!
EPR ESS Sonam Ten ST
or
oc xe .

eure ae
Also,
dy (x, 5)
a
See 5 + 2
ee
(st 2)! |
1 1
eee, ee
1
Oe De A TEENIY ©EEN hee
a 5ue Seis reel za
(sa seats 4) Savi ly se-asiy sy 4
x5*3

¥ (s + 1)(s + 3)°(s + 4)(s + 5)

1 ME 2S, tater CLARE, 1
Sete gee Se oeae Se
xot4

" (s + 1s + 3)°(s + 4s + S5)(s + 6)

and therefore,
320 CHAP. 9 / THE POWER SERIES METHOD

(Ore dy (x, S)
va Os s=-2
1 1 1 11 213
= y,(x) logx + see
ae eh + ee4 + ee 576.Yo cate
36° ee

leading to the general solution

y = Ayi(x) + Byx), x > 0,
for any constants A and B. The solution is valid for all x > 0, since x = 0 is the
only singular point of the DE.

EXERCISES 9.3

In problems 1-8, locate and classify the singular points of the DE.

1. x°@? — Dy” = x0 + Dy + @— Dy =0
Qi Katee |)Y airy es ye 0
3. x4(0x? + ID — 1)?y" + 40°e — Dy’ + & + Dy =0
4. y" +x =0
5. x*(x — 4)?y" + 3xy’ — @ — dy =0
6. x(x + 1)?y" + (x? — ly’ + 2y =0
7. xy" +(x - 3)"47 =0
8. (x + x4 — 6x>)y” + 3x’y' + — 2)y =0

In problems 9-18, show that the roots of the indicial equation do not differ by an
integer and obtain two linearly independent solutions by the method of Frobenius
about the point x = 0.

9. 2x"
+ (1+ xy’ —2y =0 10. 4xy” + 3y’ + 3y =0

11. 2x” + 5(1 + 2dy’ + Sy =0 12. xy" + x(x= 3)>"+3) =

13. 2x?y"
+ 3xy’ —y =0 14. 2x?y”
+ xy' -—y =0
15. 2x*y"— xy’ + (1+ xy =0 16. 2x’y"-—xy’ +(x — 5)y = 0
17. 2x*y" + xy’ + (x? — 3)y =0 18. 2xy” + (1 + 2x)y’ — Sy = 0

In problems 19-26, show that the roots of the indicial equation are equal and
obtain two linearly independent solutions by the method of Frobenius about the
point x = 0.

19. xy” +y'—4y =0 20. x’y”

+ 3xy' + (1 — 2x)y = 0
21. x’y"-x1+x»y'’+y =0 22. 4x*y" + (1 — 2x)y =0
23. xy” + Sxy’ + 4y =0 24. x*y" + 3xy' + (1 + 4x7)y =0
SEC. 9.3 / SOLUTIONS NEAR A REGULAR SINGULAR POINT 321

25. xy" + x(x — ly’ +(1—-—xy =0

26.; xy’ + Gd —ay’
-—7» = 0

In problems 27-36, show that the roots of the indicial equation differ by a nonzero
integer and obtain two linearly independent solutions by the method of Frobenius
about the point x = 0.
Nonlog cases
2f. x*y" + 2x(x — 2)y’ + 2(2 — 3x)y = 0 74% aah! So lbs SA yy shy aa)
29. xy” — (x + 3)y’ + 2y = 0 30. x?y" + x’y’ — 2y =0
31. x*y” + x?7y’ H& = 2)y = 0
Log cases
32. xy" +y =0 RK G3) Salers
yyy arty 0)
34. x7y" + (x? — 3x)y’ + 3y =0 BOs etcy eK lat) yeh Sven 0
36. x*y" + xy’ + @? -— Dy =0
*37. The DE
x1 — xy" +[c —- (a + b + I)x]y’ — aby = 0,
where a, b, and c are all constants, is called the hypergeometric equation, and its
solutions are called hypergeometric functions.
(a) Show that x = 0 and x = 1 are R.S.P.’s of the DE.

(b) Assuming c #0, —1, —2,... , show that one solution of this DE is

= Pie oem _ab_ Aart )b(O ele

yi(x) = F(a, b; c; x) 1+ ait oe at
+ Gea+ Dal ,

(c) When a = 1 andc = b, show that the series in (b) reduces to

yi(x) = FU, b; aes

(d) Assuming 1 — c is not an integer or zero, show that
Seabee ott aoe Oh IP Tonos 4)

is a second linearly independent solution of the DE.

"a8. The DE
Wy te Cue X) Ve ary sea,
where a and c are constants, is called the confluent hypergeometric equation, and its
solutions are likewise called confluent hypergeometric functions.
(a) Show that x = 0 is an R.S.P. of the DE.
(b) Assuming c # 0, —1, —2,... , show that one solution of this DE is
PON
yilx) = M(a;
c; x)
ea ae paa l)
r «,
coer es aeanply:

(c) When c = 4a, show that the solution in (b) reduces to

s, denominator parameters, and the
*The semicolons used in this term separate numerator parameter
argument x.
322 CHAP. 9 / THE POWER SERIES METHOD

yi(x) = M(a; a; x) = e*.

(d) Assuming 1 — c is not an integer or zero, show that
yx) = x) (MG le en)
is a second linearly independent solution of the DE.

REFERENCES

Boyce, W. E., and DiPrima, R. C. 1977. Elementary differential equations and bound-
ary value problems. 3rd ed. New York: Wiley.

