William A. Adkins, Mark G.

Davidson

ORDINARY DIFFERENTIAL
EQUATIONS
August 20, 2007
Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Basic Concepts and Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Examples of Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3 Direction Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2 First Order Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.1 Separable Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.2 Linear First Order Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.3 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.4 Miscellaneous Nonlinear First Order Equations . . . . . . . . . . . . . . 70

3 The Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

3.1 Definition and Basic Formulas for the Laplace Transform . . . . . 80
3.2 Partial Fractions: A Recursive Method for Linear Terms . . . . . . 95
3.3 Partial Fractions: A Recursive Method for Irreducible
Quadratics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.4 Laplace Inversion and Exponential Polynomials . . . . . . . . . . . . . 115
3.5 Laplace Inversion involving Irreducible Quadratics . . . . . . . . . . . 120
3.6 The Laplace Transform Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
3.7 Laplace Transform Correspondences . . . . . . . . . . . . . . . . . . . . . . . 138
3.8 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
3.9 Table of Laplace Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
3.10 Table of Convolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

4 Linear Constant Coefficient Differential Equations . . . . . . . . . 163

4.1 Notation and Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
4.2 The Existence and Uniqueness Theorem . . . . . . . . . . . . . . . . . . . . 171
4.3 Linear Homogeneous Differential Equations . . . . . . . . . . . . . . . . . 179
4.4 The Method of Undetermined Coefficients . . . . . . . . . . . . . . . . . . 184
4.5 The Incomplete Partial Fraction Method . . . . . . . . . . . . . . . . . . . 192
VIII Contents

5 System Modeling and Applications . . . . . . . . . . . . . . . . . . . . . . . 197

5.1 System Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
5.2 Spring Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.3 Electrical Circuit Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

6 Second Order Linear Differential Equations . . . . . . . . . . . . . . . . 227

6.1 The Existence and Uniqueness Theorem . . . . . . . . . . . . . . . . . . . . 228
6.2 The Homogeneous Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
6.3 The Cauchy-Euler Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
6.4 Laplace Transform Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
6.5 Reduction of Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
6.6 Variation of Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254

7 Power Series Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261

7.1 A Review of Power Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
7.2 Power Series Solutions about an Ordinary Point . . . . . . . . . . . . . 274
7.3 Orthogonal Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
7.4 Regular Singular Points and the Frobenius Method . . . . . . . . . . 290
7.5 Laplace Inversion involving Irreducible Quadratics . . . . . . . . . . . 319

8 Laplace Transform II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331

8.1 Calculus of Discontinuous Functions . . . . . . . . . . . . . . . . . . . . . . . 332
8.2 The Heaviside class H . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
8.3 The Inversion of the Laplace Transform . . . . . . . . . . . . . . . . . . . . 354
8.4 Properties of the Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . 358
8.5 The Dirac Delta Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
8.6 Impulse Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
8.7 Periodic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
8.8 Undamped Motion with Periodic Input . . . . . . . . . . . . . . . . . . . . . 384
8.9 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390

9 Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
9.1 Matrix Operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
9.2 Systems of Linear Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
9.3 Invertible Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
9.4 Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430

10 Systems of Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 441

10.1 Systems of Differential Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 441
10.1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
10.1.2 Examples of Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . 446
10.2 Linear Systems of Differential Equations . . . . . . . . . . . . . . . . . . . . 452
10.3 Linear Homogeneous Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 466
10.4 Constant Coefficient Homogeneous Systems . . . . . . . . . . . . . . . . 478
10.5 Computing eAt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 488
10.6 Nonhomogeneous Linear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 497
 Contents IX

A Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505

A.1 Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505

B Selected Answers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513

C Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 589
List of Tables

3.1 Basic Laplace Transform Formulas . . . . . . . . . . . . . . . . . . . . 94

3.9 Table of Laplace Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
3.10 Table of Convolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

C.1 Laplace Transform Rules . . . . . . . . . . . . . . . . . . . . . . . . . . 581

C.2 Table of Laplace Transforms . . . . . . . . . . . . . . . . . . . . . . 582
C.2 Table of Laplace Transforms . . . . . . . . . . . . . . . . . . . . . . 583
C.2 Table of Laplace Transforms . . . . . . . . . . . . . . . . . . . . . . 584
C.3 Table of Convolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 584
C.3 Table of Convolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 585
1
Introduction

1.1 Basic Concepts and Terminology

What is a differential equation?

Problems of science and engineering frequently require the description of some
property of interest (position, temperature, population, concentration, cur-
rent, etc.) as a function of time. However, it turns out that the scientific laws
governing such properties are very frequently expressed as equations relating
how the various rates of change of the property are related to the quantity
at a particular time. For example, Newton’s second law of motion states (in
words):
Force = Mass × Acceleration.
For the simple case of a particle of mass m moving along a straight line, if we
let y(t) denote the distance of the particle from the origin, then acceleration
is the second derivative of position, and hence Newton’s law becomes the
mathematical equation

F (t, y(t), y 0 (t)) = my 00 (t), (1)

where the function F (t, y, y 0 ) gives the force on the particle at time t, a dis-
tance y(t) from the origin, and velocity y 0 (t). Equation (1) is not an equation
for y(t) itself, but rather an equation relating the second derivative y 00 (t), the
first derivative y 0 (t), the position y(t), and t. Since the quantity of interest is
y(t), it is necessary to be able to solve this equation for y(t). Our goal in this
text is to learn techniques for the solution of equations such as (1).
In the language of mathematics, laws of nature such as Newton’s second
law of motion are expressed by means of differential equations. An ordinary
differential equation is an equation relating an unknown function y(t), some
of the derivatives of y(t), and the variable t, which in many applied problems
will represent time. Like (1), a typical ordinary differential equation involving
t, y(t), y 0 (t), and y 00 (t) would be an equation
2 1 Introduction

Φ(t, y(t), y 0 (t), y 00 (t)) = 0. (2)

In such an equation, the variable t is frequently referred to as the indepen-

dent variable, while y is referred to as the dependent variable, indicating
that y has a functional dependence on t. In writing ordinary differential equa-
tions, it is conventional to write (2) in the form

Φ(t, y, y 0 , y 00 ) = 0. (3)

That is, suppress the implicit functional evaluations y(t), y 0 (t), etc. A par-
tial differential equation is an equation relating an unknown function
u(t1 , . . . , tn ), some of the partial derivatives of u with respect to the vari-
ables t1 , . . ., tn , and possibly the variables themselves.
In contrast to algebraic equations, where the given and unknown objects
are numbers, differential equations belong to the much wider class of func-
tional equations in which the given and unknown objects are functions
(scalar functions or vector functions).
Example 1. Each of the following are differential equations:
1. y 0 = 2y 2. y 0 = ty
3. y0 = y − t 4. y0 = y − y2
5. y 0 = f (t) 6. y 00 + y = 0
7. y 00 + sin y = 0 8. ay 00 + by 0 + cy = A cos ωt
9. y 00 + ty = 0 10. y (4) = y
∂2u ∂2u ∂u ∂2u
11. + 2 =0 12. =4 2
∂t21 ∂t2 ∂t ∂x
Each of the first ten equations involves an unknown function y (or dependent
variable), the independent variable t and some of the derivatives y 0 , y 00 , y 000 ,
and y (4) . The last two equations are partial differential equations, specifically
Laplace’s equation and the heat equation, which typically occur in scientific
and engineering problems.
In this text we will generally use the prime notation, that is, y 0 , y 00 , y 000
(and y (n) for derivatives of order greater than 3) to denote derivatives, but
dy d2 y
the Leibnitz notation , , etc. will also be used when convenient. The
dt dt2
objects of study in this text are ordinary differential equations, rather than
partial differential equations. Thus, when we use the term differential equation
without a qualifying adjective, you should assume that we mean ordinary
differential equation.
The order of a differential equation is the highest order derivative which
appears in the equation. Thus, an ordinary differential equation of order n
would be an equation of the form

Φ(t, y, y 0 , y 00 , . . . , y (n) ) = 0. (4)

 1.1 Basic Concepts and Terminology 3

The first five equations above have order 1, while the others, except for the
tenth equation, have order 2, and the tenth equation has order 4. We shall
be primarily concerned with ordinary differential equations (and systems of
ordinary differential equations) of order 1 and 2, except that some topics are
no more difficult, and in some cases even simplified by considering higher order
equations. In such cases we will take up higher order equations.
The standard form for an ordinary differential equation is obtained by
solving (4) for the highest order derivative as a function of the unknown
function y, its lower order derivatives, and the independent variable t. Thus,
a first order ordinary differential equation is expressed in standard form as

y 0 = F (t, y) (5)

while a second order ordinary differential equation in standard form is written

y 00 = F (t, y, y 0 ). (6)

In Example 1, equations 1 – 5 and 10 are already in standard form, while

equations 6 – 9 can be put in standard form by solving for y 00 :
6. y 00 = −y 7. y 00 = − sin y
8. y 00 = −(b/a)y 0 − (c/a)y + (A/a) cos ωt 9. y 00 = −ty
In applications, differential equations will arise in many forms. The stan-
dard form is simply a convenient way to be able to talk about various hypothe-
ses to put on an equation to insure a particular conclusion, such as existence
and uniqueness of solutions (see Section 2.3), and to classify various types of
equations (as we do in Chapter 2, for example) so that you will know which
algorithm to apply to arrive at a solution.

What is a solution?
For an algebraic equation, such as 2x2 + 5x − 3 = 0, a solution is a particular
number which, when substituted into both the left and right hand sides of the
equation, gives the same value. Thus, x = 1/2 is a solution to this equation
since
2
2 · (1/2) + 5 · (1/2) − 3 = 0
while x = −1 is not a solution since

2 · (−1)2 + 5 · (−1) − 3 = −6 6= 0.

A solution of the ordinary differential equation (4) is a function y(t) defined

on some specific interval I ⊆ R such that substituting y(t) for y and substitut-
ing y 0 (t) for y 0 , y 00 (t) for y 00 , etc. in the equation gives a functional identity.
That is, an identity which is satisfied for all t ∈ I:

Φ(t, y(t), y 0 (t), y 00 (t), . . . , y (n) (t)) = 0 for all t ∈ I. (7)

4 1 Introduction

For low order equations expressed in standard form, this entails the following
meaning for a solution. A function y(t) defined on an interval I is a solution
of a first order equation given in standard form as y 0 = F (t, y) if
y 0 (t) = F (t, y(t)) for all t ∈ I,
while y(t) is a solution of a second order equation y 00 = F (t, y, y 0 ) on the
interval I if
y 00 (t) = F (t, y(t), y 0 (t)) for all t ∈ I.
Before presenting a number of examples of solutions of ordinary differential
equations, we remind you of what is meant by an interval in the real line R,
and the (standard) notation that will be used to denote intervals. Recall that
intervals are the primary subsets of R employed in the study of calculus. If
a < b are two real numbers, then each of the following subsets of R will be
referred to as an interval.

Name Description
(−∞, a) {x ∈ R : x < a}
(−∞, a] {x ∈ R : x ≤ a}
[a, b) {x ∈ R : a ≤ x < b}
(a, b) {x ∈ R : a < x < b}
(a, b] {x ∈ R : a < x ≤ b}
[a, b] {x ∈ R : a ≤ x ≤ b}
[a, ∞) {x ∈ R : x ≥ a}
(a, ∞) {x ∈ R : x > a}
(−∞, ∞) R

Example 2. The function y(t) = 3e−t/2 , defined on (−∞, ∞), is a solution

of the differential equation 2y 0 + y = 0 since
2y 0 (t) + y(t) = 2 · (−1/2) · 3e−t/2 + 3e−t/2 = 0
for all t ∈ (−∞, ∞), while the function z(t) = 2e−3t , also defined on
(−∞, ∞), is not a solution since
2z 0 (t) + z(t) = 2 · (−3) · 2e−3t + 2e−3t = −10e−3t 6= 0.
More generally, if c is any real number, then the function yc (t) = ce−t/2 is a
solution to 2y 0 + y = 0 since
2yc0 (t) + yc (t) = 2 · (−1/2) · ce−t/2 + ce−t/2 = 0
for all t ∈ (−∞, ∞). Thus the differential equation 2y 0 + y = 0 has an
infinite number of different solutions corresponding to the infinitely many
t ∈ (−∞, ∞).
Less obvious is the fact that every solution w(t) of the equation 2y 0 +y = 0
is one of the functions yc (t) for an appropriate choice of the parameter c. This
is easy to see by differentiating the product et/2 w(t) to get
 1.1 Basic Concepts and Terminology 5

(et/2 w(t))0 = (1/2)et/2 w(t) + et/2 w0 = (1/2)et/2 w(t) − (1/2)et/2 w(t) = 0

for all t ∈ (−∞, ∞) since w0 (t) = −(1/2)w(t) because w(t) is assumed to

be a solution of the differential equation 2y 0 + y = 0. Since a function with
derivative equal to 0 on an interval must be a constant, we conclude that
et/2 w(t) = c for some c ∈ R. Hence,

w(t) = ce−t/2 = yc (t).

Moreover, the constant c is easily identified in this case. Namely, c = w(0)

since e0 = 1.
Figure 1.1 illustrates several of the solutions yc (t) of the equation 2y 0 +y =
0.

y
c=2
c=1
c=4

c = −4
c = −1
c = −2
Fig. 1.1. The solutions yc (t) = ce−t/2 of 2y 0 + y = 0 for various c.

One thing to note from Example 2 is that it is not necessary to use y(t)
as the name of every solution of a differential equation, even if the equation
is expressed with the dependent variable y. In Example 2, we used y(t), yc (t),
and w(t) as names of different solutions to the single differential equation
2y 0 + y = 0. This is a feature that you will see frequently, particularly since
we will often be interested in more than one solution of a given equation.

Example 3. The function y(t) = t + 1 is a solution of the differential equa-

tion
y0 = y − t (8)
on the interval I = (−∞, ∞) since

y 0 (t) = 1 = (t + 1) − t = y(t) − t

for all t ∈ (−∞, ∞). The function z(t) = t + 1 − 7et is also a solution on the
same interval since

z 0 (t) = 1 − 7et = t + 1 − 7et − t = z(t) − t

6 1 Introduction

for all t ∈ (−∞.∞). Note that w(t) = y(t) − z(t) = 7et is not a solution of
(8) since
w0 (t) = 7et = w(t) 6= w(t) − t.
There are, in fact, many more solutions to y 0 = y − t. We leave it as an
exercise to check that yc (t) = t + 1 + cet, where c ∈ R is a constant, is in fact a
solution to (8) for all choices of the constant c. Moreover, you will be asked to
verify in the exercises that all of the solutions of this equation are of the form
yc (t) for some choice of the constant c ∈ R. Note that the first solution given,
y(t) = t + 1, is obtained by taking c = 0 and the second, z(t) = t + 1 − 7et , is
obtained by taking c = −7. Figure 1.2 illustrates several of the solutions yc (t)
of the equation y 0 = y − t.

c = .4

c=0

c = −1

Fig. 1.2. The solutions yc (t) = t + 1 + cet of y 0 = y − t for various c.

Example 4. The function y(t) = tan t for t ∈ I = − π2 , π

is a solution of


2
the differential equation y 0 = 1 + y 2 since
d
y 0 (t) = tan t = sec2 t = 1 + tan2 t = 1 + y(t)2
dt
for all t ∈ I. Moreover, the same formula, y(t) = tan t defines a solution in the
interval ( π2 , 3π π 3π
2 ), but y(t) = tan t is not a solution on the interval (− 2 , 2 )
π
since the function tan t is not defined for t = 2 . Note that
in this example,
the interval on which y(t) is defined, namely I = − π2 , π2 , is not apparent
from looking at the equation y 0 = 1 + y 2 . This phenomenon will be explored
further in Section 2.3.
The function z(t) = 2y(t) = 2 tan t is not a solution of the same equation
y 0 = 1 + y 2 since
z 0 (t) = 2 sec2 t = 2(1 + tan2 t) 6= 1 + 4 tan2 t = 1 + z(t)2 .
 1.1 Basic Concepts and Terminology 7
y y

t t
− π2 π
2
π
2
3π
2

Fig. 1.3. The different solutions y(t) = tan t of y 0 = 1 + y 2 on the two intervals
(− π2 , π2 ) and ( π2 , 3π
2
).

Example 5. Consider the differential equation

y 00 + 16y = 0. (9)

Let y1 (t) = cos 4t. Then

d 0 d
y100 (t) = (y (t)) = (−4 sin 4t) = −16 cos 4t = −16y1 (t)
dt 1 dt
so that y1 (t) is a solution of (9). We leave it as an exercise to check that
y2 (t) = sin 4t and y3 (t) = 2y1 (t) − y2 (t) = 2 cos 4t − sin 4t are also solutions
to (9). More generally, you should check (as usual by direct substitution) that
y(t) = c1 cos 4t + c2 sin 4t is a solution to (9) for any choice of real numbers
c1 and c2 . It is true, but not obvious, that letting c1 and c2 vary over all real
numbers produces all solutions to y 00 + 16y = 0. See Exercise 31.

Implicitly Defined Solutions

Consider a relationship between the two variables t and y determined by the
equation
t2 + y 2 = 4. (10)
Does this equation define y as a function of t? The question is: Can we solve
this equation for y in terms of t uniquely, for t in an appropriate interval I?
The question of uniqueness is important because a function has only one value
y(t) for each t in the domain. Trying to solve for y as a function of t gives
y 2 = 4 − t2 , and since we are considering real valued functions, y 2 must be
nonnegative, which forces t2 ≤ 4. The natural domain for y as a function of
8 1 Introduction

t determined by (10) is thus I = [−2, 2], and we have two equally deserving
candidates for y(t):
p p
y1 (t) = 4 − t2 or y2 (t) = − 4 − t2 .

Of course, you will recognize immediately that the graph of y1 (t) is the upper
half of the circle of radius 2 centered at the origin of the (t, y)-plane, while
the graph of y2 (t) is the lower half. We will say that either of the functions
y1 (t) or y2 (t) is defined implicitly by the equation t2 + y 2 = 4. This means
that
t2 + (y1 (t))2 = 4 for all t ∈ I,
and similarly for y2 (t). If this last equation is differentiated with respect to
the independent variable t, we get

2t + 2y1 (t)y10 (t) = 0 for all t ∈ I,

and similarly for y2 (t). That is, both functions y1 (t) and y2 (t) implicitly defined
by (10) are solutions to the same differential equation

t + yy 0 = 0. (11)

Moreover, if we change the constant 4 to any other positive constant r, implicit

differentiation of the implicit equation t2 + y 2 = r with respect to t gives
the same differential equation t + yy 0 = 0 (after dividing by 2). Thus all of
the infinitely many functions implicitly defined by the family of equations
t2 + y 2 = r, as r varies over the positive real numbers, are solutions of the
same differential equation (11).
The analysis of the previous paragraph remains true for any family of
implicitly defined functions. If f (t, y) is a function of two variables, we will
say that a function y(t) defined on an interval I is implicitly defined by the
equation
f (t, y) = r,
where r is a fixed real number if

f (t, y(t)) = r for all t ∈ I. (12)

This is a precise expression of what we mean by the statement:

Solve the equation f (t, y) = r for y as a function of t.

Since, as we have seen with the equation t2 + y 2 = r, the equation f (t, y) = r

may have more than one way to solve for y as a function of t, and hence the
same two variable function f (t, y) can determine more than one implicitly
defined function y(t) from the same constant r. It can determine infinitely
many different implicitly defined functions if we let the constant r be chosen
 1.1 Basic Concepts and Terminology 9

from an infinite subset of R. By differentiating (12) with respect to t (using

the chain rule from multiple variable calculus), we find
∂f ∂f
(t, y(t)) + (t, y(t))y 0 (t) = 0.
∂t ∂y
Since the constant r is not present in this equation, we conclude that every
function implicitly defined by the equation f (t, y) = r, for any constant r, is
a solution of the same first order differential equation

∂f ∂f 0
+ y = 0. (13)
∂t ∂y

We shall refer to (13) as the differential equation for the family of curves
f (t, y) = r. One valuable technique that we will encounter in Chapter 2 is
to solve a first order differential equation by recognizing it as the differential
equation of a particular family of curves.
Example 6. Find the first order differential equation for the family of hy-
perbolas ty = r in the (t, y)-plane.
I Solution. The defining function for the family of curves is f (t, y) = ty so
(13) gives
∂f ∂f 0
+ y = y + ty 0 = 0
∂t ∂y
as the differential equation for this family. In standard form this equation is
y 0 = −y/t. Notice that this agrees with expectations, since for this simple
family ty = r, we can solve explicitly to get y = r/t (for t 6= 0) so that
y 0 = −r/t2 = −y/t. J

Initial Value Problems

As we have seen in the examples of differential equations and their solutions
presented in this section, differential equations generally have infinitely many
solutions so to specify a particular solution of interest, it is necessary to specify
additional data. What is usually convenient to specify is an initial value y(t0 )
for a first order equation and an initial value y(t0 ) and an initial derivative
y 0 (t0 ) in the case of a second order equation, with obvious extensions to higher
order equations. When the differential equation and initial values are specified,
then one obtains what is known as an initial value problem. Thus a first
order initial value problem in standard form is

y 0 = F (t, y), y(t0 ) = y0 (14)

while a second order equation in standard form is written

y 00 = F (t, y, y 0 ), y(t0 ) = y0 , y 0 (t0 ) = y1 . (15)

10 1 Introduction

Example 7. Determine a solution to each of the following initial value prob-

lems
1. y 0 = y − t, y(0) = −3.
2. y 00 = 2 − 6t, y(0) = −1, y 0 (0) = 2.
I Solution. 1. Recall from Example 3 that for each c ∈ R, the function
yc (t) = t + 1 + cet is a solution for y 0 = y − t. Thus our strategy is just to
try to match one of the yc (t) with the required initial condition y(0) = −3.
Thus
−3 = yc (0) = 1 + ce0 = 1 + c
requires that we take c = −4. Hence,

y(t) = y−4 (t) = t + 1 − 4et

is a solution of the initial value problem.

2. The second equation is asking for a function y(t) whose second derivative
is the given function 2 − t. But this is precisely the type of problem that
you learned to solve in calculus using integration. Integration of y 00 gives
Z t Z t
y 0 (t) − y 0 (0) = y 00 (x) dx = (2 − 6x) dx = 2t − 3t2 ,
0 0

so that y 0 (t) = y(0) + 2t − 3t2 = 2 + 2t − 3t2 . Note that we have used the
variable x in the integral, since the variable t appears in the limits of the
integral. Now integrate again to get
Z t Z t
y(t) − y(0) = y 0 (x) dx = (2 + 2x − 3x2 ) dx = 2t + t2 − t3 .
0 0

Hence we get y(t) = −1 + 2t + t − t3 as the solution of our second order

2

initial value problem. J

Exercises
1. What is the order of each of the following differential equations?
(a) y 2 y 0 = t3 (b) y 0 y 00 = t3
(c) t2 y 0 + ty = et (d) t2 y 00 + ty 0 + 3y = 0
(e) 3y 0 + 2y + y 00 = t2 (f ) t(y (4) )3 + (y 000 )4 = 1
(g) y 0 + t2 y = ty 4 (h) y 000 − 2y 00 + 3y 0 − y = 0
Determine whether each of the given functions yj (t) is a solution of the corre-
sponding differential equation.
2. y 0 = 2y: y1 (t) = 0, y2 (t) = t2 , y3 (t) = 3e2t , y4 (t) = 2e3t .
3. y 0 = 2y − 10: y1 (t) = 5, y2 (t) = 0, y3 (t) = 5e2t , y4 (t) = e2t + 5.
4. ty 0 = y: y1 (t) = 0, y2 (t) = 3t, y3 (t) = −5t, y4 (t) = t3 .
5. y 00 + 4y = 0: y1 (t) = e2t , y2 (t) = sin 2t, y3 (t) = cos(2t − 1), y4 (t) = t2 .
 1.1 Basic Concepts and Terminology 11

Verify that each of the given functions y(t) is a solution of the given differential
equation on the given interval I. Note that all of the functions depend on an arbitrary
constant c ∈ R.
6. y 0 = 3y + 12; y(t) = ce3t − 4, I = (−∞, ∞)
7. y 0 = −y + 3t; y(t) = ce−t + 3t − 3 I = (−∞, ∞)
8. y 0 = y 2 − y; y(t) = 1/(1 − cet ) I = (−∞, ∞) if c < 0, I = (− ln c, ∞) if
c>0 2
9. y 0 = 2ty; y(t) = cet , I = (−∞, ∞)
10. (t + 1)y + y = 0; y(t) = c(t + 1)−1 , I = (−1, ∞)
0

11. y 0 = y 2 ; y(t) = (c − t)−1 , I = (−∞, c)

Find the general solution of each of the following differential equations by in-
tegration. See the solution of Equation (2) in Example 7 for an example of this
technique.

12. y 0 = t + 3 13. y 0 = e2t − 1 14. y 0 = te−t

t+1
15. y 0 = 16. y 00 = 2t + 1 17. y 00 = 6 sin 3t
t
Find a solution to each of the following initial value problems. See Exercises
6 through 17 for the general solutions of these equations, and see the solution of
Equation (1) in Example 7 for an example of the technique of finding a solution of
an initial value problem from the knowledge of a general family of solutions.

18. y 0 = 3y + 12, y(0) = −2 19. y 0 = −y + 3t, y(0) = 0

0 2 0
20. y = y − y, y(0) = 1/2 21. (t + 1)y + y = 0, y(1) = −9
2t
0
22. y = e − 1, y(0) = 4 0
23. y = te −t
, y(0) = −1
00 0
24. y = 6 sin 3t, y(0) = 1, y (0) = 2

Find the first order differential equation for each of the following families of
curves. In each case c denotes an arbitrary real constant. See Example 6.

25. 3t2 + 4y 2 = c 26. y 2 − t2 − t3 = c

27. y = ce2t + t 28. y = ct3 + t2

29. Show that every solution of the equation y 0 = ky, where k ∈ R, is one of the
functions yc (t) = cekt , where c is an arbitrary real constant.
Hint: See the second paragraph of Example 2.
30. In Example 3 it was shown that every function yc (t) = t + 1 + cet , where c ∈ R
is a solution of the differential equation y 0 = y − t. Show that if y(t) is any
solution of y 0 = y − t, then y(t) = yc (t) for some constant c.
Hint: Show that z(t) = y(t) − t − 1 is a solution of the differential equation
y 0 = y and apply the previous exercise.
31. If k is a positive real number, use the following sequence of steps to show that
every solution y(t) of the second order differential equation y 00 + k2 y = 0 has
the form
y(t) = c1 cos kt + c2 sin kt
for arbitrary real constants c1 and c2 .
12 1 Introduction

(a) Show that each of the functions y(t) = c1 cos kt + c2 sin kt satisfies the
differential equation y 00 + k2 y = 0.
(b) Given real numbers a and b, show that there exists a solution of the initial
value problem
y 00 + k2 y = 0, y(0) = a, y 0 (0) = b.
(c) If y(t) is any solution of y 00 + k2 y = 0, show that the function E(t) =
(ky(t))2 + (y 0 (t))2 is constant. Conclude that E(t) = (ky(0))2 + (y 0 (0))2 for
all t ∈ R, and hence, if y(0) = y 0 (0) = 0, then E(t) = 0 for all t. Conclude
that if y(0) = y 0 (0) = 0, then y(t) = 0 for all t.
(d) Given real numbers a and b, show that there exists exactly one solution of
the initial value problem

y 00 + k2 y = 0, y(0) = a, y 0 (0) = b.

1.2 Examples of Differential Equations

As we have observed in the previous section, the rules, or physical laws, gov-
erning a quantity are frequently expressed mathematically as a relationship
between the instantaneous rate of change of the quantity and the value of
the quantity. If t denotes time and y(t) denotes the value at time t of the
quantity we wish to describe, then the instantaneous rate of change of y is the
derivative y 0 (t). Thus, a relation between the instantaneous rate of change of
y and the value of y is expressed by an equation

f (y(t), y 0 (t)) = 0. (1)

This is a first order differential equation in which the time variable (or inde-
pendent variable) does not appear explicitly. An equation in which the time
variable does not appear explicitly is said to be autonomous. Frequently, the
relation between y(t) and y 0 (t) will also depend on the time t. In this case the
differential equation has the form

f (t, y(t), y 0 (t)) = 0. (2)

In the following examples, we will show how to use some basic physical
laws to arrive at differential equations like (2) that can serve as models of
physical processes. Some of the resulting differential equations will be easy
to solve using the simple observations of the previous section; others will
require the techniques developed in later chapters. Our goal at the moment
is the modeling process itself, not the solutions. The first example will be the
falling body problem, as it was understood by the seventeenth century Italian
scientist Galileo Galilei (1564 – 1642).
 1.2 Examples of Differential Equations 13

The falling body problem

Suppose that a ball is dropped from the top of a building of height h. If y(t)
denotes the height of the ball at time t, then the falling body problem is
the problem of determining y(t) for all time t in the interval 0 ≤ t ≤ T , where
T denotes the time that the ball hits the ground. See Figure 1.4.

height = h

y(t)

Fig. 1.4. A falling ball.

Galileo expressed his theory of motion for falling bodies as the statement
that the motion is uniformly accelerated. According to Galileo “A motion is
said to be uniformly accelerated, when starting from rest, it acquires, during
equal time-intervals, equal increments of speed."1 Moreover, Galileo postulated
that the rate of uniform acceleration is the same for all bodies of similar
density, which is a way of ignoring effects like air resistance.
In the language of calculus Galileo’s law states that the rate of change of
speed is independent of time. Since the speed v(t) is the rate of change of
position y(t), Galileo’s law for the motion of a falling body can be expressed
as the second order differential equation

y 00 (t) = v 0 (t) = −g, (3)

where g is an unknown positive constant (independent of the particular body)

and the negative sign is used to indicate that we are measuring height above
the ground and the ball is falling. It is a straightforward exercise in calculus
to solve this differential equation by means of integration. Indeed, integration
of Equation (3) gives
Z t Z t
y 0 (t) − y 0 (0) = y 00 (x) dx = −g dx = −gt.
0 0

1
Dialogues Concerning Two New Sciences by Galileo Galilei, translated by Henry
Crew and Alfonso de Salvio, Dover Publications, ??, Page ??
14 1 Introduction

Since y 0 (0) = v0 , the initial velocity of the body, we find

y 0 (t) = −gt + v0 .

Integrating once more we obtain

g
y(t) = − t2 + v0 t + y0 , (4)
2
where the constant y0 is the initial position y(0). If we just drop the ball,
as we have postulated in the falling body problem, then the initial velocity
v0 = 0, and since y(0) = h, the height of the building, the solution of our
problem is
g
y(t) = − t2 + h. (5)
2
There remains the problem of computing the constant g, but there is an
experimental way to do this. Since T denotes the time when the ball hits the
ground, it follows that 0 = y(T ) = −gT 2 /2 + h so that

2h
g= . (6)
T2
Experimental calculations have shown that g (known as the gravitational
constant) has the value 32 ft/sec2 in the English system of units and the value
9.8 m/sec2 in the metric system of units. Thus, using the English system of
units, the height y(t) at time t of a ball released from an initial height of h ft
with an initial velocity of v0 ft/sec is given, assuming Galileo’s law of motion,
by
y(t) = −16t2 + v0 t + h. (7)

Example 1. If a ball is dropped from a height h, what will be the speed at

which it hits the ground?

p Solution. According to (6) the ball

I will hit the ground
√ at time T =
2h/g and the speed at this time is |y 0 (T )| = |−gT | = 2gh. J

Example 2. Suppose that a ball is tossed straight up with an initial velocity

v0 . Ignoring air resistance, how high will the ball go?

I Solution. Since the initial height h = 0, the equation of motion is y(t) =

− g2 t2 + v0 t. The maximum height is obtained when the velocity is zero. But
v(t) = y 0 (t) = −gt + v0 so v(t) = 0 occurs when t = v0 /g. At this time the
height is
 2
v2
 
g v0 v0
y(v0 /g) = − + v0 = 0.
2 g g 2g
For a numerical example, this says that a ball thrown straight up at an initial
speed of 40 ft/sec will reach a height of 402 /64 = 25 ft. J
 1.2 Examples of Differential Equations 15

Radioactive decay
Certain chemicals decompose into other substances at a rate that is propor-
tional to the amount present. Such a decomposition process is referred to as
radioactive decay and it is characterized by the mathematical model

y 0 = −ky, (8)

where y(t) denotes the amount of chemical present at time t, and k is a positive
constant called the decay constant.
The solution of the radioactive decay equation (8) is

y(t) = ce−kt (9)

where c is an arbitrary real constant. This can be seen by the same argument
used in Example 2 on Page 4 (see also Exercise 29 in Section 1.1). Specifically,
if y(t) is any solution of (8), multiply by ekt to get a function z(t) = ekt y(t).
Differentiating and using the fact that y(t) is a solution of (8), gives

z 0 (t) = ekt y 0 (t) + kekt y(t) = −kekt y(t) + kekt y(t) = 0, for all t,

so that z(t) = c for some constant c ∈ R. Thus, y(t) = e−kt z(t) = ce−kt , as
required.
If there is a quantity y0 = y(t0 ) of the radioactive chemical present at the
initial time t0 , then the condition y0 = y(t0 ) = ce−kt0 gives the value of the
constant c = y0 ekt0 . Putting this value of c into (9) gives

y(t) = y0 e−k(t−t0 ) (10)

for the amount present at time t, assuming that y0 is present at time t0 .

There is still the question of determining the proportionality constant k,
which depends on the particular substance being studied. Suppose we have
measured the amount present at time t0 , namely y0 and also the amount
present at time t1 6= t0 , say y1 . Then

y1 = y(t1 ) = y0 e−k(t1 −t0 ) ,

and in this equation, all quantities are known except for k. Thus, we can solve
this equation for the unknown decay constant k. If we do the calculation, we
get
1
k= ln(y0 /y1 ). (11)
t1 − t0
The conclusion is that if the value of a radioactive quantity is known at
two different times, then it is possible to compute, using equations (10) and
(11), the quantity present at all future times. If, in this expression, we take
y1 = y0 /2 and let T = t1 − t0 , then we get
16 1 Introduction

ln 2
k= . (12)
T
That is, the decay constant can be computed if the time T that it takes the
material to decay to one-half its original amount is known. This time is usually
referred to as the half-life of the radioactive material.
Example 3. The half-life of radium is 1620 years. What is the amount y(t)
of radium remaining after t years if the initial quantity was y0 ? What per-
centage of the original quantity will remain after 500 years?
I Solution. From (12) the decay constant is calculated to be k = ln 2/1620 ≈
.000428. Thus y(t) = y0 e−.000428t . Note that we have taken the initial
time t0 to be 0. After 500 years, the amount remaining will be y(500) =
y0 e−.000428×500 ≈ .807y0 . Thus, after 500 years, approximately 80.7% of the
original amount of radium will still be present. J

Newton’s law of heating and cooling

Newton’s Law of heating and cooling states that the rate of change
of the temperature of an object is proportional to the difference between the
temperature of the object and the temperature of the surrounding medium. If
T (t) denotes the temperature of the object at time t and A is the temperature
of the surrounding medium (which will be assumed to be constant in this
simple case), then Newton’s law is expressed in mathematical language as the
differential equation
dT
(t) = k(T (t) − A). (13)
dt
If T (t) > A then the object will cool, so that dT /dt < 0, while if T (t) < A then
the object will heat up toward the ambient temperature A, so that dT /dt > 0.
In both cases, the proportionality constant k will be negative.
Letting y(t) = T (t) − A, and recalling that the temperature A of the
surrounding medium is assumed to be constant, (13) takes the form
d dT
y0 = (T − A) = = k(T − A) = ky. (14)
dt dt
This is the same equation as the equation of radioactive decay (8), and we have
already observed (Equation (9)) that this equation has the solution y(t) =
ce−kt where c = y(0). Thus, (13) has the solution
T (t) = A + (T (0) − A)ekt . (15)
Example 4. A thin plate is placed in boiling water to be sterilized in prepa-
ration for receiving a cell culture. The plate is removed from the boiling water
(100◦ C) and put in a 20◦ C room to cool. After 5 min the plate has cooled to a
temperature of 80◦ C. The culture can be introduced when the temperature is
30◦ C. How much longer will it be necessary to wait before the plate is ready
to receive the cell culture?
 1.2 Examples of Differential Equations 17

I Solution. Let T (t) denote the temperature of the plate at time t, where
t = 0 corresponds to the time when the plate is removed from the boiling
water. Thus, we are assuming that the initial temperature is T (0) = 100 deg C.
In this example, the surrounding medium is the air in the room, which is
assumed to have a temperature of A = 20 deg C. According to Newton’s law
of cooling (Equation (13)) the differential equation governing the cooling of
the plate is
dT
(t) = k(T (t) − A) = k(T (t) − 20),
dt
which, according to (15) has the solution

T (t) = A + (T (0) − A)ekt = 20 + 80ekt .

This will completely describe T (t), once the parameter k is determined, which
is accomplished by taking advantage of the fact that the temperature of the
plate is 80 deg C after 5 min, i.e., T (5) = 80. Thus,

80 = T (5) = 20 + 80e5k ,

so that 60/80 = e5k which implies that k = .2 ln .75 = −.057536414. There-

fore, the temperature of the plate at all times after it is removed from the
boiling water is
T (t) = 20 + 80e−.057536414t .
We need the time at which the temperature is 30 deg C, which we can deter-
mine by solving for t in the equation

30 = T (t) = 20 + 80e−.057536414t .

Thus, e−.057536414t = 10/80 = .125 so

ln .125
t= = 36.14 min.
−.057536414
Thus, the answer to the question asked is that since 5 min was needed it reach
80 deg C, it takes an additional 36.14 − 5 = 31.14 min to reach the desired
temperature of 30 deg C.

Mixing problems
Example 5. Consider a tank which contains 2000 gallons of water in which
10 lbs of salt are dissolved. Suppose that brine (a water-salt mixture) with a
concentration of 0.1 lb/gal enters the tank at a rate of 2 gal/min, and assume
that the well-stirred mixture flows from the tank at the same rate of 2 gal/min.
Find the amount y(t) of salt (expressed in pounds) which is present in the
tank at all times t, measured in minutes after the initial time (t = 0) when 10
lbs are present.
18 1 Introduction

I Solution. This is another example of where it is easier to describe how

y(t) changes, that is y 0 (t), than it is to directly describe y(t). Since the de-
scription of y 0 (t) will also include y(t), a differential equation will result. Start
by noticing that at time t0 , y(t0 ) lbs of salt are present and at a later time t,
the amount of salt in the tank is given by

y(t) = y(t0 ) + A(t0 , t) − S(t0 , t) (16)

where A(t0 , t) is the amount of salt added between times t0 and t and S(t0 , t)
is the amount removed between times t0 and t. To compute A(t0 , t) note that

A(t0 , t) = cin (t)rin (t)∆t

where

cin (t) = concentration of salt solution being added = 0.1 lb/gal

rin (t) = volume rate of salt solution being added = 2 gal/min
∆t = t − t0 = time interval between t0 and t,

so that
A(t0 , t) = (0.1) × (2) × ∆t. (17)
By exactly the same reasoning,

S(t0 , t) = cout (t)rout (t)∆t

where

cout (t) = concentration of salt solution being removed

rout (t) = volume rate of salt solution being removed
∆t = t − t0 = time interval between t0 and t.

The number of gallons per minute flowing out of the tank is the same as the
rate flowing in, namely, 2 gal/min, so that rout = 2 gal/min. However, cout ,
the number of pounds per gallon being removed at any given time t will be
given by y(t)/V (t), that is divide the total number of pounds of salt in the
tank at time t (namely y(t)) by the current total volume V (t) of solution in
the tank. In our case, V (t) is always 2000 gal (the flow in and the flow out
balance), but y(t) changes with time, and that is what we ultimately will want
to compute. If t is “close" to t0 then we can assume that y(t) ≈ y(t0 ) so that
 
y(t0 )
S(t0 , t) ≈ lbs/gal × (2 gal/min) × ∆t. (18)
2000
Combining (16), (17), and (18) gives

y(t) − y(t0 ) = A(t0 , t) − S(t0 , t)

y(t0 )
≈ (0.2)(t − t0 ) − 2 (t − t0 ).
2000
 1.2 Examples of Differential Equations 19

Dividing this by t − t0 and letting t → t0 gives the equation

1
y 0 (t0 ) = 0.2 − y(t0 ),
1000
which we recognize as a differential equation. Note that it is the process of
taking the limit as t → t0 that allows us to return to an equation, rather than
dealing only with an approximation. This is a manifestation of what we mean
when we indicate that it is frequently easier to describe the way something
changes, that is y 0 (t), rather than “what is," i.e. y(t) itself.
Since t0 is an arbitrary time, we can write the above equation as a differ-
ential equation
1
y 0 = (0.2) − y (19)
1000
and it becomes an initial value problem by specifying that we want y(0) = 10,
that is, there are 10 lbs of salt initially present in the tank.
The equation (19) is easily solved by factoring out −1/1000 = −10−3 to
rewrite it as
y 0 = −10−3(y − 200). (20)
This is the same type of equation as Newton’s law of cooling (Equation (13)),
where k = −10−3 and A = 200. Setting z = y − 200 transforms (20) into the
basic radioactive decay type equation z 0 = −10−3 z, which has the solution
−3
z(t) = z(0)e−10 t . Since y(t) = z(t) + 200 we get
−3
y(t) = 200 + (y(0) − 200)e−10 t
= 200 − 190e(−0.001)t. (21)

The graph of y(t) is found in Figure 1.5. Note that as t increases, the amount
of salt approaches 200 lbs, which is the same amount one would find in a 2000
gallon mixture with the given concentration 0.1 lb/gal. J

y
200
Pounds of Salt

150

100

50

0 t
0 500 1000 1500 2000 2500 3000
Time in minutes
Fig. 1.5. The amount of salt y(t) as a function of time.
20 1 Introduction

The following example summarizes a slightly more general situation than

that covered by the previous numerical example. It will lead to a differential
equation (23) more general than the equation y 0 = ay + b, where a and b
are constants, that has characterized all the examples so far presented. The
solution of equations of this type will be described in the next chapter.

Example 6 (Mixing problem). A tank initially holds V0 gal of brine(a

water-salt mixture) that contains a lb of salt. Another brine solution, contain-
ing c lb of salt per gallon, is poured into the tank at a rate of r gal/min. The
mixture is stirred to maintain uniformity of concentration of salt at all parts
of the tank, and the stirred mixture flows out of the tank at the rate of R
gal/min. Let y(t) denote the amount of salt (measured in pounds) in the tank
at time t. Find an initial value problem for y(t).

I Solution. We are searching for an equation which describes the rate of

change of the amount of salt in the tank at time t, i.e., y 0 (t). The key observa-
tion, which we shall refer to as the balance equation, is that this rate of change
is the difference between the rate at which salt is being added to the tank and
the rate at which the salt is being removed from the tank. In symbols:

y 0 (t) = Rate in − Rate out. (22)

The rate that salt is being added is easy to compute. It is rc lb/min (c lb/gal
× r gal/min = rc lb/min). Note that this is the appropriate units for a rate,
namely an amount divided by a time. We still need to compute the rate at
which salt is leaving the tank. To do this we first need to know the number of
gallons V (t) of brine in the tank at time t. But this is just the initial volume
plus the amount added up to time t minus the amount removed up to time t.
That is, V (t) = V0 + rt − Rt = V0 + (r − R)t. Since y(t) denotes the amount
of salt present in the tank at time t, the concentration of salt at time t is
y(t)/V (t) = y(t)/(V0 − (r − R)t), and the rate at which salt leaves the tank
is R × y(t)/V (t) = Ry(t)/(V0 + (r − R)t). Thus,

y 0 (t) = Rate in − Rate out

R
= rc − y(t)
V0 + (r − R)t

In the standard form of a linear differential equation, the equation for the rate
of change of y(t) is

R
y 0 (t) + y(t) = rc. (23)
V0 + (r − R)t

This becomes an initial value problem by remembering that y(0) = a. As in

the previous example, this is a first order linear differential equation, and the
solutions will be studied in Section 2.2. J
 1.2 Examples of Differential Equations 21

Remark 7. You should definitely not memorize a formula like Equation (23).
What you should remember is how it was set up so that you can set up your
own problems, even if the circumstances are slightly different from the one
given above. As one example of a possible variation, you might encounter a
situation in which the volume V (t) varies in a nonlinear manner such as, for
example, V (t) = 5 + 3e−2t .

Electric Circuits
Some of the simplest electrical circuits consist of a single electromotive force
(e.g. a battery or generator), a resistor, and either a capacitor, which stores
charge, or an inductor, which resists any change in current. These circuits
are illustrated in Figure 1.6. If the switch is closed at time t = 0, then the
current i(t) flowing through the closed circuit is governed by a mathematical
model similar to that governing cooling and mixing problems. We will start
by summarizing the elementary facts about electricity that are needed.
t=0 t=0
i(t) i(t)

+ +
V R V R
− −

C L
(a) RC Circuit (b) RL Circuit

Fig. 1.6. RC and RL circuits

An electromotive force V is denoted by the symbol −+ and

measured in volts (V). The purpose of the electromotive force is to provide
an electric charge q, measured in coulombs C, whose movement around the
circuit produces a current i(t), which is the time rate of change of charge. In
symbols:
dq
i(t) = . (24)
dt
Current is measured in amperes (A), where 1 ampere = 1 A = 1 coulomb per
sec = 1 C/s. The reference direction for a current is
A resistor of resistance R, measured in Ohms (Ω) is a circuit element,
denoted by the symbol , which results in a voltage drop VR across the
resistor governed by Ohm’s law :
22 1 Introduction

VR = Ri(t). (25)

The capacitor, denoted and measured in units of Farads (F) stores

charge, and in doing so results in a voltage drop VC across the capacitor C.
The charge stored q(t) is proportional to the voltage drop VC :

q(t) = CVC . (26)

An inductor of inductance L is a circuit element denoted by the symbol

and measured in Henrys (H), that resists changes in the current i(t).
This is manifested by a voltage drop VL that is proportional to the rate of
change of current:
di
VL = L . (27)
dt
In the simple RC (resistor-capacitor) and RL (resistor-inductor) circuits
considered here, the resistance R, capacitance C and inductance L are con-
stants while the electromotive force will be an applied voltage that could be
either constant, such as supplied by a battery, or time-varying such as that
found in an alternating current circuit. The fundamental physical law that
governs voltage drops around a closed circuit is Kirchhoff ’s voltage law:

The algebraic sum of the voltage drops around any closed loop is 0. (28)

Let us apply the basic principles expressed by Equations (24) – (28) to

the circuits of Figures 1.6 (a) and (b) to see how they lead to differential
equations governing the movement of charge around the closed circuit, once
the switch has been closed at time t = 0. In the RC circuit of Figure 1.6 (a),
there are voltage drops VR across the resistor and VC across the capacitor,
while the electromotive force V provides a positive voltage, which corresponds
to a voltage drop of −V . Kirchhoff’s voltage law then gives

VR + VC − V = 0.

Using Ohm’s law (25) and the capacitor proportionality rule (26), this gives
an equation
q(t)
Ri(t) + = V,
C
and recalling that current i(t) is the time rate of change of charge, we arrive
at the differential equation
dq q
R + = V. (29)
dt C
This differential equation describes the movement of charge around the circuit
in Figure 1.6 (a).
The RL circuit in Figure 1.6 (b) is treated similarly. Kirchhoff’s law gives
the equation
 1.2 Examples of Differential Equations 23

VR + VL = V
and applying Ohm’s law and the inductor proportionality equation (27) gives
a differential equation
di
L + Ri = V. (30)
dt
Note that the differential equation (29) determined by the RC circuit of Figure
1.6 (a) is a differential equation for the charge q(t), while the equation (30)
determined by the RL circuit of Figure 1.6 (b) is a differential equation for
the current i(t). However, if the applied voltage is a constant V , then both of
these equations have the basic form

ay 0 + by = c

where a, b, and c are constants. Note that this is exactly like the equations
considered earlier in this section.
We will conclude by considering a couple of numerical equations of the
type involved in studying RC and RL circuits. Numerical examples
here.
1. How long does it take a ball dropped from the top of a 400 ft tower to reach
the ground? What will be the speed when it hits the ground?

2. A ball is thrown straight up from the top of a building of height h. If the initial
velocity is v0 (and ignoring air resistance), how long will it take to reach the
maximum height? How long will it take to reach the ground?

3. A ball is thrown straight up from the ground with an initial velocity of 25

meters/sec. There will be two times when the ball is at a height of 15 meters.
Find those two times.

4. On planet P the following experiment is performed. A small rock is dropped

from a height of 4 feet and it is observed that it hits the ground in 1 sec.
Suppose another stone is dropped from a height of 1000 feet. What will be the
height after 5 sec.? How long will it take for the stone to hit the ground?

5. Radium decomposes at a rate proportional to the amount present. Express this

proportionality statement as a differential equation for R(t), the amount of
radium present at time t.

6. Bacteria are placed in a sugar solution at time t = 0. Assuming adequate food

and space for growth, the bacteria will grow at a rate proportional to the current
population of bacteria. Write a differential equation satisfied by the number P (t)
of bacteria present at time t.

7. Continuing with the last exercise, assume that the food source for the bacteria
is adequate, but that the colony is limited by space to a maximum population
M . Write a differential equation for the population P (t) which expresses the
assumption that the growth rate of the bacteria is proportional to the product
of the number of bacteria currently present and the difference between M and
the current population.
24 1 Introduction

8. If a bottle of your favorite beverage is at room temperature (say 70◦ F) and it

is then placed in a tub of ice at time t = 0, use Newton’s law of cooling to write
an initial value problem which is satisfied by the temperature T (t) of the bottle
at time t.
9. A turkey, which has an initial temperature of 40◦ (Fahrenheit), is placed into a
350◦ oven. After one hour the temperature of the turkey is 120◦ . Use Newton’s
Law of heating and cooling to find (1) the temperature of the turkey after 2
hours, and (2) how many hours it takes for the temperature of the turkey to
reach 250◦ .
10. A cup of coffee, brewed at 180◦ (Fahrenheit), is brought into a car with in-
side temperature 70◦ . After 3 minutes the coffee cools to 140◦ . What is the
temperature 2 minutes later?
11. The temperature outside a house is 90◦ and inside it is kept at 65◦ . A ther-
mometer is brought from the outside reading 90◦ and after 10 minutes it reads
85◦ . How long will it take to read 75◦ ? What will the thermometer read after
an hour?
12. A cold can of soda is taken out of a refrigerator with a temperature of 40◦ and
left to stand on the countertop where the temperature is 70◦ . After 2 hours the
temperature of the can is 60◦ . What was the temperature of the can 1 hour
after it was removed from the refrigerator?
13. A large cup hot of coffee is bought from a local drive through restaurant and
placed in a cup holder in a vehicle. The inside temperature of the vehicle is
70◦ Fahrenheit. After 5 minutes the driver spills the coffee on himself a receives
a severe burn. Doctors determine that to receive a burn of this severity, the
temperature of the coffee must have been about 150◦ . If the temperature of the
coffee was 142◦ 6 minutes after it was sold what was the temperature at which
the restaurant served it.
14. A student wishes to have some friends over to watch a football game. She wants
to have cold beer ready to drink when her friends arrive at 4 p.m. According to
her tastes the temperature of beer can be served when its temperature is 50◦ .
Her experience shows that when she places 80◦ beer in the refrigerator that is
kept at a constant temperature of 40◦ it cools to 60◦ in an hour. By what time
should she put the beer in the refrigerator to ensure that it will be ready for
her friends?

1.3 Direction Fields

The geometric interpretation of the derivative of a function y(t) at t0 as the
slope of the tangent line to the graph of y(t) at (t0 , y(t0 )) provides us with an
elementary and often very effective method for the visualization of the solu-
tion curves (:= graphs of solutions) for a first order differential equation. The
visualization process involves the construction of what is known as a direc-
tion field or slope field for the differential equation. For this construction
we proceed as follows.
 1.3 Direction Fields 25

Construction of Direction Fields

(1) If the equation is not already in standard form (Equation (5)) solve the
equation for y 0 to put it in the standard form y 0 = F (t, y).
(2) Choose a grid of points in a rectangle R = {(t, y) : a ≤ t ≤ b; c ≤ y ≤ d}
in the (t, y)-plane.
(3) At each grid point (t, y), the number F (t, y) represents the slope of a
solution curve through this point; for example if y 0 = y 2 − t so that
F (t, y) = y 2 − t, then at the point (1, 1) the slope is F (1, 1) = 12 − 1 = 0,
at the point (2, 1) the slope is F (2, 1) = 12 − 2 = −1, and at the point
(1, −2) the slope is F (1, −2) = 3.
(4) Through the point (t, y) draw a small line segment having the slope F (t, y).
Thus, for the equation y 0 = y 2 − t, we would draw a small line segment of
slope 0 through (1, 1), slope −1 through (2, 1) and slope 3 through (1, −2).
With a graphing calculator, one of the computer mathematics programs
Maple, Mathematica or MATLAB (which we refer to as the three M’s)
2
, or with pencil, paper, and a lot of patience, you can draw many such
line segments. The resulting picture is called a direction field for the
differential equation y 0 = F (t, y).
(5) With some luck with respect to scaling and the selection of the (t, y)-
rectangle R, you will be able to visualize some of the line segments running
together to make a graph of one of the solution curves.
(6) To sketch a solution curve of y 0 = F (t, y) from a direction field, start
with a point P0 = (t0 , y0 ) on the grid, and sketch a short curve through
P0 with tangent slope F (t0 , y0 ). Follow this until you are at or close to
another grid point P1 = (t1 , y1 ). Now continue the curve segment by using
the updated tangent slope F (t1 , y1 ). Continue this process until you are
forced to leave your sample rectangle R. The resulting curve will be an
approximate solution to the initial value problem y 0 = F (t, y), y(t0 ) = y0 .

Example 1. Draw the direction field for the differential equation yy 0 = −t.
Draw several solution curves on the direction field, then solve the differential
equation explicitly and describe the general solution.

I Solution. Before we can draw the direction field, it is necessary to first

put the differential equation yy 0 = −t into standard form by solving for y 0 .
Solving for y 0 gives the equation
t
(∗) y0 = − .
y

2
We have used the Student Edition of MATLAB, Version 6, and the
functions dfield6 and pplane6 which we downloaded from the webpage
http://math.rice.edu/dfield. To see dfield6 in action, enter dfield6 at the MATLAB
prompt
26 1 Introduction

0
y

−1

−2

−3

−4
−4 −2 0 2 4
t

Fig. 1.7. Direction Field of yy 0 = −t

Notice that this equation is not defined for y = 0, even though the original
equation is. Thus, we should be alert to potential problems arising from this
defect. We have chosen a rectangle R = {(t, y) : −4 ≤ t, y ≤ 4} for drawing
the direction field, and we have chosen to use 20 sample points in each direc-
tion, which gives a total of 400 grid points where a slope line will be drawn.
Naturally, this is being done by computer (using the dfield6 tool in MatLab),
and not by hand. Figure 1.7 gives the completed direction field, and Figure 1.8
is the same direction field with several solution curves drawn in. The solutions
which are drawn in are the solutions of the initial value problems yy 0 = −t,
y(0) = ±1, ±2, ±3. The solution curves appear to be half circles centered at
(0, 0). Since the equation yy 0 = −t is separable, we can verify that this is in
fact true by explicitly solving the equation. Writing the equation in differential
form gives ydy = −tdt and integrating gives

y2 t2
= − + c.
2 2
After multiplying by 2 and renaming the constant, we see that the solutions
of yy 0 = −t are given implicitly by y 2 + t2 = √c. Thus, there are two families
of solutions√of yy 0 = −t, specifically, y1 (t) = c − t2 (upper semicircle) and
y2 (t) = − c − t2 (lower semicircle). For both families of functions, c is a
 1.3 Direction Fields 27

0
y

−1

−2

−3

−4
−4 −2 0 2 4
t

Fig. 1.8. Solution Curves for yy 0 = −t

√ √
positive constant and the functions are defined on the interval
√ (− √ c, c). For
the solutions drawn in Figure 1.8, the constant c is √ 1, 2, and 3. Notice
that, although y1 and y2 are both defined for t = ± c, they do not satisfy
the differential equation at these points since y10 and y20 do not exist at these
2 2
points. Geometrically, this is a reflection
√ of the fact that the circle t + y = c
has a vertical tangent at the points (± c, 0) on the t-axis. This is the “defect"
that you were warned could occur because the equation yy 0 = −t, when put
in standard form y 0 = −t/y, is not defined for y = 0. J

It may happen that a formula solution for the differential equation y 0 =

F (t, y) is possible, but the formula is sufficiently complicated that it does not
shed much light on the nature of the solutions. In such a situation, it may
happen that constructing a direction field and drawing the solution curves on
the direction field gives useful insight concerning the solutions. The following
example is a situation where the picture is more illuminating than the formula.
t−2
Example 2. Solve the differential equation y 0 = .
3y 2 − 7
I Solution. The equation is separable, so we proceed as usual by separating
the variables, writing the equation in differential form, and then integrating
28 1 Introduction

0
y

−5
−4 −2 0 2 4 6 8
t
t−2
Fig. 1.9. Direction Field of y 0 =
3y 2 − 7

both sides of the equation. In the present case, the differential form of the
equation is (3y 2 − 7) dy = (t − 2) dt, so that, after integration and clearing
denominators, we find that the general solution is given by the implicit equa-
tion

(∗) 2y 3 − 14y = t2 − 4t + c.

While there is a formula for solving a cubic equation,3 it is a messy formula

which does not necessarily shed great light upon the nature of the solutions
t−2
as functions of t. However, if we compute the direction field of y 0 = 2 ,
3y − 7
and use it to draw some solution curves, we see a great deal more concerning
the nature of the solutions. Figure 1.9 is the direction field and Figure 1.10 is
the direction field with several solutions drawn in. Some observations which
can be made from the picture are:
• In the lower part of the picture, the curves seem to be deformed ovals
centered about the point P ≈ (2, −1.5).
3
The formula is known as Cardano’s formula after Girolamo Cardano (1501 –
1576), who was the first to publish it.
 1.3 Direction Fields 29

0
y

−5
−4 −2 0 2 4 6 8
t
t−2
Fig. 1.10. Solution Curves for y 0 =
3y 2 − 7

• Above the point Q ≈ (2, 2), the curves no longer are closed, but appear to
increase indefinitely in both directions.
J
We conclude our list of examples of direction fields with an example for
which the explicit solution formula, found by a method to be considered later,
gives even less insight than that considered in the last example. Nevertheless,
the direction field and some appropriately chosen solution curves drawn on the
direction field, suggest a number of properties of solutions of the differential
equation.
Example 3. The example to be considered is the differential equation

(∗∗) y 0 = y 2 − t.

This equation certainly does not look any more complicated than those con-
sidered in previous examples. In fact, the right hand side of this equation is
a quadratic which looks simple enough, certainly simpler than the right hand
side of the previous example. The parabola y 2 − t = 0 has a particularly sim-
ple meaning on the direction field. Namely, every solution of the differential
equation y 0 = y 2 − t which touches the parabola will have a horizontal tangent
30 1 Introduction

at that point. That is, for every point (t0 , y(t0 )) on the graph of a solution
y(t) for which y(t0 )2 − t0 = 0, we will have y 0 (t0 ) = 0. The curve y 2 − t = 0
is known as the nullcline of the differential equation y 0 = y 2 − t. Figure 1.11
is the direction field for y 0 = y 2 − t. Figure 1.12 shows the solution of the
equation y 0 = y 2 − t which has the initial value y(0) = 0, while Figure 1.13
shows a number of different solutions to the equation satisfying various initial
conditions y(0) = y0 . Unlike the previous examples we have considered, there
is no simple formula which gives all of the solutions of y 0 = y 2 − t. There is a
formula which involves a family of functions known as Bessel functions. Bessel
functions are themselves defined as solutions of a particular second order lin-
ear differential equation. For those who are curious, we note that the general
solution of y 0 = y 2 − t is

√ cK(− 32 , 23 t3/2 ) − I(− 32 , 23 t3/2 )

y(t) = t ,
cK( 31 , 23 t3/2 ) + I( 31 , 23 t3/2 )

where ∞
X 1
I(µ, z) := (z/2)2k+µ
Γ (k + 1)Γ (k + µ + 1)
k=0
R∞
is the modified Bessel function of the first kind, where Γ (x) := 0 e−t tx−1 dt
denotes the Gamma function, and where
π
K(µ, z) := (I(−µ, x) − I(µ, x))
2 sin(µx)

is the modified Bessel function of the second kind. 4 As we can see, even if an
analytic expression for the general solution of a first order differential equation
can be found, it might not be very helpful on first sight, and the direction field
may give substantially more insight into the true nature of the solutions.
For example, a detailed analysis of the direction field (see Figure 1.13)
reveals that the plane seems to be divided into two regions defined by some
curve fu (t). Solution curves going through points above fu (t) tend towards
infinity as t → ∞, whereas solution curves passing through points below fu (t)
seem to approach the solution curve fd (t) with y(0) = 0 as t → ∞.
The equation y 0 = y 2 − t is an example of a type of differential equation
known as a Riccati equation. A Ricatti equation is a first order differential
equation of the form
4
The solution above can be found easily with symbolic calculators like Maple,
Mathematica or MATLAB’s Symbolic Toolbox which provides a link between the
numerical powerhouse MATLAB and the symbolic computing engine Maple. The
routine dsolve is certainly one of the most useful differential equation tools in the
Symbolic Toolbox. For example, to find the solution of y 0 (t) = y(t)2 − t one simply
types
dsolve(0 Dy = y 2 − t 0 )
after the MATLAP prompt and pushes Enter.
 1.3 Direction Fields 31

-1

-2

-3

-4

-2 0 2 4 6 8 10

2
Fig. 1.11. Direction Field of y = y − t
0

x'=x -t

0
x

-1

-2

-3

-4

-2 0 2 4 6 8 10
t

Fig. 1.12. The solution curve for y 0 = y 2 − t with y(0) = 0

32 1 Introduction
x'=x -t

0
x

-1

-2

-3

-4

-2 0 2 4 6 8 10
t
2
Fig. 1.13. Solution curves for y = y − t
0

y 0 = a(t)y 2 + b(t)y + c(t),

where a(t), b(t) and c(t) are continuous functions of t. For more information on
this important class of differential equations, we refer to [Zw] and to Section
??.
As a final observation note that a number of the solution curves on Figure
1.13 appear to merge into one trajectory at certain regions of the display
window. To see that this is not the case one can use the zoom option in the
dfield6 tool, or, one might use the crucial theoretical results of Section 2.3. As
we will see there, under mild smoothness assumptions on the function F (t, y),
it is absolutely certain that the solution curves (trajectories) of an equation
y 0 = F (t, y) can never intersect.

Exercises
For each of the following differential equations, sketch a direction field on the rec-
tangle R = {(t, y) : −2 ≤ t, y ≤ 2}. You may do the direction fields by hand on
graph paper using the points in R with integer coordinates as grid points. That is
t and y are each chosen from the set {−2, −2, 0, 1, 2}. Alternatively, you may use
a graphing calculator or a computer, where you could try 20 sample values for each
of t and y, for a total of 400 grid points.
1. y 0 = y − 1

2. y 0 = t
 1.3 Direction Fields 33

3. y 0 = t2

4. y 0 = y 2

5. y 0 = y(y + 1)
In Exercises 6 – 11, a differential equation is given together with its direction
field. One solution is already drawn in. Draw at least five more representative
solutions on the direction field. You may choose whatever initial conditions
seem reasonable, or you can simply draw in the solutions with initial conditions
y(0) = −2, −1, 0, 1, and 2. Looking at the direction field can you tell if there
are any constant solutions y(t) = c? If so, list them. Are there other straight
line solutions that you can see from the direction field?

y0 = 1 − y2

3
2
1
6. 0
y

−1
−2
−3
−2 0 2
t

y0 = y − t

3
2
1

7. 0
y

−1
−2
−3
−2 0 2
t
34 1 Introduction

y 0 = −ty

3
2
1

8. 0
y

−1
−2
−3
−2 0 2
t

y 0 = y − t2

3
2
1
9. 0
y

−1
−2
−3
−2 0 2
t
 1.3 Direction Fields 35

y 0 = ty 2

3
2
1
10. 0
y

−1
−2
−3
−2 0 2
t
ty
y0 =
1+y

3
2
1
11. 0
y

−1
−2
−3
−2 0 2
t
2
First Order Differential Equations

2.1 Separable Equations

In this section and the next we shall illustrate how to obtain solutions for
two particularly important classes of first order differential equations. Both
classes of equations are described by means of restrictions on the type of
function F (t, y) which appears on the right hand side of a first order ordinary
differential equation given in standard form

y 0 = F (t, y). (1)

The simplest of the standard types of first-order equations are those with
separable variables; that is, equations of the form

y 0 = h(t)g(y). (2)

Such equations are said to be separable equations. Thus, an equation y 0 =

F (t, y) is a separable equation provided that the right hand side F (t, y) can
be written a a product of a function of t and a function of y. Most functions
of two variables are not the product of two one variable functions.
Example 1. Identify the separable equations from among the following list
of differential equations.
1. y 0 = t2 y 2
2. y 0 = t2 + y
t−y
3. y 0 =
t+y
4. y 0 = y − y 2
5. (2t − 1)(y 2 − 1)y 0 + t − y − 1 + ty = 0
6. y 0 = f (t)
7. y 0 = p(t)y
8. y 00 = ty
38 2 First Order Differential Equations

I Solution. Equations 1, 4, 5, 6, and 7 are separable. For example, in Equa-

tion 4, h(t) = 1 and g(y) = y − y 2 , while, in Example 6, h(t) = f (t) and
g(y) = 1. To see that Equation 5 is separable, we bring all terms not contain-
ing y 0 to the other side of the equation; i.e.,
(2t − 1)(y 2 − 1)y 0 = −t + y + 1 − ty = −t(1 + y) + 1 + y = (1 + y)(1 − t).
Solving this equation for y 0 gives
(1 − t) (1 + y)
y0 = · ,
(2t − 1) (y 2 − 1)
which is clearly separable. Equations 2 and 3 are not separable since neither
right hand side can be written as product of a function of t and a function
of y. Equation 8 is not a separable equation, even though the right hand side
is ty = h(t)g(y), since it is a second order equation and our definition of
separable applies only to first order equations. J
Equation 6 in the above example, namely y 0 = f (t) is particularly simple
to solve. This is precisely the differential equation that you spent half of your
calculus course understanding, both what it means and how to solve it for a
number of common functions f (t). Specifically, what we are looking for in this
case is an antiderivative of the function f (t), that is, a function y(t) such that
y 0 (t) = f (t). Recall from calculus that if f (t) is a continuous function on an
interval I = (a, b), then the Fundamental Theorem of Calculus guarantees that
there is an antiderivative of f (t) on I. Let F (t) be any antiderivative of f (t) on
I. Then, if y(t) is any solution to y 0 = f (t), it follows that y 0 (t) = f (t) = F 0 (t)
for all t ∈ I. Since two functions which have the same derivatives on an interval
I differ by a constant c, we see that the general solution to y 0 = f (t) is

y(t) = F (t) + c. (3)

There are a couple of important comments to make concerning Equation
(3).
1. The antiderivative of f exists on any interval I on which f is continuous.
This is the main point of the Fundamental Theorem of Calculus. Hence
the equation y 0 = f (t) has a solution on any interval I on which the
function f is continuous.
2. The constant c in Equation 3 can be determined by specifying y(t0 ) for
some t0 ∈ I. For example, the solution to y 0 = 6t2 , y(−1) = 3 is y(t) =
2t3 + c where 3 = y(−1) = 2(−1)3 + c so c = 5 and y(t) = 2t3 + 5.
3. The indefinite integral notation is frequently used for antiderivatives. Thus
the equation Z
y(t) = f (t) dt

just means that y(t) is an antiderivative of f (t). In this notation the

constant c in Equation 3 is implicit, although in some instances we may
write out the constant c explicitly for emphasis.
 2.1 Separable Equations 39

4. The formula y(t) = f (t) dt is valid even if the integral cannot be com-
R

puted in terms of elementary functions. In such a case, you simply leave

your answer expressed as an integral, and if numerical results are needed,
you can use numerical integration. Thus, the only way to describe the
2
solution to the equation y 0 = et is to express the answer as
Z
2
y(t) = et dt.

The indefinite integral notation we have used here has the constant of
integration implicitly included. One can be more precise by using a definite
integral notation, as in the Fundamental Theorem of Calculus. With this
notation, Z t
2
y(t) = eu du + c, y(t0 ) = c.
t0

We now extend the solution of y 0 = f (t) by antiderivatives to the case of

a general separable equation y 0 = h(t)g(y), and we provide an algorithm for
solving this equation.
Suppose y(t) is a solution on an interval I of Equation (2), which we write
in the form
1 0
y = h(t),
g(y)
1
and let Q(y) be an antiderivative of as a function of y, i.e., Q0 (y) =
g(y)
dQ 1
= and let H be an antiderivative of h. It follows from the chain rule
dy g(y)
that
d 1
Q(y(t)) = Q0 (y(t))y 0 (t) = y 0 (t) = h(t) = H 0 (t).
dt g(y(t))
This equation can be written as
d
(Q(y(t)) − H(t)) = 0.
dt
Since a function with derivative equal to zero on an interval is a constant, it
follows that the solution y(t) is implicitly given by the formula

Q(y(t)) = H(t) + c. (4)

Conversely, assume that y(t) is any function which satisfies the implicit
equation (4). Differentiation of both sides of Equation (4) gives, (again by the
chain rule),
d 1
h(t) = H 0 (t) = (Q(y(t))) = Q0 (y(t))y 0 (t) = y 0 (t).
dt g(y(t))

Hence y(t) is a solution of Equation (2).

40 2 First Order Differential Equations

Note that the analysis in the previous two paragraphs is valid as long as
1
h(t) and q(y) = have antiderivatives. From the Fundamental Theorem
g(y)
of Calculus, we know that a sufficient condition for this to occur is that h and
q are continuous functions, and q will be continuous as long as g is continuous
and g(y) 6= 0. We can thus summarize our results in the following theorem.

Theorem 2. Let g be continuous on the interval J = {y : c ≤ y ≤ d} and let

h be continuous on the interval I = {t : a ≤ t ≤ b}. Let H be an antiderivative
1
of h on I, and let Q be an antiderivative of on an interval J 0 ⊆ J for which
g
y0 ∈ J 0 and g(y0 ) 6= 0. Then y(t) is a solution to the initial value problem

y 0 = h(t)g(y); y(t0 ) = y0 (5)

if and only if y(t) is a solution of the implicit equation

Q(y(t)) = H(t) + c, (6)

where the constant c is chosen so that the initial condition is satisfied. More-
over, if y0 is a point for which g(y0 ) = 0, then the constant function y(t) ≡ y0
is a solution of Equation (5).
Proof. The only point not covered in the paragraphs preceding the theorem
is the case where g(y0 ) = 0. But if g(y0 ) = 0 and y(t) = y0 for all t, then

y 0 (t) = 0 = h(t)g(y0 ) = h(t)g(y(t))

for all t. Hence the constant function y(t) = y0 is a solution of Equation (5).
u
t

We summarize these observations in the following separable equation al-

gorithm.

Algorithm 3 (Separable Equation). To solve a separable differential equa-

tion, perform the following operations.
1. First put the equation in the form
dy
(I) y0 = = h(t)g(y),
dt
if it is not already in that form.
2. Then we separate variables in a form convenient for integration, i.e. we
formally write
1
(II) dy = h(t) dt.
g(y)
Equation (II) is known as the “differential" form of Equation (I).
 2.1 Separable Equations 41

3. Next we integrate both sides of Equation (II) (the left side with respect
to y and the right side with respect to t) and introduce a constant c, due
to the fact that antiderivatives coincide up to a constant. This yields
1
Z Z
(III) dy = h(t) dt + c.
g(y)
4. Now evaluate the antiderivatives and solve the resulting implicit equation
for y as a function of t, if you can (this won’t always be possible).
5. Additionally, the numbers y0 with g(y0 ) = 0 will give constant solutions
y(t) ≡ y0 that will not be seen from the general algorithm. u
t
t
Example 4. Find the solutions of the differential equation y 0 = .
y
I Solution. We first rewrite the equation in the form
dy t
(I) =
dt y
and then in differential form as

(II) y dy = t dt.

Integration of both sides of Equation (II) gives

Z Z
(III) y dy = t dt + c

or
1 2 1
y = t2 + c.
2 2
Multiplying by 2 we get y 2 = t2 + c, where we write c instead of 2c since twice
an arbitrary constant c is still an arbitrary constant. Thus, if a function y(t)
satisfies the differential equation yy 0 = t, then
p
y(t) = ± t2 + c (∗)

for some constant c ∈ R. On the other hand, since all functions of the form
(∗) solve yy 0 = t, it follows that the solutions are given by (∗). Figure 2.1
shows several of the curves y 2 = t2 + c which implicitly define the solutions
of yy 0 = t.√Note that each of the curves in the upper half plane is the graph
of y(t) = t2 + c for√ some c, while each curve in the lower half plane is the
graph of y(t) = − t2 + c. None of the solutions are defined on the t-axis,
i.e., when y = 0. Notice that each of the solutions is an arm of the hyperbola
y 2 − t2 = c. J

Example 5. Solve the differential equation y 0 = ky where k ∈ R is a con-

stant.
42 2 First Order Differential Equations

5 5 10 20 5
−5 20 15
15 10 −5

0
−1

−1
0 10 0

0
5 5

5
−1
−15

−5

−5
0 0

−10
−10
0
y

−2015
−1 −1 −2

−
0 0

−5
−5
0 5

5 5
10

−1
15

0
10
0 20 10 50
−5 5 15 20
−5 0 5
t
Fig. 2.1. The solutions of yy 0 = t are the level curves of y 2 = t2 + c. The constant
c is labeled on each curve.

I Solution. First note that the constant function y = 0 is one solution.

y0
When y 6= 0 we rewrite the equation in the form = k, which in differential
y
form becomes
1
dy = k dt.
y
Integrating both sides of this equation (the left side with respect to y and
right side with respect to t) gives

ln |y| = kt + c. (†)

Applying the exponential function to both sides of (†), and recalling that
eln x = x for all x > 0, we see that

|y| = eln|y| = ekt+c = ec ekt ,

so that
y = ±ec ekt . (‡)
Since c is an arbitrary constant, e is an arbitrary positive constant, so ±ec
c

is an arbitrary nonzero constant, which (as usual) we will continue to denote

by c. Thus we can rewrite Equation (‡) as

y = cekt . (7)
 2.1 Separable Equations 43

Letting c = 0 will give the solution y = 0 of y 0 = ky. Thus, as c varies over R,

Equation (7) describes all solutions of the differential equation y 0 = ky. Note
that c = y(0) is the initial value of y. Hence, the solution of the initial value
problem y 0 = ky, y(0) = y0 is

y(t) = y0 ekt . (8)

Figure 2.1 illustrates a few solution curves for this equation. J

k > 0; C > 0 k < 0; C > 0

0 0

k > 0; C < 0 k < 0; C > 0

0 0
Fig. 2.2. Some solutions of y 0 = ky for various y(0) = c. The left picture is for
k > 0, the right for k < 0.

A concrete example of the equation y 0 = ky is given by radioactive decay.

Example 6 (Radioactive Decay). Suppose that a quantity of a ra-
dioactive substance originally weighing y0 grams decomposes at a rate propor-
tional to the amount present and that half the quantity is left after a years
(a is the so-called half-life of the substance). Find the amount y(t) of the
substance remaining after t years. In particular, find the number of years it
takes such that 1/n-th of the original quantity is left.
I Solution. Since the rate of change y 0 (t) is proportional to the amount
y(t) present, we are led to the initial value problem
y 0 = −ky , y(0) = y0 ,
with solution y(t) = y0 e−kt , where k is a positive constant yet to be deter-
mined (the minus sign reflects the observation that y(t) is decreasing as t is
y0 ln 2
increasing). Since y(a) = = e−ka , it follows that k = . Thus,
2 a
t
y(t) = y0 2− a .
a ln n
This yields easily t = ln 2 as the answer to the last question by solving
t
y0 2− a = yn0 for t. J
44 2 First Order Differential Equations

Example 7. Solve the differential equation (2t−1)(y 2 −1)y 0 +t−y −1+ty =

0.

I Solution. To separate the variables in this equation we bring all terms

not containing y 0 to the right hand side of the equation, so that

(2t − 1)(y 2 − 1)y 0 = −t + y + 1 − ty = −t(1 + y) + 1 + y = (1 + y)(1 − t).

This variables can now be separated, yielding

y2 − 1 0 1−t
y = .
1+y 2t − 1
Before further simplification, observe that the constant function y(t) = −1 is
a solution of the original problem. If we now consider a solution other than
y(t) = −1, the equation can be written in differential form (after expanding
the right hand side in a partial fraction) as
 
1 1 1
(y − 1) dy = − + dt.
2 2 2t − 1

Integrating both sides of the equation gives, 12 y 2 − y = − 2t + 14 ln |2t − 1| + c.

Solving for y (and renaming the constant several times) we obtain the general
solution as either y(t) = −1 or
r
1
y(t) = 1 ± c − t + ln |2t − 1|.
2
J

Example 8. Solve the equation p0 = r(m − p)p, known as the Verhulst

population equation, where r and m are positive constants.

I Solution. Since
 
1 1 1 1
= + ,
(m − p)p m p m−p
the equation can be written with separated variables in differential form as
 
1 1 1 1
dp = + dp = r dt,
(m − p)p m p m−p
and the differential form is integrated to give
1
(ln |p| − ln |m − p|) = rt + c,
m
where c is an arbitrary constant of integration. Multiplying by m and renaming
mc as c (to denote an arbitrary constant) we get
 2.1 Separable Equations 45
 
 p 
ln 
  = rmt + c,
m − p

and applying the exponential function to both sides of the equation gives
 
 p  rmt+c
m − p = e
  = ec ermt ,

or
p
= ±ec ermt .
m−p
Since c is an arbitrary real constant, it follows that ±ec is an arbitrary real
nonzero constant, which we will again denote by c. Thus, we see that p satisfies
the equation
p
= cermt .
m−p
Solving this equation for p, we find that the general solution of the Verhulst
population equation is given by

cmermt
p(t) = . (9)
1 + cermt
Multiplying the numerator and denominator by e−rmt , we may rewrite Equa-
tion (9) in the equivalent form
cm
p(t) = . (10)
c + e−rmt
Some observations concerning this equation:
1. The constant solution p(t) = 0 is obtained by setting c = 0 in Equation
(10), even though c = 0 did not occur in our derivation.
2. The constant solution p(t) = m does not occur for any choice of c, so this
solution is an extra one.
3. Note that
cm
lim p(t) = = m,
t→∞ c
independent of c 6= 0. What this means is that if we start with a posi-
tive population, then over time, the population will approach a maximum
(sustainable) population m.
4. Figure 2.1 shows the solution of the Verhulst population equation y 0 =
y(3 − y) with initial population y(0) = 1. You can see from the graph
that y(t) approaches the limiting population 3 as t grows. It appears that
y(t) actually equals 3 after some point, but this is not true. It is simply
a reflection of the fact that y(t) and 3 are so close together that the lines
on a graph cannot distinguish them.
J
46 2 First Order Differential Equations

y ’ = y (3 − y)

3.5

2.5

1.5
y

0.5

−0.5

−3 −2 −1 0 1 2 3
t

Fig. 2.3. Solution of the population problem y 0 = y(3 − y), y(0) = 1

Exercises
In each of the following problems determine whether or not the equation is separable.
Do not solve the equations!!

1. y 0 = 2y(5 − y)

2. t2 y 0 = 1 − 2ty

3. yy 0 = 1 − y
y0
4. =y−t
y
5. ty 0 = y − 2ty

6. (t2 + 3y 2 )y 0 = −2ty

7. y 0 = ty 2 − y 2 + t − 1

8. y 0 = t2 + y 2

9. et y 0 = y 3 − y
 2.1 Separable Equations 47

Find the general solution of each of the following differential equations. If an

initial condition is given, find the particular solution which satisfies this initial con-
dition.

10. yy 0 = t, y(2) = −1.

I Solution. The variables are already separated, so integrate both sides of
1 1
the equation to get y 2 = t2 + c, which we can rewrite as y 2 − t2 = k where
2 2
k ∈ R is a constant. Since y(2) = −1, it follows that k = (−1)2 − 22 = −3 so
the solution is given implicitly
√ by the equation y 2 − t2 = −3 or we can solve
explicitly to get y = − t − 3, where the negative square root is used since
2

y(2) = −1 < 0. J
11. (1 − y 2 ) − tyy 0 = 0
I Solution. It is first necessary to separate the variables by rewriting the
equation as tyy 0 = (1 − y 2 ). This gives an equation
y 1
y0 = ,
1 − y2 t
or in the language of differentials:
y 1
dy = dt.
1 − y2 t
Integrating both sides of this equation gives
1
− ln |1 − y 2 | = ln |t| + c.
2
Multiplying by −2, and taking the exponential of both sides gives an equation
|1 − y 2 | = ±kt−2 where k is a positive constant. By considering an arbitrary
constant (which we will call c), this can be written as an implicit equation
t2 (1 − y 2 ) = c. J
12. y3y0 = t
13. y4y0 = t + 2
14. y 0 = ty 2
15. y 0 = t2 y 2
16. y 0 + (tan t)y = tan t, − π2 < t < π2
17. y 0 = tm y n , where m and n are positive integers, n 6= 1.
18. y 0 = 4y − y 2
19. yy 0 = y 2 + 1
20. y0 = y2 + 1
21. tyy 0 + t2 + 1 = 0
22. y + 1 + (y − 1)(1 + t2 )y 0 = 0
23. 2yy 0 = et
24. (1 − t)y 0 = y 2
25. ty − (t + 2)y 0 = 0

Solve the following initial value problems:

dy
26. dt
− y = y2, y(0) = 0.
48 2 First Order Differential Equations

27. y 0 = 4ty 2 , y(1) = 0

dy
28. dx = xy+2yx
, y(1) = e
29. y + 2yt = 0, y(0) = 4
0

30. y 0 = cott y , y(1) = π4

(u2 +1) dy
31. y du
= u, y(0) = 2

2.2 Linear First Order Equations

A linear first order differential equation is an equation of the form

y 0 + p(t)y = f (t). (1)

The primary objects of study in the current section are the linear first or-
der differential equations where the coefficient function p and the forcing
function f are continuous functions from an interval I into R. In some exer-
cises and in some later sections of the text, we shall have occasion to consider
linear first order differential equations in which the forcing function f is not
necessarily continuous, but for now we restrict ourselves to the case where
both p and f are continuous. Equation (1) is homogeneous if no forcing
function is present; i.e., if f (t) = 0 for all t ∈ I; the equation is inhomo-
geneous if the forcing function f is not 0, i.e., if f (t) 6= 0 for some t ∈ I.
Equation (1) is constant coefficient provided the coefficient function p is a
constant function, i.e., p(t) = p0 ∈ R for all t ∈ I.
Example 1. Consider the following list of first order differential equations.
1. y 0 = y − t
2. y 0 + ty = 0
3. y 0 = f (t)
4. y 0 + y 2 = t
5. ty 0 + y = t2
6. y 0 − 3t y = t4
7. y 0 = 7y
All of these equations except for y 0 + y 2 = t are linear. The presence of the
y 2 term prevents this equation from being linear. The second and the last
equation are homogeneous, while the first, third, fifth and sixth equations are
inhomogeneous. The first, third, and last equation are constant coefficient,
with p(t) = −1, p(t) = 0, and p(t) = −7 respectively. For the fifth and sixth
equations, the interval I on which the coefficient function p(t) and forcing
function f (t) are continuous can be either (−∞, 0) or (0, ∞). In both of these
cases, p(t) = 1/t or p(t) = −3/t fails to be continuous at t = 0. For the
first, second, and last equations, the interval I is all of R, while for the third
equation I is any interval on which the forcing function f (t) is continuous.
Note that only the second, third and last equations are separable. u
t
 2.2 Linear First Order Equations 49

Remark 2. Notice that Equation (1), which is the traditional way to express
a linear first order differential equation, is not in the standard form of Equation
(5). In standard form, Equation (1) becomes

y 0 = −p(t)y + f (t), (2)

so that the function F (t, y) of Equation (5) is F (t, y) = −p(t)y + f (t). The
standard form of the equation is useful for expressing the hypotheses which
will be used in the existence and uniqueness results of Section 2.3, while the
form given by Equation (1) is particularly useful for describing the solution
algorithm to be presented in this section. From Equation (2) one sees that
if a first order linear equation is homogeneous (i.e. f (t) = 0 for all t), then
the equation is separable (the right hand side is −p(t)y) and the technique of
the previous section applies, while if neither p(t) nor f (t) is the zero function,
then Equation (2) is not separable, and hence the technique of the previous
section is not applicable.

We will describe an algorithm for finding all solutions to the linear differ-
ential equation
y 0 + p(t)y = f (t)
which is based on first knowing how to solve homogeneous linear equations
(i.e., f (t) = 0 for all t). But, as we observed above, the homogeneous linear
equation is separable, and hence we know how to solve it.

Homogeneous Linear Equation: y 0 = h(t)y

Since the equation y 0 = h(t)y is separable, we first separate the variables and
write the equation in differential form:
1
dy = h(t) dt. (∗)
y

If H(t) = h(t) dt is any antiderivative of h(t), then integration of both sides

R
of Equation (∗) gives
ln |y| = H(t) + c
where c is a constant of integration. Applying the exponential function to
both sides of this equation gives

|y| = eln|y| = eH(t)+c = ec eH(t) .

Since c is an arbitrary constant, ec is an arbitrary positive constant. Then

y = ± |y| = ±ec eH(t) where ±ec will be an arbitrary nonzero constant, which,
as usual we will continue to denote by c. Since the constantR function y(t) = 0
is also a solution to (∗), and we conclude that, if H(t) = h(t) dt, then the
general solution to y 0 = h(t)y is
50 2 First Order Differential Equations

y(t) = ceH(t) (3)

where c denotes any real number.

3
Example 3. Solve the equation y 0 = y on the interval (0, ∞).
t
3
I Solution. In this case h(t) = so that an antiderivative on the interval
t
(0, ∞) is
3
Z
H(t) = dt = 3 ln t = ln(t3 ).
t
3
Hence then general solution of y 0 = y is
t
3
y(t) = ceH(t) = celn(t )
= ct3 .

J
We can now use the homogeneous case to transform an arbitrary first
order linear differential equation into an equation which can be solved by
antidifferentiation. What results is an algorithmic procedure for determining
all solutions to the linear first order equation

y 0 + p(t)y = f (t). (†)

The key observation is that the left hand side of this equation looks almost
like the derivative of a product. Recall that if z(t) = µ(t)y(t), then

z 0 (t) = µ(t)y 0 (t) + µ0 (t)y(t). (‡)

Comparing this with Equation (†), we see that what is missing is the coefficient
µ(t) in front of y 0 (t). If we multiply Equation (†) by µ(t), we get an equation

µ(t)y 0 (t) + µ(t)p(t)y(t) = µ(t)f (t).

The left hand side of this equation agrees with the right hand side of (‡)
provided the multiplier function µ(t) is chosen so that the coefficients of y(t)
agree in both equations. That is, choose µ(t), if possible, so that

µ0 (t) = p(t)µ(t).

But this is a homogeneous linear first order differential equation, so by Equa-

tion (3) we may take µ(t) = eP (t) where P (t) is any antiderivative of p(t) on
the given interval I. The function µ(t) is known an an integrating factor
for the equation y 0 + p(t)y = f (t), since after multiplication by µ(t), the left
hand side becomes a derivative (µ(t)y)0 and the equation itself becomes

(µ(t)y)0 = µ(t)f (t),

 2.2 Linear First Order Equations 51

which
R 0 is an equation that can be solved by integration. Recalling that
g (t) dt = g(t) + c, we see that integrating the above equation gives
Z
µ(t)y(t) = µ(t)f (t) dt.

Putting together all of our steps, we arrive at the following theorem de-
scribing all the solutions of a first order linear differential equation. The proof
is nothing more than an explicit codification of the steps delineated above into
an algorithm to follow.
Theorem 4. Let p(t), f (t) be continuous functions on an interval I. A
function y(t) is a solution of of the first order linear differential equation
y 0 + p(t)y = f (t) (Equation (1)) on I if and only if
Z
y(t) = ce−P (t) + e−P (t) eP (t) f (t) dt (4)

for all t ∈ I, where c ∈ R, and P (t) is some antiderivative of p(t) on the

interval I.
Proof. Let y(t) = ce−P (t) + e−P (t) eP (t) f (t) dt. Since P 0 (t) = p(t) and
R
d R P (t)
e f (t) dt = eP (t) f (t) (this is what it means to be an antiderivative
dt
of eP (t) f (t)) we obtain
Z
y 0 (t) = −cp(t)e−P (t) − p(t)e−P (t) eP (t) f (t) dt + e−P (t) eP (t) f (t)
 Z 
−P (t) −P (t) P (t)
= −p(t) ce +e e f (t) dt + f (t)

= −p(t)y(t) + f (t)
for all t ∈ I. This shows that every function of the form (4) is a solution of
Equation (1). Next we show that any solution of Equation (1) has a representa-
tion in the form of Equation (4). This is essentially what we have already done
in the paragraphs prior to the statement of the theorem. What we shall do
now is summarize the steps to be taken to implement this algorithm. Let y(t)
be a solution of Equation (1) on the interval I. Then we perform the follow-
ing step-by-step procedure, which will be crucial when dealing with concrete
examples.
Algorithm 5 (Solution of First Order Linear Equations). Follow the
following procedure to put any solution y(t) of Equation (1) into the form
given by Equation (4).
1. Compute an antiderivative P (t) = p(t) dt and multiply the equation
R
y 0 + p(t)y = f (t) by the integrating factor µ(t) = eP (t) . This yields
(I) eP (t) y 0 (t) + p(t)eP (t) y(t) = eP (t) f (t).
52 2 First Order Differential Equations

2. The function µ(t) = eP (t) is an integrating factor (see the paragraphs

prior to the theorem) which means that the left hand side of Equation (I)
is a perfect derivative, namely (µ(t)y(t))0 . Hence, Equation (I) becomes

d
(II) (µ(t)y(t)) = eP (t) f (t).
dt
3. Now we take an antiderivative of both sides and observe that they must
coincide up to a constant c ∈ R. This yields
Z
(III) eP (t) y(t) = eP (t) f (t) dt + c.

4. Finally, multiply by µ(t)−1 = e−P (t) to get that y(t) is of the form
Z
−P (t) −P (t)
(IV) y(t) = ce +e eP (t) f (t) dt.

This shows that any solution of Equation (1) is of the form given by
Equation (4), and moreover, the steps of Algorithm 5 tell one precisely how
to find this form. u
t
u
t

Remark 6. You should not memorize formula (4). What you should remem-
ber instead is the sequence of steps in Algorithm 5, and apply these steps to
each concretely presented linear first order differential equation (given in the
form of Equation (1)). To summarize the algorithm in words:
1. Find an integrating factor µ(t).
2. Multiply the equation by µ(t), insuring that the left hand side of the
equation is a perfect derivative.
3. Integrate both sides of the resulting equation.
4. Divide by µ(t) to give the solution y(t).

Example 7. Find all solutions of the differential equation t2 y 0 + ty = 1 on

the interval (0, ∞).

I Solution. Clearly, you could bring the equation into the standard form
of Equation (1), that is
1 1
y0 + y = 2 ,
t t
1 1
identify p(t) = and f (t) = 2 , compute an antiderivative P (t) = ln(t) of
t t
p(t) on the interval (0, ∞), plug everything into formula (4), and then compute
the resulting integral. This is a completely valid procedure if you are good in
memorizing formulas. Since we are not good at memorization, we prefer go
through the steps of Algorithm 5 explicitly.
First bring the differential equation into the standard form
 2.2 Linear First Order Equations 53

1 1
y0 + y = 2 .
t t
Then compute an antiderivative P (t) of the function in front of y and multiply
the equation by the integrating factor µ(t) = eP (t) . In our example, we take
P (t) = ln(t) and multiply the equation by µ(t) = eP (t) = eln(t) = t (we could
also take P (t) = ln(t) + c for any constant c, but the computations are easiest
if we set the constant equal to zero). This yields
1
(I) ty 0 + y = .
t
d
Next observe that the left side of this equality is equal to (ty) (see Step 2
dt
of Algorithm 5). Thus,

d 1
(II) (ty) = .
dt t
Now take antiderivatives of both sides and observe that they must coincide
up to a constant c ∈ R. Thus,

(III) ty = ln(t) + c, or

1 1
(IV) y(t) = c + ln(t).
t t
1
Observe that yh (t) = c (c ∈ R) is the general solution of the homogeneous
t
1
equation t2 y 0 +ty = 0, and that yp (t) = ln(t) is a particular solution of t2 y 0 +
t
ty = 1. Thus, all solutions are given by y(t) = yh (t) + yp (t). As the following
remark shows, this holds for all linear first order differential equations. J

Remark 8. Analyzing the general solution y(t) = ce−P (t) +e−P (t) eP (s) f (s) ds,
R

we see that this general solution is the sum of two parts. Namely, yh (t) =
ce−P (t) which is the general solution of the homogeneous problem

y 0 + p(t)y = 0,

and yp (t) = e−P (t) eP (s) f (s) ds which is a particular, i.e., a single, solution
R
of the inhomogeneous problem

y 0 + p(t)y = f (t).

The homogeneous equation y 0 + p(t)y = 0 is known as the associated ho-

mogeneous equation of the linear equation y 0 + p(t)y = f (t). That is, the
right hand side of the general linear equation is replaced by 0 to get the asso-
ciated homogeneous equation. The relationship between the general solution
54 2 First Order Differential Equations

yg (t) of y 0 + p(t)y = f (t), a particular solution yp (t) of this equation, and the
general solution yh (t) of the associated homogeneous equation y 0 + p(t)y = 0,
is usually expressed as
yg (t) = yh (t) + yp (t). (5)
What this means is that every solution to y 0 + p(t)y = f (t) can be obtained
by starting with a single solution yp (t) and adding to that an appropriate
solution of y 0 + p(t)y = 0. The key observation is the following. Suppose that
y1 (t) and y2 (t) are any two solutions of y 0 + p(t)y = f (t). Then

(y2 − y1 )0 (t) + p(t)(y2 − y1 )(t) = (y20 (t) + p(t)y2 (t)) − (y10 (t) + p(t)y1 (t))
= f (t) − f (t)
= 0,

so that y2 (t) − y1 (t) is a solution of the associated homogeneous equation

y 0 + p(t)y = 0, and y2 (t) = y1 (t) + (y2 (t) − y1 (t)). Therefore, given a solution
y1 (t) of y 0 + p(t)y = f (t), any other solution y2 (t) is obtained from y1 (t) by
adding a solution (specifically y2 (t) − y1 (t)) of the associated homogeneous
equation y 0 + p(t)y = 0.
This observation is a general property of solutions of linear equations,
whether they are differential equations of first order (as above), differential
equations of higher order (to be studied in Chapter 6), linear algebraic equa-
tions, or linear equations L(y) = f in any vector space, which is the math-
ematical concept created to handle the features common to problems of lin-
earity.. Thus, the general solution set S = yg of any linear equation L(y) = f
is of the form
yg = S = L−1 (0) + yp = yh + yp ,
where L(yp ) = f and L−1 (0) = yh = {y : L(y) = 0}.

Corollary 9. Let p(t), f (t) be continuous on an interval I, t0 ∈ I, and y0 ∈

R. Then the unique solution of the initial value problem

y 0 + p(t)y = f (t), y(t0 ) = y0 (6)

is given by
Z t
y(t) = y0 e−P (t) + e−P (t) eP (u) f (u) du, (7)
t0
Rt
where P (t) = t0 p(u) du.

Proof. Since P (t) is an antiderivative of p(t), we see that y(t) has the form of
Equation (4), and hence Theorem 4 guarantees that y(t) is a solution of the
Rt
linear first order equation y 0 +p(t)y = f (t). Moreover, P (t0 ) = t00 p(u) du = 0,
and Z t0
y(t0 ) = y0 e−P (t0 ) + e−P (t0 ) eP (u) f (u) du = y0 ,
t0
 2.2 Linear First Order Equations 55

so that y(t) is a solution of the initial value problem (6). Suppose that y1 (t)
is any other solution of Equation (6). Then y2 (t) := y(t) − y1 (t) is a solution
of the associated homogeneous equation

y 0 + p(t)y = 0, y(t0 ) = 0.

It follows from Equation (3) that y2 (t) = ce−P̃ (t) for some constant c ∈ R and
an antiderivative P̃ (t) of p(t). Since y2 (t0 ) = 0 and e−P̃ (t0 ) 6= 0, it follows that
c = 0. Thus, y(t) − y1 (t) = y2 (t) = 0 for all t ∈ I. This shows that y1 (t) = y(t)
for all t ∈ I, and hence y(t) is the only solution of Equation (6). u
t

Example 10. Find the solution of the initial value problem y 0 = −ty +
t, y(2) = 7 on R.

I Solution. Again, you could bring the differential equation into the stan-
dard form
y 0 + ty = t,
identify p(t) = t and f (t) = t, compute the antiderivative
Z t
t2
P (t) = u du = −2
2 2

of p(t), plug everything into the formula (4), and then compute the integral
in (7) to get
Z t
−P (t) −P (t)
y(t) = y0 e +e eP (u) f (u) du
t0
Z t
−t2 −t2 u2
+2 +2 −2
= 7e 2 +e 2 ue 2 du.
2

However, we again prefer to follow the steps of the algorithm. First we

proceed as in Example 7 and find the general solution of

y 0 + ty = t.
2
To do so we multiply the equation by the integrating factor et /2
and obtain
2 2 2
et /2 0
y + tet /2
y = tet /2
.
2
Since the left side is the derivative of et /2
y, this reduces to
d  t2 /2  2
e y = tet /2 .
dt
2 2
Since et /2
is the antiderivative of tet /2
, it follows that
2 2 2
et /2
y(t) = et /2
+ c, or y(t) = ce−t /2
+ 1.
56 2 First Order Differential Equations

Finally, we determine the constant c such that y(2) = 7. This yields 7 =

ce−2 + 1 or c = 6e2 . Thus, the solution is given by
t2
y(t) = 6e− 2 +2 + 1.

Corollary 11. Let f (t) be a continuous function on an interval I and p ∈ R.

Then all solution of the first order, inhomogeneous, linear, constant coeffi-
cient differential equation
y 0 + py = f (t)
are given by Z
−pt
y(t) = ce + e−p(t−u) f (u) du.

Moreover, for any t0 , y0 ∈ R, the unique solution of the initial value problem

y 0 + py = f (t), y(t0 ) = y0

is given by Z t
−p(t−t0 )
y(t) = y0 e + e−p(t−u) f (u) du.
t0

Proof. The statements follow immediately from Corollary 9. u

t

Example 12. Find the solution of the initial value problem y 0 = −y +

4, y(0) = 8 on R.

I Solution. We write the equation as y 0 + y = 4 and apply Corollary 11.

This yields
Z t Z t
−t −(t−u) −t −t
eu du = 8e−t +4e−t et − 1 = 4e−t +4.
 
y(t) = 8e + 4e ds = 8e +4e
0 0

Example 13. Find the solution of the initial value problem y 0 + y =

1
1−t , y(0) = 0 on the interval (−∞, 1).
Rt 1 u
I Solution. By Corollary 11, y(t) = e−t 0 1−u e du. Since the function
1 u
e is not integrable in closed form on the interval (−∞, 1), we might be
1−u
tempted to stop at this point and say that we have solved the equation. While
this is a legitimate statement, the present representation of the solution is of
little practical use and a further detailed study is necessary if you are “really”
interested in the solution. Any further analysis (numerical calculations, qual-
itative analysis, etc.) would be based on what type of information you are
attempting to ascertain about the solution. J
 2.2 Linear First Order Equations 57

We can use our analysis of first order linear differential equations to solve
the mixing problem set up in Example 5. For convenience we restate the
problem.

Example 14. Consider a tank that contains 2000 gallons of water in which
10 lbs of salt are dissolved. Suppose that a water-salt mixture containing
0.1 lb/gal enters the tank at a rate of 2 gal/min, and assume that the well-
stirred mixture flows from the tank at the same rate of 2 gal/min. Find the
amount y(t) of salt (expressed in pounds) which is present in the tank at all
times t measured in minutes.

I Solution. In Example 5, it was determined that y(t) satisfies the initial

value problem
y 0 + (0.001)y = 0.2, y(0) = 10. (8)
This equation has an integrating factor µ(t) = e(0.001)t , so multiplying the
equation by µ(t) gives
 0
e(0.001)t y = (0.2)e(0.001)t .

Integration of this equation gives e(0.001)t y = 200e(0.001)t + c, or after solving

for y,
y(t) = 200 + ce−(0.001)t .
Setting t = 0 gives 10 = y(0) = 200 + c so that c = −190 and the final answer
is
y(t) = 200 − 190e−(0.001)t.
J

Next we consider a numerical example of the general mixing problem con-

sidered in Example 6

Example 15. A large tank contains 100 gal of brine in which 50 lb of salt
is dissolved. Brine containing 2 lb of salt per gallon runs into the tank at the
rate of 6 gal/min. The mixture, which is kept uniform by stirring, runs out of
the tank at the rate of 4 gal/min. Find the amount of salt in the tank at the
end of t minutes.

I Solution. Let y(t) denote the number of pounds of salt in the tank after
t minutes; note that the tank will contain 100 + (6 − 4)t gallons of brine at
this time. The concentration (number of pounds per gallon) will then be

y(t)
lb/gal.
100 + 2t
Instead of trying to find the amount (in pounds) of salt y(t) at time t directly,
we will follow the analysis of Example 6 and determine the rate of change
58 2 First Order Differential Equations

of y(t), i.e., y 0 (t). But the the change of y(t) at time t is governed by the
principle
y 0 (t) = input rate − output rate,
where all three rates have to be measured in the same unit, which we take to
be lb/min. Thus,

input rate = 2 lb/gal × 6 gal/min = 12 lb/min,

y(t) 4y(t)
output rate = lb/gal × 4 gal/min = lb/min.
100 + 2t 100 + 2t
This yields the initial value problem

4y(t)
y 0 (t) = 12 − , y(0) = 50
100 + 2t
which can be solved as in the previous examples. The solution is seen to be

15(105 )
y(t) = 2(100 + 2t) − .
(100 + 2t)2

After 50 min, for example, there will be 362.5 lb of salt in the tank and 200 gal
of brine. J

Exercises
Find the general solution of the given differential equation. If an initial condition
is given, find the particular solution which satisfies this initial condition. Examples
3, 7, and 10 are relevant examples to review, and detailed solutions of a few of the
exercises will be provided for you to study.

1. y 0 (t) + 3y(t) = et , y(0) = −2.

I Solution. This equation is already in standard form (Equation (3.1.1)) with

p(t) = 3. An antiderivative of p(t) is P (t) = 3 dt = 3t. If we multiply the
differential equation y 0 (t) + 3y(t) = et by P (t), we get the equation

e3t y 0 (t) + 3e3t y(t) = e4t ,

d
and the left hand side of this equation is a perfect derivative, namely, (e3t y(t)).
dt
Thus,
d 3t
(e y(t)) = e4t .
dt
Now take antiderivatives of both sides and observe that they must coincide up
to a constant c ∈ R. This gives
1 4t
e3t y(t) = e + c.
4
 2.2 Linear First Order Equations 59

Now, multiplying by e−3t gives

1 t
y(t) = e + ce−3t (∗)
4

for the general solution of the equation y 0 (t)+3y(t) = et . To choose the constant
c to satisfy the initial condition y(0) = −2, substitute t = 0 into Equation (*)
1 9
to get −2 = y(0) = + c (remember that e0 = 1). Hence c = − , and the
4 4
solution of the initial value problem is

1 t 9 −3t
y(t) = e − e .
4 4
J

2. (cos t)y 0 (t) + (sin t)y(t) = 1, y(0) = 5

I Solution. Divide the equation by cos t to put it in the standard form

y 0 (t) + (tan t)y(t) = sec t.

In this case p(t) = tan t and an antiderivative is P (t) = tan t dt = ln(sec t). (We
do not need | sec t| since we are working near t = 0 where sec t > 0.) Now multi-
ply the differential equation y 0 (t) + (tan t)y(t) = sec t by eP (t) = eln sec t = sec t
to get (sec t)y 0 (t) + (sec t tan t)y(t) = sec2 t, the left hand side of which is a
d
perfect derivative, namely ((sec t)y(t)). Thus
dt
d
((sec t)y(t)) = sec2 t
dt
and taking antiderivatives of both sides gives

(sec t)y(t) = tan t + c

where c ∈ R is a constant. Now multiply by cos t to eliminate the sec t in front

of y(t), and we get
y(t) = sin t + c cos t
for the general solution of the equation, and letting t = 0 gives 5 = y(0) =
sin 0 + c cos 0 = c so that the solution of the initial value problem is

y(t) = sin t + 5 cos t.

3. y 0 − 2y = e2t , y(0) = 4

4. y 0 − 2y = e−2t , y(0) = 4

5. ty 0 + y = et , y(1) = 0

6. ty 0 + y = e2t , y(1) = 0.
60 2 First Order Differential Equations

7. y 0 = (tan t)y + cos t

8. y 0 + ty = 1, y(0) = 1.

9. ty 0 + my = t ln(t), where m is a constant.

10. y 0 = − yt + cos(t2 )

11. t(t + 1)y 0 = 2 + y.

12. y 0 + ay = b, where a and b are constants.

13. y 0 + y cos t = cos t, y(0) = 1

2
14. y 0 − y = (t + 1)2
t+1
2 t+1
15. y 0 − y= , y(1) = −3
t t
16. y 0 + ay = e−at , where a is a constant.

17. y 0 + ay = ebt , where a and b are constants and b 6= −a.

18. y 0 + ay = tn e−at , where a is a constant.

19. y 0 = y tan t + sec t

20. ty 0 + 2y ln t = 4 ln t
n
21. y 0 − y = et tn
t
22. y 0 − y = te2t , y(0) = a

23. ty 0 + 3y = t2 , y(−1) = 2

24. t2 y 0 + 2ty = 1, y(2) = a

Before attempting the following exercises, you may find it helpful to review the
examples in Section 1.1 related to mixing problems.

25. A tank contains 10 gal of brine in which 2 lb of salt are dissolved. Brine con-
taining 1 lb of salt per gallon flows into the tank at the rate of 3 gal/min, and
the stirred mixture is drained off the tank at the rate of 4 gal/min. Find the
amount y(t) of salt in the tank at any time t.
26. A 100 gal tank initially contains 10 gal of fresh water. At time t = 0, a brine
solution containing .5 lb of salt per gallon is poured into the tank at the rate of 4
gal/min while the well-stirred mixture leaves the tank at the rate of 2 gal/min.
a) Find the time T it takes for the tank to overflow.
b) Find the amount of salt in the tank at time T .
c) If y(t) denotes the amount of salt present at time t, what is limt→∞ y(t)?
27. A tank contains 100 gal of brine made by dissolving 80 lb of salt in water. Pure
water runs into the tank at the rate of 4 gal/min, and the mixture, which is
kept uniform by stirring, runs out at the same rate. Find the amount of salt in
the tank at any time t. Find the concentration of salt in the tank at any time t.
 2.3 Existence and Uniqueness 61

28. For this problem, our tank will be a lake and the brine solution will be polluted
water entering the lake. Thus assume that we have a lake with volume V which
is fed by a polluted river. Assume that the rate of water flowing into the lake
and the rate of water flowing out of the lake are equal. Call this rate r, let c be
the concentration of pollutant in the river as it flows into the lake, and assume
perfect mixing of the pollutant in the lake (this is, of course, a very unrealistic
assumption).
a) Write down and solve a differential equation for the amount P (t) of pol-
lutant in the lake at time t and determine the limiting concentration of
pollutant in the lake as t → ∞.
b) At time t = 0, the river is cleaned up, so no more pollutant flows into the
lake. Find expressions for how long it will take for the pollution in the lake
to be reduced to (i) 1/2 (ii) 1/10 of the value it had at the time of the
clean-up.
c) Assuming that Lake Erie has a volume V of 460 km3 and an inflow-outflow
rate of r = 175 km3 /year, give numerical values for the times found in Part
(b). Answer the same question for Lake Ontario, where it is assumed that
V = 1640 km3 and r = 209 km3 /year.
29. A 30 liter container initially contains 10 liters of pure water. A brine solution
containing 20 grams salt per liter flows into the container at a rate of 4 liters
per minute. The well stirred mixture is pumped out of the container at a rate
of 2 liters per minute.
a) How long does it take the container to overflow?
b) How much salt is in the tank at the moment the tank begins to overflow?
30. A tank holds 10 liters of pure water. A brine solution is poured into the tank at
a rate of 1 liter per minute and kept well stirred. The mixture leaves the tank
at the same rate. If the brine solution has a concentration of 1 kg salt per liter
what will the concentration be in the tank after 10 minutes.

2.3 Existence and Uniqueness

Unfortunately, only a few simple types of differential equations can be solved
explicitly in terms of well known elementary functions. In this section we will
describe the method of successive approximations, which provides one of
the many possible lines of attack for approximating solutions for arbitrary
differential equations. This method, which is quite different from what most
students have previously encountered, is the primary idea behind one of the
main theoretical results concerning existence and uniqueness of solutions of
the initial value problem

(∗) y 0 = F (t, y), y(t0 ) = y0 ,

where F (t, y) is a continuous function of (t, y) in the rectangle

R := {(t, y) : a ≤ t ≤ b , c ≤ y ≤ d}
62 2 First Order Differential Equations

and (t0 , y0 ) ∈ R. The key to the method of successive approximations is the

fact that a continuously differentiable function y(t) is a solution of (∗) if and
only if it is a solution of the integral equation
Z t
(∗∗) y(t) = y0 + F (u, y(u)) du.
t0

To see the equivalence of the initial value problem (∗) and the integral
equation (∗∗), we first integrate (∗) from t0 to t and obtain (∗∗). Con-
versely, if y(t) is a continuously differentiable solution of (∗∗), then y(t0 ) =
Rt
y0 + t00 F (u, y(u)) du = y0 . Moreover, since y(t) is a continuous function in t
and F (t, y) is a continuous function of (t, y), it follows that g(t) := F (t, y(t))
is a continuous function of t. Thus, by the Fundamental Theorem of Calculus,
 Z t   Z t 
0 d d
y (t) = y0 + F (u, y(u)) du = y0 + g(u) du = g(t) = F (t, y(t)),
dt t0 dt t0

which is what it means to be a solution of (∗).

To solve the integral equation (∗∗), mathematicians have developed a va-
riety of so-called “fixed point theorems”, each of which leads to an existence
and/or uniqueness result for solutions to the integral equation. One of the
oldest and most widely used existence and uniqueness theorems is due to
Émile Picard (1856-1941). Assuming that the function F (t, y) is sufficiently
“nice”, he first employed the method of successive approximations to prove
the existence and uniqueness of solutions of (∗∗). The method of successive
approximations is an iterative procedure which begins with a crude approx-
imation of a solution and improves it using a step by step procedure which
brings us as close as we please to an exact and unique solution of (∗∗). The
algorithmic procedure follows.
Algorithm 1 (Picard Approximation). Perform the following sequence
of steps to produce an approximate solution to the integral equation (∗∗), and
hence to initial value problem (∗).
(i) A rough initial approximation to a solution of (∗∗) is given by the constant
function
y0 (t) := y0 .
(ii) Insert this initial approximation into the right hand side of equation (∗∗)
and obtain the first approximation
Z t
y1 (t) := y0 + F (u, y0 (u)) du.
t0

(iii)The next step is to generate the second approximation in the same way;
i.e., Z t
y2 (t) := y0 + F (u, y1 (u)) du.
t0
 2.3 Existence and Uniqueness 63

(iv)At the n-th stage of the process we have

Z t
yn (t) := y0 + F (u, yn−1 (u)) du,
t0

which is defined by substituting the previous approximation yn−1 (t) into

the right hand side of (∗∗).

It is one of Picard’s great contributions to mathematics that he showed that

the functions yn (t) converge to a unique, continuously differentiable solution
y(t) of (∗∗) (and thus of (∗)) if the function F (t, y) and its partial derivative
∂
Fy (t, y) := F (t, y) are continuous functions of (t, y) on the rectangle R.
∂y
1
Theorem 2 (Picard’s Existence and Uniqueness Theorem). Let
F (t, y) and Fy (t, y) be continuous functions of (t, y) on a rectangle

R = {(t, y) : a ≤ t ≤ b , c ≤ y ≤ d} .

If (t0 , y0 ) is an interior point of R, then there exists a unique solution y(t) of

(∗) y 0 = F (t, y) , y(t0 ) = y0 ,

on some interval [a0 , b0 ] with t0 ∈ [a0 , b0 ] ⊂ [a, b]. Moreover, the successive
approximations y0 (t) := y0 ,
Z t
yn (t) := y0 + F (u, yn−1 (u)) du,
t0

computed by Algorithm 1 converge towards y(t) on the interval [a0 , b0 ]. That

is, for all  > 0 there exists n0 such that the maximal distance between the
graph of the functions yn (t) and the graph of y(t) (for t ∈ [a0 , b0 ]) is less than
 for all n ≥ n0 .

If one only assumes that the function F (t, y) is continuous on the rectangle
R, but makes no assumptions about Fy (t, y), then Guiseppe Peano (1858-
1932) showed that the initial value problem (∗) still has a solution on some
interval I with t0 ∈ I ⊂ [a, b]. This statement is known as Peano’s Existence
Theorem.2 However, in this case the solutions are not necessarily unique
(see Example 5 below). Theorem 2 is called a local existence and uniqueness
theorem because it guarantees the existence of a unique solution in some
interval I ⊂ [a, b]. In contrast, the following important variant of Picard’s
theorem yields a unique solution on the whole interval [a, b].
1
A proof of this theorem can be found in G.F. Simmons’ book Differential Equa-
tions with Applications and Historical Notes, 2nd edition McGraw-Hill, 1991.
2
For a proof see, for example, A.N. Kolmogorov and S.V. Fomin, Introductory
Real Analysis, Chapter 3, Section 11, Dover 1975.
64 2 First Order Differential Equations

Theorem 3. Let F (t, y) be a continuous function of (t, y) that satisfies a

Lipschitz condition on a strip S = {(t, y) : a ≤ t ≤ b , −∞ < y < ∞}. That
is, assume that
|F (t, y1 ) − F (t, y2 )| ≤ K|y1 − y2 |
for some constant K > 0. If (t0 , y0 ) is an interior point of S, then there exists
a unique solution of

(∗) y 0 = F (t, y) , y(t0 ) = y0 ,

on the interval [a, b].

Example 4. Let us consider the Riccati equation y 0 = y 2 −t. Here, F (t, y) =

y 2 −t and Fy (t, y) = 2y are continuous on all of R2 . Thus, by Picard’s Theorem
2, the initial value problem

(∗) y0 = y2 − t , y(0) = 0

has a unique solution on some (finite or infinite) interval I containing 0. The

direction field for y 0 = y 2 − t (see Section 1.3, Example 3) suggests that the
maximal interval Imax on which the solution exists should be of the form
Imax = (a, ∞) for some −∞ ≤ a < −1. Observe that we can not apply
Theorem 3 since

|F (t, y1 ) − F (t, y2 )| = (y12 − t) − (y22 − t) = |y12 − y22 | = |y1 + y2 ||y1 − y2 |

 

can not be bounded by K|y1 −y2 | for some constant K > 0 because this would
imply that |y1 + y2 | ≤ K for all −∞ < y1 , y2 < ∞. Thus, without further
analysis of the problem, we have no precise knowledge about the maximal
domain of the solution; i.e., we do not know if and where the solution will
“blow up”.
Next we show how Picard’s method of successive approximations works in
this example. To use this method we rewrite the initial value problem (∗) as
an integral equation; i.e., we consider
Z t
(∗∗) y(t) = (y(u)2 − u) du.
0

We start with our initial approximation y0 (t) = 0, plug it into (∗∗) and obtain
our first approximation
Z t Z t
2 1
y1 (t) = (y0 (u) − u) du = − u du = − t2 .
0 0 2
The second iteration yields
Z t Z t 
1 4 1 5 1 2
y2 (t) = (y1 (u)2 − u) du = u − u du = t − t .
0 0 4 4 ·5 2
 2.3 Existence and Uniqueness 65

Since y2 (0) = 0 and

1 1 7 1 4 1 1 7
y2 (t)2 − t = t10 − t + t − t = 2 2 t10 − t + y20 (t) ≈ y20 (t)
42 ·5 2 4·5 4 4 ·5 4·5
if t is close to 0, it follows that the second iterate y2 (t) is already a “good”
approximation of the exact solution for t close to 0. Since y2 (t)2 = 421·52 t10 −
1 7 1 4
4·5 t + 4 t , it follows that
Z t 
1 1 7 1 4 1 1 1 5 1 2
y3 (t) = 2 2
u10 − u + u −u du = 2 2
t11 − t8 + t − t .
0 4 ·5 4·5 4 11 · 4 · 5 4·5·8 4·5 2

According to Picard’s theorem, the successive approximations yn (t) converge

towards the exact solution y(t), so we expect that y3 (t) is an even better
approximation of y(t) for t close enough to 0.

Example 5. Consider the initial value problem

(∗) y 0 = 3y 2/3 , y(t0 ) = y0 .

The function F (t, y) = y 2/3 is continuous for all (t, y), so Peano’s existence
theorem shows that the initial value problem (∗) has a solution for all −∞ <
2
t0 , y0 < ∞. Moreover, since Fy (t, y) = 1/3 , Picard’s existence and uniqueness
y
theorem tells us that the solutions of (∗) are unique as long as the initial value
y0 6= 0. Since the differential equation y 0 = 3y 2/3 is separable, we can rewrite
it the differential form
1
2/3
dy = 3dt,
y
and integrate the differential form to get

3y 1/3 = 3t + c.

Thus, the functions y(t) = (t + c)3 for t ∈ R, together with the constant
function y(t) = 0, are the solution curves for the differential equation y 0 =
1/3
3y 2/3 , and y(t) = (y0 + t − t0 )3 is the unique solution of the initial value
problem (∗) if y0 6= 0. If y0 = 0, then (∗) admits infinitely many solutions of
the form 
3
(t − α)
 if t < α
y(t) = 0 if α ≤ t ≤ β (1)
3

(t − β) if t > β,


where t0 ∈ [α, β]. The graph of one of these functions (where α = −1, β = 1)
is depicted in Figure 2.4. What changes among the different functions is the
length of the straight line segment joining α to β on the t-axis.
66 2 First Order Differential Equations

0
y

−2

−4

−6

−8
−3 −2 −1 0 1 2 3
t
Fig. 2.4. A solution (where α = −1, β = 1 in Equation 1) of y 0 = 3y 2/3 , y(0) = 0.

Example 6. The differential equation

(†) ty 0 = 3y

is separable (and linear). Thus, it is easy to see that y(t) = ct3 is its general
solution. In standard form Equation (†) is
3
(‡) y0 = y
t
3
and the right hand side, F (t, y) = y, is continuous provided t 6= 0. Thus
t
Picard’s theorem applies to give the conclusion that the initial value problem
3
y 0 = y , y(t0 ) = y0 has a unique local solution if t0 6= 0 (given by y(t) =
t
y0 3
t ). However, if t0 = 0, Picard’s theorem contains no information about
t30
the existence and uniqueness of solutions. Indeed, in its standard form (‡),
it is not meaningful to talk about solutions of this equation at t = 0 since
3
F (t, y) = y is not even defined for t = 0. But in the originally designated
t
form (†), where the t appears as multiplication on the left side of the equation,
then an initial value problem starting at t = 0 makes sense, and moreover,
the initial value problem

ty 0 = 3y , y(0) = 0
 2.3 Existence and Uniqueness 67

has infinitely many solutions of the form y(t) = ct3 for any c ∈ R, whereas
the initial value problem

ty 0 = 3y , y(0) = y0

has no solution if y0 6= 0. See Figure 2.5, where one can see that all of the
function y(t) = ct3 pass through the origin (i.e. y(0) = 0), but none pass
through any other point on the y-axis.

0
y

−1

−2

−3

−4
−2 −1 0 1 2
t
Fig. 2.5. Distinct solutions of the initial value problem ty 0 = 3y, y(0) = 0.

Remark 7 (Geometric meaning of uniqueness).

1. The theorem on existence and uniqueness of solutions of differential equa-

tions (Theorem 2) has a particularly useful geometric interpretation. Sup-
pose that y 0 = F (t, y) is a first order differential equation for which Pi-
card’s existence and uniqueness theorem applies. If y1 (t) and y2 (t) denote
two different solutions of y 0 = F (t, y), then the graphs of y1 (t) and y2 (t)
can never intersect. The reason for this is just that if (t0 , y0 ) is a point of
the plane which is common to both the graph of y1 (t) and that of y2 (t),
then both of these functions will satisfy the initial value problem

y 0 = F (t, y), y(t0 ) = y0 .

68 2 First Order Differential Equations

But if y1 (t) and y2 (t) are different functions, this will violate the unique-
ness provision of Picard’s theorem. Thus the situation depicted in Figures
2.4 and 2.5 where several solutions of the same differential equation go
through the same point (in this case (0, 0)) can never occur for a differen-
tial equation which satisfies the hypotheses of Theorem 2. Similarly, the
graphs of the function y1 (t) = (t + 1)2 and the constant function y2 (t) = 1
both pass through the point (0, 1), and thus both cannot be solutions of
the same differential equation satisfying Picard’s theorem.
2. The above remark can be exploited in the following way. The constant
function y1 (t) = 0 is a solution to the differential equation y 0 = y 3 + y
(check it). Since F (t, y) = y 3 +y clearly has continuous partial derivatives,
Picard’s theorem applies. Hence, if y2 (t) is a solution of the equation for
which y2 (0) = 1, the above observation takes the form of stating that
y2 (t) > 0 for all t. This is because, in order for y(t) to ever be negative,
it must first cross the t-axis, which is the graph of y1 (t), and we have
observed that two solutions of the same differential equation can never
cross. This observation will be further exploited in the next section.

Exercises
1. a) Find the exact solution of the initial value problem

(∗) y0 = y2, y(0) = 1.

b) Apply Picard’s method (Theorem 2) to calculate the first three approxima-

tions y1 (t), y2 (t), and y3 (t) to (∗) and compare these results with the exact
solution.

I Solution. (a) The equation is separable so separate the variables to get

y −2 dy = dt. Integrating gives −y −1 = t + c and the initial condition y(0) = 1
implies that the integration constant c = −1, so that the exact solution of (∗)
is
1
y(t) = = 1 + t + t2 + t3 + t4 + · · · ; |t| < 1.
1−t
(b) To apply Picard’s method, let y0 = 1 and define
t t
y1 (t) = 1 +
(y0 (s))2 ds = 1 +
ds = 1 + t;
0 0
t t
t3
y2 (t) = 1 +
(y1 (s))2 ds = 1 +
(1 + s)2 ds = 1 + t + t2 + ;
0 0 3
t t 2
s3
y3 (t) = 1 +
(y2 (s))2 ds =
1 + s + s2 + ds
0 0  3
t
8 5 2 1
= 1+
1 + 2s + 3s2 + s3 + s4 + s5 + s6 ds
0  3 3 3 9 
2 1 1 1
= 1 + t + t2 + t3 + t4 + t5 + t6 + t7 .
3 3 9 63
 2.3 Existence and Uniqueness 69

Comparing y3 (t) to the exact solution, we see that the series agree up to order
3. J
2. Apply Picard’s method to calculate the first three approximations y1 (t), y2 (t),
y3 (t) to the solution y(t) of the initial value problem
y 0 = t − y, y(0) = 1.
3. Apply Picard’s method to calculate the first three approximations y1 (t), y2 (t),
y3 (t) to the solution y(t) of the initial value problem
y0 = t + y2 , y(0) = 0.
Which of the following initial value problems are guaranteed a unique solution
by Picard’s theorem (Theorem 2)? Explain.
4. y 0 = 1 + y 2 , y(0) = 0
√
5. y 0 = y, y(1) = 0
√
6. y 0 = y, y(0) = 1
t−y
7. y 0 = , y(0) = −1
t+y
t−y
8. y 0 = , y(1) = −1
t+y
9. a) Find the general solution of the differential equation
(†) ty 0 = 2y − t.
Sketch several specific solutions from this general solution.
b) Show that there is no solution to (†) satisfying the initial condition y(0) = 2.
Why does this not contradict Theorem 2?
10. a) Let t0 , y0 be arbitrary and consider the initial value problem
y0 = y2, y(t0 ) = y0 .
Explain why Theorem 2 guarantees that this initial value problem has a
solution on some interval |t − t0 | ≤ h.
b) Since F (t, y) = y 2 and Fy (t, y) = 2y are continuous on all of the (t, y)−plane,
one might hope that the solutions are defined for all real numbers t. Show
that this is not the case by finding a solution of y 0 = y 2 which is defined
for all t ∈ R and another solution which is not defined for all t ∈ R. (Hint:
Find the solutions with (t0 , y0 ) = (0, 0) and (0, 1).)
11. Is it possible to find a function F (t, y) that is continuous and has a continuous
partial derivative Fy (t, y) such that the two functions y1 (t) = t and y2 (t) =
t2 − 2t are both solutions to y 0 = F (t, y) on an interval containing 0?
12. Show that the function
0, for t < 0
y1 (t) =  3
t for t ≥ 0
is a solution of the initial value problem ty 0 = 3y, y(0) = 0. Show that y2 (t) = 0
for all t is a second solution. Explain why this does not contradict Theorem 2.
70 2 First Order Differential Equations

2.4 Miscellaneous Nonlinear First Order Equations

We have learned how to find explicit solutions for the standard first order
differential equation
y 0 = F (t, y)
when the right hand side of the equation has one of the particularly simple
forms:
1. F (t, y) = h(t)g(y), i.e., the equation is separable, or
2. F (t, y) = −p(t)y + f (t), i.e., the equation is linear.
Unfortunately, in contrast to the separable and first order linear differential
equations, for an arbitrary function F (t, y) it is very difficult to find closed
form “solution formulas”. In fact, most differential equations do not have closed
form solutions and one has to resort to numerical or asymptotic approximation
methods to gain information about them. In this section we discuss some other
types of first-order equations which you may run across in applications and
that allow closed form solutions in the same sense as the separable and first or-
der linear differential equations. That is, the ”explicit" solution may very well
involve the computation of an indefinite integral which cannot be expressed
in terms of elementary functions, or the solution may be given implicitly by
an equation which cannot be reasonably solved in terms of elementary func-
tion. Our main purpose in this section is to demonstrate techniques that allow
us to find solutions of these types of first-order differential equations and we
completely disregard in this section questions of continuity, differentiability,
vanishing divisors, and so on. If you are interested in the huge literature cov-
ering other special types of first order differential equations for which closed
form solutions can be found, we refer you to books like [Zw] or to one of
the three M’s (Mathematica, Maple, or MatLab) which are, most likely, more
efficient in computing closed form solutions than most of us will ever be.

Exact Differential Equations

A particularly important class of nonlinear first order differential equations
that can be solved (explicitly or implicitly) is that of exact first order equa-
tions. To explain the mathematics behind exact equations, it is necessary to
recall some facts about calculus of functions of two variables.3 Let V (t, y) be
a function of two variables defined on a rectangle

R := {(t, y) : a ≤ t ≤ b , c ≤ y ≤ d} .

The curve with equation V (t, y) = c, where c ∈ R is a constant, is a level

curve of V .
3
The facts needed will be found in any calculus textbook. For example, you may
consult Chapter 14 of Calculus: Early Transcendentals, Fourth Edition by James
Stewart, Brooks-Cole, 1999.
 2.4 Miscellaneous Nonlinear First Order Equations 71

Example 1. 1. If V (t, y) = t + 2y, the level curves are all of the lines
t + 2y = c of slope −0.5.
2. If V (t, y) = t2 + y 2√, the level curves are the circles t2 + y 2 = c centered
at (0, 0) of radius c, provided c > 0. If c = 0, then the level “curve"
t2 + y 2 = 0 consists of the single point (0, 0), while if c < 0 there are no
points at all which solve the equation t2 + y 2 = c.
3. If V (t, y) = t2 −y 2 then the level curves of V are the hyperbolas t2 −y 2 = c
if c 6= 0, while the level curve t2 − y 2 = 0 consists of the two lines y = ±t.
4. If V (t, y) = y 2 − t then the level curves are the parabolas y 2 − t = c with
axis of symmetry the t-axis and opening to the right.

Thus we see that sometimes a level curve defines y explicitly as a function

of t (for example, y = 12 (c − t) in number 1 above), sometimes t is defined
explicitly as a function of y (for example, t = −2y + c in number 1, and
t = y 2 − c in number 3 above), while in other cases it may only be possible to
define y as a function of t (or t as a function of y) implicitly by the level curve
2 2
equation V (t, y) = c. For instance, the level
√ curve t + √y = c for√ c > 0 defines
y as a function of t in two ways (y = ± c − t p 2 for − c < t < c) and it also
√ √
defines t as a function of y in two ways (t = ± c − y 2 for − c < y < c).
If we are given a two variable function V (t, y) is there anything which can
be said about all of the level curves V (t, y) = c? The answer is yes. What the
level curves of a fixed two variable function have in common is that every one
of the functions y(t) defined implicitly by V (t, y) = c, no matter what c is, is
a solution of the same differential equation. The mathematics underlying this
observation is the chain rule in two variables, which implies that
d
V (t, y(t)) = Vt (t, y(t)) + Vy (t, y(t))y 0 (t),
dt
where Vt , Vy denote the partial derivatives of V (t, y) with respect to t and y,
respectively. Thus, if a function y(t) is given implicitly by a level curve

V (t, y(t)) = c,

then y(t) satisfies the equation

d d
0= c = V (t, y(t)) = Vt (t, y(t)) + Vy (t, y(t))y 0 (t).
dt dt
This means that y(t) is a solution of the differential equation

Vt (t, y) + Vy (t, y)y 0 = 0. (1)

Notice that the constant c does not appear anywhere in this equation so that
every function y(t) determined implicitly by a level curve of V (t, y) satisfies
this same equation. An equation of the form given by Equation 1 is referred
to as an exact equation:
72 2 First Order Differential Equations

Definition 2. A differential equation written in the form

M (t, y) + N (t, y)y 0 = 0

is said to be exact if there is a function V (t, y) such that M (t, y) = Vt (t, y)

and N (t, y) = Vy (t, y).
What we observed above is that, if y(t) is defined implicitly by a level curve
V (t, y) = c, then y(t) is a solution of the exact equation 1. Moreover, the level
curves determine all of the solutions of Equation 1, so the general solution is
defined by
V (t, y) = c. (2)

Example 3. 1. The exact differential equation determined by V (t, y) =

t + 2y is
0 = Vt (t, y) + Vy (t, y)y 0 = 1 + 2y 0
so the general solution of 1 + 2y 0 = 0 is t + 2y = c.
2. The exact differential equation determined by V (t, y) = t2 + y 2 is

0 = Vt (t, y) + Vy (t, y)y 0 = 2t + 2yy 0 .

Hence, the general solution of the equation t + yy 0 = 0, which can be

written in standard form as y 0 = −t/y , is t2 + y 2 = c.

Suppose we are given a differential equation in the form

M (t, y) + N (t, y)y 0 = 0,

but we are not given apriori that M (t, y) = Vt (t, y) and N (t, y) = Vy (t, y). How
can we determine if there is such a function V (t, y), and if there is, how can
we find it? That is, is there a criterion for determining if a given differential
equation is exact, and if so is there a procedure for producing the function
V (t, y) whose level curves implicitly determine the solutions. The answer to
both questions is yes. The criterion for exactness is given by the following
theorem; the procedure for finding V (t, y) will be illustrated by example.

Theorem 4 (Criterion for exactness).

A first order differential equation

M (t, y) + N (t, y)y 0 = 0

in which M (t, y) and N (t, y) have continuous first order partial derivatives is
exact if and only if
My (t, y) = Nt (t, y) (3)

for all t, y in a square region of R2 .

 2.4 Miscellaneous Nonlinear First Order Equations 73

Proof. Recall (from your calculus course) that all functions V (t, y) whose
second partial derivatives exist and are continuous satisfy

(∗) Vty (t, y) = Vyt (t, y),

where Vty (t, y) denotes the derivative of Vt (t, y) with respect to y, and Vyt (t, y)
is the derivative of Vy (t, y) with respect to t. The equation (∗) is known as
Clairaut’s theorem (after Alexis Clairaut (1713 – 1765)) on the equality of
mixed partial derivatives. If the equation M (t, y) + N (t, y)y 0 = 0 is exact then
(by definition) there is a function V (t, y) such that Vt (t, y) = M (t, y) and
Vy (t, y) = N (t, y). Then by Clairaut’s theorem,

∂ ∂
My (t, y) = Vt (t, y) = Vty (t, y) = Vyt (t, y) = Vy (t, y) = Nt (t, y).
∂y ∂t
Hence condition 3 is satisfied.
Now assume, conversely, that condition 3 is satisfied. To verify that the
equation M (t, y) + N (t, y)y 0 = 0 is exact, we need to search for a function
V (t, y) which satisfies the equations

Vt (t, y) = M (t, y) and Vy (t, y) = N (t, y).

