Introduction to Differential Equations

Course Notes for AMath 250

J. Wainwright 1
Department of Applied Mathematics
University of Waterloo

March 9, 2010

1 c J. Wainwright, April 2003

Contents

1 First Order Differential Equations 1

1.1 DEs and Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Newton’s Second Law of Motion . . . . . . . . . . . . . . . . . . . . . 1
1.1.2 Dimensions of physical quantities . . . . . . . . . . . . . . . . . . . . 3
1.1.3 Newton’s Law of Gravitation . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Mathematical aspects of first order DEs . . . . . . . . . . . . . . . . . . . . 7
1.2.1 Types of first order DEs . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.2 Solving separable DEs . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.3 Solving linear DEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.4 Qualitative sketches of families of solutions . . . . . . . . . . . . . . . 12
1.2.5 First order linear DEs with constant coefficient . . . . . . . . . . . . 16
1.2.6 An important special case . . . . . . . . . . . . . . . . . . . . . . . . 19
1.2.7 A common error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.2.8 Initial value problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3 Other applications of first order DEs . . . . . . . . . . . . . . . . . . . . . . 22
1.3.1 Mixing problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.3.2 Population growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.3.3 Epidemics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.3.4 Cooling problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.3.5 Pursuit problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.3.6 Electrical circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2 Dimensional Analysis 33
2.1 Writing physical relations in dimensionless form . . . . . . . . . . . . . . . . 33
2.1.1 Characteristic scales and dimensionless variables . . . . . . . . . . . . 33
2.1.2 The mixing tank DE . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.1.3 The sky-diver DE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.2 Deducing physical relations using dimensional
analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.2.1 A motivating example . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.2.2 Complete sets of dimensionless variables . . . . . . . . . . . . . . . . 41
2.2.3 The Buckingham Pi Theorem . . . . . . . . . . . . . . . . . . . . . . 44

3 Second Order Linear DEs 49

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.1.1 Oscillations and Second Order DEs . . . . . . . . . . . . . . . . . . . 49

i
 3.1.2 The world’s simplest second order DE . . . . . . . . . . . . . . . . . 50
3.1.3 Types of Second Order DE . . . . . . . . . . . . . . . . . . . . . . . . 51
3.1.4 The Initial Value Problem for Second Order DEs . . . . . . . . . . . 52
3.2 Solving Second Order Linear DEs . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2.1 A fundamental property of homogeneous second order linear DEs . . 53
3.2.2 General form of the solution . . . . . . . . . . . . . . . . . . . . . . . 55
3.2.3 General Solution of the Homogeneous DE . . . . . . . . . . . . . . . 56
3.2.4 The method of undetermined coefficients . . . . . . . . . . . . . . . . 59
3.3 Analysis of the oscillator DE . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.3.1 Introductory remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.3.2 Zero driving force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.3.3 Non-zero driving force . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4 The Laplace Transform and DEs 79

4.1 Elementary properties of the Laplace transform . . . . . . . . . . . . . . . . 79
4.1.1 Definition of the Laplace transform . . . . . . . . . . . . . . . . . . . 79
4.1.2 Calculating Laplace transforms . . . . . . . . . . . . . . . . . . . . . 83
4.1.3 The Laplace transform of a derivative . . . . . . . . . . . . . . . . . . 85
4.1.4 The Inverse Laplace Transform . . . . . . . . . . . . . . . . . . . . . 86
4.1.5 The Heaviside Step Function . . . . . . . . . . . . . . . . . . . . . . . 88
4.2 Solving DEs using the Laplace Transform . . . . . . . . . . . . . . . . . . . . 91
4.2.1 First order DEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2.2 Second order DEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2.3 Discontinuous and Non-differentiable Inputs . . . . . . . . . . . . . . 95

5 Linear Vector Differential Equations 99

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.1.1 Coupled Mixing Tanks . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.1.2 The mechanical oscillator . . . . . . . . . . . . . . . . . . . . . . . . 100
5.1.3 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.1.4 Linearity and Superposition . . . . . . . . . . . . . . . . . . . . . . . 103
5.2 Solving vector DEs using eigenvectors . . . . . . . . . . . . . . . . . . . . . . 104
5.2.1 The method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.2.2 Unequal real eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.2.3 Complex eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.2.4 Equal real eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.3 Solving vector DEs using the Laplace transform . . . . . . . . . . . . . . . . 111
5.3.1 The method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.3.2 An example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.3.3 The Fundamental Matrix . . . . . . . . . . . . . . . . . . . . . . . . . 114
5.4 Orbits of a vector DE in state space . . . . . . . . . . . . . . . . . . . . . . . 116
5.4.1 Phase portraits: unequal real eigenvalues . . . . . . . . . . . . . . . . 117
5.4.2 Phase portraits: complex eigenvalues . . . . . . . . . . . . . . . . . . 120
5.4.3 Long term evolution and stability . . . . . . . . . . . . . . . . . . . . 122
5.5 Solving inhomogeneous linear vector DEs . . . . . . . . . . . . . . . . . . . . 123
5.5.1 Properties of the Fundamental Matrix . . . . . . . . . . . . . . . . . 124

ii
 5.5.2 The method of variation of parameters . . . . . . . . . . . . . . . . . 126
5.5.3 An example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

Epilogue 129

References 131

Problem Sets 133

iii
iv
Acknowledgements

I would like to thank Professor Bev Marshman for a critical reading of the manuscript and
Professors Frank Goodman and Pino Tenti for helpful discussions concerning the physical
interpretation of differential equations, and for allowing me to use their notes concerning di-
mensional analysis. Thanks are also due to Ann Puncher for her excellent work in typesetting
the notes.

v
 your knowledge of Calculusneeds refreshing. There are five problem sets,one for each
chapter in the notes, from which assignedquestions will be selected. The answers for some
ofthe routine problems are given.

vi
Chapter 1

First Order Differential Equations

1.1 DEs and Mechanics

1.1.1 Newton’s Second Law of Motion
Newton’s Second Law of Motion leads to differential equations when applied to various
problems in mechanics. We consider the simplest case of motion of a particle in a straight
line. The relevant physical quantities are

m, the mass of the particle (which may vary with time),

v(t), the velocity of the particle at time t,
F, the total force acting (time dependent, in general).

Newton’s Second Law states that the rate of change of momentum mv(t) equals the total
force, i.e.
d
[mv(t)] = F. (1.1)
dt
If the mass is constant this equation can be written
dv
m = F, (1.2)
dt
i.e. mass times acceleration equals force acting. Before (1.1) or (1.2) can be used to describe
the motion of a particle, the force F has to be specified. In the example to follow, F depends
on t through the velocity v, i.e. F = F (v), in which case (1.2) assumes the form
dv
m = F (v),
dt
which is a first order differential equation (DE) for the unknown function v(t).

Comment: Newton’s second law has been tested in countless experiments, and accurately
predicts the motion of particles, subject to one limitation, namely that the velocity of the
particle is small compared to the velocity of light i.e.
v
≪ 1.
c
1
The velocity of light is c ≈ 3 × 108 m/s ≈ 109 km/hr. For sufficiently high velocities, as occur
for example in high energy particle accelerators, Newton’s second law has to be replaced by
its relativistic counterpart, which is part of Einstein’s theory of relativity.

Example: (Terminal velocity of a sky-diver)

Gravity exerts a downward force on the sky-diver, and air-resistance exerts an upward
force, which increases as the velocity increases. Eventually we expect that air-resistance will
balance the force due to gravity, so that the total force acting is zero. Then by (1.2) the
sky-diver will fall with constant velocity, called the terminal velocity. In order to describe
the motion in detail we need to specify the force due to air-resistance. This is a complicated
matter; we make the simple assumption, supported by experiment for subsonic velocities,
that this force is proportional to the velocity and acts so as to oppose the motion. Taking
the downward direction as the positive direction, the total force acting on the sky-diver will
be
F = mg − αv, (1.3)
where g is the constant acceleration due to gravity (at the earth’s surface). The constant α,
which depends on the physical characteristics of the sky-diver or falling object, is called the
drag coefficient. 

Comment: Strictly speaking, the acceleration due to gravity is not constant, and depends
on the distance from the centre of the earth. Since the height h above the earth’s surface is
small compared to the earth’s radius R, i.e.
h
≪ 1,
R
it is a reasonable approximation to treat g as a constant. 

With (1.3), Newton’s law (1.2) assumes the form

dv
m = mg − αv, (1.4)
dt
a first order DE for v(t). This DE can be solved by separation of variables (see Section
1.2.2), but for now we draw a conclusion directly from the DE. As stated earlier we expect
the sky-diver to eventually reach a constant terminal velocity. Since a constant velocity gives
dv
dt
= 0, equation (1.4) implies that
mg
vterminal = . (1.5)
α
Note that the terminal velocity depends on the three physical parameters m, g and α, being
inversely proportional to the drag coefficient α (opening a parachute will increase α, thereby
reducing vterminal ). There is one other parameter associated with the physical system, namely
the sky-diver’s initial velocity v(0), i.e. the vertical velocity when leaving the plane at time
t = 0. It is of interest that vterminal does not depend on v(0), although we expect that v(t),
the velocity at time t, will depend on v(0). The relation between v(t) and v(0) will become
clear when you solve the DE (1.4) [see Problem Set 1].

2
1.1.2 Dimensions of physical quantities
In analyzing any physical system it is essential to keep track of the physical dimensions of
the various variables and parameters. In mechanics, the primary physical variables are taken
to be
mass M, length L, time T.
All physical quantities have dimensions of the form

M m Lℓ T t ,

where the exponents m, ℓ and t are integers. For example, velocity v has dimensions LT −1
(i.e. m s−1 , km hr −1 , depending on the units). We indicate the dimensions by writing

[velocity] = LT −1 .

Since acceleration is the rate of change of velocity with respect to time, we have

[acceleration] = LT −2 .

By Newton’s Second Law,

force = (mass)(acceleration),
and so
[force] = [mass][acceleration] = MLT −2 . (1.6)
In calculating the dimensions of force, we have implicitly used certain consistency re-
quirements that have to be satisfied by equations having physical content. These are as
follows.

D1: One can only add, subtract or equate physical quantities that have the same dimension.

This statement is referred to as the Principle of Dimensional Homogeneity.

D2: Quantities having different dimensions can only be combined by multiplication and
division, and the dimensions of a product or quotient are given by
 
A [A]
[AB] = [A][B], = ,
B [B]
where the right sides are simplified using the usual laws of exponents.

These consistency requirements are of importance for the following reasons.

1) D1 and D2 act as a check on the working in a physical problem. For example, if we

have obtained the formula
A = B + CD, (1.7)
then if we are correct the dimensions of A, B, C and D must satisfy

[A] = [B] = [C][D]. (1.8)

Of course if (1.8) holds this does not imply that (1.7) is correct: the point is that if
(1.8) does not hold then (1.7) is incorrect. 

3
 2) D1 and D2 enable us to calculate the dimension of a physical quantity.
As an illustration, we calculate the dimensions of the drag coefficient α in (1.3). The
drag force fd is given by fd = −αv. Hence by D1,

[αv] = [force],

and by D2,
[α][v] = [force].
Using (1.6) and the laws of exponents,

[force] MLT −2
[α] = = = MT −1 . 
[v] LT −1

3) D2 enables us to form dimensionless quantities, which are essential for writing physical
relations in their simplest forms.
To indicate that a quantity A is dimensionless, we write

[A] = 1.

For example, any number is dimensionless, e.g. [π] = 1. In discussing the validity of
Newton’s second law we imposed the restriction vc ≪ 1, which involves the dimension-
less quantity vc :
h v i [v] LT −1
= = = 1. 
c [c] LT −1

Comment: If t is time, i.e. [t] = T , then the expression sin(πt) does not make sense, since
the argument of sin( ) must be dimensionless. However, if ν is a frequency i.e. [ν] = T −1 ,
then 2πνt is dimensionless, and it makes sense to write sin(2πνt). Likewise, the arguments
of all other elementary functions must be dimensionless, apart from the power function
f ( ) = ( )n , n an integer (see D2). 

Exercise 1: If a particle moves in a straight line with constant acceleration g, starting

with velocity v0 and position s0 at time t = 0, the distance travelled at time t is

s = v0 t + 12 gt2 + s0 .

Verify that this formula is dimensionally consistent.

Exercise 2: The kinetic energy of a particle of mass m moving with constant velocity v
in a straight line is given by
E = 12 mv 2 .
Calculate the dimensions of energy.

Answer: [Energy] = ML2 T −2 . 

4
1.1.3 Newton’s Law of Gravitation
Will a projectile fired vertically upwards on the earth’s surface (or on the surface of the moon)
eventually fall back to earth, or will it continue travelling away indefinitely? The answer
is that it depends on the velocity with which the projectile is fired. If this initial velocity
exceeds a certain threshold called the escape velocity, then the projectile will travel away
indefinitely and “escape” from the earth’s gravitational field. As with the sky-diver, this
problem is governed by Newton’s Second Law, and also involves gravity. The key difference
is the distance scale. In the present case the distance from the earth’s surface will not be
small compared to the earth’s radius, and so it is unreasonable to treat the acceleration due
to gravity as a constant. We thus need to use Newton’s Law of Gravitation. On the other
hand, in giving a simple analysis, it is reasonable to neglect air-resistance, since the thickness
of the earth’s atmosphere is small compared to the earth’s radius (of course air resistance is
totally absent on the moon).
Newton’s Law of Gravity states that the force of attraction between two point particles
of mass m1 and m2 is proportional to the masses and inversely proportional to the square of
the distance r between them. For motion in one dimension the force is given by
Gm1 m2
F = , (1.9)
r2
where G is a constant of proportionality called the gravitational constant.

Exercise: Calculate the dimensions of G.

Answer: [G] = M −1 L3 T −2 .

Comment: We can idealize the projectile as a point particle, but not the earth. However,
it can be shown that the gravitational force exerted by a finite homogeneous sphere on a
particle is the same as if all the mass of the sphere was concentrated at the centre of the
sphere. Thus, as regards Newton’s Law of gravitation, the earth can be idealized as a point
particle. 

For a vertically moving projectile the total force acting is

GmM
F =− , (1.10)
r2
where m is the mass of the projectile and M is the earth’s mass, and r is the distance of the
projectile from the centre of the earth. We have chosen the positive direction to be vertically
up.

Comment: The acceleration g due to gravity near the earth’s surface can be related to the
gravitational constant G. Set r = R, the earth’s radius, in (1.10) to obtain
GmM
F =− = −mg,
R2
giving
GM
g= .  (1.11)
R2
5
The equation of motion for the projectile is now obtained by substituting (1.10) in Newton’s
Second Law (1.2), giving
dv GmM
m =− ,
dt r2
where v is the velocity of the projectile. We simplify this equation by cancelling m, and
using (1.11) to express G in terms of g, giving

m
1
0
0 v
1
r 0
1

11R
00
00
11
O 0
1
M

Figure 1.1:

dv gR2
=− 2 . (1.12)
dt r

Comment: In cancelling m, we are tacitly using a fundamental experimental fact about

gravity. The mass m that appears in Newton’s Second Law (1.2) is the inertial mass mI of
the particle, while the mass m that appears in Newton’s Law of Gravity is the gravitational
mass mG . It has been determined experimentally to a high degree of accuracy that

mI = mG ,

although no explanation is known. 

Equation (1.12) contains two unknown functions v(t) and r(t). We can write it as a DE
in one unknown in two ways.
First, since v = dr
dt
, we can write

d2 r gR2
= − ,
dt2 r2
which is a second order DE for r(t). The initial conditions at the launch at time t = 0 are

dr
r(0) = R, (0) = vinit ,
dt
where vinit is the initial velocity.

6
 Second, if we consider velocity v as a function of distance r, we can write (1.12) as a first
order DE, which is more convenient for determining the escape velocity. By the Chain Rule

dv dv dr dv
= = v,
dt dr dt dr

so that (1.12) becomes

dv gR2
v =− 2 , (1.13)
dr r
a first order DE for v = v(r), with initial condition

v(R) = vinit . (1.14)

The problem now is: for which values of vinit will the velocity v(r) satisfy v(r) > 0 for all
r ≥ R. The escape velocity is the smallest value of vinit with this property. In order to solve
this problem we need to solve the DE (1.13) [see Problem Set 1].

1.2 Mathematical aspects of first order DEs

1.2.1 Types of first order DEs
The general form of a first order DE is

dy
= f (x, y), (1.15)
dx

where f is a function of two variables.1 The variable y represents the unknown function,
y = y(x), and x is the independent variable. A solution of the DE (1.15) is a differentiable
function φ such that y = φ(x) satisfies (1.15) for all x in some interval.

Exercise: Verify that

2
y = Ce−x + x2 − 1, x ∈ R,

where C is a constant, is a solution of the DE

dy
= −2xy + 2x3 . 
dx

In general, it is not possible to actually find solutions of the DE (1.15), even though we
know they exist. Fortunately, the first order DEs that arise in many applications are of two
special types that can be solved, namely separable and linear.
1
It is usually assumed that f has continuous partial derivatives (i.e. is of class C 1 ); technical details such
as these are not important in AM250.

7
Separable first order DEs
The general form of a first order separable DE is
dy
= A(x)B(y), (1.16)
dx
where A(x) and B(y) are arbitrary functions.

Example 1: The sky-diver DE in Section 1.1.1,

dv
m = mg − αv, (1.17)
dt
is of the form
dv
= A(t)B(v),
dt
with
α
A(t) = 1, B(v) = g − v,
m
and hence is separable. 

Example 2: The escape velocity DE in Section 1.1.3,

dv gR2
v =− 2 , (1.18)
dr r
is of the form
dv
= A(r)B(v),
dr
with
R2 g
A(r) = , B(v) = − ,
r2 v
and hence is separable. 

Linear first order DEs

The general form of a first order linear DE is
dy
= k(x)y + f (x), (1.19)
dx
where k(x) and f (x) are arbitrary functions.

Example: The DE
dy
= xy − e−x
dx
is linear, but the DE
dy
= xy 2
dx
is non-linear. Note that the sky-diver DE (1.17) is linear, but the escape-velocity DE (1.18)
is non-linear. 

There are two special types of linear DE:

8
 • homogeneous,
dy
= k(x)y (1.20)
dx
This DE is also separable.

• constant coefficient,
dy
= ky + f (x), k = constant. (1.21)
dx

A linear DE that is homogeneous AND has a constant coefficient of the form

dy
= ky, k = constant. (1.22)
dx

This is the world’s simplest and most important first order DE. It is so simple that it can
be solved by inspection: the derivative of the unknown function y is k times y, and the only
functions with this property are

y = Cekx , C = constant. (1.23)

This is important enough to write out formally.

Proposition: Any solution of the DE

Aside: If k > 0, this DE describes
dy exponential growth and if k < 0, it
= ky, k = constant
dx describes exponential decay.
is given by
y = Cekx ,
where C is a constant.

Proof: Multiply (1.22) by e−kx and rewrite as

d −kx

e y = 0.
dx

Hence,
e−kx y = C,

giving (1.23). 

Exercise: Write the general form of a linear DE that is also separable.

dy
Answer: = k(x)(ay + b), where a and b are constants.
dx
9
1.2.2 Solving separable DEs
A separable DE
dy
= A(x)B(y) (1.24)
dx
can be solved by separation of variables, as follows. Divide (1.24) by B(y) and take the
antiderivative with respect to x:

1 dy
Z Z
dx = A(x)dx.
B(y) dx

By the change of variable algorithm, this can be written

1
Z Z
dy = A(x)dx.
B(y)

Provided that both antiderivatives can be evaluated in terms of elementary functions, one
obtains a one-parameter family of solutions i.e. depending on one constant of integration.


Equilibrium solutions:
A separable DE may have certain exceptional solutions that can be found by inspection. If
the function B(y) in (1.24) is zero at y = b, i.e.

B(b) = 0,

then the constant function y = b is a solution of the DE (1.24). Because the unknown
function is a constant function, this solution is called an equilibrium solution – one thinks of
the physical system as being in a state of equilibrium. For example the separable DE

dy
= y(1 − y)
dx
has equilibrium solutions y = 0 and y = 1. When solving a separable DE, always begin by
finding the equilibrium solutions (if any), because they are excluded by the general procedure
since one divides by B(y).

Example: The separable DE

dy xy
=
dx 1 + y2
has an equilibrium solution y = 0. To find the general solution, rewrite the DE as

1 + y 2 dy
= x.
y dx

This leads to Z  
1
Z
+ y dy = xdx,
y

10
giving
ln | y | + 21 y 2 = 12 x2 + C. 

Comment: When solving a separable DE it may not be possible to isolate y in the general
solution, as happens in the above example.

Exercise: Solve the DE

dy
= 2xe−y .
dx

Answer: y = ln(x2 + C), where C is a constant.

1.2.3 Solving linear DEs

A linear DE
dy
= k(x)y + f (x)
dx
can be solved by finding an auxiliary function called an integrating factor. The method is
illustrated in the following example.

Example: Solve the DE

dy
= y − e−x . (1.25)
dx

Solution: Transfer the y-term to the left side and multiply throughout by a function I(x),
dy
I + y(−I) = −Ie−x . (1.26)
dx
Choose I to satisfy
dI
= −I. (1.27)
dx
The reason for doing this is that (1.26) becomes
dy dI
I +y = −Ie−x ,
dx dx
which, using the Product Rule for derivatives, can be written
d
(Iy) = −Ie−x . (1.28)
dx
The DE (1.27) for I (the “world’s simplest”) can be solved by inspection:

I = Ce−x .

Since we only want a particular solution, we choose I = e−x (i.e. C = 1) for simplicity. The
DE (1.28) assumes the form
d −x

e y = −e−2x . (1.29)
dx
11
Take the antiderivative of both sides with respect to x:
d −x

Z Z
e y dx = −e−2x dx,
dx
giving
e−x y = 21 e−2x + C.
Solving for y:
y = 12 e−x + Cex , (1.30)
where C is an arbitrary constant. Equation (1.30) gives the family of all solutions of the DE
(1.25). 

Comment: The function I(x) in the previous solution is called an integrating factor for the
DE. It is always determined by solving a separable DE, which in the above example could
be solved by inspection (see equation (1.27)). The purpose of finding the integrating factor
is to write the given DE in the form (1.28), since in this form it can be solved directly by
taking the antiderivative of both sides. Note that any DE of the form
d(I(x)y)
= g(x)
dx
can be solved for y directly, by taking the antiderivative of both sides with respect to x. 

Exercise: Solve the linear DE

dy
x − y = x3 , x > 0.
dx
dy
Comment: It is essential to divide through by x so as to isolate dx
, before attempting to
find the integrating factor.

Answer: The integrating factor is I(x) = x1 , and the general solution is y = 12 x3 + Cx. 

1.2.4 Qualitative sketches of families of solutions

Two basic problems in the theory of DEs are
A: Solve the DE, i.e. find all solutions of the DE (only possible for certain classes of DE).
B: Describe the behaviour of typical solutions of a DE, e.g. how does the solution y(x)
behave as x → +∞?
In 1.2.2 and 1.2.3 we discussed the two most important special classes of first order DEs that
can be solved. In order to understand the behaviour of solutions, however, it is necessary to
give a qualitative sketch of the family of solutions, which will depend on one parameter (the
constant of integration). In this section, we discuss how to draw such a sketch.
When sketching the solutions one can obtain useful information directly from the DE,
without solving it, as follows. We think of the DE
dy
= f (x, y)
dx
12
as specifying a slope at each point of the xy-plane, namely the slope of the tangent line to
the solution y = y(x) through that point. In other words, if y = y(x) is the solution curve
through the point (x0 , y0 ), then the slope of the curve at (x0 , y0 ) is
dy
= f (x0 , y0 ).
dx x=x0

Example 1: For the DE

dy
= −2y (1.31)
dx
the family of solutions is
y = Ce−2x
(by inspection). Note that y = 0 is an equilibrium solution, with slope 0 at each point. The
special property of this DE is that the slope at (x, y) depends on y but not on x. 

slope is
−2y1 < 0
y = y1

y=0 x

y = y2

slope is
−2y2 > 0

Figure 1.2: One-parameter family of solutions

Here is an example where the pattern of the solution curves is more complicated, and
where we make use of the DE itself.

Example 2: Give a qualitative sketch of the solution curves of the DE

dy
= y − e−x .
dx
From Section 1.2.3, equation (1.30), the solutions are

y = 21 e−x + Cex .

Solution: The DE implies that the slope is zero at all points of the curve

y = e−x

13
(drawn as the dashed curve). The solution with C = 0 plays a special role (draw it in
colour). As x → −∞, y ≈ 21 e−x , i.e. all solutions approach the C = 0 solution. As
x → +∞, y ≈ Cex , and so the shape of the solution curve depends on whether C > 0 or
C < 0. 

y = 12 e−x y
C=0
C >0

y = e−x

C<0

Figure 1.3:

Exceptional solutions:
The key to sketching the family of solution curves is to identify any exceptional solutions. In
example 1, y = 0 is an exceptional solution, and in example 2, y = 12 e−x is exceptional. In
each case the exceptional solution divides the whole family of solutions into two subclasses
such that the members of a subclass have the same qualitative properties (i.e. the same
overall shape). Another feature of exceptional solutions is that they often “attract” other
solution curves, either as x → +∞ or as x → −∞.
dy
Note that for a separable DE dx = A(x)B(y), any equilibrium solution, i.e. y(x) = b,
with B(b) = 0, is an exceptional solution (as in example 1). 

A fundamental property of families of solutions:

Consider the DE
dy
= f (x, y).
dx
A fundamental theorem in the theory of DEs (the Existence-Uniqueness theorem, discussed
in AM351) states that if the function f (x, y) has continuous partial derivatives (i.e. f is of
class C 1 , in brief), then through a given point (x0 , y0) there passes one and only one solution
curve of the DE. This means that
dy
if f is of class C 1 , solution curves of the DE dx
= f (x, y) cannot intersect.

14
 This fact is very helpful when sketching families of solution curves.

Here are two examples where f is not of class C 1 and intersections do occur.

Example 3: Consider the DE

dy y
=
dx x
The solution curves are y = Cx. Here f (x, y) = xy , which is not C 1 when x = 0 (f (0, y) is
not even defined). 

x = 0 is
excluded

Figure 1.4:

Example 4: Consider the DE

dy
= 3y 2/3
dx
The solution curves are
y = (x + C)3 ,

together with the equilibrium solution

y = 0.

Here f (x, y) = 3y 2/3 , which is not C 1 when y = 0 since

∂f 2
= 1/3 .
∂y y

Intersections occur on the x-axis. 

15
 y
C=0

C >0 C<0

y=0

Figure 1.5:

1.2.5 First order linear DEs with constant coefficient

The general form of a linear DE with a constant coefficient is

dy
+ ky = f (x), (1.32)
dx
where k is a constant. The general solution of such a DE can be found by obtaining an
integrating factor as in Section 1.2.3. For certain functions f (x) that arise commonly in
applications, however, there is a quicker method for finding a particular solution (i.e. a
single solution) of the DE (1.32), called the method of undetermined coefficients.2 We first
illustrate this method, and then show how to use it to find the general solution of (1.32)
efficiently.

The method of undetermined coefficients:

Example 1: Find a particular solution of

dy
+ 2y = x. (1.33)
dx

Solution: Consider a trial function of the form

y = Ax + B, (1.34)

where A and B are constants (the “undetermined coefficients”). Substitute (1.34) in (1.33):

A + 2(Ax + B) = x.
2
This method can also be applied to second order linear DEs with constant coefficients. See Section 3.2.4.

16
Equate coefficients of x1 and x0 = 1, giving

2A = 1
A + 2B = 0.

Solve for A and B: A = 21 , B = − 14 .

By (1.34), a particular solution of the DE (1.33) is

yp = 21 x − 14 . 

Comment:
1) The trial function contains a number of constants, and the DE leads to a system of
linear algebraic equations to be solved for these constants.

2) Choosing a trial function requires some experience. If one doesn’t include enough terms
and constants, the system of linear equations will be incompatible, and one has to try
again. The method can only be applied if f (x) is composed of powers of x, sin ωx,
cos ωx and erx . The table below shows some simple cases for the DE (1.32).

f (x) trial function y

x Ax + B
x2 Ax2 + Bx + C
sin ωx or cos ωx A sin ωx + B cos ωx
erx (r 6= −k) Aerx
e−kx Axe−kx

Exercise: Find a particular solution of

dy
+ 2y = cos x.
dx

Answer: yp = 15 (2 cos x + sin x). 

Suppose we have found a particular solution yp of the DE (1.30). We can then find the
general solution of this DE without any extra work. The following Proposition shows how.
The proposition is more general than we need, since it does not depend on the coefficient k
in the DE being constant.