Coddington, E. A. 1961. An introduction to ordinary differential equations. Englewood

Cliffs, N.J.: Prentice-Hall.
Kaplan, W. 1958. Ordinary differential equations. Reading, Mass.: Addison-Wesley.
Rainville, E. D., and Bedient, P. E. 1974. Elementary differential equations. 5th ed.
New York: Macmillan.
Zill, D. G. 1979. A first course in differential equations with applications. Boston,
Mass.: Prindle, Weber & Schmidt.
‘Answers to
-Odd-Numbered
Exercises

CHAPTER 1

Section 1.2

1. linear, second-order 3. nonlinear, first-order 5. linear, third-order 7. linear, fourth-order

9. linear, first-order itl linear, second-order 13. linear in x, first-order 15. nonlinear, third-order

17. nonlinear, first-order 19. nonlinear, first-order

Section 1.3

15. m=2 7: iO 19. m=0,1, —4 21. y = Ae*

isle Se 18)
23. y= 25. y = Asinx + Bsin3x 29. xy’ — 3y = 12 31. y’ — 2y = 4x
ae EE
33. (y’? — 9y7=1 35. DVN — 2 Aeetny 37. y =O (singular solution) 39. no

Section 1.4

1. y= e? 3. y = sinhx 5. y = 3x" 7. exe)

9. y = e*(cosx — sinx) y=x 13. y =Cicosx (C, arbitrary) 15. y =0

A ee ke Oe its

CHAPTER 2

Section 2.2

mare : il ae 59 j (-aF *)
ex to = : =
oe A ley OX,

1 ie 9. P'=-+Ce™”™ 11. logN =(t - Det? -—t +C

7. —x +-sin2x — cosy =C
2 4 a

13.
(x? ik {2 = Cy ae ive.
1S: = C G7) Pea a G

19. x+y +log>=C x

91. y= ele) 23. (y — 50) = 2C(y ar x) ate (6% = 0, y= 0

323
324 ANSWERS TO ODD-NUMBERED EXERCISES

25. 4y?=9%e7+ CP 27. y =yoe™ 29. sinx — xcosx = e(y — 1)

1 — Cx? a
31. (c) y== +2 33. K (a) y = ance: 35 = ———[(1 — C’)sinx + 2Ccosx],
y= Ty ell
(Db) Vieorad y= exyo)

(singular solution)
(c) no solution

Section 2.3

1. x°-— 3x4 -—y?=C

7. u? — ue*® + sin3v =C
@ #Koonc cosxsiny — log (cosx)=
sin?@ + r°(1 — 07) =C

xy(y — 1) =C
a it y-¢
13. (xy #y2e7 =C 15. 2x — log? +y7)=C -=-x7¥°\+ C
y 2
3y 4,3 5,,3 y a Gremean
19. @+y)+—=C a. xy Grxyj=e

Y 4 2 1 2 1
5. y =Cie~* 27. y=e*|C,+ ae im Aesr ie + log(y + V1 + y’)]
y
1 u
33.
33
2logx
Ogx
+>y -—-
x
ty? =
3 2
5 log u + v*) + tan! (“) G

1 x
logy + —=tan '(
va") =C (x — y)logx +ylogy = Cx + y
& J) y
41. 2x + 2y)
+ (@ + y)log + y) =0 x’y3 — 3xy + 2y7=C
1
ae a + le ty dee yaad
a re gh Bee
4x3y? = may a
49. ES —y)= a +C
» 3
ax”
— dy? + 2bxy = 2c

Section 2.4

Jey = Cie." 5. y = Cie"? 7. y =Cie 9 y=C\x “*e

C :
1. y= rae 13. y =2e-% 15. w= —Te* 17. *y¥i= eG
2
19. y=eo* 21. y=x 23. y = 25. prob. 21, continuous at x = 0
4 v 2 hse as
prob. 22, discontinuous at x = 0
prob. 23, discontinuous at x =i
prob. 24, continuous at x = nz,
n=O Deere

Section 2.5

1. 1 Cie bs 3
y area Lee
3. y =-(sinx + C;) = cosx 5. y=x‘[@ - le*+ Ci]
xX

7% z= %t Cie-* 9. y = (x + C)cscx 11. = 1" C(x? ite Oye

Ve .
13. y=(@+Ci)e™ ecole dleeach (3 17. y= b k
Gamat Gixe
a
 ANSWERS TO ODD-NUMBERED EXERCISES 325

E 1
19. yy = 5) + Ciel*
21. i(t) = = es By Hy 23. y = 5(2x + 3)" log (2x + 3)
a5 sinx — seca 1+ 77/4
x
be?
ten 2 OS = 1+ x?’
27. = 350 yi
lat Cem
2 fe=G Ye 40 feet Ae? i pete
1+ x?

Section 2.6

LG:
3. y?=— + C,x* §&. xy =2+.Cy'"
1 ee

CHAPTER 3

Section 3.2
Con x?

3. y=ke* 5 ay =k (@) logy + og (53+1)=

i yy? =k
xt

9. = log{3 -—>]
2logy
2x?
+k Wye
5/3
= 5
ek 15. r = b(1 + sin@) 17. r?>=bsin@
uy
1 — g~2Vcgimt

23. v=1/— | ~T) Cc “ +e cg/imt

a es Cc
25. (a) v = 32011 — e “"), Oo<t<5
2 T+ 0.775¢e°"7
(b) v = 16 Tu 0775645 ete

(c) v« = 320 ft/sec if chute never opens

v« = 16 ft/sec if chute opens

yell) 1]
29. 3.1 mi/sec Sills v = 493 ft/sec 33) approx. 16 hr, 16 min
27. Wo

35. approx. 20 min, 47 s 37.

x2
1 layt! — 1 ina 1 ) ay y =-al(1- = - no
24 units, — units 43.
2 k+1 fe = Il w

o=sfres[(3) - Th .
k k]2) 1/2
y? = 2Cx + C’ (parabolic shape)
2\\a k

Section 3.3

1. 9.33 g (3) 55,820 yr (S)