The procedure will be sketched and then illustrated by means of an example.

The equation Vt (t, y) = M (t, y) means that we should be able to recover
V (t, y) from M (t, y) by indefinite integration:
Z
V (t, y) = M (t, y) dt + ϕ(y). (4)

The function ϕ(y) appears as the “integration constant" since any function of
y goes to 0 when differentiated with respect to t. The function ϕ(y) can be
determined from the equation
∂
Z
Vy (t, y) = M (t, y) dt + ϕ0 (y) = N (t, y). (5)
∂y
That is
∂
Z
ϕ0 (y) = N (t, y) − M (t, y) dt. (6)
∂y
The verification that the function on the right is really a function only of y
(as it must be if it is to be ϕ0 (y)) is where condition 3 is needed. u
t
t−y
Example 5. Solve the differential equation y 0 =
t+y
I Solution. We rewrite the equation in the form y − t + (t + y)y 0 = 0 to
get that M (t, y) = y − t and N (t, y) = y + t. Since My (t, y) = 1 = Nt (t, y), it
follows that the equation is exact and the general solution will have the form
74 2 First Order Differential Equations

V (t, y) = c, where Vt (t, y) = y − t and Vy (t, y) = y + t. Since Vt (t, y) = y − t

it follows that
t2
Z
V (t, y) = (y − t) dt + ϕ(y) = yt − + ϕ(y),
2
where ϕ(y) is a yet to be determined function depending on y, but not on t.
To determine ϕ(y) note that y + t = Vy (t, y) = t + ϕ0 (y), so that ϕ0 (y) = y.
Hence
y2
ϕ(y) = + c1
2
for some arbitrary constant c1 , and thus

t2 y2
V (t, y) = yt − + + c1 .
2 2
t−y
The general solution of y 0 = is therefore given by the implicit equation
t+y

t2 y2
V (t, y) = yt − + + c1 = c.
2 2
This is the form of the solution which we are led to by our general solution
procedure outlined in the proof of Theorem 4. However, after further simpli-
fying this equation and renaming constants several times the general solution
can be expressed implicitly by

2yt − t2 + y 2 = c,

and explicitly by p
y(t) = −t ± 2t2 + c.
J

What happens if we try to solve by equation M (t, y) + N (t, y)y 0 = 0 by

the procedure outlined above without first verifying that it is exact? If the
equation is not exact, you will discover this fact when you get to Equation 6,
since ϕ0 (y) will not be a function only of y, as the following example illustrates.

Example 6. Try to solve the equation (t − 3y) + (2t + y)y 0 = 0 by the

solution procedure for exact equations.

I Solution. Note that M (t, y) = t − 3y and N (t, y) = 2t + y. First apply

Equation 4 to get

t2
Z Z
(†) V (t, y) = M (t, y), dt = (t − 3y) dt = − 3ty + ϕ(y),
2
and then determine ϕ(y) from Equation 6:
 2.4 Miscellaneous Nonlinear First Order Equations 75

(‡)
t2
 
∂ ∂
Z
ϕ0 (y) = N (t, y) − M (t, y) dt = (2t + y) − − 3ty + ϕ(y) = y − t.
∂y ∂y 2
But we see that there is a problem since ϕ0 (y) in (‡) involves both y and
t. This is where it becomes obvious that you are not dealing with an exact
equation, and you cannot proceed with this procedure. Indeed, My (t, y) =
−3 6= 2 = Nt (t, y), so that this equation fails the exactness criterion 3. J

Bernoulli Equations
It is sometimes possible to change the variables in a differential equation
y 0 = F (t, y) so that in the new variables the equation appears in a form you
already know how to solve. This is reminiscent of the substitution procedure
for computing integrals. We will illustrate the procedure with a class of equa-
tions known as Bernoulli equations (named after Jakoub Bernoulli, (1654
– 1705)), which are equations of the form

y 0 + p(t)y = f (t)y n . (7)

If n = 0 this equation is linear, while if n = 1 the equation is both separable

and linear. Thus, it is the cases n 6= 0, 1 where a new technique is needed.
Start by dividing Equation 7 by y n to get

(∗) y −n y 0 + p(t)y 1−n = f (t),

and notice that if we introduce a new variable z = y 1−n , then the chain rule
gives
dz dz dy
z0 = = = (1 − n)y −n y 0 ,
dt dy dt
and Equation (∗), after multiplying by the constant (1 − n), becomes a linear
first order differential equation in the variables t, z:

(∗∗) z 0 + (1 − n)p(t)z = f (t).

Equation (∗∗) can then be solved by Algorithm 5, and the solution to 7 is

obtained by solving z = y 1−n for y.
Example 7. Solve the Bernoulli equation y 0 + y = y 2 .
I Solution. In this equation n = 2, so if we let z = y 1−2 = y −1 , we get
z 0 = −y −2 y 0 . After dividing our equation by y 2 we get y −2 y 0 + y −1 = 1, which
in terms of the variable z is −z 0 + z = 1. In the standard form for linear
equations this becomes
z 0 − z = −1.
We can apply Algorithm 5 to this equation. The integrating factor will be e− t.
Multiplying by the integrating factor gives (e−1 z)0 = −e−t so that e−t z =
76 2 First Order Differential Equations

e−t + c. Hence z = 1 + cet . Now go back to the original function y by solving

z = y −1 for y. Thus
1
y = z −1 = (1 + cet )−1 =
1 + cet
is the general solution of the Bernoulli equation y 0 + y = y 2 .
Note that this equation is also a separable equation, so it could have been
solved by the technique for separable equations, but the integration (and
subsequent algebra) involved in the current procedure is simpler. J
There are a number of other types of substitutions which are used to
transform certain differential equations into a form which is more amenable
for solution. We will not pursue the topic further in this text. See the book
[Zw] for a collection of many different solution algorithms.

Exercises
Exact Equations
For Exercises 1 – 9, determine if the equation is exact, and if it is exact, find the
general solution.
1. (y 2 + 2t) + 2tyy 0 = 0
I Solution. This can be written in the form M (t, y) + N (t, y)y 0 = 0 where
M (t, y) = y 2 +2t and N (t, y) = 2ty. Since My (t, y) = 2y = Nt (t, y), the equation
is exact (see Equation (3.2.2)), and the general solution is given implicitly by
F (t, y) = c where the function F (t, y) is determined by Ft (t, y) = M (t, y) = y 2 +
2t and Fy (t, y) = N (t, y) = 2ty. These equations imply that F (t, y) = t2 + ty 2
will work so the solutions are given implicitly by t2 + ty 2 = c. J

2. y − t + ty 0 + 2yy 0 = 0

3. 2t2 − y + (t + y 2 )y 0 = 0

4. y 2 + 2tyy 0 + 3t2 = 0

5. (3y − 5t) + 2yy 0 − ty 0 = 0

6. 2ty + (t2 + 3y 2 )y 0 = 0, y(1) = 1

7. 2ty + 2t2 + (t2 − y)y 0 = 0

8. t2 − y − ty 0 = 0

9. (y 3 − t)y 0 = y

10. Find conditions on the constants a, b, c, d which guarantee that the differential
equation (at + by) = (ct + dy)y 0 is exact.
Bernoulli Equations. Find the general solution of each of the following
Bernoulli equations. If an initial value is given, also solve the initial value prob-
lem.
 2.4 Miscellaneous Nonlinear First Order Equations 77

11. y 0 − y = ty 2 , y(0) = 1

12. y 0 + ty = t3 y 3

13. (1 − t2 )y 0 − ty = 5ty 2

14. y 0 + ty = ty 3

15. y 0 + y = ty 3
General Equations. The following problems may any of the types studied so
far.

16. y 0 = ty − t, y(1) = 2

17. (t2 + 3y 2 )y 0 = −2ty

√
18. t(t + 1)y 0 = 2 y
y
19. y 0 =
t2 + 2t − 3
20. sin y + y cos t + 2t + (t cos y + sin t)y 0 = 0
1
21. y 0 + y =t−1
t(t − 1)

22. y 0 − y = 21 et y −1 , y(0) = −1

8t2 − 2y
23. y 0 =
t
y2
24. y 0 = , y(1) = 1
t
3
The Laplace Transform

In this chapter we introduce the Laplace Transform and show how it gives
a direct method for solving certain initial value problems. This technique is
extremely important in applications since it gives an easily codified procedure
that goes directly to the solution of an initial value problem without first
determining the general solution of the differential equation. The same theo-
retical procedure applies to ordinary differential equations of arbitrary order
(with constant coefficients) and even to systems of constant coefficient linear
ordinary differential equations, which will be treated in Chapter 10. Moreover
the same procedure applies to linear constant coefficient equations (of any
order) for which the forcing function is not necessarily continuous. This will
be addressed in Chapter 8.
You are already familiar with certain operators which transform one func-
tion into another. One particularly important example is the differentiation
operator D which transforms each function which has a derivative into its
derivative, i.e., D(f ) = f 0 . The Laplace transform L is an integral operator
on certain spaces of functions on the interval [0, ∞). By an integral opera-
tor, we mean an operator T which takes an input function f and transforms
it into another function F = T {f } by means of integration with a kernel
function K(s, t). That is,
Z ∞
T {f (t)} = K(s, t)f (t) dt = F (s).
0

The Laplace transform is the particular integral transform obtained by using

the kernel function
K(s, t) = e−st .
When applied to a (constant coefficient linear) differential equation the
Laplace transform turns it into an algebraic equation, one that is generally
much easier to solve. After solving the algebraic equation one needs to trans-
form the solution of the algebraic equation back into a function that is the
80 3 The Laplace Transform

solution to the original differential equation. This last step is known as the
inversion problem.
This process of transformation and inversion is analogous to the use of the
logarithm to solve a multiplication problem. When scientific and engineering
calculations were done by hand, the standard procedure for doing multiplica-
tion was to use logarithm tables to turn the multiplication problem into an
addition problem. Addition, by hand, is much easier than multiplication. Af-
ter performing the addition, the log tables were used again, in reverse order,
to complete the calculation. Now that calculators are universally available,
multiplication is no more difficult than addition (one button is as easy to
push as another) and the use of log tables as a tool for multiplication is es-
sentially extinct. The same cannot be said for the use of Laplace transforms
as a tool for solving ordinary differential equations. The use of sophisticated
mathematical software (Maple, Mathematica, MatLab) can simplify many of
the routine calculations necessary to apply the Laplace transform, but it in no
way absolves us of the necessity of having a firm theoretical understanding of
the underlying mathematics, so that we can legitimately interpret the num-
bers and pictures provided by the computer. For the purposes of this course,
we provide a table (Table C.2) of Laplace transforms for many of the common
functions you are likely to see. This will provide a basis for studying many
examples.

3.1 Definition and Basic Formulas for the Laplace

Transform
Suppose f (t) is a continuous function defined for all t ≥ 0. The Laplace
transform of f is the function L {f (t)} (s) = F (s) defined by the equation
Z ∞ Z r
F (s) = L {f (t)} (s) = e−st f (t) dt = lim e−st f (t) dt (1)
0 r→∞ 0

provided the limit exists for all sufficiently large s. This means that there is
a number N , which will depend on the function f , so that the limit exists
whenever s > N . If there is no such N , then the function f will not have a
Laplace transform. It can be shown that if the Laplace transform exists at
s = N then it exists for all s ≥ N .
Let’s analyze this equation somewhat further. The function f with which
we start will sometimes be called the input function. Generally, ‘t’ will
denote the variable for an input function f , while the Laplace transform of f ,
denoted L {f (t)} (s),1 is a new function (the output function or transform
1
Technically, f is the function while f (t) is the value of the function f at t. Thus,
to be correct the notation should be L {f } (s). However, there are times when the
variable t needs to be emphasized or f is given by a formula such as in L e2t  (s).
Thus we will freely use both notations: L {f (t)} (s) and L {f } (s).
 3.1 Definition and Basic Formulas for the Laplace Transform 81

function), whose variable will usually be ‘s’. Thus Equation (1) is a formula
for computing the value of the function L {f } at the particular point s, so
that, in particular
Z ∞
F (2) = L {f } (2) = e−2t f (t) dt
0
Z ∞
and F (−3) = L {f } (−3) = e3t f (t) dt,
0

provided s = 2 and s = −3 are in the domain of L {f (t)}.

When possible, we will use a lower case letter to denote the input function
and the corresponding uppercase letter to denote its Laplace transform. Thus,
F (s) is the Laplace transform of f (t), Y (s) is the Laplace transform of y(t),
etc. Hence there are two distinct notations that we will be using for the Laplace
transform of f (t); if there is no confusion we use F (s), otherwise we will write
L {f (t)} (s).
To avoid the notation becoming too heavy-handed, we will frequently write
L {f (t)} rather than L {f (t)} (s). That is, the variable s may be suppressed
when the meaning is clear. It is also worth emphasizing that, while the in-
put function f has a well determined domain [0, ∞), the Laplace transform
L {f } (s) = F (s) is only defined for all sufficiently large s, and the domain will
depend on the particular input function f . In practice, this will not be an issue
and we will generally not emphasize the particular domain of F (s). We will
restrict our attention to continuous input functions in this chapter. In Chapter
8, we ease this restriction and consider Laplace transforms of discontinuous
functions.
A particularly useful property of the Laplace transform, both theoretically
and computationally, is that of linearity. For the Laplace transform linearity
means the following, which, because of its importance, we state formally as a
theorem.

Theorem 1. The Laplace transform is linear. In other words, if f and g are

input functions and a and b are constants then

Linearity of the Laplace transform

L {af + bg} = aL {f } + bL {g} .

Proof. This follows from the fact that (improper) integration is linear. u
t

Basic Laplace Transforms

Example 2. Verify the Laplace transform formula:
82 3 The Laplace Transform

Constant Functions
1
L {1} (s) = , s > 0.
s

I Solution. For the constant function 1 we have

Z ∞ r
−st e−ts 
L {1} (s) = e · 1 dt = lim
0 r→∞ −s 
0
e−rs − 1 1
= lim = for s > 0.
r→∞ −s s
J
Some comments are in order. The condition s > 0 is needed for the limit
e−rs − 1
lim
r→∞ −s
R ∞ −st
that defines the improper integral 0 e dt to exist. This is because
(
0 if c < 0
lim erc =
r→∞ ∞ if c > 0.

More generally, it follows from L’Hôspital’s rule that

lim tn ect = 0 if n ≥ 0 and c < 0. (2)

t→∞

This important fact (which you learned in calculus) is used in a number

∞
of calculations in the following manner. We will use the notation h(t)|a
r
as a shorthand for limr→∞ h(t)|a = limr→∞ (h(r) − h(a)). In particular, if
∞
limt→∞ h(t) = 0 then h(t)|a = −h(a), so that Equation (2) implies
(
n ct ∞
 0 if n > 0 and c < 0
t e 0 = (3)
−1 if n = 0 and c < 0.

Example 3. Assume n is a nonnegative integer. Verify the Laplace trans-

form formula:

Power Functions
n!
L {tn } (s) = , s > 0.
sn+1
 3.1 Definition and Basic Formulas for the Laplace Transform 83

I Solution. If n = 0 then f (t) = t0 = 1 and this case was given above.

Assume now that n > 0. Then
Z ∞
n
L {t } (s) = e−st tn dt
0

and this integral can be computed using integration by parts with the choice
of u and dv given as follows:

u = tn dv = e−st dt
−e−st
du = ntn−1 dt v= .
s
Using the observations concerning L’Hôspital’s rule in the previous paragraph,
we find that if n > 0 and s > 0, then
Z ∞
L {tn } (s) = e−st tn dt
0
∞
e−st  n ∞ −st n−1
Z
= −tn + e t dt
s 0 s 0
n 
= L tn−1 (s).

s
By iteration of this process (or by induction), we obtain (again assuming n > 0
and s > 0)
n  n−1
L {tn } (s) = L t (s)
s
n (n − 1)  n−2
= · L t (s)
s s
..
.
n n−1 2 1 
= · · · · · L t0 (s).
s s s s
But L t0 (s) = L {1} (s) = 1/s, by Example 2. The result now follows.

J

Example 4. Assume a ∈ R. Verify the Laplace transform formula:

Exponential Functions
1
L {eat } (s) = , s > a.
s−a
84 3 The Laplace Transform

I Solution.
∞ ∞ ∞
e−(s−a)t  1
Z Z
at −st at −(s−a)t

L e (s) = e e dt = e dt = = .
0 0 −(s − a) 0 s−a
The last equality follows from Equation (3) provided s > a. J
We note that in this example the calculation can be justified for λ =
a + ib ∈ C instead of a ∈ R, once we have noted what we mean by the
complex exponential function. The main thing to note is that the complex
exponential function ez (z ∈ C) satisfies the same rules of algebra as the
real exponential function, namely, ez1 +z2 = ez1 ez2 . This is achieved by simply
noting that the same power series which defines the real exponential makes
sense for complex values also. Recall that the exponential function ex has a
power series expansion
∞
X xn
ex =
n=0
n!
which converges for all x ∈ R. This familiar infinite series makes perfectly
good sense if x is replaced by any complex number z, and moreover, it can be
shown that the resulting series converges for all z ∈ C. Thus, we define the
complex exponential function by means of the convergent series
∞
X zn
ez := . (4)
n=0
n!

It can be shown that this function ez satisfies the functional equation

ez1 +z2 = ez1 ez2 .

Since e0 = 1, it follows that 1/ez = e−z . Taking z = it in Equation (4) leads

to an important formula for the real and imaginary parts of eit :
∞
X (it)n t2 t3 t4 t5
eit = = 1 + it − − i + + i − · · ·
n=0
n! 2! 3! 4! 5!
t2 t4 t3 t5
= (1 − + − · · · ) + i(t − + − · · · ) = cos t + i sin t,
2! 4! 3! 5!
where one has to know (from studying calculus) that the two series in the last
line above are the Taylor series expansions for cos t and sin t, respectively. In
words, this says that the real part of eit is cos t and the imaginary part of eit
is sin t. Combining this with the basic exponential functional property gives
the formula, known as Euler’s formula, for the real and imaginary parts of
eλt (λ = a + bi):

eλt = e(a+bi)t = eat+ibt = eat eibt = eat (cos bt + i sin bt).

We formally state this as a theorem.

 3.1 Definition and Basic Formulas for the Laplace Transform 85

Theorem 5 (Euler’s Formula). If λ = a + bi ∈ C and t ∈ R, then

eλt = eat (cos bt + i sin bt).

An important consequence of Euler’s formula is the limit formula

lim e(a+bi)t = 0, if a < 0.

t→∞

More generally, the analog of Equation (3) is

(
∞ 0 if n > 0 and a < 0
tn e(a+bi)t  = (5)

0 −1 if n = 0 and a < 0.

Example 6. Assume λ ∈ C. Verify the Laplace transform formula:

Compex Exponential Functions

1
L eλt (s) =

, s > Re λ.
s−λ

I Solution.
Z ∞
L eλt (s) = e−st eλt dt

0
∞ ∞
e−(s−λ)t  1
Z
= e−(s−λ)t dt = = .
0 −(s − λ) 0 s−λ

The last equality follows from Equation (5), provided the real part of the coef-
ficient of t in the exponential, i.e., −(s − Re λ), is negative. That is, provided
s > Re λ. J

Complex-Valued Functions
When we are given a problem that can be stated in the domain of the real
numbers it can sometimes be advantageous to recast that same problem in
the domain of the complex numbers. In many circumstances the resolution of
the problem is easier over the complex numbers. This important concept will
recur several times throughout this text.
Let λ = a + ib. Euler’s formula, eλt = e(a+ib)t = eat (cos bt + i sin bt),
introduces the concept of a complex-valued function. Let us say a few general
words about this notion.
A complex valued function z defined on an interval I in R is a function
of the form
86 3 The Laplace Transform

z(t) = x(t) + iy(t), t ∈ I.

Here, x(t) and y(t) are real valued functions, x(t) is called the real part of
z(t), and y(t) is called the imaginary part of z(t). Thus eibt is a complex
valued function with real part cos bt and imaginary part sin bt. The notions
of limits, continuity, differentiability, and integrability, all extend to complex
valued functions. In fact, since the complex numbers can be identified with the
Euclidean plane, R2 , with i = (0, 1), we can also identify complex-valued func-
tions with vector-valued (R2 -valued) functions written in the form (x(t), y(t)).
In the context of R2 -valued functions, differentiation and integration are per-
formed component-wise. Thus z(t) is differentiable if and only if the real
and imaginary parts of z are differentiable and

z 0 (t) = x0 (t) + iy 0 (t).

Thus, if z(t) = cos t+i sin t = eit then z 0 (t) = − sin t+i cos t = ieit . In fact, for
any complex number γ it is easy to verify that the derivative of eγt is γeγt . The
second derivative is given by z 00 (t) = x00 (t)+iy 00 (t) and higher order derivatives
are analogous. The usual product and quotient rules for differentiation apply
for complex valued functions. Similarly, z(t) is integrable on an interval [a, b]
if and only if the real and imaginary parts are integrable on [a, b]. We then
have Z b Z b Z b
z(t) dt = x(t) dt + i y(t) dt.
a a a

We define the Laplace transform of z(t) by the integral formula

Z ∞
L{z(t)}(s) = e−st z(t) dt,
0

provided the integral is finite for all s > N for some N . In this case we say z(t)
has a Laplace transform. It is easy to see that z(t) has a Laplace transform if
and only if the real and imaginary parts of z have Laplace transforms and

L {z(t)} (s) = L {x(t)} (s) + iL {y(t)} (s).

The notion of linearity introduced in Theorem 1 applies to complex-valued

functions and complex scalars. Since e−st is real-valued we have Re e−st z(t) =
e−st x(t) and Im e−st z(t) = e−st y(t) and this simple observation implies

Theorem 7. Suppose z(t) = x(t) + iy(t), where x(t) and y(t) are real-valued
functions have Laplace transforms. Then

Re L {z(t)} (s) = L {x(t)} (s)

Im L {z(t)} (s) = L {y(t)} (s).
R1
Example 8. Let z(t) = e2t + it2 . Find z 0 (t), 0 z(t) dt, and L {z(t)} (s).
 3.1 Definition and Basic Formulas for the Laplace Transform 87

I Solution.

z 0 (t) = 2e2t + 2it

1 1
e2t t3  e2 − 1 1
Z
z(t) dt = +i  = +i
0 2 3 0 2 3

1 2 s3 + i2(s − 2)
L {z(t)} (s) = +i 3 = .
s−2 s s3 (s − 2)
J

With these simple concepts and observations about complex-valued func-

tions in mind we now return to establishing basic Laplace transform formulas.

Example 9. Let b ∈ R. Verify the Laplace transform formulas:

Cosine Functions
s
L {cos bt} (s) = , for s > 0.
s2 + b2

and

Sine Functions
b
L {sin bt} (s) = , for s > 0.
s2 + b2

I Solution. A direct application of the definition of the Laplace Transform

applied to sin bt or cos bt would each require two integrations by parts; a
tedious calculation. As indicated above reframing the problem in the context
of complex-valued functions sometimes simplify matters substantially. This is
the case here. We use Equation (6) with λ = ib to get
1
L eibt (s) =

s − ib
1 s + ib s + ib
= = 2
s − ib s + ib s + b2
s b
= 2 +i 2
s + b2 s + b2
On the other hand, Euler’s formula and linearity gives
88 3 The Laplace Transform

L eibt (s) = L {cos bt + i sin bt} (s)



= L {cos bt} (s) + iL {sin bt} (s).

By equating the real and imaginary parts, as in Theorem 7, we obtain

s b
L {cos bt} (s) = and L {sin bt} (s) = , for s > 0.
s2 + b 2 s2 + b 2
J
Example 10. Let n be a nonnegative integer and a ∈ R. Verify the following
Laplace transform formula:

Simple Exponential Polynomial

n!
L {tn eat } (s) = , for s > a.
(s − a)n+1

I Solution. Notice that

Z ∞ Z ∞
L tn eat (s) = e−st tn eat dt = e−(s−a) tn dt = L {tn } (s − a).

0 0

What this formula says is that the Laplace transform of the function tn eat
evaluated at the point s is the same as the Laplace transform of the function
tn evaluated at the point s − a. Since L {tn } (s) = n!/sn+1 , we conclude
n!
L tn eat (s) = for s > a.

,
(s − a)n+1
J
The same proof gives the following
Example 11. Let n be a nonnegative integer and λ ∈ C. Verify the following
Laplace transform formula:

Simple Complex Exponential Polynomial

n!
L tn eλt (s) = for s > Re λ.

,
(s − λ)n+1

If the function tn in Example 10 is replaced by an arbitrary function f (t)

with a Laplace transform F (s), then we obtain the following:
 3.1 Definition and Basic Formulas for the Laplace Transform 89

Theorem 12. Suppose f has Laplace transform F (s). Then

First Translation Principle

L {eat f (t)} (s) = F (s − a).

Proof.
Z ∞
L eat f (t) (s) = e−st eat f (t) dt


Z0 ∞
= e−(s−a) f (t) dt
0

= L {f (t)} (s − a) = F (s − a).

u
t
In words, this formula says that to compute the Laplace transform of f (t)
multiplied by eat , then it is only necessary to take the Laplace transform of
f (t) (namely, F (s)) and replace the variable s by s−a, where a is the coefficient
of t in the exponential multiplier. Here is an example of this formula in use.
Example 13. Suppose a, b ∈ R. Verify the Laplace transform formulas

s−a
L eat cos bt (s) =

(s − a)2 + b2

and

b
L eat sin bt (s) =

.
(s − a)2 + b2

I Solution. From Example 9 we know that

s b
L {cos bt} (s) = and L {sin bt} (s) = .
s2 + b 2 s2 + b 2
Replacing s by s − a in each of these formulas gives the result. J
For a numerical example, note that
3 n √ o s−3
L e−t sin 3t = and L e3t cos 2t =

.
(s + 1)2 + 9 (s − 3)2 + 2
90 3 The Laplace Transform

Example 14. Suppose n is a nonnegative integer and λ = a+ ib ∈ C. Verify

the following Laplace transform formulas:

n! Re (s − λ)n+1
 

n at n!
L {t e cos bt} (s) = Re =
(s − λ)n+1 ((s − a)2 + b2 )n+1

and

n! Im (s − λ)n+1
 

n at n!
L {t e sin bt} (s) = Im = .
(s − λ)n+1 ((s − a)2 + b2 )n+1

I Solution. Let λ = a + bi. Then Euler’s formula gives

eλt = eat (cos bt + i sin bt)
and hence
tn eλt = tn eat cos bt + itn eat sin bt.
That is, tn eat cos bt is the real part and tn eat sin bt is the imaginary part of
tn eλt . By the First Translation principle (or, more directly, from Example 11)
n!
L tn eλt (s) =

.
(s − λ)n+1
The formulas follow now from Theorem 7. J
Notice that when n = 0 in Example 14 the formula reduces to those given
in Example 13. We need only observe that
1 1
=
s−λ (s − a) − ib
1 (s − a) + ib
= ·
(s − a) − ib (s − a) + ib
(s − a) + ib
=
(s − a)2 + b2
and equate real and imaginary parts.
If we take n = 1 in the above example, then
1 1
=
(s − λ)2 ((s − a) − ib)2
1 ((s − a) + ib)2
= ·
((s − a) − ib) ((s − a) + ib)2
2

((s − a)2 − b2 ) + i2(s − a)b

= .
((s − a)2 + b2 )2
 3.1 Definition and Basic Formulas for the Laplace Transform 91

By taking real and imaginary parts of this last expression, we conclude that

(s − a)2 − b2
L teat cos bt (s) =

((s − a)2 + b2 )2
2(s − a)b
and L teat sin bt (s) =

.
((s − a)2 + b2 )2
There are general formulas for n > 1 but they are difficult to remember and
not very instructive. We will not give them here.
We now summarize in Table 3.1 the basic Laplace transform formulas
we have derived. The student should learn these well as they will be used
frequently throughout the text and exercises. As we continue several new for-
mulas will be derived. Table C.2, in the appendix to this text, has a complete
list of Laplace transform formulas that we have derived.
With the use of the Table and linearity we can find the Laplace transform
of many functions. Consider the following examples.

Example 15. Find the Laplace transform of the following functions

1. 3 − 2t3 − 3e−3t 4. e(1+2i)t

2. 3 sin 2t + te2t 5. sin2 t
3. e6t cos 3t 6. t2 e−3t sin t

I Solution.

1. L{3 − 2t3 − 3e−3t } = 3L{1} − 2L{t3 } − 3L{e−3t }

1 3! 1 9s3 − 12s − 36
= 3 −2 4 −3 =
s s s+3 s4 (s + 3)

2. L{3 sin 2t + te2t } = 3L{sin 2t} + L{te2t }

2 1 7s2 − 24s + 28
=3 + = 2
s2
+2 2 (s − 2) 2 (s + 4)(s − 2)2
s−6 s−6
3. L{e6t cos 3t} = 2 2
= 2
(s − 6) + 3 s − 12s + 45
1 1 (s − 1) + 2i
4. L{e(1+2i)t } = = ·
s − (1 + 2i) (s − 1) − 2i (s − 1) + 2i
(s − 1) + 2i (s − 1) + 2i
= = 2
(s − 1)2 + 4 s − 2s + 5

5. The function sin2 t does not directly appear in the table. However, the
1
double angle formula cos 2t = 1 − 2 sin2 t implies sin2 t = (1 − cos 2t).
2
Thus
92 3 The Laplace Transform
 
2 1 1 1 s
L{sin t} = L{1 − cos 2t} = −
2 2 s s2 + 4
1 s2 + 4 − s2 2
= · =
2 s(s2 + 4) s(s2 + 4)

6. We apply Example 14 with λ = −3 + i. First,

2 2 ((s + 3) + i)3
3
= ·
(s − λ) ((s + 3) − i) ((s + 3) + i)3
3

(s + 3)3 − 3(s + 3)2 i + 3(s + 3)i2 − i3

=2
((s + 3)2 + 1)3
(s + 3)3 − 3(s + 3) + i(−3(s + 3)2 + 1)
=2 .
((s + 3)2 + 1)3

Thus,
2 −6(s + 3)2 + 2
L t2 e−3t sin t = Im

= .
(s − λ)3 (s2 + 6s + 10)3
J

We now introduce another useful principle that can be used to compute

some Laplace transforms.

Theorem 16. Suppose f (t) is an input function and F (s) = L {f (t)} (s) is
the transform function. Then

Transform Derivative Principle

d
L {−tf (t)} (s) = F (s).
ds

R∞
Proof. By definition, F (s) = 0
e−st f (t) dt and thus
Z ∞
0 d
F (s) = e−st f (t) dt
ds 0
Z ∞
d −st
= (e )f (t) dt
0 ds
Z ∞
= e−st (−t)f (t) dt = L {−tf (t)} (s).
0

Passing the derivative under the integral can be justified. (c.f. ??) u
t

Repeated applications of the Transform Derivative Principle gives

 3.1 Definition and Basic Formulas for the Laplace Transform 93

Transform nth -Derivative Principle

dn
(−1)n L {tn f (t)} (s) = F (s).
dsn

Example 17. Use the Transform Derivative Principle to compute

L {t sin t} (s) and L {t cos t} (s).

I Solution. A direct application of the Transform Derivative Principle

gives
d
L {t sin t} (s) = − L {sin t} (s)
ds
d 1
=−
ds s2 + 1
−2s 2s
=− 2 = 2 .
(s + 1)2 (s + 1)2

d
and L {t cos t} (s) = − L {cos t} (s)
ds
d s
=−
ds s2 + 1
(s2 + 1) − 2ss s2 − 1
=− = .
(s2 + 1)2 (s2 + 1)2
J
94 3 The Laplace Transform

Table 3.1. Basic Laplace Transform Formulas

f (t) F (s)
1
1. 1
s
n!
2. tn
sn+1
1
3. eat
s−a
n!
4. tn eat
(s − a)n
s
5. cos bt
s 2 + b2
b
6. sin bt
s 2 + b2
s−a
7. eat cos bt
(s − a)2 + b2
b
8. eat sin bt
(s − a)2 + b2
1
9. eλt
s−λ
n!
10. tn eλt
(s − λ)n+1
n!
11. tn eat cos bt Re n+1 
(s − λ)
n!
12. tn eat sin bt Im n+1 
(s − λ)
We are assuming n is a nonnegative integer, a and b are real, and λ = a + ib.

Exercises
1–4. Compute the Laplace transform of 5. 5e2t
each function given below directly from
the integral definition given in Equation 6. 3e−7t − 7t3
(1). 7. t2 − 5t + 4
1. 3t + 1
8. t3 + t2 + t + 1
2. 5t − 9et 9. e−3t + 7e−4t
3. e2t − 3e−t 10. e−3t + 7te−4t
−3t
4. te 11. cos 2t + sin 2t
5–23. Find the Laplace transform of 12. et (t − cos 2t)
each function below. Use Table 3.1 and √
the linearity of the Laplace transform. 13. e−t/3 cos 6t
 3.2 Partial Fractions: A Recursive Method for Linear Terms 95

14. (t + e2t )2 1 λt
λ
e + C. (You will need to use in-
tegration by parts.
15. 5 cos 2t − 3 sin 2t + 4
27. Product Rule: Suppose z(t) and
16. e5t (8 cos 2t + 11 sin 2t) w(t) are two complex valued func-
17. t2 sin 2t tions. Verify

18. e−at − e−bt for a 6= b. (z · w)0 = z 0 w + zw0 .

19. cos2 bt (Hint: cos2 θ = 1

2
(1 +
cos 2θ)) 28. Quotient Rule: Suppose z(t) and
w(t) are two complex valued func-
20. sin2 bt tions. Verify

21. sin bt cos bt Hint: Use an appropri- z 0 z 0 w − w0 z
= .
ate trigonometric identity. w w2

22. cosh bt (Recall that cosh bt =

29. Chain Rule: Suppose z(t) and w(t)
(ebt + e−bt )/2.)
are two complex valued functions.
23. sinh bt (Recall that sinh bt = Verify
(ebt − e−bt )/2.)
(z ◦ w)0 (t) = z 0 (w(t))w0 (t).
24–30. The following problems concern
basic properties of complex valued func-
30. Verify the limits tn e(a+bi)t
∞
tions. =
0
24. Let λ = a + ib and γ = c + id. Deter- 0 if n > 0 and a < 0

mine the real and imaginary parts of −1 if n = 0 and a < 0.
γeλt .
2
31. Verify that the function f (t) = et
25. By separating the real and imagi-
d λt does not have a Laplace transform.
nary parts verify that dt e = λeλt .
That is, show that the improper inte-
26. By separating the real and imag- gral that defines F (s) does not con-
inary parts verity that eλt dt = verge for any value of s.

3.2 Partial Fractions:

A Recursive Method for Linear Terms

A careful look at Table 3.1 reveals that the Laplace transform of each function
we considered is a rational function. The inversion of the Laplace transform,
which we will discuss in detail in Sections 3.4 and 3.5, will involve writing
rational functions as sums of those simpler ones found in the Table.
All students of calculus should be familiar with the technique of obtain-
ing the partial fraction decomposition of a rational function. Briefly, a given
rational function p(s)/q(s) is a sum of partial fractions of the form
96 3 The Laplace Transform

Aj Bk s + Ck
and ,
(s − λ)j (s2 + cs + d)k
where Aj , Bk , and Ck are constants. The partial fractions are determined by
the linear factors, s − λ, and the irreducible quadratic factors, s2 + cs + d, of
the denominator q(s), where the powers j and k occur up to the multiplicity
of the factors. After finding a common denominator and equating the numer-
ators we obtain a system of linear equations to solve for the undetermined
coefficients Aj , Bk , Ck . Notice that the degree of the denominator determines
the number of coefficients that are involved in the form of the partial fraction
decomposition. Even when the degree is relatively small this process can be
very tedious and prone to simple numerical mistakes.
Our purpose in this section and the next is to provide an alternate algo-
rithm for obtaining the partial fraction decomposition of a rational function.
This algorithm has the advantage that it is constructive (assuming the factor-
ization of the denominator), recursive (meaning that only one coefficient at a
time is determined), and self checking. This recursive method for determining
partial fractions should be well practiced by the student. It is the method we
will use throughout the text and is an essential technique in solving nonho-
mogeneous differential equations discussed in Section 4.5.
We begin by reviewing a few well known facts about polynomials and ra-
tional functions. This will also give us an opportunity to set down notation
and introduce some important concepts. Thereafter we will discuss the algo-
rithm in the linear case, i.e. when the denominator has a linear term as a
factor. In Section 3.3 we specialize to the irreducible quadratic case.

Polynomials
A real polynomial p(s) is a function of the form

p(s) = an sn + an−1 sn−1 + . . . + a1 s + a0 ,

where the coefficients a0 , . . . , an are real. If the coefficients are in the com-
plex numbers then we say p(s) is a complex polynomial. The degree
of p(s) is n if the leading coefficient, an , is nonzero. If an = 1, we say
p(s) is a monic polynomial. Since R ⊂ C, a real polynomial may be re-
garded as a complex polynomial. By separating the coefficients into their
real and imaginary parts we can always write a complex polynomial p(s) in
the from p(s) = p1 (s) + ip2 (s), where p1 (s) and p2 (s) are real polynomi-
als. The complex conjugate of p(s), written p(s), is obtained by replac-
ing the coefficients by their complex conjugate. It is not hard to see that
p(s) = p1 (s) − ip2 (s). For example, if p(s) = 1 + i + 2s + is2 then we can
write p(s) = 1 + 2s + i(1 + s2 ). The real part of p(s) is 1 + 2s, the imag-
inary part is 1 + s2 , and p(s) = 1 + 2s − i(1 + s2 ). Note that the product
of two real (complex) polynomials is again a real (complex) polynomial. It
can happen that the product of two complex polynomials is real, for example,
 3.2 Partial Fractions: A Recursive Method for Linear Terms 97

(s+i)(s−i) = s2 +1. More generally, p(s)p(s) is a real polynomial. Sometimes

we will refer to p(s) just as a polynomial if it is not necessary to distinguish
the real or complex case.
We also observe that the sum of two real (complex) polynomials is a real
(complex) polynomial. Also the product of a real scalar and a real polynomial
is again a real polynomial. In short, we say that the set of polynomials is
closed under addition and scalar multiplication. A similar statement applies
in the complex case.

Linear Spaces
This important closure property just described for polynomials can be ex-
pressed in a more general setting. A set of functions, F , is called a real
(complex) linear space if it is closed under addition and real (complex)
scalar multiplication. In other words, F is a real linear space if

f1 + f2 ∈ F closure under addition

cf ∈ F closure under real scalar multiplication,

for f1 , f2 , and f in F and c ∈ R. If the scalars used are complex numbers

then F is a complex linear space. If f1 , . . . , fk are functions then a linear
combination of f1 , . . . , fk is a sum of the form

c1 f 1 + · · · + ck f k ,

where c1 , . . . , ck are scalars. When F is a linear space, i.e., closed under ad-
dition and scalar multiplication, and all f1 , . . . , fk are in F then all linear
combinations are back in F . This very important property of linear spaces is
exploited frequently throughout this text.
Let P denote all real polynomials and PC the complex polynomials. Then,
using this new language our discussion in the preceding paragraphs implies
that P is a real linear space and PC is a complex linear space.

The fundamental theorem of algebra expresses how polynomials factor in

the real and complex cases. Let’s begin with the complex case; it’s easier.

Theorem 1 (Fundamental Theorem of Algebra over C). Let q(s) ∈

PC . Then there are finitely many distinct complex numbers λ1 , . . . , λr and
positive integers m1 , . . . , mk such that

q(s) = C(s − λ1 )m1 (s − λ2 )m2 · · · (s − λr )mk .

Each λi is a root of q(s), since p(λi ) = 0. The positive integer mi is

the multiplicity of λi (or (s − λi )) in q(s). The constant C is the leading
98 3 The Laplace Transform

coefficient of q(s). Simply put, a real or complex polynomial can always be

factored into a product of linear terms, s − λ.
The real version of this theorem, given below, is slightly more complicated
if we require that the all factors have real coefficients. Suppose t(s) = s2 +
cs + d, where c and d are real. We say t(s) is an irreducible quadratic if
t(s) cannot be factored over the real numbers, i.e., we cannot write t(s) =
s2 + cs + d = (s − a1 )(s − a2 ), with a1 and a2 real numbers. It’s easy to see
that t(s) is irreducible if and only if t(s) is strictly positive. A simple example
is t(s) = s2 + 1, since t(s) ≥ 1. To determine whether a polynomial of degree 2
is irreducible we can apply the quadratic formula and check to see whether the
discriminant, c2 − 4d, is negative. The preferred, though equivalent, method
is to complete the square. Thus
c2 c2 c 4d − c2
t(s) = s2 + cs + d = s2 + cs + +d− = (s + )2 + .
4 4 2 4
If 4d − c2 is positive then t(s) is a sum of a square and a positive number,
hence positive and hence irreducible. Thus, irreducible quadratics, t(s), can
always we written in the form

t(s) = (s − a)2 + b2 ,
√
4d−c2
with a, b real and b > 0 (a = − 2c and b = 2 ). Written in this way the
complex roots can be easily read

λ = a + ib and λ = a − ib.
For example, consider t(s) = s2 + 4s + 13. Completing the square gives
t(s) = s2 + 4s + 4 + 13 − 4 = (s + 2)2 + 9 = (s + 2)2 + 32 , a sum of squares.
Thus t(s) is an irreducible quadratic; it does not factor over the real numbers.
However, it does factor over the complex numbers. It’s complex roots are
λ = −2 + 3i and λ = −2 − 3i and we can write

s2 + 4s + 13 = (s − (−2 + 3i))(s − (−2 − 3i)).

You will note that an irreducible quadratic has non real roots related by
complex conjugation: λ, λ , which we refer to as a conjugate root pair.

More generally, if q(s) is real and λ is a non real root then so is λ, since
0 = q(λ) = q(λ) = q(λ). Clearly, both roots have the same multiplicities.
This observation can be used to get the following theorem (the real case) as
a consequence of the complex case, Theorem 1.
Theorem 2 (Fundamental Theorem of Algebra over R). Let q(s) ∈ P.
Then there are finitely many distinct real linear terms s − a1 , . . . , s − ak and
positive integers m1 , . . . , mk and finitely many irreducible quadratic terms
s2 + c1 s + d1 , · · · , s2 + cl s + dl and positive integers n1 , . . . , nl such that

q(s) = C(s − a1 )m1 · · · (s − ak )mk · (s2 + c1 s + d1 )n1 . . . (s2 + cl s + dl )nl .

 3.2 Partial Fractions: A Recursive Method for Linear Terms 99

Each ai is a root of q(s). The positive integer mi is the multiplicity of

ai (or (s − ai )) in p(s) and the positive integer nj is the multiplicity of
s2 + cj s + dj in p(s). The constant C is the leading coefficient of q(s). Thus,
this theorem says that a real polynomial can always be factored into a product
of real linear terms and/or irreducible quadratic terms.

Rational Functions
A real (complex) rational function, p(s)
q(s) , is the quotient of two real (com-
plex) polynomials and proper if the degree of the numerator is less than the
degree of the denominator. For example, each of the following are rational
functions:
s 1 + 2is s s6
, , , and .
s2 − 1 s(s2 + 1) s+i s2 + 1
The first and fourth are real while the second and third are complex rational
function. The first and second are proper while the third and fourth are not
proper. If you look carefully at Table 3.1 and Example 3.1.15 you will note that
all of the Laplace transforms are proper rational functions. Some are complex
valued and some are real valued. We let RC denote the complex proper rational
functions and R the real proper rational functions. It is sometimes useful to
regard a real rational function as a complex rational function.
If pq11(s)
(s)
and pq22 (s)
(s)
are rational functions (either real or complex) then

p1 (s) p2 (s) p1 (s)q2 (s) + p2 (s)q1 (s)

+ = .
q1 (s) q2 (s) q1 (s)q2 (s)

It is easy to see from this that R and RC are closed under addition. If c is a
scalar then clearly
p(s) cp(s)
c· = .
q(s) q(s)
This implies that R is closed under real scalar multiplication and RC is closed
under complex scalar multiplication. These observations show that R is a real
linear space and RC is a complex linear space.
Let a(s)
b(s) be a complex rational function. Then we can write

a(s) a(s)b(s)
= .
b(s) b(s)b(s)

The numerator is a complex polynomial, however, the denominator, b(s)b(s),

is real. This observation thus shows that any complex rational function can
be written
p1 (s) p2 (s)
+i ,
q(s) q(s)
100 3 The Laplace Transform

1
where p1 (s), p2 (s), and q(s) are real polynomials. For example, s−i can be
written
1 1 s+i s+i s 1
= · = 2 = 2 +i 2 .
s−i s−i s+i s +1 s +1 s +1

Partial Fraction Decompositions: The Linear

Case
We now turn our attention to the main point of this section. We describe
a recursive procedure for obtaining the decomposition of a rational function
into partial fractions with powers of linear terms in the denominator.

Lemma 3. Suppose a proper rational function can be written in the form

p0 (s)
(s − λ)n q(s)

and q(λ) 6= 0. Then there is a unique number A1 and a unique polynomial

p1 (s) such that

p0 (s) A1 p1 (s)
= + . (1)
(s − λ)n q(s) (s − λ)n (s − λ)n−1 q(s)

The number A1 and the polynomial p1 (s) are given by

p0 (λ) p0 (s) − A1 q(s)

A1 = and p1 (s) = . (2)
q(λ) s−λ

Proof. After finding a common denominator in Equation (1) and equating

numerators we get the polynomial equation p0 (s) = A1 q(s) + (s − λ)p1 (s).
Evaluating at s = λ gives p0 (λ) = A1 (q(λ)) and hence A1 = pq(λ)
0 (λ)
. Now
that A1 is determined we have p0 (s) − A1 q(s) = (s − λ)p1 (s) and hence
p1 (s) = p0 (s)−A
s−λ
1 q(s)
. u
t

The proof we have given applies to both real and complex rational func-
tions. We remark that if p0 , q, and λ are real then A1 is real and the poly-
nomial p1 (s) is real. Notice that in the calculation of p1 it is necessary that
p0 (s) − A1 q(s) have a factor of s − λ. If such a factorization does not occur
when working an example then an error has been made. This is what is meant
when we stated above that this recursive method is self-checking.
An application of Lemma 3 produces two items:
• the partial fraction of the form
A1
,
(s − λ)n
 3.2 Partial Fractions: A Recursive Method for Linear Terms 101

• and a remainder term of the form

p1 (s)
,
(s − λ)n−1 q(s)

such that the original rational function p0 (s)/(s − λ)n q(s) is the sum of these
two pieces. We can now repeat the process on the new rational function
p1 (s)/((s − λ)n−1 q(s)), where the multiplicity of (s − λ) in the denominator
has been reduced by 1, and continue in this manner until we have removed
completely (s − λ) as a factor of the denominator. In this manner we produce
a sequence,
A1 An
,...,
(s − λ)n (s − λ)
which we will refer to as the (s − λ)-chain for the rational function p0 (s)/(s−
λ)n q(s). The number of terms, n, is sometimes referred to as the length of
the chain. The chain table below summarizes the data obtained.

The s − λ -chain

p0 (s) A1
(s − λ)n q(s) (s − λ)n (s)
p1 (s) A2
(s − λ)n−1 (s)q(s) (s − λ)n−1 (s)
.. ..
. .
pn−1 (s) An
(s − λ)q(s) (s − λ)
pn (s)
q(s)

Notice that the partial fractions are placed in the second column while the
remainder terms are placed in the first column under the previous remainder
term. This form is conducive to the recursion algorithm. From the table we
get
p0 (s) A1 An pn (s)
= + ···+ + .
(s − λ)n q(s) (s − λ)n (s − λ) q(s)
By factoring another linear term out of q(s) the process can be repeated
through all linear factors of q(s).
In the examples that follow we will organize one step of the recursion
process as follows:
102 3 The Laplace Transform

p0 (s) A1 
p0 (s) 
= where A1 = q(s)  =
(s − λ)n q(s) (s − λ)n s=λ

p1 (s) 1
+ and p1 (s) = s−λ (p0 (s) − A1 q(s))=
(s − λ)n−1 q(s)
The arrowed curves indicate where the results of calculations are inserted. A1
is calculated first and inserted in two places: in the s − λ -chain and in the
calculation for p1 (s). Afterwards, p1 (s) is calculated and the result inserted
in the numerator of the remainder term. Now the process is repeated on
p1 (s)
(s−λ)n−1 q(s) until the s − λ -chain is completed.
Consider the following example.
Example 4. Find the partial fraction decomposition for
16s
.
(s + 1)3 (s − 1)2
Remark 5. Before we begin with the solution we remark that the traditional
method for computing the partial fraction decomposition introduces the equa-
tion
16s A1 A2 A3 A4 A5
= + + + +
(s + 1)3 (s − 1)2 (s + 1)3 (s + 1)2 s + 1 (s − 1)2 (s − 1)
and after finding a common denominator requires the simultaneous solution
to a system of five equations in five unknowns, a doable task but one prone
to simple algebraic errors.
I Solution. We will first compute the s + 1 -chain. According to Lemma 3
we can write
16s A1 16s 

−16
= where A 1 = (s−1) s=−1 = 4 = −4
2
(s + 1)3 (s − 1)2 (s + 1)3
p1 (s) 1
+ and p1 (s) = s+1 (16s − (−4)(s − 1)2 )
(s + 1)2 (s − 1)2
4 2
= s+1 (s + 2s + 1)

= 4(s + 1).
4(s+1)
We now repeat the process on (s+1)2 (s−1)2 to get:

4(s + 1) A2 
4(s+1) 
2 2
= where A2 = (s−1)2 s=−1 =0
(s + 1) (s − 1) (s + 1)2
p2 (s) 1
+ and p2 (s) = s+1 (4(s + 1) − (0)(s − 1)2 )
(s + 1)(s − 1)2
=4
 3.2 Partial Fractions: A Recursive Method for Linear Terms 103

Notice here that we could have canceled the (s + 1) term at the beginning
4 2
and arrived immediately at (s+1)(s−1) 2 . Then no partial fraction with (s + 1)

in the denominator would occur. We chose though to continue the recursion

4
to show the process. The recursion process is now repeated on (s+1)(s−1) 2 to

get:
4 A3 4) 

= where A 3 = 2
(s−1)  = 44 = 1
(s + 1)(s − 1)2 (s + 1) s=−1

p3 (s) 1
+ and p2 (s) = s+1 (4 − (1)(s − 1)2 )
(s − 1)2
−1 2
= s+1 (s − 22 − 3)

= −(s − 3)

Putting these calculations together gives the s + 1 -chain

The s + 1 -chain

16 −4
(s + 1)3 (s − 1)2 (s + 1)3
4(s + 1) 0
(s + 1)2 (s − 1)2 (s + 1)2
4 1
(s + 1)(s − 1)2 (s + 1)
−(s − 3)
(s − 1)2

We now compute the s − 1 -chain for the remainder −(s−3)

(s−1)2 . It is implicit
that q(s) = 1.
−(s − 3) A4 
−(s−3)) 
= where A 4 = 1 =2
(s − 1)2 (s − 1)2

s=1
p4 (s) 1
+ and p4 (s) = s−1 (−(s − 3) − (2))
(s − 1)
−1
= s−1 (s − 1) = −1

The s − 1 -chain is thus

104 3 The Laplace Transform

The s − 1 -chain

−(s − 3) 2
(s − 1)2 (s − 1)2
−1
(s − 1)

We now have
−(s − 3) 2 −1
2
= 2
+
(s − 1) (s − 1) s−1
and putting this chain together with the s + 1 -chain gives
16s −4 0 1 2 −1
= + + + + .
(s + 1)3 (s − 1)2 (s + 1)3 (s + 1)2 s + 1 (s − 1)2 s−1
J

Lemma 3 and the Fundamental Theorem of Algebra over C imply that a

proper rational function p(s)
q(s) has a partial fraction decomposition with powers
of linear terms in the denominator. In other words, we can always write p(s) q(s)
1
as a sum of simple rational functions (s−λ) k , where λ is a root of q(s) and

k = 1, . . . , m, with m the multiplicity of λ. Of course, even if q(s) is a real

polynomial the partial fraction decomposition may involve linear terms with
complex roots. This situation is considered in the following example.

Example 6. Find the partial fraction decomposition of

4
(s2 + 1)2

with powers of linear terms over the complex numbers.

I Solution. The irreducible quadratic, s2 + 1, factors over C as s2 + 1 =

(s − i)(s + i). Thus
4 4
= .
(s2 + 1) (s − i)2 (s + i)2
We begin with the s − i -chain.
 3.2 Partial Fractions: A Recursive Method for Linear Terms 105

4 A1 4

4
= where A1 = = = −1

(s − i)2 (s + i)2 (s − i)2 (s+i)2 s=i −4

p1 (s) 1
+ and p1 (s) = s−i (4 − (−1)(s + i)2 )
(s − i)(s + i)2
1 2
= s−i (s + 2is + 3)
1
= s−i (s − i)(s + 3i)

= s + 3i.
s+3i
We now repeat the process on (s−i)(s+i)2

s + 3i A2 s+3i 

4i
2
= where A2 = (s+i)2 s=i = −4 = −i
(s − i)(s + i) (s − i)
p2 (s) 1
+ and p2 (s) = s−i (s + 3i + (i)(s + i)2 )
(s + i)2
i
= s−i (−is + 3 + s2 + 2is − 1)
i 2
= s−i (s + is + 2)
i
= s−i (s − i)(s + 2i)

= i(s + 2i).
i(s+2i)
Finally, we proceed on (s+i)2 .

i(s + 2i) A3 i(s+2i) 


−1
2
= where A3 = (s+i)2 s=−i = 1 = −1
(s + i) (s + i)2
p3 (s) 1
+ and p3 (s) = s+i (i(s + 2i) − (−1)
(s + i)
i
= s+i (s + i)

= i.

Putting the resulting chains together gives

4 −1 −i −1 i
= + + + .
(s2 + 1)2 (s − i)2 s − i (s + i)2 (s + i)
J

Special Circumstances and Simplifications

Product of Distinct Linear Factors. Let p(s) q(s) be a proper rational func-
tion. Suppose q(s) is the product of distinct linear factors, i.e.

q(s) = (s − λ1 ) · · · (s − λn ),
106 3 The Laplace Transform

where λ1 , . . . λn are distinct scalars. Then each chain has length one and the
partial fraction decomposition has the form
p(s) A1 An
= + ···+ .
q(s) s − λ1 (s − λn )
The scalar Ai is the first and only entry in the (s − λi ) -chain. Thus
p(λi )
Ai = ,
qi (λi )
where qi (s) is the polynomial obtained from q(s) by factoring out (s − λi ). If
we do this for each i = 1, . . . , n, it is unnecessary to calculate any remainder
terms.
Example 7. Find the partial fraction decomposition of
−4s + 14
.
(s − 1)(s + 4)(s − 2)
I Solution. The denominator q(s) = (s − 1)(s + 4)(s − 2) is a product of
distinct linear factors. Each partial fraction is determined as follows:

A1 −4s+14  10
• For s−1 : A1 = (s+4)(s−2)  = −5 = 2.
s=1


A2 −4s+14  30
• For s+4 : A2 = (s−1)(s−2) s=−4 = 30 = 1.


A3 −4s+14  6
• For s−2 : A3 = (s−1)(s+4)  = 6 = 1.
s=2
The partial fraction decomposition is thus
−4s + 14 2 1 1
= + + .
(s − 1)(s + 4)(s − 2) s−1 s+4 s−2
J
Complex Partial Fractions of Real Rational Functions.
At times it is convenient to compute the decomposition of a real proper
rational function into linear partial fractions. It may then be necessary to
factor over the complex numbers. We did this in Example 6 and we wish to
observe in that example that the linear partial fractions came in conjugate
pairs. Recall,
4 −1 −i −1 i
= + + + .
(s2 + 1)2 (s − i)2 s − i (s + i)2 (s + i)
Since s2 + 1 is real its roots, (i, −i) come as a pair, the second root is the
complex conjugate of the first. Likewise, in the partial fraction decomposition,
there are two pairs
 3.2 Partial Fractions: A Recursive Method for Linear Terms 107

−i −1 −1
   
i
, and , .
s−i s+i (s − i)2 (s + i)2
For each pair, the second partial fraction is the complex conjugate of the first.
This will happen in general and allows a simplification of computations
when computing linear partial fraction decompositions of real rational func-
tions over the complex numbers.

Theorem 8. Suppose p(s) q(s) is a real rational function. Suppose λ is a non

A
real root of q(s) with multiplicity m. If (s−λ) k appears in the partial fraction

p(s) A
decomposition of q(s) then so does (s−λ)k
.

Proof. Since q(s) is real both λ and λ are roots with multiplicity m. Thus
A B p(s)
(s−λ)k and (s−λ)k appear in the partial fraction decomposition of q(s) , for
p(s) A
k = 1, . . . , m. However, since q(s) is real the complex conjugates, (s−λ)k
and
B
(s−λ)k , also appear in the partial fraction decomposition. It follows, by the
uniqueness of the coefficients that B = A. u
t

Theorem 8 implies that if λ is a non real root for q(s) then the s − λ -chain
is then the complex conjugate of the s − λ -chain. Thus, in Example 6 the
s + i -chain can be obtained immediately from the s − i -chain by complex
conjugation rather than by applying Lemma 3. Consider another example that
illustrates the use of Theorem 8.
Example 9. Find the partial fraction decomposition over C of
16s
.
(s2 + 1)3

I Solution. We can write (s216s 16s

+1)3 = (s−i)3 (s+i)3 . In the table below we give
the s − i -chain. We leave it to the student to make the necessary verifications.

The s − i -chain

16s −2
(s − i)3 (s + i)3 (s − i)3

2s2 + 8is + 2 −i
(s − i)2 (s + i)3 (s − i)2

is2 − 2s + 3i 0
(s − i)(s + i)3 (s − i)
108 3 The Laplace Transform

We now apply Theorem 8 to get the s+i -chain by complex conjugation. Thus
16s −2 −i −2 i
= + + + .
(s2 + 1)3 (s − i)3 (s − i)2 (s + i)3 (s + i)2
J

Exercises
1–7. Determine whether f (t) is a linear 2s
16.
combination of the given set of functions. 3s

1. {sin t, cos t} ; f (t) = 3 sin t − 4 cos t 17–27. For each exercise below com-
pute the chain table through the indi-
2. e3t , e−2t  ; f (t) = 3e3t cated linear term.
5s + 10
3. {sin 2t, cos 2t} ; f (t) = sin2 2t + 17. ; (s − 1)
(s − 1)(s + 4)
cos2 2t
10s − 2
4. sin2 t, cos2 t  ; f (t) = 1 18. ; (s − 2)
(s + 1)(s − 2)
5. e4t , e−4t  ; f (t) = cosh 4t 1
19. ; (s − 5)
3 (s + 2)(s − 5)
6. t, t , t
2 3 2
; f (t) = t + t + t + 1

7. e3t , e2t  ; f (t) = te3t + 2e2t

5s + 9
20. ; (s + 3)
(s − 1)(s + 3)
8–16. Identify each of the following 3s + 1
functions as proper rational (PR), ratio- 21. ; (s − 1)
(s − 1)(s2 + 1)
nal but not proper rational (R), or not
rational (NR). 3s2 − s + 6
2
22. ; (s + 1)
s −1 (s + 1)(s2 + 4)
8.
(s − 2)(s − 3)
s2 + s − 3
23. ; (s + 3)
2s − 1 (s + 3)3
9.
(s − 2)(s − 3)
5s2 − 3s + 10
3 2 24. ; (s + 2)
10. s − s + 1 (s + 1)(s + 2)2
1 s s
11. + (s+1)(s2 +1) 25. ; (s + 1)
s−2 (s + 2)2 (s + 1)2
1 s 16s
12. · (s+1)(s2 +1) 26. ; (s − 1)
s−2 (s − 1)3 (s − 3)2
2s + 4 1
13. 27. ; (s − 4)
s3/2 − s + 1 (s − 4)5 (s − 5)
cos(s + 1)
14.
sin(s2 + 1) 28–47. Find the decomposition of each
2 proper rational function into partial frac-
3s − 4 tions with powers of linear terms in the
15.
2s + s + 5 
2
denominator.
 3.2 Partial Fractions: A Recursive Method for Linear Terms 109
5s + 9 s2
28. 44.
(s − 1)(s + 3) (s + 2)2 (s + 1)2
8+s 45. 8s
; (s − 3)
29. (s−1)(s−2)(s−3)3
s2 − 2s − 15
s
1 46.
30. (s − 2)2 (s − 3)2
s2 − 3s + 2
16s
5s − 2 47.
31. (s − 1)3 (s − 3)2
s2 + 2s − 35
3s + 1 48–57. Find the decomposition of each
32.
s2 + s proper rational function into partial frac-
tions with powers of linear terms in the
2s + 11
33. denominator. You will need to use com-
s2 − 6s − 7
plex numbers. It may be helpful to keep
2s2 + 7 Theorem 8 in mind.
34.
(s − 1)(s − 2)(s − 3) 2s + 4
48.
2
s2 − 4s + 8
s +s+1
35. 2 + 3s
(s − 1)(s2 + 3s − 10) 49.
s2 + 6s + 13
s2 7 + 2s
36. 50.
(s − 1)3 s2 + 4s + 29
7 2s + 2
37. 51.
(s + 4)4 (s − 1)(s2 + 1)
s
38. 2s2 + 28
(s − 3)3 52.
(s + 1)(s2 + 4)
s2 + s − 3 25
39. 53.
(s + 3)3 s2 (s − 5)(s + 1)
5s2 − 3s + 10 8s2 + 48
40. 54.
(s + 1)(s + 2)2 (s − 3)(s2 + 2s + 5)

s2 − 6s + 7 6s + 12
41. 55. .
(s2 − 4s − 5)2 (s2 + 1)(s2 + 4)

81 6s2 + 8s + 8
42. 56.
s3 (s + 9) (s2 + 4)2
s 16
43. 57.
(s + 2)2 (s + 1)2 (s2 + 1)3

3.3 Partial Fractions:

110 3 The Laplace Transform

A Recursive Method for Irreducible Quadratics

We continue the discussion of Section 3.2. Here we consider the case where a
real rational function has a denominator with irreducible quadratic factors.
Lemma 1. Suppose a real proper rational function can be written in the form
p0 (s)
,
(s2 + cs + d)n q(s)

where s2 + cs + d is an irreducible quadratic that is factored completely out of

q(s). Then there is a unique linear term B1 s + C1 and a unique polynomial
p1 (s) such that
p0 (s) B1 s + C1 p1 (s)
= 2 + s . (1)
(s2 + cs + d)n q(s) (s + cs + d)n (s + cs + d)n−1 q(s)

If a + ib is a complex root of s2 + cs + d then B1 s + C1 and the polynomial

p1 (s) are given by
p0 (a + ib) p0 (s) − B1 q(s)
B1 (a + ib) + C1 = and p1 (s) = . (2)
q(a + ib) s2 + cs + d
Proof. After finding a common denominator in Equation (1) and equating
numerators we get the polynomial equation

p0 (s) = (B1 s + C1 )q(s) + (s2 + cs + d)p1 (s). (3)

Evaluating at s = a + ib gives p0 (a + ib) = (B1 (a + ib) + C1 )(q(a + ib)) and

hence
p0 (a + ib)
B1 (a + ib) + C1 = . (4)
q(a + ib)
B1 and C1 are both determined by Equation (4). Solving for p1 (s) in Equation
(3) gives
p0 (s) − (B1 s + C1 )q(s)
p1 (s) = .
s2 + cs + d
u
t
An application of Lemma 1 produces two items:
• the partial fraction of the form
B1 s + C1
,
(s2 + cs + d)n
• and a remainder term of the form
p1 (s)
,
(s2 + cs + d)n−1 q(s)
 3.3 Partial Fractions: A Recursive Method for Irreducible Quadratics 111

such that the original rational function p0 (s)/(ss + as + b)n q(s) is the sum of
these two pieces. We can now repeat the process in the same way as the linear
case.
The result is called the (s2 + cs + d)-chain for the rational function
p0 (s)/(s2 + cs + d)n q(s). The table below summarizes the data obtained.

The s2 + as + b -chain

p0 (s) B1 s + C1
(s2 + cs + d)n q(s) (s2 + cs + d)n
p1 (s) B2 s + C2
(s2 + cs + d)n−1 q(s) (s2 + cs + d)n−1
.. ..
. .

pn−1 (s) Bn s + Cn
(s2 + cs + d) q(s) (s2 + cs + d)
pn (s)
q(s)

From the table we get

p0 (s) B1 s + C1 Bn s + Cn pn (s)
= 2 + ···+ 2 + .
(s2 + cs + d)n q(s) (s + cs + d)n (s + cs + d) q(s)
Consider the following examples.

Example 2. Find the partial fraction decomposition for

30s + 40
.
(s2 + 1)2 (s2 + 2s + 2)
Remark 3. We remark that since the degree of the denominator is 6 the
traditional method of determining the partial fraction decomposition would
involve solving a system of six equations in six unknowns.

I Solution. First observe that both factors in the denominator, s2 + 1 and

s2 + 2s + s = (s + 1)2 + 1 are irreducible quadratics. We begin by determining
the s2 + 1 -chain. The roots of s2 + 1 are s = ±i. We need only focus on one
root and we will choose s = i.
According to Lemma 1 we can write
112 3 The Laplace Transform

30s + 40 B1 s + C1
= 2
(s2 + 1)2 (s2 + 2s + 2) (s + 1)2

30s+40  30i+40
where B1 i + C1 = (s2 +2s+2) s=i = 1+2i

(40+30i)(1−2i)
= (1+2i)(1−2i)

100−50i
= 5 = 20 − 10i

hence B1 = −10 and C1 = 20

p1 (s)
+
(s2 + 1)(s2 + 2s + 2)
1
and p1 (s) = s2 +1 (30s + 40−
(−10s + 20)(s2 + 2s + 2))
1 2
= s2 +1 (10s(s + 1))

= 10s.

We now repeat the process on

10s
.
(s2 + 1)(s2 + 2s + 2)
10s B2 s + C2
=
(s2 + 1)(s2 + 2s + 2) (s2 + 1)

10s 10i
where B2 i + C2 = =

(s2 +2s+2)  1+2i
s=i

(10i)(1−2i)
= (1+2i)(1−2i)

20+10i
= 5 = 4 + 2i

hence B2 = 2 and C2 = 4
p2 (s)
+
s2 + 2s + 2
1
and p2 (s) = s2 +1 (10s−
2
(2s + 4)(s + 2s + 2))
1
= s2 +1 (−2(s + 4)(s2 + 1))

= −2(s + 4).
We can now write down the s2 + 1 -chain.
 3.3 Partial Fractions: A Recursive Method for Irreducible Quadratics 113

The s2 + 1 -chain

30s + 40 −10s + 20
(s2 + 1)2 (s2 + 2s + 2) (s2 + 1)2
10s 2s + 4
(s2 + 1)(s2 + 2s + 2) (s2 + 1)
−2(s + 4)
(s2 + 2s + 2)

Since the last remainder term is already a partial fraction we obtain

30s + 40 −10s + 20 2s + 4 −2s − 8
= + 2 + .
(s2 + 1)2 (s2 + 2s + 2) (s2 + 1)2 s + 1 (s + 1)2 + 1
u
t
Example 4. Find the partial fraction decomposition for
4s + 2
.
(s + 1)2 (s2 + 1)
I Solution. We have a choice of computing the linear chain through s + 1
or the quadratic chain through s2 + 1. It is usually easier to compute linear
chains so we begin with the s + 1 -chain.
4s + 2 A1 4s+2 

−2
= where A1 = s2 +1 s=−1 = 2 = −1
(s + 1)2 (s2 + 1) (s + 1)2
p1 (s) 1
+ and p1 (s) = s+1 (4s + 2 − (−1)(s2 + 1))
(s + 1)(s2 + 1)
1 2
= s+1 (s + 4s + 3)
1
= s+1 (s + 1)(s + 3) = s + 3
s+3
We repeat the process on (s+1)(s2 +1) .

s+3 A2 s+3 

2
= where A2 = s2 +1 s=−1 = 2 =1
(s + 1)(s2 + 1) s+1
p2 (s) 1
+ and p2 (s) = s+1 (s + 3 − (1)(s2 + 1))
s2 + 1
1
= − s+1 (s2 − s − 2) = −s + 2

The s + 1 -chain is thus

114 3 The Laplace Transform

The s + 1 -chain

4s + 2 −1
(s + 1)2 (s2 + 1) (s + 1)2
s+3 1
(s + 1)(s2 + 1) s+1
−s + 2
s2 + 1

and the partial fraction decomposition is now immediate

4s + 2 −1 1 −s + 2
= + + 2
(s + 1)2 (s2 + 1) (s + 1)2 s+1 s +1
J

Exercises
s
1–6. For each exercise below compute 7.
(s2 + 1)(s − 3)
the chain table through the indicated ir-
reducible quadratic term. 4s
1 8.
1. ; (s2 + 1) (s2 + 1)2 (s + 1)
(s2 + 1)2 (s2 + 2)
s3 2
2. ; (s2 + 2) 9.
(s2 + 2)2 (s2 + 3) (s2 − 6s + 10)(s − 3)

8s + 8s2 30
3. ; (s2 + 3) 10.
(s2 + 3)3 (s2 + 1) (s2 − 4s + 13)(s − 1)

4s4 25
4. ; (s2 + 4) 11.
(s2 + 4)4 (s2 + 6) (s2 − 4s + 8)2 (s − 1)
1 s
5. ; (s2 + 12.
(s2 + 2s + 2)2 (s2 + 2s + 3)2 (s2 + 6s + 10)2 (s + 3)2
2s + 2)
s+1
5s − 5 13.
6. ; (s2 + (s2 + 4s + 5)2 (s2 + 4s + 6)2
(s2 + 2s + 2)2 (s2 + 4s + 5)
2s + 2)
s2
14. Hint: let u = s2 .
7–14. Find the decomposition of the (s2 + 5)3 (s2 + 6)2
given rational function into partial frac-
tions over R.
 3.4 Laplace Inversion and Exponential Polynomials 115

3.4 Laplace Inversion and Exponential Polynomials

Given a transform function F (s) we call an input function f (t) the inverse
Laplace transform of F (s) if L {f (t)} = F (s). We say the inverse Laplace
transform because in most circumstances it is unique. One such circumstance
is when the input function is continuous. We state this fact as a theorem. It’s
proof is beyond the level of this text.

Theorem 1. Suppose f1 (t) and f2 (t) are continuous functions with the same
Laplace transform. Then f1 (t) = f2 (t).

In Chapter 8 we will consider discontinuous input functions. For now though

we will assume that all input functions are continuous and we write L−1 {F (s)}
for the inverse Laplace transform. We can thus view L−1 as an operation on
transform functions that produces input functions.

Theorem 2 (Linearity). The inverse Laplace transform is linear. In other

words, if F (s) and G(s) are transform functions with continuous inverse
Laplace transforms and a and b are constants then

Linearity of the Inverse Laplace Transform

L−1 {aF (s) + bG(s)} = aL−1 {F (s)} + bL−1 {G(s)} .

Proof. Let f (t) = L−1 {F (s)} and g(t) = L−1 {G(s)}. Since the Laplace trans-
form is linear by Theorem 1 we have

L {(af (t) + bg(t)} = aL {f (t)} + bL {g(t)} = aF (s) + bG(s).

Since af (t) + bg(t) is continuous it follows that

L−1 {aF (s) + bG(s)} = af (t) + bg(t) = aL−1 {F (s)} + bL−1 {G(s)} .

u
t

Example 3. Find the inverse Laplace transform of the following functions:

1 1 1
, , and .
s−2 (s + 4)3 s2 + 9

I Solution. Each of these can be read from Table 3.1.

116 3 The Laplace Transform
n o
1
• L−1 s−2 = e2t .
n o n o
1 1 −1 3!
• L−1 (s+4)3 = 3! L (s−(−4))3 = 16 t2 e−4t .
n o n o
1
• L−1 s2 +9 = 13 L−1 3
s2 +32 = 1
3 sin 3t.
J

In order to use Table 3.1, we must sometimes write F (s) as a sum of

simpler proper rational functions, i.e. the partial fractions of F (s), and use
the linearity of L−1 . Consider the following example.
4s+2
Example 4. Suppose F (s) = (s+1)2 (s2 +1) . Find L−1 {F (s)} .

I Solution. Since F (s) does not appear in Table 3.1, the inverse Laplace
transform is not immediately evident. However, using the recursive partial
fraction method we found in Example 3.3.4 that
4s + 2 1 1 s 2
= − − 2 + .
(s + 1)2 (s2 + 1) (s + 1) (s + 1)2 (s + 1) s2 + 1
By linearity it is now evident that
 
−1 4s + 2
L = e−t − te−t − cos t + 2 sin t.
(s + 1)2 (s2 + 1)
J

Notice that the inverse Laplace transforms computed in the examples

above involve exponential functions, powers of t, and sin t and cos t. To be
able to speak succinctly of these kinds of functions we introduce the linear
space of exponential polynomials.

Exponential Polynomials
Let E (EC ) denote the set of continuous input functions whose Laplace trans-
form is in R (RC ), the real (complex) rational functions. We refer to E as
the space of real exponential functions and EC as the space of complex
exponential functions. The reason for the use of this terminology will be
made clear below. We note that since R ⊂ RC it follows that E ⊂ EC . As
a first step in describing these spaces more explicitly we note the following
theorem:

Theorem 5. E is a real linear space and EC is a complex linear space.

Proof. Suppose f1 and f2 are in E. Then L {f1 } (s) and L {f2 } (s) are in R
and since R is a linear space so is the sum L {f1 } (s) + L {f2 } (s). Since the
Laplace transform is linear, Theorem 1, we have
 3.4 Laplace Inversion and Exponential Polynomials 117

L {f1 + f2 } (s) = L {f2 } (s) + L {f2 } (s) ∈ R.

This implies f1 + f2 ∈ E. Now suppose c ∈ R and f ∈ E. Then L {cf } (s) =

cL {f } (s) ∈ R by Theorem 1 and since R is a linear space. It follows that E
is a linear space. The complex case is done similarly. u
t

The notion of a linear space may be new to the reader and likewise the
nature of the proof given above. However, knowing that E and EC are closed
under addition and scalar multiplication is important in getting an idea of the
kind of functions it can contain. We’ll separate the real and complex cases.
Again, the complex case is easier.

Complex Exponential Functions. By repeated application of the recur-

sive partial fraction method, Lemma 3 of Section 3.2, we see that every com-
plex rational function can be written as a linear combination of terms of the
form
1
,
(s − λ)n
called simple rational functions, where λ ∈ C and n is a positive integer.
According to Table 3.1 we have

1 tn−1 λt
L−1 { n
}= e
(s − λ) (n − 1)!

We therefore get the following theorem.

Theorem 6. Every complex exponential polynomial in EC can be written as a
finite linear combinations (with complex scalars) of terms of the form tk eλt .
We can also write each f ∈ EC as
k
X
f (t) = fi (t)eλi t , (1)
i=1

where each fi is a complex polynomial and each λi is a complex number.

The expression ‘complex exponential polynomial’ comes from Equation 1:
functions in EC are sums of products of polynomials with exponential func-
tions. A simple (complex) exponential polynomial is a term of the form
tk eλt , k a nonnegative integer and λ ∈ C. Our discussion thus far also gives

Theorem 7. The Laplace transform

L : EC → RC

establishes a one-to-one correspondence between EC and RC . Specifically, for

each proper rational function pq ∈ RC there is one and only one function
f ∈ EC such that Lf = pq .
118 3 The Laplace Transform

Real Exponential Functions. We now wish to describe the linear space

E of real exponential polynomials, that is, the input functions with Laplace
transform in R. Let p(s) p(s)
q(s) ∈ R. Since R ⊂ RC we can regard q(s) ∈ RC and
use Theorem 6 to write
  k
p(s) X
f (t) = L−1 = fi (t)eλi t ,
q(s) i=1

where each fi is a complex polynomial and each λi is a complex number.

However, according to Theorem 7 of Section 3.1, f must be a real valued
function. Therefore
k
X
Re fi (t)eλi t . (2)


f (t) = Re f (t) =
i=1

Suppose g(t) = g1 (t) + ig2 (t) is a complex polynomial with g1 and g2 its real
and imaginary parts. Suppose λ = a + ib. Using Euler’s formula we get

Re(g(t)eλt ) = Re((g1 (t) + ig2 (t))(eat cos bt + ieat sin bt))

= g1 (t)eat cos bt − g2 (t)eat sin bt.

Applying this calculation to Equation 2 gives

Theorem 8. The linear space, E, is the set of all real linear combinations of
functions of the form

p(t)eat cos bt and p(t)eat sin bt,

where p(t) is a real polynomial and a and b are real numbers. Alternately, E,
is the set of all linear combinations of functions of the form

tn eat cos bt and tn eat sin bt,

where n is a nonnegative integer and a and b are real numbers.

We will refer to terms of the form tn eat cos bt and tn eat sin bt as simple
(real) exponential polynomials. Note that if b = 0 then cos bt = 1 and
sin bt = 0. Thus the terms tn eat , with a real, is a simple exponential polyno-
mial.

Example 9. Determine which of the following are exponential polynomials

over the real or complex numbers and which are simple.

1. 1 + t + t2 + t3 3. t2 et + 2it sin 3t 6. t1/2

1 4. ln t 7. et/2
2.
t 5. cos 2t + i sin 2t 8. cosh 3t
 3.4 Laplace Inversion and Exponential Polynomials 119

9. et
2
12. tan 2t e3t
15.
10. t4 et e−3it 13. 1
e(1+2i)t + e(1−2i)t

cos 6t
2
sin 5t
11. e2t 14. (t2 − t−2 ) sin t

I Solution.

1. in E 6. not in E nor EC 11. simple in E

2. not in E nor EC 7. simple in E and EC 12. not in E nor EC
3. in EC , not in E 8. in E 13. simple in E
4. not in E nor EC 9. not in E nor EC 14. not in E nor EC
5. simple in EC 10. simple in EC 15. not in E nor EC

Comments:
5. By Euler’s formula cos 2t + i sin 2t = e2it
e3t +e−3t
8. cosh 3t = 2

10. t4 et e−3it = t4 e(1−3i)t

sin 5t
11. e2t = e−2t sin 5t
1
13. By Euler’s formula e(1+2i)t + e(1−2i)t = et cos 2t


2

Recall also that all functions in E are also in EC . J

Exercises

1–23. Compute L−1 {F (s)} (t) for the −2s

8.
given proper rational function F (s). 3s2 + 2
2
−5 9.
1. s2 + 3
s
3 3s + 2
2. 10.
s−4 3s2 + 2

4 1
3. 11.
s − 2i s2 + 6s + 9

−3 2s − 5
4. 12.
s+1+i s2 + 6s + 9

3 4 2s − 5
5. − 3 13.
s2 s (s + 3)3
4 6
6. 14.
2s + 3 s2 + 2s − 8
3s s
7. 15.
s2 + 4 s2 − 5s + 6
120 3 The Laplace Transform

2s2 − 5s + 1 27. et /t
16.
(s − 2)4
28. t sin(4t − π4 )
2s + 6
17. 2
s − 6s + 5 29. (t + et )2
4s2 30. (t + et )−2
18.
(s − 1)2 (s + 1)2
31. tet/2
27
19. 3
s (s + 3) 32. t1/2 et
8s + 16 33. sin 2t/e2t
20.
(s2 + 4)(s − 2)2
34. e2t / sin 2t
5s + 15
21. 35–38. Verify the following closure
(s2 + 9)(s − 1)
properties of the linear space of exponen-
12 tial polynomials.
22.
s2 (s + 1)(s − 2)
35. Multiplication. Show that if f and g
2s are in E (EC ), then so is f g.
23.
(s − 3)3 (s − 4)2
36. Differentiation. Show that if f is in
24–34. Determine which of the follow- E (EC ,) then so is the derivative f 0 .
ing functions are in the linear space E of
exponential polynomials. 37. Integration. Show that if f is in E ,
2 −2t
24. t e (EC ) then so is f (t) dt.

25. t−2 e2t 38. Show that E (EC ) is not closed under
inversion. That is, find a function f
26. t/et so that 1/f is not in E (EC ).

3.5 Laplace Inversion involving Irreducible Quadratics

Let’s now return to our discussion about the inverse Laplace transform. By
the recursive partial fraction method each (real or complex) rational function
p(s) 1
q(s) can be written as a linear combination of terms of the form (s−λ)k , where
λ is a (possibly complex) root of q and k is a positive integers less than or
equal to the multiplicity of λ in q(s). The inversion formula
tk
 
1
L−1 k
= eλt . (1)
(s − λ) k!
and linearity then gives a straightforward means of computing L−1 {p(s)/q(s)}.
Of course, all of this assumes that one knows the factorization of q(s).
Now, for each real rational function we would like to express its inverse
Laplace transform as a linear combination of simple real exponential poly-
nomials. The partial fraction decomposition for a real rational function is a
 3.5 Laplace Inversion involving Irreducible Quadratics 121

linear combination of the simple real rational functions

1 1 s
, , and , (2)
(s − a)k ((s − a)2 + b2 )k ((s − a)2 + b2 )k
where a, b are real and b > 0. The inversion of the first simple rational function
is straightforward, it is given by Equation 1 with λ = a. However, when
powers of irreducible quadratics are involved the inversion is a little more
complicated. To facilitate their inversion we introduce two Laplace transform
tools. Applications of these tools will reduce powers of quadratics to the cases
1 s
and . (3)
(s2 + 1)k (s2 + 1)k
We will discuss how each of these can be handled later on in this section.
The first tool is a new way of looking at the first translation principle for the
Laplace transform. By inverting both side of the formula in Theorem 12 of
Section 3.1 we get the following equivalent theorem.
Theorem 1. Suppose F (s) is a function in transform space. Then

First Translation Principle

L−1 {F (s − a)} = eat L−1 {F (s)} .

The First Translation Principle allows us to remove the shift by a in the

rational functions of Equation (2) to get a linear combination of simple partial
fractions of the from
1 s
and . (4)
(s2 + b2 )k (s2 + b2 )k
Consider the following examples where k = 1. (The case k > 1 is similar and
an example of this type will be considered later in this section.)
Example 2. Find the inverse Laplace transform of
1
.
s2 + 6s + 25
I Solution. Completing the square of the denominator gives: s2 + 6s+ 25 =
s2 + 6s + 9 + 25 − 9 = (s + 3)2 + 42 . With a = −3 in the First Translation
Principle we get
   
1 1
L−1 = L −1
s2 + 6s + 25 (s − −3)2 + 42
 
1
= e−3t L−1
s2 + 4 2
1
= e−3t sin 4t.
4
122 3 The Laplace Transform

The third line comes directly from Table 3.1. J

Example 3. Find the inverse Laplace transform of

3s + 2
.
s2 − 4s + 7
I Solution. Completing the square of the denominator gives s2 − 4s + 7 =
√ 2
s2 − 4s + 4 + 3 = (s − 2)2 + 3 . In order to apply the First Translation
Principle with a = 2 the numerator must also be rewritten with s translated
by 2. Thus 3s + 2 = 3(s − 2 + 2) + 2 = 3(s − 2) + 8. We now get
  ( )
−1 3s + 2 −1 3(s − 2) + 8
L =L √ 2
s2 − 4s + 7 (s − 2)2 + 3
( )
3s + 8
= e2t L−1 √ 2
s2 + 3
( ) ( √ )!
2t −1 s 8 −1 3
=e 3L √ 2 +√ L √ 2
s2 + 3 3 s2 + 3
√ √
 
2t 8
=e 3 cos 3t + √ sin 3t .
3
J

Notice that in each of these examples after the First Translation Principle
is applied the inversions reduce to expressions of the form given in Equation
(4). We now introduce a tool that reduces these simple partial fractions to
the case where b = 1.
Suppose f (t) is an function and b is a positive real number. The function
g(t) = f (bt) is called a dilation of f by b. If the domain of f includes
[0, ∞) then so does any dilation of f since b is positive. The following theorem
describes the Laplace transform of a dilation.

Theorem 4. Suppose f (t) is an input function and b is a positive real number.

Then

The Dilation Principle

bL {f (bt))} (s) = L {f (t)} ( sb ).

Proof. This result follows from the definition of the Laplace transform:
 3.5 Laplace Inversion involving Irreducible Quadratics 123
Z ∞
bL {f (bt)} (s) = b e−st f (bt) dt
Z0 ∞
s dr
=b e− b r f (r)
0 b
s
= L {f (t)} .
b
To get the second line we made the change of variable t = rb . Since b > 0 the
limits of integration remain unchanged. u
t

To illustrate the Dilation Principle consider the following verifications.

b 1
For f (t) = et : bL ebt (s) = s−b = L {et } sb .


• = ( s −1)
b

2
b 1 s
For f (t) = cos t:


• bL {cos bt} (s) = s2 +b2 = 2 = L {cos t} b .
(( sb ) +1)
s
bs s
For f (t) = sin t:


• bL {sin bt} (s) = s2 +b2 = b
2 = L {sin t} b .
(( sb ) +1)

 sin t
Example 5. Given that L t = cot−1 (s) determine
 
sin bt
L (s).
t

I Solution. By linearity and the Dilation principle we have

   
sin bt sin bt
L (s) = bL (s)
t bt
 
sin t
=L (s/b)
t
= cot−1 (s/b).

We now consider an application of the Dilation Principle that reduces

Equation (4) to Equation (3).
n o n o
1 s
Proposition 6. Set A(t) = L−1 (s2 +1) k+1 and B(t) = L−1 (s2 +1) k+1 .
Then
 
−1 b 1
L (t) = 2k A(bt)
(s2 + b2 )k+1 b
 
−1 s 1
L (t) = B(bt)
(s + b2 )k+1
2 b2k
124 3 The Laplace Transform

Proof. By the Dilation Principle we have

1 1
2k
L {A(bt)} (s) = 2k+1 bL {A(bt)} (s)
b b
1 s
= 2k+1 L {A(t)}
b b
b 1
= 2(k+1) s 2
b (( b ) + 1)k+1
b
= 2 .
(s + b2 )k+1

The first formula now follows by taking the inverse Laplace transform of both
side. The second formula is done similarly. u
t

A Recursive Method. We now derive a recursive method for computing

the inverse Laplace transform of
1 s
and .
(s2 + 1)k+1 (s2 + 1)k+1

Let Ak (t) and Bk (t) be functions such that

   
1 1 s 1
L−1 = Ak (t) and L−1
= k Bk (t).
(s2 + 1)k+1 2k k! (s2 + 1)k+1 2 k!

The shift by one in k and the constant 2k1k! are intentional and occur to make
the formulas for Ak and Bk that we derive below simpler. The following table
gives the formulas for Ak and Bk , for k = 0, 1, and 2. The entries for k = 0
follows directly from Table 3.1. The k = 1 case follows from Example 3.1.17.
The k = 2 case will be verified below.

k F (s) Ak (t)
1
k=0 s2 +1 sin t
1
k=1 (s2 +1)2 sin t − t cos t
1
k=2 (s2 +1)3 (3 − t2 ) sin t − 3t cos t

k F (s) Bk (t)
s
k=0 s2 +1 cos t
s
k=1 (s2 +1)2 t sin t
s
k=2 (s2 +1)3 t sin t − t2 cos t
 3.5 Laplace Inversion involving Irreducible Quadratics 125

Multiplying Ak or Bk by the constant 2k1k! gives the corresponding inverse

Laplace transform. For example, with k = 2, the constant is 81 and
 
1 1
L−1 = ((3 − t2 ) sin t − 3t cos t)
(s2 + 1)3 8
 
s 1
and L−1 = (t sin t − t2 cos t).
(s2 + 1)3 8
The following theorem describes how Ak and Bk can be recursively de-
scribed.
Theorem 7. With notation as above the functions Ak and Bk satisfy the
following recursion formulas:
Ak+2 (t) = (2k + 3)Ak+1 (t) − t2 Ak (t)

Bk+2 (t) = (2k + 1)Bk+1 (t) − t2 Bk (t).

Furthermore,
Bk+1 (t) = tAk (t).
Remark 8. A recursion relation of this type implies that once the first two
terms are known all further terms can be derived. For example to compute
A2 let k = 0 in the recursion relation and use the fact that A0 and A1 are
known. We then get
A2 (t) = 3A1 (t) − t2 A0 (t)
= 3(sin t − t cos t) − t2 (sin t)
= (3 − t2 ) sin t − 3t cos t.
Similarly, the recursion relation for Bk gives
B2 (t) = B1 (t) − t2 B0 (t)
= t sin t − t2 cos t
or, alternately, using the relation Bk+1 (t) = tAk (t) gives
B2 (t) = tA1 (t)
= t sin t − t2 cos t.
Now that A2 is known the recursion formulas can be again applied with k = 1
to derive A3 in terms of A2 and A1 and similarly for B3 . Repeated applications
of recursion gives An and Bn , for any n.

We also remark that it is possible to derive explicit formulas both Ak

and Bk . This will be done by solving a certain second order (non constant
coefficient) differential equation. We will explore this in detail in Section ??.
126 3 The Laplace Transform

Proof. By the Transform Derivative Principle, Theorem 16 of Section 3.1, we

have
1 d 1
L {tAk (t)} (s) = −
2k k! ds (s2 + 1)k+1
(k + 1)(s2 + 1)k 2s
=
(s2 + 1)2k+2
2(k + 1)s
= 2
(s + 1)k+2
2(k + 1)
= k+1 L {Bk+1 (t)} (s)
2 (k + 1)!
1
= k L {Bk+1 (t)} (s).
2 k!
By canceling the constant and applying Theorem 3.4.1 we get

Bk+1 (t) = tAk (t). (5)

Again we apply the Transform Derivative Principle to get

1 d s
L {tBk (t)} (s) = −
2k k! ds (s + 1)k+1
2

(k + 1)(s2 + 1)k 2s2 − (s2 + 1)k+1

=
(s2 + 1)2k+2
(2k + 1)s2 − 1
=
(s2 + 1)k+2
(2k + 1)(s2 + 1) − 2(k + 1)
=
(s2 + 1)k+2
1 1
= (2k + 1) 2 − 2(k + 1) 2
(s + 1)k+1 (s + 1)k+2
1 2(k + 1)
= (2k + 1) k L {Ak (t)} − k+1 L {Ak+1 (t)}
2 k! 2 (k + 1)!
1
= ((2k + 1)L {Ak (t)} − L {Ak+1 (t)}) .
2k k!
It now follows that tBk (t) = (2k + 1)Ak (t) − Ak+1 (t). Now replace k by k + 1
and solve for Ak+2 . Then

Ak+2 = (2k + 3)Ak+1 − tBk+1

= (2k + 3)Ak+1 − t2 Ak by Equation (5).

Now multiply the recursion relation for Ak by t and use Equation (5) again
to get the recursion relation for Bk . u
t
 3.5 Laplace Inversion involving Irreducible Quadratics 127

Example 9. Find the following inverse Laplace transforms:

   
1 s
L−1 (t) and L −1
(t).
(s2 + 1)4 (s2 + 1)4

I Solution. We use Theorem 7 with k = 3 in each case we get

A3 = 5A2 − t2 A1
= 5((3t2 ) sin t − 3t cos t) − t2 (sin t − t cos t)
= (15 − 6t2 ) sin t + (t3 − 15t) cos t

and

B3 = tA2 = (3t − t3 ) sin t − 3t2 cos t.

1 1
Also 2k k! |k=3 = 48 . Hence
 
1 1
(15 − 6t2 ) sin t + (t3 − 15t) cos t


L (s) =
(s2 + 1)4 48
 
s 1
and (3t − t3 ) sin t − 3t2 cos t .


L (s) =
(s2 + 1)4 48
J

We now consider a final example in this section where the First Translation
Principle, the Transform Derivative Principle, and the results of the above
recursion formula are used together.

Example 10. Find the inverse Laplace transform of

6s + 6
F (s) = .
(s2 − 4s + 13)3

I Solution. We first complete the square: s2 − 4s + 13 = s2 − 4s + 4 + 9 =

(s − 2)2 + 32 .
   
−1 −1 6s + 6 −1 6(s − 2) + 18
L {F (s)} = L (t) = L (t)
((s − 2)2 + 32 )3 ((s − 2)2 + 32 )3
     
s 3
= e2t 6L−1 2 2 3
(t) + 6L−1 (t)
(s + 3 ) (s + 32 )3
2
     
2t 6 −1 s 6 −1 1
=e L (3t) + 4 L (3t)
34 (s2 + 1)3 3 (s2 + 1)3
2e2t 1
 
2 1 2
= 3 ((3 − 9t ) sin 3t − 9t cos 3t) + (3t sin 3t − 9t cos 3t)
3 8 8
e2t
(3 + 3t − 9t2 ) sin 3t − (9t + 9t2 ) cos 3t .


=
108
128 3 The Laplace Transform

On the first line we rewrite the numerator so we can apply the First Trans-
lation Principle to get line 2. On line 3 we applied Proposition 6 and on line
4 we use the table above (note the evaluation at 3t; all instances of t in the
formula are replaced with 3t). Line 5 is a simplification. J

Below we apply Theorem 7 to construct a table of (inverse) Laplace trans-

forms.

 
1
k L−1
(s2 + 1)k+1

0 sin t
1
1 2 (sin t − t cos t)
1
(3 − t2 ) sin t − 3t cos t


2 8
1
(15 − 6t2 ) sin t − (15t − t3 ) cos t


3 48
1
(105 − 45t2 + t4 ) sin t − (105t − 10t3 ) cos t


4 384
1
(945 − 420t2 + 15t4 ) sin t − (945t − 105t3 + t4 ) cos t


5 3840
1
6 46080 (10395 − 4725t2 + 210t4 − t6 ) sin t

−(10395t − 1260t3 + 21t4 ) cos t

 3.5 Laplace Inversion involving Irreducible Quadratics 129

 
−1 s
k L
(s2 + 1)k+1

0 cos t
1
1 2 t sin t
1
t sin t − t2 cos t


2 8
1
(3t − t3 ) sin t − 3t2 cos t


3 48
1
(15t − 6t3 ) sin t − (15t2 − t4 ) cos t


4 384
1
(105t − 45t3 + t5 ) sin t − (105t2 − 10t4 ) cos t


5 3840
1
(945t − 420t3 + 15t5 ) sin t − (945t2 − 105t4 + t6 ) cos t


6 46080

Exercises
1–6. Use the First Translation Principle 8. Ei(6t); given
and Table 3.1 to find the inverse Laplace ln(s + 1)
transform. L {Ei(t)} =
s
2s
1. 2 9–18. Find the inverse Laplace trans-
s + 2s + 5
form of each rational function.
1 8s
2. 9.
s2 + 6s + 10 (s2 + 4)2
s−1 9
3. 10.
s2 − 8s + 17 (s2 + 9)2
2s + 4 2s
4. 11.
s2 − 4s + 12 (s2 + 4s + 5)2
s−1
5. 2s + 2
s2 − 2s + 10 12.
(s2 − 6s + 10)2
s−5
6. 2s
s2 − 6s + 13 13.
(s2 + 8s + 17)2
7–8. Use the Dilation Principle to find
the Laplace transform of each function. s+1
14.
The given Laplace transforms will be es- (s2 + 2s + 2)3
tablished later.
1 − cos 5t 1
7. ; given 15.
t (s2 − 2s + 5)3
1 − cos t 1 s2 8s
L
= ln 16.
 2 s + 1 
t 2 (s2 − 6s + 10)3
130 3 The Laplace Transform
s−4 1
17. 19.
(s2 − 8s + 17)4 (s2 + 1)5
2 s
18. 20.
(s2 + 4s + 8)3 (s2 + 1)5
19–20. Use Theorem 7 to compute the
following inverse Laplace transforms.

3.6 The Laplace Transform Method

The Laplace transform is particularly well suited for solving certain types of
differential equations, namely the constant coefficient linear differential
equations

an y (n) + an−1 y (n−1) + · · · + a1 y 0 + a0 y = f (t), (1)

where a0 , . . ., an are constants, the function f (t) is continuous, and the initial
values of the unknown function y(t) are also specified:

y(0) = y0 , y 0 (0) = y1 , . . . , y (n−1) (0) = yn−1 .

Equation (1), with the initial values of y(t), is known as an initial value
problem. In this chapter and the next we will assume f is an exponential
polynomial. The basic theory of this type of differential equation will be dis-
cussed in detail in Chapter 4. For now, we will only study how the Laplace
transform leads very quickly to a formula for y(t).
The Laplace transform method for solving Equation (1) is based on the
linearity property of the Laplace transform (Theorem 3.1.1) and the following
formula which expresses the Laplace transform of the derivative of a function.

Theorem 1. Suppose f (t) is an exponential polynomial and F (s) = L {f (t)} .

Then f 0 (t) is an exponential polynomial and

Input Derivative Principle:

The First Derivative

L {f 0 (t)} = sF (s) − f (0)

Proof. By Exercise 3.4.36 if f is an exponential polynomial then so is f 0 (t). We

apply integration by parts to the improper integral defining L {f 0 (t)}, taking
∞
into account the convention that g(t)|0 is a shorthand for limt→∞ (g(t)−g(0)),
provided the limit exists. Let u = e−st and dv = f 0 (t) dt then
 3.6 The Laplace Transform Method 131
Z ∞
L {f 0 (t)} (s) = e−st f 0 (t) dt
0
Z ∞

∞
= f (t)e−st 0 − (−s)e−st f (t) dt
0
Z ∞
−st
= −f (0) + e f (t) dt
0
= sL {f (t)} (s) − f (0).

The transition from the second to the third line is a result of the fact that
exponential polynomials, f (t), satisfy limt→∞ f (t)e−st = 0, for s large. (See
Equation 3.1.5.) u
t
Example 2. Verify the validity of Theorem 1 for the following functions:

f (t) = 1, f (t) = eat , and f (t) = cos 3t.

I Solution.

• If f (t) = 1, then f 0 (t) = 0 so L {f 0 (t)} = 0, and

1
sF (s) − f (0) = s − 1 = 0 = L {f 0 (t)} .
s
a
• If f (t) = eat , then f 0 (t) = aeat so L {f 0 (t)} = and
s−a
s a
sF (s) − f (0) = −1= = L {f 0 (t)} .
s−a s−a
9
• If f (t) = cos 3t then f 0 (t) = −3 sin 3t so L {f 0 (t)} = − and
s2 +9
s 9
sF (s) − f (0) = s −1=− 2 = L {f 0 (t)} .
s2 + 9 s +9
J
Example 3. Solve the first order linear differential equation:

y 0 − 3y = 1, y(0) = 1

I Solution. As is our convention, let Y (s) = L {y(t)}. We apply the Laplace

transform to both sides of the given equation and use linearity (Theorem 3.1.1)
and the input derivative formula (Theorem 1) just verified to get

L {y 0 − 3y} = L {1}
1
L {y 0 } − 3L {y} =
s
1
sY (s) − y(0) − 3Y (s) = .
s
132 3 The Laplace Transform

We now have an algebraic equation for which we solve for Y (s). Since y(0) = 1
we get  
1 1 1 1
Y (s) = 1+ = + .
s−3 s s − 3 s(s − 3)
A partial fraction decomposition gives
1 1 1 11 4 1 11
Y (s) = + − = − .
s−3 3s−3 3s 3s−3 3s
We recover y(t) from Y (s) by Laplace inversion. We obtain
   
4 1 1 1 4 1
y(t) = L−1 − L−1 = e3t − .
3 s−3 3 s 3 3
J
Let’s consider another example.
Example 4. Solve the first order linear differential equation

y 0 + y = sin t, y(0) = 0.

I Solution. Letting Y (s) = L {y(t)}, we equate the Laplace transform of

each side of the equation to obtain
1
(s + 1)Y (s) = .
s2 +1
Solving for Y (s) and decomposing Y (s) into partial fractions gives
 
1 1 s 1
Y (s) = − + .
2 s + 1 s2 + 1 s2 + 1
Inversion of the Laplace transform gives
1 −t
y(t) = L−1 {Y (s)} =


e − cos t + sin t .
2
J
As we have already remarked if f (t) is an exponential polynomial so is
f 0 (t). Inductively, all orders of the derivative of f (t) are exponential polyno-
mials. Thus we may apply Theorem 1 with f (t) replaced by f 0 (t) (so that
(f 0 )0 = f 00 ) to get

L {f 00 (t)} = sL {f 0 (t)} − f 0 (0)

= s (sL {f (t)} − f (0)) − f 0 (0)
= s2 L {f (t)} − sf (0) − f 0 (0).

Thus we have arrived at the following formula for expressing the Laplace
transform of f 00 (t) in terms of L {f (t)} and the initial values f (0) and f 0 (0).
 3.6 The Laplace Transform Method 133

Corollary 5. Suppose f (t) is an exponential polynomial. Then

Input Derivative Principle:

The Second Derivative

L {f 00 (t)} = s2 F (s) − sf (0) − f 0 (0).

The process used to determine Corollary 5 for the Laplace transform of a

second derivative can be repeated to arrive at a formula for the Laplace trans-
form of the nth derivative of an exponential polynomial, which we summarize
in the following theorem.

Theorem 6. Suppose f (t) is an exponential polynomial and L {f (t)} = F (s).

Suppose n ≥ 1. Then

Input Derivative Principle:

The nth Derivative

L f (n) (t) = sn F (s) − sn−1 f (0) − sn−2 f 0 (0) − · · · − f (n−1) (0).



For n = 3 and n = 4, this formula becomes

L {f 000 (t)} = s3 F (s) − s2 f (0) − sf 0 (0) − f 00 (0)

n o
and L f (4) (t) = s4 F (s) − s3 f (0) − s2 f 0 (0) − sf 00 (0) − f 000 (0).

If f (0) = f 0 (0) = · · · = f (n−1) = 0 then Theorem 6 has the particularly

simple form
L {f 0 (t)} = sn F (s).
In words, the operation of differentiating n-times on the space of elementary
functions with derivatives (up to order n − 1) vanishing at 0, corresponds,
under the Laplace transform, to the algebraic operation of multiplying by sn
on the space Rpr (s) of proper rational functions.
Let’s consider some examples of how Theorem 6 is used to solve some
types of differential equations.

Example 7. Solve the initial value problem

y 00 − y = 0, y(0) = 0, y 0 (0) = 1.
134 3 The Laplace Transform

I Solution. As usual, let Y (s) = L {y(t)} and apply the Laplace transform
to both sides of the differential equation to obtain

s2 Y (s) − 1 − Y (s) = 0.

Now solve for Y (s) and decompose in partial fractions to get

1 1 1 1 1
Y (s) = = − .
s2 − 1 2s−1 2s+1
Then applying the inverse Laplace transform to Y (s) gives
1 t
y(t) = L−1 {Y (s)} = (e − e−t ).
2
J
Example 8. Solve the initial value problem

y 00 + 4y 0 + 4y = 2te−2t , y(0) = 1, y 0 (0) = −3.

I Solution. Let Y (s) = L {y(t)} where, as usual, y(t) is the unknown solu-
tion. Applying the Laplace transform gives the algebraic equation
2
s2 Y (s) − s + 3 + 4(sY (s) − 1) + 4Y (s) = ,
(s + 2)2
which can be solved for Y (s) to give
s+1 2
Y (s) = 2
+ .
(s + 2) (s + 2)4
The partial fraction decomposition gives
1 1 2
Y (s) = − + .
s + 2 (s + 2)2 (s + 2)4
Taking the inverse Laplace transform y(t) = L−1 {Y (s)} then gives
1
y(t) = e−2t − te−2t + t2 e−2t .
3
J
It is worth pointing out in this last example that in solving for Y (s) we kept
s+1
the part of Y (s) that came from the initial values, namely , distinct
(s + 2)2
2
from that determined by the right-hand side of the equation, namely .
(s + 2)4
By not combining these into a single proper rational function before computing
the partial fraction decomposition, we have simplified the computation of the
partial fractions. This is a typical situation and one that you should be aware
of when working on exercises.
 3.6 The Laplace Transform Method 135

Example 9. Solve the initial value problem

y 00 + β 2 y = cos ωt, y(0) = y 0 (0) = 0,

where we assume that β 6= 0 and ω 6= 0.

I Solution. Let Y (s) = L {y(t)} and apply the Laplace transform to the
equation. We then solve algebraically for Y (s) to get
s
Y (s) = .
(s2 + β 2 )(s2 + ω 2 )

We will break our analysis into two cases: (1) β 2 6= ω 2 and (2) β 2 = ω 2 .

Case 1: β 2 6= ω 2 .
In this case we leave it as an exercise to verify that the partial fraction
decomposition of Y (s) is
 
1 s s
Y (s) = 2 − 2 ,
ω − β 2 s2 + β 2 s + ω2

so that the solution y(t) = L−1 {Y (s)} is

cos βt − cos ωt
y(t) = .
ω2 − β2

Case 2: β 2 = ω 2 .
In this case
s
Y (s) = .
(s2 + ω 2 )2
Applying the Dilation Principle and Theorem 3.5.7 gives
1
y(t) = t sin ωt.
2ω
J

Example 10. Solve the initial value problem

y 000 − y 00 + y 0 − y = 10e2t , y(0) = y 0 (0) = y 00 (0) = 0.

I Solution. Let Y (s) = L {y(t)} where y(t) is the unknown solution. We

apply the Laplace transform L to get
10
s3 Y (s) − s2 Y (s) + sY (s) − Y (s) = .
s−2
Solving for Y (s) and applying the partial fraction decomposition gives
136 3 The Laplace Transform

10
Y (s) =
(s3 − s2 + s − 1)(s − 2)
10
= 2
(s − 1)(s + 1)(s − 2)
−5 2 1 + 3s
= + + .
s − 1 s − 2 s2 + 1
Now apply the inverse Laplace transform to get
y(t) = −5et + 2e2t + sin t + 3 cos t.
J
We conclude this section by looking at what the Laplace transform tells us
about the solution of the second order linear constant coefficient differential
equation
ay 00 + by 0 + cy = f (t), y(0) = y0 , y 0 (0) = y1 , (2)
where f (t) is an exponential polynomial, and a, b, and c are real constants.
Applying the Laplace transform to Equation (2) (where Y (s) is the Laplace
transform of the unknown function y(t), as usual) gives
a(s2 Y (s) − sy0 − y1 ) + b(sY (s) − y0 ) + cY (s) = F (s).
If we let P (s) = as2 +bs+c (P (s) is known as the characteristic polynomial
of the differential equation), then the above equation can be solved for Y (s)
in the form
(as + b)y0 + ay1 F (s)
Y (s) = + = Y1 (s) + Y2 (s). (3)
P (s) P (s)
Notice that Y1 (s) depends only on P (s), which is determined by the left-
hand side of the differential equation, and the initial values y0 and y1 , while
Y2 (s) depends only on P (s) and the function f (t) on the right-hand side of
the equation. The function f (t) is usually called the input function for the
differential equation. Taking inverse Laplace transforms we can write
y(t) = L−1 {Y (s)} = L−1 {Y1 (s)} + L−1 {Y2 (s)} = y1 (t) + y2 (t).
The function y1 (t) is the solution of Equation (2) obtained by taking f (t) = 0,
while y2 (t) is the solution obtained by specifying that the initial conditions be
zero, i.e., y0 = y1 = 0. Thus, y1 (t) is referred to as the zero-input solution,
while y2 (t) is referred to as the zero-state solution. The terminology comes
from engineering applications. You will be asked to compute further examples
in the exercises, and addtional consequences of Equation (3) will be developed
in Chapter 5.

Exercises
 3.6 The Laplace Transform Method 137

1–31. For the following differential 20. y 00 + 2y 0 − 3y = 13 sin 2t,

equations use the Laplace transform y(0) = 0, y 0 (0) = 0
method. Determine both Y (s) and the
solution y(t). 21. y 00 + 6y 0 + 9y = 50 sin t,
y(0) = 0, y 0 (0) = 2
1. y 0 − 4y = 0, y(0) = 2
22. y 00 + 25y = 0,
0
2. y − 4y = 3, y(0) = 2 y(0) = 1, y 0 (0) = −1

3. y 0 − 4y = 16t, y(0) = 2 23. y 00 + 8y 0 + 16y = 0,

y(0) = 21 , y 0 (0) = 2
4. y 0 − 4y = e4t , y(0) = 0
24. y 00 − 4y 0 + 4y = 4e2t , y(0) =
5. y 0 + 2y = 3et , y(0) = 2 −1, y 0 (0) = −4
6. y 0 + 2y = te−2t , y(0) = 0 25. y 00 + y 0 + y = 0,
y(0) = 1, y 0 (0) = 0
7. y 0 + 9y = 81t2 , y(0) = 1
26. y 00 + 4y 0 + 4y = 16t sin 2t,
8. y 0 − 3y = cos t, y(0) = 0 y(0) = 1 y 0 (0) = 0
9. y 0 − 3y = 50 sin t, y(0) = 1 27. y 00 + 4y = sin 3t,
y(0) = 0 y 0 (0) = 1.
10. y 0 + y = 2t sin t, y(0) = 1
28. y 000 − y 00 = t,
11. y 0 − 3y = 2e3t cos t, y(0) = 1
y(0) = 0, y 0 (0) = 1, y 00 (0) = 0
12. y 0 − y = 10e2t sin 3t, y(0) = −2
29. y 000 − y 00 + y 0 − y = t,
13. y 00 + 3y 0 + 2y = e2t , y(0) = 0, y 0 (0) = 0, y 00 (0) = 0
y(0) = 1, y 0 (0) = 1
30. y 000 − y 0 = 6 − 3t2 ,
00 0
14. y − 4y − 5y = 25t, y(0) = 1, y 0 (0) = 1, y 00 (0) = 1
y(0) = −1 y 0 (0) = 0 31. y (4) − y = 0, y(0) = 1, y 0 (0) =
15. y 00 + 4y = 8, 0, y 00 (0) = 0, y 000 (0) = 0
y(0) = 2, y 0 (0) = 1 32–35. For each of the following differ-
00 0
16. y + 4y + 4y = e , −2t ential equations, find the zero-state solu-
y(0) = 0 y 0 (0) = 1 tion. Recall that the zero-state solution
is the solution with all initial conditions
17. y 00 + 4y 0 + 4y = 4 cos 2t, equal to zero.
y(0) = 0 y 0 (0) = 1
32. y 00 + 4y 0 + 13y = 0
00 0
18. y − 3y + 2y = 4, 33. y 00 + 4y 0 + 3y = 6
y(0) = 2, y 0 (0) = 3
34. y 00 − y = cos 3t
19. y 00 − 3y 0 + 2y = et ,
y(0) = −3, y 0 (0) = 0 35. y 00 + y = 4t sin t
138 3 The Laplace Transform

3.7 Laplace Transform Correspondences

Suppose q(s) be a real (complex) polynomial. Let Rq (Rq,C ) be the set of all
proper real (complex) rational functions with q in the denominator. Let Eq
(Eq,C ) be the set of all real (complex) exponential polynomials whose Laplace
transform is in Rq (Rq,C ). Symbolically, we would write
 
p(s)
Rq = : deg(p(s)) < deg(q(s)) and Eq = {f ∈ E : L {f (s)} ∈ Rq } ,
q(s)

where the polynomials p(s) are real. A similar description can be given for Rq,C
and Eq,C . Keep in mind that q(s) is a fixed polynomial. In this section we are
interested in obtaining an explicit description of Eq and Eq,C . Consider a simple
example. Let q(s) = s2 − 1 = (s − 1)(s + 1). How can we explicitly describe
Eq ? Observe that any proper rational function with q(s) in the denominator
p(s)
is of the form (s−1)(s+1) , where p(s) is a polynomial of degree one. A partial
fraction decomposition will have the form

p(s) A1 A2
= + ,
q(s) s−1 s+1

where A1 and A2 are in R, and its inverse Laplace transform is

A1 et + A2 e−t .

Thus Eq is the set of all real linear combinations of {et , e−t }. (It is the same
argument to show that Eq,C is the set of all complex linear combinations of
{et , e−t }.) Notice how efficient this description is: all we needed to find were
the two functions, et and e−t , and all other functions in Eq are given by taking
linear combinations. This is a much more explicit description than just saying
that Eq is the set of functions that have Laplace transform with q(s) = s2 −1 in
the denominator. This idea we want to generalize for any polynomial q. It will
introduce important notions about linear spaces, namely, spanning sets, linear
independence, and bases. The following theorem establishes a fundamental
property for the sets Eq , Eq,C , Rq , and Rq,C .

Theorem 1. Let q(s) be a real (complex) polynomial. Then Eq (Eq,C ) and Rq

(Rq,C ) are real (complex) linear spaces.

Proof. We will prove the real case; the complex case is handled in the same
way. Let’s begin with Rq . Suppose pq(s)
1 (s)
and pq(s)
2 (s)
are in Rq . Then

p1 (s) p2 (s) p1 (s) + p2 (s)

+ = ∈ Rq .
q(s) q(s) q(s)
p(s)
If c ∈ R and q(s) ∈ Rq then
 3.7 Laplace Transform Correspondences 139

p(s) cp(s)
c· = ∈ Rq .
q(s) q(s)

These two calculations show that Rq is closed under addition and scalar mul-
tiplication. This means Rq is a linear space.
Now suppose f1 (t) and f2 (t) are in Eq . Since their Laplace transforms are
both in Rq so is their sum and since the Laplace transform is linear we have

L {f1 (t) + f2 (t)} = L {f1 (t)} + L {f2 (t)} ∈ Rq .

This means f1 (t) + f2 (t) is in Eq . Let c ∈ R and f (t) ∈ Eq . Then

L {cf (t)} = cL {f (t)} ∈ Rq

and again this means that cf (t) ∈ Eq . Thus Eq is a linear space. u

t

Knowing that Eq is a linear space implies that it is closed under the op-
eration of taking linear combinations. But more is true. We will show that
there are finitely many functions f1 , . . . , fn ∈ Eq so that Eq is precisely the set
of all their linear combinations. Further, we will give an algorithm for find-
ing a canonical set of such functions based entirely on the roots of q(s) and
their multiplicities. Keep the simple example we did above in mind. When
q(s) = s2 − 1 we showed that Eq is the set of all linear combinations of two
functions, et and e−t .
To be able to speak precisely about these notions we begin by introducing
the concept of spanning sets.

Spanning Sets
Suppose F is a real or complex linear space of functions defined on a common
interval I. Suppose f1 , . . . , fn are in F . The set of all linear combinations

c1 f 1 + · · · + cn f n ,

is called the span of {f1 , . . . fn } and denoted by Span {f1 , . . . , fn }. If F is a

real linear space then the scalars c1 , . . . , cn are understood to be real numbers.
Similarly, if F is a complex linear space then the scalars c1 , . . . , cn are the
complex numbers. If it is unclear which scalars are in use we will use the
notation SpanR {f1 , . . . , fn } or SpanC {f1 , . . . , fn } to distinguish the cases. If
F = Span {f1 , . . . , fn } we will refer to the set {f1 , . . . , fn } as a spanning
set for F . In the example above where q(s) = s2 − 1 we can say that Eq is
the span of {et , e−t } or that {et , e−t } is a spanning set for Eq . However, the
notation Eq = Span {et , e−t } is much simpler.

Theorem 2. Let F = Span {f1 , . . . , fn } be as above. Then E is a linear space.

140 3 The Laplace Transform

Proof. It should be intuitively clear that F is closed under addition and scalar
multiplication: when we add two linear combinations or multiply a linear
combination by a scalar we get another such linear combination. However, to
be pedantic, here are the details. Suppose a1 f1 +· · ·+an fn and b1 f1 +· · ·+bn fn
are in F . Then

a1 f1 + · · · + an fn + b1 f1 + · · · + bn fn = (a1 + b1 )f1 + · · · (an + bn )fn ∈ F.

Thus F is closed under addition. If c is a scalar then

c · (a1 f1 + · · · + an fn ) = ca1 fn + · · · can fn ∈ F.

Thus F is closed under scalar multiplication. It follows that F is a linear space

(real or complex depending on the scalars used). u
t

We should note that a linear space can be spanned by more than one
spanning set. The following example explains what we mean.

Example 3. Let q(s) + s2 − 1. Show that

S1 = et , e−t , S2 = {sinh t, cosh t} , and S3 = et , e−t , et + e−t

 

are spanning sets for Eq .

I Solution. We need to show that each set given above is a spanning set for
Eq . We have already established that Eq = SpanR {S1 } . Recall that cosh t =
et +e−t t −t

2 and sinh t = e −e 2 . Each are linear combinations of {et , e−t } and

hence in Eq . Since Eq is a linear space it is closed under linear combinations.
It follows that SpanR {S2 } ⊂ Eq . Now observe that et = cosh t+sinh 2
t
and
−t cosh t−sinh t t −t
e = 2 . Reasoning as above, both e and e are linear combinations
of cosh t and sinh t. Hence Eq = SpanR {S1 } ⊂ SpanR {S2 }. It follows now
that Eq = SpanR {S2 } . Since each function in S3 is a linear combination
of functions in S1 it follows that SpanR {S3 } ⊂ SpanR {S1 } = Eq . On the
other hand, S1 ⊂ S3 and thus SpanR {S1 } ⊂ SpanR {S3 }. It follows now that
Eq = SpanR {S1 } = SpanR {S2 } = SpanR {S3 } . J

You will notice that in the example above S3 contains 3 functions, et ,

e and et + e−t . There is a certain redundancy in using all three functions
−t

in a spanning set when only two are necessary. Specifically, observe that the
third function depends linearly on the first two (it is their sum) so that a
linear combination of all three functions in S3 can be rewritten as a linear
combination in terms of only the first two functions:

c1 et + c2 e−t + c3 (et + e−t ) = (c1 + c3 )et + (c1 + c3 )e−t .

We root out such dependencies in spanning sets by the concept of linear

independence.
 3.7 Laplace Transform Correspondences 141

Linear Independence
A set of functions {f1 , . . . , fn }, defined on some interval I, is said to be lin-
early independent if the equation

a1 f 1 + · · · + an f n = 0 (1)

implies that all the coefficients a1 , . . . , an are zero. Otherwise, we say {f1 , . . . , fn }
is linearly dependent.2
One must be careful about this definition. We do not try to solve Equation
(1) for the variable t. Rather, we are given that this equation is valid for
all t ∈ I. With this information the focus is on what this says about the
coefficients a1 , . . . , an ; are they necessarily zero or not.
Again, we must say something about the scalars in use. We use the term
’linearly independent over R’ or the term ‘linearly independent over C’ if the
scalars in use are real or complex, respectively. However, in most cases the
context will determine which is meant.
Let’s consider an example of a set that is not linearly independent.

Example 4. Show that the set {et , e−t , et + e−t } is linearly dependent.

I Solution. This is the set S3 that we considered above. To show that a set
is linearly dependent we need only show that we can find a linear combination
that adds up to 0 with coefficients not all zero. One such is

(1)et + (1)e−t + (−1)(et + e−t ) = 0.

The coefficients, highlighted by the parentheses, are 1, 1, and −1 and are not
all zero. Thus S3 is linearly dependent. As noted above the dependency is
clearly seen in that the third function is the sum of the first two. J

Now let’s return to the basic example that began this section.
Example 5. Show that the set {et , e−t } is linearly independent.

I Solution. In this case we need to show that the only way a linear combi-
nation adds up to zero is for the coefficients to be zero. Thus we consider the
equation
a1 et + a2 e−t = 0, (2)
for all t ∈ R. In order to conclude linear independence we need to show that
a1 and a2 are zero. There are many ways this can be done. Below we show
three approaches. Of course, only one is necessary.
Method 1: Evaluation at specified points Let’s evaluate Equation (2)
at two points: t = 0 and t = ln 2.
2
A grammatical note: We say f1 , . . . , fn are linearly independent if the set
{f1 , . . . , fn } is linearly independent.
142 3 The Laplace Transform

t=0 a1 + a2 =0
t = ln 2 a1 (2) + a2 (1/2) = 0.
These two equations have only one simultaneous solution, namely, a1 = 0 and
a2 = 0. Thus {et , e−t } is linearly independent.
Method 2: Differentiation Take the derivative of Equation (2) to get
a1 et − a2 e−t = 0. Evaluating Equation (2) and the derivative at t = 0 gives
a1 + a2 = 0
a2 − a2 = 0.
The only solution to these equations is a1 = 0 and a2 = 0. Hence {et , e−t } is
linearly independent.
Method 3: The Laplace Transform Here we take the Laplace transform
of Equation (2) to get
a1 a2
+ = 0,
s−1 s+1
which is an equation valid for all s > 1, and consider the right-hand limit as s
a1
approaches 1. If a1 is not zero then s−1 has an infinite limit while the second
a2
term s+1 has a finite limit. But this cannot be as the sum is 0. It must be
that a1 = 0 and therefore a2 e−t = 0. This equation in turn implies a2 = 0.
Now it follows that {et , e−t } is linearly independent. J
We will see Method 3 used in a more general context later in this section.
In the comment following the solution to Example 3 we noted that there
was some redundancy in the linearly dependent set S3 ; it contained an extra
function that was not needed to span Eq , for q(s) = s2 − 1. This is generally
true of dependent sets; if fact, it characterizes them.
Theorem 6. A set of functions {f1 , . . . , fn } in a linear space F is linearly
dependent if and only if one of the functions is a linear combination of the
others.
Proof. Let’s assume that one of the functions, f1 say, is a linear combinations
of the others. Then we can write f1 = a2 f2 + · · · + an fn , which is equivalent
to
f1 − a2 f2 − a3 f3 − · · · − an fn = 0.
Since not all of the coefficients are zero (the coefficient of f1 is 1) it follows that
{f1 , . . . , fn } is linearly dependent. On the other hand, suppose {f1 , . . . , fn }
is linearly dependent. Then there are scalars, not all zero, such that a1 f1 +
· · · + an fn = 0. By reordering if necessary, we may assume that a1 6= 0. Now
we can solve for f1 to get
a2 −an
f1 = − f2 − · · · − fn ,
a1 a1
Thus one of the functions is a linear combination of the others. u
t
 3.7 Laplace Transform Correspondences 143

When we combine the notion of a spanning set and linear independence

we get what is called a basis.

Bases
A set of functions {f1 , . . . , fn } is said to be a basis3 of a linear space F if
{f1 , . . . , fn } is linearly independent and a spanning set for F . Thus {et , e−t }
is a basis of Eq , for q(s) = s2 − 1, because Eq = Span {et , e−t } and is linearly
independent by Example 5. Whereas {et , e−t , et + e−t } is not a basis of Eq
because, although it spans Eq it is not linearly independent.
The following theorem explains generally how a basis efficiently describes
the linear space it spans.
Theorem 7. Suppose B = {f1 , . . . , fn } is a basis of a linear space F . If f ∈ F
then there are unique scalars a1 , . . . , an such that f = a1 f1 + · · · + an fn .
Proof. The focus here is on the uniqueness of the scalars. Suppose f ∈ F.
Since B is a basis it spans F . Thus we can express f as a linear combination
of f1 , . . . , fn . Suppose now we can do this in two ways:

f = a1 f 1 + · · · + an f n
and f = b1 f1 + · · · + bn fn .

By subtracting these two equations we get (a1 − b1 )f1 + · · · (an − bn )fn = 0.

Since B is linearly independent the coefficients ai − bi , i = 1, . . . , n are all zero.
This means then that ai = bi , i = 1, . . . , n. Thus the scalars used to represent
f are unique. u
t
From Examples 5 and 3 we see that for q(s) = s2 − 1 the sets

B1 = et , e−t and B2 = {cosh t, sinh t}



are both bases of Eq . Thus a linear space can have more than one basis.
However, Theorem 7 states that once a basis of F is fixed each function in F
can be written uniquely as a linear combination of functions in the basis.

We now reiterate our goal for this section in terms of the concepts we have
introduced. For q(s) a fixed polynomial we will provide a standard method
to find a basis, Bq (Bq,C ), for Eq (Eq,C ). We will begin with the complex case
first.

The Complex Linear Space Eq,C

Let q be a fixed (complex) polynomial. Let f ∈ Eq,C and F (s) = L {f (t)}.
Then F (s) = p(s)
q(s) for some polynomial p(s). Suppose γ1 , . . . , γl are the distinct

3
A grammatical note: basis is the singular form of bases
144 3 The Laplace Transform

roots of q with multiplicities m1 , . . . , ml , respectively. Then q factors as q(s) =

C(s − γ1 )m1 · · · (s − γl )ml . The partial fraction decomposition of a rational
function pq will involve terms of the form

1
, 1 ≤ k ≤ mi .
(s − γi )k
We define the finite set Bq,C by

Bq,C = tk−1 eγi t : k = 1, . . . , mi and i = 1, . . . , l . (3)



n k−1 γ t o
Since L t (k−1)!
e i
= (s−γ1
i)
k , it follows that f (t) is a linear combination of

Bq,C and hence Eq,C ⊂ Span {Bq,C }. Note that we have dropped the factorial
terms as these may be absorbed in the scalars of any linear combination.
Since each simple exponential polynomial in Bq,C is in Eq,C we must have
Span {Bq,C } ⊂ Eq,C . It now follows that Span {Bq,C } = Eq,C .
Example 8. If q(s) = (s2 + 9)2 (s − 2) find Bq,C .
I Solution. The roots of q(s) are 3i with multiplicity 2, −3i with multi-
plicity 2, and 2 with multiplicity 1. From Equation (3) we have

Bq,C = e3it , e−3it , te3it , te−3it , et .



J
We need to show that Bq,C is a linearly independent set. To do so we
introduce an useful lemma.
Lemma 9. Suppose f1 (t), . . . , fl (t) are polynomials and λ1 , . . . , λl are distinct
complex numbers. Suppose

f1 (t)eλ1 t + · · · + fl (t)eλl t = 0. (4)

Then fi = 0, for each i = 1, . . . l.

Proof. First let’s derive a useful formula. Let f (t) = am tm + · · · + a1 t + a0 be
a polynomial of degreem and λ a complex number. Then by Example 11 of
k!
Section 3.1 we have L tk eλt = (s−λ) k+1 and hence

m
X ak k! f˜(s − λ)
L f (t)eλt =

k+1
= ,
(s − λ) (s − λ)m+1
k=0

where f˜(s − λ) = a0 (0!)(s − λ)m + a1 (1!)(s − λ)m−1 + · · · + am (m!) is a

polynomial of degree m obtained by adding the terms in the sum with common
denominator (s − λ)m+1 . We observe that f˜ is a polynomial of degree m.
Further, if f˜(s − λ) = 0 then all the coefficients a0 , . . . , am are zero and hence
f = 0. With this in mind we proceed with the proof of the lemma as stated.
 3.7 Laplace Transform Correspondences 145

Let us reorder the λi , if necessary, in such a way that λ1 is among those

roots with the largest real part. Suppose fi has degree mi , i = 1, . . . , l. We
take the Laplace transform of Equation (4) to get

f˜1 (s − λ1 ) f˜l (s − λl )
+ · · · = 0, (5)
(s − λ1 )m1 +1 (s − λl )ml +1

with notation as in the preceding paragraph. Equation 5 is valid for all complex
numbers s whose real part is larger than the real part of λ1 . Now consider the
right hand limit lims→λ+ applied to Equation (5). Suppose f˜1 (s − λ1 ) 6= 0.
1
Since λ1 , . . . , λl are distinct, all terms have finite limits except for the first,
which has an infinite limit. However, the sum of the limits must be zero, which
is impossible. We conclude that f˜1 (s − λ1 ) = 0 and hence by the remarks in
the previous paragraph f1 (s) = 0. We can now eliminate the first term from
Equation (5). Reorder the remaining λi ’s if necessary so that λ2 is among
the roots with the largest real part. By repeating the above argument we can
conclude f2 (s) = 0. Inductively, we conclude that fi (s) = 0, i = 1, . . . m. u t

We thus have the following theorem.

Theorem 10. Suppose q is a polynomial of degree n and λ1 , . . . , λl the distinct

roots with multiplicities m1 , . . . , ml , respectively. Then the set

Bq,C = tk−1 eλi t : k = 1, . . . , mi and i = 1, . . . , l



forms a basis of Eq,C .

Proof. We have already shown that Eq,C = Span {Bq,C }. To show linear inde-
pendence suppose
X
ai,k tk−1 eλi t = 0.

By first summing over k we can rewrite this as

f1 (t)eλ1 t + · · · + fl (t)eλl t = 0,

where fi (t) = m k−1

, for each i = 1, . . . , l. By Lemma 9, fi = 0, for
P l
k=1 ai,k t
i = 1, . . . , l, and hence all the coefficients {ai,j } are zero. This implies that
Bq,C is linearly independent and hence a basis of Eq,C . u
t

The finite set Bq,C is called the standard basis of Eq,C . Once the roots of
q and their multiplicities are determined it is straightforward to write down
the standard basis. Notice also that the number of terms in the standard basis
for Eq,C is the same as the degree of q.

Example 11. Find the standard basis Bq,C of Eq,C for each of the following
polynomials:
146 3 The Laplace Transform

1. q(s) = s3 − s 3. q(s) = (s − 2)4

2. q(s) = s2 + 1 4. q(s) = s2 + 4s + 5

I Solution.
1. q(s) = s3 − s = s(s − 1)(s + 1). The roots are 0, 1, and −1 and each root
occurs with multiplicity 1. The standard basis of Eq,C is Bq,C = {1, e−t , et }.
2. q(s) = s2 + 1 = (s + i)(s − i) The roots are
 i and
−i each with multiplicity
1. The standard basis of Eq,C is Bq,C = eit , e−it .
3. q(s) = (s − 2)4 . Here the
 root is 2 with multiplicity 4. Thus the standard
basis of Eq,C is Bq,C = e2t , te2t , t2 e2t , t3 e2t .

4. q(s) = s2 + 4s + 5 = (s + 2)2 + 1= (s − (−2 + i))(s − (−2 − i)). Thus the

standard basis of Eq,C is Bq,C = e(−2+i)t , e(−2−i)t . J

The Real Linear Space Eq

Suppose q(s) is a real polynomial. We want to describe a real basis Bq of Eq .
View q(s) as a complex polynomial. Recall that if λ = a + ib is a complex root
with multiplicity m then its conjugate λ = a − ib is also a root
with the same
multiplicity. Thus non real roots of q(s) come in pairs λ, λ . By switching
the order of the pair λ, λ and relabeling, if necessary, we can assume b > 0.


In the standard basis Bq,C for Eq,C the functions

n o
tk eλt , tk eλt

likewise come in pairs. Let Bq be the set obtained from Bq,C by replacing each
such pair with the real functions
 k at
t e cos bt, tk eat sin bt ,

(λ = a + ib and b > 0). We will show below that the set Bq obtained in this
way is a basis of Eq ; we call it the standard basis of Eq .
Consider the following example.
Example 12. Find Bq if q(s) = (s2 + 4)2 (s − 6)3 .
I Solution. The roots of q are ±2i with multiplicity 2 and 6 with multi-
plicity 3. Thus
Bq,C = e2it , e−2it , te2it , te−2it , e6t , te6t , t2 e6t .


The only conjugate pair of non real roots is {2i, −2i}. It gives rise to pairs
of functions e2it , e−2it and te2it , te−2it in Bq,C which are replaced by


{cos 2t, sin 2t} and {t cos 2t, t sin 2t}, respectively. We thus obtain
Bq = cos 2t, sin 2t, t cos 2t, t sin 2t, e6t , te6t , t2 e6t .