Proposition: If yp is a particular solution of the DE

dy
+ k(x)y = f (x),
dx
17
and y is any solution of this DE, then the difference
yh = y − yp (1.35)
is a solution of the homogeneous DE
dy
+ k(x)y = 0.
dx

Proof: We are given that

dy
+ k(x)y = f (x),
dx
and
dyp
+ k(x)yp = f (x).
dx
Subtract these two equations to get
d
(y − yp ) + k(x)(y − yp ) = 0,
dx
which is the desired result. 

It follows from (1.35) that the general solution y is the sum of two terms, y = yp + yh .
This is the key result, which we now write out in full.

General solution of a linear DE:

The general solution of the linear DE
dy
+ k(x)y = f (x) (1.36)
dx
has the form
y(x) = yp (x) + yh (x), (1.37)
where yp (x) is a particular solution of the inhomogeneous DE (1.36), and yh (x) is the general
solution of the homogeneous DE
dy
+ k(x)y = 0.  (1.38)
dx

Comment: The proposition depends in an essential way on the fact that the DE is linear.
In the constant coefficient case, the homogeneous DE (1.38) is
dy
+ ky = 0,
dx
(the “world’s simplest”) whose general solution we know to be
y = Ce−kx .
Thus, once we have obtained a particular solution yp (x) of (1.32) using the method of
undetermined coefficients, we can immediately write down the general solution using (1.37).


18
Return to example 1:
Find the general solution of
dy
+ 2y = x. (1.39)
dx

Solution: We have found a particular solution

yp (x) = 21 x − 14 .

The general solution of the homogeneous DE (the “world’s simplest”)

dy
+ 2y = 0
dx
is
yh (x) = Ce−2x .
Thus by (1.37), the general solution of the DE (1.39) is

y(x) = 12 x − 41 + Ce−2x .




Exercise: Find the general solution of

dy
+ 2y = cos x.
dx

Answer: y = 15 (2 cos x + sin x) + Ce−2x . 

Exercise: Find the general solution of

dy
− 3y = e2x .
dx

Answer: y = −e2x + Ce3x . 

1.2.6 An important special case

Consider the DE
dy
= ky + b, (1.40)
dx
where k and b are constants, with k 6= 0. This DE is linear (since the only y-dependence
on the right side is y itself), has a constant coefficient (since k, the coefficient of y, is a
constant), and is also separable (since b is a constant).
This DE can thus be solved using each of the techniques that we have introduced so far,
and we suggest that you do this as an exercise:

1) Solve (1.40) as a linear DE by finding an integrating factor.

19
 2) Solve as a separable DE.

3) Find a particular slution using the method of undertermined coefficients (use y = A,

a constant as the trial function), and then write down the general solution.
The recommended method for solving (1.40), and by far the quickest method, is to convert
the DE into the “world’s simplest” form and solve by inspection as follows. Since b and k
are constants and k 6= 0 we can rewrite (1.40) in the form
   
d b b
y+ =k y+ . (1.41)
dx k k
If we define
b
u=y+ (1.42)
k
this DE becomes
du
= ku,
dx
whose solution is
u = Cekx , (1.43)
where C is a constant. Using (1.42), we get
b
y+ = Cekx , (1.44)
k
Thus
b
y = − + Cekx
k
is the general solution of the DE (1.40).
In practice, there is no need to formally introduce u. Having rewritten the DE in the
form (1.41) one can immediately write down the solution (1.44).

1.2.7 A common error

Consider a first order DE
dy
= f (x, y). (1.45)
dx
Knowing that
dy
Z
dx = y + C,
dx
a naive person might be tempted to try to solve (1.45) by taking the antiderivative of both
sides with respect to x, obtaining
Z
y + C = f (x, y)dx.

This attempt to solve (1.45) fails in general because the antiderivative on the right hand side
contains y, which depends on x and hence cannot be treated as a constant (y is of course the
unknown function which we are trying to find).

20
 We note that this simple-minded approach only works in the very speical case where the
DE (1.45) has the form
dy
= f (x), (1.46)
dx
i.e. the right hand side is independent of y. We shall say that the DE (1.46) is directly
solvable, because it can be solved simply by taking the antiderivative.
Its worth noting that in solving a linear DE by finding an integrating factor, one trans-
forms the DE into a directly solvable form (see the Comment after equation (1.30)). In
particular, in the example in Section 1.2.3, we converted the DE
dy
= y − e−x
dx
into the directly solvable form
d −x
(e y) = −e−2x
dx
[see equations (1.25) and (1.29)].

1.2.8 Initial value problems

The sky-diver DE,
dv
m = mg − αv,
dt
determines the velocity of a sky-diver as a function of time t. We have seen that a first order
DE has a one-parameter family of solutions (the constant of integration is the parameter).
When applying the sky-diver DE, the physical process being described will begin at the time
when the sky-diver jumps from the plane. We label this time as t = 0, and so we have the
initial condition
v(0) = 0.
This initial condition will determine the constant of integration, leading to a unique solution
that gives the velocity of the sky-diver at time t.
In general, for a DE
dy
= f (x, y), (1.47)
dx
the initial condition will be of the form
y(x0 ) = y0 . (1.48)
where x0 and y0 are given constants. The equations (1.47) and (1.48) are said to define an
initial value problem.

Example: Find the unique solution of the initial value problem

dy
= −2y, y(0) = 3.
dx

Solution: By inspection the general solution of the DE is

y = Ce−2x

21
Substituting x = 0 and y = 3 gives
3 = Ce0 ,
i.e. C = 3. Thus the unique solution is

y = 3e−2x . 

Comment: The general solution of a first order DE (1.47) corresponds to a one-parameter

family of solution curves. The initial condition (1.48) picks out a unique curve, namely the
curve that passes through the point (x0 , y0 ).

Exercise: Find the unique solution of the initial value problem

dy
= −2xy 2 , y(1) = 31 .
dx

1
Answer: y= . 
x2 +2
Exercise: Find the unique solution of the initial value problem
dy
x2 = −2xy + 1, y(1) = 0.
dx

1 1
Answer: y= x
− x2
. 

1.3 Other applications of first order DEs

1.3.1 Mixing problems
Various problems in biology and engineering can be put in the following framework. Consider
a tank containing a chemical solution. The contents are kept well-mixed, so that the con-
centration is uniform. There is an inflow of the chemical solution of specified concentration,
and and outflow of chemical solution, whose concentration at time t equals the concentration
of solution in the tank at time t.
The goal is to predict the amount of chemical in the tank at time t, or perhaps to adjust
the inflow and outflow as as to achieve a desired concentration in the tank.
Let m(t) denote the amount of chemical in the tank at time t. The rate of change of
m(t) equals the difference between the rate of inflow and rate of outflow:
dm
= rin − rout , (1.49)
dt
where rin is the rate at which chemical is added by the inflow and rout is the rate at which
chemical is removed by the outflow. By the Principle of Dimensional Homogeneity, each
term in equation (1.39) will have the same dimensions, namely MT −1 (i.e. mass of chemical
per unit time). We shall refer to equation (1.49) as the mass balance equation for a mixing
tank.

22
 mixer

rin
inflow

rout
outflow
m(t)

Figure 1.6:

Example:
A tank contains m0 kg of salt dissolved in 100 litres of water. A salt solution containing
1
4
kg per litre is added at 3 litre/min., and the well-stirred mixture leaves the tank at the
same rate. Find the amount of salt in the tank at time t.

Solution: From the given data, the rate at which salt is added to the tank is

rin = (3) 14 kg min.−1 .

Let m(t) denote the amount of salt in the tank at time t. Then the concentration at time t
is
m(t)
kg/litre
100
Thus, the rate at which salt leaves the tank is
 
m(t)
rout = (3) kg min.−1 .
100

The mass balance equation (1.49) gives

dm 3 3 3
= 4
− 100
m = − 100 (m − 25). (1.50)
dt
The initial condition is
m(0) = m0 . (1.51)
The DE (1.50) can be written

d 3
(m − 25) = − 100 (m − 25),
dt
23
and hence can be solved by inspection:
3
m − 25 = Ce− 100 t .
On setting t = 0, the initial condition (1.51) gives C = m0 − 25, leading to the solution
3
m(t) = 25 + (m0 − 25)e− 100 t . (1.52)

Interpretation: Based on the physical set-up, one expects that as time passes the con-
centration of the solution in the tank will approach the concentration of the inflow i.e. 41
kg/litre. Thus the amount of salt in the tank will approach 41 (100) = 25 kg as t → ∞, in

agreement with equation (1.52). 

Comment: One can imagine problems such as the above arising in different contexts, e.g.
(1) nutrients flowing into and out of a cell (which plays the role of the tank),
(2) carbon-monoxide seeping into a room and then being dispersed.

Overview:
In a mixing tank problem, the unknown function is the mass of chemical in the tank at time
t, denoted by m(t), with dimensions
[m(t)] = M.
There are two flow rates, the inflow rate fin and the outflow rate fout , with dimensions
[fin ] = [fout ] = L3 T −1 .
The flow rates fin and fout are given, and in general could be functions of time t, but in
simple problems they will be constants, and may even be equal. If they are equal, then the
volume V of solution in the tank will be constant in time.
There are two concentrations, the concentration of the inflow cin and the concentration
of the outflow cout , with
[cin ] = [cout ] = ML−3 .
The inflow concentration is given, and will be constant in the simplest situation. The outflow
concentration is the concentration of the solution in the tank at time t, and is hence given
by the key relation
m(t)
cout = ,
V (t)
where V (t) is the volume at time t. Finally, the rates of mass inflow rin and mass outflow
rout that appear in the mass balance equation (1.49) are given by
rin = cin fin , rout = cout fout .
Verify that these equations are dimensionally consistent.
References: Boyce & Diprima, pg. 51, #18-22.
Braun, pg. 56, #6-11.

24
1.3.2 Population growth
Let x(t) be the population of some species (e.g. fish, bacteria, etc.) at time t. The simplest
hypothesis is the Malthusian Law i.e. that the rate of change of x at time t is proportional
to the population at time t:
dx
= rx, (1.53)
dt
where r is a constant with [r] = T −1 , called the rate of growth (rate of decline, if r < 0). By
inspection the general solution of the DE (1.53) is
x(t) = Cert ,
and the initial condition x(0) = x0 leads to
x(t) = x0 ert ,
describing exponential growth (r > 0) or decay (r < 0).
It is clear that exponential growth can only continue for a restricted time due to resource
limitations. A more realistic model takes into account that a given environment can support
at most a finite number of a particular species, denoted by K, and called the carrying
capacity. A realistic model thus requires that the rate of change dxdt
approach zero as x gets
close to K. A simple way to accomplish this is to assume that the rate of growth is not
simply a constant r as in (1.53), but depends on x according to
 x
r 1− ,
K
where K is the constant carrying capacity. The DE governing the population x(t) then
assumes the form
dx  x
=r 1− x, (1.54)
dt K
called the logistic equation, a separable, non-linear DE.
Comment: The variable x is dimensionless, [x] = 1, since it represents a number. There is,
however, a choice of scale associated with x, since one can measure x in thousands (say) or
millions (if the population is very large). The DEs (1.53) and (1.54) are in fact independent
of the choice of scale, since if x (and K) are multiplied by the same constant factor, the DEs
are unchanged.
Exercise: Solve the DE (1.54) using two different methods:
x
(i) by separation of variables (you will find that letting u = K
, a simple rescaling of x,
simplifies the algebra);
(ii) by making the change of variable y = Kx . This has the effect of changing (1.54) into
the simple form dtd (y − 1) = −r(y − 1), which can be solved by inspection.

Answer: The solution can be written in the form

x0 ert
x(t) = ,
1 − xK0 + xK0 ert
after imposing the initial condition x(0) = x0 .

25
1.3.3 Epidemics
Consider a population of N individuals containing a number of individuals having an in-
fectious disease. The goal is to determine how rapidly the disease will spread if no control
measures are taken.
Let x(t) denote the number of infectious individuals at time t, and let N − x(t) denote
the number of susceptible individuals. Assume that the rate of spread of the disease, dx dt
,
is proportional to the number of contacts between infectious and susceptible individuals.
Assume that both groups mingle freely so that the number of contacts is proportional to
x(N − x). Then x(t) satisfies
dx α
= x(N − x), (1.55)
dt N
where α is a constant with [α] = T −1 . We have written the constant of proportion as Nα , in
order that the DE remain unchanged if x and N are rescaled (see the comment in Section
1.3.2). Equation (1.55) is a first order DE for x(t), with initial condition x(0) = x0 , which
can be solved by separation of variables.

Reference: Boyce & DiPrima, page 65.

1.3.4 Cooling problems

Consider an object whose temperature at time t is T (t), in surroundings whose temperature
is TA (called the ambient temperature). If T (t) is larger than TA the object will lose heat to
its surroundings and its temperature will decrease. Newton’s Law of Cooling states that the
rate of decrease is proportional to the difference between T (t) and the ambient temperature
TA :
dT
= −k(T − TA ), (1.56)
dt
where k is a constant with dimensions [k] = T −1 . (Newton’s Law is an approximation which
is reasonable provided the temperature differences are not too great.)
If the ambient temperature is constant (not necessary in general), the DE (1.56) can be
written
d
(T − TA ) = −k(T − TA ),
dt
which can be solved by inspection:

T (t) − TA = Ce−kt ,

giving an exponential rate of cooling.

Reference: Boyce & DiPrima, page 48.

1.3.5 Pursuit problems

This class of problems refers to the situation in which one participant tries to catch a second
moving participant, e.g. a bull charging a running person, a hawk chasing a flying pigeon, a
gunboat chasing a smuggler. These problems are difficult. Here is an example.

26
Example: A dog swims across a river towards his master on the far bank, but is carried
downstream by the current. The dog always paddles towards his master. What path will
the dog follow? Assume that the speed of the river is w and of the dog in still water is u,
with w < u.

Solution: Since the dog always swims towards its master, its velocity is

vdog = (−u cos θ, −u sin θ).

The velocity of the river is

vriver = (0, w).
The rate of change of the dog’s position x(t) = (x(t), y(t)) is the sum of the two velocities:

path of
dog vriver
(x(t), y(t))

11
00
11
00
00
11
00
11
vdog
11
00
0 θ
1 0
1
111
000
0
1
00
11
0
1 000
111
x
0
1
0
1 000
111
000
111
master dog’s
starting
point

Figure 1.7:

dx
= vdog + vriver .
dt
In components,
dx dy
= −u cos θ, = −u sin θ + w. (1.57)
dt dt
But
x y
cos θ = p , sin θ = p . (1.58)
x2 + y2 x2 + y2

27
We can write the dog’s path in the form y = y(x). Then

dy
dy dt
= dx
.
dx dt

By (1.57) and (1.58),

p
dy uy − w x2 + y 2
= .
dx ux
Let k = wu , a dimensionless quantity. Since x > 0 we can rearrange the DE to get
r
dy y y2
= −k 1+ , (1.59)
dx x x2

a first order DE for the path y = y(x) of the dog. The initial condition is y(b) = 0, where b
is the width of the river.

Comment: [x] = L, [y] = L and [k] = 1, making (1.59) dimensionally consistent.

Exercise: The form of the DE (1.59) suggests that we introduce a new dependent variable
z defined by
y
z= .
x
The DE (1.59) assumes the form
√
dz 1 + z2
= −k ,
dx x

which is separable. The final solution for y = y(x) is

  
1 x 1−k  x 1+k
y= 2
b − .
b b

The algebra involved in deriving this is somewhat tricky. An intermediate step is

√  x −k
z+ 1 + z2 = ,
b

which can be solved for z giving

   x k 
1 x −k
z= 2
− . 
b b

Reference: Borelli & Coleman, pages 112-5.

28
 R L C

V = RI V = L dI
dt
Q = VC

Figure 1.8: Resistor, inductor & capacitor

1.3.6 Electrical circuits

We consider simple electrical circuits which are constructed from a voltage source and three
basic elements, resistors (of resistance R), inductors (of inductance L) and capacitors (of
capacitance C), indicated symbolically in figure 1.8.

V = RI V = L dI
dt
Q = VC (1.60)

Here I denotes the electric current through an element, V is the electrical potential
difference across the element, and Q is the electric charge on the plate of the conductor onto
which the current I is assumed to flow; I is therefore related to Q by
dQ
I= . (1.61)
dt
The three formulas in the figure may be regarded as defining R, L and C.
The current I in the electrical circuit varies with time t, and in simple cases is governed
by a DE. The form of the DE is determined by one of Kirchoff ’s Laws, which states that the
potential difference across the terminals (i.e. the voltage source) must equal the sum of the
potential differences across the various elements:
X
Vterminals = Velements . (1.62)

Example: The RC circuit.

R C

I(t)

V (t)

Figure 1.9:

By (1.62) and (1.60),

Q
V (t) = VR + VC = RI + . (1.63)
C
29
 dQ
Differentiate (1.63) with respect to t and use (1.61) to eliminate dt
, giving

dI 1 1 dV
+ I= . (1.64)
dt RC R dt
Equation (1.64) is a first order linear DE for I(t), the source voltage V (t) being regarded as
given. 

Example: The RL circuit.

R L

I(t)

V (t)

Figure 1.10:

By (1.62) and (1.60),

dI
V (t) = VR + VL = RI + L ,
dt
that is
dI R 1
+ I = V (t). (1.65)
dt L L
Equation (1.55) is a first order linear DE for I(t). 

Example: The RLC circuit.

R L

I(t) C

V (t)

Figure 1.11:

By (1.62) and (1.60),

dI Q
V (t) = VR + VL + VC = RI + L + .
dt C
Differentiate with respect to t and use (1.61), as in passing from (1.63) to (1.64). We obtain

30
 d2 I R dI 1 1 dV
+ + I = , (1.66)
dt2 L dt LC L dt
a second order linear DE (to be studied later in the course).

References: Boyce & DiPrima, pages 180-1.

Borelli & Coleman, pages 176-86.

31
32
Chapter 2

Dimensional Analysis

2.1 Writing physical relations in dimensionless form

2.1.1 Characteristic scales and dimensionless variables
Problems in the sciences, arising for example in quantum mechanics, oceanography, the
solar system and cosmology involve widely different time scales and length scales. Indeed,
in studying any physical problem it is important to identify a typical or characteristic time,
denoted tc , which is the length of time over which the physical system changes in a significant
way. It may also be necessary to define a characteristic length ℓc and a characteristic mass
mc .
The characteristic time is usually defined in terms of the physical parameters associated
with the system in question, for example the physical parameters that appear in the DE
that describes the evolution of the system. Here’s the simplest example.
Example 1: Consider exponential decay of a radioactive substance. The system is de-
scribed by the DE
dm
= −km, (2.1)
dt
where k is a positive constant and m denotes the mass of the radioactive substance at time
t.
By dimensional homogeneity (see property D1 in Section 1.1.2),
 
dm
[−km] = = MT −1 ,
dt
giving
MT −1
[k] = = T −1 ,
M
since [m] = M. It is thus reasonable to define a characteristic time by
1
tc = . (2.2)
k
To get a physical interpretation of tc , note that the solution of the DE (2.1) with initial
condition m(0) = m0 is
m(t) = m0 e−kt . (2.3)

33
It follows that
m(tc ) = m0 e−ktc = m0 e−1 ,
by (2.2). Thus, tc is the length of time during which the mass decreases by a factor of 1e .
Having defined a characteristic time, one can use it to write the DE (2.1) in a simpler
form. The idea is to define a dimensionless time variable τ by
t
τ= , (2.4)
tc
[t] T
(τ is dimensionless, since [τ ] = [tc ]
= T
= 1). Then, regarding m as a function of τ ,

m = m(τ ) = m(τ (t)),

the Chain Rule gives

dm dm dτ 1 dm
= = .
dt dτ dt tc dτ
Thus, using (2.2) the DE (2.1) becomes simply
dm
= −m.
dτ

Comment: There is no unique way of defining a characteristic time. For example, for a
radioactive substance one can define the half-life t1/2 , which is the time taken for half the
substance to decay, i.e.
m(t1/2 ) = 21 m0 .
By (2.3), 21 m0 = m0 e−kt1/2 , giving ekt1/2 = 2, and hence
1
t1/2 = ln 2.
k
This time could equally well be used as the characteristic time, and it is of the same order
of magnitude as tc , since by (2.2),

t1/2 = tc ln 2. 

In the next example, we write a given physical relation in dimensionless form, by defining
a characteristic time and characteristic length.

Example 2: Consider the motion of a baseball thrown vertically up with initial velocity
v0 , ignoring air resistance. The height of the ball at time t is given by

h(t) = v0 t − 21 gt2 , (2.5)

where g is the acceleration due to gravity, assumed constant. [This equation can be derived
2
by integrating ddt2h = −g twice, assuming that dh
dt
(0) = v0 , h(0) = 0.] In this situation there
are two physical parameters v0 and g, with

[v0 ] = LT −1 , [g] = LT −2 .

34
 h i
v0
It follows that g
= T , and so it is reasonable to define a characteristic time by

v0
tc = . (2.6)
g
h 2i
v0
Further, g
= L, and so we define a characteristic length:

v02
ℓc = .
g

As regards the physical interpretation, tc is the time for a deceleration of g to reduce a

velocity v0 to zero, and since
ℓc = v0 tc , (2.7)

ℓc is the distance travelled in time tc by an object moving with velocity v0 .

We now define a dimensionless time and height by

t h
τ= , H= .
tc ℓc

In terms of these variables the relation (2.5) becomes

1
v0 tc τ − 21 gt2c τ 2 ,


H=
ℓc

which simplifies using (2.6) and (2.7) to

H = τ − 21 τ 2 . (2.8)

This is the dimensionless form of the physical relation (2.5), and captures its essential content
in the simplest possible form. 

Overview:

These two examples illustrate the approach used to define a characteristic time/length/mass:

1) List the physical parameters in the DE or physical relation and give their dimensions.

2) By inspection find a combination using products and quotients that has the dimensions
of time/length/mass, to play the role of characteristic quantity.

3) Deduce the physical interpretation of the characteristic quantity as a check that your
definition is appropriate.

35
 f , cin V , m(t)

m
f , cout = V

Figure 2.1:

2.1.2 The mixing tank DE

Consider the DE describing the constant volume mixing tank as shown (see Section 1.3.1):

dm f
= − m(t) + f cin . (2.9)
dt V
Here m(t) denotes the mass of chemical in the tank at time t, f denotes the constant flow
rate and V the constant volume. The inflow concentration cin may depend on time t.
The two physical constants are f and V , with dimensions

[f ] = L3 T −1 , [V ] = L3 . (2.10)

It follows that
L3
 
V [V ]
= = 3 −1 = T.
f [f ] LT
So it is reasonable to define a characteristic time by
V
tc = . (2.11)
f
Since V = f tc , the interpretation is that tc is the time taken for an inflow at a given rate f
to fill a tank of volume V .
Introducing a dimensionless time τ by
t
τ=
tc
as before, the DE (2.9) assumes the simpler form

dm
+ m = V cin , (2.12)
dτ
(verify the details; the calculation is similar to example 1 in Section 2.1.1).
In the special case when cin is a constant we can simplify the DE (2.12) further by
introducing a characteristic mass, as follows.

36
 Since
[cin ] = ML−3 ,
it follows from (2.10) that
[V cin ] = [V ][cin ] = M.
It is thus reasonable to define a characteristic mass by

mc = V cin .

The interpretation is that mc is the mass of chemical in a tank of volume V if the concen-
tration of the chemical is cin .
We now define a dimensionless mass variable M by
m(t)
M(t) = . (2.13)
mc
On dividing (2.12) by mc = V cin and using (2.13), the DE assumes the simplest possible
form
dM
+ M = 1. (2.14)
dτ

Comment: This DE can be solved by inspection by writing it in the form

d
(M − 1) = −(M − 1),
dτ
giving
M − 1 = Ce−τ . 

2.1.3 The sky-diver DE

The DE is equation (1.4):
dv
m
= mg − αv. (2.15)
dt
The physical parameters are the mass m, the acceleration due to gravity g, and the drag
coefficient α, having dimensions

[m] = M, [g] = LT −2 , [α] = MT −1 . (2.16)

By inspection hmi
= T.
α
It is thus reasonable to define a characteristic time by
m
tc = . (2.17)
α
In order to describe the physical significance of tc , we recall that the terminal velocity is
mg
vterm = , (2.18)
α
37
 dv
as follows by setting dt
= 0 in (2.15). Using (2.17)
vterm = gtc . (2.19)
Thus tc is the time over which an acceleration of g must act in order to produce a velocity
vterm . Introducing a dimensionless time
t
τ= (2.20)
tc
in the usual way, the DE (2.15) assumes the simpler form
dv
+ v = vterm , (2.21)
dτ
(verify the details).
One can simplify the DE further by regarding vterm as a characteristic velocity, and using
it to define a dimensionless velocity function
v(t)
V (t) = . (2.22)
vterm
The DE (2.21) then simplifies further to
dV
+ V = 1. (2.23)
dτ
This DE has precisely the same form as the mixing tank DE (2.14), showing that the mixing
tank problem and the sky-diver problem have the same mathematical content!
In the sky-diver problem one can also define a characteristic length (distance). By in-
spection of (2.16) we observe that  2 
mg
= L.
α2
So it is reasonable to define a characteristic length by
m2 g
ℓc = . (2.24)
α2
By (2.17), (2.18) and (2.24),
ℓc = vterm tc ,
i.e. ℓc is the distance fallen in time tc when falling at the terminal velocity.

The height fallen at time t, h(t), is related to v(t) in the usual way,
dh(t)
v(t) = . (2.25)
dt
One can now define a dimensionless height fallen H by
h
H= . (2.26)
ℓc
It follows that (2.25) assumes the form
dH(τ )
V (τ ) =
dτ
in terms of the dimensionless variables (2.20), (2.22) and (2.26) (Exercise). 

38
Typical values:

For a skydiver of mass 75kg, typical values of vterm are

vterm ≈ 55m/sec without a parachute

vterm ≈ 5m/sec with a parachute.

These values lead to (using g ≈ 10m s−2 )

tc ≈ 5.5s, ℓc ≈ 300m without a parachute,

and
tc ≈ 0.5s, ℓc ≈ 2.5m with a parachute.

A typical training jump begins at 1200m above ground level. After a free fall of approxi-
mately 10s, the chute is opened at approximately 760m above ground. The rest of the jump
takes approximately 3 minutes, with a landing velocity of 5m/sec.

Reference:
Meade, D.B. 1998, DE models for the parachute problem, SIAM Review 40, 327-32. 

Consistency check:

In setting up the DE (2.15), we made two simplifying assumptions (see Section 1.1.1):

(i) distances are small compared to the radius R of the earth, so that g can be regarded
as constant,

(ii) velocities are small compared to the velocity of light c, so that Newton’s Second Law
can be used.

We can use the above characteristic values to give a consistency check i.e.

ℓc vterm
≪ 1, ≪ 1. 
R c

2.2 Deducing physical relations using dimensional

analysis
The goal in this section is to use arguments based on dimensions to obtain useful information
about the dependence of one physical quantity on other physical quantities, without giving
a detailed analysis of the physical problem in question.

39
2.2.1 A motivating example
Consider the problem of determining a formula for the terminal velocity of a sky-diver. This
formula is by now quite familiar (see equation (2.18)), and, as we have seen, can be derived
using the sky-diver DE. However, the problem is an excellent one for illustrating the method
of dimensional analysis,1 which is based on the Buckingham Pi Theorem.

By thinking about the physical system, we convince ourselves that the terminal velocity
vterm should depend only on the mass m of the sky-diver, the gravitational acceleration g,
and the drag coefficient α (we assume the force due to airdrag is of the form αv, where v is
the velocity):
vterm = F (m, g, α), (2.27)
where F is an unknown function of 3 variables.
In order to apply the Buckingham Pi Theorem, we have to construct all possible inde-
pendent dimensionless quantities from the physical variables

{vterm , m, g, α}, (2.28)

and so we begin by discussing this procedure. The most general dimensionless quantity will
be a product of integer powers of these variables:2

Π = v Pv mPm g Pg αPα , (2.29)

(see requirement D2 in Section 1.1.2.) The dimensions of the physical variables are

[v] = LT −1 , [m] = M, [g] = LT −2 , [α] = MT −1 . (2.30)

The dimensions of Π can be calculated as follows:

[Π] = (LT −1 )Pv (M)Pm (LT −2 )Pg (MT −1 )Pα
= M Pm +Pα LPv +Pg T −Pv −2Pg −Pα .

(see requirement D2 in Section 1.1.2). Since we require Π to be dimensionless, i.e. [Π] = 1,

the exponents of M, L and T must be zero, giving the following equations for the 4 unknowns
Pv , Pm , Pg and Pα in equation (2.29):

Pm +Pα = 0
Pv +Pg =0 (2.31)
−Pv −2Pg −Pα = 0

In matrix form  
 Pv 
0 1 0 1 Pm 
 1 0 1 0
 Pg  = 0. (2.32)

−1 0 −2 −1
Pα
1
In practice one would do the dimensional analysis before setting up the DE.
2
For brevity, we write v in place of vterm .