/
1.142 x 10 32 Ti UP@y= (LOO) +1Po = 3) ci
+ 1 + 100b(3 sint — cost)
steady-state solution

dM
M(0)=M, 11. v(t) = 400 — els)
(b) k = 1: M(10) = 1.790Mo (c) ae = 0.06M,
9.
M(t) = Meo
k = 4: M(10) = 1.814Mo
k = 365: M(10) = 1.821 Mo
M(10) = 1.822Mo
EoCw w
m Doc
13. (a) 1 = hoa(1+2) 15.) i(t) = 1 — R°C2w? T= R200.
Cc mg
(b) 0.884
 ANSWERS TO ODD-NUMBERED EXERCISES

E,
(a) i(t) = rl aa emtey te ine —Rt/L 19. 58.5°

: ER oe E,Lw 21. x(t) = 600 — 550e7"'°

(b) i(t) = Rn isin + R?2 = La?
(e —RUL
— cos wt) + ine *””
x(60) = 231.31b
xx» = 600 Ib
Ne pear < 10
(c)
Oeaiee tes LO)
3) @ x(t) = 1200 — 1150e~"” (b) x(t) = 400 — 350e°"'
x(60) = 258.5 1b x (60) = 207.91b
Xe = 12001b xX» = 400 lb

CHAPTER 4

Section 4.2

1. independent 3. dependent 5. independent 9. Ce* 11

15. Yar Cex e™

Section 4.3

y= e 2 3. y= xe>*
1. 5. y2 = cosx 7. yo = e*sin2x
Ih sp oe ae
9. yo=x WW. y=x'? 13. y. = log 15. y.’=x ““cosx
re

Section 4.4

y= Ci =f Gen Si y= Cie Cer: ,

< Il e*(C,cosh2V2x + C)sinh2V2x) ile = (Gi Ain Cx )e*

y (C; se (ay e© 11, y =(C; #Cyx)e*

ae V3 NB
y=e* SONS + ee 15. y =e*(C,cos4x + Csin4x)

2 Bt AVA]
y =e SNS ye! + re! 23. 4D* — 17D — 15 25. D?-1

DEAN tx" 29. x°D*> + 2xD — 6 Sian — Crosses Goemure

y= Gene + (Grew
35. «y = Cie + Cye™" + Ge *
1
37. y =e 3
9 =
ee ny!
en! —_¥ e
)

Section 4.5

dependent 3. independent 5. dependent

y = (Cyt Cox + Cyx”)e* 11. ye (C\ = Cox + Gan )\eas + (Cs + Cax\en*

V3 V3
y = (C, + Cox + Cax2)e™* 15, “yas Cer i e.* (cscos*Ss + Cassin)

a y=
y=
C, aia Gene
Gien ae Grew:
ar C3e*
te Gren
19. ye C\ + (C2 =i Crew

21. 23. Be (Cy ae Gen ae (C3 ae Cixj\enee

 ANSWERS TO ODD-NUMBERED EXERCISES 327

25. .y = Clem + Coen + Cre? + Cre” 27. y =C, + Cox + Cx? + (Ca + Csx) cos 3x + (Co + C7x) sin 3x
+ (Cg + Cox + Cyox? + Cix*)e*
29. y = Ci + (C2 + C3x) cosx 31. y =(L—}x)e™
+ (Cy + Csx) sinx 33. y =2+ 3x + 2x? + 3x?
QS) y = (Ci + Cox + Cax? + Cux*Je™ + e*(Cs + Cox) cosx + (Cy + Cax)sinx + Coe + Cyoe™*
37. (D*® — 7D? + 17D — 15)y =0 39. y =e *[{(C, + Cox) cos2x + (C3 + Cyx) sin2x]
41. (a) y=0 (b) y = Csinnzx, where C is arbitrary andn = 1, 2,...

Section 4.6

1. y =C, + Core™ + 3(cosx — sinx) 3. y =(C, + Coxje™ + je"

5. y= C, cos (2V2x) == (65 sin(2V2x) 7. y =C,cos3x + C,sin3x + #xsin3x — $x? cos 3x
+ ge * + 3x 9. y =C,cos2x
+ C)sin2x — 4xcos2x
-}
Ailemeye G1 ereia Ooeme 2x (xl )e~ 13. y =C, + Coe + gxt t+ 3x? — 2x
3 — =—2% = =x,
5) y=e
=e */2 C, cos 5 eaetaG>
v3 Sin
a bell aei W7.a.y) = Gye (Ga Gaxye =) (ey)e

: can x Ome enh

1
- PAS — 2sin2x) 19. y = Ce"? + Coe? + Cscos= af Casin5 i ve

24 yp = Ax? + Bx + C +\Dx‘e*| - 23) yp = (Ax? + Bx2,-+-Gx + D)e™* + Ex? + Fx

1 fe Pe 11 8
25. eS no — X)sinx — cosx 27, y = —mcosx — ree) aa BOS + 2xcosx

29. y =AG@— le “+ (1 — 2x) 31. y =(C + x)sinx, where C is arbitrary

6 3+4e
7 Zsin(4)] | 1 A
ce. pS SSense 22 |Sn SO
5 2 cos (4) 5

Section 4.7

1. y =Cye™* + Coe* + 3(2x — le" 3. y =C,cos3x + Cysin3x

&) y =(C, — log|cosx}|)cosx + (C2 + x) sinx — 500s 3x(x — gsin6x) + qgsin?3x
7. y =C,cosx + Cysinx + gsec?x 9. y =C,cosx + (C2 — log|cscx + cotx|) sinx
— 3+ 4sinxlog|secx + tanx|
11. y =(C, — log|secx + tanx|)cosx (3) y =(C, + Cox + 3x%)e*
+ (C2 — log|escx + cotx|) sinx 5) y =(C, + Cox + 3x? log|x| — 3x2%e*
ieee Gie a C,ema sn (Can) e COS(es,) 19. y=(C, + $cot?x)cosx + (C2 — 4cot>x) sinx