J
 3.7 Laplace Transform Correspondences 147

Let’s now show that Bq is basis of Eq , for a real polynomial q(s). Since
we can regard q(s) as a complex polynomial we have Rq ⊂ Rq,C and hence
Eq ⊂ Eq,C . By Theorem 7 of Section 3.1, Eq is precisely the set of all real
valued functions in Eq,C . We begin by showing Eq = SpanR {Bq }. Let λ be a
root of q(s). Either λ is real or is part of a non real root pair (λ, λ). If λ = a
is a real root of q(s) with multiplicity m the function tk eat , k = 0, . . . , m − 1,
is in Bq and Bq,C , hence in Eq,C . Consider a non real root pair (λ, λ) of q(s),
where λ = a + ib with multiplicity m and b > 0. Let k be an integer between
0 and m − 1. By Euler’s equation the associated pairs of functions are related
by the following equations:
 
tk eλt = tk eat (cos bt + i sin bt) tk eat cos bt = 21 tk eλt + tk eλt
and   (6)
1
tk eλt = tk eat (cos bt − i sin bt) tk eat sin bt = 2i tk eλt − tk eλt

Notice that tk eat cos bt and tk eat sin bt are linear combinations of tk eλt and
tk eλt and hence in Eq,C . If follows now that Bq ⊂ Eq,C and hence any function
in the real span of Bq is a real function in Eq,C and hence in Eq . This implies
SpanR {Bq } ⊂ Eq .
Now suppose f ∈ Eq . Then it is also in Eq,C and we can write f as a
(complex)
P k λt linear combination of functions in Bq,C . Briefly, we can write f =
µt e , where the λ’s are among the roots of q(s), k is a nonnegative integer
up to the multiplicity of λ less one, and µ is complex number for each pair
(k, λ). Consider the two cases. Suppose λ is real. Let µ = α + iβ. Then
Re(µtk eλt ) = αtk eλt is a real scalar multiple of a function in Bq . Suppose
λ = a + ib is non real, with b > 0. Then

Re(µtk eλt ) = Re((α + iβ)tk eat (cos bt + i sin bt))

= αtk eat cos bt − βtk eat sin bt,

a real linear combination of functions in Bq . Since

P f is kreal these preced-
ing calculations show that f (t) = Re(f (t)) = Re(µt eλt ) is a real lin-
ear combination of functions in Bq . Thus f (t) ∈ SpanR {Bq }. It follows that
Eq ⊂ SpanR {Bq } and hence Eq = SpanR {Bq }.
We proceed now to show that Bq is linearly independent. First consider a
preliminary calculation. Let c1 and c2 be real numbers and λ = a + ib, with
b > 0. Then by Equation (6) we have
c1 − ic2 k λt c1 + ic2 k λt
c1 tk eat cos bt + c2 tk eat sin bt = t e + t e .
2 2
If c1 −ic
2
2
= 0 or c1 +ic
2
2
= 0 then c1 = c2 = 0. Now suppose f (t) is a real linear
combination of Bq and f (t) = 0. Each pair of terms c1 tk eat cos bt+c2 tk eat sin bt
can be replaced as above by c1 −ic 2 t e + c1 +ic
2 k λt
2
2 k λt
t e . We thus reexpress f
as a complex linear combination of Bq,C . But Bq,C is linearly independent by
Theorem 10 and hence all coefficients of f expressed as a linear combination
148 3 The Laplace Transform

over Bq,C are zero. The above remark then implies that the coefficients of f
expressed as a linear combination over Bq are zero. This implies Bq is linearly
independent.
We summarize our discussion in the following theorem.
Theorem 13. Suppose q is a real polynomial. Let Bq be the set obtained from
Bq,C by replacing the each occurrence of pairs of functions

(tk eλt , tk eλt )

with the real functions

(tk eat cos bt, tk eat sin bt),

where λ = a + ib with b > 0. Then Bq is a basis of Eq .

Example 14. Find the standard basis for Eq for each polynomial q below:

1. q(s) = s(s − 1)2 3. q(s) = s3 − s2 + s − 1

2. q(s) = s2 + 1 4. q(s) = (s2 − 2s + 5)2

I Solution.
1. The roots of q are 0 and 1 with multiplicity 1 and 2 respectively. Thus
the standard basis for Eq is {1, et , tet }.
2. The roots
 of q are the pair i and −i each with multiplicity one. Since
Bq,C = eit , e−it it follows that Bq = {cos t, sin t}.
3. It is easy to check that q factors as q(s) = (s − 1)(s2 + 1). The roots are
1, i, and −i, each with multiplicity one. Thus Bq,C = et , eit , e−it . From

this we see that the standard basis for Eq is Bq = {et , cos t, sin t}.
4. Observe that q(s) = (s2 − 2s + 5)2 = ((s − 1)2 + 4)2 and has roots 1 + 2i
and 1 − 2i, each with multiplicity two. It follows the standard basis for Eq
is Bq = {et cos(2t), et sin(2t), tet cos 2t, tet sin 2t}.
J

Exercises

1–14. Find the complex basis Bq,C for 4. q(s) = s2 − 6s + 9

each polynomial q(s) given.
5. q(s) = s2 − 9s + 14
1. q(s) = s − 4

2. q(s) = s + 6 6. q(s) = s2 − s − 6

3. q(s) = s2 − 3s − 4 7. q(s) = s2 + 9
 3.8 Convolution 149

8. q(s) = s2 − 10s + 25 20. q(s) = s2 − s − 6

9. q(s) = s2 + 4s + 13 21. q(s) = s2 + 9

10. q(s) = s2 + 9s + 18 22. q(s) = s2 − 10s + 25

11. q(s) = s3 − s 23. q(s) = s2 + 4s + 13

12. q(s) = s4 − 1 24. q(s) = s2 + 9s + 18

13. q(s) = (s − 1)3 (s + 7)2 25. q(s) = s3 − s

14. q(s) = (s2 + 1)3 26. q(s) = s4 − 1

15–28. Find the standard basis Bq of 27. q(s) = (s − 1)3 (s + 7)2

Eq for each polynomial q(s).
28. q(s) = (s2 + 1)3
15. q(s) = s − 4
29. Let q1 (s) and q2 (s) be polynomi-
16. q(s) = s + 6 als. The polynomial q1 (s) is said to
divide q2 (s) if q2 (s) = r(s)q1 (s)
17. q(s) = s2 − 3s − 4
for some polynomial r(s), in which
18. q(s) = s2 − 6s + 9 case we write q1 (s)|q2 (s). Show that
q1 (s)|q2 (s) if and only if Bq1 ⊂ Bq2
19. q(s) = s2 − 9s + 14 if and only if Eq1 ⊂ Eq2 .

3.8 Convolution
The Laplace transform L : E → R provides a one-to-one linear correspondence
between the input space E of exponential polynomials and the transform space
R of proper rational functions. We have seen how operations defined on the
input space induce via the Laplace transform a corresponding operation on
transform space, and vice versa. The main examples have been

Linearity af (t) + bg(t) ←→ aF (s) + bG(s)

The Translation Princible eat f (t) ←→ F (s − a)
d
Transform Derivative Principle −tf (t) ←→ F (s)
ds

The Dilation Principle bf (bt) ←→ F sb
Input Derivative Principle f 0 (t) ←→ sF (s) − f (0)

where the double arrow ←→ indicates the Laplace transform correspondence

and, as usual, F (s) and G(s) are the Laplace transforms of f (t) and g(t).
Our goal in this section is to study another operational identity of this
type. Specifically, we will be concentrating on the question of what is the
150 3 The Laplace Transform

effect on the input space E (EC ) of ordinary multiplication of functions in the

transform space R ( RC ). Thus we are interested in the following question:
Given functions F (s) and G(s) in R and their inverse Laplace transforms
f (t) and g(t) in E, what is the exponential polynomial h(t) such that h(t)
corresponds to H(s) = F (s)G(s) under the Laplace transform? In other words,
how do we fill in the following question mark in terms of f (t) and g(t)?

h(t) = ? ←→ F (s)G(s)

You might guess that h(t) = f (t)g(t). That is, you would be guessing that
multiplication in the input space corresponds to multiplication in the trans-
form space. This guess is wrong as you can quickly see by looking at almost
any example. For a concrete example, let
1 1
F (s) = and G(s) = .
s s2
Then H(s) = F (s)G(s) = 1/s3 and h(t) = t2 /2. However, f (t) = 1, g(t) = t,
and, hence, f (t)g(t) = t. Thus h(t) 6= f (t)g(t).
Let’s modify this example slightly. Again suppose that F (s) = 1/s so
that f (t) = 1, but assume now that G(s) = n!/sn+1 so that g(t) = tn . Now
determine which function h(t) has F (s)G(s) as its Laplace transform:
 
n! n! 1 n+1
h(t) = L−1 {F (s)G(s)} = L−1 = tn+1 = t .
sn+2 (n + 1)! n+1

What is the relationship between f (t), g(t), and h(t)? One thing that we can
observe is that h(t) is an integral of g(t):
Z t Z t
1 n+1
h(t) = t = τ n dτ = g(τ ) dτ.
n+1 0 0

Let’s try another example. Again let F (s) = 1/s so f (t) = 1, but now let
G(s) = s/(s2 + 1) which implies that g(t) = cos t. Then
   
−1 −1 1 s −1 1
h(t) = L {F (s)G(s)} = L =L = sin t,
s s2 + 1 s2 + 1

and again what we can observe is that h(t) is an integral of g(t):

Z t Z t
h(t) = sin t = cos τ dτ = g(τ ) dτ.
0 0

What these examples suggest is that multiplication of G(s) by 1/s = L {1}

in transform space corresponds to integration of g(t) in the input space E. In
fact, it is easy to see that this observation is legitimate by a calculation with
the input derivative principle, Theorem 3.6.1. Here’s the argument. Suppose
 3.8 Convolution 151
Rt
that G(s) ∈ R. Let h(t) = 0 g(τ ) dτ . Then h0 (t) = g(t) and h(0) = 0 so the
input derivative principle gives

G(s) = L {g(t)} = L {h0 (t)} = sH(s) − h(0) = sH(s).

Hence, H(s) = (1/s)G(s) and

Z t 
1
L g(τ ) dτ = G(s).
0 s

We summarize this discussion in the following theorem.

Theorem 1. Let g(t) be an exponential polynomial and G(s) its Laplace trans-
form. Then

Input Integral Principle

nR o
t
L 0 g(τ ) dτ = 1s G(s).

We have thus determined the effect on the input space E of multiplication

on the transform space by the Laplace transform of the function 1. Namely,
the effect is integration. If we replace the function 1 by an arbitrary function
f (t) ∈ E, then the effect on E of multiplication by F (s) is more complicated,
but it can still be described by means of an integral operation. To describe this
operation precisely, suppose that f (t) and g(t) are exponential polynomials.
The convolution product or convolution of f (t) and g(t), is a new expo-
nential polynomial denoted by the symbol f ∗ g, and defined by the integral
formula Z t
(f ∗ g)(t) = f (t − τ )g(τ ) dτ. (1)
0
What this formula means is that f ∗g is the name of a new function constructed
from f (t) and g(t) and the value of f ∗ g at the arbitrary point t is denoted by
(f ∗ g)(t) and it is computed by means of the integral formula (1). Consider
the following simple example.

Example 2. Use Equation (1) to compute

1 ∗ t.

I Solution. Here we let f (t) = 1 and g(t) = t in Equation (1) to get

t t
t2
Z Z
1∗t= f (t − τ )g(τ ) dτ = τ dτ = .
0 0 2
J
152 3 The Laplace Transform

We will give more examples below but let’s consider the main point that
we want: the convolution product of f (t) and g(t) on the input space E cor-
responds to the ordinary multiplication of F (s) and G(s) on the transform
space R. That is the content of the following theorem, the proof of which we
will postpone until Chapter 8, where the Convolution Theorem is proved for
a broader class of functions.

Theorem 3 (The Convolution Theorem). Let f (t), g(t) ∈ E. Then

The Convolution Principle

L {(f ∗ g)(t)} = L {f (t)} L {g(t)} .

In terms of inverse Laplace transforms, this is equivalent to the following

statement. If F (s) and G(s) are in transform space then

L−1 {F (s)G(s)} = L−1 {F (s)} ∗ L−1 {G(s)} .

An important special case of Theorem 3 that is worth pointing out explic-

itly is that when f (t) = 1 we get Theorem 1:
Z t 
1
L g(τ ) dτ = L {1 ∗ g(t)} = L {1} L {g(t)} = G(s).
0 s

Properties of the Convolution Product. The convolution product f ∗ g

behaves in many ways like an ordinary product:

commutative property: f ∗g = g∗f

associative property: (f ∗ g) ∗ h = f ∗ (g ∗ h)
distributive property: f ∗ (g + h) = f ∗ g + f ∗ h
f ∗0 = 0∗f =0

Indeed, these properties of convolution are easily verified from the defini-
tion given in Equation (1). There is one significant difference, however. From
Example 2 we found that t∗1 = t2 /2 6= t so that, in general, f ∗1 6= f . In other
words, convolution by the constant function 1 does not behave like a multi-
plicative identity. In fact, no such function exists, however, in Chapter 8 we
will discuss a socalled ‘generalized function’ that will act as a multiplicative
identity. Let’s now consider a few examples.

Example 4. Compute the convolution product eat ∗ ebt where a 6= b.

 3.8 Convolution 153

I Solution. Use the defining equation (1) to get

Z t Z t
at bt aτ b(t−τ ) bt eat − ebt
e ∗e = e e dτ = e e(a−b)τ dτ = .
0 0 a−b
Observe that
 at
e − ebt
  
1 1 1 1
= L eat L ebt ,
 
L = − =
a−b a−b s−a s−b (s − a)(s − b)
so this calculation is in agreement with what is expected from Theorem 3. J
Example 5. Compute the convolution product eat ∗ eat .
I Solution. Computing from the definition:
Z t Z τ
eat ∗ eat = eaτ ea(t−τ ) dτ = eat dτ = teat .
0 0
As with the previous example, note that the calculation
1
L teat = = L eat L eat
  
(s − a)2

agrees with the expectation of Theorem 3. J

Remark 6. Since
eat − ebt d at
lim = e = teat ,
a→b a − b da
the previous two examples show that
lim eat ∗ ebt = teat = eat ∗ eat ,
a→b

so that the convolution product is, in some sense, a continuous operation.

The convolution theorem is particularly useful in computing the inverse
Laplace transform of a product.
s
Example 7. Compute the inverse Laplace transform of .
(s − 1)(s2 + 9)
s 1
I Solution. The inverse Laplace transforms of 2 and are cos 3t
s +9 s−1
t
and e , respectively. The convolution theorem now gives
 
−1 s
L = cos 3t ∗ et
(s − 1)(s2 + 9)
Z t
= cos 3τ et−τ dτ
0
Z t
= et cos 3τ e−τ dτ
0
et
t
= −e−τ cos 3τ + 3e−τ sin 3τ 0
10
1
= (− cos 3t + 3 sin 3t + et ).
10
154 3 The Laplace Transform

The computation of the integral involves integration by parts. We leave it to

the student to verify this calculation. Of course, this calculation agrees with
Laplace inversion using the method of partial fractions. J
In the Table of Section C.3 a list is given of convolutions of some common
functions. You may want to familiarize yourself with this table so as to know
when you will be able to use it. The example above appears in the table (a = 1
and b = 3). Verify the answer.
Example 8. Verify the following convolution product where m, n ≥ 0:

m! n!
tm ∗ tn = tm+n+1 .
(m + n + 1)!

I Solution. Our method for this computation is to use the Convolution

theorem, Theorem 3. We get
m! n! m! n!
L {tm ∗ tn } = L {tm } L {tn } = = .
sm+1 sn+1 sm+n+2
Now take the inverse Laplace transform to conclude
 
m! n! m! n!
tm ∗ tn = L−1 {L {tm ∗ tn }} = L−1 = tm+n+1 .
sm+n+2 (m + n + 1)!
J
As special cases of this formula note that
1 6 1 6
t2 ∗ t3 = t and t ∗ t4 = t .
60 30
1
Example 9. Find the inverse Laplace transform of .
s(s2 + 1)
I Solution. This could be done using partial fractions, but instead we will
do the calculation using the input integral formula:
  Z t
1
L−1 = sin τ dτ = − cos t + 1.
s(s2 + 1) 0

J
Example 10. Express the solution to the following initial value problem in
terms of the convolution product:

y 00 + a2 y = f (t), y(0) = 0, y 0 (0) = 0, (2)

where f (t) ∈ E is an arbitrary exponential polynomial.

 3.8 Convolution 155

I Solution. If we apply the Laplace transform L to this equation we obtain

(s2 + a2 )Y (s) = F (s),
so that
1
Y (s) = F (s).
s2 + a 2
Since  
1 1
L−1 = sin at,
(s2 + a2 ) a
the convolution theorem expresses y(t) as a convolution product
1
y(t) = sin at ∗ f (t).
a
J
Remark 11. This allows for the expression of y(t) as an integral
1 t
Z
y(t) = f (τ ) sin a(t − τ ) dτ.
a 0
This integral equation can be thought of dynamically as starting from an arbi-
trary input function f (t) and producing the output function y(t) determined
by the differential equation (2). Schematically, we write
f (t) 7−→ y(t).
Moreover, although we arrived at this equation via the Laplace transform, it
was never actually necessary to compute F (s).
In the next example, we revisit a simple rational function whose inverse
Laplace transform can be computed by the techniques of Section 3.5.
b
Example 12. Compute the inverse Laplace transform of 2 .
(s + b2 )2
I Solution. The inverse Laplace transform of 1/(s2 + b2 ) is (1/b) sin bt. By
the convolution theorem we have
 
−1 b b
L = 2 sin bt ∗ sin bt.
(s2 + b2 )2 b
1 t
Z
= sin bτ sin b(t − τ ) dτ
b 0
1 t
Z
= sin bτ (sin bt cos bτ − sin bτ cos bt) dτ
b 0
Z t Z t
1
= sin bt sin bτ cos bτ dτ − cos bt sin2 bτ dτ
b 0 0
sin2 bt
 
1 bt − sin bt cos at
= sin bt − cos bt
b 2b 2b
1
= 2 (sin bt − bt cos bt).
2b
156 3 The Laplace Transform

Now, one should see how to handle b/(s2 + b2 )3 and even higher powers:
repeated applications of convolution. Let f ∗k denote the convolution of f
with itself k times. In other words

f ∗k = f ∗ f ∗ · · · ∗ f, k times.

Then it is easy to see that

 
b 1
L−1 = k sin∗(k+1) bt
(s2 + b2 )k+1 b
 
s 1
and L−1 = k cos bt ∗ sin∗k bt.
(s2 + b2 )n b

These rational functions with powers of irreducible quadratics in the de-

nominator were introduced in Section 3.5 and recursion formulas were derived
to calculate their inverse Laplace transforms. In fact, using the Dilation Prin-
ciple and the Convolution Principle it is an easy exercise to calculate these
convolutions in terms of Ak and Bk defined in Section 3.5.
1
sin∗(k+1) bt = Ak (bt)
(2b)k k!
1
and cos bt ∗ sin∗k bt = Bk (bt).
(2b)k k!

Recursive formulas for these convolutions can now be computed via the
recursive formulas for Ak and Bk . As mentioned earlier we will derive explicit
formulas for Ak and Bk in Section ?? through the use of second order (non
constant coefficient) differential equations.

Exercises
1–15. Compute the convolution prod- 8. cos t ∗ cos 2t
uct of the following functions.
9. e2t ∗ e−4t
1. t ∗ t
10. t ∗ tn
2. t ∗ t3

3. 3 ∗ sin t 11. eat ∗ sin bt

4. (3t + 1) ∗ e4t 12. eat ∗ cos bt

5. sin 2t ∗ e3t 13. sin at ∗ sin bt

6. (2t + 1) ∗ cos 2t 14. sin at ∗ cos bt

7. t2 ∗ e−6t 15. cos at ∗ cos bt

 3.8 Convolution 157

16–22. Compute the Laplace trans- 1

27.
form of each of the following functions. (s + 6)s3
t 2
16. f (t) = 0
(t − τ ) cos 2τ dτ 28.
(s − 3)(s2 + 4)
t
17. f (t) = 0
(t − τ )2 sin 2τ dτ s
29.
t
(s − 4)(s2 + 1)
18. f (t) = 0
(t − τ )3 e−3τ dτ
1
t 3 −3(t−τ )
30. a 6= b
19. f (t) = τ e dτ (s − a)(s − b)
0

t 1
20. f (t) = 0
cos 5τ e4(t−τ ) dτ 31.
s2 (s2 + a2 )
t
21. f (t) = 0
sin 2τ cos(t − τ ) dτ G(s)
32.
t s+2
22. f (t) = 0
sin 2τ sin 2(t − τ ) dτ
s
33. G(s)
23–33. In each of the following exer- s2 + 2
cises compute the inverse Laplace trans-
34–37. Write the zero-state solution of
form of the given function by use of the
each of the following differential equa-
convolution theorem.
tions in terms of a convolution integral
1 involving the input function f (t). You
23.
(s − 2)(s + 4) may wish to review Example ?? before
proceeding.
1
24.
s2 − 6s + 5 34. y 00 + 3y = f (t)
1 35. y 00 + 4y 0 + 4y = f (t)
25.
(s2 + 1)2
36. y 00 + 2y 0 + 5y = f (t)
s
26.
(s2 + 1)2 37. y 00 + 5y 0 + 6y = f (t)
158 3 The Laplace Transform

3.9 Table of Laplace Transforms

Input function f Laplace transform F

Basic Functions
1 1 2 n!
1) 1, t, t2 , · · · , tn , 2 , 3 , · · · , n+1
s s s s
1
2) eat
s−a
b
3) sin(bt)
s + b2
2
s
4) cos(bt)
s2 + b 2

Translation of Basic Functions

eat f (t) F(s − a)

n!
5) eat tn , n = 0, 1, 2, ...
(s − a)n+1
b
6) eat sin(bt)
(s − a)2 + b2
s−a
7) eat cos(bt)
(s − a)2 + b2
 3.9 Table of Laplace Transforms 159

Input function f Laplace transform F

Heaviside Expansion Formula: 1

rk
1e
r1 t
rk
ne
rn t
sk
q0 (r1 ) + ··· + q0 (rn ) , q(s)=(s−r1 )···(s−rn ) (s−r1 )···(s−rn ) , r1 ,...,rn , distinct

eat ebt 1
8) a−b + b−a (s−a)(s−b) a 6= b
aeat bebt s
9) a−b + b−a (s−a)(s−b) a 6= b
eat ebt ect 1
10) (a−b)(a−c) + (b−a)(b−c) + (c−a)(c−b) (s−a)(s−b)(s−c) a, b, c distinct

aeat bebt cect s

11) (a−b)(a−c) + (b−a)(b−c) + (c−a)(c−b) (s−a)(s−b)(s−c) a, b, c distinct

a2 eat b2 ebt c2 ect s2

12) (a−b)(a−c) + (b−a)(b−c) + (c−a)(c−b) (s−a)(s−b)(s−c) a, b, c distinct

Heaviside Expansion Formula: 2

k k−l tn−l−1 sk
( kl=0
P
at
l a (n−l−1)! )e (s − a)n
1
13) teat
(s − a)2
s
14) (1 + at)eat
(s − a)2
t2 at 1
15) e
2 (s − a)3
at2 at s
16) (t + )e
2 (s − a)3
a2 t2 at s2
17) (1 + 2at + )e
2 (s − a)3
160 3 The Laplace Transform

Input function f Laplace transform F

General Rules

18) eat f (t) F (s − a)

19) f 0 (t) sF (s) − f (0)

20) f 00 (t) s2 F (s) − sf (0) − f 0 (0)

21) −tf (t) F 0 (s)

22) t2 f (t) F 00 (s)

23) (−t)n f (t) F (n) (s)

1 R∞
24) f (t) s F (u)du
t
Rt F (s)
25) 0 f (v)dv
s
26) (f ∗ g)(t) F (s)G(s)
1 s
27) f (at) F
a a
 3.10 Table of Convolutions 161

3.10 Table of Convolutions

f (t) g(t) f ∗ g(t)

tn+2
t tn
(n + 1)(n + 2)
at − sin at
t sin at
a2
2 2 2
t2 sin at 3
(cos at − (1 − a 2t ))
a
1 − cos at
t cos at
a2
2
t2 cos at (at − sin at)
a3
at
e − (1 + at)
t eat
a2
2 at 2 2
t2 eat 3
(e − (a + at + a 2t ))
a
1
eat ebt (ebt − eat ) a 6= b
b−a
eat eat teat
1
eat sin bt (beat − b cos bt − a sin bt)
a2 + b2
1
eat cos bt (aeat − a cos bt + b sin bt)
a + b2
2
1
sin at sin bt (b sin at − a sin bt) a 6= b
b − a2
2
1
sin at sin at (sin at − at cos at)
2a
1
sin at cos bt (a cos at − a cos bt) a 6= b
b 2 − a2
1
sin at cos at t sin at
2
1
cos at cos bt (a sin at − b sin bt) a 6= b
a2 − b 2
1
cos at cos at (at cos at + sin at)
2a
4
Linear Constant Coefficient Differential
Equations

In Section 2.2 we introduced the first order linear differential equation

a1 (t)y 0 + a0 (t)y = f (t)

and discussed how the notion of linearity led us to a concise description of the
structure of the solution set. We also provided an algorithm for obtaining the
solution set explicitly.
In this and subsequent chapters we will consider the higher order linear
differential equations and, in the second order case, consider applications to
mechanical and electrical problems and wave oscillations. Again, the notion of
linearity is of fundamental importance for it implies a rather simple structure
of the solution set. Unfortunately, unlike the first order case, there is no general
solution method, except when the coefficients are constant. It is for this reason
that in this chapter we focus on the important case of constant coefficient
linear differential equations. Many important physical phenomena are modeled
by linear constant coefficient differential equations. In chapter 6 we consider
the situation where the coefficients are functions. Much of the general theory
we discuss here will carry over. In chapter 7 we will consider power series
methods for solving such.

4.1 Notation and Definitions

Suppose a0 , . . . , an are scalars, an 6= 0, and f is a function defined on some
interval I. A differential equation of the form

an y (n) + an−1 y (n−1) + · · · + a1 y 0 + a0 y = f (t), (1)

is called a constant coefficient nth order linear differential equation.

The constants a0 , . . . , an are called the coefficients and an is called the lead-
ing coefficient. By dividing by an , if necessary, we can assume that the
leading coefficient is 1. We will frequently make this assumption; there is no
164 4 Linear Constant Coefficient Differential Equations

loss in generality. When the leading coefficient is 1 the function f is called

the forcing function. When f = 0, we call Equation (1) homogeneous,
otherwise, it is called nonhomogeneous.
Example 1. Consider the following list of differential equations.

1. 3y 00 + 2y 0 − 5y = 0 4. y 000 + y 0 + y = t
2. y 00 − y 0 =0 5. 4y 0 + y = 1
3. y 00 − 4y = 0 6. y (4) + 4ty = 1t

Examples (1.1) to (1.5) are constant coefficient linear differential equations.

The presence of 4t in front of the y term makes Example (1.6) non constant
coefficient. Examples (1.1) to (1.3) are homogeneous while the others are
nonhomogeneous. Examples (1.1) to (1.3) are second order, Example (1.4) is
third order, Example (1.5) is first order, and Example(1.6) is fourth order.
The left hand side of Equation (1) is made up of a linear combination of
differentiations and multiplications by constants. Let D denote the deriva-
tive operator: D(y) = y 0 . If C n (I) denotes the set of functions that have a
continuous nth derivative on the interval I then D : C n (I) → C n−1 (I). We
are using the convention that C 0 (I) is the set of continuous functions on the
interval I. In a similar way D 2 will denote the second derivative operator.
Thus D 2 (y) = y 00 and D 2 : C n (I) → C n−2 (I). Let
L = an D n + · · · + a1 D + a0 , (2)
where a0 , . . . an are the same constants given in Equation (1). Then L(y) =
an y (n) + · · · + a1 y 0 + a0 y and Equation (1) can be rewritten
L(y) = f.
For example, y 00 + 2y 0 + y can be rewritten Ly where L = D2 + 2D + 1. We call
L in Equation (2) a linear constant coefficient differential operator of
order n, (assuming, of course, that an 6= 0). Since L is a linear combination
of powers of D we will sometimes refer to it as a polynomial differential
operator. More specifically, if
q(s) = an sn + an−1 sn−1 + · · · + a1 s + a0 ,
then q is a polynomial of order n and L is obtained from q by substituting
D for s and we will write L = q(D). The polynomial q is referred as the
characteristic polynomial of L. The operator L can be thought of as tak-
ing a function y that has at least n continuous derivatives and producing a
continuous function.
Example 2. Suppose L = D2 − 4D + 3. Find
L(tet ), L(et ), and L(e3t )
.
 4.1 Notation and Definitions 165

I Solution.

• L(tet ) = D2 (tet ) − 4D(tet ) + 3(tet )

= D(et + tet ) − 4(et + tet ) + 3tet
= 2et + tet − 4et − 4tet + 3tet
= −2et .
• L(et ) = D2 (et ) − 4D(et ) + 3(et )
= et − 4et + 3et
= 0.
• L(e3t ) = D2 (e3 t) − 4D(e3t ) + 3(e3t )
= 9e3t − 12e3t + 3e3t
= 0.
J
Example 3. Suppose L = D3 + D. Find

L(sin t), L(1), and L(et )

I Solution.

• L(sin t) = D3 (sin t) + D(sin t)

= cos t − cos t
= 0.
• L(1) = D3 (1) + D(1)
= 0.
• L(et ) = D3 (et ) + D(et )
= (et ) + (et )
= 2et .
J
In each case, L takes a function and produces a new function. The case
where that new function is identically 0 is important as we will see below.

Linearity
The adjective ‘linear’ describes a very important property that L satisfies. To
explain this let’s start with the familiar derivative operator D. One learns
early on in calculus the following two properties:
1. If y1 and y2 are continuously differentiable functions then

D(y1 + y2 ) = D(y1 ) + D(y2 ).

2. If y is a continuously differentiable function and c is a scalar then

D(cy) = cD(y).
166 4 Linear Constant Coefficient Differential Equations

Simply put, D preserves addition and scalar multiplication of functions. When

an operation on functions satisfies these two properties we call it linear.
Observe closely the following proposition.
Proposition 4. The operator
L = an Dn + . . . + a1 D + a0
given by Equation 2 is linear. Specifically,
1. If y1 and y2 have sufficiently many derivatives then
L(y1 + y2 ) = L(y1 ) + L(y2 ).
2. If y has sufficiently many derivatives and c is a scalar then
L(cy) = cL(y).
Proof. Suppose y, y1 , and y2 are n-times differentiable functions and c is a
scalar. If k ≤ n then by induction we have
Dk (y1 + y2 ) = Dk−1 (Dy1 + Dy2 ) = Dk y1 + Dk y2 .
Also by induction, if c is a scalar we have
Dk (cy) = Dk−1 D(cy) = Dk−1 cD(y) = cDk (y). (3)
k
Thus D is linear. We also observe that multiplication by a constant ak like-
wise preserves multiplication and addition. We thus have
L(y1 + y2 ) = an Dn (y1 + y2 ) + · · · + a1 D(y1 + y2 ) + a0 (y1 + y2 )
= an (Dn y1 + Dn y2 ) + · · · + a1 (Dy1 + Dy2 ) + a0 (y1 + y2 )
= an Dn y1 + · · · + a1 Dy1 + a0 y1 + an Dn y2 + · · · + a1 Dy2 + a0 y2
= Ly1 + Ly2 .
Thus L preserves addition. In a similar way L preserves scalar multiplication
and hence is linear.
To illustrate the power of linearity consider the following example.
Example 5. Let L = D2 − 4D + 3. Use linearity to determine
L(3et + 4tet + 5e3t ).
I Solution. Recall from Example 2 that
• L(et ) = 0
• L(tet ) = −2et .
• L(e3t ) = 0.
Using linearity we obtain
L(3et + 4tet + 5e3t ) = 3L(et ) + 4L(tet ) + 5L(e3t )
= 3 · 0 + 4 · (−2et ) + 5 · 0
= −8et .
J
 4.1 Notation and Definitions 167

Solutions
A solution to Equation (1) is a function y defined on some interval I with
sufficiently many continuous derivatives so that Ly = f . The solution set or
the general solution is the set of all solutions.
Linearity is particulary useful for finding solutions in the homogeneous
case. The following proposition shows that linear combinations of known so-
lutions produce new ones.
Proposition 6. Suppose L is a linear differential operator then the solution
set to Ly = 0 is a linear space. Specifically, suppose y1 and y2 are solutions
to Ly = 0. Then c1 y1 + c2 y2 is a solution to Ly = 0 for all scalars c1 and c2 .
Proof. By Proposition 4 we have
L(c1 y1 + c2 y2 ) = c1 L(y1 ) + c2 L(y2 )
= c1 · 0 + c2 · 0 = 0.
u
t
Example 7. Use Example 2 to find homogeneous solutions to Ly = 0, where
L = D2 − 4D + 3
I Solution. In Example 2 we found that L(et ) = 0 and L(e3t ) = 0. Now
using Proposition 6 we have
c1 et + c2 e3t
is a solution to Ly = 0, for all scalars c1 and c2 . In other words,
L(c1 et + c2 e3t ) = 0,
for all scalars c1 and c2 . We will later show that all of the homogeneous
solutions are of this form. J
We cannot underscore more the importance of Proposition 6. Once a few
specific solutions to Ly = 0 are known then all linear combinations are likewise
solutions.
Given a constant coefficient linear differential equation Ly = f we call
Ly = 0 the associated homogeneous differential equation. The following
theorem considers the nonhomogeneous case. Note it carefully.
Theorem 8. Suppose L is a linear differential operator and f is a continuous
function. If yp is a fixed particular solution to Ly = f and yh is any solution
to the associated homogeneous differential equation Ly = 0 then
yp + yh
is a solution to Ly = f . Furthermore, any solution y to Ly = f has the form
y = yp + yh .
168 4 Linear Constant Coefficient Differential Equations

Proof. Suppose yp satisfies Lyp = f and yh satisfies Lyh = 0. Then by linearity

L(yp + yh ) = Lyp + Lyh = f + 0 = f.

Thus yp + yh is a solution to Ly = f . On the other hand, suppose y is any

solution to Ly = f . Let yh = y − yp . Then, again by linearity,

Lyh = L(y − yp ) = Ly − Lyp = f − f = 0.

Thus yh is a solution to Ly = 0 and y = yp + yh . u

t

This important theorem actually provides an effective strategy for finding

the solution set to a linear differential equation:

Algorithm 9. The general solution to a linear differential equation

Ly = f

can be found as follows:

1. Find all the solutions to the associated homogeneous differential equation
Ly = 0,
2. Find one particular solution yp ,
3. Add the particular solution to the homogeneous solutions:

yp + yh .

As yh varies over all homogeneous solution we obtain all solution to Ly =

f.
This is the strategy we will follow. Section 4.3 will be devoted to determining
solutions to the associated homogeneous differential equation. Sections 4.4
and 4.5 will show methods for finding a particular solution.

Example 10. Use Algorithm 9 and Example 3 to find solutions to

y 000 + y 0 = 2et .

I Solution. The left-hand side can be written Ly, where L is the differential
operator
L = D 3 + D.
From Example 3 we found
• L(et ) = 2et ,
• L(sin t) = 0,
• L(1) = 0
 4.1 Notation and Definitions 169

Notice that the first equation tells us that a particular solution is yp = et .

The second and third equations give sin t and 1 as solutions to the associ-
ated homogeneous differential equation Ly = 0. Thus, by Proposition 6, for
each scalar c1 and c2 , the function yh = c1 sin t + c2 (1) = c1 sin t + c2 is a
homogeneous solution for all scalars c1 , c2 ∈ R. Now, by Theorem 8 we have

yp + yh = et + c1 sin t + c2

is a solution to Ly = 2et , for all scalars c1 and c2 . At this point we cannot

say that this is the solution set. In fact, there are other solutions that we will
later find. J

Initial Value Problems

Suppose L is a constant coefficient linear differential operator of order n and
f is a function defined on an interval I. Let t0 ∈ I. To the equation

Ly = f

we can associate initial conditions of the form

y(t0 ) = y1 , y 0 (t0 ) = y1 , . . . , y (n−1) (t0 ) = yn−1 .

We refer to the initial conditions and the differential equation Ly = f as an

initial value problem, just as in the first order case. In some applications
y(t0 ) represents initial position, y 0 (t0 ) represents initial velocity, and y 00 (t0 )
represents initial acceleration.
After finding the general solution the initial conditions are used to deter-
mine the arbitrary constants that appear. Consider the following example.

Example 11. Use Example 2 to find the solution to the following initial
value problem
Ly = −2et , y(0) = 1, y 0 (0) = −2,
where L = D2 − 4D + 3

I Solution. In Example 2 we verified

• L(tet ) = −2et ,
• L(et ) = 0,
• L(e3t ) = 0.
Linearity gives
y = tet + c1 et + c2 e3t ,
is a solution to Ly = −2et for every c1 , c2 ∈ R. We will later show all of the
solutions are of this form. Observe that y 0 = et + tet + c1 et + 3c2 e3t . Setting
t = 0 in both y and y 0 gives
170 4 Linear Constant Coefficient Differential Equations

1 = y(0) = c1 + c2
−2 = y 0 (0) = 1 + c1 + 3c2 .

Solving these equations gives c1 = 3, and c2 = −2. Thus y = tet + 3et − 2e3t
is a solution to the given initial value problem. It turns out that y is the only
solution to this initial value problem. Uniqueness of solutions to initial value
problems is the subject of our next section.

Exercises
1-10 Determine which of the following 13. L = D − 2.
are constant coefficient linear differential (a) y = e−2t
equations. In these cases write the equa- (b) y = 3e2t
tion is the form Ly = f , determine the (c) y = tan t
order, the characteristic polynomial, and
14. L = D2 + 1.
whether they are homogeneous.
(a) y = −4 sin t
1. y 00 − yy 0 = 6 (b) y = 3 cos t
(c) y = 1
2. y 000 − 3y 0 = et
15. L = D3 − 4D.
3. y 000 0
+ y + 4y = 0 (a) y = e2t
(b) y = e−2t
4. y 00 + sin(y) = 0 (c) y = 2
5. ty 0 + y = ln t 16. L = D2 − 4D + 8.
(a) y = e2t
6. y 00 + 2y 0 = et (b) y = e2t sin 2t
(c) y = e2t cos 2t
7. y 00 − 7y 0 + 10t = 0
17. Suppose L = D2 − 5D + 4. Verify
8. y 0 + 8y = t that
• L(cos 2t) = 10 sin 2t
9. y 00 + 2 = cos t
• L(et ) = 0
10. 2y 00 − 12y 0 + 18y = 0 • L(e4t ) = 0,
11-16: For the linear operator L, deter- and use this information to find
mine L(y). other solutions to Ly = 10 sin 2t.

11. L = D2 + 3D + 2. 18. Suppose L = D2 −D−6. Verity that

(a) y = et • L(te3t ) = 5e3t
(b) y = e−t • L(e3t ) = 0
(c) y = sin t • L(e−2t ) = 0
and use this information to find
12. L = D2 − 2D + 1.
other solutions to Ly = 5e3t .
(a) y = 4et
(b) y = cos t 19. Suppose L = D3 − 4D. Verity that
(c) y = −e2t • L(te2t ) = 8e2t
 4.2 The Existence and Uniqueness Theorem 171

• L(e2t ) = 0 22. Let L be as in Exercise 18. Use the

• L(e−2t ) = 0 results there to solve the initial value
• L(1) = 0 problem
and use this information to find Ly = 5e3t ,
other solutions to Ly = 8e2t . where y(0) = −1 and y 0 (0) = 8.
20. Suppose L = D4 − 16. Verity that
23. Let L be as in Exercise 19. Use the
• L(sin t) = −15 sin t
results there to solve the initial value
• L(e2t ) = 0
problem
• L(e−2t ) = 0
Ly = 8e2t ,
• L(sin 2t) = 0
• L(cos 2t) = 0 where y(0) = 2, y 0 (0) = 7, and
and use this information to find y 00 (0) = 0.
other solutions to Ly = −15 sin t.
24. Let L be as in Exercise 20. Use the
21. Let L be as in Exercise 17. Use the results there to solve the initial value
results there to solve the initial value problem
problem
Ly = −15 sin t,
Ly = 10 sin 2t,
where y(0) = 0, y 0 (0) = 3, y 00 (0) =
where y(0) = 1 and y 0 (0) = −3. −16, and y 000 (0) = −9.

4.2 The Existence and Uniqueness Theorem

In this section we prove a fundamental theorem for this chapter: The Exis-
tence and Uniqueness Theorem for constant coefficient linear differential
equations. Its statement follows.

Theorem 1 (The Existence and Uniqueness Theorem). Suppose p is

an nth -order real polynomial, L = p(D), and f is a continuous real-valued
function on an interval I. Let t0 ∈ I. Then there is a unique real-valued
function y defined on I satisfying

Ly = f y(t0 ) = y0 , y 0 (t0 ) = y1 , . . . , y (n−1) (t0 ) = yn−1 (1)

where y0 , y1 , . . . , yn−1 ∈ R. If I ⊃ [0, ∞) and f has a Laplace transform

then so does the solution y. Furthermore, if f is in E then y is in E.

You will notice in the statement that when f has a Laplace transform so
does the solution. This theorem thus justifies the use of the Laplace transform
method in Section 3.6. Specifically, when we considered a differential equation
of the form Ly = f and applied the Laplace transform we implicitly assumed
that y, the solution, was in the domain of the Laplace transform. This theorem
now establishes this when the forcing function f has a Laplace transform.
172 4 Linear Constant Coefficient Differential Equations

The proof of this theorem is by induction on the order of L. The main

idea is to factor L = p(D) into linear terms and use what we know about first
order linear differential equations. Consider, for example, L = D 2 − 3D + 2.
The characteristic polynomial is q(s) = s2 − 3s + 2 and factors as q(s) =
(s − 1)(s − 2). Observe,

(D − 1)(D − 2)y = (D − 1)(y 0 − 2y)

= (D − 1)y 0 − (D − 1)(2y)
= y 00 − y 0 − (2y 0 − 2y)
= y 00 − 3y 0 + 2y
= (D 2 − 3D + 2)y = q(D)y.

Therefore q(D) = (D − 1)(D − 2) and since there was nothing special about
the order we can also write q(D) = (D − 2)(D − 1). The key observation why
this factorization works in this way is that the coefficients are scalars and not
functions. (The latter case will be considered in Chapter 6.) More generally, if a
polynomial, q, factors as q(s) = q1 (s)q2 (s) then the corresponding differential
operator, q(D), factors as q(D) = q1 (D)q2 (D).
To get a feel for the inductive proof consider the following example.

Example 2. Solve the second order differential equation

y 00 − 3y 0 + 2y = et , y(0) = 0, y 0 (0) = 1. (2)

I Solution. Let L = D2 − 3D + 2. Then L = (D − 1)(D − 2), as discussed

above. The differential equation in Equation (2) can now be written in the
form
(D − 1)(D − 2)y = et . (3)
Let
u = (D − 2)y = y 0 − 2y. (4)
0
Then u(0) = y (0) − 2y(0) = 1. Substituting u into Equation (3) gives the
first order initial value problem

(D − 1)u = et , u(0) = 1.

An easy calculation using Algorithm ?? of Chapter 2 gives u(t) = (t + 1)et .

Equation (4) now gives

y 0 − 2y = u = (t + 1)et y(0) = 0.

Again an easy calculation gives y = −(t + 2)et + 2e2t . J

Higher order differential equations are done similarly, As you well know,
however, factoring a polynomial into linear terms sometimes introduces com-
plex numbers. For example, s2 + 1 factors into s + i and s − i. Hence, we will
 4.2 The Existence and Uniqueness Theorem 173

need to consider complex valued functions and not just real valued functions.
It is not uncommon in mathematics to achieve a desired result by proving
more than is needed. This is certainly the case here. We will end up proving
a complex-valued analogue of Theorem 1 and on the way you will see the
advantage of recasting this problem in the domain of the complex numbers.

Complex-Valued Solutions
For the most part we will be concerned with a real valued function y defined on
some interval that satisfies Ly = f . However, as we indicated above, there are
times when it is useful to consider a complex valued function z that satisfies
Lz = f . Under the right circumstances complex valued solutions are more
readily obtained and one can then find real valued solutions from the complex
valued solutions. A similar idea was seen in Chapter 3 where we calculated the
Laplace transform of cos t. There we introduced the complex-valued function
eit which, by Euler’s formula, has real part cos t. The Laplace transform of eit
was easy to compute and its real part gave the desired result. Likewise, the
computation of the real standard basis Bq , for q a real polynomial, is simple
to determine via the complex standard basis Bq,C .
Suppose L is a constant coefficient linear differential operator and f (t)
is a complex valued function. A complex valued solution to Lz = f is a
complex valued function z such that Lz = f . Our primary interest will be
when L has real coefficients. However, there will be occasions when it will be
convenient and necessary to allow L to have complex coefficients. The main
results of the previous section, i.e. linearity and its consequences, are also valid
over the complex numbers.
Example 3. Show that z(t) = teit is a solution to

z 00 + z = 2ieit .

I Solution. We use the product rule to compute

z 0 (t) = eit + iteit

and z 00 (t) = 2ieit − teit .

Then we obtain
z 00 + z = 2ieit − teit + teit = 2ieit .
J
Let’s discuss this example a little further. Let L = D2 + 1. The coefficients
are real and hence, when we separate the differential equation Lz = z 00 + z =
2ieit into its real and imaginary parts we obtain

Lx = x00 + x = −2 sin t
Ly = y 00 + y = 2 cos t.
174 4 Linear Constant Coefficient Differential Equations

Now observe that the solution we considered, z(t) = teit , has real part x(t) =
t cos t and imaginary part y(t) = t sin t. Furthermore,

Lx(t) = (t cos t)00 + t cos t = (−2 sin t − t cos t) + t cos t = −2 sin t.

A similar calculation gives, Ly(t) = 2 cos t. Thus the real and imaginary parts
of z(t) = teit satisfy, respectively, the equations above. The following theorem
tells us that this happens more generally.

Theorem 4. Suppose L = q(D), where q is a polynomial with real coefficients.

Suppose f (t) = α(t) + iβ(t) is a complex valued function and z(t) is a complex
valued solution to
Lz = f.
If x(t) and y(t) are the real and imaginary parts of z(t) then x(t) satisfies

Lx = α

and y(t) satisfies

Ly = β.

Proof. Since Dk z = Dk x+iDk y the real part of Dk z is Dk x and the imaginary

part is Dk y. Now suppose ak is a real number then

ak Dk z = ak Dk x + iak Dk y;

the real part is ak Dk x and the imaginary part is ak Dk y. From this it follows
that the real part of Lz is Lx and the imaginary part is Ly. Since Lz = α + iβ
and since two complex valued functions are equal if and only if their real and
complex parts are equal we have

Lx = α

and
Ly = β.
u
t

Remark 5. It is an essential hypothesis that q be a real polynomial. Do you

see where we used this assumption in the proof? Exercises 13-15 explore by
examples what can be said when q is not real.

The First Order Case

We begin the proof of Existence and Uniqueness Theorem by expanding the
first order case established in Section ?? to include complex valued functions.
 4.2 The Existence and Uniqueness Theorem 175

Theorem 6 (First Order). Suppose f is a continuous complex valued func-

tion on an interval I. Let t0 ∈ I and γ ∈ C. Then there is a unique complex
valued function z on I satisfying

z 0 + γz = f z(t0 ) = z0 ∈ C. (5)

If I ⊃ [0, ∞) and f has a Laplace transform then so does the solution y.

Furthermore, if f is in EC then z ∈ EC .

Proof. The proof of this theorem follows the same line of thought given in
Theorem ?? and Corollaries ??. Suppose z is a solution to the initial value
problem given in 5. Using the integrating factor eγt we get

(eγt z)0 = eγt f.

Integrating both sides gives

Z t
γt
e z(t) = eγx f (x) dx + C
t0

and hence Z t
−γt
z(t) = e eγx f (x) dx + e−γt C.
t0

The initial condition z(t0 ) = z0 implies that C = eγt0 z0 and we arrive at the
formula Z t
z(t) = e−γt eγx f (x) dx + e−γ(t−t0 ) z0 . (6)
t0

Since a solution must be expressed by this formula, it must be unique. Fur-

thermore, it is a simple exercise to show that this formula defines a solution,
thus proving existence. Now suppose the domain of f includes [0, ∞) and f
has a Laplace transform. The First Translation Principle, Theorem 3.1.12,
implies that eγt f (t) has a Laplace transform and Theorem ?? implies that its
antiderivative has a Laplace transform. The formula for z in Equation (5) is
made up of a sequence of such operations applied to f . It follows then that z
has a Laplace transform. Furthermore, if f ∈ EC a similar argument implies
z ∈ EC . u
t

There is little value in memorizing Equation 6. To find a solution in the

complex-valued case it is easier to follow Algorithm ??, given earlier in the
real case.

Example 7. Find the solution to

z 0 − iz = 1 z(0) = 1 + i.
176 4 Linear Constant Coefficient Differential Equations

I Solution. Observe that since f (t) = 1 is in EC Theorem 6 implies that the

solution is in EC . Indeed, the integrating factor is e  −i dt = e−it . Multiplying
by the integrating factor gives

(e−it z)0 = e−it .

Integration gives

e−it
e−it z = +C
−i
= ie−it + C.

Hence,
z = i + Ceit ∈ EC .
The initial condition z(0) = 1 + i implies that C = 1 and we thus obtain the
solution
z = i + eit = cos t + i(1 + sin t).
J

The General Case

We proceed by induction to prove the Existence and Uniqueness theorem for
complex-valued functions.

Theorem 8. Suppose p is an nth order complex valued polynomial and f is a

continuous complex valued function on an interval I. Let t0 ∈ I. Then there
is a unique complex valued function z satisfying

p(D)z = f z(t0 ) = z0 , z 0 (t0 ) = z1 , . . . , z (n−1) (t0 ) = zn−1 (7)

where z0 , z1 , . . . , zn−1 ∈ C. If I ⊃ [a, ∞) and f has a Laplace transform

then so does the solution z. Furthermore, if f is in EC then z ∈ EC .

Proof. We may assume p has leading coefficient one. Over the complex num-
bers p has a root γ ∈ C and thus factors as p(s) = (s − γ)p1 (s), where p1 (s)
is a polynomial of degree n − 1. We can write

p(D) = (D − γ)p1 (D).

Suppose p1 (s) = sn−1 + an−2 sn−2 + · · · + a1 s + a0 . Consider the differential

equation

(D − γ)u = f u(t0 ) = zn−1 + an−2 zn−2 + · · · + a1 z1 + a0 z0 .

Theorem 6 implies there is a unique solution u. Now consider the differential

equation
 4.2 The Existence and Uniqueness Theorem 177

p1 (D)z = u z(t0 ) = z0 , z 0 (t0 ) = z1 , . . . , z (n−2) (t0 ) = zn−2 .

By induction there is a unique solution z. Now observe that

p(D)z = (D − γ)p1 (D)z = (D − γ)u = f.

Furthermore,

z (n−1) (t0 ) = (Dn−1 − p1 (D))z(t0 ) + p1 (D)z(t0 )

= −(an−2 Dn−2 + · · · + a1 D + a0 )z(t0 ) + p1 (D)z(t0 )
= −(an−2 zn−2 + · · · + a0 z0 ) + u(t0 ) = zn−1 .

Therefore z is a solution to Equation 7 and this proves existence. If the domain

of f includes [0, ∞) and f has a Laplace transform, by induction, so does u
and hence so does z, by Theorem 6. Further, if f ∈ EC so is u, by induction,
and again z ∈ EC , by Theorem 6.
Now suppose w is another solution to Equation 7. Let v = p1 (D)w. Then
(D − γ)v = (D − γ)p1 (D)w = f and v(t0 ) = p1 (D)w(t0 ) = Dn−1 w(t0 ) +
an−2 Dn−1 w(t0 ) + · · · + a1 Dw(t0 ) + a0 w(t0 ) = zn−1 + an−1 zn−1 + · · · + a0 z0 .
Thus v and u satisfy the same (n − 1)st order initial value problem and by
induction are equal. Therefore p1 (D)w = u and w0 (t0 ) = z0 . Thus z and w
satisfy the same first order initial value problem and again, by the uniqueness
part of Theorem 6, we have w = z. u
t

The existence and uniqueness theorem, Theorem 1, that we stated at the

beginning of this section now follows from Theorems 4 and 8.
To illustrate the inductive proof we provide the following second order
example for which the complex numbers naturally arise. The student is asked
to work out the first order initial value problems that arise. Keep in mind that
we will provide far more efficient methods in the sections to come.

Example 9. Solve the following initial problem

y 00 + y = sin t, y(0) = 0, y 0 (0) = 1.

I Solution. Notice that all values are real and sin t is defined on the interval
(−∞, ∞). Theorem 1 tells us that there is a unique real-valued solution on
(−∞, ∞). Notice that sin t is the imaginary part of eit . We will consider the
complex-valued differential equation

z 00 + z = eit , z(0) = 0, z 0 (0) = i. (8)

Theorem 4 applies and the imaginary part of the solution to Equation (8) is
the solution to Example 9. Let L = D2 + 1 = (D − i)(D + i). Equation (8)
can be written
(D − i)(D + i)z = eit .
178 4 Linear Constant Coefficient Differential Equations

Let u = (D + i)z. Then u(0) = Dz(0) = iz(0) = i. We now solve the first
order initial value problem

(D − i)u = eit , u(0) = i,

and get u(t) = (t + i)eit , (see Exercise 7). Now that u is determined we return
to u = (D + i)z and determine z. The first order equation is

(D + i)z = (t + i)eit , z(0) = 0

and the solution is

3 − 2it it
e ,
z(t) =
4
(see Exercise 8). The imaginary part is
3 t
y(t) = sin t − cos t
4 2
and this gives the solution we seek. J

Exercises

1-6 Use the inductive method, as illus- 12. Let f and g be two complex-
trated in Example 2, to solve each initial valued differential functions. Prove
value problem. the product and quotient rules,
1. y 00 +y 0 −6y = 2et , y(0) = 1, y 0 (0) = 1 namely,

2. y 00 −7y 0 +12y = 0, y(0) = 2,y 0 (0) = 7 (f g)0 = f 0 g + g 0 f

and
3. y 00 − 2y 0 + y = et , y(0) = 1, y 0 (0) = 0 0
f f 0 g − g0f
= .
g  g2
00 0
4. y + y = t, y(0) = 1, y (0) = 1

5. y 00 −2y 0 +2y = 0, y(0) = 0, y 0 (0) = 1 13-15 (Systems of Differential

000 0 0
Equations): Suppose x and y are real-
6. y − y = 1, y(0) = 1, y (0) = −2, valued functions. A system of differen-
y 00 (0) = −1. tial equations consists of two differential
7-10 Solve each initial value problem. equations of the form
7. z 0 − iz = eit , z(0) = i
x0 (t) = ax(t) + by(t) + f (t)
8. z 0 + iz = (t + i)eit , z(0) = 0
y 0 (t) = cx(t) + dy(t) + g(t),
0
9. z − 3iz = sin t, z(0) = 1
where a, b, c, and d, are real numbers
10. z 00 + iz 0 = 1, z(0) = 1, z 0 (0) = i and f and g are real-valued functions.
11. Let γ be a complex number. Show A solution is a pair of functions x and
that y that satisfy both equations simultane-
d γt ously. In Chapter 10 we will discuss this
e = γeγt .
dt topic thoroughly.
 4.3 Linear Homogeneous Differential Equations 179

In the following problems a differen- 17. More generally, suppose φ1 and φ2

tial equation in a complex variable and are solutions to
a solution are given. Deduce the system
and its solution by separating the differ-
ential equation into its real and imagi- y 00 + ay 0 + by = f,
nary parts.
13. z 0 − iz = eit , z = teit
where a and b are real constants and
14. z 0 + (1 + i)z = 0, z = e−(1+i)t f is a continuous function on an in-
terval I. Show that if the graphs of
15. z 0 +(1−i)z = 2(t+i), z = (1+i)t−1 φ1 and φ2 are tangent at some point
16. Suppose φ(t) is a solution to then φ1 = φ2 .
y 00 + ay 0 + by = 0,
where a and b are real constants.
Show that if the graph of φ is tan- φ1 φ2
gent to the t-axis then φ = 0.

4.3 Linear Homogeneous Differential Equations

Suppose L = q(D) where q(s) = an sn + · · · + a1 s + a0 . As we will see in this
section the characteristic polynomial, q, plays a decisive role in determining
the solution set to Ly = 0.
Theorem 1. Suppose L = q(D) is a polynomial differential operator. The
solution set to
Ly = 0
is Eq . This means then that if {y1 , . . . , yn } is the standard basis of Eq then any
solution y to Ly = 0 is of the form

y = a1 y 1 + · · · + an y n .

Proof. The forcing function f (t) = 0 is in E. Thus by Theorem 4.2.1 any

solution to L{y} = 0 is in E. Suppose y is a solution. Then L{Ly} =
q(s)L{y} − p(s) = 0 where p(s) is a polynomial of degree at most n − 1 de-
pending on the initial values y(0), y 0 (0), . . . , y (n−1) (0). Solving for L{y} gives
180 4 Linear Constant Coefficient Differential Equations

p(s)
L{y}(s) = ∈ Rq .
q(s)

This implies y ∈ Eq . On the other hand, suppose y ∈ Eq . Then L{y}(s) =

p(s)
q(s) ∈ Rq and

p(s)
L{Ly}(s) = q(s) − p1 (s) = p(s) − p1 (s),
q(s)

where p1 (s) is a polynomial that depends on the initial conditions. Note, how-
ever, that p(s) − p1 (s) is a polynomial in R and therefore must be identically
0. Thus L{Ly} = 0 and this implies Ly = 0. u
t

Example 2. Find the general solution to the following differential equa-

tions:
1. y 00 + 3y 0 + 2y = 0
2. y 00 + 2y 0 + y = 0
3. y 00 − 6y 0 + 10y = 0
4. y 000 + y 0 = 0

I Solution.
1. The characteristic polynomial for y 00 + 3y 0 + 2y = 0 is

q(s) = s2 + 3s + 2 = (s + 1)(s + 2).

The roots are −1 and −2. The standard basis for Eq is e−t , e−2t . Thus


the solution set is

Eq = c1 e−t + c2 e−2t : c1 , c2 ∈ R .


2. The characteristic polynomial of y 00 + 2y 0 + y = 0 is

q(s) = s2 + 2s + 1 = (s + 1)2 ,

which has root s = −1 with multiplicity 2. The standard basis for Eq is

{e−t , te−t }. The solution set is

Eq = c1 e−t + c2 te−t : c1 , c2 ∈ R .


3. The characteristic polynomial for y 00 − 6y 0 + 10y = 0 is

q(s) = s2 − 6s + 10 = (s − 3)2 + 1.

From this we see

 that the roots of q are 3 + i and 3 − i. The standard
basis for Eq is e3t cos t, e3t sin t . Thus the solution set is

Eq = c1 e3t cos t + c2 e3t sin t : c1 , c2 ∈ R .


 4.3 Linear Homogeneous Differential Equations 181

4. The characteristic polynomial for y 000 + y 0 = 0 is

q(s) = s3 + s = s(s2 + 1).

In this case the roots of q are 0, i, and −i. The standard basis for Eq is
{1, cos t, sin t}. Thus the solution set is

Eq = {c1 + c2 cos t + c3 sin t : c1 , c2 , c3 ∈ R} .

These examples show that conceptually it is a relatively easy process to

write down the solution set to a homogeneous constant coefficient linear dif-
ferential equation. We codify the process in the following algorithm.

Algorithm 3. Given a constant coefficient linear differential equation Ly = 0

the solution set is determined as follows:
1. Determine the characteristic polynomial, q.
2. Determine the roots of q and their multiplicities. This is usually done by
factoring q.
3. Use the roots to find the standard basis of Eq .
4. The solution set is the set of all linear combinations of the functions in
the standard basis.

Initial Value Problems

Let q(D)y = 0 be a homogeneous constant coefficient differential equation,
where deg p = n. Suppose y(t0 ) = y0 , . . . , y (n−1) (t0 ) = yn−1 are initial con-
ditions. The Existence and Uniqueness Theorem 1 states there is a unique
solution to the initial value problem. However, the general solution is a linear
combination of the n functions in the standard basis Eq . It follows then that
the initial conditions uniquely determine the coefficients.

Example 4. Find the solution to

y 000 − 4y 00 + 5y 0 − 2y = 0 y(0) = 1, y 0 (0) = 2, y 00 (0) = 0.

I Solution. The characteristic polynomial is q(s) = s3 − 4s2 + 5s − 2 and

factors as
q(s) = (s − 1)2 (s − 2).
Here, only real roots arise. The basis for Eq is et , tet , e2t and the general


solution is
y = c1 et + c2 tet + c3 e2t .
We first calculate the derivatives and simplify to get,
182 4 Linear Constant Coefficient Differential Equations

y(t) = c1 et + c2 tet + c3 e2t

y 0 (t) = (c1 + c2 )et + c2 tet + 2c3 e2t
y 00 (t) = (c1 + 2c2 )et + c2 tet + 4c3 e2t .

To determine the coefficients c1 , c2 , and c3 we use the initial conditions.

Evaluating at t = 0 gives

1 = y(0) = c1 + c3
2 = y 0 (0) = c1 + c2 + 2c3
0 = y 00 (0) = c1 + 2c2 + 4c3 .

Solving these equations gives c1 = 4, c2 = 4, and c3 = −3. The unique solution

is thus
y(t) = 4et + 4tet − 3e2t .
J

Complex-Valued Solutions
There is a complex-valued analogue to Theorem 1. It’s proof is essentially the
same; we will not repeat it.

Theorem 5. Suppose L = q(D) is a polynomial differential operator with

complex coefficients. The solution set to

Lz = 0

is Eq,C . Thus if {z1 , . . . , zn } is the standard basis of Eq,C then any solution z
to Lz = 0 is of the form

z = a1 z 1 + · · · + an z n .

Example 6. Find the solution to

y 00 + 4y 0 + 5y = 0 y(0) = 1, y 0 (0) = 1.

I Solution. The characteristic polynomial is

q(s) = s2 + 4s + 5 = (s + 2)2 + 1.

We see that q has complex roots γ1 = −2 + i and γ2 = −2 − i. We will

approach this problem in two different ways. First, we will used the complex
basis Bq,C . Second, we will use the real basis Bq . Each have their particular
advantages.
Method 1 The complex basis is Bq,C = {eγ1 t , eγ2 t }. The general solution over
the complex numbers is
 4.3 Linear Homogeneous Differential Equations 183

y(t) = c1 eγ1 t + c2 eγ2 t .

We determine c1 and c2 by the initial conditions. Observe first that
y(t) = c1 eγ1 t + c2 eγ2 t
y 0 (t) = c1 γ1 eγ1 t + c2 γ2 eγ2 t .
The initial conditions imply
1 = y(0) = c1 + c2
1 = y 0 (0) = c1 γ1 + c2 γ2 .
We solve for c1 by multiplying the first equation by γ2 and subtracting. We get
γ2 −1 = c1 (γ2 −γ1 ). Thus c1 = (γ2 −1)/(γ2 −γ1 ) = (−3−i)/(−2i) = (1−3i)/2.
Substituting γ1 into the first equation gives c2 = 1 − c1 = (1 + 3i)/2. Using
Euler’s equation and simplifying we get
1 − 3i γ1 t 1 + 3i γ2 t
y(t) = e + e .
2 2
−2t
= e (cos t + 3 sin t)

Method 2 The real basis for Eq is Bq = e−2t cos t, e−t sin t . The general


solutions over the reals is

y(t) = c1 e−2t cos t + c2 e−2t sin t,
where c1 and c2 are real. We determine c1 and c2 by the initial conditions.
Observe first that
y(t) = = c1 e−2t cos t + c2 e−2t sin t
y 0 (t) = (−2c1 + c2 )e−2t cos t + (−c1 − 2c2 )e−2t sin t.
The initial conditions imply
1 = y(0) = c1
1 = y 0 (0) = −2c1 + c2
from which we find c1 = 1 and c2 = 3. The solution is thus
y = e−2t (cos t + 3 sin t).
J
Remark 7. In the complex case, method 1, it is generally easier to compute
the derivative but harder to deal with the complex entries that arise. While
in the real case, method 2, it is generally harder to compute the derivative
but easier to deal with only the real entries the arise.

Exercises
184 4 Linear Constant Coefficient Differential Equations

1-19 Determine the solution set to the 14. y 00 − 4y 0 − 21y = 0

following homogeneous differential equa-
tions. Write your answer as a linear com- 15. y 000 − y = 0
bination of functions from the standard
16. y 000 − 6y 00 + 12y 0 − 8y = 0
basis.
17. y (4) − y = 0
1. y 0 − 6y = 0
18. y 000 + 2y 00 + y 0 = 0
2. y 0 + 4y = 0
19. y (4) − 5y 00 + 4y = 0
3. y 00 − y 0 − 2y = 0
20-25 Solve the following initial value
4. y 00 + y 0 − 12y = 0 problems.
00 0
5. y + 10y + 24y = 0 20. y 00 − y = 0, y(0) = 0, y 0 (0) = 1
6. y 00 − 4y 0 + 12y = 0 21. y 00 − 3y 0 − 10y = 0, y(0) = 5,
y 0 (0) = 4
7. y 00 + 8y 0 + 16y = 0
22. y 00 − 10y 0 + 25y = 0, y(0) = 0,
8. y 00 − 3y 0 − 10y = 0
y 0 (0) = 1
9. y 00 + 2y 0 + 5y = 0
23. y 00 + 4y 0 + 13y = 0, y(0) = 1,
10. 2y 00 − 12y 0 + 18y = 0 y 0 (0) = −5

11. y 00 + 13y 0 + 36y = 0 24. y 000 + y 00 − y 0 − y = 0, y(0) = 1,

y 0 (0) = 4, y 00 (0) = −1
12. y 00 + 8y 0 + 25y = 0
25. y (4) − y = 0, y(0) = −1, y 0 (0) = 6,
00 0
13. y + 10y + 25y = 0 y 00 (0) = −3, y 000 (0) = 2

4.4 The Method of Undetermined Coefficients

In Section 4.2 we discussed extensively the existence and uniqueness theo-

rem for a nonhomogeneous linear differential equation. Given a polynomial
differential operator q(D), a solution to

q(D)y = f (1)

could be ‘built up’ by factoring q and successively applying first order meth-
ods. Admittedly, this procedure can be tedious. In this section we introduce a
method that exploits linearity and the fact that such solutions are in E when
the forcing functions f is in E. The method is called the method of unde-
termined coefficients. By Theorem 4.1.8, all solutions to Equation (1) are
of the form
y = yp + yh ,
 4.4 The Method of Undetermined Coefficients 185

where yp is a fixed particular solution and yh is any homogenous solution.

In Section 4.3 we found the set of solutions associated to the homogenous
equation Ly = 0. The method of undetermined coefficients is a method to
find a particular solution yp .
Here is the basic idea. We suppose f ∈ E. Theorem 4.2.1 implies a solution
y = yp to Equation (1) is in E. Therefore, the form of yp is

y p = c1 y 1 + · · · + cn y n ,

where each yi , i = 1, . . . n, is a simple exponential polynomial and the co-

efficients c1 , . . . , cn are to be determined. We call yp a test function. The
method of undetermined coefficients can be broken into two parts. First, de-
termine which exponential polynomials, y1 , . . . , yn , will arise in a test function.
Second, determine the coefficients, c1 , . . . , cn .
Before giving the general procedure lets consider the essence of the method
in a simple example.

Example 1. Find the general solution to

y 00 − y 0 − 6y = e−t .

I Solution. Let us begin by finding the solution to the associated homoge-

nous equation
y 00 − y 0 − 6y = 0. (2)
Observe that the characteristic polynomial is q(s) = s2 − s− 6 = (s− 3)(s+ 2).
Thus, the set of homogenous solutions are of the form

yh = c1 e3t + c2 e−2t . (3)

Let us now find a particular solution to y 00 −y 0 −6y = e−t . Since any particular
solution will do, consider the case where y(0) = 0 and y 0 (0) = 0. Since f (t) =
e−t ∈ E we conclude by the Uniqueness and Existence Theorem, Theorem
4.2.1, that the solution y is in E. We apply the Laplace transform to both
sides and get
1
q(s)L {y} (s) = ,
s+1
Solving for L {y} gives
1 1
L {y} (s) = = .
(s + 1)q(s) (s + 1)(s − 3)(s + 2)

It follows that y ∈ Ep , where p is the polynomial p(s) = (s + 1)(s − 3)(s + 2).

Thus
y = c1 e−t + c2 e3t + c3 e−2t ,
for some c1 , c2 , c3 . Observe now that c2 e3t +c3 e−2t is a homogeneous solution.
This means then that leftover piece yp = c1 e−t is a particular solution. This is
186 4 Linear Constant Coefficient Differential Equations

the test function. Let’s now determine c1 by plugging yp into the differential
equation. First, observe that yp0 = −c1 e−t and yp00 = c1 e−t . Thus, Equation
(2) gives
e−t = yp00 − yp0 − 6yp
= c1 e−t + c1 e−t − 6c1 e−t
= −4c1 e−t .
From this we conclude 1 = −4c1 or c1 = −1/4. Therefore
−1 −t
yp = e
4
is a particular solution and the general solution is obtained by adding the
homogeneous solution to it. Thus, the functions
−1 −t
y = yp + yh = e + Ae3t + Be−2t ,
4
where A and B are real numbers, make up the set of real solutions. J
−t
Remark 2. Let v(s) = (s+1) be the denominator of L {e  }. Then v(s)q(s) =
−t 3t −2t
(s + 1)(s − 3)(s + 2) and the standard basis for is . The

E vq e , e , e
standard basis for Eq is e3t , e−2t . Observe that yp is made up of functions

from the standard basis of Evq that are not in the standard basis of Eq . This
will always happen. The general argument is in the proof of the following
theorem.
Theorem 3. Suppose L = q(D) is a polynomial differential operator and
f ∈ E. Suppose Lf = uv . Let Bq denote the standard basis for Eq . Then there
is a particular solution yp to
Ly = f
which is a linear combination of terms in Bqv but not in Bq .
Proof. By Theorem 4.2.1, any solution to Ly = f is in E and hence has a
Laplace Transform. Thus
u(s)
L {q(D)} y) = L {f } =
v(s)
u(s)
q(s)L {y} − p(s) = by Theorem ??
v(s)
p(s) u(s) u(s) + p(s)v(s)
L {y} = = + = .
q(s) q(s)v(s) q(s)v(s)
It follows that y is in Eqv and hence a linear combination of terms in Bqv .
Since Bq ⊂ Bqv , we can write y = yh + yp , where yh is the linear combination
of terms in Bq and yp is a linear combination of terms in Bqv but not in Bq .
Since yh is a homogenous solution it follows that yp = y − yh is a particular
solution of the required form. u
t
 4.4 The Method of Undetermined Coefficients 187

Theorem 3 is the basis for the following algorithm.

Algorithm 4 (The method of undetermined coefficients). Given

q(D)y = f

1. Compute the standard basis, Bq , for Eq .

2. Determine v so that L {f } = uv . I.e. f ∈ Ev .
3. Compute the standard basis, Bvq , for Evq .
4. The test function, yp is the linear combination with arbitrary coefficients
of functions in Bvq that are not in Bq .
5. The coefficients in yp are determined by plugging yp into the differential
equation q(D)y = f .
6. The general solution is given by

yp + yh ,

where yh ∈ Eq .

Example 5. Find the general solution to y 00 − 5y 0 + 6y = 4e2t .

2
I Solution. The characteristic polynomial  2tis q(s)
= s −5s+6
 = (s−2)(s−
4
3) and the standard basis for Eq is Bq = e , e . Since L 4e2t = s−2
3t
we
2
have v(s)q(s)
 2t 2t 3t = (s − 2) (s − 3) and the standard basis for E vq is B vq =
e , te , e . The only function in Bvq that is not in Bq is te2t . Therefore
our test function is yp = c1 te2t . A simple calculation gives

yp = c1 te2t
yp0 = c1 e2t + 2c1 te2t
yp00 = 4c1 e2t + 4c1 te2t .

Substitution into y 00 − 5y 0 + 6y = 4e2t gives

4e2t = 4c1 e2t + 4c1 te2t − 5(c1 e2t + 2c1 te2t ) + 6(c1 te2t )
= −c1 e2t .

From this it follows that c1 = −4 and yp = −4te2t . The general solution is

given by
yp = −4te2t + Ae2t + Be3t ,
where A and B are real. J

Remark 6. Based on Example 1 one might have expected that the test func-
tion in Example 5 would be yp = c1 e2t . But this cannot be as e2t is a ho-
mogenous solution. Observe that v(s) = s − 2 and q(s) = (s − 2)(s + 3)
share a common root, namely s = 2, so that the product has root s = 2
with multiplicity 2. This produces te2t in the standard basis for Evq that does
188 4 Linear Constant Coefficient Differential Equations

not appear in the standard basis for Eq . In Example 1 all the roots of vq
are distinct so this phenomenon does not occur. There is thus a qualitative
difference between the cases when v and q have common roots and the cases
where they do not. However, Algorithm 4 does not distinguish this difference.
It will always produce a test function that leads to a particular solution.

Example 7. Find the general solution to

y 00 − 3y 0 + 2y = 2tet .
2
I Solution. The characteristic polynomial is q(s) = s − 3s + 2 = (s −
t 2t
1)(s − 2). Hence the standard basis for Eq is e , e . The Laplace transform
of 2tet is 2/(s − 1)2 . If v(s) = (s − 1)2 then v(s)q(s) = (s − 1)3 (s − 2)
and y ∈ Evq , which has standard basis et , tet , t2 et , e2t . Our test function
is yp (t) = c1 tet + c2 t2 et . We determine c1 and c2 by plugging yp into the
differential equation. A calculation of derivatives gives

yp = c1 tet + c2 t2 et
yp0 = c1 et + (c1 + 2c2 )tet + c2 t2 et
yp00 = (2c1 + 2c2 )et + (c1 + 4c2 )tet + c2 t2 et .

Substitution into y 00 − 3y 0 + 2y = 2tet gives

2tet = yp00 − 3yp0 + 2yp

= (−c1 + 2c2 )et − 2c2 tet .

Hence, the coefficients c1 and c2 satisfy

−c1 + 2c2 = 0
−2c2 = 2.

From this we find c2 = −1 and c1 = 2c2 = −2. Hence, a particular solution is

yp = −2te−t − t2 e−t

and the general solution is

y = yp + yh = −2te−t − t2 e−t + Aet + Be2t .

Linearity is particularly useful when the forcing function f is a sum of

terms. Here is the general principle, sometimes referred to as the Superpo-
sition Principle.

Theorem 8 (Superposition Principle). Suppose yp1 is a solution to Ly =

f1 and yp2 is a solution to Ly = f2 . Then yp1 +yp2 is a solution to Ly = f1 +f2 .
 4.4 The Method of Undetermined Coefficients 189

Proof. By linearity

L(yp1 + yp2 ) = L(yp1 ) + L(yp2 ) = f1 + f2 .

u
t

Example 9. Find the general solution to

y 00 − 5y 0 + 6y = 12 + 4e2t . (4)

I Solution. Theorem 8 allows us to find a particular solution by adding

together the particular solutions to

y 00 − 5y 0 + 6y = 12 (5)

and
y 00 − 5y 0 + 6y = 4e2t . (6)
In both cases the characteristic polynomial is q(s) = s2 −5s+6 = (s−2)(s−3).
In Equation (5) the Laplace transform of 12 is 12/s. Thus v(s) = s, q(s)v(s) =
s(s − 2)(s − 3), and Bqv = 1, e2t , e3t . The test function is yp1 = c. It is easy
to see that yp1 = 2. In Example 5 we found that yp2 = −4te2t is a particular
solution to Equation (6). By Theorem 8, yp = 2 − 4te2t is a particular solution
to Equation (4). Thus

y = 2 − 4te2t + Ae2t + Be3t

is the general solution. J

There is a complex-valued analogue to Theorem 3 that is proved in exactly

the same way.

Theorem 10 (Method of Undetermined Coefficients: The complex-

valued version). Suppose L = q(D) is a polynomial differential operator
and f ∈ EC . Suppose Lf = uv . Let Bq,C denote the standard basis for Eq,C .
Then there is a particular solution zp to

Lz = f

which is a linear combination of terms in Bqv,C but not in Bq,C .

Example 11. Find the general solution to

y 00 + y = cos t. (7)

I Solution. The characteristic polynomial is q(s) = s2 + 1 which has ±i as

its roots. We will approach this problem in two ways. First, we view Equation
(7) as a real differential equation and apply the Method of Undetermined
Coefficients (real version). Second, we solve the differential equation z 00 + z =
190 4 Linear Constant Coefficient Differential Equations

eit using the complex version, Theorem 10. The real part of that solution gives
the solution to Equation (7) by Theorem 4.

Method 1 The real basis for Eq is Bq = {cos t, sin t}. With v(s) = s2 + 1 as
above we find that the real basis for Evq is Bvq = {cos t, sin t, t cos t, t sin t} . It
follows now that the test function is
yp (t) = c1 t cos t + c2 t sin t.
To determine the coefficients first observe
yp (t) = c1 t cos t + c2 t sin t
yp0 (t) = c1 cos t − c1 t sin t + c2 sin t + c2 t cos t
yp00 (t) = −2c1 sin t − c1 t cos t + 2c2 cos t − c2 t sin t.
Plugging these formulas into y 00 + y = cos t gives
cos t = y 00 + y
= −2c1 sin t − c1 t cos t + 2c2 cos t − c2 t sin t + c1 t cos t + c2 t sin t
= −c1 sin t + 2c2 cos t.
It follows that c1 = 0 and c2 = 1/2. Therefore yp (t) = (1/2)t sin t and the
general solution is
1
y = yp + yh = t sin t + A cos t + B sin t.
2

Method 2 We first consider the complex-valued differential equation z 00 +z =

eit . The standard basis for Eq,C is Bq,C = eit , e−it . The Laplace transform

1
of eit is s−i . Let v(s) = s2 + 1 = (s − i)(s + i). Then v(s)q(s) = (s − i)2 (s + i)
and the standard basis for Evq,C is Bqv,C = eit , teit , e−it . It follows that the


test function is given by zp = cteit . To determine c observe that

zp = cteit
zp0 = c(eit + iteit )
zp00 = c(2ieit − teit ).

Plugging these formulas into z 00 + z = eit gives

eit = zp00 + zp
= 2iceit .
1
Therefore c = 2i = − 2i and zp = − 2i eit . By Theorem 4 the real part of zp is
a solution to y + y = cos t. Thus yp = 21 t sin t. The general solution is
00

1
y = yp + yh = t sin t + A cos t + B sin t.
2
J
 4.4 The Method of Undetermined Coefficients 191

Exercises

1-14: Given q and v below determine 25. y 00 − y = t2

the test function yp for the differential
equation q(D)y = f , where Lf = uv . 26. y 00 − 4y 0 + 4y = et

1. q(s) = s + 1 v(s) = s − 4 27. y 00 − 4y 0 + 4y = e2t

2. q(s) = s − 4 v(s) = s2 + 1 28. y 00 + y = 2 sin t

3. q(s) = s2 − s − 2 v(s) = s − 3 29. y 00 + 6y 0 + 9y = 25te2t

4. q(s) = s2 + 6s + 8 v(s) = s + 3 30. y 00 + 6y 0 + 9y = 25te−3t

5. q(s) = s2 − 5s + 6 v(s) = s − 2 31. y 00 + 6y 0 + 13y = e−3t cos(2t)

6. q(s) = s2 − 7s + 12 v(s) = (s − 4)2 32. y 00 − 8y 0 + 25y = 36te4t sin(3t)

7. q(s) = (s − 5)2 v(s) = s2 + 25 33. y 000 − y 0 = et

8. q(s) = s2 + 1 v(s) = s2 + 4 34. y 000 − y 0 + y 0 − y = 4 cos t

9. q(s) = s2 + 4 v(s) = s2 + 4 35. y (4) − 5y 00 + 4y = e2t

10. q(s) = s2 + 4s + 5 v(s) = (s − 1)2 36. y (4) − y = et + e−t

11. q(s) = (s − 1)2 v(s) = s2 + 4s + 5 37-41 Solve each of the following initial
value problems.
12. q(s) = s3 − s v(s) = s + 1
37. y 00 − 5y 0 − 6y = e3t , y(0) = 2,
3 2
13. q(s) = s − s − s + 1 v(s) = s − 1 y 0 (0) = 1
4
14. q(s) = s − 81 v(s) = s2 + 9 38. y 00 + 2y 0 + 5y = 8e−t , y(0) = 0,
y 0 (0) = 8
15-36 Find the general solution for
each of the differential equations given 39. y 00 + y = 10e2t , y(0) = 0, y 0 (0) = 0
below.
40. y 00 − 4y = 2 − 8t, y(0) = 0, y 0 (0) = 5
0
15. y + 6y = 3
41. y 000 − y 00 + 4y 0 − 4y = et , y(0) = 0,
16. y 0 + y = 4e3t y 0 (0) = 0, y 00 (0) = 0

17. y 0 − 4y = 2e4t
42-51 (The Annihilator) Suppose f
18. y 0 − y = 2 cos t is a function on some interval. If v(D)f =
0 for some polynomial v then we say
19. y 00 + 3y 0 − 4y = e2t that f is annihilated by v(D). For ex-
ample, e3t is annihilated by D − 3 and
20. y 00 − 3y 0 − 10y = 7e−2t D2 − 5D + 6. (check this!) The annihi-
21. y 00 + 2y 0 + y = et lator of f is the polynomial differential
operator v(D) of least degree and lead-
22. y 00 + 3y 0 + 2y = 4 ing coefficient 1 that annihilates f . For
example, the annihilator of e3t is D − 3.
23. y 00 + 4y 0 + 5y = e−3t
Find the annihilator of each function be-
24. y 00 + 4y = 1 + et low.
192 4 Linear Constant Coefficient Differential Equations

42. e5t 50. Suppose L = q(D) is a polynomial

differential operator and y is a solu-
43. te−6t tion to
44. cos(3t)
Ly = f,
2t
45. e cos(4t)
where f ∈ E . Suppose v(D) is
46. t2 sin t the annihilator of f . Show that
47. tn eat q(D)v(D) annihilates y. Give an ex-
ample to show that q(D)v(D) is not
48. Show that f has an annihilator if
necessarily the annihilator of y.
and only if f ∈ E .
49. Suppose f ∈ E and L{f } = uv , re- 51. (continued) What does knowing
duced. Show that the annihilator of that y is annihilated by q(D)v(D)
f is v(D). tell you about the form y takes?

4.5 The Incomplete Partial Fraction Method

In this section we provide an alternate method for finding the solution set to
the nonhomogeneous differential equation

q(D)y = f, (1)
where f ∈ E. This alternate method begins with the Laplace transform
method and exploits the efficiency of the partial fraction decomposition algo-
rithm. However the partial fraction decomposition applied to Y (s) = L{y}(s)
is not needed to its completion. We therefore refer to this method as the
Incomplete Partial Fraction Method. For purposes of illustration and
comparison to the method of undetermined coefficients let’s reconsider Ex-
ample 5 of Section 4.4.
Example 1. Find the general solution to

y 00 − 5y 0 + 6y = 4e2t . (2)

I Solution. Our goal is to find a particular solution yp to which we will

add the homogeneous solutions yh . Since any particular solution will do we
begin by applying the Laplace transform with initial conditions y(0) = 0 and
y 0 (0) = 0. The 2
characteristic polynomial is q(s) = s − 5s + 6 = (s − 2)(s − 3)
2t
and L 4e = 4/(s − 2). If, as usual, we let Y (s) = L{y}(s) then q(s)Y (s) =
4/(s − 2) and
4 4
Y (s) = = .
(s − 2)q(s) (s − 2)2 (s − 3)
In the table below we compute the (s − 2)-chain for Y (s) but stop when the
denominator reduces to q(s) = (s − 2)(s + 3).
 4.5 The Incomplete Partial Fraction Method 193

Incomplete (s − 2)-chain

4 −4
(s − 2)2 (s − 3) (s − 2)2

p(s)
(s − 2)(s − 3)

The table tells us that

4 −4 p(s)
2
= 2
+ .
(s − 2) (s − 3) (s − 2) (s − 2)(s − 3)

There is no need to compute p(s) and finish out the table since the inverse
p(s)
Laplace transform of (s−2)(s−3) = p(s)
q(s) is a homogenous solution. If Yp (s) =
−4 −1
(s−2)2 then yp (t) = L {Yp }(t) = −4te2t is a particular solution. By linearity
we get the general solution:

y(t) = −4te2t + Ae2t + Be3t .

Observe that the particular solution yp we have obtained here is exactly

what we derived using the Method of Undetermined Coefficients in Exam-
ple 5 of Section 4.4 and is obtained by one iteration of the Partial Fraction
Decomposition Algorithm.

The Incomplete Partial Fraction Method

We now proceed to describe the procedure generally. Consider the differential
equation
q(D)y = f,
where q is a real polynomial and f ∈ E. Suppose L{f } = uv . Since we are only
interested in finding a particular solution we can choose initial conditions that
are convenient for us. Assume y(0) = 0, y 0 (0) = 0, . . . , y (n−1) (0) = 0 where n
is the degree of q. As usual let Y (s) = L{y}(s). Then

u(s)
Y (s) = .
v(s)q(s)

Let’s consider the linear case where v(s) = (s − γ)m and u(s) has no factors
of s − γ. (The quadratic case, v(s) = (s2 + cs + d)m is handled similarly.) This
194 4 Linear Constant Coefficient Differential Equations

means f (t) = p(t)eγt , where the degree of p is m − 1. It could be the case that
γ is a root of q with multiplicity j, in which case we can write

q(s) = (s − γ)j qγ (s),

where qγ (γ) 6= 0. Thus

u(s)
Y (s) = .
(s − γ)m+j qγ (s)

For convenience let p0 (s) = u(s). We now iterate the Partial Fraction Decom-
position Algorithm until the denominator is q. This occurs after m iterations.
The incomplete (s − γ)-chain is given by the table below.

Incomplete (s − γ)-chain

p0 (s) A1
(s − γ)m+j qγ (s) (s − γ)m+j

p1 (s) A2
(s − γ)m+j−1 qγ (s) (s − γ)m+j−1

.. ..
. .

pm−1 (s) Am
(s − γ)j+1 qγ (s) (s − γ)j+1

pm (x)
(s − γ)j qγ (s)

The last entry in the first column has denominator q(s) = (s − γ)j qγ (s)
and hence its inverse Laplace transform in a homogeneous solution. It follows
A1 Am
that if Yp (s) = (s−γ) m+j + · · · + (s−γ)j+1 then

A1 Am j−1 γt
yp (t) = tm+j−1 eγt + · · · + t e
(m + j − 1)! (j)!

is a particular solution.

To illustrate this general procedure let’s consider two further examples.

 4.5 The Incomplete Partial Fraction Method 195

Example 2. Find the general solution to

y 000 − 2y 00 + y 0 = tet . (3)

I Solution. The characteristic polynomial is q(s) = s3 − 2s2 + s = s(s − 1)2

1
and L{tet } = (s−1) 0
2 . Again assume y(0) = 0 and y (0) = 0 then

1
Y (s) = .
(s − 1)4 s
The incomplete (s − 1)-chain for Y (s) is

Incomplete (s − 1)-chain

1 1
(s − 1)4 s (s − 1)4
−1 −1
(s − 1)3 s (s − 1)3
p(s)
(s − 1)2 s

1 1
Let Yp (s) = (s−1)4 − (s−1)3 . Then a particular solution is

1 3 t 1
yp (t) = L−1 {Yp }(t) = t e − t2 e t .
3! 2!
The homogeneous solution can be read off from the roots of q(s) = s(s − 1)2 .
Thus the general solution is
1 3 t 1 2 t
y= t e − t e + c1 + c2 et + c2 tet .
6 2
J
Example 3. Find the general solution to

y (4) − y = 4 cos t. (4)

I Solution. The characteristic polynomial is q(s) = s4 −1 = (s2 −1)(s2 +1)

and L {4 cos t} = s24s 0
+1 . Again assume y(0) = 0 and y (0) = 0. Then

4s
Y (s) = .
(s2 − 1)(s2 + 1)2
Using the irreducible quadratic partial fraction method we obtain the incom-
plete (s2 + 1)-chain for Y (s):
196 4 Linear Constant Coefficient Differential Equations

Incomplete (s2 + 1)-chain

4s −2s
(s2 − 1)(s2 + 1)2 (s2+ 1)2
p(s)
(s2 − 1)(s2 + 1)

By Table ?? we have
 
−2s
yp = L−1 = −t sin t.
(s2 + 1)2

The homogeneous solution as yh = c1 et + c2 e−t + c3 cos t + c4 sin t and thus

the general solution is

y = yp + yh = −t sin t + c1 et + c2 e−t + c3 cos t + c4 sin t.

Exercises

1-20 Use the Incomplete Partial Frac- 10. y 00 + 3y 0 − 10y = sin t

tion Method to solve the following differ-
ential equations. 11. y 000 − 9y 0 = 18t

1. y 0 + 4y = 2e7t 12. y 00 + 6y 0 + 9y = 25te2t

2. y 0 − 3y = e3t 13. y 00 − 5y 0 − 6y = 10te4t

3. y 0 + 5y = te4t 14. y 00 − 8y 0 + 25y = 36te4t sin(3t)

4. y 0 − 2y = t2 et 15. y 00 − 4y 0 + 4y = te2t

5. y 00 − 4y = e−6t 16. y 00 + 2y 0 + y = cos t

6. y 00 + 2y 0 − 15y = 16et 17. y 00 + 2y 0 + 2y = et cos t

7. y 00 + 5y 0 + 6y = e−2t 18. y 000 + 4y 0 = t cos t

8. y 00 + 3y 0 + 2y = 4 19. y 000 − y 0 = et

9. y 00 + 2y 0 − 8y = 6e−4t 20. y (4) − y = et + e−t

5
System Modeling and Applications

Many laws in physics and engineering are best expressed in terms of how a
physical quantity changes with time. For example, Newton’s law of heating
and cooling, discussed in Chapter 2, was expressed in terms of the change in
temperature and led to a first order differential equation: dT /dt = k(T − T◦ ).
In this chapter we will discuss two other important physical laws: Newton’s
law of motion and Kirchhoff’s electrical laws. Each of these laws are likewise
expressed in terms of change of a physical quantity.
Newton’s law of motion will be used to model the motion of a body un-
der the influence of gravity, a spring, a dashpot, and some external force.
Kirchhoff’s laws will be used to model the charge in a simple electrical circuit
influenced by a resistor, a capacitor, and inductor, and an externally applied
voltage. Each model leads to a second order constant coefficient differential
equation.
To begin with, it is worthwhile discussing modeling of physical systems
apart from a particular application.

5.1 System Modeling

Modeling involves understanding how a system works mathematically. By a
system we mean something that takes inputs and produces outputs such as
might be found in the biological, chemical, engineering, and physical sciences.
The core of modeling thus involves expressing how the outputs of a system
can be mathematically described as a function of the inputs. The following
system diagram represents the inputs coming in from the left of the system
and outputs going out on the right.

f (t) System y(t)

198 5 System Modeling and Applications

For the most part, inputs and outputs will be quantities that are time-
dependent; they will be represented as functions of time. There are occasions,
however, where other parameters such as position or frequency are used in
place of time. An input-output pair, (f (t), y(t)), implies a relationship which
we denote by
y(t) = Φ(f )(t).
Notice that Φ is an operation on the input function f and produces the output
function y. In a certain sense, understanding Φ is equivalent to understanding
the workings of the system. Frequently we identify the system under study
with Φ itself. Our goal in modeling then is to give an explicit mathematical
description of Φ.
In many settings a mathematical model can be described implicitly by
a constant coefficient linear differential equation and its solution gives an
explicit description. That is the assumption we will make about systems in
this section. In Chapter 2 we modeled a mixing problem by a first order
constant coefficient linear differential equation. In this chapter we will model
spring systems and electrical circuit systems by second order equations. All of
these models have common features which we explore in this section. To get
a better idea of what we have in mind let’s reconsider the mixing problem as
an example.

Example 1. Suppose a tank holds 10 liters of a brine solution with initial

concentration a/10 grams of salt per liter. (Thus, there are a grams of salt
in the tank, initially.) Pure water flows in an intake tube at a rate of b liters
per minute and the well mixed solution flows out of the tank at the same
rate. Attached to the intake is a hopper containing salt. The amount of salt
entering the intake is controlled by a valve and thus varies as a function of
time. Let f (t) be the rate (in grams of salt per minute) at which salt enters
the system. Let y(t) represent the amount of salt in the tank at time t. Find
a mathematical model that describes y.

Salt

Pure Water
 5.1 System Modeling 199

I Solution. As in Chapter 2 our focus is on the way y changes. Observe

that
y 0 = Input Rate − Output Rate.
By Input Rate we mean the rate at which salt enters the system and this is
just f (t). The Output Rate is the rate at which salt leaves the system and is
given by the product of the concentration of salt in the tank, y(t)/10, and the
flow rate b. We are thus led to the following initial value problem:
b
y0 + y = f (t), y(0) = a. (1)
10
In this system f (t) is the input function and y(t) is the output function. In-
corporating this mathematical model in a system diagram gives the following
implicit description:

b
Solve y 0 + 10 y = f (t),
f (t) y(t).
y(0) = a

Using Algorithm ?? to solve Equation (1) gives

Z t
−bt bt bx
y(t) = ae 10 + e− 10 f (x)e 10 dx
0
Z t
−bt b(t−x)
= ae 10 + f (x)e− 10 dx
0
= ah(t) + f ∗ h(t),
−bt
where h(t) = e 10 and f ∗ h denotes the convolution of f with h. We therefore
arrive at an explicit mathematical model Φ for the mixing system:

Φ(f )(t) = ah(t) + f ∗ h(t).

J
As we shall see this simple example illustrates many of the main features
shared by all systems modeled by a constant coefficient differential equation.
You notice that the output consists of two pieces: ah(t) and f ∗ h(t). If a = 0
then the initial state of the system is zero, i.e. there is no salt in the tank at
time t = 0. In this case, the output is

y(t) = Φ(f )(t) = f ∗ h(t). (2)

This output is called the zero-state response; it represents the response

of the system by purely external forces of the system and not on any ini-
tial condition or state. The zero-state response is the particular solution to
b
p(D)y = f , y(0) = 0, where p(D) = D + 10 .
200 5 System Modeling and Applications

On the other hand, if f = 0 the output is ah(t). This output is called

the zero-input response; it represents the response based on purely inter-
nal conditions of the system and not on any external inputs. The zero-input
response is the homogenous solution to p(D)y = 0 with initial condition
y(0) = a. The total response is the sum of the zero-state response and the
zero-input response.
The function h is called the unit-impulse response function. It is the
zero-input response with a = 1 and completely characterizes the system. Al-
ternatively, we can understand h in terms of the mixing system by opening
the hopper for a very brief moment just before t = 0 and letting 1 gram of salt
enter the intake. At t = 0 the amount of salt in the tank is 1 gram and no salt
enters the system thereafter. The expression ‘unit-impulse’ refers to the fact
that a unit of salt (1 gram) enters the system and does so instantaneously,
i.e. as an impulse. Such impulsive inputs will be discussed more thoroughly in
Chapter 8. Over time the amount of salt in the tank will diminish according
−bt
to the unit-impulse response h = e 10 as illustrated in the graph below with
b = 5:

Once h is known the zero-state response for an input f is completely

determined by Equation (2). One of the main points of this section will be to
show that in all systems modeled by a constant coefficient differential equation
the zero-state response is given by this same formula, for some h. We will also
show how to find h.
With a view to a more general setting let q(D) be an nth -order differential
operator and let a = (a0 , a1 , . . . , an−1 ) be a vector of n scalars. Suppose Φ
is a system. We say Φ is modeled by q(D) with initial state a if for
each input function f the output function y = Φ(f ) satisfies q(D)y = f ,
and y(0) = a0 , y 0 (0) = a1 , . . . , y (n−1) (0) = an−1 . Sometimes we say that the
initial state of Φ is a. By the Uniqueness and Existence Theorem, Theorem
4.2.1, y is unique. In terms of a system diagram we have

Solve q(D)y = f
f (t) with initial state a y(t).

The definitions of zero-state response, zero-input response, and total response

given above naturally extend to this more general setting.
 5.1 System Modeling 201

The Zero-Input Response

First, we consider the response of the system with no external inputs, i.e.
f (t) = 0. This is the zero-input response and is a result of internal initial
conditions of the system only. Such initial conditions could represent the initial
position and velocity in a system where the motion of an object is under
consideration. In electrical systems the initial conditions could include the
amount of initial stored energy. In the mixing system we described earlier the
only initial condition is the amount of salt a in the tank at time t = 0.
The zero-input response is the solution to

q(D)y = 0, y(0) = a0 , y 0 (0) = a1 , . . . , y (n−1) (0) = an−1 .

By Theorem 4.3.1 y is a linear combination of functions in the standard basis

Bq . The roots λ1 , . . . , λk ∈ C of q are called the characteristic values and
the functions in the standard basis are called the characteristic modes of
the system . If the corresponding multiplicities of the characteristic values are
m1 , . . . , mk , then the zero-input response is of the form

y(t) = p1 (t)eλ1 t + · · · + pk (t)eλk t ,

where the degree of pi is less than mi , 1 ≤ i ≤ k. Consider the following

example.

Example 2. For each problem below a system is modeled by q(D) with

initial state a. Plot the characteristic values in the complex plane, determine
the zero-input response, and graph the response.
a. q(D) = D + 3, a = 2
b. q(D) = (D − 3)(D + 3), a = (0, 1)
c. q(D) = (D + 1)2 + 4, a = (1, 3)
d. q(D) = D 2 + 4, a = (1, 2)
e. q(D) = (D 2 + 4)2 , a = (0, 0, 0, 1).

I Solution. The following table summarizes the calculations that we ask

the reader to verify.
Characteristic Zero-input Graph
Values Response

-3

(a) y = 2e−3t
202 5 System Modeling and Applications

-3 3

1 3t
(b) y= 6 (e − e−3t )

-1+2i

-1-2i
(c) y = e−t (cos(2t) + 2 sin(2t))

2i

-2i
(d) y = cos(2t) + sin(2t)

2i 2

-2i 2
1
(e) y= 16 (2t cos(2t) + sin(2t))

In part (e) we have indicated the multiplicity of ±2i by a 2 to the right of the
characteristic value. J
 5.1 System Modeling 203

The location of the characteristic values in the complex plane is related to

an important notion called system stability.

stability
System stability has to do with the long term behavior of a system. If the
zero-input response of a system, for all initial states, tends to zero over time
then we say the system is asymptotically stable. This behavior is seen
in the mixing system of Example 1. For any initial state a the zero-input
−bt
response y(t) = ae 10 has limiting value 0 as t → ∞. In the case a = 2 and
b = 30 the graph is the same as that given in Example 2a. Notice that the
system in Example 2c is also asymptotically stable since zero-input response
always takes the form y(t) = e−t (A sin(2t) + B cos(2t)). The function t →
A sin(2t) + B cos(2t) is bounded, so the presence of e−t guarantees that the
limit value of y(t) is 0 as t → ∞. In both Example 2a and 2c the characteristic
values lie in the left-half side of the complex plane. More generally, suppose
tk eλt is a characteristic mode for a system. If λ = α + iβ and α < 0 then

lim tk eλt = 0,
t→∞

for all nonnegative integers k. On the other hand, if α > 0 then the charac-
teristic mode, tk eλt is unbounded. Thus a system is asymptotically stable if
and only if all characteristic values lie to the left of the imaginary axis.
If a system is not asymptotically stable but the zero-input response is
bounded for all possible initial states then we say the system is marginally
stable. Marginal stability is seen in Example 2d and occurs when one or more
of the characteristic values lie on the imaginary axis and have multiplicity
exactly one. Those characteristic values that are not on the imaginary axis
must be to the left of the imaginary axis.
We say a system is unstable if there is an initial state in which the zero-
input response is unbounded over time. This behavior can be seen in Example
2b and 2e. Over time the response becomes unbounded. Of course, in a real
physical system this cannot happen. The system will break or explode when
it passes a certain threshold. Unstable systems occur for two distinct reasons.
First, if one of the characteristic values is λ = α + iβ and α > 0 then λ
lies in the right-half side of the complex plane. In this case the characteristic
mode is of the form tk eλt . This function is unbounded as a function of t. This
is what happens in Example 2b. Second, if one of the characteristic values
λ = iβ lies on the imaginary axis and, in addition, the multiplicity is greater
than one then the characteristic mode is of the form tk eiβt , k ≥ 1. This mode
oscillates unboundedly as a function of t > 0, as in Example 2e. Remember,
it only takes one unbounded characteristic mode for the whole system to be
unstable.

Example 3. Determine the stability of each system modeled by q(D) below.

204 5 System Modeling and Applications

1. q(D) = (D + 1)2 (D + 3)
2. q(D) = (D 2 + 9)(D + 4)
3. q(D) = (D + 4)2 (D − 5)
4. q(D) = (D 2 + 1)(D2 + 9)2

I Solution.
1. The characteristic values are λ = −1 with multiplicity 2 and λ = −3. The
system is asymptotically stable.
2. The characteristic values are λ = ±3i and λ = −4. The system is mar-
ginally stable.
3. The characteristic values are λ = −4 with multiplicity 2 and λ = 5. The
system is unstable.
4. The characteristic values are λ = ±i and λ = ±3i with multiplicity 2. The
system is unstable.

the unit-impulse response function

Suppose Φ is a system modeled by q(D), an nth order constant coefficient
differential operator. The unit-impulse response function h(t) is the zero-
input response to Φ when the initial state of the system is a = (0, . . . , 0, 1).
More specifically, h(t) is the solution to

q(D)y = 0 y(0) = 0, . . . , y (n−2) (0) = 0, y (n−1) (0) = 1.

If n = 1 then y(0) = 1 as in the mixing system discussed in the beginning

of this section. In this simple case h(t) is a multiple of a single characteristic
mode. For higher order systems, however, the unit-impulse response function
is a homogeneous solution to q(D)y = 0 and, hence, a linear combination of
the characteristic modes of the system.

Example 4. Find the unit-impulse response function for a system Φ mod-

eled by
q(D) = (D + 1)(D2 + 1).

I Solution. It is an easy matter to apply the Laplace transform method to

q(D)y = 0 with initial condition y(0) = 0, y 0 (0) − 0, and y 00 (0) = 1. We get
q(s)Y (s) − 1 = 0. A short calculation gives
 
1 1 1 2 −2s + 2
Y (s) = = = + .
q(s) (s + 1)(s2 + 1) 4 s+1 s2 + 1
1
The inverse Laplace transform is y(t) = (e−t +sin t−cos t). The unit-impulse
2
response function is thus
1 −t
h(t) = (e + sin t − cos t).
2
 5.1 System Modeling 205

J
1
Observe in this example that h(t) = L−1 { q(s) }(t). It is not hard to see
that this formula extends to the general case. We record this in the following
theorem. The proof is left to the reader.
Theorem 5. Suppose a system Φ is modeled by a constant coefficient differ-
ential operator q(D). The unit-impulse response function, h(t), of the system
Φ is given by  
1
h(t) = L−1 (t).
q(s)

The Zero-State Response

Let’s now turn our attention to a system Φ in the zero-state and consider the
zero-state response. This occurs precisely when a = 0, ie. all initial conditions
of the system are zero. We should think of the system as initially being at rest.
We continue to assume that Φ is modeled by an nth -order constant coefficient
differential operator q(D). Thus, for each input f (t) the output y(t) satisfies
q(D)y = f with y and its higher derivatives up to order n − 1 all zero at t = 0.
An important feature of Φ in this case is its linearity.
Proposition 6. Suppose Φ is a system modeled by q(D) in the zero-state.
Then Φ is linear. Specifically, if f , f1 and f2 are input functions and c is a
scalar then
1. Φ(f1 + f2 ) = Φ(f1 ) + Φ(f2 )
2. Φ(cf ) = cΦ(f )
Proof. If f is any input function then Φ(f ) = y if and only if q(D)y = f and
y(0) = y 0 (0) = · · · = y (n−1) (0) = 0. If y1 and y2 are the zero-state response
functions to f1 and f2 , respectively, then the linearity of q(D) implies
q(D)(y1 + y2 ) = q(D)y1 + q(D)y2 = f1 + f2 .
Furthermore, since the initial state of both y1 and y2 are zero so is the initial
state of y1 + y2 . This implies Φ(f1 + f2 ) = y1 + y2 . In a similar way,
q(D)(cy) = cq(D)y = cf,
by the linearity of q(D). The initial state of cy is clearly zero. So Φ(cf ) =
cΦ(f ). u
t
Theorem 7. Suppose Φ is a system modeled by q(D) in the zero-state. If f is
a continuous input function on an interval which includes zero then the zero-
state response is given by the convolution of f with the unit-impulse response
function h. I.e.
Z t
φ(f )(t) = f ∗ h(t) = f (x)h(t − x) dx.
0
206 5 System Modeling and Applications

Before we proceed with the proof let’s introduce the following helpful
lemma.

Lemma 8. Suppose f is continuous on a interval I containing 0. Suppose h

is differentiable on I. Then

(f ∗ h)0 (t) = f (t)h(0) + f ∗ h0 .

Proof. Let y(t) = f ∗ h(t). Then

∆y y(t + ∆t) − y(t)

=
∆t ∆t !
Z t+∆t Z t
1
= f (x)h(t + ∆t − x) dx − f (x)h(t − x) dx
∆t 0 0
Z t Z t+∆t
h(t + ∆t − x) − h(t − x) 1
= f (x) dx + f (x)h(t + ∆t − x) dx.
0 ∆t ∆t t

Rt
We now let ∆t go to 0. The first summand has limit 0 f (x)h0 (t − x) dx =
f ∗ h0 (t). By the Fundamental Theorem of Calculus the limit of the second
summand is obtained by evaluating the integrand at x = t, thus getting
f (t)h(0). The lemma now follows by adding these two terms. u
t

Proof (of Theorem 7). Let h(t) be the unit-impulse response function. Then
h(0) = h0 (0) = · · · = h(n−2) (0) = 0 and h(n−1) (0) = 1. Set y(t) = f ∗ h(t).
Repeated applications of Lemma 8 to y gives

y 0 = f ∗ h0 + h(0)f = f ∗ h
y 00 = f ∗ h00 + h0 (0)f = f ∗ h00
..
.
y (n−1) = f ∗ h(n−1) + h(n−2) (0)f = f ∗ h(n−1)
y (n) = f ∗ h(n) + h(n−1) (0)f = f ∗ h(n) + f.

From this it follows that

q(D)y = f ∗ q(D)h + f = f,

since q(D)h = 0. It is easy to check that y is in the zero-state. Therefore

Φ(f ) = y = f ∗ h.

u
t
 5.1 System Modeling 207

At this point let us make a few remarks on what this theorem tells us. The
most remarkable thing is the fact that Φ is precisely determined by the unit-
impulse response function h. From a mathematical point of view, knowing h
means you know how the system Φ works in the zero-state. Once h is deter-
mined all output functions, i.e. system responses, are given by the convolution
product, f ∗ h, for an input function f . Admittedly, convolution is an unusual
product. It is not at all like the usual product of functions where the value (or
state) at time t is determined by knowing just the value of each factor at time
t. Theorem 7 tells us that the state of a system response at time t depends
on knowing the values of the input function f for all x between 0 and t. The
system ‘remembers’ the whole of the input f up to time t and ‘meshes’ those
inputs with the internal workings of the system, as represented by impulse
response function h, to give f ∗ h(t).
Since the zero-state response, f ∗ h, is a solution to q(D)y = f , with zero
initial state, it is a particular solution. This formula thus gives an alternative
to the method of undetermined coefficients and the incomplete partial fraction
method. We can also take the opposite view. The method of undetermined
coefficients and the incomplete partial fraction method are alternative ways
to compute a convolution f ∗ h, when f and h are in E. In practice, computing
convolutions can be time consuming and tedious. In the following examples
we will limit the inputs to functions in EC and use Table ??.

Example 9. A system Φ is modeled by q(D). Find the zero-state response

for the given input function.
a) q(D) = D + 2 and f (t) = 1
b) q(D) = D 2 + 4D + 3 and f (t) = e−t
c) q(D) = D 2 + 4 and f (t) = cos(2t)

I Solution.
a) The characteristic polynomial is q(s) = s + 2 and therefore the character-
istic mode is e−2t . It follows that h(t) = Ae−2t and with initial condition
h(0) = 1 we get h(t) = e−2t . The system response, y(t), for the input 1 is
1
y(t) = e−2t ∗ 1(t) = (1 − e−2t ).
2
b) The characteristic polynomial is q(s) = s2 + 4s + 3 = (s + 1)(s + 3). The
characteristic modes are e−3t and e−t . Thus h(t) has the form h(t) =
Ae−t + Be−3t . The initial conditions h(0) = 0 and h0 (0) = 1 implies
1 −t
h(t) = (e − e−3t ).
2
The system response to the input function f (t) = e−t is
1 −t 1 1 1
y(t) = (e − e−3t ) ∗ e−t = te−t − e−t + e−3t .
2 2 4 4
208 5 System Modeling and Applications

1
c) It is easy to verify that the impulse response function is h(t) = 2 sin(2t).
The system response to the input function f (t) = cos(2t) is
1 1
y(t) = sin(2t) ∗ cos(2t) = t sin(2t).
2 4
J

bounded-in bounded-out
In Example 9a we introduce a bounded input, f (t) = 1, and the response
y(t) = 12 (1 − e−2t ) is also bounded, by 21 , in fact. On the other hand, in
Example 9c, we introduce a bounded input, f (t) = cos 2t, yet the response
y(t) = 14 t sin 2t oscillates unboundedly. We say that a system Φ is BIBO-
stable if for every bounded input f (t) the response function y(t) is likewise
bounded. (BIBO stands for ‘bounded input bounded output’.) Note the fol-
lowing theorem. An outline of the proof is given in the exercises.
Theorem 10. Suppose Φ is an asymptotically stable system. Then Φ is BIBO-
stable.
Unstable systems are of little practical value to an engineer designing a
‘safe’ system. In an unstable system a set of unintended initial states can lead
to an unbounded response that destroys the system entirely. Even marginally
stable systems can have bounded input functions that produce unbounded
output functions. This is seen in Example 9c, where the system response to
the bounded input f (t) = cos(2t) is the unbounded function y(t) = 14 t sin(2t).
Asymptotically stable systems are thus the ‘safest’ systems since they are
BIBO-stable. They produce at worst a bounded response to a bounded input.
However, this is not to say that the response can not be destructive. We will
say more about this in the following topic.

resonance
We now come to a very interesting and important phenomenon called res-
onance. Loosely speaking, resonance is the phenomenon that occurs when
a system reacts very energetically to a relatively mild input. Resonance can
sometimes be catastrophic for the system. For example, a wine glass has a
natural frequency at which it will vibrate. You can hear this frequency by
rubbing your moistened finger around the rim of the glass to cause it to vi-
brate. An opera singer who sings a note at this same frequency with sufficient
intensity can cause the wine glass to vibrate so much that it shatters. Reso-
nance can also be used to our advantage as is familiar to a musician tuning a
musical instrument to a standard frequency given, for example, by a tuning
fork. Resonance occurs when the instrument is ‘in tune’.
The characteristic values of a system Φ are sometimes referred to as the
characteristic frequencies. As we saw earlier the internal workings of a
 5.1 System Modeling 209

system are governed by these frequencies. A system that is energized tends

to operate at these frequencies. Thus, when an input function matches an
internal frequency the system response will generally be quite energetic, even
explosive.
A dramatic example of this occurs when a system is marginally stable.
Consider the following example.

Example 11. A zero-state system Φ is modeled by the differential equation

(D 2 + 1)y = f.

Determine the impulse response function h and the system response to the
following inputs.
1. f (t) = sin(πt)
2. f (t) = sin(1.25t)
3. f (t) = sin(t)
Discuss the resonance that occurs.

I Solution. The characteristic values are ±i with multiplicity one. Thus the
system is marginally stable. The unit-impulse response h(t) is the solution to
(D 2 + 1)y = 0 with initial conditions y(0) = 0 and y 0 (0) = 1. If q(s) = s2 + 1
then Bq = {sin t, cos t} and h(t) = A cos(t) + B sin(t). The initial conditions
imply A = 0 and B = 1 and therefore h(t) = sin(t).
The three inputs all have amplitude 1 but different frequencies. We will
use the convolution formula

a sin(bt) − b sin(at)
if b 6= a


a2 − b 2

sin(at) ∗ sin(bt) =
 sin(at) − at cos(at)

if a = b

2a
from Table ??.
1. Consider the input function f (t) = sin(πt). Since sin(πt) = =eπit =
−=e−πit its characteristic value is ±πi. The system response is
π sin(t) − sin(πt)
y(t) = sin(πt) ∗ sin(t) = .
π2 − 1
The graph of the input function together with the response is given in
Figure 5.1. The graph of the input function is dashed and has amplitude
1 while the response function has an amplitude less than 1. No resonance is
occurring here and this is reflected in the fact that the inputs characteristic
frequency ±πi is far from the systems characteristic frequency ±i. We also
note that the response function is not periodic. This is reflected in the fact
that the quotient of the frequencies πi i = π is not rational. We will say
more about periodic functions in Chapter 8.
210 5 System Modeling and Applications

Fig. 5.1. Input and Response functions with dissimilar characteristic values

2. Next we take the input function to be f (t) = sin(1.25t). Its characteristic

frequency is ±1.25i. The system response is
1.25 sin(t) − sin(1.1t)
y(t) = sin(1.25t) ∗ sin(t) = .
.5625
The graph of the input function together with the response is given in
Figure 5.2. In this graph we needed to scale back significantly to see the

Fig. 5.2. Input and Response functions with similar yet unequal characteristic
values. Beats occur.

response function. Notice how the amplitude of the response is signifi-

cantly higher than that of the input. Also notice how the response comes
in pulses. This phenomenon is known as beats and is familiar to musi-
cians who try to tune an instrument. When the frequency of vibration of
the string is close but not exactly equal to that of the tuning fork, one
hears a pulsating beat. The instrument is out of tune.

3. We now consider the input function f (t) = sin t. Here the input frequency
matches exactly the system frequency. The system response is
sin(t) − t cos(t)
y(t) = sin(t) ∗ sin(t) = .
2
 5.1 System Modeling 211

The presence of t in t cos(t) implies that the response will oscillate without
bound as seen in Figure 5.3.

Fig. 5.3. Input and Response functions with equal characteristic values. Resonance
occurs.

Again in this graph we needed to scale back to see the enormity of the re-
sponse function. This is resonance in action. In physical systems resonance
can be so energetic that the system may fall apart. Because of the significant
damage that can occur systems designers must be well aware of the internal
characteristic values or frequencies of their system and the likely kinds of in-
puts it may need to deal with. J
As a final example we consider resonance in an asymptotically stable sys-
tem.
Example 12. A zero-state system Φ is modeled by the differential equation

((D + .1)2 + 1)y = f.

Determine the impulse response function h and the system response to the
input function f (t) = e−.1t cos(t). Discuss the resonance that occurs.
I Solution. The characteristic values or frequencies are −.1 ± i and lie in
the left-hand side of the complex plane. Thus the system is asymptotically
stable. The characteristic modes are e−.1t sin t, e−.1t cos t . A straightforward


calculation gives the unit-impulse response function

h(t)) = e−.1t sin(t).

The input function f (t) = e−.1t cos(t) is a characteristic mode and the re-
sponse function is
1
y(t) = h ∗ f (t) = te−.1t sin(t).
2
Figure 5.4 shows the graph. Notice the initial energetic response. This is a
manifestation of resonance even though the response dies out in time. If the
212 5 System Modeling and Applications

response passes a certain threshold the system may break. On the other hand,
resonance can be used in a positive way as in tuning a radio to a particular
frequency. Again, system designers must be well aware of the resonance effect.

Fig. 5.4. Asymptotically stable system with resonance.

Exercises
1-12 For each problem a system is mod- 12. q(D) = D 4 − 1, a = (0, 1, 0, −1)
eled by q(D) with initial state a. De-
termine the zero-input response. Deter- 13-18 For each problem a system is
mine whether the system is asymptoti- modeled by q(D). Determine the unit-
cally stable, marginally stable, or unsta- impulse response function.
ble.
1. q(D) = D + 5, a = 10 13. q(D) = D + 1

2. q(D) = D − 2, a = 2 14. q(D) = D 2 + 4

3. q(D) = D 2 − 4D + 3, a = (2, 4) 15.

2
4. q(D) = D + 5D + 4, a = (0, 3) 16. q(D) = D 2 − 4

5. q(D) = D 2 + 4D + 5, a = (0, 1) 17. q(D) = D 2 + 2D + 5

6. q(D) = D 2 + 9, a = (1, 1) 18. q(D) = D 3 + D

7. q(D) = D 2 + 6D + 9, a = (1, 1)
19-20 For each problem a system is
8. q(D) = D 2 + D − 2, a = (1, −2) modeled by the given differential equa-
2 tion. Determine those a for which the
9. q(D) = D − 2D + 2, a = (1, 2) system is asymptotically stable.
10. q(D) = D 3 + D 2 , a = (1, −1, 1) 19. q(D) = D 2 + aD + 6
2
11. q(D) = (D + 1)(D + 1),
a = (1, −1, 1) 20.
 5.2 Spring Systems 213

21-22 In this set of problems we estab- and suppose f is a bounded function.

lish that an asymptotically stable system Show that eλt ∗ f is a bounded func-
modeled by q(D), for some constant co- tion.
efficient differential operator is BIBO-
stable. 22. Suppose a system modeled by a con-
stant coefficient differential operator
21. Suppose λ is a complex number ly- is asymptotically stable. Show it is
ing to the left of the imaginary axis BIBO-stable.

5.2 Spring Systems

In this section we illustrate how a second order constant coefficient differential

equation arises from modeling a spring-body-dashpot system. This model may
arise in a simplified version of a suspension system on a vehicle or a washing
machine. Consider the three main objects in the diagram below: the spring,
the body, and the dashpot (shock absorber).

We assume the body only moves vertically without any twisting. Our goal
is to determine the motion of the body in such a system. Various forces come
into play. These include the force of gravity, the restoring force of the spring,
the damping force of the dashpot, and perhaps an external force. Let’s examine
each of these forces and how they contribute to the overall motion of the body.

force of gravity
First, assume that the body has mass m. The force of gravity, FG , acts on the
body by the familiar formula
214 5 System Modeling and Applications

FG = mg, (1)
where g is the acceleration due to gravity. Our measurements will be positive
in the downward direction so FG is positive.

restoring force
When a spring is suspended with no mass attached the end of the spring will
lie at a reference point (u = 0). When the spring is stretched or compressed
we will denote the displacement by u. The force exerted by the spring that
acts in the opposite direction to a force that stretches or compresses a spring
is called the restoring force. It depends on the displacement and is denoted
by FR (u). Hooke’s law says that the restoring force of many springs is pro-
portional to the displacement, as long as the displacement is not too large.
We will assume this. Thus, if u is the displacement we have
FR (u) = −ku, (2)
where k is a positive constant, called the spring constant. When the dis-
placement is positive (downward) the restoring force pulls the body upward,
hence the negative sign.
To determine the spring constant k consider the effect of a body of mass m
attached to the spring and allowed to come to equilibrium (i.e., no movement).
It will stretch the spring a certain distance, u0 , as illustrated below:

u=0 
u0
u = u0
mass m
(at equilibrium)

At equilibrium the restoring force of the spring will cancel the gravitational
force on the mass m. Thus we get
FR (u0 ) + FG = 0. (3)
Combining Equations (1), (2), and (3) gives us mg − ku0 = 0 and hence
mg
k= . (4)
u0
 5.2 Spring Systems 215

damping force
In any practical situation there will be some kind of resistance to the motion
of the body. If our spring-body-dashpot system were under water the viscosity
of the water would dampen the motion (no pun intended) to a much greater
extent than in air. In our system this resistance is represented by a dashpot,
which in many situations is a shock absorber. The force exerted by the dashpot
is called the damping force, FD . It depends on a lot of factors but an
important factor is the velocity of the body. To see that this is reasonable
imagine the difference in the forces against your body when you dive into
a swimming pool off a 3 meter board and when you dive from the side of
the pool. The greater the velocity when you enter the pool the greater the
force that decelerates your body. We will assume that the damping force is
proportional to the velocity. We thus have

FD = −µv = −µu0 ,

where v = u0 is velocity and µ is a positive constant known as the damping

constant. The damping force acts in a direction opposite the velocity, hence
the negative sign.

external forces and newton’s law of motion

We will let f (t) denote an external force acting on the body. For example, this
could be the varying forces acting on a suspension system due to driving over
a bumpy road. If a = u00 is acceleration then Newton’s second law of motion
says that the total force of a body, given by mass times acceleration, is the
sum of the forces acting on that body. We thus have

Total Force = FG + FR + FD + External Force,

which implies the equation

mu00 = mg − ku − µu0 + f (t).

Equation 4 implies mg = −ku0 . Substituting and combining terms gives

mu00 + µu0 + k(u − u0 ) = f (t).

If y = u − u0 then y measures the displacement of the body from the equilib-

rium point, u0 . In this new variable we obtain

my 00 + µy 0 + ky = f (t). (5)

This second order constant coefficient differential equation is a mathemat-

ical model for the spring-body-dashpot system. The solutions that can be
216 5 System Modeling and Applications

obtained vary dramatically depending on the constants m, µ, and k, and, of

course, the external force, f (t). The initial conditions,

y(0) = y0 and y 0 (0) = v0

represent the initial position, y(0) = y0 , and the initial velocity, y 0 (0) = v0 of
the given body. Once the constants, external force, and initial conditions are
determined the Uniqueness and Existence Theorem, Theorem 1, guarantees
a unique solution. Of course, we should always keep in mind that Equation 5
is a mathematical model of a real phenomenon and its solution is an approx-
imation to what really happens. However, as long as our assumptions about
the spring and damping constants are in effect, which usually require that
y(t) and y 0 (t) be relatively small in magnitude, and the mass of the spring is
negligible compared to the mass of the body, the solution will be a reasonably
good approximation.

units of measurement
Before we consider specific examples we summarize the two commonly used
units of measurement: The English and Metric systems. The following table
summarizes the units.
System Time Distance Mass Force

Metric seconds (s) meters (m) kilograms (kg) Newtons (N)

English seconds (s) feet (ft) slugs (sl) pounds (lbs)

The main units of the English system are kilograms and meters while in
the English system they are pounds and feet. The time unit is common to
both. The next table summarizes quantities derived from these units.
Quantity Formula

velocity (v) distance / time

acceleration (a) velocity /time
force (F) mass · acceleration
spring constant (k) force / distance
damping constant (µ) force /velocity

In the metric system one Newton of force (N) will accelerate a one kilo-
gram mass (kg) one m/s2 . In the English system a one pound force (lb) will
accelerate a one slug mass (sl) one ft/s2 . To compute the mass of a body in the
English system one must divide the weight by the acceleration due to gravity,
which is g = 32 ft/sec2 near the surface of the earth. Thus a body weighing 64
lbs has a mass of 2 slugs. To compute the force a body exerts due to gravity
 5.2 Spring Systems 217

in the metric system one must multiply the mass by the acceleration due to
gravity, which is g = 9.8 m/sec2 . Thus a 5 kg mass exerts a gravitational force
of 49 N.

Example 1.
1. A body exerts a damping force of 10 pounds when the velocity of the mass
is 2 feet per second. Find the damping constant.
2. A body exerts a damping force of 6 Newtons when the velocity is 40
centimeters per second. Find the damping constant
3. A body weighing 4 pounds stretches a spring 2 inches. Find the spring
constant
4. A mass of 8 kilograms stretches a spring 20 centimeters. Find the spring
constant.

I Solution.
1. The force is 10 pounds and the velocity is 2 feet per second. The damping
constant is given by µ = Force/Velocity = 10/2 = 5.
2. The force is 6 Newtons and the velocity is .4 meters per second The
damping constant is given by µ = Force/Velocity = 6/.4 = 15.
3. The force is 4 pounds. A length of 2 inches is 1/6 foot. The spring constant
is k = Force/Distance = 4/(1/6) = 24.
4. The force exerted by a mass of 8 kilograms is 8 · 9.8 = 78.4 Newtons.
A length of 20 centimeters is .2 meters. The spring constant is given by
k = Force/Distance = 78.4/.2 = 392.

Example 2. A spring is stretched 20 centimeters by a force of 5 Newtons. A

body of mass 4 kilogram is attached to such a spring with an accompanying
dashpot. At t = 0 the mass is pulled down from its equilibrium position a
distance of 50 centimeters and released with a downward velocity of 1 meter
per second. Suppose the damping force is 5 Newtons when the velocity of the
body is .5 meter per second. Find a mathematical model that represents the
motion of the body.

I Solution. We will model the motion by Equation 5. Units are converted

to the kilogram-meter units of the metric system. The mass is m = 4. The
spring constant k is given by k = 5/(.2) = 25. The damping constant is given
by µ = 5/.5 = 10. Since no external force is mentioned we may assume it
is zero. The initial conditions are y(0) = .5 and y 0 (0) = 1. The following
equation

4y 00 (t) + 10y 0 (t) + 25y = 0, y(0) = .5, y 0 (0) = 1

represents the model for the motion of the body. J

Example 3. A body weighing 4 pounds will stretch a spring 3 inches. This

same body is attached to such a spring with an accompanying dashpot. At
t = 0 the mass is pulled down from its equilibrium position a distance of 1
218 5 System Modeling and Applications

foot and released. Suppose the damping force is 8 pounds when the velocity
of the body is 2 feet per second. Find a mathematical model that represents
the motion of the body.

I Solution. Units are converted to the pound-foot units in the English

system. The mass is m = 4/32 = 1/8 slugs. The spring constant k is given by
k = 4/(3/12) = 16. The damping constant is given by µ = 8/2 = 4. Since no
external force is mentioned we may assume it is zero. The initial conditions
are y(0) = 1 and y 0 (0) = 0. The following equation
1 00
y (t) + 4y 0 (t) + 16y = 0, y(0) = 1, y 0 (0) = 0
8
models the motion of the body. J

Let’s now turn our attention to an analysis of Equation (5). The zero-input
response models the motion of the body with no external forces. We refer to
this motion as free motion (f (t) ≡ 0). Otherwise we refer to the motion as
forced motion (f (t) 6= 0). In turn each of these are divided into undamped
(µ = 0) and damped (µ 6= 0).

Undamped Free Motion

When the damping constant is zero and there is no externally applied force
the resulting motion of the object is called undamped free motion. This is
an idealized situation, for seldom if ever will a system be free of any damping
effects. Nevertheless, Equation (5) becomes

my 00 + ky = 0, (6)

with m > 0 and k > 0. The characteristic polynomial of this p equation is

q(s) = ms2 + k. Its characteristic values are ±iβ where β = k/m, and its
characteristic modes are {sin βt, cos βt}. Hence Equation (14) has the general
solution
y = c1 cos βt + c2 sin βt. (7)
Such is system is marginally stable as the characteristic values are purely com-
plex. Using the trigonometric identity cos(βt−δ) = cos(βt) cos δ+sin(βt) sin δ,
Equation (7) can be rewritten as

y = A cos(βt − δ) (8)

where c1 =p A cos δ and c2 = A sin δ. Solving these equations for A and δ

gives A = c21 + c22 and tan δ = c2 /c1 . Therefore, the graph of y(t) satisfying
Equation (14) is a pure cosine function with frequency β and with period
 5.2 Spring Systems 219
r
2π m
T = = 2π .
β k
The numbers A and δ are commonly referred to as the amplitude and phase
angle of the system. From Equation (8) we see that A is the maximum pos-
sible value of the function y(t), and hence the maximum displacement from
equilibrium, and that |y(t)| = A precisely when t = (δ + nπ)/β, where n ∈ Z.
The graph of Equation (8) is given below and well represents the oscillating
motion of the body. This kind of motion occurs ubiquitously in the sciences
and is usually referred to as simple harmonic motion.
y(t)

T
A

δ
β

Example 4. Find the amplitude, phase angle, and frequency of the damped
free motion of a spring-body-dashpot system with unit mass and spring con-
stant 3. Assume the initial conditions are y(0) = −3 and y 0 (0) = 3.

I Solution. The initial value problem that models such a system is

y 00 + 3y = 0, y(0) = −3, y 0 (0) = 3.

√ √ √
An easy calculation gives y = −3 cos 3t+ 3 sin 3t. The amplitude is given
√
1 √
by A = (−3)2 + ( 3)2 2 = 2 3 and the phase angle is given implicitly by
√ √
tan δ = 3/ − 3 = −1/ 3, hence δ = −π/6. Thus
√ √ π
y = 2 3 cos 3t + .
6
J

Damped Free Motion

In this case we include the damping term µ y 0 with µ > 0. In applications the
coefficient µ represents the presence of friction or resistance, which can never
be completely eliminated. Thus we want solutions to the differential equation

my 00 + µy 0 + ky = 0. (9)

The characteristic polynomial q(s) = ms2 + µs + k has roots r1 and r2 given

by the quadratic formula
220 5 System Modeling and Applications
p
−µ ± µ2 − 4mk
r1 , r2 = . (10)
2m
The nature of the solutions of Equation (9) are determined by whether the
discriminant D = µ2 − 4mk is negative (complex roots), zero (double root),
or positive (distinct roots). We say that the system is
• Underdamped if D < 0
• Critically damped if D = 0
• Overdamped if D > 0.
Let’s consider each of these cases separately.
√
Underdamped Systems: When µ is between 0 and 4mk then D < 0.
The damping is not sufficient to overcome the oscillatory behavior that we
saw in the undamped case, µ = 0. Indeed, observe that in this case the roots
of the characteristic polynomial q(s) are a pair of conjugate complex numbers
α ± iβ where
√ p
µ −D 4mk − µ2
α=− < 0 and β = = .
2m 2m 2m
The solution to Equation (9) is

y(t) = eαt (c1 cos βt + c2 sin βt) ,

which can be rewritten as

y(t) = Aeαt cos(βt − δ) (11)

p
where A = c21 + c22 and tan δ = c2 /c1 , as we did earlier for the undamped
case. A typical graph of Equation (11) is given below.

Aeαt

y(t)

−Aeαt

Notice that y appears to be a cosine curve in which the amplitude oscillates

between Aeαt and −Aeαt . The motion of the body passes through equilibrium
at regular intervals (see Problem ??). Since α < 0 the amplitude decreases
with time. One may imagine the suspension on an automobile with rather
weak shock absorbers. A push on the fender will send the vehicle oscillating
as in the graph above.
 5.2 Spring Systems 221
√
Critically damped Systems: If µ = 4mk then the discriminant D is
zero. At this critical point the damping is just large enough to overcome oscil-
latory behavior. Observe that the characteristic polynomial q(s) has a single
µ
root, namely r = − , with multiplicity two. Furthermore, r is negative
2m
since m and µ are positive. The general solution of Equation (9) is

y(t) = c1 ert + c2 tert = (c1 + c2 t)ert . (12)

In this case there is no oscillatory behavior. In fact, the system will pass
through equilibrium only if t = −c1 /c2 and since t > 0 this only occurs if c1
and c2 have opposite signs. (See problem ??) The following graph represents
the two possibilities.

system does not pass

through equililbrium

system passes through

equililbrium

√
Overdamped Systems: When µ > 4mk then the discriminant D is
positive. The characteristic polynomial q(s) has two distinct roots:
p p
−µ + µ2 − 4mk −µ − µ2 − 4mk
r1 = and r2 = .
2m 2m
Both roots are negative. The general solution of Equation (9) is

y(t) = c1 er1 t + c2 er2 t (13)

The graphs shown for the critically damped case are representative of the
possible graphs for the present case as well.

Notice that in all three cases the real parts of the characteristic values are
negative and hence the spring-body-dashpot system is asymptotically stable.
It follows that
lim y(t) = 0.
t→∞

and thus the motion y(t) dies out as t increases.

Undamped Forced Motion

Undamped forced motion refers to a system governed by a differential equation
222 5 System Modeling and Applications

my 00 + ky = f (t),

where f (t) is a nonzero forcing function. We will only consider the special case
where the forcing function is given by f (t) = F0 cos ωt where F0 is a nonzero
constant. Thus we are interested in describing the solutions of the differential
equation
my 00 + ky = F0 cos ωt (14)
where, as usual, m > 0 and k > 0. Imagine an engine embedded within a
spring-body-dashpot system. The spring system has a characteristic frequency
β while the engine exerts a cyclic force with frequency ω.
00
From Equation (7) we know that p a general solution to my + ky = 0 is
yh = c1 cos βt + c2 sin βt where β = k/m, so if we can find a single solution
yp (t) to Equation (14), then Theorem 4.1.8 shows that the entire solution set
is given by

S = {yp (t) + c1 cos βt + c2 sin βt : c1 , c2 ∈ R} .

To find yp (t) we shall solve Equation (14) subject to the initial conditions
y(0) = 0, y 0 (0) = 0. As usual, if Y = L(y)(s), then we apply the Laplace
transform to Equation (14) and solve for Y (s) to get

1 F0 s F0 β s
Y (s) = = . (15)
ms2 2
+ks +ω 2 mβ s + β s + ω 2
2 2 2

Then the convolution theorem, Theorem 3.8.3, shows that

F0
y(t) = L−1 (Y (s)) = sin βt ∗ cos ωt. (16)
mβ
The following convolution formula comes from Table C.3:
β


 2 (cos ωt − cos βt) if β 6= ω
 β − ω2

sin βt ∗ cos ωt = (17)
 1 t sin ωt


if β = ω.

2
Combining Equations (16) and (17) gives
 F0
 (cos ωt − cos βt) if β 6= ω
 m(β 2 − ω 2 )


y(t) = (18)
 F0 t sin ωt


if β = ω.