40
Since (2.31) is a linear system of 3 equations in 4 unknowns, the solution is not unique. In
fact, one obtains
Pv = Pα , Pm = −Pα , Pg = −Pα , (2.33)
i.e. one can choose Pα arbitrarily and then Pv , Pm and Pg are determined. The dimensionless
quantity (2.29) assumes the form
 Pα
Pα −Pα −Pα Pα vα
Π=v m g α = .
mg

There is only one independent dimensionless quantity, obtained by choosing Pα = 1 for

simplicity, i.e.
vα
Π= . (2.34)
mg
The Buckingham Pi Theorem asserts that the physical relation (2.27) is equivalent to a
relation involving only the dimensionless variables. Since in the present case there is only one
dimensionless variable, the conclusion is that this variable must equal a numerical constant:
vα
Π = C, i.e. = C. (2.35)
mg
This can be rearranged to give
mg
vterm = C . (2.36)
α
Thus, we have shown, purely by consideration of dimensions, that the formula (2.27) must
have the form (2.36). The only uncertainty is in the numerical constant C, which has to be
determined by a more detailed mathematical analysis or by experiments.

2.2.2 Complete sets of dimensionless variables

In this Section we discuss the process for constructing dimensionless variables in more detail.
The matrix in equation (2.32) is called the dimensional matrix associated with the physical
variables {v, m, g, α}. Its entries are the dimensions of these variables, which can be read off
from equation (2.30), which we repeat here:

[v] = LT −1 , [m] = M, [g] = LT −2 , [α] = MT −1 .

The matrix is

v m g α
 
M 0 1 0 1
L1 0 1 0
T −1 0 −2 −1

It is a 3 × 4 matrix, corresponding to 3 distinct dimensions M, L and T , and 4 physical

variables. The equation (2.32) (or (2.31)) that determines the exponents can be written in
the form
DP = 0 ,

41
where D is the dimensional matrix, and

P = (Pv , Pm , Pg , Pα )T .

We have seen that this equation has only one independent solution

P = (1, −1, −1, 1)T ,

(see eq. (2.33) with Pα = 1) giving one dimensionless quantity

vα
Π = v 1 m−1 g −1 α1 = . 
mg

The key point is that the form of the dimensional matrix D enables one to predict how
many independent dimensionless quantities there will be. In the previous example, the matrix
D has rank 3 (since there are 3 linearly independent columns, as one can see by inspection)
and hence the linear system
DP = 0
has 4 − 3 = 1 independent solution, leading to 1 dimensionless scalar.

The general situation is this:

if there are N physical variables and the dimensional matrix D has rank r, then
the equation DP = 0 will have N − r linearly independent solutions, which will
lead to N − r independent dimensionless scalars.

Here is another example.

Example: A baseball is thrown vertically upwards. We anticipate that subsequently its

height h will depend on time t, the acceleration g due to gravity, its mass m, and initial
velocity v0 :

Figure 2.2:

h = F (t, v0 , g, m). (2.37)

42
Find the dimensional matrix and construct a complete set of independent dimensionless
variables.

Solution: List the physical variables in the order

{h, t, v0 , g, m}. (2.38)

Their dimensions are

[h] = L, [t] = T, [v0 ] = LT −1 , [g] = LT −2 , [m] = M.

The dimensional matrix D is thus

h t v0 g m
 
M 0 0 0 0 1
L 1 0 1 1 0
T 0 1 −1 −2 0

By inspection there are 3 linearly independent columns, and hence the rank of D is 3. Thus
the linear system
DP = 0 , (2.39)
where
P = (Ph , Pt , Pv , Pg , Pm ),
will have 5 − 3 = 2 linearly independent solutions, implying that we can construct 2 inde-
pendent dimensionless quantities of the form:

Π = hPh tPt v0Pv g Pg mPm (2.40)

from the physical variables (2.38). The linear system (2.39) reads

Pm = 0 
Ph +Pv +Pg =0 . (2.41)
Pt −Pv −2Pg =0


Regarding Ph and Pg as arbitrary, we can write the solution as

Pv = −Ph − Pg , Pt = −Ph + Pg , Pm = 0.

Choosing Pg = 0, Ph = 1 and substituting in (2.40), we get

h
Π1 = h1 t−1 v0−1 g 0 m0 =
v0 t
as one dimensionless quantity. A second linearly independent solution of (2.41) is obtained
by choosing Pg = 1, Ph = 0. Substituting in (2.40) gives
gt
Π2 = h0 t1 v0−1 g 1 m0 = .
v0

43
Thus, a complete set of independent dimensionless variables is
h gt
Π1 = , Π2 = . (2.42)
v0 t v0


Comment: The above set of dimensionless variables is not the only possible choice. If we
regard Ph and Pt as arbitrary when solving (2.41) we get

Pg = Ph + Pt , Pv = −2Ph − Pt , Pm = 0.

Choosing Ph = 1, Pt = 0 and then Ph = 0, Pt = 1, and using (2.40) gives the two dimen-
sionless variables:
gh gt
Π̂1 = 2 , Π̂2 = , (2.43)
v0 v0
which also form a complete independent set.

Summary of the procedure:

1) List the N physical variables and their dimensions.
2) Set up the dimensional matrix D, and calculate its rank r.
3) There will be N − r independent dimensionless variables. These can be found either
by using the solution procedure as illustrated in the example OR by inspection i.e.
by studying the dimensions of the set of physical quantities and forming obvious com-
binations. This is always possible in simple cases and is the recommended method.


2.2.3 The Buckingham Pi Theorem

Loosely speaking, the Buckingham Pi Theorem asserts that
any physical relation involving (N)
dimensional variables can be written
in terms of a complete set of (N − r)
independent dimensionless variables,
where r is the rank of the dimensional matrix D.
To see why this theorem is useful it is necessary to consider some examples.

Case 1: N = 4, r = 3

Suppose we have 4 physical variables

Q1 , Q2 , Q3 , Q4 ,

and we know that one of them, say Q1 , depends on the others. We write

Q1 = F (Q2 , Q3 , Q4 ), (2.44)

44
which represents the physical relation we wish to investigate.
If the rank of the dimensional matrix D is r = 3, then one can form only N −r = 4−3 = 1
dimensionless quantity from the Q’s, which we denote by Π. The Buckingham Pi Theorem
then implies that the physical relation (2.44) must have the simple form

Π = C, (2.45)

where C is a numerical constant. 

Example 1: Referring to the terminal velocity problem in Section 2.2.1, we write

Q1 = vterm , Q2 = m, Q3 = g, Q4 = α. (2.46)

Equation (2.44) has the form

vterm = F (m, g, α). (2.47)

As shown in Section 2.2.1 we can form only one dimensionless quantity from the physical
variables (2.46), namely
vterm α
Π= . (2.48)
mg
The Buckingham Pi Theorem asserts that (2.47) can be written in the form (2.45) i.e.
vterm α
= C.
mg
Rearranging gives
mg
vterm = C . (2.49)
α
What has been gained? Well, the theorem has determined the function F (m, g, α) in (2.47)
almost completely, i.e. up to the unknown numerical constant C. This represents a remark-
able simplification and provides considerable information. For example, we can conclude
that if m and g remain fixed, the terminal velocity varies inversely as the drag coefficient α.


Case 2: N = 5, r = 3

We suppose that Q1 depends on Q2 , . . . , Q5 , and we write

Q1 = F (Q2 , Q3 , Q4 , Q5 ). (2.50)

If the rank of the dimensional matrix is r = 3, then we can form precisely N − r = 5 − 3 = 2

dimensionless quantities from the Q’s, which we denote

Π1 , Π2 .

The Buckingham Pi Theorem asserts that (2.50) can be written in the form

Π1 = f (Π2 ), (2.51)

45
where f is an unknown function of one variable. 

Here is an example to illustrate the above case.

Example 2: The period of a simple pendulum.

We wish to obtain information about the period P of a simple pendulum, ignoring all
frictional effects.
The relevant physical variables and their dimensions are
the period P , [P ] = T
the mass m, [m] = M
the length ℓ, [ℓ] = L
the acceleration due to gravity g, [g] = LT −2
the amplitude of swing θ [θ] = 1.

θ
ℓ

mg

Figure 2.3:

We assume that
P = F (m, ℓ, g, θ). (2.52)
The dimensional matrix is

P m ℓ g θ
 
M 0 1 0 0 0
L 0 0 1 1 0 .
T 1 0 0 −2 0

By inspection, the rank is 3 (the matrix has 3 linearly independent columns). We can thus
form precisely two independent dimensionless quantities from the physical variables. By
2
inspection of the list of dimensions, we see that P ℓ g and θ are dimensionless, and so we write

P 2g
Π1 = , Π2 = θ. (2.53)
ℓ
46
The Buckingham Pi Theorem asserts that the physical relation (2.52) can be written in the
form (2.51), i.e.
P 2g
= f (θ),
ℓ
where f (θ) is an unknown function of one variable. Rearranging gives
ℓ
P 2 = f (θ). (2.54)
g
What has been gained? Well, we know precisely how the period P depends on ℓ and g,
and that it does not depend on m. What is not determined is the dependence on θ, which
is governed by the unknown function f . A detailed analysis of the pendulum is needed in
order to determine f (θ). 
Case 3: N = 6, r = 3

In this case the Buckingham Pi Theorem will restrict a physical relation

P1 = F (P2 , . . . , P6 ), (2.55)
to have the form
Π1 = f (Π2 , Π3 ), (2.56)
since there are N − r = 6 − 3 = 3 independent dimensionless variables. Some examples of
this case are on Problem Set 2. 
Example 3: A baseball thrown vertically upwards (N = 5, r = 3).

Referring to equation (2.37) we have

h = F (t, v0 , g, m). (2.57)
We found two independent dimensionless variables (see equation (2.42)):
h gt
Π1 = , Π2 = .
v0 t v0
The Buckingham Pi Theorem asserts that (2.57) can be written in the form
 
h gt
=f , (2.58)
v0 t v0
where the function f ( ) is not restricted. 
Comments:
1) We in fact know the explicit formula for h, namely
h = v0 t − 21 gt2 , (2.59)
(see equation (2.5)). By comparing (2.59) and (2.58) we see that the function f in
(2.58) is
f (z) = 1 − 21 z.
It is important to note that this function cannot be obtained from the Buckingham Pi
Theorem.

47
2) We also gave a different set of dimensionless quantities for this problem (see equation
(2.43)):
gh gt
Π̂1 = 2 , Π̂2 = . (2.60)
v0 v0
Observe that
Π̂1 = Π1 Π2 , Π̂2 = Π2 .
The theorem asserts that (2.57) can be written
 
gh gt
2
= fˆ , (2.61)
v0 v0

where the function fˆ is not specified. Comparison of (2.61) and (2.59) shows that
ˆ = z − 1 z2.
f(z) (2.62)
2

This example shows that when there is more than one dimensionless variable, the
information obtained from the Buckingham Pi Theorem can be written in more than
one way, e.g. (2.58) and (2.61).

3) The dimensionless variables Π̂1 , Π̂2 in equation (2.62) in fact arose in the discussion of
the baseball problem in Section 2.1.1 (see equations (2.6)-(2.8)). There we defined a
dimensionless time τ and height H by
t h
τ= , H= ,
tc ℓc
where
v0 v02
tc = , ℓc = .
g g
It follows from equation (2.60) that

Π̂1 = H, Π̂2 = τ,

and the relation

Π̂1 = fˆ(Π̂2 )
where fˆ is given by equation (2.62), becomes

H = τ − 12 τ 2

which is equation (2.8). 

48
Chapter 3

Second Order Linear DEs

3.1 Introduction
3.1.1 Oscillations and Second Order DEs
Oscillations are the single most important physical phenomenon described by second order
DEs. As motivation, we consider a simple mechanical system consisting of a spring to which
is attached a trolley that can run smoothly on a straight track.

y
0 y(t)

Figure 3.1: A mechanical oscillator

If the trolley is displaced from it equilibrium position y = 0 and released, the spring will
cause it to run to and fro on the track. It is this motion that we wish to describe.
Let y(t) be the displacement of the trolley from its equilibrium position at time t = 0.
When a spring is stretched or compressed it exerts a restoring force which, for small displace-
ments can be assumed to be proportional to the displacement (Hooke’s Law). The constant
of proportionality is called the stiffness constant, and is denoted by k. The force exerted by
the spring on the trolley is thus
Fspring = −ky. (3.1)
We also assume a damping force (due to frictional effects, air resistance) that is proportional
to the velocity, and acts so as to slow the motion:
dy
Fdamp = −c , (3.2)
dt
where c > 0 is the damping constant.

49
Comment: If dydt
> 0, the trolley is moving to the right and the force acts to the left, while if
dy
dt
< 0 the trolley is moving to the left and the force acts to the right, in both cases opposing
the motion, as required.

The motion of the trolley is governed by Newton’s Second Law. Since its mass m is
constant, we have
d2 y
m 2 = Fspring + Fdamp .
dt
On account of (3.1) and (3.2) this equation gives

d2 y dy
m 2
+ c + ky = 0. (3.3)
dt dt
We shall show how to solve this DE (see Section 3.2.3) and shall find that if the damping
constant c is non-zero, the solutions show two types of behaviour as t → ∞:

i) exponential monotone decay to zero, or

ii) exponential oscillatory decay to zero.

We shall also consider the situation in which an external force F (t) acts on the trolley,
in which case (3.3) is replaced by

d2 y dy
m 2
+ c + ky = F (t). (3.4)
dt dt
Of particular interest is the case in which F (t) is periodic, which can lead to a variety of
interesting behaviour (see Section 3.3).

Comment: The DEs (3.3) and (3.4) describes many different physical systems e.g. a pen-
dulum with small amplitude, a buoy bobbing up and down in the ocean, electrical circuits,
two connected mixing tanks, etc.

3.1.2 The world’s simplest second order DE

On setting the damping constant c to zero in the DE (3.3) in Section 3.1.1, we obtain

d2 y
m + ky = 0.
dt2
On defining ω 2 = k/m, this becomes

d2 y
+ ω 2y = 0. (3.5)
dt2
By dimensional homogeneity, the dimension of ω is (time)−1 , i.e. [ω] = T −1 , and so we can
define a dimensionless time τ by
τ = ωt. (3.6)

50
It follows from (3.6) and the Chain Rule that
dy dy d2 y 2
2d y
=ω , = ω .
dt dτ dt2 dτ 2
The DE (3.5) thus becomes
d2 y
+ y = 0, (3.7)
dτ 2
which is indubitably the world’s simplest and most important second order DE.
The solutions of this DE can be found by inspection, by recalling the differentiation
property of sin and cos:
d2 d2
(sin τ ) = − sin τ, (cos τ ) = − cos τ.
dτ 2 dτ 2
Thus y1 = cos τ and y2 = sin τ are solutions of (3.7). It is easy to verify that

y = c1 cos τ + c2 sin τ, (3.8)

for any constants c1 and c2 , is also a solution of (3.7). We shall see later (Section 3.2.1) that
(3.8) is in fact the general solution of the DE (3.7). By (3.6) it follows that

y = c1 cos(ωt) + c2 sin(ωt)

is the general solution of (3.5). We summarize this most important result:

the general solution of the DE

d2 y
+ ω2y = 0
dt2
is
y = c1 cos(ωt) + c2 sin(ωt),
where c1 and c2 are arbitrary constants. 
Comment: Since this solution is periodic of period 2π ω
, the solution describes the trolley
moving to and fro with constant period. This type of motion is referred to as simple harmonic
motion.

3.1.3 Types of Second Order DE

The general form of a second order DE is
d2 y
 
dy
=f x, y, . (3.9)
dx2 dx
dy


The simplest class are the linear DEs, which are characterized by f x, y, dx being linear in
dy
y and dx , i.e.  
dy dy
f x, y, = F (x) + A(x)y + B(x) .
dx dx

51
 dy
A linear DE is usually written with the y and dx terms on the left side. Replacing B(x) by
−P (x) and A(x) by −Q(x), the DE (3.9) becomes

d2 y dy
2
+ P (x) + Q(x)y = F (x),
dx dx
or more concisely,
y ′′ + P (x)y ′ + Q(x)y = F (x). (3.10)
This is the general form of a non-homogeneous linear second order DE.
If F (x) = 0, the DE is said to be homogeneous:

y ′′ + P (x)y ′ + Q(x)y = 0. (3.11)

This is the general form of a homogeneous linear second order DE.

Finally, if P (x) = p and Q(x) = q, where p and q are constants, (3.10) and (3.11) become

y ′′ + py ′ + qy = F (x), (3.12)

and
y ′′ + py ′ + qy = 0. (3.13)
This is the general form of a second order linear DE with constant coefficients, (3.12) being
non-homogeneous and (3.13) being homogeneous. In this course we shall be concerned almost
exclusively with this type of second order DE.
Note that the DE (3.4), namely

d2 y dy
m + c + ky = F (t),
dt2 dt
is of the form (3.12) – just divide by m to put this DE in standard form. Finally the world’s
simplest second order DE, namely

d2 y
+ ω 2y = 0,
dt2
is of the form (3.13) with p = 0 and q = ω 2.

3.1.4 The Initial Value Problem for Second Order DEs

Recall the spring and trolley problem from Section 3.1.1, described by the DE

my ′′ + cy ′ + ky = 0, (3.14)

where y(t) is the displacement of the trolley at time t, from the equilibrium position and ′
denotes dtd .
What are the different ways of setting the trolley in motion? The simplest way is to pull
the trolley from its equilibrium position, hold it at rest, and then release it at time t = 0.
This procedure corresponds to the initial conditions

y(0) = y0 , y ′(0) = 0, (3.15)

52
i.e. the initial velocity is zero. Another possibility is to give the trolley a tap with a hammer
while it is at rest in its equilibrium position. The impact will set the trolley in motion, and
the initial condition will be
y(0) = 0, y ′(0) = v0 , (3.16)
i.e. the initial velocity is non-zero. The most general possibility is to pull the trolley from
equilibrium and give it a little jerk as you release it, thereby imparting an initial displacement
and initial velocity to it. The initial conditions will be

y(0) = y0 , y ′(0) = v0 . (3.17)

With initial conditions such as (3.15)-(3.17) we certainly expect the physical system to move
in a uniquely determined way, and hence we expect the DE (3.14) to have a unique solution.
This simple example illustrates what holds in general for a linear second order DE:

y ′′ + P (x)y ′ + Q(x)y = F (x), (3.18)

namely, that the appropriate initial conditions are

y(x0 ) = y0 , y ′(x0 ) = v0 , (3.19)

and that these initial conditions determine a unique solution of the DE (3.18) (subject to
appropriate restrictions on P, Q and F ). This existence-uniqueness result will be discussed
in AM351. 

3.2 Solving Second Order Linear DEs with constant

coefficients
3.2.1 A fundamental property of homogeneous second order linear
DEs
Consider the homogeneous DE

y ′′ + P (x)y ′ + Q(x)y = 0. (3.20)

The fact that this DE is linear AND homogeneous has an immediate and important conse-
quence.

Proposition: If y1 (x) and y2 (x) are solutions of the homogeneous linear DE (3.20), then

c1 y1 (x) + c2 y2 (x)

is also a solution, for any constants c1 and c2 .

Proof: Since y1 and y2 are solutions of (3.20),

y1′′ + P (x)y1′ + Q(x)y1 = 0 (3.21)

y2′′ + P (x)y2′ + Q(x)y2 = 0. (3.22)

53
Multiply (3.21) by c1 and (3.22) by c2 and add, to get

(c1 y1 + c2 y2 )′′ + P (x)(c1 y1 + c2 y2 )′ + Q(x)(c1 y1 + c2 y2 ) = 0,

as required. 

The fundamental property referred to in the title is this:

Any solution of the homogeneous linear DE (3.20) has the form

y = c1 y1 (x) + c2 y2 (x), (3.23)

where y1 (x) and y2 (x) are linearly independent1 solutions of (3.20), and
c1 , c2 are constants.

Example: Consider the world’s simplest second order DE

y ′′ + ω 2 y = 0. (3.24)

Two solutions are

y1 = sin ωt, y2 = cos ωt,

and these are linearly independent, since cos( ) is not a multiple of sin( ). The general
solution of (3.24) is thus
y = c1 sin ωt + c2 cos ωt, (3.25)

where c1 and c2 are constants. 

The fundamental property is a consequence of the existence and uniqueness theorem

for second order linear DEs (AM351). The general solution (3.23) contains TWO arbitrary
constants because the initial conditions are TWO in number:

y(x0) = y0 , y ′ (x0 ) = v0 .

When one solves an initial value problem, c1 and c2 are determined in terms of y0 and v0 .

Exercise: Show that the unique solution of the initial value problem

y ′′ + ω 2y = 0; y(0) = y0 , y ′ (0) = v0

is
v0
y= sin ωt + y0 cos ωt. 
ω
1
y1 (x) and y2 (x) are linearly independent means that no linear combination of y1 (x) and y2 (x) equals
zero, for all x.

54
3.2.2 General form of the solution
Consider a non-homogeneous second order linear DE
y ′′ + P (x)y ′ + Q(x)y = F (x). (3.26)
The linearity of this DE has an important consequence.

Proposition: If y1 and y2 are solutions of the non-homogeneous DE (3.26), then the

difference
y1 (x) − y2 (x)
is a solution of the associated homogeneous DE
y ′′ + P (x)y ′ + Q(x)y = 0. (3.27)

Proof: Same as the proposition on page 16 for first order DEs. 

Suppose yp (x) is a particular solution of (3.26) and y(x) is any solution of (3.26), i.e. it
represents the general solution. Then by the Proposition,
yh (x) = y(x) − yp (x)
is a solution of the homogeneous DE (3.27). We can rewrite this equation as
y(x) = yh (x) + yp (x),
giving the general solution. 

General solution of a second order linear DE:

The general solution of the DE (3.26) is of the form

y(x) = yh (x) + yp (x),

where yh (x) is the general solution of the homogeneous DE (3.27), and

yp (x) is a particular solution of the non-homogeneous DE (3.26).
(compare with page 17).
It follows from Section 3.2.1 that
yh (x) = c1 y1 (x) + c2 y2 (x),
where y1 and y2 are two linearly independent solutions of the homogeneous DE (3.27).

We thus need to develop two algorithms:

(1) An algorithm to give the general solution of a homogeneous linear DE. This is possible
only for the case of constant coefficients
y ′′ + py ′ + qy = 0,
where p and q are constants. We develop this algorithm in Section 3.2.3.

55
 (2) An algorithm to find a particular solution of a non-homogeneous linear DE

y ′′ + py ′ + qy = f (x).

Here we simply extend the method of undetermined coefficients that we used in the
first order case (see Section 1.2.5). This extension is discussed in Section 3.2.4. 

3.2.3 General Solution of the Homogeneous DE

A homogeneous second order linear DE with constant coefficients is of the form

y ′′ + py ′ + qy = 0, (3.28)

where p and q are constants. Our goal is to find the general solution of any such DE.
We begin by considering a trial function of the form

y = emx , (3.29)

where m is a constant. Since y ′ = memx and y ′′ = m2 emx , equation (3.28) yields

(m2 + pm + q)emx = 0.

Since emx > 0 for all x, (3.29) is a solution of (3.28) if and only if m is a solution of

m2 + pm + q = 0. (3.30)

This quadratic equation is called the characteristic equation of the DE (3.28). Its roots are
h p i
m1,2 = 21 −p ± p2 − 4q . (3.31)

There are three distinct cases, each of which has to be treated separately:

A) Distinct real roots (p2 > 4q)

B) Distinct complex roots (p2 < 4q)

C) Equal real roots (p2 = 4q).

Case A: Distinct real roots

In this case, substituting the two roots (3.31) into (3.29), we get two solutions

em1 x and em2 x .

em1 x
Since the ratio = e(m1 −m1 )x is not constant, these solutions are linearly independent.
em2 x
By Section 3.2.1, the general solution of the DE (3.28) is thus

y = c1 em1 x + c2 em2 x , (3.32)

56
where c1 and c2 are arbitrary constants and m1 , m2 are the real roots of the characteristic
equation (3.30). 

Example 1: Find the general solution of the DE

y ′′ + 5y ′ + 6y = 0. (3.33)

Solution: Substituting the trial function y = emx in (3.33) gives the characteristic equation

m2 + 5m + 6 = 0,

which factors as
(m + 2)(m + 3) = 0.
There two distinct real roots m = −2 and m = −3, give two independent solutions

e−2x and e−3x .

The general solution is thus

y = c1 e−2x + c2 e−3x . 

Case B: Distinct complex roots

The roots m1 and m2 , as given by (3.31), are distinct and complex if and only if

p2 − 4q < 0.

In this case it is convenient to write m1 and m2 in the form

m1 = a + ib, m2 = a − ib,

where a and b are real. We substitute m1 and m2 into (3.29) to obtain two complex solutions
which we can decompose into real and imaginary parts using Euler’s formula

eiθ = cos θ + i sin θ.

The two solutions of the DE (3.28) are

em1 x = e(a+ib)x = eax eibx = eax (cos bx + i sin bx) (3.34)

and
em2 x = e(a−ib)x = eax e−ibx = eax (cos bx − i sin bx). (3.35)
Since we want real solutions we consider the linear combinations
1 m1 x
2
(e + em2 x ) = eax cos bx
1
2i
(em1 x − em2 x ) = eax sin bx,

as follows from (3.34) and (3.35). These solutions are linearly independent. Thus, the general
solution of the DE (3.28) is

y = (c1 cos bx + c2 sin bx)eax , (3.36)

57
where c1 and c2 are arbitrary constants, and

a ± ib

are the roots of the characteristic equation. 

Example 2: Find the general solution of the DE

y ′′ + 2y ′ + 5y = 0. (3.37)

Solution: Substituting the trial function y = emx into (3.37) gives

m2 + 2m + 5 = 0.

Completing the square leads to

(m + 1)2 + 22 = 0,
giving the complex roots
m = −1 ± 2i.
A complex solution is

e(−1+2i)x = e−x (cos 2x + i sin 2x),

(by Euler’s formula), from which one can read off the two independent real solutions

e−x cos 2x and e−x sin 2x.

Thus, the general solution of the DE (3.37) is

y = (c1 cos 2x + c2 sin 2x)e−x . 

Case C: Equal real roots

The roots m1 and m2 ,as given by (3.31), are real and equal if and only if

p2 − 4q = 0.

In this case, we obtain only one solution from (3.29) and (3.31), namely

y1 = emx ,

where m = − 21 p.
To find a second linearly independent solution we consider a trial function of the form

y = v(x)emx . (3.38)

We substitute (3.38) into the DE (3.28) to obtain

v ′′ + (2m + p)v ′ + (m2 + pm + q)v = 0, (3.39)

58
after collecting terms and cancelling a factor of emx (fill in the details). Now a miracle
happens . . . since m = − 12 p and m is a solution of the characteristic equation m2 +pm+q = 0,
equation (3.39) reduces to
v ′′ = 0.
Choose v(x) = x as a particular solution, and then (3.38) gives

y2 = xemx (3.40)

as a second linearly independent solution. Thus, the general solution of the DE (3.28) is

y = (c1 + c2 x)emx ,

where c1 and c2 are arbitrary constants, and m is the single solution of the characteristic
equation. 

Example 3: Find the general solution of the DE

y ′′ + 6y ′ + 9y = 0. (3.41)

Solution: Substituting the trial function y = emx in (3.41) gives

m2 + 6m + 9 = 0,

which is a perfect square,

(m + 3)2 = 0
giving a single root m = −3, and one solution y1 = e−3x . A second solution is obtained from
(3.40), which gives
y2 = xe−3x .
(There is no need to repeat the whole derivation – the second solution is simply x times the
first one.) Thus, the general solution of (3.41) is

y = (c1 + c2 x)e−3x . 

3.2.4 The method of undetermined coefficients

We now show how to find a particular solution of the non-homogeneous linear DE with
constant coefficients, whose general form is

y ′′ + py ′ + qy = f (x),

where p and q are constants, and f is a given function.

We use the method of undetermined coefficients, as introduced in Section 1.2.5, for first
order DEs. This means that we restrict our considerations to the case where f is

• an exponential ebx

• a sine or cosine

59
 • a polynomial
or
• sums of such functions.

Example 1: Find the general solution of the DE

y ′′ + 5y ′ + 6y = e2x . (3.42)

Solution: We consider a trial function

y = Ae2x , (3.43)

where A is a constant. Since y ′ = 2Ae2x and y ′′ = 4Ae2x , substituting (3.43) in (3.42) yields

[4A + 5(2A) + 6(A)]e2x = e2x .