21. y = (Gx? -4x + Per + fe 23. y =(1 + m — 2x)sinx — (3 + 2log|sinx|)cosx

25. y=C\x + Coe*7 + G —x)e” 27. y! =10x? + Ce. +e = oe"

DOME Cyt Calog 1 1

—— — x 37) y= Cie ™ HiGee? + Cet te"
= og x

Section 4.8

Vy = Cx + Cox" 3. y =(Ci + Crlogx)x

Sey = Cy + Cx? 7. y =x[C,cos(4logx) + C,sin(4logx)]
9. y =(C; + Crlogx)x? + x*(logx)’ disy = Cine Ox 42x =)
days 'Cyx Cx 413% oes 15. y =Cix + Cox? + x°(logx — 3)
328 ANSWERS TO ODD-NUMBERED EXERCISES

7s Yar ice Xs 19. y = 4x?[9cos(3logx) — 7sin(3logx)] + 7x°

ai. y= Cx + Cox 23 y SC tO
25. y =C, + (C2 + Cslogx)x? 27. y = Cx + Crcos(logx) + C3sin (log x)
29. ya (C, a0 Cr log x)x at Gaxn ot ix? 31. Va Gia 5m a (ABS sue 5)

CHAPTER 5

Section 5.2

5 27 - 7 1 ‘
paar —ed
—Hz, 5) OUOM a
(a) —=s OE
(b eee
= —cos(8V3t 7. 29 ticks

9. 8 lb 11. Be 13. 2.845 15. 0.364s, -—3.20 rad/s

17. (c) 6=27, max att =0,27,47,... 19. vo + cyo/2m #0

min att = 7,37,57,... 21 (ay 13 (b)o oy = "se"? (c) = s

23. 1.069; no

Section 5.3

‘ 2s cosi2 ta OL tae
eae ee ae pene 4) 2003 | ae oO
i aa) + — = L y — a —

P
7. 0.0056 Hz, 2m 13. 5.56s, 8.85 15. e ‘sint (transient), 2sint(steady-state) 17. ’

Section 5.4

Se (a) 0. (b) ip(t) = cos2t + Hsin 2t 5. (a) i(t) =e *(3sin4t — 4cos4t) + 4cosSt
(b) i(t) = Se = sint
7. q(t) =3—4e ' (cos
10¢ + sin102), 3 coulombs 9. R = 9.41 X 10°-*ohms

Section 5.5

1. y =sint 3. y = 2e*
— 3e' 5. y=3-e"
1
7. y = (2 — 3logt)t? 9. g(t, T)=t—T 11. g,(t, rT) = —=sinh
V5(t — 7)
F : V5
t
13. gi(t, 7) = —e 7? — e 2 15. g(t, T) = =|() - (:)
3 8L\r t
1 (ett) Fa) 1
17. gilt,t, T)
T) =
= 5
-(1 — eet
7?) log Ki en 19. as
SS 211. ye Cs
=e al
3) Wiis 00 ental = ote mat (Se tiem en! 29. y=aro—gr' +t
31. y=pr?+ eet %— Zlogt 37. gi(t, rT) = He" — e” ) — (t — 7)
(a + 1)e""? — (b + Ie" + (b — ade —3 — V33 —3 + V33
3900210.) ee
—— where a = ————_,,_ 6 = —————
(GE 70) (Gaeta) (Dasa) 4 4
41. git, 7) = pe 7 — ce 9 + Fe 45. y =jcos2t + ¢sin2t — 4
 ANSWERS TO ODD-NUMBERED EXERCISES 329

Section 5.6

@) vc Saat Ve eee cc atam ee), O—S—err.

Ae e ‘(cost + 2sint)-—e “™sint, t >a A PAU} y

: (cost — coswt)—sint, t > 7
on— 1 .

2sint, O<t ee
5. y(t) = a Son
2sint — A costae tf ae

V3
(t + 3)e'— 2eN cos——t, Oma

V3
7% yM=4(4+3)e'- 2e" cost +e!

= | Leh — 1) + — ayesin—(t
cos——(f Ney, — | : fal
2 se

CHAPTER 6

Section 6.2

1 a 3 n! 5 Ss i= @ §

pe hs? rad "52 + k? (s t ays + b) 5? — k?

if es : 5 13 s? — 2k? ne s-a - 1 D 2s(s? — 3k?)
hee eae ” s(s? — 4k?) " (s-aP-kh " (s —ay * (52 +
s? + 2k? > DG = Q-) (Ul ae eu) (s —2)
21. : b) ——— ———— + ——e
s(s? + 4k?) sea te, eocet! 0 ca eo

Section 6.3

3 2k(Bs? — k? s? 3(s + 4 +4
12(s
1. 3. ae ft rarer @ oT
(s +4?+16 x [(s + 4)? + 16]
(s — 2) (s? +k’? s* + 4k
s+1 (s +17? +2 fl F 1 i
9. | 144.. ——_—_—_—_—. 13. (a) -+—-(l-e” b) -(l-e?
ae 1 (s + I[(s + 1)? + 4] (a) 5 sa ) ( A )

ise
3 caret
17 . —arctan— 25. : (b) T er(== )
s? + 16 sae 2Vs

Section 6.4

CYR sO —4
Whe 1
\ —t
4,..=3
g 5. e'sint Te Wr
1. aie (3) 2 e

9. Whossinst
e-2'( 7cosSt ~ = | —21/3
41) aS : 2
Scos=V5t ae
= 4V5sin5V51 iksk il>@ a

t berae 35 - =.) ad 19. as

-(e ta e') + 5
—(e — E ‘)