2mω
We will first consider the case β 6= ω in Equation (18). Notice that, in this
case, the solution y(t) is the sum of two cosine functions with equal amplitude
 5.2 Spring Systems 223

= F0 /m(β 2 − ω 2 ) , but different frequencies β and ω. Recall the trigonomet-

ric identity
cos(θ − ϕ) − cos(θ + ϕ) = 2 sin θ sin ϕ.
If we set θ − ϕ = ωt and θ + ϕ = βt and solve for θ = (β + ω)t/2 and
ϕ = (β − ω)t/2, we see that we can rewrite the first part of Equation (18) in
the form
2F0 (β − ω)t (β + ω)t
y(t) = sin sin . (19)
a(β 2 − ω 2 ) 2 2
One may think of the function y(t) as a sine function, namely sin(β + ω)t/2
(with frequency (β + ω)/2) which is multiplied by another function, namely

2F0 (β − ω)t
sin ,
a(β 2 − ω 2 ) 2

which functions as a time varying amplitude function.

An interesting case is when β is close to ω so that β + ω is close to 2ω and
β − ω is close to 0. In this situation, one sine function changes very rapidly,
while the other, which represents the change in amplitude, changes very slowly
as is illustrated below.

One might have observed this type of motion in an unbalanced washing ma-
chine. The spinning action exerts a cyclic force, F0 cos ωt, on the spring sys-
tem with characteristic frequency close to ω. The chaotic motion that results
settles down momentarily only to repeat itself again. In music, this type of
phenomenon, known as beats, can be heard when one tries to tune a piano.
When the frequency of vibration of the string is close to that of the tuning
fork, one hears a pulsating beat which disappears when the two frequencies
coincide. The piano is slightly out of tune.
In the case where the input frequency and characteristic frequency are
equal, β = ω, in Equation (18), the solution

F0
y(t) = t sin ωt
2aω
is unbounded as t → ∞ as illustrated below.
224 5 System Modeling and Applications

The resulting amplification of vibration eventually becomes large enough to

destroy the mechanical system. This is a manifestation of resonance as dis-
cussed in Section 5.1.

Damped Forced Motion

We now assume the damping constant µ is positive and consider forcing func-
tions of the form f (t) = F0 cos ωt where F0 is a constant, as before. Thus we
are interested in analyzing the solutions of the equation

my 00 + µy 0 + ky = F0 cos ωt (20)

where m, µ, k and F0 are positive constants. It is a straightforward (albeit

tedious) calculation to check that the function
F0
(k − ω 2 m) cos ωt + µω sin ωt


yp (t) =
(k − ω m)2
2 + µ2 ω 2
is a solution of Equation (20). Using Equation (17), this can be rewritten as
F0
yp (t) = p cos(ωt − δ)
(k − ω m)2 + µ2 ω 2
2

µω
where tan δ = . Combining this with Equation (??), the general
k − ω2m
solution to Equation (??) is
F0
y(t) = eαt (c1 cos βt + c2 sin βt) + p cos(ωt − δ) (21)
(c − ω 2 a)2 + b2 ω 2
b
where α = − < 0. Notice that this implies that limt→∞ (y(t) − ϕp (t)) = 0,
2a
which says that every general solution of Equation (??) converges asymptot-
ically to the particular solution ϕp (t). For this reason, the solution ϕp (t) is
usually referred to as the steady state solution to the equation, while the so-
lution y(t) = eαt (c1 cos βt + c2 sin βt) of the associated homogeneous equation
is referred to as a transient solution.
 5.3 Electrical Circuit Systems 225

5.3 Electrical Circuit Systems

6
Second Order Linear Differential Equations

In this chapter we consider the class of second order linear differential equa-
tions. In particular, we will consider differential equations of the following
form:
a2 (t)y 00 + a1 (t)y 0 + a0 (t)y = f (t). (1)
Notice that the coefficients a0 (t), a1 (t), and a2 (t) are functions of the inde-
pendent variable t and not necessarily constants. This difference has many
important consequences, the main one being that there is no general solution
method as in the constant coefficient case. Nevertheless, it is still linear and,
as we shall see, this implies that the solution set has a structure similar to the
constant coefficient case.
In order to find solution methods one must put some rather strong restric-
tions on the coefficient functions a0 (t), a1 (t) and a2 (t). For example, in the
following list the coefficient functions are polynomial of a specific form. The
equations in this list are classical and have important uses in the physical and
engineering sciences.

t2 y 00 + ty 0 + (t2 − ν 2 )y = 0 Bessel’s Equation of index ν

ty 00 + (1 − t)y 0 + λy = 0 Laguerre’s Equation of index λ
(1 − t2 )y 00 − 2ty 0 + α(α + 1)y = 0 Legendre’s Equation of index α
(1 − t2 )y 00 − ty 0 + α2 y = 0 Chebyshev’s Equation of index α
y 00 − 2ty 0 + 2λy = 0 Hermite’s Equation of index λ

Except for a few simple cases the solutions to these equations are not
exponential polynomials nor are they expressible in terms of algebraic com-
binations of polynomials, trigonometric or exponential functions, nor their
inverses. Nevertheless, the general theory implies that solutions exist and tra-
ditionally have been loosely categorized as special functions. In addition, to
satisfying the differential equation for the given index, there are other inter-
228 6 Second Order Linear Differential Equations

esting and important functional relations as the index varies. We will explore
some of these relations.

6.1 The Existence and Uniqueness Theorem

In this chapter we will assume that the coefficient functions a0 (t), a1 (t), and
a2 (t) and the forcing function f (t) are continuous functions on some com-
mon interval I. We also assume that a2 (t) 6= 0 for all t ∈ I. By dividing by
a2 (t), when convenient, we may assume that the leading coefficient function
is 1. In this case we say that the differential equation is in standard form.
We will adopt much of the notation that we used in Section 4.1. In particular,
let D denote the derivative operator and let

L = a2 (t)D2 + a1 (t)D + a0 (t). (1)

Then Equation 1 in the introductory paragraph can be written as Ly = f . The

equation Ly = 0 is called homogeneous. Otherwise it is non homogeneous.
We can think of L as an operation on functions. If y ∈ C 2 (I), in other words,
if y is a function on an interval I having a second order continuous derivative,
then Ly produces a continuous function.

Example 1. If L = D2 + 4tD + 1 then

• L(et ) = (et )00 + 4t(et )0 + 1(et ) = (2 + 4t)et
• L(sin t) = − sin t + 4t cos t + sin t = 4t cos t
• L(t2 ) = 2 + 4t(2t) + (t2 ) = 9t2 + 2
• L(t + 2) = 0 + 4t(1) + (t + 2) = 5t + 2

The most important property that we can say about L, generally, is that
it is linear.

Proposition 2. The operator

L = a2 (t)D2 + a1 (t)D + a0 (t)

given by Equation 1 is linear. Specifically,

1. If y1 and y2 have second order continuous derivatives then

L(y1 + y2 ) = L(y1 ) + L(y2 ).

2. If y has a second order continuous derivative and c is a scalar then

L(cy) = cL(y).
 6.1 The Existence and Uniqueness Theorem 229

Proof. The proof of this proposition is essentially the same as the proof of
Proposition 4 in Section 4.1. We only need to remark that multiplication by
a function ak (t) preserves addition and scalar multiplication in the same way
as multiplication by a constant ak .

We call L a second order linear differential operator. Proposition

4.1.6 and Theorem 4.1.8 are two important consequences of linearity for the
constant coefficient case. The statement and proof are essentially the same for
the second order linear differential operators considered here in this chapter.
We consolidate these results and the algorithm that followed into the following
theorem:

Theorem 3. Suppose L is a second order linear differential operator and y1

and y2 are solutions to Ly = 0. Then c1 y1 +c2 y2 is a solution to Ly = 0, for all
scalars c1 and c2 . Suppose f is a continuous function. If yp is a fixed particular
solution to Ly = f and yh is any solution to the associated homogeneous
differential equation Ly = 0 then

yp + yh

is a solution to Ly = f . Furthermore, any solution to Ly = f have this same

form. Thus to solve Ly = f we proceed as follows:
1. Find all the solutions to the associated homogeneous differential equation
Ly = 0,
2. Find one particular solution yp ,
3. Add the particular solution to the homogeneous solutions.

As an application of Theorem 3 consider the following example.

Example 4. Let L = tD2 +2D+t. Show that sint t and cost t are homogeneous
solutions. Show that t + 1 is a solution to Ly = t2 + t + 2. Use this information
and linearity to write the most general solution.

I Solution. It is a straightforward calculation to verify that sint t and cost t

are homogeneous solutions. Theorem 3 now implies that c1 sint t + c2 cost t is a
homogeneous solution for all scalars c1 and c2 . Again it is straightforward to
verify that L(t + 1) = t2 + t + 2. Theorem 3 implies

sin t cos t
y = t + 1 + c1 + c2
t t
is a solution and we will soon see it is the general solution. J

Again, suppose L is a second order linear differential operator and f is a

function defined on an interval I. Let t0 ∈ I. To the equation

Ly = f
230 6 Second Order Linear Differential Equations

we can associate initial conditions of the form

y(t0 ) = y0 , and y 0 (t0 ) = y1 .

We refer to the initial conditions and the differential equation Ly = f as an

initial value problem.

Example 5. Let L = tD2 + 2D + t. Solve the initial value problem

Ly = t2 + t + 2, y(π) = 1, y 0 (π) = 1.

I Solution. By Example 4 the general solution is of the form y = t + 1 +

c1 sint t + c2 cost t . The initial conditions lead to the equations
c2
π+1− =1
π
c1 1
1− + = 1,
π π
which imply c1 = π and c2 = π 2 . The solution to the initial value problem is
sin t cos t
y =t+1+π + π2 .
t t
J

In the case where the coefficient functions of L are constant we proved the
existence and uniqueness theorem by an inductive method in Section 4.2. The
inductive method relied on factoring L, which is not always possible in the
present, more general, context. Nevertheless, we still have the existence and
uniqueness theorem. Its proof is beyond the scope of this book. We refer to
Coddington [?] for a proof and more detailed information.

Theorem 6. Suppose a0 (t), a1 (t), a2 (t) and f are continuous functions on

an open interval I and an (t) 6= 0 for all t ∈ I. Suppose t0 ∈ I. Let L =
a2 (t)D2 + a1 (t)D + a0 (t). Then there is one and only one solution to the
initial value problem

Ly = f, y(t0 ) = y0 , y 0 (t0 ) = y1 .

Theorem 6 does not tell us how to find any solution. We must develop
procedures for this. Let’s explain in more detail what this theorem does say.
Under the conditions stated the Uniqueness and Existence theorem says that
there always is a solution to the given initial value problem. The solution is at
least twice differentiable on I and there is no other solution. In Example 5 we
found y = t + 1 + π sint t + π 2 cost t is a solution to ty 00 + 2y 0 + ty = t2 + t + 2. with
initial conditions y(π) = 2 and y 0 (π) = 1. Notice, in this case, that y is, in fact,
infinitely differentiable on any interval not containing 0. This is precisely where
a2 (t) = t is zero. The uniqueness part of Theorem 6 implies that there are no
 6.1 The Existence and Uniqueness Theorem 231

other solutions. In other words, there are no potentially hidden solutions, so

that if we can find enough solutions to take care of all possible initial values,
then Theorem 6 provides the theoretical underpinnings to know that we have
found all possible solutions and need look no further. Compare this theorem
with the discussion in Section 2.3 where we saw examples (in the nonlinear
case) of initial value problems which had infinitely many distinct solutions.

Exercises
1–12. For each of the following dif- 15. L(y) = 2y 00 + y 0 − 3y
ferential equations, determine if it is
linear (yes/no). For each of those 16. L = D 2 + 6D + 5
which is linear, further determine if
17. L = D 2 − 4
the equation is homogeneous (homo-
geneous/nonhomogeneous) and constant 18. L = t2 D 2 + tD − 1
coefficient (yes/no). Do not solve the
equations. 19. If L = aD 2 + bD + c where a, b,
00 0
1. y + y y = 0 c are real numbers, then show that
L(ert ) = (ar 2 + br + c)ert . That
2. y 00 + y 0 + y = 0 is, the effect of applying the oper-
ator L to the exponential function
3. y 00 + y 0 + y = t2 ert is to multiply ert by the number
4. y 00 + ty 0 + (1 + t2 )y 2 = 0 ar 2 + br + c.

5. 3y 00 + 2y 0 + y = e2 20. The differential equation t2 y 00 +ty 0 −

1
y = t 2 , t > 0 has a solution of the
6. 3y 00 + 2y 0 + y = et 1
form ϕp (t) = Ct 2 . Find C.
√
7. y 00 + y 0 + y = t 21. The differential equation y 00 + 3y 0 +
√ 2y = t has a solution of the form
8. y 00 + y 0 + y = t
ϕp (t) = C1 + C2 t. Find C1 and C2 .
9. y 00 − 2y = ty
22. Does the differential equation y 00 +
00
10. y + 2y + t sin y = 0 3y 0 + 2y = e−t have a solution of the
form ϕp (t) = Ce−t ? If so find C.
11. y 00 + 2y 0 + (sin t)y = 0
23. Does the differential equation y 00 +
12. t2 y 00 + ty 0 + (t2 − 5)y = 0 3y 0 + 2y = e−t have a solution of the
13–18. For each of the following linear form ϕp (t) = Cte−t ? If so find C.
differential operators L compute L(1), Let L(y) = y 00 + y.
L(t), L(e−t ), and L(cos 2t). That is,
evaluate L(y) for each of the given input 24–26.
functions. 24. Let L(y) = y 00 + y
13. L(y) = y + y00 1. Check that y(t) = t2 − 2 is one
solution to the differential equa-
14. L(y) = ty 00 + y tion L(y) = t2 .
232 6 Second Order Linear Differential Equations

2. Check that y1 (t) = cos t and 3. Using the results of Parts (a)
y2 (t) = sin t are two solutions to and (b), find a solution to each
the differential equation L(y) = of the following initial value
0. problems.
3. Using the results of Parts (a) a) t2 y 00 − 4ty 0 + 6y = t5 ,
and (b), find a solution to each y(1) = 1, y 0 (1) = 0.
of the following initial value b) t2 y 00 − 4ty 0 + 6y = t5 ,
problems. y(1) = 0, y 0 (1) = 1.
a) y 00 + y = t2 , y(0) = 1, c) t2 y 00 − 4ty 0 + 6y = t5 ,
y 0 (0) = 0. y(1) = −1, y 0 (1) = 3.
b) y 00 + y = t2 , y(0) = 0, d) t2 y 00 − 4ty 0 + 6y = t5 ,
y 0 (0) = 1. y(1) = a, y 0 (1) = b, where
c) y 00 + y = t2 , y(0) = −1, a, b ∈ R.
y 0 (0) = 3. 27–32. For each of the following differ-
d) y 00 + y = t2 , y(0) = a, ential equations, find the largest inter-
y 0 (0) = b, where a, b ∈ R. val on which a unique solution of the
initial value problem a0 (t)y 00 + a1 (t)y 0 +
25. Let L(y) = y 00 − 5y 0 + 6y. a3 (t)y = f (t), y(t0 ) = y1 , y 0 (t0 ) = y1
1. Check that y(t) = 12 et is one is guaranteed by Theorem 6. Note that
solution to the differential equa- your interval may depend on the choice
tion L(y) = et . of t0 .
2. Check that y1 (t) = e2t and 27. t2 y 00 + 3ty 0 − y = t4
y2 (t) = e3t are two solutions to
the differential equation L(y) = 1 + t2
28. y 00 − 2y 0 − 2y =
0. 1 − t2
3. Using the results of Parts (1)
29. (sin t)y 00 + y = cos t
and (2), find a solution to each
of the following initial value 30. (1 + t2 )y 00 − ty 0 + t2 y = cos t
problems. √ √
a) y 00 − 5y 0 + 6y = et , y(0) = 31. y 00 + ty 0 − t − 3y = 0
1, y 0 (0) = 0. 32. t(t2 − 4)y 00 + y = et
t
b) y − 5y + 6y = e ,
00 0
y(0) = 33. The functions y1 (t) = t2 and y2 (t) =
0, y 0 (0) = 1. t3 are two distinct solutions of the
c) y 00 − 5y 0 + 6y = et , y(0) = initial value problem
−1, y 0 (0) = 3. t2 y 00 −4ty 0 +6y = 0, y(0) = 0, y 0 (0) = 0.
d) y 00 − 5y 0 + 6y = et , y(0) = Why doesn’t this violate the unique-
a, y 0 (0) = b, where a, b ∈ ness part of Theorem 6?
R.
34. Let y(t) be a solution of the differ-
26. Let L(y) = t2 y 00 − 4ty 0 + 6y. ential equation
1. Check that y(t) = 16 t5 is one
y 00 + a1 (t)y 0 + a0 (t)y = 0.
solution to the differential equa-
tion L(y) = t5 . We assume that a1 (t) and a0 (t) are
2. Check that y1 (t) = t2 and continuous functions on an interval
y2 (t) = t3 are two solutions to I, so that Theorem 6 implies that
the differential equation L(y) = y is defined on I. Show that if the
0. graph of y(t) is tangent to the t-axis
 6.2 The Homogeneous Case 233

at some point t0 of I, then y(t) = 0 where, as usual we assume that

for all t ∈ I. Hint: If the graph of a1 (t), a0 (t), and f (t) are continuous
y(t) is tangent to the t-axis at (t0 , 0), functions on an interval I, so that
what does this say about y(t0 ) and Theorem 6 implies that y1 and y2
y 0 (t0 )? are defined on I. Show that if the
graphs of y1 (t) and y2 (t) are tan-
35. More generally, let y1 (t) and y2 (t) gent at some point t0 of I, then
be two solutions of the differential y1 (t) = y2 (t) for all t ∈ I.
equation
y 00 + a1 (t)y 0 + a0 (t)y = f (t),

6.2 The Homogeneous Case

In this section we are mainly concerned with the homogeneous case:

L(y) = a2 (t)y 00 + a1 (t)y 0 + a0 (t)y = 0 (1)

The main result, Theorem 4 given below, shows that we will in principle be
able to find two functions y1 and y2 such that all solutions to Equation (1)
are of the form c1 y1 + c2 y2 , for some constants c1 and c2 . This is just like
the second order constant coefficient case. In fact, if q(s) is a characteristic
polynomial of degree 2 then Bq = {y1 , y2 } is a set of two linearly indepen-
dent functions that span the solution set of the corresponding homogeneous
constant coefficient differential equation. In section 3.7 we introduced the con-
cept of linear independence for a set of n functions. Let’s recall this important
concept in the case n = 2.

Linear Independence
Two functions y1 and y2 defined on some interval I are said to be linearly
independent if the equation

c1 y 1 + c2 y 2 = 0 (2)

implies that c1 and c2 are both 0. Otherwise, we call y1 and y2 linearly

dependent.
One must be careful about the meaning of this definition. We do not try
solve Equation (2) for the variable t. Rather, we are given that this equation
is valid for all t ∈ I. With this information the focus is on what this says
about the constants c1 and c2 : are they necessarily both zero or not.
Let’s consider two examples.
234 6 Second Order Linear Differential Equations

Example 1. Let y1 (t) = t and y2 (t) = t2 be defined on I = R. If the

equation
c1 t + c2 t2 = 0,
is valid for all t ∈ R, then this implies, in particular, that

c1 + c2 = 0 (let t = 1)
−c1 + c2 = 0 (let t = −1)

Now this system of linear equations is easy to solve. We obtain c1 = 0 and

c2 = 0. Thus t and t2 are linearly independent.
Example 2. In this second example let y1 (t) = t and y2 (t) = −2t defined
on I = R. Then there are many sets of constants c1 and c2 such that c1 t +
c2 (−2t) = 0. For example, we could choose c1 = 2 and c2 = 1. So the equation
c1 t + c2 (−2t) = 0 does not necessarily mean that c1 and c2 are zero. Hence t
and −2t are not independent. They are linearly dependent.
Remark 3. Notice that y1 and y2 are linearly dependent precisely when one
function is a scalar multiple of the other, i.e., y1 = αy2 or y2 = βy1 for
α ∈ R or β ∈ R. In Example 1, y2 6= cy1 while in Example 2, y2 = −2y1 .
Furthermore, given two linearly independent functions neither of them can be
zero.

The main theorem for the homogeneous case

Theorem 4. Let L = a2 (t)D 2 + a1 (t)D + a0 (t), where a0 (t), a1 (t), and a2 (t)
are continuous functions on an interval I. Assume a2 (t) 6= 0 for all t ∈ I.
1. There are two linearly independent solutions to Ly = 0.
2. If y1 and y2 are independent solution Ly = 0 then any homogeneous so-
lution y can be written y = c1 y1 + c2 y2 , for some c1 , c2 ∈ R.
Proof. Let t0 ∈ I. By Theorem 6, there are functions, ψ1 and ψ2 , that are
solutions to the initial value problems L(y) = 0, with initial conditions y(t0 ) =
1, y 0 (t0 ) = 0 and y(t0 ) = 0, y 0 (t0 ) = 1, respectively. Suppose c1 ψ1 + c2 ψ2 = 0.
Then
c1 ψ1 (t0 ) + c2 ψ2 (t0 ) = 0.
Since ψ1 (t0 ) = 1 and ψ2 (t0 ) = 0 it follows that c1 = 0. Similarly we have,

c1 ψ10 (t0 ) + c2 ψ20 (t0 ) = 0.

Since ψ10 (t0 ) = 0 and ψ20 (t0 ) = 1 it follows that c2 = 0. Therefore ψ1 and ψ2
are linearly independent. This proves (1).
Suppose y is a homogeneous solution. Let r = y(t0 ) and s = y 0 (t0 ). By
Theorem 3 the function rψ1 + sψ2 is a solution to Ly = 0. Furthermore,
 6.2 The Homogeneous Case 235

rψ1 (t0 ) + sψ2 (t0 ) = r

and rψ10 (t0 ) + sψ20 (t0 ) = s.

This means the rψ1 + sψ2 and y satisfy the same initial conditions. By the
uniqueness part of Theorem 6 they are equal. Thus y = rψ1 + sψ2 , i.e., every
homogeneous solution is a linear combination of ψ1 and ψ2 .
Now suppose y1 and y2 are any two linearly independent homogeneous
solutions and suppose y is any other solution. From the argument above we
can write

y1 = aψ1 + bψ2
y2 = cψ1 + dψ2 ,

which in matrix form can be written

    
y1 a b ψ1
= .
y2 c d ψ2

d −b
We multiply both sides of this matrix equation by the adjoint to
−c a
obtain
       
d −b y1 ad − bc 0 ψ1 ψ
= = (ad − bc) 1 .
−c a y2 0 ad − bc ψ2 ψ2

Suppose ad − bc = 0. Then

dy1 − by2 = 0
and − cy1 + ay2 = 0.

But since y1 and y2 are independent this implies that a, b, c, and d are zero
which in turn implies that y1 and y2 are both zero. But this cannot be. We
conclude that ad − bc 6= 0. We can now write ψ1 and ψ2 each as a linear
combination of y1 and y2 . Specifically,
    
ψ1 1 d −b y1
= .
ψ2 ad − bc −c a y2

Since y is a linear combination of ψ1 and ψ2 it follows the y is a linear com-

bination of y1 and y2 . u
t
 
ab
Remark 5. The matrix that appears in the proof above appears in
cd
other contexts as well. If y1 and y2 are homogeneous solutions to Ly = 0 we
define the Wronskian matrix by
 
y (t) y2 (t)
W (y1 , y2 )(t) = 10
y1 (t) y20 (t)
236 6 Second Order Linear Differential Equations

and the Wronskian by

w(y1 , y2 )(t) = det W (y1 , y2 ).

The relations

y1 = aψ1 + bψ2
y2 = cψ1 + dψ2 ,

in the proof, when evaluated at t0 imply that

y1 (t0 ) y10 (t0 )

   
ab
= = W (y1 , y2 )t (t0 ).
cd y1 (t0 ) y20 (t0 )

Since it was shown that ad − bc 6= 0 we have shown the following proposi-

tion.

Proposition 6. Suppose L satisfies the conditions of Theorem 4. Suppose y1

and y2 are linearly independent solutions to Ly = 0. Then

w(y1 , y2 ) 6= 0.

On the other hand, given any two differentiable functions, y1 and y2 , (not
necessarily homogeneous solutions) whose Wronskian is a nonzero function
then it is easy to see that y1 and y2 are independent. For suppose, t0 is chosen
so that w(y1 , y2 )(t0 ) 6= 0 and c1 y1 + c2 y2 = 0. Then c1 y10 + c2 y20 = 0 and we
have      
0 c1 y1 (t0 ) + c2 y2 (t0 ) c
= = W (y1 , y2 ) 1 .
0 c1 y10 (t0 ) + c2 y20 (t0 ) c2
Simple matrix algebra 1 gives c1 = 0 and c2 = 0. Hence y1 and y2 are linearly
independent.
Although one could check independence in this way it is simpler and more
to the point to use the observation given in Remark 3.

Remark 7. Let’s now summarize what Theorems 6.1.3, 6.1.6, and 4 tell us.
In order to solve L(y) = f (satisfying the continuity hypotheses) we first need
to find a particular solution yp , which exists by the Uniqueness and Existence
Theorem 6. Next, Theorem 4 says that if y1 and y2 are any two linearly
independent solutions of the associated homogeneous equation L(y) = 0, then
all of the solutions of the associated homogeneous equation are of the form
c1 y1 + c2 y2 . Theorem 3 now tells us that all solutions to L(y) = f are of the
form yp + c1 y1 + c2 y2 for some choice of the constants c1 and c2 . Furthermore,
any set of initial conditions uniquely determine the constants c1 and c2 .

1
c.f. Chapter 9 for a discussion of matrices
 6.2 The Homogeneous Case 237

A set {y1 , y2 } of linearly independent solutions to the homogeneous equa-

tion L(y) = 0 is called a fundamental set for the second order linear differ-
ential operator L. A fundamental set is a basis of the linear space of homoge-
neous solutions (c.f. Section 3.7 for the definition of a basis). Furthermore, the
standard basis, Bq , in the context of constant coefficient differential equations,
is a fundamental set.
In the following sections we will develop methods, under suitable assump-
tions, for finding a fundamental set for L and a particular solution to the
differential equation L(y) = f . For now, let’s illustrate the main theorems
with an example.

Example 8. Consider the differential equation

t2 y 00 + ty 0 + y = 2t.

Suppose
• A particular solution is yp (t) = t.
• Two independent solutions of the homogeneous equation L(y) = 0 are
y1 (t) = cos(ln t) and y2 (t) = sin(ln t).
Determine the solution set and the largest interval on which these solutions
are valid.

I Solution. In this case we divide by t2 to rewrite the equation in standard

form as y 00 + (1/t)y 0 + (1/t2 )y = 2/t and observe that the coefficients are
continuous on the interval (0, ∞). Here L = D2 + (1/t)D + (1/t2 ) and the
forcing function is f (t) = 1/t2 . It is easy to see that {cos(ln t), sin(ln t)} is
linearly independent and thus a fundamental set for L(y) = 0. It now follows
that the solution set to L(y) = f is given by

{t + c1 cos(ln t) + c2 sin(ln t) : c1 , c2 ∈ R} .

Exercises
1–8. Determine if each of the following 5. y1 (t) = ln(2t), y2 (t) = ln(5t)
pairs of functions are linearly indepen-
dent or linearly dependent. 6. y1 (t) = ln t2 , y2 (t) = ln t5

1. y1 (t) = 2t, y2 (t) = 5t 7. y1 (t) = sin 2t, y2 (t) = sin t cos t

2. y1 (t) = t2 , y2 (t) = t5 8. y1 (t) = cosh t, y2 (t) = 3et (1+e−2t )

3. y1 (t) = e2t , y2 (t) = e5t 9. 1. Verify that y(t) = t3 and y2 (t) =

|t3 | are linearly independent on
4. y1 (t) = e2t+1 , y2 (t) = e2t−3 (−∞, ∞).
238 6 Second Order Linear Differential Equations

2. Show that the Wronskian, tial equation t2 y 00 − 2ty 0 = 0,

w(y1 , y2 )(t) = 0 for all t ∈ R. y(0) = 0, y 0 (0) = 0.
3. Explain why Parts (a) and (b) 5. Explain why Parts (a), (b),
do not contradict Proposition 6. and (d) do not contradict the
4. Verify that y1 (t) and y2 (t) are Uniqueness and Existence theo-
solutions to the linear differen- rem, Theorem 6 of Section 6.1.

6.3 The Cauchy-Euler Equations

When the coefficient functions of a second order linear differential equation are
nonconstant the corresponding equation can become very difficult to solve. In
order to expect to find solutions one must put certain restrictions on the coeffi-
cient functions. A class of nonconstant coefficient linear differential equations,
known as Cauchy-Euler equations, have solutions that are easy to obtain.
A Cauchy-Euler equation is a second order linear differential equation
of the following form:

at2 y 00 + bty 0 + cy = 0, (1)

where a, b and c are real constants and a 6= 0. When put in standard form we
obtain:
b c
y 00 + y 0 + 2 y = 0.
at at
b
The functions at and atc2 are continuous everywhere except at 0. Thus by the
Uniqueness and Existence Theorem solutions exist in either of the intervals
(−∞, 0) or (0, ∞). Of course, a solution need not, and in general, will not
exist on the entire real line R. To work in a specific interval we will assume
t > 0. We will refer to L = at2 D 2 + btD + c as a Cauchy-Euler operator.
The Laplace transform method does not work in any simple fashion here.
However, the simple change in variable t = ex will transform Equation (1) into
a constant coefficient linear differential equation. To see this let Y (x) = y(ex ).
Then the chain rule gives

Y 0 (x) = ex y 0 (ex )
and Y 00 (x) = ex y 0 (ex ) + (ex )2 y 00 (ex )
= Y 0 (x) + (ex )2 y 00 (ex ).

Thus

a(ex )2 y 00 (ex ) = aY 00 (x) − aY 0 (x)

bex y 0 (ex ) = bY 0 (x)
cy(ex ) = cY (x).
 6.3 The Cauchy-Euler Equations 239

Addition of these terms gives

a(ex )2 y 00 (ex ) + bex y 0 (ex ) + cy(ex ) = aY 00 (x) − aY 0 (x) + bY 0 (x) + cY (x)

= aY 00 (x) + (b − a)Y 0 (x) + cY (x).

With t replaced by ex in Equation 1 we now obtain

aY 00 (x) + (b − a)Y 0 (x) + cY (x) = 0. (2)

The polynomial
q(s) = as2 + (b − a)s + c
is the characteristic polynomial of Equation 2 and known as the indicial
polynomial of Equation 1. Equation (2) is a second order constant coefficient
differential equation and by now routine to solve. Its solutions depends on the
way q(s) factors. We consider the three possibilities.

q has distinct real roots

Suppose r1 and r2 are distinct roots to the indicial polynomial q(s). Then
er1 x and er2 x are solutions to Equation (2). Solutions to Equation (1) are
obtained by the substitution x = ln t: we have er1 x = er1 ln t = tr1 and similarly
er2 x = tr2 . Since tr1 is not a multiple of tr2 they are independent and hence
{tr1 , tr2 } is a fundamental set for L(y) = 0.

Example 1. Find a fundamental set for the equation t2 y 00 − 2y = 0.

I Solution. The indicial polynomial is s2 −s−2 = (s−2)(s+1) and it has 2

2 −1
and −1 as roots and thus t , t is a fundamental set for this Cauchy-Euler

equation. J

q has a double root

Suppose r is a double root of q. Then erx and xerx are independent solutions to
Equation (2). The substitution x = ln t then gives tr and tr ln t as independent
solutions to Equation (1). Hence {tr , tr ln t} is a fundamental set for L(y) = 0.

Example 2. Find a fundamental set for the equation 4t2 y 00 + 8ty 0 + y = 0.

I Solution. The indicial polynomialn is 4s2 + 4so+ 1 = (2s + 1)2 and has − 12
1 1
as a root with multiplicity 1. Thus t− 2 , t− 2 ln t is a fundamental set. J
240 6 Second Order Linear Differential Equations

q has conjugate complex roots

Suppose q has complex roots α ± iβ, where β 6= 0. Then eαx cos βx and
eαx sin βx are independent solutions to Equation (2). The substitution x = ln t
then gives {tα cos(β ln t), tα sin(β ln t)} as a fundamental set for Ly = 0.

Example 3. Find a fundamental set for the equation t2 y 00 + ty 0 + y = 0.

I Solution. The indicial polynomial is s2 +1 which has ±i as complex roots.

Theorem 4 implies that {cos ln t, sin ln t} is a fundamental set. J

We now summarize the above results into one theorem.

Theorem 4. Let L = at2 D 2 + btD + c, where a, b, c ∈ R and a 6= 0. Let
q(s) = as2 + (b − a)s + c be the indicial polynomial.
1. If r1 and r2 are distinct real roots of q(s) then

{tr1 , tr2 }

is a fundamental set for L(y) = 0.

2. If r is a double root of q(s) then

{tr , tr ln t}

is a fundamental set for L(y) = 0.

3. If α ± iβ are complex conjugate roots of q(s), β 6= 0 then

{tα sin(β ln t), tα cos(β ln t)}

is a fundamental set for L(y) = 0.

Exercises
1–11. Find the general solution of each 6. t2 y 00 − 3ty 0 − 21 = 0
of the following homogeneous Cauchy-
Euler equations on the interval (0, ∞). 7. t2 y 00 + 7ty 0 + 9y = 0

8. t2 y 00 − y = 0
1. t2 y 00 + 2ty 0 − 2y = 0
9. t2 y 00 + ty 0 − 4y = 0
2. 2t2 y 00 − 5ty 0 + 3y = 0
10. t2 y 00 + ty 0 + 4y = 0
3. 9t2 y 00 + 3ty 0 + y = 0
11. t2 y 00 − 3ty 0 + 13y = 0
4. t2 y 00 + ty 0 − 2y = 0
12–15. Solve each of the following ini-
5. 4t2 y 00 + y = 0 tial value problems.
 6.4 Laplace Transform Methods 241

12. t2 y 00 + 2ty 0 − 2y = 0, y(1) = 0, 14. t2 y 00 + ty 0 + 4y = 0, y(1) = −3,

y 0 (1) = 1 y 0 (1) = 4

13. 4t2 y 00 + y = 0, y(1) = 2, y 0 (1) = 15. t2 y 00 − 4ty 0 + 6y = 0, y(0) = 1,

0 y 0 (0) = −1

6.4 Laplace Transform Methods

In this section we will develop some further properties of the Laplace transform
and use them to solve some linear differential equations with non constant
coefficient functions. The use of the Laplace transform is limited, however, as
we shall see.
To begin with we need to consider a class of functions wider than the
exponential-polynomials. A continuous function f on [0, ∞) is said to be of
exponential type with order a if there is a constant K such that

|f (t)| ≤ Keat

for all t ∈ [0, ∞). If the order is not important to the discussion we will just say
f is of exponential type. This definition applies to both real and complex
valued functions. The idea here is to limit the kind of growth that we allow
f to have; it cannot grow faster than a multiple of an exponential function.
The above inequality means

−Keat ≤ f (t) ≤ Keat ,

for all t ∈ [0, ∞) as illustrated in Figure 6.1, where the boldfaced curve, f (t),
lies between the upper and lower exponential functions.
As we will see below, limiting growth in this way will assure us that f
has a Laplace transform. It is an easy exercise to show that the sum and
product of functions of exponential type are again of exponential type. Also
any continuous bounded function on [0, ∞) is of exponential type. Since tn ≤
 n λt
e for t ≥ 0 it follows that t e  ≤ e(n+a)t , where a = Re λ. Hence, the
nt 
exponential-polynomials are of exponential type and thus everything we say
about functions of exponential type applies to the exponential-polynomials.
Proposition 1. Let f be of exponential type with order a. Then the Laplace
transform L{f (t)}(s) exists for all s > a.
Proof. Let f be exponential type of order a. Then |f | ≤ Keat for some K and
Z ∞  Z ∞ Z ∞
−st −st K
e−(s−a)t dt =
 
 e f (t) dt ≤ e |f (t)| dt ≤ K ,

0

0 0 s−a
provided s > a. u
t
242 6 Second Order Linear Differential Equations

Fig. 6.1. The exponential function Keat bounds f (t).

Note here that the Laplace transform is a function of s for all s > a. Unless
necessary we will not mention this restriction. This is also a good place to note
that many functions are not of exponential type. For example, consider the
2
function y(t) = et . If a ∈ R then
2 2 2 −a2
+at+ a4
et eat = et e 4

a 2 −a2
= e(t− 2 ) e 4

2 −a2
= eu e 4 ,
2
where u = t − a2 . As t approaches infinity so does u. Since limu→∞ eu = ∞
2 2
it is clear that limt→∞ et eat = ∞, for all a ∈ R, and hence y(t) = et is not
of exponential type. This same argument implies that the Laplace transform
2
of et does not exist.

Asymptotic Values
An interesting property of the Laplace transform is that certain limiting values
of f (t) can be deduced from its Laplace transform and vice versa.

Lemma 2. Suppose f is of exponential type and F (s) = L{f (t)}(s). Then

lim F (s) = 0.
s→∞

Proof. Suppose f is of exponential type with order a. As in the Proof of

Theorem 1 we have
K
|F (s)| ≤ .
s−a
Taking limits gives the result. u
t
 6.4 Laplace Transform Methods 243

Theorem 3 (Initial Value Theorem). Suppose f and its derivative f 0 are

of exponential order. Let F (s) = L{f (t)}(s). Then

Initial Value Principle

f (0) = lim sF (s).

s→∞

Proof. Let H(s) = L{f 0 (t)}(s). By Lemma 2 we have

0 = lim H(s) = lim (sF (s) − f (0)) = lim (sF (s) − f (0)).
s→∞ s→∞ s→∞

This implies the result. u

t

Example 4. Verify the Initial Value Theorem for f (t) = cos at.

I Solution. On the one hand, cos at|t=0 = 1. On the other hand,

s2
sL{cos at}(s) =
s2 + a 2
which has limit 1 as s → ∞. J

Theorem 5 (Final Value Theorem). Suppose f and f 0 are of exponential

type and limt→∞ f (t) exists. If F (s) = L{f (t)}(s) then

Final Value Principle

lim f (t) = lim sF (s).

t→∞ s→0

Proof. Let H(s) = L{f 0 (t)}(s) = sF (s) − f (0). Then sF (s) = H(s) + f (0)
and

lim sF (s) = lim H(s) + f (0)

s→0 s→0
Z M
= lim lim e−st f 0 (t) dt + f (0)
s→0 M→∞ 0
Z M
= lim f 0 (t) dt + f (0)
M→∞ 0
= lim f (M ) − f (0) + f (0)
M→∞
= lim f (M ).
M→∞

u
t
244 6 Second Order Linear Differential Equations

Integration in Transform Space

The Transform Derivative Principle, Theorem 3.1.16, tells us that multiplica-
tion by −t induces differentiation of the Laplace transform. One might expect
then that division by −t will induce integration in the transform space. This
idea is valid but we must be careful about assumptions. First if f (t) has a
Laplace transform it is not necessarily the case that f (t)/t will likewise. For
example, the constant function f (t) = 1 has Laplace transform 1s but f (t)
t = t
1

does not have a Laplace transform. Second, integration produces an arbitrary

constant of integration. What is this constant? The precise statement is as
follows:

Theorem 6 (Integration in Transform Space). Suppose f is of exponen-

tial type with order a and f (t)
t has a continuous extension to 0, i.e. limt→0
+
f (t)
t
exists. Then f (t)
t is of exponential type with order a and

Tranform Integral Principle

n o R∞
L f (t)
t (s) = s F (σ) dσ,

where s > a.
f (t)
Proof. Let L = limt→0+ t and define
(
f (t)
t if t > 0
h(t) = .
L if t = 0

Since f is continuous so is h. Since f is of exponential type with order a there

is a K so that |f (t)| ≤ Keat . Since 1t ≤ 1 on [1, ∞)
 
 f (t) 
|h(t)| = 
  ≤ |f (t)| ≤ Keat ,
t 

for all t ≥ 1. Since h is continuous on [0, 1] it is bounded by B, say. Thus

|h(t)| ≤ B ≤ Beat for all t ∈ [0, 1]. If M is the larger of K and B then

|h(t)| ≤ M eat ,

for all t ∈ [0, ∞) and hence h is of exponential type. Let H(s) = L{h(t)}(s)
and F (s) = L {f (t)} (s). Then, since −th(t) = −f (t) we have, by Theorem
3.1.16, H 0 (s) = −F (s). Thus H is an antiderivative of −F and we have
Z s
H(s) = − F (σ) dσ + C.
a
 6.4 Laplace Transform Methods 245
R∞
Lemma 2 implies 0 = lims→∞ H(s) = − a
F (σ) dσ + C and hence C =
R∞
a
F (σ) dσ. Therefore
 
f (t)
L (s) = H(s)
t
Z a Z ∞
= F (σ) dσ + F (σ) dσ
Zs ∞ a

= F (σ) dσ.
s

u
t

The Laplace transform of several new functions can now be deduced from
this theorem. Consider an example.

Example 7. Find L sint t .



I Solution. Since limt→0 sint t = 1 Theorem 6 applies to give

  Z ∞
sin t 1
L (s) = dσ
t s σ2 + 1
= tan−1 σ|∞
s
π −1
= − tan (s)
2
1
= tan−1 .
s
J

Solving Linear Differential Equations

We now consider by example how one can use the Laplace transform method
to solve some differential equations.

Example 8. Find a solution of exponential type that solves

ty 00 − (1 + t)y 0 + y = 0.

I Solution. Note that the Existence and Uniqueness theorem implies that
solutions exist on intervals that do not contain 0. We presume that such a
solution has a continuous extension to t = 0 and is of exponential type. Let y
be such a solution. Let y(0) = y0 , y 0 (0) = y1 and Y (s) = L {y(t)} (s). Appli-
cation of Transform Derivative Principle, Theorem 3.116, to each component
of the differential equation gives
246 6 Second Order Linear Differential Equations

L {ty 00 } = −(s2 Y (s) − sy(0) − y 0 (0))0

= −(2sY (s) + s2 Y 0 (s) − y(0))
L {−(1 + t)y 0 } = L{−y 0 } + L{−ty 0 }
= −sY (s) + y0 + (sY (s) − y0 )0
= sY 0 (s) − (s − 1)Y (s) + y0
L {y} = Y (s).

The sum of the left-hand terms is given to be 0. Thus adding the right-hand
terms and simplifying gives

(s − s2 )Y 0 (s) + (−3s + 2)Y (s) + 2y0 = 0,

which can be rewritten in the following way:

3s − 2 2y0
Y 0 (s) + Y (s) = .
s(s − 1) s(s − 1)
3s−2
This equation is a first order linear differential equation in Y (s). Since s(s−1) =
2 1 2
s + s−1 it is easy to see that an integrating factor is I = s (s − 1) and hence

(IY (s))0 = 2y0 s.

Integrating and solving for Y gives

y0 c
Y (s) = + .
s − 1 s2 (s − 1)

The inverse Laplace transform is y(t) = y0 et + c(et − t − 1). For simplicity, we

can write this solution in the form

y(t) = Aet + B(t + 1),

where A = y0 + c and B = −c. It is easy to verify that et and t + 1 are linearly

independent and solutions to the given differential equation. J

Example 9. Find a solution of exponential type that solves

ty 00 + 2y 0 + ty = 0.

I Solution. Again, assume y is a solution of exponential type. Let Y (s) =

L{y(t)}(s). As in the preceding example we apply the Laplace transform and
simplify. The result is a simple linear differential equation:
−y0
Y 0 (s) = .
s2 + 1
Integration gives Y (s) = y0 (− tan−1 s + C). By Lemma 2 we have 0 =
lims→∞ Y (s) = y0 (− π2 + C) which implies C = π2 and
 6.4 Laplace Transform Methods 247

π 1
Y (s) = y0 ( − tan−1 s) = y0 tan−1 .
2 s
By Example 7 we get
sin t
y(t) = y0 .
t
J
Theorem 6.2.4 implies that there are two linearly independent solutions.
The Laplace transform method has found only one, namely, sint t . In the Sec-
tion 6.5 we will introduce a technique that will find another independent
solution. When applied to this example we will find y(t) = cost t is another
solution. (c.f. Example 6.1.4.) It is easy to check that the Laplace transform
of cost t does not exist and thus the Laplace transform method cannot find it
as a solution. Furthermore, the constant of integration, C, in this example
cannot be arbitrary, because of Lemma 2. It frequently happens in examples
that C must be carefully chosen.
We observe that the presence of the linear factor t in Examples 8 and 9
produces a differential equation of order 1 which can be solved by techniques
learned in Chapter 2. Correspondingly, the presence of higher order terms, tn ,
produce differential equations of order n. For example, the Laplace transform
applied to the differential equation t2 y 00 + 6y = 0 gives, after a short calcu-
lation, s2 Y 00 (s) + 4sY 0 (s) + 8Y (s) = 0. The resulting differential equation in
Y (s) is still second order and no simpler than the original. In fact, both are
Cauchy-Euler. Thus, when the coefficient functions are polynomial of order
greater then one, the Laplace transform method will generally be of little use.
For this reason we will usually limit our examples to second order linear dif-
ferential equations with coefficient functions that are linear terms, i.e. of the
form at + b. Even with this restriction we still will need to solve a first order
differential equation in Y (s) and determine its inverse Laplace transform; not
always easy problems. We provide a list of Laplace transform pairs at the end
of this chapter for quick reference. This list extends the list given in Chapter
3 and should facilitate the Laplace inversion for the exercises given here.

Laguerre Polynomials
The Laguerre polynomial, `n (t), of order n is the polynomial solution to
Laguerre’s differential equation

ty 00 + (1 − t)y 0 + ny = 0,

where y(0) = 1 and n is a nonnegative integer.

Proposition 10. The nth Laguerre polynomial is given by
n   k
X n t
`n (t) = (−1)k
k k!
k=0
248 6 Second Order Linear Differential Equations

and
(s − 1)n
L {`n (t)} (s) = .
sn+1
Proof. Taking the Laplace transform of Laguerre’s differential equation gives
(s2 − s)Y 0 (s) + (s − (1 + n))Y (s) = 0
n
and hence Y (s) = C (s−1)
sn+1 . By the Initial Value Theorem

s(s − 1)n
1 = y(0) = lim C = C.
s→∞ sn+1
Now using the binomial theorem we get (s − 1)n = nk=0 nk (−1)k sn−k and
P

Pn k n 1
hence Y (s) = k=0 (−1) k sk+1 . It now follows by inversion that y(t) =


Pn k
`n (t) = k=0 (−1)k nk tk! .


u
t
It is easy to see that the first five Laguerre polynomials are:
`0 (t) = 1
`1 (t) = 1 − t
t2
`2 (t) = 1 − 2t +
2
3t2 t3
`3 (t) = 1 − 3t + −
2 6
3
2t t4
`4 (t) = 1 − 4t + 3t2 − + .
3 24
Below are their graphs on the interval [0, 6].

`2

`3

`◦

`4

`1
 6.4 Laplace Transform Methods 249

Define the following differential operators

E◦ = 2tD2 + (2 − 2t)D − 1
E+ = tD2 + (1 − 2t)D + (t − 1)
E− = tD2 + D.
Theorem 11. We have the following differential relationships amongst the
Laguerre polynomials.
1. E◦ `n = −(2n + 1)`n
2. E+ `n = −(n + 1)`n+1
3. E− `n = −n`n−1 .
Proof.
1. Let An = tD2 +(1−t)D+n be Laguerre’s differential equation. Then from
the defining equation of the Laguerre polynomial `n we have An `n = 0.
Multiply this equation by 2 and add −(1 + 2n)`n to both sides. This gives
E◦ `n = −(2n + 1)`n

2. A simple observation gives E+ = An − tD − (1 − t − n). Since An `n = 0

it is enough to verify that t`0n + (1 − t − n)`n = (n + 1)`n + 1. This we do
in transform space. Let Ln = L {`n } .
L {t`0n + (1 − t − n)`n } (s) = −(sLn (s) − `n (0))0 + (1 − n)Ln (s) + L0n (s)
= −(Ln + sL0n ) + (1 − n)Ln + L0n
= −(s − 1)L0n − nLn
−(s − 1)n
= (ns − (n + 1)(s − 1) − ns)
sn+2
= (n + 1)L {`n+1 } (s).
3. This equation is proved in a similar manner as above. We leave the details
to the exercises. J
For a differential operator A let A2 y = A(Ay), A3 y = A(A(Ay)), etc. It is
easy to verify by induction and the use of Theorem 11 that
(−1)n n
E+ `◦ = `n .
n!
The operator E+ is called a creation operator since successive applications
to `◦ creates all the other Laguerre polynomials. In a similar way it is easy to
verify that
m
E− `n = 0,
for all m > n. The operator E− is called an annihilation operator.

Exercises
250 6 Second Order Linear Differential Equations

1–4. For each of the following functions 19. The Laguerre polynomial of order
show that Theorem 6 applies and use it n can be defined in another way:
1 t dn
to find its Laplace transform. `n (t) = n! e dtn (e−t tn ). Show that
this definition is consistent with the
ebt − eat definition in the text.
1.
t
20. Verify Equation (3) in Theorem 11:
cos bt − cos at
2. 2
t E− `n = −n`n−1 .
cos bt − cos bt
3. 2
t2 21. The Lie bracket [A, B] of two dif-
sin at ferential operators A and B is de-
4. fined by
t
5–13. Use the Laplace transform to [A, B] = AB − BA.
find solutions to each of the following dif-
Show the following:
ferential equations.
• [E◦ , E+ ] = −2E+
5. (t − 1)y 00 − ty 0 + y = 0 • [E◦ , E− ] = 2E−
• [E+ , E− ] = E◦ .
6. ty 00 + (t − 1)y 0 − y = 0
22. Show that the Laplace n
transform of
7. ty 00 + (1 + t)y 0 + y = 0 `n (at), a ∈ R, is (s−a)
n+1 .
s

8. ty 00 + (2 + 4t)y 0 + (4 + 4t)y = 0 23. Verify that

9. ty 00 − 2y 0 + ty = 0 n
n k
a `k (t)(1 − a)n−k = `n (at).
k
10. (t − 2)y 00 − (t − 1)t0 + y = 0 k=0 

11. ty 00 + (2 + 2t)y 0 + (2 + t)y = 0 24. Show that

12. ty − 4y + ty = 0, assume y(0) = 0.
00 0 t

`n (x) dx = `n (t) − `n (t).
0
13. (1 + t)y 00 + (1 + 2t)y 0 + ty = 0

14–18. Use the Laplace transform to 25. Verify the following recursion for-
find solutions to each of the following dif- mula:
ferential equations. Use the results of Ex- 1
ercises 1 to 4 or Table ?? `n+1 (t) = ((2n + 1 − t)`n (t) − n`n−1 (t)) .
n+1
14. −ty 00 + (t − 2)y 0 + y = 0
26. Show that
15. −ty 00 − 2y 0 + ty = 0 t

`n (x)`m (t−x) dx = `m+n (t)−`m+n+1 (t).
16. ty 00 + (2 − 5t)y 0 + (6t − 5)y = 0 0

17. ty 00 + 2y 0 + 9ty 0 = 0 27. Show that

00 0 ∞
18. ty + (2 + t)y + y = 0
e−x `n (x) dx = e−t (`n (t) − `n−1 (t)) .
t
19–28. Laguerre polynomials: Each
of these problems develops further prop-
erties of the Laguerre polynomials. 28.
 6.5 Reduction of Order 251

29–33. Functions of Exponential 32. Show that if f is of exponential type

type: Verify the following claims. then any antiderivative is also of ex-
29. Show that the sum of two functions ponential type.
of exponential type is of exponential 2
type. 33. Let y(t) = sin(et ). Why is y(t)
of exponential type? Compute y 0 (t)
30. Show that the product of two expo- and show that it is not of exponen-
nential type functions is of exponen- tial type. The moral: The derivative
tial type. of a function of exponential type is
not necessarily of exponential type.
31. Show that any bounded continuous
function on [0, ∞) is of exponential
type.

6.5 Reduction of Order

It is a remarkable feature of linear differential equations that one nonzero
homogeneous solution can be used to obtain a second independent solution.
Suppose L = a2 (t)D2 + a1 (t)D + a0 (t) and suppose y1 (t) is a known nonzero
solution . Let
y2 (t) = u(t)y1 (t), (1)
where u(t) will be determined by substituting y2 into Ly = 0. This imposes
conditions on u that are, in principle, solvable and hence will lead to a formula
for y2 . The product rule for differentiation gives

y20 = u0 y1 + uy10
and y200 = u00 y1 + 2u0 y10 + uy100 .

Substituting these equations into Ly2 gives

Ly2 = a2 y200 + a1 y20 + a0 y2

= a2 (u00 y1 + 2u0 y10 + uy100 ) + a1 (u0 y1 + uy10 ) + a0 uy1
= u00 a2 y1 + 2u0 a2 y10 + u0 a1 y1 + u(a2 y100 + a1 y10 + a0 y1 )
= u00 a2 y1 + u0 (2a2 y10 + a1 y1 ).

In the third line above the coefficient of u is zero because y1 is assumed to be

a solution to Ly = 0. The equation Ly2 = 0 implies

u00 a2 y1 + u0 (2a2 y10 + a1 y1 ) = 0, (2)

another second order differential equation in u. One obvious solution to Equa-

tion 2 is u(t) a constant, implying y2 is a multiple of y1 . To find another
independent solution we use the substitution v = u0 to get
252 6 Second Order Linear Differential Equations

v 0 a2 y1 + v(2a2 y10 + a1 y1 ) = 0,

a first order separable differential equation in v. This substitution gives this

procedure its name: reduction of order. It is now straightforward to solve
for v. In fact, separating variables gives
v0 −2y10 a1
= − .
v y1 a2
From this we get
1 − a1 /a2
v= e  .
y12
Since v = u0 we integrate v to get
1 − a1 /a2
Z
u= e  , (3)
y12
which is independent of the constant solution. Substituting Equation 3 into
Equation 1 then gives a new solution independent of y1 . Admittedly, Equa-
tion 3 is difficult to remember and not very enlightening. In the exercises we
recommend following the procedure we have outlined above. This is what we
shall do in the examples to follow.

Example 1. The function y1 (t) = et is a solution to

(t − 1)y 00 − ty 0 + y = 0.

Use reduction of order to find another independent solution and write down
the general solution.

I Solution. Let y2 (t) = u(t)et . Then

y20 (t) = u0 (t)et + u(t)et

y200 (t) = u00 (t)et + 2u0 (t)et + u(t)et .

Substitution into the differential equation (t − 1)y 00 − ty 0 + y = 0 gives

(t − 1)(u00 (t)et + 2u0 (t)et + u(t)et ) − t(u0 (t)et + u(t)et ) + u(t)et = 0

which simplifies to
(t − 1)u00 + (t − 2)u0 = 0.
Let v = u0 . Then we get (t − 1)v 0 + (t − 2)v = 0. Separating variables gives

v0 −(t − 2) 1
= = −1 +
v t−1 t−1
with solution v = e−1 (t − 1). Integration by parts gives u(t) =
R
v(t) dt =
−te−t . Substitution gives
 6.5 Reduction of Order 253

y2 (t) = u(t)et = −te−t et = −t.

It is easy to verify that this is indeed a solution. Since our equation is homoge-
neous we know −y2 (t) = t is also a solution. Clearly t and et are independent.
By Theorem 4 the general solution is
y(t) = c1 t + c2 et .
J
sin t
Example 2. In Example 6.4.9 we showed that y1 = t is a solution to
ty 00 + 2y 0 + ty = 0.
Use reduction of order to find a second independent solution and write down
the general solution.
I Solution. Let y2 (t) = u(t) sint t . Then
sin t t cos t − sin t
y20 (t) = u0 (t) + u(t)
t t2
sin t t cos t − sin t −t2 sin t − 2t cos t + 2 sin t
y200 (t) = u00 (t) + 2u0 (t) 2
+ u(t) .
t t t3
We next substitute y2 into ty 00 + wy 0 + ty = 0 and simplify to get
u00 (t) sin t + 2u0 (t) cos t = 0.
Let v = u0 . Then we get v 0 (t) sin t + 2v(t) cos t = 0. Separating variables gives
v0 −2 cos t
=
v sin t
with solution
v(t) = csc2 (t).
Integration gives u(t) = v(t) dt = − cot(t) and hence
R

sin t − cos t
y2 (t) = − cot t
= .
t t
By Theorem 6.2.4, the general solution can be written as
sin t cos t
+ c2
c1 .
t t
Compare this result with Examples 6.1.4 and 6.4.9. J
We remark that the constant of integration in the computation of u was
chosen to be 0 in both examples. There is no loss in this for if a nonzero
constant, c say, is chosen then y2 = uy1 + cy1 . But cy1 is already known to be
a homogeneous solution. We gain nothing by including it in y2 .

Exercises
254 6 Second Order Linear Differential Equations

1–15. For each differential equation use 8. t2 y 00 − 4ty 0 + (t2 + 6)y = 0, y1 (t) =
the given solution to find a second in- t2 cos t
dependent solution and write down the
general solution. 9. ty 00 − y 0 + 4t3 y = 0, y1 (t) = sin t2
1. t2 y 00 − 3ty 0 + 4y = 0 y1 (t) = t2 10. ty 00 − 2(t + 1) + 4y = 0, y1 (t) = e2t
2. t2 y 00 + 2ty 0 − 2y = 0, y1 (t) = t 11. ty 00 − 2(sec2 t) y = 0, y1 (t) = tan t
2 00 2
3. (1 − t )y + 2y = 0 y1 (t) = 1 − t
12. ty 00 + (t − 1)y 0 − y = 0, y1 (t) = e−t
4. (1 − t2 )y 00 − 2ty 0 + 2y = 0, y1 (t) = t
√ 13. y 00 −(tan t)y 0 −(sec2 t)y = 0, y1 (t) =
5. 4t2 y 00 + y = 0, y1 (t) = t tan t

6. t2 y 00 + 2ty 0 = 0, y1 (t) = t 14. (1 + t2 )y 00 − 2ty 0 + 2y = 0, y1 (t) = t

7. t2 y 00 − t(t + 2)y 0 + (t + 2)y = 0, 15. t2 y 00 − 2ty 0 + (t2 + 2)y = 0, y1 (t) =
y1 (t) = t t cos t

6.6 Variation of Parameters

Let L be a second order linear differential operator. In this section we address
the issue of finding a particular solution to a nonhomogeneous linear differ-
ential equation L(y) = f , where f is continuous on some interval I. It is a
pleasant feature of linear differential equations that the homogeneous solu-
tions can be used decisively to find a particular solution. The procedure we
use is called variation of parameters and, as you shall see, is akin to the
method of reduction of order.
Suppose, in particular, that L = D2 + a1 (t)D + a0 (t), i.e. we will assume
that the leading coefficient function is 1 and it is important to remember
that this assumption is essential for the method we develop below. Suppose
{y1 , y2 } is a fundamental set for L(y) = 0. We know then that all solutions
of the homogeneous equation L(y) = 0 are of the form c1 y1 + c2 y2 . To find
a particular solution yp to L(y) = f the method of variation of parameters
makes two assumptions. First, the parameters c1 and c2 are allowed to vary
(hence the name). We thus replace the constants c1 and c2 by functions u1 (t)
and u2 (t), and assume that the particular solution yp , takes the form

yp (t) = u1 (t)y1 (t) + u2 (t)y2 (t). (1)

The second assumption is

u01 (t)y1 (t) + u02 (t)y2 (t) = 0. (2)

What’s remarkable is that these two assumptions consistently lead to explicit

formulas for u1 (t) and u2 (t) and hence a formula for yp .
 6.6 Variation of Parameters 255

To simplify notation in the calculations that follow we will drop the ‘t’
in expression like u1 (t), etc. Before substituting yp into L(y) = f we first
calculate yp0 and yp00 .
yp0 = u01 y1 + u1 y10 + u02 y2 + u2 y20
= u1 y10 + u2 y20 (by Equation 2).
Now for the second derivative
yp00 = u01 y10 + u1 y100 + u02 y20 + u2 y200 .
We now substitute yp into L(y).
L(yp ) = yp00 + a1 yp0 + a0 yp
= u01 y10 + u1 y100 + u02 y20 + u2 y200
+a1 (u1 y10 + u2 y20 )
+a0 (u1 y1 + u2 y2 )
= u01 y10 + u02 y20 + u1 (y100 + a1 y10 + a0 y1 ) + u2 (y200 + a1 y20 + a0 y2 )
= u01 y10 + u02 y20 .
In the second to the last equation the coefficients of u1 and u2 are zero because
y1 and y2 are assumed to be homogeneous solutions. The second assumption,
Equation 2 and the equation L(yp ) = f now lead to the following system:
u01 y1 + u02 y2 = 0
u01 y10 + u02 y20 = f
which can be rewritten in matrix form as
y1 y2 u01
    
0
= . (3)
y10 y20 u02 f
The left most matrix in Equation (3) is none other than the Wronskian matrix,
W (y1 , y2 ), which has a nonzero determinant because {y1 , y2 } is a fundamental
set (c.f. Theorem 6.2.4 and Proposition 6.2.6). By Cramer’s rule, we can solve
for u01 and u02 . We obtain
−y2 f
u01 =
w(y1 , y2 )
y1 f
u02 = .
w(y1 , y2 )
We now obtain an explicit formula for a particular solution:
yp (t) = u1 y1 + u2 y2
−y2 f
y1 f

Z Z
= y1 + y2 .
w(y1 , y2 ) w(y1 , y2 )
The following theorem consolidates these results with Theorem 6.1.6.
256 6 Second Order Linear Differential Equations

Theorem 1. Let L = D2 + a1 (t)D + a0 (t), where a1 (t) and a0 (t) are contin-
uous on an interval I. Suppose {y1 , y2 } is a fundamental set of solutions for
L(y) = 0. If f is continuous on I then a particular solution, yp , to L(y) = f
is given by the formula
−y2 f
y1 f

Z Z
yp = y1 + y2 . (4)
w(y1 , y2 ) w(y1 , y2 )

Furthermore, the solution set to L(y) = f becomes

{yp + c1 y1 + c2 y2 : c1 , c2 ∈ R} .

Remark 2. Equation (4), which gives an explicit formula for a particular

solution, is too complicated to memorize and we do not recommend students
to do this. Rather the point of variation of parameters is the method that
leads to Equation (4) and our recommended starting point is Equation (3).
You will see such matrix equations as we proceed in the text.

We will illustrate the method of variation of parameters with two examples.

Example 3. Find the general solution to the following equation:

t2 y 00 − 2y = t2 ln t.

I Solution. In standard form this becomes

2
y 00 − y = ln t.
t2
The associated homogeneous equation is y 00 − (2/t2 )y = 0 or, equivalently,
t2 y 00 − 2y = 0 and is a Cauchy-Euler equation. The indicial polynomial is
q(s) = s2 − s − 2 = (s − 2)(s + 1), which has 2 and −1 as roots. Thus
t−1 , t2 is a fundamental set to the homogeneous equation y 00 − (2/t2 )y = 0,


by Theorem 4 of Section 6.3. Let yp = t−1 u1 (t) + t2 u2 (t). Our starting point
is the matrix equation
 −1 2   0   
t t u1 0
=
−t−2 2t u02 ln t

which is equivalent to the system

t−1 u01 + t2 u02 = 0

−t−2 u01 + 2tu02 = ln t.

Multiplying the bottom equation by t and then adding the equations together
gives 3t2 u02 = t ln t and hence
1
u02 = ln t.
3t
 6.6 Variation of Parameters 257

Substituting u02 into the first equation and solving for u01 gives

t2
u01 = − ln t.
3
Integration by parts leads to

1 t3 t3
u1 = − ( ln t − )
3 3 9
and a simple substitution gives
1
u2 = (ln t)2 .
6
We substitute u1 and u2 into Equation (1) to get

1 t3 t3 1 t2
yp (t) = − ( ln t − )t−1 + (ln t)2 t2 = (9(ln t)2 − 6 ln t + 2).
3 3 9 6 54
It follows that the solution set is
 2 
t 2 −1 2
(9(ln t) − 6 ln t + 2) + c1 t + c2 t : c1 , c2 ∈ R .
54
J

Example 4. Find the general solution to

ty 00 + 2y 0 + ty = 1.

Use the results of Example 6.5.2.

I Solution. Example 6.4.2 showed that

sin t cos t
y1 (t) = and y2 (t) =
t t
are homogeneous solutions to ty 00 +2y 0 +ty = 0. Let yp = sint t u1 (t)+ cost t u2 (t).
Then " sin t cos t
#" # " #
t t u01 (t) 0
t cos t−sin t −t sin t−cos t 0
= 1 .
t2 t2
u2 (t) t

(We get 1/t in the last matrix because the differential equation in standard
form is y 00 + (2/t)y 0 + y = 1/t.) From the matrix equation we get the following
system
sin t 0 cos t 0
u1 (t) + u (t) = 0
t t 2
t cos t − sin t 0 −t sin t − cos t 0 1
2
u1 (t) + 2
u2 (t) = .
t t t
258 6 Second Order Linear Differential Equations

The first equation gives

u01 (t) = −(cot t)u02 (t).

If we multiply the first equation by t, the second equation by t2 , and then add
we get
(cos t)u01 (t) − (sin t)u02 (t) = 1.
Substituting in u01 (t) and solving for u02 (t) gives u02 (t) = − sin t and thus
u01 (t) = cos t. Integration gives

u1 (t) = sin t
u2 (t) = cos t.

We now substitute these functions into yp to get

sin t cos t
yp (t) = sin t + cos t
t t
sin2 t + cos2 t
=
t
1
= .
t
The general solution is
1 sin t cos t
y(t) = + c1 + c2 .
t t t
J

Exercises
1–5. Use Variation of Parameters to 5. y 00 − 3y 0 + 2y = e3t
find a particular solution and then write
down the general solution. Next solve 6–16. Use Variation of Parameters to
each using the method of undetermined find a particular solution and then write
coefficients or the incomplete partial down the general solution. In some exer-
fraction method. cises a fundamental set {y1 , y2 } is given.

1. y 00 + y = sin t 6. y 00 + y = tan t

2. y 00 − 4y = e2t et
7. y 00 − 2y 0 + y =
t
3. y 00 − 2y 0 + 5y = et
8. y 00 + y = sec t
00 0 −3t
4. y + 3y = e
9. t2 y 00 − 2ty 0 + 2y = t4
 6.6 Variation of Parameters 259

10. ty 00 − y 0 = 3t2 − 1 y1 (t) = 1 and y1 (x) y2 (x)

y2 (t) = t2
t y1 (t) y2 (t) 
yp =
 f (x)dx
y1 (x) y2 (x)
0 0
11. t2 y 00 − ty 0 + y = t y2 (x) y20 (x)

e2t
12. y 00 − 4y 0 + 4y = t2 +1
18–21. For each problem below use the
00 0 2 result of Problem 17 to obtain a particu-
13. y − (tan t)y − (sec t)y = t
lar solution to the given differential equa-
y1 (t) = tan t and y2 (t) = sec t
tion in the form given. Solve the differen-
14. ty 00 + (t − 1)y 0 − y = t2 e−t y1 (t) = tial equation using the Laplace transform
t − 1 and y2 (t) = e−t method and compare.
18. y 00 + a2 y = f (t) yp (t) = 1
a
f (t) ∗
15. ty 00 − y 0 + 4t3 y = 4t5 y1 = cos t2 sin at
and y2 (t) = sin t2
19. y 00 − a2 y = f (t) yp (t) = 1
a
f (t) ∗
00 1
16. y − y = 1+e−t
sinh at

20. y 00 − 2ay 0 + a2 y = f (t) yp (t) =

1
17. Show that the constants of integra- a
f (t) ∗ te−at
tion in the formula for yp in Theorem
1 can be chosen so that a particular 21. y 00 − (a + b)y 0 + aby = f (t), a 6=
solution can be written in the form:
1
b yp (t) = b−a f (t) ∗ (ebt − eat )
7
Power Series Methods

Thus far in our study of linear differential equations we have imposed severe
restrictions on the coefficient functions in order to find solution methods. Two
special classes of note are the constant coefficient and Cauchy-Euler differen-
tial equations. The Laplace transform method was also useful in solving some
differential equations where the coefficients were linear. Outside of special
cases such as these, linear second order differential equations with variable
coefficients can be very difficult to solve.
In this chapter we introduce the use of power series in solving differential
equations. Here’s the main idea. Suppose a second order differential equation

a2 (t)y 00 + a1 (t)y 0 + a0 (t)y = f (t)

is given. Under the right conditions on the coefficient functions a solution can
be expressed in terms of a power series which takes the form
∞
X
y(t) = cn (t − t0 )n ,
n=0

for some fixed t0 . Substituting the power series into the differential equation
∞
gives relationships amongst the coefficients {cn }n=0 , which when solved gives
a power series solution. This technique is called the power series method.
While we may not enjoy a closed form solution, as in the special cases thus
far considered, power series methods imposes the least restrictions on the
coefficient functions.

7.1 A Review of Power Series

We begin with a review of the main properties of power series that are usually
learned in a first year calculus course.
262 7 Power Series Methods

Definitions and convergence

A power series centered at t0 in the variable t is a series of the form
∞
X
cn (t − t0 )n = c0 + c1 (t − t0 ) + c2 (t − t0 )2 + · · · . (1)
n=0

The center of the power series is t0 and the coefficients are the constants
∞
{cn }n=0 . Frequently we will simply refer to Equation (1) as a power series.
Let I be the set of real numbers where the series converges. Obviously, t0 is in
I, so I is nonempty. It turns out that I is an interval and is called the interval
of convergence. It contains an open interval of the form (t0 − R, t0 + R) and
possibly one or both of the endpoints. The number R is called the radius of
convergence and can frequently be determined by the ratio test.
P∞ n
The Ratio Test for Power Series   Let n=0 cn (t − t0 ) be a given power
 cn+1 
series and suppose L = limn→∞  cn  . Define R in the following way:

R=0 if L = ∞
R=∞ if L = 0.
R = L1 if 0 < L < ∞

Then
i. The power series converges only at t = t0 if R = 0.
ii. The power series converges absolutely for all t ∈ R if R = ∞.
iii. The power series converges absolutely when |t − t0 | < R and diverges when
|t − t0 | > R if 0 < R < ∞.
If R = 0 then I is the degenerate interval [t0 , t0 ] and if R = ∞ then I =
(−∞, ∞). If 0 < R < ∞ then I is the interval (t0 − R, t0 + R) and possibly
the endpoints, t0 − R and t0 + R, which one must check separately using other
tests of convergence.
Recall that absolute convergence means that ∞ n
n=0 |cn (t − t0 ) | con-
P
verges and implies the original series converges. One of the important advan-
tages absolute convergence gives us is that we can add up the terms in a series
in any order we please and still get the same result. For example, we can add
all the even terms and then the odd terms separately. Thus
∞
X X X
cn (t − t0 )n = cn (t − t0 )n + cn (t − t0 )n
n=0 n odd n even
X∞ ∞
X
= c2n+1 (t − t0 )2n+1 + c2n (t − t0 )2n .
n=0 n=0
 7.1 A Review of Power Series 263

Example 1. Find the interval of convergence of the power series

∞
X (t − 4)n
.
n=1
n2n

I Solution. The ratio test gives

n2n
 
 cn+1  n 1
 cn  = (n + 1)2n+1 = 2(n + 1) → 2
 

as n → ∞. The radius of convergence is 2. The interval of convergence has

4 as the center and thus the endpoints are 2 and 6. When t = 2 the power
P∞ n
series reduces to n=1 (−1)n , which is the alternating harmonic series and
known to converge. When t = 6 the power series reduces to ∞ 1
n=0 n , which is
P
the harmonic series and known to diverge. The interval of convergence is thus
I = [ 2, 6).

Example 2. Find the interval of convergence of the power series

∞
X (−1)n t2n
J0 (t) = .
n=0
22n (n!)2
P∞ (−1)n un
I Solution. Let u = t2 . We apply the ratio test to n=0 22n (n!)2 to get

22n (n!)2
 
 cn+1  1
 cn  22(n+1) ((n + 1)!)2 = 4(n + 1)2 → 0
 =

as n → ∞. It follows that R = ∞ and the series converges for all u. Hence

P∞ (−1)n t2n
n=0 22n (n!)2 converges for all t and I = (−∞, ∞). J

In each example the power series defines a function on its interval of con-
vergence. In Example 2, the function J0 (t) is known as the Bessel function
of order 0 and plays P∞an important role in many physical problems. More
generally, let f (t) = n=0 cn (t − t0 )n for all t ∈ I. Then f is a function on
the interval of convergence I and Equation (1) is its power series repre-
sentation. A simple example of a power series representation is a polynomial
defined on R. In this case the coefficients are all zero except for finitely many.
Other well known examples from calculus are:
264 7 Power Series Methods
∞ n
X t t2 t3
et = = 1 + t+ + + ··· (2)
n=0
n! 2 3!
∞
X (−1)n t2n t2 t4
cos t = = 1 − + − ··· (3)
n=0
(2n)! 2 4!
∞
X (−1)n t2n+1 t3 t5
sin t = = t− + − ··· (4)
n=0
(2n + 1)! 3! 5!
∞
1 X
= tn = 1 + t + t2 + · · · (5)
1 − t n=0
∞
X (−1)(n+1) (t − 1)n (t − 1)2 (t − 1)3
ln t = = (t − 1) − + − · · · (6)
n=1
n! 2 3

Equations 2, 3 and 4 are centered at 0 and have interval of convergence

(−∞, ∞). Equation 5, known as the geometric series, is centered at 0 and
has interval of convergence (−1, 1). Equation 6 is centered at 1 and has interval
of convergence (0, 2 ].

Index Shifting
Rb
In calculus, the variable x in a definite integral a f (x) dx is called a dummy
variable because the value of the integral is independent of x. Sometimes it
is convenientR to change the variable. For example, if we replace x by x − 1 in
2 1
the integral 1 x+1 dx we obtain
2 x−1=2 3
1 1 1
Z Z Z
dx = d(x − 1) = dx.
1 x+1 x−1=1 x−1+1 2 x

In like manner, the index n in a power series is referred to as a dummy variable

because the sum is independent of n. It is also sometimes convenient to make
a change of variable, which, for series is called an index shift. For example,
∞
in the series (n + 1)tn+1 we replace n by n − 1 to obtain
P
n=0

∞
X n−1=∞
X ∞
X
(n + 1)tn+1 = (n − 1 + 1)tn−1+1 = ntn .
n=0 n−1=0 n=1

The lower limit n = 0 is replaced by n − 1 = 0 or n = 1. The upper limit

n = ∞ is replace by n − 1 = ∞ or n = ∞. The terms (n + 1)tn+1 in the series
go to (n − 1 + 1)tn−1+1 = ntn .
When a power series is given is such a way that the index n in the sum is
the power of (t − t0 ) we say the power series is written in standard form.
∞ ∞
Thus ntn is in standard form while (n + 1)tn+1 is not.
P P
n=1 n=0
 7.1 A Review of Power Series 265
∞
tn−2
Example 3. Make an index shift so that the series is expressed as
P
n2
n=2
a series in standard form.
I Solution. We replace n by n + 2 and get
∞ n−2 n+2=∞
X tn+2−2 ∞
X t X tn
2
= 2
= .
n=2
n n+2=2
(n + 2) n=0
(n + 2)2

Differentiation and Integration of Power Series

If a function can be represented by a power series then we can compute its
derivative and integral by differentiating and integrating each term in the
power series as noted in the following theorem.
Theorem 4. Suppose
∞
X
f (t) = cn (t − t0 )n
n=0
is defined by a power series with radius of convergence R > 0. Then f is
differentiable and integrable on (t0 − R, t0 + R) and
∞
X
f 0 (t) = ncn (t − t0 )n−1 (7)
n=1

and ∞
(t − t0 )n+1
Z X
f (t) dt = cn + C. (8)
n=0
n+1
Furthermore, the radius of convergence for the power series representations of
f 0 and f are both R.
R

We note that the presence of the factor n in f 0 (t) allows us to write

∞
X
f 0 (t) = ncn (t − t0 )n−1
n=0

since the term at n = 0 is zero. This observation is occasionally used. Consider

the following examples.
1
Example 5. Find a power series representation for (1−t)2 in standard form.
1 1
I Solution. If f (t) = 1−t then f 0 (t) = (1−t) 2 . If follows from Theorem 4

that
∞ ∞ ∞
1 d X n X n−1 X
= t = nt = (n + 1)tn .
(1 − t)2 dt n=0 n=1 n=0
J
266 7 Power Series Methods

Example 6. Find the power series representation for ln(1 − t) in standard

form.
R 1
I Solution. For t ∈ (−1, 1), ln(1 − t) = − 1−t + C. Thus
∞ ∞ ∞ n
tn+1 t
Z X X X
ln(1 − t) = C − tn dt = C − =C− .
n=0 n=0
n + 1 n=1
n

Evaluating both side at t = 0 gives C = 0. It follows that

∞ n
X t
ln(1 − t) = − .
n=1
n

The algebra of power series

P∞ P∞
Suppose f (t) = n=0 an (t−t0 )n and g(t) = n=0 bn (t−t0 )n are power series
representation of f and g and converge on the interval (t0 − R, t0 + R) for
some R > 0. Then

f (t) = g(t) if and only if an = bn ,

for all n = 1, 2, 3, . . . . Let c ∈ R. Then the power series representation of f ±g,

cf , f g, and f /g are given by
∞
X
f (t) ± g(t) = (an ± bn )(t − t0 )n (9)
n=0
X∞
cf (t) = can (t − t0 )n (10)
n=0
X∞
f (t)g(t) = cn (t − t0 )n where cn = a0 bn + a1 bn−1 + · · · + an b0 (11)
n=0
∞
f (t) X
and = dn (t − t0 )n , g(a) 6= 0, (12)
g(t) n=0

where each dn is determined by the equation f (t) = g(t) ∞ n

n=0 dn (t − t0 ) . In
P
Equations 9, 10, and 11, the series converges on the interval (t0 − R, t0 + R).
For division of power series, Equation 12, the radius of convergence is positive
but may not be as large as R.

Example 7. Compute the power series representations of

et + e−t et − e−t
cosh t = and sinh t = .
2 2
 7.1 A Review of Power Series 267

I Solution. We write out the terms in each series, et and e−t , and get
t2 t3 t4
et = 1 + t + + + + ···
2! 3! 4!
t2 t3 t4
e−t = 1 − t+ − + − ···
2! 3! 4!
t2 t4
et + e−t = 2 + 2 + 2 + ···
2! 4!
t3 t5
et − e−t = 2t + 2 + 2 + · · ·
3! 5!
It follows that
∞ ∞
X t2n X t2n+1
cosh t = and sinh t = .
n=0
(2n)! n=0
(2n + 1)!

J
∞
Example 8. Let y(t) = cn tn . Compute
P
n=0

(1 + t2 )y 00 + 4ty 0 + 2y

as a power series.
I Solution. We differentiate y twice to get
∞
X ∞
X
0 n−1 00
y (t) = cn nt and y (t) = cn n(n − 1)tn−2 .
n=1 n=2

In the following calculations we shift indices as necessary to obtain series in

standard form.
∞
X ∞
X
t2 y 00 = cn n(n − 1)tn = cn n(n − 1)tn
n=2 n=0
X∞ X∞
y 00 = cn n(n − 1)tn−2 = cn+2 (n + 2)(n + 1)tn
n=2 n=0
X∞ ∞
X
4ty 0 = 4cn ntn = 4cn ntn
n=1 n=0
X∞
2y = 2cn tn
n=0

Notice that the presence of the factors n and n − 1 in the first series allows
us to write it with a starting point n = 0 instead of n = 2; similarly for the
third series. Adding these results and simplifying gives
268 7 Power Series Methods
∞
X
(1 + t2 )y 00 + 4ty 0 + 2y = ((cn+2 + cn )(n + 2)(n + 1)) tn .
n=0

A function f is said to be an odd function if f (−t) = −f (t) and even if

f (−t) = f (t). If f is odd and has a power series representation with center 0
then all coefficients of even powers of t are zero. Similarly, if f is even then
all the coefficients of odd powers are zero. Thus f has the following form:
∞
f (t) = a0 + a2 t2 + a4 t4 + . . . = a2n t2n f -even
P
n=0
∞
f (t) = a1 t + a3 t3 + a5 t5 + . . . = a2n+1 t2n+1 f -odd.
P
n=0

For example, the power series representations of cos t and cosh t reflect that
they are even and while those of sin t and sinh t reflect that they are odd
functions.

Example 9. Compute the first four nonzero terms in the power series rep-
resentation of
sinh t
tanh t = .
cosh t
I Solution. Division of power series is generally complicated. To make
things a little simpler here we observe that tanh t is an odd function. Thus
∞
its power series expansion is of the form tanh t = d2n+1 t2n+1 and satisfies
P
n=1
∞
2n+1
. By Example 7 this means
P
sinh t = cosh t d2n+1 t
n=0

t3 t5 t2 t4
   
d1 t + d3 t3 + d5 t5 + · · ·


t+ + + ··· = 1 + + + ···
3! 5! 2! 4!
 
d1 3 d3 d1 5
= d1 t + (d3 + )t + (d5 + + )t + · · · .
2! 2! 4!

We now equate coefficients to get the following sequence of equations:

d1 =1
d1 1
d3 + =
2! 3!
d3 d1 1
d5 + + =
2! 4! 5!
d5 d3 d1 1
d7 + + + =
2! 4! 6! 7!
..
.
 7.1 A Review of Power Series 269

Recursively solving these equations gives d1 = 1, d3 = −1 2

3 , d5 = 15 , and
−17
d7 = 315 . The first four nonzero terms in the power series expansion for
tanh t is thus
1 2 17 7
tanh t = t − t3 + t5 − t + ··· .
3 15 315
J

Identifying Power Series

Given a power series, with positive radius of convergence, it is sometimes
possible to identify it with a known function. When we can do this we will
say that it is written in closed form. Usually, such identification come by
using a combination of differentiation, integration, or the algebraic properties
of power series discussed above. Consider the following examples.

Example 10. Write the power series

∞ 2n+1
X t
n=0
n!

in closed form.

I Solution. Observe that we can factor out t and associate the term t2 to
get
∞ 2n+1 ∞
X t X (t2 )n 2
=t = tet ,
n=0
n! n=0
n!

from Equation 2. J

Example 11. Write the power series

∞
X
n(−1)n t2n
n=1

in closed form.
∞
I Solution. Let z(t) = n(−1)n t2n . Then dividing both sides by t gives
P
n=1

∞
z(t) X
= n(−1)n t2n−1 .
t n=1

Integration will now simplify the sum:

270 7 Power Series Methods
∞ Z
z(t)
Z X
dt = n(−1)n t2n−1 dt
t n=1
∞
X t2n
= (−1)n
n=1
2
∞
1 X
= (−t2 )n
2 n=1
 
1 1
= − 1 + c,
2 1 + t2
where the last line is obtained from the geometric series by adding and sub-
tracting the n = 0 term and c is a constant of integration. We now differentiate
this equation to get
 
z(t) 1 d 1
= −1
t 2 dt 1 + t2
−t
= .
(1 + t2 )2
It follows now that
−t2
z(t) = .
(1 + t2 )2
It is straightforward to check that the radius of convergence is 1 so we get the
equality
∞
−t2 X
= n(−1)n t2n ,
(1 + t2 )2 n=1

on the interval (−1, 1). J

Taylor Series
P∞
Suppose f (t) = n=0 cn (t − t0 )n , with positive radius of converge. Theorem
4 implies that the derivatives, f (n) , exist for all n = 0, 1, . . .. Furthermore, it
(n)
is easy to check that f (n) (t0 ) = n!cn and thus cn = f n!(t0 ) . Therefore, if f is
represented by a power series then it must be that
∞
X f (n) (t0 )
f (t) = (t − t0 )n . (13)
n=0
n!

This series is called the Taylor Series of f centered at t0 .

Now let’s suppose that f is a function on some domain D and we wish to
find a power series representation centered at t0 ∈ D. By what we have just
argued f will have to be given by its Taylor Series, which, of course, means
that that all higher order derivatives of f at t0 must exist. However, it can
 7.1 A Review of Power Series 271

happen that the Taylor series may not converge to f on any interval containing
t0 (See Problems 28–32 where such an example is considered). When Equation
13 is valid on an open interval containing t0 we call f analytic at t0 . The
properties of power series listed above shows that the sum, difference, scalar
multiple and product of analytic functions is again analytic. The quotient of
analytic functions is likewise analytic at points where the denominator is not
zero. Derivatives and integrals of analytic functions are again analytic.

Example 12. Verify that the Taylor series of sin t centered at 0 is that given
in Equation 4.

I Solution. The first four derivatives of sin t and their values at 0 are as
follows:
order n sin(n) (t) sin(n) (0)
n=0 sin t 0
n=1 cos t 1
n=2 − sin t 0
n=3 − cos t −1
n=4 sin t 0
The sequence 0, 1, 0, −1 thereafter repeats. Hence, the Taylor series of sin t is
∞
t2 t3 t4 t5 X t2n+1
0 + 1t + 0 − + 0 + 1 + ··· = ,
2! 3! 4! 5! n=0
(2n + 1)!

as in Equation 4. J

Given a function f , it can sometimes be difficult to compute the Taylor

series by computing f (n) (t0 ) for all n = 0, 1, . . .. For example, compute the
first few derivatives of tanh t, considered in Example 9, to see how complicated
the derivatives become. Additionally, determining whether the Taylor series
converges to f requires some additional information, for example, the Taylor
remainder theorem. We will not include this in our review. Rather we will stick
to examples where we derive new power series representations from existing
ones as we did in Examples 5 – 9.

Rational Functions
A rational function is the quotient of two polynomials and is analytic at
all points where the denominator is nonzero. Rational functions will arise in
many examples. It will be convenient to know what the radius of convergence
about a point t0 . The following theorem allows us to determine this without
going through the work of determining the power series. The proof is beyond
the scope of this text.
272 7 Power Series Methods

Theorem 13. Suppose p(t) q(t) is a quotient of two polynomials p and q. Suppose
q(t0 ) 6= 0. Then the power series expansion for pq about t0 has radius of con-
vergence equal to the closest distance from t0 to the roots (including complex
roots) of q.

Example 14. Find the radius of convergence for each rational function
about the given point.
t
1. 4−t about t0 = 1
1−t
2. 9−t2 about t0 = 2
t3
3. t2 +1 about t0 = 2

I Solution.
1. The only root of 4 − t is 4. Its distance to t0 = 1 is 3. The radius of
convergence is 3.
2. The roots of 9 − t2 are 3 and −3. Their distances to t0 = 2 is 1 and 5,
respectively. The radius of convergence is 1.
2
√
3. The roots of t√ + 1 are i and −i. Their distances√
to t0 = 2 are |2 − i| = 5
and |2 + i| = 5. The radius of convergence is 5.
J

Exercises
1–9. Compute the radius of conver- 
∞ (−1)n t2n+1
7.
gence for the given power series. n=0 (2n)!


∞ 
∞ nn tn
1. n2 (t − 2)n 8.
n=0 n=0 n!


∞ tn 
∞ n!tn
2. 9.
1 · 3 · 5 · · · (2n + 1)
n=1 n
n=0


∞ (t − 1)n 10–17. Find the Taylor series for each
3. function with center t0 = 0.
n=0 2n n!
1

∞ 3n (t − 3)n 10.
4. 1 + t2
n=0 n+1
1

∞ 11.
5. n!t n t−a
n=0
12. eat

∞ (−1) t n 2n
sin t
6. 13.
n=0 (2n + 1)! t
 7.1 A Review of Power Series 273

et − 1 and suppose L = limn→∞ n
|cn |. Define
14.
t R in the following way:
15. tan−1 t R = L1 if 0 < L < ∞
16. ln(1 + t ) 2 R=0 if L = ∞
R=∞ if L = 0.
17. Then
18–21. Find the first four nonzero i. The power series converges only at
terms in the Taylor series with center 0 t = t0 if R = 0.
for each function. ii. The power series converges for all
18. tan t t ∈ R if R = ∞.
19. sec t iii. The power series converges if
|t − t0 | < R and diverges if |t − t0 | >
Solving these equations gives R.
d0 = 1 26. Use the root test to determine the

∞
tn
1 radius of convergence of .
d2 = n=0
nn
2
5 27. Let cn = 1 if n is odd and cn = 2 if
d4 =
4!
69 is
n even. Consider the power series
n
n=0 cn t . Show that the ratio test
∞
d6 = .
6! does not apply. Use the root test to
Thus sec t = 1 + 12 t2 + 5 4
4!
t + 61 6
6!
t + determine the radius of convergence.
···. 28–32. In this sequence of exercises we
20. et sin t consider a function that is infinitely dif-
21. et cos t ferentiable but not analytic. Let

26–27. Find a closed form expression 0 if t ≤ 0

f (t) =  −1 .
for each power series. e t if t > 0
∞ n n+1 n
22. n=0 (−1) n!
t 28. Compute f 0 (t) and f 00 (t) and ob-
−1
serve that f (n) (t) = e t pn ( 1t ) where
23. pn is a polynomial, n = 1, 2. Find p1
24. Redo Exercises 20 and 21 in the fol- and p2 .
lowing way. Recall Euler’s formula
eit = cos t + i sin t and write et cos t 29. Use mathematical −1
induction to show
and et sin t as the real and imaginary that f (n) (t) = e t pn ( 1t ) where pn is
parts of et eit = e(1+i)t expanded as a polynomial.
a power series. 30. Show that lim f (n) (t) = 0. To do
t→0+
25. Use the power series (with center −1
this let u = 1t in f (n) (t) = e t pn ( 1t )
0) for the exponential function and and let u → ∞. Apply L’Hospitals
expand both sides of the equation rule.
eat ebt = e(a+b)t . What well-known
formula arises when the coefficients 31. Show that f n (0) = 0 for all n =
n
of tn! are equated? 0, 1, . . ..
26–27. A test similar to the ratio test 32. Conclude that f is not analytic at
is the root test. t = 0 though all derivatives at t = 0
The exist.
∞ Root Testn for Power Series Let
n=0 cn (t − t0 ) be a given power series
274 7 Power Series Methods

7.2 Power Series Solutions about an Ordinary Point

A point t0 is called an ordinary point of Ly = 0 if we can write the differ-
ential equation in the form

y 00 + a1 (t)y 0 + a0 (t)y = 0, (1)

where a0 (t) and a1 (t) are analytic at t0 . If t0 is not an ordinary point we call
it a singular point.

Example 1. Determine the ordinary and singular points for each of the
following differential equations.
1. y 00 + t21−9 y 0 + t+1
1
y=0
2. (1 − t )y − 2ty 0 + n(n + 1)y = 0, where n is an integer.
2 00

3. ty 00 + sin ty 0 + (et − 1)y = 0

I Solution.
1. Here a1 (t) = t21−9 is analytic except at t = ±3. The function a0 = t+1
1
is
analytic except at t = −1. Thus the singular points at −3, 3, and −1. All
other points are ordinary.
−2t
2. This is Legendre’s equation. In standard form we find a1 (t) = 1−t 2 and

a0 (t) = n(n+1)
1−t2 . They are analytic except at 1 and −1. These are the
singular points and all other points are ordinary.
t
3. In standard form a1 (t) = sint t and a0 (t) = e −1
t . Both of these are analytic
everywhere. (See Exercises 13 and 14 in Section 7.) It follows that all points
are ordinary.

In this section we restrict our attention to ordinary points. Their impor-

tance is underscored by the following theorem. It tells us that there is always
a power series solution about ordinary points.

Theorem 2. Suppose a0 (t) and a1 (t) are analytic at t0 , both of which con-
verge for |t − t0 | < R. Then there is a unique solution y(t), analytic at t0 , to
the initial value problem

y 00 + a1 (t)y 0 + a0 (t)y = 0, y(t0 ) = α, y 0 (t0 ) = β. (2)

If
∞
X
y(t) = cn (t − t0 )n
n=0

then c0 = α, c1 = β, and all other ck , k = 2, 3, . . ., are determined by c0 and

c1 . Furthermore, the power series for y converges for |t − t0 | < R.
 7.2 Power Series Solutions about an Ordinary Point 275

Of course the Uniqueness and Existence Theorem, Theorem 6 of Section

6.1, implies there is a unique solution. What is new here is that the solution is
analytic at t0 . Since the solution is necessarily unique it is not at all surprising
that the coefficients are determined by the initial conditions. The only hard
part about the proof, which we omit, is showing that the solution converges
for |t − t0 | < R. Let y1 be the solution with initial conditions y(t0 ) = 1 and
y 0 (t0 ) = 0 and y2 the solution with initial condition y(t0 ) = 0 and y 0 (t0 ) = 1.
Then it is easy to see that y1 and y2 are independent solutions and hence
all solutions are of the form c1 y1 + c2 y2 . (See the proof of Theorem 4 in
Section 6.2.) The power P∞series method refers to the use of this theorem
by substituting y(t) = n=0 cn (t − t0 )n into Equation 2 and determining the
coefficients.
We illustrate the use of Theorem 2 with a few examples. Let’s begin with
a familiar constant coefficient differential equation.
Example 3. Use the power series method to solve
y 00 + y = 0.
I Solution. Of course, we can solve this equation by the characteristic poly-
nomial s2 + 1 to get c1 sin t + c2 cos t. Let’s see how the power series method
gives the same answer. Since the coefficients are constant they are analytic
everywhere with infinite radius of convergence. Theorem P∞2 implies that the
power series solutions converge everywhere. Let y(t) = n=0 cn tn be a power
series about t0 = 0. Then
∞
X
y 0 (t) = cn ntn−1
n=1
X∞
and y 00 (t) = cn n(n − 1)tn−2 .
n=2
00
P∞ n
An index shift, n → n+2, gives y (t) = n=0 cn+2 (n+2)(n+1)t . Therefore,
the equation y 00 + y = 0 gives
∞
X
(cn + cn+2 (n + 2)(n + 1))tn = 0,
n=0

which implies cn + cn+2 (n + 2)(n + 1) = 0, or, equivalently,

−cn
cn+2 = for all n = 0, 1, . . .. (3)
(n + 2)(n + 1)
Equation 3 is an example of a recurrence relation: terms of the sequence
are determined by earlier terms. Since the difference in indices between cn and
cn+2 is 2 it follows that even terms are determined by previous even terms
and odd terms are determined by previous odd terms. Let’s consider these
two cases separately.
276 7 Power Series Methods

The even case The odd case

−c0 −c1
n=0 c2 = 2·1 n=1 c3 = 3·2
−c2 c0 c0 −c3 c1 c1
n=2 c4 = 4·3 = 4·3·2·1 = 4! n=3 c5 = 5·4 = 5·4·3·2 = 5!
−c4 −c0 −c5 −c1
n=4 c6 = 6·5 = 6! n=5 c7 = 7·6 = 7!
−c6 c0 −c7 c1
n=6 c8 = 8·7 = 8! n=7 c9 = 9·8 = 9!
.. .. .. ..
. . . .

More generally, we can see that Similarly, we see that

c0 c1
c2n = (−1)n . c2n+1 = (−1)n .
(2n)! (2n + 1)!

Now, as we mentioned in Section 7, we can change the order of absolutely

convergent
P∞ sequences without affecting the sum. Thus let’s rewrite y(t) =
n
n=0 cn t in terms of odd and even indices to get
∞
X ∞
X
y(t) = c2n t2n + c2n+1 t2n+1
n=0 n=0
∞ ∞
X (−1)n 2n X (−1)n 2n+1
= c0 t + c1 t .
n=0
(2n)! n=0
(2n + 1)!

The first power series in the last line is that of cos t and the second power
series is that of sin t (See Equations 3 and 4 in Section 7). J
Example 4. Use the power series method with center t0 = 0 to solve

(1 + t2 )y 00 + 4ty 0 + 2y = 0.

What is a lower bound on the radius of convergence?

I Solution. We write the given equation in standard form to get
4t 0 2
y 00 + y + y = 0.
1 + t2 1 + t2
4t 2
Since the coefficient functions a1 (t) = 1+t 2 and a2 (t) = 1+t2 are rational func-

tions with nonzero denominators they are analytic at all points. By Theorem
13 of Section 7, it’s not hard to see that they have power series expansions
about t0 = 0 with radius of convergence
P∞ 1. By Theorem 2 the radius of con-
n
vergence for a solution, y(t) = n=0 cn t is at least 1. To determine the
coefficients it is easier to substitute y(t) directly into (1 + t2 )y 00 + 4ty 0 + 2y = 0
instead of its equivalent standard form. The details were worked out in Ex-
ample 8 of Section 7. We thus obtain
∞
X
2 00 0
(1 + t )y + 4ty + 2y = ((cn+2 + cn )(n + 2)(n + 1)) tn = 0.
n=0
 7.2 Power Series Solutions about an Ordinary Point 277

From this equation we get cn+2 +cn = 0 for all n = 0, 1, · · · . Again we consider
even and odd cases.
278 7 Power Series Methods

The even case The odd case

n=0 c2 = −c0 n=1 c3 = −c1
n=2 c4 = −c2 = c0 n=3 c5 = −c3 = c1
n=4 c6 = −c0 n=5 c7 = −c1

More generally, we can see that Similarly, we see that

c2n = (−1)n c0 . c2n+1 = (−1)n c1 .

It follows now that

∞
X ∞
X
y(t) = c0 (−1)n t2n + c1 (−1)n t2n+1 .
n=0 n=0

As we observed earlier each of these series has radius of convergence at least

1. In fact, the radius of convergence of each is 1. J

A couple of observations are in order for this example. First, we can relate
the power series solutions to the geometric series, Equation 7.5, and write
them in closed form. Thus
∞ ∞
X
n 2n
X 1
(−1) t = (−t2 )n =
n=0 n=0
1 + t2
∞ ∞
X X t
(−1)n t2n+1 = t (−t2 )n = .
n=0 n=0
1 + t2

1 t
It follows now that the general solution is y(t) = c0 1+t 2 +c1 1+t2 . Second, since

a1 (t) and a0 (t) are continuous on R the Uniqueness and Existence Theorem,
Theorem 6 of Section 6.1, guarantees the existence of solutions defined on all
1 t
of R. It is easy to check that these closed forms, 1+t 2 and 1+t2 , are defined

on all of R and satisfy the given differential equation.

This example illustrates that there is some give and take between the
uniqueness and existence theorem, Theorem 6.1.6, and Theorem 2 above. On
the one hand, Theorem 6.1.6 may guarantee a solution, but it may be diffi-
cult or impossible to find without the power series method. The power series
method, Theorem 2, on the other hand, may only find a series solution on
the interval of convergence, which may be quite smaller than that guaranteed
by Theorem 6.1.6. Further analysis of the power series may reveal a closed
form solution valid on a larger interval as in the example above. However, it
is not always be possible to do this. Indeed, some recurrence relations can
be difficult to solve and we must be satisfied with writing out only a finite
number of terms in the power series solution.

Example 5. Discuss the radius of convergence of the power series solution

about t0 = 0 to
 7.2 Power Series Solutions about an Ordinary Point 279

(1 − t)y 00 + y = 0.
Write out the first five terms given the initial conditions

y(0) = 1 and y 0 (0) = 0.

I Solution. In standard form the differential equation is

1
y 00 + y = 0.
1−t
1
Thus a1 (t) = 0, a0 (t) = 1−t , and t0 = 0 is an ordinary point. Since a0 (t)
is represented by the geometric series, which has radius of convergence 1,
it follows that any solution will have radius of convergence at least 1. Let
∞
cn tn . Then y 00 and −ty 00 are given by
P
y(t) =
n=0

∞
X
y 00 (t) = cn+2 (n + 2)(n + 1)tn
n=0
X∞
−ty 00 (t) = −cn n(n − 1)tn−1
n=2
X∞
= −cn+1 (n + 1)ntn .
n=0

It follows that
∞
X
(1 − t)y 00 + y = (cn+2 (n + 2)(n + 1) − cn+1 (n + 1)n + cn )tn
n=0

which leads to the recurrence relations

cn+2 (n + 2)(n + 1) − cn+1 (n + 1)n + cn = 0,

for all n = 0, 1, 2, . . . . This recurrence relation is not easy to solve generally.

We can, however, compute any finite number of terms. First, we solve for
cn+2 :
n 1
cn+2 = cn+1 − cn . (4)
n+2 (n + 2)(n + 1)
The initial conditions y(0) = 1 and y 0 (0) = 0 imply that c0 = 1 and c1 = 0.
Recursively applying Equation 4 we get
n=0 c2 = c1 · 0 − c0 12 = − 12
n=1 c3 = c2 13 − c1 16 = − 16
n=2 c4 = c3 12 − c2 121 1
= − 24
n=3 c5 = c4 35 − c3 201 1
= − 60
280 7 Power Series Methods

It now follows that the first five terms of y(t) is

1 1 1 1
y(t) = 1 − t2 − t3 − t4 − t5 .
2 6 24 60
J

In general, it may not be possible to find a closed form description of cn .

Nevertheless, we can use the recurrence relation to find as many terms as we
desire. Although this may be tedious it may suffice to give an approximate
solution to a given differential equation.
We note that the examples we gave are power series solutions about t0 = 0.
We can always reduce to this case by a substitution. To illustrate consider the
differential equation

ty 00 − (t − 1)y 0 − ty = 0. (5)

It has t0 = 1 as an ordinary point. Suppose we wish to derive a power series

solution about t0 = 1. Let y(t) be a solution and let Y (x) = y(x + 1). Then
Y 0 (x) = y 0 (x + 1) and Y 00 (x) = y 00 (x + 1). In the variable x, Equation 5
becomes (x + 1)Y 00 (x) − xY 0 (x) − (x + 1)Y (x) = 0 and x0 = 0 is an ordinary
∞
point. We solve Y (x) = cn xn as before. Now let x = t − 1. I.e. y(t) =
P
P∞ n=0
Y (t − 1) = n=0 cn (t − 1)n is the series solution to Equation 5 about t0 = 1.

Chebyshev Polynomials
We conclude this section with the following two related problems: For a non-
negative integer n expand cos nx and sin nx in terms of just cos x and sin x. It
is an easy exercise (see Exercises 13 and 14) to show that we can write cos nx
as a polynomial in cos x and we can write sin nx as a product of sin x and
a polynomial in cos x. More specifically, we will find polynomials Tn and Un
such that
cos nx = Tn (cos x)
(6)
sin(n + 1)x = sin x Un (cos x).

(The shift by 1 in the formula defining Un is intentional.) The polynomials

Tn and Un are called the Chebyshev polynomials of the first and second
kind, respectively. They each have degree n. For example, if n = 2 we have
cos 2x = cos2 x − sin2 x = 2 cos2 x − 1. Thus T2 (t) = 2t2 − 1; if t = cos x we
have
cos 2x = T2 (cos x).
Similarly, sin 2x = 2 sin x cos x. Thus U1 (t) = 2t and

sin 2x = sin xU1 (cos x).

 7.2 Power Series Solutions about an Ordinary Point 281

More generally, we can use the trigonometric summation formulas

sin(x + y) = sin x cos y + cos x sin y

cos(x + y) = cos x cos y − sin x sin y

and the basic identity sin2 x + cos2 x = 1 to expand

cos nx = cos((n − 1)x + x) = cos((n − 1)x) sin x − sin((n − 1)x) sin x.

Now expand cos((n − 1)x) and sin((n − 1)x) and continue inductively to the
point where all occurrences of cos kx and sin kx, k > 1, are removed. Whenever
sin2 x occurs replace it by 1 − cos2 x. In the table below we have done just that
for some small values of n. We include in the table the resulting Chebyshev
polynomials of the first kind, Tn .

n cos nx Tn (t)

0 cos 0x = 1 T0 (t) = 1

1 cos 1x = cos x T1 (t) = t

2 cos 2x = 2 cos2 x − 1 T2 (t) = 2t2 − 1

3 cos 3x = 4 cos3 x − 3 cos x T3 (t) = 4t3 − 3t

4 cos 4x = 8 cos4 x − 8 cos2 x + 1 T4 (t) = 8t4 − 8t2 + 1

In a similar way we expand sin(n + 1)x. The following table gives the
Chebyshev polynomials of the second kind, Un , for some small values of n.

n sin(n + 1)x Un (t)

0 sin 1x = sin x U0 (t) = 1

1 sin 2x = sin x(2 cos x) U1 (t) = 2t

2 sin 3x = sin x(4 cos2 x − 1) U2 (t) = 4t2 − 1

3 sin 4x = sin x(8 cos3 x − 4 cos x) U3 (t) = 8t3 − 4t

4 sin 5x = sin x(16 cos4 x − 12 cos2 x + 1) U4 (t) = 16t4 − 12t2 + 1

The method we used for computing the tables is not very efficient. We
will use the interplay between the defining equations, Equation 6, to derive
second order differential equations that will determine Tn and Un . This theme
of using the interplay between two related families of functions will come up
again is Section 7.5.
282 7 Power Series Methods

Let’s begin by differentiating the equations that define Tn and Un in Equa-

tion 6. For the first equation we get
d
LHS: cos nx = −n sin nx = −n sin xUn−1 (cos x)
dx
d
RHS: Tn (cos x) = Tn0 (cos x)(− sin x).
dx
Equating these results, simplifying, and substituting t = cos x gives

Tn0 (t) = nUn−1 (t). (7)

For the second equation we get

d
LHS: sin(n + 1)x = (n + 1) cos(n + 1)x = (n + 1)Tn+1 (cos x)
dx
d
RHS: sin xUn (cos x) = cos xUn (cos x) + sin xUn0 (cos x)(− sin x)
dx
= cos xUn (cos x) − (1 − cos2 x)Un0 (cos x).

It now follows that (n + 1)Tn+1 (t) = tUn (t) − (1 − t2 )Un0 (t). Replacing n by
n − 1 gives

nTn (t) = tUn−1 (t) − (1 − t2 )Un−1

0
(t). (8)

We now substitute Equation 7 and its derivative Tn00 (t) = nUn−1

0
(t) into Equa-
tion 8. After simplifying we get that Tn satisfies

(1 − t2 )Tn00 (t) − tTn0 + n2 Tn (t) = 0. (9)

By substituting Equation 7 into the derivative of Equation 8 and simplifying

we get that Un satisfies

(1 − t2 )Un00 (t) − 3tUn0 + n(n + 2)Un (t) = 0. (10)

The differential equations

(1 − t2 )y 00 (t) − ty 0 + α2 y(t) = 0 (11)

(1 − t )y 00 (t) − 3ty 0 + α(α + 2)y(t) = 0.
2
(12)

are known as Chebyshev’s differential equations. Each have t0 = ±1 as

singular points and t0 = 0 is an ordinary point. The Chebyshev polynomial
Tn is a polynomial solution to Equation 11 and Un is a polynomial solution
to Equation 12, when α = n.

Theorem 6. We have the following explicit formulas for Tn and Un .

 7.2 Power Series Solutions about an Ordinary Point 283
n
X (n + k − 1)! (2t)2k
T2n (t) = n(−1)n (−1)k
(n − k)! (2k)!
k=0
n
2n + 1 X (n + k)! (2t)2k+1
T2n+1 (t) = (−1)n (−1)k
2 (n − k)! (2k + 1)!
k=0

n
X (n + k)! (2t)2k
U2n (t) = (−1)n (−1)k
(n − k)! (2k)!
k=0
n
X (n + k + 1)! (2t)2k+1
U2n+1 (t) = (−1)n (−1)k
(n − k)! (2k + 1)!
k=0

Proof. Let’sPfirst consider Chebyshev’s first differential equation, for general α.

∞
Let y(t) = k=0 ck tk . Substituting y(t) into Equation ?? we get the following
relation for the coefficients:
−(α2 − k 2 )ck
ck+2 = .
(k + 2)(k + 1)
Let’s consider the even and odd cases.

The even case

2 2
c2 = − α 2·1
−0
c0
2 2 (α2 −22 )(α2 −02 )
c4 = − α 4·3
−2
c2 = 4! c0
2 2 (α2 −42 )(α2 −22 )(α2 −02 )
c6 = − α 6·5
−4
c4 = − 6! c0
..
.
More generally, we can see that
(α2 − (2k − 2)2 ) · · · (α2 − 02 )
c2k = (−1)k c0 .
(2n)!
2 2
By factoring

α each
expression α − (2j) that appears in the numerator into
2 α
2 2 + j 2 − j and rearranging terms we can write
α α



kα 2 + k − 1 ··· 2 − k + 1
c2k = (−1) 22k c0 .
2 (2k)!
The odd case
2 2
c3 = − (α 3·2
−1 )
c1
2 2
(α2 −32 )(α2 −12 )
c5 = − (α 5·4
−3 )
c3 = 5! c1
2 2
(α2 −52 )(α2 −32 )(α2 −12 )
c7 = − (α 7·6
−5 )
c5 = − 7! c1
..
.
284 7 Power Series Methods

Similarly, we see that

(α2 − (2k − 1)2 ) · · · (α2 − 12 )
c2k+1 = (−1)k c1 .
(2k + 1)!
By factoring each
expression α2
− (2j − 1)2 that appears in the numerator
into 22 α−1
2 + j α−1
2 − (j − 1) we get
α−1 α−1



k 2 + k ··· 2 − (k − 1)
c2k+1 = (−1) 22k c1 .
(2k + 1)!
Let ∞ α
+ k − 1 · · · α2 − k + 1



αX
y0 = (−1)k 2
(2t)2k
2 (2k)!
k=0
and
∞ α−1
+ k · · · α−1



1X 2 − (k − 1)
y1 = (−1)k 2
(2t)2k+1 .
2 (2k + 1)!
k=0
Then the general solution to Chebyshev’s first differential equation is
y = c0 y 0 + c1 y 1 .
It is clear that neither y0 nor y1 is a polynomial if α is not an integer.

The case α = 2n: In this case y0 is a polynomial while y1 is not. In fact,

for k > n the numerator in the sum for y0 is zero and hence
n
X (n + k − 1) · · · (n − k + 1)
y0 (t) = n (−1)k (2t)2k
(2k)!
k=0
n
X (n + k − 1)! (2t)2k
=n (−1)k .
(n − k)! (2k)!
k=0

It follows T2n (t) = c0 y0 (t), where c0 = T2n (0). To determine T2n (0) we eval-
uate the defining equation T2n (cos x) = cos 2nx at x = π2 to get T2n (0) =
cos nπ = (−1)n . The formula for T2n now follows.

The case α = 2n+1: In this case y1 is a polynomial while y0 is not. Further,

n
1X (n + k) · · · (n − k + 1)
y1 (t) = (−1)k (2t)2k+1
2 (2k + 1)!
k=0
n
1X (n + k)! (2t)2k+1
= (−1)k .
2 (n − k)! (2k + 1)!
k=0

It now follows that T2n+1 (t) = c1 y1 (t). There is no constant coefficient term
in y1 . However, y10 (0) = 1 is the coefficient of t in y1 . Differentiating the
defining equation T2n+1 (cos x) = cos((2n + 1)x) at x = π2 gives T2n+1 0
(0) =
n n
(2n + 1)(−1) . Let c1 = (2n + 1)(−1) . The formula for T2n+1 now follows.
The verification of the formulas for U2n and U2n+1 are left as exercises.
 7.2 Power Series Solutions about an Ordinary Point 285

Exercises

1–4. Use the power series method for any integer n.

about t0 = 0 to solve the given differ-
ential equation. Identify the power se- 13. Assume n is a nonnegative inte-
ries with known functions. Since each are ger. Use the binomial theorem on de
constant coefficient use the characteristic Moivre’s formula to show that cos nx
polynomial to solve and compare. is a polynomial in cos x.

1. y 00 − y = 0 14. Assume n is a nonnegative inte-

ger. Use the binomial theorem on de
2. y 00 − 2y 0 + y = 0
Moivre’s formula to show that sin nx
3. y 00 + k2 y = 0, where k ∈ R is a product of sin x and a polyno-
mial in cos x.
4. y 00 − 3y 0 + 2y = 0
15. Show that
5–11. Use the power series method
about t0 = 0 to solve each of the follow- (1 − t2 )Un (t) = tTn+1 (t) − Tn+2 (t).
ing differential equations. Write the solu-
tion in the form y(t) = c0 y0 (t) + c1 y1 (t),
where y0 (0) = 1, y00 (0) = 0 and y1 (0) = 16. Show that
0, y10 (t) = 1. Find y0 and y1 in closed
Un+1 (t) = tUn (t) + Tn+1 (t).
form where possible. (**) indicates that
both y0 and y1 can be written in closed
form and (*) indicates that one of them 17. Show that
can be written in closed form.
5. (**) (1 − t2 )y 00 + 2y = 0 Tn+1 (t) = 2tTn (t) − Tn−1 (t).

6. (1 − t2 )y 00 − 2ty 0 + 2y = 0
18. Show that
7. (t − 1)y 00 − ty 0 + y = 0
Un+1 (t) = 2tUn (t) − Un−1 (t).
8. (1 + t2 )y 00 − 2ty 0 + 2y = 0

9. (1 + t2 )y 00 − 4ty 0 + 6y = 0 19. Show that

10. (1 − t2 )y 00 − 6ty 0 − 4y = 0 1
Tn (t) = (Un (t) − Un−2 (t)) .
2
11.
20. Show that
12–21. Chebyshev Polynomials:
12. Use Euler’s formula to derive the fol- 1
Un+2 (t) = (Tn (t) − Tn+2 (t)) .
lowing identity known as de Moivre’s 2
formula:

(cos x + i sin x)n = cos nx + i sin nx, 21.

286 7 Power Series Methods

7.3 Orthogonal Functions

Let w be a positive continuous function defined on an interval [a, b]. We say a
family of functions {y1 , y2 , . . .} defined on [a, b] are orthogonal with respect
to w if Z b
(yi | yj ) = yi (t)yj (t)w(t) dt = 0, (1)
a
whenever i 6= j. The function w is called the weight function. The norm of
yi is given by s
Z b
kyi k = yi2 (t)w(t) dt.
a
If, in addition to orthogonality, we have kyi k = 1 then we call the family of
functions, {y1 , y2 , . . .}, orthonormal with respect of w.
Equation 1 is an example on a more general concept. Suppose F is a
linear space of functions. A real inner product, denoted ( | ), associates a
real number to each pair of functions f, g ∈ F, such that
1. (f | g) = (g | f )
2. (f1 + f2 | g) = (f1 | g) + (f2 | g)
3. (αf | g) = α (f | g)
4. (f | f ) ≥ 0 and (f | f ) = 0 if and only if f = 0.
It is not hard to see that Equation 1 defines an inner product. In fact, one
can relax the condition that w be positive on [a, b] and allow w(a) = w(b) =
and positive elsewhere. This assumption is made in some examples below.
Families of orthogonal functions play a very important role in many fields
of the engineering sciences. They have been used to solve classical problems in
conduction of heat, vibrations, and fluid flow, More recently they have been
used to solve problems is data compression, electronic storage of information,
and image reconstruction to name a few. In this section we will explore, by
example, several classical families of orthogonal functions that arise by solving
certain second order differential equations.

Sturm-Louiville Problems
A second order differential equation that can be written in the form
(ry 0 )0 + qy = λy,
where r and q are continuous functions, is called a Sturm-Louiville equation.

Fourier Trig Functions

Arguably, the most well known and important family of orthogonal functions
is that given by the solutions to
y 00 + n2 y = 0, [0, 2π]
 7.3 Orthogonal Functions 287

The Laguerre Polynomials

Let f and g be continuous functions on the interval [0, ∞). Define
Z ∞
(f | g) = f (t)g(t)e−t dt. (2)
0

Of course, we must assume that f and g are chosen in such a way that the
improper integral converges. For example, if f and g are of exponential type
with order less than 12 then an easy argument shows that the improper integral
in Equation 2 converges.
In of Section 6.4 we defined the Laguerre polynomials and established
important differential properties which we list here below for quick reference
as a Proposition.
Proposition 1.
1. The Laguerre polynomial, `n (t), is defined to be the polynomial solution
to
ty 00 + (1 − t)y 0 + ny = 0,
with initial condition y(0) = 1.
2. The Laguerre polynomial is given by
n   k
X
k n t
`n (t) = (−1)
k k!
k=0

3. The Laplace transform of the Laguerre polynomial is given by

(s − 1)n
L {`n (t)} (s) = .
sn+1
4. For the differential operators
E◦ = 2tD2 + (2 − 2t)D − 1
E+ = tD2 + (1 − 2t)D + (t − 1)
E− = tD2 + D.
we have
E◦ `n = −(2n + 1)`n
E+ `n = −(n + 1)`n+1
E− `n = −n`n−1 .
We now establish that the Laguerre polynomials form an orthonormal
family of functions with respect to the inner product given in Equation 2.
Lemma 2. Suppose f and g are of exponential type of order less than 12 . Then
lim tf (t)g(t)e−t = 0.
t→∞
288 7 Power Series Methods

Proof. There are numbers a and b, each less than 12 , such that the order of
f and g are a and b, respectively. Since a + b < 1 there is a c > 0 so that
a + b + c < 1. Let r = −1 + a + b + c. Then r < 0 and since |t| < Kect for K
sufficiently large we have
tf (t)g(t)e−t  < Kert .
 

The result follows. u

t

Let C denote the set of continuous functions f on [0, ∞) that have con-
tinuous second derivatives and f and f 0 are of exponential type of order less
than 1/2. Clearly polynomials are in C.

Theorem 3. Suppose f and g are in C. Then

(E− f | g) = (f | E+ g) .
(E◦ f | g) = (f | E◦ g)

Proof. We will establish the first formula leaving the second as an exercise.
Let f, g ∈ C. Then E− f = tf 00 + f 0 = (tf 0 )0 . Let u = g(t)e−t and dv = (tf 0 )0 .
Then du = (g 0 (t) − g(t))e−t and v = tf 0 . Integration by parts and Lemma 2
gives
Z ∞
(E− f | g) = (tf 0 (t))0 g(t)e−t dt
0
Z ∞
∞
= tf 0 (t)g(t)e−t 0 − tf 0 (t)(g 0 (t) − g(t))e−t dt
0
Z ∞
= tf 0 (t)(g(t) − g 0 (t))e−t dt
0

Let u = t(g(t) − g 0 (t))e−t and dv = f 0 . A straightforward calculation gives

u0 = −(tg 00 + (1 − 2t)g 0 + (t − 1)g)e−t = E+ g and v = f . Again, integration
by parts and Lemma 2 gives
Z ∞
(E− f | g) = tf 0 (t)(g(t) − g 0 (t))e−t dt
0
Z ∞
∞
= t(g(t) − g 0 (t))f (t)e−t 0 + f (t)E+ g(t)e−t dt
0
= (f | E+ g) .

u
t

Theorem 4. The collection of Laguerre polynomials, {`n , n = 0, 1, 2, . . .}, forms

an orthonormal set.

Proof. Let n and m be nonnegative integers. First suppose n 6= m. By Theo-

−1
rem 3 we have 2n+1 E◦ `n = `n and hence
 7.3 Orthogonal Functions 289

−1
(`n | `m ) = (E◦ `n | `m )
2n + 1
−1
= (`n | E◦ `m )
2n + 1
2m + 1
= (`n | `m ) .
2n + 1
Since n 6= m it follows that (`n | `m ) = 0.
−1
Now suppose n = m. By Proposition 1 we can write `n = n E+ `n−1 .
Then by Theorem 3 we have
−1
(`n | `n ) = (E+ `n−1 | `n )
n
−1
= (`n−1 | E− `n )
n
n
= (`n−1 | `n−1 ) = (`n−1 | `n−1 ) .
n
By repeating in this way we get

(`n | `n ) = (`0 | `0 )
Z ∞
∞
= e−t dt = −e−t 0 = 1.
0

u
t

Chebyshev Polynomials
Exercises

1–2.
1. F

2.
3–6. Laguerre Polynomials:
3. Verify the second formula in Theorem 3:

(E◦ f | g) = (f | E◦ g) ,

for f and g in C.

4. Show that Z ∞
e−t `n (at) dt = (1 − a)n .
0
290 7 Power Series Methods

5. Show that
∞  
n m
Z
−t
e `m (t)`n (at) dt = a (1 − a)n−m ,
0 m
if m ≤ n and 0 if m > n.

6.

7.4 Regular Singular Points and the Frobenius Method

Recall that the singular points of a differential equation

y 00 + a1 (t)y 0 + a0 (t)y = 0 (1)

are those points for which either a1 (t) or a0 (t) is not analytic. Generally, they
are few in number, but tend to be the most important and interesting. In this
section we will describe a modified power series method, called the Frobenius
Method, that can be applied to differential equations with certain kinds of
singular points.
We say that the point t0 is a regular singular point of Equation 1 if
1. t0 is a singular point and
2. A1 (t) = (t − t0 )a1 (t) and A0 (t) = (t − t0 )2 a0 (t) are analytic at t0 .
Note that by multiplying a1 (t) by t − t0 and a0 (t) by (t − t0 )2 we ‘restore’
the analyticity at t0 . In this sense a regular singularity at t0 is not too bad.
A singular point that is not regular is called irregular.

Example 1. Show t0 = 0 is a regular singular point for the differential

equation
t2 y 00 + t sin ty 0 − 2(t + 1)y = 0. (2)

I Solution. Let’s rewrite Equation 2 by dividing by 2t2 . We get

sin t 0 2(t + 1)
y 00 + y − .
t t2

While sint t is analytic a 0 (see Exercise 7.13) the coefficient function −2(t+1)
t2
is not. However, both t sint t = sin t and t2 −2(t+1)
t2 = −2(1 + t) are analytic at
t0 = 0. It follows that t0 = 0 is a regular singular point. J

In the case of a regular singular point we will rewrite Equation 1: multiply

both sides by (t − t0 )2 and note that

(t − t0 )2 a1 (t) = (t − t0 )A1 (t) and (t − t0 )2 a0 (t) = A0 (t).

 7.4 Regular Singular Points and the Frobenius Method 291

We then get
(t − t0 )2 y 00 + (t − t0 )A1 (t)y 0 + A0 (t)y = 0. (3)
We will refer to this equation as the standard form of the differential equa-
tion when t0 is a regular singular point. By making a change of variable, if
necessary, we can assume that t0 = 0. We will restrict our attention to this
case. Equation 3 then becomes

t2 y 00 + tA1 (t)y 0 + A0 (t)y = 0. (4)

When A1 (t) and A0 (t) are constants then Equation 4 is a Cauchy-Euler

equation. We would expect that any reasonable adjustment to the power series
method should able to handle this simplest case. Before we describe the ad-
justments let’s explore, by an example, what goes wrong when we apply the
power series method to a Cauchy-Euler equation. Consider the differential
equation
2t2 y 00 + 5ty 0 − 2y = 0. (5)
∞
Let y(t) = cn tn . Then
P
n=0

∞
X
2t2 y 00 = 2n(n − 1)cn tn
n=0
X∞
5ty 0 = 5ncn tn
n=0
X∞
−2y = −2cn tn .
n=0

Thus
∞
X ∞
X
2t2 y 00 + 5ty 0 − 2y = (2n(n − 1) + 5n − 2)cn tn = (2n − 1)(n + 2)cn tn .
n=0 n=0

Equation 5 now implies (2n − 1)(n + 2)cn = 0, and hence cn = 0, for all
n = 0, 1, . . .. The power series method has failed; it has only given us the
trivial solution. With a little forethought we could have seen the problem.
The indicial polynomial for Equation 5 is 2s2 +n5s − 2 =o (2s − 1)(s + 2). The
1
roots are 12 and −2. Thus, a fundamental set is t 2 , t−2 and neither of these
functions is analytic at t0 = 0. Our assumption that there was a power series
solution centered at 0 was wrong!
Any modification of the power series method must take into account that
solutions to differential equations about regular singular points can have frac-
tional or negative powers of t, as in the example above. It is thus natural to
consider solutions of the form
292 7 Power Series Methods
∞
X
y(t) = tr c n tn , (6)
n=0

where r is a constant to be determined. This is the starting point for the

Frobenius method. We may assume that c0 is nonzero for if c0 = 0 we could
factor out a power of t and incorporate it into r. Under this assumption, we
call Equation 6 a Frobenius series. Of course, if r is a nonnegative integer
then a Frobenius series is a power series.
Recall that the fundamental sets for Cauchy-Euler equations take the form

{tr1 , tr2 } , {tr , tr ln t} , and {tα cos β ln t, tα sin β ln t} .

(c.f. Section 6.3.) The power of t depends on the roots of the indicial polyno-
mial. For differential equations with regular singular points something similar
occurs. Suppose A1 (t) and A0 (t) have power series expansions about t0 = 0
given by

A1 (t) = a0 + a1 t + a2 t2 + · · · and A0 (t) = b0 + b1 t + b2 t2 + · · · .

The polynomial
q(s) = s(s − 1) + a0 s + b0 (7)
is called the indicial polynomial associated to Equation 4 and extends the
definition given in the Cauchy-Euler case. 1 Its roots are called the exponents
of singularity and, as in the Cauchy-Euler equations, indicate the power to
use in the Frobenius series. A Frobenius series that solves Equation 4 is called
a Frobenius series solution.

Theorem 2 (The Frobenius Method). Suppose t0 = 0 is a regular sin-

gular point of the differential equation

t2 y 00 + tA1 (t)y 0 + A0 (t)y = 0.

Suppose r1 and r2 are the exponents of singularity.

The real case: Assume r1 and r2 are real and r1 ≥ r2 . There are three cases
to consider:
1. If r1 − r2 is not an integer then there are two Frobenius solutions of the
form
∞
X ∞
X
y1 (t) = tr1 cn tn and y2 (t) = tr2 Cn tn .
n=0 n=0

2. If r1 −r2 is a positive integer then there is one Frobenius series solution

of the form

1
If the coefficient of t2 y 00 is a number c other than 1 we take the indicial poly-
nomial to be q(s) = cs(s − 1) + a0 s + b0 .
 7.4 Regular Singular Points and the Frobenius Method 293
∞
X
y1 (t) = tr1 c n tn
n=0

and a second independent solution of the form

∞
X
y2 (t) = y1 (t) ln t + tr2 Cn tn .
n=0

It can be arranged so that  is either 0 or 1. When  = 0 there are two

Frobenius series solution. When  = 1 then a second independent solution
is the sum of a Frobenius series and a logarithmic term. We refer to these
as the non logarithmic and logarithmic cases, respectively.
3. If r1 − r2 = 0 let r = r1 = r2 . Then there is one Frobenius series solution
of the form
∞
X
y1 (t) = tr c n tn
n=0

and a second independent solution of the form

∞
X
y2 (t) = y1 (t) ln t + tr Cn tn .
n=0

The second solution y2 is also referred to as a logarithmic case.

The complex case: If the roots of the indicial polynomial are distinct complex
numbers, r and r say, then there is a complex valued Frobenius series solution
of the form
∞
X
y(t) = tr c n tn ,
n=0

where the coefficients cn may be complex. The real and imaginary parts of
y(t), y1 (t) and y2 (t) respectively, are linearly independent solutions.
Each series given for all five different cases have a positive radius of con-
vergence.

Remark 3. You will notice that in each case there is at least one Frobenius
solution. When the roots are real there is a Frobenius solution for the larger
of the two roots. If y1 is a Frobenius solution and there is not a second Frobe-
nius solution then a second independent solution is the sum of a logarithmic
expression y1 (t) ln t and a Frobenius series. This fact is obtained by applying
reduction of order. We will not provide the proof as it is long and not very
enlightening. However, we will consider an example of each case mentioned
in the theorem. Read these examples carefully. They will reveal some of the
subtleties of involved in the general proof and, of course, are a guide through
the exercises.
294 7 Power Series Methods

Example 4 (Real roots not differing by an integer). Use

Theorem 2 to solve the following differential equation:

2t2 y 00 + 3t(1 + t)y 0 − y = 0. (8)

I Solution. We can identify A1 (t) = 3 + 3t and A0 (t) = −1. It is easy to

see that t0 = 0 is a regular singular point and the indicial equation

q(s) = 2s(s − 1) + 3s − 1 = (2s2 + s − 1) = (2s − 1)(s + 1).

The exponents of singularity are thus 12 and −1 and since their difference is
not an integer Theorem 2 tells us there are two Frobenius solutions: one for
each exponent of singularity. Before we specialize to each case we will first
derive the general recurrence relation from which the indicial equation falls
out. Let ∞ ∞
X X
y(t) = tr c n tn = cn tn+r .
n=0 n=0
Then
∞
X ∞
X
y 0 (t) = (n + r)cn tn+r−1 and y 00 (t) = (n + r)(n + r − 1)cn tn+r−2 .
n=0 n=0

It follows that
∞
X
2t2 y 00 (t) = tr 2(n + r)(n + r − 1)cn tn
n=0
X∞
3ty 0 (t) = tr 3(n + r)cn tn
n=0
X∞ ∞
X
3t2 y 0 (t) = tr 3(n + r)cn tn+1 = tr 5(n − 1 + r)cn−1 tn
n=0 n=1
X∞
−y(t) = tr −cn tn .
n=0

The sum of these four expressions is 2t2 y 00 + 3t(1 + t)y 0 − 1y = 0. Notice that
each term has tr as a factor. It follows that the sum of each corresponding
power series is 0. They are each written in standard form so the sum of the
coefficients with the same powers must likewise be 0. For n = 0 only the first,
second and fourth series contribute constant coefficients (t0 ) while for n ≥ 1
all four series contribute coefficients for tn . We thus get
n=0 (2r(r − 1) + 3r − 1)c0 = 0
n≥1 (2(n + r)(n + r − 1) + 3(n + r) − 1)cn + 3(n − 1 + r)cn−1 = 0.
Now observe that for n = 0 the coefficient of c0 is the indicial polynomial
q(r) = 2r(r − 1) + 3r − 1 = (2r − 1)(r + 1) and for n ≥ 1 the coefficient of cn is
 7.4 Regular Singular Points and the Frobenius Method 295

q(n + r). This will happen routinely. We can therefore rewrite these equations
in the form
n=0 q(r)c0 = 0
(9)
n≥1 q(n + r)cn + 3(n − 1 + r)cn−1 = 0.

Since a Frobenius series has a nonzero constant term it follows that q(r) =
0 implies r = 12 and r = −1, the exponents of singularity derived in the
beginning. Let’s now specialize to these cases individually. We start with the
larger of the two.
The case r = 12 . Let r = 21 in the recurrence relation given in Equation 9.
Observe that q(n + 12 ) = 2n(n + 32 ) = n(2n + 3) and is nonzero for all positive
n since the only roots are 12 and −1. We can therefore solve for cn in the
recurrence relation and get

−3(n − 21 )
 
−3 (2n − 1)
cn = cn−1 = cn−1 . (10)
n(2n + 5) 2 n(2n + 5)
Recursively applying Equation 10 we get

c1 = −3

1 −3

3
n=1 2 5 c0 = 2 c0
−3

3
1·(5·3)3
−3 2
n=2 c2 = 2 2·7 c1 = 2 c0
−3

5
2·(7·5)
−3 3 5·3 −3 3

3
n=3 c3 = 2 3·9 c2 = 2 c0 = (3!)(9·7) c0
−3

7
(3·2)(9·7·5)
−3 4 3
2
n=4 c4 = 2 4·11 c2 = 2 (4!)(11·9) c0

n
Generally, we have cn = −3 2
3
n!(2n+3)(2n+1) c0 . We let c0 = 1 and substitute
cn into the Frobenius series to get
∞  n
1
X −3 3
y1 (t) = t 2 tn .
n=0
2 n!(2n + 3)(2n + 1)

The case r = −1. In this case q(n + r) = q(n − 1) = (2n − 3)(n) is

again nonzero for all positive integers n. The recurrence relation in Equation
9 simplifies to
n−2
cn = −3 . (11)
(2n − 3)(n)
Recursively applying Equation 11 we get
−1
n=1 c1 = −3 −1·1 c0 = −3c0
n=2 c2 = 0c1 = 0
n=3 c3 = 0
.. ..
. .

Again, we let c0 = 1 and substitute cn into the Frobenius series to get

296 7 Power Series Methods

1 − 3t
y2 (t) = t−1 (1 − 3t) = .
t
Since y1 and y2 are linearly independent the solutions to Equation 8 is the set
of all linear combinations. J

Before proceeding to our next example let’s make a couple of observations

that will apply in general. It is not an accident that the coefficient of c0 is the
indicial polynomial q. This will happen in general and since we assumed from
the outset that c0 6= 0 it follows that if a Frobenius series solution exists then
q(r) = 0; i.e. r must be an exponent of singularity. It is also not an accident
that the coefficient of cn is q(n+r) in the recurrence relation. This will happen
in general as well. If we can guarantee that q(n + r) is not zero for all positive
integers n then we obtain a consistent recurrence relation, i.e. we can solve for
cn to obtain a Frobenius series solution. This will always happen for r1 , the
larger of the two roots. For the smaller root r2 we need to be more careful. In
fact, in the previous example we were careful to point out that q(n + r2 ) 6= 0
for n > 0. However, if the roots differ by an integer then the consistency of
the recurrence relation comes into question in the case of the smaller root.
The next two examples consider this situation.