1
The e2x cancels, and solving for A gives A = 20
. Thus (3.43) gives the particular solution
1 2x
yp = 20
e . (3.44)

The general solution of (3.42) has the form

y = yh (x) + yp (x), (3.45)

where yh is the general solution of the homogeneous DE y ′′ + 5y ′ + 6y = 0. We solved this

problem in Example 1 of Section 3.2.3:

yh = c1 e−2x + c2 e−3x . (3.46)

By (3.44), (3.45) and (3.46), the general solution of the DE (3.42) is

y = c1 e−2x + c2 e−3x + 1 2x
20
e . 

It is of interest to consider the previous example with an exponential driving term, in

greater generality. Consider
y ′′ + py ′ + qy = eax (3.47)
where p, q and a are constants. As in the previous example, we consider a trial function

y = Aeax . (3.48)

Substituting (3.48) in (3.47) yields

A(a2 + pa + q)eax = eax

(fill in the details). Thus

1
A= ,
a2 + pa + q

60
giving a particular solution (3.48). However, it is clear that A is undefined for certain values
of the constant a, namely, those values which satisfy

a2 + pa + q = 0. (3.49)

This is precisely the characteristic equation of the homogeneous DE associated with (3.47).
Thus, the trial function (3.48) does not give a solution if a is a root of the characteristic
equation.
How do we find a particular solution in this special case? Based on our experience in
case C in the previous section when we had to multiply by x to find a second solution, we
consider the trial function
y = Axeax . (3.50)

On substituting (3.50) in (3.47) and simplifying, we get

A(a2 + pa + q)xeax + A(2a + p)eax = eax .

The first term in brackets is zero, because we are assuming that a satisfies (3.49). Thus

1
A= .
2a + p

Using (3.50), this value of A gives a particular solution unless

a = − 21 p. (3.51)

What is the significance of this condition? Well, if the characteristic equation (3.49) has a
double real root, i.e. if p2 = 4q, then (3.49) becomes

2
a + 21 p = 0.

Thus, (3.51) states that a is a double root of the characteristic equation. In this case we
generalize (3.50) and use
y = Ax2 eax (3.52)

as a trial function. On substituting (3.52) in (3.47) we get

A(a2 + pa + q)x2 eax + 2A(2a + p)xeax + 2Aeax = eax .

Since we are assuming a is a double root of the characteristic equation, equations (3.49) and
(3.51) hold, and hence the terms in brackets are zero, leaving

A = 21 .

Thus, in this case equation (3.52) gives a particular solution of the DE (3.47).

61
Summary:
Consider the DE
y ′′ + py ′ + qy = eax , (3.53)
with characteristic equation
a2 + pa + q = 0. (3.54)

• If a is not a root of (3.54), (3.53) has a particular solution of the form y = Aeax .

• If a is a single root of (3.54), (3.53) has a particular solution of the form y = Axeax .

• If a is a double root of (3.54), (3.53) has a particular solution of the form y = Ax2 eax .


We next consider the case of a polynomial driving term:

y ′′ + py ′ + qy = a0 + a1 x + · · · + an xn . (3.55)

Since the derivative of a polynomial is a polynomial, we consider a trial function of the form

yp = A0 + A1 x + · · · + An xn , (3.56)

where A0 , A1 , . . . , An are the undetermined coefficients. This choice works unless q = 0 in

(3.55), in which case one multiplies (3.56) by x. 

Exercise 1: Find a particular solution of

y ′′ + 5y ′ + 6y = 6x.

Answer: yp = − 65 + x

Exercise 2: Find a particular solution of

y ′′ + y ′ = 2x.

Answer: yp = −2x + x2 . 

We finally consider a sin or cosine driving term:

y ′′ + py ′ + qy = sin ωx. (3.57)

Since the derivatives of sin ωx are multiples of sin ωx and cos ωx, we take a trial function of
the form
yp = A sin ωx + B cos ωx. (3.58)
This choice works unless p = 0 and ω 2 = q. This special case is

y ′′ + ω 2 y = sin ωx. (3.59)

62
To get a suitable trial function we multiply (3.58) by x:

yp = x(A sin ωx + B cos ωx).

Exercise 3: Find a particular solution of

y ′′ − 3y ′ − 4y = 2 sin x.

5 3
Answer: yp = − 17 sin x + 17
cos x.

Exercise 4: Find a particular solution of

y ′′ + 4y = sin 2x.

Answer: yp = − 14 x cos 2x. 

If the driving term is a sum one can use the following proposition to find a particular
solution.

Proposition:

If y1 is a solution of y ′′ + py ′ + qy = f1 (x),
and y2 is a solution of y ′′ + py ′ + qy = f2 (x),
then y = y1 + y2 is a solution of

y ′′ + py ′ + qy = f1 (x) + f2 (x).

Proof: Exercise (a consequence of linearity. . .). 

3.3 Analysis of the oscillator DE

3.3.1 Introductory remarks
In this section we describe the properties and physical interpretation of the solutions of the
DE that describes the mass-spring system with damping and periodic driving force:

my ′′ + cy ′ + ky = F0 cos ωt. (3.60)

We divide by m and write the DE in the form

y ′′ + 2λy ′ + ω02 y = f0 cos ωt, (3.61)

where
c k F0
λ= , ω02 = , f0 = . (3.62)
2m m m
63
Since the DE (3.61) describes a variety of oscillating systems, both mechanical and electrical,
we shall refer to it as the oscillator DE.
The parameter λ and ω0 characterize the system itself (while f0 and ω characterize the
driving force), and are thus called the system parameters. λ is called the damping parameter
and ω0 is called the natural frequency (the latter name will be justified in Section 3.3.2).
Both system parameters have dimensions T −1 , as follows from (3.61) and can thus be used
to define two characteristic times ω10 and λ1 . We shall see that the ratio ωλ0 of these two
parameters, which is dimensionless, plays an important role in determining the behaviour of
the solutions of the DE.
The DE (3.61) also describes the current in an RLC electrical circuit. In Section 1.3.6
we saw that the current in such a circuit satisfies the DE
1
LI ′′ + RI ′ + I = V ′ (t),
C
where V (t) is the applied voltage. With

V (t) = V0 sin(ωt)

this DE becomes
1
LI ′′ + RI ′ + I = V0 ω cos ωt. (3.63)
C
Dividing by L gives
R ′ 1 V0 ω
I ′′ + I + I= cos ωt,
L LC L
which is of the form (3.61) with

R 1 V0 ω
λ= , ω02 = , f0 = . (3.64)
2L LC L

3.3.2 Zero driving force

In this section we discuss the oscillator DE (3.61) with zero force, i.e.

y ′′ + 2λy ′ + ω02 y = 0. (3.65)

There are four qualitatively different cases, depending on the dimensionless ratio

λ
ζ= .
ω0

A) The undamped case: ζ = 0

B) The underdamped case: 0 < ζ < 1

C) The critically damped case: ζ = 1

D) The overdamped case: ζ > 1

64
 (c1 , c2 )
R
φ

Figure 3.2: Definition of the phase angle φ.

These cases correspond to the different possibilities for the roots of the characteristic equation
of the DE (3.65).

Case A: The undamped case (ζ = 0)

The DE (3.65) specializes to

y ′′ + ω02 y = 0,
the “world’s simplest”. By inspection the general solution is (see Section 3.1.2)

y = c1 cos(ω0 t) + c2 sin(ω0 t). (3.66)

In order to interpret the solution it is necessary to use the trig identity

cos(A − B) = cos A cos B + sin A sin B (3.67)

to rewrite (3.66) as single cosine.

Think
p of the constants c1 , c2 in (3.66) as defining a point (c1 , c2 ) on a circle of radius
R = c21 + c22 , and let φ be the radian measure of the angle defined by (c1 , c2 ) (see Figure
3.2).
It follows that
c1 = R cos φ, c2 = R sin φ. (3.68)
The solution (3.66) becomes

y = R[cos(ω0 t) cos φ + sin(ω0 t) sin φ],

which, using (3.67), can be written as

y = R cos(ω0 t − φ). (3.69)

Thus, the system (e.g. the trolley attached to the spring) oscillates with period

2π
P =
ω0

(recall that cos t is periodic of period 2π and hence cos(ω0 t − φ) is periodic of period ω2π0 ).
We say that the system undergoes simple harmonic motion (SHM). The constant R in the

65
solution (3.69), which represents the maximum displacement of the system, is called the
amplitude of the SHM. The frequency of the SHM is
1 ω0
ν= = ,
P 2π
and the constant ω0 is called the circular frequency. The dimensionless constant φ is called
the phase.
It should be kept in mind that the frequency ω0 , which is one of the system parameters,
is an intrinsic property of the system. Since the initial conditions

y(0) = y0 , y ′(0) = v0 (3.70)

enter into the solution (3.69) through the constants c1 and c2 , the frequency ω0 is independent
of the initial conditions.
On the other hand, the amplitude R and phase φ do depend on the initial conditions. It
follows from equations (3.69) and (3.70) that
s
v2
R = y02 + 02 , (3.71)
ω0

and that φ is determined by

y0 v0
cos φ = , sin φ = , (3.72)
R ω0 R
(exercise on Problem Set 3).
The special case of “release from rest”, i.e. y0 6= 0, v0 = 0, gives
(
0, if y0 > 0
R =| y0 | and φ = .
π, if y0 < 0

The other special case of “a kick while in equilibrium”, i.e. y0 = 0, v0 6= 0, gives

(
π
| v0 | 2
, if v0 > 0
R= and φ = 3π .
ω0 2
, if v0 < 0

Case B: The underdamped case (0 < ζ < 1)

The characteristic equation for the DE (3.65) is

m2 + 2λm + ω02 = 0.

The roots are q

m = −λ ± λ2 − ω02 ,
which we rewrite as h p i
m = ω0 −ζ ± i 1 − ζ 2

66
 λ
using ζ = ω0
. The general solution of the DE is thus
h p  p i
y = e−ζω0 t c1 cos 1 − ζ 2 ω0 t + c2 sin 1 − ζ 2 ω0 t ,

(as in Section 3.2.3). As in Case A, we can write the expression in square brackets as a single
cosine, giving hp i
y = Re−ζω0 t cos 1 − ζ 2 ω0 t − φ . (3.73)
The constants R and φ are related to c1 and c2 by equation (3.68) and are thus determined
by the initial conditions via equations analogous to, but more complicated than, equations
(3.71) and (3.72). We do not need the specific form of these equations.
Equation (3.73) gives the general solution of the oscillator DE (3.65) in the underdamped
case. The function y(t) satisfies

−Re−ζω0 t ≤ y ≤ Re−ζω0 t .

The graph of y(t) thus oscillates between these two exponential curves. The zeros are equally
spaced, with the spacing determined by the period of the cosine function i.e. the spacing is
2π
p .
ω0 1 − ζ 2
Of course y(t) itself is not periodic. We say that y(t) describes a system that is performing
damped oscillations. The graph of y(t) is shown below.

y = Re−ζω0 t

∆t = √π
ω0 1−ζ 2
−ζω0 t
y = −Re

hp i
−ζω0 t 2
Figure 3.3: Graph of y = Re cos 1 − ζ ω0 t − φ .

Comment: Damped oscillations can approximate SHM under the following conditions. If
the dimensionless constant
λ
ζ=
ω0
is very small, the system will perform a significant number of oscillations before the damp-
ing has had sufficient time to decrease the amplitude of the oscillations appreciably. This

67
behaviour can be explained heuristically in terms of the characteristic times associated with
the damping and with the oscillations, namely
1 1
tdamp = , tosc = . (3.74)
λ ω0
The restriction ζ ≪ 1 is equivalent to

tosc ≪ tdamp .

For example, let ∆t1% be the time for the amplitude factor e−λt to decay by 1%. Show that
if ζ = 10−5 , there will be approximately 160 oscillations during this time interval.

Case C: The critically damped case (ζ = 1)

The characteristic equation for the DE (3.65) is

m2 + 2ω0 m + ω02 = 0,

with a single repeated root m = −ω0 . As in Section 3.2.3, the general solution is

y = (c1 + c2 t)e−ω0 t .

It follows (Exercise) that the unique solution which satisfies the initial conditions y(0) = y0
and y ′(0) = v0 , is
y = [y0 + (v0 + ω0 y0 )t] e−ω0 t . (3.75)
For all initial conditions, lim y = 0, i.e. the system returns to a state of equilibrium. The
t→∞
shape of the graph of y(t) (i.e. the qualitative behaviour) depends on y0 and v0 . Figure 3.4
shows the different possibilities for y0 > 0.

y0

Figure 3.4: Displacement of a critically damped oscillator.

Case D: The overdamped case (ζ > 1)

68
 The characteristic equation for the DE (3.65) is

m2 + 2λm + ω02 = 0,

with distinct real roots q

m1,2 = −λ ± λ2 − ω02 .
In terms of ζ = λ/ω0 , h i
p
m1,2 = ω0 −ζ ± ζ 2 − 1 . (3.76)

The general solution of the DE (3.65) is

h √ √ i
2 2
y = e−ζω0 t c1 e ζ −1 ω0 t + c2 e− ζ −1 ω0 t . (3.77)

The constants c1 and c2 are determined by the initial conditions. The expressions are a bit
complicated, and are not important. The essential point is that for all initial conditions, i.e.
for all c1 and c2 ,
lim y = 0,
t→+∞

i.e. the system eventually returns to a state of equilibrium.

Summary:
We have analyzed the solutions of the DE

y ′′ + 2λy ′ + ω02 y = 0, (3.78)

which governs the displacement of a mechanical oscillator and also the current in an electrical
circuit. The first essential result is a consequence of the form of the general solutions (3.73),
(3.75) and (3.77).

Proposition: If λ > 0 and ω0 > 0 then all solutions of the DE

y ′′ + 2λy ′ + ω02y = 0

satisfy
lim y(t) = 0. 
t→+∞

The interpretation is as follows. The DE (3.78) admits the equilibrium solution y(t) = 0
for all t, corresponding to the physical system being in a state of equilibrium. The proposition
implies that if the system starts at time t = 0 with the initial conditions y(0) = y0 and
y ′(0) = v0 , then it will eventually return arbitrarily closely to the equilibrium state y = 0, no
matter what the initial conditions are. The physical cause of this behaviour is the damping,
either the mechanical damping or the electrical damping (the resistor R in the electrical
circuit).
The second essential result is this: the system undergoes (damped) oscillations while
returning to equilibrium if an only if the dimensionless system parameter ζ = ωλ0 satisfies
0 < ζ < 1.

69
 A final question of practical concern arises from the Proposition, namely, how rapidly
does the system approach equilibrium, i.e. how rapidly does y tend to zero? The decay rate
of y is governed by the exponential term in equations (3.73), (3.75) and (3.77):

−ζω0 t
e
 , if 0 < ζ < 1
−ω0 t
y(t) ∼ e , if ζ = 1 (3.79)
 −(ζ−√ζ 2 −1)ω0 t

e , if ζ > 1
p
Since ζ > 1 implies ζ − ζ 2 − 1 < 1, it follows that the displacement y decays most rapidly
for ζ = 1, i.e. in the critically damped case. Thus, if one wishes to design a mechanical
or electrical system that will return rapidly to a state of equilibrium after being disturbed,
he/she should arrange that the damping is close to critical i.e. ζ = ωλ0 ≈ 1.

3.3.3 Non-zero driving force

The oscillator DE with non-zero driving force is

y ′′ + 2λy ′ + ω02 y = f0 cos(ωt), (3.80)

[see equation (3.61) in Section 3.3.1]. We know from Section 3.2.2 that the general solution
of this DE is of the form
y(t) = yh (t; c1 , c2 ) + yp (t), (3.81)
where yh is the general solution of the homogeneous DE

y ′′ + 2λy + ω02 y = 0,

(and hence depends on two arbitrary constants c1 and c2 ), and yp is a particular solution of
(3.80). We also know from Section 3.2.4 that yp (t) is of the form

yp (t) = a1 cos ωt + a2 sin ωt,

(use the method of undetermined coefficients). This solution can also be written in the form

yp (t) = A cos(ωt − δ), (3.82)

where the constants a1 and a2 have been replaced by the amplitude A and phase δ (see
equations (3.66) and (3.69) in Section 3.3.2).
One of the main results from Section 3.3.2 is that

lim yh (t; c1 , c2 ) = 0.
t→∞

Thus, by (3.81) and (3.82), for sufficiently large t

y(t) ≈ A cos(ωt − δ),

i.e. for sufficiently large t the response y(t) of the system is periodic, having the same period
as the driving force. Heuristically, one can say that the driving force “overcomes” the effects

70
of the system parameters, namely the natural frequency ω0 and the damping constant λ,
and compels the system to oscillate at the driving force’s frequency, namely ω.

Referring to equation (3.81), the term yh , which dies away, is called the transient part
of the solution, while the term yp , which is periodic and hence persists indefinitely, is called
the steady state part of the solution. In many applications, the transient part will die out
rapidly (depending on ζ = λ/ω0 and ω0 ; see equation (3.79) in Section 3.3.2), in which case
the behaviour of the physical system is essentially described by the steady state part, which
we refer to as the response of the system to the driving (i.e. external) force.

The essential question then is: what is the amplitude A of the steady state term (3.82)?
Or more precisely, how does the amplitude A depend on the system parameters λ and ω0 ,
and on the driving force parameters, namely the frequency ω and the amplitude f0 ?

We shall show, by deriving the particular solution yp , that A depends in a critical way
on the dimensionless ratio ωω0 and on the dimensionless damping parameter ζ = ωλ0 . This
dependence has important implications for the design of many physical systems.

1. Transforming the DE to dimensionless form

In order to find the steady state amplitude A we have to find the explicit form of the
particular solution (3.82) of the DE (3.80). Since there are 4 parameters in this DE it makes
sense to first simplify the DE by introducing a dimensionless time τ and a dimensionless
displacement y. It is natural to use
1
tc =
ω0
as the characteristic time, and to define the dimensionless time by
t
τ= = ω0 t. (3.83)
tc
To make the change in the DE (3.80) divide through by ω02, obtaining

1 d2 y
   
λ 1 dy f0 ω
+2 + y = 2 cos (ω0 t) .
ω02 dt2 ω0 ω0 dt ω0 ω0

In the usual way the chain rule and equation (3.83) lead to

d2 y dy f0
2
+ 2ζ + y = 2 cos(Ωτ ) (3.84)
dτ dτ ω0
where we have introduced the dimensionless parameters
λ ω
ζ= , Ω= . (3.85)
ω0 ω0
We note for future reference that
Ωτ = ωt, (3.86)
as follows from (3.83) and (3.85).

71
 The next step is to note that the requirement of dimensional homogeneity applied to
equation (3.84) implies  
f0
= [y].
ω02
We do not specify the dimensions of y, since it could represent a displacement, an electric
current, charge, etc. It now follows that the new variable Y defined by
y ω02
Y = = y (3.87)
f0 f0
ω02

is a dimensionless variable. Thus, on dividing equation (3.84) by ωf02 this equation assumes
0
the simpler form
d2 Y dY
2
+ 2ζ + Y = cos(Ωτ ),
dτ dτ
involving only two dimensionless parameters ζ and Ω. Finally, we write dτd as ′ to obtain:

Y ′′ + 2ζY ′ + Y = cos(Ωτ ). (3.88)

2. Derivation of the steady state solution

A particular solution of the DE (3.88) can be found in the usual way, using “undetermined
coefficients”. Here we take the opportunity to show that the solution can be found more
quickly by using complex numbers.
Consider the complex DE
z ′′ + 2ζz ′ + z = eiΩτ . (3.89)
Since eiΩτ = cos Ωτ + i sin Ωτ , the DE (3.88) is the real part of (3.89), with Y = Re(z).
Assume a trial function of the form

z(τ ) = Aei(Ωτ −δ) (3.90)

where A and δ are the undetermined coefficients (they are assumed to be real). When (3.90)
is substituted in (3.89) one finds after a non-trivial calculation that
− 1
A = (1 − Ω2 )2 + 4ζ 2Ω2 2 ,

(3.91)

and
cos δ = A(1 − Ω2 ), sin δ = 2AζΩ.
[The details are an exercise on Problem Set 3.]
Taking the real part of (3.90) gives the particular solution of (3.88)

Y (τ ) = A cos(Ωτ − δ), (3.92)

where A and δ are given by equations (3.91). We can use (3.86) and (3.87) to transform the
particular solution (3.92) to a particular solution of the DE (3.80) in terms of the physical
variables y and t, obtaining
f0
y(t) = A 2 cos(ωt − δ). (3.93)
ω0

72
In summary, the steady state solution of the oscillator DE
y ′′ + λy ′ + ω02 y = f0 cos(ωt),
is given by (3.93), with
− 1
A = (1 − Ω2 )2 + 4ζ 2 Ω2 2 ,

(3.94)
cos δ = A(1 − Ω2 ), sin δ = 2AζΩ.
Thus the steady state response is simple harmonic motion with amplitude
f0
A= A, (3.95)
ω02
angular frequency ω (the same as the driving force), and phase δ.

3. Analysis of the amplitude of the steady state solution

Consider a fixed system, with system parameters λ and ω0 , governed by the oscillator DE
(3.80). We are interested in the response of the system to a driving force of frequency ω and
amplitude f0 . The amplitude A of the response is given by (3.95) and (3.94):
 
f0 f0 ω λ
A = 2 A(Ω, ζ) = 2 A , . (3.96)
ω0 ω0 ω0 ω0
For a given system ω0 and λ, and hence ζ = ωλ0 , are fixed. The dependence of A on the
amplitude f0 of the driving force is simple: A is proportional to f0 , by (3.96). The dependence
of A on the frequency ω of driving force is governed by the function A(Ω, ζ) in (3.96), and is
complicated. One can imagine changing ω. How does A change? We will be able to answer
this question at a glance if we sketch the graph of A(Ω, ζ) versus Ω, for fixed ζ. This graph
is called the frequency response curve of the system.

The graph can be drawn using the following information:

(i) A(0, ζ) = 1, for all ζ ≥ 0
(ii) lim A(Ω, ζ) = 0, for all ζ ≥ 0.
Ω→∞
These results follow immediately from (3.94).
(iii) If 2ζ 2 ≥ 1, then A < 1 for all Ω > 0. This follows by writing (3.94) in the form
− 1
A = 1 + Ω4 + 2(2ζ 2 − 1)Ω2 2 .

(3.97)
Thus, if 2ζ 2 ≥ 1, the maximum of A on the interval 0 ≤ Ω < ∞ occurs at the endpoint
Ω = 0.
p
(iv) If 2ζ 2 < 1, then A equals 1 for Ω = 0 and Ω = 2(1 − 2ζ 2), and is greater than 1
between these values. This result follows by writing (3.97) in the form
− 1
A = 1 + Ω2 {Ω2 − 2(1 − 2ζ 2 )} 2 .

(3.98)
Thus, if 0 < 2ζ 2 < 1, A must have a maximum value greater than 1.

73
 (v) If 0 < 2ζ 2 < 1 the maximum value of A is
1
A= p , (3.99)
2ζ 1 − ζ 2

and occurs for p

Ω= 1 − 2ζ 2. (3.100)
This result is obtained by showing that the derivative of A with respect to Ω has the
form
∂A 3
= −2Ω[Ω2 − (1 − 2ζ 2)]A− 2 ,
∂Ω
so that
∂A p
= 0 implies Ω = 0 or Ω = 1 − 2ζ 2 .
∂Ω
p
Then substitute Ω = 1 − 2ζ 2 into (3.98) to get (3.99).

A(Ω, ζ) A(Ω, ζ)

Ares
1 1

Ω Ω
1 Ωres 1

2ζ 2 ≥ 1 0 < 2ζ 2 < 1

Figure 3.5: Typical frequency response curves in the cases 2ζ 2 ≥ 1 and 0 < 2ζ 2 < 1.

When A attains a maximum greater than 1, one says the the system undergoes resonance.
We thus label the values of A and Ω with a subscript “res”, i.e. we write (3.99) and (3.100)
as
1 p
Ares = p , Ωres = 1 − 2ζ 2 . (3.101)
2ζ 1 − ζ 2
By (3.95) and (3.85), the physical amplitude and frequency at resonance are

f0 p
Ares = p , ωres = ω0 1 − 2ζ 2 , (3.102)
2ω02 ζ 1 − ζ 2

with ζ = λ/ω0. Observe that

ωres < ω0 for all ζ with 2ζ 2 < 1,

74
i.e. the resonant frequency is less than the natural frequency. The second result is that Ares
can be arbitrarily large if the dimensionless damping parameter ζ is sufficiently close to 0.
Indeed, if 0 < ζ ≪ 1, equation (3.102) gives the approximations
f0
Ares ≈ , ωres ≈ ω0 . (3.103)
2ω02ζ
It is instructive to draw the family of frequency response curves for different values of ζ,
in the ΩA-plane. For this purpose it is useful to note that
1
Ares = p , (3.104)
1 − Ω4res

as follows from (3.101).

A A= √ 1
1−Ω4

ζ≪1

1
1
ζ= 2

ζ= √1
ζ≫1 2

Ω
1

Figure 3.6: The frequency response diagram. Note that the maxima lie on the curve (3.104).

Comment: If 0 < ζ ≪ 1, the frequency response curve has a steep and narrow peak, with
1
Ares ≈ , ωres ≈ ω0
2ζ
showing how resonance provides amplification. Engineers refer to the quantity
1
Q=
2ζ
as the Q-factor of the system (Q for Quality). In some situations, a high Q-factor is needed,
as in a radio tuning circuit (an RLC circuit, see Section 3.3.1). In this situation, the driving

75
frequency ω would be the frequency of the station you wish to receive, and you would “tune
1
in” the station by varying ω0 = √LC to obtain resonance i.e. the circuit selectively amplifies
the signal of the station. In other situations, it is desirable to have a flat frequency response
curve. The value ζ = √12 , i.e. A = √12 , which is the value of ζ which just avoids resonance,
gives the flattest response curve, i.e. very flat for 0 < Ω < 1 (0 < ω < ω0 ). Such a system
would act as a filter, excluding frequencies with ω ≫ ω0 .

4. Analysis of the phase of the steady state solution

Recall from equation (3.94) that the phase δ of the steady state solution is given by
cos δ = A(1 − Ω2 ), sin δ = 2AζΩ, (3.105)
where − 1
A = (1 − Ω2 )2 + 4ζ 2Ω2 2 .


We are going to sketch the graph of δ as a function of Ω, for various values of ζ.

We need the following properties of the function δ = δ(Ω, ζ), which are a consequence of
equation (3.105):
π
(i) δ(0, ζ) = 0 and δ(1, ζ) = 2
for all ζ > 0,
(ii) lim δ(Ω, ζ) = π,
Ω→∞

∂δ ∂δ
(iii) (0, ζ) = 2ζ and (1, ζ) = 1ζ ,
∂Ω ∂Ω
2ζΩ
obtained by differentiating tan δ = with respect to Ω.
1 − Ω2
δ(Ω, ζ)

ζ≫1
π
2
ζ= √1
2

0<ζ≪1
ω
Ω= ω0
0 1

Figure 3.7: The phase response diagram.

The main results are:

(1) the response of the system always lags the driving force, since δ > 0,
(2) if ζ ≪ 1, resonance occurs for Ω ≈ 1 and then the phase is close to π2 .

76
5. The case of zero damping
We finally consider the idealized case of zero damping (λ = 0), so that the oscillator DE
(3.80) reduces to
y ′′ + ω02 y = f0 cos(ωt). (3.106)
If ω 6= ω0 , the general solution is
f0
y = c1 cos(ω0 t) + c2 sin(ω0 t) + cos(ωt), (3.107)
ω02 − ω2
and if ω = ω0 , the general solution is
f0
y = c1 cos(ω0 t) + c2 sin(ω0 t) + t sin(ω0 t). (3.108)
2ω0
(exercise using undetermined coefficients, preferably in complex form for efficiency).
The first significant difference from the damped case is that when ω = ω0 , i.e. the driving
frequency equals the natural frequency, the response y grows without bound as t → +∞,
due to the t cos(ω0 t) term in (3.108). This is an extreme but idealized form of resonance.
The second significant difference is that the general solution does not separate into a
transient term and a steady state term, since there are no (decaying) exponential terms
in the general solution. This implies that the long term behaviour depends on the initial
conditions.
We consider in detail the case where the system starts in equilibrium, i.e. the initial
conditions are
y(0) = 0, y ′ (0) = 0.
It follows from equations (3.107) and (3.108) that the unique solutions are
f0
y= (cos ωt − cos ω0 t), ω 6= ω0 , (3.109)
ω02 − ω 2
and
f0
y= t sin(ω0 t), ω = ω0 . (3.110)
2ω0
In the second case, the response is a linearly growing oscillation, as shown in Figure 3.8.
In the first case, we can use the trig identity
   
A−B A+B
cos A − cos B = −2 sin sin ,
2 2
to write the solution (3.109) in the form
 
2f0
y= 1
sin 2 (ω0 − ω)t sin 21 (ω0 + ω)t. (3.111)
ω02 − ω 2
If | ω0 − ω |≪ 1 and ω0 + ω ≫| ω0 − ω |, then sin 21 (ω0 + ω)t is a rapidly oscillating function
compared to sin 12 (ω0 − ω)t. Thus, the solution (3.111) represents a rapid oscillation with
frequency 21 (ω0 + ω), but with a slowly varying sinusoidal amplitude with frequency 21 (ω0 − ω)
[see Figure 3.9]. This type of response, with a periodic variation of amplitude, exhibits what
are called beats. This phenomenon can be heard when two tuning forks of nearly equal
frequency are sounded simultaneously. In electronics, the periodic variation of amplitude
with time is called amplitude modulation.