12
«(3
3
ay ee ser

a ie 4s
2

OP
—2t a

Sa Lie has 2)
330 ANSWERS TO ODD-NUMBERED EXERCISES

1 2 8) 1 lee ; =iy2 1/2

D4 Se3 hateraan GgHOE 23. —e* ——e-
+ —e '(9sin2t — cos 21) (mt) '* cos 2(at
10 Eoteaais 50. * 25 50 G9

Section 6.5

y(t) = 3(e7 — 3) 3. y(t) = sinht 5. y(t) =sint 7. y(t) =e' — (cost + sint)
1.
ti 1 sae
y(t) =(e ‘— 1)sint 11. pipet cet +e (eos Vir_ V21)
sin
3V2
1
3) y(t) =¢(1
8 + 20) + fe7"(1 — 20 15. y(t) = 8e '— 9e-* + 3e°*
17. yQ=Fe'—Se WV +he *+ he" 19. y@= 4(sinht Se sinit) at

Section 6.6
\ x Site POD)

Ol eens 11.
se s+

es SYO@) = 3) == ale == 1h)

49) en — 3)
21. cos2i[1 + h(t — 7)
Ws 26] Qo = 2) Se
23. y#=e'+[1l—e “ne - 1) 95) y(t) = cos2t +t —3sint — [ft — 1 — 5sin2(t — IJa(t — 1)
7) y(t) = (cos2t — cos 4t)[1 + A(t — )] + 3sin2r 29. y =e" = e% + 2[1 + 20 —3e™ he
31) y@)i— ex (Cost + sind) —e) “aa sinth( = am)
33. y(t) =2te'+[1—e *? —(¢ —2m)e h(t — 277)
Ip 1 1 k
5) y(t) = — (“sin Wot — —sin ar)- sin wo(t — 7), where wo =—
m(@~ — wo) \@Wo @ MW m

Section 6.7

2 S = ies P
a.) ——— Si So 5 Cede ie ee — tea Sig
s(s* + 4) (GS se AIG? ae 1}

9 a.75| 2 + St)e"!
(4 ak 50
Ree ad cos Di eosin
air sin f) 15 A ee at
pain
oe 17 * y(t)
) = =t + aeaL
sindt

2 P

19. y(#)=1+ x 23. (a) y(t)= a — COS Wot)

(db) yO) = 55,

—— (C08 WL = Cosiangt) (c) y(t)= tsin Wot
m(wo — w°) 2MWo

CHAPTER 7

Section 7.2

ln x OiSiCle IGsee Ey she, ene ya

y(t) = Cye’ + 3C2e ' V(t) = $C +t 2636
5. x(t) = e(C,cos2t + C2sin 2?) a x(t) = e*(C, cost + Crsint)
y(t) = $e(C, sin2t — Cycos 21) y(t) = e*[(2C, — Cr) cost + (2C2 + C;)sint]
 ANSWERS TO ODD-NUMBERED EXERCISES 331

9. x(t) = Cie’ + Cre-* + Cre” 11. x(t) = Crea ak Gre cate

y(t) = Crewe)! ah. Gre cae
y(t) = —4Cy\e' — Cre?! + 2C3e*"
z(t) = —Cy\e' — Coe** + C3e*
13. x(t) =Cie'+ Coe‘ -1 15. BY =n? as eng
yQ@) = Cie’ 3C.e* +t — 2 y@)= —3Cje7* — $C2e*"
17. x(t) =2C, + 6Cre* 195 x(t) = Gren!

y(t) = C, + Cre + Ce” y(t) = C.e-™"

21. x(@) = Cye'+ Cre" —3t -—& 23. x(t) = Cye' — sint
y(t) = —3Cye' + 3Qne *+5 t+H y(t) = -3Cie' + $sint
25. x(t) = Cie! + Cre" + 3C3e"% +3 27. x(t) = —6C\e' — 3C.e7 + 2C3e*
y(t) = 3C\e* — Cse" + 5 y()) =e) 4+ Coetn ti Cae,
z(t) = 5C\e' + Coe* + Cse™
29. no solutions 31. infinitely many solutions

Section 7.3

1. x() =3(e'—e*) 3. x(t) = —cos3t — $sin3t

y@ =s@e"+ e7*) y(t) = 2cos3t — tsin3t
5. x(t)= —e'— pe ' + pe (2lcost — 13sin) 7 x(t) =S5e'-4+5¢-2° +348
y(t) = e' —ge ' + $e7(—4cost + 7sinn) y(t) = —5e '+5 — St + 22?
9 x(t) = —-3r- 3V2sinV2t Ws. xQH=tr+2 +e + sint
y() = —$¢ +3V2sinV2t yi) = 1 —1 — 3e * = cost
3 2 2 V6 13 145. 25a
13. yi) = 5 sin! + 500s! — 5008 V6 - ss sin Vr 15.9.0) =—sint #7 sin 27 S sind
42 42 14
6 4 1 V6 3)
y2(t) = poe
145 | LOR
= Pinta oe Zee
y(t) = =sint + —cost + -cos V6t — —sin V6t
5 5 5 30
1
19907, — 1 es" (S sin 50V3t — cos sov31) 23. (a) 0.69,1.66 (b) 70:72, 1:38
(ce) 0.67, 2.100) 0.56, 1.46
1
DG) cn (cos50V3t + —=sin sov31) (e) 0.56, 1.46
V3

Section 7.4

Kee =%
3. x1 5. u’=w
x3 = —kx, + Psinwt a Xa Demat aa 10
Xj =Xq nae AM Gn:
xq = k*x, + fd)
(7) ees 9. x(t) = Cie’ + 3C2e' Tioex) = Cres ot ere
1 ka y(t) = Cie’ ae SGoeme y(t) = —3Cie "+ Cre™
x3 = =——(ki AP kz) x, 3
m m