Example 5 (Real roots differing by an integer: the non

logarithmic case). Use Theorem 2 to solve the following differential
equation:
ty 00 + 2y 0 + ty = 0.

I Solution. We first multiply both sides by t to put in standard form. We

get
t2 y 00 + 2ty 0 + t2 y = 0 (12)
and it is easy to verify that t0 = 0 is a regular singular point. The indicial
polynomial is q(s) = s(s − 1) + 2s = s2 + s = s(s + 1). It follows that 0 and
−1 are the exponents of singularity. They differ by an integer so there may or
may not be a second Frobenius solution. Let
∞
X
y(t) = tr c n tn .
n=0

Then
∞
X
t2 y 00 (t) = tr (n + r)(n + r − 1)cn tn
n=0
X∞
2ty 0 (t) = tr 2(n + r)cn tn
n=0
X∞
t2 y(t) = tr cn−2 tn
n=2
 7.4 Regular Singular Points and the Frobenius Method 297

The sum of the left side of each of these equations is t2 y 00 + 2ty 0 + t2 y = 0

and, therefore, the sum of the series is zero. We separate the n = 0, n = 1,
and n ≥ 2 cases and simplify to get

n=0 (r(r + 1))c0 = 0

n=1 ((r + 1)(r + 2))c1 = 0 (13)
n≥2 ((n + r)(n + r + 1))cn + cn−2 = 0,

The n = 0 case tells us that r = 0 or r = −1, the exponents of singularity.

The case r = 0. If r = 0 is the larger of the two roots then the n = 1 case
in Equation 13 implies c1 = 0. Also q(n + r) = q(n) = n(n + 1) is nonzero for
all positive integers n. The recurrence relation simplifies to
−cn−2
cn = .
(n + 1)n

Since the difference in indices in the recurrence relation is 2 it follows that all
the odd terms, c2n+1 , are zero. For the even terms we get
−c0
n=2 c2 = 3·2
−c2 c0
n=4 c4 = 5·4 = 5!
−c4 −c0
n=6 c6 = 7·6 = 7! ,

(−1)n
and generally, c2n = (2n+1)! c0 . If we choose c0 = 1 then

∞
X (−1)n t2n
y1 (t) =
n=0
(2n + 1)!

is a Frobenius solution (with exponent of singularity 0).

The case r = −1. In this case we see something different in the recurrence
relation. For if n = 1 in Equation 13 then we get the equation

0 · c1 = 0.

This equation is satisfied for all c1 . There is no restriction on c1 so we will

choose c1 = 0, as this is most convenient. The recurrence relation becomes
−cn−2
cn = , n ≥ 2.
(n − 1)n

A calculation similar to what we did above gives all the odd terms c2n+1 zero
and
(−1)n
c2n = c0 .
(2n)!
If we set c0 = 1 we find the Frobenius solution with exponent of singularity
−1
298 7 Power Series Methods
∞
X (−1)n t2n−1
y2 (t) = .
n=0
(2n)!

It is easy to verify that y1 (t) = sint t and y2 (t) = cost t . Since y1 and y2 are
linearly independent the solutions to Equation 12 is the set of all linear com-
binations of y1 and y2 . J

The main difference that we saw in the previous example from that of the
first example was in the case of the smaller root r = −1. We had q(n − 1) = 0
when n = 1 and this lead to the equation c1 · 0 = 0. We were fortunate in that
any c1 is a solution and choosing c1 = 0 lead to a second Frobenius solution.
The recurrence relation remained consistent. In the next example we will not
be so fortunate. (If c1 were chosen to be a fixed nonzero number then the odd
terms would add up to a multiple of y1 ; nothing is gained.)

Example 6 (Real roots differing by an integer: the loga-

rithmic case). Use Theorem 2 to solve the following differential equation:

t2 y 00 − t2 y 0 − (3t + 2)y = 0. (14)

I Solution. It is easy to verify that t0 = 0 is a regular singular point. The

indicial polynomial is q(s) = s(s− 1)− 2 = s2 − s− 2 = (s− 2)(s+ 1). It follows
that 2 and −1 are the exponents of singularity. They differ by an integer so
there may or may not be a second Frobenius solution. Let
∞
X
y(t) = tr c n tn .
n=0

Then
∞
X
t2 y 00 (t) = tr (n + r)(n + r − 1)cn tn
n=0
X∞
−t2 y 0 (t) = tr (n − 1 + r)cn−1 tn−1
n=1
X∞
−3ty(t) = tr −3cn−1 tn
n=1
X∞
−2y(t) = tr −2cn tn
n=0

As in the previous examples the sum of the series is zero. We separate the
n = 0 and n ≥ 1 cases and simplify to get

n=0 (r − 2)(r + 1)c0 = 0

(15)
n≥1 ((n + r − 2)(n + r + 1))cn + (n + r + 2)cn−1 = 0,
 7.4 Regular Singular Points and the Frobenius Method 299

The n = 0 case tells us that r = 2 or r = −1, the exponents of singularity.

The case r = 2. If r = 2 is the larger of the two roots then the n ≥ 1 case in
n+4
Equation 15 implies cn = n(n+3) cn−1 . We then get
4
n=1 c1 = 1·4 c0
6 6
n=2 c2 = 2·5 c1 = 2!·4 c0
7 8
n=3 c3 = 3·6 c2 = 3!·4 c0 ,
n+4
and generally, cn = n!·4 c0 . If we choose c0 = 4 then
∞ ∞
X n + 4 n X n + 4 n+2
y1 (t) = t2 t = t (16)
n=0
n! n=0
n!

is a Frobenius series solution; the exponent of singularity is 2. (It is easy to

see that y1 (t) = (t3 + 4t2 )et but we will not use this fact.)
The case r = −1. The recurrence relation in Equation 15 simplifies to
n(n − 3)cn = −(n + 1)cn−1 .
In this case there is a problem when n = 3. Observe
n=1 −2c1 = −2c0 hence c1 = c0
n=2 −2c2 = −3c1 hence c2 = 32 c0
n=3 0 · c3 = −4c2 = −6c0 , ⇒⇐
In the n = 3 case there is no solution since c0 6= 0. The recurrence relation
is inconsistent and there is no second Frobenius series solution. However,
Theorem 2 tells us there is a second independent solution of the form
∞
X
y(t) = y1 (t) ln t + t−1 c n tn . (17)
n=0

Although the calculations that follow are straightforward they are more in-
volved than in the previous examples. The idea is simple though: substitute
Equation 17 into Equation 14 and solve for the coefficients cn , n = 0, 1, 2, . . ..
If y(t) is as in Equation 17 then a calculation gives
∞
X
t2 y 00 = t2 y100 ln t + 2ty10 − y1 + t−1 (n − 1)(n − 2)cn tn
n=0
∞
X
−t2 y 0 = −t2 y10 ln t − ty1 + t−1 −(n − 2)cn−1 tn
n=1
∞
X
−3ty = −3ty1 ln t + t−1 cn−1 tn
n=1
∞
X
−2y = −2y1 ln t + t−1 −2cn tn .
n=0
300 7 Power Series Methods

The sum of the terms on the left is zero since we are assuming a y is a solution.
The sum of the terms with ln t as a factor is also zero since y1 is the solution,
Equation 16, we found in the case r = 2. Observe also that the n = 0 term
only occurs in the first and fourth series. In the first series the coefficient is
(−1)(−2)c0 = 2c0 and in the fourth series the coefficient is (−2)c0 = −2c0 .
We can thus start all the series at n = 1 since the n = 0 terms cancel. Adding
these terms together and simplifying gives

0 = 2ty10 − (t + 1)y1
∞
(18)
+ t−1 (n(n − 3))cn − (n + 1)cn−1 ) tn .
P
n=1

Now let’s calculate the power series for 2ty10 − (t + 1)y1 and factor t−1 out of
the sum. A short calculation and some index shifting gives
∞ ∞
X 2(n + 4)(n + 1) n+2 X 2(n + 3)(n + 1) n
2ty10 = t = t−1 t
n=0
n! n=1
(n − 1)!
∞ ∞
X n + 4 n+2 X n+3 n
−y1 = − t = t−1 − t
n=0
n! n=1
(n − 1)!
∞ ∞
X n + 4 n+3 X (n + 2)(n − 1) n
−ty1 = − t = t−1 − t .
n=0
n! n=1
(n − 1)!

Adding these three series and simplifying gives

∞
X (n + 1)(n + 5) n
2ty10 − (t + 1)y1 = t−1 t .
n=1
(n − 1)!

We now substitute this calculation into Equation 18, cancel out the common
factor t−1 , and get
∞ ∞
X (n + 1)(n + 5) n X
t + (n(n − 3)cn − (n + 1)cn−1 )tn = 0
n=1
(n − 1)! n=1

from which we derive the recurrence relation

(n + 1)(n + 5)
+ n(n − 3)cn − (n + 1)cn−1 = 0, n ≥ 1.
(n − 1)!

We can rewrite this as

(n + 1)(n + 5)
n(n − 3)cn = (n + 1)cn−1 − n ≥ 1. (19)
(n − 1)!

You should notice that the coefficient of cn is zero when n = 3. As observed

earlier this lead to an inconsistency of the recurrence relation for a Frobenius
 7.4 Regular Singular Points and the Frobenius Method 301

series. However, the additional term (n+1)(n+5)

(n−1)! that comes from the logarith-
mic term results in consistency but only for a specific value of c0 . To see this
lets consider the first few terms:
n=1 −2c1 = 2c0 − 2 · 6
c1 = −c0 + 6
n=2 −2c2 = 3c1 − 3 · 7
c2 = −3
2 (−c0 + 6) +
21
2 = 23 c0 + 3
2
4·8
n=3 0c3 = 4c2 − 2 = 6c0 − 10.

The only way that the third equation can be consistent is for c0 = 10
6 in which
case the equation 0c3 = 0 implies that c3 can take on any value we so choose.
We will let c3 = 0. For n ≥ 4 we can write Equation 19 as
(n + 1) (n + 1)(n + 5)
cn = cn−1 − n ≥ 4. (20)
n(n − 3) (n − 3)(n)!
Such recurrence relations are generally very difficult to solve in a closed form.
However, we can always solve any finite number of terms. In fact, the following
terms are easy to check:
10 5 −15
n=0 c0 = 6 = 3 n=4 c4 = 8
13 −11
n=1 c1 = 3 n=5 c5 = 8
−77
n=2 c2 = 4 n=6 c6 = 135
−1193
n=3 c3 = 0 n=7 c7 = 7560 .
We substitute these values into Equation 17 to obtain (an approximation to)
a second linearly independent solution
5 13 15 11 77 5 1193 6
y2 (t) = y1 (t) ln t + t−1 + + 4t − t3 − t4 − t − t + ··· .
3 3 8 8 135 7560
J
A couple of remarks are in order. By far the logarithmic cases are the most
tedious and long winded. In the case just considered the difference in the roots
is 2 − (−1) = 3 and it was precisely at n = 3 in the recurrence relation where
c0 is determined in order to achieve consistency. After n = 3 the coefficients
are nonzero so the recurrence relation is consistent. In general, it is at the
difference in the roots where this junction occurs. If we choose c3 to be a
nonzero fixed constant the terms that would arise with c3 as a coefficient
would be a multiple of y1 and thus nothing is gained. Choosing c3 = 0 does
not exclude any critical part of the solution.
Example 7 (Real roots: Coincident). Use Theorem 2 to solve the
following differential equation:
t2 y 00 − t(t + 3)y 0 + 4y = 0. (21)
302 7 Power Series Methods

I Solution. It is easy to verify that t0 = 0 is a regular singular point. The

indicial polynomial is q(s) = s(s−1)−3s+4 = s2 −4s+4 = (s−2)2 . It follows
that 2 is a root with multiplicity two. Hence, r = 2 is the only exponent of
singularity. There will be only one Frobenius series solution. A second solution
will involve a logarithmic term. Let
∞
X
y(t) = tr c n tn .
n=0

Then
∞
X
t2 y 00 (t) = tr (n + r)(n + r − 1)cn tn
n=0
X∞
−t2 y 0 (t) = tr (n + r − 1)cn−1 tn
n=1
X∞
−3ty 0 (t) = tr −3cn (n + r)tn
n=0
X∞
4y(t) = tr 4cn tn
n=0

As in previous examples the sum of the series is zero. We separate the n = 0

and n ≥ 1 cases and simplify to get

n=0 (r − 2)2 c0 = 0
(22)
n≥1 (n + r − 2)2 cn = (n + r − 1)cn−1 .

The case r = 2. Equation 22 implies r = 2 and

n+1
cn = cn−1 n ≥ 1.
n2
We then get
2
n=1 c1 = 12 c0
3 3·2
n=2 c2 = 22 c1 = 22 ·12 c0
4 4·3·2
n=3 c3 = 32 c2 = 32 ·22 ·12 c0 ,
and generally,
(n + 1)! n+1
cn = c0 = c0 .
(n!)2 n!
If we choose c0 = 1 then
∞
X n+1 n
2
y1 (t) = t t (23)
n=0
n!
 7.4 Regular Singular Points and the Frobenius Method 303

is a Frobenius series solution (with exponent of singularity 2). This is the only
Frobenius series solution.
A second independent solution takes the form
∞
X
y(t) = y1 (t) ln t + t2 c n tn . (24)
n=0

The ideas and calculations are very similar to the previous example. A
straightforward calculation gives
∞
X
t2 y 00 = t2 y100 ln t + 2ty10 − y1 + t2 (n + 2)(n + 1)cn tn
n=0
∞
X
−t2 y 0 = −t2 y10 ln t − ty1 + t2 −(n + 1)cn−1 tn
n=1
X∞
−3ty 0 = −3ty10 ln t − 3y1 + t2 −3(n + 2)cn tn
n=0
∞
X
4y = 4y1 ln t + t2 4cn tn .
n=0

The sum of the terms on the left is zero since we are assuming a y is a solution.
The sum of the terms with ln t as a factor is also zero since y1 is a solution.
Observe also that the n = 0 terms occur in the first, third, and fourth series.
In the first series the coefficient is 2c0 , in the third series the coefficient is
−6c0 , and in the fourth series the coefficient is 4c0 . We can thus start all the
series at n = 1 since the n = 0 terms cancel. Adding these terms together and
simplifying gives
∞
0 = 2ty10 − (t + 4)y1 + t2 n2 cn − (n + 1)cn−1 tn .


(25)
P
n=1

Now let’s calculate the power series for 2ty10 − (t + 4)y1 and factor t2 out of
the sum. A short calculation and some index shifting gives
∞
X n+2 n
2ty10 − (t + 4)y1 = t2 t . (26)
n=1
(n − 1)!

We now substitute this calculation into Equation 25, cancel out the common
factor t2 , and equate coefficients to get
n+2
+ n2 cn − (n + 1)cn−1 = 0.
(n − 1)!
Since n ≥ 1 we can solve for cn to get the following recurrence relation
n+1 n+2
cn = cn−1 − , n ≥ 1.
n2 n(n!)
304 7 Power Series Methods

As in the previous example such recurrence relations are difficult (but not
impossible ) to solve in a closed form. There is no restriction on c0 so we may
assume it is zero. The first few terms thereafter are as follows:
−173
n=0 c0 = 0 n=4 c4 = 288
−187
n=1 c1 = −3 n=5 c5 = 1200
−13 −463
n=2 c2 = 4 n=6 c6 = 14400
−31 −971
n=3 c3 = 18 n=7 c7 = 176400 .

We substitute these values into Equation 24 to obtain (an approximation to)

a second linearly independent solution
13 4 31 5 173 6 187 7 463 8
y2 (t) = y1 (t) ln t − (3t3 + t + t + t + t + t + · · · ).
4 18 288 1200 14400
J

Since the roots in this example are coincident their difference is 0. The juncture
mentioned in the example that proceeded this one thus occurs at n = 0 and
so we can make the choice c0 = 0. If c0 chosen to be nonzero then y2 will
include an extra term c0 y1 . Thus nothing is gained.

Example 8 (Complex roots). Use Theorem 2 to solve the following

differential equation:

t2 (t + 1)y 00 + ty 0 + (t + 1)3 y = 0. (27)

I Solution. In standard form this equation becomes

1 0
t2 y 00 + t y + (t + 1)2 y = 0.
t+1
From this it is easy to see that t0 = 0 is a regular singular point. However, we
will substitute
X∞
y(t) = tr c n tn (28)
n=0
1
into Equation 27 to avoid using the infinite power series expansion for t+1 =
2 3 2
1 − t + t − t + · · · . The indicial polynomial is q(s) = s(s − 1) + s + 1 = s + 1
and has roots r = ±i. Thus there is a complex valued Frobenius solution.
Theorem 4.2.4 tells us that in the real constant coefficient case the real and
imaginary parts of a complex valued solution are likewise solutions. It is not
hard to modify the proof so that it applies in the case where the coefficient
functions of a linear differential equation are real valued functions. Thus the
real and imaginary parts of the above Frobenius solution will be two linear
independent solutions to Equation 27. A straightforward substitution gives
 7.4 Regular Singular Points and the Frobenius Method 305
∞
X
t3 y 00 = tr (n + r − 1)(n + r − 2)cn−1 tn
n=1
X∞
t2 y 00 = tr (n + r)(n + r − 1)cn tn
n=0
X∞
ty 0 = tr (n + r)cn tn
n=0
X∞
y = tr c n tn
n=0
X∞
3ty = tr 3cn−1 tn
n=1
X∞
3t2 y = tr 3cn−2 tn
n=2
X∞
t3 y = tr cn−3 tn
n=3

As usual the sum of these series is zero. Since the index of the sums have
different starting points we separate the cases n = 0, n = 1, n = 2, and n ≥ 3
to get, after some simplification, the following
n=0 (r2 + 1)c0 = 0
n=1 ((r + 1)2 + 1)c1 + (r(r − 1) + 3)c0 = 0
n=2 ((r + 2)2 + 1)c2 + ((r + 1)r + 3)c1 + 3c0 = 0
n≥3 ((r + n)2 + 1)cn + ((r + n − 1)(r + n − 2) + 3)cn−1
+ 3cn−2 + cn−3 = 0
The n = 0 case implies that r = ±i. We will let r = i (the r = −i case will
give equivalent results). As usual c0 is arbitrary but nonzero. For simplicity
let’s fix c0 = 1. Substituting these values into the cases, n = 1 and n = 2
above, gives c1 = i and c2 = −12 . The general recursion relation is

((i + n)2 + 1)cn + ((i + n − 1)(i + n − 2) + 3)cn−1 + 3cn−2 + cn−3 = 0 (29)

from which we see that cn is determined as long as we know the previous
three terms. Since c0 , c1 , and c2 are known it follows that can determine all
cn ’s. Although Equation 29 is somewhat tedious to work with straightforward
calculations give the following values
−i i3
n=0 c0 = 1 n=3 c3 = 6 = 3!
1 i4
n=1 c1 = i n=4 c4 = 24 = 4!
−1 i2 i i5
n=2 c2 = 2 = 2! n=5 c5 = 120 = 5!
306 7 Power Series Methods

We will leave it as an exercise to verify by mathematical induction that cn =

in
n! . It follows now that
∞ n
X i n
y(t) = ti t = ti eit
n=0
n!

We assumed t > 0. Therefore we can write ti = ei ln t and

y(t) = ei(t+ln t) .
By Euler’s formula the real and imaginary parts are
y1 (t) = cos(t + ln t) and y2 (t) = sin(t + ln t).
It is easy to see that these functions are linearly independent solutions. We
remark that the r = −i case gives the solution y(t) = e−i(t+ln t) . Its real and
imaginary parts are, up to sign, the same as y1 and y2 given above. J

Exercises
1–5. For each problem determine the singular points. Classify them as regular
or irregular.
t 1
1. y 00 + 1−t2 y
0
+ 1+t y =0

t
Solution: The function 1−t 2 is analytic except at t = 1 and t = −1. The
1
function 1+t is analytic except at t = −1. It follows that t = 1 and t = −1
 
t −t
are the only singular points. Observe that (t − 1) 1−t 2 = 1+t is analytic
 
1
at 1 and (t − 1)2 1+t is analytic at t = 1. It follows that 1 is a regular
 
t t
singular point. Also observe that (t + 1) 1−t 2 = 1−t is analytic at −1
 
1
and (t + 1)2 1+t = (1 + t) is analytic at t = −1. It follows that −1 is a
regular singular point. Thus 1 and −1 are regular points.
1−t 0 1−cos t
2. y 00 + t y + t3 y =0

Solution: The functions 1−tt and

1−cos t
t3 are analytic except
at t = 0. So
t = 0 is the only singular point.
Observe that (t − 0) 1−t t = 1 − t is
analytic at 0 and (t − 0)2 1−cos
t3
t
= 1−cos t
t is analytic at t = 0. It follows
that 0 is a regular singular point.
1−et
3. y 00 + 3t(1 − t)y 0 + t y =0

t
Solution: Both 3t(1−t) and 1−e
t are analytic. There are no singular points
and hence no regular points.
 7.4 Regular Singular Points and the Frobenius Method 307

4. y 00 + 1t y 0 + 1−t
t3 y =0

Solution: Clearly, t = 0
is the only singular point. Observe that t 1t = 1

is analytic while t2 t13 = 1t is not. Thus t = 0 is an irregular singular

point.
5. ty 00 + (1 − t)y 0 + 4ty = 0

Solution: We first write it in standard form: y 00 + 1−t 0

t y +4y = 0. While the
0 1−t
coefficient of y is analytic the coefficient of y is t is analytic except at
t = 0. It follows that t = 0 is a singular point. Observe that t 1−t


t = 1 − t
is analytic and 4t2 is too. It follows that t = 0 is a regular point.
6–10. Each differential equation has a regular singular point at t = 0. De-
termine the indicial polynomial and the exponents of singularity. How many
Frobenius solutions are guaranteed by Theorem 2.
6. 2ty 00 + y 0 + ty = 0

Solution: In standard form the equation is 2t2 y 00 +ty 0 +t2 y = 0. The indicial
equation is p(s) = 2s(s − 1) + s = 2s2 − s = s(2s − 1) The exponents of
singularity are 1 and 12 . Theorem 7.42 guarantees two Frobenius solutions.
7. t2 y 00 + 2ty 0 + t2 y = 0

Solution: The indicial equation is p(s) = s(s − 1) + 2s = s2 + s = s(s + 1)

The exponents of singularity are 0 and −1. Theorem 2 guarantees one
Frobenius solution but there could be two.
8. t2 y 00 + tet y 0 + 4(1 − 4t)y = 0

Solution: The indicial equation is p(s) = s(s − 1) + s + 4 = s2 + 4 The

exponents of singularity are ±2i. Theorem 2 guarantees that there are two
complex Frobenius solutions.
9. ty 00 + (1 − t)y 0 + λy = 0

Solution: In standard form the equation is t2 y 00 + t(1 − t)y 0 + λty = 0. The

indicial equation is p(s) = s(s − 1) + s = s2 The exponent of singularity is
0 with multiplicity 2. Theorem 2 guarantees that there is one Frobenius
solution. The other has a logarithmic term.
10. t2 y 00 + 3t(1 + 3t)y 0 + (1 − t2 )y = 0

Solution: The indicial equation is p(s) = s(s − 1) + 3s + 1 = s2 + 2s + 1 =

(s + 1)2 The exponent of singularity is −1 with multiplicity 2. Theorem 2
308 7 Power Series Methods

guarantees that there is exactly Frobenius solution. There is a logarithmic

solution as well.
11–14. Verify the following claims that were made in the text.
sin t cos t
11. In Example 5 verify the claims that y1 (t) = t and y2 (t) = t .

∞ ∞
(−1)n t2n 1 (−1)n t2n 1
sin t. y2 is done simi-
P P
Solution: y1 (t) = (2n+1)! = t (2n)! = t
n=0 n=0
larly.
12. In Example 8 we claimed that the solution to the recursion relation
((i + n)2 + 1)cn + ((i + n − 1)(i + n − 2) + 3)cn−1 + 3cn−2 + cn−3 = 0
in
was cn = n! . Use Mathematical Induction to verify this claim.

n
Solution: The entries in the table in Example 8 confirm cn = in! for
n = 0, . . . 5. Assume the formula is true for all k ≤ n. Let’s consider the
recursion relation with n replaced by n + 1 to get the following formula
for cn+1 :
((i + n + 1)2 + 1)cn+1 = −((i + n)(i + n − 1) + 3)cn + 3cn−1 + cn−2
in 3in−1 in−2
(n + 1)(2i + n + 1)cn+1 = −(n2 − n + 2in − i + 2) + +
n! (n − 1)! (n − 2)!
in−2 2 2
(n + 1)(2i + n + 1)cn+1 = − i (n − n + 2in − i + 2) + 3in + n(n − 1) 
n! 
in
(n + 1)(2i + n + 1)cn+1 = (i(n + 1 + 2i)).
n!
On the second line we use the induction hypothesis. Now divide both
sides by (n + 1 + 2i)(n + 1) in the last line to get
in+1
cn+1 = .
(n + 1)!
in
We can now conclude by Mathematical Induction that cn = n! , for all
n ≥ 0.
13. In Remark 3 we stated that in the logarithmic case could be obtained by
a reduction of order argument. Consider the Cauchy-Euler equation
t2 y 00 + 5ty 0 + 4y = 0.
One solution is y1 (t) = t−2 . Use reduction of order to show that a second
independent solution is y2 (t) = t−2 ln t, in harmony with the statement in
the coincident case of the theorem.

Solution: Let y(t) = t−2 v(t). Then y 0 (t) = −2t−3 v(t) + t−2 v 0 (t) and
y 00 (t) = 6t−4 v(t) − 4t−3 v 0 (t) + t−2 v 00 (t). From which we get
 7.4 Regular Singular Points and the Frobenius Method 309

t2 y 00 = 6t−2 v(t) − 4t−1 v 0 (t) + v 00 (t)

5ty 0 = −10t−2v(t) + 5t−1 v 0 (t)
4y = 4t−2 v(t).

Adding these terms and remembering that we are assuming the y is a

solution we get
0 = t−1 v 0 (t) + v 00 (t).
00
From this we get vv0 = −1 0
t . Integrating we get ln v (t) = − ln t and hence
1
v (t) = t . Integrating again gives v(t) = ln t. It follows that y(t) = t−2 ln t
0

is a second independent solution.

14. Verify the claim made in Example 6 that y1 (t) = (t3 + 4t2 )et

Solution:
∞
X n + 4 n+2
y1 (t) = t
n=0
n!
∞ ∞
X n n+2 X 4 n+2
= t + t
n=0
n! n=0
n!
∞ ∞ n
3
Xtn−1 2
X t
=t + 4t
n=1
(n − 1)! n=0
n!
= t3 et + 4t2 et = (t3 + 4t2 )et .

15–26. Use the Frobenius method to solve each of the following differential
equations. For those problems marked with a (*) one of the independent so-
lutions can easily be written in closed form. For those problems marked with
a (**) both independent solutions can easily be written in closed form.
15. ty 00 − 2y 0 + ty = 0 (**) (real roots; differ by integer; two Frobenius solu-
tions)

Solution: Indicial polynomial: p(s) = s(s − 3); exponents of singularity

s = 0 and s = 3.
n = 0 c0 (r)(r − 3) = 0
n = 1 c1 (r − 2)(r + 1) = 0
n ≥ 1 cn (n + r)(n + r − 3) = −cn−1

r=3:
n odd cn = 0
m
n = 2m c2m = 3c0 (−1) (2m+2)
(2m+3)!
P∞ (−1)m (2m+2)t2m+3
y(t) = 3c0 m=0 (2m+3)! = 3c0 (sin t − t cos t).
310 7 Power Series Methods

r=0: One is lead to the equation 0c3 = 0 and we can take c3 = 0. Thus
n odd cn = 0
m+1
n = 2m c2m = c0 (−1) (2m)!
(2m−1)

P∞ (−1)m+1 (2m−1)t2m
y(t) = c0 m=0 (2m)! = c0 (t sin t + cos t).
General Solution: y = c1 (sin t − t cos t) + c2 (t sin t + cos t).
16. 2t2 y 00 − ty 0 + (1 + t)y = 0 (**) (real roots, do not differ by an integer, two
Frobenius solutions)

Solution: Indicial polynomial: p(s) = (2s − 1)(s − 1); exponents of singu-

larity s = 1/2 and s = 1.
n = 0 c0 (2r − 1)(r − 1) = 0
n ≥ 1 cn (2(n + r) − 1)(n + r − 1) = −cn−1

−cn−1
r = 1: cn = (2n+1)n and hence

(−1)n 2n c0
cn = , n ≥ 1.
(2n + 1)!
From this we get
∞
X (−1)n 2n tn+1
y(t) = c0
n=0
(2n + 1)!
√ ∞ √
t X (−1)n ( 2t)2n+1
= c0 √
2 n=0 (2n + 1)!
c0 √ √
= √ t sin 2t.
2

r = 1/2: The recursion relation becomes cn (2n − 1)(n) = −cn−1 . The

−cn−1
coefficient of cn is never zero for n ≥ 1 hence cn = n(2n−1) and hence

(−1)n 2n
cn = .
(2n)!
We now get
∞
X (−1)n 2n tn+1/2
y(t) =
n=0
(2n)!
∞ √
√ X (−1)n ( 2t)2n
= t
n=0
(2n)!
√ √
= t cos 2t.
 7.4 Regular Singular Points and the Frobenius Method 311
√ √ √ √
General Solution: y = c1 t sin 2t + c2 t cos 2t.
17. t2 y 00 − t(1 + t)y 0 + y = 0, (*) (real roots, coincident, logarithmic case)

Solution: Indicial polynomial: p(s) = (s − 1)2 ; exponents of singularity

s = 1, multiplicity 2. There is one Frobenius solution.
P∞
r = 1 : Let y(t) = n=0 cn tn+1 . Then

n≥1 n2 cn − ncn−1 = 0.
1
This is easy to solve. We get cn = n! c0 and hence
∞
X 1 n+1
y(t) = c0 t = c0 tet .
n=0
n!

Logarithmic Solution: Let y1 (t) = tet . The second independent solution

is necessarily of the form
∞
X
y(t) = y1 (t) ln t + cn tn+1 .
n=0

Substitution into the differential equation leads to

∞
X
t2 e t + (n2 cn − ncn−1 )tn+1 = 0.
n=1

We write out the power series for t2 et and add corresponding coefficients
to get
1
n2 cn − ncn−1 + .
(n − 1)!
The following list is a straightforward verification:

n=1 c1 = −1
−1 1


n=2 c2 = 2 1+ 2
−1 1 1


n=3 c3 = 3! 1+ 2 + 3
−1 1 1 1


n=4 c4 = 4! 1+ 2 + 3 + 4 .

1
Let sn = 1 + 2 + · · · + n1 . Then an easy argument gives that

−sn
cn = .
n!
We now have a second independent solution
312 7 Power Series Methods
∞
X s n tn
y2 (t) = tet ln t − t .
n=1
n!

General Solution:
∞
!
X s n tn
y = c1 tet + c2 tet ln t − t .
n=1
n!

18. 2t2 y 00 − ty 0 + (1 − t)y = 0 (**) (real roots, do not differ by an integer, two
Frobenius solutions)

Solution: Indicial polynomial: p(s) = (2s − 1)(s − 1); exponents of singu-

larity s = 1/2 and s = 1.
n = 0 c0 (2r − 1)(r − 1) = 0
n ≥ 1 cn (2(n + r) − 1)(n + r − 1) = cn−1
cn−1
r = 1: cn = (2n+1)n and hence

2 n c0
cn = , n ≥ 1.
(2n + 1)!
From this we get
∞
X 2n tn+1
y(t) = c0
n=0
(2n + 1)!
√ ∞ √ 2n+1
t X ( 2t)
= c0 √
2 n=0 (2n + 1)!
c0 √ √
= √ t sinh 2t.
2

r = 1/2: The recursion relation becomes cn (2n − 1)(n) = cn−1 . The

cn−1
coefficient of cn is never zero for n ≥ 1 hence cn = n(2n−1) and hence

2n
cn = .
(2n)!
We now get
∞
X 2n tn+1/2
y(t) =
n=0
(2n)!
∞ √
√ X ( 2t)2n
= t
n=0
(2n)!
√ √
= t cosh 2t.
 7.4 Regular Singular Points and the Frobenius Method 313
√ √ √ √
General Solution: y = c1 t sinh 2t + c2 t cosh 2t.
19. t2 y 00 + t2 y 0 − 2y = 0 (**) (real roots; differ by integer; two Frobenius
solutions)

Solution: Indicial polynomial: p(s) = (s−2)(s+1); exponents of singularity

s = 2 and s = −1.
n = 0 c0 (r − 2)(r + 1) = 0
n ≥ 1 cn (n + r − 2)(n + r + 1) = −cn−1 (n + r − 1)

s=2:
(−1)n (n + 1)
cn = 6c0 , n≥1
(n + 3)!
 
P∞ (−1)n (n+1)tn −t
y(t) = 6c0 n=0 (n+3)! = 6c0 (t+2)e
t + t−2
t .
s=-1: The recursion relation becomes cn (n − 3)(n) = cn−1 (n − 2) = 0.
Thus
n = 1 c1 = − c20
n = 2 c2 = 0
n = 3 0c3 = 0
We can take c3 = 0 and
then cn = 0 for all n ≥ 2. We now have y(t) =
c0 t−1 (1 − 2t ) = c20 2−t
t .
−t
(t+2)e
General Solution: y = c1 2−t
t + c2 t .
20. t2 y 00 + 2ty 0 − a2 t2 y = 0 (**) (real roots, differ by integer, two Frobenius
Solutions)

Solution: Indicial polynomial: p(s) = s(s + 1); exponents of singularity

r = 0 and r = −1.
n = 0 c0 r(r + 1) = 0
n = 1 c1 (r + 1)(r + 2) = 0
n ≥ 1 cn (n + r)(n + r + 1) = a2 cn−2

r = 0: In the case c1 = 0 and c2n+1 = 0 for all n ≥ 0. We also get

a2 c0 c0 a2n
c2n = 2n(2n+1) , n ≥ 1 which gives c2n = (2n+1)! . Thus
∞
X (at)2n
y(t) = c0
n=0
(2n + 1)!
∞
c0 X (at)2n+1
=
at n=0 (2n + 1)!
c0
= sinh at.
at
314 7 Power Series Methods

r = −1: The n = 1 case gives 0c1 = 0 so c1 may be arbitrary. There

is no inconsistency. We choose c1 = 0. The recursion relation becomes
cn (n − 1)(n) = a2 cn−2 and it is straightforward to see that c2n+1 = 0
2n
while c2n = a(2n)!
c0
. Thus

∞
X a2n t2n−1
y(t) = c0
n=0
(2n)!
∞
c0 X (at)2n
=
t n=0 (2n)!
c0
= cosh at.
t

General Solution: y = c1 cosht at + c2 sinht at .

21. ty 00 + (t − 1)y 0 − 2y = 0, (*) (real roots, differ by integer, logarithmic case)

Solution: Indicial polynomial: p(s) = s(s − 2); exponents of singularity

r = 0 and r = 2.
n=0 c0 r(r − 2) = 0
n≥1 cn (n + r)(n + r − 2) = −cn−1 (n + r − 3)

n−1
r=2: The recursion relation becomes cn = − n(n+2) cn−1 . For n = 1 we
see that c1 = 0 and hence cn = 0 for all n ≥ 1. It follows that y(t) = c0 t2
is a solution.

r=0: The recursion relation becomes cn (n)(n − 2) = −cn−1 (n − 3) = 0.

Thus
c0
n = 1 −c1 = − −2
n = 2 0c2 = −2c0 ⇒⇐
The n = 2 case implies an inconsistency in the recursion relation since
c0 6= 0. Since y1 (t) = t2 is a Frobenius solution a second independent
solution can be written in the form
∞
X
y(t) = t2 ln t + c n tn .
n=0

Substitution leads to
∞
X
t3 + 2t2 + (−c1 − 2c0 )t + (cn (n)(n − 2) + cn−1 (n − 3))tn = 0
n=2

and the following relations:

 7.4 Regular Singular Points and the Frobenius Method 315

n=1 −c1 − 2c0 = 0

n=2 2 − c1 = 0
n=3 1 + 3c3 = 0
n≥4 n(n − 2)cn + (n − 3)cn−3 = 0.

We now have c0 = −1, c1 = 2, c3 = −1/3. c2 can be arbitrary so we choose

c2 = 0, and cn = −(n−3)c
n(n−2)
n−1
, for n ≥ 4. A straightforward calculation
gives
2(−1)n
cn = .
n!(n − 2)
A second independent solution is
∞
!
t3 X 2(−1)n tn
y2 (t) = t2 ln t + −1 + 2t − + .
3 n=4 n!(n − 2)

General Solution: y = c1 t2 + c2 y2 (t).

22. ty 00 − 4y = 0 (real roots, differ by an integer, logarithmic case)

Solution: Write the differential equation in standard form: t2 y 00 − 4ty = 0.

Clearly, the indicial polynomial is p(s) = s(s− 1); exponents of singularity
are r = 0 and r = 1.
r=1: Let ∞
X
y(t) = cn tn+1 , c0 6= 0.
n=0

Then
n≥1 cn (n + 1)(n) = 4cn−1
n
4 c0
It follows that cn = (n+1)!n! , for all n ≥ 1 (notice that when n = 0 we get
P∞ 4n tn+1 P∞ (4t)n+1
c0 ). Hence y(t) = c0 n=0 (n+1)!n! = c40 n=0 (n+1)!n! .

r = 0: When r = 0 the recursion relations become

cn (n)(n − 1) = 4cn−1 , n≥1

and hence when n = 1 we get 0c1 = 4c0 , an inconsistency as c0 6= 0.

There is no second Frobenius solution. However, there isPa second linearly
independent solution of the form y(t) = y1 (t) ln t + ∞ n
n=0 cn t , where
P∞ (4t)n+1
y1 (t) = n=0 (n+1)!n! . Substituting this equation into the differential
equation and simplifying leads to the recursion relation

(2n − 1)4n
n(n − 1)cn = 4cn−1 − , n ≥ 1.
(n + 1)!n!
316 7 Power Series Methods

When n = 1 we get 0 = 0c1 = 4c0 − 2 and this implies c0 = 12 . Then the

recursion relation for n = 1 becomes 0c1 = 0, no inconsistency; we will
choose c1 = 0. The above recursion relation is difficult to solve. We give
the first few terms:
c0 = 12 c1 = 0
−46
c2 = −2 c3 = 27
−251
c4 = 405
It now follows that
 
1 46 251 4
y2 (t) = y1 (t) ln t + − 2t2 − t3 − t + ··· .
2 27 405

23. t2 (−t + 1)y 00 + (t + t2 )y 0 + (−2t + 1)y = 0 (**) (complex)

Solution: Indicial polynomial: p(s) = (s2 + 1); exponents of singularity

r = ±i. Let r = i (the case r = −i
P gives equivalent results). The recursion
relation that arises from y(t) = ∞n=0 c n t n+i
is c n ((n+i) 2
+1)−cn−1 ((n−
2
2 + i) + 1) = 0 and hence

(n − 2)(n − 2 + 2i)
cn = cn−1 .
n(n + 2i)

A straightforward calculation gives the first few terms as follows:

1−2i
n = 1 c1 = 1+2i c0
n = 2 c2 = 0
n = 3 c3 = 0
 
1−2i
and hence cn = 0 for all n ≥ 2. Therefore y(t) = c0 (ti + 1+2i) t1+i .
Since t > 0 we can write ti = ei ln t = cos ln t+i sin ln t, by Euler’s formula.
Separating the real and imaginary parts we get two independent solutions

y1 (t) = −3 cos ln t − 4 sin ln t + 5t cos ln t

y2 (t) = −3 sin ln t + 4 cos ln t + 5t sin ln t.

24. t2 y 00 + t(1 + t)y 0 − 2y = 0, (**) (real roots, differ by an integer, two

Frobenius solutions)

Solution: Indicial polynomial: p(s) = (s−1)(s+1); exponents of singularity

r = 1 and r = −1.

n = 0 c0 (r − 1)(r + 1) = 0
n ≥ 1 cn (n + r − 1)(n + r + 1) = −cn−1 (n + r − 1)
 7.4 Regular Singular Points and the Frobenius Method 317

r = 1 : The recursion relation becomes cn (n(n+2)) = −ncn−1 and hence

2(−1)n c0
cn = .
(n + 2)!
We then get
∞
X
y(t) = cn tn+1
n=0
∞
X (−1)n tn+1
= 2c0
n=0
(n + 2)!
∞
2c0 X (−t)n+2
=
t n=0 (n + 2)!
2c0 −t
= (e − (1 − t))
t 
e−t − (1 − t)

= 2c0 .
t

r = −1: The recursion relation becomes cn (n − 2)(n) = −cn−1 (n − 2).

Thus
n = 1 c1 = −c0
n = 2 0c2 = 0
Thus c2 can be arbitrary so we can take c2 = 0 and then cn = 0 for all
n ≥ 2. We now have
∞
X 1−t
y(t) = cn tn−1 = c0 (t−1 − 1) = c0 .
n=0
t

−t
General Solution: y = c1 1−t e
t + c2 t .

25. t2 y 00 + t(1 − 2t)y 0 + (t2 − t + 1)y = 0 (**) (complex)

Solution: Indicial polynomial: p(s) = (s2 + 1); exponents of singularity

r = ±i. Let r = i (the case r = −i
P∞ gives equivalent results). The recursion
relation that arises from y(t) = n=0 cn tn+i is

n = 1 c1 = c0
n ≥ 2 cn ((n + i)2 + 1) + cn−1 (−2n − 2i + 1) + cn−2 = 0
A straightforward calculation gives the first few terms as follows:
n=1 c1 = c0
1
n=2 c2 = 2! c0
1
n=3 c3 = 3! c0
1
n=4 c4 = 4! c0 .
318 7 Power Series Methods

An easy induction argument gives

1
cn = c0 .
n!
We now get
∞
X
y(t) = cn tn+i
n=0
∞ n+i
X t
= c0
n=0
n!
i t
= c0 t e .

Since t > 0 we can write ti = ei ln t = cos ln t+i sin ln t, by Euler’s formula.

Now separating the real and imaginary parts we get two independent
solutions
y1 (t) = et cos ln t and y2 (t) = et sin ln t.

26. t2 (1 + t)y 00 − t(1 + 2t)y 0 + (1 + 2t)y = 0 (**) (real roots, equal, logarithmic
case)

Solution: Indicial polynomial: p(s) = (s − 1)2 ; exponents of singularity

s = 1 with multiplicity 2.
r=1: Let
∞
X
y(t) = cn tn+1 , c0 6= 0.
n=0

Then
n=1 c1 = 0
n≥2 n2 cn = −cn−1 (n − 2)
It follows that cn = 0 for all n ≥ 1. Hence y(t) = c0 t. Let y1 (t) = t.
P∞
logarithmic case: Let y(t) = t ln t+ n=0 cn tn+1 . The recursion relations
becomes
n = 0 −c0 + c0 = 0
n = 1 c1 = 1
n ≥ 2 n2 cn = −(n − 1)(n − 2)cn−1
The n = 0 case allows us to choose c0 arbitrarily. We choose c0 = 0. For
n ≥ 2 it is easily to see the cn = 0. Hence y(t) = t ln t + t2 .

general solution: y(t) = a1 t + a2 (t ln t + t2 ).

 7.5 Laplace Inversion involving Irreducible Quadratics 319

7.5 Application of the Frobenius Method:

Laplace Inversion involving Irreducible Quadratics
In Section 3.8 we considered the problem of finding the inverse Laplace trans-
form of simple rational functions with powers of an irreducible quadratic in
the denominator. After some reductions involving the First Translation Prin-
ciple and the Dilation Principle we reduced the problem to simple rational
functions of the form
1 s
and ,
(s2 + 1)k+1 (s2 + 1)k+1

where k is a nonnegative integer. Thereafter we derived a recursion formula

that determines the inversion for all such rational functions. In this section we
will derive a closed formula for the inversion by solving a distinguished second
order differential equation, with a regular singular point at t = 0, associated
to each simple rational function given above.
Recall from Section 3.5 that Ak (t) and Bk (t) were defined by the relations
1 1
L {Ak (t)} = 2
2k k! (s + 1)k+1
(1)
1 s
L {Bk (t)} = 2 .
2k k! (s + 1)k+1

Also, recall from Theorem 3.5.7 that

Bk+1 (t) = tAk (t). (2)

Lemma 1. Suppose k ≥ 1. Then

Ak (0) = 0 and Bk (0) = 0.

Proof. By the Initial Value Theorem, Theorem 6.4.3,

s
Ak (0) = 2k k! lim =0
(s2 + 1)k+1
s→∞

s2
Bk (0) = 2k k! lim 2 =0
s→∞ (s + 1)k+1

u
t

Proposition 2. Suppose k ≥ 1. Then

A0k (t) = Bk (t) (3)

Bk0 (t) = 2kAk−1 (t) − Ak (t). (4)
320 7 Power Series Methods

Proof. By the Input Derivative Principle and the Lemma above we have
1 1
L {A0k (t)} = (sL {Ak (t)} − Ak (0))
2k k! 2k k!
s
=
(s2 + 1)k+1
1
= k L {Bk (t)} .
2 k!
Equation (3) now follows. In a similar way we have
1 1
L {Bk0 (t)} = (sL {Bk (t)} − Bk (0))
2k k! 2k k!
s2
=
(s2
+ 1)k+1
1 1
= 2 k
− 2
(s + 1) (s + 1)k+1
2k 1
= k L {Ak−1 (t)} − k L {Ak (t)} .
2 k! 2 k!
Equation (4) now follows. u
t

Proposition 3. With notation as above we have

tA00k − 2kA0k + tAk = 0 (5)

2
t Bk00 − 2ktBk0 2
+ (t + 2k)Bk = 0. (6)

Proof. We first differentiate Equation (3) and then substitute in Equation (4)
to get
A00k = Bk0 = 2kAk−1 − Ak .
Now multiply this equation by t and simplify using Equations (2) and (3) to
get
tA00k = 2ktAk−1 − tAk = 2kBk − tAk = 2kA0k − tAk .
Equation (5) now follows. To derive Equation (6) first differentiate Equation
(5) and then multiply by t to get

t2 A000 00 2 0
k + (1 − 2k)tAk + t Ak + tAk = 0.

From Equation (5) we get tAk = −tA00k + 2kA0k which we substitute into the
equation above to get

t2 A000 00 2 0
k − 2ktAk + (t + 2k)Ak = 0.

Equation (6) follows now from Equation (3). u

t
 7.5 Laplace Inversion involving Irreducible Quadratics 321

Proposition 3 tells us that Ak and Bk both satisfy differential equations

with a regular singular point at t0 = 0. We know from Section 3.5 that Ak
and Bk are sums of products of polynomials with sin t and cos t (see Example
3.5.9, for the case k = 3). Thus the Frobenius method will produce rather
complicated power series and it will be very difficult to recognize these poly-
nomial factors. Let’s introduce a simplifying feature. Recall that the complex
roots of s2 + 1 are ±i. There are thus only two chains in the linear partial frac-
tion decomposition of each rational functions in Equation (1). By Theorem
3.2.8 the linear partial fractions come in conjugate pairs
 
cj cj
, ,
(s − i)j (s + i)j

for some cj ∈ C, and hence the terms in the inverse Laplace transform will be
a sum of terms of the form
tj−1 it tj−1 −it tj−1 it
 
cj e + cj e = 2 Re cj e ,
(j − 1)! (j − 1)! (j − 1)!

for each j = 1, . . . , k + 1. It follows from this that there are complex polyno-
mials ak (t) and bk (t), each of degree k, such that
 
1 1
L−1 = k Re(ak (t)eit )
(s2 + 1)k+1 2 k!
  (7)
−1 s 1 it
L = k Re(bk (t)e ).
(s2 + 1)k+1 2 k!

By the uniqueness of the Laplace transform this means then that

Ak (t) = Re(ak (t)eit )

(8)
Bk (t) = Re(bk (t)eit ).

Example 4. Compute a1 (t) and b1 (t).

I Solution. The relations

   
1 1 s 1
L−1 = Re(a 1 (t)e it
) and L−1
= Re(b1 (t)eit )
(s2 + 1)2 2 (s2 + 1)2 2

determines a1 (t) and b1 (t). Over the complex numbers the irreducible quadratic
factors as s2 + 1 = (s − i)(s + i). The following are the respective s − i chains,
which we will let the reader verify.
322 7 Power Series Methods

The s − i -chain The s − i -chain

1 (−1/4) s (−i/4)
(s − i)2 (s + i)2 (s − i)2 (s − i)2 (s + i)2 (s − i)2
(−i/4)(s + 3i) (−i/4) (i/4)(s − i) 0
(s − i)(s + i)2 (s − i) (s − i)(s + i)2 (s − i)
p2 (s) p2 (s)
(s + i)2 (s + i)2

It is unnecessary to compute p2 (s) for by Theorem 3.2.8 the s + i -chain is

given by the complex conjugation of the s − i -chain. It now follows that
1 (−1/4) (−1/4) (−i/4) (i/4)
= + + +
(s2 + 1)2 (s − i)2 (s + i)2 s−i s+i
s (−i/4) (i/4)
= +
(s2 + 1)2 (s − i)2 (s + i)2
and
 
−1 1 1 1
L = ((−t − i)eit + (−t + i)e−it ) = Re((−t − i)eit )
(s2 + 1)2 4 2
 
s 1 1
L−1 = (−iteit + ite−it ) = Re(−iteit ).
(s2 + 1)2 4 2
It follows that
a1 (t) = −t − i and b1 (t) = −it.
J
The following lemma will prove useful in determining ak and bk in general.
Among other things it shows that ak and bk are uniquely defined by Equation
8.
Lemma 5. Suppose p1 (t) and p2 (t) are polynomials with complex coefficients
and
Re(p1 (t)eit ) = Re(p2 (t)eit ).
Then p1 (t) = p2 (t).
λ+λ
Proof. For a complex number λ we have Re λ = 2 . We can write our
assumption as Re((p1 (t) − p2 (t))eit ) = 0 and hence

(p1 (t) − p2 (t))eit + (p1 (t) − p2 (t))e−it = 0.

It follows from Lemma 3.7.9 that p1 (t) − p2 (t) = 0 and the result follows. u
t
 7.5 Laplace Inversion involving Irreducible Quadratics 323

We now proceed to show that ak (t) and bk (t) satisfy second order differen-
tial equations with a regular singular point at t0 = 0. The Frobenius method
will give only one polynomial solution in each case, which we identify with
ak (t) and bk (t). From there it is an easy matter to use Equation (8) to find
Ak (t) and Bk (t).
Proposition 6. The polynomials ak (t) and bk (t) satisfy
ta00k + 2(it − k)a0k − 2kiak = 0
t2 b00k + 2t(it − k)b0k − 2k(it − 1)bk = 0.
Proof. Let’s start with Ak . Since differentiation respects the real and imagi-
nary parts of complex-valued functions we have
Ak (t) = Re(ak (t)eit )
A0k (t) = Re((a0k (t) + iak (t))eit )
A00k (t) = Re((a00k (t) + 2ia0k (t) − ak (t))eit ).
It follows now from Proposition 3 that
0 = tA00k (t) − 2kA0k (t) + tAk (t)
= t Re(ak (t)eit )00 − 2k Re(ak (t)eit )0 + t Re(ak (t)eit )
= Re((t(a00k (t) + 2ia0k (t) − ak (t)) − 2k(a0k (t) + iak (t)) + tak (t))eit )
= Re (ta00k (t) + 2(it − k)a0k (t) − 2kiak (t))eit .

Now, Lemma 5 implies

ta00k (t) + 2(it − k)a0k (t) − 2kiak (t) = 0.
The differential equation in bk is done similarly:
0 = t2 Bk00 (t) − 2ktBk0 (t) + (t2 + 2k)Bk (t)
= t2 Re(bk (t)eit )00 − 2kt Re(bk (t)eit )0 + (t2 + 2k) Re(bk (t)eit )
= Re((t2 (b00k (t) + 2ib0k (t) − bk (t)) − 2kt(b0k (t) + ibk (t)) + (t2 + 2k)bk (t))eit )
= Re (t2 b00k (t) + 2t(it − k)b0k (t) − 2k(it − 1)bk (t))eit .

Again, we apply Lemma 5 to get

t2 b00k (t) + 2t(it − k)b0k (t) − 2k(it − 1)bk (t) = 0.
u
t
Each differential equation involving ak and bk in Proposition 6 is second
order and has t0 = 0 as a regular singular point. Do not be troubled by
the presence of complex coefficients; the Frobenius method applies over the
complex numbers as well.
The following lemma will be useful in determining the coefficient needed
in the Frobenius solutions given below for ak (t) and bk (t).
324 7 Power Series Methods

Lemma 7. The constant coefficient of ak (t) is given by

(2k)!
ak (0) = −i .
k!2k
The coefficient of t in bk is given by
(2(k − 1))!
b0k (0) = −i .
(k − 1)!2(k−1)
Proof. By Theorem 3.5.7 Ak satisfies the recursion relation
Ak+2 = (2k + 3)Ak+1 − t2 Ak ,
for all t and k ≥ 1. A straightforward calculation and Lemma 5 gives that ak
satisfies the same recursion relation:
ak+2 (t) = (2k + 3)ak+1 (t) − t2 ak (t).
If t = 0 then we get ak+2 (0) = (2k + 3)ak+1 (0). Replace k by k − 1 to get
ak+1 (0) = (2k + 1)ak (0).
By Example 4 we have a1 (0) = −i. The above recursion relation gives the
following first four terms:
a1 (0) = −i a3 (0) = 5a2 (0) = −5 · 3 · i
a2 (0) = 3a1 (0) = −3i a4 (0) = 7a3 (0) = −7 · 5 · 3 · i

Inductively, we get
ak (0) = −(2k − 1) · (2k − 3) · · · 5 · 3 · i
(2k)!
= −i .
k!2k
For any polynomial p(t) the coefficient of t is given by p0 (0). Thus b0k (0) is the
coefficient of t in bk (t). On the other hand, Equation (2) implies bk+1 = tak (t)
and hence the coefficient of t in bk+1 (t) is the same as the constant coefficient,
ak (0) of ak (t). Replacing k by k − 1 and using the formula for ak (0) derived
above we get
(2(k − 1))!
b0k (0) = ak−1 (0) = −i .
(k − 1)!2(k−1)
J
Proposition 8. With the notation as above we have
k
−i X (2k − n)!
ak (t) = k (−2it)n
2 n=0 n!(k − n)!
k−1
−it X (2(k − 1) − n)!
and bk (t) = (−2it)n .
2k−1 n=0 n!(k − 1 − n)!
 7.5 Laplace Inversion involving Irreducible Quadratics 325

Proof. By Proposition 6, ak (t) is the polynomial solution to the differential

equation
ty 00 + 2(it − k)y 0 − 2kiy = 0 (9)
with constant coefficient ak (0) as given in Lemma 7. Multiplying Equation
(9) by t gives t2 y 00 + 2t(it − k)y 0 − 2kity = 0 which is easily seen to have a
regular singular point at t = 0. The indicial polynomial is given by q(s) =
s(s − 1) − 2ks = s(s − (2k + 1)). It follows that the exponents of singularity
are 0 and 2k + 1. We will show below that the r = 0 case gives a polynomial
solution. We leave it to the exercises to verify that the r = 2k + 1 case gives
a non polynomial Frobenius solution. We thus let
∞
X
y(t) = c n tn .
n=0

Then
∞
X
ty 00 (t) = (n)(n + 1)cn+1 tn
n=1
X∞
2ity 0 (t) = 2incn tn
n=1
X∞
−2ky 0 (t) = −2k(n + 1)cn+1 tn
n=0
X∞
−2iky(t) = −2ikcn tn
n=0

By assumption the sum of the series is zero. We separate the n = 0 and n ≥ 1

cases and simplify to get
n=0 −2kc1 − 2ikc0 = 0
(10)
n≥1 (n − 2k)(n + 1)cn+1 + 2i(n − k)cn = 0
The n = 0 case tells us that c1 = −c0 . For 1 ≤ n ≤ 2k − 1 we have
−2i(k − n)
cn+1 = cn .
(2k − n)(n + 1)
Observe that ck+1 = 0 and hence cn = 0 for all k + 1 ≤ n ≤ 2k − 1. For n = 2k
we get from the recursion relation 0c2k+1 = 2ikck = 0. This implies that c2k+1
can be arbitrary. We will choose c2k+1 = 0. Then cn = 0 for all n ≥ k + 1
and hence the solution y is a polynomial. We make the usual comment that if
ck+1 is chosen to be nonzero then the those terms with ck+1 as a factor will
make up the Frobenius solution for exponent of singularity r = 2k + 1. Let’s
now determine the coefficients cn for 0 ≤ n ≤ k. From the recursion relation,
Equation (10), we get
326 7 Power Series Methods

n=0 c1 = −ic0
−2i(k−1) 2(−i)2 (k−1) (−2i)2 k(k−1)
n=1 c2 = (2k−1)2 c1 = (2k−1)2 = (2k)(2k−1)2
−2i(k−2) (−2i)3 k(k−1)(k−2)
n=2 c3 = (2k−2)3 c2 = 2k(2k−1)(2k−2)3! c0
−2i(k−3) (−2i)4 k(k−1)(k−2)(k−3)
n=3 c4 = (2k−3)4 c3 = 2k(2k−1)(2k−2)(2k−3)4! c0

and generally,

(−2i)n k(k − 1) · · · (k − n + 1)
cn = n = 1, . . . , k.
2k(2k − 1) · · · (2k − n + 1)n!

We can write this more compactly in terms of binomial coefficients as

(−2i)n nk


cn = 2k

.
n n!

It follows now that

k
(−2i)n nk n


X
y(t) = c0 2k

t .
n=0 n n!

The constant coefficient is c0 . Thus we choose c0 = ak (0) as given in Lemma

7. Then y(t) = ak (t) is the polynomial solution we seek. We get
k
(−2i)n nk n


X
ak (t) = ak (0) 2k

t
n=0 n n!
k k
(2k)! (−2i)n n n


X
= −i 2k
t
k!2k


n=0 n n!
k
−i X (2k − n)!
= (−2it)n .
2k n=0 (k − n)!n!

It is easy to check that Equation (2) has as an analogue the equation bk+1 (t) =
tak (t). Replacing k by k − 1 in the formula for ak (t) and multiplying by t thus
establishes the formula for bk (t). J

For x ∈ R we let [x] denote the greatest integer function. It is defined to

be the greatest integer less than or equal to x. For example, [1.5] = 1 and
[2] = 2. We are now in a position to give a closed formula for the inverse
Laplace transforms of the simple rational functions given in Equation (7).

Theorem 9. For the simple rational functions we have:

 7.5 Laplace Inversion involving Irreducible Quadratics 327

Simple Irreducible Quadratics

n o [P
2]
k

1 sin t 2k−2m (2t)2m

L −1
(s2 +1)k+1
(t) = 22k
(−1)m k (2m)!
m=0

2 ]
[ k−1
cos t 2k−2m−1 (2t)2m+1


(−1)m
P
− 22k k (2m+1)!
m=0

n o 2 ]
[ k−1
s t sin t 2k−2m−2 (2t)2m


(−1)m
P
L −1
(s2 +1)k+1
(t) = k·22k−1 k−1 (2m)!
m=0

2 ]
[ k−2
t cos t 2k−2m−3 (2t)2m+1


(−1)m
P
− k·22k−1 k−1 (2m+1)!
.
m=0

Proof. From Proposition 8 we can write

k
1 X (2k − n)!
ak (t) = (−1)n+1 (i)n+1 (2t)n .
2k n=0 n!(k − n)!

It is easy to see that the real part of ak consists of those terms where n is odd.
The imaginary part (up to the factor of i) consists of those terms where n is
even. The odd  integers from n = 0, . . . , k can be written n = 2m + 1 where
m = 0, . . . , k−1 . Similarly, the even integers can be written n = 2m, where

 2
m = 0, . . . , k2 . We thus have

[ k−1
2 ]
1 X (2k − 2m − 1)!
Re(ak (t)) = k (−1)2m+2 (i)2m+2 (2t)2m+1
2 m=0 (2m + 1)!(k − 2m − 1)!
2 ]
[ k−1
−1 X (2k − 2m − 1)!
= k (−1)m (2t)2m+1
2 m=0 (2m + 1)!(k − 2m − 1)!

and
328 7 Power Series Methods

[ k2 ]
1 1 X (2k − 2m)!
Im(ak (t)) = (−1)2m+1 (i)2m+1 (2t)2m
i 2k m=0 (2m)!(k − 2m)!
[ k2 ]
−1 X (2k − 2m)!
= k (−1)m (2t)2m .
2 m=0 (2m)!(k − 2m)!

Now Re(ak (t)eit ) = Re(ak (t)) cos t − Im(ak (t)) sin t. It follows from Equation
7 that
 
−1 1 1
L = k Re(ak (t)eit )
(s2 + 1)k+1 2 k!
1 1
= k Re(ak (t)) cos t − k Im(ak (t)) sin t
2 k! 2 k!
[ k−1
2 ]
− cos t X (2k − 2m − 1)!
= (−1)m (2t)2m+1
22k m=0 k!(2m + 1)!(k − 2m − 1)!
[ k2 ]
sin t X (2k − 2m)!
+ 2k (−1)m (2t)2m
2 m=0 k!(2m)!(k − 2m)!
[ k−1
2 ]
(2t)2m+1

− cos t X 2k − 2m − 1
= 2k
(−1)m
2 m=0
k (2m + 1)!
[ k2 ] 
(2t)2m

sin t X 2k − 2m
+ 2k (−1)m .
2 m=0 k (2m)!
n o
s
A similar calculation gives the formula for L−1 (s2 +1)k+1 . u
t

Exercises
1–3.
1. Verify the statement made that the Frobenius solution to

ty 00 + 2(it − k)y 0 − 2kiy = 0

with exponent of singularity r = 2k + 1 is not a polynomial.

2. Use the Frobenius method on the differential equation

t2 y 00 + 2t(it − k)y 0 − 2k(it − 1) = 0.

Verify that the indicial polynomial is q(s) = (s − 1)(s − 2k). Show that for
the exponent of singularity r = 1 there is a polynomial solution while for the
exponent of singularity r = 2k there is not a polynomial solution. Verify the
formula for bk (t) given in Proposition 8.
 7.5 Laplace Inversion involving Irreducible Quadratics 329
s
3. Verify the formula for the inverse Laplace transform of given in
(s2 + 1)k+1
Theorem 9.

4–11. This series of exercises leads to closed formulas for the inverse
Laplace transform of
1 s
and .
(s2 − 1)k+1 (s2 − 1)k+1

Define Ck (t) and Dk (t) by the formulas

1 1
L {Ck (t)} (s) = 2
2k k! (s − 1)k+1
and
1 s
L {Dk (t)} (s) = 2 .
2k k! (s − 1)k+1
4. Show the following:

1. Ck+2 (t) = t2 Ck (t) − (2k + 3)Ck+1 (t)

2. Dk+2 (t) = t2 Dk (t) − (2k + 1)Dk+1 (t)
3. tCk (t) = Dk+1 (t).

5. Show the following: For k ≥ 1

1. Ck0 (t) = Dk (t)

2. Dk0 (t) = 2kCk−1 (t) + Ck (t).

6. Show the following:

1. tCk00 (t) − 2kCk0 (t) − tCk (t) = 0

2. t2 Dk00 (t) − 2ktDk0 (t) = (2k − t2 )Dk (t) = 0

7. Show that there are polynomials ck (t) and dk (t), each of degree at most k, such
that

1. Ck (t) = ck (t)et − ck (−t)e−t

2. Dk (t) = dk (t)et + dk (−t)e−t

8. Show the following:

1. tc00k (t) + (2t − 2k)c0k (t) − 2kck (t) = 0

2. td00k (t) + 2t(t − k)d0k (t) − 2k(t − 1)dk (t) = 0

9. Show the following:

(−1)k (2k)!
1. ck (0) = .
2k+1 k!

(−1)k−1 (2(k − 1))!

2. d0k (0) = .
2k (k − 1)!
330 7 Power Series Methods

10. Show the following:

(−1)k (2k − n)!

k
1. ck (t) = (−2t)n .
2k+1 n=0 n!(k − n)!

(−1)k (2k − n − 1)!

k
2. dk (t) = k+1
(−2t)n .
2 n=1
(n − 1)!(k − n)!

11. Show the following:

(−1)k (2k − n)!

k
1
1. L−1
(t) = (−2t)n et − (2t)n e−t .
(s2 − 1)k+1  22k+1 k! n=0 n!(n − k)! 

(−1)k (2k − n − 1)!

k
s
2. L−1
(t) = 2k+1 (−2t)n et + (2t)n e−t .
(s2 − 1) k+1  2 k! n=1 (n − 1)!(n − k)! 
8
Discontinuous Functions and the Laplace
Transform

For many applications the set of elementary functions, as defined in Chapter

3, or even the set of continuous functions is not sufficiently large to deal
with some of the common applications we encounter. Consider two examples.
Imagine a mixing problem (see Example 5 in Section 1.1 and the discussion
that follows for a review of mixing problems) where there are two sources of
incoming salt solutions with different concentrations. Initially, the first source
may be flowing for several minutes. Then the second source is turned on at
the same time the first source is turned off. The graph of the input function
may well be represented by Figure 8.1. The most immediate observation is

0
t

Fig. 8.1. A discontinuous input function

that the input function is discontinuous. Nevertheless, the Laplace transform

methods we will develop will easily handle this situation, leading to a formula
for the amount of the salt in the tank as a function of time. As a second
example, consider a sudden force that is applied to a spring-mass-dashpot
332 8 Laplace Transform II

system (see Section ?? for a discussion of spring-mass-dashpot systems). To

model this we imagine that a large force is applied over a small interval. As
the interval gets smaller the force gets larger so that the total force always
remains a constant. By a limiting process we obtain an instantaneous input
called an impulse function. This idea is graphed in Figure 8.2. Such impulse

Fig. 8.2. An impulse function

functions have predicable effects on the system. Again the Laplace transform
methods we develop here will lead us to the motion of the body without much
difficulty.
These two examples illustrate the need to extend the Laplace transform
beyond the set of elementary functions that we discussed in Chapter 3. We will
do this in two stages. First we will identify a suitably larger class of functions,
the Heaviside class, that includes discontinuous functions and then extend
the Laplace transform method to this larger class. Second, we will consider
the Dirac delta function, which models the impulse function we discussed
above. Even though it is called a function the Dirac delta function is actually
not a function at all. Nevertheless, its Laplace transform can be defined and
the Laplace transform method can be extended to differential equations that
involve impulse functions.

8.1 Calculus of Discontinuous Functions

Our focus in the next few sections is a study of first and second order linear
constant coefficient differential equations with possibly discontinuous input or
forcing function f :
y 0 + ay = f (t)
y + ay 0 + by = f (t)
00

Allowing f to have some discontinuities introduces some technical difficul-

ties as to what we mean by a solution to such a differential equation. To get
 8.1 Calculus of Discontinuous Functions 333

an idea of these difficulties and motivate some of the definitions that follow
we consider two elementary examples.
First, consider the simple differential equation

y 0 = f (t),

where (
0 if 0 ≤ t < 1
f (t) =
1 if 1 ≤ t < ∞.
Simply stated what we are seeking is a function y whose derivative is the
discontinuous function f . If y is a solution then y must also be a solution when
restricted to any subinterval. In particular, let’s restrict to the subintervals
(0, 1) and (1, ∞), where f is continuous separately. On the interval (0, 1) we
obtain y(t) = c1 , where c1 is a constant and on the interval (1, ∞) the solution
is y(t) = t + c2 , where c2 is a constant. Piecing these solutions together gives
(
c1 if 0 < t < 1
y=
t + c2 if 1 < t < ∞.

Notice that this family has two arbitrary parameters, c1 and c2 and unless
c1 and c2 are chosen just right y will not extend to a continuous function.
In applications, like the mixing problem introduced in the introduction, it is
reasonable to seek a continuous solution. Thus suppose an initial condition
is given, y(0) = 1, say, and suppose we wish to find a continuous solution.
Since limt→0+ y(t) = c1 , continuity and the initial condition forces c1 = 1
and therefore y(t) = 1 on the interval [0, 1). Now since limt→1− y(t) = 1,
continuity forces that we define y(1)=1. Repeating this argument we have
limt→1+ y(t) = 1 + c2 and this forces c2 = 0. Therefore y(t) = t on the interval
(1, ∞). Putting these pieces together gives a continuous solution whose graph
is given in Figure 8.3. Nevertheless, no matter how we choose the constants
c1 and c2 the "solution" y is never differentiable at the point t = 1. Therefore
the best that we can expect for a solution to y 0 = f (t) is a continuous function
y which is differentiable at all points except t = 1.
As a second example consider the differential equation

y 0 = f (t),

where (
1
(1−t)2 if 0 ≤ t < 1
f (t) =
1 if 1 ≤ t < ∞.
We approach this problem as we did above and obtain that a solution must
have the form (
1
+ c1 if 0 < t < 1
y(t) = 1−t
t + c2 if 1 < t < ∞,
334 8 Laplace Transform II

3
y
2

0 1 2 3 4
t
Fig. 8.3. A continuous solution to y 0 = f (t).

3
y
2

0 1 2 3 4
t
Fig. 8.4. Always a discontinuous solution

where c1 and c2 are arbitrary constants. The graph of this function for c1 = 1
and c2 = 1 is given in Figure 8.4. For us this situation is very undesirable
in that no matter how we choose the constants, the solution y will always be
discontinuous at t = 1. The asymptotic behavior at t = 1 for the solution
results in the fact that f has a vertical asymptote at t = 1.
These examples illustrate the need to be selective in the kinds of disconti-
nuities we allow. In particular, we will require that if f does have a disconti-
nuity it must be a jump discontinuity. The function f in our first example has
a jump discontinuity at t = 1 while in the second example the discontinuity at
t = 1 is a result of a vertical asymptote. We also must relax our definition of
what we mean by a solution to allow solutions y that have some points where
the derivative may not exist. We will be more precise about this later.
 8.1 Calculus of Discontinuous Functions 335

Jump Discontinuities
We say f has a jump discontinuity at a point a if

f (a+ ) 6= f (a− )

where f (a+ ) = limt→a+ f (t) and f (a− ) = limt→a− f (t). In other words, the
left hand limit and the right hand limit at a exist but are not equal. Examples
of such functions are typically given piecewise, that is, a different formula is
used to define f on different subintervals of the domain. For example, consider

2
t
 if 0 ≤ t < 1,
f (t) = 1 − t if 1 ≤ t < 2,

1 if 2 ≤ t ≤ 3


whose graph is given in Figure 8.5. We see that f is defined on the interval

0.5

0 0.5 1 1.5 2 2.5 3

x

–0.5

–1

Fig. 8.5. A piecewise continuous function

[0, 3] and has a jump discontinuity at a = 1 and a = 2: f (1− ) = 1 6= f (1+ ) = 0

and f (2− ) = −1 6= f (2+ ) = 1. On the other hand, the function
(
t if 0 ≤ t < 1,
g(t) = 1
t−1 if 1 ≤ t ≤ 2,

whose graph is given in Figure 8.6, is defined on the interval [0, 2] and has
a discontinuity at a = 1. However, this is not a jump discontinuity because
limt→1+ g(t) does not exist.
For our purposes we will say that a function f is piecewise continuous
on an interval [α, β] if f is continuous except for possibly finitely many
jump discontinuities. If an interval is not specified it will be understood that
f is defined on [0, ∞) and f is continuous on all subintervals of the form
[0, N ] except for possibly finitely many jump discontinuities. For convenience
it will not be required that f be defined at the jump discontinuities. Suppose
336 8 Laplace Transform II

1 2
t

Fig. 8.6. A discontinuous function but not a jump discontinuity

a1 , . . . , an are the locations of the jump discontinuities in the interval [0, N ]

and assume ai < ai+1 , for each i. On the interval (ai , ai+1 ) we can extend f
to a continuous function on the closed interval [ai , ai+1 ]. Since a continuous
function on a closed interval is bounded and there are only finitely many jump
discontinuities we have the following proposition.
Proposition 1. If f is a piecewise continuous function on [0, N ] then f is
bounded.

Integration of Piecewise Continuous Functions

Suppose f is a piecewise continuous function on the interval [0, N ] and the
jump discontinuities are located at a1 , . . . , ak . We may assume ai < ai+1 and
we will let a0 = 0 and ak+1 = N . In this case the definite integral of f on
[0, N ] exists and
Z N Z a1 Z a2 Z ak+1
f (t) dt = f (t) dt + f (t) dt + · · · + f (t) dt.
0 a0 a1 ak

On each interval of the form (ai , ai+1 ) f is continuous and therefore an an-
tiderivative Fi exists. Since f is bounded so is Fi and thus may be extended
to the closed interval [ai , ai+1 ]. When necessary we will denote the extended
values of Fi at ai by Fi (a+ −
i ) and at ai+1 by Fi (ai+1 ). We then have
Z ai+1
f (t) dt = Fi (a+ −
i ) − Fi (ai+1 ).
ai
R5
Example 2. Find −1 f (t) dt, if


 t if −1≤t<1
if

2 t=1
f (t) = 1
t

 if 1≤t<3
2 if 3≤t<5

 8.1 Calculus of Discontinuous Functions 337

I Solution. The function f is piecewise continuous and

Z 5 Z 1 Z 3 Z 5
1
f (t) dt = t dt + dt + 2 dt
−1 −1 1 t 3
= 0 + ln 3 + 4 = 4 + ln 3.

We note that the value of f at t = 1 played no role in the computation of the

integral. J
Rt
Example 3. Find 0 f (u) du for the piecewise function f given by

2
t
 if 0 ≤ t < 1,
f (t) = 1 − t if 1 ≤ t < 2,
if 2 ≤ t < ∞.

1


I Solution. The function f is given piecewise on the intervals [0, 1), [1, 2)
and [2, ∞). We will therefore consider three cases. If t ∈ [0, 1) then
Z t Z t
t3
f (u) du = u2 du = .
0 0 3
It t ∈ [1, 2) then
Z t Z 1 Z t
f (u) du = f (u) du + f (u) du
0 0 1
Z t
1
= + (1 − u) du
3 1
t2 t2
 
1 1 1
= + t− − =− +t− .
3 2 2 2 6
Finally, if t ≥ 2 then
Z t Z 2 Z t
f (u) du = f (u) du + f (u) du
0 0 2
Z t
1 13
=− + 1 du = t −
6 2 6
Piecing these functions together gives
 3
t
Z t 3 2
 if 0 ≤ t < 1
f (u) du = − t2 + t − 1
6 if 1 ≤ t < 2
0
t − 13

if 2 ≤ t < ∞.

6

It is continuous as can be observed in the graph given in Figure 8.7 J

The discussion above leads to the following proposition.

338 8 Laplace Transform II

1.5

y 1

0.5

0 1 2 3 4
t
–0.5

–1
Fig. 8.7. The graph of the integral of the discontinuous function in Example 3

Proposition 4. If f is a piecewise continuous function on an interval [α, β]

Rt
and a, t ∈ [α, β] then the integral a f (u) du exists and is a continuous function
in the variable t.
Rt
Proof. The integral exists as discussed above. Let F (t) = a f (u) du. Since f
is piecewise continuous on [α, β] it is bounded by Proposition 1. We may then
suppose |f (t)| ≤ B, for some B > 0. Let  > 0. Then
Z t+ Z t+
|F (t + ) − F (t)| ≤ |f (u)| du ≤ B du = B.
t t

Therefore lim→0 F (t + ) = F (t) and hence F (t+ ) = F (t). In a similar way

F (t− ) = F (t). This establishes the continuity of F . u
t

Differentiation of Piecewise Continuous

Functions
In the applications, we will consider (continuous) functions that are differen-
tiable on intervals [0, N ) except at finitely many points. In this case we will
use the symbol f 0 to denote the derivative of f though it may not be defined
at some points. For example, consider
(
0 if 0 ≤ t < 1
f (t) =
t−1 if 1 ≤ t < ∞.

This function is continuous on [0, ∞) and differentiable at all points except

t = 1. A simple calculation gives
 8.1 Calculus of Discontinuous Functions 339
(
0 if 0 < t < 1
f 0 (t) =
1 if 1 < t < ∞.

Notice that f 0 is not defined at t = 1.

Differential Equations and Piecewise

Continuous Functions
We are now in a position to consider some examples of constant coefficient
linear differential equations with piecewise continuous forcing functions. Let
f (t) be a piecewise continuous function. We say that a function y is a solution
to y 0 +ay = f (t) if y is continuous and satisfies the differential equation except
at the location of the discontinuities of the input function.
Example 5. Find a continuous solution to

1
 if 0 ≤ t < 1
0
y + 2y = f (t) = t − 1 if 1 ≤ t < 3 y(0) = 1. (1)

0 if 3 ≤ t < ∞,


I Solution. We will apply the method that we discussed at the beginning of

this section. That is, we will consider the differential equation on the subinter-
vals where f is continuous and then piece the solution together. In each case
the techniques discussed in Section 2.2 apply. The first subinterval is (0, 1),
where f (t) = 1. Multiplying both sides of y 0 +2y = 1 by the integrating factor,
e2t , leads to (e2t y)0 = e2t . Integrating and solving for y gives y = 21 + ce−2t .
The initial condition y(0) = 1 implies that c = 21 and therefore
1 1 −2t
y= + e , 0 ≤ t < 1.
2 2
We let y(1) = y(1− ) = 12 + 12 e−2 . This value becomes the initial condition for
y 0 + 2y = t − 1 on the interval [1, 3). Following a similar procedure leads to
t − 1 1 3 −2(t−1) 1 −2t
y= − + e + e , 1 ≤ t < 3.
2 4 4 2
We define y(3) = y(3− ) = 34 + 43 e−4 + 12 e−6 . This value becomes the initial
condition for y 0 + 2y = 0 on the interval [3, ∞). Its solution there is the
function
3 3 1
y = e−2(t−3) + e−2(t−1) + e−2t .
4 4 2
Putting these pieces together gives the solution

1 1 −2t
2 + 2e
 if 0 ≤ t < 1
t−1 1 3 −2(t−1) 1 −2t
y(t) = 2 − 4 + 4e + 2e if 1 ≤ t < 3
 3 −2(t−3) 3 −2(t−1) 1 −2t

4e + 4e + 2e if 3 ≤ t < ∞.
340 8 Laplace Transform II

By making the initial value on each subinterval the left hand limit of the
solution on the previous interval we guarantee continuity. The graph of this
solution is shown in Figure 8.8. The discontinuity of the derivative of y at
t = 1 and t = 3 is evident by the kinks at those points. J

1.2

0.8

y 0.6

0.4

0.2

0 1 2 3 4
t

Fig. 8.8. The graph of the solution to Example 5

The method we used here insures that the solution we obtain is continuous
and the initial condition at t = 0 determines the subsequent initial conditions
at the points of discontinuity of f . We also note that the initial condition at
t = 0, the left hand endpoint of the domain, was chosen only for convenience;
we could have taken the initial value at any point t0 ≥ 0 and pieced together
a continuous function on both sides of t0 . That this can be done in general is
stated in the following theorem.
Theorem 6. Suppose f is a piecewise continuous function on an interval
[α, β] and t0 ∈ [α, β]. There is a unique continuous function y which satisfies
y 0 + ay = f (t), y(t0 ) = y0 .
Recall that this means that y 0 will not exist at the points of discontinuity of f .
Proof. We follow the method illustrated in the example above to construct a
continuous solution. To prove uniqueness suppose y1 and y2 are two continuous
solutions. If y = y1 − y2 then y(t0 ) = 0 and y is a continuous solution to
y 0 + ay = 0.
On the interval containing t0 on which f is continuous, y = 0 by the uniqueness
and existence theorem. The initial value at the endpoint of adjacent intervals
is thus 0. Continuing in this way we see that y is identically 0 on [α, β] and
hence y1 = y2 . u
t

We now consider a second order constant coefficient differential equation

with a piecewise continuous forcing function. Our method is similar to the one
 8.1 Calculus of Discontinuous Functions 341

above, however, we demand more out of our solution. If f (t) be a piecewise

continuous function then we say a function y is a solution to y 00 + ay 0 + by =
f (t) if y is continuous, has a continuous derivative, and satisfies the differential
equation except at the discontinuities of the forcing function f .
Example 7. Find a solution y to
(
00 t if 0 ≤ t < 2π
y + y = f (t) = y(0) = 0 y 0 (0) = 1.
2 if 2π ≤ t ≤ 4π,
I Solution. We begin by considering the differential equation y 00 + y = t
on the interval [0, 2π). The homogenous solution is yh (t) = a cos t + b sin t and
the method of undetermined coefficients or variation of parameters leads to a
particular solution yp (t) = t−1. The general solution is y = t−1+a cos t+b sin t
and incorporating the initial conditions leads to y = t−1+cos t on the interval
[0, 2π). We calculate that y 0 = 1 − sin t. In order to piece together a solution
that is continuous at t = 2π we must have y(2π) = y(2π − ) = 2π. In order for
the derivative y 0 to be continuous at t = 2π we must have y 0 (2π) = y 0 (2π − ) =
1. We use these values for the initial conditions on the interval [2π, 4π]. The
general solution to y 00 + y = 2 is y = 2 + a cos t + b sin t. The initial conditions
imply a = 2π − 2 and b = 1 and thus y = 2 + (2π − 2) cos t + sin t. Piecing
these two solutions together gives
(
t − 1 + cos t if 0 ≤ t < 2π
y(t) =
2 + (2π − 2) cos t + sin t if 2π ≤ t ≤ 4π.
Its derivative is
(
1 − sin t if 0 ≤ t < 2π
y 0 (t) =
−(2π − 2) sin t + cos t if 2π ≤ t ≤ 4π.
Figure 8.9 gives (a) the graph of the solution and (b) the graph of its derivative.
The solution is differentiable on the interval [0, 4π] and the derivative is

6 4

y
4 2
y

2 0 2 4 6 8 10 12
t
0 –2
2 4 6 8 10 12
t
–2 –4

(a) (b)

Fig. 8.9. The solution (a) and its derivative (b) to Example 7

continuous on [0, 4π). However, the kink in the derivative at t = 2π indicates

that the second derivative is not continuous. J
342 8 Laplace Transform II

In direct analogy to the first order case we considered above we are lead
to the following theorem. The proof is omitted.
Theorem 8. Suppose f is a piecewise continuous function on an interval
[α, β] and t0 ∈ [α, β]. There is a unique continuous function y which satisfies

y 00 + ay 0 + by = f (t), y(t0 ) = y0 , y 0 (t0 ) = y1 .

Furthermore, y is differentiable and y 0 is continuous.

Piecing together solutions in the way that we described above is at best
tedious. As we proceed we will extend the Laplace transform to a class of
functions that includes piecewise continuous function. The Laplace transform
method extends as well and will provide an alternate method for solving dif-
ferential equations like the ones above. It is one of the hallmarks of the Laplace
transform.

Exercises
Match the following functions that are given piecewise with their graphs and
determine where jump discontinuities occur.
1
 if 0 ≤ t < 4
1. f (t) = −1 if 4 ≤ t < 5


0 if 5 ≤ t < ∞.
t
 if 0 ≤ t < 1
2. f (t) = 2 − t if 1 ≤ t < 2


1 if 2 ≤ t < ∞.
t if 0 ≤ t < 1
3. f (t) = 
2−t if 1 ≤ t < ∞.

t if 0 ≤ t < 1

t − 1

 if 1 ≤ t < 2
4. f (t) = if 2 ≤ t < 3
t − 2



..
. .
1 if 2n ≤ t < 2n + 1
5. f (t) = 
0 if 2n + 1 ≤ t < 2n + 2.
t2
 if 0 ≤ t < 2
6. f (t) = 4 if 2 ≤ t < 3


7−t if 3 ≤ t < ∞.

1−t if 0 ≤ t < 2

3 − t

 if 2 ≤ t < 4
7. f (t) = 5−t if 4 ≤ t < 6




..
. .
 8.1 Calculus of Discontinuous Functions 343


1 if 0 ≤ t < 2


3 − t if 2 ≤ t < 3
8. f (t) =

2(t − 3) if 3≤t<4


2 if 4 ≤ t < ∞.
Graphs for problems 1 through 8

1 2

0.5 1

0 0 2 4 6 8
1 2 t 3 4 5 t
(a) (b)

1 4

2
0 2 4 6 8
t
0 2 4 6 8
–1 t
(c) (d)

1
1 2 t 3 4 5
0
0.5

–2
0 1 2 3 4 5
t
(e) (f)

1 1

0.5
0 2 4 6 8
t

0 2 4 6 8
t –1
(g) (h)
In problems 9 through 12 calculate the indicated integral.
t2 − 4
 if 0 ≤ t < 2
5
9. 0 f (t) dt, where f (t) = 0 if 2 ≤ t < 3


−t + 3 if 3 ≤ t < 5.
2 2−u if 0 ≤ u < 1
10. 0
f (u) du, where f (u) =  3
u if 1 ≤ u < 2.
344 8 Laplace Transform II
2π
11. 0
|sin(x)| dx.
w
 if 0 ≤ w < 1
3
12. f (w) dw where f (w) = w1 if 1 ≤ w < 2
0 
1
2
if 2 ≤ w < ∞.
In problems 13 through 16 find the indicated integral. (See problems 1 through
9 for the appropriate formula.)
5
13. 2 f (t) dt, where the graph of f is:
2

0 2 4 6 8
t
8
14. 0
f (t) dt, where the graph of f is:
1

0 2 4 6 8
t

–1
6
15. 0
f (u) du, where the graph of f is:
1

0.5

0 2 4 6 8
t
7
16. 0
f (t) dt, where the graph of f is:
4

0 2 4 6 8
t

17. Of the following four piecewise defined functions determine which ones (A)
satisfy the differential equation

4 if 0 ≤ t < 2
y 0 + 4y = f (t) = 
8t if 2 ≤ t < ∞,

except at the point of discontinuity of f , (B) are continuous, and (C) are con-
tinuous solutions to the differential equation with initial condition y(0) = 2. Do
not solve the differential equation.
 8.1 Calculus of Discontinuous Functions 345

1 if 0 ≤ t < 2
a) y(t) =  1
2t − 2
− 52 e−4(t−2) if 2 ≤ t < ∞
1+e −4t
if 0 ≤ t < 2
b) y(t) = 
2t − 12 − 52 e−4(t−2) + e−4t if 2 ≤ t < ∞
1 + e−4t if 0 ≤ t < 2
c) y(t) =  5e−4(t−2)
2t − 12 − 2
if 2 ≤ t < ∞
2e−4t if 0 ≤ t < 2
d) y(t) = 
2t − 12 − 52 e−4(t−2) + e−4t if 2 ≤ t < ∞
18. Of the following four piecewise defined functions determine which ones (A)
satisfy the differential equation

et if 0 ≤ t < 1
y 00 − 3y 0 = 2y = f (t) =  2t
e if 1 ≤ t < ∞,

except at the point of discontinuity of f , (B) are continuous, and (C) have con-
tinuous derivatives, and (D) are continuous solutions to the differential equation
with initial conditions y(0) = 0 and y 0 (0) = 0 and have continuous derivatives.
Do not solve the differential equation.
−tet − et + e2t if 0 ≤ t < 1
a) y(t) =  2t t
te − 2e if 1 ≤ t < ∞
−tet − et + e2t if 0 ≤ t < 1
b) y(t) =  2t
te − 3et − 12 e2t if 1 ≤ t < ∞
−tet − et + e2t if 0 ≤ t < 1
c) y(t) =  2t
te + et+1 − et − e2t − e2t−1 if 1 ≤ t < ∞
−tet + et − e2t if 0 ≤ t < 1
d) y(t) =  2t
te + et+1 + et − e2t−1 − 3e2t if 1 ≤ t < ∞
Solve the following differential equations.
t if 0 ≤ t < 1
19. y 0 + 3y =  y(0) = 0.
1 if 1 ≤ t < ∞,

0 if 0≤t<1


−1
t if 1≤t<2
20. y 0 − y = y(0) = 0.

3−t if 2≤t<3


0 if 3 ≤ t < ∞,
sin t if 0 ≤ t < π
21. y 0 + y =  y(π) = −1.
0 if π ≤ t < ∞
t if 0 ≤ t < 1
22. y 00 − y =  y(0) = 0, y 0 (0) = 1.
0 if 1 ≤ t < ∞,
0 if 0 ≤ t < 2
23. y 00 − 4y 0 + 4y =  y(0) = 1, y 0 (0) = 0
4 if 2 ≤ t < ∞
24. Suppose f is a piecewise continuous function on an interval [α, β]. Let a ∈ [α, β]
t
and define y(t) = y0 + a f (u) du. Show that y is a continuous solution to

y 0 = f (t) y(a) = y0 .
346 8 Laplace Transform II

25. Suppose f is a piecewise continuous function on an interval [α, β]. Let a ∈ [α, β]
t
and define y(t) = y0 + e−at a eau f (u) du. Show that y is a continuous solution
to
y 0 + ay = f (t) y(a) = y0 .
sin(1/t) if t 6= 0
26. Let f (t) = 
0 if t = 0.
a) Show that f is bounded.
b) Show that f is not continuous at t = 0.
c) Show that f is not piecewise continuous.

8.2 The Heaviside class H

In this section we will extend the definition of the Laplace transform beyond
the set of elementary functions E to include piecewise continuous functions.
The Laplace transform method will extend as well and provide a rather simple
means of dealing with the differential equations we saw in Section 8.1.
Since the Laplace transform
Z ∞
L {f } (s) = e−st f (t) dt
0

is defined by means of an improper integral, we must be careful about the issue

RN
of convergence. Recall that this definition means we compute 0 e−st f (t) dt
and then take the limit as N goes to infinity. To insure convergence we must
take into consideration the kinds of functions we feed into it. There are two
main reasons why such improper integrals may fail to exist. First, if the dis-
tribution of the discontinuities of f is ‘too bad’ then even the finite integral
R N −st
0
e f (t) dt may fail to exist. Second, if the finite integral exists the limit
as N goes to ∞ may not. This has to do with how fast f grows. These two
issues will be handled separately by (1) identifying the particular type of dis-
continuities allowed and (2) restricting the type of growth that f is allowed.
What will result is a class of functions large enough to handle most of the
applications one is likely to encounter.
The first issue is handled for us by restricting to the piecewise continuous
functions defined in section 8.1. If f is a piecewise continuous function (on
[0, ∞)) then so is t 7→ e−st f (t). By Proposition 4 the integral
Z N
e−st f (t) dt
0

exists and is a continuous function in the variable N . Now to insure conver-

gence as N goes to ∞ we must place a further requirement on f .
 8.2 The Heaviside class H 347

Functions of Exponential Type

RN
We now want to put further restrictions on f to assure that limN →∞ 0 e−st f (t) dt
exists. As we indicated this can be achieved by making sure that f doesn’t
grow too fast.
A function y = f (t) is said to be of exponential type if

|f (t)| ≤ Keat

for all t ≥ M , where M , K, and a are positive real constants. The idea here
is that functions of exponential type should not grow faster than a multiple
of an exponential function Keat . Visually, we require the graph of |f | to lie
below such an exponential function from some point on, t ≥ M , as illustrated
in Figure 8.10.

f
_
_

M
Fig. 8.10. The exponential function Keat eventually overtakes |f | for t ≥ M .

We note here that in the case where f is also piecewise continuous then f
is bounded on [0, M ] and one can find a constant K 0 such that

|f (t)| ≤ K 0 eat

for all t > 0.

The Heaviside class is the set H of all piecewise continuous functions
of exponential type. One can show it is closed under addition and scalar
multiplication. (see Exercise ??) It is to this class of functions that we extend
the Laplace transform. The set of elementary functions E that we introduced
348 8 Laplace Transform II

in Chapter 3 are all examples of functions in the Heaviside class. Recall that
f is an elementary function if f is a sum of functions of the form ctn eat sin(bt)
and ctn eat cos(bt), where a, b, c are constants and n is a nonnegative integer.
Such functions are continuous. Since sin and cos are bounded by 1 and tn ≤ ent
it follows that |ctn eat sin bt| ≤ ce(a+n)t and likewise |ctn eat cos bt| ≤ ce(a+n)t .
Thus elementary functions are of exponential type, i.e., E ⊂ H. Although the
Heaviside class is much bigger than the set of elementary functions there are
2
many important functions which are not in H. An example is f (t) = et . For
if b is any positive constant, then
2
et 2 b 2 b2

bt
= et −bt = e(t− 2 ) − 4
e
and therefore,
2
et
lim bt = ∞.
t→∞ e
2
This implies that f (t) = et grows faster than any exponential function and
thus is not of exponential type.

Existence of the Laplace transform

Recall that for an elementary function the Laplace transform exists and has
the further property that lims→∞ F (s) = 0. These two properties extend to
all functions in the Heaviside class.
Theorem 1. For f ∈ H the Laplace transform exists and
lim F (s) = 0.
s→∞
RN
Proof. The finite integral 0 e−st f (t) dt exists because f is piecewise contin-
uous on [0, N ]. Since f is also of exponential type there are constants K and
a such that |f (t)| ≤ Keat for all t ≥ 0. Thus, for all s > a,
Z ∞ Z ∞
−st
|e f (t)| dt ≤ |e−st Keat | dt
0 0
Z ∞
=K e−(s−a)t dt
0
K
= .
s−a
This shows that the integral converges absolutely and hence the Laplace trans-
K K
form exists for s > a. Since |L {f } (s)| ≤ s−a and lims→∞ s−a = 0 it follows
that
lim L {f } (s) = 0.
s→∞
u
t
 8.2 The Heaviside class H 349

As might be expected computations using the definition to compute

Laplace transforms of even simple functions can be tedious. To illustrate the
point consider the following example.
Example 2. Use the definition to compute the Laplace transform of
(
t2 if 0 ≤ t < 1
f (t) =
2 if 1 ≤ t < ∞.

I Solution. Clearly f is piecewise continuous and bounded, hence it is in

the Heaviside class. We can thus proceed with the definition confident, by
Theorem 1, that the improper integral will converge. We have
Z ∞
L {f } (s) = e−st f (t) dt
0
Z 1 Z ∞
= e−st t2 dt + e−st 2 dt
0 1

For the first integral we need integration by parts twice:

Z 1
t2 e−st 1 2 1 −st
Z
e−st t2 dt = |0 + e t dt
0 −s s 0
e−s 2 te−st 1 1 1 −st
 Z 
= + |0 + e dt
−s s −s s 0
−s
 −s 
e 2 e 1 −st 1
=− + − − 2 e |0
s s s s
−s −s
e 2e 2 2e−s
=− − 2 + 3− 3 .
s s s s
The second integral is much simpler and we get
Z ∞
2e−s
e−st 2 dt =
1 s

Now putting things together and simplifying gives

 
2 2 2 1
L {f } (s) = 3 + e−s − 3 − 2 + .
s s s s
J

Do not despair. The Heaviside function that we introduce next will lead
to a Laplace transform principle that will make unnecessary calculations like
the one above.
350 8 Laplace Transform II

The Heaviside Function

In order to effectively manage piecewise continuous functions in H it is useful
to introduce an important auxiliary function called the unit step function
or Heaviside function:
(
0 if 0 ≤ t < c,
hc (t) =
1 if c ≤ t.

The graph of this function is given in Figure 8.11.

Fig. 8.11. The Heaviside Function hc (t)

Clearly, it is piecewise continuous, and since it is bounded it is of exponen-

tial type. Thus hc ∈ H. Frequently we will write h(t) = h0 (t). Observe also
that hc (t) = h(t − c). More complicated functions can be built from the Heav-
iside function. First consider the model for an on-off switch, χ[a,b) , which is
1 (the on state) on the interval [a, b) and 0 (the off state) elsewhere. Its graph
is given in Figure 8.12. Observe that χ[a,b) = ha − hb and χ[a,∞) = ha . Now
using on-off switches we can easily describe functions defined piecewise.

Example 3. Write the piecewise defined function

(
t2 if 0 ≤ t < 1,
f (t) =
2 if 1 ≤ t < ∞.

in terms of on-off switches and in terms of Heaviside functions.

I Solution. In this piecewise function t2 is in the on state only in the in-

terval [0, 1) and 2 is in the on state only in the interval [1, ∞).Thus

f (t) = t2 χ[0,1) + 2χ[1,∞) .

 8.2 The Heaviside class H 351

a b
t
Fig. 8.12. The On/Off Switch χa,b (t)

Now rewriting the on-off switches in terms of the Heaviside functions we ob-
tain:

f (t) = t2 (h0 − h1 ) + 2h1

= t2 h0 + (2 − t2 )h1
= t2 + (2 − t2 )h(t − 1).

The Laplace Transform on the Heaviside class

The importance of writing piecewise continuous functions in terms of Heav-
iside functions is seen by the ease of computing its Laplace transform. For
simplicity when f ∈ H we will extend f by defining f (t) = 0 when t < 0. This
extension does not effect the Laplace transform for the Laplace transform only
involves values f (t) for t > 0.
Theorem 4 (The Second Translation Principle). Suppose f ∈ H is a
function with Laplace transform F . Then

L {f (t − c)h(t − c)} = e−sc F (s).

In terms of the inverse Laplace transform this is equivalent to

L−1 e−sc F (s) = f (t − c)h(t − c).



Proof. The calculation is straightforward and involves a simple change of vari-

ables:
352 8 Laplace Transform II
Z ∞
L {f (t − c)h(t − c)} (s) = e−st f (t − c)h(t − c) dt
Z0 ∞
= e−st f (t − c) dt
Zc ∞
= e−s(t+c) f (t) dt (t 7→ t + c)
0
Z ∞
−sc
=e e−st f (t) dt
0
= e−sc F (s)

u
t

Frequently, we encounter expressions in the form g(t)h(t−c). If f (t) is replaced

by g(t + c) in Theorem 4 then we obtain
Corollary 5.
L {g(t)h(t − c)} = e−sc L {g(t + c)} .
−sc
A simple example of this is when g = 1. Then L {hc } = e−sc L {1} = e s .
When c = 0 then L {h0 )} = 1s which is the same as the Laplace transform of
the constant function 1. This is consistent since h0 = h = 1 for t ≥ 0.
(
t2 if 0 ≤ t < 1
Example 6. Find the Laplace transform of f (t) =
2 if 1 ≤ t < ∞
given in Example 8.2.

I Solution. In Example 3 we found f (t) = t2 + (2 − t2 )h(t − 1). By the

Corollary we get
2 −s
 2

L {f } = + e L 2 − (t + 1)
s3
2
= 3 + e−s L −t2 − 2t + 1

s  
2 −s 2 2 1
= 3 +e − 3− 2+
s s s s
J

Example 7. Find the Laplace transform of


cos t if 0 ≤ t < π

f (t) = 1 if π ≤ t < 2π
if 2π ≤ t < ∞.

0


I Solution. First writing f in terms of on-off switches gives

f = cos t χ[0,π) + 1 χ[π,2π) + 0 χ[2π,∞) .

 8.2 The Heaviside class H 353

Now rewrite this expression in terms of Heaviside functions:

f = cos t (h0 − hπ ) + (hπ − h2π ) = cos t + (1 − cos t)hπ − h2π .

Since hc (t) = h(t − c) the corollary gives

s e−2sπ
F (s) = + e−sπ L {1 − cos(t + π)} −
s2
+1 s
−2sπ
 
s 1 s e
= 2 + e−sπ + − .
s +1 s s2 + 1 s

In the second line we have used the fact that cos(t + π) = − cos t. J

Exercises
Graph each of the following functions defined by means of the unit step function
h(t − c) and/or the on-off switches χ[a, b) .
1. f (t) = 3h(t − 2) − h(t − 5)
2. f (t) = 2h(t − 2) − 3h(t − 3) + 4h(t − 4)
3. f (t) = (t − 1)h(t − 1)
4. f (t) = (t − 2)2 h(t − 2)
5. f (t) = t2 h(t − 2)
6. f (t) = h(t − π) sin t
7. f (t) = h(t − π) cos 2(t − π)
8. f (t) = t2 χ[0, 1) + (1 − t)χ[1, 3) + 3χ[3, ∞)
For each of the following functions f (t), (a) express f (t) in terms of on-off switches,
(b) express f (t) in terms of Heaviside functions, and (c) compute the Laplace trans-
form F (s) = L {f (t)}.

0 if 0 ≤ t < 2,
9. f (t) = 
t−2 if 2 ≤ t < ∞.
0 if 0 ≤ t < 2,
10. f (t) = 
t if 2 ≤ t < ∞.
0 if 0 ≤ t < 2,
11. f (t) = 
t+2 if 2 ≤ t < ∞.
0 if 0 ≤ t < 4,
12. f (t) = 
(t − 4)2 if 4 ≤ t < ∞.
0 if 0 ≤ t < 4,
13. f (t) =  2
t if 4 ≤ t < ∞.
0 if 0 ≤ t < 4,
14. f (t) =  2
t −4 if 4 ≤ t < ∞.
354 8 Laplace Transform II

0 if 0 ≤ t < 2,
15. f (t) = 
(t − 4)2 if 2 ≤ t < ∞.
0 if 0 ≤ t < 4,
16. f (t) =  t−4
e if 4 ≤ t < ∞.
0 if 0 ≤ t < 4,
17. f (t) =  t
e if 4 ≤ t < ∞.
0 if 0 ≤ t < 6,
18. f (t) =  t−4
e if 6 ≤ t < ∞.
0 if 0 ≤ t < 4,
19. f (t) =  t
te if 4 ≤ t < ∞.

1 if 0 ≤ t < 4
20. f (t) = −1 if 4 ≤ t < 5


0 if 5 ≤ t < ∞.
t
 if 0 ≤ t < 1
21. f (t) = 2 − t if 1 ≤ t < 2


1 if 2 ≤ t < ∞.
t if 0 ≤ t < 1
22. f (t) = 
2−t if 1 ≤ t < ∞.

t if 0 ≤ t < 1

t − 1

 if 1 ≤ t < 2
23. f (t) = t−2 if 2 ≤ t < 3




..
. .
1 if 2n ≤ t < 2n + 1
24. f (t) = 
0 if 2n + 1 ≤ t < 2n + 2.
t

2
if 0 ≤ t < 2
25. f (t) = 4 if 2 ≤ t < 3


7−t if 3 ≤ t < ∞.

1−t if 0 ≤ t < 2

3 − t

 if 2 ≤ t < 4
26. f (t) = 5 − t if 4 ≤ t < 6




..
. .

1 if 0 ≤ t < 2


3 − t if 2 ≤ t < 3
27. f (t) =
2(t − 3)
 if 3≤t<4


2 if 4 ≤ t < ∞.

8.3 The Inversion of the Laplace Transform

We now turn our attention to the inversion of the Laplace transform. In
Chapter 3 we established a one-to-one correspondence between elementary
 8.3 The Inversion of the Laplace Transform 355

functions and proper rational functions: for each proper rational function its
inverse Laplace transform is a unique elementary function. For the Heaviside
class the matter is complicated by our allowing discontinuity. Two functions
f1 and f2 are said to be essentially equal if for each interval [0, N ) they are
equal as functions except at possibly finitely many points. For example, the
functions

1 if 0 ≤ t < 1
( (
1 if 0 ≤ t < 1 1 if 0 ≤ t ≤ 1

f1 (t) = f2 (t) = 3 if t = 1 f3 (t) =
2 if 1 ≤ t < ∞  2 if 1 < t < ∞.
2 if 1 < t < ∞


are essentially equal for they are equal everywhere except at t = 1. Two
functions that are essentially equal have the same Laplace transform. This is
because the Laplace transform is an integral operator and integration cannot
distinguish functions that are essentially equal. The Laplace transform of f1 ,
−s
f2 , and f3 in our example above are all 1s + e s . Here is our problem: Given a
−s
transform, like 1s + e s , how do we decide what ‘the’ inverse Laplace transform
is. It turns out that if F (s) is the Laplace transform of functions f1 , f2 ∈ H
then f1 and f2 are essentially equal. For most practical situations it does not
matter which one is chosen. However, in this text we will consistently use the
one that is right continuous at each point. A function f in the Heaviside class
is said to be right continuous at a point a if we have

f (a) = f (a+ ) = lim+ f (t),

t→a

and it is right continuous on [0, ∞) if it is right continuous at each point

in [0, ∞). In the example above, f1 is right continuous while f2 and f3 are not.
The function f3 is, however, left continuous, using the obvious definition of
left continuity. If we decide to use right continuous functions in the Heaviside
class then the correspondence with its Laplace transform is one-to-one. We
summarize this discussion as a theorem:
Theorem 1. If F (s) is the Laplace transform of a function in H then there
is a unique right continuous function f ∈ H such that L {f } = F . Any two
functions in H with the same Laplace transform are essentially equal.
Recall from our definition that hc is right continuous. So piecewise func-
tions written as sums of products of a continuous function and a Heaviside
function are right continuous.
Example 2. Find the inverse Laplace transform of

e−s e−3s
F (s) = 2
+
s s−4
and write it as a right continuous piecewise function.
356 8 Laplace Transform II

1
I Solution. The inverse Laplace transforms of s12 and s−4 are, respectively,
4t
t and e . By Theorem 4 the inverse Laplace transform of F (s) is

(t − 1)h1 + e4(t−3) h3 .

On the interval [0, 1) both t − 1 and e4(t−3) are off. On the interval [1, 3) only
t − 1 is on. On the interval [3, ∞) both t − 1 and e4(t−3) are on. Thus

0
 if 0 ≤ t < 1
−1
L {F (s)} = t − 1 if 1 ≤ t < 3 .
t − 1 + e4(t−3) if 3 ≤ t < ∞



The Laplace Transform of tα and the Gamma

function
n!
We showed in Chapter 3 that the Laplace transform of tn is sn+1 , for each
nonnegative integer n. One might conjecture that the Laplace transform of
tα , for α an arbitrary nonnegative real number, is given by a similar formula.
Such a formula would necessarily extend the notion of ‘factorial’. We define
the gamma function by the formula
Z ∞
Γ (α) = e−t tα−1 dt.
0

It can be shown that the improper integral that defines the gamma function
converges as long as α is greater than 0. The following proposition, whose proof
is left as an exercise, establishes the fundamental properties of the gamma
function.

Proposition 3.
1. Γ (α + 1) = αΓ (α) (The fundamental recurrence relation)
2. Γ (1) = 1
3. Γ (n + 1) = n!

The third formula in the proposition allows us to rewrite the Laplace transform
of tn in the following way:

Γ (n + 1)
L {tn } = .
sn+1
If α > −1 we obtain
Γ (α + 1)
L {tα } = .
sα+1
 8.3 The Inversion of the Laplace Transform 357

(Even though tα is not in the Heaviside class for −1 < α < 0 its Laplace
transform still exists.) To establish this formula fix α > −1. By definition
Z ∞
L {tα } = e−st tα dt.
0

We make the change of variable u = st. Then du = sdt and

Z ∞
u du
L {tα } (s) = e−u
0 s s
Z ∞
1
= α+1 e−u uα du
s 0
Γ (α + 1)
= .
sα+1
Of course, in order to actually compute the Laplace transform of some non
integer positive power of t one must know the value of the gamma function √
for the corresponding power. For example, it is known that Γ ( 21 ) = π. By
√
the fundamental recurrence relation Γ ( 23 ) = 12 Γ ( 12 ) = 2π . Therefore
√
n 1o π
L t2 = 3 .
2s 2

Exercises
Compute the inverse Laplace transform of each of the following functions.
e−3s
1.
s−1
e−3s
2.
s2
e−3s
3.
(s − 1)3
e−πs
4.
s2 + 1
se−3πs
5.
s2 + 1
e−πs
6. 2
s + 2s + 5
e−s e−2s
7. +
s2 (s − 1)3
e−2s
8.
s2 + 4
e−2s
9.
s2 − 4
se−4s
10. 2
s + 3s + 2
358 8 Laplace Transform II

e−2s + e−3s
11.
s2 − 3s + 2
1 − e−5s
12.
s2
1 + e−3s
13.
s4
2s + 1
14. e−πs 2
s + 6s + 13
2s + 1
15. 1 − e−πs  2
 s + 6s + 13

8.4 Properties of the Laplace Transform

Many of the properties of the Laplace transform that we discussed in Chapter
3 for elementary functions carry over to the Heaviside class. Their proofs are
the same. These properties are summarized below.
Linearity L {af + bg} = aL {f } + bL {g} .

The First Translation Principle L {e−at f } = L {f } (s − a).

Differentiation in Transform Space L(−tf (t)) = F 0 (s)

L {(−t)n f (t)} = F (n) (s).
nR o
t F (s)
Integration in Domain Space L( 0 f (u) du = s .

There are a few properties though that need some clarifications. In partic-
ular, we need to discuss the meaning of the fundamental derivative formula

L {f 0 } = sL {f } − f (0),

when f is in the Heaviside class. You will recall that the derivative of an ele-
mentary function is again an elementary function. However, for the Heaviside
class this is not necessarily the case. A couple of things can go wrong. First,
there are examples of functions in H for which the derivative does not exist
at any point. Second, even when the derivative exists there is no guarantee
that it is back in H. As an example, consider the function
2
f (t) = sin et .

This function is in H because it is bounded (between −1 and 1) and contin-

uous. However, its derivative is
2 2
f 0 (t) = 2tet cos et ,
2
which is continuous but not of exponential type. To see this recall that et is
2
not of exponential type. Thus at those values of t where cos et = 1, |f 0 (t)| is
 8.4 Properties of the Laplace Transform 359

not bounded by an exponential function and hence f 0 ∈ / H. Therefore, in order

to extend the derivative formula to H we must include in the hypotheses the
requirement that both f and f 0 be in H. Recall that for f in H the symbol
f 0 is used to denote the derivative of f if f is differentiable except at a finite
number of points on each interval of the form [0, N ].

The Laplace Transform of a Derivative

With these understandings we now have
Theorem 1. If f is continuous and f and f 0 are in H then

L {f 0 } = sL {f } − f (0).
RN
Proof. We begin by computing 0 e−st f 0 (t) dt. This integral requires that we
consider the points where f 0 is discontinuous. There are only finitely many
on [0, N ), a1 , . . . , ak , say, and we may assume ai < ai+1 . If we et a0 = 0 and
ak+1 = N then we obtain
Z N k Z
X ai+1
−st 0
e f (t) dt = e−st f 0 (t) dt,
0 i=0 ai

and integration by parts gives

Z N k 
X Z ai+1 
e−st f 0 (t) dt = f (t)e−st |aai+1
i
+s e−st f (t) dt
0 i=0 ai
k
X Z N
= (f (a−
i+1 )e
−sai+1
− f (a+
i )e
−sai
) + e−st f (t) dt
i=0 0
Z N
= f (N )e−N s − f (0) + s e−st f (t) dt.
0

We have used the continuity of f to make the evaluations at ai and ai+1 ,

which allows for the collapsing sum in line 2. We now take the limit as N goes
to infinity and the result follows. u
t

The following corollary is immediate:

Corollary 2. If f and f 0 are continuous and f , f 0 , and f 00 are in H then

L {f 00 } = s2 L {f } − sf (0) − f 0 (0).
360 8 Laplace Transform II

The Laplace Transform Method

The differential equations that we will solve by means of the Laplace transform
are first and second order constant coefficient linear differential equations with
a forcing function f in H:
y 0 + ay = f (t)
y + ay 0 + by = f (t).
00

In order to apply the Laplace transform method we will need to know

that there is a solution y which is continuous in the first equation and both
y and y 0 are continuous in the second equation. These facts were proved in
Theorems 6 and 8.
We are now in a position to illustrate the Laplace transform method to
solve differential equations with possibly discontinuous forcing functions f .
Example 3. Solve the following first order differential equation:
y 0 + 2y = f (t), y(0) = 1,
where (
0 if 0 ≤ t < 1
f (t) =
t if 1 ≤ t < ∞.
I Solution. We first rewrite f in terms of Heaviside functions: f (t) =
t χ[1,∞) (t) = t h1 (t). By Corollary 5 its Laplace transform is F (s) = e−s L {t + 1} =
e−s ( s12 + 1s ) = e−s ( s+1
s2 ). The Laplace transform of the differential equation
yields
s+1
sY (s) − y(0) + 2Y (s) = e−s ( 2 ),
s
and solving for Y gives
1 s+1
Y (s) = + e−s 2 .
s+2 s (s + 2)
A partial fraction decomposition gives
s+1 1 1 1 1 1 1
= + − ,
s2 (s + 2) 4 s 2 s2 4 s+2
and the second translation principle (Theorem 4) gives
       
1 1 1 1 1 1 1
y(t) = L−1 + L−1 e−s + L−1 e−s 2 − L−1 e−s
s+2 4 s 2 s 4 s+2
1 1 1
= e−2t + h1 + (t − 1)h1 − e−2(t−1) h1 .
( 4 2 4
e−2t if 0 ≤ t < 1
= −2t 1 1 −2(t−1)
e + 4 (2t − 1) − 4 e if 1 ≤ t < ∞.
J
 8.4 Properties of the Laplace Transform 361

We now consider a mixing problem of the type mentioned in the introduc-

tion to this chapter.
Example 4. Suppose a tank holds 10 gallons of pure water. There are two
input sources of brine solution: the first source has a concentration of 2 pounds
of salt per gallon while the second source has a concentration of 3 pounds of
salt per gallon. The first source flows into the tank at a rate of 1 gallon per
minute for 5 minutes after which it is turned off and simultaneously the second
source is turned on at a rate of 1 gallon per minute. The well mixed solution
flows out of the tank at a rate of 1 gallon per minute. Find the amount of salt
in the tank at any time t.
I Solution. The principles we considered in Chapter 1.1 apply here:

y 0 (t) = Rate in − Rate out.

Recall that the input and output rates of salt are the product of the concen-
tration of salt and the flow rates of the solution. The rate at which salt is
input depends on the interval of time. For the first five minutes, source one
inputs salt at a rate of 2 lbs per minute, and after that, source two inputs
salt at a rate of 3 lbs per minute. Thus the input rate is represented by the
function (
2 if 0 ≤ t < 5
f (t) =
3 if 5 ≤ t < ∞.

The rate at which salt is output is y(t)

10 lbs per minute. We therefore have the
following differential equation and initial condition:
y(t)
y 0 = f (t) − , y(0) = 0.
10
Rewriting f in terms of Heaviside functions gives f = 2χ[0,5) + 3χ[5,∞) =
2(h0 − h5 ) + 3h5 = 2 + h5 . Applying the Laplace transform to the differential
equation and solving for Y (s) = L {y} (s) gives
2 + e−s
  
1
Y (s) = 1
s + 10 s
2 1
= 1 + e−5s 1
(s + 10 )s (s + 10 )s
20 20 10 10
= − 1 + e−5s − e−5s 1 .
s s + 10 s s + 10

Taking the inverse Laplace transform of Y (s) gives

t t−5
y(t) = 20 − 20e− 10 + 10h5 (t) − 10e− 10 h5 (t)
( t
20 − 20e− 10 if 0 ≤ t < 5
= t t−5
30 − 20e− 10 − 10e− 10 if 5 ≤ t < ∞.
362 8 Laplace Transform II

The graph of y is given in Figure 8.13. As expected we observe that the

solution is continuous, but the small kink at t = 5 indicates that there is a
discontinuity of the derivative at this point. This occurred when the flow of
the second source, which had a higher concentration of salt, was turned on.
J

30

25

20

15

10

0 10 20 30 40 50
t
Fig. 8.13. The solution to a mixing problem with discontinuous input function.

Exercises
Solve each of the following initial value problems.
0 if 0 ≤ t < 1
1. y 0 + 2y = f (t) where f (t) =  y(0) = 0.
−3 if t ≥ 1
−2 if 0 ≤ t < 1
2. y 0 + 2y = f (t) where f (t) =  y(0) = 0.
2 if t ≥ 1
0
 if 0 ≤ t < 1
3. y 0 + 2y = f (t) where f (t) = 2 if 1 ≤ t < 3 y(0) = 0.


0 if t ≥ 3
t if 0 ≤ t < 1
4. y 0 + 2y = f (t) where f (t) =  y(0) = 0.
0 if t ≥ 1
5. y 00 + 9y = h(t − 3), y(0) = 0, y 0 (0) = 0.
1 if 0 ≤ t < 5
6. y 00 − 5y 0 + 4y = f (t) where f (t) =  y(0) = 0, y 0 (0) = 1.
0 if t ≥ 5
0
 if 0 ≤ t < 1
7. y 00 + 5y 0 + 6y = 2 if 1 ≤ t < 3 y(0) = 0, y 0 (0) = 0.


0 if t ≥ 3
 8.5 The Dirac Delta Function 363

8. y 00 + 9y = h(t − 2π) sin t, y(0) = 1, y 0 (0) = 0.

9. y 00 + 2y 0 + y = h(t − 3), y(0) = 0, y 0 (0) = 1.
10. y 00 + 2y 0 + y = h(t − 3)et , y(0) = 0, y 0 (0) = 1.
11. y 00 + 6y 0 + 5y = 1 − h(t − 2) + h(t − 4) + h(t − 6), y(0) = 0, y 0 (0) = 0.

8.5 The Dirac Delta Function

In applications we may encounter an input into a system we wish to study that
is very large in magnitude, but applied over a short period of time. Consider,
for example, the following mixing problem:
Example 1. A tank holds 10 gallons of a brine solution in which each gallon
contains 2 pounds of dissolved salt. An input source begins pouring fresh
water into the tank at a rate of 1 gallon per minute and the thoroughly mixed
solution flows out of the tank at the same rate. After 5 minutes 3 pounds of
salt are poured into the tank where it instantly mixes into the solution. Find
the amount of salt at any time t.
This example introduces a sudden action, namely, the sudden input of 3
pounds of salt at time t = 5 minutes. If we imagine that it actually takes 1
second to do this then the average rate of input of salt would be 3 lbs/ sec =
180 lbs/min. Thus we see a high magnitude in the rate of input of salt over
a short interval. Moreover, the rate multiplied by the duration of input gives
the total input.
More generally, if r(t) represents the rate of input over a time interval
Rb
[a, b] then a r(t) dt would represent the total input over that interval. A unit
input means that this integral is 1. Let t = c ≥ 0 be fixed and let  be a small
positive number. Imagine a constant input rate over the interval [c, c + ) and
0 elsewhere. The function dc, = 1 χ[c,c+) represents such an input rate with
constant input ( 1 ) over the interval [c, c + ) ( c.f. section 8.2 where the on-off
switch χ[a,b) is discussed). The constant 1 is chosen so that the total input is
∞ c+
1 1
Z Z
dc, dt = 1 dt =  = 1.
0  c 
1
For example, if  = 60 min, then 3d5, would represent the input of 3 lbs of
salt over a 1 second interval beginning at t = 5.
Figure 8.14 shows the graphs of dc, for a few values of . The main idea
will be to take smaller and smaller values of , i.e. we want to imagine the
total input being concentrated at the point c. Formally, we define the Dirac
delta function by δc (t) = lim→0+ dc, (t). Heuristically, we would like to
write (
∞ if t = c
δc (t) =
0 elsewhere,
364 8 Laplace Transform II

10

2
1

6s 30s 60s
t
Fig. 8.14. Approximation to a delta function

R∞ R∞
with the property that 0 δc (t) dt = lim→0 0 dc, dt = 1. Of course, there
is really no such function with this property. (Mathematically, we can make
precise sense out of this idea by extending the Heaviside class to a class that
includes distributions or generalized functions. We will not pursue dis-
tributions here as it will take us far beyond the introductory nature of this
text.) Nevertheless, this is the idea we want to develop, at least formally. We
will consider first order constant coefficient differential equations of the form
y 0 + ay = f (t)
where f involves the Dirac delta function δc . It turns out that the main prob-
lem lies in the fact that the solution is not continuous, so Theorem 1 does not
apply. Nevertheless, we will justify that we can apply the usual Laplace trans-
form method in a formal way to produce the desired solutions. The beauty
of doing this is found in the ease in which we can work with the "Laplace
transform" of δc .
We define the Laplace transform of δc by the formula:
L {δc } = lim L {dc, } .
→0

Theorem 2. The Laplace transform of δc is

L {δc } = e−cs .
Proof. We begin with dc, .
1
L {dc, } = L {hc − hc+ }

1 e−cs − e−(c+)s
 
=
 s
−cs
1 − e−s
 
e
= .
s 
 8.5 The Dirac Delta Function 365

We now take limits as  goes to 0 and use L’Hospitals rule to obtain:

e−cs 1 − e−s e−cs
 
L {δc } = lim L {dc, } = lim = · s = e−cs .
→0 s →0  s
u
t

We remark that when c = 0 we have L {δ0 } = 1. By Theorem 1 there

is no Heaviside function with this property. Thus, to reiterate, even though
L {δc } is a function, δc is not. We will frequently write δ = δ0 . Observe that
δc (t) = δ(t − c).
The mixing problem from Example 1 gives rise to a first order linear dif-
ferential equation involving the Dirac delta function.
I Solution. Let y(t) be the amount of salt in the tank at time t. Then
y(0) = 20 and y 0 is the difference of the input rate and the output rate. The
only input of salt occurs at t = 5. If the salt were input over a small interval,
[5, 5 + ) say, then 3 χ[5,5+) would represent the input of 3 pounds of salt over
a period of  minutes. If we let  go to zero then 3δ5 would represent the input
rate. The output rate is y(t)/10. We are thus led to the differential equation:
y
y0 + = 3δ5 , y(0) = 20.
10
J

The solution to this differential equation will fall out of the slightly more
general discussion we give below.

Differential Equations of the form y 0 + ay = kδc

We will present progressively four methods for solving

y 0 + ay = kδc , y(0) = y0 . ?

The last method, the formal Laplace Transform Method, is the simplest
method and is, in part, justified by the methods that precede it. The formal
method will thereafter be used to solve equations of the form ? and will work
for all the problems introduced in this section. Keep in mind though that in
practice a careful analysis of the limiting processes involved must be done to
determine the validity of the formal Laplace Transform method.
Method 1. In our first approach we solve the equation
k
y 0 + ay = χ[c,c+), y(0) = y0

and call the solution y . We let y(t) = lim→0 y . Then y(t) is the solution to
y 0 + ay = kδc , y(0) = y0 . Recall from Exercise ?? the solution to
366 8 Laplace Transform II

y 0 + ay = Aχ[α, β), y(0) = y0 ,

is 
0 if 0 ≤ t < α
A
y(t) = y0 e−at + 1 − e−a(t−α) if α ≤ t < β
a
 −a(t−β)
e − e−a(t−α) if β ≤ t < ∞.
We let A = k , α = c, and β = c +  to get

0 if 0 ≤ t < c
k 
y (t) = y0 e−at + 1 − e−a(t−c) if c ≤ t < c + 
a 
 −a(t−c−)
e − e−a(t−c) if c +  ≤ t < ∞.

The computation of lim→0 y is done on each interval separately. If 0 ≤ t ≤ c

then y = y0 e−at is independent of  and hence

lim y (t) = y0 e−at 0 ≤ t ≤ c.

→0

If c < t < ∞ then for  small enough, c +  < t and thus

k −a(t−c−) k ea − 1
y (t) = y0 e−at + (e − e−a(t−c) ) = y0 e−at + e−a(t−c) .
a a 
Therefore
lim y (t) = y0 e−at + ke−a(t−c) c < t < ∞.
→0

We thus obtain
(
y0 e−at if 0 ≤ t ≤ c
y(t) =
y0 e−at + ke−a(t−c) if c < t < ∞.

In the mixing problem above the infusion of 3 pounds of salt after five
minutes will instantaneously increase the amount of salt by 3; a jump dis-
continuity at t = 5. This is seen in the solution y above. At t = c there is a
jump discontinuity of jump k. Of course, the solution to the mixing problem
1
is obtained by setting a = 10 , k = 3, c = 5, and y0 = 20:
( t
20e− 10 if 0 ≤ t ≤ 5
y(t) = t t−5
20e− 10 + 3e− 10 if 5 < t < ∞,

whose graph is given in Figure 8.15. We observe that y(5− ) = 20e−1/2 ' 12.13
and y(5+ ) = 20e−1/2 + 3 ' 15.13. Also notice that y(5+ ) is y(5− ) plus the
jump 3.
Method 2. Our second approach realizes that the mixing problem stated
1
above can be thought of as the differential equation, y 0 + 10 y = 0, defined on
two separate intervals; (1) on the interval [0, 5) with initial value y(0) = 20
 8.5 The Dirac Delta Function 367

20
18
16
14
12
y 10
8
6
4
2
0 2 4 6 8 10 12 14 16 18 20
t
Fig. 8.15. Graph of the Solution to the Mixing Problem

and (2) on the interval [5, ∞) where the initial value y(5) is the value of the
solution given in part (1) at t = 5, plus the jump 3. We apply this idea to our
more generic initial value problem, Equation ?.
On the interval [0, c) we solve y 0 + ay = 0 with initial value y(0) = y0 . The
general solution is easily seen to be y = be−at . The initial value y(0) = y0
gives b = y0 . The solution on [0, c) is thus
y = y0 e−at .
On the interval [c, ∞) we solve y 0 +ay = 0 with initial value y(c) = y0 e−ac +k.
(y(c) is the value of the solution just obtained at t = c plus the jump k.)
Again the general solution is y = be−at and the initial condition implies
be−ac = y0 e−ac + k. Solving for b gives b = y0 + keac . Thus
y = y0 e−at + ke−a(t−c) ,
on the interval [c, ∞). Piecing these two solutions together yields
(
y0 e−at if 0 ≤ t < c
y= −at −a(t−c)
,
y0 e + ke if c ≤ t < ∞

which, as it should be, is the same solution we obtained by method 1.

Method 3. In this method we want to focus on the differential equation,
y 0 + ay = 0 on the entire interval [0, ∞) with the a priori knowledge that
there is a jump discontinuity at t = c. Recall from Theorem 1 that when y is
continuous and both y and y 0 are in H we have the formula
L {y 0 } (s) = sY (s) − y(0).
We cannot apply this theorem as stated for y is not continuous. But if y has
a single jump discontinuity at t = c we can prove a slight generalization of
Theorem 1.
368 8 Laplace Transform II

Theorem 3. Suppose y and y 0 are in H and y is continuous except for one

jump discontinuity at t = c with jump k. Then

L {y 0 } (s) = sY (s) − y(0) − ke−cs .

Proof. Let N > c. Then integration by parts gives

Z N Z c Z N
e−st y 0 (t) dt = e−st y(t) dt + e−st y(t) dt
0 0 c
Z c Z N
= e−st y(t)|c0 + s e−st y(t) dt + e−st y(t)|N
c +s e−st y(t) dt
0 c
Z N
=s e−st y(t) dt + e−sN y(N ) − y(0) − e−sc (y(c+ ) − y(c− ).
0

We take the limit as N goes to infinity and obtain:

L {y 0 } = sL {y} − y(0) − ke−sc .

u
t

We apply this theorem to the initial value problem

y 0 + ay = 0, y(0) = y0

with the knowledge that the solution y has a jump discontinuity at t = c with
jump k. Apply the Laplace transform to to the differential equation to obtain:

sY (s) − y(0) − ke−ac + aY (s) = 0.

Solving for Y gives

y0 e−as
Y (s) =
+k .
s+a s+a
Applying the inverse Laplace transform gives the solution

y(t) = y0 e−at + ke−a(t−c) hc (t)

(
y0 e−at if 0 ≤ t < c
=
y0 e−at + ke−a(t−c) if c ≤ t < ∞.

Method 4: The Formal Laplace Transform Method. We now return

to the differential equation

y 0 + ay = kδc , y(0) = y0

and apply the Laplace transform method directly. That we can do this is partly
justified by method 3 above. From Theorem 2 the Laplace transform of kδc is
ke−sc . This is precisely the term found in Theorem 3 where the assumption of
a single jump discontinuity is assumed. Thus the presence of kδc automatically
 8.6 Impulse Functions 369

encodes the jump discontinuity in the solution. Therefore we can (formally)

proceed without any advance knowledge of jump discontinuities. The Laplace
transform of
y 0 + ay = kδc , y(0) = y0
gives
sY (s) − y(0) + kY (s) = ke−sc
and one proceeds as at the end of method 3 to get
(
y0 e−at if 0 ≤ t < c
y(t) = −at −a(t−c)
y0 e + ke if c ≤ t < ∞.

8.6 Impulse Functions

An impulsive force is a force with high magnitude introduced over a short
period of time. For example, a bat hitting a ball or a spike in electricity
on an electric circuit both involve impulsive forces and are best represented
by the Dirac delta function. In this section we will consider the effect of the
introduction of impulsive forces into such systems and how they lead to second
order differential equations of the form

my 00 + µy 0 + ky = Kδc (t).

As we will soon see the effect of an impulsive force introduces a discontinuity

not in y but its derivative y 0 .
If F (t) represents a force which is 0 outside a time interval [a, b] then
R∞ Rb
0 F (t) dt = a F (t) dt represents the total impulse of the force F (t) over
that interval. A unit impulse means that this integral is 1. If F is given by
the acceleration of a constant mass then F (t) = ma(t), where m is the mass
and a(t) is the acceleration. The total impulse
Z b Z b
F (t) dt = ma(t) dt = mv(b) − mv(a)
a a

represents the change of momentum. (Momentum is the product of mass and

velocity). Now imagine this force is introduced over a very short period of
time, or even instantaneously. As in the previous section, we could model the
force by dc, = 1 χ[c,c+) and one would naturally be lead to the Dirac delta
function to represent the instantaneous change of momentum. Since momen-
tum is proportional to velocity we see that such impacts lead to discontinuities
in the derivative y 0 .
370 8 Laplace Transform II

Example 1. (see Chapter ?? for a discussion of spring-mass-dashpot sys-

tems) A spring is stretched 49 cm when a 1 kg mass is attached. The body is
pulled to 10 cm below its spring-body equilibrium and released. We assume
the system is frictionless. After 3 sec the mass is suddenly struck by a hammer
in a downward direction with total impulse of 4 kg·m/sec. Find the motion of
the mass.

I Solution. We will work in units of kg, m, and sec. Thus the spring con-
49
stant k is given by 1(9.8) = k 100 , so that k = 20. The initial conditions are
0
given by y(0) = .10 and y (0) = 0, and since the system is frictionless the
rewritten initial value problem is

y 00 + 20y = 4δ3 , y(0) = .10, y 0 (0) = 0.

We will return to the solution of this problem after we discuss the more
general second order case.

Differential Equations of the form

y 00 + ay 0 + by = Kδc
Our goal is to solve

y 00 + ay 0 + by = Kδc , y(0) = y0 , y 0 (0) = y1 ?

using the formal Laplace transform method that we discussed in Method 4 of

Section 8.5.
As we discussed above the effect of Kδc is to introduce a single jump
discontinuity in y 0 at t = c with jump K. Therefore the solution to (?) is
equivalent to solving
y 00 + ay 0 + by = 0
with the advanced knowledge that y 0 has a jump discontinuity at t = c. If we
apply Theorem 3 to y 0 we obtain

L {y 00 } = sL {y 0 } − y 0 (0) − Ke−sc
= s2 Y (s) − sy(0) − y 0 (0) − Ke−sc

Therefore, the Laplace transform of y 00 + ay 0 + by = 0 leads to

(s2 + as + b)Y (s) − sy(0) − y 0 (0) − Ke−sc = 0.

On the other hand, if we (formally) proceed with the Laplace transform of

Equation (?) without foreknowledge of discontinuities we obtain the equivalent
equation
(s2 + as + b)Y (s) − sy(0) − y 0 (0) = Ke−sc .
 8.6 Impulse Functions 371

Again, the Dirac function δc encodes the jump discontinuity automatically. If

we proceed as usual we obtain

sy(0) + y 0 (0) Ke−sc

Y (s) = 2
+ 2 .
s + as + b s + as + b
The inversion will depend on the way the characteristic polynomial factors.

We now return to the example given above. The equation we wish to solve
is
y 00 + 20y = 4δ3 , y(0) = .10, y 0 (0) = 0.
I Solution. We apply the formal Laplace transform to obtain

.1s e−3s
Y (s) = + .
s2 + 20 s2 + 20
The inversion gives
1 √ 1 √
y(t) = cos( 20 t) + √ sin( 20 (t − 3))h3 (t)
10 20
(
1 √ 0 if 0 ≤ t < 3
= cos( 20 t) + 1
√
10 √
20
sin( 20 (t − 3)) if 3 ≤ t < ∞.