77
 y
f0
y= 2ω0 t

t
π
ω0

Figure 3.8: Undamped response when the driving frequency equals the natural frequency.

y = sin 12 (ω0 − ω)t

t
0

Figure 3.9: Graph of y = sin 12 (ω0 − ω)t sin 21 (ω0 + ω)t, showing the phenomenon of beats.
 

78
Chapter 4

The Laplace Transform and DEs

The idea behind the Laplace transform is this:

transform a linear differential equation for y(t) into a linear algebraic equation
for Y (s), thereby making it easier to solve.

linear DE transform linear algebraic

for y(t) equation for Y (s)

standard solve
methods easily

solution y(t) solution Y (s)

inverse
transform

The function Y (s) is called the Laplace transform of the function y(t).

Comment: In its simplest form, the method is restricted to linear differential equations
with constant coefficients. It provides an alternative to the standard methods that we have
developed so far. We’ll mention its advantages as we proceed.

4.1 Elementary properties of the Laplace transform

4.1.1 Definition of the Laplace transform
The Laplace transform is one of several “transforms” that are used in engineering mathe-
matics. They are defined using the integral.

79
Definition: Given a real-valued (or complex valued) function y(t) the Laplace transform
of y is defined to be Z ∞
Y (s) = e−st y(t)dt, (4.1)
0
for all values of s for which the improper integral converges.

Recall the definition of “converges”:

Z ∞ Z r
The improper integral g(t)dt converges means that lim g(t)dt exists. 
a r→∞ a

Before discussing the significance of this definition we work out the simplest and most
important example.

Example: Find the Laplace transform of y(t) = eαt , where α is a constant.

Solution: Referring to equation (4.1), consider

Z r Z r
−st αt
e e dt = e−(s−α)t dt,
0 0
r
1 −(s−α)t
=− e , assuming s 6= α, (4.2)
s−α 0
1
= [1 − e−(s−α) ].
s−α
If s > α, then lim e−(s−α)r = 0, which by (4.2) implies that
r→∞

r
1
Z
lim e−st eαt dt = (exists).
r→∞ 0 s−α

Thus, if s > α, the Laplace transform of y(t) = eαt is

1
Y (s) = ,
s−α
by the definition. 

We think of equation (4.1) as defining an operator L which acts on a function y(t) to give
a new function Y (s). We write
L[y] = Y,
or, if we wish to indicate the arguments,

L[y(t)] = Y (s),

where Y (s) is given by (4.1). We shall refer to L as the Laplace transform operator.
Referring to the Example, we can use the operator notation to write
1
L[eαt ] = , s > α. (4.3)
s−α

80
 1
This equation is read as “the Laplace transform of eαt is s−α ”.
The idea of an “operator” which acts on functions is not unfamiliar. One can think of
the process of differentiation as defining an operator D that acts on functions, i.e. D acts
on a differentiable function f to give its derivative:

D[f ] = f ′ .

The operator D and L have one important property in common. The operator D satisfies

D[f1 + f2 ] = Df + D[f2 ],

and
D[cf ] = cD[f ],
where c is a constant. These equations are simply the sum property and “multiplication by
a constant” property of derivatives. These equations mean that D is a linear operator.1

Proposition: The Laplace transform operator L is a linear operator:

L[y1 + y2 ] = L[y1] + L[y2],

and
L[cy] = cL[y],
where c is a constant.

Proof: These results follow from the linearity of the integral:

Z b Z b Z b
[y1 (t) + y2 (t)]dt = y1 (t)dt + y2 (t)dt,
a a a

and Z b Z b
cy(t)dt = c y(t)dt,
a a

by taking limits to obtain the improper integrals. We skip the details. 

This proposition is used extensively when working with the Laplace transform.

Before continuing, we pause briefly to consider a theoretical question: what conditions

on f are needed to ensure that L[f ] exists?
Speaking intuitively, we can say that L[f ] exists provided that

(i) f (t) does not grow too rapidly as t → +∞,

and

(ii) the only discontinuities of f for t ≥ 0 are jump discontinuities.

1
This is the same concept as in linear algebra, where a 3 × 3 matrix A acting on vectors x in R3 (for
example) satisfies A(x1 + x2 ) = Ax1 + Ax2 , and A(λx) = λAx, thereby defining a linear transformation.

81
 First, to describe the growth condition it is convenient to use the big-O notation. Recall
the definition:
f (t) = O(g(t)) as t → ∞,
means that there exist constants B and b such that

| f (t) |≤ B | g(t) | for all t ≥ b.

The required growth condition is that

f (t) = O(eat ) as t → ∞, (4.4)

for some constant a. Intuitively (4.4) means that f (t) does not grow more rapidly than eat
as t → ∞.
Second, f has a jump discontinuity at c means that the 1-sided limits

lim f (t) and lim f (t)

t→c+ t→c−

exist and are unequal. The value of f at c is unimportant. The statement “f is piecewise
continuous on a finite interval I” means that f is continuous on I except for a finite number
of jump discontinuities.

f (t)

t
0 t1 t2 b

Figure 4.1: A piecewise continuous function with two jump discontinuities.

A function such as this (a pulse function) could act as the driving force of an oscillator.

Proposition: (Existence of the Laplace transform)

If H1 : f is piecewise continuous on each interval 0 ≤ t ≤ r,

H2 : there is a constant a such that

f (t) = O(eat ) as t → ∞,

then the Laplace transform L[f ] exists for s > a.

82
Proof: Outline.
Z r
By H1 , f (t)dt exists (just divide the interval into subintervals on which f is continu-
0
ous).
By H2 , | f (t) |≤ Beat for t ≥ b. Hence

| e−st f (t) |≤ Be−(s−a)t , for t ≥ b.

Z ∞
If s > a, Be−(s−a)t dt converges. Hence by the Comparison Test for improper integrals,
Z ∞ 0
−st
e f (t)dt converges i.e. L[f ] exists by definition. 
0

Examples:

(i) f (t) = (1 + t)t does not satisfy the growth condition H2 , and L[f ] does not exist.
1
(ii) f (t) = t
is not piecewise continuous on any interval [0, r], and L[f ] does not exist. 

4.1.2 Calculating Laplace transforms

Our goal is to build up a table of Laplace transforms of elementary functions, which we can
use to solve differential equations.

So far we have shown that

1
L[eαt ] = , for s > α. (4.5)
s−α
This formula includes the important special case of the constant function f (t) = 1, i.e.
choose α = 0:
1
L[1] = , for s > 0. (4.6)
s
Another useful example is the Laplace transform of tn .

Example: Show that

n!
L[tn ] = , for s > 0, (4.7)
sn+1
where n ≥ 0 is an integer.

Solution: Outline (fill in the details as an exercise).

Use induction on n, with (4.6) giving the result for n = 0.

Z b
Apply integration by parts to e−st tk+1 dt to reduce the power of t by 1, and then let
0
b → ∞. Then the case n = k will imply the case n = k + 1. You will also need the result

tk+1
lim = 0 for s > 0,
b→∞ esb

83
i.e. an exponential dominates a power as b → ∞. 

The direct evaluation of Laplace transforms using the definition can be time-consuming,
and so one seeks short cuts, for example

(i) by using complex functions,

(ii) by developing theorems to give new Laplace transforms from old, without extra calcu-
lation.

We first illustrate (i). The derivation leading to (4.5) is valid if α is complex, provided
that we make one small change:

lim e−(s−α)r = 0 provided that s > Re(α),

r→∞

where Re(α) is the real part of α. We can now use Euler’s formula

eibt = cos bt + i sin bt (4.8)

and the linearity of L to calculate L[cos bt] and L[sin bt] directly from (4.5).

Example: Show that

s b
L[cos bt] = , L[sin bt] = . (4.9)
s2 + b2 s2 + b2

Solution: By linearity of L and (4.8):

L[eibt ] = L[cos bt] + iL[sin bt]. (4.10)

The right hand side of (4.5), with α = ib is

1 s + ib s + ib
= = 2 . (4.11)
s − ib (s − ib)(s + ib) s + b2

Choosing α = ib in (4.5) and substituting (4.10) and (4.11) gives

s b
L[cos bt + i sin bt] = + i .
s2 + b2 s2 + b2

Equating real and imaginary parts leads to (4.9). 

We now give a theorem which provides a quick way of calculating L[ect f (t)] if L[f (t)] is
known.

84
First Shift Theorem:
If L[f (t)] = F (s) exists for s > a ≥ 0, then

L[ect f (t)] = F (s − c), for s > a + c, (4.12)

where c is a constant.

Proof: Consider Z r Z r
−st ct
e e f (t)dt = e−(s−c)t f (t)dt. (4.13)
0 0
Z r
By the hypothesis, lim e−(s−c)t f (t)dt exists for s − c > a, and equals F (s − c). The result
r→∞ 0
follows by letting r → ∞ in (4.13). 

Example: Calculate the Laplace transform of g(t) = tect .

1
Solution: By (4.7), L[t] = s2
for s > 0. Thus by the Shift Theorem,
1
L[tect ] = for s > c. 
(s − c)2

Exercise: Calculate L[e−2t sin πt].

π
Answer: . 
(s + 2)2 + π 2

4.1.3 The Laplace transform of a derivative

In using the Laplace transform to solve a linear DE,

y ′ + ky = f (t),

where k is a constant, we begin by applying the operator L to the DE:

L[y ′ + ky] = L[f ].

By linearity of L, we obtain
L[y ′] + kL[y] = L[f ].
To proceed further we need to relate L[y ′ ] to L[y]. The next proposition does just that.

Proposition 1:

If H1 : f (t) = O(eat ) as t → ∞, for some a,

H2 : f ′ is piecewise continuous and f is continuous, on any
interval [0, r],
then L[f ′ (t)] exists, and

L[f ′ (t)] = sL[f (t)] − f (0), for s > a. (4.14)

85
Proof: For simplicity, we assume that f ′ is continuous.
Using integration by parts:
Z r Z r
−st ′
 −sr
(−s)e−st f (t)dt.

e f (t)dt = e f (r) − f (0) − (4.15)
0 0

By H1 , | e−sr f (r) |≤ Be−(s−a)r . Thus, if s > a, lim e−sr f (r) = 0.

r→∞
The result now follows by letting r → ∞ in (4.15). 

Proposition 1 extends in a natural way to the second derivative.

Proposition 2:

If H1 : f (t) = O(eat ) and f ′ (t) = O(eat ) as t → ∞,

H2 : f ′′ is piecewise continuous and f ′ , f are continuous, on
any interval [0, r],
then L[f ′′ (t)] exists, and
L[f ′′ (t)] = s2 L[f (t)] − sf (0) − f ′ (0), for s > a. (4.16)

Proof: Apply Proposition 1, with f replaced by f ′ , obtaining

L[f ′′ (t)] = sL[f ′ (t)] − f ′ (0).

Now apply Proposition 1 to the right side directly to get

L[f ′′ (t)] = s {sL[f (t)] − f (0)} − f ′ (0). 

4.1.4 The Inverse Laplace Transform

Referring to the diagram on the first page of this Chapter, the final (and most difficult) step
in using the Laplace transform to solve a DE is this:

given a Laplace transform Y (s), find the function y(t).

One should first ask: given Y (s), is there a unique y(t)?

The next proposition gives the answer YES.

Proposition: If y1 and y2 are continuous functions of order eat as t → ∞, then

y1 6= y2 implies L[y1 ] 6= L[y2 ]. 

Proof: The proof is beyond the scope of this course. See for example, Brauer & Nohel,
page 391-2. 

86
 This proposition states that the Laplace transform operator L is a one-to-one operator,
and hence has an inverse operator L−1 which maps a Laplace transform Y (s) onto the original
function y(t):
L−1 [F (s)] = f (t) means that L[f (t)] = F (s).
Equivalently we can write

L−1 [L[f (t)]] = f (t) and L[L−1 [F (s)]] = F (s).

We shall call L−1 the inverse Laplace transform operator, and shall call f (t) = L−1 [F (s)]
the inverse Laplace transform of F (s). Note that L−1 is also linear. 

In elementary discussions, inverse Laplace transforms are found by referring to a table

of Laplace transforms – calculating them directly is difficult (and sometimes impossible).
Based on the results of Section 4.1.2 we can construct the following Table:

f (t) F (s) = L[f (t)]

1
eαt
s−α
s
cos bt
s + b2
2

b
sin bt
s + b2
2

n!
tn n+1
s
ect f (t) F (s − c)

For example, we can conclude directly from the Table that

   
−1 1 −t −1 s
L =e , L = cos πt.
s+1 s2 + π 2
In the examples we shall encounter, the Laplace transform F (s) will mostly be a rational
function, i.e.
p(s)
F (s) = ,
q(s)
where p(s) and q(s) are polynomial functions. It will be necessary to use partial fraction
methods to write F (s) as a sum of the simple terms appearing in the Table. The first shift
theorem (summarized in the last line of the Table) will also be useful.
 
−1 s
Example: Find L .
s2 + 5s + 6
Solution: Using partial fractions
s s 3 2
= = − .
s2 + 5s + 6 (s + 2)(s + 3) s+3 s+2

87
Using the Table and linearity of L−1 :
     
−1 s −1 3 −1 2
L =L −L
s2 + 5s + 6 s+3 s+2
−3t −2t
= 3e − 2e . 

 
−1 s
Example: Find L .
s2 + 4s + 5

Solution: Since the denominator does not factor, we complete the square:
s s (s + 2) 2
= = − .
s2 + 4s + 5 2
(s + 2) + 1 (s + 2) + 1 (s + 2)2 + 1
2

From the Table (using the Shift Theorem),

s+2 1
L[e−2t cos t] = , L[e−2t sin t] = .
(s + 2)2 + 1 (s + 2)2 + 1
It thus follows that
 
−1 s
L = e−2t cos t − 2e−2t sin t. 
s2 + 4s + 5

In view of this example it is convenient to restate the First Shift Theorem in terms of
the inverse Laplace transform operator L−1 .

First Shift Theorem (inverse form):

If L−1 [F (s)] = f (t), then L−1 [F (s − c)] = ect f (t).

   
−1 s−3 −1 s+4
Exercise: Find (i) L (ii) L
s2 − 2s + 5 s2 + 2s

Answer: (i) et (cos 2t − sin 2t) (ii) 2 − e−2t .

4.1.5 The Heaviside Step Function

A driving force f (t) [or input function] that changes abruptly with time can be modelled by
using a discontinuous function, and one that does not change smoothly can be modelled by
a non-differentiable function.

When working with the Laplace transform of functions such as these it is convenient to
introduce the Heaviside or unit step function H(t), defined by
(
0, if t < 0
H(t) = (4.17)
1, if t ≥ 0.

88
 y y

y = f1 (t) y = f2 (t)
1 1

t t
a a

Figure 4.2: a step function f1 (t) a ramp function f2 (t).

Then the above step function f1 (t) is given by

f1 (t) = H(t − a), (4.18)

(translation to the right).

Example: Express the ramp function f2 (t) above in terms of H(t).

Solution: We have
f2 (t) = a1 t, H(t − a) = 0 for 0 ≤ t < a,
f2 (t) = 1, H(t − a) = 1, for t ≥ a.

Thus  
1 1
f2 (t) = t + − t + 1 H(t − a), for t ≥ 0.
a a
Thus the ramp function f2 (t) can be written

f2 (t) = a1 [t − (t − a)H(t − a)]. (4.19)

The next step is to calculate the Laplace transform of H(t − a).

Example: Show that

e−as
L[H(t − a)] = , (4.20)
s
for s > 0, where a > 0 is a constant.

Solution: Consider, with r > a,

Z r Z r
−st
e H(t − a)dt = e−st dt, since H(t − a) = 0 if t < a,
0 a
1
= − (e−sr − e−sa ),
s
89
by the Fundamental Theorem of Calculus. Thus, if s > 0,
Z r
lim e−st H(t − a)dt
r→∞ 0

exists and equals 1s e−sa . The result follows by definition of the Laplace transform. 

We now give a theorem which provides a quick way of calculating L[H(t − c)f (t − c)] if
L[f (t)] is known.

Second Shift Theorem:

If L[f (t)] = F (s) exists for s > a ≥ 0, and c is a positive constant, then

L[H(t − c)f (t − c)] = e−cs F (s), s > a. (4.21)

Proof: Consider
Z r Z r
−st
e H(t − c)f (t − c)dt = e−st f (t − c)dt, by definition of H,
0
Zc r−c
= e−s(u+c) f (u)du, by making a change of variable u = t − c,
0
Z r−c
−cs
=e e−su f (u)du.
0
(4.22)
Z r−c
Since L[f (t)] exists, lim e−su f (u)du exists and equals F (s) (note that r − c → ∞,
r→∞ 0
and the integration variable u can be relabelled t). The result follows by letting r → ∞ in
(4.22). 

Example: Calculate the Laplace transform of the ramp function (see Figure 4.2)

f2 (t) = a1 [t − (t − a)H(t − a)].

Solution: By linearity of L,
1
L[f2 (t)] = a
{L[t] − L[(t − a)H(t − a)]} .
1
Since L[t] = s2
(see the Table in Section 4.1.4), the Second Shift Theorem gives

e−as
 
1 1 1
L[f2 (t)] = 2
− 2 = 2 (1 − e−as ). 
a s s as

Another classic example is the saw-tooth function:

90
 y

y = f3 (t)
1

t
a 2a

Figure 4.3: The saw-tooth function.

Exercise: Show that

f3 (t) = a1 [t − 2(t − a)H(t − a) + (t − 2a)H(t − 2a)].

Exercise: Show that 2

1 − e−as

1
L[f3 (t)] = .
a s

The Second Shift Theorem can be restated in terms of the inverse operator L−1 , thereby
providing a useful tool for calculating inverse Laplace transforms.

Second Shift Theorem (inverse form):

If L−1 [F (s)] = f (t), then L−1 [e−cs F (s)] = H(t − c)f (t − c).
h −2s i
Example: Find L−1 es+3 .

1
We know L−1 = e−3t , from the Table. Thus, by the Second Shift Theorem,
 
Solution: s+3
(
−2s
 
e 0, 0≤t<2
L−1 = H(t − 2)e−3(t−2) = −3(t−2)

s+3 e , t ≥ 2.

4.2 Solving DEs using the Laplace Transform

We have now developed the machinery needed to implement the algorithm outlined on the
first page of this Chapter. In this Section we give examples to illustrate its use.

4.2.1 First order DEs

Example: Solve the first order initial value problem

y ′ + 3y = 13 sin(2t); y(0) = y0 . (4.23)

91
Solution: Apply the Laplace transform operator L to the DE and use linearity to obtain

L[y ′] + 3L[y] = 13L[sin(2t)]. (4.24)

By Proposition 1 in Section 4.1.3,

L[y ′ ] = sL[y] − y(0),

and by the Table in Section 4.1.4,

2
L[sin(2t)] = .
s2 +4
Writing L[y(t)] = Y (s) as usual, and using the initial condition (4.23), equation (4.24)
becomes
26
sY − y0 + 3Y = 2 .
s +4
Solving for Y (s) gives
y0 26
Y (s) = + . (4.25)
s + 3 (s + 3)(s2 + 4)
The next step is to calculate y(t) = L−1 [Y (s)], the inverse Laplace transform of Y (s). Before
applying L−1 to (4.25), we expand the second term in partial fractions (exercise):
 
26 1 s 3
=2 − + . (4.26)
(s + 3)(s2 + 4) s + 3 s2 + 4 s2 + 4

Thus, applying L−1 to (4.25) and using (4.26) gives

       
−1 −1 1 −1 1 −1 s −1 2
L [Y (s)] = y0 L + 2L − 2L + 3L .
s+3 s+3 s2 + 4 s2 + 4
Referring to the Table, we get

y(t) = (y0 + 2)e−3t − 2 cos(2t) + 3 sin(2t).

as the solution of the initial value problem (4.23). 

Exercise: Solve the first order initial value problem

y ′ − y = 2et , y(0) = y0 .

Answer: y(t) = y0 et + 2tet .

1
Hint: You will need to use the First Shift Theorem with L[t] = s2
. 

Here’s an example that involves arbitrary parameters.

Exercise: Solve the initial value problem:

y ′ + ky = kA cos(ωt); y(0) = y0 .

92
Answer:
kA  −kt
+ y0 e−kt .

y(t) = 2 2
k(cos ωt − e ) + ω sin ωt
k +ω

Hint: You will need the following partial fraction expansion:

ω2
 
s 1 k ks
= 2 − + + .
(s + k)(s2 + ω 2) k + ω2 s + k s2 + ω 2 s2 + ω 2

Technical comment:

In applying L to the DE we used Proposition 1 in Section 4.1.3. You may ask: are the
hypotheses of the Proposition satisfied in this application? We can be sure that H2 is satis-
fied, since the solution of any constant coefficient linear DE with continuous input function
is continuous and hence has a continuous derivative (y ′ = −ky + f (t), both continuous).
What about H1 : y(t) = O(eat ) as t → ∞ for some a? We have no way of verifying this in
advance, since y(t) is the unknown function. So we proceed “on faith”, and having found
the solution, verify after the fact that the hypothesis is satisfied. 

4.2.2 Second order DEs

Example: Solve the second order initial value problem:

y ′′ + 2y ′ + 2y = 3e−t ; y(0) = y0 , y ′(0) = 0. (4.27)

Solution: Apply the operator L to the DE and use linearity to obtain

L[y ′′ ] + 2L[y ′] + 2L[y] = 3L[e−t ]. (4.28)

As in Section 4.1.3 the transforms of the derivatives are

L[y ′] = sY (s) − y(0),
L[y ′′] = s2 Y (s) − sy(0) − y ′ (0),

where Y (s) = L[y(t)], and the Table gives

1
L[e−t ] = , s > −1.
s+1
Thus (4.28) becomes
3
s2 Y − sy0 + 2[sY − y0 ] + 2Y = ,
s+1
which is a linear algebraic equation for the unknown Y (s). Collecting like terms gives

3
(s2 + 2s + 2)Y (s) = (s + 2)y0 + .
s+1

93
Solving for Y (s) gives

s+2 3
Y (s) = y 0 + . (4.29)
s2 + 2s + 2 (s + 1)(s2 + 2s + 2)

Since s2 + 2s + 2 = (s + 1)2 + 1, we express both terms in terms of (s + 1). First,

s+2 s+1 1
= + . (4.30)
s2 + 2s + 2 (s + 1) + 1 (s + 1)2 + 1
2

Second,
3 3 3(s + 1)
= − . (4.31)
(s + 1)(s2+ 2s + 2) s + 1 (s + 1)2 + 1

Aside: The simple way to get (4.31) is to observe that

1 1 u
= − 2 ,
u(u2 + 1) u u +1

and choose u = (s + 1).

We now use the First Shift Theorem, knowing
   
−1 s −1 1
L = cos t, L = sin t.
s2 + 1 s2 + 1

It follows from (4.30) and (4.31) that

 
−1 s+2
L = e−t cos t + e−t sin t,
s2 + 2s + 2
  (4.32)
−1 3
L = 3e−t − 3e−t cos t.
(s + 1)(s2 + 2s + 2)

We now apply the operator L−1 to (4.29) and use (4.32) to obtain

y(t) = L−1 [Y (s)] = y0 e−t (cos t + sin t) + 3e−t (1 − cos t),

as the solution of the initial value problem (4.27). 

Exercise: Solve
y ′′ + 2y ′ + y = t; y ′ (0) = 1, y(0) = 0.

Answer: y = −2 + t + 2e−t (1 + t). 

Comment: The examples in this Section illustrate one advantage of the Laplace transform
method, namely that the initial conditions are automatically incorporated.

94
 Figure 4.4: A step function and a saw-tooth function.

4.2.3 Discontinuous and Non-differentiable Inputs

A second advantage of the Laplace transform method is that non-smooth input functions
e.g. step functions and saw-tooths, can be dealt with easily.
Consider a system (e.g. a mixing tank or electrical circuit) whose state y(t) is governed
by the DE
y ′ + y = f (t),
where f (t) is the input function (assume that dimensionless variables have been introduced).
If the input function is f (t) = 0, the solution is

y(t) = y(0)e−t ,

showing that the system decays exponentially to its natural equilibrium state y(t) = 0. If
the input function is a constant, say f (t) = 1, the solution is

y(t) = 1 + [y(0) − 1]e−t ,

i.e. the system exponentially approaches a new equilibrium state y(t) = 1. The question we
now ask is: what is the response y(t) of the system if the input function is a step function:
(
1, 0 ≤ t < b
f (t) = (4.33)
0, t ≥ b,

i.e. a constant input of 1 for a finite time?

Example: Solve the initial value problem

y ′ + y = f (t); y(0) = y0 , (4.34)

where f (t) is the step function (4.33).

Solution: We begin by writing the input function in terms of the Heaviside step function

f (t) = 1 − H(t − b). (4.35)

Apply the operator L to the DE (4.34) with (4.35), using linearity to obtain

L[y ′ ] + L[y] = L[1] − L[H(t − b)]. (4.36)

We know from equation (4.20) that

e−bs
L[H(t − b)] = , s>0
s
and from Section 4.1.3 that
L[y ′ ] = sL[y] − y(0).

95
Also, by equation (4.6)
1
L[1] = .
s
Thus, writing L[y] = Y (s) as usual, equation (4.36) becomes

1 e−bs
sY − y0 + Y = − .
s s
Solving for Y (s) leads to
y0 1 − e−bs
Y (s) = + . (4.37)
s + 1 s(s + 1)
In order to find the inverse transform, we first use the partial fraction expansion
1 1 1
= −
s(s + 1) s s+1

to write (4.37) in the form

 
y0 − 1 1 1 1
Y (s) = + − e−bs − . (4.38)
s+1 s s s+1

We now use the Second Shift Theorem (Section 4.1.5). We know that
 
−1 1 1
L − = 1 − e−t ,
s s+1

from which follows

  
−1 −bs 1 1
L e − = H(t − b)(1 − e−t+b ).
s s+1

Thus, applying L−1 to (4.38) gives

y(t) = L−1 [Y (s)] = (y0 − 1)e−t + 1 − H(t − b)(1 − e−t+b ). (4.39)

as the solution of the initial value problem (4.34). 

Analysis of the solution:

Without further analysis the solution (4.39) does not give one much insight. The solution
can be written in the form
(
1 + (y0 − 1)e−t , 0 ≤ t < b
y(t) = (4.40)
(y0 − 1 + eb )e−t , t ≥ b.

The derivative is (
−(y0 − 1)e−t , 0≤t<b
y ′ (t) = b −t
(4.41)
−(y0 − 1 + e )e , t ≥ b.
Note that y ′(b) does not exist, but that y(t) is continuous at t = b.

96
 It follows from (4.40) that

lim y(t) = 0 for all values of y0 ,

t→∞

i.e. the system approaches its natural equilibrium state y(t) = 0 once the input is switched
off at t = b. Equation (4.41) shows that the intermediate behaviour depends on y0 , and that
there are two special values of y0 , namely

y0 = 1 and y0 = −eb + 1.

By considering the sign of y ′ (t), we reach the following conclusions.

• If y0 − 1 > 0, then y0 − 1 + eb > 0 and y(t) is decreasing for t ≥ 0.

• If y0 − 1 + eb < 0, then y0 − 1 < 0 and y(t) is increasing for t ≥ 0.

• If −eb + 1 < y0 < 1, then the response attains a maximum value

ymax = 1 + (y0 − 1)e−b > 0

at t = b.

These results enable us to sketch the graphs in Figure 4.5.

Exercise: Solve y ′ + y = f (t); y(0) = y0 .

Answer: y(t) = (y0 − 1)e−t + 1 − 2H(t − b)(1 − e−t+b ) + H(t − 2b)(1 − e−t+2b ) 

97
 y

y0 > 1

decreasing
for t ≥ 0

y0 = 1

0 < y0 < 1

y0 = 0 t
0

−eb + 1 < y0 < 0

y0 = −eb + 1
increasing
for t ≥ 0 t=b
y0 < −eb + 1

Figure 4.5: The family of solutions (4.40) with initial state y0 as parameter. The exceptional
solutions given by y0 = 1 and y0 = −eb + 1 are shown in bold.