X3 = X4
ky kp
xj =—xX1 - —%3

13. x(t) = e*(C,cos2t + Czsin2t) 15. x(t) = 2e7"(C, cos3t + C2sin3t)

y(t) = e*(C,sin2t — C2,cos 2t) y(t) = e2[(3C2 — Ci) cos 3t — (BC, + C2) sin3¢]
332 ANSWERS TO ODD-NUMBERED EXERCISES

17. x(t) = Cie’ + Cote’ 19. x@ = —3C\e + C3e”

y(t) = —Cye' — Co(t + 4)e y(t) = 4C\e™' + Cre + Cote”
21. independent 30) = 20.2) = Cre + Cad = es
23. independent 27. x(t) = Cicost + Csint + $(cos2t + sin2t) — St
y(t) = 4(2C, — C2)cost + $(2C, + Cy)sint + 4sin2t — 2¢ + 1
29. x(t) = Cie~* + Coe! — ge’ 31. x(t) = e'[Csint — 2C,cost — 2costlog|sect + tant|]
y(t) = —4Cie~*" + Core™ — ge" y(t) = e'[$C, cost + Cysint — 1 + sintlog|sect + tan t|]

Section 7.5

1 voa(Jera(er @ vu=cl(}) +a(})e"

3
1

ue 2cos 3t , 2 sin 3t 24
SE EA Bae, = ee sf“Aer ie sna)

(1 1
1h Lew
7 Y@=Gl + Cl 9! |e*
6: |)46 —2: |e" 9. veo =35 (flew +(
=13 i =|
ioe 1.543 1.175 cosht sinht
ee) i ‘| (b) tee a | (c) ea on
13. Y= (2)cost + (?)sint|ar cl
c;| ()cost + (3)sint|+ ie, aa + (axa )

15. Y(1) = Ae 7 colte = (te

aia Oe ee dee eae

1 wo- Ce (ere
17. Y@=

Section 7.6

S700) 7. (0, 0) 9. (0,0), (-3, 1)

W. y-@+1P=C 13. ( —y’ =C@+y) 15. x°-3xy +y°=C
17. 3xy —x’y?-—2y?=C 19. node, asymptotically stable 21. saddle point, unstable
23. foci, unstable 25. unstable 27. unstable saddle point
29. no conclusion 31. asymptotically stable
33. (0,0) unstable 35. (1,1) unstable
(2, 4) unstable (—1, -1) asymptotically stable

CHAPTER 8

Section 8.2

1. 2.9278 (Euler), 3. 3.2261 (Euler) 5. 1.8370 (Euler)

3.4509 (Improved Euler) 3.8254 (Improved Euler) 2.0486 (Improved Euler)
 ANSWERS TO ODD-NUMBERED EXERCISES 333

7. 1.2194 (Euler) 9. 0.4198 (Euler) 11. answers given above in 1-9

1.3260 (Improved Euler) 0.4053 (Improved Euler)
13. y(1) = 2.4883, 15. y(1) = 2.7182, 17. y(1) = 0.7839, aw = 3.1355
e = 2.71828 . . . (actual value) e = 2.71828. . . (actual value) 23. y(0.5) = 0.7971

Section 8.3

1. x(0.1)=1.1, y(O.1) = —-0.1 <j cA) S11, psi) Tal

x(0.2) = 1.25, y(0.2) = —0.22 x(0.2) = 1.451, y(0.2) = 1.32
5. x(0.1) =0.8, y(0.1) = 1.1 7. x(0.1) = 1.3110, y(0.1) = 1.9197
a(O-2 = 0.5825>¢ (0:2) = 1.18 9. x(0.1) = 1.1305, y(0.1) = 0.3851
41. x01) = 1.2, x(0.2) = 1.39 13. x(0.1) = 1.1947

CHAPTER 9

Section 9.2
Suh ah
ra Q > — exe
1 ne ce 2)(n +k >)
Se a us
. ee
= aera ae (oi \)))!

9. y-= co > C1yx" 11. y = col — $x? + ex — >> -)

@) Se Ue 3 n=0
~ (—1)"(2x)" e(—1)x) |

13, y= to, S|x| <

: 2s (2n)! 1d (2n + D!
+ (x + 9x9 — Bx
15. y =coll + 4x2) +--+), [x] <3
Wf Y= coll — 6x? + 3x*—---) + e1(x — 3x3), |x| <a

+ gx5----), [x] <e .

19.. y = col or? $+ 9x — +) FG —4y3
+ 5x +---), \<1 23) y= ley ae tee te
2. y=cot ee + 5x?
950 = 4 + Ox xt t+ ox° to -
27. y =coll —8@ — I tee — DS-- tele -)-a@- Di +e —D'-- +] 4
1) at elie — 4) = i(x — 1)°] y,
209 y= cll — 3.
31 =e (l tae tek tT ae ye eT EN ea ax tape tt)
S350) ty = col — x7 ee) — wx) ex
(b) y = co(l — 4x? + $x*) tar — x? + wx - + °°)

Section 9.3

0 aR s. Peat | Byers Prats =!) iets Prax = 0

east
ater Se 4 Te Rest aater— 083
BR, ISSIR, aise = ©, LSP.
9. y =A(1 + 2x + 4x2) + Bx'(1 + 3x + ax? + --)
—- +) + Bx O71 — 10x) 13m SeAce = Bx 1/2

11. y =A(U —x + x?
15, y = Axl —x + bx? —- ++) + Bx(l — 3x 4 ox’ — °°")
17 ays AG (let 3h ak oe ) + Bx2(1 — yx? + oex*— ++ *)
19. + 4x + 4x7 +--—
y =(A + Blogx)(1 BBx + 12x? + ex? +--+ -)
+)
334 ANSWERS TO ODD-NUMBERED EXERCISES