Figure 8.16 gives the graph of the solution. You will note that y is continuous

0.15

0.1

0.05

0 2 4 6 8 10
x
–0.05

–0.1

–0.15

Fig. 8.16. Harmonic motion with impulse function

but the little kink at t = 3 indicates the discontinuity of y 0 . This is precisely

when the impulse to the system was delivered. J
372 8 Laplace Transform II

Exercises
Solve each of the following initial value problems.
1. y 0 + 2y = δ1 (t), y(0) = 0
2. y 0 + 2y = δ1 (t), y(0) = 1
3. y 0 + 2y = δ1 (t) − δ3 (t), y(0) = 0
4. y 00 + 4y = δπ (t), y(0) = 0, y 0 (0) = 1
5. y 00 + 4y = δπ (t) − δ2π (t), y(0) = 0, y 0 (0) = 0
6. y 00 + 4y = δπ (t) − δ2π (t), y(0) = 1, y 0 (0) = 0
7. y 00 + 4y 0 + 4y = 3δ1 (t), y(0) = 0, y 0 (0) = 0
8. y 00 + 4y 0 + 4y = 3δ1 (t), y(0) = −1, y 0 (0) = 3
9. y 00 + 4y 0 + 5y = 3δ1 (t), y(0) = 0, y 0 (0) = 0
10. y 00 + 4y 0 + 5y = 3δ1 (t), y(0) = −1, y 0 (0) = 3
11. y 00 + 4y 0 + 20y = δπ (t) − δ2π (t), y(0) = 1, y 0 (0) = 0
12. y 00 − 4y 0 − 5y = 2e−t + δ3 (t), y(0) = 0, y 0 (0) = 0

8.7 Periodic Functions

In modelling mechanical and other systems it frequently happens that the
forcing function repeats over time. Periodic functions best model such repeti-
tion.
A function f defined on [0, ∞) is said to be periodic if there is a positive
number p such that f (t + p) = f (t) for all t in the domain of f . We say p is a
period of f . If p > 0 is a period of f and there is no smaller period then we
say p is the fundamental period of f although we will usually just say the
period. The interval [0, p) is called the fundamental interval. If there is no
such smallest positive p for a periodic function then the period is defined to
be 0. The constant function f (t) = 1 is an example of a periodic function with
period 0. The sine function is periodic with period 2π: sin(t + 2π) = sin(t).
Knowing the sine on the interval [0, 2π) implies knowledge of the function
everywhere. Similarly, if we know f is periodic with period p > 0 and we
know the function on the fundamental interval then we know the function
everywhere. Figure 8.17 illustrates this point.

The Sawtooth Function

A particularly useful periodic function is the sawtooth function. With it we
can express other periodic functions simply by composition. Let p > 0. The
saw tooth function is given by


 t if 0 ≤ t < p
if p ≤ t < 2p

t − p

< t >p = .
 t − 2p if 2p ≤ t < 3p
..



.
 8.7 Periodic Functions 373

p 2p 3p 4p
t

Fig. 8.17. An example of a periodic function with period p. Notice how the interval
[0, p) determines the function everywhere.

p 2p 3p 4p
t

Fig. 8.18. The Sawtooth Function < t >p with period p

It is periodic with period p. Its graph is given in Figure 8.18.

The sawtooth function < t >p is obtained by extending the function y = t
on the interval [0, p) periodically to [0, ∞). More generally, given a function
f defined on the interval [0, p), we can extend it periodically to [0, ∞) by the
formula 

 f (t) if 0 ≤ t < p
if p ≤ t < 2p

f (t − p)

.
 f (t − 2p) if 2p ≤ t < 3p
..



. .
This complicated piecewise definition can be expressed simply by the compo-
sition of f and < t >p :
f (< t >p ).
For example, Figure 8.19 is the graph of y = sin(< t >π ). This function, which
is periodic with period π, is known as the rectified sine wave.
374 8 Laplace Transform II

π 2π 3π 4π
x

Fig. 8.19. The Rectified Sine Wave: sin(< t >π )

The Staircase Function

Another function that will be particularly useful is the staircase function.
For p > 0 it is defined as follows:


 0 if t ∈ [0, p)
if t ∈ [p, 2p)

p

[t]p = .
 2p if t ∈ [2p, 3p)
..



.
Its graph is given in Figure 8.20. The staircase function is not periodic. It

5p

4p

3p

2p

p 2p 3p 4p 5p
t
Fig. 8.20. The Staircase Function: [t]p

is useful in expressing piecewise functions that are like steps on intervals of

length p. For example, if f is a function on [0, ∞) then f ([t]p ) is a function
whose value on [np, (n + 1)p) is the constant f (np). Figure 8.21 illustrates this
idea with the function f (t) = 1 − e−t and p = 0.5.
 8.7 Periodic Functions 375
1

0.8

0.6

0.4

0.2

0 1 2 3 4
t
Fig. 8.21. The graph of 1 − e−t and 1 − e−[t].5

Observe that the staircase function and the sawtooth function are related
by
< t >p = t − [t]p .

The Laplace Transform of Periodic Functions

Not surprisingly, the formula for the Laplace transform of a periodic function
is determined by the fundamental interval.
Theorem 1. Let f be a periodic function in H and p > 0 a period of f . Then
Z p
1
L {f } (s) = e−st f (t) dt.
1 − e−sp 0
Proof.
Z ∞
L {f } (s) = e−st f (t) dt
0
Z p Z ∞
= e−st f (t) dt + e−st f (t) dt
0 p

However, the change of variables t → t + p in the second integral and the

periodicity of f gives
Z ∞ Z ∞
e−st f (t) dt = e−s(t+p) f (t + p) dt
p 0
Z ∞
= e−sp e−st f (t) dt
0
= e−sp L {f } (s).
376 8 Laplace Transform II

Therefore Z p
L {f } (s) = e−st f (t) dt + e−sp L {f } (s).
0
Solving for L {f } gives the desired result. u
t
Example 2. Find the Laplace transform of the square-wave function swc
given by
(
1 if t ∈ [2nc, (2n + 1)c)
swc (t) = for each integer n.
0 if t ∈ [(2n + 1)c, (2n + 2)c)

I Solution. The square-wave function swc is periodic with period 2c. Its
graph is given in Figure 8.22 and, by Theorem 1, its Laplace transform is

c 2c 3c 4c 5c 6c 7c
t

Fig. 8.22. The graph of the square wave function swc

Z 2c
1
L {swc } (s) = e−st swc (t) dt
1 − e−2cs 0
Z c
1
= e−st dt
1 − e−2cs 0
1 1 − e−sc
= −sc 2
1 − (e ) s
1 1
= .
1 + e−sc s
J
Example 3. Find the Laplace transform of the sawtooth function < t >p .
I Solution. Since the sawtooth function is periodic with period p and since
< t >p = t for 0 ≤ t < p, Theorem 1 gives
Z p
1
L {< t >p } (s) = e−st t dt.
1 − e−sp 0
 8.7 Periodic Functions 377

Integration by parts gives

Z p Z p
te−st p 1 pe−sp 1 −st p pe−sp e−sp − 1
e−st t dt = |0 − e−st dt = − − 2 e |0 = − − .
0 −s −s 0 s s s s2
With a little algebra we obtain
1 spe−sp
L {< t >p } (s) = 2
(1 − ).
s 1 − e−sp
J

As mentioned above it frequently happens that we build periodic functions

by restricting a given function f to the interval [0, p) and then extending it to
be periodic with period p: f (< t >p ). Suppose now that f ∈ H. We can then
express the Laplace transform of f (< t >p ) in terms of the Laplace transform
of f . The following corollary expresses this relationship and simplifies unnec-
essary calculations like the integration by parts that we did in the previous
example.

Corollary 4. Let p > 0 Suppose f ∈ H. Then

1
L {f (< t >p )} (s) = L {f − f hp } .
1 − e−sp
Proof. The function f − f hp = f (1 − hp ) is the same as f on the interval [0, p)
and 0 on the interval [p, ∞).. Therefore
Z p Z ∞
e−st f (t) dt = e−st (f (t) − f (t)hp (t)) dt = L {f − f hp } .
0 0

The result now follows from Theorem 1. u

t

Let’s return to the sawtooth function in Example 3 and see how Corollary
4 simplifies the calculation of its Laplace transform.
1
L {< t >p } (s) = L {t − thp }
1 − e−sp
 
1 1 −sp
= − e L {t + p}
1 − e−sp s2
 
1 1 −sp 1 + sp
= −e
1 − e−sp s2 s2
−sp
 
1 spe
= 2 1− .
s 1 − e−sp
The last line requires a few algebraic steps.

Example 5. Find the Laplace transform of the rectified sine wave sin(<
t >π ). See Figure 8.19.
378 8 Laplace Transform II

I Solution. Corollary 4 gives

1
L {sin(< t >π )} = L {sin t − sin t hπ (t)}
1 − e−πs
 
1 1 −πs
= − e L {sin(t + π)}
1 − e−πs s2 + 1
1 + e−πs
 
1
= ,
1 − e−πs s2 + 1

where we use the fact that sin(t + π) = − sin(t). J

The inverse Laplace transform

The inverse Laplace transform of functions of the form
1
F (s)
1 − e−sp
is not always a straightforward matter to find unless, of course, F (s) is of the
form L {f − f hp } so that Corollary 4 can be used. Usually though this is not
the case. Let r be a fixed real or complex number. Recall that the geometric
series
∞
X
rn = 1 + r + r2 + r3 + · · ·
n=0
1
converges to 1−r when |r| < 1. Since e−sp < 1 for s > 0 we can write
∞
1 X
= e−snp
1 − e−sp n=0

and therefore
∞
1 X
F (s) = e−snp F (s).
1 − e−sp n=0

If f = L−1 {F } then a termwise computation gives

  X ∞ ∞
−1 1 −1
 −snp X
L F (s) = L e F (s) = f (t − np)hnp (t).
1 − e−sp n=0 n=0

On an interval of the form [N p, (N +1)p) the function hnp is 1 for n = 0, . . . , N

and 0 otherwise. We thus obtain
∞ N
  X !
−1 1 X
L F (s) = f (t − np) χ[N p,(N +1)p) .
1 − e−sp n=0
N =0
 8.7 Periodic Functions 379

A similar argument gives

∞ N
  X !
−1 1 X
n
L F (s) = (−1) f (t − np) χ[N p,(N +1)p) .
1 + e−sp n=0
N =0

For reference we record these results in the following theorem:

Theorem 6. Let p > 0 and suppose L {f (t)} = F (s). Then
n o P P 
∞ N
1. L−1 1−e1−sp F (s) = N =0 n=0 f (t − np) χ[N p,(N +1)p) .
n o P P 
∞ N
2. L−1 1+e1−sp F (s) = N =0 n=0 (−1) n
f (t − np) χ[N p,(N +1)p) .

Example 7. Find the inverse Laplace transform of

1
.
(1 − e−2s )s
I Solution. If f (t) = 1 then F (s) = 1s is its Laplace transform. We thus
have
∞ N
  !
−1 1 X X
L = f (t − 2n) χ[2N,2(N +1))
(1 − e−2s )s n=0
N =0
X∞
= (N + 1)χ[2N,2(N +1))
N =0
∞
1 X
= 1+ 2N χ[2N,2(N +1))
2
N =0
1
= 1 + [t]2 .
2
J

Mixing Problems with Periodic Input

We now turn our attention to two examples. Both are mixing problems with
periodic input functions.
Example 8. Suppose a tank contains 10 gallons of pure water. Two input
sources alternately flow into the tank for 1 minute intervals. The first input
source is a brine solution with concentration 1 pound salt per gallon and flows
(when on) at a rate of 5 gallons per minute. The second input source is pure
water and flows (when on) at a rate of 5 gallons per minute. The tank has
a drain with a constant outflow of 5 gallons per minute. Let y(t) denote the
total amount of salt at time t. Find y(t) and for large values of t determine
how y(t) fluctuates.
380 8 Laplace Transform II

I Solution. The input rate of salt is given piecewise by the formula

(
5 if 2n ≤ t < 2n + 1)
= 5 sw1 (t).
0 if 2n + 1 ≤ t < 2n + 2
The output rate is given by
y(t)
· 5.
10
This leads to the first order differential equation
1
y 0 + y = 5 sw1 (t) y(0) = 0.
2
A calculation using Example 2 gives that the Laplace transform is
1 1
Y (s) = 5 ,
1 + e−s s(s + 12 )
and a partial fraction decomposition gives
1 1 1 1
Y (s) = 10 − 10 .
1 + e−s s 1 + e−s s + 12
Now apply the inverse Laplace transform. By Theorem 6 the inverse Laplace
transform of the first expression is
∞ X
X N ∞
X
10 (−1)n χ[N,N +1) = 10 χ[2N,2N +1) = 10 sw1 (t).
N =0 n=0 N =0

By Theorem 6 the inverse Laplace transform of the second expression is

∞ X
N ∞ X
N
1 1 1
X X
10 (−1)n e− 2 (t−n) χ[N,N +1) = 10e− 2 t (−e 2 )n χ[N,N +1)
N =0 n=0 N =0 n=0
∞ 1
− 12 t
X 1 − (−e 2 )N +1
= 10e 1 χ[N,N +1)
N =0
1 + e2
1
( N +1
10e− 2 t 1+e 2 if t ∈ [N, N + 1) (N even)
= 1 N +1 .
1+e 2 1−e 2 if t ∈ [N, N + 1) (N odd)
Finally, we put these two expression together to get our solution

1
( N +1
10e− 2 t 1+e 2 if t ∈ [N, N + 1) (N even)
y(t) = 10 sw1 (t) − 1 N +1 (1)
1+e 2 1−e 2 if t ∈ [N, N + 1) (N odd)
−1t −t+N +1


 10 − 10 e 2 +e 1 2 if t ∈ [N, N + 1) (N even)
1+e 2


= .
 1 −t+N +1
−10 e− 2 t −e 1 2

if t ∈ [N, N + 1) (N odd)

1+e 2
 8.7 Periodic Functions 381

0 2 4 6 8 10 12 14 16 18 20
t
Fig. 8.23. A mixing problem with square wave input function.

The graph of y(t), obtained with the help of a computer, is presented in

Figure 8.23. The solution is sandwiched in between a lower and upper curve.
The lower curve, l(t), is obtained by setting t = m to be an even integer in
the formula for the solution and then continuing it to all reals. We obtain
1 −m+m+1 1 1
e− 2 m + e 2 e− 2 m + e 2
l(m) = 10 − 10 1 = 10 − 10 1
1+e 2 1+e 2

and thus 1 1
e− 2 t + e 2
l(t) = 10 − 10 1
1 + e2
In a similar way, the upper curve, u(t), is obtained by setting t = m− to be
an odd integer and continuing to all reals. We obtain
1 1
e− 2 t − e 2
u(t) = −10 1 .
1 + e2
An easy calculation gives
1 1
10e 2 10e 2
limt→∞ l(t) = 10 − 1 ' 3.78 and limt→∞ u(t) = 1 ' 6.22.
1+e 2 1+e 2

This means that the salt fluctuation in the tank varies between 3.78 and 6.22
pounds for large values of t. J
In practice it is not always possible to know the input function, f (t), pre-
cisely. Suppose though that it is known that f is periodic with period p. Then
R (n+1)p
the total input on all intervals of the form [np, (n + 1)p) is np f (t) dt = h,
a constant. On the interval [0, p) we could model the input with a Dirac delta
function concentrated at a point, c say, and then extend it periodically. We
would then obtain a sum of Dirac delta functions of the form
382 8 Laplace Transform II

a(t) = h(δc + δc+p + δc+2p + · · · )

that may adequately represent the input for the system we are trying to model.
Additional information may justify distributing the total input over two or
more points in the interval and extend periodically. Whatever choices are made
the solution will need to be analyzed in the light of empirical data known about
the system. Consider the example above. Suppose that it is known that the
input is periodic with period 2 and total input 5 on the fundamental interval.
Suppose additionally that you are told that the distribution of the input of
salt is on the first half of each interval. We might be led to try to model the
input on [0, 2) by 52 δ0 + 52 δ1 and then extend periodically to obtain
∞
5X
a(t) = δn .
2 n=0

Of course, the solution modelled by the input function a(t) will differ from the
actual solution. What is true though is that both exhibit similar long term
behavior. This can be observed in the following example.

Example 9. Suppose a tank contains 10 gallons of pure water. Pure water

flows into the tank at a rate of 5 gallons per minute. The tank has a drain
with a constant outflow of 5 gallons per minute. Suppose 52 pounds of salt is
put in the tank each minute whereupon it instantly and uniformly dissolves.
Assume the level of fluid in the tank is always 10 gallons. Let y(t) denote the
total amount of salt at time t. Find y(t) and for large values of t determine
how y(t) fluctuates.
P∞
I Solution. As discussed above the input function is 52 n=1 δn and there-
fore the differential equation that models this system is
∞
1 5X
y0 + y = δn , y(0) = 0.
2 2 n=1

The Laplace transform leads to

∞
5 X −sn 1
Y (s) = e ,
2 n=0 s + 12

and inverting the Laplace transform gives

 8.7 Periodic Functions 383
∞
5 X − 1 (t−n)
y(t) = e 2 hn (t)
2 n=0
∞
5 −1t X 1 n
= e 2 (e 2 ) hn (t)
2 n=0
∞ N
!
5 1 X X 1
= e− 2 t (e )
2 n
χ[N,N +1)
2 n=0
N =0
∞ N +1
5 −1t X 1−e 2
= e 2 1 χ[N,N +1)
2 1 − e2
N =0
1 1
5(e− 2 t − e− 2 (t−[t]−1) )
= 1 .
2(1 − e 2 )
The graph of this equation is given in Figure 8.24. The solution is sandwiched

0 2 4 6 8 10 12 14 16 18 20
x
Fig. 8.24. A mixing problem with a periodic Dirac delta function: The solution to
the differential equation y 0 + 12 y = 52 ∞
n=1 δn y(0) = 0.

in between a lower and upper curve. The lower curve, l(t), is obtained by
setting t = m to be an integer in the formula for the solution and then
continuing it to all reals. We obtain
5 m −m+m+1 5 m 1
l(m) = 1 (e− 2 − e 2 )= 1 (e− 2 − e 2 )
2(1 − e− 2 ) 2(1 − e− 2 )
and thus
5 t 1
l(t) = 1 (e− 2 − e 2 )
2(1 − e− 2 )
In a similar way, the upper curve, u(t), is obtained by setting t = (m + 1)−
(an integer slightly less than m + 1) and continuing to all reals. We obtain
384 8 Laplace Transform II

5 t
u(t) = 1 (e− 2 − 1)
2(1 − e− 2 )
An easy calculation gives
1
−5e 2 −5
limt→∞ l(t) = 1 ' 3.85 and limt→∞ u(t) = 1 ' 6.35.
2(1−e 2 ) 2(1−e 2 )

This means that the salt fluctuation in the tank varies between 3.85 and 6.35
pounds for large values of t. J
A comparison of the solutions in these examples reveals similar long term
behavior in the fluctuation of the salt content in the tank. Remember though
that each problem that is modelled must be weighed against hard empirical
data to determine if the model is appropriate or not. Also, we could have
modelled the instantaneous input by assuming the input was concentrated at
a single point, rather than two points. The results are not as favorable. These
other possibilities are explored in the exercises.

8.8 Undamped Motion with Periodic Input

In Section ?? we discussed various kinds of harmonic motion that can result
from solutions to the differential equation
ay 00 + by 0 + cy = f (t).
Undamped motion led to the differential equation
ay 00 + cy = f (t). (1)
In particular, we explored the case where f (t) = F0 cos ωt and were led to the
solution  F0
 (cos ωt − cos βt) if β 6= ω
 a(β 2 − ω 2 )


y(t) = (2)
F

0

t sin ωt if β = ω,


2aω
pc
where β = a . The case where β 6= ω gave rise to the notion of beats, while
the case β = ω gave us resonance. Since cos ωt is periodic the system that led
to Equation 1 is an example of undamped motion with periodic input.
In this section we will explore this phenomenon with two further examples:
P∞ a
square wave periodic function, swc and a periodic impulse function, n=0 δnc .
Both examples are algebraically tedious, so you will be asked to fill in some of
the algebraic details in the exercises. To simplify the notation we will rewrite
Equation (1) as
y 00 + β 2 y = g(t)
and assume y(0) = y 0 (0) = 0.
 8.8 Undamped Motion with Periodic Input 385

Undamped Motion with square wave forcing

function
Example 1. A constant force of r units for c units of time is applied to a
mass-spring system with no damping force that is initially at rest. The force
is then released for c units of time. This on-off force is extended periodically
to give a periodic forcing function with period 2c. Describe the motion of the
mass.

I Solution. The differential equation which describes this system is

y 00 + β 2 y = r swc (t), y(0) = 0, y 0 (0) = 0 (3)

where swc is the square wave function with period 2c and β 2 is the spring
constant. By Example 2 the Laplace transform leads to the equation
 
1 1 r 1 1 s
Y (s) = r = − (4)
1 + e−sc s(s2 + β 2 ) β 2 1 + e−sc s s2 + β 2
r 1 1 r 1 s
= 2 −sc
− 2 −sc
β 1+e s β 1+e s + β2
2

Let
r 1 1 r 1 s
F1 (s) = and F2 (s) = .
β 2 1 + e−sc s β 2 1 + e−sc s2 + β 2
Again, by Example 2 we have
r
f1 (t) = swc (t). (5)
β2
By Theorem 6 we have
∞ N
!
r X X
n
f2 (t) = 2 (−1) cos(βt − nβc) χ[N c,(N +1)c) . (6)
β n=0
N =0

We consider two cases.

βc is not an odd multiple of π

sin(v)
Lemma 2. Suppose v is not an odd multiple of π and let α = 1+cos(v) . Then
PN
1. n=0 (−1)n cos(u+nv) = 21 cos u + α sin u + (−1)N (cos(u + N v) − α sin(u + N v)


PN
2. n=0 (−1)n sin(u+nv) = 21 sin u − α cos(u) + (−1)N (sin(u + N v) + α cos(u + N v) .

Proof. The proof of the lemma is left as an exercise. u

t
386 8 Laplace Transform II
− sin(βc)
Let u = βt and v = −βc. Then α = 1+cos (βc) . In this case we can apply
part (1) of the lemma to Equation (6) to get
∞
r X
cos βt + α sin βt + (−1)N (cos β(t − N c) − α sin β(t − N c) χ[N c,N +1)c


f2 (t) = 2
2β
N =0
r r
= 2
(cos βt + α sin βt) + 2 (−1)[t/c]1 (cos β < t >c −α sin β < t >c ). (7)
2β 2β
Let
r r
y1 (t) = swc (t) − 2 (−1)[t/c]1 (cos β < t >c −α sin β < t >c )
β2 2β
r  [t/c]1

= 2 sw c (t) − (−1) (cos β < t > c −α sin β < t > c )
2β 2
and
r
y2 (t) = − (cos βt + α sin βt).
2β 2
Then

y(t) = f1 (t) − f2 (t) = y1 (t) + y2 (t)

r  [t/c]1

= 2 sw c (t) − (−1) (cos β < t > c −α sin β < t > c )
2β 2
r
− 2 (cos βt + α sin βt). (8)
2β
A quick check shows that y1 is periodic with period 2c and y2 is periodic with
period 2πβ . Clearly y2 is continuous and since the solution y(t) is continuous
by Theorem 8, so is y1 . The following lemma will help us determine when y
is a periodic solution.
Lemma 3. Suppose g1 and g2 are continuous periodic functions with periods
p1 > 0 and p2 > 0, respectively. Then g1 + g2 is periodic if and only if pp21 is
a rational number.
Proof. If pp12 = m
n is rational then np1 = mp2 is a common period of g1 and g2
and hence is a period of g1 +g2 . It follows that g1 +g2 is periodic. The opposite
implication, namely, that the periodicity of g1 + g2 implies pp21 is rational, is a
nontrivial fact. We do not include a proof. u
t
Using this lemma we can determine precisely when the solution y = y1 +
2c
y2 is periodic. Namely, y is periodic precisely when 2π/β = cβ
π is rational.
Consider the following illustrative example. Set r = 2, c = 3π
2 , and β = 1.
Then α = 1 and

y(t) = 2 swc (t) − (−1)[t/c]1 (cos < t >c − sin < t >c ) − (cos t + sin t). (9)

This function is graphed simultaneously with the forcing function in Figure

8.25. The solution is periodic with period 4c = 6π. Notice that there is an
 8.8 Undamped Motion with Periodic Input 387

4
3
2
1

0 10 20 30 40 50 60
x
–1
–2

Fig. 8.25. The graph of equation 9

interval where the motion of the mass is stopped. This occurs in the interval
[3c, 4c). The constant force applied on the interval [2c, 3c) gently stops the
motion of the mass by the time t = 3c. Since the force is 0 on [3c, 4c) there
is no movement. At t = 4c the force is reapplied and the process thereafter
repeats itself. This phenomenon occurs in all cases where the solution y is
periodic. (cf. Exercise ??)
In Section ?? we observed that when the natural frequency of the spring
is close to but not equal to the frequency of the forcing function, cos(ωt),
then one observes vibrations that exhibit a beat. This phenomenon likewise
occurs for the square wave forcing function. Let r = 2, c = 9π 8 , and β = 1.
Recall that frequency is merely the reciprocal of the period so when these
frequencies are close so are their periods. The natural period of the spring is
2π 9π
β = 2π while the period of the forcing function is 2c = 4 : their periods are
close and likewise their frequencies. Figure 8.26 gives a graph of y in this case.
Again it is evident that the motion of the mass stops on the last subinterval
before the end of its period. More interesting is the fact that y oscillates with
an amplitude that varies with time and produces ’beats’.

βc is an odd multiple of π
We now return to equation (6) in the case βc is an odd multiple of π. Things
reduce substantially because cos(βt − N βc) = (−1)N cos(βt) and we get
388 8 Laplace Transform II

10

0 20 40 60 80 100
x
–5

–10
Fig. 8.26. The graph of equation 9: the beats are evident here.

∞ N
r XX
f2 (t) = cos(βt)χ[N c,(N +1)c)
β2 n=0
N =0
∞
r X
= 2 (N + 1)χ[N c,(N +1)c) cos(βt)
β
N =0
r
= 2 ([t/c]1 + 1) cos(βt).
β
The solution now is
y(t) = f1 (t) − f2 (t)
r
= 2 (swc (t) − [t/c]1 cos(βt) − cos(βt)) . (10)
β
Figure 8.27 gives the graph of this in the case where r = 2, β = π and
c = 1. Resonance is clearly evident. Of course, this is an idealized situation;
the spring would eventually fail.

Undamped Motion with period impulses

Example 4. A mass-spring system with no damping force is acted upon at
rest by an impulse force of r units at all multiples of c units of time starting
at t = 0. (Imagine a hammer exerting blows to the mass at regular intervals.)
Describe the motion of the mass.
I Solution. The differential equation that describes this system is given by
∞
X
00 2
y +β y =r δnc y(0) = 0, y 0 (0) = 0,
n=0
 8.8 Undamped Motion with Periodic Input 389

0 5 10 15 20 25 30
x
–2

–4

–6
Fig. 8.27. The graph of equation 10: resonance is evident here.

where, again, β 2 is the spring constant. The Laplace transform gives

∞
r X −ncs β
Y (s) = e .
β n=0 s2 + β 2

By Theorem 6
∞
r X
y(t) = sin β(t − nc)hnc
β n=0
∞ N
r XX
= sin(βt − nβc)χ[N c,(N +1)c) (11)
β n=0
N =0

Again we will consider two cases.

βc is not a multiple of 2π
sin v
Lemma 5. Suppose v is not a multiple of 2π. Let α = 1−cos v . Then
PN
1. n=0 sin(u + nv) = 21 (sin u + α cos u + sin(u + N v) − α cos(u + N v)) .
PN
2. n=0 cos(u + nv) = 12 (cos u − α sin u + cos(u + N v) + α sin(u + N v)) .

Let u = βt and v = −βc. By the first part of Lemma 5 we get

∞
r X
y(t) = (sin βt + α cos βt + sin β(t − N c) − α cos β(t − N c)) χ[N c,(N +1)c)
2β
N =0
r
= (sin βt + α cos βt + sin β < t >c −α cos β < t >c ) , (12)
2β
390 8 Laplace Transform II
− sin βc
where α = 1−cos βc . Lemma 3 implies that the solution will be periodic when
c βc
2π/β = is rational. Consider the following example. Let r = 2, β = 1 and
2π
3π
c = 2 .
The graph of the solution, Equation (12), in this case is given in
Figure 8.28. The period is 6π = 4c. Observe that on the interval [3c, 4c) the

0 10 20 30 40 50
x
–1

–2
Fig. 8.28. The graph of equation 12

motion of the mass is completely stopped. At t = 3c the hammer strikes and

imparts a velocity that stops the mass dead in its track. At t = 4c the process
repeats itself. As in the previous example this phenomenon occurs in all cases
c βc
where the solution y is periodic, i.e. when 2π/(β) = 2π is rational.
When the period of the forcing function is close to that of the natural
period of the spring the beats in the solution can again be seen. For example,
Figure 8.29 shows the graph when c = 89 (2π), β = 1, and r = 2.

βc is a multiple of 2π
In this case Equation (11) simplifies to
r
y(t) = (sin βt + [t/c]1 sin βt) . (13)
β
Figure 8.30 gives a graph of the solution when c = 2π, β = 1, and r = 2. In
this case resonance occurs. J

8.9 Convolution
In this section we extend to the Heaviside class the definition of the convolu-
tion that we introduced in Section 3.4. The importance of the convolution is
 8.9 Convolution 391

0 20 40 60 80 100
x
–2

–4

Fig. 8.29. A solution that demonstrates beats.

30
20
10

0 20 40 60 80 100
x
–10
–20
–30

Fig. 8.30. A solution with resonance.

that it provides a closed formula for the inverse Laplace transform of a prod-
uct of two functions. This is the essence of the convolution theorem which we
give here. We will then consider further extensions to the delta functions δc
and explore some very pleasant properties.
Given two functions f and g in H the function

u 7→ f (u)g(t − u)

is continuous except for perhaps finitely many points on each interval of the
form [0, t]. Therefore the integral
Z t
f (u)g(t − u) du
0

exists for each t > 0. The convolution of f and g is given by

392 8 Laplace Transform II
Z t
f ∗ g(t) = f (u)g(t − u) du.
0

We will not make the argument but it can be shown that f ∗ g is in fact
continuous. Since there are numbers K, L, a, and b such that

|f (t)| ≤ Keat and |g(t)| ≤ Lebt

it follows that
Z t
|f ∗ g(t)| ≤ |f (u)| |g(t − u)| du
0
Z t
≤ KL eau eb(t−u) du
0
Z t
= KLebt e(a−b)u du
( 0
tebt if a = b
= KL eat −ebt .
a−b if a 6= b

This shows that f ∗ g is of exponential type and therefore is back in H.

The linearity properties we listed in Section 3.4 extend to H. We restate
them here: Suppose f , g, and h are in H. Then
1. f ∗g ∈H
2. f ∗g =g∗f
3. (f ∗ g) ∗ h = f ∗ (g ∗ h)
4. f ∗ (g + h) = f ∗ g + f ∗ h
5. f ∗ 0 = 0.

The sliding window and an example

Let’s now break the convolution up into its constituents to get a better idea
of what it does. The function u 7→ g(−u) has a graph that is folded or flipped
across the y-axis. The function u 7→ g(t − u) shifts the flip by t ≥ 0. The
convolution measures the amount of overlap between f and the flip and shift
of g by positive values t. One can think of g(t − u) as a horizontally sliding
window by which f is examined and measured.

Example 1. Let f (t) = tχ[0,1) (t) and g(t) = χ[1,2) (t). Find the convolution
f ∗ g.

I Solution. The flip of g is g(−u) = χ[−2,−1) (u) while the flip and shift of
g is g(t − u) = χ[t−2,t−1) (u). See Figure 8.31.
 8.9 Convolution 393
1

t–2 t–1 t–2 t–1 0

Fig. 8.31. The flip and shift of g = χ[1,2) . Fig. 8.32. The window g(t − u) and f (u) have no
overlap: 0 ≤ t < 1

1 1

t–2 0 t–1 0 t–2 t–1

Fig. 8.33. The window g(t − u) and f (u) overlap:Fig. 8.34. The window g(t − u) and f continue to
1 ≤ t < 2. overlap: 2 ≤ t < 3.

If t < 1 then there is no overlap of the window u 7→ g(t − u) with f , i.e.

u 7→ f (u)g(t − u) = 0 and hence f ∗ g(t) = 0. See Figure 8.32. Now suppose
1 ≤ t < 2. Then there is overlap between the window and f as seen in Figure
8.33.
The product of f (u) and g(t − u) is the function u 7→ u, 0 ≤ u < t − 1
2
and hence f ∗ g(t) = (t−1)2 . Now if 2 ≤ t < 3 there is still overlap between
the window and f as seen in Figure 8.34. The product of f (u) and g(t − u)
2
is u 7→ u, t − 2 ≤ u < 1 and f ∗ g(t) = 1−(t−2)
2 = −(t−1)(t−3)
2 . Finally, when
3 ≤ t < ∞ the window shifts past f as illustrated in Figure 8.35. The product
of f (u) and g(t − u) is 0 and f ∗ g(t) = 0.

1 0.5
0.4
0.3
0.2
0.1

–1 0 1 2 3 4
0 t–2 t–1 t

Fig. 8.35. Again, there is no overlap: 3 ≤ t < ∞ Fig. 8.36. The convolution f ∗ g.

We can now piece these function together to get

394 8 Laplace Transform II

if 0 ≤ t < 1

0

 (t−1)2 (t − 1)2
if 1 ≤ t < 2

2
(t − 1)(t − 3)
f ∗g(t) = = χ[1,2) − χ[2,3) .
 −(t−1)(t−3) if 2 ≤ t < 3 2 2

 2
0 if 3 ≤ t < ∞


Its graph is given in Figure 8.36. Notice that the convolution is continuous;
in this case it is not differentiable at t = 2, 3. J

Theorem 2 (The Convolution Theorem). Suppose f and g are in H and

F and G are their Laplace transforms, respectively. Then
L {f ∗ g} (s) = F (s)G(s)
or, equivalently,
L−1 {F (s)G(s)} (t) = (f ∗ g)(t).
Proof. For any f ∈ H we will define f (t) = 0 for t < 0. By Theorem 4
e−st G(s) = L {g(u − t)ht } .
Therefore,
Z ∞
F (s)G(s) = e−st f (t) dtG(s)
0
Z ∞
= e−st G(s)f (t) dt
Z0 ∞
= L {g(u − t)ht (u)} (s)f (t) dt
0
Z ∞Z ∞
= e−su g(u − t)h(u − t)f (t) du dt (1)
0 0

A theorem in calculus 1 tells us that we can switch the order of integration

in (1) when f and g are in H. Thus we obtain
Z ∞Z ∞
F (s)G(s) = e−su g(u − t)h(u − t)f (t) dt du
0 0
Z ∞Z t
= e−su g(u − t)f (t) dt du
0 0
Z ∞
= e−su (f ∗ g)(u) du
0
= L {f ∗ g} (s)
u
t
1
c.f. Vector Calculus, Linear Algebra, and Differential Forms, J.H. Hubbard and
B.B Hubbard, page 444
 8.9 Convolution 395

There are a variety of uses for the convolution theorem. For one it is
sometimes a convenient way to compute the convolution of two functions f
and g; namely (f ∗ g)(t) = L−1 {F (s)G(s)} .

Example 3. Compute the convolution of the functions given in 1:

f (t) = tχ[0,1) and g(t) = χ[1,2) .

In the following example, which is a reworking of Example 1, instead of keeping

track of the sliding window g(t−u) the convolution theorem turns the problem
into one that is primarily algebraic.

I Solution. The Laplace transforms of f and g are, respectively,

e−s − e−2s
 
1 −s 1 1
F (s) = 2 − e + and G(s) = .
s s2 s s

The product simplifies to

   
1 −s 2 1 1 1
F (s)G(s) = e − + 2 e−2s + + 2 e−3s .
s3 s3 s s3 s

Its inverse Laplace transform is

(f ∗ g)(t) = L−1 {F (s)G(s)} (t)

(t − 1)2 (t − 3)(t − 1)
= h1 (t) − ((t − 2)(t − 1))h2 (t) + h3 (t)
2 2
(t − 1)2 (t − 1)(t − 3)
= χ[1,2) (t) − χ[2,3) (t)
2 2
J

Convolution and the Dirac Delta Function

We would like to extend the definition of convolution to include the delta
functions δc , c ≥ 0. Recall that we formally defined the delta function by

δc (t) = lim dc, (t),

→0

where dc, = 1 χ[c,c+) . In like manner, for f ∈ H, we define

f ∗ δc (t) = lim f ∗ dc, (t).

→0
396 8 Laplace Transform II

Theorem 4. For f ∈ H

f ∗ δc (t) = f (t − c)hc ,

where the equality is understood to mean essentially equal.

Proof. Let f ∈ H. Then

Z t
f ∗ dc, (t) = f (u)dc, (t − u) dt
0
1 t
Z
= f (u)χ[c,c+) (t − u) du
 0
Z t
1
= f (u)χ[t−c−,t−c) (u) du
 0

Now suppose t < c. Then χ[t−c−,t−c)(u) = 0, for all u ∈ [0, t). Thus f ∗ dc, =
0. On the other hand if t > c then for  small enough we have

1 t−c
Z
f ∗ dc, (t) = f (u) du.
 t−c−

Let t be such that t − c is a point of continuity of f . Then by the Fundamental

Theorem of Calculus
1 t−c
Z
lim f (u) du = f (t − c).
→0  t−c−

Since f has only finitely many removable discontinuities on any finite interval
it follows that f ∗ δc is essentially equal to f (t − c)hc . u
t

The special case c = 0 produces the following pleasant corollary.

Corollary 5. For f ∈ H we have

f ∗ δ0 = f.

This corollary tells us that this extension to the Dirac delta function gives
an identity under the convolution product. We thus have a correspondence
between the multiplicative identities in domain and transform space under
the Laplace transform since L {δ0 } = 1.

The Impulse Response Function

Let f ∈ H. Let us return to our basic second order differential equation

ay 00 + by 0 + cy = f (t), y(0) = y0 and y 0 (0) = y1 . (2)

 8.9 Convolution 397

By organizing terms in its Laplace transform in the right manner we can

express the solution in terms of convolution of a special function called the
impulse response function and f . To explain the main idea let’s begin by
considering the following special case

ay 00 + by 0 + cy = 0 y(0) = 0 and y 0 (0) = 1.

This corresponds to a system in initial position but with a unit velocity. Our
discussion in Section 8.6 shows that this is exactly the same thing as solving

ay 00 + by 0 + cy = δ0 , y(0) = 0 and y 0 (0) = 0

the same system at rest but with unit impulse at t = 0. The Laplace transform
of either equation above leads to
1
Y (s) = .
as2 + bs + c
The inverse Laplace transform is the solution and will be denoted by ζ(t); it
is called the impulse response function.
The Laplace transform of Equation 2 leads to

(as + b)y0 + y1 F (s)

Y (s) = 2
+ 2 .
as + bs + c as + bs + c
Let
(as + b)y0 + y1 F (s)
H(s) = and G(s) = .
as2 + bs + c as2 + bs + c
Then Y (s) = H(s) + G(s). The inverse Laplace transform of H corresponds
to the solution to Equation 2 when f = 0. It is the homogeneous solution. On
the other hand, G can be written as a product
 
1
G(s) = F (s)
as2 + bs + c

and its inverses Laplace transform g(t) is

g(t) = f ∗ ζ(t),

by the convolution theorem.

We summarize this discussion in the following theorem:
Theorem 6. Let f ∈ H. The solution to Equation 2 can be expressed as

h(t) + f ∗ ζ(t),

where h is the homogenous solution to Equation 2 and ζ is the impulse re-

sponse function.
398 8 Laplace Transform II

Example 7. Solve the following differential equation:

y 00 + 4y = χ[0,1) y(0) = 0 and y 0 (0) = 0.

I Solution. The homogeneous solution to

y 00 + 4y = 0 y(0) = 0 and y 0 (0) = 0

is the trivial solution h = 0. The impulse response function is

 
1 1
ζ(t) = L−1 = sin 2t.
s2 + 4 2

By Theorem 6 the solution is

y(t) = ζ ∗ χ[0,1)
Z t
1
= sin(2u)χ[0,1) (t − u) du
0 2
1 t
Z
= sin(2u)χ[t − 1, t)(u) du
2 0
(R t
1 sin 2u du if 0 ≤ t < 1
= R0t
2 t−1 sin 2u du if 1 ≤ u < ∞
(
1 1 − cos 2t if 0 ≤ t < 1
=
4 cos 2(t − 1) − cos 2t if 1 ≤ t < ∞

J
9
Matrices

Most students by now have been exposed to the language of matrices. They
arise naturally in many subject areas but mainly in the context of solving
a simultaneous system of linear equations. In this chapter we will give a re-
view of matrices, systems of linear equations, inverses, and determinants. The
next chapter will apply what is learned here to linear systems of differential
equations.

9.1 Matrix Operations

A matrix is a rectangular array of entities and is generally written in the

following way:  
x11 · · · x1n
X =  ... . . . ...  .
 

xm1 · · · xmn
We let R denote the set of entities that will be in use at any particular time.
Each xij is in R and in this text R can be one of the following sets:

R or C The scalars
R[t] or C[t] Polynomials with real or complex entries
R(s) or C(s) The real or complex rational functions
C n (I, R) or C n (I, C) Real or complex valued functions
with n continuous derivatives

Notice that addition and scalar multiplication is defined on R. Below we will

extend these operations to matrices. (In Chapter 10 we will see an instance
where R will even be matrices themselves; thus matrices of matrices. But we
will avoid that for now.)
The following are examples of matrices.
400 9 Matrices

Example 1.
   
1 0 3   i 2−i
A= B = 1 −1 9 C =
2 −1 4 1 0
 s 1 
t2 e2t
 
D= E = s2 −1 s2 −1
−1 s+2
t3 cos t s2 −1 s2 −1
It is a common practice to use capital letters, like A, B, C, D, and E, to
denote matrices. The size of a matrix is determined by the number of rows
m and the number of columns n and written m × n. In Example 1 A is a
2 × 3 matrix, B is a 1 × 3 matrix, C and E are 2 × 2 matrices, and D is a
2 × 1 matrix. A matrix is square if the number of rows is the same as the
number of columns. Thus, C and E are square matrices. An entry in a matrix
is determined by its position. If X is a matrix the (i, j) entry is the entry that
appears in the ith row and j th column. We denote it in two ways: entij (X)
or more simply Xij . Thus, in Example 1, A1 3 = 3, B1 2 = −1, and C2 2 = 0.
We say that two matrices X and Y are equal if the corresponding entries are
equal, i.e. Xi j = Yi j , for all indices i and j. Necessarily X and Y must be
the same size. The main diagonal of a square n × n matrix X is the vector
formed from the entries Xi i , for i = 1, . . . , n. The main diagonal of C is (i, 0)
and the main diagonal of E is ( s2s−1 , ss+2
2 −1 ). In this book all scalars are either

real or complex. A matrix is said to be a real matrix if each entry is real

and a complex matrix if each entry is complex. Since every real number is
also complex, every real matrix is also a complex matrix. Thus A and B are
real ( and complex) matrices while C is a complex matrix.
Even though a matrix is a structured array of entities in R it should be
viewed as a single object just as a word is a single object though made up
of many letters. We let Mm,n (R) denote the set of all m × n matrices with
entries in R. If the focus is on matrices of a certain size and not the entries
we will sometimes just write Mm,n .
The following definitions highlights various kinds of matrices that com-
monly arise.
1. A diagonal matrix D is a square matrix in which all entries off the main
diagonal are 0. We can say this in another way:
Di j = 0 if i 6= j.
Examples of diagonal matrices are:
1 
 t  s0 0 0
  e 0 0 0 2 0 0 
10  0 e4t 0  s−1
0 0 0 0  .

04
0 0 1 1
0 0 0 − s−2

It is convenient to write diag(d1 , . . . , dn ) to represent the diagonal n × n

matrix with (d1 , . . . , dn ) on the diagonal. Thus the diagonal matrices
 9.1 Matrix Operations 401

2
listed above are diag(1, 4), diag(et , e4t , 1) and diag( 1s , s−1 1
, 0, − s−2 ), re-
spectively.
2. The zero matrix 0 is the matrix with each entry 0. The size is usually
determined by the context. If we need to be specific we will write 0m,n to
mean the m × n zero matrix. Note that the square zero matrix, 0n,n is
diagonal and is diag(0, . . . , 0).
3. The identity matrix, I, is the square matrix with ones on the main di-
agonal and zeros elsewhere. The size is usually determined by the context,
but if we want to be specific, we write In to denote the n × n identity
matrix. The 2 × 2 and the 3 × 3 identity matrices are
 
  100
10
I2 = I3 = 0 1 0 .
01
001

4. We say a square matrix is upper triangular if each entry below the main
diagonal is zero. We say a square matrix is lower triangular if each entry
above the main diagonal is zero.
 
  13 5
12
and 0 0 3  are upper triangular
03
0 0 −4

and  
  00 0
40
and 2 0 0  are lower triangular.
11
1 1 −7
5. Suppose A is an m × n matrix. The transpose of A, denoted At , is the
n × m matrix obtained by turning the rows of A into columns. In terms
of the entries we have more explicitly,

(At )i j = Aj i .

This expression reverses the indices of A and thus changes rows to columns
and columns to rows. Simple examples are
 t
2 3 t 1 2 t 1 2
et
      
 9 0 = 2 9 −1 = et e−t
 
s s3 = s s2 .
2 3 2 3
30 4 e−t s2 s s3 s
−1 4

Matrix Algebra
There are three matrix operations that make up the algebraic structure of
matrices: addition, scalar multiplication, and matrix multiplication.
402 9 Matrices

Addition
Suppose A and B are two matrices of the same size. We define matrix ad-
dition, A + B, entrywise by the following formula

(A + B)i j = Ai j + Bi j .

Thus if    
1 −2 0 4 −1 0
A= and B =
4 5 −3 −3 8 1
then    
1 + 4 −2 − 1 0 + 0 5 −3 0
A+B = = .
4 − 3 5 + 8 −3 + 1 1 13 −2
Corresponding entries are added. Addition preserves the size of matrices. We
can symbolize this in the following way: + : Mm,n (R)×Mm,n (R) → Mm,n (R).
Addition satisfies the following properties:

Proposition 2. Suppose A, B, and C are m × n matrices. Then

A+B =B+A (commutative)

(A + B) + C = A + (B + C) (associative)
A+0=A (additive identity)
A + (−A) = 0 (additive inverse)

Scalar Multiplication
Suppose A is an matrix and c ∈ R. We define scalar multiplication, c · A,
(but usually we will just write cA), entrywise by the following formula

(cA)i j = cAi j .

Thus if  
1 9
c = −2 and A = −3 0
2 5
then  
−2 −18
cA =  6 0  .
−4 −10
Scalar multiplication preserves the size of matrices. Thus · : R × Mm,n (R) →
Mm,n (R). In this context we will call c ∈ R a scalar. Scalar multiplication
satisfies the following properties:
 9.1 Matrix Operations 403

Proposition 3. Suppose A and B are matrices whose sizes are such that each
line below is defined. Suppose c1 , c2 ∈ R. Then
c1 (A + B) = c1 A + c1 B (distributive)
(c1 + c2 )A = c1 A + c2 A (distributive)
c1 (c2 A) = (c1 c2 )A (associative)
1A = A
0A = 0

Matrix Multiplication
Matrix multiplication is more complicated than addition and scalar multipli-
cation. We will define it in two stages: first on row and column matrices and
then on general matrices.
A row matrix or row vector is a matrix which has only one row. Thus
row vectors are in M1,n . Similarly, a column matrix or column vector is
a matrix which has only one column. Thus column vectors are in Mm,1 . We
frequently will denote column and row vectors by lower case boldface letters
like v or x instead of capital letters. It is unnecessary to use double subscripts
to indicate the entries of a row or column matrix: if v is a row vector then we
write vi for the ith entry instead of v1 i . Similarly for column vectors. Suppose
v ∈ M1,n and w ∈ Mn,1 . We define the product v · w (or preferably vw) to
be the scalar given by
vw = v1 w1 + · · · + vn wn .

Even though this formula looks like the scalar product or dot product that
you likely have seen before, keep in mind that v is a row vector while w is a
column vector. For example, if
 
1
3
v = 1 3 −2 0 and w = 
 

0
9
then
vw = 1 · 1 + 3 · 3 + (−2) · 0 + 0 · 9 = 10.
Now suppose that A is any matrix. It is often convenient to distinguish
the rows of A in the following way: If Rowi (A) denotes the ith row of A then
 
Row1 (A)
 Row2 (A) 
A= .. .
 
 . 
Rowm (A)
404 9 Matrices

Clearly Rowi (A) is a row vector. In a similar way, if B is another matrix we

can distinguish the columns of B: Let Colj (B) denote the j th column of B
then  
B = Col1 (B) Col2 (B) · · · Colp (B) .
Each Colj (B) is a column vector.
Now let A ∈ Mmn and B ∈ Mnp . We define the matrix product of A and
B to be the m × p matrix given entrywise by enti j (AB) = Rowi (A) Colj (B).
In other words, the (i, j)-entry of the product of A and B is the ith row of A
times the j th column of B. We thus have
 
Row1 (A) Col1 (B) Row1 (A) Col2 (B) · · · Row1 (A) Colp (B)
 Row2 (A) Col1 (B) Row2 (A) Col2 (B) · · · Row2 (A) Colp (B) 
AB =  .. .. .. .. .
 
 . . . . 
Rowm (A) Col1 (B) Rowm (A) Col2 (B) · · · Rowm (A) Colp (B)
Notice that each entry of AB is given as a product of a row vector and a
column vector. Thus it is necessary that the number of columns of A (the first
matrix) match the number of rows of B (the second matrix). This common
number is n. The resulting product is an m × p matrix. Symbolically, · :
Mm,n (R) × Mn,p (R) → Mm,p (R). In terms of the entries of A and B we have
n
X n
X
enti j (AB) = Rowi (A) Colj (B) = enti k (A)entk j (B) = Ai,k Bk,j .
k=1 k=1

Example 4.
1. If  
2 1  
2 1
A = −1 3  and B =
2 −2
4 −2
then AB is defined because the number of columns of A is the number of
rows of B. Further AB is a 3 × 2 matrix and
    
  2   1
 21 2 21
−2 
 
   


 
 2 

 1 
 6 0
 −1 3 2 −1 3
AB =   = 4 −7 .
−2 



 4 8
    
  2   1 
4 −2 4 −2
2 −2
 t
e 2et
  
−2
2. If A = 2t 2t and B = then
e 3e 1
 t
e (−2) + 2et (1)
  
0
AB = 2t = 2t .
e (−2) + 3e2t (1) e
 9.1 Matrix Operations 405

Notice in the definition (and the example) that in a given column of AB

the corresponding column of B appears as the second factor. Thus

Colj (AB) = A Colj (B). (1)

Similarly, in each row of AB the corresponding row of A appears and we get

Rowi (A)B = Rowi (AB). (2)

Notice too that even though the product AB is defined it is not necessarily
true that BA is defined. This is the case in part 1 of the above example due
to the fact that the number of columns of B (2) does not match the number
of rows of A (3). Even when AB and BA are defined it is not necessarily true
that they are equal. Consider the following example:
Example 5. Suppose
   
12 2 1
A= and B = .
03 4 −1

Then     
12 2 1 10 −1
AB = =
03 4 −1 12 −3
yet     
2 1 12 27
BA = = .
4 −1 0 3 45
These products are not the same. This example show that matrix multiplica-
tion is not commutative. However, the other properties that we are used to
in an algebra are valid. We summarize them in the following proposition.

Proposition 6. Suppose A, B, and C are matrices whose sizes are such that
each line below is defined. Suppose c1 , c2 ∈ R. Then

A(BC) = (AB)C (associatvie)

A(c1 B) = (c1 A)B = c1 (AB) (associative)
(A + B)C = AC + BC (distributive)
A(B + C) = AB + AC (distributive)
IA = AI = I (I is a multiplicative identity)

We highlight two useful formulas that follow from these algebraic proper-
ties. If A is an m × n matrix then
 
x1
 .. 
Ax = x1 Col1 (A) + · · · xn Coln (A), where x =  .  (3)
xn
406 9 Matrices

and

where y = y1 · · · ym . (4)
 
yA = y1 Row1 (A) + · · · ym Rowm (A),

Henceforth, we will use these algebraic properties without explicit refer-

ence. The following result expresses the relationship between multiplication
and transposition of matrices
Theorem 7. Let A and B be matrices such that AB is defined. Then B t At
is defined and
B t At = (AB)t .

Proof. The number of columns of B t is the same as the number of rows of

B while the number of rows of At is the number of columns of A. These
numbers agree since AB is defined so B t At is defined. If n denotes these
common numbers then
n
X n
X
(B t At )i j = (B t )i k (At )k j = Aj k Bk i = (AB)j i = ((AB)t )i j .
k=1 k=1

u
t

Exercises
1 −1 0 2
2 −1 3  2 3 , −3 4. Compute the following
Let A = , B=  and C = 
 0 4
1
−1 2  1 1
matrices.

1. B + C, B − C, 2B − 3C

2. AB, AC, BA, CA

3. A(B + C), AB + AC, (B + C)A

2 1 1 2
4. Let A = 
3 4 and B = 
−1 1. Find C so that 3A + C = 4B.
−1 0 1 0

21 2
3 −1 1 3
2 1 1 −3  1
Let A = 0 −2, B = , and C = 
0 1
. Find the following
0 −1 4 −1 8

1 2
11 7
products

5. AB

6. BC
 9.1 Matrix Operations 407

7. CA

8. B t At

9. ABC.
1
0 


10. Let A = 1 4 3 1! and B =  . Find AB and BA.
−1
−2

1 2 5
 2 4 10 , 1 0 3 −2
Let A =  B= , C= . Verify the following facts:
−1 −2 −5 4 −1  −2
3

11. A2 = 0

12. B 2 = I2

13. C 2 = C

Compute AB − BA in each of the following cases.

01 10
14. A = , B=
1 1 1 1
2 10 3 1 −2
 1 1 1,
15. A =   3 −2 4 
B= 
−1 2 1 −3 5 −1

1a 10
16. Let A = and B = . Show that there are no numbers a and b so that
0 1  b 1
AB − BA = I, where I is the 2 × 2 identity matrix.

17. Suppose that A and B are 2 × 2 matrices.

a) Show by example that it need not be true that (A + B)2 = A2 + 2AB + B 2 .
b) Find conditions on A and B to insure that the equation in Part (a) is valid.

01
18. If A = , compute A2 and A3 .
1 1
11
19. If B = , compute B n for all n.
0 1
a0
20. If A = , compute A2 , A3 , and more generally, An for all n.
0 b 
v1
21. Let A = be a matrix with two rows v1 and v2 . (The number of columns of
v2 
A is not relevant for this problem) Describe the effect of multiplying A on the
left by the following matrices:

01 1c 10 a0 10
(a) (b) (c) (d) (e)
1 0  0 1  c 1  0 1  0 a
408 9 Matrices

cos θ sin θ
22. Let E(θ) = . Show that E(θ1 + θ2 ) = E(θ1 )E(θ2 ).
 sin θ cos θ 
−

cosh θ sinh θ
23. Let (θ) = . Show that F (θ1 + θ2 ) = F (θ1 )F (θ2 ).
sinh θ cosh θ 
24. Let D = diag(d1 , . . . , dn ) and E = diag(e1 , . . . , e2 ). Show that

DE = diag(d1 e1 , . . . , dn en )

9.2 Systems of Linear Equations

Most students have learned various techniques for finding the solution of a
system of linear equations. They usually include various forms of elimination
and substitutions. In this section we will learn the Gauss-Jordan elimination
method. It is essentially a highly organized method involving elimination and
substitution that always leads to the solution set. This general method has
become the standard for solving systems. At first reading it may seem to be
a bit complicated because of its description for general systems. However,
with a little practice on a few examples it is quite easy to master. We will
as usual begin with our definitions and proceed with examples to illustrate
the needed concepts. To make matters a bit cleaner we will stick to the case
where R = R. Everything we do here will work for R = C , R(s), or C(s) as
well. (A technical difficulty for general R is the lack of inverses.)
If x1 , . . . , xn are variables then the equation

a1 x1 + · · · + an xn = b

is called a linear equation in the unknowns x1 , . . . , xn . A system of linear

equations is a set of m linear equations in the unknowns x1 , . . . , xn and is
written in the form

a1 1 x1 + a1 2 x2 + · · · + a1 n xn = b1
a2 1 x1 + a2 2 x2 + · · · + a2 n xn = b2
.. .. .. .. (1)
. . . .
am 1 x1 + am 2 x2 + · · · + am n xn = bm .
The entries ai j are in R and are called coefficients. Likewise, each bj is
in R. A key observation is that Equation (1) can be rewritten in matrix form
as:
Ax = b, (2)
where
 9.2 Systems of Linear Equations 409
     
a1 1 a1 2 · · · a1 n x1 b1
 a2 1 a2 2 · · · a2 n   x2   b2 
A= . .. ..  x= .  and b =  . .
     
 .. . .   ..   .. 
am 1 am 2 · · · am n , xn bm

We call A the coefficient matrix, x the variable matrix, and b the

output matrix. Any column vector x with entries in R that satisfies (1) (or
(2)) is called a solution. If a system has a solution we say it is consistent;
otherwise, it is inconsistent. The solution set, denoted by SA b
, is the set of
all solutions. The system (1) is said to be homogeneous if b = 0, otherwise
it is called nonhomogeneous. Another important matrix associated with (2)
is the augmented matrix:
 
a1 1 a1 2 · · · a1 n b 1
 a2 1 a2 2 · · · a2 n b 2 
[A| b ] =  .. .. .. ..  ,
 
 . . . .
am 1 am 2 · · · am n b m

where the vertical line only serves to separate A from b.

Example 1. Write the coefficient, variable, output, and augmented matrices

for the following system:

−2x1 + 3x2 − x3 = 4
x1 − 2x2 + 4x3 = 5.

Determine whether the following vectors are solutions:

     
−3 7 10  
2
(a) x = 0   (b) x = 7
  (c) x = 7  (d) x = .
1
2 3 1
 
−2 3 −1
I Solution. The coefficient matrix is A = , the variable matrix
  1 −2 4
x1  
4
is x = x2 , the output matrix is b =
  , and the augmented matrix is
5
x3
 
−2 3 −1 4
. The system is nonhomogeneous. Notice that
1 −2 4 5
     
−3   7   10  
4 4 0
A 0 =
  and A 7 =
  while A 7 =
  .
5 5 0
2 3 1

Therefore (a) and (b) are solutions, (c) is not a solution and the matrix in (d)
is not the right size and thus cannot be a solution. J
410 9 Matrices

Remark 2. When only 2 or 3 variables are involved in an example we will

frequently use the variables x, y, and z instead of the subscripted variables
x1 , x2 , and x3 .

Linearity
It is convenient to think of Rn as the set of column vectors Mn,1 (R). If A is
an m × n real matrix then for each column vector x ∈ Rn , the product, Ax,
is a column vector in Rm . Thus the matrix A induces a map which we also
denote just by A : Rn → Rm given by matrix multiplication. It satisfies the
following important property.

Proposition 3. The map A : Rn → Rm is linear. In other words,

1. A(x + y) = A(x) + A(y)
2. A(cx) = cA(x),
for all x, y ∈ Rn and c ∈ R.

Proof. This follows directly from Propositions 3 and 6. u

t

Linearity is an extremely important property for it allows us to describe

the structure of the solution set to Ax = b in a particularly nice way. Recall
that SAb
denotes the solution set to the equation Ax = b.

Proposition 4. With A as above we have two possibilities:

b
1. SA = ∅ or
b b 0
2. there is an xp ∈ SA and SA = xp + SA .
b
In other words, when SA is not the empty set then each solution to Ax = b
has the form
xp + xh ,
where xp is a fixed particular solution to Ax = b and xh is a solution to
Ax = 0.

Proof. Suppose xp is a fixed particular solution and xh ∈ SA

0
. Then A(xp +
xh ) = Axp + Axh = b + 0 = b. This implies that each column vector of
the form xp + xh is in SA b
. On the other hand, suppose x is in SAb
. Then
A(x − xp ) = Ax − Axp = b − b = 0. This means that x − xp is in SA 0
.
Therefore x = xp + xh , for some vector xh ∈ SA .
0
u
t

Remark 5. The system of equations Ax = 0 is called the associated ho-

mogeneous system. Case (1) is a legitimate possibility. For example, the
simple equation 0x = 1 has empty solution set. When the solution set is not
empty it should be mentioned that the particular solution xp is not necessarily
 9.2 Systems of Linear Equations 411

unique. In Chapter 6 we saw a similar theorem for a second order differential

equation Ly = f . That theorem provided a strategy for solving such differen-
tial equations: First we solved the homogeneous equation Ly = 0 and second
found a particular solution (using variation of parameters or undetermined
coefficients). For a linear system of equations the matter is much simpler; the
Gauss-Jordan method will give the whole solution set at one time. We will
see that it has the above form.

Homogenous Systems
The homogeneous case, Ax = 0, is of particular interest. Observe that x = 0
is always a solution so SA
0
is never the empty set, i.e. case (1) is not possible.
But much more is true.
0
Proposition 6. The solution set, SA , to a homogeneous system is closed un-
der addition and multiplication by scalars. In other words, if x and y are
solutions to the homogeneous system and c is a scalar then x + y and cx are
also solutions.

Proof. Suppose x and y are in SA 0

. Then A(x + y) = Ax + Ay = 0 + 0 = 0.
This shows that x + y is in SA . Now suppose c ∈ R. Then A(cx) = cAx =
0

c0 = 0. Hence cx ∈ SA 0
. This shows that SA0
is closed under addition and
scalar multiplication. u
t

Corollary 7. The solution set to a general system of linear equations, Ax =

b, is either
1. empty
2. unique
3. or infinite.

Proof. The associated homogeneous system Ax = 0 has solution set, SA 0

, that
is either equal to the trivial set {0} or an infinite set. To see this suppose that
x is a nonzero solution to Ax = 0 then by Proposition 6 all multiples, cx, are
in SA0
as well. Therefore, by Proposition 4, if there is a solution to Ax = b it
is unique or there are infinitely many. u
t

The Elementary Equation and Row Operations

We say that two systems of equations are equivalent if their solution sets
are the same. This definition implies that the variable matrix is the same for
each system.
412 9 Matrices

Example 8. Consider the following systems of equations:

2x + 3y = 5 x= 1
and
x − y =0 y = 1.
The solution set to the second system is transparent. For the first system there
are some simple operations that easily lead to the solution: First, switch the
two equations around. Next, multiply the equation x − y = 1 by −2 and add
the result to the first. We then obtain
x− y =0
5y = 5

Next, multiply the second equation by 15 to get y = 1. Then add this equation
to the first. We get x = 1 and y = 1. Thus they both have the same solution
1
set, namely the single vector . They are thus equivalent. When used in
1
the right way these kinds of operations can transform a complicated system
into a simpler one. We formalize these operations in the following definition:
Suppose Ax = b is a given system of linear equations. The following three
operations are called elementary equation operations.
1. Switch the order in which two equations are listed
2. Multiply an equation by a nonzero scalar
3. Add a multiple of one equation to another
Notice that each operation produces a new system of linear equations but
leaves the size of the system unchanged. Furthermore we have the following
proposition.
Proposition 9. An elementary equation operation applied to a system of lin-
ear equations is an equivalent system of equations.
Proof. This means that the system that arises from an elementary equation
operation has the same solution set as the original. We leave the proof as an
exercise. u
t
The main idea in solving a system of linear equations is to perform a
finite sequence of elementary equation operations to transform a system into
simpler system where the solution set is transparent. Proposition 9 implies
that the solution set of the simpler system is the same as original system.
Let’s consider our example above.
Example 10. Use elementary equation operations to transform
2x + 3y = 5
x − y =0
into
x= 1
y = 1.
 9.2 Systems of Linear Equations 413

I Solution.
2x + 3y = 5
x − y =0

Switch the order of the two equations x − y =0

2x + 3y = 5

Add −2 times the first equation x− y =0

to the second equation 5y = 5

1
Multiply the second equation by 5 x−y=0
y=1

Add the second equation to the first x =1

y=1

Each operation produces a new system equivalent to the first by Proposi-

tion (9). The end result is a system where the solution is transparent. Since
y = 1 is apparent in the fourth system we could have stopped and used the
method of back substitution, that is, substitute y = 1 into the first equation
and solve for x. However, it is in accord with the Gauss-Jordan elimination
method to continue as we did to eliminate the variable y in the first equation.
You will notice that the variables x and y play no prominent role here. They
merely serve as placeholders for the coefficients, some of which change with
each operation. We thus simplify the notation (and the amount of writing)
by performing the elementary operations on just the augmented matrix. The
elementary equation operations become the elementary row operations
which act on the augmented matrix of the system.
The elementary row operations on a matrix are
1. Switch two rows.
2. Multiply a row by a nonzero constant.
3. Add a multiple of one row to another.
The following notations for these operations will be useful.
1. pij - switch rows i and j.
2. mi (a) - multiply row i by a 6= 0.
3. tij (a) - add to row j the value of a times row i.
The effect of pij on a matrix A is denoted by pij (A). Similarly for the other
elementary row operations.
The corresponding operations when applied to the augmented matrix for
the system in example 10 becomes:
414 9 Matrices
         
2 3 5 1 −1 0 1 −1 0 1 −1 0 10 1
p1 2 t1 2 (−2) m2 (1/5) t2 1 (1)
1 −1 0 −−
→ 2 3 5 −−−−−→ 0 5 5 −−−−−→ 0 1 1 −−−− → 01 1

Above each arrow is the notation for the elementary row operation performed
to produce the next augmented matrix. The sequence of elementary row op-
erations chosen follows a certain strategy: Starting from left to right and top
down one tries to isolate a 1 in a given column and produce 0’s above and
below it. This corresponds to isolating and eliminating variables.
Let’s consider three illustrative examples. The sequence of elementary row
operation we perform is in accord with the Gauss-Jordan method which we
will discuss in detail later on in this section. For now verify each step. The
end result will be an equivalent system for which the solution set will be
transparent.

Example 11. Consider the following system of linear equations

2x + 3y + 4z = 9
.
x + 2y − z = 2

Find the solution set and write it in the form xp + SA

0
.

I Solution. We first will write the augmented matrix and perform a se-
quence of elementary row operations:
     
23 4 9 1 2 −1 2 1 2 −1 2
p t1 2 (−2)
1 2 −1 2 −−1→2 23 4 9 −−−−−→ 0 −1 6 5
   
1 2 −1 2 1 0 11 12
m2 (−1) t2 1 (−2)
−−−−−→ 0 1 −6 −5 −−−−−→ 0 1 −6 −5

The last augmented matrix corresponds to the system

x+ 11z = 12
y − 6z = −5.

In the first equation we can solve for x in terms of z and in the second equation
we can solve for y in terms of z. We refer to z as a free variable and let
z = α be a parameter in R. Then we obtain

x = 12 − 11α
y = −5 + 6α
z= α

In vector form we write

       
x 12 − 11α 12 −11
x =  y  =  −5 + 6α  = −5 + α  6  .
z α 0 1
 9.2 Systems of Linear Equations 415
 
12
The vector, xp = −5 is a particular solution ( corresponding to α = 0) while
0
 
−11
the vector xh =  6  generates the homogeneous solutions as α varies over
1
0
R. We have thus written the solution in the form xp + SA . J
Example 12. Find the solution set for the system
3x + 2y + z = 4
2x + 2y + z = 3
x + y + z = 0.
I Solution. Again we start with the augmented matrix and apply elemen-
tary row operations. Occasionally we will apply more than one operation at
a time. When this is so we stack the operations above the arrow with the
topmost operation performed first followed in order by the ones below it.
     
321 4 111 0 1 1 1 0
t1 2 (−2) 
 2 2 1 3  p1 3 2 2 1 3
t1 3 (−3)
0 0 −1 3 
−−
→
111 0 321 4 −−−−−→ 0 −1 −2 4
   
1 1 1 0 111 0
 0 −1 −2 4  m2 (−1)  0 1 2 −4 
p2 3
−−
→ m3 (−1)
0 0 −1 3 −−−−−→ 0 0 1 −3
   
110 3 100 1
t3 2 (−2) 
t3 1 (−1)
010 2 t2 1 (−1)  0 1 0 2
−−−−−→
−−−−−→ 0 0 1 −3 0 0 1 −3
The last augmented matrix corresponds to the system
x= 1
y= 2
z = −3.
 
1
The solution set is transparent: x =  2  . J
−3
0
In this example we note that SA = {0} so that the solution set SA
b
consists of
a single point: The system has a unique solution.
Example 13. Solve the following system of linear equations:
x + 2y + 4z = −2
x + y + 3z = 1
2x + y + 5z = 2
416 9 Matrices

I Solution. Again we begin with the augmented matrix and perform ele-
mentary row operations.
     
1 2 4 −2 1 2 4 −2 1 2 4 −2
1 1 3 1 12 t (−1)  0 −1 −1 3  m2 (−1)  0 1 1 −3 
t1 3 (−2) −−−−−→
2 1 5 2 −−−−−→ 0 −3 −3 6 0 −3 −3 6
   
1 2 4 −2 102 6
 0 1 1 −3  m3 (−1/3) 
t2 3 (3)
t2 1 (−2)
0 1 1 −3 
−−−− →
0 0 0 −3 −
−−−−−−→ 000 1
 
102 0
t3 1 (−6) 
t3 2 (3)
0 1 1 0.
−−−−−→ 000 1

The system that corresponds to the last augmented matrix is

x + 2z = 0
y+ z = 0
0 = 1.

The last equation, which is shorthand for 0x + 0y + 0z = 1, clearly has no

solution. Thus the system has no solution. In this case we write SA
b
= ∅. J

Reduced Matrices
These last three examples typify what happens in general and illustrate the
three possible outcomes discussed in Corollary 7: infinitely many solutions,
a unique solution, or no solution at all. The most involved case is when the
solution set has infinitely many solutions. In Example 11 a single parameter
α was needed to generate the set of solutions. However, in general, there may
be many parameters needed. We will always want to use the least number of
parameters possible, without dependencies amongst them. In each of the three
preceding examples it was transparent what the solution was by considering
the system determined by the last listed augmented matrix. The last matrix
was in a certain sense reduced as simple as possible.
We say that a matrix A is in row echelon form (REF) if the following
three conditions are satisfied.
1. The nonzero rows lie above the zero rows.
2. The first nonzero entry in a non zero row is 1. (We call such a 1 a leading
one.)
3. For any two adjacent nonzero rows the leading one of the upper row is to
the left of the leading one of the lower row. (We say the leading ones are
in echelon form.)
 9.2 Systems of Linear Equations 417

We say A is in row reduced echelon form (RREF) if it also satisfies

4 The entries above each leading one are zero.

Example 14. Determine which of the following matrices are row echelon
form, row reduced echelon form, or neither. For the matrices in row echelon
form determine the columns (C) of the leading ones. If a matrix is not in row
reduced echelon form explain which conditions are violated.
     
1 0 −3 11 2 01014 010
(1) 0 0 1 0 3 (2) 0 0 1 0 2 (3) 0 0 0
00 0 1 4 00000 001
   
10043 0   010 2
1 1 2 4 −7
(4) 0 2 1 2 0 2  (5) (6) 1 0 0 −2
0000 1
000000 001 0

I Solution. 1. (REF): leading ones are in the first, third and fourth col-
umn. It is not reduced because there is a nonzero entry above the leading
one in the third column.
2. (RREF): The leading ones are in the second and third column.
3. neither: The zero row is not at the bottom.
4. neither: The first non zero entry in the second row is not 1.
5. (REF): leading ones are in the first and fifth column. It is not reduced
because there is a nonzero entry above the leading one in the fifth column.
6. neither: The leading ones are not in echelon form.
J

The definitions we have given are for arbitrary matrices and not just ma-
trices that come from a system of linear equations; i.e. the augmented matrix.
Suppose though that a system Ax = b which has solutions is under con-
sideration. If the augmented matrix [A|b] is transformed by elementary row
operations to a matrix which is in row reduced echelon form the variables that
correspond to the columns where the leading ones occur are called the lead-
ing variables or dependent variables. All of the other variables are called
free variables. The free variables are sometimes replaced by parameters, like
α, β, . . .. Each leading variable can be solved for in terms of the free variables
alone. As the parameters vary the solution set is generated. The Gauss-Jordan
elimination method which will be explained shortly will always transform an
augmented matrix into a matrix that is in row reduced echelon form. This we
did in Examples 11, 12, and 13. In Example 11 the augmented matrix was
transformed to  
1 0 11 12
.
0 1 −6 −5
The leading variables are x and y while there is only one free variable, z. Thus
we obtained
418 9 Matrices
     
12 − 11α 12 −11
x =  −5 + 6α  = −5 + α  6  ,
α 0 1
where z is replace by the parameter α. In example 12 the augmented matrix
was transformed to  
100 1
0 1 0 2.
0 0 1 −3
In this case x, y, and z are leading variables; there are no free variables. The
solution set is  
1
x =  2 .
−3
In Example 13 the augmented matrix was transformed to
 
102 0
0 1 1 0.
000 1
In this case there are no solutions; the last row corresponds to the equation
0 = 1. There are no leading variable nor free variables.
These examples illustrate the following proposition which explains Corol-
lary 7 in terms of the augmented matrix in row reduced echelon form.
Proposition 15. Suppose Ax = b is a system of linear equations and the
augmented matrix [A|b] is transformed by elementary row operations to a
matrix [A0 |b0 ] which is in row reduced echelon form.
1. If a row of the form 0 . . . 0 | 1 appears in [A0 |b0 ] then there are no
 

solutions.  
2. If there are no rows of the form 0 . . . 0 | 1 and no free variables asso-
ciated with [A0 |b0 ] then there is a unique solution.
0 0
 or more free variables associated with [A |b ] and no rows
3. If there is one
of the form 0 . . . 0 | 1 then there are infinitely many solution.
Example 16. Suppose the following matrices are obtained by transforming
the augmented matrix of a system of linear equations using elementary row
operations. Identify the leading and free variables and write down the solution
set. Assume the variables are x1 , x2 , . . ..
   
1140 2 103 1 2  
110 1
(1) 0 0 0 1 3 (2) 0 1 1 −1 3 (3)
   
000 0
0000 0 000 0 0
 
  1 0 1  
100 3 0
 1 2 0120 2
(4)  0 1 0 4  0
(5)  0 1 (6)  0 0 0 1 0 
001 5 0 0 0 0000 0
0 0 0
 9.2 Systems of Linear Equations 419

I Solution. 1. The zero row provides no information and can be ignored.

The variables are x1 , x2 , x3 , and x4 . The leading ones occur in the first
and fourth column. Therefore x1 and x4 are the leading variables. The
free variables are x2 and x3 . Let α = x2 and β = x3 . The first row
implies the equation x1 + x2 + 4x3 = 2. We solve for x1 and obtain
x1 = 2 − x2 − 4x3 = 2 − α − 4β. The second row implies the equation
x4 = 3. Thus
         
x1 2 − α − 4β 2 −1 −4
x2   α  0 1 0
x= x3  = 
   =   + α  + β  ,
β  0 0 1
x4 3 3 0 0

where α and β are arbitrary parameters in R.

2. The leading ones are in the first and second column therefore x1 and x2
are the leading variables. The free variables of x3 and x4 . Let α = x3 and
β = x4 . The first row implies x1 = 2 − 3α − β and the second row implies
x2 = 3 − α + β. The solution is
       
2 − 3α − β 2 −3 −1
 3 − α + β  3 −1 1
x= 
 = 0 + α  1  + β  0  ,
      
α
β 0 0 1

where α, β are in R.
3. x1 is the leading variable. α = x2 and β = x3 are free variables. The first
row implies x1 = 1 − α. The solution is
       
1−α 1 −1 0
x =  α  = 0 + α  1  + β 0 ,
β 0 0 1

where α and β are in R.

4. The leading variables are x1 , x2 , and x3 . There are no free variables. The
solution set is  
3
x = 4 .
5
5. The row 0 0 1 implies the solution set is empty.
 

6. The leading variables are x2 and x4 . The free variables are α = x1 and
β = x3 . The first row implies x2 = 2 − 2β and the second row implies
x4 = 0. The solution set is
       
α 0 1 0
2 − 2β  2 0 −2
x= β  = 0 + α 0 + β  1  ,
      

0 0 0 0
420 9 Matrices

where α and β are in R.

J

The Gauss-Jordan Elimination Method

Now that you have seen several examples we present the Gauss-Jordan Elim-
ination Method for any matrix. It is an algorithm to transform any matrix to
row reduced echelon form using a finite number of elementary row operations.
When applied to an augmented matrix of a system of linear equations the so-
lution set can be readily discerned. It has other uses as well so our description
will be for an arbitrary matrix.
Algorithm 17. The Gauss-Jordan Elimination Method Let A be a ma-
trix. There is a finite sequence of elementary row operations that transform A
to a matrix in row reduced echelon form. There are two stages of the process:
(1) The first stage is called Gaussian elimination and transforms a given
matrix to row echelon form and (2) The second stage is called Gauss-Jordan
elimination and transforms a matrix in row echelon form to row reduced ech-
elon form.
From A to REF: Gaussian elimination
1. Let A1 = A. If A1 = 0 then A is in row echelon form.
2. If A1 6= 0 then in the first nonzero column from the left, ( say the j th
column) locate a nonzero entry in one of the rows: (say the ith row with
entry a.)
a) Multiply that row by the reciprocal of that nonzero entry. (mi (1/a))
b) Permute that row with the top row. (p1 i ) There is now a 1 in the
(1, j) entry.
c) If b is a nonzero entry in the (i, j) position for i 6= 1, add −b times the
first row to the ith row.(t1 j (−b)) Do this for each row below the first.
The transformed matrix will have the following form
 
0 ··· 0 1 ∗ ··· ∗
0 · · · 0 0 
 .. . . .. .
 
. . . A 
2
0 ··· 0 0

The *’s in the first row are unknown entries and A2 is a matrix with
fewer rows and columns than A1 .
3. If A2 = 0 we are done. The above matrix in in row echelon form.
4. If A2 6= 0, apply step (2) to A2 . Since there are zeros to the left of A2 and
the only elementary row operations we apply effect the rows of A2 (and
not all of A) there will continue to be zeros to the left of A2 . The result
will be a matrix of the form
 9.2 Systems of Linear Equations 421
 
0 ··· 0 1 ∗ ··· ∗ ∗ ∗ ··· ∗
 00 ··· 0 1 ∗ ··· ∗
 .. . . ..
 
. . 00

. ··· 0 0 .
.. .. .. ..
 
 
 . . . . A3 
0 ··· 0 0 0 ··· 0 0

5. If A3 = 0, we are done. Otherwise continue repeating step (2) until a

matrix Ak = 0 is obtained.
From REF to RREF: Gauss-Jordan Elimination
1. The leading ones now become apparent in the previous process. We begin
with the rightmost leading one. Suppose it is in the k th row and lth column.
If there is a nonzero entry (b say) above that leading one we add −b times
the k th row to it. (tk j (−b).) We do this for each nonzero entry in the lth
column. The result is zeros above the rightmost leading one. (The entries
to the left of a leading one are zeros. This process preserves that property.)
2. Now repeat the process described above to each leading one moving right
to left. The result will be a matrix in row reduced echelon form.

Example 18. Use the Gauss-Jordan method to row reduce the following
matrix to echelon form:  
2 3 8 04
 3 4 11 1 8
 1 2 5 1 6 .
 

−1 0 −1 0 1

I Solution. We will first write out the sequence of elementary row opera-
tions that transforms A to row reduced echelon form.
422 9 Matrices
   
2 3 8 0 4 1 2 5 16
 3 4 11 1 8  3 4 11 1 8 t1 2 (−3)
  p13   t1 3 (−2)
1 2 5 1 6 −→  2 3 8 0 4 t1 4 (1)
−−−−−→
−1 0 −1 0 1 −1 0 −1 0 1
   
1 2 5 1 6 1 2 5 1 6
0 −2 −4 −2 −10
 2 (−1/2) 0 1 2 1 5  t2 3 (1)
 
0 −1 −2 −2 −8  m

−−−−−−→ 0 −1 −2 −2 −8 t2 4 (−2)
− 
−−−−−→
0 2 4 1 7 0 2 4 1 7
   
12 5 1 6 125 16
0 1 2 1 5 m3 (−1)
0 1 2 1 5 t3 2 (−1)
   
0 0 0 −1 −3 t3 4 (1)
−−−−−→
0 0 0 1 3 t3 1 (−1)
−−−−−→
00 0 −1 −3 000 00
   
12 503 101 0 −1
0 1 2 0 2 0 1 2 0 2
  t2 1 (−2)  .
0 0 0 1 3 −−−−−→ 0 0 0 1 3
00 000 000 0 0

In the first step we observe that the first column is nonzero so it is possible to
produce a 1 in the upper left hand corner. This is most easily accomplished by
p1,3 . The next set of operations produces 0’s below this leading one. We repeat
this procedure on the submatrix to the right of the zeros’s. We produce a one in
the 2, 2 position by m2 (− 21 ) and the next set of operations produce zeros below
this second leading one. Now notice that the third column below the second
leading one is zero. There are no elementary row operations that can produce
a leading one in the (3, 3) position that involve just the third and fourth row.
We move over to the fourth column and observe that the entries below the
second leading one are not both zero. The elementary row operation m3 (−1)
produces a leading one in the (3, 4) position and the subsequent operation
produces a zero below it. At this point A has been transformed to row echelon
form. Now starting at the rightmost leading one, the 1 in the 3, 4 position,
we use operations of the form t3 i (a) to produce zeros above that leading one.
This is applied to each column that contains a leading one. J

The student is encouraged to go carefully through Examples 11, 12, and

13. In each of those examples the Gauss-Jordan Elimination method was used
to transform the augmented matrix to the matrix in row reduced echelon form.

Exercises
1. For each system of linear equations identify the coefficient matrix A, the variable
matrix x, the output matrix b and the augmented matrix [A|b].
 9.2 Systems of Linear Equations 423

x + 4y + 3z = 2
x + y − z =4 2x1 − 3x2 + 4x3 + x4 = 0
(a) (b)
2x + z =1 3x1 + 8x2 − 3x3 − 6x4 = 1
y − z =6

x1
1 0 −1 4 3  2
 x2 
 5 3 −3 −1 −3    

1

2. Suppose A = 
 3 −2
, x = x3 , and b =
 
. Write out the
8 4 −3 x4 
3 

−8 2 0 2 1 −4
x5
system of linear equations that corresponds to Ax = b.
In the following matrices identify those that are in row reduced echelon form.
If a matrix is not in row reduced echelon form find a single elementary row
operation that will transform it to row reduced echelon form and write the new
matrix.
10 1
3. 
0 0 0 
0 1 −4 
10 4
4.
0 1 2 
1210 1
5.
0 1 3 1 1 
010 3
6. 
0 0 2 6 
0000
0110 3
7. 0 0 0 1 2 

00000
1010 3
8. 0 1 3 4 1 

30309
Use elementary row operations to row reduce each matrix to row reduced echelon
form.
1 23 1
9. −1 0 3 −5

0 11 0 
2 1 31 0
10. 
1 −1 1 2 0 
0 2 112
0 −2 3 2 1

 
11.  2 −1 4 0 
0
0 6 −7 0 −2  
0 4 −6 −4 −2
1211 5

 
12. 2 4 0 0 6 
1 2 0 1 3 

00112
−1 0 1 1 0 0
13. −3 1 3 0 1 0 

7 −1 −4 0 0 1 
424 9 Matrices

1 2 4

2 4 8 
 
14. 
−1 2 0 

1 6 8
0 44
5181
1 1 4 0 

15.  
2 0 2 1

4171
2 8 006
16. 
 1 4 1 1 7
−1 −4 0 1 0
1 −1 1 −1 1
 1 1 −1 −1 1 

17.  
−1 −1 1 1 −1 
1 1 −1 1 −1
Solve the following systems of linear equations:
x + 3y = 2
18. 5x + 3z = −5
3x − y + 2z = −4

3x1 + 2x2 + 9x3 + 8x4 = 10

x1 + x3 + 2x4 = 4
19.
−2x1 + x2 + x3 − 3x4 = −9
x1 + x2 + 4x3 + 3x4 = 3

−x + 4y = −3x
20.
x − y = −3y

−2x1 − 8x2 − x3 − x4 = −9
21. −x1 − 4x2 − x4 = −8
x1 + 4x2 + x3 + x4 = 6

2x + 3y + 8z = 5
22. 2x + y + 10z = 3
2x + 8z = 4

x1 + x2 + x3 + 5x4 = 3
x2 + x3 + 4x4 = 1
23. x1 + x3 + 2x4 = 2
2x1 + 2x2 + 3x3 + 11x4 = 8
2x1 + x2 + 2x3 + 7x4 = 7

x1 + x2 = 3 + x1
24. x2 + 2x3 = 4 + x2 + x3
x1 + 3x2 + 4x3 = 11 + x1 + 2x2 + 2x3
 9.3 Invertible Matrices 425

1
1
25. Suppose the homogeneous system Ax = 0 has the following two solutions: 
2
1 5
and −1. Is 
−1 a solution? Why or why not?
0 4
26. For what value of k will the following system have a solution:
x1 + x2 − x3 = 2
2x1 + 3x2 + x3 = 4
x1 − 2x2 + 8x3 = k

1 34 1 1 1
27. Let A = −2 1 7, b1 =  0, b2 =  1,and b3 = 
1 .
1 1 0 0 0 1
a) Solve Ax = bi , for each i = 1, 2, 3.
b) Solve the above systems simultaneously by row reducing

134 11 1
−2 1 7 0 1 1 
[A|b1 |b2 |b3 ] = 
110 001 

9.3 Invertible Matrices

Let A be a square matrix. A matrix B is said to be an inverse of A if
BA = AB = I. In this case we say A is invertible or nonsingular. If A is
not invertible we say A is singular.
Example 1. Suppose  
3 1
A= .
−4 −1
Show that A is invertible and an inverse is
 
−1 −1
B= .
4 3
I Solution. Observe that
    
3 1 −1 −1 10
AB = =
−4 −1 4 3 01
and     
−1 −1 3 1 10
BA = = .
4 3 −4 −1 01
J
The following proposition says that when A has an inverse there can only
be one.
426 9 Matrices

Proposition 2. Let A be an invertible matrix. Then the inverse is unique.

Proof. Suppose B and C are inverses of A. Then
B = BI = B(AC) = (BA)C = IC = C.
u
t
Because of uniqueness we can properly  say the inverse of A when A is
−1 −1
invertible. In Example 1, the matrix B = is the inverse of A; there
4 3
are no others. It is standard convention to denote the inverse of A by A−1 .
For many matrices it is possible to determine their inverse by inspection.
For example, the identity matrix In is invertible and its inverse is In : In In =
In . A diagonal matrix diag(a1 , . . . , an ) is invertible if each ai 6= 0, i = 1, . . . , n.
The inverse then is simply diag( a11 , . . . , a1n ). However, if one of the ai is zero
then the matrix in not invertible. Even more is true. If A has a zero row, say
the ith row, then A is not invertible. To see this we get from Equation (2) in
Section 9.1 that Rowi (AB) = Rowi (A)B = 0. Hence, there is no matrix B for
which AB = I. Similarly, a matrix with a zero column cannot be invertible.
Proposition 3. Let A and B be invertible matrices. Then
1. A−1 is invertible and (A−1 )−1 = A.
2. AB is invertible and (AB)−1 = B −1 A−1 .
Proof. Suppose A and B are invertible. The symmetry of the equation
A−1 A = AA−1 = I says that A−1 is invertible and (A−1 )−1 = A. Also
(B −1 A−1 )(AB) = B −1 (A−1 A)B = B −1 IB = B −1 B = I and (AB)(B −1 A−1 ) =
A(B −1 B)A−1 = AA−1 = I. This shows (AB)−1 = B −1 A−1 . u
t
The following corollary easily follows:
Corollary 4. If A = A1 · · · Ak is the product of invertible matrices then A is
invertible and A−1 = A−1 −1
k · · · A1 .

Inversion Computations
Let ei be the column vector with 1 in the ith position and 0’s elsewhere. By
Equation (1) of Section 9.1 the equation AB = I implies that A Coli (B) =
Coli (I) = ei . This means that the solution to Ax = ei is the ith column of
the inverse of A, when A is invertible. We can thus compute the inverse of
A one column at a time using the Gauss-Jordan elimination method on the
augmented matrix [A|ei ]. Better yet, though, is to perform the Gauss-Jordan
elimination method on the matrix [A|I]. If A is invertible it will reduce to a
matrix of the form [I|B] and B will be A−1 . If A is not invertible it will not
be possible to produce the identity in the first slot.
We illustrate this in the following two examples.
 9.3 Invertible Matrices 427

Example 5. Determine whether the matrix

 
2 0 3
A = 0 1 1
3 −1 4

is invertible. If it is compute the inverse.

I Solution. We will augment A with I and follow the procedure outlined

above:
   
2 0 3 100 1 −1 1 −1 0 1
0 1 1 0 1 0 t13 (−1) 0 1 1 0 1 0 t1 3 (−2)
p1 3 t2 3 (−2)
3 −1 4 0 0 1 −−−−→ 2 0 3 1 0 0 −−−−→
   
1 −1 1 −1 0 1 m3 (−1) 1 −1 0 2 −2 −1
 0 1 1 0 1 0  t3 2 (−1)  0 1 0 3 −1 −2  t2 1 (1)
−−−→
0 0 −1 3 −2 −2 t−3−1− (−1)
−→ 0 0 1 −3 2 2
 
1 0 0 5 −3 −3
 0 1 0 3 −1 −2  .
0 0 1 −3 2 2
 
5 −3 −3
It follows that A is invertible and A−1 =  3 −1 −2 . J
−3 2 2
 
1 −4 0
Example 6. Let A = 2 1 3 . Determine whether A is invertible. If it is
0 −7 3
find its inverse.

I Solution. Again, we augment A with I and row reduce:

   
1 −4 0 100 1 −4 0 1 0 0
2 1 3 0 1 0  tt12 23 (−2)
(−1)
 0 9 3 −2 1 0 
0 9 3 0 0 1 −−−−→ 0 0 0 2 −1 1

We can stop at this point. Notice that the row operations produced a 0 row
in the reduction of A. This implies A cannot be invertible. J

Solving a System of Equations

Suppose A is a square matrix with a known inverse. Then the equation Ax = b
implies x = A−1 Ax = A−1 b and thus gives the solution.
428 9 Matrices

Example 7. Solve the following system:

2x + + 3z = 1
y+ z = 2
3x − y + 4z = 3.

I Solution. The coefficient matrix is

 
2 0 3
A = 0 1 1
3 −1 4

whose inverse we computed in the example above:

 
5 −3 −3
A−1 =  3 −1 −2 .
−3 2 2

The solution to the system is thus

    
5 −3 −3 1 −10
x = A−1 b =  3 −1 −2 2 =  −5  .
−3 2 2 3 7
J

Exercises
Determine whether the following matrices are invertible. If so, find the inverse:
1 1
1.
3 4 
3 2
2.
4 3 
1 −2
3.
 −4
2
1 −2
4.
 −4 
3
12 4
5. 0 1 −3

25 5 
11 1
6. 
0 1 2 
001
1 2 3
7. 4 5 1 

−1 −1 1 
 9.3 Invertible Matrices 429

1 0 −2
8. 2 −2 0 
1 2 −1
1 3 0 1
2 2 −2 0 

9. 1 −1 0 4


1 2 3 9
−1 1 1 −1
 1 −1 1 −1 

10.  1 1 −1 −1


−1 −1 −1 1
0100
1 0 1 0 

11. 0 1 1 1


1111
−3 2 −8 2

 0 2 −3 5 
12.  
1 2 3 5  
1 −1 1 −1
Solve each system Ax = b, where A and b are given below, by first computing
A−1 and and applying it to Ax = b to get x = A−1 b.
1 1 2
13. A = b=
3 4   
3
11 1 1
14. A =  0 1 2  b =  0 
001 −3 
1 0 −2 −2
15. A =  2 −2 0  b =  1 
1 2 −1 2
1 −1 1 1
16. A =  2 5 −2 b =  1
− 2 −1 1
1 3 0 1 1
2 2 −2 0 
 0 

17. A =    b =  

1 −1 0 4  −1
1 2 3 9 2
0100 1

1 0 1 0  −1 
18. A =   b=   
0 1 1 1
 −2 
1111 1
19. Suppose A is an invertible matrix. Show that At is invertible and give a formula
for the inverse.
cos θ sin θ
20. Let E(θ) = . Show E(θ) is invertible and find its inverse.
 sin θ cos θ 
−
sinh θ cosh θ
21. Let F (θ) = . Show F (θ) is invertible and find its inverse.
cosh θ sinh θ 
22. Suppose A is invertible and AB = AC. Show that B = C. Give an example of
a nonzero matrix A (not invertible) with AB = AC, for some B and C, but
B 6= C.
430 9 Matrices

9.4 Determinants
In this section we will discuss the definition of the determinant and some
of its properties. For our purposes the determinant is a very useful number
that we can associate to a square matrix. The determinant has an wide range
of applications. It can be used to determine whether a matrix is invertible.
Cramer’s rule gives the unique solution to a system of linear equations as
the quotient of determinants. In multidimensional calculus, the Jacobian is
given by a determinant and expresses how area or volume changes under a
transformation. Most students by now are familiar
 with the definition of the
ab
determinant for a 2 × 2 matrix: Let A = . The determinant of A is
cd
given by
det(A) = ad − bc.
It is the product of the diagonal
  entries minus the product of the off diagonal
1 3
entries. For example, det = 1 · (−2) − 5 · 3 = −17.
5 −2
The definition of the determinant for an n × n matrix is decidedly more
complicated. We will present an inductive definition. Let A be an n×n matrix
and let A(i, j) be the matrix obtained from A by deleting the ith row and j th
column. Since A(i, j) is an (n − 1) × (n − 1) matrix we can inductively define
the (i, j) minor, Minori j (A), to be the determinant of A(i, j):

Minori j (A) = det(A(i, j)).

The following theorem, whose proof is extremely tedious and we omit, is

the basis for the definition of the determinant.
Theorem 1 (Laplace expansion formulas). Suppose A is an n × n ma-
trix. Then the following numbers are all equal and we call this number the
determinant of A:
n
X
det A = (−1)i+j ai,j Minori j (A) for each i
j=1

and
n
X
det A = (−1)i+j ai,j Minori j (A) for each j.
i=1

Any of these formulas can thus be taken as the definition of the determi-
nant. In the first formula the index i is fixed and the sum is taken over all j.
The entries ai,j thus fill out the ith row. We therefore call this formula the
Laplace expansion of the determinant along the ith row or simply a
 9.4 Determinants 431

row expansion . Since the index i can range from 1 to n there are n row
expansions. In a similar way, the second formula is called the Laplace ex-
pansion of the determinant along the j th column or simply a column
expansion and there are n column expansions. The presence of the factor
(−1)i+j alternates the signs along the row or column according as i + j is
even or odd. The sign matrix
 
+ − + ···
− + − · · ·
 
+ − + · · ·
.. .. .. . .
 
. . . .

is a useful tool to organize the signs in an expansion.

It is common to use the absolute value sign |A| to denote the determinant
of A. This should not cause confusion unless A is a 1 × 1 matrix, in which
case we will not use this notation.

Example 2. Find the determinant of the matrix

 
1 2 −2
A = 3 −2 4  .
1 0 5

I Solution. For purposes of illustration we compute the determinant in two

ways. First, we expand along the first row.
     
−2 4
 − 2 3 4 + (−2) 3 −2 = 1 · (−10) − 2 · (11) − 2(2) = −36.
   
det A = 1 · 
0 5 15   1 0

Second, we expand along the second column.

     
3 4 1 −2 1 −2
det A = (−)2   + (−2) 
    (−)0 
  = (−2) · 11 − 2 · (7) = −36.
15 1 5 3 4

Of course, we get the same answer; that’s what the theorem guarantees. Ob-
serve though that the second column has a zero entry which means that we
really only needed to compute two minors. In practice we usually try to use
an expansion along a row or column that has a lot of zeros. Also note that we
use the sign matrix to adjust the signs on the appropriate terms. J

Properties of the determinant

The determinant has many important properties. The three listed below show
how the elementary row operations effect the determinant. They are used
extensively to simplify many calculations.
432 9 Matrices

Corollary 3. Let A be an n × n matrix. Then

1. det pi,j A = − det A.
2. det mi (a)A = a det A.
3. det ti,j (a) = det A.
 
rs
Proof. We illustrate the proof for the 2 × 2 case. Let A = . We then
tu
have
 
t u 
1. |p1,2 (A)| =   = ts − ru = −|A|.
r s  
 r s 
2. |t1,2 (a)(A)| =  
 = r(u + as) − s(t + ar) = |A|.
 t + ar
 u + as
ar as
3. |m1 (a)(A) =   = aru − ast = a|A|.
t u
u
t
Further important properties include:
1. If A has a zero row (or column) then det A = 0.
2. If A has two equal rows (or columns) then det A = 0.
3. det A = det At .
Example 4. Use elementary row operations to find det A if
 
  1 0 5 1
2 4 2 −1 2 1 3 
1) A = −1 3 5  and 2) A =   2 2 16 6  .

0 11
3 1 0 1
I Solution. Again we will write the elementary row operation that we have
used above the equal sign.
     
 2 4 2
 m1 ( 12 )  1 2 1  t12 (1)
  1 2 1 
  
1) −1 3 5 = 2 −1 3 5  = 2 0 5 6 
 0 1 1 0 11 0 1 1 
   
1 2 1  1 2 1 
p23   t23 (−5)  
= −2 0 1 1  = −2 0 1 1  = − 2.
0 5 6  0 0 1 
In the last equality we have used the fact that the last matrix is upper
triangular and its determinant is the product of the diagonal entries.

 1 0 5 1  t12 (1) 1 0 5 1 
   
  t13 (−2)  
−1 2 1 3 
 t14 (−3) 0 2 6 4  = 0,
 
2) 
 2 2 16 6  0 2 6 4 
  
3 1 0 1 = 0 1 −15 −2

because two rows are equal. J

 9.4 Determinants 433

In the following example we use elementary row operations to zero out

entries in a column and then use a Laplace expansion formula.
Example 5. Find the determinant of
 
1 42 −1
2 23 0
A= −1 1 2
.
4
0 13 2

I Solution.
   
 1 4 2 −1 1 4 2 −1
   
 2 2 3 0  t1,2 (−2) 0 −6 −1 2 
det(A) = 

 t1,3 (1) 
  
−1 1 2 4  = 0 5 4 3 

 0 13 2  0 1 3 2 
   
−6 −1 2
 t3,1 (6) 0 17 14 
 

=  5 4 3 t3,2 (−5) 0 −11 −7
 1 3 2 = 1 3 2 
   
 
 17 14 
−11 −7 = −119 + 154 = 35
=  

The following theorem contains two very important properties of the de-
terminant. We will omit the proof.

Theorem 6.
1. A square matrix A is invertible if and only if det A 6= 0.
2. If A and B are square matrices of the same size then

det(AB) = det A det B.

The cofactor and adjoint matrices

Again, let A be a square matrix. We define the cofactor matrix, Cof(A), of A
to be the matrix whose (i, j)-entry is (−1)i+j Minori,j . We define the adjoint
matrix, Adj(A), of A by the formula Adj(A) = (Cof(A))t . The important role
of the adjoint matrix is seen in the following theorem and its corollary.

Theorem 7. For A a square matrix we have

A Adj(A) = Adj(A) A = det(A)I.

434 9 Matrices

Proof. The (i, j) entry of A Adj(A) is

n
X n
X
Ai k (Adj(A))k j = (−1)k+j Ai k Minork j (A).
k=0 k=0

When i = j this is a Laplace expansion formula and is hence det A by Theorem

1. When i 6= j this is the expansion of a determinant for a matrix with two
equal rows and hence is zero. u
t

The following corollary immediately follows.

Corollary 8 (The adjoint inversion formula). If det A 6= 0 then
1
A−1 = Adj(A).
det A
 
ab
The inverse of a 2 × 2 matrix is a simple matter: Let A = . Then
cd
 
d −b
Adj(A) = and if det(A) = ad − bd 6= 0 then
−c a
 
1 d −b
A−1 = . (1)
ad − bc −c a
 
1 −3
For an example suppose A = . Then det(A) = 1 − (6) = −5 6= 0 so
−2 1
  " −1 −3 #
13 5 5
A is invertible and A−1 = −1 5 = −2 .
21 −1
5 5
The general formula for the inverse of a 3 × 3 is substantially more com-
plicated and difficult to remember. Consider though an example.

Example 9. Let  
1 2 0
A= 1 4 1 .
−1 0 3
Find its inverse if it is invertible.

I Solution. We expand  along

 the first
 row
 to compute the determinant
41 1 1
and get det(A) = 1 det − 2 det = 1(12) − 2(4) = 4. Thus A
03 −1 3
 
12 −4 4
is invertible. The cofactor of A is Cof(A) = −6 3 −2 and Adj(A) =
  2 −1 2
12 −6 2
Cof(A)t = −4 3 −1 . The inverse of A is thus
4 −2 2
 9.4 Determinants 435
 −3 1 

12 −6 2
 3 2 2
1
A−1 = −4 3 −1 = −1 3 −1 
4 .
 
4 4
4 −2 2 1 −1 1
2 2

J
In our next example we will consider a matrix with entries in R = R[s].
Such matrices will arise naturally in Chapter 10.
Example 10. Let  
121
A = 0 1 3 .
112
Find the inverse of the matrix
 
s − 1 −2 −1
sI − A =  0 s − 1 −3 
−1 −1 s − 2
.
I Solution. A straightforward computation gives det(sI −A) = (s−4)(s2 +
1). The matrix of minors for sI − A is
 
(s − 1)(s − 2) − 3 −3 s−1
 −2(s − 2) − 1 (s − 1)(s − 2) − 1 −(s − 1) − 2 .
6 + (s − 1) −3(s − 1) (s − 1)2
After simplifying somewhat we obtain the cofactor matrix
 2 
s − 3s − 1 3 s−1
 2s − 3 s2 − 3s + 1 s + 1  .
s+5 3s − 3 (s − 1)2
The adjoint matrix is
s2 − 3s − 1 2s − 3
 
s+5
 3 s2 − 3s + 1 3s − 3  .
s−1 s+1 (s − 1)2
Finally, we obtain the inverse:
 s2 −3s−1 2s−3 s+5

(s−4)(s2 +1) (s−4)(s2 +1) (s−4)(s2 +1)
s2 −3s+1
 
3 3s−3
(sI − A)−1 =  (s−4)(s2 +1) (s−4)(s2 +1) (s−4)(s2 +1)  .
 
(s−1)2
 
s−1 s+1
(s−4)(s2 +1) (s−4)(s2 +1) (s−4)(s2 +1)

J
436 9 Matrices

Cramer’s Rule
We finally consider a well known theoretical tool used to solve a system Ax =
b when A is invertible. Let A(i, b) denote the matrix obtained by replacing
the ith column of A with the column vector b. We then have the following
theorem:

Theorem 11. Suppose det A 6= 0. Then the solution to Ax = b is given

coordinate wise by the formula:

det A(i, b)
xi = .
det A
Proof. Since A is invertible we have
n
X
xi = (A−1 b)i = (A−1 )i k bk
k=1
n
1 X
= (−1)i+k Minork i (A)bk
det A
k=1
n
1 X det A(i, b)
= (−1)i+k bk Minork i (A) = .
det(A) det A
k=1

u
t

The following example should convince you that Cramer’s Rule is mainly
a theoretical tool and not a practical one for solving a system of linear equa-
tions. The Gauss-Jordan elimination method is usually far more efficient than
computing n + 1 determinants for a system Ax = b, where A is n × n.

Example 12. Solve the following system of linear equations using Cramer’s
Rule.

x + y +z= 0
2x + 3y − z = 11
x + z = −2

I Solution. We have
 
1 1 1 
 
det A = 2 3 −1 = −3,
1 0 1 
 
0 1 1
 
det A(1, b) =  11 3 −1 = −3,
−2 0 1 
 9.4 Determinants 437
 
1 0 1 
 
det A(2, b) = 2 11 −1 = −6,
1 −2 1 
 
1 1 0 
 
and det A(3, b) = 2 3 11  = 9,
1 0 −2
 
0
where b =  11 . Since det A 6= 0 Cramer’s Rule gives
−2

det A(1, b) −3
x1 = = = 1,
det A −3
det A(2, b) −6
x2 = = = 2,
det A −3
and
det A(3, b) 9
x3 = = = −3.
det A −3
J

Exercises
Find the determinant of each matrix given below in three ways: a row expansion,
a column expansion, and using row operations to reduce to a triangular matrix.
14
1.
2 9
11
2.
4 4
34
3.
2 6
1 1 −1
4. 1 4 0 

23 1 
40 3
5. 
8 1 7 
341
3 98 100
6. 
0 2 99 
0 0 1 
0 1 −2 4
2 3 9 2 

7.  
1 4 8 3  
−2 3 −2 4
438 9 Matrices

−4 9 −4 1
 2 3 0 −4 


8.  
−2 3 5 −6 
−3 2 0 1
24 2 3
1 2 1 4 

9. 
4 8 4 6 


1 9 11 13
Find the inverse of (sI −A) and determine for which values of s det(sI −A) = 0.
1 2
10.
1 2 
3 1
11.
1 3 
1 1
12.
 1
−1
10 1
0 1 0 
13. 
031
1 −3 3
−3 1 3 
14. 
3 −3 1 
0 4 0
15. −1 0 0 

1 4 −1 
Use the adjoint formula for the inverse for the matrices given below.
14
16.
2 9
11
17.
4 4
34
18.
2 6
1 1 −1
19. 
1 4 0 
23 1 
40 3
20. 
8 1 7 
341
3 98 100
21. 0 2 99 

0 0 1 
0 1 −2 4

 
22.  3 9 2 
2
1 4 8 3  
−2 3 −2 4
−4 9 −4 1
 2 3 0 −4 

23.  
−2 3 5 −6 
−3 2 0 1
 9.4 Determinants 439

24 2 3

1 2 1 4 

24.  
48 4 6 

1 9 11 13
10
Systems of Differential Equations

10.1 Systems of Differential Equations

10.1.1 Introduction
In the previous chapters we have discussed ordinary differential equations in
a single unknown function. These are adequate to model real world systems
as they evolve in time, provided that only one state, i.e., one number y(t),
is necessary to describe the system. For instance, we might be interested in
the way that the population of a species changes over time, the way the
temperature of an object changes over time, the way the concentration of
a pollutant in a lake changes over time, or the displacement over time of a
weight attached to a spring. In each of these cases, the system we wish to
describe is adequately represented by a single number. In the examples listed,
the number is the population p(t) at time t, the temperature T (t) at time t,
the concentration c(t) of a pollutant at time t, or the displacement y(t) of
the weight from equilibrium. However, a single ordinary differential equation
is inadequate for describing the evolution over time of a system which needs
more than one number to describe its state at a given time t. For example,
an ecological system consisting of two species will require two numbers p1 (t)
and p2 (t) to describe the population of each species at time t, i.e., to describe
a system consisting of a population of rabbits and foxes, you need to give
the population of both rabbits and foxes at time t. Moreover, the description
of the way this system changes with time will involve the derivatives p01 (t),
p02 (t), the functions p1 (t), p2 (t) themselves, and possibly the variable t. This
is precisely what is intended by a system of ordinary differential equations.
A system of ordinary differential equations is a system of equations
relating several unknown functions yi (t) of an independent variable t, some
of the derivatives of the yi (t), and possibly t itself. As for a single differential
equation, the order of a system of differential equations is the highest order
derivative which appears in any equation.
442 10 Systems of Differential Equations

Example 1. The following two equations

y10 = ay1 − by1 y2
(1)
y20 = −cy1 + dy1 y2
constitute a system of ordinary differential equations involving the unknown
functions y1 and y2 . Note that in this example the number of equations is
equal to the number of unknown functions. This is the typical situation which
occurs in practice.
Example 2. Suppose that a particle of mass m moves in a force field
F = (F1 , F2 , F3 ) that depends on time t, the position of the particle x(t) =
(x1 (t), x2 (t), x3 (t)) and the velocity of the particle x0 (t) = (x01 (t), x02 (t), x03 (t)).
Then Newton’s second law of motion states, in vector form, that F = ma,
where a = x00 is the acceleration. Writing out what this says in components,
we get a system of second order differential equations
mx001 (t) = F1 (t, x1 (t), x2 (t), x3 (t), x01 (t), x02 (t), x03 (t))
mx002 (t) = F2 (t, x1 (t), x2 (t), x3 (t), x01 (t), x02 (t), x03 (t)) (2)
mx003 (t) = F3 (t, x1 (t), x2 (t), x3 (t), x01 (t), x02 (t), x03 (t)).

In this example, the state at time t is described by six numbers, namely the
three coordinates and the three velocities, and these are related by the three
equations described above. The resulting system of equations is a second order
system of differential equations since the equations include second order deriv-
atives of some of the unknown functions. Notice that in this example we have
six states, namely the three coordinates of the position vector and the three
coordinates of the velocity vector, but only three equations. Nevertheless, it
is easy to put this system in exactly the same theoretical framework as the
first example by renaming the states as follows. Let y = (y1 , y2 , y3 , y4 , y5 , y6 )
where y1 = x1 , y2 = x2 , y3 = x3 , y4 = x01 , y5 = x02 , and y6 = x03 . Using these
new function names, the system of equations (2) can be rewritten using only
first derivatives:
y10 = y4
y20 = y5
y30 = y6
1
y40 = F1 (t, y1 , y2 , y3 , y4 , y5 , y6 ) (3)
m
1
y50 = F2 (t, y1 , y2 , y3 , y4 , y5 , y6 )
m
1
y60 = F3 (t, y1 , y2 , y3 , y4 , y5 , y6 ).
m
Note that this can be expressed as a vector equation

y0 = f (t, y)
 10.1 Systems of Differential Equations 443

where
1 1 1
f (t, y) = (y4 , y5 , y6 , F1 (t, y), F2 (t, y), F3 (t, y)).
m m m
The trick used in Example 2 to reduce the second order system to a first
order system in a larger number of variables works in general, so that it is
only really necessary to consider first order systems of differential equations.
As with a single first order ordinary differential equation, it is convenient
to consider first order systems in a standard form for purposes of describing
properties and solution algorithms for these systems.
Definition 3. The standard form for a first order system of ordinary dif-
ferential equations is a vector equation of the form

y0 = f (t, y) (4)

where f : U → Rn is a function from an open subset U of Rn+1 to Rn . If an

initial point t0 and an initial vector y 0 are also specified, then one obtains an
initial value problem:

y0 = f (t, y), y(t0 ) = y 0 . (5)

A solution of Equation (4) is a differentiable vector function y : I → Rn

where I is an open interval in R and the function y satisfies Equation (4) for
all t ∈ I. This means that

y0 (t) = f (t, y(t)) (6)

for all t ∈ I. If also y(t0 ) = y 0 , then y(t) is a solution of the initial value
problem (5).
Equation (1) is a system in standard form where n = 2. That is, there
are two unknown functions y1 and y2 which can be incorporated into a two
dimensional vector y = (y1 , y2 ), and if f (t, y) = (f1 (t, y1 , y2 ), f2 (t, y1 ,2 )) =
(ay1 − by1 y2 , −cy1 + dy1 y2 ), then Equation (5) is a short way to write the
system of equations

y10 = ay1 − by1 y2 = f1 (t, y1 , y2 )

(7)
y20 = −cy1 + dy1 y2 = f2 (t, y1 , y2 ).

Equation (3) of Example 2 is a first order system with n = 6. We shall

primarily concentrate on the study of systems where n = 2 or n = 3, but
Example 2 shows that even very simple real world systems can lead to systems
of differential equations with a large number of unknown functions.
Example 4. Consider the following first order system of ordinary differen-
tial equations:
y10 = 3y1 − y2
(8)
y20 = 4y1 − 2y2 .
444 10 Systems of Differential Equations

1. Verify that y(t) = (y1 (t), y2 (t)) = (e2t , e2t ) is a solution of Equation (8).
I Solution. Since y1 (t) = y2 (t) = e2t ,
y10 (t) = 2e2t = 3e2t − e2t = 3y1 (t) − y2 (t)
and y20 (t) = 2e2t = 4e2t − 2e2t = 4y1 (t) − 2y2 (t),
which is precisely what it means for y(t) to satisfy (8). J
2. Verify that z(t) = (z1 (t), z2 (t)) = (e−t , 4e−t ) is a solution of Equation
(8).
I Solution. As above, we calculate
z10 (t) = −e−t = 3e−t − 4e−t = 3z1 (t) − z2 (t)
and z20 (t) = −4e−t = 4e−t − 2 · 4e−t = 4z1 (t) − 2z2 (t),
which is precisely what it means for z(t) to satisfy (8). J
3. If c1 and c2 are any constants, verify that w(t) = c1 y(t) + c2 z(t) is also
a solution of Equation (8).
I Solution. Note that w(t) = (w1 (t), w2 (t)), where w1 (t) = c1 y1 (t) +
c2 z1 (t) = c1 e2t + c2 e−t and w2 (t) = c1 y2 (t) + c2 z2 (t) = c1 e2t + c2 4e−t .
Then
w10 (t) = 2c1 e2t − c2 e−t = 3w1 (t) − w2 (t)
and w20 (t) = 2c1 e−t − 4c2 e−t = 4w1 (t) − 2w2 (t).
Again, this is precisely what it means for w(t) to be a solution of (8).
We shall see in the next section that w(t) is, in fact, the general solution
of Equation (8). That is, any solution of this equation is obtained by a
particular choice of the constants c1 and c2 . J
Example 5. Consider the following first order system of ordinary differen-
tial equations:
y10 = 3y1 − y2 + 2t
(9)
y20 = 4y1 − 2y2 + 2.
Notice that this is just Equation (8) with one additional term (not involving
the unknown functions y1 and y2 ) added to each equation.
1. Verify that y p (t) = (yp1 (t), yp2 (t)) = (−2t + 1, −4t + 5) is a solution of
Equation (9).
I Solution. Since yp1 (t) = −2t + 1 and yp2 (t) = −4t + 5, direct calcu-
0 0
lation gives yp1 (t) = −2, yp2 (t) = −4 and
0
3yp1 (t) − yp2 (t) + 2t = 3(−2t + 1) − (−4t + 5) + 2t = −2 = yp1 (t)
0 0 0
and 4yp1 (t) − 2yp2 (t) + 2 = 4(−2t + 1) − 2(−4t + 5) + 2 = −4 = yp2 (t).
Hence y p (t) is a solution of (9). J
 10.1 Systems of Differential Equations 445

2. Verify that z p (t) = 2y p (t) = (zp1 (t), zp2 (t)) = (−4t + 2, −8t + 10) is not
a solution to Equation (9).
I Solution. Since
0
3zp1 (t)−zp2 (t)+2t = 3(−4t+2)−(−8t+10)+2t = −2t−4 6= −4 = zp1 (t),

z p (t) fails to satisfy the first of the two equations of (9), and hence is not
a solution of the system. J
3. We leave it as an exercise to verify that y g (t) = w(t)+y p (t) is a solution of
(9), where w(t) is the general solution of Equation (8) from the previous
example.
We will now list some particular classes of first order systems of ordinary
differential equations. As for the case of a single differential equation, it is
most convenient to identify these classes by describing properties of the right
hand side of the equation when it is expressed in standard form.
Definition 6. The first order system in standard form

y0 = f (t, y)

is said to be
1. autonomous if f (t, y) is independent of t;
2. linear if f (t, y) = A(t)y + q(t) where A(t) = [aij (t)] is an n × n matrix
of functions and q(t) = (q1 (t), . . . , qn (t)) is a vector of functions of t;
3. constant coefficient linear if f (t, y) = Ay + q(t) where A = [aij ] is an
n × n constant matrix and q(t) = (q1 (t), . . . , qn (t)) is a vector of functions
of t;
4. linear and homogeneous if f (t, y) = A(t)y. That is, a system of linear
ordinary differential equations is homogeneous provided the term q(t) is
0.
In the case n = 2, a first order system is linear if it can be written in the
form
y10 = a(t)y1 + b(t)y2 + q1 (t)
y20 = c(t)y1 + d(t)y2 + q2 (t);

this linear system is homogeneous if

y10 = a(t)y1 + b(t)y2

y20 = c(t)y1 + d(t)y2 ,

and it is a constant coefficient linear system if

y10 = ay1 + by2 + q1 (t)

y20 = cy1 + dy2 + q2 (t)
446 10 Systems of Differential Equations
 
a(t) b(t)
In the first two cases the matrix of functions is A(t) = , while in the
  c(t) d(t)
ab
third case, the constant matrix is A = . Notice that the concepts constant
cd
coefficient and autonomous are not identical for linear systems of differential
equations. The linear system y0 = A(t)y+q(t) is constant coefficient provided
all entries of A(t) are constant functions, while it is autonomous if all entries
of both A(t) and q(t) are constants.
Example 7. 1. The linear system (1) of Example 1 is autonomous, but
not linear.
2. The system
y10 = y2
1
y20 = −y1 − y2
t
is linear and homogeneous, but not autonomous.
3. The system
y10 = −y2
y20 = y1
is linear, constant coefficient, and homogeneous (and hence autonomous).
4. The system
y10 = −y2 + 1
y20 = y1
is linear and autonomous (and hence constant coefficient) but not homo-
geneous.
5. The system
y10 = −y2 + t
y20 = y1
is constant coefficient, but not autonomous or homogeneous.
Note that the term autonomous applies to both linear and nonlinear sys-
tems of ordinary differential equations, while the term constant coefficient
applies only to linear systems of differential equations, and as the examples
show, even for linear systems, the terms constant coefficient and autonomous
do not refer to the same systems.

10.1.2 Examples of Linear Systems

In this section we will look at some situations which give rise to systems of or-
dinary differential equations. Our goal will be to simply set up the differential
equations; techniques for solutions will come in later sections.
 10.1 Systems of Differential Equations 447

Example 8. The first example is simply the observation that a single or-
dinary differential equation of order n can be viewed as a first order system
of n equations in n unknown functions. We will do the case for n = 2; the
extension to n > 2 is straightforward. Let

y 00 = f (t, y, y 0 ), y(t0 ) = a, y 0 (t0 ) = b (10)

be a second order initial value problem. By means of the identification y1 = y,

y2 = y 0 , Equation (10) can be identified with the system

y10 = y2 y1 (t0 ) = a
(11)
y20 = f (t, y1 , y2 ) y2 (t0 ) = b

For a numerical example, consider the second order initial value problem

(∗) y 00 − y = t, y(0) = 1, y 0 (0) = 2.

According to Equation (11), this is equivalent to the system

y10 = y2 y1 (0) = 1
(∗∗)
y20 = y1 + t y2 (0) = 2.

Equation (∗) can be solved by the techniques of Chapter 6 to give a solution

y(t) = 2et − e−t − t. The corresponding solution to the system (∗∗) is the
vector function

y(t) = (y1 (t), y2 (t)) = (2et − e−t − t, 2et + e−t − 1),

where y1 (t) = y(t) = 2et − e−t − t and y2 (t) = y 0 (t) = 2et + e−t − 1.

Example 9 (Predator-Prey System). In Example ?? we intro-

duced two differential equation models for the growth of population of a single
species. These were the Malthusian proportional growth model, given by the
differential equation p0 = kp (where, as usual, p(t) denotes the population at
time t), and the Verhulst model which is governed by the logistic differential
equation p0 = c(m − p)p, where c and m are constants. In this example we
will consider an ecological system consisting of two species, where one species,
which we will call the prey, is the food source for another species which we
will call the predator. For example we could have coyotes (predator) and rab-
bits (prey) or sharks (predators) and food fish (prey). Let p1 (t) denote the
predator population at time t and let p2 (t) denote the prey population at time
t. Using some assumptions we may formulate potential equations satisfied by
the rates of change of p1 and p2 . To talk more succinctly, we will assume that
the predators are coyotes, and the prey are rabbits. Let us assume that if there
are no coyotes then the rabbit population will increase at a rate proportional
to the current population, that is p02 (t) = ap2 (t) where a is a positive constant.
Since the coyotes eat the rabbits, we may assume that the rate at which the
448 10 Systems of Differential Equations

rabbits are eaten is proportional to the number of contacts between coyotes

and rabbits, which we may assume is proportional to p1 (t)p2 (t); this will, of
course, have a negative impact upon the rabbit population. Combining the
growth rate (from reproduction) and the rate of decline (from being eaten by
coyotes), we arrive at p02 (t) = ap2 (t)−bp1(t)p2 (t) where b is a positive constant
as a formula expressing the rate of change of the rabbit population. A similar
reasoning will apply to the coyote population. If no rabbits are present, then
the coyote population will die out, and we will assume that this happens at a
rate proportional to the current population. Thus p01 (t) = −cp1 (t) where c is
a positive constant is the first approximation. Moreover, the increase in the
population of coyotes is dependent upon interactions with their food supply,
i.e., rabbits, so a simple assumption would be that the increase is propor-
tional to the number of interactions between coyotes and rabbits, which we
can take to be proportional to p1 (t)p2 (t). Thus, combining the two sources of
change in the coyote population gives p01 (t) = −cp1 (t)+dp1 (t)p2 (t). Therefore,
the predator and prey populations are governed by the first order system of
differential equations

p01 (t) = −cp1 (t) + dp1 (t)p2 (t)

(12)
p02 (t) = ap2 (t) − bp1 (t)p2 (t).
 
p (t)
If we let p(t) = 1 then Equation (12) can be expressed as the vector
p2 (t)
equation
p0 (t) = f (t, p(t)), (13)
where  
−cu1 + du1 u2
f (t, u) =
au2 − bu1 u2
 
u1
and u = , which is a more succinct way to write the system (12) for many
u2
purposes. We shall have more to say about this system in a later section.

Example 10 (Mixing problem). Example 6 considers the case of com-

puting the amount of salt in a tank at time t if a salt mixture is flowing into
the tank at a known volume rate and concentration and the well-stirred mix-
ture is flowing out at a known volume rate. What results is a first order linear
differential equation for the amount y(t) of salt at time t (Equation (23)). The
current example expands upon the earlier example by considering the case of
two connected tanks. See Figure 10.1. Tank 1 contains 200 gallons of brine
which 50 pounds of salt are initially dissolved ; Tank 2 initially contains 200
gallons of pure water. Moreover, the mixtures are pumped between the two
tanks, 6 gal/min from Tank 1 to Tank 2 and 2 gal/min going from Tank 2
back to Tank 1. Assume that a brine mixture containing .5 lb/gal enters Tank
1 at a rate of 4 gal/min, and the well-stirred mixture is removed from Tank
2 at the same rate of 4 gal/min. Let y1 (t) be the amount of salt in Tank 1 at
 10.1 Systems of Differential Equations 449

4 Gal/Min
6 Gal/Min

Tank 1 Tank 2

4 Gal/Min
200 Gal 200 Gal
2 Gal/Min

Fig. 10.1. A Two Tank Mixing Problem.

time t and let y2 (t) be the amount of salt in Tank 2 at time t. Find a system
of differential equations which relates y1 (t) and y2 (t).
I Solution. The underlying principle is the same as that of the single tank
mixing problem. Namely, we apply the balance equation
(†) y 0 (t) = Rate in − Rate out
to the amount of salt in each tank. If y1 (t) denotes the amount of salt at
time t in Tank 1, then the concentration of salt at time t in Tank 1 is
c1 (t) = (y1 (t)/200) lb/gal. Similarly, the concentration of salt in Tank 2 at
time t is c2 (t) = (y2 (t)/200) lb/gal. The relevant rates of change can be
summarized in the following table.
From To Rate
Outside Tank 1 (0.5 lb/gal)·(4 gal/min) = 2 lb/min
y1 (t)
Tank 1 Tank2 · 6 gal/min = 0.03y1 (t) lb/min
200
y2 (t)
Tank 2 Tank 1 lb/gal · 2 gal/min = 0.01y2 (t) lb/min
200
y2 (t)
Tank 2 Outside lb/gal · 4 gal/min = 0.02y2 (t) lb/min
200

The data for the balance equations (†) can then be read from the following
table:
Tank Rate in Rate out
1 2 + 0.01y2 (t) 0.03y1 (t)
2 0.03y1(t) 0.02y2 (t)
450 10 Systems of Differential Equations

Putting these data in the balance equations then gives

y10 (t) = 2 + 0.01y2(t) − 0.03y1 (t)

y20 (t) = 0.03y1(t) − 0.02y2 (t)

as the first order system of ordinary differential equations satisfied by the

vector function whose two components are the amount of salt in tank 1 and
in tank 2 at time t. This system is a nonhomogeneous, constant coefficient,
linear system. We shall address some techniques for solving such equations in
Section 10.4, after first considering some of the theoretical underpinnings of
these equations in the next two sections. J

Exercises
For each of the following systems of differential equations, determine if it is
linear (yes/no) and autonomous (yes/no). For each of those which is linear, fur-
ther determine if the equation is homogeneous/nonhomogeneous and constant
coefficient (yes/no). Do not solve the equations.

y10 = y2
1.
y20 = y1 y2

y10 = y1 + y2 + t2
2.
y20 = −y1 + y2 + 1

y10 = (sin t)y1 − y2

3.
y20 = y1 + (cos t)y2

y10 = t sin y1 − y2
4.
y20 = y1 + t cos y2

y10 = y1
y0 = 2y1 + y4
5. 20
y3 = y4
y40 = y2 + 2y3

1
y10 = y1 − y2 + 5
6. 2
1
y20 = −y1 + y2 − 5
2
y1 (t)
7. Verify that y(t) = , where y1 (t) = et − e3t and y2 (t) = 2et − e3t is a
y2 (t)
solution of the initial value problem

5 −2 0
y0 = y; y(0) = .
 −1
4 1
 10.1 Systems of Differential Equations 451

Solution: First note that y1 (0) = 0 and y2 (0) = 1, so the initial condi-
y10 (t) et − 3e3t 5 −2
tion is satisfied. Then y 0 (t) = = t 3t while y(t) =
0
y2 (t)3t 2e − 3e   −1
4
t 3t t 3t t
5(e − e ) − 2(2e − e ) e − 3e 5 −2
3t . Thus y (t) = y, as required.
0
t 3t t 3t = t
 4(e − e ) − (2e − e )  2e − 3e   −1
4

y1 (t)
8. Verify that y(t) = , where y1 (t) = 2e4t − e−2t and y2 (t) = 2e4t + e−2t is
y2 (t)
a solution of the initial value problem

13 1
y0 = y; y(0) = .
3 1 3
y1 (t)
9. Verify that y(t) = , where y1 (t) = et + 2tet and y2 (t) = 4tet is a solution
y2 (t)
of the initial value problem

3 −1 1
y0 = y; y(0) = .
4 −1 0
y1 (t)
10. Verify that y(t) = , where y1 (t) = cos 2t − 2 sin 2t and y2 (t) = − cos 2t is
y2 (t)
a solution of the initial value problem

1 5 1
y0 = y; y(0) = .
 −1
−1  
−1

Rewrite each of the following initial value problems for an ordinary differen-
tial equation as an initial value problem for a first order system of ordinary
differential equations.

11. y 00 + 5y 0 + 6y = e2t , y(0) = 1, y 0 (0) = −2.

Solution: Let y1 = y and y2 = y 0 . Then y10 = y 0 = y2 and y20 = y 00 = −5y 0 −
y1
6y + e2t = −6y1 − 5y2 + e2t . Letting y = , this can be expressed in vector
y2 
form (see Equation (6.1.7)) as

0 1 0 1
y0 = y + 2t ; y(0) = .
−6 −5   
e  
−2

12. y 00 + k2 y = 0, y(0) = −1, y 0 (0) = 0

13. y 00 − k2 y = 0, y(0) = −1, y 0 (0) = 0

14. y 00 + k2 y = A cos ωt, y(0) = 0, y 0 (0) = 0

15. ay 00 + by 0 + cy = 0, y(0) = α, y 0 (0) = β

16. ay 00 + by 0 + cy = A sin ωt, y(0) = α, y 0 (0) = β

17. t2 y 00 + 2ty 0 + y = 0, y(1) = −2, y 0 (1) = 3

452 10 Systems of Differential Equations

10.2 Linear Systems of Differential Equations

This section and the next will be devoted to the theoretical underpinnings
of linear systems of ordinary differential equations which accrue from the
main theorem of existence and uniqueness of solutions of such systems. As
with Picard’s existence and uniqueness theorem (Theorem 2) for first order
ordinary differential equations, and the similar theorem for linear second order
equations (Theorem ??), we will not prove this theorem, but rather show how
it leads to immediately useful information to assist us in knowing when we
have found all solutions.
A first order system y 0 = f (t, y) in standard form is linear provided
f (t, y) = A(t)y + q(t) where A(t) = [aij (t)] is an n × n matrix of functions,
while    
q1 (t) y1
 ..   .. 
q(t) =  .  and y =  . 
qn (t) yn
are n× 1 matrices. Thus the standard description of a first order linear system
in matrix form is
y 0 = A(t)y + q(t), (1)
while, if the matrix equation is written out in terms of the unknown functions
y1 , y2 , . . ., yn , then (1) becomes

y10 = a11 (t)y1 + a12 (t)y2 + · · · + a1n (t)yn + q1 (t)

y20 = a21 (t)y1 + a22 (t)y2 + · · · + a2n (t)yn + q2 (t)
(2)
.................................................
yn0 = an1 (t)y1 + an2 (t)y2 + · · · + ann (t)yn + qn (t).

For example, the matrix equation

   
0 1 −t cos t
y = −t y+
e −1 0

and the system of equations

y10 = y1 − ty2 + cos t

y20 = e−t y1 − y2

have the same meaning.

It is convenient to state most of our results on linear systems of ordinary
differential equations in the language of matrices and vectors. To this end the
following terminology will be useful. A property P of functions will be said to
be satisfied for a matrix A(t) = [aij (t)] of functions if it is satisfied for all of
the functions aij (t) which make up the matrix. In particular:
1. A(t) is defined on an interval I of R if each aij (t) is defined on I.
 10.2 Linear Systems of Differential Equations 453

2. A(t) is continuous on an interval I of R if each aij (t) is continuous on

I. For instance, the matrix
1
 
cos 2t
A(t) =  t + 2
 
−2t 1 
e 2
(2t − 3)

is continuous on each of the intervals I1 = (−∞, −2), I2 = (−2, 3/2) and

I3 = (3/2, ∞), but it is not continuous on the interval I4 = (0, 2).
3. A(t) is differentiable on an interval I of R if each aij (t) is differentiable
on I. Moreover, A0 (t) = [a0ij (t)]. That is, the matrix A(t) is differentiated
by differentiating each entry of the matrix. For instance, for the matrix
A(t) in the previous item,
 −1 
2
−2 sin 2t
 (t + 2)
A0 (t) =  −4  .

−2t
−2e
(2t − 3)3

4. A(t) is integrable on an interval I of R if each aij (t) is integrable on I.

Moreover, the integral of A(t) on the interval [a, b] is computed
hR by comput-
i
Rb b
ing the integral of each entry of the matrix, i.e., a A(t) dt = a aij (t) dt .
For the matrix A(t) of item 2 above, this gives
1
R 
1 R1
Z 1 dt cos 2t dt 
ln 32 1
sin 2

 0 t+2 0
2
A(t) dt =  R 1 = ,

1 1 −2 1
2 (1 − e )
−2t
R 1 
0 e dt 0 dt
0 3
(2t − 3) 2

while, if t ∈ I2 = (−2, 3/2), then

1 t+2
R   
t Rt 1
Z t du cos 2u du ln 2 sin 2t
 0 0
A(u) du =  R t u + 2  = 1 2 .
  
−2u
R 1 1 −2t −1 1
0
0
e du 0 (2u − 3)2
du 2 (1 − e ) − 6
2(2t − 3)

5. If each entry aij (t) of A(t) is of exponential type (see the definition on page
347), we can take the Laplace transform of A(t), by taking the Laplace
transformof each entry. That  is L(A(t))(s) = [L(aij (t))(s)]. For example,
−2t
te cos 2t
if A(t) = 3t , this gives
e sin t (2t − e)2

1 2
 
L te−2t (s) L(cos 2t)(s)
"
#
 (s + 2)2 s2 + 4 
L(A(t))(s) = = 3t  .
 