+1 y = f (t)

t
b 2b

−1

Figure 4.6:

98
Chapter 5

Linear Vector Differential Equations

In this Chapter we generalize the concept of differential equations to the case where the
unknown is a vector function in R2 . We begin by discussing two familiar physical systems
from a different point of view, in order to motivate the idea.

5.1 Introduction
5.1.1 Coupled Mixing Tanks
Consider a system of two coupled mixing tanks each of volume V , with flow rates as shown,
and constant inflow concentration c. Let m1 (t) and m2 (t) denote the mass of chemical
in the two tanks at time t, respectively. Introduce a characteristic time tc = V /f and a
dimensionless time τ = t/tc (see equation (2.11)). Then the mass balance equation (1.48)
applied to each tank separately leads to the two DEs

f, c f

2f f
m1 (t), V m2 (t), V

Figure 5.1: A system of two coupled mixing tanks.

m′1 = −2m1 + m2 + cV
m′2 = 2m1 − 2m2 ,

99
where ′ denotes differentiation with respect to τ . Fill in the details as an exercise, referring
to Sections 1.3.1 and 2.1.3, if necessary. We describe the state of this system at time τ by
the vector  
m1 (τ )
x(τ ) = ∈ R2 .
m2 (τ )

The two scalar DEs can be written as one vector DE

 ′     
m1 −2 1 m1 cV
= + , (5.1)
m2 2 −2 m2 0

using the 2 × 2 coefficient matrix. Using the state vector x we write (5.1) in vector notation
as
x′ = Ax + f, (5.2)
where    
−2 1 cV
A= , and f = .
2 −2 0
We refer to (5.2) as a first order vector DE in R2 . 

Comment: The notation used in equation (5.2) makes sense in Rn . For example, one can
imagine a system of 3 coupled mixing tanks leading to a vector DE in R3 .

5.1.2 The mechanical oscillator

The motion of a damped mass-spring system with applied force is described by the second
order DE
my ′′ + cy ′ + ky = F (t),
(see equation (3.4)). The dimensionless version of this DE is

k m
F (t)

0 y

damping constant c

Figure 5.2: A damped mass-spring system.

y ′′ + 2ζy ′ + y = f (τ ), (5.3)
where ζ is the dimensionless damping constant (see equation (3.88) in Section 3.3.3).

100
 To describe the state of the system at an instant of time it is not enough to give the
dy dy
displacement y: one has also to give the velocity , or , in terms of the dimensionless
dt dτ
time τ .
So we introduce the state vector
 
y
x= ∈ R2 ,
y′
with components
x1 = y and x2 = y ′, (5.4)
where ′ denotes differentiation with respect to τ .
But what vector DE does the state vector satisfy? Well, from (5.4),
x′1 = y ′ = x2 ,
and from (5.3) and (5.4)
x′2 = y ′′ = −2ζy ′ − y + f (τ )
= −2ζx2 − x1 + f (τ ),
Collecting the results, we have
x′1 = x2
x′2 = −x1 − 2ζx2 + f (τ ).
In vector form this reads
x′ = Ax + f, (5.5)
where    
0 1 0
A= , f= . 
−1 −2ζ f

Comment: One can obtain vector DEs in higher dimensions in this context. The system
shown will have a state vector

 
y1
 y2  4
y1′  ∈ R .
x= 

y2′

Exercise: The procedure leading from the second order linear DE (5.3) to the linear
vector DE (5.5) can be used to write any second order linear DE as a vector DE. Show that
y ′′ + py ′ + qy = 0 (5.6)
is equivalent to
   
′ y 0 1
x = Ax, with x = , and A = . (5.7)
y′ −q −p

101
 k1 m k2 m k1

y1 y2

Figure 5.3: A two-mass oscillator.

5.1.3 Overview
From a mathematical point of view the object of study in this chapter is a linear vector DE
of the form
dx
= Ax + f(t),
dt
or more concisely,
x′ = Ax + f(t). (5.8)
Here  
x1 (t)
x= ∈ R2 (5.9)
x2 (t)
is the unknown vector-valued function,
 
a11 a12
A= (5.10)
a21 a22

is the 2 × 2 coefficient matrix, whose entries are constants, and

 
f1 (t)
f(t) = (5.11)
f2 (t)

is a given vector-valued function. The initial condition is of the form

x(0) = a, (5.12)
 
a1
where a = is a given vector.
a2
The DE (5.8) should be thought of as describing the evolution in time of a physical system
(e.g. coupled mixing tanks or mechanical oscillator). The unknown vector-valued function
x is the state vector of the system. The constant coefficient matrix A describes the internal
characteristics of the system (e.g. flow rate, damping), and the vector-valued function f(t)
describes the external input to the system.
The system can be represented symbolically in a so-called

The goal is to determine the state x(t) of the system at time t (the “output”), given the
input f(t) and initial state x(0) = a. In this Chapter we shall develop algorithms to solve
this problem.

102
 x(0) = a

input physical system output

f(t) (matrix A) x(t)
block diagram:

Figure 5.4: Block diagram for the physical system described by the linear vector DE x′ =
Ax + f(t).

Finally we note that it is sometimes helpful to write a vector DE (5.7) in component

form. Using (5.9)-(5.11), (5.7) can be written as
(
x′1 = a11 x1 + a12 x2 + f1 (t)
(5.13)
x′2 = a21 x1 + a22 x2 + f2 (t),

which is referred to as a system of linear DEs. Indeed the terms “linear vector DE” and
“system of linear DEs” have the same meaning. We prefer the former, because it emphasizes
the fact that the unknown is a vector-valued function.

5.1.4 Linearity and Superposition

We shall begin be considering the case of zero external input, in which case the DE (5.7)
reduces to
x′ = Ax, (5.14)
referred to as a homogeneous linear vector DE (the zero function x(t) = 0 is a solution).
As in the case of a homogeneous linear DE for a scalar (see Section 3.2.1), the Principle
of Superposition holds for a homogeneous linear vector DE.

Proposition: If x(1) (t) and x(2) (t) are solutions of

x′ = Ax,
then
c1 x(1) (t) + c2 x(2) (t)
is also a solution for any constant scalars c1 and c2 .

Proof: We are given that

′ ′
x(1) = Ax(1) and x(2) = Ax(2) .
Multiply the first by c1 , the second by c2 , and add, using the matrix property
A(c1 x(1) + c2 x(2) ) = c1 Ax(1) + c2 Ax(2) ,

103
to get
(c1 x(1) + c2 x(2) )′ = A(c1 x(1) + c2 x(2) ),
as required. 

For a linear vector DE (5.14) in R2 , the general solution must contain two arbitrary
constants because the initial condition (5.12) contains two arbitrary constants, i.e. the
components of the vector a. We can thus state:

the general solution of the homogeneous linear vector DE in R2 ,

x′ = Ax,

is of the form
x = c1 x(1) (t) + c2 x(2) (t), (5.15)
where x(1) and x(2) are two linearly independent solutions of the DE, and c1 , c2
are arbitrary constants.

5.2 Solving vector DEs using eigenvectors

5.2.1 The method
Consider a homogeneous linear vector DE in R2 :

x′ = Ax. (5.16)

As discussed in Section 5.1.4 we need to obtain two linearly independent solutions. We

consider a trial function containing an exponential (surprise!)

x = eλt v, (5.17)

where v ∈ R2 is a constant vector, and λ is a constant scalar (which may be complex).

The derivative of (5.17) is
x′ = λeλt v.
Substituting in (5.16) gives
λeλt v = A(eλt v) = eλt Av.
Multiply by the scalar e−λt obtaining

Av = λv. (5.18)

This equation states that the scalar λ in the trial function (5.17) is an eigenvalue of the
coefficient matrix A, and v is an associated eigenvector.
The eigenvalues can be found by rewriting (5.18) in the form

(A − λI)v = 0,

where  
1 0
I=
0 1

104
is the 2 × 2 identity matrix. This equation will have a non-zero solution for v iff the matrix
A − λI is non-invertible i.e., the inverse matrix (A − λI)−1 does not exist. This is the case
iff the determinant of A − λI is zero. We let

h(λ) = det(A − λI). (5.19)

Then the eigenvalues of A are the roots of the equation

h(λ) = 0, (5.20)

which is called the characteristic equation of A. The function h(λ) defined by (5.19) is in
fact a polynomial of degree two:
 
a11 − λ a12
h(λ) = det = (a11 − λ)(a22 − λ) − a12 a21 .
a21 a22 − λ

Here we used the usual formula for the determinant of a 2 × 2 matrix. The function h(λ) is
called the characteristic polynomial of the matrix A.
If λ is an eigenvalue of A (i.e. a solution of (5.20)) then (5.18) will have a non-zero
solution for v, and (5.17) will be a solution of the DE (5.16). There are three cases:

A) Unequal real eigenvalues

B) Complex eigenvalues

C) Equal real eigenvalues,

which we illustrate with examples in the rest of this Section. In case A) we immediately
get two linearly independent solutions of the DE (5.16) (since eigenvectors associated with
distinct eigenvalues are linearly independent). In case B) we have to take real and imaginary
parts of the solutions, while case C) requires special treatment.

5.2.2 Unequal real eigenvalues

Example 1: Find the general solution of the vector DE
 
′ −4 1
x = Ax, A= . (5.21)
−2 −1

Solution: We consider a trial function

x = eλt v. (5.22)

Substituting (5.22) in (5.21) leads to

(A − λI)v = 0. (5.23)

Thus λ must satisfy

h(λ) = 0,

105
where the characteristic polynomial is
 
−4 − λ 1
h(λ) = det(A − λI) = det
−2 −1 − λ
= (−4 − λ)(−1 − λ) − (−2)(1)
= λ2 + 5λ + 6
= (λ + 2)(λ + 3).
Thus the eigenvalues of A are
λ = −2 and − 3.
First, considering λ = −2, equation (5.23) becomes
    
−2 1 v1 0
= ,
−2 1 v2 0
leading to
−2v1 + v2 = 0
−2v1 + v2 = 0,
which requires that v2 = 2v1 (the two equations are identical). In other words, any vector
 
v1
v=
2v1
with v1 6= 0, is an eigenvector of A associated with the eigenvalue λ = −2. Choosing v1 = 1,
it follows from (5.22) that  
−2t 1
x=e (5.24)
2
is a solution of the DE (5.21).
Second, considering λ = −3, equation (5.23) becomes
    
−1 1 v1 0
= ,
−2 2 v2 0
which implies v1 = v2 . In other words, any vector
 
v
v= 1 ,
v1
with v1 6= 0, is an eigenvector of A with eigenvalue λ = −3. Choosing v1 = 1 it follows from
(5.22) that  
−3t 1
x=e (5.25)
1
is a solution of the DE (5.21).   
1 1
Finally, since the vectors and are linearly independent, the solutions (5.24)
2 1
and (5.25) are linearly independent. Thus by Section 5.1.4 (see equation (5.15)) the general
solution of the DE (5.21) is
   
−2t 1 −3t 1
x = c1 e + c2 e , (5.26)
2 1
where c1 and c2 are arbitrary constants. 

106
Initial conditions:
The constants c1 and c2 are determined by the initial condition
 
a1
x(0) = a = .
a2

Setting t = 0 in (5.26) gives  

c1 + c2
a= ,
2c1 + c2
which can be solved to give    
c1 a2 − a1
= . (5.27)
c2 2a1 − a2
   
1 0
Two special cases will be important later, namely a = and a = . From (5.26)
0 1
and (5.27) we obtain:
 −2t
−e + 2e−3t
  
1
x= −2t −3t , with x(0) = , (5.28)
−2e + 2e 0

and

e−2t − e−3t
   
0
x= , with x(0) = . (5.29)
2e−2t − e−3t 1

5.2.3 Complex eigenvalues

Example 2: Find the general solution of the vector DE
 
′ 1 5
x = Ax, A= . (5.30)
−1 −3

Solution: We consider a trial function

x = eλt v. (5.31)

As in Example 1, λ must be an eigenvalue of A i.e. λ must be a solution of

h(λ) = det(A − λI) = 0,

and v must be an eigenvector, i.e. a non-zero solution of

(A − λI)v = 0 . (5.32)

107
The characteristic equation is
 
1−λ 5
h(λ) = det = (λ + 1)2 + 1,
−1 −3 − λ

after simplifying and completing the square. Setting h(λ) = 0, we obtain the eigenvalues

λ = −1 + i, −1 − i.

Choosing λ = −1 + i, equation (5.32) becomes

    
2−i 5 v1 0
= ,
−1 −2 − i v2 0
leading to

(2 − i)v1 + 5v2 = 0,
−v1 − (2 + i)v2 = 0.

These two equations are essentially the same, since multiplying the second by −(2 − i) yields
the first. Thus
v2 = − 51 (2 − i)v1 ,
and choosing v1 = 5 gives  
5
v=
−2 + i
as a solution of equation (5.32) in the case λ = −1 + i.
Then, using (5.31), we obtain
 
(−1+i)t 5
x=e (5.33)
−2 + i

as a complex solution of the DE (5.30). The real and imaginary parts of this solution are
themselves solutions of (5.30). Using Euler’s formula we decompose (5.33) into its real and
imaginary parts:
   
−t 5 0
x = e (cos t + i sin t) +i
−2 1
         
−t 5 0 5 0
=e cos t − sin t + i sin t + cos t
−2 1 −2 1
   
5 cos t 5 sin t
= e−t + ie−t .
−2 cos t − sin t cos t − 2 sin t
These two solutions are linearly independent since one is not a multiple of the other. Thus,
by equation (5.15), the general solution of the DE (5.30) is
    
−t 5 cos t 5 sin t
x=e c1 + c2 , (5.34)
−2 cos t − sin t cos t − 2 sin t
where c1 and c2 are arbitrary constants. 

108
5.2.4 Equal real eigenvalues
Example 3: Find the general solution of the vector DE
 
′ −3 4
x = Ax, A = . (5.35)
−1 1

Solution: We consider a trial function

x = eλt v. (5.36)

As in Example 1, λ must be a solution of

h(λ) = det(A − λI) = 0,

and v must be a non-zero solution of

(A − λI)v = 0 . (5.37)

The characteristic equation is

 
−3 − λ 4
h(λ) = det = (λ + 1)2 ,
−1 1−λ

after simplifying, and so we have equal eigenvalues

λ1 = λ2 = −1.

Equation (5.37) becomes     

−2 4 v1 0
= ,
−1 2 v2 0
giving −v1 + 2v2 = 0. Choosing v1 = 2 gives
 
2
v= , (5.38)
1

and so by (5.36),  
−t 2
x=e (5.39)
1
is a solution of the DE (5.35).
We are now faced with the problem of finding a second linearly independent solution. In
the past we have found that in such situations “multiplying by t” was a good thing to do.
So we consider
x = te−t v
as a new trial function (“plan A”). When substituted in the DE (5.35), this choice leads to a
contradiction. We presumably need “more constants” in the trial function. So we try “plan
B”, a trial function of the form
x = te−t v + e−t w. (5.40)

109
Differentiating and simplifying gives

x′ = e−t [−tv + (v − w)].

Substitute in the DE (5.35) and multiply by et to obtain

−tv + (v − w) = tAv + Aw,

which must hold for all t. Equating coefficients gives

Av = −v and Aw = v − w,

i.e.
(A + I)v = 0 and (A + I)w = v.
 
2
We can use (5.38) as a solution of the first equation, i.e. v = . Then the second equation
1
becomes     
−2 4 w1 2
= .
−1 2 w2 1
Solving this linear system gives w1 = −1, w2 = 0. Thus, the trial function (5.40) gives the
solution    
−t −1 2
x=e +t . (5.41)
0 1
of the DE (5.35).
The solutions (5.39) and (5.41) are linearly independent by inspection. Thus by equation
(5.15) the general solution of the DE (5.35) is
     
−t 2 −t −1 2
x = c1 e + c2 e +t ,
1 0 1

which can be rearranged as

    
−t 2 −1
x=e (c1 + tc2 ) + c2 . 
1 0

Exercises:
 
′ 3 −2
1) Solve x = x.
2 −2
   
1 −t 2t 2
Answer: x = c1 e + c2 e .
2 1
 
′ −1 −4
2) Solve x = x.
1 −1
    
−t −2 sin 2t 2 cos 2t
Answer: x=e c1 + c2
cos 2t sin 2t

110
  
′ 2 1
3) Solve x = x.
−1 4
    
3t 1 0
Answer: x=e (c1 + tc2 ) + c2 . 
1 1

5.3 Solving vector DEs using the Laplace transform

5.3.1 The method
Consider a homogeneous linear vector DE
x′ = Ax, (5.42)
with initial condition
x(0) = a. (5.43)
We wish to apply the Laplace transform operator L to this DE. In order to do this we first
have to define the Laplace transform of a vector-valued function x(t).
 
x1 (t)
Definition: The Laplace transform of a vector-valued function x(t) = is defined
x2 (t)
by  
L[x1 (t)]
L[x(t)] = ,
L[x2 (t)]
provided that the Laplace transform of each component function exists. 
Note: The Laplace transform of x(t) is itself a vector-valued function, which we denote by
X(s), i.e.
X(s) = L[x(t)].

We now need to relate L[x′ ] and L[Ax] to L[x]. These matters are taken care of in the
following propositions.
 ′ 
′ x1 (t)
Proposition 1: Given a vector-valued function with derivative x (t) = . If
x′2 (t)
L[x′1 (t)] and L[x′2 (t)] exist, then
L[x′ (t)] = sL[x(t)] − x(0).

Proof: By the definition,

L[x′1 (t)]
 
′
L[x (t)] = ,
L[x′2 (t)]
 
sL[x1 (t)] − x1 (0)
= , (Laplace transform of the derivative of a scalar function)
sL[x2 (t)] − x2 (0)
   
L[x1 (t)] x1 (0)
=s − , (using standard operations on vectors)
L[x2 (t)] x2 (0)
= sL[x(t)] − x(0),

111
again using the definition. 

Proposition 2: If A is a constant matrix, and L[x(t)] exists, then

L[Ax(t)] = AL[x(t)].

Proof: Consider
  
a11 a12 x1 (t)
L[Ax(t)] = L ,
a21 a22 x2 (t)
 
a11 x1 + a12 x2
=L , (a matrix acting on a vector)
a21 x1 + a22 x2
 
L[a11 x1 + a12 x2 ]
= , (by the definition)
L[a21 x1 + a22 x2 ]
 
a11 L[x1 ] + a12 L[x2 ]
= , (since L is a linear operator)
a21 L[x1 ] + a22 L[x2 ]
  
a11 a12 L[x1 ]
= (a matrix acting on a vector)
a21 a22 L[x2 ]

= AL[x(t)], by the definition. 

We can now apply L to the DE (5.42):

L[x′ ] = L[Ax].

By Propositions 1 and 2 this becomes

sL[x(t)] − x(0) = AL[x(t)].

Writing L[x(t)] = X(s), and using equation (5.43), we get

sX(s) − a = AX(s),

which can be rearranged to read

(sI − A)X(s) = a, (5.44)

 
1 0
where I = is the identity matrix. Equation (5.44) is a system of linear algebraic
0 1
equations with coefficient matrix (sI − A). Provided that sI − A is invertible, the solution
of (5.44) is
X(s) = (sI − A)−1 a. (5.45)
Having obtained X(s), the solution x(t) of the DE is obtained by taking the inverse Laplace
transform:
x(t) = L−1 [X(s)],

112
i.e. apply L−1 to the components of X(s). 

Comment: Finding the inverse of a 2 × 2 matrix to obtain the solution (5.45) is easy:
   
a b −1 1 d −b
if B = , then B = , (5.46)
c d det(B) −c a

where det(B) = ad − bc.

This can be verified by showing that BB −1 = I. 

We now illustrate the method with an example.

5.3.2 An example
Example: Solve the vector DE
 
′ −4 1
x = Ax, A= , (5.47)
−2 −1

with initial condition  

a1
x(0) = a = .
a2

Solution: Apply L to the DE (5.47) and use Propositions 1 and 2 to obtain

sX(s) − x(0) = AX(s)

where we have written L[x(t)] = X(s), as usual. Rearrange and use the initial condition to
get
(sI − A)X(s) = a,
i.e.  
s + 4 −1
X(s) = a. (5.48)
2 s+1
To find the inverse matrix we note
 
s + 4 −1
det = (s + 2)(s + 3),
2 s+1

after simplifying. Thus by equation (5.46),

 −1  
s + 4 −1 1 s+1 1
= ,
2 s+1 (s + 2)(s + 3) −2 s + 4

and so the solution of (5.48) is

  
1 s+1 1 a1
X(s) = , (5.49)
(s + 2)(s + 3) −2 s + 4 a2

113
for s 6= −2, −3. In component form
s+1 1
X1 (s) = a1 + a2 ,
(s + 2)(s + 3) (s + 2)(s + 3)
−2 s+4
X2 (s) = a1 + a2 .
(s + 2)(s + 3) (s + 2)(s + 3)
Performing the partial fraction expansions yields
   
2 1 1 1
X1 (s) = − a1 + − a2 ,
s+3 s+2 s+2 s+3
   
1 1 2 1
X2 (s) = 2 − a1 + − a2 .
s+3 s+2 s+2 s+3
 1 
We can calculate the inverse Laplace transforms using L−1 s−α = eαt . This gives

x1 (t) = L−1 [X1 (s)] = (2e−3t − e−2t )a1 + (e−2t − e−3t )a2 ,
(5.50)
x2 (t) = L−1 [X2 (s)] = 2(e−3t − e−2t )a1 + (2e−2t − e−3t )a2 .

These equations can be written in vector form as

 −3t
2e − e−2t e−2t − e−3t

x(t) = a, (5.51)
2(e−3t − e−2t ) 2e−2t − e−3t

giving the solution of the vector DE (5.47) which satisfies the initial condition x(0) = a.


Exercise: Use the Laplace transform method to solve the vector DE

 
′ 1 5
x = Ax, A = ,
−1 −3

with initial condition x(0) = a. This is example 2 in Section 5.2.3.

 
−t cos t + 2 sin t 5 sin t
Answer: x(t) = e a.
− sin t cos t − 2 sin t

s+3 (s + 1) + 2
Hint: = . 
s2 + 2s + 2 (s + 1)2 + 1

5.3.3 The Fundamental Matrix

The solution (5.51) has the following form

x(t) = Φ(t)a, (5.52)

where Φ(t) is the 2 × 2 matrix in (5.51). This form of the solution shows directly how the
state x(t) at time t depends on the initial state x(0) = a. The matrix Φ(t) is called the
fundamental matrix of the DE (5.47).

114
 A different form of the solution was obtained using the eigenvalue method for the same
example (see Example 1 in Section 5.2.2). The solution was obtained in the form (see
equation (5.26))    
−2t 1 −3t 1
x = c1 e + c2 e . (5.53)
2 1
This form of the solution is simpler algebraically and shows clearly that there are two distinct
rates of decay i.e. e−2t and e−3t .

How are the two forms of the solution related? By equation (5.51) the columns of the
fundamental matrix are
 −3t
2e − e−2t
 −2t
e − e−3t
 
1 2
Φ (t) = , Φ (t) = . (5.54)
2(e−3t − e−2t ) 2e−2t − e−3t
1
Each column vector is a solution
 of the DE. The column Φ (t) is the solution corresponding
1
to the initial condition a = , and the column Φ2 (t) is the solution corresponding to
0
 
0
the initial condition a = . [Choose a1 = 1, a2 = 0 and a1 = 0, a2 = 1 in (5.50) to
1
get these solutions.] If we have the solution in eigenvector form (5.53), we can construct
the fundamental matrix
  Φ(t) by finding
  the two special solutions (5.54): impose the initial
1 0
conditions x(0) = and x(0) = successively and determine the constants c1 and c2
0 1
in (5.53) (this was done in equations (5.28) and (5.29) in Section 5.2.2). 

In the Laplace transform method, the Laplace transform of the solution x(t), denoted
X(s), is given by
X(s) = (sI − A)−1 a (5.55)
The solution x(t) = L−1 [X(s)] is obtained by applying L−1 to (5.55). In the example in this
section we wrote (5.55) in component form and applied L−1 to each component. One can
apply L−1 directly to (5.55) obtaining

x(t) = L−1 [(sI − A)−1 ]a,

where the Laplace transform of the matrix is obtained by applying L−1 to each entry. In the
example,
2 1 1 1 !
s+3
− s+2 s+2
− s+3
X(s) = a.
2 2 2 1
s+3
− s+2 s+2
− s+3
Thus
L−1 [ ] L−1 [ ]
 
−1
x(t) = L X(s) = a
L−1 [ ] L−1 [ ]

2e−3t − e−2t e−2t − e−3t

 
= a,
2e−3t − 2e−2t 2e−2t − e−3t

in agreement with equation (5.51). 

115
 We now summarize the Laplace transform method schematically, writing L[x(t)] = X(s),
as usual.

x′ = Ax apply L apply L−1
−−−−−−−−−−→ X(s) = (sI − A)−1 a −−−−−−−−−→ x(t) = Φ(t)a
x(0) = a and rearrange

Thus, the fundamental matrix Φ(t) is related to the coefficient matrix A by

L[Φ(t)] = (sI − A)−1 ,

or equivalently
L−1 [(sI − A)−1 ] = Φ(t). 

5.4 Orbits of a vector DE in state space

In this Section we show how to sketch the solutions of the homogeneous,linear vector DE

x′ = Ax

as a family of curves in the state space R2 . These curves are called the orbits of the DE.
The goal is to use this picture of the solutions (sometimes called the “phase portrait”) to
understand the behaviour of the underlying physical system (e.g. oscillator or coupled mixing
tanks):

x2
a = x(0)
initial state
x(t1 )
x(t2 ) state at time t1

state at time t2

x1

Figure 5.5: An orbit of the DE x′ = Ax.

Oscillator: Coupled mixing tanks:

 
y
 
x= m1
y′ x=
m2

The evolution of a physical system described by the DE x′ = Ax is thus represented by

a moving point x(t) in the state space R2 , i.e. as time passes, the point x(t) moves along the
orbit, as shown in Figure 5.5. From the shape of the orbit one can draw various conclusions

116
e.g. are the individual components x1 and x2 increasing or decreasing, do they attain a
maximum or minimum etc.?
We have seen that the second order oscillator DE with zero driving force, i.e.

y ′′ + 2ζy ′ + y = 0

can be written as a first order vector DE

 
′ y
x = Ax, x= ∈ R2 , (5.56)
y′
where  
0 1
A= , (5.57)
−1 −2ζ
and ζ is the dimensionless damping constant (see Section 5.1.2). We are going
to sketch the
orbits for two values of ζ, one for which the system is overdamped ζ = 45 > 1 and one for
which the system is underdamped ζ = 53 < 1 .

5.4.1 Phase portraits: unequal real eigenvalues

Example 1: Give a qualitative sketch of the orbits of the vector DE
 
′ 0 1
x = Ax, A = (5.58)
−1 − 52

in the state space R2 .

Solution: It is best to use the form of the solution obtained using the eigenvector method.
One obtains (exercise):    
− 21 t 2 −2t 1
x = c1 e + c2 e , (5.59)
−1 −2
where c1 and c2 are arbitrary constants. This equation determines a solution for any initial
state x(0) = a, and thus determines an orbit through each point of R2 . To sketch the orbits
we rely on three properties of the family of solutions.

1) Exceptional solutions:

There are three exceptional solutions in the family (5.59). First, the equilibrium solution

x = 0,

given by c1 = 0 = c2 . Second, the solution given by c2 = 0, c1 6= 0, i.e.

 
− 12 t 2
x = c1 e . (5.60)
−1
Third, the solution given by c1 = 0, c2 6= 0, i.e.
 
−2t 1
x = c2 e . (5.61)
−2

117
 x2

c2 = 0, c1 < 0

x1

c2 = 0, c1 > 0

Figure 5.6: Three exceptional solutions.

The solutions (5.60) and (5.61) are represented by straight lines in the state space: elimi-
nating t gives x2 = − 12 x1 for (5.60) and x2 = −2x1 for (5.61).

2) Asymptotic behaviour as t → +∞:

It follows immediately from (5.59) that

lim x(t) = 0 ,
t→+∞

for all values of the constants c1 and c2 . This result means that every orbit approaches the
1
equilibrium orbit x = 0 as t → +∞. Moreover, since e−2t is small compared to e− 2 t as
t → +∞, we can write  
− 21 t 2
x ≈ c1 e as t → +∞,
−1
i.e. any solution with c1 6= 0 is approximated by the exceptional solution (5.60) as t → ∞.
This result means that all orbits with c1 6= 0 approach the origin along the line x2 = − 21 x1
i.e. this line “attracts” other orbits as they approach the origin. This property of the orbits
is shown in Figure 5.8.