21. y =(A + Blogx)xe* — B(x? + 3x? + 3x4 ee) 23. y =(A + Blogx)x”
25. y =(A + Blogx)x — B(x? — 4x° + 7gx* — + -*)
27. yy = colx — 2x? + 2x?) + Cn(x* — Fx? + 4x°— >>")
29. y =co(l + 3x + x2) + ca(x* + 3x5 + oxo t+: °°) 31. y = cox”! + c3(x? + 3x? + xt +: °
1 4
33. y=(A + Blogry ~ 12+ 32x ~ 40s? +) + B(5 42-4 16x ++)
go
5 1
35. y =(A + Blogs)(~3e ~ 4x 3? = +++) + a(S -2-x 3+)
x
ABEL, NEILS H., 89 Curves of linear pursuit, 63-65
Abel’s formula, 89, 103 Cycloid, 69
system of DEs, 248
Amplitude, 131 Damped motion, 130, 198
maximum, 142 forced, 141-144
Asymptotic stability, 262 free, 134-137
Autonomous system, 261 Decay, radioactive, 70
Auxiliary conditions, 14, 127 Delta sequence, 164
Auxiliary equation, 94, 104 Differential equation, 2
system of DEs, 239 classification of, 2
exact, 25—28
Beats, 140 first-order, 19-54
BERNOULLI, JAKOB, 52 general solution of, 9-12
Bernoulli’s equation, 52-54 higher-order, 82-125
Bessel function, 187, 201, 214 homogeneous, 31-33, 36-39, 83, 101
first kind, 314 linear, 3
second kind, 314 nonhomogeneous, 36, 41-50, 83, 101, 108-114
zeros of, 168 nonlinear, 3
Bessel’s equation, 83, 201 order of, 3
standard form, 166 ordinary, 3
Boundary value problem, 14-16 partial, 3
Brachistochrone, 69 solution of, 6-12
systems of, 220-273
Catenary, 69 Differential operator, 36, 98
Cauchy-Euler equation, 121-123 normal, 36, 101, 127
Center, 264 Dirac, PAut A. M., 160
Characteristic equation. See Auxiliary equation Dirac delta function. See Impulse function
CHEBYSHEV, PAFNUTI LvovicH, 305 Discontinuous function, 49, 201-208
Chebyshev’s equation, 305
Clairaut’s equation, 35 Eigenvalues, 250
Comparison theorem, Sturm, 167 Eigenvectors, 250
Complementary solution, 108 Electrical networks, 230-232
Confluent hypergeometric equation, 321 Electric circuits, 75, 146-148
Confluent hypergeometric function, 321 Electromotive force, 147
Constant-coefficient equations, 94-99, 104-106
Equidimensional equation, 121-123
linear system of DEs, 239-247 Error truncation, 276
Convolution theorem, 210-212 Error function, 47, 186
Critical damping, 135, 198 complementary, 186
Critical point, 261 Escape velocity, 62

335
336 INDEX

Euler constant, 314 Impulse function, 159-160, 207-208

Euler equation, 121-123 Indicial equation, 308
Euler formulas, 96, 242 Influence function, 161
Euler method, 276-279 Initial conditions, 36, 104
improved, 279-280 Initial value problems, 14-16
Exact equation, 25-28 first-order DE, 36
Existence-uniqueness theorem: higher-order DE, 127
first-order DEs, 19 system of DEs, 236
linear DEs, 48, 127 Input function, 129
system of DEs, 236 Integral equation, Volterra, 216
Exponential order, 174 Integrating factor:
first-order DEs, 28-30
Focus, 264 linear DE, 30
Forced motion, 130, 139-144 Inverse Laplace transform, 187-191
Forcing function, 41, 129 Inverse matrix, 253
Free-falling body, 59-62 Irregular singular point, 306
Free motion, 130-137 Isogonal trajectory, 66
Frequency:
angular, 131 Jump discontinuity, 162, 201
natural, 131
FROBENIUS, GEORG F., 307 KIRCHHOFF, GUSTAV R., 75
Frobenius, method of, 307-320 Kirchhoff’s laws, 75, 146
Fundamental matrix, 252 Kutta, WILHELM, 281
Fundamental set, 102
Laguerre polynomials, 186
Gamma function, 177, 215 LAPLACE, PIERRE SIMON, 172
Generalized function, 160 Laplace transform, 172
General solution, 9-12 convolution theorem, 210-212
first-order linear, 41 existence of, 175
higher-order linear, 84 initial value problems, 194—200
system of DEs, 238, 250, 254 inverse, 187-191
Gravitational constant, 59 linearity property, 178
GREEN, GEORGE, 150 operational properties, 178-184
Green’s function, 161-162 periodic functions, 185
one-sided, 150-155, 213-214 table of, 217-218
table of one-sided, 155 Legendre’s equation, 83, 296
LEIBNIZ, GOTTFRIED WILHELM VON, 2
Half-life, 71 Leibniz formula, 152
Harmonic motion, simple, 131 Lerche’s theorem, 188
Heaviside unit function, 163, 201 Limiting velocity, 60
HERMITE, CHARLES, 305 Linear DE, 3
Hermite’s equation, 305 auxiliary equation, 94, 104
Homogeneous equation: Cauchy-Euler, 121-123
first-order DE, 31-33 constant-coefficients, 94-99, 104-106
linear DE, 36-39, 83, 101 existence-uniqueness theorem, 48, 127
system of DEs, 235, 249 first-order, 30-31, 36-50
Homogeneous function, 31 general solution of, 41, 84
HOOKE, ROBERT, 128 higher-order, 82-125
Hooke’s law, 128 homogeneous, 36-39, 83, 101
Hypergeometric equation, 321 nonhomogeneous, 36, 41-50, 83, 101, 108-114
Hypergeometric function, 321 normal form, 36, 84, 101, 127
particular solution, 41, 108
Identity matrix, 250 undetermined coefficients, 109-114
Implicit solution, 8 variation of parameters, 42, 115-119
Improved Euler method, 279-280 Linear dependence, 84, 102, 237
INDEX 337