3t 2 1 8e


L(e sin t)(s) L (2t − 3) (s)  2 
2
(s − 3) + 1 s 3
454 10 Systems of Differential Equations

If A(t) and q(t) are continuous matrix functions on an interval I, then a

solution to the linear differential equation
y 0 = A(t)y + q(t)
on the subinterval J ⊆ I, is a continuous matrix function y(t) on J such that
y 0 (t) = A(t)y(t) + q(t)
for all t ∈ J. If moreover, y(t0 ) = y 0 , then y(t) is a solution of the initial
value problem
y 0 = A(t)y + q(t), y(t0 ) = y 0 . (3)
 3t 
e
Example 1. Verify that y(t) = is a solution of the initial value
3e3t
problem (3) on the interval (−∞, ∞) where
     
0 1 0 1
A(t) = , q(t) = , t0 = 0 and y 0 = .
6 −1 6e3t 3
I Solution. All of the functions in the matrices A(t), y(t), and q(t)
 are
1
differentiable on the entire real line (−∞, ∞) and y(t0 ) = y(0) = = y0 .
3
Moreover,
 3t 
3e
(∗) y 0 (t) =
9e3t
and
e3t
      3t 
0 1 0 3e
(∗∗) A(t)y(t) + q(t) = + = .
6 −1 3e3t 6e3t 9e3t
Since (∗) and (∗∗) agree, y(t) is a solution of the initial value problem. J

The Existence and Uniqueness Theorem

The following result is the fundamental foundational result of the current
theory. It is the result which guarantees that if we can find a solution of a
linear initial value problem by any means whatsoever, then we know that we
have found the only possible solution.
Theorem 2 (Existence and Uniqueness). 1 Suppose that the n× n matrix
function A(t) and the n × 1 matrix function q(t) are both continuous on an
interval I in R. Let t0 ∈ I. Then for every choice of the vector y 0 , the initial
value problem
y 0 = A(t)y + q(t), y(t0 ) = y 0
has a unique solution y(t) which is defined on the same interval I.
1
A Proof of this result can be found in the text An Introduction to Ordinary
Differential Equations by Earl Coddington, Prentice-Hall, (1961), Page 256.
 10.2 Linear Systems of Differential Equations 455

Remark 3. How is this theorem related to existence and uniqueness theorems

we have stated previously?
• If n = 1 then this theorem is just Corollary 9. In this case we have actually
proved the result by exhibiting a formula for the unique solution. We are
not so lucky for general n. There is no formula like Equation (7) which is
valid for the solutions of linear initial value problems if n > 1.
• Theorem ?? is a corollary of Theorem 2. Indeed, if n = 2,
 
0 1
A(t) = ,
−b(t) −a(t)
     
0 y0 y
q(t) = , y0 = , and y = , then the second order linear
f (t) y1 y0
initial value problem
y 00 + a(t)y 0 + b(t)y = f (t), y(t0 ) = y0 , y 0 (t0 ) = y1
has the solution y(t) if and only if the first order linear system
y 0 = A(t)y + q(t), y(t0 ) = y 0
 
y(t)
has the solution y(t) = 0 . You should convince yourself of the validity
y (t)
of this statement.
Example 4. Let n = 2 and consider the initial value problem (3) where
1
 
−t  −t 
e
 
1
A(t) =  1
 t + 1  , q(t) = cos t , t0 = 0, y 0 = 2 .

2
t
t2 − 2
Determine the largest interval I on which a solution to (3) is guaranteed by
Theorem 2.
I Solution. All of the entries in all the matrices above are continuous on
1 1
the entire real line except that is not continuous for t = −1 and 2
√ t + 1 t −2
is not continuous for t = ± 2. Thus the largest interval √ I containing 0 for
which all of the matrix entries are continuous is I = (−1, 2). The theorem
applies on this interval, and on no larger interval containing 0. J
For first order differential equations, the Picard approximation algorithm
(Algorithm 1) provides an algorithmic procedure for finding an approximate
solution to a first order initial value problem y 0 = f (t, y), y 0 (t0 ) = y0 . For first
order systems, the Picard approximation algorithm also works. We will state
the algorithm only for linear first order systems and then apply it to constant
coefficient first order systems, where we will be able to see an immediate
analogy to the simple linear equation y 0 = ay, y 0 (0) = c, which, as we know
from Chapter 2, has the solution y(t) = ceat .
456 10 Systems of Differential Equations

Algorithm 5 (Picard Approximation for Linear Systems). Perform

the following sequence of steps to produce an approximate solution to the
initial value problem (3).
(i) A rough initial approximation to a solution is given by the constant func-
tion
y 0 (t) := y 0 .
(ii) Insert this initial approximation into the right hand side of Equation (3)
and obtain the first approximation
Z t
y 1 (t) := y 0 + (A(u)y 0 (u) + q(u)) du.
t0

(iii)The next step is to generate the second approximation in the same way;
i.e., Z t
y 2 (t) := y 0 + (A(u)y 1 (u) + q(u)) du.
t0

(iv)At the n-th stage of the process we have

Z t
y n (t) := y 0 + (A(u)y n−1 (u) + q(u)) du,
t0

which is defined by substituting the previous approximation y n−1 (t) into

the right hand side of Equation (3).

As in the case of first order equations, under the hypotheses of the Ex-
istence and Uniqueness for linear systems, the sequence of vector functions
y n (t) produced by the Picard Approximation algorithm will converge on the
interval I to the unique solution of the initial value problem (3).

Example 6. We will consider the Picard Approximation algorithm in the

special case where the coefficient matrix A(t) is constant, so that we can write
A(t) = A, the function q(t) = 0, and the initial point t0 = 0. In this case the
initial value problem (3) becomes

y 0 = Ay, y(0) = y 0 , (4)

and we get the following sequence of Picard approximations y n (t) to the

solution y(t) of (4).
 10.2 Linear Systems of Differential Equations 457

y 0 (t) = y 0
Z t
y 1 (t) = y 0 + Ay 0 du
0
= y 0 + Ay 0 t
Z t
y 2 (t) = y 0 + Ay 1 (u) du
0
Z t
= y0 + A(y 0 + Ay 0 u) du
0
1
= y 0 + Ay 0 t + A2 y 0 t2
2
Z t  
1 2 2
y 3 (t) = A y 0 + Ay 0 u + A y 0 u du
0 2
1 1
= y 0 + Ay 0 t + A2 y 0 t2 + A3 y 0 t3
2 6
..
.
1 1
y n (t) = y 0 + Ay 0 t + A2 y 0 t2 + · · · + An y 0 tn .
2 n!
Notice that we may factor a y 0 out of each term on the right hand side of
y n (t). This gives the following expression for the function y n (t):
 
1 2 2 1 3 3 1 n n
y n (t) = In + At + A t + A t + · · · + A t y 0 (5)
2 3! n!

where In denotes the identity matrix of size n. If you recall the Taylor series
expansion for the exponential function eat :
1 1 1
eat = 1 + at + (at)2 + (at)3 + · · · + (at)n + · · ·
2 3! n!
you should immediately see a similarity. If we replace the scalar a with the
n× n matrix A and the scalar 1 with the identity matrix In then we can define
eAt to be the sum of the resulting series. That is,
1 1 1
eAt = In + At + (At)2 + (At)3 + · · · + (At)n + · · · . (6)
2 3! n!
It is not difficult (but we will not do it) to show that the series we have
written down for defining eAt in fact converges for any n × n matrix A, and
the resulting sum is an n × n matrix of functions of t. That is
 
h11 (t) h12 (t) · · · h1n (t)
 h21 (t) h22 (t) · · · h2n (t) 
eAt =  . .. .. ..  .
 
 .. . . . 
hn1 (t) hn2 (t) · · · hnn (t)
458 10 Systems of Differential Equations

It is not, however, obvious what the functions hij (t) are. Much of the remain-
der of this chapter will be concerned with precisely that problem. For now, we
simply want to observe that the functions y n (t) (see Equation (5)) computed
from the Picard approximation algorithm converge to eAt y 0 , that is the ma-
trix function eAt multiplied by the constant vector y 0 from the initial value
problem (4). Hence we have arrived at the following fact: The unique solution
to (4) is

y(t) = eAt y 0 . (7)

Following are a few examples where we can compute the matrix exponential
eAt with only the definition.
Example 7. Compute eAt for each of the following constant matrices A.
 
00
1. A = = 02 . (In general 0k denotes the k × k matrix, all of whose
00
entries are 0.)
 
00
I Solution. In this case An tn = for all n. Hence,
00
1 1
e02 t = eAt = I2 + At + A2 t2 + A3 t3 + · · ·
2 3!
= I2 + 02 + 02 + · · ·
= I2 .
Similarly, e0n t = In . This is the matrix analog of the fact e0 = 1. J
 
20
2. A = .
03
I Solution. In this case the powers of the matrix A are easy to compute.
In fact
     n 
40 8 0 2 0
A2 = , A3 = , · · · An = ,
09 0 27 0 3n
so that
1 1
eAt = I + At + A2 t2 + A3 t3 + · · ·
  2  3!
1 4t2 0

10 2t 0
= + + +
01 0 3t 2 0 9t2
1 8t3 0 1 2 n tn 0
   
+ + · · · + + ···
3! 0 27t3 n! 0 3 n tn
1 + 2t + 12 4t2 + · · · + n! 1 n n
 
2 t + ··· 0
=
0 1 + 3t + 12 9t2 + · · · + 1 n n
n! 3 t + ···
 2t 
e 0
= .
0 e3t
 10.2 Linear Systems of Differential Equations 459

J

a0
3. A = .
0b
I Solution. There is clearly nothing special about the numbers 2 and 3
on the diagonal of the matrix in the last example. The same calculation
shows that

#a 0%t  at 
At 0 b$ e 0
e =e" = bt . (8)
0 e

J
 
01
4. A = .
00
    
01 01 00
I Solution. In this case, check that A2 = = = 02 .
00 00 00
Then An = 02 for all n ≥ 2. Hence,
1 1
eAt = I + At + A2 t2 + A3 t3 + · · ·
2 3!
= I + At
 
1t
= .
01

Note that in this case, the individual entries of eAt do not look like expo-
nential functions eat at all. J
 
0 −1
5. A = .
1 0
I Solution. We leave it as an exercise to compute
 the powers
 of the
2 −1 0 3 0 1 4 10
matrix A. You should find A = ,A = ,A = = I2 ,
0 −1 −1 0 01
A5 = A, A6 = A2 , etc. That is, the powers repeat with period 4. Then
1 1
eAt = I + At + A2 t2 + A3 t3 + · · ·
  2  3!
1 −t2 0 1 0 t3 1 t4 0
    
10 0 −t
= + + + + + ···
01 t 0 2 0 −t2 3! −t3 0 4! 0 t4
1 − 12 t2 + 4!
1 4 1 3 1 5
 
t + · · · −t + 3! t − 5! t + ···
= 
1 3 1 5 1 2 1 4
t − 3! t + 5! t − · · · 1 − 2 t + 4! t + · · ·
 
cos t − sin t
= .
sin t cos t
460 10 Systems of Differential Equations

In this example also the individual entries of eAt are not themselves ex-
ponential functions. J
Example 8. Use Equation (7) and the calculation of eAt from the corre-
sponding item in the previous example to solve the initial value problem
 
c
y 0 = Ay, y(0) = y 0 = 1
c2
for each of the following matrices A.
 
00
1. A = = 02 .
00
I Solution. By Equation (7), the solution y(t) is given by
 
c
y(t) = e y 0 = I2 y 0 = y 0 = 1 .
At
(9)
c2
That is, the solution of the vector differential equation y 0 = 0, y(0) = y 0
is the constant function y(t) = y 0 . In terms of the component functions
y1 (t), y2 (t), the system of equations we are considering is
y10 = 0, y1 (0) = c1
y20 = 0, y2 (0) = c2
and this clearly has the solution y1 (t) = c1 , y2 (t) = c2 , which agrees with
(9). J
 
20
2. A = .
03
 2t 
At e 0
I Solution. Since in this case, e = , the solution of the initial
0 e3t
value problem is
 2t     2t 
e 0 c1 c e
At
y(t) = e y 0 = = 1 3t .
0 e3t c2 c2 e
Again, in terms of the component functions y1 (t), y2 (t), the system of
equations we are considering is
y10 = 2y1 , y1 (0) = c1
y20 = 3y2 , y2 (0) = c2 .
Since the first equation does not involve y2 and the second equation does
not involve y1 , what we really have is two independent first order linear
equations. The first equation clearly has the solution y1 (t) = c1 e2t and the
second clearly has the solution y2 (t) = c2 e3t , which agrees with the vector
description provided by Equation (7). (If the use of the word clearly is not
clear, then you are advised to review Section 2.2.) J
 10.2 Linear Systems of Differential Equations 461

a0
3. A = .
0b
eat 0
 
At
I Solution. Since in this case, e = , the solution of the initial
0 ebt
value problem is

eat 0
     at 
c1 c e
At
y(t) = e y 0 = = 1 bt .
0 ebt c2 c2 e

J
 
01
4. A = .
00
 
1t At
I Solution. In this case e = , so the solution of the initial value
01
problem is     
At 1 t c1 c1 + tc2
y(t) = e y 0 = = .
0 1 c2 c2
Again, for comparative purposes, we will write this equation as a system
of two equations in two unknowns:

y10 = y2 , y1 (0) = c1
y20 = 0, y2 (0) = c2 .

In this case also, it is easy to see directly what the solution of the system is
and to see that it agrees with that computed by Equation (7). Indeed, the
second equation says that y2 (t) = c2 , and then the first equation implies
that y1 (t) = c1 + tc2 by integration.
J
 
0 −1
5. A = .
1 0
I Solution. The solution of the initial value problem is
    
At cos t − sin t c1 c1 cos t − c2 sin t
y(t) = e y 0 = = .
sin t cos t c2 c1 sin t + c2 cos t
J

Exercises
Compute the derivative of each of the following matrix functions.

cos 2t sin 2t
1. A(t) =
 sin 2t cos 2t 
−
462 10 Systems of Differential Equations

e−3t t
2. A(t) = 2 2t
t e 

e−t te−t t2 e−t

3. A(t) =  0 e−t te−t 

0 0 e−t 

t
t2 
4. y(t) = 
ln t

12
5. A(t) =
3 4
6. v(t) = e−2t ln(t2 + 1) cos 3t!

For each of the following matrix functions, compute the requested integral.

π cos 2t sin 2t
7. Compute A(t) dt if A(t) = .
− sin 2t cos 2t
0

1 1 e2t + e−2t e2t − e−2t

8. Compute A(t) dt if A(t) =
0 2
e
−2t
− e2t e2t + e−2t 
2
9. Compute 1
y(t) dt for the matrix y(t) of Exercise 4.
5
10. Compute 1
A(t) dt for the matrix A(t) of Exercise 5.

t (t + 1)−1
11. On which of the following intervals is the matrix function A(t) =
(t − 1)
−2
t+6 
continuous?
(a) I1 = (−1, 1) (b) I2 = (0, ∞) (c) I3 = (−1, ∞)
(d) I4 = (−∞, −1) (e) I5 = (2, 6)

If A(t) = [aij (t)] is a matrix of functions, then the Laplace transform of A(t)
can be defined by taking the Laplace transform of each function aij (t). That is,

L(A(t))(s) = [L(aij (t))(s)] .

e2t sin 2t
For example, if A(t) = 2t , then
e cos 3t t 
1 2
s−2 s2 +4
L(A(t))(s) = & s−2 1 .
(s−2)2 +9 s2 '

Compute the Laplace transform of each of the following matrix functions.

1 t
12. A(t) = 2 2t
t e 
 10.2 Linear Systems of Differential Equations 463

cos t sin t
13. A(t) =
 sin t cos t
−

t3 t sin t te−t
14. A(t) = 2 3t
t − t e cos 2t 3 
t
t2 
15. A(t) = 
t3 

1 −1 −1 1
16. A(t) = et + e−t
−1 1   1 −1
1 sin t 1 − cos t
17. A(t) = 
0 cos t sin t 
0 − sin t cos t 

The inverse Laplace transform of a matrix function is also defined by taking the
inverse Laplace transform of each entry of the matrix. For example,
1 1 1 t
)
( s s2  ,  2 4 .
* -= 
+
−1
L
1 1 t t 
s3 s 4 2 6
Compute the inverse Laplace transform of each matrix function:
1 2 6
18. .
s s2 s3 /
1 1

 s s 2 
19. 
s s  
s2 − 1 s2 + 1
1 1

 s − 1 s 2 − 2s + 1

 
 4 1 

20.  3 
s + 2s − 3s
2 2
s +1 
 
3s 1
s2 + 9 s−3
2s 2
s2 − 1 s2 − 1 

21.  
2 2s 
s2 − 1 s2 − 1
For each matrix A given below:
(i) Compute (sI − A)−1 .
(ii) Compute the inverse Laplace transform of (sI − A)−1 .

10
22. A =
0 2
464 10 Systems of Differential Equations

1 −1
23. A =
−2 2 

01 1
24. A = 
0 0 1
00 0

0 1
25. A =
−1 0
0t
26. Let A(t) = and consider the initial value problem
t 0
1
y0 = Ay, y(0) = .
1
a) Use Picard’s method to calculate the first four terms, y0 , · · · , y3 .
b) Make a conjecture about what the n-th term will be. Do you recognize the
series?
2
et /2
c) Verify that y(t) = & t2 /2 is a solution. Are there any other solutions
e '
possible? Why or Why not?

0 t
27. Let A(t) = and consider the initial value problem
 0
−t

1
y0 = Ay, y(0) = .
0
a) Use Picard’s method to calculate the first four terms, y0 , · · · , y3 .
cos t2 /2
b) Verify that y(t) = 2 is a solution. Are there any other solutions
 sin t /2
−
possible? Why or Why not?

t t
28. Let A(t) = and consider the initial value problem
 −t
−t

1
y0 = Ay, y(0) = .
1
a) Use Picard’s method to calculate the first four terms, y0 , · · · , y3 .
b) Deduce the solution.

29. Verify the product rule for matrix functions. That is, if A(t) and B(t) are matrix
functions which can be multiplied and C(t) = A(t)B(t) is the product, then

C 0 (t) = A0 (t)B(t) + A(t)B 0 (t).

r
Hint: Write the ij term of C(t) as cij (t) = k=1 aik (t)bkj (t) (where r is the
number of columns of A(t) = the number of rows of B(t)), and use the ordinary
product and sum rules for derivatives.
 10.2 Linear Systems of Differential Equations 465

What is the largest interval containing 0 on which the initial value problem

2
y 0 = A(t)y, y(0) =
 
−1

is guaranteed by Theorem 4.2.2 to have a solution, assuming:

0 1
30. A(t) = 2
(t + 2) −1
cos t

(t + 4)−2 t2 + 4
31. A(t) =
ln(t − 3) (t + 2) 
−4

t+2
2 −5t+6 t
32. A(t) = &t 2
t t3 '

1 −1
33. A(t) =
2 5 
010
34. Let N = 0 0 1.
0 0 0
a) Show that
00 1 00 0
N2 = 
0 0 0 and N3 = 
0 0 0 .
00 0 00 0
b) Using the above calculations, compute eNt .
c) Solve the initial value problem

1
y 0 = N y, 2 .
y(0) = 
3

d) Compute the Laplace transform of eNt , which you calculated in Part (b).
e) Compute the matrix (sI − N )−1 . Do you see a similarity to the matrix
computed in the previous part?

11
35. Let A = .
0 1
1n
a) Verify that An = for all natural numbers n.
0 1 
b) Using part (a), verify, directly from the definition, that

et tet
eAt = t .
0 e 
c) Now solve the initial value problem y 0 = Ay, y(0) = y 0 for each of the
following initial conditions y 0 .
1 0 −2 c1
(i) y 0 = , (ii) y 0 = , (iii) y 0 = , (iv) y 0 =
0  1   5  c2 
466 10 Systems of Differential Equations

d) Compute the Laplace transform of eAt .

e) Compute (sI − A)−1 and compare to the matrix computed in Part (d).

36. One of the fundamental properties of the exponential function is the formula
ea+b = ea eb . The goal of this exercise is to show, by means of a concrete ex-
ample, that the analog of this fundamental formula is not true for the matrix
exponential function (at least without some additional assumptions). From the
20 01
calculations of Example 4.2.7, you know that if A = and B = , then
0 3  0 0
e2t 0 1 t
eAt = 3t and e
Bt
= .
0 e  0 1
e2t te3t
a) Show that eAt eBt = 3t .
0 e 
0
b) Let y 0 = and let y(t) = eAt eBt y 0 . Compute y 0 (t) and (A + B)y(t).
1 
Are these two functions the same?
c) What do these calculations tell you about the possible equality of the matrix
functions eAt eBt and e(A+B)t for these particular A and B?
d) We will see later that the formula e(A+B)t = eAt eBt is valid provided that
AB = BA. Check that AB 6= BA for the matrices of this exercise.

10.3 Linear Homogeneous Equations

This section will be concerned with using the fundamental existence and
uniqueness theorem for linear systems (Theorem 2) to describe the solution
set for a linear homogeneous system of ordinary differential equations

y 0 = A(t)y. (1)

The main result will be similar to the description given by Theorem ?? for
linear homogeneous second order equations.
Recall that if A(t) is a continuous n × n matrix function on an interval
I, then a solution to system (1) is an n × 1 matrix function y(t) such that
y 0 (t) = A(t)y(t) for all t ∈ I. Since this is equivalent to the statement

y 0 (t) − A(t)y(t) = 0 for all t ∈ I,

to be consistent with the language of solution sets used in Chapter 6, we will

0
denote the set of all solutions of (1) by SL , where L = D −A(t) is the (vector)
differential operator which acts on the vector function y(t) by the rule

L(y(t)) = (D − A(t))(y(t)) = y 0 (t) − A(t)y(t).

Thus
 10.3 Linear Homogeneous Equations 467
0
SL = {y(t) : L(y(t)) = 0} .
Let y 1 (t) and y 2 (t) be two solutions of system (1), i.e., y 1 (t) and y 2 (t)
0
are in SL and let y(t) = c1 y 1 (t) + c2 y 2 (t) where c1 and c2 are scalars (either
real or complex). Then

y 0 (t) = (c1 y 1 (t) + c2 y 2 (t))0

= c1 y 01 (t) + c2 y 02 (t)
= c1 A(t)y 1 (t) + c2 A(t)y 2 (t)
= A(t) (c1 y 1 (t) + c2 y 2 (t))
= A(t)y(t).

Thus every linear combination of two solutions of (1) is again a solution. which
0
in the language of linear algebra means that SL is a vector space. We say
that a set of vectors
B = {v 1 , . . . , v k }
in a vector space V is a basis of V if the set B is linearly independent and
if every vector v in V can be written as a linear combination

v = λ1 v 1 + · · · + λk v k .

The number k of vectors in a basis B of V is known as the dimension of

V . Thus R2 has dimension 2 since it has a basis e1 = (1, 0), e2 = (0, 1)
consisting of 2 vectors, R3 has dimension 3 since it has a basis e1 = (1, 0, 0),
e2 = (0, 1, 0), e3 = (0, 0, 1) consisting of 3 vectors, etc. The main theorem on
solutions of linear homogeneous systems can be expressed most conveniently
in the language of vector spaces.

Theorem 1. If the n × n matrix A(t) is continuous on an interval I, then the

0
solution set SL of the homogeneous system

y 0 = A(t)y (2)

is a vector space of dimension n. In other words,

0
1. There are n linearly independent solutions of (2) in SL .
0
2. If ϕ1 , ϕ2 , . . ., ϕn ∈ SL are independent solutions of (2), and ϕ is any
0
function in SL , then ϕ can be written as

ϕ = c1 ϕ1 + · · · + cn ϕn

for some scalars c1 , . . ., cn ∈ R.

Proof. To keep the notation as explicit as possible, we will only present the
proof in the case n = 2. You should compare this proof with that of Theorem
??. To start with, let
468 10 Systems of Differential Equations
   
1 0
e1 = and e2 = ,
0 1
and let t0 ∈ I. By Theorem 2 there are vector functions ψ 1 (t) and ψ 2 (t)
defined for all t ∈ I and which satisfy the initial conditions

ψ i (t0 ) = ei for i = 1, 2. (3)

Suppose there is a dependence relation c1 ψ 1 + c2 ψ 2 = 0. This means that

c1 ψ 1 (t) + c2 ψ 2 (t) = 0

for all t ∈ I. Applying this equation to the particular point t0 gives

     
1 0 c
0 = c1 ψ 1 (t0 ) + c2 ψ 2 (t0 ) = c1 e1 + c2 e2 = c1 + c2 = 1 .
0 1 c2

Thus c1 = 0 and c2 = 0 so that ψ 1 and ψ 2 are linearly independent. This

proves (1).
0
Now suppose that ϕ ∈ SL . Evaluating at t0 gives
 
r
ϕ(t0 ) = .
s
0 0 0
Now define ψ ∈ SL by ψ = rψ 1 + sψ 2 . Note that ψ ∈ SL since SL is a vector
space. Moreover,
 
r
ψ(t0 ) = rψ 1 (t0 ) + sψ 2 (t0 ) = re1 + se2 = = ϕ(t0 ).
s
0
This means that ϕ and ψ = rψ 1 + sψ 2 are two elements of SL which have
the same value at t0 . By the uniqueness part of Theorem 2, they are equal.
Now suppose that ϕ1 and ϕ2 are any two linearly independent solutions
0
of (2) in SL . From the argument of the previous paragraph, there are scalars
a, b, c, d so that

ϕ1 = aψ 1 + bψ2
ϕ2 = cψ 1 + dψ 2

which in matrix form can be written

 
    ac
ϕ1 ϕ2 = ψ 1 ψ 2 .
bd

We
 multiply
 both sides of this matrix equation on the right by the adjoint
d −c
to obtain
−b a
   
  d −c   ad − bc 0  
ϕ1 ϕ2 = ψ1 ψ2 = ψ 1 ψ 2 (ad − bc).
−b a 0 ad − bc
 10.3 Linear Homogeneous Equations 469

Suppose ad − bc = 0. Then

dϕ1 − bϕ2 = 0
and − cϕ1 + aϕ2 = 0.

But since ϕ1 and ϕ2 are independent this implies that a, b, c, and d are zero
which in turn implies that ϕ1 and ϕ2 are both zero. But this cannot be. We
conclude that ad − bc 6= 0. We can now write ψ 1 and ψ 2 each as a linear
combination of ϕ1 and ϕ2 . Specifically,
 
  1   d −c
ψ 1 ψ2 = ϕ1 ϕ2 .
ad − bc −b a

Since ϕ is a linear combination of ψ 1 and ψ 2 it follows that ϕ is a linear

combination of ϕ1 and ϕ2 . u
t
 
ac
The matrix that appears in the above proof is a useful theoreti-
bd
0
cal criterion for determining if a pair of solutions ϕ1 , ϕ2 in SL is linearly
indpependent and hence a basis in the case n = 2. The proof shows:
 
ac
1. ϕ1 (t0 ) ϕ2 (t0 ) =
 
bd
 
0 ac
2. If the solutions ϕ1 (t), ϕ2 (t) are a basis of SL , then ad−bc = det 6= 0,
bd
and moreover, this is true for any t0 ∈ I.
3. The converse of the above statement is also true (and easy). Namely, if
 
det ϕ1 (t0 ) ϕ2 (t0 ) 6= 0
0
for some t0 ∈ I, then ϕ1 , ϕ2 in SL is a basis of the solution space (always
assuming n = 2).
0
Now assume that ϕ1 and ϕ2 are any two solutions in SL . Then we can
form a 2 × 2 matrix of functions
 
Φ(t) = ϕ1 (t) ϕ2 (t)

where each column is a solution to y 0 = A(t)y. We will say that Φ(t) is a

fundamental matrix for y 0 = A(t)y if the columns are linearly independent,
0
and hence form a basis of SL . Then the above discussion is summarized in
the following result.

Theorem  2. If A(t) is a 2 × 2 matrix of continuous functions on I, and if

Φ(t) = ϕ1 (t) ϕ2 (t) where each column is a solution to y 0 = A(t)y, then
Φ(t) is a fundamental matrix for y 0 = A(t)y if and only if det Φ(t) 6= 0 for
at least one t ∈ I. If this is true for one t ∈ I, it is in fact true for all t ∈ I.
470 10 Systems of Differential Equations

Remark 3. The above theorem is also true, although we will not prove it,
for n × n matrix systems y 0 = A(t)y, where a solution matrix consists of an
n × n matrix  
Φ(t) = ϕ1 (t) · · · ϕn (t)
where each column ϕi (t) is a solution to y 0 = A(t)y. Then Φ(t) is a funda-
0
mental matrix, that is the columns are a basis for SL if and only if det Φ(t) 6= 0
for at least one t ∈ I.

Note that if a matrix B is written in columns, say

 
B = b1 · · · bn ,

then the matrix multiplication AB, if it is defined (which means the number
of columns of A is the number of rows of B), can be written as
   
AB = A b1 · · · bn = Ab1 · · · Abn .

In other words, multiply A by each column of B separately. For example, if

   
1 −1 10 2
A= and B = ,
2 3 0 1 −1

then
  
1 −1 1 0 2
AB =
2 3 0 1 −1
         
1 −1 1 1 −1 0 1 −1 2
=
2 3 0 2 3 1 2 3 −1
 
1 −1 3
= .
2 3 1

Now suppose that Φ(t) = ϕ1 (t) · · · ϕn (t) is a fundamental matrix of

 

solutions for y 0 = A(t)y. Then

Φ0 (t) = ϕ01 (t) · · · ϕ0n (t)

 
 
= A(t)ϕ1 (t) · · · A(t)ϕn (t)
 
= A(t) ϕ1 (t) · · · ϕn (t)
= A(t)Φ(t).

Thus the n × n matrix Φ(t) satisfies the same differential equation, namely
y 0 = A(t)y, as each of its columns. We summarize this discussion in the
following theorem.
Theorem 4. If A(t) is a continuous n × n matrix of functions on an interval
I, then an n × n matrix of functions Φ(t) is a fundamental matrix for the
homogeneous linear equation y 0 = A(t)y if and only if
 10.3 Linear Homogeneous Equations 471

Φ0 (t) = A(t)Φ(t)
(4)
and det Φ(t) 6= 0.

The second condition need only be checked for one value of t ∈ I.

Example 5. Show that

e2t e−t
 
Φ(t) =
2e2t −e−t
 
01
is a fundamental matrix for the system y 0 = y.
21
I Solution. First check that Φ(t) is a solution matrix, i.e., check that the
first condition of Equation (4) is satisfied. To see this, we calculate
 2t −t 0  2t
2e −e−t

0 e e
Φ (t) = =
2e2t −e−t 4e2t e−t

and    2t −t   2t
2e −e−t
  
01 01 e e
Φ(t) = = .
21 2 1 2e2t −e−t 4e2t e−t
Since these two matrices of functions are the same, Φ(t) is a solution matrix.
To check that it is a fundamental matrix, pick t = 0 for example. Then
 
1 1
Φ(0) =
2 −1

and this matrix has determinant −3, so Φ(t) is a fundamental matrix. J

Example 6. Show that
te2t (t + 1)e2t
 
Φ(t) =
e2t e2t
 
0 21
is a fundamental matrix for the system y = y.
02

I Solution. Again we check that the two conditions of Equation (4) are
satisfied. First we calculate Φ0 (t):
 2t 0 
te (t + 1)e2t (2t + 1)e2t (2t + 3)e2t

Φ0 (t) = 2t = .
e e2t 2e2t 2e2t
 
21
Next we calculate A(t)Φ(t) where A(t) = :
02
   2t
2 1 te (t + 1)e2t (2t + 1)e2t (2t + 3)e2t
  
= .
0 2 e2t e2t 2e2t 2e2t
472 10 Systems of Differential Equations

Since these two matrices of functions are the same, Φ(t) is a solution matrix.
Next check the second condition of (4) at t = 0:
 
01
det Φ(0) = det = −1 6= 0.
11
 
21
Hence Φ(t) is a fundamental matrix for y 0 = y. J
02

Example 7. Show that

t2 t3
 
Φ(t) =
2t 3t2
is a fundamental matrix for the system y 0 = A(t)y where

0 1
" #
A(t) = 6 4 .
− 2
t t
I Solution. Note that
 " 0 1# 2 3 
2t 3t2

0 t t
Φ (t) = = 6 4 ,
2 6t − 2 2t 3t2
t t
while  
11
det Φ(1) = = 1 6= 0.
23
 
00
Hence Φ(t) is a fundamental matrix. Note that Φ(0) = which has deter-
00
minant 0. Why does this not prevent Φ(t) from being a fundamental matrix?
J

In Exercise 29, Page 464 you were asked to verify the product rule for
differentiating a product of matrix functions. The rule is

(B(t)C(t))0 = B 0 (t)C(t) + B(t)C 0 (t).

Since matrix multiplication is not commutative, it is necessary to be careful of

the order. If one of the matrices is constant, then the product rule is simpler:

(B(t)C)0 = B 0 (t)C

since C 0 = 0 for a constant matrix. We apply this observation in the following

way. Suppose that Φ1 (t) is a fundamental matrix of a homogeneous system
y 0 = A(t)y where A(t) is an n × n matrix of continuous functions on an
interval I. According to Theorem 4 this means that

Φ01 (t) = A(t)Φ1 (t) and det Φ1 (t) 6= 0.

 10.3 Linear Homogeneous Equations 473

Now define a new n × n matrix of functions Φ2 (t) := Φ1 (t)C where C is an

n × n constant matrix. Then

Φ02 (t) = (Φ1 (t)C)0 = Φ01 (t)C = A(t)Φ1 (t)C = A(t)Φ2 (t),

so that Φ2 (t) is a solution matrix for the homogeneous system y 0 = A(t)y.

To determine if Φ2 (t) is also a fundamental matrix, it is only necessary to
compute the determinant:

det Φ2 (t) = det(Φ1 (t)C) = det Φ1 (t) det C.

Since det Φ1 (t) 6= 0, it follows that det Φ2 (t) 6= 0 if and only if det C 6= 0, i.e.,
if and only if C is a nonsingular n × n matrix.
Example 8. In Example 5 it was shown that
 2t −t 
e e
Φ(t) =
2e2t −e−t
   
0 01 1 1 1
is a fundamental matrix for the system y = y. Let C = 3 . Then
21 2 −1
det C = −1/3 6= 0 so C is invertible, and hence
1 e2t e−t 1 e2t + 2e−t e2t − e−t
    
1 1
Ψ (t) = Φ(t)C = =
3 2e2t −e−t 2 −1 3 2e2t − 2e−t 2e2t + e−t
 
0 01
is also a fundamental matrix for y = y. Note that Ψ (t) has the partic-
21
ularly nice feature that its value at t = 0 is
 
10
Ψ (0) = = I2
01
the 2 × 2 identity matrix. u
t
Example 9. In Example 6 it was shown that
 2t
te (t + 1)e2t

Φ(t) = 2t
e e2t
   
0 21 −1 1
is a fundamental matrix for the system y = y. Let C = . Then
02 1 0
det C = −1 6= 0 so C is invertible, and hence
 2t
te (t + 1)e2t −1 1
   2t 2t 
e te
Ψ (t) = Φ(t)C = 2t =
e e2t 1 0 0 e2t
 
0 21
is also a fundamental matrix for y = y. As in the previous example
02
Ψ (0) = I2 is the identity matrix. u
t
474 10 Systems of Differential Equations

If Φ(t) is a fundamental matrix for the linear system y 0 = A(t)y on the

interval I and t0 ∈ I, then Φ(t0 ) is an invertible matrix by Theorem 4 so if
we take C = (Φ(t0 ))−1 , then

Ψ (t) = Φ(t)C = Φ(t)(Φ(t0 ))−1

is a fundamental matrix which satisfies the extra condition

Ψ (t0 ) = Φ(t0 )(Φ(t0 ))−1 = In .

Hence, we can always arrange for our fundamental matrices to be the identity
at the initial point t0 . Moreover, the uniqueness part of the existence and
uniqueness theorem insures that there is only one solution matrix satisfying
this extra condition. We record this observation in the following result.
Theorem 10. If A(t) is a continuous n×n matrix of functions on an interval
I and t0 ∈ I, then there is an n × n matrix of functions Ψ (t) such that
1. Ψ (t) is a fundamental matrix for the homogeneous linear equation y 0 =
A(t)y and
2. Ψ (t0 ) = In ,
3. Moreover, Ψ (t) is uniquely determined by these two properties.
4. If y 0 is a constant vector, then y(t) = Ψ (t)y 0 is the unique solution of
the homogeneous initial value problem y 0 = A(t)y, y(t0 ) = y 0 .

Proof. Only the last statement was not discussed in the preceding paragraphs.
Suppose that Ψ (t) satisfies conditions (1) and (2) and let y(t) = Ψ (t)y 0 . Then
y(t0 ) = Ψ (t0 )y 0 = In y 0 = y 0 . Moreover,

y 0 (t) = Ψ 0 (t)y 0 = A(t)Ψ (t)y 0 = A(t)y(t),

so y(t) is a solution of the initial value problem, as required. u

t
   
01 3
Example 11. Solve the initial value problem y 0 = y, y(0) = .
21 −6
 
01
I Solution. In Example 8 we found a fundamental matrix for y 0 = y
21
satisfying (1) and (2) of the above theorem, namely

1 e2t + 2e−t e2t − e−t

 
Ψ (t) = .
3 2e2t − 2e−t 2e2t + e−t

Hence the unique solution of the initial value problem is

1 e2t + 2e−t e2t − e−t

    2t
−e + 4e−t
   
3 3
y(t) = Ψ (t) = = .
−6 3 2e2t − 2e−t 2e2t + e−t −6 −2e2t − 4e−t
J
 10.3 Linear Homogeneous Equations 475
   
0 21 −2
Example 12. Solve the initial value problem y = y, y(0) = .
02 3
 
0 21
I Solution. In Example 9 we found a fundamental matrix for y = y
02
satisfying (1) and (2) of the above theorem, namely
 2t 2t 
e te
Ψ (t) = .
0 e2t
Hence the solution of the initial value problem is
   2t 2t    
(3t − 2)e2t

−2 e te −2
y(t) = Ψ (t) = = .
3 0 e2t 3 3e2t
J
We conclude this section by observing that for a constant matrix function
A(t) = A, at least in principle, it is easy to describe the fundamental matrix
Ψ (t) from Theorem 10. It is in fact the matrix function we have already
encountered in the last section, i.e., the matrix exponential eAt . Recall that
eAt (for a constant matrix A) is defined by substituting A for a in the Taylor
series expansion of eat :
1 1 1
(∗) eAt = In + At + A2 t2 + A3 t3 + · · · + An tn + · · ·
2 3! n!
We have already observed (but not proved) that the series on the right hand
side of (∗) converges to a well defined matrix function for all matrices A. Let
Ψ (t) = eAt . If we set t = 0 in the series we obtain Ψ (0) = eA0 = In and if
we differentiate the series terms by term (which can be shown to be a valid
operation), we get
d At
Ψ 0 (t) = e
dt
1 1
= 0 + A + A2 t + A3 t2 + · · · + An tn−1 + · · ·
2 (n − 1)!
 
1 2 2 1 n−1 n−1
= A In + At + A t + · · · + A t + ···
2 (n − 1)!
= AeAt
= AΨ (t).

Thus we have shown that Ψ (t) = eAt satisfies the first two properties of
Theorem 10, and hence we have arrived at the important result:
Theorem 13. Suppose A is an n × n constant matrix.
1. A fundamental matrix for the linear homogeneous problem y 0 = Ay is
Ψ (t) = eAt .
476 10 Systems of Differential Equations

2. If y 0 is a constant vector, then the unique solution of the initial value

problem y 0 = Ay, y(0) = y 0 is

y(t) = eAt y 0 . (5)

3. If Φ(t) is any fundamental matrix for the problem y 0 = Ay, then

−1
eAt = Φ(t) (Φ(0)) . (6)

Example
 14.
 1. From the calculations in Example 8 we conclude that if
01
A= , then
21

1 e2t + 2e−t e2t − e−t

 
At
e = .
3 2e2t − 2e−t 2e2t + e−t
 
21
2. From the calculations in Example 9 we conclude that if A = then
02

e2t te2t
 
eAt = .
0 e2t

What we have seen in this section is that if we can solve y 0 = Ay (where

A is a constant matrix), then we can find eAt by Equation (6), and conversely,
if we can find eAt by some method, then we can find all solutions of y 0 = Ay
by means of Equation (5). Over the next few sections we will learn a couple
of different methods for calculating eAt .

Exercises
1. For each of the following pairs of matrix functions Φ(t) and A(t), determine if
Φ(t) is a fundamental matrix for the system y 0 = A(t)y. It may be useful to
review Examples 4.3.5 – 4.3.7.
 10.3 Linear Homogeneous Equations 477

Φ(t) A(t)

cos t sin t 0 1
(a)
 sin t cos t
−  0
−1
cos t sin t 0 1
(b)
− sin(t + π/2) cos(t + π/2)  0
−1
e−t e2t −2 1
(c) 2t
e 4e  −4 3
−t

e − e2t e2t
−t
−2 1
(d) 2t 2t
e − 4e 4e  −4 3
−t

t 2t
e e 12
(e) 3t 4t
 e 
e 3 4
e2t 3e3t 5 −3
(f) 2t 3t
e 2e   0 
2
3e2t e6t 33
(g) 2t 6t
−e e  1 5
−2e3t (1 − 2t)e3t 1 −4
(h) 3t
 e te3t   5 
1
2 2
sin(t /2) cos(t /2) 0 t
(i) 2 2
cos(t /2) − sin(t /2) −t 0
1 + t2 3 + t2 t t
(j) 2 2
1 − t −1 − t  −t −t
2 2
et /2 e−t /2 0 t
(k) & t2 /2 2
e −e−t /2 ' t 0
2. For each of the matrices A in parts (a), (c), (f), (g), (h) of Exercise 1:
a) Find a fundamental matrix Ψ (t) for the system y 0 = Ay satisfying the
condition Ψ (0) = I2 . (See Examples 4.3.8 and 4.3.9.)
3
b) Solve the initial value problem y 0 = Ay, y(0) = .
−2
At
c) Find e .

3. For each of the matrices A(t) in parts (i), (j) and (k) of Exercise 1:
a) Find a fundamental matrix Ψ (t) for the system y 0 = A(t)y satisfying the
condition Ψ (0) = I2 .
3
b) Solve the initial value problem y 0 = A(t)y, y(0) = .
 
−2
A(t)t
c) Is e = Ψ (t)? Explain.

4. In each problem below determine whether the given functions are linearly inde-
pendent.
1 t
a) y1 = y2 = .
2   
−t
1 t
b) y1 = y2 = .
1
 t t −t
te e
c) y1 = t y2 = −t .
e  te 
478 10 Systems of Differential Equations

1 0 0
t 
d) y1 =  1
y2 =  0 .
y3 = 
t2  t 1

10.4 Constant Coefficient Homogeneous Systems

In previous sections we studied some of the basic properties of the homoge-
neous linear system of differential equations

y 0 = A(t)y. (1)

In the case of a constant coefficient system, i.e., A(t) = A = a constant ma-

trix, this analysis culminated in Theorem 13 which states that a fundamental
matrix for y 0 = Ay is the matrix exponential function eAt and the unique
solution of the initial value problem y 0 = Ay, y(0) = y 0 is

y(t) = eAt y 0 .

That is, the solution of the initial value problem is obtained by multiplying
the fundamental matrix eAt by the initial value vector y 0 . The problem of
how to compute eAt for a particular constant matrix A was not addressed,
except for a few special cases where eAt could be computed directly from the
series definition of eAt . In this section we will show how to use the Laplace
transform to solve the constant coefficient homogeneous system y 0 = Ay and
in the process we will arrive at a Laplace transform formula for eAt .
As we have done previously, we will do our calculations in detail for the case

ab
of a constant coefficient linear system where the coefficient matrix A =
cd
is a 2 × 2 constant matrix so that Equation (1) becomes

y10 = ay1 + by2

(2)
y20 = cy1 + dy2 .

The calculations are easily extended to systems with more than 2 unknown
functions. According to theexistence  and uniqueness theorem (Theorem 2)
y1 (t)
there is a solution y(t) = for system (2), and we assume that the
y2 (t)
functions y1 (t) and y2 (t) have Laplace transforms. From Chapter 3, we know
that this is a relatively mild restriction on these functions, since, in particular,
all functions of exponential growth have Laplace transforms. Our strategy
will be to use the Laplace transform of the system (2) to determine what the
solution must be.
Let Y1 (s) = L(y1 ) and Y2 (s) = L(y2 ). Applying the Laplace transform to
each equation in system (2) and using the formulas from Table C.2 gives a
system of algebraic equations
 10.4 Constant Coefficient Homogeneous Systems 479
sY1 (s) − y1 (0) = aY1 (s) + bY2 (s)
(3)
sY2 (s) − y2 (0) = cY1 (s) + dY2 (s).
 
Y
Letting Y = 1 , the system (3) can be written compactly in matrix form
Y2
as
sY (s) − y(0) = AY (s)
which is then easily rewritten as the matrix equation

(sI − A) Y (s) = y(0). (4)

If the matrix sI − A is invertible, then we may solve Equation (4) for Y (s),
and then apply the inverse Laplace transform to the entries of Y (s) to find
the unknown functions y(t). But
 
s − a −b
sI − A = (5)
−c s − d

so p(s) = det(sI − A) = (s − a)(s − b) − bc = s2 − (a + d)s + (ad − bc) =

s2 − Tr(A)s + det(A). Hence p(s) is a nonzero polynomial function of degree
2, so that the matrix sI − A is invertible as a matrix of rational functions,
although one should note that for certain (the ≤ 2 roots of p(s)) values of
s the numerical matrix will not be invertible. For the purposes of Laplace
transforms, we are only interested in the inverse of sI − A as a matrix of
rational functions. Hence we may solve Equation (4) for Y (s) to get
−1
Y (s) = (sI − A) y(0). (6)

Now  
−1 1 s−d b
(sI − A) = (7)
p(s) c s−a
   
1 s−d 1 b
so let Z 1 (s) = and Z 2 (s) = be the first and second
p(s) c p(s) s − a
−1
columns of (sI − A) := Z(s), respectively. Since each entry of Z 1 (s) and
Z 2 (s) is a rational function of s with denominator the quadratic polynomial
p(s), the analysis of inverse Laplace transforms of rational functions of s with
quadratic denominator from Section 6.4 applies to show that each entry of
     
−1 s−d −1 b
L L
 p(s)   p(s) 
−1 −1
   
z 1 (t) = L Z 1 (s) =    and z 2 (t) = L Z 2 (s) = 
   
  
 −1 c s−a 
L−1
 
L
p(s) p(s)
will be of the form
1. c1 er1 t + c2 er2 t if p(s) has distinct real roots r1 6= r2 ;
480 10 Systems of Differential Equations

2. c1 ert + c2 tert if p(s) has a double root r; or

3. c1 eαt cos βt + c2 eαt sin βt if p(s) has complex roots α ± iβ,
where c1 and c2 are appropriate constants. Equation (6) shows that

Y (s) = Z(s)y(0)
= y1 (0)Z 1 (s) + y2 (0)Z 2 (s), (8)

and by applying the inverse Laplace transform we conclude that the solution
to Equation (2) is
y(t) = y1 (0)z 1 (t) + y2 (0)z 2 (t). (9)
If we let  
z(t) = z 1 (t) z 2 (t) = L−1 (sI − A)−1 , (10)
 

then Equation (9) for the solution y(t) of system (2) has a particularly nice
and useful matrix formulation:

y(t) = z(t)y(0). (11)

Before analyzing Equation (11) further to extract theoretical conclusions,

we will first see what the solutions look like in a few numerical examples.

Example 1. Find all solutions of the constant coefficient homogeneous lin-

ear system:

y10 = y2
(12)
y20 = 4y1 .
 
01
I Solution. In this system the coefficient matrix is A = . Thus sI −
  40
s −1
A= so that p(s) = det(sI − A) = s2 − 4 and
−4 s

s 1
 
 s2 − 4 s2 − 4 
−1
(sI − A) =  . (13)
 
 4 s 
s2 − 4 s2 − 4
Since
   
1 1 1 1 s 1 1 1
= − and = + ,
s2 − 4 4 s−2 s+2 s2 − 4 2 s−2 s+2

we conclude from our Laplace transform formulas (Table C.2) that the matrix
z(t) of Equation (10) is
 10.4 Constant Coefficient Homogeneous Systems 481
1
1 2t

e2t + e−2t e − e−2t
2 4 
z(t) =  . (14)
1 2t
 
2t −2t
e + e−2t


e −e
2
Hence, the solution of the system (12) is
1
1 2t

e2t + e−2t e − e−2t  
2 4  c1
y(t) =  
 c2
1

e2t − e−2t e2t + e−2t


2 (15)
1 1  1 1 
c1 + c2 c1 − c2
2 4  2t  2 4  −2t
=  e +  e ,
1 1
   
c1 + c2 −c1 + c2
2 2
   
y (0) c
where y(0) = 1 = 1 .
y2 (0) c2
Let’s check that we have, indeed, found a solution to the system of differ-
ential equations (2). From Equation (15) we see that
   
1 1 2t 1 1
y1 (t) = c1 + c2 e + c1 − c2 e−2t ,
2 4 2 4
and    
1 1
y2 (t) = c1 + c2 e2t + −c1 − c2 e−2t .
2 2
Thus y10 (t) = y2 (t) and y20 (t) = 4y1 (t), which is what it means to be a solution
of system (12).
The solution to system (12) with initial conditions y1 (0) = 1, y2 (0) = 0 is
1 1
   

y 1 (t) =  2  e2t +  2  e−2t

   

1 −1

while the solution with initial conditions y1 (0) = 0, y2 (0) = 1 is

1   1 
−
 4  2t  4  −2t
y 2 (t) =   e + 
   e .
1 1

2 2
The solution with initial conditions y1 (0) = c1 , y2 (0) = c2 can then be written

y(t) = c1 y 1 (t) + c2 y 2 (t),

482 10 Systems of Differential Equations

that is, every solution y of system (12) is a linear combination ofthe two
1 2t
particular solution y 1 and y 2 . Note, in particular, that y 3 (t) = e is a
  2
1
solution (with c1 = 1, c2 = 2), while y 4 = e−2t is also a solution (with
−2
c1 = 1, c2 = −2). The solutions y 3 (t) and y 4 (t) are notably simple solutions
in that each of these solutions is of the form

y(t) = veat (16)

where v ∈ R2 is a constant vector and a is a scalar. Note that

           
1 2 1 1 −2 1
A = =2 and A = = −2 .
2 4 2 −2 4 −2

That is, the vectors v and scalars a such that y(t) = veat is a solution to
y 0 = Ay are related by the algebraic equation

Av = av. (17)

A vector-scalar pair (v, a) which satisfies Equation 17 is known as a eigenvector-

eigenvalue pair for the matrix A. Finally, compare these two solutions y 3 (t)
and y 4 (t) of the matrix differential equation y 0 = Ay with the solution of
the scalar differential equation y 0 = ay, which we recall (see Section 2.2) is
y(t) = veat where v = y(0) ∈ R is a scalar. In both cases one gets either a
scalar or a vector multiplied by a pure exponential function eat . J
Example 2. Find all solutions of the linear homogeneous system

y10 = y1 + y2
(18)
y20 = −4y1 − 3y2 .
 
1 1
I Solution. For this system, the coefficient matrix is A = . We
−4 −3
will solve this equation by using Equation (11). Form the matrix
 
s − 1 −1
sI − A = .
4 s+3

Then p(s) = det(sI − A) = (s − 1)(s + 3) + 4 = (s + 1)2 , so that

 
1 s+3 1
(sI − A)−1 =
(s + 1)2 −4 s − 1
 1 2 1 
+ (19)
 s + 1 (s + 1)2 (s + 1)2 
= .
 
 −4 1 2 
2
− 2
(s + 1) s + 1 (s + 1)
 10.4 Constant Coefficient Homogeneous Systems 483

Thus the matrix z(t) from Equation (10) is, using the inverse Laplace formulas
from Table C.2
 −t
e + 2te−t te−t

−1 −1


z(t) = L (sI − A) = .
−4te−t e−t − 2te−t

The general solution to system (18) is therefore

 −t
e + 2te−t te−1
  
y(t) = z(t)y(0) = c1 + c2 −t . (20)
−4te−t e − 2te−t

Taking c1 = 1 and c2 = −2 in this equation gives a solution

 −t   
e 1
y 1 (t) = = e−t ,
−2e−t −2
 
at 1
which is a solution of the form y(t) = ve where v = is a constant
−2
vector and a = −1 is a scalar. Note that (v, −1) is an eigenvector-eigenvalue
pair for the matrix A (see Example 1). That is,
      
1 1 1 −1 1
Av = A = = (−1) · .
−1 −3 −2 2 −2
J

Example 3. Find the solution of the linear homogeneous initial value prob-
lem:
y10 = y1 + 2y2
0 , y1 (0) = c1 , y2 (0) = c2 . (21)
y2 = −2y1 + y2
 
1 2
I Solution. For this system, the coefficient matrix is A = . We will
−2 1
solve this equation by using Equation (11), as was done for the previous
examples. Form the matrix
 
s − 1 −2
sI − A = .
2 s−1

Then p(s) = det(sI − A) = (s − 1)2 ) + 4, so that

 
1 s−1 2
(sI − A)−1 =
(s − 1)2 + 4 −2 s − 1
 s−1 2 
 (s − 1)2 + 4 (s − 1)2 + 4 
= .
 
 −2 s−1 
2 2
(s − 1) + 4 (s − 1) + 4
484 10 Systems of Differential Equations

Hence, using the inverse Laplace transform formulas from Table C.2, the ma-
trix z(t) of Equation (10) is
 t
e cos 2t et sin 2t

z(t) = L−1 (sI − A)−1 =


,
−et sin 2t et cos 2t

and the solution of system (21) is

c1 et cos 2t + c2 et sin 2t
   
y (t)
y(t) = 1 = .
y2 (t) −c1 et sin 2t + c2 et cos 2t

Now we return briefly to the theoretical significance of Equation (11).

According to the analysis leading to (11), the unique solution of the initial
value problem

(∗) y 0 = Ay, y(0) = y 0 ,

where A is a 2 × 2 constant matrix, is

y(t) = z(t)y(0) = z(t)y 0 ,

where  
−1
z(t) = L−1 (sI − A) .

But according to Theorem 13, the unique solution of the initial value problem
(*) is
y(t) = eAt y 0 .
These two descriptions of y(t) give an equality of matrix functions
 
(∗∗) L−1 (sI − A)−1 y 0 = z(t)y 0 = eAt y 0

which holds for all choices of the constant vector y 0 . But if C is a 2 × 2 matrix
1 0
then Ce1 = C is the first column of C and Ce2 = C is the second
0 1
column of C (check this!). Thus, if B and C are two 2 × 2 matrices such that
Bei = Cei for i = 1, 2, then B = C (since column i of B = column i of C for
i = 1, 2). Taking y 0 = ei for i = 1, 2, and applying this observation to the
matrices of (∗∗), we arrive at the following result:
Theorem 4. If A is a 2 × 2 constant matrix, then
 
−1
eAt = L−1 (sI − A) . (22)
 10.4 Constant Coefficient Homogeneous Systems 485
 
−1
Example 5. From the calculations of L−1 (sI − A) done in Examples
At
1, 2 and 3 this theorem gives the following values of e :

A eAt
1
1 2t

e2t + e−2t
e − e−2t
2 4
 
01 


40 
1 2t

2t −2t −2t
e −e e +e
2
 −t
e + 2te−t te−t
  
1 1
−4 −3 −4te−t e−t − 2te−t

et cos 2t et sin 2t
   
1 2
−2 1 −et sin 2t et cos 2t

While our derivation of the formula for eAt in Theorem 4 was done for 2×2
matrices, the formula remains valid for arbitrary constant n × n matrices A,
At
and moreover,
 once
 one can guess that there is a relationship between e and
−1
L−1 (sI − A) , it is a simple matter to verify it by computing the Laplace
transform of the matrix function eAt . This computation is, in fact, almost the
same as the computation of L(eat ) in Example ??.

Theorem 6. If A is an n × n constant matrix (whose entries can be either

real numbers or complex numbers), then

 
−1
eAt = L−1 (sI − A) . (23)

Proof. Note that if B is an n × n invertible matrix (of constants), then

d d
B −1 eBt = B −1 eBt = B −1 BeBt = eBt ,


dt dt
so that
Z t
(†) eBτ dτ = B −1 (eBt − I).
1

Note that this is just the matrix analog of the integration formula
Z t
ebτ dτ = b−1 (ebt − 1).
0
486 10 Systems of Differential Equations

Now just mimic the scalar calculation from Example ??, and note that formula
(†) will be applied with B = A − sI, where, as usual, I will denote the n × n
identity matrix.
Z ∞
L eAt (s) = eAt e−st dt


0
Z ∞
= eAt e−stI dt
0
Z ∞
= e(A−sI)t dt
0
Z N
= lim e(A−sI)t dt
N →∞ 0
 
−1
= lim (A − sI) e(A−sI)N − I
N →∞
−1
= (sI − A) .

The last equality is justified since limN →∞ e(A−sI)N = 0 if s is large enough.

This fact, analogous to the fact that e(a−s)t converges to 0 as t → ∞ provided
s > a, will not be proved. u
t

Example 7. Compute eAt for the matrix

 
1 −3 3
A = −3 1 3 ,
3 −3 1
At 0
and using
 the calculation of e , solve the initial value problem y = Ay,
1
y(0) = 1.
1
 
−1
I Solution. According to Theorem 6, eAt = L−1 (sI − A) , so we need
−1
to begin by computing (sI − A) , which is most conveniently done (by hand)
by using the adjoint formula for a matrix inverse (see Corollary 8). Recall
that this formula says that if B is an n × n matrix with det B 6= 0, then
B −1 = (det B)−1 [Cij ] where the term Cij is (−1)i+j times the determinant
of the matrix obtained by deleting the j th row and ith column from B. We
apply this with B = sI − A. Start by calculating
 
s−1 3 −3
p(s) = det(sI − A) = det  3 s − 1 −3  = (s − 1)(s + 2)(s − 4).
−3 3 s−1

In particular, sI − A is invertible whenever p(s) 6= 0, i.e., whenever s 6= 1, −2,

or 4. Then a tedious, but straightforward calculation, gives
 10.4 Constant Coefficient Homogeneous Systems 487

(s − 1)2 + 9 −3(s + 2)
 
3(s − 4)
−1 1 
(sI − A) = −3(s − 4) (s − 1)2 − 9 3(s − 4) 
p(s)
3(s + 2) −3(s + 2) (s − 1)2 − 9
(s − 1)2 + 9 −3 3
 
 p(s) (s − 1)(s + 4) (s − 1)(s + 2) 
 −3 1 3 
=
 
 (s − 1)(s + 2) s−1 (s − 1)(s + 2) 

 3 −3 1 
(s − 1)(s − 4) (s − 1)(s − 4) s−1
 1 1 1 1 1 1 1 
− + + − −
 s − 1 s + 2 s − 4 s − 1 s − 4 s + 1 s + 2
 −1 1 1 1 1 
= + − .
 s − 1 s + 2 s − 1 s + 1 s + 2
 −1 1 1 1 1 
+ −
s−1 s−4 s−1 s−4 s−1
By applying the inverse Laplace transform to each function in the last matrix
gives
 t
−e + e−2t + e4t et − e4t et − e−2t

 
−1
eAt = L−1 (sI − A) =  −et + e−2t et et − e−2t  .
t 4t t 4t
−e + e e −e et
 
1
Then the solution of the initial value problem is given by y(t) = eAt 1 =
  1
1
1 et . J
1
Remark 8. Most of the examples of numerical systems which we have dis-
cussed in this section are first order constant coefficient linear systems with
two unknown functions, i.e. n = 2 in Definition 3. Nevertheless, the same
analysis works for first order constant coefficient linear systems in any num-
ber of unknown functions, i.e. arbitrary n. Specifically, Equations (6) and (11)
apply to give the Laplace transform Y (s) and the solution function y(t) for
the constant coefficient homogeneous linear system

y 0 = Ay

where A is an n × n constant matrix. The practical difficulty in carrying out

this program is in calculating (sI − A)−1 . This can be done by programs
like Mathematica, MatLab, or Maple if n is not too large. But even if the
calculations of specific entries in the matrix (sI − A)−1 are difficult, one can
extract useful theoretical information concerning the nature of the solutions
of y 0 = Ay + q(t) from the formulas like Equation (11) and from theoretical
algebraic descriptions of the inverse matrix (sI − A)−1 .
488 10 Systems of Differential Equations

Exercises
For each of the following matrices A, (a) find the matrix z(t) = L−1 (sI − A)−1 

from Equation (4.4.10) and (b) find the general solution of the homogeneous system
y = Ay. It will be useful to review the calculations in Examples 4.4.1 – 4.4.3.
0

−1 0 0 2 21
1. 2. 3.
 0 3  −2 0  0 2
−1 2 2 −1 3 −4
4. 5. 6.
−2 −1  −2
3 1 −1
2 −5 −1 −4 2 1
7. 8. 9.
 −2
1  1 −1 1 2
−1 0 3 0 4 0
5 2
10.  0 2 0
11.  −1 0 0 
12. 
−8 −3 0 0 1 1 4 −1
−2 2 1 0 1 1 3 1 −1
 0 −1 0 
13.   1 1 −1
14.  0 3 −1
15. 
2 −2 −1 −2 1 3  00 3

10.5 Computing eAt

In this section we will present a variant of a technique due to Fulmer2 for
computing the matrix exponential eAt . It is based on the knowledge of what
type of functions are included in the individual entries of eAt . This knowledge
is derived from our understanding of the Laplace transform table and the
fundamental formula  
−1
eAt = L−1 (sI − A)
which was proved in Theorem 6.
To get started, assume that A is an n × n constant matrix. The matrix
sI − A is known as the characteristic matrix of A and its determinant

p(s) := det(sI − A)

is known as the characteristic polynomial of A. The following are some

basic properties of sI −A and p(s) which are easily derived from the properties
of determinants in Section 9.4.
1. The polynomial p(s) has degree n, when A is an n × n matrix.
2. The characteristic matrix sI − A is invertible except when p(s) = 0.
3. Since p(s) is a polynomial of degree n, it has at most n roots (exactly n if
multiplicity of roots and complex roots are considered). The roots of p(s)
are called the eigenvalues of A.
2
Edward P. Fulmer, Computation of the Matrix Exponential, American Mathe-
matical Monthly, 82 (1975) 156–159.
 10.5 Computing eAt 489

4. The inverse of sI − A is given by the adjoint formula (Corollary 8)

 
−1 1 Cij (s)
(∗) (sI − A) = [Cij (s)] =
p(s) p(s)
where Cij (s) is (−1)i+j times the determinant of the matrix obtained from
sI − A by deleting the ith column and j th row. For example, if n = 2 then
we get the formula
 −1  
s − a −b 1 s−d b
=
−c s − d p(s) c s−a
which we used in Section 10.4.
5. The functions Cij (s) appearing in (∗) are polynomials of degree at most
n − 1. Therefore, the entries
Cij (s)
pij (s) =
p(s)
of (sI − A)−1 are proper rational functions with denominator of degree n.
6. Since    

−1 Cij (s)
eAt = L−1 (sI − A) = L−1 ,
p(s)
the form of the functions
 
−1 Cij (s)
hij (t) = L ,
p(s)
which are the individual entries of the matrix exponential eAt , are com-
pletely determined by the roots of p(s) and their multiplicities via the
analysis of inverse Laplace transforms of rational functions as described
in Section 3.3.
7. Suppose that r is an eigenvalue of A of multiplicity k. That is, r is a root
of the characteristic polynomial p(s) and (s − r)k divides p(s), but no
higher power of s − r divides p(s). We distinguish two cases:
Case 1: The eigenvalue r is real.
In this case r will contribute a linear combination of the functions
(∗real ) ert , tert , · · · , tk−1 ert
to each hij .
Case 2: The eigenvalue r = α + iβ has nonzero imaginary part β 6= 0.
In this case r = α + iβ and its complex conjugate r = α − iβ will
contribute a linear combination of the functions
eαt cos βt, teαt cos βt, · · · , tk−1 eαt cos βt
(∗Imag. )
eαt sin βt, teαt sin βt, · · · , tk−1 eαt sin βt
to each hij .
490 10 Systems of Differential Equations

8. The total number of functions listed in (∗real ) and (∗Imag. ) counting all
eigenvalues is n = deg p(s). If we let φ1 , . . ., φn be these n functions, then
it follows from our analysis above, that each entry hij (t) can be written
as a linear combination

(∗) hij (t) = mij1 φ1 (t) + · · · + mijn φn (t)

of φ1 , . . ., φn . We will define an n × n matrix Mk = [mijk ] whose ij th

entry is the coefficient of φk (t) in the expansion of hij (t) in (∗). Then we
have a matrix equation expressing this linear combination relation:

(∗∗) eAt = [hij (t)] = M1 φ1 (t) + · · · + Mn φ(t).

Example 1. As a specific example of the decomposion given by (∗∗), con-

sider the matrix eAt from Example 7:
 t
−e + e−2t + e4t et − e4t et − e−2t


eAt =  −et + e−2t et et − e−2t  .

t 4t t 4t
−e + e e −e et

In this case (refer to Example 7 for details), p(s) = (s − 1)(s + 2)(s − 4) so

the eigenvalues are 1, −2 and 4 and the basic functions φi (t) are φ1 (t) = et ,
φ2 (t) = e−2t and φ3 (t) = e4t . Then (∗∗) is the identity
     
−1 1 1 1 0 −1 1 −1 0
eAt = −1 1 1 et + 1 0 −1 e−2t + 0 0 0 e4t ,
−1 1 1 00 1 1 −1 0

where      
−1 1 1 1 0 −1 1 −1 0
M1 = −1 1 1 , M2 = 1 0 −1 and M3 = 0 0 0 .
−1 1 1 00 1 1 −1 0

With the notational preliminaries out of the way, we can give the variation
on Fulmer’s algorithm for eAt .
Algorithm 2 (Fulmer’s method). The following procedure will compute
eAt where A is a given n × n constant matrix.
1. Compute p(s) = det(sI − A).
2. Find all roots and multiplicities of the roots of p(s).
3. From the above observations we have

(‡) eAt = M1 φ1 (t) + · · · + Mn φn (t),

where Mi i = 1, . . . , n are n × n matrices. We need to find these matrices.

By taking derivatives we obtain a system of linear equations (with matrix
coefficients)
 10.5 Computing eAt 491

eAt = M1 φ1 (t) + · · · + Mn φn (t)

AeAt = M1 φ01 (t) + · · · + Mn φ0n (t)
..
.
(n−1)
An−1 eAt = M1 φ1 (t) + · · · + Mn φ(n−1)
n (t).

Now we evaluate this system at t = 0 to obtain

I = M1 φ1 (0) + · · · + Mn φn (0)
A = M1 φ01 (0) + · · · + Mn φ0n (0)
.............................................. (1)
(n−1) (n−1)
An−1 = M1 φ1 (0) + · · · + Mn φn (0).

Let
φ1 (0) ... φn (0)
 
.. .. ..
W = . . .
 

(n−1) (n−1)
φ1 (0) . . . φn (0)
Then W is a nonsingular n × n matrix; its determinant is just the Wron-
skian evaluated at 0. So W has an inverse. The above system of equations
can now be written:    
I M1
 A   M2 
 ..  = W  ..  .
   
 .   . 
An−1 Mn
Therefore,   
I M1
 A   M2 
W −1  ..  =  ..  .
   
 .   . 
An−1 Mn
Having solved for M1 , . . . , Mn we obtain eAt from (‡). u
t
Remark 3. Note that this last equation implies that each matrix Mi is a
polynomial in the matrix A since W −1 is a constant matrix. Specifically,
Mi = pi (A) where  
1
 s 
pi (s) = Rowi (W −1 )  .  .
 
 .. 
sn−1
 
1
Example 4. Solve y 0 = Ay with initial condition y(0) = , where A =
  2
2 −1
.
1 0
492 10 Systems of Differential Equations

I Solution. The characteristic polynomial is p(s) = (s − 1)2 . Thus there

is only one eigenvalue r = 1 with multiplicity 2 so only case (∗real ) occurs
and all of the entries hij (t) from eAt are linear combinations of et , tet . That
is φ1 (t) = et while φ2 (t) = tet . Therefore, Equation (∗∗) is

eAt = M et + N tet .

Differentiating we obtain

AeAt = M et + N (et + tet )

= (M + N )et + N tet .

Now, evaluate each equation at t = 0 to obtain:

I= M
A = M + N.

Solving for M and N we get

M= I
N = A − I.

Thus,

eAt = Iet + (A − I)tet

  
10 t 1 −1
= e + tet
01 1 −1
 t
e + tet −tet

=
tet et − tet

We now obtain
 
1
y(t) = eAt y 0 = eAt
2
 t t t
 
e + te −te 1
=
tet et − tet 2
 t
e − tet

= .
−tet + 2et

J
− 21
 
1 0
Example 5. Compute eAt where A = 1 1 −1 using Fulmer’s method.
0 12 1
 10.5 Computing eAt 493

I Solution. The characteristic polynomial is p(s) = (s − 1)(s2 − 2s + 2).

The eigenvalues of A are thus r = 1 and r = 1 ± i. From (∗real ) and (∗Imag. )
each entry of eAt is a linear combination of

φ1 (t) = et , φ2 (t) = et sin t, and φ3 (t) = et cos t.

Therefore
eAt = M et + N et sin t + P et cos t.
Differentiating twice and simplifying we get the system:

eAt = M et + N et sin t + P et cos t

At
Ae = M e + (N − P )et sin t + (N + P )et cos t
t

A2 eAt = M et − 2P et sin t + 2N et cos t.

Now evaluating at t = 0 gives

I= M +P
A= M +N +P
A2 = M + 2N.

Solving gives

N= A−I
2
M = A − 2A + 2I
P = −A2 + 2A − I.
1
−1 12

2
Since A2 =  2 0 −2, it follows that
1 1
2 1 2
 −1 1
0 12
   1 −1

0 2 0 2 2 0 2
N = 1 0 −1 M =  0 0 0  and P =  0 1 0  .
0 12 0 1
2 0 12 −1
2 0 2
1

Hence,
1
0 12
  −1   1 −1

2 0 2 0 2 0 2
eAt =  0 0 0  et + 1 0 −1 et sin t +  0 1 0  et cos t
1 1
2 0 2 0 12 0 −1
2 0 2
1
 t
e + et cos t −et sin t et − et cos t

1
= 2et sin t 2et cos t −2et sin t  .
2 t
e − et cos t et sin t et + et cos t
J
The technique of this section is convenient for giving an explicit formula
for the matrix exponential eAt when A is either a 2 × 2 or 3 × 3 matrix.
494 10 Systems of Differential Equations

eA for 2 × 2 matrices.
Suppose that A is a 2 × 2 real matrix with characteristic polynomial p(s) =
det(sI − A) = s2 + as + b. We distinguish three cases.
1. p(s) = (s − r1 )(s − r2 ) with r1 6= r2 .
Then the basic functions are φ1 (t) = et1 t and φ2 (t) = er2 t so that eAt =
M er1 t + N er2 t . Equation (1) is then

I = M +N
A = r1 M + r2 N

which are easily solved to give

(A − r2 I) (A − r1 I)
M= and N = .
r1 − r2 r2 − r1
Hence, if p(s) has distinct roots, then

(A − r2 I) r1 t (A − r1 I) r2 t
eAt = e + e . (2)
r1 − r2 r2 − r1

2. p(s) = (s − r)2 .
In this case the basic functions are ert and tert so that

(∗) eAt = M ert + N tert .

This time it is more convenient to work directly from (∗) rather than Equa-
tion (1). Multiplying (∗) by e−rt and observing that eAt e−rt = eAt e−rtI =
e(A−rI)t (because A commutes with rI), we get

M + N t = e(A−rI)t
1
= I + (A − rI)t + (A − rI)2 t2 + · · · .
2
Comparing coefficients of t on both sides of the equation we conclude that

M = I, N = (A − rI) and (A − rI)n = 0 for all n ≥ 2.

Hence, if p(s) has a single root of multiplicity 2, then

eAt = (I + (A − rI)t) ert . (3)

 10.5 Computing eAt 495

3. p(s) = (s−α)2 +β 2 where β 6= 0, i.e., p(s) has a pair of complex conjugate

roots α ± β.
In this case the basic functions are eαt cos βt and eαt sin βt so that

eAt = M eαt cos βt + M eαt sin βt.

Equation (1) is easily checked to be

I=M
A = αM + βN.

Solving for M and N then gives

(A − αI) αt
eAt = Ieαt cos βt + e sin βt. (4)
β

eA for 3 × 3 matrices.
Suppose that A is a 3 × 3 real matrix with characteristic polynomial p(s) =
det(sI − A). As for 2 × 2 matrices, we distinguish three cases.
1. p(s) = (s − r1 )(s − r2 )(s − r3 ) with r1 . r2 , and r3 distinct roots of p(s).
This is similar to the first case done above. The basic functions are er1 t ,
er2 t , and er3 t so that

eAt = M er1 t + N er2 t + P er3 t

and the system of equations (1) is

I = M + N + P
A = r1 M + r2 N + r3 P (5)
A2 = r12 M + r22 N + r32 P.

We will use a very convenient trick for solving this system of equations.
Suppose that q(s) = s2 + as + b is any quadratic polynomial. Then in
system (5), multiply the first equation by b, the second equation by a,
and then add the three resulting equations together. You will get

A2 + aA + bI = q(A) = q(r1 )M + q(r2 )N + q(r3 )P.

Suppose that we can choose q(s) so that q(r2 ) = 0 and q(r3 ) = 0. Since a
quadratic can only have 2 roots, we will have q(r1 ) 6= 0 and hence

q(A)
M= .
q(r1 )
496 10 Systems of Differential Equations

But it is easy to find the required q(s), namely, use q(s) = (s − r2 )(s − r3 ).
This polynomial certainly has roots r2 and r3 . Thus, we find

(A − r2 I)(A − r3 I)
M= .
(r1 − r2 )(r1 − r3 )

Similarly, we can find N by using q(s) = (s − r1 )(s − r3 ) and P by using

q(s) = (s − r1 )(s − r2 ). Hence, we find the following expression for eAt :

(A − r2 I)(A − r3 I) r1 t (A − r1 I)(A − r3 I) r2 t (A − r1 I)(A − r2 I) r3 t

eAt = e + e + e .
(r1 − r2 )(r1 − r3 ) (r2 − r1 )(r2 − r3 ) (r3 − r1 )(r3 − r2 )

(6)

2. p(s) = (s − r)3 , i.e, there is a single eigenvalue of multiplicity 3.

In this case the basic functions are ert , tert , and t2 ert so that

eAt = M ert + N tert + P t2 ert .

As for the case of 2 × 2 matrices, multiply by e−rt to get

M + N t + P t2 = eAt e−rt = e(A−rI)t

1 1
= I + (A − rI)t + (A − rI)2 t2 + (A − rI)3 t3 + · · · .
2 3!
Comparing powers of t on both sides of the equation gives
(A−rI)2
M = I, N = (A − rI), P = 2 and (A − rI)n = 0 if n ≥ 3.

Hence,
 
1
eAt = I + (A − rI)t + (A − rI)2 t2 ert . (7)
2

3. p(s) = (s − r1 )2 (s − r2 ) where r1 6= r2 . That is, A has one eigenvalue with

multiplicity 2 and another with multiplicity 1.
The derivation is similar to that of the case p(s) = (s−r)3 . We will simply
record the result:

(A − r1 I)2 (A − r1 I)2 (A − r1 I)2 r2 t

eAt = I− er1 t + (A − r1 I) − ter1 t + e .
 (r2 − r1 ) 
2
 r2 − r1  (r2 − r1 )2

(8)
 10.6 Nonhomogeneous Linear Systems 497

10.6 Nonhomogeneous Linear Systems

This section will be concerned with the nonhomogeneous linear equation

(∗) y 0 = A(t)y + q(t),

where A(t) and q(t) are matrix functions defined on an interval J in R. The
strategy will be analogous to that of Section
 6.6 in that we will assume that
we have a fundamental matrix Φ(t) = ϕ1 (t) · · · ϕn (t) of solutions of the
associated homogeneous system

(∗h ) y 0 = A(t)y

and we will then use this fundamental matrix Φ(t) to find a solution y p (t) of
(∗) by the method of variation of parameters. Suppose that y 1 (t) and y 2 (t)
are two solutions of the nonhomogeneous system (∗). Then

(y 1 −y2 )0 (t) = y 01 (t)−y 02 (t) = (A(t)y 1 (t)+q(t))−(A(t)y 2 (t)+q(t)) = A(t)(y 1 (t)−y 2 (t))

so that y 1 (t) − y 2 (t) is a solution of the associated homogeneous system (∗h ).

Since Φ(t) is a fundamental matrix of (∗h ), this means that

y 1 (t) − y 2 (t) = Φ(t)c = c1 ϕ1 (t) + · · · + cn ϕn (t)

for some constant matrix  

c1
c =  ...  .
 

cn
Thus it follows that if we can find one solution, which we will call y p (t), then
all other solutions are determined by the equation

y(t) = y p (t) + Φ(t)c = y p (t) + y h (t)

where y h (t) = Φ(t)c (c an arbitrary constant vector) is the solution of the

associated homogeneous equation (∗h ). This is frequently expressed by the
mnemonic:

y gen (t) = y p (t) + y h (t), (1)

or in words: The general solution of a nonhomogeneous equation is the sum of

a particular solution and the general solution of the associated homogeneous
equation. The strategy for finding a particular solution of (∗), assuming that
we already know y h (t) = Φ(t)c, is to replace the constant vector c with an
unknown vector function
498 10 Systems of Differential Equations
 
v1 (t)
v(t) =  ...  .
 

vn (t)
That is, we will try to choose v(t) so that the vector function

(†) y(t) = Φ(t)v(t) = v1 (t)ϕ1 (t) + · · · + vn (t)ϕn (t)

is a solution of (∗). Differentiating y(t) gives y 0 (t) = Φ0 (t)v(t) + Φ(t)v 0 (t),

and substituting this expression for y 0 (t) into (∗) gives

Φ0 (t)v(t) + Φ(t)v 0 (t) = y 0 (t) = A(t)y(t) + q(t) = A(t)Φ(t)v(t) + q(t).

But Φ0 (t) = A(t)Φ(t) (since Φ(t) is a fundamental matrix for (∗h )) so

Φ0 (t)v(t) = A(t)Φ(t)v(t) cancels from both sides of the equation to give

Φ(t)v 0 (t) = q(t).

Since Φ(t) is a fundamental matrix Theorem 4 implies that Φ(t)−1 exists, and
we arrive at an equation

(‡) v 0 (t) = Φ(t)−1 q(t)

for v 0 (t). Given an initial point t0 ∈ J, we can then integrate (‡) to get
Z t
v(t) − v(t0 ) = Φ(u)−1 q(u) du,
t0

and multiplying by Φ(t) gives

Z t
y(t) − Φ(t)v(t0 ) = Φ(t) Φ(u)−1 q(u) du.
t0

But if y(t0 ) = y 0 , then y 0 = y(t0 ) = Φ(t0 )v(t0 ), and hence v(t0 ) =

Φ(t0 )−1 y 0 . Substituting this expression in the above equation, we arrive at
the following result, which we formally record as a theorem.
Theorem 1. Suppose that A(t) and q(t) are continuous on an interval J and
t0 ∈ J. If Φ(t) is a fundamental matrix for the homogeneous system y 0 =
A(t)y then the unique solution of the nonhomogeneous initial value problem

y 0 = A(t)y + q(t), y(t0 ) = y 0

is
Z t
−1
y(t) = Φ(t) (Φ(t0 )) y 0 + Φ(t) Φ(u)−1 q(u) du. (2)
t0
 10.6 Nonhomogeneous Linear Systems 499

Remark 2. The procedure described above is known as variation of para-

meters for nonhomogeneous systems. It is completely analogous to the
technique of variation of parameters previously studied for a single second
order linear nonhomogeneous differential equation. See Section 6.6.

Remark 3. How does the solution of y 0 = A(t)y+q(t) expressed by Equation

(2) correlate to the general mnemonic expressed in Equation (1)? If we let the
initial condition y 0 vary over all possible vectors in Rn , then y h (t) is the
first part of the expression on the right of Equation (2). That is y h (t) =
Φ(t) (Φ(t0 ))−1 y 0 . The second part of the expression on the right of Equation
(2) is the particular solution of y 0 = A(t)y + q(t) corresponding to the specific
initial condition y(t0 ) = 0. Thus, in the language of (1)
Z t
y p = Φ(t) Φ(u)−1 q(u) du.
t0

Finally, y gen (t) is just the function y(t), and the fact that it is the general
solution is just the observation that the initial vector y 0 is allowed to be
arbitrary.

Example 4. Solve the initial value problem

     
01 0 1
y0 = y+ , y(0) = . (3)
21 −et −1

I Solution. From Example 5 we have that

 2t −t 
e e
Φ(t) =
2e2t −e−t
 
01
is a fundamental matrix for the homogeneous system y 0 = y, which is
21
the associated homogeneous system y 0 = Ay for the nonhomogeneous system
(3). Then det Φ(t) = −3et and
 −t
−e −e−t 1 e−2t e−2t
  
1
Φ(t)−1 = = .
−3et −2e2t e2t 3 2et −et
Then       
−1 1 1 1 1 1 0
Φ(0) = = ,
−1 3 2 −1 −1 1
and hence
   2t −t     −t 
−1 1 e e 0 e
Φ(t)Φ(0) = = .
−1 2e2t −e−t 1 −e−t

which is the first part of y(t) in Equation (2). Now compute the second half
of Equation (2):
500 10 Systems of Differential Equations
t  Z t  −2u −2u  
e2t e−t
 
1 e
Z
−1 e 0
Φ(t) Φ(u) q(u) du = du
t0 2e2t −e−t 0 3 2eu −eu −eu
 2t −t  Z t  −u 
e e 1 −e
= du
2e2t −e−t 0 3 e2u
 2t −t  1 (e−t − 1)
 
e e 3
=
2e2t −e−t
 
1 2t
6 (e − 1)
1 t 2t 1 t −t

3 (e − e ) + 6 (e − e )
= 
2 t 2t 1 t −t
(e − e ) − (e − e )
 31 t 1 2t 16 −t 
2e − 3e − 6e
= .
1 t 2 2t 1 −t
2e − 3e + 6e

Putting together the two parts which make up y(t) in Equation (2) we get
 1 t 1 2t 1 −t 
2e − 3e − 6e
 −t 
e
y(t) = −t +
 .
−e 1 t 2 2t 1 −t
2e − 3e + 6e

We will leave it as an exercise to check our work by substituting the above

expression for y(t) back into the system (3) to see that we have in fact found
the solution. J
If the linear system y 0 = Ay + q(t) is constant coefficient, then a funda-
mental matrix for the associated homogeneous system is Φ(t) = eAt . Since
(eAt )−1 = e−At , it follows that eAt (eAt0 )−1 = eA(t−t0 ) and hence Theorem 1
has the following form in this situation.
Theorem 5. Suppose that A is a constant matrix and q(t) is a continuous
vector function on an interval J and t0 ∈ J. Then the unique solution of the
nonhomogeneous initial value problem
y 0 = Ay + q(t), y(t0 ) = y 0
is
Z t
y(t) = eA(t−t0 ) y 0 + eAt e−Au q(u) du. (4)
t0

You should compare the statement of this theorem with the solution of
the first order linear initial value problem as expressed in Corollary 9.
Example 6. Solve the initial value problem
   t  
−1 1 e 1
y0 = y+ , y(0) = .
−4 3 2et 0
 10.6 Nonhomogeneous Linear Systems 501
   t
−1 1 e
I Solution. In this example, A = , q(t) = t , t0 = 0, and y(0) =
  −4 3 2e
1
. Since the characteristic polynomial of A is p(s) = det(sI − A) = (s − 1)2 ,
0
the fundamental matrix eAt can be computed from Equation (3):
eAt = (I + (A − I)t)et
 
1 − 2t t
= et .
−4t 1 + 2t

Since e−At = eA·(−t) , we can compute e−At by simply replacing t by −t in the

formula for eAt :  
−At 1 + 2t −t
e = e−t .
4t 1 − 2t
Then applying Equation (4) give
Z t
At At
y(t) = e y 0 + e e−Au q(u) du
0
      Z t   u
1 − 2t t t 1 1 − 2t t t 1 + 2u −u −u e
= e + e e du
−4t 1 + 2t 0 −4t 1 + 2t 0 4u 1 − 2u 2eu
 Z t 
(1 − 2t)et
  
1 − 2t t 1
= + et du
−4tet −4t 1 + 2t 0 2
(1 − 2t)et
     
1 − 2t t t
= + et
−4tet −4t 1 + 2t 2t
(1 − 2t)et
   t
te
= + .
−4tet 2tet
J
If we take the initial point t0 = 0 in Theorem 5, then we can get a further
refinement of Equation (4) by observing that
Z t Z t
At −Au
e e q(u) du = eA(t−u) q(u) du = eAt ∗ q(t)
0 0
At
where e ∗ q(t) means the matrix of functions obtained by a formal matrix
multiplication in which ordinary product of entries arereplaced by the
 con- 
h11 (t) h12 (t) q (t)
volution product. For example, if B(t) = and q(t) = 1 ,
h21 (t) h22 (t) q2 (t)
then by B(t) ∗ q(t) we mean the matrix
t t
(h11 ∗ q1 )(t) + (h12 ∗ q2 )(t) h (t − u)q1 (u) du +
0 11 0
h12 (t − u)q2 (u) du
B(t)∗q(t) = 
 = 
 .
t t
(h21 ∗ q1 )(t) + (h22 ∗ q2 )(t) h (t − u)q1 (u) du +
0 21
h (t − u)q2 (u) du
0 22

With this observation we can give the following formulation of Theorem 5

in terms of the convolution product.
502 10 Systems of Differential Equations

Theorem 7. Suppose that A is a constant matrix and q(t) is a continuous

vector function on an interval J and 0 ∈ J. Then the unique solution of the
nonhomogeneous initial value problem

y 0 = Ay + q(t), y(t0 ) = y 0

is

y(t) = eAt y 0 + eAt ∗ q(t). (5)

Remark 8. For low dimensional examples, the utility of this result is greatly
enhanced by the use of the explicit formulas (2) – (8) from the previous section
and the table of convolution products (Table C.3).

Example 9. Solve the following constant coefficient non-homogeneous lin-

ear system:
y10 = y2 + e3t
(6)
y20 = 4y1 + et .
 
01
I Solution. In this system the coefficient matrix is A = and q(t) =
 3t  40
e
. The associated homogeneous equation y 0 = Ay has already been stud-
et
ied in Example 1 where we found that a fundamental matrix is
1
1 2t

e2t + e−2t e − e−2t
2 4 
z(t) =  . (7)
1 2t
 
e2t − e−2t −2t


e +e
2
 
10
Since, z(0) = it follows that z(t) = eAt . If we write eAt in terms of its
01
columns, so that eAt = z 1 (t) z 2 (t) then we conclude that a solution of the
 

initial value problem

y 0 = Ay + q(t), y(0) = 0

is given by
 10.6 Nonhomogeneous Linear Systems 503

y p (t) = eAt ∗ q(t)

= z 1 (t) ∗ q1 (t) + z 2 (t) ∗ q2 (t)
= z1 ∗ e3t + z2 ∗ et
"1
# "1
#
2t −2t 2t −2t
e + e e − e
= 2 2t ∗ e3t + 41 2t
∗ et
e − e−2t 2 e + e −2t

"1 1


# " 1 2t 1


#
2 e3t − e2t + 5 e3t − e−2t t t
4 e −e − 3 e −e
−2t
= 1
+ 1 2t 1
e3t − e2t − e3t − e−2t t 3t −2t




5 2 e −e + 3 e −e
"3 1 2t 1 −2t
"1 1 −2t
− 13 et
# #
3t 2t
5e − 2e − 10 e 4 e + 12 e
= 4 3t 1 −2t
+ 1 2t 1 −2t
2t
5e − e + 5e 2e − 6e − 13 et
"3 #
5e
3t
− 13 et − 14 e2t − 1 −2t
60 e
= 4 3t
.
5e − 13 et − 12 e2t + 1 −2t
30 e

The general solution to (6) is then obtained by taking y p (t) and adding
 
c
to it the general solution y h (t) = eAt 1 of the associated homogeneous
c2
equation. Hence,
"3 1 t 1 2t 1 −2t
#
3t
5 e − 3 e − 4 e − 60 e
1
c1 + 14 c2 2t
1
c1 − 14 c2 −2t
 
ygen = 2 e + 2 e + 4 3t 1 t 1 2t .
c1 + 12 c2 −c1 + 12 c2 e − e − e + 1 e−2t
5 3 2 30

Exercises
In part (a) of each exercise in Section 4.4, you were asked to find eAt for the given
matrix A. Using your answer to that exercise, solve the nonhomogeneous equation

y 0 = Ay + q(t), y(0) = 0,

where A is the matrix in the corresponding exercise in Section 4.4 and q(t) is the
following matrix function. (Hint: Theorem 4.6.5 and Example 4.6.7 should prove
particularly useful to study for these exercises.)
e−t 0 t
1. q(t) = t 2. q(t) = 3. q(t) =
2e   t
cos 
1
et
et 0
5. q(t) = 7. q(t) = 11. q(t) = e2t 

  sin t
−t
e
e−t 
A
Complex Numbers

A.1 Complex Numbers

The history of numbers starts in the stone age, about 30,000 years ago. Long
before humans could read or write, a caveman who counted the deer he killed
by a series of notches carved into a bone, introduced mankind to the natural
counting numbers 1, 2, 3, 4, · · · . To be able to describe quantities and their
relations among each other, the first human civilizations expanded the number
system first to rational numbers (integers and fractions) √ and then to real
numbers (rational numbers and irrational numbers like 2 and π). Finally in
1545, to be able to tackle more advanced computational problems in his book
about The Great Art (Ars Magna), Girolamo Cardano √ brought the complex
numbers (real numbers and “imaginary” numbers like −1) into existence.
Unfortunately, 450 years later and after changing the whole of mathematics
forever, complex numbers are still greeted by the general public with suspicion
and confusion.
The problem is that most folks still think of numbers as entities that are
used solely to describe quantities. This works reasonably well if one restricts
the number universe to the real numbers,√but fails miserably if one considers
complex numbers: no one will ever catch −1 pounds of crawfish, not even a
mathematician.
In mathematics, numbers are used to do computations, and it is a matter
of fact that nowadays almost all serious computations in mathematics require
somewhere along the line the use of the largest possible number system given
to mankind: the complex numbers. Although complex numbers are useless
to describe the weight of your catch of the day, they are indispensable if,
for example, you want to make a sound mathematical prediction about the
behavior of any biological, chemical, or physical system in time.
Since the ancient Greeks, the algebraic concept of a real number is asso-
ciated with the geometric concept of a point on a line (the number line), and
these two concepts are still used as synonyms. Similarly, complex numbers
can be given a simple, concrete, geometric interpretation as points in a plane;
506 A Complex Numbers

i.e., any complex number z corresponds to a point in the plane (the number
plane) and can be represented in Cartesian coordinates as z = (x, y), where
x and y are real numbers.
We know from Calculus II that every point z = (x, y) in the planep can be
described also in polar coordinates as z = [α, r], where r = |z| = x2 + y 2
denotes the radius (length, modulus, norm, absolute value, distance
to the origin) of the point z, and where α = arg(z) is the angle (in radians)
between the positive x-axis and the line joining 0 and z. Note that α can be
determined by the equation tan α = y/x, when x 6= 0, and knowledge of which
quadrant the number z is in. Be aware that α is not unique; adding 2πk to α
gives another angle (argument) for z.
We identify the real numbers with the x-axis in the plane; i.e., a real
number x is identified with the point (x, 0) of the plane, and vice versa. Thus,
the real numbers are a subset of the complex numbers. As pointed out above,
in mathematics the defining property of numbers is not that they describe
quantities, but that we can do computations with them; i.e., we should be
able to add and multiply them. The addition and multiplication of points in
the plane are defined in such a way that
(a) they coincide on the x-axis (real numbers) with the usual addition and
multiplication of real numbers, and
(b) all rules of algebra for real numbers (points on the x-axis) extend to com-
plex numbers (points in the plane).
Addition: we add complex numbers coordinate-wise in Cartesian coordi-
nates. That is, if z1 = (x1 , y1 ) and z2 = (x2 , y2 ), then

z1 + z2 = (x1 , y1 ) + (x2 , y2 ) := (x1 + x2 , y1 + y2 ).

Multiplication: we multiply complex numbers in polar coordinates by

adding their angles α and multiplying their radii r (in polar coordinates).
That is, if z1 = [α1 , r1 ] and z2 = [α2 , r2 ], then

z1 z2 := [α1 + α2 , r1 r2 ].

The definition of multiplication of points in the plane is an extension of the

familiar rule for multiplication of signed real numbers: plus times plus is plus,
minus times minus is plus, plus times minus is minus. To see this, we identify
the real numbers 2 and −3 with the complex numbers z1 = (2, 0) = [0, 2] and
z2 = (−3, 0) = [π, 3]. Then

z1 z2 = [0 + π, 2 · 3] = [π, 6] = (−6, 0) = −6
z22 = [π, 3][π, 3] = [π + π, 3 · 3] = [2π, 9] = (0, 9) = 9,

which is not at all surprising since we all know that 2·−3 = −6, and (−3)2 = 9.
What this illustrates is part (a); namely, the arithmetic of real numbers is the
 A.1 Complex Numbers 507

same whether considered in their own right, or considered as a subset of the

complex numbers.
To demonstrate the multiplication of complex numbers (points in √ the
plane) which are not real (not on the x-axis), consider z1 = (1, 1) = [ π4 , 2]
√
and z2 = (1, −1) = [− π4 , 2]. Then
π π √ √
z1 z2 = [ − , 2 · 2] = [0, 2] = (2, 0) = 2.
4 4
If one defines multiplication of points in the plane as above, the point
i := (0, 1) = [ π2 , 1] has the property that
π π
i2 = [ + , 1 · 1] = [π, 1] = (−1, 0) = −1.
2 2
Thus, one defines √
−1 := i = (0, 1).
√
Notice that −1 is not on the x-axis and is therefore not a real number.
Employing i and identifying the point (1, 0) with the real number 1, one can
now write a complex number z = (x, y) in the standard algebraic form
z = x + iy; i.e.,

z = (x, y) = (x, 0) + (0, y) = x(1, 0) + (0, 1)y = x + iy.

If z = (x, y) = x + iy, then the real number x := Re z is called the real part
and the real number y := Im z is called the imaginary part of z (which is
one of the worst misnomers in the history of science since there is absolutely
nothing imaginary about y).
The basic rules of algebra carry over to complex numbers if we simply
remember the identity i2 = −1. In particular, if z1 = x1 +iy1 and z2 = x2 +iy2 ,
then

z1 z2 = (x1 + iy1 )(x2 + iy2 ) = x1 x2 + iy1 x2 + x1 iy2 + iy1 iy2

= (x1 x2 − y1 y2 ) + i(x1 y2 + x2 y1 ) = (x1 x2 − y1 y2 , x1 y2 + x2 y1 ).

This algebraic rule is often easier to use than the geometric definition of
multiplication given above. For example, if z1 = (1, 1) = 1 + i and z2 =
(1, −1) = 1 − i, then the computation z1 z2 = (1 + i)(1 − i) = 1 − i2 = 2 is
more familiar than the one given above using the polar coordinates of z1 and
z2 .
The formula for division of two complex numbers (points in the plane)
is less obvious, and is most conveniently expressed in terms of the complex
conjugate z := (x, −y) = x − iy of a complex number z = (x, y) = x + iy.
Note that z + w = z + w, zw = z w, and
z+z z−z
|z|2 = x2 + y 2 = zz, Re z = and Im z = .
2 2i
508 A Complex Numbers

Using complex conjugates, we divide complex numbers using the formula

z z w zw
= · = .
w w w |w|2
As an example we divide the complex number z = (1, 1) = 1 + i by
w = (3, −1) = 3 − i. Then
z 1+i (1 + i)(3 + i) 2 + 4i 1 2 1 2
= = = = + i = ( , ).
w 3−i (3 − i)(3 + i) 10 5 5 5 5
Let z = (x, y)pbe a complex number with polar coordinates z = [α, r].
Then |z| = r = x2 + y 2 , Re z = x = |z| cos α, Im z = y = |z| sin α, and
tan α = y/x. Thus we obtain the following exponential form of the complex
number z; i.e.,
z = [α, r] = (x, y) = |z|(cos α, sin α) = |z|(cos α + i sin α) = |z|eiα ,
where the last identity requires Euler’s formula relating the complex exponen-
tial and trigonometric functions. The most natural means of understanding
the validity of Euler’s formula is via the power series expansions of ex , sin x,
and cos x, which were studied in calculus. Recall that the exponential function
ex has a power series expansion
∞
X xn
ex =
n=0
n!

which converges for all x ∈ R. This infinite series makes perfectly good sense
if x is replaced by any complex number z, and moreover, it can be shown
that the resulting series converges for all z ∈ C. Thus, we define the complex
exponential function by means of the convergent series
∞
X zn
ez := . (1)
n=0
n!
It can be shown that this function ez satisfies the expected functional equation,
that is
ez1 +z2 = ez1 ez2 .
1
Since e0 = 1, it follows that z = e−z . Euler’s formula will be obtained by
e
taking z = it in Definition 1; i.e.,
∞
X (it)n t2 t3 t4 t5
eit = = 1 + it − − i + + i − · · ·
n=0
n! 2! 3! 4! 5!
t2 t4 t3 t5
= (1 − + − · · · ) + i(t − + − · · · ) = cos t + i sin t = (cos t, sin t),
2! 4! 3! 5!
where one has to know that the two series following the last equality are the
Taylor series expansions for cos t and sin t, respectively. Thus we have proved
Euler’s formula, which we formally state as a theorem.
 A.1 Complex Numbers 509

Theorem 1 (Euler’s Formula). For all t ∈ R we have

eit = cos t + i sin t = (cos t, sin t) = [t, 1]. u

t

Example 2. Write z = −1 + i in exponential form.

p Solution. Note
I √ that z = (−1, 1) so that x = −1, y = 1, 7π r = |z| =
(−1)2 + 12 = 2, and tan α = y/x = −1. Thus, α = 3π 4 or α = 4 . But √
z is
3π 3π
in the 2nd quadrant, so α = 4 . Thus the polar coordinates of z are [ 4 , 2]
√ 3π
and the exponential form of z is 2ei 4 . J
πi
Example 3. Write z = 2e 6 in Cartesian form.

I Solution.
√ !
π π 3 1 √ √
z = 2(cos + i sin ) = 2 +i = 3 + i = ( 3, 1).
6 6 2 2

Using the exponential form of a complex number gives yet another de-
scription of the multiplication of two complex numbers. Suppose that z1 and
z2 are given in exponential form, that is, z1 = r1 eiα1 and z2 = r2 eiα2 . Then

z1 z2 = (r1 eiα1 )(r2 eiα2 ) = (r1 r2 )ei(α1 +α2 ) .

Of course, this is nothing more than a reiteration of the definition of mul-

tiplication of complex numbers; i.e., if z1 = [α1 , r1 ] and z2 = [α2 , r2 ], then
z1 z2 := [α1 + α2 , r1 r2 ].
√
Example 4. Find z = i. That is, find all z such that z 2 = i.
π π
I Solution. Observe that i = (0, 1) = [π/2, 1] = e 2 i . Hence, if z = ei 4 then
π π
z 2 = (ei 4 )2 = ei 2 = i so that
√ √ √
π π 2 2 2
z = cos + i sin = +i = (1 + i).
4 4 2 2 2
π π π π
Also note that i = e( 2 +2π)i so that w = e( 4 +π)i = e 4 eπi = −e 4 = −z is
another square root of i. J

Example 5. Find all complex solutions to the equation z 3 = 1.

I Solution. Note that 1 = e2πki for any integer k. Thus the cube roots of
1 are obtained by dividing the possible arguments of 1 by 3 since raising a
complex number to the third power multiplies the argument by 3 (and also
2π
cubes √the modulus). Thus the possible
√
cube roots of 1 are 1, ω = e 3 i =
4π
− 12 + 23 i and ω 2 = e 3 i = − 21 − 23 . J
510 A Complex Numbers

We will conclude this section by summarizing some of the properties of the

complex exponential function. The proofs are straight forward calculations
based on Euler’s formula and are left to the reader.
Theorem 6. Let z = x + iy. Then
1. ez = ex+iy = ex cos y + iex sin y. That is Re ez = ex cos y and Im ez =
ex sin y.
2. |ez | = ex . That is, the modulus of ez is the exponential of the real part of
z.
eiy + e−iy
3. cos y =
2
eiy − e−iy
4. sin y = u
t
2i
Example 7. Compute the real and imaginary parts of the complex function
5t
z(t) = (2 + 3i)ei 2 .

I Solution. Since z(t) = (2 + 3i)(cos 5t 5t 5t 5t

2 + i sin 2 ) = (2 cos 2 − 3 sin 2 ) +
5t 5t 5t 5t
(3 cos 2 + 2 sin 2 )i, it follows that Re z(t) = 2 cos 2 − 3 sin 2 and Im z(t) =
3 cos 5t 5t
2 + 2 sin 2 . J

Exercises
√
1. Let z = (1, 1) and w = (−1, 1). Find z · w, z
w
, w
z
, z2 , z and z 11 using
(a) the polar coordinates,
(b) the standard forms x + iy,
(c) the exponential forms.

2. Find
1 1 4 − 2i
(a) (1 + 2i)(3 + 4i) (b) (1 + 2i)2 (c) (d) (e)
2 + 3i (2 − 3i)(2 + 4i) 2+i
.

3. Solve each of the following equations for z and check your result.
z−1 2 2+i
(a) (2 + 3i)z + 2 = i (b) = (c) +1 = 2+i (d)
z
z − i 3 z
e = −1.

4. Find the modulus of each of the following complex numbers.

13 1 + 2it − t2
a) 4 + 3i (b) (2 + i)2 (c) (d) where t ∈ R.
5 + 12i 1 + t2
5. Find all complex numbers z such that |z − 1| = |z − 2|. What does this equation
mean geometrically?

6. Determine the region in the complex plane C described by the inequality

|z − 1| + |z − 3| < 4.

Give a geometric description of the region.

 A.1 Complex Numbers 511
√ √
7. Compute: (a) 2 + 2i (b) 3 + 4i

8. Write each of the following complex numbers in exponential form.

1
a) 3 + 4i (b) 3 − 4i (c) (3 + 4i)2 (d) (e) −5 (f) 3i
3 + 4i
9. Find the real and imaginary parts of each of the following functions.
a) (2 + 3i)e(−1+i)t (b) ie2it+π (c) e(2+3i)t e(−3−i)t

10. a) Find the value of the sum

1 + ez + e2z + · · · + e(n−1)z .

Hint: Compare the sum to a finite geometric series.

b) Compute sin( 2π
n
) + sin( 4π
n
) + · · · + sin( (n−1)π
n
)

11. Find all of the cube roots of 8i. That is, find all solutions to the equation z 3 = 8i.

12. By multiplying out eiθ eiφ and comparing it to ei(θ+φ) , rederive the addition
formulas for the cosine and sine functions.
B
Selected Answers

Chapter 1

Section 1.1
1. (a) 1 (b) 2 (c) 1 (d) 2 (e) 2 (f ) 4 (g) 1 (h) 3
2. y1 (t) and y3 (t) are solutions. 3. y1 (t) and y4 (t) are solutions.
4. y1 (t), y2 (t), and y3 (t) are solutions. 5. y2 (t) and y3 (t) are solutions.
12. y(t) = 12 t2 + 3t + c 13. y(t) = 21 e2t − t + c 14. y(t) = −e−t (t + 1) + c
3 2
15. y(t) = t + ln |t| + c 16. y(t) = t3 + t2 + c1 t + c2 17. y(t) = − 23 sin 3t + c1 t + c2
18. y(t) = 2e3t − 4 19. y(t) = 3e−t + 3t − 3 20. y(t) = 1/(1 + et )
−1 1 2t 7
21. y(t) = −18(t + 1) 22. y(t) = 2 e − t + 2 23. y(t) = −e−t (t + 1)
2 0
24. y(t) = − 3 sin 3t + 4t + 1 25. 4yy + 3t = 0 26. 2yy 0 = 3t2 + 2t
0 0 2
27. y = 2y − 2t + 1 28. ty = 3y − t

Section 1.2

1. From Example 1, the ball will hit the ground after T = 400/16 = 5 sec. The
speed at this time is |y 0 (T )| = gT = 32 × 5 = 160 ft/sec.

2. According to Equation (4) the height at time t is y(t) = −(g/2)t2 + v0 t + h

and the maximum height is obtained when y 0 (t) = −gt + v0 = 0. That is, when
t = v0 /g. The ball will hit the ground when y(t)= 0, so solve the quadratic
equation −(g/2)t2 + v0 t + h = 0 to get t = (v0 + v02 + 2gh)/g.

3. t1 ≈ .695 sec, t2 ≈ 4.407 sec.

4. From Equation (6) the gravitational constant is g = 2h/T 2 = 8 ft/sec2 , so the

second stone will have the height equation y(t) = −4t2 + 1000. When t = 5
the height
 is y(5)
√ = −4 × 25 + 1000 = 900 ft. It will hit the ground at time
T = 2h/g = 250 ≈ 15.81 sec.

5. R0 = kR where k is a proportionality constant.

6. P 0 = kP where k is a proportionality constant.

514 B Selected Answers

7. P 0 = kP (M − P ) where k is a proportionality constant.

8. Recall that Newton’s Law of heating and cooling states: The change in the
temperature of an object is proportional to the difference between the temperature
of the object and the temperature of the surrounding medium. Thus, if T (t) is
the temperature of the beverage at time t and A is the temperature of the
surrounding medium then
T 0 (t) = k(T − A),
for some proportionality constant k. In this case the surrounding medium is
the tub of ice, which has a temperature of A = 32◦ F, so that the differential
equation is T 0 = k(T − 32) with the initial value given by T (0) = 70◦ F. The
proportionality constant k cannot be determined from the given data.

9. In this case, the oven is the surrounding medium and has a constant temperature
of A = 350◦ F. Thus the differential equation, determined by Newton’s law of
cooling and heating, that describes the temperature of the turkey T (t) at time
t is
T 0 = k(T − 350),
where k is a proportionality constant. The initial temperature of the turkey is
40◦ , so Equation (15) gives the solution of this differential equation as

T (t) = A + (T (0) − A)ekt = 350 + (40 − 350)ekt = 350 − 310ekt ,

where the proportionality constant k is yet to be determined. To determine k

note that we are given T (1) = 120. This implies 120 = T (1) = 350 − 310ek
and solving for k gives k = ln 2331
≈ −.298. To answer question (1), compute
T (2) = 350 − 310e2k ≈ 179.35◦ . To answer question (2), we want to find t
so that T (t) = 250, i.e, solve 250 = T (t) = 350 − 310ekt . Solving this gives
kt = ln 10
31
so t ≈ 3.79 hours.

10. The initial temperature is T (0) = 180◦ and the ambient temperature is A = 70◦ .
Thus Equation (15) gives T (t) = A + (T (0) − A)ekt = 70 + 110ekt for the
temperature of the coffee at time t. The constant k is determined from the
temperature at a second time: T (3) = 140 = 70 + 110e3k so k ≈ −.1507. Thus
T (t) = 70 + 110e−0.1507t . The temperature requested is T (5) = 121.8◦ .

Section 1.3

Section 2.1
1. separable

2. not separable

3. separable
 B Selected Answers 515

4. not separable

5. separable

6. not separable

7. separable

8. not separable

9. separable

12. y 4 = 2t2 + c

13. 2y 5 = 5(t + 2)2 + c

14. y(t2 + c) = −2, y = 0

−3
15. y = ,y=0
t3 +c
16. y = 1 − c cos t, y = 1
1 − n m+1
17. y 1−n = t + c, y = 0
1+m
4ce4t
18. y = ,y=4
1 − ce4t
19. y 2 + 1 = ce2t

20. y = tan(t + c)

21. t2 + y 2 + 2 ln |t| = c

22. tan−1 t + y − 2 ln |y + 1| = c, y = −1

23. y 2 = et + c

24. y ln |c(1 − t)| = 1

25. cet = y(t + 2)2

26. y = 0

27. y = 0

28. y = x2 ex
2
29. y = 4e−t
√
30. y = sec−1 ( 2t2 )
516 B Selected Answers
√
31. y = 2 u2 + 1

Section 2.2
3. y(t) = te2t + 4e2t
1 17
4. y(t) = − e−2t + e2t
4 4
1 t e
5. y(t) = e −
t t
1 2t
6. y(t) = [e − e2 ]
2t
t+sin t cos t
7. y(t) = 2 cos t + c sec t
2 Rt 2 2
8. y(t) = e−t /2
0
es /2
ds + e−t /2

t ln t t
9. y(t) = − + ct−m
m + 1 (m + 1)2
sin(t2 )+c
10. y(t) = 2t
1
11. y(t) = t+1 (−2 + ct)

12. y(t) = b/a + ce−at

13. y(t) = 1

14. y(t) = t(t + 1)2 + c(t + 1)2

1 3 2
15. y(t) = −t − − t
2 2
16. y(t) = te−at + ce−at
1 bt
17. y(t) = e + ce−at
a+b
tn+1 −at
18. y(t) = e + ce−at
n+1
t+c
19. y(t) = cos t
2
20. y = 2 + ce−(ln t)

21. y(t) = tn et + ctn

22. y(t) = (t − 1)e2t + (a + 1)et

 B Selected Answers 517

t2 9
23. y(t) = − t−3
5 5
 
2(2a − 1)
24. y(t) = 1t 1 +
t
t 4
25. y(t) = (10 − t) − 8(1 − 10 ) . Note that y(10) = 0, so the tank is empty
after 10 min.

26. (a) T = 45 min; (b) y(t) = 12 (10 + 2t) − 50(10 + 2t)−1 for 0 ≤ t ≤ 45
so y(45) = 50 − 12 = 49.5 lb. (c) limt→∞ y(t) = 50. Once the tank is
full, the inflow and outflow rates will be equal and the brine in the tank
will stabilize to the concentration of the incoming brine, i.e., .5 lb/gal.
Since the tank holds 100 gal, the total amount present will approach
.5 × 100 = 50 lb.

27. If y(t) is the amount of salt present at time t (measured in pounds), then
y(t) = 80e−.04t , and the concentration c(t) = .8e−.04t lb/gal.

28. a) Differential equation: P 0 (t) + (r/V )P (t) = rc. If P0 denotes the initial
amount of pollutant in the lake, then P (t) = V c + (P0 − V c)e−(r/V )t .
The limiting concentration is c.
b) (i) t1/2 = (V /r) ln 2; (ii) t1/10 = (V /r) ln 10
c) Lake Erie: t1/2 = 1.82 years, t1/10 = 6.05 years, Lake Ontario: t1/2 =
5.43 years, t1/10 = 18.06 years

29. a) 10 minutes
b) 1600/3 grams

30. 1 − e−1 grams/liter

Section 2.3
t2
2. y1 (t) = 1 − t +
2
t3
y2 (t) = 1 − t + t2 −
6
2 t3 t4
y3 (t) = 1 − t + t − +
3 4!
t2
3. y1 (t) =
2
t2 t5
y2 (t) = +
2 20
t2 t5 t8 t11
y3 (t) = + + +
2 20 160 4400
4. Unique solution
5. Not guaranteed unique
6. Unique solution
518 B Selected Answers

7. Unique solution
8. Not guaranteed unique
9. a) y(t) = t + ct2
b) Every solution satisfies y(0) = 0. There is no contradiction to Theorem
2
2 since, in normal form, the equation is y 0 = y − 1 = F (t, y) and
t
F (t, y) is not continuous for t = 0.
10. a) F (t, y) = y 2 so both F (t, y) = y 2 and Fy (t, y) = 2y are continuous for
any (t0 , y0 ). Hence Theorem 2 applies.
1
b) y(t) = 0 is defined for all t; y(t) = is only defined on (−∞, 1).
1−t
11. No. Both y1 (t) and y2 (t) would be solutions to the initial value problem
y 0 = F (t, y), y(0) = 0. If F (t, y) and Fy (t, y) are both continuous near
(0, 0), then the initial value problem would have a unique solution by
Theorem 2.
3
12. There is no contraction to Theorem 2 since, in the normal form y 0 = y =
t
F (t, y) has a discontinuous F (t, y) near (0, 0).

Section 2.4
1
2. ty + y 2 − t2 = c
2
3. Not Exact
4. ty 2 + t3 = c
5. Not Exact
6. t2 y + y 3 = 2
7. (y − t2 )2 − 2t4 = c
1 c
8. y = t2 −
3 t
9. y 4 = 4ty + c
10. b+c=0
11. y = (1 − t)−1
2
12. y 2 (tl2 √
+ 1 + et ) = 1
13. y = (c 1 − t2 − 5)−1
2
14. y 2 = (1 + cet )−1
15. y 2 = (t + 12 + ce2t )−1
√
16. y = − 2e2t − et
17. y = 2t2 + ct−2
18. y = (1 − ln t)−1
1 2
19. y(t) = e 2 (t −1) + 1
20. t2 y + y 3 = c 
 t 
21. y = (ln   + c)2
t + 1
 t − 1 1/4
 
22. y = c  
t + 3
 B Selected Answers 519

sin t + t2 = c
23. t sin y + y  
t 1 2
24. y = t − 2t + ln |t| + c
t−1 2

Section 3.1
1. 4.
L {3t + 1} (s) L te−3t  (s)
∞ ∞
=
(3t + 1)e−st dt =
e−st te−3t dt
0 0
∞ ∞ ∞
= 3
te −st
dt +
e −st
dt =
te−(s+3)t | dt
0 0 0

t −st −1 −st te
∞ ∞ ∞−(s+3)t
1 ∞ 1
=3 e +
e−st dt + e = |∞
0 +
e−(s+3)t dt

 −s 0 s 0 s −(s
0
+ 3) s+3 0
1 −1 ∞
1 1
=3 e−st + =
(s + 3)2
.
 s   s  0  s
3 1 5
5. s−2
= 2 + .
s s
3
6. s+7 − s144
2.
2 5 4
7. − +
L 5t − 9et  (s)
s3 s2 s
6 2 1 1
∞ 8. s4
+ s3
+ s2
+ s
t
=
e −st
(5t − ie dt 1 7
0 9. s+3
+ s+4
∞ ∞
= 5
te−st dt − 9
e−st et dt 10. 1
s+3
+ 7
(s+4)2
0 0
−t −st ∞ 1 ∞ −st ∞ s+2
=5 e |0 +
e dt − 9 11.

es−(s−1)t
2 +4 dt
s s 0  0
1 s−1
1 1 12. (s−1)2 − (s−1) 2 +4
= 5 0+ 2 −9
 s  s − 1 s+ 1
13. (s+ 1 )32 +6
5 9 3
= 2 − .
s s−1
14. s23 + s−2 2 1
+ s−4
3. 15. 5s−6
+ 4
s2 +4 s

L e2t − 3e−t  (s) 16. 8s−18

∞ s2 +4
=
e−st (e2t − 3e−t ) dt 12s2 −16
0 17. (s2 +4)3
∞ ∞
=
e−st e2t dt − 3
e−st e−t dt 18. b−a
(s+b)(s+a)
0 0
∞ ∞
s2 +2b2
=
e−(s−2)t dt − 3
e−(s+1)t dt 19. s(s2 +4b2 )
0 0
1 3 2b2
= − . 20. s(s2 +4b2 )
; We use the identity
s−2 s+1 2
sin bt = 12 (1 − cos 2θ).
520 B Selected Answers

b b
21. s2 +4b2
; We use the identity sin 2θ = 23. s2 −b2
2 sin θ cos θ.
s
22. s2 −b2

Section 3.2
1. yes.
The s − 2 -chain
2. yes; the coefficient of e−2t is 0.
10s−2 6
18. (s+1)(s−2) (s−2)
3. no.
4
(s+1)
4. yes, sin2 t + cos2 t = 1

5. yes; cosh 4t = 21 e4t + 12 e−4t .

6. no; the constant function 1 is not in The s − 5 -chain

the given set.
1 1/7
19. (s+2)(s−5) (s−5)
7. no, te3t is not a scalar multiple of e3t
−1/7
(s+2)
8. R

9. PR

10. R: a polynomial may be viewed as The s + 3 -chain

rational (with denominator 1)
5s+9 3/2
20. (s−1)(s+3) s+3
11. R
7/2
s−1
12. R

13. NR: the denominator is not a poly-

nomial
The s − 1 -chain
14. NR
3s+1 2
21. (s−1)(s2 +1) s−1
15. PR
−2s+1
16. NR s2 +1

The s − 1 -chain
The s + 1 -chain
5s+10 3
17. (s−1)(s+4) s−1 3s2 −s+6 2
22. (s+1)(s2 +4) (s+1)
2
s+4 s−2
s2 +4
 B Selected Answers 521

The s + 3 -chain The s − 5 -chain

s2 +s−3 3 1 −1
(s+3)3 (s+3)3 (s−5)5 (s−6) (s−5)5

s−2 −5 1 −1
23. (s+3)2 (s+3)2 (s−5)4 (s−6) (s−5)4

1 1 1 −1
s+3 s+3 27. (s−5)3 (s−6) (s−5)3

1 −1
0 (s−5)2 (s−6) (s−5)2

1 −1
(s−5)(s−6) s−5

1
s−6

The s + 2 -chain

5s2 −3s+10 3/2 7/2

(s+1)(s+2)2
−36
(s+2)2
28. s+3
+ s−1
24. 13/8 5/8
5s+23 −13 29. −
(s+1)(s+2) s+2 s−5 s+3

−1 1
18 30. s−1
+ s−2
s+1
23 37
31. 12(s−5)
+ 12(s+7)

1 2
32. s
+ s+1

25 9
33. 8(s−7)
− 8(s+1)
The s + 1 -chain
25 9 15
s −1
34. 2(s−3)
+ 2(s−1)
− s−2
(s+2)2 (s+1)2 (s+1)2
1 1 1
25. 35. 2(s+5)
− 2(s−1)
+ s−2
s+4 3
(s+2)2 (s+1) s+1
1 2 1
36. (s−1)3
+ (s−1)2
+ s−1
−3s−8
(s+2)2
7
37. (s+4)4

3 1
38. (s−3)3
+ (s−3)2

3 1 5
39. (s+3)3
+ s+3
− (s+3)2
The s − 1 -chain
−36 13 18
40. (s+2)2
− s+2
+ s+1
16s 4
(s−1)3 (s−3)2 (s−1)3 
1 5 21 3 5
41. + (s+1)2
+ (s−5)2
−
−4s+36 8
54 s−5 s+1

26. (s−1)2 (s−3)2 (s−1)2
9 1 1 1 1
42. s3
− s2
+ 9s
− 9 s+1
−8s+36 7
(s−1)(s−3)2 (s−1)
−2 3 1 3
43. (s+2)2
− s+2
− (s+1)2
+ s+1
−7s+27
(s−3)2
4 4 1 4
44. (s+2)2
+ s+2
+ (s+1)2
− s+1
522 B Selected Answers

12 −14 15 −16 1 2 1 1
45. (s−3)3
+ (s−3)2
+ s−3
+ s−2
+ s−1
51. s−1
− s+i
− s−i

2 5 3 5 6 2+i
46. (s−2)3
+ s−2
+ (s−3)2
− s−3 52. s+1
+ −2+i
s+2i
− s−2i

4 8 7 6 7
47. (s−1)3
+ (s−1) 2 + s−1 + (s−3)2 − s−3 53. 1 1
− 25 1
− 5
+ 4
6 s−5 6 s+1 s2 s

6 1−2i 1+2i
48. 2s+4
= 2s+4
= 1−2i
+ 54. s−3
+ s+1+2i
+ s+1−2i
s2 −4s+8 (s−2)2 +22 s−(2+2i)
1+2i
1+2i 1−2i −1−i −1+i
s−(2−2i) 55. s+i
+ s−i
+ s+2i
+ s−2i
2+3s 2+3s 1−i
49. s2 +6s+13
= (s+3)2 +32
= s+3+2i
+ i i 1+i 1−i
1+i
56. s+2i
− s−2i
+ (s+2i)2
+ (s−2i)2
s+3−2i
−3i 3 2i 3i 3
7+2s 7+2s 1+3i 57. − (s−i) 2 + (s−i) 3 + s+i − (s+i)2 −
50. s2 +4s+29
= (s+2)2 +52
= s+2+5i
+ s−i
2i
1−3i (s+i)3
s+2−5i

Section 3.3

The s2 + 1 -chain The s2 + 3 -chain

1 1 8s+8s2 12−4s
(s2 +1)2 (s2 +2) (s2 +1)2 (s+ 3)3 (s2 +1) (s2 +3)3
1.
−1 −1 4(s−1) 2−2s
(s2 +1)(s2 +2) (s2 +1) 3. (s2 +3)2 (s2 +1) (s2 +3)2

1 2(s−1) 1−s
s2 +2 (s2 +3)(s2 +1) s2 +3

s−1
s2 +1

The s2 + 4 -chain

4s4 32
(s2 +4)4 (s2 +6) (s2 +4)4

4s2 −48
The s2 + 2 -chain (s2 +4)3 (s2 +6)
−32
(s2 +4)3
4.
s3 −2s 36 18
(s2 +2)2 (s2 +3) (s2 +2)2 (s2 +4)2 (s2 +6) (s2 +4)2
2.
3s 3s −18 −9
(s2 +2)(s2 +3) s+ 2 (s2 +4)(s2 +6) s2 +4

−3s 9
s2 +3 s2 +6
 B Selected Answers 523

1 3 1−3s
7. +
The s2 + 2s + 2 -chain 10 s−3 s2 +1 

2s+2 s−1 1
1 1 8. (s2 +1)2
+ s2 +1
− s+1
(s2 +2s+2)2 (s2 +2s+3)2 (s2 +2s+2)2
5.
−(s2 +2s+4) −2 2 6−2s
(s2 +2s+2)(s2 +2s+3)2 s2 +2s+2
9. s−3
+ (s−3)2 +1

2s2 +4s+7 9−3s 3

(s2 +2s+3)2 10. ((s−2)2 +9)
+ s−1

−5s+15 −s+3 1
11. (s2 −4s+8)2
+ s2 −4s+8
+ s−1

The s2 + 2s + 2 -chain 12. −s s

− s2 +6s+10 3
− (s+3) 1
(s2 +6s+10)2 2 + s+3

5s−5 5s+5
s+1
(s2 +2s+2)2 (s2 +4s+5) (s2 +2s+2)2 13. (s2 +4s+6)2
+ s22s+2
+4s+6
+ s+1
(s2 +4s+5)2
−
6. 2s+2
−5s−15 3s−1
s2 +4s+5
(s2 +2s+2)(s2 +4s+5) s2 +2s+2

−3s−5 14. −5
(s2 +5)3
+ (s211
+5)2
− s217
+5
6
+ (s2 +6) 2 +
s2 +4s+5
17
s2 +6

Section 3.4
1. −5 14. e2t − e−4t

2. 3e4t 15. 3e3t − 2e2t

−1 3 2t
3. 4e2it 16. 6
t e + 32 t2 e2t + 2te2t

4. −3e−t e−it 17. 4e5t − 2et

5. 3t − 2t2 18. tet + et + te−t − e−t

3t 9t2
6. 2e− 2 19. 2
− 3t + 1 − e−3t

7. 3 cos 2t 20. 4te2t − e2t + cos 2t − sin 2t

8. −2
cos 023 t 21. −2 cos 3t + sin 3t + 2et
3

√ 22. 3 − 6t + e2t − 4e−t

9. √2 sin 3t
3
23. (3t2 + 14t + 22)e3t + (8t − 22)e4t
10. cos 023 t + 023 sin 032 t
24. Yes.
−3t
11. te 25. No; t−2 is not a polynomial.
t
12. −11te−3t + 2e−3t 26. Yes; et
= te−t .
−11 2 −3t 1
13. 2
t e + 2te−3t 27. No; t
is not a polynomial.
524 B Selected Answers
√
π 2
28. Yes; t sin(4t − 4
) = t( 2
sin 4t − 31. Yes.
√
2
cos 4t). 1
2 32. No; t 2 is not a polynomial.
sin 2t
29. Yes; (t + et )2 = t2 + 2tet + e2t . 33. Yes; e2t
= sin 2te−2t .