3) Sign of the slope of the orbits:

In component form the DE (5.58) reads

x′1 = x2
x′2 = −x1 − 52 x2 .
 ′
′ x1
Keeping in mind that the vector x = is tangent to the orbit x = x(t), and that their
x′2
x ′
dx2
slope is dx 1
= x′2 , we conclude that
1

(i) the tangent line to an orbit is vertical whenever an orbit crosses the line x2 = 0 (i.e.
x′1 = 0), called the vertical isocline,

118
 (ii) the tangent line to an orbit is horizontal whenever an orbit crosses the line x2 = − 52 x1
(i.e. x′2 = 0), called the horizontal isocline, and
(iii) the slope of an orbit is positive when it lies in the region defined by
− 25 x1 < x2 < 0,
or
0 < x2 < − 25 x1 ,
i.e. the shaded region in Figure 5.7.

x2
x2 = − 52 x1

x1

x2 = 0

Figure 5.7: Region in which the slope of an orbit is positive.

The information obtained under 1)–3) leads to Figure 5.8.

Comment: The phase portrait for any DE x′ = Ax with unequal negative real eigenvalues
will have the general form shown in Figure 5.8. The specific shape will depend on the
location of the attracting orbit and of the horizontal and vertical isoclines. If the matrix A is
diagonal, the horizontal and vertical isoclines coincide with the exceptional solutions giving
the simpl picture shown in Figure 5.9.

Physical interpretation of the phase portrait:

 
y
In example 1, the state vector x is given by x = , where y is the displacement of
y′
the trolley and y ′ is its velocity. Thus, when an orbit cuts the horizontal axis (x2 = 0) the
velocity is zero and when an orbit cuts the vertical axis (x1 = 0) the displacement is zero.
In Figure 5.8, the dark orbit represents the situation in which the trolley is given a positive
displacement (x1 > 0) and a negative velocity (x2 < 0) initially, of sufficient magnitude that
it overshoots the equilibrium before approaching a state of rest. One can see from Figure
5.8 that in order for “overshoot” to occur the initial state must lie in a restricted region in
state space, namely the region between the vertical axis and the line x2 = −2x1 , i.e.
y>0 and y ′ < −2y, or

y<0 and y ′ > −2y.

119
 x2

vertical
isocline

x1

x2 = − 25 x1
horizontal
isocline

x2 = − 21 x1
attracting
orbit

initial state x2 = −2x1

5


Figure 5.8: Orbits of the oscillator DE (5.58): the overdamped case ζ = 4
>1 . 

5.4.2 Phase portraits: complex eigenvalues

We begin by considering an example with complex eigenvalues that is simple algebraically.

Example 2: Give a qualitative analysis of the orbits of the DE

 
′ −1 1
x = Ax, A = . (5.62)
−1 −1

Solution: We use the eigenvalue method to get the general solution

    
−t cos t sin t
x=e c1 + c2
− sin t cos t
 
c1 cos t + c2 sin t
= e−t ,
−c1 sin t + c2 cos t

(exercise). We can write each component as a single cosine or sine respectively:

 
−t cos(t − δ)
x = be , (5.63)
− sin(t − δ)
where
c1 c2
q
b= c21 + c22 , cos δ = , sin δ = .
b b
120
 x2

x1

 
−1 0
′
Figure 5.9: Orbits of the DE x = x.
0 −2

In this case, the only exceptional solution is the equilibrium solution x = 0 (i.e. b = 0),
and all other solutions are asymptotic to it as t → +∞:
lim x(t) = 0 .
t→+∞

Because of the cos and sin in (5.63), the

 orbitsspiral around in a clockwise manner as they
cos t
approach the origin. [Recall that x = describes a circle traversed clockwise.] The
− sin t
shape of the spirals is determined by the horizontal and vertical isoclines, which are obtained
by writing the DE (5.62) in component form
x′1 = −x1 + x2
x′2 = −x1 − x2 .
The horizontal isocline (x′2 = 0) is x2 = −x1 and the vertical isocline (x′1 = 0) is x2 = x1 .
The orbits are shown in Figure 5.10.

Comment: The phase portrait for any DE x′ = Ax having complex eigenvalues with
negative real parts, is qualitatively the same as Figure 5.10 – a family of spirals focusing on
the origin. The specific shape will depend on the horizontal and vertical isoclines, which in
general will not be orthogonal, unlike the above special case. 

We now consider the example that is a special case of the oscillator DE (5.56)-(5.57).

Example 3: Give a qualitative sketch of orbits of the DE

 
′ 0 1
x = Ax, A = .
−1 − 65

Solution: We use the eigenvalue method to find the general solution:

    
− 35 t 4 4

1 4 4

0
x=e c1 cos 5 t + c2 sin 5 t + −c1 sin 5 t + c2 cos 5 t 4 ,
− 53 5

121
 x2 vertical isocline
x2 = x1

x1

horizontal isocline
x2 = −x1

Figure 5.10: Orbits of a first order linear vector DE with complex eigenvalues: the horizontal
and vertical isoclines are orthogonal.

(exercise).
Since the real part of the eigenvalues is negative Re(λ) = − 35 , the solution contains

a decaying exponential, and so the orbits spiral into the origin. The shape of the spiral is
determined by the isoclines. In component form the DE is
x′1 = x2
x′2 = −x1 − 56 x2 .
Thus the horizontal isocline (x′2 = 0) is x2 = − 56 x1 , and the vertical isocline (x′1 = 0) is
x2 = 0. The orbits are shown in Figure 5.11.

5.4.3 Long term evolution and stability

The examples of homogeneous vector DEs x′ = Ax in this section have the property that all
solutions satisfy
lim x(t) = 0 . (5.64)
t→∞
Thus, for all initial states the system approaches the equilibrium state x = 0 in the long
term. In this case we say that the equilibrium state is asymptotically stable.
One can predict whether or not the system is asymptotically stable without finding the
solutions explicitly, simply by considering the eigenvalues of the coefficient matrix A. The
solutions are linear combinations of the trial functions
x = eλt v. (5.65)
We have seen that λ has to be an eigenvalue of the matrix A. If all eigenvalues of A satisfy
Re(λ) < 0,

122
 x2

vertical isocline
x2 = 0

x1

horizontal isocline
x2 = − 65 x1

Figure 5.11: Orbits of a first order linear vector DE with complex eigenvalues: the horizontal
and vertical isoclines are not orthogonal.

then every solution (5.65) will contain a decaying exponential and hence all solutions will
satisfy (5.64). We summarize this result in the following Proposition.

Proposition: If all eigenvalues of the matrix A satisfy

Re(λ) < 0,

then all solutions of the DE x′ = Ax satisfy

lim x(t) = 0 ,
t→∞

i.e. the equilibrium state x = 0 is asymptotically stable. 

5.5 Solving inhomogeneous linear vector DEs

In this Section we show how to solve the inhomogeneous vector DE

x′ = Ax + f(t). (5.66)

This DE should be thought of as describing the state x(t) of a linear physical system, with
input function f(t) (see Section 5.1.3).

As with any linear DE, the general solution will be of the form

x(t) = xh (t) + xp (t),

where xh (t) is the general solution of the homogeneous DE x′ = Ax, and xp (t) is a particular
solution of the inhomogeneous DE (5.66)

Three methods are available to find xp (t):

123
 (i) the method of undetermined coefficients (a generalization of the method used for second
order scalar DEs),

(ii) the Laplace transform method (an extension of the method used in Section 5.3), and

(iii) the method of “variation of parameters”, which makes use of the fundamental matrix
Φ(t).
Methods (i) and (ii) are limited as regards the possible forms of the input function f(t),
and also entail extensive algebraic manipulation.
Method (iii) is generally applicable, and has a very simple formulation in terms of the
fundamental matrix Φ(t) (although the algebra can be tedious in some examples). Because
of these advantages it is the method we use.
Before we can develop the method we need to discuss some properties of Φ(t).

5.5.1 Properties of the Fundamental Matrix

In Section 5.3.3 we showed that the general solution of the homogeneous DE x′ = Ax with
initial condition x(0) = a has the form

x(t) = Φ(t)a, (5.67)

where Φ(t) is a 2 × 2 time-dependent matrix called the fundamental matrix of the DE. It
follows by setting t = 0 in (5.67) that Φ(0)a = a. Since a is arbitrary, this equation implies
that
Φ(0) = I, (5.68)
where I is the 2 × 2 identity matrix.
We need two additional properties of Φ(t), the first of which follows directly from the
fact that (5.67) is a solution of the homogeneous DE

x′ = Ax. (5.69)

Proposition 1: The fundamental matrix Φ(t) of the DE x′ = Ax satisfies

Φ′ (t) = AΦ(t). (5.70)

Proof: Since (5.67) is a solution of (5.69) we have

[Φ(t)a]′ = A[Φ(t)a],

which yields
[Φ′ (t) − AΦ(t)]a = 0 ,
after rearranging. Since this holds for all a ∈ R2 , it follows that

Φ′ (t) − AΦ(t) = 0,

where “0” denotes the zero matrix, which gives the required result. 

124
 We shall refer to (5.70) as the derivative property of Φ(t).

The second property of Φ(t) follows from the geometric interpretation of equation (5.67).
One can think of Φ(t) as an operator that transforms an initial state x(0) = a into the state
x(t) = Φ(t)a at time t.

Consider the situation shown in Figure 5.12. At time t = 0 the system is in state a, and
after a time t1 it evolves into state b, given by

x2 x2

c = Φ(t2 )b
Φ(t)a
b = Φ(t1 )a

a a
x1 x1

Figure 5.12: The action of the fundamental matrix.

b = Φ(t1 )a. (5.71)

After a further time t2 , it evolves into state c, so that

c = Φ(t2 )b, (5.72)

thinking of b as the initial state. On the other hand, in a time t2 + t1 the system will evolve
from a to c so that
c = Φ(t2 + t1 )a. (5.73)
Substituting (5.71) in (5.72) and comparing with (5.73) gives

Φ(t2 + t1 )a = Φ(t2 )Φ(t1 )a.

Since a is an arbitrary vector, it follows that

Φ(t2 + t1 ) = Φ(t2 )Φ(t1 ),

a matrix equation.

We summarize this result and the result (5.68) in the following Proposition.

Proposition 2: The fundamental matrix Φ(t) of the DE x′ = Ax satisfies

Φ(0) = I,

125
and
Φ(t2 + t1 ) = Φ(t2 )Φ(t1 ), (5.74)
for all t1 , t2 ∈ R. 

The property we need is an immediate consequence of this Proposition. Choosing t2 =

−t1 and writing t1 = t, equation (5.74) with (5.68) gives

Φ(−t)Φ(t) = I.

In other words, Φ(t) is invertible for any t, and

[Φ(t)]−1 = Φ(−t). (5.75)

We shall refer to (5.75) as the inverse property of the fundamental matrix. This result is
quite remarkable. It states that to find the inverse matrix of a fundamental matrix Φ(t), one
simply replaces t by −t.

5.5.2 The method of variation of parameters

Given the DE
x′ = Ax + f(t), (5.76)
let Φ(t) be the fundamental matrix of the related homogeneous DE x′ = Ax. We know that
x(t) = Φ(t)a is a solution of x′ = Ax for any a ∈ R2 . This suggests that in order to find a
particular solution of (5.76) we consider a trial function of the form

x(t) = Φ(t)v(t), (5.77)

where v(t) is an arbitrary vector-valued function. In other words, we replace the constant
vector a (whose components are two arbitrary parameters) by a time-dependent vector v(t)
(whose components are two arbitrary scalar functions), i.e. we “vary the parameters”. This
choice of trial function is thus called the method of variation of parameters.

Differentiate (5.77) with respect to t using the Product Rule and the derivative property
(5.70) of Φ(t) to obtain

x′ (t) = Φ′ (t)v(t) + Φ(t)v′ (t)

= AΦ(t)v(t) + Φ(t)v′ (t)
= Ax(t) + Φ(t)v′ (t),

by (5.77). Equating this expression for x′ with x′ in (5.76) gives

Φ(t)v′ (t) = f(t).

We now multiply by the matrix Φ(−t) and use the inverse property (5.75) to get

v′ (t) = Φ(−t)f(t). (5.78)

126
Equations (5.77) and (5.78) constitute the method of variation of parameters. Given f(t),
and having calculated Φ(t), one obtains v(t) by taking the antiderivative of (5.78). Then
(5.77) gives a particular solution of the DE (5.76). 

Comment: When taking the antiderivative of (5.78), a constant of integration, which is a

constant vector, arises. If one simply wants a particular solution one can choose this vector
to be zero. Alternately, one can choose this vector so that

v(0) = 0 .

Then the particular solution

xp (t) = Φ(t)v(t)
satisfies
xp (0) = 0 . (5.79)
With this choice of particular solution, the general solution x(t) = xh (t) + xp (t) can be
written
x(t) = Φ(t)a + xp (t). (5.80)
Equation (5.79) ensures that

x(0) = Φ(0)a + xp (0)

= Ia + 0 = a,

as required. 

5.5.3 An example
Example: Solve the DE
x′ = Ax + f(t), x(0) = a, (5.81)
with    
1 1 cos t
A= , f(t) = .
−1 1 − sin t
The fundamental matrix of x′ = Ax is
 
t cos t sin t
Φ(t) = e . (5.82)
− sin t cos t

Solution: The solution is of the form (5.80), i.e.

x(t) = Φ(t)a + xp (t). (5.83)

Substituting a trial function

xp (t) = Φ(t)v(t) (5.84)

127
in the DE leads to (as in Section 5.5.2)

v′ (t) = Φ(−t)f(t)
  
−t cos t − sin t cos t
=e
sin t cos t − sin t
 
1
= e−t
0

Taking the antiderivative yields

 
−t 1
v(t) = −e + c. (5.85)
0
   
1 1
Setting t = 0 gives v(0) = − + c. We want v(0) = 0, and so we choose c = . Thus
0 0
(5.85) becomes  
−t 1
v(t) = (1 − e ) . (5.86)
0
By (5.82), (5.84) and (5.86) the particular solution is
    
t cos t sin t 1 t cos t
xp (t) = (e − 1) = (e − 1) ,
− sin t cos t 0 − sin t

and by (5.83) the full solution is

   
t cos t sin t t cos t
x(t) = e a + (e − 1) . 
− sin t cos t − sin t

Exercise: Solve the DE

x′ = Ax + f(t), x(0) = a,
with    
−3 2 −3t 1
A= f =e .
2 −3 1
The fundamental matrix is
 −t
e + e−5t e−t − e−5t

1
Φ(t) = .
e−t − e5t e−t + e−5t
2

 
1 −t −3t 1
Answer: x(t) = Φ(t)a + 2
(e −e ) . 
1

128
Epilogue

Differential equations is a large subject, and in AMath250 we have essentially examined the
“tip of a large iceberg”.
The subject of DEs splits into two branches:
I: Linear DEs
II: Non-linear DEs
“Solving a DE” means one of three things:

i) finding an exact (i.e. explicit) solution of the DE

ii) finding an approximate numerical solution of the DE using a computer, e.g. MATLAB,

iii) finding an approximate analytic solution, e.g. an expansion in terms of a small param-
eter, referred to as perturbation methods.

In all three approaches it is desirable to have a graphical representation of the solution.

In AMath250, we have almost exclusively worked with linear DEs, in particular linear
DEs with constant coefficients, and have shown how to solve them explicitly. For other DEs,
there are no general solution algorithms, but one can find solutions in special cases by making
use of

- tables of solutions and solution methods

e.g. D. Zwillinger, Handbook of DEs, 2nd ed., Section II A, pp. 185-363.

- computer software, e.g. MAPLE.

As regards the other methods, numerical solutions are considered in AMath342 and
perturbation methods are introduced in AMath351.
A completely different approach, which complements solving a DE, is to study the prop-
erties of general classes of solutions, e.g. their long term behaviour. This approach, which
dates back to Henri Poincaré at the turn of the 20th century, is called the qualitative analysis
of DEs. We had a glimpse of this topic when we sketched orbits in state space in Chapter
5. To progress further one needs to study various theoretical issues, which begins in AM-
ath351. The emphasis in qualitative analysis is on non-linear DEs, and this subject is the
main topic in AMath451.

129
130
References

Borelli, R.L. and Coleman, C.S., 1987, Differential Equations: A Modeling Approach, Prentice-
Hall.

Goldberg, J. and Potter, M.C., 1998, Differential Equations: A Systems Approach, Prentice-
Hall.

Simmons, G.F., 1972, Differential Equations – with applications and historical notes, McGraw-
Hill.

Reiss, E.L., Callegari, A.J., Ahluwalia, D.S., 1976, Ordinary differential equations with ap-
plications, Holt, Rinehart & Winston.

Brauer, F. and Nohel, J.A., 1967, Ordinary Differential Equations, W.A. Benjamin.

Braun, M., 1983, Differential Equations and their Applications, Springer-Verlag.

Boyce, W.E. and diPrima, R.C., 1997, Elementary Differential Equations and Boundary
Value Problems, 6th edition, J. Wiley & Sons.

131
132
Problem Sets

133
134 Review Problem Set

Review Problem Set

Note: Questions 1 and 2 are intended to give you practice with antiderivatives and curve-
sketching, two aspects of Calculus that will arise frequently. Questions 3 and 4 are
intended to give you practice with logs, exponentials and complex number. Questions
5-7 will also be relevant; they require more thought.

1. Express the following antiderivatives in elementary terms. Check your answers by

differentiation.

1
Z Z Z
−t −t2
(i) te dt (ii) te dt (iii) dt
1+t
t 1−t 1
Z Z Z
(iv) dt (v) dt (vi) dt
1−t t t(1 − t)
1 1
Z Z Z
(vii) dt (viii) dt (ix) t sin(t2 ) dt
4 + t2 4 − t2
Z Z Z
2 2
(x) sin t dt (xi) t sin t dt (xii) t3 et dt

Hint for (xii): Make a change of variable u = t2 .

2. Give a qualitative sketch of the graphs of the following functions. The goal is to give
a sketch of the graph, not drawn accurately to scale, but which shows the essential
properties of the function. Think about symmetry, asymptotes and the behaviour as
x → ±∞, where appropriate.
1 1 1
(i) f (x) = (ii) f (x) = (iii) f (x) = x +
1 + x2 1 − x2 x
1 2
(iv) f (x) = x − (v) f (x) = e−x (vi) f (x) = e−|x|
x

(vii) f (x) = x3 + x (viii) f (x) = x3 − x (ix) f (x) = sin2 x

ex
(x) f (x) = e−x + e2x (xi) f (x) = e−x − e2x (xii) f (x) =
2 + ex

3. (i) Simplify the expression

ea ln b−b ln a ,
where a and b are positive constants i.e. rewrite the expression so as to eliminate
the logs and exponentials.
(ii) Find all solutions of the equation ex − e−x = 2.
Hint: Let u = ex .
Review Problem Set 135

(iii) Suppose that

T (t) = T0 e−kt
and
T (t1 ) = T1 , T (t2 ) = T2 ,
where k, t1 and t2 are positive. Verify that
ln(T2 /T0 )
t2 = t1 .
ln(T1 /T0 )

4. Use Euler’s formula

eiθ = cos θ + i sin θ
to derive the cosine and sine addition identities:
cos(A + B) = cos A cos B − sin A sin B
sin(A + B) = sin A cos B + cos A sin B.

Hint: Let θ = A + B and use ei(A+B) = eiA eiB .

2n
5. What does the graphZ πof f (x) = sin x look like for large n? Make a conjecture
about the value of sin2n x dx for large n.
0

6. We shall find that functions such as

f (t) = Ae−kt sin(ωt + β) ,

where A, k > 0, ω > 0 and β are constants, describe damped oscillations.

a) (i) Sketch the graph of f (t) for A = 1, k = 1, ω = 1 and β = 0.

(ii) Describe how the graph changes as ω increases, and as k increases.
b) Show that f (t) satisfies

f ′′ + 2kf ′ + (ω 2 + k 2 )f = 0 .

7. Exponential growth of a population i.e. N̂(t) = N0 ert , r > 0, can only occur if
the resources are essentially unlimited. If there are limited resources one encounters
functions such as
N0 ert
N(t) = ,
1 − NM0 + NM0 ert
where r, N0 , and M are positive constants.

(i) Evaluate lim N(t).

t→+∞
136 Review Problem Set

(ii) Sketch the graph of N(t) for N0 = 21 M, r = 1. How does the shape of the graph
change as r increases?
(iii) You will notice that N(t) in (ii) is close to N̂ (t) = N0 et over a subinterval of
the t-axis. Find the restriction on t that will ensure that

N̂ (t) − N(t) 1
< .
N̂(t) 100
 
N̂ (t)−N (t)
(iv) In the general case show that if 0 < t < 1r ln ǫM
N0
+ 1 , then 0 < N̂ (t)
<ǫ.

Note: In the section of the course dealing with Laplace transforms we shall work with
improper integrals. We shall use the following results.

8. Show that if s > 0 and a is constant, then

Z ∞ Z ∞
−st a s
e sin at dt = 2 2
, e−st cos at dt = 2 .
0 s +a 0 s + a2

Hint: Evaluate the complex antiderivative

Z
e(−s+ia)t dt,

and use Euler’s formula.

9. (i) Derive the reduction formula

1 −st n n
Z Z
n −st
t e dt = − e t + tn−1 e−st dt , for s > 0 .
s s

(ii) Let In (s) denote the improper integral

Z ∞
In (s) = tn e−st dt , s > 0.
0

Show that
n
In (s) = In−1 (s) , n = 1, 2, · · ·
s
Hence show that
n!
In (s) = .
sn+1
Problem Set 1 137

Problem Set 1
First Order Differential Equations
1. Each equation below defines a one-parameter family of curves. The parameter k as-
sumes all real values.

(i) y = ex + k (ii) y = 2 + ke−2x (iii) y = x + ke−x

k
(iv) y =x+ x
(v) y = ln(ex + k) (vi) k(x2 + y 2 ) = 2y

a) Derive the DE that is satisfied by the family of curves. State whether the DE is
separable, linear or neither.
b) Use information deduced from the DE and from the given equation to give a
qualitative sketch of each family.

2. Consider the following first order DEs:

dy dy dy
(i) dx
= −2y + e−x (ii) dx
= y sin x (iii) dx
= x(1 − y)
dy dy y dy
(iv) dx
= y(1 − y) (v) dx
= 1+x2
(vi) x2 dx + 3xy = 1
dy dy dy
(vii) dx
= −y − x (viii) dx
= −2xy + 2x3 ix) dx
= 1 − y2

a) Find the general solution of each DE.

Hint: For (viii), see the Review Set, #1 (xii).

b) By using the form of the DE and the solution, give a qualitative sketch of the
family of solutions of each DE, showing some typical solutions and all exceptional
solutions. In your diagram, indicate as a dotted curve the set of all points at
dy
which the slope dx is zero.

3. Repeat question 2 for the DE

dy
x + (x + 1)y = e−x .
dx
Note: The sketch of the family of solution curves has an intricate structure, and it will
take some effort to discover it.

4. Solve each initial value problem. Specify the interval in which your solution is valid.
Sketch your solution.

dy 1 dy
(i) dx
= y 2 cos x, y(0) = 2
(ii) dx
= y 2 cos x, y(0) = 2
dy 2y dy
(iii) dx
= x
+ x, y(1) = e (iv) dx
= ey−x , y(0) = ln 2
138 Problem Set 1
dy 2
5. Suppose that y = y(x) is a solution of the DE dx
+ 2xy = 2e−x .
If y(0) = 2, find y(1). Evaluate lim y(x), if the limit exists. Does y(x) attain a
x→+∞
maximum value for x ≥ 0? Sketch the graph of y(x).

6. Use the method of undetermined coefficients to find the general solution for each DE
where applicable. Give the reason if not applicable, but do not solve.

dy dy
(i) + 2y = 3 − 2x (ii) − 2y = 2 + e−x
dx dx
dy dy
(iii) + y = sin 2x (iv) + 2xy = x2
dx dx
dy dy
(v) + y = e−x (vi) + 3y = y 2
dx dx
dy dy
(vii) + 2y = xex (viii) + y = tan x
dx dx
7. Solve the DE
dy
+ ky = A sin ωt,
dt
with initial condition y(0) = y0 , where k, A and ω are positive constants, with [ω] = T −1
[k] = T −1 , [A] = [y]T −1 .

a) Are all/any of the solutions periodic in t?

b) Evaluate lim y(t), if the limit exists.
t→+∞

c) Write the solution y(t) as the sum of a transient term and a steady state term.

8. a) When a coil of steel is removed from an annealing furnace its temperature is 684◦
C. Four minutes later its temperature is 246◦ C. How long will it take to reach
100◦ C? Assume that Newton’s law of cooling holds, which states that the time
rate of change of temperature of a cooling body is proportional to the difference
between the temperature of the body and the temperature of the surrounding
medium. Assume that room temperature is 27◦ .
b) You will find it quite tedious to solve part a) because of all the numbers. The
problem can be solved efficiently by formulating it more generally, as follows.
Let TA be the temperature of the surrounding medium (called the ambient tem-
perature). Let T0 be the temperature of the coil when it is removed from the
furnace at time t = 0. The temperature is measured to be T1 at time t1 . The
problem is to find the time t2 at which the temperature is T2 . So the given quan-
tities are TA , T0 , T1 , T2 and t1 and the unknown is t2 . The idea is to solve the DE
for the temperature function T (t) and show that
  
T2 −TA
ln T0 −TA
t2 = t1    .
T1 −TA
ln T0 −TA
Problem Set 1 139

9. The velocity v(t) of a sky-diver falling towards the earth’s surface satisfies the DE
dv
m = mg − αv,
dt
where m is the mass, g the acceleration due to gravity (assumed constant), and α is
the drag coefficient (see the notes).
a) Find v(t) assuming an initial velocity v(0) = v0 > 0.
b) Show that as time passes, v(t) approaches the terminal velocity vterm = mg/α.
Does v(t) ever equal vterm ?
c) Find the distance y fallen as a function of time t and verify that your formula is
dimensionally consistent.
(d) Find the distance y fallen as a function of velocity v, and verify that your formula
is dimensionally consistent.
10. The velocity v of a projectile fired vertically up from the surface of a planet and
travelling only under the influence of gravity, satisfies the DE.
dv gR2
v =− 2 ,
dr r
where r is the distance of the projectile from the centre of the planet, R is the radius
of the planet and g is the acceleration due to gravity on the surface of the planet (see
the notes).

a) Find v as a function of r, assuming the initial condition v(R) = vinit .

b) Find the escape velocity for the planet. Verify that your formula is dimensionally
consistent.

11. A student carrying a flu virus returns to an isolated college campus of 1,000 students.
The rate at which the virus spreads is proportional to the product of the number
of infected students and the number not infected. Predict the number of infected
students after 7 days, if it is found that after 3 days 40 students are infected. When is
the infection spreading most rapidly? Illustrate your solution by graphing the number
of infected students as a function of time t.
Suggestion: Formulate and solve this problem more generally, as in #8b).
12. Suppose that a corpse with temperature 27◦ C is discovered at midnight, and that the
ambient temperature is a constant 17◦ C. The body is moved quickly to a morgue where
the ambient temperature is 7◦ . After one hour the body temperature is 20◦ C. Estimate
the time of death.
Suggestion: Formulate and solve this problem more generally, as in #8b).
13. A tank is used in certain hydrodynamic experiments. After one experiment, the tank
contains 200 litres of a dye solution with a concentration of 1 g/litre. To prepare for
the next experiment, the tank is to be rinsed with clear water flowing in at a rate of 2
litres/min, the well-stirred mixture flowing out at the same rate. How long will it take
to reduce the concentration of dye to 1% of its original value?
140 Problem Set 1

14. The DE  
dN N
=r 1− N − h,
dt K
where r, K and h are positive constants, describes a population of fish with natural
growth rate coefficient r, carrying capacity K and a constant harvest rate h.

a) Show that if h < 41 rK there are two equilibrium solutions N(t) = N1 and N(t) =
N2 where N1 and N2 are constants with 0 < N1 < N2 . Find N1 and N2 .
b) Give a qualitative sketch of the solution curves in the case h < 14 rK. Discuss the
longterm behaviour of N(t) in the two cases (i) N(0) > N1 and (ii) N(0) < N1 .

15. A projectile of mass m is fired straight up from the earth’s surface with initial velocity
v0 . If v0 is small compared to the escape velocity it is reasonable to assume that the
acceleration due to gravity g is constant throughout the motion. After rising to a
certain height, the projectile will return to the earth’s surface. The return time tr is
the total time spent in flight. Calculate tr in two cases:

a) neglecting air resistance,

b) assuming that the force due to air drag is proportional to the velocity.