Linear independence, 84, 102, 237, 249 Operator, differential, 36, 98

Linearity property (see also Superposition Orbit, 258
principle): Order, 3
of Laplace transforms, 178 Ordinary point, 294
Linear operator, 36, 98, 101, 127 Orthogonal trajectory, 56-59
Linear system of DEs, 221 Oscillation theory, 164-169
auxiliary equation, 239 Overdamping, 135
constant-coefficients, 239-247
existence-uniqueness theorem, 236 Particular solution, 41, 108, 245
general solution of, 238 Pendulum, 5, 133-134
homogeneous, 235, 249 double, 234
matrix solution of, 248-256 stability of, 268-270
nonhomogeneous, 235, 254 Period, 131
stability of, 263-267 quasi, 136
variation of parameters, 245, 254 Phase angle, 131
Logarithmic decrement, 138 Phase plane, 258
Logistic curves, 72 Piecewise continuous function, 174
Logistic equation, 72 Population growth, 71
Lotka-Volterra equations, 222, 272 Power series, 291
method of, 292-303
Matrix: Predator-prey, 221, 272
column, 249 Predictor-corrector method, 279, 283
fundamental, 252
identity, 250 Quasiperiod, 136
inverse, 253
methods, 248-256 Radioactive decay, 70
Mixing problems, 76-78 Radius of convergence, 291
Model, mathematical, 56 Ratio test, 291
Recurrence formula, 293
NEWTON, SIR ISAAC, 2 Regular singular point, 306
Newton’s law of cooling, 76 Resonance:
Newton’s law of gravitation, 60 electrical, 147, 150
Newton’s second law of motion, 59, 73, 129 mechanical, 141, 200
Node, 264 Riccati equation, 52
Nonhomogeneous equation: RUNGE, CARL, 281
first-order DE, 36, 41-50 Runge-Kutta method, 281-283
higher-order, 83, 101, 108-114
linear system of, 235 Saddle point, 264
Nonlinear DE, 4 Separation of variables, 20-23
exact equation, 25-28 Separation theorem, 164
existence-uniqueness theorem, 19 Series solution:
homogeneous equation, 31-33 existence theorem, 296, 307
integrating factor method, 28-30 indicial equation, 308
Nonlinear system of DEs, 258-270 near ordinary point, 296-303
autonomous, 261 near regular singlular point, 307-320
phase plane, 258 recurrence formula, 293
stability of, 267-270 Shifting theorem, 179
Simple harmonic motion, 131
trajectory, 258
Normal form, 84, 101, 127 SIMPSON, THOMAS, 281
Simpson’s rule, 281
Normal linear operator, 36, 101, 127
Null function, 188 Sine integral, 185
Numerical methods, 274-288 Singular point, 294
Euler method, 276-279 irregular, 306
regular, 306
Euler method, improved, 279-280
Singular solution, 10, 20, 23, 54
Runge-Kutta, 281-283
INDEX

Solution, 6-12, 236 Terminal velocity, 60

complementary, 108 Three-term Taylor series method, 285
fundamental set, 102 TORRICELLI, EVANGELISTA, 62
general, 9-12, 41, 84, 238 Torricelli’s law, 62
homogeneous, 41, 108, 150, 245 Tractrix, 68
implicit, 8 Trajectory, 258
particular, 41, 108, 150, 245 isogonal, 66
singular, 10, 20, 23, 54 orthogonal, 56-59
steady-state, 51, 142 Transient solution, 142
transient, 142 Trivial solution, 7, 37
trivial, 7, 37 Truncation error, 276
Special function, 47 Two-point boundary value problem, 16
Spring constant, 128
Spring-mass system, 128-144, 197-200 Undamped motion, 130-134, 139-141
coupled, 228-230 Underdamping, 135, 198
Stability, 258 Undetermined coefficients, 109-114
linear system, 263-267 Unit step function, 163, 201
nonlinear system, 267-270 Unstable critical point, 262
Stable critical point, 261
Standard form, 166
Variables, separable, 20—23
Bessel’s equation, 166
Variation of parameters, 42, 115-119, 245, 254
Steady-state solution, 51, 142
Step function, 163, 201
Velocity:
escape, 62
Sturm comparison theorem, 167
terminal or limiting, 60
Sturm separation theorem, 164
Superposition principle, 11, 83, 236 Volterra integral equation, 216
System of DEs, 220-270 (see also Linear systems
of DEs): WRONSKI, JOZEF M. H., 87
autonomous, 261 Wronskian, 86—90, 103
existence-uniqueness theorem, 236 system of DEs, 238, 249
linear, 220-256
nonlinear, 258-270
numerical methods, 274-288
—
 4 ~

i
= a

XN

i,
“ as 2: 4 .
: “Oe
ns ee re
a a
<
; pete a na
= ost ’

x
: x
;
; i.
> \ Ss
1S,
3
‘-
f ‘
%

2cat 4 <a

ee
~
ie
¥. i
=) * +

ce x ite ra a
pe

ss med : 3

+ oe : if y /
f os =

i a

+ \ ‘
rhe :
oe
oo
f

ia
: ; ©
| 7
: i. ‘
el Aa te
- : Rohs Nz

TT, FORESMAN AND COMPANY

5] Fal CRE >| a a ay , t ae -\- P77 S 5 ion ai z .

ii
a‘ i
;
eae
1OOU
:

@)
yA aoe Ene : . ; a i . ; :

pote oa & fas ‘ es at 4 hues ieee (ie ‘