34. No.
30. No.

Section 3.5
1
1. 2e−t cos 2t − e−t sin 2t 10. 6
sin 3t − 21 t cos 3t

2. e−3t sin t 11. 2te−2t cos t + (t − 2)e−2t sin t

3. e4t cos t + 3e4t sin t 12. −4te3t cos t + (t + 4)e3t sin t

√ √ √
4. 2e2t cos 8 t + 8 e2t sin 8 t 13. 4te−4t cos t + (t − 4)e−4t sin t

5. et cos 3t 14. 1
8
(te−t sin t − t2 e−t cos t)

6. e3t cos 2t − e3t sin 2t 15. 1

(3 − 4t2 )et sin 2t − 6tet cos 2t 
256

16. (−3t2 + t + 9)e3t sin t − (t2 +
5t  5t  9t)e3t cos t
7. L 1−cos
t
= 5L 1−cos
5t
=
2
1 (s/5) 1 s2
2
ln (s/5)2 +1
= 2
ln s2 +25
. 17. 1
(−t3 + 3t)e4t cos t − 3t2 e4t cos t 
48

8. L {Ei(6t)} = 61 (6L {Ei(6t)}) = 1
18. 128 (4t2 + 2t − 3)e−2t sin 2t − (4t2 − 6t)e−2t cos 2t 
1 ln((s/6)+1) 
6 s/6
= ln(s+6)−ln
s
6
1
19. 384 (t4 − 45t2 + 105) sin t + (10t3 − 105t) cos t 

9. 2t sin 2t
1
20. 384 (t4 − 15t2 ) cos t − (6t3 − 15t) sin t 


Section 3.6
1. Y (s) = 2
s−4
and y(t) = 2e4t 5. Y (s) = s+22
+ 3
(s−1)(s+2)
and y(t) =
−2t t
e +e
2 3
2. Y (s) = s−4 + s(s−4)
and y(t) =
11 4t 3 6. Y (s) = 1
(s+2)3
and y(t) = 21 t2 e−2t
4
e − 4
162 1
2 16
7. Y (s) = s3 (s+9)
+ s+9 and y(t) =
3. Y (s) = s−4 + s2 (s−4)
and y(t) = 7 −9t 2 2
4t 9
e + 9
− 2t + 9t
3e − 1 − 4t
s
8. Y (s) = (s2 +1)(s−3) and y(t) =
1 4t
4. Y (s) = (s−4)2
and y(t) = te 3 3t 3 1
e − 10 cos t + 10 sin t
10
 B Selected Answers 525

1 50 2 50
9. Y (s) = s−3 + (s−3)(s 2 +1) and y(t) = 21. Y (s) = (s+3) 2 + (s2 +1)(s+3)2 =
3t 3 7
6e − 5 cos t − 15 sin t s+3
+ s2 +1 + (s+3)2 and y(t) =
−3s+4

3e−3t + 7te−3t − 3 cos t + 4 sin t

10. Y (s) = (s2 +1)4s2 (s+1) + s+1 1
=
s−1
s 1 s 1
2 (s+ 1)2 +2 (s2 +1)2 +2 s2 +1 − s2 +1 and 22. Y (s) = s2 +25
and y(t) = −1
5
sin 5t +
y(t) = t sin t − t cos t + cos t cos 5t
s+12 1 −4t
2 1 23. Y (s) = 2(s+4)2
and y(t) = e +
11. Y (s) = (s−3) 2 +1 + s−3
and y(t) = 2

3t 3t 4te−4t
2e sin t + e
s+1
25. Y (s) = and y(t) =
12. Y (s) = ((s−2)230 2
− s−1 and √ −t
√s2 +s+1 −t √
3
+9)(s−1)
e 2 sin 2 + e 2 cos 23t
3t
y(t) = −3e cos 3t + e sin 3t + et
2t 2t 3
 
s+4
12 26. Y (s) = (s+2)2
+ (s2 +4)64s
2 (s+2)2 =
13. Y (s) = (s−2)(s+1)(s+2)
+ 16 s
s+4 1 1 1 + s2 +4 and y(t) =
(s+1)(s+2)
= s−2
+ (s+2)
+ s+1
and (s2 +4)2
−2t cos 2t + cos 2t + sin 2t
y(t) = e2t + e−2t − e−t
27. Y (s) = s21+4 + (s2 +9)(s3
2 +4) =
−s+4 25
14. Y (s) = (s+1)(s−5) + s2 (s+1)(s−5) = 8 1 3 1 4
− 5 s2 +9 and y(t) = 5 sin 2t −
5 s2 +4
5 4
−5
s2
− s+1
+ s
and y(t) = 4−5t−5e −t 1
sin 3t
5

15. Y (s) = s2s+1 8

2 +4 + s(s2 +4) =
1
s2 +4
+ 2
s
28. Y (s) = s12 + s4 (s−1)
1 1
= s−1 − 1s −
1
and y(t) = 2 sin 2t + 2
1
s3
− s4 and y(t) = e − 1 − 2 t − 16 t3
1 t 1 2

1
16. Y (s) = (s+2) 1
and y(t) = 29. Y (s) = (s−1)(s12 +1)s2 = 21 s−1 1
+
2 + (s+2)3 1 s+1 1 1 1 t
te −2t 1 2 −2t
+ 2t e 2
2 s +1
− s2 − s
and y(t) = 2
e +
1 1
2
sin t + 2
cos t − t − 1
1 4s 1
17. Y (s) = (s+2) 2 + (s2 +4)(s+2)2 = s2 +4 2 2
1
and y(t) = 2 sin 2t 30. Y (s) = ss3 +s
−s
+ s6(s −1)
3 (s3 −s) =
1
s−1
+ 6
s4
and y(t) = et + t3
2s−3 4
18. Y (s) = (s−1)(s−2) + s(s−1)(s−2) = 3
s 1 s
2
+ 3
− 3
and y(t) = 2+3e 2t
−3e t 31. Y (s) = (s−1)(s+1)(s2 +1) = 2 s2 +1
+
s s−3 s−1 1 1 1 1 1
4 s−1
+ 4 s+1
and y(t) = 2
cos t +
1 t 1 −t
−3s+9
19. Y (s) = (s−1)(s−2) + (s−1)21(s−2) = 4
e + 4
e
−7
+ s−2 − (s−1)2 and y(t) = −tet +
4 1
s−1 32. y(t) = 0
4e − 7et
2t

33. y(t) = −3e−t + e−3t + 2

26 1
20. Y (s) = (s2 +4)(s+3)(s−1) = −1
2 s+3
+ 1 1 t 1 −t
13 1 1 −4s−14 13 t 34. y(t) = − 10 cos 3t + e + e
10 s−1
+ 5 s2 +4 and y(t) = 20 e − 20 20
1 −3t 4
2
e − 5
cos 2t − 57 sin 2t 35. y(t) = t sin t − t2 cos t

Section 3.7
526 B Selected Answers

1. Bq,C = e4t  15. Bq = e4t 

2. Bq,C = e−6t  16. Bq = e−6t 

3. Bq,C = e4t , e−t  17. Bq = e−4t , e−t 

4. Bq,C = e3t , te3t  18. Bq = e3t , te3t 

5. Bq,C = e2t , e7t  19. Bq = e2t , e7t 

6. Bq,C = e3t , e−2t  20. Bq = e3t , e−2t 
7. Bq,C = e3it , e−3it  21. Bq = {cos 3t, sin 3t}
5t 
8. Bq,C = e , te 22. Bq = e5t , te5t 
5t

9. Bq,C = 1e(−2+3i)t , e(−2−3i)t 2 23. Bq = e−2t cos 3t, e−2t sin 3t 

24. Bq = e−3t , e−6t 

10. Bq,C = e−3t , e−6t 
25. Bq = et , cos t, sin t 
11. Bq,C = 1, et , e−t 
26. Bq = et , e−t , cos t, sin t 
12. Bq,C = et , e−t , eit , e−it 
27. Bq = et , tet , t2 et , e−7t , te−7t 
13. Bq,C = et , tet , t2 et , e−7t , te−7t 
28. Bq = cos t, t cos t, t2 cos t, sin t, t sin t, t2 sin t 
14. Bq,C = eit , teit , t2 eit , e−it , te−it , t2 e−it 

Section 3.8
t3
1. 6
12. 1
a2 +b2
(aeat − a cos bt + b sin bt)
t5
2. 20
b sin at−a
2 2
sin bt
if b 6= a
b −a
13. 
3. 3(1 − cos t) sin at−at cos at if b = a
2a
4t
7e
a cos at−a
−12t−7
4. 16
cos bt
if b 6= a
b2 −a2
14. 
5. 1
13
(2e3t − 2 cos 2t − 3 sin 2t) 1 t sin at if b = a
2
1
6. (1 − cos 2t + sin 2t)
2 a sin at−b sin bt
if b 6= a
a2 −b2
7. 1
(18t2 − 6t + 1 − e −6t
) 15. 
108 1 (at cos at + sin at) if b = a
2a
1
8. (2 sin 2t − sin t)
3 16. The key is to recognize the inte-
9. 1
(e2t − e−4t ) gral defining f (t) as the convolu-
6
tion integral of two functions. Thus
10. tn+2 f (t) = (cos 2t) ∗ t so that F (s) =
(n+2)(n+1)
L {(cos 2t) ∗ t} = L {cos 2t} L {t} =
s 1
11. 1
a2 +b2
(beat − b cos bt − a sin bt) s2 +4 s2
= s(s21+4) .
 B Selected Answers 527

17. F (s) = 4
s3 (s2 +4)
29. 1
17
(4e4t − 4 cos t + sin t)
6 eat −ebt
18. F (s) = s4 (s+3)
30. a−b

6 at−sin at
19. F (s) = s4 (s+3)
31. a3

s t
20. F (s) = (s2 +25)(s−4) 32.
g(τ )e−2(t−τ ) dτ
0
2s
21. F (s) = (s2 +4)(s2 +1) t √
33.
g(τ ) cos 2(t − τ ) dτ
4
22. F (s) = (s2 +4)2
0

1 t √
23. 1
6
(e2t − e−4t ) 34. √
sin 3(t − τ ) f (τ ) dτ
3 0
24. 1
4
(−et + e5t ) t
35.
(t − τ )e−2(t−τ ) f (τ ) dτ
1
25. 2
(sin t − t cos t) 0

t
1
26. 2
t sin t 36.
e−(t−τ ) sin 2(t − τ ) f (τ ) dτ
0
27. 1
216
(−e−6t + 1 − 6t + 18t2 ) t 
37.
e−2(t−τ ) − e−3(t−τ ) f (τ ) dτ
28. 1
13
(2e3t − 2 cos 2t − 3 sin 2t) 0 

Section 4.1

1. no 10. yes; (2D 2 −12D +18)y = 0, order 2,

c(s) = 2s2 − 12s + 18, homogeneous
2. yes; (D 3 −3D)y = et , order 3, c(s) =
s3 − 3s, nonhomogeneous 11. (a) 6et
(b) 0
3. yes; (D 3 + D + 4)y = 0, order 3, (c) sin t − 3 cos t
c(s) = s3 + s + 4, homogeneous
12. (a) 0
4. no (b) 2 sin t
5. no (c) −e2t

6. yes; (D 2 + 2D)y = 0, order 2, c(s) = 13. (a) −4e−2t

s2 + 2s, nonhomogeneous (b) 0
(c) sec2 t − 2 tan t
7. yes; (D 2 − 7D + 10)y = 0, order 2,
c(s) = s2 − 7s + 10, homogeneous 14. (a) 0
(b) 0
8. yes; (D + 8)y = t, order 1, c(s) = (c) 1
s + 8, nonhomogeneous
15. (a) 0
9. yes; D 2 y = −2+cos t, order 2, c(s) = (b) 0
s2 , nonhomogeneous (c) 0
528 B Selected Answers

16. (a) 4e2t 21. y(t) = cos 2t + et − e4t

(b) 0
(c) 0 22. y(t) = te3t + e3t − 2e−2t
17. y(t) = cos 2t + c1 et + c2 e4t 23. y(t) = te2t + e2t − 2e−2t + 3
3t 3t −2t
18. y(t) = te + c1 e + c2 e
24. y(t) = sin t − e2t − e−2t + sin 2t +
19. y(t) = te 2t
+ c1 e 2t
+ c2 e −2t
+ c3 2 cos 2t
2t
20. y(t) = sin t + c1 e + c2 e−2t +
c2 sin 2t + c4 cos 2t

Section 4.2

1. y = et 8. z(t) = 3−2it it
4
e − 34 e−it

2. y = e3t + e4t 9. z = 18 (cos t + 3i sin t) + 87 e3it

2
3. y = et ( t2 − t + 1) 10. z = 3 − it − 2e−it
4. y = t + cos t

7. z(t) = (t + i)eit

Section 4.3
1. y(t) = ce6t 12. y(t) = c1 e−4t cos 3t + c2 e−4t sin 3t

2. y(t) = ce−4t 13. y(t) = c1 e−5t + c2 te−5t

3. y(t) = c1 e2t + c2 e−t 14. y(t) = c1 e7t + c2 e−3t

√
4. y(t) = c1 e−4t + c2 e3t 15. y(t) = c1 e−t + c2 e− 2 t cos
1 3
t +
√ 2
1
5. y(t) = c1 e −4t
+ c2 e −6t c3 e− 2 t sin 23 t

6. y(t) = c1 e6t + c2 e−2t 16. y(t) = (c1 + c2 t + c3 t2 )e2t

7. y(t) = c1 e−4t + c2 te−4t 17. y(t) = c1 et +c2 e−t +c3 sin t+c4 cos t

8. y(t) = c1 e5t + c2 e−2t 18. y(t) = c1 + c2 e−t + c3 te−t

9. y(t) = c1 e−t cos 2t + c2 e−t sin 2t 19. y(t) = c1 et + c2 e−t + c3 e2t + c4 e−2t
et −e−t
10. y(t) = c1 e3t + c2 te3t 20. y = 2

11. y(t) = c1 e−9t + c2 e−4t 21. y = 2e5t + 3e−2t

 B Selected Answers 529

22. y = te5t 24. y = 2et − e−t + te−t

23. y = e−2t cos 3t − e−2t sin 3t 25. y = cos t + 2 sin t + et − 3e−t

Section 4.4
1. yp (t) = Ce4t 29. y = te2t − 52 e2t + Ae−3t + Bte−3t
25 3 −3t
30. y = 6
t e + Ae−3t + Bte−3t

31. y = 14 te−3t sin(2t) + Ae−3t cos(2t) +

Be−3t sin(2t)

1 32. y = −t2 e4t cos(3t) + te4t sin(3t) +

15. y = + Ae−6t
2 Ae4t cos(3t) + Be4t sin 3t
16. y = e3t + Ae−t
33. y = 12 tet + Ae−t + Bet + C
17. y = 2te4t + Ae4t
34. y = −t(sin t+cos t)+Aet +B cos t+
18. y = sin t − cos t + Aet C sin t

19. y = 16 e2t + Aet + Be−4t

1
35. y = 12 te2t + Aet + Be−t + Ce2t +
De−2t
20. y = −te−2t + Ae−2t + Be5t
36. y = 4t (et − e−t ) + Aet + Be−t +
21. y = 14 et + Ae−t + Bte−t C cos t + D sin t

22. y = 2 + Ae−t + Be−2t −1 3t 10 6t 135 −t

37. y = 12
e + 21
e + 84
e

23. y = 12 e−3t + Ae−2t cos t + Be−2t sin t 38. y = 2e−t − 2e−t cos 2t + 4e−t sin 2t
24. y = 1
4
+ 15 et + A cos(2t) + B sin(2t) 39. y = 2e2t − 2 cos t − 4 sin t
25. y = −t2 − 2 + Aet + Be−t 40. y = −1 −2t
2
e + e2t + 2t − 1
2

26. y = et + Ae2t + Bte2t

27. y = 12 t2 e2t + Ae2t + Bte2t

28. y = −t cos t + A sin t + B cos t

Section 4.5
2 7t
1. y = 11
e + Ae−4t 3. y = 19 te4t − 1 4t
81
e + Ae−5t

2. te3t + Ae3t 4. y = 13 t3 e2t + Ae2t

530 B Selected Answers

5. y = 1 −6t
32
e + Ae2t + Be−2t 14. y = −t2 e4t cos(3t) + te4t sin(3t) +
Ae4t cos(3t) + Be4t sin 3t
6. y = −et + Ae−3t + Be5t
15. y = 16 t3 e2t + Ae2t + Bte2t
7. y = 15 te−2t + Ae−2t + Be−3t
1
16. y = sin t + Ae−t + Bte−t
8. y = 2 + Ae−t + Be−2t 2

9. y = −te−4t + Ae2t + Be−4t 17. y = 12 tet sin t

1 1
3
10. y = − 130 cos t − 11
sin t 18. 3
t sin t
+ 9
cos t + A + B cos(2t) +
130
C sin(2t)
11. y = −t2 + A + Be−3t + Ce3t
19. y = 12 tet + Ae−t + Bet + C
2t 2 2t −3t −3t
12. y = te − 5
e + Ae + Bte
20. y = 4t (et − e−t ) + Aet + Be−t +
4t 3 4t
13. y = −te − 10
e + Ae −t
+ Be6t C cos t + D sin t
 B Selected Answers 531

Section 6.1

1. no L(e−t ) = (t2 − t − 1)e−t

L(cos 2t) = (−4t2 − 1) cos 2t −
2. yes, yes, homogeneous 2t sin 2t
3. yes, yes, nonhomogeneous 19. L(ert ) = a(ert )00 + b(ert )0 + cert =
ar 2 ert +brert +cert = (ar 2 +br+c)ert
4. no
−3
20. C = 4
5. yes, yes, nonhomogeneous
1
21. C1 = −3
4
and C2 = 2
6. yes, yes, nonhomogeneous
22. no
7. no
23. yes, C = 1.
8. yes, yes, nonhomogeneous
24. Parts (1) and (2) are done by com-
9. yes, no, homogeneous puting y 00 + y where y(t) = t2 − 2,
y(t) = cos t, or y(t) = sin t. Then
10. no by Theorem 3, every function of the
11. yes, no, homogeneous form y(t) = t2 − 2 + c1 cos t + c2 sin t
is a solution to y 00 + y = t2 , where
12. yes, no, homogeneous c1 and c2 are constants. If we want a
solution to L(y) = t2 with y(0) = a
13. L(1) = 100 + 1 = 1 and y 0 (0) = b, then we need to solve
L(t) = t00 + t = t for c1 and c2 :
L(e−t ) = (e−t )00 + e−t = 2e−t
L(cos 2t) = (cos 2t)00 + cos 2t = a = y(0) = −2 + c1
−4 cos 2t + cos 2t = −3 cos 2t. b = y 0 (0) = c2
14. L(1) = 1 These equations give c1 = a + 2,
L(t) = t c2 = b. Particular choices of a and b
L(e−t ) = (t + 1)e−t give the answers for Part (3).
L(cos 2t) = (−4t + 1) cos 2t
(3)a. y = 12 et + 2e2t − 23 e3t
15. L(1) = −3
L(t) = 1 − 3t (3)b. y = 12 et − 2e2t + 32 e3t
L(e−t ) = −2e−t (3)c. y = 12 et − 7e2t + 11 3t
2
e
L(cos 2t) = −11 cos 2t − 2 sin 2t (3)d. y = 12 et + (−1 + 3a − b)e2t +
1
− 2a + b  e3t
16. L(1) = 5 2
L(t) = 5t + 6 25.(3)a. y = 21 et + 2e2t − 23 e3t
L(e−t ) = 0 1 t
(3)b. e − 2e2t + 32 e3t
L(cos 2t) = cos 2t − 12 sin 2t 2
1 t
(3)c. 2
e + 11
2
e2t − 7e3t
17. L(1) = −4 1 t
(3)d. 2
e + (3a − b − 1)e2t + (b − 2a +
L(t) = −4t 1
2
)e3t
L(e−t ) = −3e−t
L(cos 2t) = −8 cos 2t 26.(3)a. y = 61 t5 + 10 2
3
t − 52 t3
18. L(1) = −1 (3)b. y = 16 t5 − 2 2
3
t + 12 t3
L(t) = 0 (3)c. y = 16 t5 − 17 2
3
t + 92 t3
532 B Selected Answers

(3)d. y = 16 t5 + 13 + 3a − b  t2 + 30. (−∞, ∞)


− 1 − 2a + b  t3
 2 31. (3, ∞)
27. Write the equation in the standard
form: 32. (−∞, −2), (−2, 0), (0 2), (2, ∞)
3 1
y 00 + y 0 − 2 y = t2 .
t t 33. The initial condition occurs at t = 0
3 1 which is precisely where a2 (t) = t2
Then a1 (t) = , a2 (t) = − 2 , and has a zero. Theorem 6 does not ap-
t t
f (t) = t2 . These three functions ply.
are all continuous on the intervals
34. In this case y(t0 ) = 0 and y 0 (t0 ) = 0.
(0, ∞) and (−∞, 0). Thus, Theorem
The function y(t) = 0, t ∈ I is a so-
6 shows that if t0 ∈ (0, ∞) then the
lution to the initial value problem.
unique solution is also defined on the
By the uniqueness part of Theorem
interval (0, ∞), and if t0 ∈ (−∞, 0),
6 y = 0 is the only solution.
then the unique solution is defined
on (−∞, 0). 35. The assumptions say that y1 (t0 ) =
28. Maximal intervals are (−∞, −1), y2 (t0 ) and y10 (t0 ) = y20 (t0 ). Both y1
(−1, 1), (1, ∞) and y2 therefore satisfies the same
initial conditions. By the uniqueness
29. (kπ, (k + 1)π) where k ∈ Z part of Theorem 6 y1 = y2 .

Section 6.2
1. dependent; 2t and 5t are multiples of 9. 1. Suppose at3 + b t3 = 0 on
each other. (−∞, ∞). Then for t = 1 and
t = −1 we get
2. independent
a+b = 0
3. independent −a + b = 0.
2 1 2t
4. dependent; e t + 1 = e e and These equations imply a = b =
e2t−3 = e−3 e2t , they are multiples 0. So y1 and y2 are linearly in-
of each other. dependent.
2. Observe that y10 (t) = 3t2 and
5. independent −3t2 if t < 0
y20 (t) =  2 If
6. dependent; ln t2 = 2 ln t and ln t5 = 3t if t ≥ 0.
5 ln t, they are multiples of each t < 0 then w(y1 , y2 )(t) =
other. t3 −t3
2 2 = 0. If t ≥ 0 then
3t −3t 
7. dependent; sin 2t = 2 sin t cos t, they t3 t3
w(y1 , y2 )(t) = = 0.
are multiples of each other. 2
3t 3t 
2

It follows that the Wronskian is

8. dependent; cosh t = 12 (et + e−t ) and zero for all t ∈ (−∞, ∞).
3et (1 + e−2t ) = 3(et + e−t ), they are 3. The condition that the coeffi-
multiples of each other. cient function a2 (t) be nonzero
 B Selected Answers 533

in Theorem 4 and Proposition 4. Consider the cases t < 0 and

6 is essential. Here the coeffi- t ≥ 0. The verification is then
cient function, t2 , of y 00 is zero at straightforward.
t = 0, so Proposition 6 does not
apply on (−∞, ∞). The largest 5. Again the condition that
open intervals on which t2 is the coefficient function a2 (t)
nonzero are (−∞, 0) and (0, ∞). be nonzero is essential. The
On each of these intervals y1 and Uniqueness and Existence theo-
y2 are linearly dependent. rem does not apply.

Section 6.3

1. The indicial polynomial is q(s) = 6. The indicial polynomial is q(s) =

s2 + s − 2 = (s + 2)(s − 1). There s2 −4s−21 = (s−7)(s+3). The roots
are two distinct roots 1 and −2. The are 7 and −3. The fundamental set
fundamental set is t, t−2 . The gen- is t7 , t−3 . The general solution is
eral solution is y(t) = c1 t + c2 t−2 . y(t) = c1 t7 + c2 t−3 .

2. The indicial polynomial is q(s) = 8. The indicial polynomial is q(s) =

2s2 − 7s + 3 = (2s − 1)(s − 3). s2 − s + 1 = s2 − s +√1/4 +
There are two distinct roots 12 and 3/4 = (s − 1/2)2 + ( 3/2)√2 .
3. The fundamental set is 1t 2 , t3 2.
1
There are two complex roots, 1+i2 3
√
1
The general solution is y(t) = c1 t 2 + and 1−i2 3 . The fundamental set
is 1t 2 sin 3 t, t 2 cos 3 t 2. The gen-
√ √
c2 t3 . 1 1
2 2
1 √
3
3. The indicial polynomial is q(s) = eral solution is y(t) = c1 t 2 sin 2
t+
√
9s2 − 6s + 1 = (3s − 1)2 . There is 1
c2 t 2 cos 23 t.
one root, 1/3, with multiplicity 2.
The fundamental set is 1t 3 , t 3 ln t 2.
1 1
9. The indicial polynomial is q(s) =
1 s2 − 4 = (s − 2)(s + 2). There are two
The general solution is y(t) = c1 t + 3
distinct roots, 2 and −2. The funda-
1
c2 t 3 ln t. mental set is t2 , t−2 . The general
solution is y(t) = c1 t2 + c2 t−2 .
4. The indicial √polynomial
√ is q(s) =
s2 − 2 = (s − 2)(s√+ 2). There
√ are 10. The indicial polynomial is q(s) =
two distinct roots 2√and −√ 2. The
s2 + 4. There are two complex roots,
fundamental set is 1t 2 , t− 2 2. The 2i and −2i. The fundamental set
√
general solution is y(t) = c1 t 2
+ is {cos(2 ln t), sin(2 ln t)}. The gen-
√
c2 t − 2
. eral solution is y(t) = c1 cos(2 ln t) +
c2 sin(2 ln t).
5. The indicial polynomial is q(s) =
4s2 − s + 1 = (2s − 1)2 . The root is 12 11. The indicial polynomial is q(s) =
with multiplicity 2. The fundamen- s2 − 4s + 13 = (s − 2)2 + 9.
tal set is 1t 2 , t 2 ln t 2. The general There are two complex roots, 2 +
1 1

1 1
3i and 2 − 3i. The fundamen-
solution is y(t) = c1 t 2 + c2 t 2 ln t. tal set is t2 cos(3 ln t), t2 sin(3 ln t) .
534 B Selected Answers

The general solution is y(t) = 13. y = 2t1/2 − t1/2 ln t

c1 t2 cos(3 ln t) + c2 t2 sin(3 ln t).
14. y = −3 cos(2 ln t) + 2 sin(2 ln t)
1
12. y = (t − t−2 ) 15. No solution is possible.
3

Section 6.4

s−a t2 −t
1. ln e ). The general solution can be
s−b
 2
 written y(t) = Ae−t + B(te−t +
s2 +b2 t2 −t
2. ln s2 +a2 2
e ).

 14. Y 0 (s) = y0 ( 1s − s−1
1
) Then Y (s) =
2 2 
3. s ln ss2+a + 2b tan−1 b
s
− s
+b2   y0 ln s−1 + C. Take C = 0. Hence
2a tan−1 as   t
 y(t) = y0 1−e
t
.
a
4. tan−1 
 s
15. Y 0 (s) = −y0
and Y (s) =
 s2 −1
5. y = at + bet y0 s+1
ln s−1 + C. Take C = 0 Then
2

6. y = a(t + 1) + be−t t
y(t) = y20 e −e
−t
.
t
1 C
7. Y 0 (s)+ s+1 Y (s) = 0 and Y (s) = s+1 16. Y 0 (s) = −y0
and Y (s) =
and y(t) = Ce−t  (s2 −5s+6)

y0 ln s−2 + C. Take C = 0. Then

y0
s−3


2 , Y (s) = , and
−y0
8. Y 0 (s) = (s+2) s+2 e3t −e2t
y(t) = y0 .
y(t) = e .
−2t t

17. Y 0 (s) = −y0
so Y (s) =
9. Y 0 (s) + 4s
s2 +1
Y (s) = s3y 2 +1 . Solving
0 s2 +9

y0 (s3 +3s)+C
−y0
3
−1
(tan (s/3) + C). Since
gives Y (s) = (s2 +1)2 and y(t) = lims→∞ Y (s) = 0 we have C = − π2 .
A(t cos t − sin t) + B(t sin t + cos t). Hence Y (s) = y30 tan−1 ( 3s ) and
2s2 +s−2 y(t) = y30 sint 3t .
10. Y 0 (s) + s(s−1)
= 2(ys(s−1)
0 s+y1 )
Y (s) =
2s2 −2s+1 18. Y 0 (s) = s−y2 +s and hence y(s) =
2s−1 0
y + 1s2 (s−1) y1 y(t) = (−t+
2s2 (s−1) 0 s+1
1 + e ) + (t − 1 + et )y1 The general
t y0 ln s  + C. But C = 0 and
 −t
solution can be written A(t−1)+Be t y(t) = y0 1−et .
y0 19. We use the formula
2 , Y (s) = , and
−y0
11. Y 0 (s) = (s+1) s+1
y(t) = y0 e .
−t
dn n
n dk dn−k
(f (t)g(t)) = f (t)· n−k g(t).
dt n
k=0 
m dt k dt
12. Y 0 (s) + s26s+1 Y (s) = 0,
Y (s) = (s2 C +1) 3 , and y(t) = Observe that
C (3 − t2 ) sin t − 3t cos t  dk −t
 e = (−1)k e−t
dtk
13. Y 0 (s) − s−2
s+1
Y (s) = −sy 0 −y1
(s+1)2
, Y (s) =
2 and
y0 s(s+1)
+3s+3
3
s+2
+ y1 (s+1) 3 , and y(t) =
dn−k n
y0 (e−t + te−t + t2 −t
e ) + y1 (te−t + t = n(n − 1) · · · (k + 1)tk .
2 dtn−k
 B Selected Answers 535

It now follows that 1 (s − 1)n

L {`n ∗ 1} (s) =
s sn+1
s − 1 (s − 1)n
1 t dn −t n = 1−
e (e t )  s  sn+1
n! dtn
(s − 1)n (s − 1)n+1
1 t n dk −t dn−k n
n
= −
= e e t s n+1 sn+2
n! k  dtk dtn−k
k=0  = L {`n } (s) − L {`n+1 } (s).
n
n k −t n(n − 1) · · · (k + 1) The
=e t
(−1) e t k result follows by inversion.
k=0 
k n!
25. We compute the Laplace transform
n
k n t
k
of both sides. We’ll do a piece at a
= (−1)
k  k! time.
k=0 
= `n (t). L {(2n + 1)`n } (s)
(s − 1)n
= (2n + 1) n+1
s
n−1
20. This follows in a similar manner to (s − 1)
= (2n + 1)(s(s − 1)).
the proof of Equation (2) given in sn+2
Theorem 11.
L {−t`n } (s)
21. F (s − 1)n
0
=
 s
n+1 
22. n−1
(s − 1)
= (n + 1 − s).
sn+2
L {`n (at)} (s)
n −nL {`n−1 } (s)
n tk
= (−1)k ak L
(s − 1)n−1
k=0 
k k!  = −n
sn
n
n k k 1 (s − 1)n−1
= (−1) a k+1 = (−ns2 ).
k s sn+2
k=0 
We have written each n−1 so that the
1 n
n
= n+1 (−a)k sn−k common factor is (s−1) sn+2
. The co-
s k=0 
k efficients are
(s − a)n
= . n + 1 − s + (2n + 1)(s(s − 1)) − ns2
sn+1
= (n + 1)(s2 − 2s + 1)
The last line follows from the bi- = (n + 1)(s − 1)2
nomial theorem. Note: the dilation The right hand side is now
principle gives the same formula for
a > 0. 1 (s − 1)n−1
(n + 1)(s − 1)2
n+1  sn+2 
23. Hint: Take the Laplace transform of (s − 1) n+1

each side. Use the previous exercise =

sn+2
and the binomial theorem. = L {`n+1 } (s).
t
24. We have that `n ∗1(t) = 0 `n (x) dx. Taking the inverse Laplace trans-
By the convolution theorem form completes the verification.
536 B Selected Answers

26. We have that `n ∗ `m (t) = By the convolution theorem

t
0 n
` (x)`m (t − x) dx. By the convo-
lution theorem
L et ∗ `n  (s)
L {`n ∗ `m } (s)
1 (s − 1)n
(s − 1)m (s − 1)n =
= s − 1 sn+1
sm+1 sn+1
m+n
(s − 1)n−1
(s − 1) s−1 =
= 1− sn+1
sm+n+1  s 
(s − 1)n−1 s−1
(s − 1)m+n (s − 1)m+n+1 = 1−
= − sn  s 
sm+n+1 sm+n+2
(s − 1)n−1 (s − 1)n
= L {`m+n (s)} − L {`m+n+1 } (s). = n
−
s sn+1
The result follows by inversion. = L {`n−1 (t)} − L−1 {`n (t)} .
−1

∞
27. First of all 0
e−t `n (t) dt =
L {`n } (1) = 0. Thus It follows by inversion that et ∗ `n =
∞
`n−1 − `n and substituting this for-

e−x `n (x) dt
t
mula into the previous calculation
t gives the needed result.
= −
e−x `n (x) dx
0
∞ 28. F
t−x
= −e −t

e `n (x) dx
0
t
= −e−t (e ∗ `n (t)).

Section 6.5

1. Let y2 (t) = t2 u(t). Then t4 u00 + 4. y2 (t) = −1 + t
ln 1+t
. The gen-
2
 1−t
t3 u0 = 0, which gives u0 = t−1 eral solution can be written y(t) =
 
and u(t) = ln t. Substituting gives
c1 t + c2 −1 + 2t ln 1−t
1+t
.
y2 (t) = t2 ln t. The general solution 
can be written y(t) = c1 t2 + c2 t2 ln t. 1 5
5. Let y2 (t) = t 2 u(t). Then 4t 2 u00 +
2. y2 (t) = −1
3t2
.
The general solution can 3
4t 2 u0 = 0 leads to u0 = 1/t and
be written y(t) = c1 t + c2 t12 . hence u(t) = ln t. Thus y2 (t) =
√
3. Let y2 (t) = (1−t2 )u(t). Substitution t ln t. The general
√ solution√ can be
gives (1 − t2 )2 u00 − 4t(1 − t2 )u0 = 0 written y(t) = c1 t + c + 2 t ln t.
00
and hence uu0 = −2 1−t −2t
2 . From this
1
6. y2 (t) = 1. The general solution can
we get u = (1−t)2 . Integrating u0 by
0
be written y(t) = c1 t + c2 .
t
partial fractions give u = −1 +
 2 1−t2
1 1+t 7. y2 (t) = tet . The general solution can
ln 1−t and hence
4
 be written y(t) = c1 t + c2 tet .
−1 1 1+t
y2 (t) = t + (1 − t2 ) ln . 8. Let y2 (t) = t2 cos t u(t). Then
2 4 1 − t 
t4 cos t u00 − 2t4 sin t u0 = 0 which
 B Selected Answers 537

gives u0 (t) = sec2 t and hence u(t) = 12. y2 (t) = t − 1. The general solution
tan t. Thus y2 (t) = t2 sin t. The gen- can be written y(t) = c1 e−t + c2 (t −
eral solution can be written y(t) = 1).
c1 t2 cos t + c2 t2 sin t.
13. y2 (t) = − sec t. The general solu-
2
tion can be written y(t) = c1 tan t +
9. y2 (t) = cos t . The general solu-
−1
2 c2 sec t.
tion can be written y(t) = c1 sin t2 +
c2 cos t2 . 14. Let y2 (t) = tu(t). Then (t3 + t)u00 +
2
2u0 = 0 which gives u0 = t t+1 2 and
10. y2 (t) = te2t . The general solution u = t − 1t . Thus y2 (t) = t2 − 1.
can be written y(t) = c1 e2t + c2 te2t . The general solution can be written
y(t) = c1 t + c2 (t2 − 1).
11. y2 (t) = −1 − t tan t. The gen- 15. y2 (t) = t sin t. The general solution
eral solution can be written y(t) = can be written y(t) = c1 t cos t +
c1 tan t + c2 (1 + t tan t). c2 t sin t.

Section 6.6
1. sin t and cos t form a fundamental set See Example 3 in Section 4.5 for a
for the homogeneous solutions. Let related problem.
yp (t) = u1 cos t + u2 sin t. Then the
matrix equation 2. A fundamental set for y 00 − 4y = 0
cos t sin t u01
=
0
im- is e2t , e−2t . Let yp (t) = u1 (t)e2t +
   t
0
− sin t cos t u 2 sin u2 (t)e−2t . Then the matrix equation
2 1
plies u1 (t) = − sin t = 2 (cos 2t − 1)
0
e2t e−2t u01 0
= im-
and u02 (t) = cos t sin t = 12 (sin 2t). 2e 2t
−2e −2t
 u 0
2   
e 2t

Integration give u1 (t) = 14 (sin(2t) − 4t

plies u01 (t) = 41 and u02 (t) = −e4 and
2t) = 12 (sin t cos t − t) and u2 (t) = 4t
hence u1 (t) = 4t and u2 (t) = −e .
−1
4
cos 2t = −1 4
(2 cos 2t−1). This im- 2t
16
2t
plies yp (t) = 14 sin t − 21 t cos t. Since Now yp (t) = 4t e2t − e16 . Since e16
1
sin t is a homogeneous solution we is a homogeneous solution we can
4
can write the general solution in the write the general solution as y(t) =
t 2t
form y(t) = −1 t cos t + c1 cos t + 4
e + c1 e2t + c2 e−2t . On the other
2
c2 sin t. We observe that a particu- hand, the incomplete partial fraction
lar solution is the imaginary part of method gives Y (s) = (s−2)12 (s+2 =
a solution to y 00 +y = eit . We use the 1
4 p(s)
+ (s−2)(s+2) . From this we see
(s−2)2
incomplete partial fraction method
that a particular solution is yp (t) =
and get Y (s) = (s−i)21(s+i) . This 1 2t
4
te . The general solution is y(t) =
1 1
can be written Y (s) = 2i (s−i)2
+ 1 2t
te + c1 e2t + c2 e−2t .
4
p(s)
. From this we get yp (t) =

(s−i)(s+i)
3. The functions et cos 2t and et sin 2t
Im 1
L −1 1
{ (s−i) 2} = Im −i teit =
2i
 2
form a fundamental set. Let yp (t) =
−1
2
t cos t. The general solution is c1 et cos 2t + c2 et sin 2t. Then the ma-
y(t) = −1 2
t cos t + c1 cos t + c2 sin t. trix equation
538 B Selected Answers

u01 0 6. sin t and cos t form a fundamental set

W (et cos 2t, et sin 2t) = t
u2  e 
0
for the homogeneous solutions. Let
implies that u1 (t) = 2 sin 2t and
0 −1
yp (t) = u1 cos t + u2 sin t. Then the
u02 (t) = 21 cos 2t. Hence, u1 (t) = matrix equation
1
4
cos2t and u2 (t) = 41 sin 2t. From cos t sin t u01 0
= im-
this we get yp (t) = 41 et cos2 2t + − sin t cos t  u 0
2   t
tan
1 t
4
e sin2 2t = 41 et . On the other plies that u1 (t) = cos t − sec t
0

hand, the method of undetermined and u02 (t) = sin t. From this we
coefficients implies that a particular get u1 (t) = sin t − ln |sec t + tan t|
solution is of the form yp (t) = Cet . and u2 (t) = − cos t. Therefore
Substitution gives 4Cet = et and yp (t) = − cos t ln |sec t + tan t|. The
hence C = 41 . It follows that yp (t) = general solution is thus y(t) =
1 t
4
e . Furthermore, the general solu- − cos t ln |sec t + tan t| + c1 cos t +
tion is y(t) = 14 et + c1 et cos 2t + c2 sin t.
c2 et sin 2t.
7. A fundamental set is et , tet . The
matrix equation
4. A fundamental set is 1, e−3t . The
et tet u01 0
matrix equation t t t = et implies
e e + te  u2  t 
0
1 e−3t u01 0
= im- u01 (t) = −1 and u02 (t) = 1t . Hence,
 u02  
−3t −3t
 0 −3e e
0 1 −3t
plies u1 (t) = 3 e and u2 (t) = −1
0
. u1 (t) = −t, u2 (t) = ln t, and yp (t) =
3
−e−3t −tet + t ln tet . Since −tet is a homo-
Thus u1 (t) = , u2 (t) = −t ,
9 3 geneous solution we can write the
−e−3t t −3t
and yp (t) = 9
− 3
e . Ob- general solution as y(t) = t ln tet +
−3t
serve though that −e9 is a ho- c1 et + c2 tet .
mogeneous solution and so the gen-
eral solution can be written y(t) = 8. A fundamental set is {cos t, sin t}.
− 3t e−3t + c1 + c2 e−3t . The incom- The matrix equation
plete partial fraction method gives cos t sin t u01 0
= im-
Y (s) = (s+3) 1
−1
p(s)  − sin t cos t  u 0
2   t
sec
2 s = (s+3)2 + (s+3)s
3
plies u1 (t) = − tan t and u2 (t) = 1.
0 0
which implies that 3 te
−1 −3t
is a par- Hence u1 (t) = ln(cos t), u2 (t) = t,
ticular solution. The general solution and yp (t) = cos t ln(cos t) + t sin t.
is as above. The general solution is y(t) =
cos t ln(cos t) + t sin t + c1 cos t +
5. A fundamental set is et , e2t . The c2 sin t.
matrix equation
et e2t u01 0 9. The associated homogeneous equa-
t 2t = 3t implies tion is Cauchy-Euler with indicial
   
0
e 2e  u 2 e
u01 (t) = −e2t and u02 (t) = et . Hence equation s2 − 3s + 2 = (s − 2)(s − 1).
u1 (t) = −1 e2t , u2 (t) = et , and It follows that t, t2  forms a funda-
2
yp (t) = −1 e 2t t
e + et e2t = 12 e3t . The mental set. We put the given equa-
2
general solution is y(t) = 12 e3t + tion is standard form to get y 00 −
c1 et + c2 e2t . The method of undeter- t
y + t22 y = t2 . Thus f (t) = t2 . The
2 0

mined coefficients implies that a par- matrix equation

ticular solution is of the form yp = t t2 u01 0
= 2 implies
 u2  t 
0
Ce3t . Substitution gives 2Ce3t =  1 2t
2
3e3t and hence C = 21 . The general u1 (t) = −t and u2 (t) = t. Hence
0 0
3 2
solution is as above. u1 (t) = −t3 , u2 (t) = t2 , and yp (t) =
 B Selected Answers 539

−t3 2 4
3
t + t2 t2 = t6 . It follows that the tan t is a homogeneous solution we
4
general solution is y(t) = t6 + c1 t + can write the general solution as
2
c2 t2 . y(t) = t2 tan t + t + c1 tan t + c2 sec t.

10. In standard form we get y 00 − 14. When put in standard form one sees
1 0
t
y = 3t − 1t . The matrix equation that f (t) = te−t . The matrix equa-
1 t2 u01 0 tion
= 1 implies t − 1 e−t u01 0
 u2  3t − t 
0
 0 2t = im-
 u2  te 
−t 0 −t
−3 2
u1 (t) = 2 t + 2 and u2 (t) = 23 − 2t12 .
0 1 0  1 −e
Hence u1 (t) = −1 t3 + 12 t, u2 (t) = plies u1 (t) = e and u2 (t) = 1 − t.
0 −t 0
2 2
3
2
t + 2t , and yp (t) = t3 + t. The gen-
1 Hence u1 (t) = −e−t , u2 (t) = t − t2 ,
eral solution is y(t) = t3 +t+c1 +c2 t2 . and yp (t) = −(t − 1)e + (t −−t

t2 2
)e−t = −t2 e−t + e−t . Since e−t
11. The homogeneous equation is 2
is a homogeneous solution we can
Cauchy-Euler with indicial equation
write the general solution as y(t) =
s2 − 2s + 1 = (s − 1)2 . It follows that −t2 −t
e + c1 (t − 1) + c2 e−t .
{t, t ln t} is a fundamental set. After 2

writing in standard form we see the 15. After put in standard form the forc-
forcing function f (t) is 1t . The ma- ing function f is 4t4 . The matrix
trix equation equation
t t ln t u01 0 cos t2 sin t2 u01
= 1 implies =
  t 
0
 1 ln t + 1  u 2
−2t sin t 2t cos 2t  u2 
2 0
− ln t 1
u1 (t) = t and u2 (t) = t . Hence
0 0
0
2
u1 (t) = − ln2 t , u2 (t) = ln t, and 4 implies u01 (t) = −2t3 sin t2
4t 
yp (t) = 2 ln2 t + t ln2 t = 2t ln2 t.
−t
and u02 (t) = 2t3 cos t2 . Integration by
The general solution is y(t) = parts gives u1 (t) = t2 cos t2 − sin t2
2
t
ln2 t + c1 t + c2 t ln t. and u2 (t) = t2 sin t2 + cos t2 . Hence
yp (t) = t2 cos t2 − cos t2 sin t2 +
12. A fundamental set is e2t , te2t .
t2 sin t2 + cos t2 sin t2 = t2 . The gen-
The matrix equation
eral solution is y(t) = t2 + c1 cos t2 +
e2t te2t u01 0
c2 sin t2 .
2t 2t = e2t
2e e (1 + 2t) u2  
0
t +1
2

implies u01 (t) = t2−t and u0

2 (t) = 16. A fundamental set is et , e−t . The
+1
1
. Hence u1 (t) = −1 ln(t2 + 1), matrix equation
t2 +1 2
et e−t u01 0
u2 (t) = tan−1 t, and yp (t) = t = 1 im-
  1+e−t 
−t 0
−1 2t  e −e u
2
e ln(t2 + 1) + t tan−1 e2t . 2
1 1
The general solution is y(t) = plies u1 (t) = 2 1+et and u2 (t) =
0 0

−1 et
−1 2t
2
e ln(t2 +1)+t tan−1 e2t +c1 e2t + 2 1+et
. Hence u1 (t) = 12 (t − ln(1 +
c2 te2t . t
e ), u2 (t) = −1 2
(et − ln(1 + et )), and
yp (t) = 21 (tet − 1 − (et − e−t ) ln(1 +
13. The matrix equation t
e )). (Note: in the integrations of
tan t sec t u01 0
2 = im- u01 and u02 use the substitution u =
 u2  t 
0
sec t sec t tan t
et .) The general solution can now
plies u1 (t) = t and u2 (t) = −t sin t.
0 0
2 be written y(t) = 12 (tet − 1 − (et −
Hence u1 (t) = t2 , u2 (t) = t cos t − e−t ) ln(1 + et )) + c1 et + c2 e−t .
2
sin t, and yp (t) = t2 tan t + (t cos t −
2
sin t) sec t = t2 tan t + t − tan t. Since
540 B Selected Answers

Section 7
(n+1)2 ((n+1)!)2 2n+1 (2n+1)!
1. The ratio test gives n2
→ 1. test gives (n!)2 2n (2n+3)!
=
R = 1. (n+1) 2 2
1
(2n+3)(2n+2)
→ 2
. R = 2.
n
2. The ratio test gives n+1
→ 1. R = 1
10. Use the geometric series to get

∞
2n n! 1 1
= 1−(−t = (−t2 )n =
3. The ratio test gives 2n+1 (n+1)!
= 1+t2 2)
n=0
1
2(n+1)
→ 0. R = ∞. 
∞
n 2n
(−1) t .
n=0
3n+1 n+1
4. The ratio test gives n+2 3n
=
3(n+1) 11. Use the geometric series to get
n+2
→ 3. R = 31 . 
∞
1 1 t n
a
= −1 = −1 =
n=0 
t−a a 1− t a
a
(n+1)!
5. The ratio test gives n!
= n+1 → 
∞
tn
∞. R = 0. − an+1
.
n=0

∞
(−1)n un
6. Let u = t to get2
The 12. Replace t by at in the power se-
n=0
(2n+1)!

∞ (at)n
(−1)n+1 (2n+1)!
= ries for et to get eat = =
ratio test gives (−1)n (2n+3)! n=0 n!
1
 a t
∞ n n
(2n+2)(2n+3)
→ 0. It follows that the .
radius of convergence in u is ∞ and n=0 n!

hence the radius of convergence in t 

∞
sin t 1 (−1)n t2n+1
is ∞. 13. t
= t (2n+1)!
=
n=0

∞ n 2n
(−1) t
7. t is a factor in this series which .

∞
(−1)n t2n n=0
(2n+1)!
we factor out to get t (2n)!
.
n=0
et −1

∞
tn
Since t is a polynomial its presence 14. = 1
−1 =
will not change the radius of con-
t t
n=0
n! 

∞ 
∞
vergence. Let u = t2 in the new tn−1
= tn

∞
(−1)n un n=1
n!
n=0
(n+1)!
powers series to get (2n)!
.
n=0 1
15. Recall that tan−1 t = 1+t 2 dt. Us-
The ratio test gives (−1)
(−1)n+1 (2n)!
n (2n+2)! = ing the result of Problem 1 we get

1
→ 0. The radius of con- 
∞
(2n+1)(2n+2) tan−1 t = (−1)n t2n dt + C =
vergence in u and hence t is ∞. n=0

∞ 2n+1
(n+1) (−1)n t2n+1 + C. Since tan−1 0 =
8. The ratio test gives (n+1)
nn (n+1)!
n!
= n=0
n+1 n 0 it follows that C = 0. Thus
 → e. Thus R = 1
. 
∞
 n e
tan−1 t =
2n+1
(−1)n t2n+1 .
n=0
9. The expression in the denomina-
tor can be written 1·3·5···(2n+1)
1
= 16. It is easy to verify that ln(1 + t2 ) =
1·2·3·4···(2n+1) (2n+1)! 2t
2·4···2n
= n
2 (1·2·3···n)
= 1+t2
dt. Using the result of Prob-
2t
(2n+1)!
2n n!
and the given power se- lem 10 it is easy to see that 1+t 2 =
∞
(n!)2 2n tn 
∞
ries is (2n+1)!
. The ratio 2 (−1)2 t2n+1 . Hence ln(1 + t2 ) =
n=0 n=0
 B Selected Answers 541

∞ 2n+2
2 (−1)n t2n+2 + C. Evaluating at d0 =1
n=0 d2
t = 0 give C = 0. Thus ln(1 + d2 − =0

∞ 2
2n+2
t2 ) = 2 (−1)n t2n+2 . An index d2 d2
n=0 d4 − + =0
2! 4!
shift n → n − 1 gives ln(1 + t2 ) = d4 d2 d0

∞ 2n 
∞ 2n d6 − + − = 0.
2 (−1)n−1 t2n = (−1)n−1 t n . 2! 4! 6!
n=1 n=1

18. Since tan t is odd we can write Solving these equations gives

∞
tan t = d2n+1 t2n+1 and hence
n=0 d0 = 1

∞
sin t = cos t d2n+1 t 2n+1
. Writing 1
d2 =
n=0
3 5
2
out a few terms gives t − t3! + t5! − 5
2 4 d4 =
· · · = (1 − t2! + t4! − · · · )(d1 t + d3 t3 + 4!
d5 t5 · · · ). Collecting like powers of 69
d6 = .
t gives the following recursion rela- 6!
tions
Thus sec t = 1 + 12 t2 + 4! 5 4
t + 61 6!
t6 +
d1 =1 ···.
2 3 4
d1 −1 20. et sin t = (1 + t + t2! + t3! + t4! +
d3 − = 3 5
2! 3! · · · )(t − t3! + t5! − · · · ) = t + (1)t2 +
3
d3 d1 1 −1 1 −1
( 3! + 2! )t +( 3! + 3! 1
)t4 +( 5!
1 1 1
− 2! +
d5 − + = 1 5 2 1 3 1 5
3!
2! 4! 5! 4!
)t · · · = t + t + 3
t − 30
t .
d5 d3 d1 −1
d7 − + − = . 2 3
2! 4! 6! 7! 21. et cos t = (1 + t + t2! + t3! + · · · )(1 −
t2 4
Solving these equations gives 2!
+ t4! −· · · ) = 1+(1)t+( 2! 1
− 2!1
)t2 +
1
( 3! − 2! 1
)t3 + ( 4!
1 1 1
− 2! 2!
+ 1
4!
)t 4
· ·· =
d1 = 1 1 + t − 13 t3 − 16 t4 + · · · .
1
d3 =
3
2 24.
d5 =
15

∞
(1 + i)n tn
17 et eit = e(1+i)t = .
d7 = . n!
315 n=0

Thus tan t = 1 + 13 t3 + 15
2 5 17 7
t + 315 t + Now let’s compute a few powers of
···. (1 + i).
19. Since sec t is even we can write
n=0 (1 + i)0 =1
its power series in the form

∞ n=1 (1 + i)1 =1+i
d2n t2n . The equation sec t =
n=0 n=2 (1 + i)2 = 2i

∞
n=3 (1 + i)3 = −2 + 2i
1
cos t
= d2n t2n implies 1 =
n=0 n=4 (1 + i)4 = −4

∞
(−1)n t2n 
∞ 2
(2n)!
d2n t2n = (1 − t2 + n=3 (1 + i)5 = −4 − 4i
n=0 n=0
t4 6 .. ..
4!
− t6! +· · · )·(d0 +d2 t2 +d4 t4 +d6 t6 + . .
· · · ). Collecting like powers of t gives
the following recursion relations It now follows that
542 B Selected Answers

et cos t = Re e(1+i)t n approaches ∞ both even and odd

∞
tn terms approach 1. It follows that the
= (Re(1 + i)n ) radius of convergence is 1.
n=0
n!
−1
t3 t4 t528. y 0 (t) = e t t12  and y 00 (t) =
= 1 + t + 0t2 − 2 − 4 − 4 + · · −1 · 
3! 4! 5! e t t14 − t23 . Here p1 (x) = x2 and
t 3
t 4
t 5  4 3
= 1+t− − − + ··· p2 (x) = x − 2x .
3 6 30
−1

Similarly, 29. Suppose f (n) (t) = e t pn ( 1t ) where

pn is a polynomial. Then f (n+1) (t) =
et sin t = Im e(1+i)t −1
e t t12  pn ( 1t ) + p0n ( 1t ) −1
−1
e t =
∞   t2
tn −1
e t pn+1 ( 1t ), where pn+1 (x) =
= (Im(1 + i)n )
n=0
n! 2
x (pn (x) − p0n (x)). By mathematical
(n)
t2 t3 t4 t5 induction it follows that f (t) =
= 0 + t + 2 + 2 + 0 − 4 + ·e·−1 ·t p 1  for all n = 1, 2, . . ..
2! 3! 4! 5! n t

3 5
t t
= t + t2 + − +··· 30. f (n) (u) = pne(u) u . Repeated ap-
3 30
plications of L’Hospitals rule gives
lim pn (u)
u = 0.
The binomial theorem: (a + b)n =
25.  u→∞ e
n n k n−k
k=0 k a b . 31. Since f (t) = 0 for t ≤ 0 clearly

lim f (n) (t) = 0 The previous prob-
26. The root test gives n0n1n = n1 → 0. t→0−
lems imply that the right hand limits
R = ∞.
are also zero. Thus f (n) (0) exist and
cn+1 1
27. The ratio cn is 2 if n is even and is 0.
2 if n is odd. Thus lim n+1 does
c
32. Since all derivatives at t = 0 are zero
n→∞ cn
not exist. The ratio test does not ap- the Taylor series for f at t = 0 gives
√ the zero function and not f . Hence
ply. The root test√gives that n cn is
1 if n is odd and n 2 if n is even. As f is not analytic at 0.

Section 7.2

∞ t −t t −t
1. Let y(t) = cn tn . Then cn+2 = cosh t = e +e
2
and sinh t = e −e2
;
cn
n=0 the set {cosh t, sinh t} is also a fun-
(n+2)(n+1)
. Consider even and odd damental set.
c0
cases to get c2n = (2n)! and c2n+1 =
c1 
∞
t2n
(2n+1)!
. Thus y(t) = c0 (2n)!
+
n=0

∞
t2n+1
c1 (2n+1)!
= c0 cosh t + c1 sinh t.
n=0
(see Example 7 of Section 7) We ob- 
∞

serve that the characteristic polyno- 2. Let y(t) = cn tn . Then cn+2 =

n=0
mial is s2 − 1 = (s − 1)(s + 1) so 1 cn
(2cn+1 − n+1 ). We writeout sev-
et , e−t  is a fundamental set. But
n+2
eral terms. After simplifying we get:
 B Selected Answers 543
1 2
n=0 c2 = 2!
(2c1 − c0 ) n=0 c2 = − k(2·1
c0
1
n=1 c3 = 3!
(3c1 − 2c0 ) n=2
2
c4 = − k4·3c2 = k 4 c0
1 4!
n=2 c4 = (4c1 − 3c0 ) 6
4!
1
n=4 c6 = − k 6!c0
n=3 c5 = (5c1 − 4c0 )
5! .. ..
.. .. . .
. .
From this it follows that c2n =
A clear pattern emerges that can be 2n
(−1)n k(2n)!
c0
. The odd case is similar.
verified by induction: 2n+1
We get c2n+1 = (−1)n (2n+1)!
k
. The
1 power series expansion becomes
cn = (nc1 − (n − 1)c0 ).
n!
∞
y(t) = cn tn
It follows now that
n=0

∞ ∞
k2n t2n
cn tn = = c0 (−1)n
n=0
(2n)!
n=0
∞
1 n ∞
1 ∞
k2n+1 t2n+1
c1 nt − c0 (n − 1)tn . + c1 (−1)n
n! n! n=0
(2n + 1)!
n=0 n=0
= c0 cos kt + c1 sin kt.
The first series simplifies to

∞
∞
1 n ∞
1 4. Let y(t) = cn tn . Then cn+2 (n +
nt = tn n=0
n=0
n! n=1
(n − 1)! 2)(n + 1) − 3(n + 1)cn+1 + 2cn = 0
1

∞
tn+1 or cn+2 = (n+2)(n+1) (3(n + 1)cn+1 −
= 2cn ). We write out several terms and
n=0
n!
simplify:
∞ n
t
=t = tet . n=0 c2 = 1
2!
(3c1 − 2c0 )
n=0
n! 1
n=1 c3 = 3!
(7c1 − 6c0 )
1
The second series simplifies to n=2 c4 = 4!
(15c1 − 14c0 )
1
n=3 c5 = 5!
(31c1 − 30c0 )
∞
1 1
(n − 1)tn n=4 c6 = 6!
(63c1 − 62c0 )
n=0
n! .. ..
. .

∞
n n tn
∞
= t − After careful observation we get
n!
n=0
n! n=0
1
= te − et .
t
cn = ((2n − 1)c1 − (2n − 2)c0 ).
n!
It follows that (Admittedly, not easy to notice.
However, the remark below shows
y(t) = c0 (1 − t)et + c1 tet . why the presence of 2n is not unex-
pected.). From this we get

∞

∞
3. Let y(t) = cn tn . Then cn+2 (n + (2n − 1)tn
n=0
y(t) = c1
2 n!
2)(n + 1) + k cn = 0 or cn+2 = n=0
k 2 cn
− (n+2)(n+1) . We consider first the
∞
(2 − 2)tn
n
−c0 .
even case. n=0
n!
544 B Selected Answers

The first series can be written n=1 c3 = −1 c

3 1
n=3 c5 1
= 51 c3 = − 5·3 c1

∞
2n tn
∞
tn 3 1
− = e2t − et . n=5 c7 = 7 c5 = − 7·5 c1
n=0
n! n=0
n! n=7 c9 5 1
= 9 c5 = − 9·7 c1
.. ..
Similarly, the second series is e2t − . .
2et . The general solution is
From this we see that c2n+1 =
t 2t
y(t) = c0 (2e − e ) + c1 (e − 2e ). 2t t −c1
(2n+1)(2n−1)
. Thus

The characteristic polynomial is s2 −

∞
∞
t2n+1
c2n+1 t2n+1 = −c1
3s + 2 = (s − 1)(s − 2). It follows (2n + 1)(2n − 1)
that et , e2t  is a fundamental set
n=0 n=0

and so is 2et − e2t , e2t − 2et . The 

∞
t2n+1
presence of e2t in the general solu- and hence y1 (t) = − (2n+1)(2n−1)
.
n=0
tion implies that 2n tn appears in a We can write y1 in closed form by
power series solution. Thus the pres- observing
ence of 2n in the formula for cn is
necessary.
∞
t2n
y10 (t) = −
(2n − 1)

∞ n=0
5. Let y(t) = cn tn . Then the recur- y10 (t) t2n−1
∞
n=0 =−
rence relation is t 2n − 1
n=0

y10
0 2n−2 ∞
(n+2)(n+1)cn+2 −(n−2)(n+1)cn = 0 =− t
t  n=0
or
1 2n
∞
n−2 =− t
cn+2 = cn . t2 n=0
n+2
Since there is a difference of two in 1 1
=−
the indices we consider the even and t2 1 − t2
odd case. We consider first the even
case. Now integrate by parts, multiply by
t, and integrate again. We get
n=0 c2 = −c0
t t2 − 1 1−t
n=2 c4 = 40 c2 = 0 y1 (t) = + ln .
2 2 1 + t 
n=4 c6 = 62 c4 = 0
.. .. The general solution is
. .
t t2 − 1 1−t
It follows that c2n = 0 for all n = y(t) = c0 (1 − t2 ) − c1 + ln .
2 2 1 + t 
2, 3, . . .. Thus
(See also Exercise 3 in Section 6.5.)

∞
c2n t 2n 2 4
= c0 + c2 t + 0t + · · · 
∞
6. Let y(t) = cn tn . Then the recur-
n=0
n=0
= c0 (1 − t2 ) rence relation is cn+2 = n−1 c . The
n+1 n
index step is two so we consider the
and hence y0 (t) = 1 − t2 . We now even an odd cases. In the even case
consider the odd case. we get
 B Selected Answers 545

n=0 c2 = −c0 
∞
7. Let y(t) = cn tn . Then the recur-
n=2 c4 = 31 c2 = − 31 c0 n=0
rence relation is
n=4 c6 = 53 c4 = − 51 c0
2 n−1
n=4 c8 = 75 c6 = − 71 c0 cn+2 = cn+1 − cn .
n+2 (n + 2)(n + 1)
.. ..
. . For the first several terms we get
In general, c2n = −1
c
2n−1 0
and n=0 c2 = 0c1 + 21 c0 = 2!1
c0
1 1
n=1 c3 = 2 c2 − 0 = 3! c0

∞
∞
t2n
c2n t2n = −c0 . n=2 c4 1
= 42 c3 − 4·3 1
c2 = 4! c0
n=0 n=0
2n − 1 n=3 c5 3 2
= 5 c4 − 5·4 c3 = 5!3
c0 − 2
c = 1
c
5! 0 5! 0
.. ..
Thus . .

∞
t2n In general,
y0 (t) = − .
n=0
2n − 1 1
cn = c0 , n = 2, 3, . . . .
n!
Now observe that
We now get
y0 (t) t2n−1 ∞
=−
∞
t n=0
2n − 1 y(t) = cn tn
y0 (t)
0 ∞ n=0
= t2n−2
∞
 t  n=0 = c0 + c1 t + cn tn
1 1 n=2
=
∞
t2 1 − t2 tn
= (c1 − c0 )t + c0 + c0 t + c0
y0 (t) 1 1 1−t n=2
n!
= + ln
t t 2 1 + t  = (c1 − c0 )t + c0 et
t 1−t
y0 (t) = 1 + ln . = c0 (et − t) + c1 t.
2 1 + t 
For the odd terms we get 
∞
8. Let y(t) = cn tn . Then the recur-
n=0
n=1 c3 = 20 c1 rence relation is
n=3 c5 = 42 c3 = 0 (n − 2)(n − 1)
cn+2 = − cn .
n=5 c7 = 0 (n + 2)(n + 1)
.. ..
. . We separate into even and odd cases.
The even case gives
Thus
n=0 c2 = − 22 c0 = −c0

∞
n=2 c4 = 0
c2n+1 t2n+1 = c1 t + 0t3 + · · · = c1 t
n=0
n=4 c6 = 0
.. ..
and hence y1 (t) = t. The general so- . .
lution is
Generally,
t 1−t
y(t) = c0 1 + ln + c1 t. c2n = 0 n = 2, 3, . . . .
 2 1 + t 
Thus
546 B Selected Answers

∞
∞
t3
c2n t2n = c0 + c2 t2 = c0 (1 − t2 ). c2n+1 t2n+1 = c1 t + c3 t3 = c1 (t − ).
n=0 n=0
3

For the odd indices The general solution is

n=1 c3 = 0 t3
y(t) = c0 (1 − 3t2 ) + c1 (t − ).
n=3 c5 = 0 3
n=5 c7 = 0
.. .. 
∞
10. Let y(t) = cn tn . Then the recur-
. . n=0
rence relation is
Generally,
(n + 4)
cn+2 = cn .
c2n+1 = 0, n = 1, 2, . . . . (n + 2)

Thus The even case gives

∞
n=0 c2 = 24 c0 = 2c0
c2n+1 t2n+1 = c1 t.
n=0
n=2 c4 = 46 c2 = 3c0
n=4 c6 = 68 c4 = 4c0
It follows that .. ..
. .
y(t) = c0 (1 − t2 ) + c1 t.
More generally,

∞
9. Let y(t) = cn tn . Then the recur- c2n = (n + 1)c0
n=0
rence relation is and
(n − 2)(n − 3)
∞
∞
cn+2 =− cn . c2n t2n = c0 (n + 1)t2n .
(n + 2)(n + 1)
n=0 n=0
The even case gives:

∞
Let y0 (t) = (n + 1)t2n and ob-
n=0 c2 = − 26 c0 = −3c0 n=0
n=2 c4 = 0 serve that
n=4 c6 = 0
∞
ty0 (t) = (n + 1)t2n+1
.. ..
. . n=0

1
∞

Hence
ty0 (t) dt = t2n+2

2 n=0

∞
2
∞
c2n t2n = c0 + c2 t2 = c0 (1 − 3t2 ). t
= t2n
n=0 2 n=0
The odd case gives t2 1
=
2 1 − t2
n=1 c3 = − 31 c1
d t2 1
n=3 c5 = 0 ty0 (t) =
dt  2 1 − t2 
n=5 c7 = 0
t
.. .. =
. . (1 − t2 )2
1
Hence y0 (t) = .
(1 − t2 )2
 B Selected Answers 547

The odd case gives 13. By de Moivre’s formula cos nx is the

real part of (cos x + i sin x)n . The bi-
n=1 c3 = 35 c1 nomial theorem gives
n=3 c5 = 57 c3 = 73 c1
(cos x + i sin x)n
n=5 c7 = 79 c5 = 93 c1
.. .. n
n
. . = cosn−k xik sink x
k 
k=0 

More generally,
Only the even powers of i contribute
2n + 3
c2n+1 = c1 to the real part. It follows that
3
cos nx
and

[n/2]

∞
∞
2n + 3 = cosn−2j x(i2j ) sin2j x
c2n+1 t2n+1 = c1 t2n+1 . j=0
n=0 n=0
3

[n/2]


∞ = cosn−2j x(−1)j (1 − cos2 x)j ,
2n+3 2n+1
Let y1 (t) = 3
t and ob- j=0
n=0
serve that where we use the greatest integer

∞ function [x] to denote the greatest
3ty1 (t) = (2n + 3)t2n+2 integer less than or equal to x. It
n=0 follows that cos nx is a polynomial
∞
in cos x.

3ty1 (t) dt = t2n+3
n=0 14. By de Moivre’s formula sin nx is the

∞ imaginary part of (cos x + i sin x)n .
3 2n
=t t The binomial theorem gives
n=0

t 3 (cos x + i sin x)n

=
1 − t2 n
n
d t3 = cosn−k xik sink x
k 
3ty1 (t) = k=0 
dt (1 − t2 )
t2 (3 − t2 ) Only the odd powers of i contribute
=
(1 − t2 )2 to the imaginary part. It follows that
t(3 − t2 )
y1 (t) = . sin nx
3(1 − t2 )2

[n−1/2]

The general solution is = Im cosn−(2j+1) x(i2j+1 ) sin2j+1 x

j=0
1 t(3 − t2 )
y(t) = c0 + c1 .
[n−1/2]
2
(1 − t ) 2 3(1 − t2 )2 = cosn−2j−1 x(−1)j (1 − cos2 x)j sin x,
j=0

12. Let n be an integer. Then einx = where we use the greatest integer
(eix )n . By Euler’s formula this is function [x] to denote the greatest
integer less than or equal to x. It fol-
(cos x + i sin x)n = cos nx + i sin nx. lows that sin nx is a product of sin x
and a polynomial in cos x.
548 B Selected Answers

15. We have
19. By using the sum and difference for-
cos x cos(n + 1)x = cos xTn+1 (cos x)
mula it is easy to verify the following
sin x sin(n + 1)x = sin2 xUn (cos x). trigonometric identity:
Subtracting gives
2 sin a cos b = sin(a + b) − sin(b − a).
cos(n+2)x = cos xTn+1 (cos x)−(1−cos2 x)Un (cos x)
Let a = x and b = nx. Then
and from this we get
Tn+2 (t) = tTn+1 (t) − (1 − t2 )Un (t). 2 sin x cos nx = sin((n+1)x)−sin((n−1)x)

16. We have and hence

sin((n+2)x) = sin((n+1)x+x) = sin((n+1)x) cosx)T

2(sin x+cos((n+1)x) sin x
n (cos x) = (sin x)Un (cos x)−(sin x)Un−2 (cos x).

and hence
Now divide by 2 sin x and let t =
(sin x)Un+1 (cos x) = (sin x)Un (x) cos x+(sincos
x)Tx.n+1 (cos x).
Now divide by sin x and let t = cos x. 20. By using the sum and difference for-
We get mula it is easy to verify the following
Un+1 (t) = tUn (t) + Tn+1 (t). trigonometric identity:

2 sin a sin b = cos(b − a) − cos(a + b).

17. By using the sum and difference for-
mula it is easy to verify the following
Let a = x and b = nx. Then
trigonometric identity:
2 cos a cos b = cos(a + b) + cos(a − b). 2 sin x sin nx = cos((n−1)x)−cos((n+1)x)

Let a = x and b = nx. Then and hence

2 cos x cos nx = cos((n+1)x)+cos((n−1)x)
2Un−1 (cos x) = Tn−1 (cos x)−Tn+1 (cos x).
(use the fact that cos is even) and
hence Now let t = cos x, replace n by n+1,
2(cos x)Tn (cos x) = Tn+1 (cos x)+Tn−1 (cos x). divide by 2.
and

Now let t = cos x. 21. F

Section 7.3
1–2.
1. F

2.
3–6. Laguerre Polynomials:
 B Selected Answers 549

3. Verify the second formula in Theorem 3:

(E◦ f | g) = (f | E◦ g) ,

for f and g in C.

4. Show that Z ∞
e−t `n (at) dt = (1 − a)n .
0

5. Show that
∞  
n m
Z
−t
e `m (t)`n (at) dt = a (1 − a)n−m ,
0 m
if m ≤ n and 0 if m > n.

6.

Section 7.4
Exercises
1–5. For each problem determine the singular points. Classify them as regular
or irregular.
t 1
1. y 00 + 1−t2 y
0
+ 1+t y =0

t
Solution: The function 1−t 2 is analytic except at t = 1 and t = −1. The
1
function 1+t is analytic except at t = −1. It follows that t = 1 and t = −1
 
t −t
are the only singular points. Observe that (t − 1) 1−t 2 = 1+t is analytic
 
2 1
at 1 and (t − 1) 1+t is analytic at t = 1. It follows that 1 is a regular
 
t t
singular point. Also observe that (t + 1) 1−t 2 = 1−t is analytic at −1
 
1
and (t + 1)2 1+t = (1 + t) is analytic at t = −1. It follows that −1 is a
regular singular point. Thus 1 and −1 are regular points.
1−t 0 1−cos t
2. y 00 + t y + t3 y =0

Solution: The functions 1−tt and

1−cos t
t3 are analytic except
at t = 0. So
t = 0 is the only singular point.
Observe that (t − 0) 1−t t = 1 − t is
analytic at 0 and (t − 0)2 1−cos
t3
t
= 1−cos t
t is analytic at t = 0. It follows
that 0 is a regular singular point.
550 B Selected Answers

1−et
3. y 00 + 3t(1 − t)y 0 + t y =0

t
Solution: Both 3t(1−t) and 1−e
t are analytic. There are no singular points
and hence no regular points.
4. y 00 + 1t y 0 + 1−t
t3 y =0

Solution: Clearly, t = 0
is the only singular point. Observe that t 1t = 1

is analytic while t2 t13 = 1t is not. Thus t = 0 is an irregular singular

point.
5. ty 00 + (1 − t)y 0 + 4ty = 0

Solution: We first write it in standard form: y 00 + 1−t 0

t y +4y = 0. While the
0 1−t
coefficient of y is analytic the coefficient of y is t is analytic except at
t = 0. It follows that t = 0 is a singular point. Observe that t 1−t


t = 1 − t
is analytic and 4t2 is too. It follows that t = 0 is a regular point.
6–10. Each differential equation has a regular singular point at t = 0. De-
termine the indicial polynomial and the exponents of singularity. How many
Frobenius solutions are guaranteed by Theorem 2.
6. 2ty 00 + y 0 + ty = 0

Solution: In standard form the equation is 2t2 y 00 +ty 0 +t2 y = 0. The indicial
equation is p(s) = 2s(s − 1) + s = 2s2 − s = s(2s − 1) The exponents of
singularity are 1 and 12 . Theorem 7.42 guarantees two Frobenius solutions.
7. t2 y 00 + 2ty 0 + t2 y = 0

Solution: The indicial equation is p(s) = s(s − 1) + 2s = s2 + s = s(s + 1)

The exponents of singularity are 0 and −1. Theorem 2 guarantees one
Frobenius solution but there could be two.
8. t2 y 00 + tet y 0 + 4(1 − 4t)y = 0

Solution: The indicial equation is p(s) = s(s − 1) + s + 4 = s2 + 4 The

exponents of singularity are ±2i. Theorem 2 guarantees that there are two
complex Frobenius solutions.
9. ty 00 + (1 − t)y 0 + λy = 0

Solution: In standard form the equation is t2 y 00 + t(1 − t)y 0 + λty = 0. The

indicial equation is p(s) = s(s − 1) + s = s2 The exponent of singularity is
0 with multiplicity 2. Theorem 2 guarantees that there is one Frobenius
solution. The other has a logarithmic term.
 B Selected Answers 551

10. t2 y 00 + 3t(1 + 3t)y 0 + (1 − t2 )y = 0

Solution: The indicial equation is p(s) = s(s − 1) + 3s + 1 = s2 + 2s + 1 =

(s + 1)2 The exponent of singularity is −1 with multiplicity 2. Theorem 2
guarantees that there is exactly Frobenius solution. There is a logarithmic
solution as well.
11–14. Verify the following claims that were made in the text.
sin t cos t
11. In Example 5 verify the claims that y1 (t) = t and y2 (t) = t .

∞ ∞
(−1)n t2n 1 (−1)n t2n 1
sin t. y2 is done simi-
P P
Solution: y1 (t) = (2n+1)! = t (2n)! = t
n=0 n=0
larly.
12. In Example 8 we claimed that the solution to the recursion relation

((i + n)2 + 1)cn + ((i + n − 1)(i + n − 2) + 3)cn−1 + 3cn−2 + cn−3 = 0

in
was cn = n! . Use Mathematical Induction to verify this claim.

n
Solution: The entries in the table in Example 8 confirm cn = in! for
n = 0, . . . 5. Assume the formula is true for all k ≤ n. Let’s consider the
recursion relation with n replaced by n + 1 to get the following formula
for cn+1 :
((i + n + 1)2 + 1)cn+1 = −((i + n)(i + n − 1) + 3)cn + 3cn−1 + cn−2
in 3in−1 in−2
(n + 1)(2i + n + 1)cn+1 = −(n2 − n + 2in − i + 2) + +
n! (n − 1)! (n − 2)!
in−2 2 2
(n + 1)(2i + n + 1)cn+1 = − i (n − n + 2in − i + 2) + 3in + n(n − 1) 
n! 
in
(n + 1)(2i + n + 1)cn+1 = (i(n + 1 + 2i)).
n!
On the second line we use the induction hypothesis. Now divide both
sides by (n + 1 + 2i)(n + 1) in the last line to get

in+1
cn+1 = .
(n + 1)!
in
We can now conclude by Mathematical Induction that cn = n! , for all
n ≥ 0.
13. In Remark 3 we stated that in the logarithmic case could be obtained by
a reduction of order argument. Consider the Cauchy-Euler equation

t2 y 00 + 5ty 0 + 4y = 0.
552 B Selected Answers

One solution is y1 (t) = t−2 . Use reduction of order to show that a second
independent solution is y2 (t) = t−2 ln t, in harmony with the statement in
the coincident case of the theorem.

Solution: Let y(t) = t−2 v(t). Then y 0 (t) = −2t−3 v(t) + t−2 v 0 (t) and
y 00 (t) = 6t−4 v(t) − 4t−3 v 0 (t) + t−2 v 00 (t). From which we get
t2 y 00 = 6t−2 v(t) − 4t−1 v 0 (t) + v 00 (t)
5ty 0 = −10t−2v(t) + 5t−1 v 0 (t)
4y = 4t−2 v(t).
Adding these terms and remembering that we are assuming the y is a
solution we get
0 = t−1 v 0 (t) + v 00 (t).
00
From this we get vv0 = −1 0
t . Integrating we get ln v (t) = − ln t and hence
1
v (t) = t . Integrating again gives v(t) = ln t. It follows that y(t) = t−2 ln t
0

is a second independent solution.

14. Verify the claim made in Example 6 that y1 (t) = (t3 + 4t2 )et

Solution:
∞
X n + 4 n+2
y1 (t) = t
n=0
n!
∞ ∞
X n n+2 X 4 n+2
= t + t
n=0
n! n=0
n!
∞ ∞ n
X tn−1 X t
= t3 + 4t2
n=1
(n − 1)! n=0
n!
= t3 et + 4t2 et = (t3 + 4t2 )et .

15–26. Use the Frobenius method to solve each of the following differential
equations. For those problems marked with a (*) one of the independent so-
lutions can easily be written in closed form. For those problems marked with
a (**) both independent solutions can easily be written in closed form.
15. ty 00 − 2y 0 + ty = 0 (**) (real roots; differ by integer; two Frobenius solu-
tions)

Solution: Indicial polynomial: p(s) = s(s − 3); exponents of singularity

s = 0 and s = 3.
n = 0 c0 (r)(r − 3) = 0
n = 1 c1 (r − 2)(r + 1) = 0
n ≥ 1 cn (n + r)(n + r − 3) = −cn−1
 B Selected Answers 553

r=3:
n odd cn = 0
m
n = 2m c2m = 3c0 (−1) (2m+2)
(2m+3)!
P∞ (−1)m (2m+2)t2m+3
y(t) = 3c0 m=0 (2m+3)! = 3c0 (sin t − t cos t).
r=0: One is lead to the equation 0c3 = 0 and we can take c3 = 0. Thus

n odd cn = 0
m+1
n = 2m c2m = c0 (−1) (2m)!
(2m−1)

P∞ (−1)m+1 (2m−1)t2m
y(t) = c0 m=0 (2m)! = c0 (t sin t + cos t).
General Solution: y = c1 (sin t − t cos t) + c2 (t sin t + cos t).
16. 2t2 y 00 − ty 0 + (1 + t)y = 0 (**) (real roots, do not differ by an integer, two
Frobenius solutions)

Solution: Indicial polynomial: p(s) = (2s − 1)(s − 1); exponents of singu-

larity s = 1/2 and s = 1.

n = 0 c0 (2r − 1)(r − 1) = 0
n ≥ 1 cn (2(n + r) − 1)(n + r − 1) = −cn−1

−cn−1
r = 1: cn = (2n+1)n and hence

(−1)n 2n c0
cn = , n ≥ 1.
(2n + 1)!

From this we get

∞
X (−1)n 2n tn+1
y(t) = c0
n=0
(2n + 1)!
√ ∞ √
t X (−1)n ( 2t)2n+1
= c0 √
2 n=0 (2n + 1)!
c0 √ √
= √ t sin 2t.
2

r = 1/2: The recursion relation becomes cn (2n − 1)(n) = −cn−1 . The

−cn−1
coefficient of cn is never zero for n ≥ 1 hence cn = n(2n−1) and hence

(−1)n 2n
cn = .
(2n)!

We now get
554 B Selected Answers
∞
X (−1)n 2n tn+1/2
y(t) =
n=0
(2n)!
∞ √
√ X (−1)n ( 2t)2n
= t
n=0
(2n)!
√ √
= t cos 2t.
√ √ √ √
General Solution: y = c1 t sin 2t + c2 t cos 2t.
17. t2 y 00 − t(1 + t)y 0 + y = 0, (*) (real roots, coincident, logarithmic case)

Solution: Indicial polynomial: p(s) = (s − 1)2 ; exponents of singularity

s = 1, multiplicity 2. There is one Frobenius solution.
P∞
r = 1 : Let y(t) = n=0 cn tn+1 . Then

n≥1 n2 cn − ncn−1 = 0.
1
This is easy to solve. We get cn = n! c0 and hence
∞
X 1 n+1
y(t) = c0 t = c0 tet .
n=0
n!

Logarithmic Solution: Let y1 (t) = tet . The second independent solution

is necessarily of the form
∞
X
y(t) = y1 (t) ln t + cn tn+1 .
n=0

Substitution into the differential equation leads to

∞
X
t2 e t + (n2 cn − ncn−1 )tn+1 = 0.
n=1

We write out the power series for t2 et and add corresponding coefficients
to get
1
n2 cn − ncn−1 + .
(n − 1)!
The following list is a straightforward verification:

n=1 c1 = −1
−1 1


n=2 c2 = 2 1+ 2
−1 1 1


n=3 c3 = 3! 1+ 2 + 3
−1 1 1 1


n=4 c4 = 4! 1+ 2 + 3 + 4 .
 B Selected Answers 555

1
Let sn = 1 + 2 + · · · + n1 . Then an easy argument gives that
−sn
cn = .
n!
We now have a second independent solution
∞
X s n tn
y2 (t) = tet ln t − t .
n=1
n!

General Solution:
∞
!
t t
X s n tn
y = c1 te + c2 te ln t − t .
n=1
n!

18. 2t2 y 00 − ty 0 + (1 − t)y = 0 (**) (real roots, do not differ by an integer, two
Frobenius solutions)

Solution: Indicial polynomial: p(s) = (2s − 1)(s − 1); exponents of singu-

larity s = 1/2 and s = 1.

n = 0 c0 (2r − 1)(r − 1) = 0
n ≥ 1 cn (2(n + r) − 1)(n + r − 1) = cn−1

cn−1
r = 1: cn = (2n+1)n and hence

2 n c0
cn = , n ≥ 1.
(2n + 1)!
From this we get
∞
X 2n tn+1
y(t) = c0
n=0
(2n + 1)!
√ ∞ √ 2n+1
t X ( 2t)
= c0 √
2 n=0 (2n + 1)!
c0 √ √
= √ t sinh 2t.
2

r = 1/2: The recursion relation becomes cn (2n − 1)(n) = cn−1 . The

cn−1
coefficient of cn is never zero for n ≥ 1 hence cn = n(2n−1) and hence

2n
cn = .
(2n)!
We now get
556 B Selected Answers
∞
X 2n tn+1/2
y(t) =
n=0
(2n)!
∞ √
√ X ( 2t)2n
= t
n=0
(2n)!
√ √
= t cosh 2t.
√ √ √ √
General Solution: y = c1 t sinh 2t + c2 t cosh 2t.
19. t2 y 00 + t2 y 0 − 2y = 0 (**) (real roots; differ by integer; two Frobenius
solutions)

Solution: Indicial polynomial: p(s) = (s−2)(s+1); exponents of singularity

s = 2 and s = −1.

n = 0 c0 (r − 2)(r + 1) = 0
n ≥ 1 cn (n + r − 2)(n + r + 1) = −cn−1 (n + r − 1)

s=2:
(−1)n (n + 1)
cn = 6c0 , n≥1
(n + 3)!
 
P∞ (−1)n (n+1)tn −t
y(t) = 6c0 n=0 (n+3)! = 6c0 (t+2)e
t + t−2
t .
s=-1: The recursion relation becomes cn (n − 3)(n) = cn−1 (n − 2) = 0.
Thus
n = 1 c1 = − c20
n = 2 c2 = 0
n = 3 0c3 = 0
We can take c3 = 0 and
then cn = 0 for all n ≥ 2. We now have y(t) =
c0 t−1 (1 − 2t ) = c20 2−t
t .
−t
(t+2)e
General Solution: y = c1 2−t
t + c2 t .
20. t2 y 00 + 2ty 0 − a2 t2 y = 0 (**) (real roots, differ by integer, two Frobenius
Solutions)

Solution: Indicial polynomial: p(s) = s(s + 1); exponents of singularity

r = 0 and r = −1.

n = 0 c0 r(r + 1) = 0
n = 1 c1 (r + 1)(r + 2) = 0
n ≥ 1 cn (n + r)(n + r + 1) = a2 cn−2

r = 0: In the case c1 = 0 and c2n+1 = 0 for all n ≥ 0. We also get

a2 c0 c0 a2n
c2n = 2n(2n+1) , n ≥ 1 which gives c2n = (2n+1)! . Thus
 B Selected Answers 557
∞
X (at)2n
y(t) = c0
n=0
(2n + 1)!
∞
c0 X (at)2n+1
=
at n=0 (2n + 1)!
c0
= sinh at.
at

r = −1: The n = 1 case gives 0c1 = 0 so c1 may be arbitrary. There

is no inconsistency. We choose c1 = 0. The recursion relation becomes
cn (n − 1)(n) = a2 cn−2 and it is straightforward to see that c2n+1 = 0
2n
while c2n = a(2n)!
c0
. Thus
∞
X a2n t2n−1
y(t) = c0
n=0
(2n)!
∞
c0 X (at)2n
=
t n=0 (2n)!
c0
= cosh at.
t

General Solution: y = c1 cosht at + c2 sinht at .

21. ty 00 + (t − 1)y 0 − 2y = 0, (*) (real roots, differ by integer, logarithmic case)

Solution: Indicial polynomial: p(s) = s(s − 2); exponents of singularity

r = 0 and r = 2.
n=0 c0 r(r − 2) = 0
n≥1 cn (n + r)(n + r − 2) = −cn−1 (n + r − 3)

n−1
r=2: The recursion relation becomes cn = − n(n+2) cn−1 . For n = 1 we
see that c1 = 0 and hence cn = 0 for all n ≥ 1. It follows that y(t) = c0 t2
is a solution.

r=0: The recursion relation becomes cn (n)(n − 2) = −cn−1 (n − 3) = 0.

Thus
c0
n = 1 −c1 = − −2
n = 2 0c2 = −2c0 ⇒⇐
The n = 2 case implies an inconsistency in the recursion relation since
c0 6= 0. Since y1 (t) = t2 is a Frobenius solution a second independent
solution can be written in the form
∞
X
2
y(t) = t ln t + c n tn .
n=0
558 B Selected Answers

Substitution leads to
∞
X
t3 + 2t2 + (−c1 − 2c0 )t + (cn (n)(n − 2) + cn−1 (n − 3))tn = 0
n=2

and the following relations:

n=1 −c1 − 2c0 = 0

n=2 2 − c1 = 0
n=3 1 + 3c3 = 0
n≥4 n(n − 2)cn + (n − 3)cn−3 = 0.

We now have c0 = −1, c1 = 2, c3 = −1/3. c2 can be arbitrary so we choose

c2 = 0, and cn = −(n−3)c
n(n−2)
n−1
, for n ≥ 4. A straightforward calculation
gives
2(−1)n
cn = .
n!(n − 2)
A second independent solution is
∞
!
2 t3 X 2(−1)n tn
y2 (t) = t ln t + −1 + 2t − + .
3 n=4 n!(n − 2)

General Solution: y = c1 t2 + c2 y2 (t).

22. ty 00 − 4y = 0 (real roots, differ by an integer, logarithmic case)

Solution: Write the differential equation in standard form: t2 y 00 − 4ty = 0.

Clearly, the indicial polynomial is p(s) = s(s− 1); exponents of singularity
are r = 0 and r = 1.
r=1: Let
∞
X
y(t) = cn tn+1 , c0 6= 0.
n=0

Then
n≥1 cn (n + 1)(n) = 4cn−1
n
4 c0
It follows that cn = (n+1)!n! , for all n ≥ 1 (notice that when n = 0 we get
P∞ 4n tn+1
P∞ (4t)n+1
c0 ). Hence y(t) = c0 n=0 (n+1)!n! = c40 n=0 (n+1)!n! .

r = 0: When r = 0 the recursion relations become

cn (n)(n − 1) = 4cn−1 , n≥1

and hence when n = 1 we get 0c1 = 4c0 , an inconsistency as c0 6= 0.

There is no second Frobenius solution. However, there is a second linearly
 B Selected Answers 559

independent solution of the form y(t) = y1 (t) ln t + ∞ n

n=0 cn t , where
P
P∞ (4t)n+1
y1 (t) = n=0 (n+1)!n! . Substituting this equation into the differential
equation and simplifying leads to the recursion relation

(2n − 1)4n
n(n − 1)cn = 4cn−1 − , n ≥ 1.
(n + 1)!n!

When n = 1 we get 0 = 0c1 = 4c0 − 2 and this implies c0 = 12 . Then the

recursion relation for n = 1 becomes 0c1 = 0, no inconsistency; we will
choose c1 = 0. The above recursion relation is difficult to solve. We give
the first few terms:
c0 = 12 c1 = 0
−46
c2 = −2 c3 = 27
−251
c4 = 405
It now follows that
 
1 46 251 4
y2 (t) = y1 (t) ln t + − 2t2 − t3 − t + ··· .
2 27 405

23. t2 (−t + 1)y 00 + (t + t2 )y 0 + (−2t + 1)y = 0 (**) (complex)

Solution: Indicial polynomial: p(s) = (s2 + 1); exponents of singularity

r = ±i. Let r = i (the case r = −i
P gives equivalent results). The recursion
relation that arises from y(t) = ∞n=0 c n t n+i
is c n ((n+i) 2
+1)−cn−1 ((n−
2
2 + i) + 1) = 0 and hence

(n − 2)(n − 2 + 2i)
cn = cn−1 .
n(n + 2i)

A straightforward calculation gives the first few terms as follows:

1−2i
n = 1 c1 = 1+2i c0
n = 2 c2 = 0
n = 3 c3 = 0
 
1−2i
and hence cn = 0 for all n ≥ 2. Therefore y(t) = c0 (ti + 1+2i) t1+i .
Since t > 0 we can write ti = ei ln t = cos ln t+i sin ln t, by Euler’s formula.
Separating the real and imaginary parts we get two independent solutions

y1 (t) = −3 cos ln t − 4 sin ln t + 5t cos ln t

y2 (t) = −3 sin ln t + 4 cos ln t + 5t sin ln t.
560 B Selected Answers

24. t2 y 00 + t(1 + t)y 0 − 2y = 0, (**) (real roots, differ by an integer, two

Frobenius solutions)

Solution: Indicial polynomial: p(s) = (s−1)(s+1); exponents of singularity

r = 1 and r = −1.
n = 0 c0 (r − 1)(r + 1) = 0
n ≥ 1 cn (n + r − 1)(n + r + 1) = −cn−1 (n + r − 1)

r = 1 : The recursion relation becomes cn (n(n+2)) = −ncn−1 and hence

2(−1)n c0
cn = .
(n + 2)!
We then get
∞
X
y(t) = cn tn+1
n=0
∞
X (−1)n tn+1
= 2c0
n=0
(n + 2)!
∞
2c0 X (−t)n+2
=
t n=0 (n + 2)!
2c0 −t
= (e − (1 − t))
t 
e−t − (1 − t)

= 2c0 .
t

r = −1: The recursion relation becomes cn (n − 2)(n) = −cn−1 (n − 2).

Thus
n = 1 c1 = −c0
n = 2 0c2 = 0
Thus c2 can be arbitrary so we can take c2 = 0 and then cn = 0 for all
n ≥ 2. We now have
∞
X 1−t
y(t) = cn tn−1 = c0 (t−1 − 1) = c0 .
n=0
t

−t
General Solution: y = c1 1−t e
t + c2 t .

25. t2 y 00 + t(1 − 2t)y 0 + (t2 − t + 1)y = 0 (**) (complex)

Solution: Indicial polynomial: p(s) = (s2 + 1); exponents of singularity

r = ±i. Let r = i (the case r = −i
P gives equivalent results). The recursion
relation that arises from y(t) = ∞ c
n=0 n t n+i
is
 B Selected Answers 561

n = 1 c1 = c0
n ≥ 2 cn ((n + i)2 + 1) + cn−1 (−2n − 2i + 1) + cn−2 = 0

A straightforward calculation gives the first few terms as follows:

n=1 c1 = c0
1
n=2 c2 = 2! c0
1
n=3 c3 = 3! c0
1
n=4 c4 = 4! c0 .

An easy induction argument gives

1
cn = c0 .
n!
We now get
∞
X
y(t) = cn tn+i
n=0
∞ n+i
X t
= c0
n=0
n!
i t
= c0 t e .

Since t > 0 we can write ti = ei ln t = cos ln t+i sin ln t, by Euler’s formula.

Now separating the real and imaginary parts we get two independent
solutions
y1 (t) = et cos ln t and y2 (t) = et sin ln t.

26. t2 (1 + t)y 00 − t(1 + 2t)y 0 + (1 + 2t)y = 0 (**) (real roots, equal, logarithmic
case)

Solution: Indicial polynomial: p(s) = (s − 1)2 ; exponents of singularity

s = 1 with multiplicity 2.
r=1: Let
∞
X
y(t) = cn tn+1 , c0 6= 0.
n=0

Then
n=1 c1 = 0
n≥2 n2 cn = −cn−1 (n − 2)
It follows that cn = 0 for all n ≥ 1. Hence y(t) = c0 t. Let y1 (t) = t.

logarithmic case: Let y(t) = t ln t+ ∞ n+1

. The recursion relations
P
n=0 cn t
becomes
562 B Selected Answers

n = 0 −c0 + c0 = 0
n = 1 c1 = 1
n ≥ 2 n2 cn = −(n − 1)(n − 2)cn−1
The n = 0 case allows us to choose c0 arbitrarily. We choose c0 = 0. For
n ≥ 2 it is easily to see the cn = 0. Hence y(t) = t ln t + t2 .

general solution: y(t) = a1 t + a2 (t ln t + t2 ).

Section 7.5
Exercises
1–3.
1. Verify the statement made that the Frobenius solution to

ty 00 + 2(it − k)y 0 − 2kiy = 0

with exponent of singularity r = 2k + 1 is not a polynomial.

2. Use the Frobenius method on the differential equation

t2 y 00 + 2t(it − k)y 0 − 2k(it − 1) = 0.

Verify that the indicial polynomial is q(s) = (s − 1)(s − 2k). Show that for
the exponent of singularity r = 1 there is a polynomial solution while for the
exponent of singularity r = 2k there is not a polynomial solution. Verify the
formula for bk (t) given in Proposition 8.
s
3. Verify the formula for the inverse Laplace transform of given in
(s2 + 1)k+1
Theorem 9.

4–11. This series of exercises leads to closed formulas for the inverse
Laplace transform of
1 s
and .
(s2 − 1)k+1 (s2 − 1)k+1

Define Ck (t) and Dk (t) by the formulas

1 1
L {Ck (t)} (s) = 2
2k k! (s − 1)k+1
and
1 s
L {Dk (t)} (s) = 2 .
2k k! (s − 1)k+1
4. Show the following:

1. Ck+2 (t) = t2 Ck (t) − (2k + 3)Ck+1 (t)

 B Selected Answers 563

2. Dk+2 (t) = t2 Dk (t) − (2k + 1)Dk+1 (t)

3. tCk (t) = Dk+1 (t).

5. Show the following: For k ≥ 1

1. Ck0 (t) = Dk (t)

2. Dk0 (t) = 2kCk−1 (t) + Ck (t).

6. Show the following:

1. tCk00 (t) − 2kCk0 (t) − tCk (t) = 0

2. t2 Dk00 (t) − 2ktDk0 (t) = (2k − t2 )Dk (t) = 0

7. Show that there are polynomials ck (t) and dk (t), each of degree at most k, such
that

1. Ck (t) = ck (t)et − ck (−t)e−t

2. Dk (t) = dk (t)et + dk (−t)e−t

8. Show the following:

1. tc00k (t) + (2t − 2k)c0k (t) − 2kck (t) = 0

2. td00k (t) + 2t(t − k)d0k (t) − 2k(t − 1)dk (t) = 0

9. Show the following:

(−1)k (2k)!
1. ck (0) = .
2k+1 k!

(−1)k−1 (2(k − 1))!

2. d0k (0) = .
2k (k − 1)!

10. Show the following:

(−1)k (2k − n)!

k
1. ck (t) = (−2t)n .
2k+1 n=0 n!(k − n)!

(−1)k (2k − n − 1)!

k
2. dk (t) = (−2t)n .
2k+1 n=1 (n − 1)!(k − n)!

11. Show the following:

(−1)k (2k − n)!

k
1
1. L−1
(t) = (−2t)n et − (2t)n e−t .
(s2 − 1)k+1  22k+1 k! n=0 n!(n − k)! 

(−1)k (2k − n − 1)!

k
s
2. L−1
(t) = (−2t)n et + (2t)n e−t .
(s2 − 1)k+1  22k+1 k! n=1 (n − 1)!(n − k)! 
564 B Selected Answers

Chapter 8

Section 8.1
1. (c)
2. (g)
3. (e)
4. (a)
5. (f)
6. (d)
7. (h)
8. (b)
−22
9. 3
9
10. 4
11. 4
12. 1 + ln 2
11
13. 2
14. 0
15. 5
44
16. 3
17. a) A,B
b) A,B,C
c) A
d) none
18. a) A,B
b) A,C
c) A,B,C,D
d) A,B,C
( −3t
t
−1+e if 0 ≤ t < 1
19. y(t) = 13 9e−3(t−1)9 e−3t
− + if 1≤t<∞
3 9 9

 0 if 0 ≤ t < 1
−t + et−1

if 1 ≤ t < 2
20. y(t) = t−2 t−1


 t − 2 − 2e + e if 2 ≤ t < 3
t−3 t−2 t−1
e − 2e +e if 3 ≤ t < ∞

(
sin cos t 3e−(t−π)
− 2 − if 0 ≤ t < π
21. y = 2 −(t−π) 2
−e if π ≤ t < ∞
(
t −t
−t + e − e if 0 ≤ t < 1
22. y(t) =
et − et−1 − e−t if 1 ≤ t < ∞
(
e2t − 2te2t if 0 ≤ t < 2
23. y(t) =
1 + e2t − 5e2(t−2) − 2te2t + 2te2(t−2) if 2 ≤ t < ∞
 B Selected Answers 565

Section 8.2
9. (a) (t − 2)χ[2, ∞) ; (b) (t − 2)h(t − 2); (c) e−2s /s2 .
10. (a) tχ[2, ∞) ; (b) th(t − 2); (c) e−2s s12 + 2s .

11. (a) (t + 2)χ[2, ∞) ; (b) (t + 2)h(t − 2); (c) e−2s s12 + 4s .

12. (a) (t − 4)2 χ[4, ∞) ; (b) (t − 4)2 h(t − 4); (c) e−4s s23 .
2 2 2
3 + s2 + s .
8 16
13. (a) t χ[4, ∞) ; (b) t h(t − 4); (c) e −4s
2 2 s 2
3 + s2 + s .
8 12
14. (a) (t − 4)χ[4, ∞) ; (b) (t − 4)h(t − 4); (c) e −4s
s
15. (a) (t − 4)2 χ[2, ∞) ; (b) (t − 4)2 h(t − 2); (c) e−2s s23 − s42 + 4s .

16. (a) et−4 χ[4, ∞) ; (b) et−4 h(t − 4); (c) e−4s s−11
.
t t −4(s−1) 1
17. (a) e χ[4, ∞) ; (b) e h(t − 4); (c) e s−1
.
18. (a) et−4 χ[6, ∞) ; (b) et−4 h(t − 6); (c) e−6s+2 s−1 1
.

t t 1 4
19. (a) te χ[4, ∞) ; (b) te h(t − 4); (c) e −4(s−1)
(s−1)2
+ s−1 .
 −5s
2e−4s
20. (a) χ[0,4) (t) − χ[4,5) (t); (b) 1 − 2h4 + h5 ; (c) s − s + e s .
1

21. (a) tχ[0,1) (t) + (2 − t)χ[1,2) (t) + χ[2,∞) (t); (b) t − (2 − 2t)h1 + (t − 1)h2 ; (c)
−s
1
s2
+ 2es2 + e−2s s12 + 1s .
 −s
22. (a) tχ[0,1) (t) + (2 − t)χ[1,∞) (t); (b) t + (2 − 2t)h1 ; (c) s12 + 2es2 .
∞ ∞ −s
23. (a) n=0 (t − n)χ[n,n+1) (t); (b) t − n=1 hn ; (c) s12 − s(1−e e
−s ) .
∞ ∞ n 1
24. (a) n=0 χ[2n,2n+1) (t); (b) n=0 (−1) hn ; (c) s(1+e−s ) .
25. (a) t2 χ[0,4) (t)+4χ[2,3) (t)+(7−t)χ[3,4) (t); (b) t2 +(4−t2 )h2 +(3−t)h3 +(7−t)h4 ;
−3s
2 2
(c) s3 − e
−2s
s3
− s42  − e s2 − e−4s s12 − 4s . ∞
  1 1
26. (a) ∞ n=0 (2n + 1 − t)χ[2n,2n+2) (t); (b) −(t + 1) + 2 n=0 h2n ; (c) − s2 − s +
2
s(1−e−2s )
.
27. (a) χ[0,2) (t) + (3 − t)χ[2,3) (t) + 2(t − 3)χ[3,4) (t) + 2χ[4,∞) (t); (b) 1 + (2 − t)h2 +
−2s −3s −4s
(3t − 9)h3 − (2t − 8)h4 ; (c) 1s − e s2 + 3es2 − 2es2 .

Section 8.3
0 if 0 ≤ t < 3,
1. et−3 h(t − 3) =  t−3
e if t ≥ 3.
0 if 0 ≤ t < 3,
2. (t − 3)h(t − 3) = 
t−3 if t ≥ 3.
0 if 0 ≤ t < 3,
3. 1
2
(t − 3)2 et−3 h(t − 3) = 1
2
(t − 3)2 et−3 if t ≥ 3.
0 if 0 ≤ t < π, 0 if 0 ≤ t < π,
4. h(t − π) sin(t − π) =  = 
sin(t − π) if t ≥ π − cos t if t ≥ π.
0 if 0 ≤ t < 3π, 0 if 0 ≤ t < 3π,
5. h(t − 3π) cos(t − 3π) =  = 
cos(t − 3π) if t ≥ 3π − cos t if t ≥ 3π.
1 −(t−π) 0 if 0 ≤ t < π,
6. 2
e sin 2(t − π)h(t − π) = 1
2
e −(t−π)
sin 2t if t ≥ π.
7. (t − 1)h(t − 1) + 12 (t − 2)2 et−2 h(t − 2)
0 if 0 ≤ t < 2,
8. 12 h(t − 2) sin 2(t − 2) = 1
2
sin 2(t − 2) if t ≥ 2.
566 B Selected Answers

 0 if 0 ≤ t < 2,
9. 1
− 2) e2(t−2) − e−2(t−2) = 1  2(t−2)
h(t
4
 e − e−2(t−2) if t ≥ 2.
4

 0 if 0 ≤ t < 5,
10. h(t − 5) 2e−2(t−5) − e−(t−5) =  −2(t−5)
 2e − e−(t−5) if t ≥ 5.
 
11. h(t − 2) e2(t−2) − et−2 + h(t − 3) e2(t−3) − et−3
 
t if 0 ≤ t < 5,
12. t − (t − 5)h(t − 5) = 
5 if t ≥ 5.
1 3
1 3 t if 0 ≤ t < 3,
13. 6
t + 61 (t − 3)3 h(t − 3) = 16
6
t3 + 16 (t − 3)3 if t ≥ 3.
5 0 if 0 ≤ t < π,
14. h(t−π)e−3(t−π) 2 cos 2(t − π) − sin 2(t − π)  =  −3(t−π)
 2
e 2 cos 2t − 25 sin 2t  if t ≥ π.
5

15. e−3t 2 cos 2t − sin 2t  − h(t − π)e−3(t−π) 2 cos 2(t − π) − 52 sin 2(t − π) 
 2


Section 8.4

1. y = − 23 h(t − 1) 1 − e−2(t−1)
 
2. y = e−2t − 1 + 2h(t − 1) 1 − e−2(t−1)
 
3. y = h(t − 1) 1 − e−2(t−1) − h(t − 3) 1 − e−2(t−3)
  
4. y = 14 e−2t − 14 + 21 t+h(t−1) 14 e−2(t−1) − 14 + 21 (t − 1) − 12 h(t−1) 1 − e−2(t−1)
 
5. − 91 h(t − 3) (−1 + cos 3(t − 3)) 
6. y = − 23 et + 125 4t
e + 14 + 121
h(t − 5) −3 + 4et−5 − e4(t−5)
  
7. y = 13 h(t−1) 1 − 3e−2(t−1) + 2e−3(t−1) + 13 h(t−3) −1 + 3e−3(t−3) − 2e−3(t−3)
 
1
8. y = cos 3t + 24 h(t −2π) (3 sin t − sin 3t)
9. y = te−t + h(t − 3) 1 − (t − 2)e−(t−3)

10. y = te−t − 41 h(t − 3) −et − 5e−t+6 + 2te−t+6  

1 −5t
11. y = 20 e − 14 e−t + 15 + 20 1
h(t − 2) 4 + e−5(t−2) − 5e−(t−2) + 20 1
h(t −
  
1
4) 4 + e−5(t−4) − 5e−(t−4) + 20 h(t − 6) 4 + e−5(t−6) − 5e−(t−6)
 

Sections 8.5 and 8.6

1. y = h(t − 1)e−2(t−1)
2. y = (1 + h(t − 1))e−2(t−1)
3. y = h(t − 1)e−2(t−1) − h(t − 3)e−2(t−3)
1
sin 2t if 0 ≤ t < π,
4. y = 12 (1 + h(t − π)) sin 2t = 2
sin 2t if t ≥ π.
1
sin 2t if π ≤ t < 2π,
5. y = 12 χ[π, 2π) sin 2t = 2
0 otherwise.
6. y = cos 2t + 12 χ[π, 2π) sin 2t
 B Selected Answers 567

7. y = (t − 1)e−2(t−1) h(t − 1)
8. y = (t − 1) e−2t + e−2(t−1) h(t − 1)

9. y = 3h(t − 1)e−2(t−1) sin(t − 1)
10. y = e−2t (sin t − cos t) + 3h(t − 1)e−2(t−1)
 sin(t − 1)
11. y = e−2t cos 4t + 21 sin 4t  + 14 sin 4t h(t − π)e−2(t−π) − h(t − 2π)e−2(t−2π)
  
1
12. y = 18 e5t − e−t − 6te−t  + 16 h(t − 3) e5(t−3) − e−(t−3)
 

Chapter 9

Section 9.1
   
    1 −1 −1 2 0 8
−3 1 63
2. AB = , AC = , BA = 5 −2 18 , CA = −2 3 7
−3 5 46
0 1 5 3 −1 7
 
  3 −1 7
3 4
3. A(B + C) = AB + AC = , (B + C)A = 3 1 25
1 11
3 2 12
 
−2 5
4. C = −13 −8
7 0 
6 4 −1 −8
5. AB = 0 2 −8 2 
 2 −1 9 −5
2 3 −8
6. BC =
−2 0 24
 
8 0
 4 −5
7. CA =   8 14 


10 11
 
6 0 2
 4 2 −1
8. B t At = −1 −8 9 


−8 2 −5
8 9 −48
9. ABC =  4 0 −48 .
−2 3 40
 
1 4 3 1
0 0 0 0
10. AB = −4 and BA =  −1 −4 −3 −1


  −2 −8 −6 −2
1 0
14.
1 −1
568 B Selected Answers
 
0 0 −1
15. 3 −5 −1
0 0 5
 
ab 0
16. AB − BA = . It is not possible to have ab = 1 and −ab = 1 since
0 −ab
1 6= −1.    
01 00
17. (a) Choose, for example, A = and B = .
00 10
2 2 2
(b) (A+ B)  = A +2AB  + B precisely when AB = BA.
11 12
18. A2 = , A3 =
1
 2 23
1 n
19. Bn =
01
 n 
a 0
20. An = n
 0 b    
01 v2 1c
21. (a) A = ; the two rows of A are switched. (b) A =
10 v1 01
 
v1 + cv2
; to the first row is added c times the second row while the
v2
second row is unchanged, (c) to the second row is added c times the first
row while the first row is unchanged. (d) the first row is multiplied by a
while the second row is unchanged, (e) the second row is multiplied by a
while the first row is unchanged.

Section 9.2
     
1 4 3   2 14 3 2
x
1 1 −1 , x =  y , b = 4, and [A|b] =  1 1 −1 4 .
   
1. (a) A = 

2 0 1   1   2 0 1 1
z
0 1 −1   6 0 1 −1 6
  x 1    
2 −3 4 1 x2 
, b = 0 , and [A|b] = 2 −3 4 1 0 .
(b) A = ,x=
3 8 −3 −6 x3  1 3 8 −3 −6 1
x4
x1 − x3 + 4x4 + 3x5 = 2
5x1 + 3x2 − 3x3 − x4 − 3x5 = 1
2.
3x1 − 2x2 + 8x3 + 4x4 − 3x5 = 3
−8x1 + 2x2 + 2x4 + x5 = −4
 
10 1
3. p2,3 (A) = 0 1 4 
000
4. RREF  
1 0 −5 −2 −1
5. t2,1 (−2)(A) =
01 3 1 1
 B Selected Answers 569
 
010 3
6. m2 (1/2)(A) = 0 0 1 3 
0000
7. RREF  
1010 3
8. t1,3 (−3)(A) = 0 1 3 4 1 
  00000
100 2
9. 0 1 0 1 
0 0 1 −1
 
1 0 0 −11 −8
10. 0 1 0 −4 −2
001 9 6
0 1 0 27 14
 
0 0 1 3 1 
11. 
0 0 0 0 0 
2

0 0 0 0 0 
1200 3
0 0 1 0 2 
12. 
0 0 0 1 0 


0 0 0 0 0 
100 1 1 1
13. 0 1 0 −1 3 1 
0 0 1 2 1 1
10 2
0 1 1 
 
14. 0 0 0 
 
0 0 0 
000
 
1010
0 1 3 0
15. 
0 0 0 1


0000
 
14003
16. 0 0 1 0 1
00013
 
1 0 0 0 0
0 ‘1 −1 0 0 
17. 
0 0 0 1 −1


0 0 0 0 0
     
x −1 −3
18. y =  1  + α 1 
z 0 5
570 B Selected Answers
       
x1 4 −1 −2
 x2  −1 −3 −1
19.   =   + α  + β 
 x3   0  1 0
x4 0 0 1
   
x −2
20. =α
y 1
     
x1 3 −4
 x2   0  1
21.   =   + α 
 x3  −2 0
x4 5 0
   
x 14/3
22.  y  =  1/3 
z −2/3
23. no solution
   
0 1
24.  3  + α  0 
4 0
     
5 1 1
25. The equation −1 = a  1  + b −1 has solution a = 2 and b = 3. By
4 2 0
 
5
Proposition 6 −1 is a solution.
4
26. k = 2  
−7/2
27. a) If xi is the solution set for Ax = bi then x1 =  7/2 , x2 =
    −3/2
−3/2 7
 3/2 , and x3 = −6.
−1/2 3
b) The augmented matrix [A|b1 |b2 |b3 ] reduces to
 
1 0 0 −7/2 −3/2 7
 0 1 0 7/2 3/2 −6  .
0 0 1 −3/2 −1/2 3

The last three columns correspond in order to the solutions.

Section 9.3
 
4 −1
1.
−3 1
 B Selected Answers 571
 
3 −2
2.
−4 3
3. not invertible
 
−2 1
4.
−3/2 1/2
5. not invertible
 
1 −1 1
6. 0 1 −2
0 0 1
 
−6 5 13
7.  5 −4 −11
−1 1 3 
−1/5 2/5 2/5
8. −1/5 −1/10 2/5 
−3/5 1/5 1/5 
−29 39/2 −22 13
 7 −9/2 5 −3 
9. 
−22 29/2 −17 10 


9 −6 7 −4 
−1 0 0 −1
1  0 −1 0 −1 
 
10. 2  0 0 −1 −1 
−1 −1 −1 −1
 
0 0 −1 1
1 0 0 0
11. 
 0 1 1 −1


−1 −1 0 1
12. not invertible
 
5
13. b=
−3
 
−2
14. b= 6 
−3
 
16
1 
15. b = 10 11 
  18
1
16. b = 1
1 
19
 −4 
17. b=  
15 
−6
572 B Selected Answers
 
3
1
18. b= −4 .


1
19. (At )−1 = (A−1 )t
20. (E(θ))−1 = E(−θ)
21. F (θ)−1 = F (−θ)
22.

Section 9.4
1. 1
2. 0
3. 10
4. 8
5. −21
6. 6
7. 2
8. 15
9. 0  
1 −2 + s 2
10. s2 −3s s = 0, 3
1 −1 + s
 
1 s−3 1
11. s2 −6s+8 s = 2, 4
 1 s − 3
1 s−1 1
12. s2 −2s+s s=1±i
−1 s − 1
2
 
(s − 1) 3 s−1
1
13. (s−1)3
 0 (s − 1)2 0  s=1
0 3(s − 1) (s − 1)2
s2 − 2s + 10 −3s − 6

3s − 12
1  −3s + 12 s2 − 2s − 8 3s − 12  s = −2, 1, 4
14. s3 −3s2 −6s+8
2
 2 3s + 6 −3s −
 6 s − 2s − 8
s + s 4s + 4 0
1 −s − 1 s2 + s 0  s = −1, ±2i
15. 3 2
s +s +4s+4
  s − 4 4s + 4 s2 + 4
9 −4
16.
−2 1
17. no inverse 
1 6 −4
18. 10 −2 3
 
4 −4 4
1 
19. 8 −1 3 −1

−5 −1 3
 B Selected Answers 573
 
27 −12 3
1
20. 21
−13 5 4 
−29 16 −4
 
2 −98 9502
1 
21. 6 0 3 −297

0 0 6
 
−13 76 −80 35
1 −14 76 −80 36 
 
22. 2  6 −34 36 −16 

7 −36 38 −17
 
55 −95 44 −171
1 50 −85 40 −150 
 
23. 15 70 −125 59 −216 

65 −115 52 −198
24. no inverse

Chapter 10

Section 10.1
1. nonlinear, autonomous
2. linear, constant coefficient, not autonomous, not homogeneous
3. linear, homogeneous, but not constant coefficient or autonomous
4. nonlinear and not autonomous
5. linear, constant coefficient, homogeneous, and autonomous
6. linear, constant coefficient, not homogeneous, but autonomous
   
y y
In all of the following solutions, y = 1 = 0 .
y2 y
   
0 0 1 −1
12. y = y, y(0) =
−k 2 0 0
   
0 1 −1
13. y 0 = 2 y, y(0) =
k 0 0
     
0 1 0 0
14. y 0 = y + , y(0) =
−k 2 0 A cos ωt 0
0 1
" #  
α
15. y 0 = c b y, y(0) =
− − β
" a a#
0 1 
0
  
α
0
16. y = c b y+ , y(0) =
− − A sin ωt β
" a a#
0 1  
−2
17. y 0 = 1 2 y, y(1) =
− 2 − 3
t t
574 B Selected Answers

Section 10.2
 
0 −2 sin 2t 2 cos 2t
1. A (t) =
−2 cos
−3t
2t −2sin 2t
−3e 1
2. A0 (t) =
2t 2e2t
 −t
−e (1 − t)e−t (2t − t2 )e−t


3. A0 (t) =  0 −e−t (1 − t)e−t 

0 0 −e−t
 
1
4. y 0 (t) =  2t 
t−1
 
0 00
5. A (t) =
00
6. v 0 (t) = −2e−2t t22t
 
+1 −3 sin 3t
 
0 1
7.
−1 0
 2
e − e−2 e2 + e−2 − 1

1
8. 2 −2
 1 − e − e
4 e2 − e−2
3/2
9.  7/3 
ln 4 − 1
 
4 8
10.
12 16
11. " 1 1 # on I1 , I4 , and I5 .
Continuous
s s2
12. 2 1
" s32 s−2 1 #
s2 +1 s2 +1
13. −1 s
s2 +1 s2 +1

3! 2s 1
" #
s4 (s2 +1)2 (s+1)2
14. 2−s s−3 3
3 2
 1s  s −6s+13 s

s2
2
15.  s3 
6
s4

 
2 1 −1
16. s2 −1 −1 1
 B Selected Answers 575
1 1 1 
s s2+1 s(s2 +1)
s 1
17.  0 s2 +1 s2 +1 
 

0 s−1
+1
s
s2 +1
2
18. 1 2t 3t
 
" #
1 t
19. t −t
e +e
2 cos t
et tet
 

20.  −4 e−3t
3 + + et sin t
 
3

" 3 cos 3t e#3t

et + e−t et − e−t
21.
et − e−t et + e−t
" 1 #
0
 t 
s−1 e 0
22. (sI − A)−1 = and L −1
(sI − A) −1
=
0 1 0 e2 t
" s−2 s−2 −1 # "2 1 #
3t 1 1 3t
−1 s(s−3) s(s−3) −1 −1 3 + 3e 3 − 3e
23. (sI − A) = −2 s−1 and L (sI − A) = 2 2 3t 1 2 3t
s(s−3) s(s−3) 3 − 3e + e
 1 1 s+1   2
 3 3
t
s s2 s3 1 t t+ 2
−1  1 1  −1 −1
24. (sI − A) =  0 s s2  and L (sI − A) = 0 1 t 
 

0 0 1s 00 1
" s 1
#  
−1 s2 +1 s2 +1 −1 −1 cos t sin t
25. (sI − A) = −1 and L (sI − A) =
s − sin t cos t
s2 +1 s2 +1
t2 t2 2
1 + ( 2 ) + 12 ( t2 )2
" # " #
1+(2)
 
1
26. a) y0 = , y1 = , y2 = ,
1 2
1 + ( t2 )
2 2
1 + ( t2 ) + 12 ( t2 )2
2 2
1 t2 3
" #
1 + ( t2 ) + 21 ( t2 )2 + 3! (2)
y3 = .
t2 1 t2 2 1 t2 3
1 + ( 2 ) + 2 ( 2 ) + 3! ( 2 )
b) The nth term is
2
1 t2 n
" #
1 + ( t2 ) + · · · + n! (2)
y= 2
1 t2 n
1 + ( t2 ) + · · · + n! (2)

c) By the Existence and " Uniqueness theorem there are

# no other solutions.
1 t2 2
# "
 
1 1 1 − 2( 2 )
27. a) y0 = , y1 = , y2 = ,
0 t2 2
−( 2 ) −( t2 )
" 2 #
1 − 12 ( t2 )2
y3 = 2
.
1 t2 3
−( t2 ) + 3! (2)
b) By the Uniqueness and Existence Theorem there are no other solu-
tions.
576 B Selected Answers
" #
1 + t2
 
1
28. a) y0 = , y1 = ,
1 1 − t2
" # " #
1 + t2 1 + t2
y2 = , y3 = .
1 − t2 1 − t2
" #
1 + t2
b) y = .
1 − t2
30. (−∞, ∞)
31. (−2, 3)
32. (−∞, 2)
33. (−∞, ∞) 
2 2
  
1 t t2 1 + 2t + 3t2
34. b) eN t = 0 1 t . (c) y(t) =  2 + 3t .
   

00 1 3
1 1 1 
s s2 s3
1 1 
d) and (e): Both matrices are  0

s s2 
0 0 1s
(c1 + c2 t)et
 
35. c) (iv) y(t) = .
c2 e t
1 1
" #
s−1 (s−1)2
d) and (e): Both matrices are 1
0 s−1

Section 10.3
1. All except (b) and (e) are fundamental matrices.
2. A Ψ (t) = eAt y(t)
   
cos t sin t 3 cos t − 2 sin t
(a)
− sin t cos t −3 sin t + 2 cos t
 −t 2t −t 2t
14e−t − 5e2t
  
1 4e − e −e + e 1
(c) 3
4e−t − 4e2t −e−t + 4e2t 3 14e−t − 20e2t
−2e2t + 3e3t 3e2t − 3e3t −12e2t + 15e3t
   
(f)
−2e2t + 2e3t 3e2t − 2e3t −12e2t + 10e3t
 2t
3e + e6t −3e2t + 3e6t 1 15e2t − 3e6t
  
(g) 41
−e2t + e6t e2t + 3e6t 4 −5e2t − 3e6t

(1 − 2t)e3t −4te3t (3 + 2t)e3t

   
(h)
te3t (1 + 2t)e3t −(2 + t)e3t
 B Selected Answers 577

3. A(t) Ψ (t) y(t)

cos(t2 /2) − sin(t2 /2) 3 cos(t2 /2) − 2 sin(t2 /2)

   
(i)
sin(t2 /2) cos(t2 /2) −3 sin(t2 /2) − 2 cos(t2 /2)
2
2t2 2
   
1 4 + 2t 1 12 + 2t
(j) 4 −2t2 4 − 2t2 4 −8 − 2t2
" # " #
1 t2 /2 2 2 2
1 t2 /2 2
(e + e−t /2 ) 12 (et /2 − e−t /2 ) e + 52 e−t /2
(k) 1 t2 /2
2 2 2 2
2
1 t2 /2 2

2 (e − e−t /2 ) 12 (et /2 + e−t /2 ) 2e − 52 e−t /2

4. (a), (c), and (d) are linearly independent.

Section 10.4
In the following, c1 , c2 , and c3 denote arbitrary real constants.
 −t   −t 
e 0 c1 e
1. (a) ; (b) y(t) =
0 e3t c e3t
   2 
cos 2t sin 2t c1 cos 2t + c2 sin 2t
2. (a) ; (b)
−2tsin 2t 2t
cos 2t  2t −c1 sin  2t + c2 cos 2t
e te c1 e + c2 e2t
3. (a) ; (b)
0 e2t c2 e2t
 −t
e cos 2t e−t sin 2t c1 e−t cos 2t + c2 e−t sin 2t
  
4. (a) ; (b)
−e−t sin 2t e−t cos 2t −c1 e−t sin 2t + c2 e−t cos 2t
 t −t t −t
(3c1 − c2 )et + (−c2 + c2 )e−t
  
3e − e −e + e
5. (a) 12 t ; (b) 1

 3e − 3e−t −et +3e−t 2

 t (3c1 − c2 )e +
t
3(−c1 + c2 )e
−t
t t t t
e + 2te −4te c e + (2c1 − 4c2 )te
6. (a) ; (b) 1 t
tet et − 2tet c e + (c1 − 2c2 )tet
  2  
cos t + 2 sin t −5 sin t c2 cos t + (2c1 − 5c2 ) sin t
7. (a) ; (b)
" sin t cos t − 2 sin #t " c2 cos t + (c1 − 2c2 ) sin #t
−t −t −t
e cos 2t −2e sin 2t e (c1 cos 2t − 2c2 sin 2t)
8. (a) 1 −t −t
; (b) 1 −t
2e sin 2t e cos 2t 2 e (c1 sin 2t + 2 cos2t)
t 3t 3t t
c2 )et + (c1 + c2 )e3t
 
e + e e − e (c 1 −
9. (a) 12 3t t t 3t ; (b) 12 3t
 t e − et e + te   (ct1 + c2 )e + (c2 − c1 )et
c e + (4c1 + 2c2 )tet

e + 4te 2te
10. (a) ; (b) 1 t
−8tet et − 4tet c2 e − (8c1 + 4c2 )tet
3 t
(c1 − 32 c3 )e−t + 23 c3 et
 −t −t
  
e 0 2 (e − e )
11. (a)  0 e2t 0 ; (b)  c2 e2t 
t t
0 0 e c3 e
cos 2t 2 sin 2t 0 c1 cos 2t + 2c2 sin 2t
   
1
12. (a)  − 2 sin 2t cos 2t 0 ; (b)  − 21 c1 sin 2t + c2 cos 2t 
−e−t + cos 2t 2 sin 2t e−t (c3 − c1 )e−t + c1 cos 2t + 2c2 sin 2t
578 B Selected Answers
 −3t
+ 1 3e−t − 3e−3t 1 − e−3t

2e
13. (a) 13  0 3e−t 0 ;
−3t
2 − 2e −3e + 3e−3t e−3t + 2
−t

(2c1 − 3c2 − c3 )e−3t + 3c2 e−t + (c1 + c3 )

 

(b) 13  c2 e−t 
−3t
(−2c1 + 3c2 + c3 )e − 3c2 e−t + (2c1 + 2c3 )
−tet + et tet tet c1 et + (c2 + c3 − c1 )tet
   

14. (a)  e2t − et et −e2t + et ; (b)  (−c1 + c2 + c3 )et + (c1 − c3 )e2t 

t 2t t t 2t t t t 2t
−te − e
 3t 3t 1 2 3t + e te e + te  3t 1 c e + (−c 1 + c 2 + c 3 )te + (c 3 − c 1 )e
e te − 2 t e − te3t c1 e + (c1 − c3 )e3t − 21 c3 t2 e3t
 

15. (a)  0 e3t −te3t ; (b)  c2 e3t − c3 te3t 

3t 3t
0 0 e c3 e

Section 10.6
te−t t(e2t − 1)
     
12 cos t − 2 cos 2t 1
1. 2. 3.
e3t − et 3 2 sin 2t − sin t 2e2t − 1
 t
5e − 5e−t − 6te−t

3tet + te−t + e−t − et
   
1 1 5t cos t − 5 sin t
5. 2 7. 2 11. 14  4te2t
3tet + 3te−t + 2e−t − 2et

t sin t + 2t cos t − 2 sin t t −t
2e − 2e

Appendix A

Section A.1
√ √ π √
1. z = (1, 1) = 1 + i = [ π4 , 2] = 2ei 4 , w = (−1, 1) = −1 + i = [ 3π 4 , 2] =
√ i 3π √ p √
2e 4 , z · w = −2, wz = −i, wz = i, z 2 = 2i, z = ±( 2 cos( π8 ) +
p√ p √ p √ p √
i 2 sin( π8 )) = ±( 21 2 2 + 2+i 21 2 2 − 2) since cos( π8 ) = 12 2 + 2
p √
and sin( π8 ) = 12 2 − 2 , z 11 = −32 + 32i.
2 3 8 1 6
2. (a) −5 + 10i (b) −3 + 4i (c) 13 − 13 i (d) 130 − 130 i (e) 5 − 85 i.

3. (a) - (c) check your result (d) z = (3π/2 + 2kπ)i for all integers k.

4. Always either 5 or 1.
3
5. The vertical line x = . The distance between two points z, w in the plane
2
is given by |z − w|. Hence, the equation describes the set of points z in
the plane which are equidistant from 1 and 2.

6. This is the set of points inside the ellipse with foci (1, 0) and (3, 0) and
major axis of length 4.
 B Selected Answers 579
p√ p√ 
7. (a) ± 2+1+i 2−1 (b) ±(2 + i)
−1 −1 −1
8. (a) 5ei tan (4/3) ≈ 5e0.927i (b) 5e−i tan (4/3)
(c) 25e2i tan (4/3)
(d)
1 −i tan−1 (4/3) iπ
5e (e) 5eiπ (f) 3e 2

9. (a) Real: 2e−t cos t−3e−t sin t; Imaginary: 3e−t cos t+2e−t sin t (b) Real:
−eπ sin 2t; Imaginary: eπ cos 2t (c) Real: e−t cos 2t; Imaginary: e−t sin 2t

10. (a) If z = 2πki for k an integer, the sum is n. Otherwise the sum is
1 − enz
. (b) 0
1 − ez
√ √
11. −2i, 3 + i, − 3 + 2i
C
Tables

Table C.1: Laplace Transform Rules

f (t) F (s)

1. a1 f1 (t) + a2 f2 (t) a1 F1 (s) + a2 F2 (s)

1 s
2. f (at) F
a a
3. eat f (t) F (s − a)

4. f (t − c)h(t − c) e−sc F (s)

5. f 0 (t) sF (s) − f (0)

6. f 00 (t) s2 F (s) − sf (0) − f 0 (0)

7. f (n) (t) sn F (s)−sn−1 f (0)−sn−2 f 0 (0)−· · ·−

sf (n−2) (0) − f (n−1) (0)
8. tf (t) −F 0 (s)

9. t2 f (t) F 00 (s)

10. tn f (t) (−1)n F (n) (s)

1 R∞
11. f (t) s F (u)du
t
Rt F (s)
12. f (v)dv
0 s
Rt
13. (f ∗ g)(t) = 0 f (τ )g(t − τ ) dτ F (s)G(s)
582 C Tables

Table C.2: Table of Laplace Transforms

F (s) f (t)

1
1. 1
s
1
2. t
s2
1 tn−1
3. (n = 1, 2, 3, . . .)
sn (n − 1)!
1
4. eat
s−a
1
5. teat
(s − a)2
1 tn−1 eat
6. (n = 1, 2, 3, . . .)
(s − a)n (n − 1)!
b
7. sin bt
s2 + b 2
s
8. cos bt
s + b2
2

b
9. eat sin bt
(s − a)2 + b2
s−a
10. eat cos bt
(s − a)2 + b2
 
n!
11. Re (n = 0, 1, 2, . . .) tn eat cos bt
(s − (a + bi))n+1
 
n!
12. Im (n = 0, 1, 2, . . .) tn eat sin bt
(s − (a + bi))n+1
1 eat − ebt
13. (a 6= b)
(s − a)(s − b) a−b
s aeat − bebt
14. (a 6= b)
(s − a)(s − b) a−b
1 eat ebt
15. a, b, c distinct + +
(s − a)(s − b)(s − c) (a − b)(a − c) (b − a)(b − c)
ect
(c − a)(c − b)
 C Tables 583

Table C.2: Table of Laplace Transforms

F (s) f (t)

s aeat bebt
16. a, b, c distinct + +
(s − a)(s − b)(s − c) (a − b)(a − c) (b − a)(b − c)
cect
(c − a)(c − b)
s2 a2 eat b2 ebt
17. a, b, c distinct + +
(s − a)(s − b)(s − c) (a − b)(a − c) (b − a)(b − c)
c2 ect
(c − a)(c − b)
s
18. (1 + at)eat
(s − a)2

at2
 
s
19. t+ eat
(s − a)3 2

s2 a 2 t2
 
20. 1 + 2at + eat
(s − a)3 2

1 sin bt − bt cos bt
21.
(s2 + b2 )2 2b3
s t sin bt
22.
(s2 + b2 )2 2b

23. 1 δ(t)

24. e−cs δc (t)

e−cs
25. h(t − c)
s
1 tα−1
26. (α > 0)
sα Γ (α)

Laplace Transforms of periodic functions

p
1
Z
27. e−st f (t) dt f (t) where f (t + p) = f (t) for all t
1 − e−sp 0
1
28. swc (t)
s(1 + e−sc )
584 C Tables

Table C.2: Table of Laplace Transforms

F (s) f (t)
p
1
Z
29. e−st t dt htip
1 − e−sp 0

Table C.3: Table of Convolutions

f (t) g(t) (f ∗ g)(t)

Z t
1. 1 g(t) g(τ ) dτ
0
m!n!
2. tm tn tm+n+1
(m + n + 1)!
at − sin at
3. t sin at
a2
2 2 2
4. t2 sin at (cos at − (1 − a 2t ))
a3
1 − cos at
5. t cos at
a2
2
6. t2 cos at (at − sin at)
a3
eat − (1 + at)
7. t eat
a2
2 at 2 2
8. t2 eat (e − (a + at + a 2t ))
a3
1
9. eat ebt (ebt − eat ) a 6= b
b−a
10. eat eat teat
1
11. eat sin bt (beat − b cos bt − a sin bt)
a2 + b 2
1
12. eat cos bt (aeat − a cos bt + b sin bt)
a + b2
2
 C Tables 585

Table C.3: Table of Convolutions

f (t) g(t) (f ∗ g)(t)

1
13. sin at sin bt (b sin at − a sin bt) a 6= b
b 2 − a2
1
14. sin at sin at (sin at − at cos at)
2a
1
15. sin at cos bt (a cos at − a cos bt) a = 6 b
b 2 − a2
1
16. sin at cos at t sin at
2
1
17. cos at cos bt (a sin at − b sin bt) a 6= b
a2 − b 2
1
18. cos at cos at (at cos at + sin at)
2a
19. f (t) δc (t) f (t − c)h(t − c)

Inverse Laplace Transforms of Quadratics

 
1
k L−1
(s2 + 1)k+1

0 sin t
1
1 2 (sin t − t cos t)
1
(3 − t2 ) sin t − 3t cos t


2 8
1
(15 − 6t2 ) sin t − (15t − t3 ) cos t


3 48
1
(105 − 45t2 + t4 ) sin t − (105t − 10t3 ) cos t


4 384
1
(945 − 420t2 + 15t4 ) sin t − (945t − 105t3 + t4 ) cos t


5 3840
1
6 46080 (10395 − 4725t2 + 210t4 − t6 ) sin t

−(10395t − 1260t3 + 21t4 ) cos t

586 C Tables

 
−1 s
k L
(s2 + 1)k+1

0 cos t
1
1 2 t sin t
1
t sin t − t2 cos t


2 8
1
(3t − t3 ) sin t − 3t2 cos t


3 48
1
(15t − 6t3 ) sin t − (15t2 − t4 ) cos t


4 384
1
(105t − 45t3 + t5 ) sin t − (105t2 − 10t4 ) cos t


5 3840
1
(945t − 420t3 + 15t5 ) sin t − (945t2 − 105t4 + t6 ) cos t


6 46080
 C Tables 587

 
−1 1
k L
(s − 1)k+1
2

1
0 2 (et − e−t )
1
1 4 ((−1 + t)et − (−1 − t)e−t )
1
(3 − 3t + t2 )et − (3 + 3t + t2 )e−t


2 16
1
(−15 + 15t − 6t2 + t3 )et − (−15 − 15t − 6t2 − t3 )e−t


3 96
1
4 768 105 − 105t + 45t2 − 10t3 + t4 )et

−(105 + 105t + 45t2 + 10t3 + t4 )e−t

1
5 7680 (−945 + 945t − 420t2 + 105t3 − 15t4 + t5 )et

−(−945 − 945t − 420t2 − 105t3 − 15t4 − t5 )e−t

 
−1 s
k L
(s2 − 1)k+1

1
0 2 (et + e−t )
1
1 4 (tet + −te−t )
1
(−t + t2 )et + (t + t2 )e−t


2 16
1
(3t − 3t2 + t3 )et + (−3t − 3t2 − t3 )e−t


3 96
1
(−15t + 15t2 − 6t3 + t4 )et + (15t + 15t2 + 6t3 + t4 )e−t


4 768
1
5 7680 105t − 105t2 + 45t3 − 10t4 + t5 )et

+(−105t − 105t2 − 45t3 − 10t4 − t5 )e−t

Index

adjoint, 435 addition, 404

adjoint inversion formula, 436 augmented, 411
augmented matrix, 411 coefficient, 411
coefficient matrix, 411 column matrix, 405
coefficients, 410 column vector, 405
cofactor, 435 inverse, 427
column expansion, 433 product, 406
Cramer’s Rule, 438 row matrix, 405
dependent variables, 419 row vector, 405
determinant scalar multiplication, 404
definition: 2 × 2 case, 432 variable, 411
definition: general case, 432 vector product, 405
elementary row operations, 434 matrix
properties, 433 adjoint, 435
echelon form, 418 minor, 432
elementary equation operations, 414 nonhomogeneous, 411
elementary row operations, 415 nonsingular, 427
notation, 415 output matrix, 411
equivalent, 413 reduced matrice, 418
free variables, 419 row echelon form, 418
Gauss-Jordon elimination, 423 row expansion, 433
Gaussian elimination, 422 row reduced echelon form, 419
homogeneous, 411 sign matrix, 433
inconsistent, 411 singular, 427
inverse, 427 solution set, 411
inversion computations, 428 system of linear equations, 410
invertible, 427 variable matrix, 411
Laplace expansion formula, 432
leading one, 418 acceleration, 219
leading variables, 419 adjoint formula, 491
linear, 412 algorithm
linear equation, 410 first order linear equations, 53
matrix Fulmer’s method for eAt , 492
590 Index

Picard Approximation, 63 Cosine function, 89

Picard approximation, 458 Cramer’s rule, 257
separable equation, 40
amplitude, 221 damped forced motion, 226
antiderivative, 38 damped free motion, 221
approximate solution, 458 damping force, 217
associated homogeneous differential damping term, 221
equation, 169 dependent variable, 2
associated homogeneous equation, 55 derivative operator, 166
associated homogeneous system, 499 differential equation
exact, 72
balance equation, 20, 451 first order
basis, 469 solution curve, 24
beats, 225 linear
Bernoulli equation, 76 constant coefficient, 49, 57
Bernoulli, Jakoub, 76 first order, 49
Bessel function, 30 homogeneous, 49, 50, 166
first kind, 30 inhomogeneous, 49
modified second kind, 30 nonhomogeneous, 166
order, 2
Cardano’s formula, 28 ordinary, 1
Cardano, Girolamo, 28, 507 partial, 2
Cauchy-Euler equation, 240 separable, 27, 37, 50
fundamental set, 242 solution, 3
characteristic matrix, 490 standard form, 3
characteristic polynomial, 138, 490 system
discriminant, 222 order, 443
Clairaut’s theorem, 74 differential operator
Clairaut, Alexis, 74 vector, 468
coefficient function, 49 dimension, 469
complex conjugate, 509 direction field, 25
complex exponential function, 86, 510 displacement, 216
complex number
exponential form, 510 eigenvalue, 490
imaginary part, 509 eigenvector-eigenvalue pair, 484
real part, 509 English system, 218
complex numbers, 507 equality of mixed partial derivatives, 74
addition, 508 Euler’s formula, 86, 510, 511
division, 509 exactness criterion, 74
multiplication, 508 Existence and Uniqueness Theorem
standard algebraic form, 509 Picard, 64
concentration, 451 existence and uniqueness theorem, 240,
consistent, 411 456
constant function, 83 Existence Theorem
constant of integration, 44 Peano, 65
continuous function, 40, 49, 63 eAt , 459
convolution, 153 calculation by Laplace transform,
convolution product, 503 486, 490
convolution theorem, 154, 224 fundamental matrix for y 0 = Ay, 477
 Index 591

eAt for 2 × 2 matrices, 496 fundamental matrix, 471

eAt for 3 × 3 matrices, 497 linear homogeneous systems
exponential function, 85, 86, 510 dimension of solution space, 469
complex, 87 linear systems
Exponential matrix, 459 linear
exponential type, 349 existence and uniqueness theorem,
468
falling bodies, 13 linearly dependent, 143, 235
forcing function, 49, 166 linearly independent, 143, 235, 469
Fulmer’s method, 492
functional equation, 2 Maple, 25
functional identity, 3 mass, 219
fundamental matrix, 471, 499 Mathematica, 25
criterion for, 473 MATLAB, 25
fundamental set, 239, 256 matrix, 401
Fundamental Theorem of Calculus, 38 cofactor, 435
complex, 402
Galileo Galilei, 12 diagonal, 402
Gamma function, 30 identity, 403
general solution, 55 inverse, 491
general solution set, 55 lower triangular, 403
gravitational force, 218 main diagonal, 402
real, 402
half-life, 43 size, 402
transpose, 403
independent variable, 2 upper triangular, 403
indicial polynomial, 241 zero, 403
conjugate complex roots, 242 matrix
distinct real roots, 241 size, 402
double root, 241 matrix function
initial conditions, 171, 232 continuous, 455
initial value problem, 9, 55, 132, 171, differentiable, 455
232, 445, 456 integrable, 455
input function, 82, 138 Laplace transform, 455
integral equation, 63 product rule for differentiation, 466,
integrating factor, 52, 53 474
interval, 4 metric system, 218
inverse Laplace transform mixing problem, 17, 20, 58, 61, 450
linearity, 117 modulus, 508

Laplace transform, 81, 82 Newton, 218

differentiation formula nullcline, 30
first order, 132
linearity, 83 output function, 83
length, 508
level curve, 72 particular solution, 55
linear combination of solutions, 469 Peano, Guiseppe, 65
linear equation, 55 phase angle, 221
linear homogeneous system Picard approximation algorithm, 457
592 Index

Picard, Émile, 63 solution, 445

polar coordinates, 508 standard form, 445
power function, 84 linear constant coefficient
predator-prey system, 449 exponential solution, 460
proper rational function, 491 linear homogeneous, 468
linear nonhomogeneous, 499
radioactive decay, 43 associated homogeneous system,
radius, 508 499
rate in − rate out, 20 system of ordinary differential
Rate in - Rate out, 451 equations, 443
rate of change, 20, 43
real numbers, 508 Taylor series expansion, 459
restoring force, 216 transform function, 83
Ricatti equation, 30 transient solution, 226
Riccati equation, 65
undamped forced motion, 223
Sine function, 89 undamped free motion, 220
slope field, 24 uniqueness
slug, 218 geometric meaning, 69
solution, 411 units of measurement, 218
constant, 45
steady state solution, 226 variation of parameters, 256, 501
successive approximations, 62 vector space, 55, 469
system of constant coefficient linear basis, 469
differential equations dimension, 469
solution by Laplace transform, 480 Verhulst population equation, 44
system of differential equations
first order Wronskian, 238
autonomous, 447 Wronskian matrix, 237, 257
constant coefficient linear, 447
linear, 447, 454 zero-input solution, 138
linear and homogeneous, 447 zero-state solution, 138