Which return time is shorter?

16. An object of mass m is released from rest at a height of h metres above the earth’s
surface, and strikes the ground after falling for th seconds. Assume that the force
due to air drag is proportional to the velocity. It is possible to determine the drag
coefficient and knowing m, h and th ?

17. Experiments show that the rate of decrease of atmospheric pressure p with height h is
proportional to the product of the acceleration due to gravity (assumed constant) and
the pressure.

a) Express the above result as a differential equation for the unknown function p(h).
b) Let p0 denote the pressure at sea-level, p1 denote the pressure at a reference
altitude of h1 . Show that the dependence of height h on pressure p can be written
in the form
ln[p/p0 ]
h = h1 .
ln[p1 /p0 ]
c) Suppose that the pressure at sea level is 104 kPa and at an altitude of 3000 m is
70 kPa. Most people will lose consciousness if the pressure falls below 50 kPa. At
what altitude does this occur? (kPa means kilopascals, and 1 Pa = 1 N/m2 ).

18. Consider a population of size N(t) which grows exponentially at a rate r ([r] = T −1 ),
but is harvested at a constant rate H per day. Then N(t) satisfies the DE

dN
= rN − H.
dt
Problem Set 1 141

a) Suppose that r = 0.01 days−1 and N(0) = 4000. Show that if H > 40 per
day, then the population becomes extinct in a finite time, but if H < 40, the
population will increase without bound.
b) Referring to a), if H > 40 per day, find the time of extinction.
c) In a), H = 40 represents the critical harvest rate. Find the critical harvest rate
for arbitrary r > 0 and arbitrary N(0).

19. A mixing tank of capacity Vmax litres initially contains V0 litres of pure water. A salt
solution of constant concentration cin gm/litres flows in at a rate 2f litres/min and the
contents of the tank flow out at f litres/min. For this physical situation it is reasonable
to define a characteristic time by tc = Vf0 (the time to drain the initial volume, with
zero inflow) and a characteristic mass by mc = Vmax cin (the mass of salt in the tank
if it were completely filled by the inflow). Then one can define a dimensionless time τ
and dimensionless mass M by
t m
τ= , M= ,
tc mc
where m is the mass of salt in the tank at time t.

a) Show that M satisfies the DE

dM M
=− + 2b,
dτ 1+τ
V0
where b = Vmax
.
b) Show that when the tank is filled it contains (1 − b2 )Vmax cin grams of salt.

Answers:
2.
(i) y = e−x + Ce−2x (ii) y = Ce− cos x
1 2 1
(iii) y = 1 − Ce− 2 x (iv) y= and y = 0
1 + Ce−x
1 C
(v) y = Cearctan x (vi) y= + 3
2x x
2
(vii) y = −x + 1 + Ce−x (viii) y = x2 − 1 + Ce−x
1 − Ce−2x
(ix) y= and y = −1.
1 + Ce−2x
142 Problem Set 2

Problem Set 2
Dimensional Analysis
1. Calculate the dimensions of
(i) pressure gradient (iv) velocity gradient
(ii) density gradient (v) shear stress
(iii) surface tension (vi) viscosity

Note: “Surface tension” is the work done in creating unit area of a fluid surface. A
velocity gradient in a fluid (called the ‘shear rate” by engineers) has associated with
it a shear stress (a force per unit area). The shear stress is proportional to the shear
rate, and the constant of proportionality is called the “viscosity” of the fluid.

2. The Duffing oscillator is described by the non-linear second order DE

d2 x
+ αx + βx3 = F0 sin ωt.
dt2
where [x] = L and [t] = T . Calculate the dimensions of α, β, F0 and ω.

3. The diffusion of heat in a slab of material is governed by the partial differential equation
(the heat equation)
∂T ∂2T
=κ 2,
∂t ∂x
where the temperature T varies only in the x-direction. The constant κ is called the
thermal diffusivity, and depends on the nature of the material. Find the dimensions of
κ.

4. Consider a sky-diver who falls from rest under the influence of gravity, with air-drag
proportional to velocity.

a) Introduce a characteristic time tc and length ℓc , and use them to define dimen-
sionless variables.
b) Write the DE for the velocity as a function of time, using dimensionless variables,
and solve to find the velocity as a function of time. Then find distance as a
function of time, also using dimensionless variables.
c) Derive a DE for the velocity regarded as a function of distance fallen, using
dimensionless variables. Solve to obtain distance fallen as a function of velocity.
d) Determine how far the sky-diver has fallen and for how long, at the instant when
9
his velocity is 10 of the terminal velocity. Give your answer in two forms,
(i) in terms of dimensionless distance and time, and
(ii) in terms of actual distance and time, assuming the skydiver has a mass of 80
kg and a drag coefficient α = 10 kg s−1 .
Problem Set 2 143

e) Use b) to write asymptotic forms for the physical velocity and distance fallen in
the limiting cases (i) 0 < t ≪ tc and (ii) t ≫ tc .

∗5. a) Repeat #4 (all parts), assuming that air-drag is proportional to the square of the
velocity. Let β denote the new-air drag coefficient.
b) Does an object with air-drag of the form βv 2 approach terminal velocity more
rapidly than an object with air-drag of the form αv?

6. The goal is to obtain information about the period of a satellite in a circular orbit
around a planet. The relevant physical quantities are the period P , the planetary
mass M, the satellite mass m, the radius of the orbit r, and, since the process is
governed by gravity, by the gravitational constant G.

a) Set up the dimensional matrix and determine its rank.

b) Find a complete set of dimensionless quantities.
c) Deduce the relation between P and r for fixed masses m and M. This result is
one of Kepler’s laws for planetary motion.

7. a) List the three physical variables involved in the escape velocity problem. Set up
the dimensional matrix to show that there is only one dimensionless variable.
2
Hence write down an expression for vesc .
b) Suppose that planet X has radius 6 times that of earth and that the acceleration
due to gravity on the surface is twice that on earth. How does the escape velocity
for planet X compare to that for the earth?

8. An explosive charge releases energy E in a soil medium of mass density ρ. Assuming

that the only other important quantity is the acceleration g of gravity, investigate the
length scale ℓ of the resulting crater. If it is desired to double the length scale of the
crater, by how many factors must the energy be increased?

9. Consider a star as being a fluid sphere held together by its own gravity. A star may
undergo vibrations, the most important being those in which all particles of the star
execute simple harmonic motions which are in phase with each other, with frequency
ν. Assume that ν could depend on only the mass density ρ of the star, its radius
r and the gravitational constant G (the viscosity of the star is not important here).
Investigate ν as a function of ρ, r and G.

10. A little cooking problem ... how does the cooking time of a roast depend on the size
of the roast?
The physical quantities are the cooking time ∆t, the volume of the roast V , the thermal
diffusivity κ of the roast (see #3), the initial temperature difference ∆Ti = Ti − T0 and
the final temperature difference ∆Tf = Tf − T0 . Here T0 is the oven temperature, and
Ti , Tf denote the initial and final temperature of the centre of the roast, respectively.
Show that ∆t is proportional to V 2/3 .
144 Problem Set 2

11. Consider the motion of gravity waves in deep water (the “ocean swell”). The physical
parameters are the speed of the wave v, the wavelength of the wave λ, the density
√ of
water ρ and the acceleration g due to gravity. Show that v is proportional to λg.

12 Referring to #11, if the water is shallow, we must add the depth h to the list of essential
parameters. Show that in this case

p
v = f λh λg,

where f is an arbitrary function of one variable.

∗13. Suppose we are interested in the power P necessary to keep a ship of length L moving
at constant velocity U. Energy must be supplied to replace energy wasted in making
waves, and which is lost because of water viscosity (friction), and a worker in the field
may be led to assume that
P = P (L, U, g, ρ, η) , (1)
where g is the acceleration of gravity, ρ is the mass density of water, and η is its
viscosity. Assuming that (1) is correct, prove that

P = ρ L2 U 3 f (F r, Re),

where f is an unknown function of two variables, F r is the “Froude number”, and Re

is the “Reynolds number”:

F r ≡ U 2 /(Lg) , Re ≡ U L ρ/η .

∗14. Consider steady nonturbulent incompressible flow of a fluid, of mass density ρ and
viscosity η, in a cylindrical pipe, of length L and radius R. The pressure difference
∆P between the ends should depend on only L, R, ρ, η, and the maximum speed U of
the flow. If this is so, show that

∆P = 21 ρ U 2 f (ReL , ReR ) , (2)

where f is an unknown function of two variables, and the two Reynolds numbers are
defined by
ReL ≡ U L ρ/η , ReR ≡ U R ρ/η .
Suggest a possible great simplification of the result (2) if L/R is sufficiently large.

Acknowledgement: Thanks to F.O. Goodman & G. Tenti for many of these problems.
Problem Set 3 145

Problem Set 3
Second Order Linear Differential Equations
1. Find the general solution of each homogeneous linear DE:

(i) y ′′ + y ′ − 2y = 0 (ii) y ′′ + 4y ′ + 4y = 0
(iii) y ′′ + 2y ′ + 5y = 0 (iv) y ′′ − 6y ′ + 10y = 0 .
2. Find the general solution of the following inhomogeneous linear DEs.
(i) y ′′ + y ′ − 6y = 6t
(ii) y ′′ + y ′ − 6y = 5 cos t
(iii) y ′′ + y ′ − 6y = 5e4t
3. a) Find the unique solution of the initial value problems

(i) y ′′ + 4y ′ + 4y = 0, y(0) = 0, y ′(0) = 1

(ii) y ′′ + 2y ′ + 5y = 0, y(0) = 0 , y ′(0) = 1.
b) Give a qualitative sketch of the solutions in a).
4. Write the general solution of the linear DE

y ′′ + ω 2 y = b,

where ω and b are constants, by inspection.

5. A particle undergoes simple harmonic motion (SHM) described by

y ′′ + ω 2 y = 0,

with initial conditions y(0) = y0 , y ′ (0) = v0 . Show that the amplitude of the SHM is
q
v2
ymax = y02 + ω02 .

6. Find the general solution of the DE

d2 y dy
2
+ 2k + b2 y = 0,
dt dt
where k and b are constants.
Note: There are a number of different cases, depending on the values of k and b.
7. Find the general solution of the DE

y ′′ + y ′ − 6y = αert ,

where r and α are constants. Which values of r have to be treated as special cases?
146 Problem Set 3

8. Find the general solution of the DE

y ′′ + 2y ′ + 2y = α cos ωt

where α and ω are constants.

9. a) Find the general solution of the DE

y ′′ + ω02 y = α cos ωt,

where ω0 , α and ω are constants. Which values(s) of ω have to be treated as

special cases?
b) Find the unique solution of the DE in a) with ω 2 6= ω02 , subject to the initial
condition y(0) = 0, y ′(0) = 0,
c) Show that the solution in b) can be written in the form

y = A(t) sin 12 (ω0 + ω)t ,

 

where  
2α 1
A(t) = ω02 −ω 2
sin (ω0 − ω)t .
2
Give a qualitative sketch of the graph of y(t) in the case where ω0 − ω is small
compared to ω0 . If you want to use specific values, use ω0 = 11, ω = 9, but don’t
try to sketch the curve exactly.
d) Show that the unique solution of the DE in a) with ω 2 = ω02 , subject to the initial
condition y(0) = 0, y ′(0) = 0 is
α
y= t sin ω0 t.
2ω0
Give a qualitative sketch of the graph.

10. The homogeneous linear DE

y ′′ + py ′ + qy = 0 (1)
has y(t) = 0 as an equilibrium solution. In studying a physical system one is interested
under what conditions the general solution will approach the equilibrium solution as
time t → ∞. Show that the general solution of (1) approaches the equilibrium solution
as t → +∞ if and only if p > 0 and q > 0.

11. a) An object of mass 2 kg extends a spring 0.1 m from its equilibrium position.
Calculate the spring constant k in Newtons per metre. Assuming that the mass
moves with no damping and no driving force, calculate the period of the resulting
Simple Harmonic Motion (SHM).
b) A resistive medium exerts a damping force of 200 Newtons when acting on the
above object moving with a velocity of 10 metres per second. Calculate the
damping constant in kilograms per second.
Problem Set 3 147

c) Consider a damped mass–spring system with spring as in a) and damping as in

b). Will the system undergo damped oscillations?
Note: A force of 1 Newton will give a mass of 1 kilogram an acceleration of
1 metres/sec2 .
12. A mechanical system undergoes damped oscillations governed by the DE

y ′′ + 2λy ′ + w02 y = 0.

Suppose that ζ = wλ0 = 10−5. How many oscillations will take place in the time interval
during which the amplitude decays by 1%? In this situation the motion of the system
approximates SHM over a restricted time interval.
13. Consider the dimensionless DE

Y ′′ + 2ζY ′ + Y = 0,

with initial conditions Y (0) = 1, Y ′ (0) = −V0 , where ′ is differentiation with respect
to a dimensionless time variable τ , Y is a dimensionless displacement, and ζ is a
dimensionless constant satisfying ζ ≥ 1, so that oscillations do not occur.

(i) One expects that if V0 is sufficiently large, then the mass will pass through the
equilibrium position before coming to rest. Find the restriction on V0 that will
ensure that this happens, in the case of critical damping ζ = 1. Also find the
time τzero at which Y is zero.
(ii) Find the time τcrit at which Y attains its minimum value Ymin. Show that τcrit
satisfies
τcrit = τzero + 1,
and that
Ymin = −(V0 − 1)e−τcrit .
(iii) Sketch the graphs of the displacement Y (τ ) and the velocity Y ′ (τ ), and label τzero ,
τcrit , V0 and Ymin.

14. Referring to #13(i) show that in the case ζ > 1, the restriction on V0 is
p
V0 > ζ + ζ 2 − 1

15. We wish to find a particular solution of the DE

Y ′′ + 2ζY ′ + Y = cos(Ωτ ). (2)

Consider the complex DE

Z ′′ + 2ζZ ′ + Z = eiΩτ (3)
The DE (2) is the real part of this DE. Consider a trial function

Z = Rei(Ωτ −φ) , (4)

where R and φ are real constants.

148 Problem Set 3

(i) Show that (4) is a solution of (3) if and only if

R = [(1 − Ω2 )2 + 4ζ 2 Ω2 ]−1/2 , (5)

and
cos φ = R(1 − Ω2 ), sin φ = 2RζΩ (6)
(ii) Hence conclude that
Y = R cos(Ωτ − φ),
where R and φ are given by (5) and (6), is a particular solution of (2).

16. The charge Q(t) on the capacitor in the electrical circuit shown satisfies
d2 Q dQ 1
L 2
+R + Q = V (t),
dt dt C

L = 0.5 henrys is the coil’s inductance

R = 6 ohms is the resistor’s resistance
C = 0.02 farads is the capacitor’s capacitance.

and V (t) is the applied voltage.

V (t) L

(i) Is the circuit oscillatory?

(ii) If V (t) = 24 sin 10t volts and Q(0) = 0 = Q′ (0), find Q(t).
(iii) Sketch the transient solution, the steady state solution and the full solution Q(t).

17. Two mixing tanks of volume V are coupled together with flow rates as shown. Let
m1 (t) and m2 (t) denote the mass of chemical in each tank at time t.

(i) Write down the first order differential equations satisfied by m1 and m2 . Use a di-
mensionless time variable τ defined in the usual way in terms of the characteristic
time
1 V
tc = p .
k(1 + k) f
Problem Set 3 149

f kf

(k + 1)f f
m1 (t) m2 (t)

(ii) Show that both m1 and m2 satisfy the second order DE

m′′ + 2qm′ + (q 2 − 1)m = (q 2 − 1)V cin

where r
1+k
q= ,
k
cin is the inflow concentration, and ′ denotes differentiation with respect to τ .
(iii) Assume that both tanks are initially filled with pure water. Show that

(q 2 − 1)V cin  −(q−1)τ

− e−(q+1)τ

m1 (τ ) − m2 (τ ) = e
2q

(iv) What is the long term behaviour of the physical system?

18. A car suspension system may be modelled by a mass-spring-damper model as shown;

the wheel is assumed to ride on a sinusoidally-undulating road, that is the wheel’s
position w(t) is assumed to be given by w(t) = w0 cos(wt) at time t.

a) Show, using Newton’s Second Law, that a suitable differential equation for de-
scribing the motion of the mass m is

mẍ + c(ẋ − ẇ) + k(x − w − ℓ) = 0, (∗)

where ℓ is the equilibrium length of the spring, and where ξ˙ ≡ dξ/dt.

b) Show that the problem may be solved by considering it to be the same as that
discussed in the lectures, but with a modified applied force; that is, show that (∗)
may be written as
mz̈ + cż + kz = F (t),
with suitable choices of z and F . Find F (t) in the form F (t) = F0 cos(ωt + ψ).
150 Problem Set 3

body of car

c (damper)

k
x(t)

wheel

w(t)

road surface mean level of road

19. An electric charge q of mass m moves with velocity v = (v1 , v2 , v3 ) in a region of space
where there is a magnetic field B = (0, 0, B) of constant strength. According to the
laws of physics, the motion of the charge is given by the following equations:
dv1 qB dv2 qB dv3
m = v2 , m = − v1 , m = 0.
dt c dt c dt
where c is the velocity of light in a vacuum. Assume an initial condition of the form
v(0) = (u1 , u2 , u3 ).

a) Eliminate v2 from the first two equations and get a single second order DE for v1 .
qB
b) Show that the quantity ωc = mc has the dimensions of a frequency, thus justifying
its name, which is the “cyclotron frequency” (or “gyrofrequency”).
c) Solve the resulting DE for v1 , and then find v2 .
d) Identify the curve representing the path of the particle.
Problem Set 3 151

Answers:
1. (i) y = c1 ex + c2 e−2x (ii) y = (c1 + c2 x)e−2x
(iii) y = (c1 cos 2x + c2 sin 2x)e−x (iv) y = (c1 cos x + c2 sin x)e3x .

2. (i) yp (t) = −t− 16 (ii) 7

yp (t) = − 10 1
cos t+ 10 sin t (iii) yp (t) = 5 4t
14
e .
In all cases, the general solution is

y = c1 e2t + c2 e−3t + yp (t).

152 Problem Set 4

Problem Set 4
Laplace Transforms and Differential Equations
1. Calculate the Laplace transform Y = L(y) of each function y, using the definition.
Give the domain of validity. Use complex variables for the trig. functions.

(i) y(t) = t2 (ii) y(t) = te−2t (iii) y(t) = t cos πt (iv) y(t) = t sin πt

2. Use the fact that the Laplace transform operator L is linear, and the results from #1
to write down the Laplace transform of y(t) = t(2t + 3e−2t + π sin πt).
1
3. Knowing L(eαt ) = s−α
for s > Re(α), find the Laplace transforms of

L(eat cos bt) and L(eat sin bt).

n!
4. Show that L[tn ] = , for s > 0, where n is a non-negative integer.
sn+1
Hint: See #9 on the Review Problem Set.

5. Use the Formula Sheet to find the inverse Laplace transforms of

1 1 1
(i) Y (s) = (ii) Y (s) = 2 , (iii) Y (s) = ,
(s + a)(s + b) (s + a2 )(s2 + b2 ) (s + a)(s2 + b2 )
where a and b are real constants with a 6= b.

6. Find each inverse Laplace transform using the first shift theorem:

If L−1 [(F (s)] = f (t) then L−1 [F (s − c)] = ect f (t).

     
5 s − 1 1
(i) L−1 (ii) L−1 (iii) L−1 2
s(s2 + 4s + 5) 2s2 + 5s + 2 s − 4s + 8
7. Find the Laplace transform of each piecewise defined function f (t) in two ways:

(a) by directly using the definition,

(b) by expressing f (t) in terms of the Heaviside step function, H(t).
Problem Set 4 153

(i) (ii) (iii)

y y = f (t) y y = f (t) y y = f (t)
1 1 1

t t t
0 a b 0 b 2b 0 b 2b

−1

8. Find the inverse Laplace transform using the second shift theorem, and sketch the
graph of the resulting function:

If L−1 [F (s)] = f (t), then L−1 [e−cs F (s)] = H(t − c)f (t − c).
 −3s   −πs   −πs 
−1 e −1 e −1 e
(i) L (ii) L 2
(iii) L .
s+2 s +1 s2 − 1

9. Derive the equation

L[f ′′ (t)] = s2 L[f (t)] − sf (0) − f ′ (0)
for a suitably restricted function f (t). State the hypothesis on f (t).
Hint: Very little work is required.

10. Solve the following initial value problems using the Formula Sheet

(i) y ′ + 2y = e−3t , y(0) = 5 (ii) y ′ + 2y = 4 cos 2t; y(0) = 1.

(iii) y ′′ − y ′ − 2y = 0; y(0) = 1, y ′ (0) = 0.
(iv) y ′′ + y = 3 sin 2t; y(0) = 6, y ′(0) = 1.

11. A mixing tank with constant volume V0 and flow rate k is initially filled with pure
water. If the inflow concentration is a constant cin for 0 ≤ t < T , and is then zero
afterwards, calculate the mass of chemical in the tank at time 2T .

12. Consider a population y(t) which undergoes exponential growth with growth factor k.
Starting at time t = 0, the population is harvested at a constant rate h (number per
unit time), for a period of time T .

(i) By introducing a dimensionless time τ and a suitably scaled population function

z(τ ), show that this system is governed by the DE

z ′ − z = −1 + H(τ − b),

where b = kT and H is the Heaviside step function.

(ii) If the initial population is y0 , find the scaled population z at time τ . Let z0 be
the initial value of z.
154 Problem Set 4

(iii) Sketch the family of dimensionless solutions with z0 as parameter. What happens
if z0 < 1 − e−b ? What is the physical interpretation?
(iv) For what values of y0 does the population y fall below y0 during the interval
0 < t < T ? In this case, at what time does y recover to its initial value?

13. Consider the system described by the dimensionless DE

y ′ + y = g(t),

where the input function g(t) is the saw-tooth function in #7(iii), and the initial
condition is y(0) = 0.

(i) Based on the graph of the input function, make an educated guess as to the graph
of the response y(t).
(ii) Show that the response at time t is

y(t) = 1b [f (t) − 2H(t − b)f (t − b) + H(t − 2b)f (t − 2b)],

where f (t) = t − 1 + e−t .

(iii) Confirm your “educated guess” in (i) by appropriate analysis. In particular show
that the maximum response occurs at time tmax = ln(2eb − 1), and is given by
ymax = 1b (2b − tmax ).

14. Consider the undamped oscillator DE with a driving force that is a rectangular pulse
of duration b:
y ′′ + y = 1 − H(τ − b),
where H is the Heaviside step function.

(i) Use the Laplace transform to show that the unique solution satisfying the initial
conditions y(0) = 0 = y ′ (0) is

y(τ ) = 1 − cos τ − H(τ − b)[1 − cos(τ − b)],

(ii) Show that for τ > b, the response y(τ ) can be expressed as

y(τ ) = A sin(τ − δ),

i.e. simple harmonic motion. Express the amplitude A and phase δ in terms of
the pulse duration b.
(iii) Sketch the full response for b = 2π, 3π and 2π + ǫ, where 0 < ǫ ≪ 1.
(iv) For what values(s) of b is the amplitude of the long-term response a maximum?
zero?
Problem Set 4 155

Answers:
2 1
i) L[t2 ] = s3
, s>0 ii) L[te−2t ] = (s+2)2
, s > −2
1.
2 2
iii) L[t cos πt] = (ss2 +π
−π
2 )2 iv) L[t sin πt] = 2πs
(s2 +π 2 )2
.
h i
1 1
5. i) L (s+a)(s+b) = a−b (e−bt − e−at )
1
s−1
= e−2t − 12 e− 2 t
 
6. ii) L−1 2s2 +5s+2


0,
h
e−3s
i if t < 3
8. i) L−1 s+2
= H(t − 3)e−2(t−3) = .

e−2(t−3) , if t ≥ 3


10. i) y(t) = 6e−2t − e−3t iv) y(t) = 6 cos t + 3 sin t − sin 2t.
156 Problem Set 5

Problem Set 5
Linear Vector DEs
1. In each case verify that the given vector-valued function satisfies the vector DE x′ =
Ax.
   
2 6 −t −2
i) A = ; x(t) = e
−2 −5 1
   
1 5 −t 5 cos t
ii) A = ; x(t) = e
−1 −3 −2 cos t − sin t
     
3 −4 t 1 2
iii) A = ; x(t) = e +t .
1 −1 0 1

2. a) Give a description of the eigenvalue method for solving a vector DE.

b) Find the general solution of each vector DE using the eigenvalue method, referring
to your explanation in a).
   
′ −3 2 ′ −1 −4
i) x = x ii) x = x
2 −3 1 −1
   
−1 2 ′ −1 −1
iii) x = x iv) x = x.
0 −1 −1 −1

3. a) Give a description of the Laplace transform method for finding the fundamental
matrix of a linear vector DE.
b) Find the fundamental matrix for each DE in #2, referring to your description in
#3 a).
c) Verify that Φ(t) has the properties

Φ(0) = I, Φ(−t) = [Φ(t)]−1 .

Find the unique solution for each DE in #2 that satisfies each given initial con-
ditions (i.e. for each DE, two initial conditions are given):
       
1 1 1 π

1
i) x(0) = ; x(ln 2) = ii) x(0) = ; x 4
=
1 0 1 0
       
1 1 1 1
iii) x(0) = ; x(1) = iv) x(0) = ; x(ln 2) =
1 0 1 0
4. Give a qualitative sketch of the orbits for each DE in #2.

5. a) Give a description of the variation of parameters method for solving an inhomo-

geneous linear vector DE
x′ = Ax + f(t). (*)
Explain how to find a particular solution, and how to find the solution which
satisfies an arbitrary initial condition x(0) = a.
Problem Set 5 157

b) Find a particular solution of the DE (*) for the 2 × 2 matrices given in #2, and
the input functions given below:
   
1 2 cos 2t
(i) f(t) = e−3t (ii) f(t) =
1 sin 2t
   
−t 0 1
(iii) f(t) = e (iv) f(t) = cos t .
1 −1
6. Consider the coupled constant volume mixing tank system as shown with inflow con-
centration cin (t), with state vector
 
m1
x= ,
m2
where m1 and m2 denote the mass of chemical in tanks 1 and 2 respectively. Let V be
the volume of each tank.
1
f 3
f

Tank 1 Tank 2

4
3
f f

a) Show that the vector DE governing the state of the system is

   
′ −4 1 3cin V
x = Ax + f, A= , f= ,
4 −4 0
f
where ′ denotes differentiation with respect to τ = 3V
t. (The factor “3” is included
to simplify the numbers.)
b) Find the fundamental matrix for the homogeneous DE.
c) Find the solution for the following initial conditions, assuming cin (t) = 0:
i) m1 (0) = M, m2 (0) = 0 ii) m1 (0) = 13 M, m2 (0) = 23 M iii) m1 (0) = 0, m2 (0) = M.
In each case give a qualitative sketch of the mass functions m1 (τ ) and m2 (τ ) on
the same axes. Use the graphs to give a physical interpretation of the behaviour
of the system, discussing whether the mass of chemical in each tank is increasing
or decreasing and whether the masses are ever equal.
d) Referring to c), in which case does the system flush most rapidly, i.e. in which
case does the total mass in the system tend to zero most rapidly? First make an
“educated guess”, and then give a mathematical analysis.
158 Problem Set 5

e) Sketch typical orbits of the DE in R2 , subject to the restriction m1 ≥ 0, m2 ≥ 0.

(i) Mark the orbits corresponding to the three solutions in part c) on your sketch.
(ii) Consider an initial state with m2 (0) < m1 (0). Use the sketch to describe the
future evolution of the system.
(iii) Do the same for an initial state with m2 (0) > 4m1 (0).
f) Find the solution of the non-homogeneous DE assuming cin (t) = c, a constant,
and an arbitrary initial state x(0) = a. What is the asymptotic behaviour as
t → +∞?
Problem Set 5 159

Answers:
   
−t 1 −5t 1
2. b) i) x(t) = c1 e + c2 e .
1 −1
    
−t cos 2t −2 sin 2t
ii) x(t) = e c1 1 + c2 .
2
sin 2t cos 2t
    
−t 1 0
iii) x(t) = e (c1 + 2c2 t) + c2 .
0 1
   
1 −2t 1
iv) x(t) = c1 + c2 e .
−1 1

3. b)
e−t + e−5t e−t − e−5t
   
1 cos 2t −2 sin 2t
i) Φ(t) = ii) Φ(t) = e−t 1
2e−t − e−5t e−t + e−5t 2
sin 2t cos 2t
 −2t
e + 1 e−2t − 1
  
−t 1 2t 1
iii) Φ(t) = e iv) Φ(t) = 2 −2t .
0 1 e − 1 e−2t + 1